diff options
author | Quincey Koziol <koziol@hdfgroup.org> | 2016-12-02 16:07:04 (GMT) |
---|---|---|
committer | Quincey Koziol <koziol@hdfgroup.org> | 2016-12-02 16:07:04 (GMT) |
commit | 64a339183f0e4532597744351548308203e993c8 (patch) | |
tree | ddfff2f15c0f3a01782f191a847665c6eceadd42 /src | |
parent | 5d7d029b97b36d1c380cef82d637342921bf3a1d (diff) | |
download | hdf5-64a339183f0e4532597744351548308203e993c8.zip hdf5-64a339183f0e4532597744351548308203e993c8.tar.gz hdf5-64a339183f0e4532597744351548308203e993c8.tar.bz2 |
Bring SWMR support in to the main development branch. (Finally!) More tests
and the tool and API wrappers will be coming in over the weekend.
Diffstat (limited to 'src')
90 files changed, 7568 insertions, 2368 deletions
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt index a4c6367..55de5ea 100644 --- a/src/CMakeLists.txt +++ b/src/CMakeLists.txt @@ -47,6 +47,7 @@ set (H5AC_SOURCES ${HDF5_SRC_DIR}/H5ACdbg.c ${HDF5_SRC_DIR}/H5AClog.c ${HDF5_SRC_DIR}/H5ACmpio.c + ${HDF5_SRC_DIR}/H5ACproxy_entry.c ) set (H5AC_HDRS @@ -73,6 +74,8 @@ set (H5B2_SOURCES ${HDF5_SRC_DIR}/H5B2dbg.c ${HDF5_SRC_DIR}/H5B2hdr.c ${HDF5_SRC_DIR}/H5B2int.c + ${HDF5_SRC_DIR}/H5B2internal.c + ${HDF5_SRC_DIR}/H5B2leaf.c ${HDF5_SRC_DIR}/H5B2stat.c ${HDF5_SRC_DIR}/H5B2test.c ) @@ -209,6 +212,7 @@ set (H5FA_SOURCES ${HDF5_SRC_DIR}/H5FAdblkpage.c ${HDF5_SRC_DIR}/H5FAdblock.c ${HDF5_SRC_DIR}/H5FAhdr.c + ${HDF5_SRC_DIR}/H5FAint.c ${HDF5_SRC_DIR}/H5FAstat.c ${HDF5_SRC_DIR}/H5FAtest.c ) @@ -231,6 +235,7 @@ set (H5FD_SOURCES ${HDF5_SRC_DIR}/H5FDsec2.c ${HDF5_SRC_DIR}/H5FDspace.c ${HDF5_SRC_DIR}/H5FDstdio.c + ${HDF5_SRC_DIR}/H5FDtest.c ${HDF5_SRC_DIR}/H5FDwindows.c ) @@ -271,6 +276,7 @@ set (H5FS_SOURCES ${HDF5_SRC_DIR}/H5FS.c ${HDF5_SRC_DIR}/H5FScache.c ${HDF5_SRC_DIR}/H5FSdbg.c + ${HDF5_SRC_DIR}/H5FSint.c ${HDF5_SRC_DIR}/H5FSsection.c ${HDF5_SRC_DIR}/H5FSstat.c ${HDF5_SRC_DIR}/H5FStest.c @@ -135,6 +135,7 @@ static const char *H5AC_entry_type_names[H5AC_NTYPES] = "fixed array data block pages", "superblock", "driver info", + "proxy entry", "test entry" /* for testing only -- not used for actual files */ }; @@ -604,6 +605,52 @@ done: /*------------------------------------------------------------------------- + * Function: H5AC_evict + * + * Purpose: Evict all entries except the pinned entries + * in the cache. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Vailin Choi + * Dec 2013 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_evict(H5F_t *f, hid_t dxpl_id) +{ + hbool_t log_enabled; /* TRUE if logging was set up */ + hbool_t curr_logging; /* TRUE if currently logging */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity check */ + HDassert(f); + HDassert(f->shared); + HDassert(f->shared->cache); + + /* Check if log messages are being emitted */ + if(H5C_get_logging_status(f->shared->cache, &log_enabled, &curr_logging) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, "unable to get logging status") + + /* Evict all entries in the cache except the pinned superblock entry */ + if(H5C_evict(f, dxpl_id) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTFREE, FAIL, "can't evict cache") + +done: + + /* If currently logging, generate a message */ + if(curr_logging) + if(H5AC__write_evict_cache_log_msg(f->shared->cache, ret_value) < 0) + HDONE_ERROR(H5E_CACHE, H5E_LOGFAIL, FAIL, "unable to emit log message") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5AC_evict() */ + + +/*------------------------------------------------------------------------- * Function: H5AC_expunge_entry * * Purpose: Expunge the target entry from the cache without writing it @@ -1004,6 +1051,84 @@ done: /*------------------------------------------------------------------------- + * Function: H5AC_mark_entry_clean + * + * Purpose: Mark a pinned entry as dirty. The target + * entry MUST be pinned. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * 7/23/16 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_mark_entry_clean(void *thing) +{ +#if H5AC__TRACE_FILE_ENABLED + char trace[128] = ""; + FILE * trace_file_ptr = NULL; +#endif /* H5AC__TRACE_FILE_ENABLED */ + hbool_t log_enabled; /* TRUE if logging was set up */ + hbool_t curr_logging; /* TRUE if currently logging */ + H5AC_info_t *entry_ptr = NULL; /* Pointer to the cache entry */ + H5C_t *cache_ptr = NULL; /* Pointer to the entry's associated metadata cache */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity check */ + HDassert(thing); + +#if H5AC__TRACE_FILE_ENABLED + /* For the mark pinned or protected entry clean call, only the addr + * is really necessary in the trace file. Write the result to catch + * occult errors. + */ + if(NULL != (trace_file_ptr = H5C_get_trace_file_ptr_from_entry(thing))) + sprintf(trace, "%s 0x%lx", FUNC, + (unsigned long)(((H5C_cache_entry_t *)thing)->addr)); +#endif /* H5AC__TRACE_FILE_ENABLED */ + + entry_ptr = (H5AC_info_t *)thing; + cache_ptr = entry_ptr->cache_ptr; + + /* Check if log messages are being emitted */ + if(H5C_get_logging_status(cache_ptr, &log_enabled, &curr_logging) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, "unable to get logging status") + +#ifdef H5_HAVE_PARALLEL +{ + H5AC_aux_t *aux_ptr; + + aux_ptr = (H5AC_aux_t *)H5C_get_aux_ptr(cache_ptr); + if((!entry_ptr->is_dirty) && (!entry_ptr->is_protected) && + (entry_ptr->is_pinned) && (NULL != aux_ptr)) + if(H5AC__log_cleaned_entry(entry_ptr) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTMARKCLEAN, FAIL, "can't log cleaned entry") +} +#endif /* H5_HAVE_PARALLEL */ + + if(H5C_mark_entry_clean(thing) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTMARKCLEAN, FAIL, "can't mark pinned or protected entry clean") + +done: +#if H5AC__TRACE_FILE_ENABLED + if(trace_file_ptr) + HDfprintf(trace_file_ptr, "%s %d\n", trace, (int)ret_value); +#endif /* H5AC__TRACE_FILE_ENABLED */ + + /* If currently logging, generate a message */ + if(curr_logging) + if(H5AC__write_mark_clean_entry_log_msg(cache_ptr, entry_ptr, ret_value) < 0) + HDONE_ERROR(H5E_CACHE, H5E_LOGFAIL, FAIL, "unable to emit log message") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5AC_mark_entry_clean() */ + + +/*------------------------------------------------------------------------- * Function: H5AC_move_entry * * Purpose: Use this function to notify the cache that an object's @@ -2619,3 +2744,67 @@ done: FUNC_LEAVE_NOAPI(ret_value) } /* end H5AC_reset_ring() */ + +/*------------------------------------------------------------------------- + * Function: H5AC_remove_entry() + * + * Purpose: Remove an entry from the cache. Must be not protected, pinned, + * dirty, involved in flush dependencies, etc. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_remove_entry(void *_entry) +{ + H5AC_info_t *entry = (H5AC_info_t *)_entry; /* Entry to remove */ + H5C_t *cache = NULL; /* Pointer to the entry's associated metadata cache */ +#if H5AC__TRACE_FILE_ENABLED + char trace[128] = ""; + FILE * trace_file_ptr = NULL; +#endif /* H5AC__TRACE_FILE_ENABLED */ + hbool_t log_enabled; /* TRUE if logging was set up */ + hbool_t curr_logging; /* TRUE if currently logging */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(entry); + cache = entry->cache_ptr; + HDassert(cache); + +#if H5AC__TRACE_FILE_ENABLED + /* For the remove entry call, only the addr is really necessary + * in the trace file. Also write the result to catch occult errors. + */ + if(NULL != (trace_file_ptr = H5C_get_trace_file_ptr_from_entry(entry))) + sprintf(trace, "%s 0x%lx", FUNC, (unsigned long)(entry->addr)); +#endif /* H5AC__TRACE_FILE_ENABLED */ + + /* Check if log messages are being emitted */ + if(H5C_get_logging_status(cache, &log_enabled, &curr_logging) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTGET, FAIL, "unable to get logging status") + + /* Remove the entry from the cache*/ + if(H5C_remove_entry(entry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "can't remove entry") + +done: +#if H5AC__TRACE_FILE_ENABLED + if(trace_file_ptr) + HDfprintf(trace_file_ptr, "%s %d\n", trace, (int)ret_value); +#endif /* H5AC__TRACE_FILE_ENABLED */ + + /* If currently logging, generate a message */ + if(curr_logging) + if(H5AC__write_remove_entry_log_msg(cache, entry, ret_value) < 0) + HDONE_ERROR(H5E_CACHE, H5E_LOGFAIL, FAIL, "unable to emit log message") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5AC_remove_entry() */ + diff --git a/src/H5AClog.c b/src/H5AClog.c index 980b105..1cdaa00 100644 --- a/src/H5AClog.c +++ b/src/H5AClog.c @@ -206,6 +206,51 @@ done: /*------------------------------------------------------------------------- + * Function: H5AC__write_evict_cache_log_msg + * + * Purpose: Write a log message for eviction of cache entries. + * + * Return: Success: SUCCEED + * Failure: FAIL + * + * Programmer: Dana Robinson + * Sunday, March 16, 2014 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC__write_evict_cache_log_msg(const H5AC_t *cache, + herr_t fxn_ret_value) +{ + char msg[MSG_SIZE]; + herr_t ret_value = SUCCEED; + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(cache); + + /* Create the log message string */ + HDsnprintf(msg, MSG_SIZE, +"\ +{\ +\"timestamp\":%lld,\ +\"action\":\"evict\",\ +\"returned\":%d\ +},\n\ +" + , (long long)HDtime(NULL), (int)fxn_ret_value); + + /* Write the log message to the file */ + if(H5C_write_log_message(cache, msg) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, "unable to emit log message") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* H5AC__write_evict_cache_log_msg() */ + + +/*------------------------------------------------------------------------- * Function: H5AC__write_expunge_entry_log_msg * * Purpose: Write a log message for expunge of cache entries. @@ -404,6 +449,53 @@ done: /*------------------------------------------------------------------------- + * Function: H5AC__write_mark_clean_entry_log_msg + * + * Purpose: Write a log message for marking cache entries as clean. + * + * Return: Success: SUCCEED + * Failure: FAIL + * + * Programmer: Quincey Koziol + * Saturday, July 23, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC__write_mark_clean_entry_log_msg(const H5AC_t *cache, const H5AC_info_t *entry, + herr_t fxn_ret_value) +{ + char msg[MSG_SIZE]; /* Log message buffer */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(cache); + HDassert(entry); + + /* Create the log message string */ + HDsnprintf(msg, MSG_SIZE, +"\ +{\ +\"timestamp\":%lld,\ +\"action\":\"clean\",\ +\"address\":0x%lx,\ +\"returned\":%d\ +},\n\ +" + , (long long)HDtime(NULL), (unsigned long)entry->addr, (int)fxn_ret_value); + + /* Write the log message to the file */ + if(H5C_write_log_message(cache, msg) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, "unable to emit log message") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* H5AC__write_mark_clean_entry_log_msg() */ + + +/*------------------------------------------------------------------------- * Function: H5AC__write_move_entry_log_msg * * Purpose: Write a log message for moving a cache entry. @@ -870,3 +962,51 @@ done: FUNC_LEAVE_NOAPI(ret_value) } /* H5AC__write_set_cache_config_log_msg() */ + +/*------------------------------------------------------------------------- + * Function: H5AC__write_remove_entry_log_msg + * + * Purpose: Write a log message for removing a cache entry. + * + * Return: Success: SUCCEED + * Failure: FAIL + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC__write_remove_entry_log_msg(const H5AC_t *cache, const H5AC_info_t *entry, + herr_t fxn_ret_value) +{ + char msg[MSG_SIZE]; + herr_t ret_value = SUCCEED; + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(cache); + HDassert(entry); + + /* Create the log message string */ + HDsnprintf(msg, MSG_SIZE, +"\ +{\ +\"timestamp\":%lld,\ +\"action\":\"remove\",\ +\"address\":0x%lx,\ +\"returned\":%d\ +},\n\ +" + , (long long)HDtime(NULL), (unsigned long)entry->addr, + (int)fxn_ret_value); + + /* Write the log message to the file */ + if(H5C_write_log_message(cache, msg) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, "unable to emit log message") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* H5AC__write_remove_entry_log_msg() */ + diff --git a/src/H5ACmpio.c b/src/H5ACmpio.c index 89a8f9d..570783a 100644 --- a/src/H5ACmpio.c +++ b/src/H5ACmpio.c @@ -805,6 +805,63 @@ done: /*------------------------------------------------------------------------- * + * Function: H5AC__log_cleaned_entry() + * + * Purpose: Treat this operation as a 'clear' and remove the entry + * from both the cleaned and dirtied lists if it is present. + * Reduces the dirty_bytes count by the size of the entry. + * + * Return: Non-negative on success/Negative on failure. + * + * Programmer: Quincey Koziol + * 7/23/16 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC__log_cleaned_entry(const H5AC_info_t *entry_ptr) +{ + H5AC_t * cache_ptr; + H5AC_aux_t * aux_ptr; + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Sanity check */ + HDassert(entry_ptr); + HDassert(entry_ptr->is_dirty == FALSE); + cache_ptr = entry_ptr->cache_ptr; + HDassert(cache_ptr != NULL); + aux_ptr = (H5AC_aux_t *)H5C_get_aux_ptr(cache_ptr); + HDassert(aux_ptr != NULL); + HDassert(aux_ptr->magic == H5AC__H5AC_AUX_T_MAGIC); + + if(aux_ptr->mpi_rank == 0) { + H5AC_slist_entry_t *slist_entry_ptr; + haddr_t addr = entry_ptr->addr; + + /* Sanity checks */ + HDassert(aux_ptr->d_slist_ptr != NULL); + HDassert(aux_ptr->c_slist_ptr != NULL); + + /* Remove it from both the cleaned list and the dirtied list. */ + if(NULL != (slist_entry_ptr = (H5AC_slist_entry_t *)H5SL_remove(aux_ptr->c_slist_ptr, (void *)(&addr)))) + slist_entry_ptr = H5FL_FREE(H5AC_slist_entry_t, slist_entry_ptr); + if(NULL != (slist_entry_ptr = (H5AC_slist_entry_t *)H5SL_remove(aux_ptr->d_slist_ptr, (void *)(&addr)))) + slist_entry_ptr = H5FL_FREE(H5AC_slist_entry_t, slist_entry_ptr); + + } /* end if */ + + /* Decrement the dirty byte count */ + aux_ptr->dirty_bytes -= entry_ptr->size; + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* H5AC__log_cleaned_entry() */ + + +/*------------------------------------------------------------------------- + * * Function: H5AC__log_flushed_entry() * * Purpose: Update the clean entry slist for the flush of an entry -- diff --git a/src/H5ACpkg.h b/src/H5ACpkg.h index a298f85..3900ff1 100644 --- a/src/H5ACpkg.h +++ b/src/H5ACpkg.h @@ -412,6 +412,7 @@ typedef struct H5AC_aux_t /* Parallel I/O routines */ H5_DLL herr_t H5AC__log_deleted_entry(const H5AC_info_t *entry_ptr); H5_DLL herr_t H5AC__log_dirtied_entry(const H5AC_info_t *entry_ptr); +H5_DLL herr_t H5AC__log_cleaned_entry(const H5AC_info_t *entry_ptr); H5_DLL herr_t H5AC__log_flushed_entry(H5C_t *cache_ptr, haddr_t addr, hbool_t was_dirty, unsigned flags); H5_DLL herr_t H5AC__log_inserted_entry(const H5AC_info_t *entry_ptr); @@ -432,6 +433,8 @@ H5_DLL herr_t H5AC__open_trace_file(H5AC_t *cache_ptr, const char *trace_file_na /* Cache logging routines */ H5_DLL herr_t H5AC__write_create_cache_log_msg(H5AC_t *cache); H5_DLL herr_t H5AC__write_destroy_cache_log_msg(H5AC_t *cache); +H5_DLL herr_t H5AC__write_evict_cache_log_msg(const H5AC_t *cache, + herr_t fxn_ret_value); H5_DLL herr_t H5AC__write_expunge_entry_log_msg(const H5AC_t *cache, haddr_t address, int type_id, @@ -447,6 +450,8 @@ H5_DLL herr_t H5AC__write_insert_entry_log_msg(const H5AC_t *cache, H5_DLL herr_t H5AC__write_mark_dirty_entry_log_msg(const H5AC_t *cache, const H5AC_info_t *entry, herr_t fxn_ret_value); +H5_DLL herr_t H5AC__write_mark_clean_entry_log_msg(const H5AC_t *cache, + const H5AC_info_t *entry, herr_t fxn_ret_value); H5_DLL herr_t H5AC__write_move_entry_log_msg(const H5AC_t *cache, haddr_t old_addr, haddr_t new_addr, @@ -482,6 +487,9 @@ H5_DLL herr_t H5AC__write_unprotect_entry_log_msg(const H5AC_t *cache, H5_DLL herr_t H5AC__write_set_cache_config_log_msg(const H5AC_t *cache, const H5AC_cache_config_t *config, herr_t fxn_ret_value); +H5_DLL herr_t H5AC__write_remove_entry_log_msg(const H5AC_t *cache, + const H5AC_info_t *entry, + herr_t fxn_ret_value); #endif /* _H5ACpkg_H */ diff --git a/src/H5ACprivate.h b/src/H5ACprivate.h index 2251af4..76013a3 100644 --- a/src/H5ACprivate.h +++ b/src/H5ACprivate.h @@ -86,7 +86,8 @@ typedef enum { H5AC_FARRAY_DBLK_PAGE_ID, /* (24) fixed array data block page */ H5AC_SUPERBLOCK_ID, /* (25) file superblock */ H5AC_DRVRINFO_ID, /* (26) driver info block (supplements superblock)*/ - H5AC_TEST_ID, /* (27) test entry -- not used for actual files */ + H5AC_PROXY_ENTRY_ID, /* (27) cache entry proxy */ + H5AC_TEST_ID, /* (28) test entry -- not used for actual files */ H5AC_NTYPES /* Number of types, must be last */ } H5AC_type_t; @@ -199,6 +200,25 @@ typedef H5C_cache_entry_t H5AC_info_t; /* Typedef for metadata cache (defined in H5Cpkg.h) */ typedef H5C_t H5AC_t; +/* Metadata cache proxy entry type */ +typedef struct H5AC_proxy_entry_t { + H5AC_info_t cache_info; /* Information for H5AC cache functions */ + /* (MUST be first field in structure) */ + + /* General fields */ + haddr_t addr; /* Address of the entry in the file */ + /* (Should be in 'temporary' address space) */ + + /* Parent fields */ + H5SL_t *parents; /* Skip list to track parent addresses */ + + /* Child fields */ + size_t nchildren; /* Number of children */ + size_t ndirty_children; /* Number of dirty children */ + /* (Note that this currently duplicates some cache functionality) */ +} H5AC_proxy_entry_t; + + #define H5AC_RING_NAME "H5AC_ring_type" /* Dataset transfer property lists for metadata calls */ @@ -353,11 +373,14 @@ H5_DLL herr_t H5AC_unprotect(H5F_t *f, hid_t dxpl_id, const H5AC_class_t *type, haddr_t addr, void *thing, unsigned flags); H5_DLL herr_t H5AC_flush(H5F_t *f, hid_t dxpl_id); H5_DLL herr_t H5AC_mark_entry_dirty(void *thing); +H5_DLL herr_t H5AC_mark_entry_clean(void *thing); H5_DLL herr_t H5AC_move_entry(H5F_t *f, const H5AC_class_t *type, haddr_t old_addr, haddr_t new_addr, hid_t dxpl_id); H5_DLL herr_t H5AC_dest(H5F_t *f, hid_t dxpl_id); +H5_DLL herr_t H5AC_evict(H5F_t *f, hid_t dxpl_id); H5_DLL herr_t H5AC_expunge_entry(H5F_t *f, hid_t dxpl_id, const H5AC_class_t *type, haddr_t addr, unsigned flags); +H5_DLL herr_t H5AC_remove_entry(void *entry); H5_DLL herr_t H5AC_get_cache_auto_resize_config(const H5AC_t * cache_ptr, H5AC_cache_config_t *config_ptr); H5_DLL herr_t H5AC_get_cache_size(H5AC_t *cache_ptr, size_t *max_size_ptr, @@ -381,6 +404,15 @@ H5_DLL herr_t H5AC_set_ring(hid_t dxpl_id, H5AC_ring_t ring, H5P_genplist_t **dx H5_DLL herr_t H5AC_reset_ring(H5P_genplist_t *dxpl, H5AC_ring_t orig_ring); H5_DLL herr_t H5AC_expunge_tag_type_metadata(H5F_t *f, hid_t dxpl_id, haddr_t tag, int type_id, unsigned flags); +/* Virtual entry routines */ +H5_DLL H5AC_proxy_entry_t *H5AC_proxy_entry_create(void); +H5_DLL herr_t H5AC_proxy_entry_add_parent(H5AC_proxy_entry_t *pentry, void *parent); +H5_DLL herr_t H5AC_proxy_entry_remove_parent(H5AC_proxy_entry_t *pentry, void *parent); +H5_DLL herr_t H5AC_proxy_entry_add_child(H5AC_proxy_entry_t *pentry, H5F_t *f, + hid_t dxpl_id, void *child); +H5_DLL herr_t H5AC_proxy_entry_remove_child(H5AC_proxy_entry_t *pentry, void *child); +H5_DLL herr_t H5AC_proxy_entry_dest(H5AC_proxy_entry_t *pentry); + #ifdef H5_HAVE_PARALLEL H5_DLL herr_t H5AC_add_candidate(H5AC_t * cache_ptr, haddr_t addr); #endif /* H5_HAVE_PARALLEL */ diff --git a/src/H5ACproxy_entry.c b/src/H5ACproxy_entry.c new file mode 100644 index 0000000..ab5edb9 --- /dev/null +++ b/src/H5ACproxy_entry.c @@ -0,0 +1,796 @@ +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * + * Copyright by The HDF Group. * + * Copyright by the Board of Trustees of the University of Illinois. * + * All rights reserved. * + * * + * This file is part of HDF5. The full HDF5 copyright notice, including * + * terms governing use, modification, and redistribution, is contained in * + * the files COPYING and Copyright.html. COPYING can be found at the root * + * of the source code distribution tree; Copyright.html can be found at the * + * root level of an installed copy of the electronic HDF5 document set and * + * is linked from the top-level documents page. It can also be found at * + * http://hdfgroup.org/HDF5/doc/Copyright.html. If you do not have * + * access to either file, you may request a copy from help@hdfgroup.org. * + * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ + +/*------------------------------------------------------------------------- + * + * Created: H5ACproxy_entry.c + * + * Purpose: Functions and a cache client for a "proxy" cache entry. + * A proxy cache entry is used as a placeholder for entire + * data structures to attach flush dependencies, etc. + * + *------------------------------------------------------------------------- + */ + +/****************/ +/* Module Setup */ +/****************/ +#include "H5ACmodule.h" /* This source code file is part of the H5AC module */ + +/***********/ +/* Headers */ +/***********/ +#include "H5private.h" /* Generic Functions */ +#include "H5ACpkg.h" /* Metadata cache */ +#include "H5Eprivate.h" /* Error handling */ +#include "H5MFprivate.h" /* File memory management */ + + +/****************/ +/* Local Macros */ +/****************/ + + +/******************/ +/* Local Typedefs */ +/******************/ + + +/********************/ +/* Package Typedefs */ +/********************/ + + +/********************/ +/* Local Prototypes */ +/********************/ + +/* Metadata cache (H5AC) callbacks */ +static herr_t H5AC__proxy_entry_image_len(const void *thing, size_t *image_len); +static herr_t H5AC__proxy_entry_serialize(const H5F_t *f, void *image_ptr, + size_t len, void *thing); +static herr_t H5AC__proxy_entry_notify(H5AC_notify_action_t action, void *thing); +static herr_t H5AC__proxy_entry_free_icr(void *thing); + +/*********************/ +/* Package Variables */ +/*********************/ + +/* H5AC proxy entries inherit cache-like properties from H5AC */ +const H5AC_class_t H5AC_PROXY_ENTRY[1] = {{ + H5AC_PROXY_ENTRY_ID, /* Metadata client ID */ + "Proxy entry", /* Metadata client name (for debugging) */ + H5FD_MEM_SUPER, /* File space memory type for client */ + 0, /* Client class behavior flags */ + NULL, /* 'get_initial_load_size' callback */ + NULL, /* 'get_final_load_size' callback */ + NULL, /* 'verify_chksum' callback */ + NULL, /* 'deserialize' callback */ + H5AC__proxy_entry_image_len, /* 'image_len' callback */ + NULL, /* 'pre_serialize' callback */ + H5AC__proxy_entry_serialize, /* 'serialize' callback */ + H5AC__proxy_entry_notify, /* 'notify' callback */ + H5AC__proxy_entry_free_icr, /* 'free_icr' callback */ + NULL, /* 'fsf_size' callback */ +}}; + + +/*****************************/ +/* Library Private Variables */ +/*****************************/ + + +/*******************/ +/* Local Variables */ +/*******************/ + +/* Declare a free list to manage H5AC_proxy_entry_t objects */ +H5FL_DEFINE_STATIC(H5AC_proxy_entry_t); + + + +/*------------------------------------------------------------------------- + * Function: H5AC_proxy_entry_create + * + * Purpose: Create a new proxy entry + * + * Return: Success: Pointer to the new proxy entry object. + * Failure: NULL + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +H5AC_proxy_entry_t * +H5AC_proxy_entry_create(void) +{ + H5AC_proxy_entry_t *pentry = NULL; /* Pointer to new proxy entry */ + H5AC_proxy_entry_t *ret_value = NULL; /* Return value */ + + FUNC_ENTER_NOAPI(NULL) + + /* Allocate new proxy entry */ + if(NULL == (pentry = H5FL_CALLOC(H5AC_proxy_entry_t))) + HGOTO_ERROR(H5E_CACHE, H5E_CANTALLOC, NULL, "can't allocate proxy entry") + + /* Set non-zero fields */ + pentry->addr = HADDR_UNDEF; + + /* Set return value */ + ret_value = pentry; + +done: + /* Release resources on error */ + if(!ret_value) + if(pentry) + pentry = H5FL_FREE(H5AC_proxy_entry_t, pentry); + + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_proxy_entry_create() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC_proxy_entry_add_parent + * + * Purpose: Add a parent to a proxy entry + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_proxy_entry_add_parent(H5AC_proxy_entry_t *pentry, void *_parent) +{ + H5AC_info_t *parent = (H5AC_info_t *)_parent; /* Parent entry's cache info */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(parent); + HDassert(pentry); + + /* Add parent to the list of parents */ + if(NULL == pentry->parents) + if(NULL == (pentry->parents = H5SL_create(H5SL_TYPE_HADDR, NULL))) + HGOTO_ERROR(H5E_CACHE, H5E_CANTCREATE, FAIL, "unable to create skip list for parents of proxy entry") + + /* Insert parent address into skip list */ + if(H5SL_insert(pentry->parents, parent, &parent->addr) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTINSERT, FAIL, "unable to insert parent into proxy's skip list") + + /* Add flush dependency on parent */ + if(pentry->nchildren > 0) { + /* Sanity check */ + HDassert(H5F_addr_defined(pentry->addr)); + + if(H5AC_create_flush_dependency(parent, pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTDEPEND, FAIL, "unable to set flush dependency on proxy entry") + } /* end if */ + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_proxy_entry_add_parent() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC_proxy_entry_remove_parent + * + * Purpose: Removes a parent from a proxy entry + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_proxy_entry_remove_parent(H5AC_proxy_entry_t *pentry, void *_parent) +{ + H5AC_info_t *parent = (H5AC_info_t *)_parent; /* Pointer to the parent entry */ + H5AC_info_t *rem_parent; /* Pointer to the removed parent entry */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(pentry); + HDassert(pentry->parents); + HDassert(parent); + + /* Remove parent from skip list */ + if(NULL == (rem_parent = (H5AC_info_t *)H5SL_remove(pentry->parents, &parent->addr))) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "unable to remove proxy entry parent from skip list") + if(!H5F_addr_eq(rem_parent->addr, parent->addr)) + HGOTO_ERROR(H5E_CACHE, H5E_BADVALUE, FAIL, "removed proxy entry parent not the same as real parent") + + /* Shut down the skip list, if this is the last parent */ + if(0 == H5SL_count(pentry->parents)) { + /* Sanity check */ + HDassert(0 == pentry->nchildren); + + if(H5SL_close(pentry->parents) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CLOSEERROR, FAIL, "can't close proxy parent skip list") + pentry->parents = NULL; + } /* end if */ + + /* Remove flush dependency between the proxy entry and a parent */ + if(pentry->nchildren > 0) + if(H5AC_destroy_flush_dependency(parent, pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTUNDEPEND, FAIL, "unable to remove flush dependency on proxy entry") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_proxy_entry_remove_parent() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC__proxy_entry_add_child_cb + * + * Purpose: Callback routine for adding an entry as a flush dependency for + * a proxy entry. + * + * Return: Success: Non-negative on success + * Failure: Negative + * + * Programmer: Quincey Koziol + * Thursday, September 22, 2016 + * + *------------------------------------------------------------------------- + */ +static int +H5AC__proxy_entry_add_child_cb(void *_item, void H5_ATTR_UNUSED *_key, void *_udata) +{ + H5AC_info_t *parent = (H5AC_info_t *)_item; /* Pointer to the parent entry */ + H5AC_proxy_entry_t *pentry = (H5AC_proxy_entry_t *)_udata; /* Pointer to the proxy entry */ + int ret_value = H5_ITER_CONT; /* Callback return value */ + + FUNC_ENTER_STATIC + + /* Add flush dependency on parent for proxy entry */ + if(H5AC_create_flush_dependency(parent, pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTDEPEND, H5_ITER_ERROR, "unable to set flush dependency for virtual entry") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC__proxy_entry_add_child_cb() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC_proxy_entry_add_child + * + * Purpose: Add a child a proxy entry + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_proxy_entry_add_child(H5AC_proxy_entry_t *pentry, H5F_t *f, hid_t dxpl_id, + void *child) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(pentry); + HDassert(child); + + /* Check for first child */ + if(0 == pentry->nchildren) { + /* Get an address, if the proxy doesn't already have one */ + if(!H5F_addr_defined(pentry->addr)) + if(HADDR_UNDEF == (pentry->addr = H5MF_alloc_tmp(f, 1))) + HGOTO_ERROR(H5E_CACHE, H5E_CANTALLOC, FAIL, "temporary file space allocation failed for proxy entry") + + /* Insert the proxy entry into the cache */ + if(H5AC_insert_entry(f, dxpl_id, H5AC_PROXY_ENTRY, pentry->addr, pentry, H5AC__PIN_ENTRY_FLAG) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTINSERT, FAIL, "unable to cache proxy entry") + + /* Proxies start out clean (insertions are automatically marked dirty) */ + if(H5AC_mark_entry_clean(pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTCLEAN, FAIL, "can't mark proxy entry clean") + + /* If there are currently parents, iterate over the list of parents, creating flush dependency on them */ + if(pentry->parents) + if(H5SL_iterate(pentry->parents, H5AC__proxy_entry_add_child_cb, pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_BADITER, FAIL, "can't visit parents") + } /* end if */ + + /* Add flush dependency on proxy entry */ + if(H5AC_create_flush_dependency(pentry, child) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTDEPEND, FAIL, "unable to set flush dependency on proxy entry") + + /* Increment count of children */ + pentry->nchildren++; + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_proxy_entry_add_child() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC__proxy_entry_remove_child_cb + * + * Purpose: Callback routine for removing an entry as a flush dependency for + * proxy entry. + * + * Return: Success: Non-negative on success + * Failure: Negative + * + * Programmer: Quincey Koziol + * Thursday, September 22, 2016 + * + *------------------------------------------------------------------------- + */ +static int +H5AC__proxy_entry_remove_child_cb(void *_item, void H5_ATTR_UNUSED *_key, void *_udata) +{ + H5AC_info_t *parent = (H5AC_info_t *)_item; /* Pointer to the parent entry */ + H5AC_proxy_entry_t *pentry = (H5AC_proxy_entry_t *)_udata; /* Pointer to the proxy entry */ + int ret_value = H5_ITER_CONT; /* Callback return value */ + + FUNC_ENTER_STATIC + + /* Remove flush dependency on parent for proxy entry */ + if(H5AC_destroy_flush_dependency(parent, pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTUNDEPEND, H5_ITER_ERROR, "unable to remove flush dependency for proxy entry") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC__proxy_entry_remove_child_cb() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC_proxy_entry_remove_child + * + * Purpose: Remove a child a proxy entry + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_proxy_entry_remove_child(H5AC_proxy_entry_t *pentry, void *child) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(pentry); + HDassert(child); + + /* Remove flush dependency on proxy entry */ + if(H5AC_destroy_flush_dependency(pentry, child) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTUNDEPEND, FAIL, "unable to remove flush dependency on proxy entry") + + /* Decrement count of children */ + pentry->nchildren--; + + /* Check for last child */ + if(0 == pentry->nchildren) { + /* Check for flush dependencies on proxy's parents */ + if(pentry->parents) + /* Iterate over the list of parents, removing flush dependency on them */ + if(H5SL_iterate(pentry->parents, H5AC__proxy_entry_remove_child_cb, pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_BADITER, FAIL, "can't visit parents") + + /* Unpin proxy */ + if(H5AC_unpin_entry(pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTUNPIN, FAIL, "can't unpin proxy entry") + + /* Remove proxy entry from cache */ + if(H5AC_remove_entry(pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "unable to remove proxy entry") + } /* end if */ + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_proxy_entry_remove_child() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC_proxy_entry_dest + * + * Purpose: Destroys a proxy entry in memory. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_proxy_entry_dest(H5AC_proxy_entry_t *pentry) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(pentry); + HDassert(NULL == pentry->parents); + HDassert(0 == pentry->nchildren); + HDassert(0 == pentry->ndirty_children); + + /* Free the proxy entry object */ + pentry = H5FL_FREE(H5AC_proxy_entry_t, pentry); + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_proxy_entry_dest() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC__proxy_entry_image_len + * + * Purpose: Compute the size of the data structure on disk. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5AC__proxy_entry_image_len(const void H5_ATTR_UNUSED *thing, size_t *image_len) +{ + FUNC_ENTER_STATIC_NOERR + + /* Check arguments */ + HDassert(image_len); + + /* Set the image length size to 1 byte */ + *image_len = 1; + + FUNC_LEAVE_NOAPI(SUCCEED) +} /* end H5AC__proxy_entry_image_len() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC__proxy_entry_serialize + * + * Purpose: Serializes a data structure for writing to disk. + * + * Note: Should never be invoked. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5AC__proxy_entry_serialize(const H5F_t H5_ATTR_UNUSED *f, void H5_ATTR_UNUSED *image, + size_t H5_ATTR_UNUSED len, void H5_ATTR_UNUSED *thing) +{ + FUNC_ENTER_STATIC_NOERR /* Yes, even though this pushes an error on the stack */ + + /* Should never be invoked */ + HDassert(0 && "Invalid callback?!?"); + + HERROR(H5E_CACHE, H5E_CANTSERIALIZE, "called unreachable fcn."); + + FUNC_LEAVE_NOAPI(FAIL) +} /* end H5AC__proxy_entry_serialize() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC__proxy_entry_notify + * + * Purpose: Handle cache action notifications + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5AC__proxy_entry_notify(H5AC_notify_action_t action, void *_thing) +{ + H5AC_proxy_entry_t *pentry = (H5AC_proxy_entry_t *)_thing; + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_STATIC + + /* Sanity check */ + HDassert(pentry); + + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + break; + + case H5AC_NOTIFY_ACTION_AFTER_LOAD: +#ifdef NDEBUG + HGOTO_ERROR(H5E_CACHE, H5E_BADVALUE, FAIL, "invalid notify action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Invalid action?!?"); +#endif /* NDEBUG */ + break; + + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: +#ifdef NDEBUG + HGOTO_ERROR(H5E_CACHE, H5E_BADVALUE, FAIL, "invalid notify action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Invalid action?!?"); +#endif /* NDEBUG */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* Sanity checks */ + HDassert(0 == pentry->ndirty_children); + + /* No action */ + break; + + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + /* Sanity checks */ + HDassert(pentry->ndirty_children > 0); + + /* No action */ + break; + + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + /* Sanity checks */ + HDassert(0 == pentry->ndirty_children); + + /* No action */ + break; + + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + /* Increment # of dirty children */ + pentry->ndirty_children++; + + /* Check for first dirty child */ + if(1 == pentry->ndirty_children) + if(H5AC_mark_entry_dirty(pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTDIRTY, FAIL, "can't mark proxy entry dirty") + break; + + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* Sanity check */ + HDassert(pentry->ndirty_children > 0); + + /* Decrement # of dirty children */ + pentry->ndirty_children--; + + /* Check for last dirty child */ + if(0 == pentry->ndirty_children) + if(H5AC_mark_entry_clean(pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTCLEAN, FAIL, "can't mark proxy entry clean") + break; + + default: +#ifdef NDEBUG + HGOTO_ERROR(H5E_CACHE, H5E_BADVALUE, FAIL, "unknown notify action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC__proxy_entry_notify() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC__proxy_entry_free_icr + * + * Purpose: Destroy/release an "in core representation" of a data + * structure + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5AC__proxy_entry_free_icr(void *_thing) +{ + H5AC_proxy_entry_t *pentry = (H5AC_proxy_entry_t *)_thing; + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_STATIC + + /* Destroy the proxy entry */ + if(H5AC_proxy_entry_dest(pentry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTFREE, FAIL, "unable to destroy proxy entry") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* H5AC__proxy_entry_free_icr() */ + +#ifdef OLD_CODE + +/*------------------------------------------------------------------------- + * Function: H5AC_virt_entry_dirty_parent + * + * Purpose: Indicate that a virtual entry's parent became dirty + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * July 23, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_virt_entry_dirty_parent(H5AC_virt_entry_t *ventry) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity check */ + HDassert(ventry); + HDassert(ventry->track_parents); + HDassert(ventry->nparents > 0); + + /* If this is the first dirty parent or child, mark the virtual entry dirty */ + if(ventry->in_cache && 0 == ventry->ndirty_parents && 0 == ventry->ndirty_children) + if(H5AC_mark_entry_dirty(ventry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTDIRTY, FAIL, "can't mark virtual entry dirty") + + /* Increment the number of dirty parents */ + ventry->ndirty_parents++; + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_virt_entry_dirty_parent() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC_virt_entry_clean_parent + * + * Purpose: Indicate that a virtual entry's parent became clean + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * July 23, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_virt_entry_clean_parent(H5AC_virt_entry_t *ventry) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity check */ + HDassert(ventry); + HDassert(ventry->track_parents); +// HDassert(ventry->nparents > 0); + HDassert(ventry->ndirty_parents > 0); + + /* Decrement the number of dirty parents */ + ventry->ndirty_parents--; + + /* If this is the last dirty parent or child, mark the virtual entry clean */ + if(ventry->in_cache && 0 == ventry->ndirty_parents && 0 == ventry->ndirty_children) + if(H5AC_mark_entry_clean(ventry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTCLEAN, FAIL, "can't mark virtual entry clean") + + /* Destroy the skip list, if no more parents */ + if(0 == ventry->nparents && 0 == ventry->ndirty_parents) { +// /* Sanity check */ +// HDassert(0 == ventry->ndirty_parents); + + if(H5SL_close(ventry->parents) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CLOSEERROR, FAIL, "can't close parent list") + ventry->parents = NULL; + } /* end if */ + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_virt_entry_clean_parent() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC_virt_entry_dirty_child + * + * Purpose: Indicate that a virtual entry's child became dirty + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * July 24, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_virt_entry_dirty_child(H5AC_virt_entry_t *ventry) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity check */ + HDassert(ventry); + HDassert(ventry->nchildren > 0); + + /* If this is the first dirty parent or child, mark the virtual entry dirty */ + if(ventry->in_cache && 0 == ventry->ndirty_parents && 0 == ventry->ndirty_children) + if(H5AC_mark_entry_dirty(ventry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTDIRTY, FAIL, "can't mark virtual entry dirty") + + /* Increment the number of dirty children */ + ventry->ndirty_children++; + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_virt_entry_dirty_child() */ + + +/*------------------------------------------------------------------------- + * Function: H5AC_virt_entry_clean_child + * + * Purpose: Indicate that a virtual entry's child became clean + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * July 24, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5AC_virt_entry_clean_child(H5AC_virt_entry_t *ventry) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity check */ + HDassert(ventry); + HDassert(ventry->nchildren > 0); + HDassert(ventry->ndirty_children > 0); + + /* Decrement the number of dirty children */ + ventry->ndirty_children--; + + /* If this is the last dirty parent or child, mark the virtual entry clean */ + if(ventry->in_cache && 0 == ventry->ndirty_parents && 0 == ventry->ndirty_children) + if(H5AC_mark_entry_clean(ventry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTCLEAN, FAIL, "can't mark virtual entry clean") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5AC_virt_entry_clean_child() */ + +#endif /* OLD_CODE */ + + @@ -290,7 +290,7 @@ H5B2_insert(H5B2_t *bt2, hid_t dxpl_id, void *udata) hdr = bt2->hdr; /* Insert the record */ - if(H5B2__insert_hdr(hdr, dxpl_id, udata) < 0) + if(H5B2__insert(hdr, dxpl_id, udata) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into B-tree") done: @@ -336,17 +336,17 @@ H5B2_update(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_modify_t op, void *op_ /* Check if the root node is allocated yet */ if(!H5F_addr_defined(hdr->root.addr)) { /* Create root node as leaf node in B-tree */ - if(H5B2__create_leaf(hdr, dxpl_id, &(hdr->root)) < 0) + if(H5B2__create_leaf(hdr, dxpl_id, hdr, &(hdr->root)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "unable to create root node") } /* end if */ /* Attempt to insert record into B-tree */ if(hdr->depth > 0) { - if(H5B2__update_internal(hdr, dxpl_id, hdr->depth, NULL, &hdr->root, &status, H5B2_POS_ROOT, udata, op, op_data) < 0) + if(H5B2__update_internal(hdr, dxpl_id, hdr->depth, NULL, &hdr->root, &status, H5B2_POS_ROOT, hdr, udata, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update record in B-tree internal node") } /* end if */ else { - if(H5B2__update_leaf(hdr, dxpl_id, &hdr->root, &status, H5B2_POS_ROOT, udata, op, op_data) < 0) + if(H5B2__update_leaf(hdr, dxpl_id, &hdr->root, &status, H5B2_POS_ROOT, hdr, udata, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update record in B-tree leaf node") } /* end else */ @@ -354,9 +354,19 @@ H5B2_update(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_modify_t op, void *op_ HDassert(H5B2_UPDATE_UNKNOWN != status); /* Use insert algorithm if nodes to leaf full */ - if(H5B2_UPDATE_INSERT_CHILD_FULL == status) - if(H5B2__insert_hdr(hdr, dxpl_id, udata) < 0) + if(H5B2_UPDATE_INSERT_CHILD_FULL == status) { + if(H5B2__insert(hdr, dxpl_id, udata) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into B-tree") + } /* end if */ + else if(H5B2_UPDATE_SHADOW_DONE == status || H5B2_UPDATE_INSERT_DONE == status) { + /* Mark B-tree header as dirty */ + if(H5B2__hdr_dirty(hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTMARKDIRTY, FAIL, "unable to mark B-tree header dirty") + } /* end else-if */ + else { + /* Sanity check */ + HDassert(H5B2_UPDATE_MODIFY_DONE == status); + } /* end else */ done: FUNC_LEAVE_NOAPI(ret_value) @@ -432,7 +442,7 @@ H5B2_iterate(H5B2_t *bt2, hid_t dxpl_id, H5B2_operator_t op, void *op_data) /* Iterate through records */ if(hdr->root.node_nrec > 0) { /* Iterate through nodes */ - if((ret_value = H5B2__iterate_node(hdr, dxpl_id, hdr->depth, &hdr->root, op, op_data)) < 0) + if((ret_value = H5B2__iterate_node(hdr, dxpl_id, hdr->depth, &hdr->root, hdr, op, op_data)) < 0) HERROR(H5E_BTREE, H5E_CANTLIST, "node iteration failed"); } /* end if */ @@ -469,6 +479,7 @@ H5B2_find(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_found_t op, { H5B2_hdr_t *hdr; /* Pointer to the B-tree header */ H5B2_node_ptr_t curr_node_ptr; /* Node pointer info for current node */ + void *parent = NULL; /* Parent of current node */ uint16_t depth; /* Current depth of the tree */ int cmp; /* Comparison value of records */ unsigned idx; /* Location of record which matches key */ @@ -522,6 +533,10 @@ H5B2_find(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_found_t op, /* Current depth of the tree */ depth = hdr->depth; + /* Set initial parent, if doing swmr writes */ + if(hdr->swmr_write) + parent = hdr; + /* Walk down B-tree to find record or leaf node where record is located */ cmp = -1; curr_pos = H5B2_POS_ROOT; @@ -530,12 +545,18 @@ H5B2_find(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_found_t op, H5B2_node_ptr_t next_node_ptr; /* Node pointer info for next node */ /* Lock B-tree current node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node_ptr.addr, curr_node_ptr.node_nrec, depth, H5AC__READ_ONLY_FLAG))) + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, &curr_node_ptr, depth, FALSE, H5AC__READ_ONLY_FLAG))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to load B-tree internal node") + /* Unpin parent if necessary */ + if(parent) { + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + parent = NULL; + } /* end if */ + /* Locate node pointer for child */ - if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, - udata, &idx, &cmp) < 0) { + if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) { /* Unlock current node before failing */ H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, H5AC__NO_FLAGS_SET); HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") @@ -566,9 +587,13 @@ H5B2_find(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_found_t op, } /* end if */ /* Unlock current node */ - if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, H5AC__NO_FLAGS_SET) < 0) + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, (unsigned)(hdr->swmr_write ? H5AC__PIN_ENTRY_FLAG : H5AC__NO_FLAGS_SET)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree node") + /* Keep track of parent if necessary */ + if(hdr->swmr_write) + parent = internal; + /* Set pointer to next node to load */ curr_node_ptr = next_node_ptr; } /* end if */ @@ -598,16 +623,22 @@ H5B2_find(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_found_t op, H5B2_leaf_t *leaf; /* Pointer to leaf node in B-tree */ /* Lock B-tree leaf node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node_ptr.addr, curr_node_ptr.node_nrec, H5AC__READ_ONLY_FLAG))) + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, &curr_node_ptr, FALSE, H5AC__READ_ONLY_FLAG))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + /* Unpin parent if necessary */ + if(parent) { + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + parent = NULL; + } /* end if */ + /* Locate record */ - if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, - udata, &idx, &cmp) < 0) { - /* unlock current node before failing */ + if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) { + /* Unlock current node before failing */ H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, curr_node_ptr.addr, leaf, H5AC__NO_FLAGS_SET); HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - } + } /* end if */ if(cmp != 0) { /* Unlock leaf node */ @@ -655,6 +686,12 @@ H5B2_find(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_found_t op, } /* end block */ done: + if(parent) { + HDassert(ret_value < 0); + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + } /* end if */ + FUNC_LEAVE_NOAPI(ret_value) } /* H5B2_find() */ @@ -683,6 +720,7 @@ H5B2_index(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, hsize_t idx, { H5B2_hdr_t *hdr; /* Pointer to the B-tree header */ H5B2_node_ptr_t curr_node_ptr; /* Node pointer info for current node */ + void *parent = NULL; /* Parent of current node */ uint16_t depth; /* Current depth of the tree */ herr_t ret_value = SUCCEED; /* Return value */ @@ -712,6 +750,10 @@ H5B2_index(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, hsize_t idx, /* Current depth of the tree */ depth = hdr->depth; + /* Set initial parent, if doing swmr writes */ + if(hdr->swmr_write) + parent = hdr; + /* Check for reverse indexing and map requested index to appropriate forward index */ if(order == H5_ITER_DEC) idx = curr_node_ptr.all_nrec - (idx + 1); @@ -723,9 +765,16 @@ H5B2_index(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, hsize_t idx, unsigned u; /* Local index variable */ /* Lock B-tree current node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node_ptr.addr, curr_node_ptr.node_nrec, depth, H5AC__READ_ONLY_FLAG))) + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, &curr_node_ptr, depth, FALSE, H5AC__READ_ONLY_FLAG))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to load B-tree internal node") + /* Unpin parent if necessary */ + if(parent) { + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + parent = NULL; + } /* end if */ + /* Search for record with correct index */ for(u = 0; u < internal->nrec; u++) { /* Check if record is in child node */ @@ -734,9 +783,13 @@ H5B2_index(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, hsize_t idx, next_node_ptr = internal->node_ptrs[u]; /* Unlock current node */ - if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, H5AC__NO_FLAGS_SET) < 0) + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, (unsigned)(hdr->swmr_write ? H5AC__PIN_ENTRY_FLAG : H5AC__NO_FLAGS_SET)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree node") + /* Keep track of parent if necessary */ + if(hdr->swmr_write) + parent = internal; + /* Set pointer to next node to load */ curr_node_ptr = next_node_ptr; @@ -776,9 +829,13 @@ H5B2_index(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, hsize_t idx, next_node_ptr = internal->node_ptrs[u]; /* Unlock current node */ - if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, H5AC__NO_FLAGS_SET) < 0) + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, (unsigned)(hdr->swmr_write ? H5AC__PIN_ENTRY_FLAG : H5AC__NO_FLAGS_SET)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree node") + /* Keep track of parent if necessary */ + if(hdr->swmr_write) + parent = internal; + /* Set pointer to next node to load */ curr_node_ptr = next_node_ptr; } /* end if */ @@ -795,9 +852,16 @@ H5B2_index(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, hsize_t idx, H5B2_leaf_t *leaf; /* Pointer to leaf node in B-tree */ /* Lock B-tree leaf node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node_ptr.addr, curr_node_ptr.node_nrec, H5AC__READ_ONLY_FLAG))) + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, &curr_node_ptr, FALSE, H5AC__READ_ONLY_FLAG))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + /* Unpin parent if necessary */ + if(parent) { + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + parent = NULL; + } /* end if */ + /* Sanity check index */ HDassert(idx < leaf->nrec); @@ -816,6 +880,12 @@ H5B2_index(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, hsize_t idx, } /* end block */ done: + if(parent) { + HDassert(ret_value < 0); + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + } /* end if */ + FUNC_LEAVE_NOAPI(ret_value) } /* H5B2_index() */ @@ -859,8 +929,9 @@ H5B2_remove(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_remove_t op, if(hdr->depth > 0) { hbool_t depth_decreased = FALSE; /* Flag to indicate whether the depth of the B-tree decreased */ - if(H5B2__remove_internal(hdr, dxpl_id, &depth_decreased, NULL, hdr->depth, - &(hdr->cache_info), NULL, H5B2_POS_ROOT, &hdr->root, udata, op, op_data) < 0) + if(H5B2__remove_internal(hdr, dxpl_id, &depth_decreased, NULL, NULL, + hdr->depth, &(hdr->cache_info), NULL, H5B2_POS_ROOT, &hdr->root, + udata, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree internal node") /* Check for decreasing the depth of the B-tree */ @@ -878,7 +949,7 @@ H5B2_remove(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_remove_t op, } /* end for */ } /* end if */ else { - if(H5B2__remove_leaf(hdr, dxpl_id, &hdr->root, H5B2_POS_ROOT, udata, op, op_data) < 0) + if(H5B2__remove_leaf(hdr, dxpl_id, &hdr->root, H5B2_POS_ROOT, hdr, udata, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree leaf node") } /* end else */ @@ -941,8 +1012,9 @@ H5B2_remove_by_idx(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, if(hdr->depth > 0) { hbool_t depth_decreased = FALSE; /* Flag to indicate whether the depth of the B-tree decreased */ - if(H5B2__remove_internal_by_idx(hdr, dxpl_id, &depth_decreased, NULL, hdr->depth, - &(hdr->cache_info), NULL, &hdr->root, H5B2_POS_ROOT, idx, op, op_data) < 0) + if(H5B2__remove_internal_by_idx(hdr, dxpl_id, &depth_decreased, NULL, NULL, + hdr->depth, &(hdr->cache_info), NULL, &hdr->root, H5B2_POS_ROOT, + idx, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree internal node") /* Check for decreasing the depth of the B-tree */ @@ -960,7 +1032,7 @@ H5B2_remove_by_idx(H5B2_t *bt2, hid_t dxpl_id, H5_iter_order_t order, } /* end for */ } /* end if */ else { - if(H5B2__remove_leaf_by_idx(hdr, dxpl_id, &hdr->root, H5B2_POS_ROOT, (unsigned)idx, op, op_data) < 0) + if(H5B2__remove_leaf_by_idx(hdr, dxpl_id, &hdr->root, H5B2_POS_ROOT, hdr, (unsigned)idx, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree leaf node") } /* end else */ @@ -1055,11 +1127,11 @@ H5B2_neighbor(H5B2_t *bt2, hid_t dxpl_id, H5B2_compare_t range, void *udata, /* Attempt to find neighbor record in B-tree */ if(hdr->depth > 0) { - if(H5B2__neighbor_internal(hdr, dxpl_id, hdr->depth, &hdr->root, NULL, range, udata, op, op_data) < 0) + if(H5B2__neighbor_internal(hdr, dxpl_id, hdr->depth, &hdr->root, NULL, range, hdr, udata, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "unable to find neighbor record in B-tree internal node") } /* end if */ else { - if(H5B2__neighbor_leaf(hdr, dxpl_id, &hdr->root, NULL, range, udata, op, op_data) < 0) + if(H5B2__neighbor_leaf(hdr, dxpl_id, &hdr->root, NULL, range, hdr, udata, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "unable to find neighbor record in B-tree leaf node") } /* end else */ @@ -1094,6 +1166,7 @@ H5B2_modify(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_modify_t op, { H5B2_hdr_t *hdr; /* Pointer to the B-tree header */ H5B2_node_ptr_t curr_node_ptr; /* Node pointer info for current node */ + void *parent = NULL; /* Parent of current node */ H5B2_nodepos_t curr_pos; /* Position of current node */ uint16_t depth; /* Current depth of the tree */ int cmp; /* Comparison value of records */ @@ -1122,6 +1195,10 @@ H5B2_modify(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_modify_t op, /* Current depth of the tree */ depth = hdr->depth; + /* Set initial parent, if doing swmr writes */ + if(hdr->swmr_write) + parent = hdr; + /* Walk down B-tree to find record or leaf node where record is located */ cmp = -1; curr_pos = H5B2_POS_ROOT; @@ -1131,16 +1208,22 @@ H5B2_modify(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_modify_t op, H5B2_node_ptr_t next_node_ptr; /* Node pointer info for next node */ /* Lock B-tree current node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node_ptr.addr, curr_node_ptr.node_nrec, depth, H5AC__NO_FLAGS_SET))) + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, &curr_node_ptr, depth, FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to load B-tree internal node") + /* Unpin parent if necessary */ + if(parent) { + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + parent = NULL; + } /* end if */ + /* Locate node pointer for child */ - if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, - udata, &idx, &cmp) < 0) { - /* Unlock current node */ + if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) { + /* Unlock current node before failing */ H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, H5AC__NO_FLAGS_SET); HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - } + } /* end if */ if(cmp > 0) idx++; @@ -1168,9 +1251,13 @@ H5B2_modify(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_modify_t op, } /* end if */ /* Unlock current node */ - if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, H5AC__NO_FLAGS_SET) < 0) + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, (unsigned)(hdr->swmr_write ? H5AC__PIN_ENTRY_FLAG : H5AC__NO_FLAGS_SET)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree node") + /* Keep track of parent if necessary */ + if(hdr->swmr_write) + parent = internal; + /* Set pointer to next node to load */ curr_node_ptr = next_node_ptr; } /* end if */ @@ -1209,15 +1296,22 @@ H5B2_modify(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_modify_t op, hbool_t changed = FALSE;/* Whether the 'modify' callback changed the record */ /* Lock B-tree leaf node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node_ptr.addr, curr_node_ptr.node_nrec, H5AC__NO_FLAGS_SET))) + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, &curr_node_ptr, FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + /* Unpin parent if necessary */ + if(parent) { + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + parent = NULL; + } /* end if */ + /* Locate record */ - if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, - udata, &idx, &cmp) < 0) { + if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) { + /* Unlock current node before failing */ H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, curr_node_ptr.addr, leaf, H5AC__NO_FLAGS_SET); HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - } + } /* end if */ if(cmp != 0) { /* Unlock leaf node */ @@ -1278,6 +1372,12 @@ H5B2_modify(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_modify_t op, } /* end block */ done: + if(parent) { + HDassert(ret_value < 0); + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + } /* end if */ + FUNC_LEAVE_NOAPI(ret_value) } /* H5B2_modify() */ @@ -1451,6 +1551,59 @@ done: /*------------------------------------------------------------------------- + * Function: H5B2_depend + * + * Purpose: Make a child flush dependency between the v2 B-tree's + * header and another piece of metadata in the file. + * + * Return: SUCCEED/FAIL + * + * Programmer: Dana Robinson + * Fall 2012 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2_depend(H5B2_t *bt2, hid_t dxpl_id, H5AC_proxy_entry_t *parent) +{ + /* Local variables */ + H5B2_hdr_t *hdr = bt2->hdr; /* Header for B-tree */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(SUCCEED) + + /* + * Check arguments. + */ + HDassert(bt2); + HDassert(hdr); + HDassert(parent); + HDassert(hdr->parent == NULL || hdr->parent == parent); + + /* + * Check to see if the flush dependency between the parent + * and the v2 B-tree header has already been setup. If it hasn't, then + * set it up. + */ + if(NULL == hdr->parent) { + /* Sanity check */ + HDassert(hdr->top_proxy); + + /* Set the shared v2 B-tree header's file context for this operation */ + hdr->f = bt2->f; + + /* Add the v2 B-tree as a child of the parent (proxy) */ + if(H5AC_proxy_entry_add_child(parent, hdr->f, dxpl_id, hdr->top_proxy) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSET, FAIL, "unable to add v2 B-tree as child of proxy") + hdr->parent = parent; + } /* end if */ + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2_depend() */ + + +/*------------------------------------------------------------------------- * Function: H5B2_patch_file * * Purpose: Patch the top-level file pointer contained in bt2 diff --git a/src/H5B2cache.c b/src/H5B2cache.c index 51ad32b..6954e6c 100644 --- a/src/H5B2cache.c +++ b/src/H5B2cache.c @@ -72,6 +72,7 @@ static void *H5B2__cache_hdr_deserialize(const void *image, size_t len, static herr_t H5B2__cache_hdr_image_len(const void *thing, size_t *image_len); static herr_t H5B2__cache_hdr_serialize(const H5F_t *f, void *image, size_t len, void *thing); +static herr_t H5B2__cache_hdr_notify(H5AC_notify_action_t action, void *thing); static herr_t H5B2__cache_hdr_free_icr(void *thing); static herr_t H5B2__cache_int_get_initial_load_size(void *udata, size_t *image_len); @@ -81,6 +82,7 @@ static void *H5B2__cache_int_deserialize(const void *image, size_t len, static herr_t H5B2__cache_int_image_len(const void *thing, size_t *image_len); static herr_t H5B2__cache_int_serialize(const H5F_t *f, void *image, size_t len, void *thing); +static herr_t H5B2__cache_int_notify(H5AC_notify_action_t action, void *thing); static herr_t H5B2__cache_int_free_icr(void *thing); static herr_t H5B2__cache_leaf_get_initial_load_size(void *udata, size_t *image_len); @@ -90,6 +92,7 @@ static void *H5B2__cache_leaf_deserialize(const void *image, size_t len, static herr_t H5B2__cache_leaf_image_len(const void *thing, size_t *image_len); static herr_t H5B2__cache_leaf_serialize(const H5F_t *f, void *image, size_t len, void *thing); +static herr_t H5B2__cache_leaf_notify(H5AC_notify_action_t action, void *thing); static herr_t H5B2__cache_leaf_free_icr(void *thing); /*********************/ @@ -109,7 +112,7 @@ const H5AC_class_t H5AC_BT2_HDR[1] = {{ H5B2__cache_hdr_image_len, /* 'image_len' callback */ NULL, /* 'pre_serialize' callback */ H5B2__cache_hdr_serialize, /* 'serialize' callback */ - NULL, /* 'notify' callback */ + H5B2__cache_hdr_notify, /* 'notify' callback */ H5B2__cache_hdr_free_icr, /* 'free_icr' callback */ NULL, /* 'fsf_size' callback */ }}; @@ -127,7 +130,7 @@ const H5AC_class_t H5AC_BT2_INT[1] = {{ H5B2__cache_int_image_len, /* 'image_len' callback */ NULL, /* 'pre_serialize' callback */ H5B2__cache_int_serialize, /* 'serialize' callback */ - NULL, /* 'notify' callback */ + H5B2__cache_int_notify, /* 'notify' callback */ H5B2__cache_int_free_icr, /* 'free_icr' callback */ NULL, /* 'fsf_size' callback */ }}; @@ -145,7 +148,7 @@ const H5AC_class_t H5AC_BT2_LEAF[1] = {{ H5B2__cache_leaf_image_len, /* 'image_len' callback */ NULL, /* 'pre_serialize' callback */ H5B2__cache_leaf_serialize, /* 'serialize' callback */ - NULL, /* 'notify' callback */ + H5B2__cache_leaf_notify, /* 'notify' callback */ H5B2__cache_leaf_free_icr, /* 'free_icr' callback */ NULL, /* 'fsf_size' callback */ }}; @@ -433,6 +436,91 @@ H5B2__cache_hdr_serialize(const H5F_t *f, void *_image, size_t H5_ATTR_UNUSED le /*------------------------------------------------------------------------- + * Function: H5B2__cache_hdr_notify + * + * Purpose: Handle cache action notifications + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Neil Fortner + * nfortne2@hdfgroup.org + * Apr 24 2012 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5B2__cache_hdr_notify(H5AC_notify_action_t action, void *_thing) +{ + H5B2_hdr_t *hdr = (H5B2_hdr_t *)_thing; + herr_t ret_value = SUCCEED; + + FUNC_ENTER_STATIC + + /* + * Check arguments. + */ + HDassert(hdr); + + /* Check if the file was opened with SWMR-write access */ + if(hdr->swmr_write) { + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + case H5AC_NOTIFY_ACTION_AFTER_LOAD: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: + /* Increment the shadow epoch, forcing new modifications to + * internal and leaf nodes to create new shadow copies */ + hdr->shadow_epoch++; + break; + + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* If hdr->parent != NULL, hdr->parent is used to destroy + * the flush dependency before the header is evicted. + */ + if(hdr->parent) { + /* Sanity check */ + HDassert(hdr->top_proxy); + + /* Destroy flush dependency on object header proxy */ + if(H5AC_proxy_entry_remove_child((H5AC_proxy_entry_t *)hdr->parent, (void *)hdr->top_proxy) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency between v2 B-tree and proxy") + hdr->parent = NULL; + } /* end if */ + + /* Detach from 'top' proxy for extensible array */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_remove_child(hdr->top_proxy, hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency between header and v2 B-tree 'top' proxy") + /* Don't reset hdr->top_proxy here, it's destroyed when the header is freed -QAK */ + } /* end if */ + break; + + default: +#ifdef NDEBUG + HGOTO_ERROR(H5E_BTREE, H5E_BADVALUE, FAIL, "unknown action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + } /* end if */ + else + HDassert(NULL == hdr->parent); + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__cache_hdr_notify() */ + + +/*------------------------------------------------------------------------- * Function: H5B2__cache_hdr_free_icr * * Purpose: Destroy/release an "in core representation" of a data @@ -573,9 +661,8 @@ H5B2__cache_int_deserialize(const void *_image, size_t H5_ATTR_UNUSED len, HDassert(udata); /* Allocate new internal node and reset cache info */ - if(NULL == (internal = H5FL_MALLOC(H5B2_internal_t))) + if(NULL == (internal = H5FL_CALLOC(H5B2_internal_t))) HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, NULL, "memory allocation failed") - HDmemset(&internal->cache_info, 0, sizeof(H5AC_info_t)); /* Increment ref. count on B-tree header */ if(H5B2__hdr_incr(udata->hdr) < 0) @@ -583,6 +670,8 @@ H5B2__cache_int_deserialize(const void *_image, size_t H5_ATTR_UNUSED len, /* Share B-tree information */ internal->hdr = udata->hdr; + internal->parent = udata->parent; + internal->shadow_epoch = udata->hdr->shadow_epoch; /* Magic number */ if(HDmemcmp(image, H5B2_INT_MAGIC, (size_t)H5_SIZEOF_MAGIC)) @@ -775,6 +864,80 @@ done: /*------------------------------------------------------------------------- + * Function: H5B2__cache_int_notify + * + * Purpose: Handle cache action notifications + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Neil Fortner + * nfortne2@hdfgroup.org + * Apr 25 2012 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5B2__cache_int_notify(H5AC_notify_action_t action, void *_thing) +{ + H5B2_internal_t *internal = (H5B2_internal_t *)_thing; + herr_t ret_value = SUCCEED; + + FUNC_ENTER_NOAPI_NOINIT + + /* + * Check arguments. + */ + HDassert(internal); + HDassert(internal->hdr); + + /* Check if the file was opened with SWMR-write access */ + if(internal->hdr->swmr_write) { + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + case H5AC_NOTIFY_ACTION_AFTER_LOAD: + /* Create flush dependency on parent */ + if(H5B2__create_flush_depend((H5AC_info_t *)internal->parent, (H5AC_info_t *)internal) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDEPEND, FAIL, "unable to create flush dependency") + break; + + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* Destroy flush dependency on parent */ + if(H5B2__destroy_flush_depend((H5AC_info_t *)internal->parent, (H5AC_info_t *)internal) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency") + + /* Detach from 'top' proxy for v2 B-tree */ + if(internal->top_proxy) { + if(H5AC_proxy_entry_remove_child(internal->top_proxy, internal) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency between internal node and v2 B-tree 'top' proxy") + internal->top_proxy = NULL; + } /* end if */ + break; + + default: +#ifdef NDEBUG + HGOTO_ERROR(H5E_BTREE, H5E_BADVALUE, FAIL, "unknown action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + } /* end if */ + else + HDassert(NULL == internal->top_proxy); + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__cache_int_notify() */ + + +/*------------------------------------------------------------------------- * Function: H5B2__cache_int_free_icr * * Purpose: Destroy/release an "in core representation" of a data @@ -915,9 +1078,8 @@ H5B2__cache_leaf_deserialize(const void *_image, size_t H5_ATTR_UNUSED len, HDassert(udata); /* Allocate new leaf node and reset cache info */ - if(NULL == (leaf = H5FL_MALLOC(H5B2_leaf_t))) + if(NULL == (leaf = H5FL_CALLOC(H5B2_leaf_t))) HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, NULL, "memory allocation failed") - HDmemset(&leaf->cache_info, 0, sizeof(H5AC_info_t)); /* Increment ref. count on B-tree header */ if(H5B2__hdr_incr(udata->hdr) < 0) @@ -925,6 +1087,8 @@ H5B2__cache_leaf_deserialize(const void *_image, size_t H5_ATTR_UNUSED len, /* Share B-tree header information */ leaf->hdr = udata->hdr; + leaf->parent = udata->parent; + leaf->shadow_epoch = udata->hdr->shadow_epoch; /* Magic number */ if(HDmemcmp(image, H5B2_LEAF_MAGIC, (size_t)H5_SIZEOF_MAGIC)) @@ -1027,7 +1191,7 @@ H5B2__cache_leaf_image_len(const void *_thing, size_t *image_len) *------------------------------------------------------------------------- */ static herr_t -H5B2__cache_leaf_serialize(const H5F_t *f, void *_image, size_t H5_ATTR_UNUSED len, +H5B2__cache_leaf_serialize(const H5F_t H5_ATTR_UNUSED *f, void *_image, size_t H5_ATTR_UNUSED len, void *_thing) { H5B2_leaf_t *leaf = (H5B2_leaf_t *)_thing; /* Pointer to the B-tree leaf node */ @@ -1086,6 +1250,80 @@ done: /*------------------------------------------------------------------------- + * Function: H5B2__cache_leaf_notify + * + * Purpose: Handle cache action notifications + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Neil Fortner + * nfortne2@hdfgroup.org + * Apr 25 2012 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5B2__cache_leaf_notify(H5AC_notify_action_t action, void *_thing) +{ + H5B2_leaf_t *leaf = (H5B2_leaf_t *)_thing; + herr_t ret_value = SUCCEED; + + FUNC_ENTER_NOAPI_NOINIT + + /* + * Check arguments. + */ + HDassert(leaf); + HDassert(leaf->hdr); + + /* Check if the file was opened with SWMR-write access */ + if(leaf->hdr->swmr_write) { + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + case H5AC_NOTIFY_ACTION_AFTER_LOAD: + /* Create flush dependency on parent */ + if(H5B2__create_flush_depend((H5AC_info_t *)leaf->parent, (H5AC_info_t *)leaf) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDEPEND, FAIL, "unable to create flush dependency") + break; + + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* Destroy flush dependency on parent */ + if(H5B2__destroy_flush_depend((H5AC_info_t *)leaf->parent, (H5AC_info_t *)leaf) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency") + + /* Detach from 'top' proxy for v2 B-tree */ + if(leaf->top_proxy) { + if(H5AC_proxy_entry_remove_child(leaf->top_proxy, leaf) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency between leaf node and v2 B-tree 'top' proxy") + leaf->top_proxy = NULL; + } /* end if */ + break; + + default: +#ifdef NDEBUG + HGOTO_ERROR(H5E_BTREE, H5E_BADVALUE, FAIL, "unknown action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + } /* end if */ + else + HDassert(NULL == leaf->top_proxy); + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__cache_leaf_notify() */ + + +/*------------------------------------------------------------------------- * Function: H5B2__cache_leaf_free_icr * * Purpose: Destroy/release an "in core representation" of a data diff --git a/src/H5B2dbg.c b/src/H5B2dbg.c index ddff9e9..19ca89a 100644 --- a/src/H5B2dbg.c +++ b/src/H5B2dbg.c @@ -188,6 +188,7 @@ H5B2__int_debug(H5F_t *f, hid_t dxpl_id, haddr_t addr, FILE *stream, int indent, { H5B2_hdr_t *hdr = NULL; /* B-tree header */ H5B2_internal_t *internal = NULL; /* B-tree internal node */ + H5B2_node_ptr_t node_ptr; /* Fake node pointer for protect */ unsigned u; /* Local index variable */ char temp_str[128]; /* Temporary string, for formatting */ herr_t ret_value=SUCCEED; /* Return value */ @@ -217,9 +218,10 @@ H5B2__int_debug(H5F_t *f, hid_t dxpl_id, haddr_t addr, FILE *stream, int indent, /* * Load the B-tree internal node */ - H5_CHECK_OVERFLOW(nrec, unsigned, uint16_t); H5_CHECK_OVERFLOW(depth, unsigned, uint16_t); - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, addr, (uint16_t)nrec, (uint16_t)depth, H5AC__READ_ONLY_FLAG))) + node_ptr.addr = addr; + H5_CHECKED_ASSIGN(node_ptr.node_nrec, unsigned, nrec, uint16_t) + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, NULL, &node_ptr, (uint16_t)depth, FALSE, H5AC__READ_ONLY_FLAG))) HGOTO_ERROR(H5E_BTREE, H5E_CANTLOAD, FAIL, "unable to load B-tree internal node") /* Print opening message */ @@ -298,6 +300,7 @@ H5B2__leaf_debug(H5F_t *f, hid_t dxpl_id, haddr_t addr, FILE *stream, int indent { H5B2_hdr_t *hdr = NULL; /* B-tree header */ H5B2_leaf_t *leaf = NULL; /* B-tree leaf node */ + H5B2_node_ptr_t node_ptr; /* Fake node pointer for protect */ unsigned u; /* Local index variable */ char temp_str[128]; /* Temporary string, for formatting */ herr_t ret_value = SUCCEED; /* Return value */ @@ -328,7 +331,9 @@ H5B2__leaf_debug(H5F_t *f, hid_t dxpl_id, haddr_t addr, FILE *stream, int indent * Load the B-tree leaf node */ H5_CHECK_OVERFLOW(nrec, unsigned, uint16_t); - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, addr, (uint16_t)nrec, H5AC__READ_ONLY_FLAG))) + node_ptr.addr = addr; + H5_CHECKED_ASSIGN(node_ptr.node_nrec, unsigned, nrec, uint16_t) + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, NULL, &node_ptr, FALSE, H5AC__READ_ONLY_FLAG))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") /* Print opening message */ diff --git a/src/H5B2hdr.c b/src/H5B2hdr.c index 49ffb5b..251a33d 100644 --- a/src/H5B2hdr.c +++ b/src/H5B2hdr.c @@ -133,10 +133,6 @@ H5B2__hdr_init(H5B2_hdr_t *hdr, const H5B2_create_t *cparam, void *ctx_udata, HDassert(cparam->split_percent > 0 && cparam->split_percent <= 100); HDassert(cparam->merge_percent < (cparam->split_percent / 2)); - /* Initialize basic information */ - hdr->rc = 0; - hdr->pending_delete = FALSE; - /* Assign dynamic information */ hdr->depth = depth; @@ -207,6 +203,13 @@ H5B2__hdr_init(H5B2_hdr_t *hdr, const H5B2_create_t *cparam, void *ctx_udata, } /* end for */ } /* end if */ + /* Determine if we are doing SWMR writes. Only enable for data chunks for now. */ + hdr->swmr_write = (H5F_INTENT(hdr->f) & H5F_ACC_SWMR_WRITE) > 0 + && (hdr->cls->id == H5B2_CDSET_ID || hdr->cls->id == H5B2_CDSET_FILT_ID); + + /* Reset the shadow epoch */ + hdr->shadow_epoch = 0; + /* Create the callback context, if the callback exists */ if(hdr->cls->crt_context) if(NULL == (hdr->cb_ctx = (*hdr->cls->crt_context)(ctx_udata))) @@ -285,6 +288,7 @@ H5B2__hdr_create(H5F_t *f, hid_t dxpl_id, const H5B2_create_t *cparam, void *ctx_udata) { H5B2_hdr_t *hdr = NULL; /* The new v2 B-tree header information */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ haddr_t ret_value = HADDR_UNDEF; /* Return value */ FUNC_ENTER_PACKAGE @@ -307,17 +311,40 @@ H5B2__hdr_create(H5F_t *f, hid_t dxpl_id, const H5B2_create_t *cparam, if(HADDR_UNDEF == (hdr->addr = H5MF_alloc(f, H5FD_MEM_BTREE, dxpl_id, (hsize_t)hdr->hdr_size))) HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, HADDR_UNDEF, "file allocation failed for B-tree header") + /* Create 'top' proxy for extensible array entries */ + if(hdr->swmr_write) + if(NULL == (hdr->top_proxy = H5AC_proxy_entry_create())) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCREATE, HADDR_UNDEF, "can't create v2 B-tree proxy") + /* Cache the new B-tree node */ if(H5AC_insert_entry(f, dxpl_id, H5AC_BT2_HDR, hdr->addr, hdr, H5AC__NO_FLAGS_SET) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, HADDR_UNDEF, "can't add B-tree header to cache") + inserted = TRUE; + + /* Add header as child of 'top' proxy */ + if(hdr->top_proxy) + if(H5AC_proxy_entry_add_child(hdr->top_proxy, f, dxpl_id, hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSET, HADDR_UNDEF, "unable to add v2 B-tree header as child of array proxy") /* Set address of v2 B-tree header to return */ ret_value = hdr->addr; done: - if(!H5F_addr_defined(ret_value) && hdr) - if(H5B2__hdr_free(hdr) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTRELEASE, HADDR_UNDEF, "unable to release v2 B-tree header") + if(!H5F_addr_defined(ret_value)) + if(hdr) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(hdr) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTREMOVE, HADDR_UNDEF, "unable to remove v2 B-tree header from cache") + + /* Release header's disk space */ + if(H5F_addr_defined(hdr->addr) && H5MF_xfree(f, H5FD_MEM_BTREE, dxpl_id, hdr->addr, (hsize_t)hdr->hdr_size) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTFREE, HADDR_UNDEF, "unable to free v2 B-tree header") + + /* Destroy header */ + if(H5B2__hdr_free(hdr) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTRELEASE, HADDR_UNDEF, "unable to release v2 B-tree header") + } /* end if */ FUNC_LEAVE_NOAPI(ret_value) } /* end H5B2__hdr_create() */ @@ -503,6 +530,7 @@ H5B2__hdr_protect(H5F_t *f, hid_t dxpl_id, haddr_t hdr_addr, void *ctx_udata, unsigned flags) { H5B2_hdr_cache_ud_t udata; /* User data for cache callbacks */ + H5B2_hdr_t *hdr = NULL; /* v2 B-tree header */ H5B2_hdr_t *ret_value = NULL; /* Return value */ FUNC_ENTER_PACKAGE @@ -520,11 +548,32 @@ H5B2__hdr_protect(H5F_t *f, hid_t dxpl_id, haddr_t hdr_addr, void *ctx_udata, udata.ctx_udata = ctx_udata; /* Protect the header */ - if(NULL == (ret_value = (H5B2_hdr_t *)H5AC_protect(f, dxpl_id, H5AC_BT2_HDR, hdr_addr, &udata, flags))) + if(NULL == (hdr = (H5B2_hdr_t *)H5AC_protect(f, dxpl_id, H5AC_BT2_HDR, hdr_addr, &udata, flags))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, NULL, "unable to load v2 B-tree header, address = %llu", (unsigned long long)hdr_addr) - ret_value->f = f; /* (Must be set again here, in case the header was already in the cache -QAK) */ + hdr->f = f; /* (Must be set again here, in case the header was already in the cache -QAK) */ + + /* Create top proxy, if it doesn't exist */ + if(hdr->swmr_write && NULL == hdr->top_proxy) { + /* Create 'top' proxy for v2 B-tree entries */ + if(NULL == (hdr->top_proxy = H5AC_proxy_entry_create())) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCREATE, NULL, "can't create v2 B-tree proxy") + + /* Add header as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, f, dxpl_id, hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSET, NULL, "unable to add v2 B-tree header as child of proxy") + } /* end if */ + + /* Set return value */ + ret_value = hdr; done: + /* Clean up on error */ + if(!ret_value) { + /* Release the header, if it was protected */ + if(hdr && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_HDR, hdr_addr, hdr, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, NULL, "unable to unprotect v2 B-tree header, address = %llu", (unsigned long long)hdr_addr) + } /* end if */ + FUNC_LEAVE_NOAPI(ret_value) } /* end H5B2__hdr_protect() */ @@ -623,6 +672,13 @@ H5B2__hdr_free(H5B2_hdr_t *hdr) if(hdr->max_native_rec) hdr->max_native_rec = H5MM_xfree(hdr->max_native_rec); + /* Destroy the 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_dest(hdr->top_proxy) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTRELEASE, FAIL, "unable to destroy v2 B-tree 'top' proxy") + hdr->top_proxy = NULL; + } /* end if */ + /* Free B-tree header info */ hdr = H5FL_FREE(H5B2_hdr_t, hdr); @@ -671,7 +727,7 @@ H5B2__hdr_delete(H5B2_hdr_t *hdr, hid_t dxpl_id) /* Delete all nodes in B-tree */ if(H5F_addr_defined(hdr->root.addr)) - if(H5B2__delete_node(hdr, dxpl_id, hdr->depth, &hdr->root, hdr->remove_op, hdr->remove_op_data) < 0) + if(H5B2__delete_node(hdr, dxpl_id, hdr->depth, &hdr->root, hdr, hdr->remove_op, hdr->remove_op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to delete B-tree nodes") /* Indicate that the heap header should be deleted & file space freed */ diff --git a/src/H5B2int.c b/src/H5B2int.c index a8fa59f..47100fd 100644 --- a/src/H5B2int.c +++ b/src/H5B2int.c @@ -37,16 +37,13 @@ #include "H5private.h" /* Generic Functions */ #include "H5B2pkg.h" /* v2 B-trees */ #include "H5Eprivate.h" /* Error handling */ -#include "H5MFprivate.h" /* File memory management */ -#include "H5MMprivate.h" /* Memory management */ #include "H5VMprivate.h" /* Vectors and arrays */ + /****************/ /* Local Macros */ /****************/ -/* Uncomment this macro to enable extra sanity checking */ -/* #define H5B2_DEBUG */ /******************/ /* Local Typedefs */ @@ -61,44 +58,15 @@ /********************/ /* Local Prototypes */ /********************/ +static herr_t H5B2__update_child_flush_depends(H5B2_hdr_t *hdr, hid_t dxpl_id, + unsigned depth, const H5B2_node_ptr_t *node_ptrs, unsigned start_idx, + unsigned end_idx, void *old_parent, void *new_parent); -/* Helper functions */ -static herr_t H5B2__split1(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, - H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx); -static herr_t H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - H5B2_internal_t *internal, unsigned idx); -static herr_t H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx); -static herr_t H5B2__merge2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, - H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx); -static herr_t H5B2__merge3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, - H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx); -static herr_t H5B2__swap_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx, - void *swap_loc); -static herr_t H5B2__create_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, - H5B2_node_ptr_t *node_ptr, uint16_t depth); -#ifdef H5B2_DEBUG -/* Don't label these with H5_ATTR_PURE or you'll get even more warnings... */ -static herr_t H5B2__assert_leaf(const H5B2_hdr_t *hdr, const H5B2_leaf_t *leaf); -static herr_t H5B2__assert_leaf2(const H5B2_hdr_t *hdr, const H5B2_leaf_t *leaf, const H5B2_leaf_t *leaf2); -static herr_t H5B2__assert_internal(hsize_t parent_all_nrec, const H5B2_hdr_t *hdr, const H5B2_internal_t *internal); -static herr_t H5B2__assert_internal2(hsize_t parent_all_nrec, const H5B2_hdr_t *hdr, const H5B2_internal_t *internal, const H5B2_internal_t *internal2); -#endif /* H5B2_DEBUG */ /*********************/ /* Package Variables */ /*********************/ -/* Declare a free list to manage the H5B2_internal_t struct */ -H5FL_DEFINE(H5B2_internal_t); - -/* Declare a free list to manage the H5B2_leaf_t struct */ -H5FL_DEFINE(H5B2_leaf_t); - /*****************************/ /* Library Private Variables */ @@ -137,7 +105,7 @@ H5FL_SEQ_EXTERN(H5B2_node_info_t); */ herr_t H5B2__locate_record(const H5B2_class_t *type, unsigned nrec, size_t *rec_off, - const uint8_t *native, const void *udata, unsigned *idx, int *cmp) + const uint8_t *native, const void *udata, unsigned *idx, int *cmp) { unsigned lo = 0, hi; /* Low & high index values */ unsigned my_idx = 0; /* Final index value */ @@ -179,7 +147,7 @@ done: * *------------------------------------------------------------------------- */ -static herr_t +herr_t H5B2__split1(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx) @@ -195,7 +163,7 @@ H5B2__split1(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, unsigned left_child_flags = H5AC__NO_FLAGS_SET, right_child_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting child nodes */ herr_t ret_value = SUCCEED; /* Return value */ - FUNC_ENTER_STATIC + FUNC_ENTER_PACKAGE /* Check arguments. */ HDassert(hdr); @@ -214,19 +182,20 @@ H5B2__split1(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Create new internal node */ internal->node_ptrs[idx + 1].all_nrec = internal->node_ptrs[idx + 1].node_nrec = 0; - if(H5B2__create_internal(hdr, dxpl_id, &(internal->node_ptrs[idx + 1]), (uint16_t)(depth - 1)) < 0) + if(H5B2__create_internal(hdr, dxpl_id, internal, &(internal->node_ptrs[idx + 1]), (uint16_t)(depth - 1)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "unable to create new internal node") /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_INT; - left_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Protect both leaves */ - if(NULL == (left_int = H5B2__protect_internal(hdr, dxpl_id, left_addr, internal->node_ptrs[idx].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + /* (Shadow left node if doing SWMR writes) */ + if(NULL == (left_int = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - if(NULL == (right_int = H5B2__protect_internal(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_int = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], (uint16_t)(depth - 1), FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for child nodes */ left_child = left_int; @@ -243,19 +212,20 @@ H5B2__split1(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Create new leaf node */ internal->node_ptrs[idx + 1].all_nrec = internal->node_ptrs[idx + 1].node_nrec = 0; - if(H5B2__create_leaf(hdr, dxpl_id, &(internal->node_ptrs[idx + 1])) < 0) + if(H5B2__create_leaf(hdr, dxpl_id, internal, &(internal->node_ptrs[idx + 1])) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "unable to create new leaf node") /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_LEAF; - left_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Protect both leaves */ - if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, left_addr, internal->node_ptrs[idx].node_nrec, H5AC__NO_FLAGS_SET))) + /* (Shadow the left node if doing SWMR writes) */ + if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for child nodes */ left_child = left_leaf; @@ -295,7 +265,7 @@ H5B2__split1(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Determine total number of records in new child nodes */ if(depth > 1) { - unsigned u; /* Local index variable */ + unsigned u; /* Local index variable */ hsize_t new_left_all_nrec; /* New total number of records in left child */ hsize_t new_right_all_nrec; /* New total number of records in right child */ @@ -329,6 +299,12 @@ H5B2__split1(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, if(parent_cache_info_flags_ptr) *parent_cache_info_flags_ptr |= H5AC__DIRTIED_FLAG; + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, right_node_ptrs, + 0, (unsigned)(*right_nrec + 1), left_child, right_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + #ifdef H5B2_DEBUG H5B2__assert_internal((hsize_t)0, hdr, internal); if(depth > 1) { @@ -407,11 +383,11 @@ H5B2__split_root(H5B2_hdr_t *hdr, hid_t dxpl_id) /* Create new internal node to use as root */ hdr->root.node_nrec = 0; - if(H5B2__create_internal(hdr, dxpl_id, &(hdr->root), hdr->depth) < 0) + if(H5B2__create_internal(hdr, dxpl_id, hdr, &(hdr->root), hdr->depth) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "unable to create new internal node") /* Protect new root node */ - if(NULL == (new_root = H5B2__protect_internal(hdr, dxpl_id, hdr->root.addr, hdr->root.node_nrec, hdr->depth, H5AC__NO_FLAGS_SET))) + if(NULL == (new_root = H5B2__protect_internal(hdr, dxpl_id, hdr, &hdr->root, hdr->depth, FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") /* Set first node pointer in root node to old root node pointer info */ @@ -444,7 +420,7 @@ done: * *------------------------------------------------------------------------- */ -static herr_t +herr_t H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5B2_internal_t *internal, unsigned idx) { @@ -458,7 +434,7 @@ H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, unsigned left_child_flags = H5AC__NO_FLAGS_SET, right_child_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting child nodes */ herr_t ret_value = SUCCEED; /* Return value */ - FUNC_ENTER_STATIC + FUNC_ENTER_PACKAGE /* Check arguments. */ HDassert(hdr); @@ -471,14 +447,15 @@ H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_INT; - left_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Lock left & right B-tree child nodes */ - if(NULL == (left_internal = H5B2__protect_internal(hdr, dxpl_id, left_addr, internal->node_ptrs[idx].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + /* (Shadow both nodes if doing SWMR writes) */ + if(NULL == (left_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - if(NULL == (right_internal = H5B2__protect_internal(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for child nodes */ left_child = left_internal; @@ -496,14 +473,15 @@ H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_LEAF; - left_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Lock left & right B-tree child nodes */ - if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, left_addr, internal->node_ptrs[idx].node_nrec, H5AC__NO_FLAGS_SET))) + /* (Shadow both nodes if doing SWMR writes) */ + if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for child nodes */ left_child = left_leaf; @@ -549,7 +527,7 @@ H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Handle node pointers, if we have an internal node */ if(depth > 1) { hsize_t moved_nrec = move_nrec; /* Total number of records moved, for internal redistrib */ - unsigned u; /* Local index variable */ + unsigned u; /* Local index variable */ /* Count the number of records being moved */ for(u = 0; u < move_nrec; u++) @@ -564,6 +542,12 @@ H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, HDmemmove(&(right_node_ptrs[0]), &(right_node_ptrs[move_nrec]), sizeof(H5B2_node_ptr_t) * (new_right_nrec + (unsigned)1)); } /* end if */ + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, left_node_ptrs, + (unsigned)(*left_nrec + 1), (unsigned)(*left_nrec + move_nrec + 1), right_child, left_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Update number of records in child nodes */ *left_nrec = (uint16_t)(*left_nrec + move_nrec); *right_nrec = new_right_nrec; @@ -599,7 +583,7 @@ H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Handle node pointers, if we have an internal node */ if(depth > 1) { hsize_t moved_nrec = move_nrec; /* Total number of records moved, for internal redistrib */ - unsigned u; /* Local index variable */ + unsigned u; /* Local index variable */ /* Slide node pointers in right node up */ HDmemmove(&(right_node_ptrs[move_nrec]), &(right_node_ptrs[0]), sizeof(H5B2_node_ptr_t) * (size_t)(*right_nrec + 1)); @@ -614,6 +598,12 @@ H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5_CHECKED_ASSIGN(right_moved_nrec, hssize_t, moved_nrec, hsize_t) } /* end if */ + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, right_node_ptrs, + 0, (unsigned)move_nrec, left_child, right_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Update number of records in child nodes */ *left_nrec = new_left_nrec; *right_nrec = (uint16_t)(*right_nrec + move_nrec); @@ -674,7 +664,7 @@ done: * *------------------------------------------------------------------------- */ -static herr_t +herr_t H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx) { @@ -695,7 +685,7 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, unsigned middle_child_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting child nodes */ herr_t ret_value = SUCCEED; /* Return value */ - FUNC_ENTER_STATIC + FUNC_ENTER_PACKAGE /* Check arguments. */ HDassert(hdr); @@ -710,17 +700,18 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_INT; - left_addr = internal->node_ptrs[idx - 1].addr; - middle_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Lock B-tree child nodes */ - if(NULL == (left_internal = H5B2__protect_internal(hdr, dxpl_id, left_addr, internal->node_ptrs[idx - 1].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + /* (Shadow all nodes if doing SWMR writes) */ + if(NULL == (left_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx - 1], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - if(NULL == (middle_internal = H5B2__protect_internal(hdr, dxpl_id, middle_addr, internal->node_ptrs[idx].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx - 1].addr; + if(NULL == (middle_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - if(NULL == (right_internal = H5B2__protect_internal(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + middle_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for child nodes */ left_child = left_internal; @@ -743,17 +734,18 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_LEAF; - left_addr = internal->node_ptrs[idx - 1].addr; - middle_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Lock B-tree child nodes */ - if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, left_addr, internal->node_ptrs[idx - 1].node_nrec, H5AC__NO_FLAGS_SET))) + /* (Shadow all nodes if doing SWMR writes) */ + if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx - 1], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - if(NULL == (middle_leaf = H5B2__protect_leaf(hdr, dxpl_id, middle_addr, internal->node_ptrs[idx].node_nrec, H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx - 1].addr; + if(NULL == (middle_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, H5AC__NO_FLAGS_SET))) + middle_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for child nodes */ left_child = left_leaf; @@ -802,7 +794,7 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Move node pointers also if this is an internal node */ if(depth > 1) { - hsize_t moved_nrec; /* Total number of records moved, for internal redistrib */ + hsize_t moved_nrec; /* Total number of records moved, for internal redistrib */ unsigned move_nptrs; /* Number of node pointers to move */ unsigned u; /* Local index variable */ @@ -820,6 +812,12 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, HDmemmove(&(middle_node_ptrs[0]), &(middle_node_ptrs[move_nptrs]), sizeof(H5B2_node_ptr_t) * ((*middle_nrec - move_nptrs) + 1)); } /* end if */ + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, left_node_ptrs, + (unsigned)(*left_nrec + 1), (unsigned)(*left_nrec + moved_middle_nrec + 1), middle_child, left_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Update the current number of records in middle node */ curr_middle_nrec = (uint16_t)(curr_middle_nrec - moved_middle_nrec); @@ -847,7 +845,7 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Move node pointers also if this is an internal node */ if(depth > 1) { - hsize_t moved_nrec; /* Total number of records moved, for internal redistrib */ + hsize_t moved_nrec; /* Total number of records moved, for internal redistrib */ unsigned u; /* Local index variable */ /* Slide the node pointers in right node up */ @@ -863,6 +861,12 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, middle_moved_nrec -= (hssize_t)(moved_nrec + right_nrec_move); } /* end if */ + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, right_node_ptrs, + 0, (unsigned)right_nrec_move, middle_child, right_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Update the current number of records in middle node */ curr_middle_nrec = (uint16_t)(curr_middle_nrec - right_nrec_move); @@ -890,7 +894,7 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Move node pointers also if this is an internal node */ if(depth > 1) { - hsize_t moved_nrec; /* Total number of records moved, for internal redistrib */ + hsize_t moved_nrec; /* Total number of records moved, for internal redistrib */ unsigned u; /* Local index variable */ /* Slide the node pointers in middle node up */ @@ -906,6 +910,12 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, middle_moved_nrec += (hssize_t)(moved_nrec + left_nrec_move); } /* end if */ + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, middle_node_ptrs, + 0, (unsigned)left_nrec_move, left_child, middle_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Update the current number of records in middle node */ curr_middle_nrec = (uint16_t)(curr_middle_nrec + left_nrec_move); @@ -932,7 +942,7 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Move node pointers also if this is an internal node */ if(depth > 1) { - hsize_t moved_nrec; /* Total number of records moved, for internal redistrib */ + hsize_t moved_nrec; /* Total number of records moved, for internal redistrib */ unsigned u; /* Local index variable */ /* Move right node pointers into middle node */ @@ -948,6 +958,12 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, HDmemmove(&(right_node_ptrs[0]), &(right_node_ptrs[right_nrec_move]), sizeof(H5B2_node_ptr_t) * (size_t)(new_right_nrec + 1)); } /* end if */ + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, middle_node_ptrs, + (unsigned)(curr_middle_nrec + 1), (unsigned)(curr_middle_nrec + right_nrec_move + 1), right_child, middle_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Mark nodes as dirty */ middle_child_flags |= H5AC__DIRTIED_FLAG; right_child_flags |= H5AC__DIRTIED_FLAG; @@ -979,46 +995,6 @@ H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Mark parent as dirty */ *internal_flags_ptr |= H5AC__DIRTIED_FLAG; -#ifdef QAK -{ - unsigned u; - - HDfprintf(stderr, "%s: Internal records:\n", FUNC); - for(u = 0; u < internal->nrec; u++) { - HDfprintf(stderr, "%s: u = %u\n", FUNC, u); - (hdr->cls->debug)(stderr, hdr->f, dxpl_id, 3, 4, H5B2_INT_NREC(internal, hdr, u), NULL); - } /* end for */ - - HDfprintf(stderr, "%s: Left Child records:\n", FUNC); - for(u = 0; u < *left_nrec; u++) { - HDfprintf(stderr, "%s: u = %u\n", FUNC, u); - (hdr->cls->debug)(stderr, hdr->f, dxpl_id, 3, 4, H5B2_NAT_NREC(left_native, hdr, u), NULL); - } /* end for */ - - HDfprintf(stderr, "%s: Middle Child records:\n", FUNC); - for(u = 0; u < *middle_nrec; u++) { - HDfprintf(stderr, "%s: u = %u\n", FUNC, u); - (hdr->cls->debug)(stderr, hdr->f, dxpl_id, 3, 4, H5B2_NAT_NREC(middle_native, hdr, u), NULL); - } /* end for */ - - HDfprintf(stderr, "%s: Right Child records:\n", FUNC); - for(u = 0; u < *right_nrec; u++) { - HDfprintf(stderr, "%s: u = %u\n", FUNC, u); - (hdr->cls->debug)(stderr, hdr->f, dxpl_id, 3, 4, H5B2_NAT_NREC(right_native, hdr, u), NULL); - } /* end for */ - - for(u = 0; u < internal->nrec + 1; u++) - HDfprintf(stderr, "%s: internal->node_ptrs[%u] = (%Hu/%u/%a)\n", FUNC, u, internal->node_ptrs[u].all_nrec, internal->node_ptrs[u].node_nrec, internal->node_ptrs[u].addr); - if(depth > 1) { - for(u = 0; u < *left_nrec + 1; u++) - HDfprintf(stderr, "%s: left_node_ptr[%u] = (%Hu/%u/%a)\n", FUNC, u, left_node_ptrs[u].all_nrec, left_node_ptrs[u].node_nrec, left_node_ptrs[u].addr); - for(u = 0; u < *middle_nrec + 1; u++) - HDfprintf(stderr, "%s: middle_node_ptr[%u] = (%Hu/%u/%a)\n", FUNC, u, middle_node_ptrs[u].all_nrec, middle_node_ptrs[u].node_nrec, middle_node_ptrs[u].addr); - for(u = 0; u < *right_nrec + 1; u++) - HDfprintf(stderr, "%s: right_node_ptr[%u] = (%Hu/%u/%a)\n", FUNC, u, right_node_ptrs[u].all_nrec, right_node_ptrs[u].node_nrec, right_node_ptrs[u].addr); - } /* end if */ -} -#endif /* QAK */ #ifdef H5B2_DEBUG H5B2__assert_internal((hsize_t)0, hdr, internal); if(depth > 1) { @@ -1062,7 +1038,7 @@ done: * *------------------------------------------------------------------------- */ -static herr_t +herr_t H5B2__merge2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx) @@ -1076,7 +1052,7 @@ H5B2__merge2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, unsigned left_child_flags = H5AC__NO_FLAGS_SET, right_child_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting child nodes */ herr_t ret_value = SUCCEED; /* Return value */ - FUNC_ENTER_STATIC + FUNC_ENTER_PACKAGE /* Check arguments. */ HDassert(hdr); @@ -1091,14 +1067,15 @@ H5B2__merge2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_INT; - left_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Lock left & right B-tree child nodes */ - if(NULL == (left_internal = H5B2__protect_internal(hdr, dxpl_id, left_addr, internal->node_ptrs[idx].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + /* (Shadow the left node if doing SWMR writes) */ + if(NULL == (left_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - if(NULL == (right_internal = H5B2__protect_internal(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], (uint16_t)(depth - 1), FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for accessing child node information */ left_child = left_internal; @@ -1116,14 +1093,15 @@ H5B2__merge2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_LEAF; - left_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Lock left & right B-tree child nodes */ - if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, left_addr, internal->node_ptrs[idx].node_nrec, H5AC__NO_FLAGS_SET))) + /* (Shadow the left node if doing SWMR writes) */ + if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for accessing child node information */ left_child = left_leaf; @@ -1146,12 +1124,20 @@ H5B2__merge2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, if(depth > 1) HDmemcpy(&(left_node_ptrs[*left_nrec + 1]), &(right_node_ptrs[0]), sizeof(H5B2_node_ptr_t) * (size_t)(*right_nrec + 1)); + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, left_node_ptrs, + (unsigned)(*left_nrec + 1), (unsigned)(*left_nrec + *right_nrec + 2), right_child, left_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Update # of records in left node */ *left_nrec = (uint16_t)(*left_nrec + *right_nrec + 1); /* Mark nodes as dirty */ left_child_flags |= H5AC__DIRTIED_FLAG; - right_child_flags |= H5AC__DIRTIED_FLAG | H5AC__DELETED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; + right_child_flags |= H5AC__DELETED_FLAG; + if(!(hdr->swmr_write)) + right_child_flags |= H5AC__DIRTIED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; } /* end block */ /* Update # of records in child nodes */ @@ -1215,7 +1201,7 @@ done: * *------------------------------------------------------------------------- */ -static herr_t +herr_t H5B2__merge3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx) @@ -1236,7 +1222,7 @@ H5B2__merge3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, unsigned middle_child_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting child nodes */ herr_t ret_value = SUCCEED; /* Return value */ - FUNC_ENTER_STATIC + FUNC_ENTER_PACKAGE /* Check arguments. */ HDassert(hdr); @@ -1252,17 +1238,18 @@ H5B2__merge3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_INT; - left_addr = internal->node_ptrs[idx - 1].addr; - middle_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Lock B-tree child nodes */ - if(NULL == (left_internal = H5B2__protect_internal(hdr, dxpl_id, left_addr, internal->node_ptrs[idx - 1].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + /* (Shadow left and middle nodes if doing SWMR writes) */ + if(NULL == (left_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx - 1], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - if(NULL == (middle_internal = H5B2__protect_internal(hdr, dxpl_id, middle_addr, internal->node_ptrs[idx].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx - 1].addr; + if(NULL == (middle_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx], (uint16_t)(depth - 1), hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - if(NULL == (right_internal = H5B2__protect_internal(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) + middle_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], (uint16_t)(depth - 1), FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for accessing child node information */ left_child = left_internal; @@ -1285,17 +1272,18 @@ H5B2__merge3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Setup information for unlocking child nodes */ child_class = H5AC_BT2_LEAF; - left_addr = internal->node_ptrs[idx - 1].addr; - middle_addr = internal->node_ptrs[idx].addr; - right_addr = internal->node_ptrs[idx + 1].addr; /* Lock B-tree child nodes */ - if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, left_addr, internal->node_ptrs[idx - 1].node_nrec, H5AC__NO_FLAGS_SET))) + /* (Shadow left and middle nodes if doing SWMR writes) */ + if(NULL == (left_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx - 1], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - if(NULL == (middle_leaf = H5B2__protect_leaf(hdr, dxpl_id, middle_addr, internal->node_ptrs[idx].node_nrec, H5AC__NO_FLAGS_SET))) + left_addr = internal->node_ptrs[idx - 1].addr; + if(NULL == (middle_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx], hdr->swmr_write, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, right_addr, internal->node_ptrs[idx + 1].node_nrec, H5AC__NO_FLAGS_SET))) + middle_addr = internal->node_ptrs[idx].addr; + if(NULL == (right_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx + 1], FALSE, H5AC__NO_FLAGS_SET))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + right_addr = internal->node_ptrs[idx + 1].addr; /* More setup for accessing child node information */ left_child = left_leaf; @@ -1344,6 +1332,12 @@ H5B2__merge3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, HDmemmove(&(middle_node_ptrs[0]), &(middle_node_ptrs[middle_nrec_move]), sizeof(H5B2_node_ptr_t) * (size_t)((unsigned)(*middle_nrec + 1) - middle_nrec_move)); } /* end if */ + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, left_node_ptrs, + (unsigned)(*left_nrec + 1), (unsigned)(*left_nrec + middle_nrec_move + 1), middle_child, left_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Update # of records in left & middle nodes */ *left_nrec = (uint16_t)(*left_nrec + middle_nrec_move); *middle_nrec = (uint16_t)(*middle_nrec - middle_nrec_move); @@ -1366,12 +1360,20 @@ H5B2__merge3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Copy node pointers from right node into middle node */ HDmemcpy(&(middle_node_ptrs[*middle_nrec + 1]), &(right_node_ptrs[0]), sizeof(H5B2_node_ptr_t) * (size_t)(*right_nrec + 1)); + /* Update flush dependencies for grandchildren, if using SWMR */ + if(hdr->swmr_write && depth > 1) + if(H5B2__update_child_flush_depends(hdr, dxpl_id, depth, middle_node_ptrs, + (unsigned)(*middle_nrec + 1), (unsigned)(*middle_nrec + *right_nrec + 2), right_child, middle_child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child nodes to new parent") + /* Update # of records in middle node */ *middle_nrec = (uint16_t)(*middle_nrec + (*right_nrec + 1)); /* Mark nodes as dirty */ middle_child_flags |= H5AC__DIRTIED_FLAG; - right_child_flags |= H5AC__DIRTIED_FLAG | H5AC__DELETED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; + right_child_flags |= H5AC__DELETED_FLAG; + if(!(hdr->swmr_write)) + right_child_flags |= H5AC__DIRTIED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; } /* end block */ /* Update # of records in child nodes */ @@ -1429,98 +1431,7 @@ done: /*------------------------------------------------------------------------- - * Function: H5B2__swap_leaf - * - * Purpose: Swap a record in a node with a record in a leaf node - * - * Return: Success: Non-negative - * - * Failure: Negative - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Mar 4 2005 - * - *------------------------------------------------------------------------- - */ -static herr_t -H5B2__swap_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - H5B2_internal_t *internal, unsigned *internal_flags_ptr, - unsigned idx, void *swap_loc) -{ - const H5AC_class_t *child_class; /* Pointer to child node's class info */ - haddr_t child_addr; /* Address of child node */ - void *child = NULL; /* Pointer to child node */ - uint8_t *child_native; /* Pointer to child's native records */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_STATIC - - /* Check arguments. */ - HDassert(hdr); - HDassert(internal); - HDassert(internal_flags_ptr); - HDassert(idx <= internal->nrec); - - /* Check for the kind of B-tree node to swap */ - if(depth > 1) { - H5B2_internal_t *child_internal; /* Pointer to internal node */ - - /* Setup information for unlocking child node */ - child_class = H5AC_BT2_INT; - child_addr = internal->node_ptrs[idx].addr; - - /* Lock B-tree child nodes */ - if(NULL == (child_internal = H5B2__protect_internal(hdr, dxpl_id, child_addr, internal->node_ptrs[idx].node_nrec, (uint16_t)(depth - 1), H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - - /* More setup for accessing child node information */ - child = child_internal; - child_native = child_internal->int_native; - } /* end if */ - else { - H5B2_leaf_t *child_leaf; /* Pointer to leaf node */ - - /* Setup information for unlocking child nodes */ - child_class = H5AC_BT2_LEAF; - child_addr = internal->node_ptrs[idx].addr; - - /* Lock B-tree child node */ - if(NULL == (child_leaf = H5B2__protect_leaf(hdr, dxpl_id, child_addr, internal->node_ptrs[idx].node_nrec, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - - /* More setup for accessing child node information */ - child = child_leaf; - child_native = child_leaf->leaf_native; - } /* end else */ - - /* Swap records (use disk page as temporary buffer) */ - HDmemcpy(hdr->page, H5B2_NAT_NREC(child_native, hdr, 0), hdr->cls->nrec_size); - HDmemcpy(H5B2_NAT_NREC(child_native, hdr, 0), swap_loc, hdr->cls->nrec_size); - HDmemcpy(swap_loc, hdr->page, hdr->cls->nrec_size); - - /* Mark parent as dirty */ - *internal_flags_ptr |= H5AC__DIRTIED_FLAG; - -#ifdef H5B2_DEBUG - H5B2__assert_internal((hsize_t)0, hdr, internal); - if(depth > 1) - H5B2__assert_internal(internal->node_ptrs[idx].all_nrec, hdr, (H5B2_internal_t *)child); - else - H5B2__assert_leaf(hdr, (H5B2_leaf_t *)child); -#endif /* H5B2_DEBUG */ - -done: - /* Unlock child node */ - if(child && H5AC_unprotect(hdr->f, dxpl_id, child_class, child_addr, child, H5AC__DIRTIED_FLAG) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree child node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* end H5B2__swap_leaf() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__insert_hdr + * Function: H5B2__insert * * Purpose: Adds a new record to the B-tree. * @@ -1533,7 +1444,7 @@ done: *------------------------------------------------------------------------- */ herr_t -H5B2__insert_hdr(H5B2_hdr_t *hdr, hid_t dxpl_id, void *udata) +H5B2__insert(H5B2_hdr_t *hdr, hid_t dxpl_id, void *udata) { herr_t ret_value = SUCCEED; /* Return value */ @@ -1546,7 +1457,7 @@ H5B2__insert_hdr(H5B2_hdr_t *hdr, hid_t dxpl_id, void *udata) /* Check if the root node is allocated yet */ if(!H5F_addr_defined(hdr->root.addr)) { /* Create root node as leaf node in B-tree */ - if(H5B2__create_leaf(hdr, dxpl_id, &(hdr->root)) < 0) + if(H5B2__create_leaf(hdr, dxpl_id, hdr, &(hdr->root)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "unable to create root node") } /* end if */ /* Check if we need to split the root node (equiv. to a 1->2 node split) */ @@ -1558,11 +1469,11 @@ H5B2__insert_hdr(H5B2_hdr_t *hdr, hid_t dxpl_id, void *udata) /* Attempt to insert record into B-tree */ if(hdr->depth > 0) { - if(H5B2__insert_internal(hdr, dxpl_id, hdr->depth, NULL, &hdr->root, H5B2_POS_ROOT, udata) < 0) + if(H5B2__insert_internal(hdr, dxpl_id, hdr->depth, NULL, &hdr->root, H5B2_POS_ROOT, hdr, udata) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into B-tree internal node") } /* end if */ else { - if(H5B2__insert_leaf(hdr, dxpl_id, &hdr->root, H5B2_POS_ROOT, udata) < 0) + if(H5B2__insert_leaf(hdr, dxpl_id, &hdr->root, H5B2_POS_ROOT, hdr, udata) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into B-tree leaf node") } /* end else */ @@ -1572,834 +1483,7 @@ H5B2__insert_hdr(H5B2_hdr_t *hdr, hid_t dxpl_id, void *udata) done: FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__insert_hdr() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__insert_leaf - * - * Purpose: Adds a new record to a B-tree leaf node. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Mar 3 2005 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__insert_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, - H5B2_nodepos_t curr_pos, void *udata) -{ - H5B2_leaf_t *leaf; /* Pointer to leaf node */ - int cmp; /* Comparison value of records */ - unsigned idx; /* Location of record which matches key */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - - /* Lock current B-tree node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node_ptr->addr, curr_node_ptr->node_nrec, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - - /* Must have a leaf node with enough space to insert a record now */ - HDassert(curr_node_ptr->node_nrec < hdr->node_info[0].max_nrec); - - /* Sanity check number of records */ - HDassert(curr_node_ptr->all_nrec == curr_node_ptr->node_nrec); - HDassert(leaf->nrec == curr_node_ptr->node_nrec); - - /* Check for inserting into empty leaf */ - if(leaf->nrec == 0) - idx = 0; - else { - /* Find correct location to insert this record */ - if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - if(cmp == 0) - HGOTO_ERROR(H5E_BTREE, H5E_EXISTS, FAIL, "record is already in B-tree") - if(cmp > 0) - idx++; - - /* Make room for new record */ - if(idx < leaf->nrec) - HDmemmove(H5B2_LEAF_NREC(leaf, hdr, idx + 1), H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size * (leaf->nrec - idx)); - } /* end else */ - - /* Make callback to store record in native form */ - if((hdr->cls->store)(H5B2_LEAF_NREC(leaf, hdr, idx), udata) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into leaf node") - - /* Update record count for node pointer to current node */ - curr_node_ptr->all_nrec++; - curr_node_ptr->node_nrec++; - - /* Update record count for current node */ - leaf->nrec++; - - /* Check for new record being the min or max for the tree */ - /* (Don't use 'else' for the idx check, to allow for root leaf node) */ - if(H5B2_POS_MIDDLE != curr_pos) { - if(idx == 0) { - if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) { - if(hdr->min_native_rec == NULL) - if(NULL == (hdr->min_native_rec = H5MM_malloc(hdr->cls->nrec_size))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "memory allocation failed for v2 B-tree min record info") - HDmemcpy(hdr->min_native_rec, H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size); - } /* end if */ - } /* end if */ - if(idx == (unsigned)(leaf->nrec - 1)) { - if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) { - if(hdr->max_native_rec == NULL) - if(NULL == (hdr->max_native_rec = H5MM_malloc(hdr->cls->nrec_size))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "memory allocation failed for v2 B-tree max record info") - HDmemcpy(hdr->max_native_rec, H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size); - } /* end if */ - } /* end if */ - } /* end if */ - -done: - /* Release the B-tree leaf node (marked as dirty) */ - if(leaf && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, curr_node_ptr->addr, leaf, H5AC__DIRTIED_FLAG) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release leaf B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__insert_leaf() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__insert_internal - * - * Purpose: Adds a new record to a B-tree node. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Mar 2 2005 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__insert_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - unsigned *parent_cache_info_flags_ptr, H5B2_node_ptr_t *curr_node_ptr, - H5B2_nodepos_t curr_pos, void *udata) -{ - H5B2_internal_t *internal = NULL; /* Pointer to internal node */ - unsigned internal_flags = H5AC__NO_FLAGS_SET; - unsigned idx; /* Location of record which matches key */ - H5B2_nodepos_t next_pos = H5B2_POS_MIDDLE; /* Position of node */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(depth > 0); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - - /* Lock current B-tree node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node_ptr->addr, curr_node_ptr->node_nrec, depth, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - - /* Sanity check number of records */ - HDassert(internal->nrec == curr_node_ptr->node_nrec); - - /* Split or redistribute child node pointers, if necessary */ - { - int cmp; /* Comparison value of records */ - unsigned retries; /* Number of times to attempt redistribution */ - size_t split_nrec; /* Number of records to split node at */ - - /* Locate node pointer for child */ - if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, - udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - if(cmp == 0) - HGOTO_ERROR(H5E_BTREE, H5E_EXISTS, FAIL, "record is already in B-tree") - if(cmp > 0) - idx++; - - /* Set the number of redistribution retries */ - /* This takes care of the case where a B-tree node needs to be - * redistributed, but redistributing the node causes the index - * for insertion to move to another node, which also needs to be - * redistributed. Now, we loop trying to redistribute and then - * eventually force a split */ - retries = 2; - - /* Determine the correct number of records to split child node at */ - split_nrec = hdr->node_info[depth - 1].split_nrec; - - /* Preemptively split/redistribute a node we will enter */ - while(internal->node_ptrs[idx].node_nrec == split_nrec) { - /* Attempt to redistribute records among children */ - if(idx == 0) { /* Left-most child */ - if(retries > 0 && (internal->node_ptrs[idx + 1].node_nrec < split_nrec)) { - if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__split1(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to split child node") - } /* end else */ - } /* end if */ - else if(idx == internal->nrec) { /* Right-most child */ - if(retries > 0 && (internal->node_ptrs[idx - 1].node_nrec < split_nrec)) { - if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, (idx - 1)) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__split1(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to split child node") - } /* end else */ - } /* end if */ - else { /* Middle child */ - if(retries > 0 && ((internal->node_ptrs[idx + 1].node_nrec < split_nrec) || - (internal->node_ptrs[idx - 1].node_nrec < split_nrec))) { - if(H5B2__redistribute3(hdr, dxpl_id, depth, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__split1(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to split child node") - } /* end else */ - } /* end else */ - - /* Locate node pointer for child (after split/redistribute) */ - /* Actually, this can be easily updated (for 2-node redistrib.) and shouldn't require re-searching */ - if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, - udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - if(cmp == 0) - HGOTO_ERROR(H5E_BTREE, H5E_EXISTS, FAIL, "record is already in B-tree") - if(cmp > 0) - idx++; - - /* Decrement the number of redistribution retries left */ - retries--; - } /* end while */ - } /* end block */ - - /* Check if this node is left/right-most */ - if(H5B2_POS_MIDDLE != curr_pos) { - if(idx == 0) { - if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) - next_pos = H5B2_POS_LEFT; - } /* end if */ - else if(idx == internal->nrec) { - if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) - next_pos = H5B2_POS_RIGHT; - } /* end else */ - } /* end if */ - - /* Attempt to insert node */ - if(depth > 1) { - if(H5B2__insert_internal(hdr, dxpl_id, (uint16_t)(depth - 1), &internal_flags, &internal->node_ptrs[idx], next_pos, udata) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into B-tree internal node") - } /* end if */ - else { - if(H5B2__insert_leaf(hdr, dxpl_id, &internal->node_ptrs[idx], next_pos, udata) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into B-tree leaf node") - } /* end else */ - - /* Update record count for node pointer to current node */ - curr_node_ptr->all_nrec++; - - /* Mark node as dirty */ - internal_flags |= H5AC__DIRTIED_FLAG; - -done: - /* Release the B-tree internal node */ - if(internal && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, internal_flags) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__insert_internal() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__update_leaf - * - * Purpose: Insert or modify a record in a B-tree leaf node. - * If the record exists already, it is modified as if H5B2_modify - * was called). If it doesn't exist, it is inserted as if - * H5B2_insert was called. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@hdfgroup.org - * Dec 23 2015 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__update_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, - H5B2_update_status_t *status, H5B2_nodepos_t curr_pos, void *udata, - H5B2_modify_t op, void *op_data) -{ - H5B2_leaf_t *leaf; /* Pointer to leaf node */ - unsigned leaf_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting the leaf node */ - int cmp = -1; /* Comparison value of records */ - unsigned idx; /* Location of record which matches key */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - - /* Lock current B-tree node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node_ptr->addr, curr_node_ptr->node_nrec, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - - /* Sanity check number of records */ - HDassert(curr_node_ptr->all_nrec == curr_node_ptr->node_nrec); - HDassert(leaf->nrec == curr_node_ptr->node_nrec); - - /* Check for inserting into empty leaf */ - if(leaf->nrec == 0) - idx = 0; - else { - /* Find correct location to insert this record */ - if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - - /* Check for inserting a record */ - if(0 != cmp) { - /* Check if the leaf node is full */ - if(curr_node_ptr->node_nrec == hdr->node_info[0].split_nrec) { - /* Indicate that the leaf is full, but we need to insert */ - *status = H5B2_UPDATE_INSERT_CHILD_FULL; - - /* Let calling routine handle insertion */ - HGOTO_DONE(SUCCEED) - } /* end if */ - - /* Adjust index to leave room for record to insert */ - if(cmp > 0) - idx++; - - /* Make room for new record */ - if(idx < leaf->nrec) - HDmemmove(H5B2_LEAF_NREC(leaf, hdr, idx + 1), H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size * (leaf->nrec - idx)); - } /* end if */ - } /* end else */ - - /* Check for modifying existing record */ - if(0 == cmp) { - hbool_t changed = FALSE; /* Whether the 'modify' callback changed the record */ - - /* Make callback for current record */ - if((op)(H5B2_LEAF_NREC(leaf, hdr, idx), op_data, &changed) < 0) { - /* Make certain that the callback didn't modify the value if it failed */ - HDassert(changed == FALSE); - - HGOTO_ERROR(H5E_BTREE, H5E_CANTMODIFY, FAIL, "'modify' callback failed for B-tree update operation") - } /* end if */ - - /* Mark the node as dirty if it changed */ - leaf_flags |= (changed ? H5AC__DIRTIED_FLAG : 0); - - /* Indicate that the record was modified */ - *status = H5B2_UPDATE_MODIFY_DONE; - } /* end if */ - else { - /* Must have a leaf node with enough space to insert a record now */ - HDassert(curr_node_ptr->node_nrec < hdr->node_info[0].max_nrec); - - /* Make callback to store record in native form */ - if((hdr->cls->store)(H5B2_LEAF_NREC(leaf, hdr, idx), udata) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into leaf node") - - /* Mark the node as dirty */ - leaf_flags |= H5AC__DIRTIED_FLAG; - - /* Indicate that the record was inserted */ - *status = H5B2_UPDATE_INSERT_DONE; - - /* Update record count for node pointer to current node */ - curr_node_ptr->all_nrec++; - curr_node_ptr->node_nrec++; - - /* Update record count for current node */ - leaf->nrec++; - } /* end else */ - - /* Check for new record being the min or max for the tree */ - /* (Don't use 'else' for the idx check, to allow for root leaf node) */ - if(H5B2_POS_MIDDLE != curr_pos) { - if(idx == 0) { - if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) { - if(hdr->min_native_rec == NULL) - if(NULL == (hdr->min_native_rec = H5MM_malloc(hdr->cls->nrec_size))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "memory allocation failed for v2 B-tree min record info") - HDmemcpy(hdr->min_native_rec, H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size); - } /* end if */ - } /* end if */ - if(idx == (unsigned)(leaf->nrec - 1)) { - if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) { - if(hdr->max_native_rec == NULL) - if(NULL == (hdr->max_native_rec = H5MM_malloc(hdr->cls->nrec_size))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "memory allocation failed for v2 B-tree max record info") - HDmemcpy(hdr->max_native_rec, H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size); - } /* end if */ - } /* end if */ - } /* end if */ - -done: - /* Release the B-tree leaf node */ - if(leaf && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, curr_node_ptr->addr, leaf, leaf_flags) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release leaf B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__update_leaf() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__update_internal - * - * Purpose: Insert or modify a record in a B-tree leaf node. - * If the record exists already, it is modified as if H5B2_modify - * was called). If it doesn't exist, it is inserted as if - * H5B2_insert was called. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@hdfgroup.org - * Dec 24 2015 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__update_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - unsigned *parent_cache_info_flags_ptr, H5B2_node_ptr_t *curr_node_ptr, - H5B2_update_status_t *status, H5B2_nodepos_t curr_pos, void *udata, - H5B2_modify_t op, void *op_data) -{ - H5B2_internal_t *internal = NULL; /* Pointer to internal node */ - unsigned internal_flags = H5AC__NO_FLAGS_SET; - int cmp; /* Comparison value of records */ - unsigned idx; /* Location of record which matches key */ - H5B2_nodepos_t next_pos = H5B2_POS_MIDDLE; /* Position of node */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(depth > 0); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - - /* Lock current B-tree node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node_ptr->addr, curr_node_ptr->node_nrec, depth, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - - /* Sanity check number of records */ - HDassert(internal->nrec == curr_node_ptr->node_nrec); - - /* Locate node pointer for child */ - if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - - /* Check for modifying existing record */ - if(0 == cmp) { - hbool_t changed = FALSE; /* Whether the 'modify' callback changed the record */ - - /* Make callback for current record */ - if((op)(H5B2_INT_NREC(internal, hdr, idx), op_data, &changed) < 0) { - /* Make certain that the callback didn't modify the value if it failed */ - HDassert(changed == FALSE); - - HGOTO_ERROR(H5E_BTREE, H5E_CANTMODIFY, FAIL, "'modify' callback failed for B-tree update operation") - } /* end if */ - - /* Mark the node as dirty if it changed */ - internal_flags |= (changed ? H5AC__DIRTIED_FLAG : 0); - - /* Indicate that the record was modified */ - *status = H5B2_UPDATE_MODIFY_DONE; - } /* end if */ - else { - /* Adjust index to leave room for node to insert */ - if(cmp > 0) - idx++; - - /* Check if this node is left/right-most */ - if(H5B2_POS_MIDDLE != curr_pos) { - if(idx == 0) { - if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) - next_pos = H5B2_POS_LEFT; - } /* end if */ - else if(idx == internal->nrec) { - if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) - next_pos = H5B2_POS_RIGHT; - } /* end else */ - } /* end if */ - - /* Attempt to update record in child */ - if(depth > 1) { - if(H5B2__update_internal(hdr, dxpl_id, (uint16_t)(depth - 1), &internal_flags, &internal->node_ptrs[idx], status, next_pos, udata, op, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update record in internal B-tree node") - } /* end if */ - else { - if(H5B2__update_leaf(hdr, dxpl_id, &internal->node_ptrs[idx], status, next_pos, udata, op, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update record in leaf B-tree node") - } /* end else */ - - /* Take actions based on child's status report */ - switch(*status) { - case H5B2_UPDATE_MODIFY_DONE: - /* No action */ - break; - - case H5B2_UPDATE_INSERT_DONE: - /* Mark node as dirty */ - internal_flags |= H5AC__DIRTIED_FLAG; - - /* Update total record count for node pointer to current node */ - curr_node_ptr->all_nrec++; - break; - - case H5B2_UPDATE_INSERT_CHILD_FULL: - /* Split/redistribute this node */ - if(internal->nrec == hdr->node_info[depth].split_nrec) { - hbool_t could_split = FALSE; /* Whether the child node could split */ - -#ifdef QAK -HDfprintf(stderr, "%s: idx = %u, internal->nrec = %u\n", FUNC, idx, internal->nrec); -HDfprintf(stderr, "%s: hdr->node_info[%u].split_nrec = %u\n", FUNC, (unsigned)depth, (unsigned)hdr->node_info[depth].split_nrec); -HDfprintf(stderr, "%s: hdr->node_info[%u].split_nrec = %u\n", FUNC, (unsigned)(depth - 1), (unsigned)hdr->node_info[depth - 1].split_nrec); -#endif /* QAK */ - if(idx == 0) { /* Left-most child */ -#ifdef QAK -HDfprintf(stderr, "%s: Left-most child\n", FUNC); -HDfprintf(stderr, "%s: internal->node_ptrs[%u].node_nrec = %u\n", FUNC, (unsigned)idx, (unsigned)internal->node_ptrs[idx].node_nrec); -HDfprintf(stderr, "%s: internal->node_ptrs[%u].node_nrec = %u\n", FUNC, (unsigned)(idx + 1), (unsigned)internal->node_ptrs[idx + 1].node_nrec); -#endif /* QAK */ - /* Check for left-most child and its neighbor being close to full */ - if((internal->node_ptrs[idx].node_nrec + internal->node_ptrs[idx + 1].node_nrec) - >= ((hdr->node_info[depth - 1].split_nrec * 2) - 1)) - could_split = TRUE; - } /* end if */ - else if(idx == internal->nrec) { /* Right-most child */ -#ifdef QAK -HDfprintf(stderr, "%s: Right-most child\n", FUNC); -HDfprintf(stderr, "%s: internal->node_ptrs[%u].node_nrec = %u\n", FUNC, (unsigned)(idx - 1), (unsigned)internal->node_ptrs[idx - 1].node_nrec); -HDfprintf(stderr, "%s: internal->node_ptrs[%u].node_nrec = %u\n", FUNC, (unsigned)idx, (unsigned)internal->node_ptrs[idx].node_nrec); -#endif /* QAK */ - /* Check for right-most child and its neighbor being close to full */ - if((internal->node_ptrs[idx - 1].node_nrec + internal->node_ptrs[idx].node_nrec) - >= ((hdr->node_info[depth - 1].split_nrec * 2) - 1)) - could_split = TRUE; - } /* end else-if */ - else { /* Middle child */ -#ifdef QAK -HDfprintf(stderr, "%s: Middle child\n", FUNC); -HDfprintf(stderr, "%s: internal->node_ptrs[%u].node_nrec = %u\n", FUNC, (unsigned)(idx - 1), (unsigned)internal->node_ptrs[idx - 1].node_nrec); -HDfprintf(stderr, "%s: internal->node_ptrs[%u].node_nrec = %u\n", FUNC, (unsigned)idx, (unsigned)internal->node_ptrs[idx].node_nrec); -HDfprintf(stderr, "%s: internal->node_ptrs[%u].node_nrec = %u\n", FUNC, (unsigned)(idx + 1), (unsigned)internal->node_ptrs[idx + 1].node_nrec); -#endif /* QAK */ - /* Check for middle child and its left neighbor being close to full */ - if((internal->node_ptrs[idx - 1].node_nrec + internal->node_ptrs[idx].node_nrec) - >= ((hdr->node_info[depth - 1].split_nrec * 2) - 1)) - could_split = TRUE; - /* Check for middle child and its right neighbor being close to full */ - else if((internal->node_ptrs[idx].node_nrec + internal->node_ptrs[idx + 1].node_nrec) - >= ((hdr->node_info[depth - 1].split_nrec * 2) - 1)) - could_split = TRUE; - } /* end if */ - - /* If this node is full and the child node insertion could - * cause a split, punt back up to caller, leaving the - * "insert child full" status. - */ - if(could_split) { - /* Release the internal B-tree node */ - if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, internal_flags) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") - internal = NULL; - -#ifdef QAK -HDfprintf(stderr, "%s: Punting back to caller\n", FUNC); -#endif /* QAK */ - /* Punt back to caller */ - HGOTO_DONE(SUCCEED); - } /* end if */ - } /* end if */ - - /* Release the internal B-tree node */ - if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, internal_flags) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") - internal = NULL; - - /* Indicate that the record was inserted */ - *status = H5B2_UPDATE_INSERT_DONE; - - /* Dodge sideways into inserting a record into this node */ - if(H5B2__insert_internal(hdr, dxpl_id, depth, parent_cache_info_flags_ptr, curr_node_ptr, curr_pos, udata) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into internal B-tree node") - break; - - case H5B2_UPDATE_UNKNOWN: - default: - HDassert(0 && "Invalid update status"); - HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "invalid update status") - } /* end switch */ - } /* end else */ - -done: - /* Release the internal B-tree node */ - if(internal && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, internal_flags) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__update_internal() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__create_leaf - * - * Purpose: Creates empty leaf node of a B-tree and update node pointer - * to point to it. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Feb 2 2005 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__create_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *node_ptr) -{ - H5B2_leaf_t *leaf = NULL; /* Pointer to new leaf node created */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(node_ptr); - - /* Allocate memory for leaf information */ - if(NULL == (leaf = H5FL_MALLOC(H5B2_leaf_t))) - HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree leaf info") - - /* Set metadata cache info */ - HDmemset(&leaf->cache_info, 0, sizeof(H5AC_info_t)); - - /* Increment ref. count on B-tree header */ - if(H5B2__hdr_incr(hdr) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINC, FAIL, "can't increment ref. count on B-tree header") - - /* Share B-tree header information */ - leaf->hdr = hdr; - - /* Allocate space for the native keys in memory */ - if(NULL == (leaf->leaf_native = (uint8_t *)H5FL_FAC_MALLOC(hdr->node_info[0].nat_rec_fac))) - HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree leaf native keys") - HDmemset(leaf->leaf_native, 0, hdr->cls->nrec_size * hdr->node_info[0].max_nrec); - - /* Set number of records */ - leaf->nrec = 0; - - /* Allocate space on disk for the leaf */ - if(HADDR_UNDEF == (node_ptr->addr = H5MF_alloc(hdr->f, H5FD_MEM_BTREE, dxpl_id, (hsize_t)hdr->node_size))) - HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "file allocation failed for B-tree leaf node") - - /* Cache the new B-tree node */ - if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_BT2_LEAF, node_ptr->addr, leaf, H5AC__NO_FLAGS_SET) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "can't add B-tree leaf to cache") - -done: - if(ret_value < 0) { - if(leaf) - if(H5B2__leaf_free(leaf) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTFREE, FAIL, "unable to release v2 B-tree leaf node") - } /* end if */ - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__create_leaf() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__protect_leaf - * - * Purpose: "Protect" an leaf node in the metadata cache - * - * Return: Pointer to leaf node on success/NULL on failure - * - * Programmer: Quincey Koziol - * koziol@hdfgroup.org - * May 5 2010 - * - *------------------------------------------------------------------------- - */ -H5B2_leaf_t * -H5B2__protect_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, haddr_t addr, uint16_t nrec, - unsigned flags) -{ - H5B2_leaf_cache_ud_t udata; /* User-data for callback */ - H5B2_leaf_t *ret_value = NULL; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(H5F_addr_defined(addr)); - - /* only H5AC__READ_ONLY_FLAG may appear in flags */ - HDassert((flags & (unsigned)(~H5AC__READ_ONLY_FLAG)) == 0); - - /* Set up user data for callback */ - udata.f = hdr->f; - udata.hdr = hdr; - H5_CHECKED_ASSIGN(udata.nrec, uint16_t, nrec, unsigned) - - /* Protect the leaf node */ - if(NULL == (ret_value = (H5B2_leaf_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_BT2_LEAF, addr, &udata, flags))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, NULL, "unable to protect B-tree leaf node") - -done: - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__protect_leaf() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__create_internal - * - * Purpose: Creates empty internal node of a B-tree and update node pointer - * to point to it. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Feb 3 2005 - * - *------------------------------------------------------------------------- - */ -static herr_t -H5B2__create_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *node_ptr, - uint16_t depth) -{ - H5B2_internal_t *internal = NULL; /* Pointer to new internal node created */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_STATIC - - /* Check arguments. */ - HDassert(hdr); - HDassert(node_ptr); - HDassert(depth > 0); - - /* Allocate memory for internal node information */ - if(NULL == (internal = H5FL_MALLOC(H5B2_internal_t))) - HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree internal info") - - /* Set metadata cache info */ - HDmemset(&internal->cache_info, 0, sizeof(H5AC_info_t)); - - /* Increment ref. count on B-tree header */ - if(H5B2__hdr_incr(hdr) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINC, FAIL, "can't increment ref. count on B-tree header") - - /* Share B-tree header information */ - internal->hdr = hdr; - - /* Allocate space for the native keys in memory */ - if(NULL == (internal->int_native = (uint8_t *)H5FL_FAC_MALLOC(hdr->node_info[depth].nat_rec_fac))) - HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree internal native keys") - HDmemset(internal->int_native, 0, hdr->cls->nrec_size * hdr->node_info[depth].max_nrec); - - /* Allocate space for the node pointers in memory */ - if(NULL == (internal->node_ptrs = (H5B2_node_ptr_t *)H5FL_FAC_MALLOC(hdr->node_info[depth].node_ptr_fac))) - HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree internal node pointers") - HDmemset(internal->node_ptrs, 0, sizeof(H5B2_node_ptr_t) * (hdr->node_info[depth].max_nrec + 1)); - - /* Set number of records & depth of the node */ - internal->nrec = 0; - internal->depth = depth; - - /* Allocate space on disk for the internal node */ - if(HADDR_UNDEF == (node_ptr->addr = H5MF_alloc(hdr->f, H5FD_MEM_BTREE, dxpl_id, (hsize_t)hdr->node_size))) - HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "file allocation failed for B-tree internal node") - - /* Cache the new B-tree node */ - if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_BT2_INT, node_ptr->addr, internal, H5AC__NO_FLAGS_SET) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "can't add B-tree internal node to cache") - -done: - if(ret_value < 0) { - if(internal) - if(H5B2__internal_free(internal) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTFREE, FAIL, "unable to release v2 B-tree internal node") - } /* end if */ - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__create_internal() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__protect_internal - * - * Purpose: "Protect" an internal node in the metadata cache - * - * Return: Pointer to internal node on success/NULL on failure - * - * Programmer: Quincey Koziol - * koziol@hdfgroup.org - * Aug 25 2006 - * - *------------------------------------------------------------------------- - */ -H5B2_internal_t * -H5B2__protect_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, haddr_t addr, - uint16_t nrec, uint16_t depth, unsigned flags) -{ - H5B2_internal_cache_ud_t udata; /* User data to pass through to cache 'deserialize' callback */ - H5B2_internal_t *ret_value = NULL; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(H5F_addr_defined(addr)); - HDassert(depth > 0); - - /* only H5AC__READ_ONLY_FLAG may appear in flags */ - HDassert((flags & (unsigned)(~H5AC__READ_ONLY_FLAG)) == 0); - - /* Set up user data for callback */ - udata.f = hdr->f; - udata.hdr = hdr; - udata.nrec = nrec; - udata.depth = depth; - - /* Protect the internal node */ - if(NULL == (ret_value = (H5B2_internal_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_BT2_INT, addr, &udata, flags))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, NULL, "unable to protect B-tree internal node") - -done: - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__protect_internal() */ +} /* H5B2__insert() */ /*------------------------------------------------------------------------- @@ -2421,13 +1505,15 @@ done: */ herr_t H5B2__iterate_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - const H5B2_node_ptr_t *curr_node, H5B2_operator_t op, void *op_data) + const H5B2_node_ptr_t *curr_node, void *parent, H5B2_operator_t op, + void *op_data) { const H5AC_class_t *curr_node_class = NULL; /* Pointer to current node's class info */ void *node = NULL; /* Pointers to current node */ uint8_t *node_native; /* Pointers to node's native records */ uint8_t *native = NULL; /* Pointers to copy of node's native records */ H5B2_node_ptr_t *node_ptrs = NULL; /* Pointers to node's node pointers */ + hbool_t node_pinned = FALSE; /* Whether node is pinned */ unsigned u; /* Local index */ herr_t ret_value = H5_ITER_CONT; /* Iterator return value */ @@ -2443,7 +1529,7 @@ H5B2__iterate_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5B2_internal_t *internal; /* Pointer to internal node */ /* Lock the current B-tree node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node->addr, curr_node->node_nrec, depth, H5AC__READ_ONLY_FLAG))) + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, (H5B2_node_ptr_t *)curr_node, depth, FALSE, H5AC__READ_ONLY_FLAG))) /* Casting away const OK -QAK */ HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") /* Set up information about current node */ @@ -2462,7 +1548,7 @@ H5B2__iterate_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5B2_leaf_t *leaf; /* Pointer to leaf node */ /* Lock the current B-tree node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node->addr, curr_node->node_nrec, H5AC__READ_ONLY_FLAG))) + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, (H5B2_node_ptr_t *)curr_node, FALSE, H5AC__READ_ONLY_FLAG))) /* Casting away const OK -QAK */ HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") /* Set up information about current node */ @@ -2479,15 +1565,18 @@ H5B2__iterate_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, HDmemcpy(native, node_native, (hdr->cls->nrec_size * curr_node->node_nrec)); /* Unlock the node */ - if(H5AC_unprotect(hdr->f, dxpl_id, curr_node_class, curr_node->addr, node, H5AC__NO_FLAGS_SET) < 0) + if(H5AC_unprotect(hdr->f, dxpl_id, curr_node_class, curr_node->addr, node, (unsigned)(hdr->swmr_write ? H5AC__PIN_ENTRY_FLAG : H5AC__NO_FLAGS_SET)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree node") - node = NULL; + if(hdr->swmr_write) + node_pinned = TRUE; + else + node = NULL; /* Iterate through records, in order */ for(u = 0; u < curr_node->node_nrec && !ret_value; u++) { /* Descend into child node, if current node is an internal node */ if(depth > 0) { - if((ret_value = H5B2__iterate_node(hdr, dxpl_id, (uint16_t)(depth - 1), &(node_ptrs[u]), op, op_data)) < 0) + if((ret_value = H5B2__iterate_node(hdr, dxpl_id, (uint16_t)(depth - 1), &(node_ptrs[u]), node, op, op_data)) < 0) HERROR(H5E_BTREE, H5E_CANTLIST, "node iteration failed"); } /* end if */ @@ -2500,11 +1589,15 @@ H5B2__iterate_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Descend into last child node, if current node is an internal node */ if(!ret_value && depth > 0) { - if((ret_value = H5B2__iterate_node(hdr, dxpl_id, (uint16_t)(depth - 1), &(node_ptrs[u]), op, op_data)) < 0) + if((ret_value = H5B2__iterate_node(hdr, dxpl_id, (uint16_t)(depth - 1), &(node_ptrs[u]), node, op, op_data)) < 0) HERROR(H5E_BTREE, H5E_CANTLIST, "node iteration failed"); } /* end if */ done: + /* Unpin the node if it was pinned */ + if(node_pinned && H5AC_unpin_entry(node) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "can't unpin node") + /* Release the node pointers & native records, if they were copied */ if(node_ptrs) node_ptrs = (H5B2_node_ptr_t *)H5FL_FAC_FREE(hdr->node_info[depth].node_ptr_fac, node_ptrs); @@ -2516,867 +1609,6 @@ done: /*------------------------------------------------------------------------- - * Function: H5B2__remove_leaf - * - * Purpose: Removes a record from a B-tree leaf node. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Mar 3 2005 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__remove_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, - H5B2_nodepos_t curr_pos, void *udata, H5B2_remove_t op, void *op_data) -{ - H5B2_leaf_t *leaf; /* Pointer to leaf node */ - haddr_t leaf_addr = HADDR_UNDEF; /* Leaf address on disk */ - unsigned leaf_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting leaf node */ - unsigned idx; /* Location of record which matches key */ - int cmp; /* Comparison value of records */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - - /* Lock current B-tree node */ - leaf_addr = curr_node_ptr->addr; - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, leaf_addr, curr_node_ptr->node_nrec, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - - /* Sanity check number of records */ - HDassert(curr_node_ptr->all_nrec == curr_node_ptr->node_nrec); - HDassert(leaf->nrec == curr_node_ptr->node_nrec); - - /* Find correct location to remove this record */ - if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - if(cmp != 0) - HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "record is not in B-tree") - - /* Check for invalidating the min/max record for the tree */ - if(H5B2_POS_MIDDLE != curr_pos) { - /* (Don't use 'else' for the idx check, to allow for root leaf node) */ - if(idx == 0) { - if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) { - if(hdr->min_native_rec) - hdr->min_native_rec = H5MM_xfree(hdr->min_native_rec); - } /* end if */ - } /* end if */ - if(idx == (unsigned)(leaf->nrec - 1)) { - if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) { - if(hdr->max_native_rec) - hdr->max_native_rec = H5MM_xfree(hdr->max_native_rec); - } /* end if */ - } /* end if */ - } /* end if */ - - /* Make 'remove' callback if there is one */ - if(op) - if((op)(H5B2_LEAF_NREC(leaf, hdr, idx), op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record into leaf node") - - /* Update number of records in node */ - leaf->nrec--; - - /* Mark leaf node as dirty also */ - leaf_flags |= H5AC__DIRTIED_FLAG; - - if(leaf->nrec > 0) { - /* Pack record out of leaf */ - if(idx < leaf->nrec) - HDmemmove(H5B2_LEAF_NREC(leaf, hdr, idx), H5B2_LEAF_NREC(leaf, hdr, (idx + 1)), hdr->cls->nrec_size * (leaf->nrec - idx)); - } /* end if */ - else { - /* Let the cache know that the object is deleted */ - leaf_flags |= H5AC__DELETED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; - - /* Reset address of parent node pointer */ - curr_node_ptr->addr = HADDR_UNDEF; - } /* end else */ - - /* Update record count for parent of leaf node */ - curr_node_ptr->node_nrec--; - -done: - /* Release the B-tree leaf node */ - if(leaf && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, leaf_addr, leaf, leaf_flags) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release leaf B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__remove_leaf() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__remove_internal - * - * Purpose: Removes a record from a B-tree node. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Mar 3 2005 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__remove_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, hbool_t *depth_decreased, - void *swap_loc, uint16_t depth, H5AC_info_t *parent_cache_info, - unsigned *parent_cache_info_flags_ptr, H5B2_nodepos_t curr_pos, - H5B2_node_ptr_t *curr_node_ptr, void *udata, H5B2_remove_t op, void *op_data) -{ - H5AC_info_t *new_cache_info; /* Pointer to new cache info */ - unsigned *new_cache_info_flags_ptr = NULL; - H5B2_node_ptr_t *new_node_ptr; /* Pointer to new node pointer */ - H5B2_internal_t *internal; /* Pointer to internal node */ - H5B2_nodepos_t next_pos = H5B2_POS_MIDDLE; /* Position of next node */ - unsigned internal_flags = H5AC__NO_FLAGS_SET; - haddr_t internal_addr; /* Address of internal node */ - size_t merge_nrec; /* Number of records to merge node at */ - hbool_t collapsed_root = FALSE; /* Whether the root was collapsed */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(depth > 0); - HDassert(parent_cache_info); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - - /* Lock current B-tree node */ - internal_addr = curr_node_ptr->addr; - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, internal_addr, curr_node_ptr->node_nrec, depth, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - - /* Determine the correct number of records to merge at */ - merge_nrec = hdr->node_info[depth - 1].merge_nrec; - - /* Check for needing to collapse the root node */ - /* (The root node is the only internal node allowed to have 1 record) */ - if(internal->nrec == 1 && - ((internal->node_ptrs[0].node_nrec + internal->node_ptrs[1].node_nrec) <= ((merge_nrec * 2) + 1))) { - - /* Merge children of root node */ - if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, 0) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") - - /* Let the cache know that the object is deleted */ - internal_flags |= H5AC__DELETED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; - - /* Reset information in header's root node pointer */ - curr_node_ptr->addr = internal->node_ptrs[0].addr; - curr_node_ptr->node_nrec = internal->node_ptrs[0].node_nrec; - - /* Indicate that the level of the B-tree decreased */ - *depth_decreased = TRUE; - - /* Set pointers for advancing to child node */ - new_cache_info = parent_cache_info; - new_cache_info_flags_ptr = parent_cache_info_flags_ptr; - new_node_ptr = curr_node_ptr; - - /* Set flag to indicate root was collapsed */ - collapsed_root = TRUE; - - /* Indicate position of next node */ - next_pos = H5B2_POS_ROOT; - } /* end if */ - /* Merge or redistribute child node pointers, if necessary */ - else { - unsigned idx; /* Location of record which matches key */ - int cmp = 0; /* Comparison value of records */ - unsigned retries; /* Number of times to attempt redistribution */ - - /* Locate node pointer for child */ - if(swap_loc) - idx = 0; - else { - if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, - udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - if(cmp >= 0) - idx++; - } /* end else */ - - /* Set the number of redistribution retries */ - /* This takes care of the case where a B-tree node needs to be - * redistributed, but redistributing the node causes the index - * for removal to move to another node, which also needs to be - * redistributed. Now, we loop trying to redistribute and then - * eventually force a merge */ - retries = 2; - - /* Preemptively merge/redistribute a node we will enter */ - while(internal->node_ptrs[idx].node_nrec == merge_nrec) { - /* Attempt to redistribute records among children */ - /* (NOTE: These 2-node redistributions should actually get the - * record to promote from the node with more records. - QAK) - */ - /* (NOTE: This code is the same in both H5B2__remove_internal() and - * H5B2__remove_internal_by_idx(), fix bugs in both places! - QAK) - */ - if(idx == 0) { /* Left-most child */ - if(retries > 0 && (internal->node_ptrs[idx + 1].node_nrec > merge_nrec)) { - if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") - } /* end else */ - } /* end if */ - else if(idx == internal->nrec) { /* Right-most child */ - if(retries > 0 && (internal->node_ptrs[idx - 1].node_nrec > merge_nrec)) { - if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, (idx - 1)) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, (idx - 1)) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") - } /* end else */ - } /* end if */ - else { /* Middle child */ - if(retries > 0 && ((internal->node_ptrs[idx + 1].node_nrec > merge_nrec) || - (internal->node_ptrs[idx - 1].node_nrec > merge_nrec))) { - if(H5B2__redistribute3(hdr, dxpl_id, depth, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__merge3(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") - } /* end else */ - } /* end else */ - - /* Locate node pointer for child (after merge/redistribute) */ - if(swap_loc) - idx = 0; - else { -/* Actually, this can be easily updated (for 2-node redistrib.) and shouldn't require re-searching */ - if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - if(cmp >= 0) - idx++; - } /* end else */ - - /* Decrement the number of redistribution retries left */ - retries--; - } /* end while */ - - /* Handle deleting a record from an internal node */ - if(!swap_loc && cmp == 0) - swap_loc = H5B2_INT_NREC(internal, hdr, idx - 1); - - /* Swap record to delete with record from leaf, if we are the last internal node */ - if(swap_loc && depth == 1) - if(H5B2__swap_leaf(hdr, dxpl_id, depth, internal, &internal_flags, idx, swap_loc) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSWAP, FAIL, "Can't swap records in B-tree") - - /* Set pointers for advancing to child node */ - new_cache_info_flags_ptr = &internal_flags; - new_cache_info = &internal->cache_info; - new_node_ptr = &internal->node_ptrs[idx]; - - /* Indicate position of next node */ - if(H5B2_POS_MIDDLE != curr_pos) { - if(idx == 0) { - if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) - next_pos = H5B2_POS_LEFT; - } /* end if */ - else if(idx == internal->nrec) { - if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) - next_pos = H5B2_POS_RIGHT; - } /* end if */ - } /* end if */ - } /* end else */ - - /* Attempt to remove record from child node */ - if(depth > 1) { - if(H5B2__remove_internal(hdr, dxpl_id, depth_decreased, swap_loc, (uint16_t)(depth - 1), - new_cache_info, new_cache_info_flags_ptr, next_pos, new_node_ptr, udata, op, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree internal node") - } /* end if */ - else { - if(H5B2__remove_leaf(hdr, dxpl_id, new_node_ptr, next_pos, udata, op, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree leaf node") - } /* end else */ - - /* Update record count for node pointer to current node */ - if(!collapsed_root) - new_node_ptr->all_nrec--; - - /* Mark node as dirty */ - internal_flags |= H5AC__DIRTIED_FLAG; - -#ifdef H5B2_DEBUG - H5B2__assert_internal((!collapsed_root ? (curr_node_ptr->all_nrec - 1) : new_node_ptr->all_nrec), hdr, internal); -#endif /* H5B2_DEBUG */ - -done: - /* Release the B-tree internal node */ - if(internal && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, internal_addr, internal, internal_flags) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__remove_internal() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__remove_leaf_by_idx - * - * Purpose: Removes a record from a B-tree leaf node, according to the - * offset in the B-tree records. - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@hdfgroup.org - * Nov 14 2006 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__remove_leaf_by_idx(H5B2_hdr_t *hdr, hid_t dxpl_id, - H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, - unsigned idx, H5B2_remove_t op, void *op_data) -{ - H5B2_leaf_t *leaf; /* Pointer to leaf node */ - haddr_t leaf_addr = HADDR_UNDEF; /* Leaf address on disk */ - unsigned leaf_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting leaf node */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - - /* Lock B-tree leaf node */ - leaf_addr = curr_node_ptr->addr; - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, leaf_addr, curr_node_ptr->node_nrec, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - - /* Sanity check number of records */ - HDassert(curr_node_ptr->all_nrec == curr_node_ptr->node_nrec); - HDassert(leaf->nrec == curr_node_ptr->node_nrec); - HDassert(idx < leaf->nrec); - - /* Check for invalidating the min/max record for the tree */ - if(H5B2_POS_MIDDLE != curr_pos) { - /* (Don't use 'else' for the idx check, to allow for root leaf node) */ - if(idx == 0) { - if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) { - if(hdr->min_native_rec) - hdr->min_native_rec = H5MM_xfree(hdr->min_native_rec); - } /* end if */ - } /* end if */ - if(idx == (unsigned)(leaf->nrec - 1)) { - if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) { - if(hdr->max_native_rec) - hdr->max_native_rec = H5MM_xfree(hdr->max_native_rec); - } /* end if */ - } /* end if */ - } /* end if */ - - /* Make 'remove' callback if there is one */ - if(op) - if((op)(H5B2_LEAF_NREC(leaf, hdr, idx), op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record into leaf node") - - /* Update number of records in node */ - leaf->nrec--; - - /* Mark leaf node as dirty also */ - leaf_flags |= H5AC__DIRTIED_FLAG; - - if(leaf->nrec > 0) { - /* Pack record out of leaf */ - if(idx < leaf->nrec) - HDmemmove(H5B2_LEAF_NREC(leaf, hdr, idx), H5B2_LEAF_NREC(leaf, hdr, (idx + 1)), hdr->cls->nrec_size * (leaf->nrec - idx)); - } /* end if */ - else { - /* Let the cache know that the object is deleted */ - leaf_flags |= H5AC__DELETED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; - - /* Reset address of parent node pointer */ - curr_node_ptr->addr = HADDR_UNDEF; - } /* end else */ - - /* Update record count for parent of leaf node */ - curr_node_ptr->node_nrec--; - -done: - /* Release the B-tree leaf node */ - if(leaf && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, leaf_addr, leaf, leaf_flags) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release leaf B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__remove_leaf_by_idx() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__remove_internal_by_idx - * - * Purpose: Removes a record from a B-tree node, according to the offset - * in the B-tree records - * - * Return: Non-negative on success/Negative on failure - * - * Programmer: Quincey Koziol - * koziol@hdfgroup.org - * Nov 14 2006 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__remove_internal_by_idx(H5B2_hdr_t *hdr, hid_t dxpl_id, - hbool_t *depth_decreased, void *swap_loc, uint16_t depth, - H5AC_info_t *parent_cache_info, unsigned *parent_cache_info_flags_ptr, - H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, hsize_t n, - H5B2_remove_t op, void *op_data) -{ - H5AC_info_t *new_cache_info; /* Pointer to new cache info */ - unsigned *new_cache_info_flags_ptr = NULL; - H5B2_node_ptr_t *new_node_ptr; /* Pointer to new node pointer */ - H5B2_internal_t *internal; /* Pointer to internal node */ - H5B2_nodepos_t next_pos = H5B2_POS_MIDDLE; /* Position of next node */ - unsigned internal_flags = H5AC__NO_FLAGS_SET; - haddr_t internal_addr; /* Address of internal node */ - size_t merge_nrec; /* Number of records to merge node at */ - hbool_t collapsed_root = FALSE; /* Whether the root was collapsed */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(depth > 0); - HDassert(parent_cache_info); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - - /* Lock current B-tree node */ - internal_addr = curr_node_ptr->addr; - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, internal_addr, curr_node_ptr->node_nrec, depth, H5AC__NO_FLAGS_SET))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - HDassert(internal->nrec == curr_node_ptr->node_nrec); - HDassert(depth == hdr->depth || internal->nrec > 1); - - /* Determine the correct number of records to merge at */ - merge_nrec = hdr->node_info[depth - 1].merge_nrec; - - /* Check for needing to collapse the root node */ - /* (The root node is the only internal node allowed to have 1 record) */ - if(internal->nrec == 1 && - ((internal->node_ptrs[0].node_nrec + internal->node_ptrs[1].node_nrec) <= ((merge_nrec * 2) + 1))) { - HDassert(depth == hdr->depth); - - /* Merge children of root node */ - if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, 0) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") - - /* Let the cache know that the object is deleted */ - internal_flags |= H5AC__DELETED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; - - /* Reset information in header's root node pointer */ - curr_node_ptr->addr = internal->node_ptrs[0].addr; - curr_node_ptr->node_nrec = internal->node_ptrs[0].node_nrec; - - /* Indicate that the level of the B-tree decreased */ - *depth_decreased = TRUE; - - /* Set pointers for advancing to child node */ - new_cache_info = parent_cache_info; - new_cache_info_flags_ptr = parent_cache_info_flags_ptr; - new_node_ptr = curr_node_ptr; - - /* Set flag to indicate root was collapsed */ - collapsed_root = TRUE; - - /* Indicate position of next node */ - next_pos = H5B2_POS_ROOT; - } /* end if */ - /* Merge or redistribute child node pointers, if necessary */ - else { - hsize_t orig_n = n; /* Original index looked for */ - unsigned idx; /* Location of record which matches key */ - hbool_t found = FALSE; /* Comparison value of records */ - unsigned retries; /* Number of times to attempt redistribution */ - - /* Locate node pointer for child */ - if(swap_loc) - idx = 0; - else { - /* Search for record with correct index */ - for(idx = 0; idx < internal->nrec; idx++) { - /* Check which child node contains indexed record */ - if(internal->node_ptrs[idx].all_nrec >= n) { - /* Check if record is in this node */ - if(internal->node_ptrs[idx].all_nrec == n) { - /* Indicate the record was found and that the index - * in child nodes is zero from now on - */ - found = TRUE; - n = 0; - - /* Increment to next record */ - idx++; - } /* end if */ - - /* Break out of loop early */ - break; - } /* end if */ - - /* Decrement index we are looking for to account for the node we - * just advanced past. - */ - n -= (internal->node_ptrs[idx].all_nrec + 1); - } /* end for */ - } /* end else */ - - /* Set the number of redistribution retries */ - /* This takes care of the case where a B-tree node needs to be - * redistributed, but redistributing the node causes the index - * for removal to move to another node, which also needs to be - * redistributed. Now, we loop trying to redistribute and then - * eventually force a merge */ - retries = 2; - - /* Preemptively merge/redistribute a node we will enter */ - while(internal->node_ptrs[idx].node_nrec == merge_nrec) { - /* Attempt to redistribute records among children */ - /* (NOTE: These 2-node redistributions should actually get the - * record to promote from the node with more records. - QAK) - */ - /* (NOTE: This code is the same in both H5B2__remove_internal() and - * H5B2__remove_internal_by_idx(), fix bugs in both places! - QAK) - */ - if(idx == 0) { /* Left-most child */ - if(retries > 0 && (internal->node_ptrs[idx + 1].node_nrec > merge_nrec)) { - if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") - } /* end else */ - } /* end if */ - else if(idx == internal->nrec) { /* Right-most child */ - if(retries > 0 && (internal->node_ptrs[idx - 1].node_nrec > merge_nrec)) { - if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, (idx - 1)) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, (idx - 1)) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") - } /* end else */ - } /* end if */ - else { /* Middle child */ - if(retries > 0 && ((internal->node_ptrs[idx + 1].node_nrec > merge_nrec) || - (internal->node_ptrs[idx - 1].node_nrec > merge_nrec))) { - if(H5B2__redistribute3(hdr, dxpl_id, depth, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") - } /* end if */ - else { - if(H5B2__merge3(hdr, dxpl_id, depth, curr_node_ptr, - parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") - } /* end else */ - } /* end else */ - - /* Locate node pointer for child (after merge/redistribute) */ - if(swap_loc) - idx = 0; - else { - /* Count from the orginal index value again */ - n = orig_n; - - /* Reset "found" flag - the record may have shifted during the - * redistribute/merge - */ - found = FALSE; - - /* Search for record with correct index */ - for(idx = 0; idx < internal->nrec; idx++) { - /* Check which child node contains indexed record */ - if(internal->node_ptrs[idx].all_nrec >= n) { - /* Check if record is in this node */ - if(internal->node_ptrs[idx].all_nrec == n) { - /* Indicate the record was found and that the index - * in child nodes is zero from now on - */ - found = TRUE; - n = 0; - - /* Increment to next record */ - idx++; - } /* end if */ - - /* Break out of loop early */ - break; - } /* end if */ - - /* Decrement index we are looking for to account for the node we - * just advanced past. - */ - n -= (internal->node_ptrs[idx].all_nrec + 1); - } /* end for */ - } /* end else */ - - /* Decrement the number of redistribution retries left */ - retries--; - } /* end while */ - - /* Handle deleting a record from an internal node */ - if(!swap_loc && found) - swap_loc = H5B2_INT_NREC(internal, hdr, idx - 1); - - /* Swap record to delete with record from leaf, if we are the last internal node */ - if(swap_loc && depth == 1) - if(H5B2__swap_leaf(hdr, dxpl_id, depth, internal, &internal_flags, idx, swap_loc) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTSWAP, FAIL, "can't swap records in B-tree") - - /* Set pointers for advancing to child node */ - new_cache_info_flags_ptr = &internal_flags; - new_cache_info = &internal->cache_info; - new_node_ptr = &internal->node_ptrs[idx]; - - /* Indicate position of next node */ - if(H5B2_POS_MIDDLE != curr_pos) { - if(idx == 0) { - if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) - next_pos = H5B2_POS_LEFT; - } /* end if */ - else if(idx == internal->nrec) { - if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) - next_pos = H5B2_POS_RIGHT; - } /* end if */ - } /* end if */ - } /* end else */ - - /* Attempt to remove record from child node */ - if(depth > 1) { - if(H5B2__remove_internal_by_idx(hdr, dxpl_id, depth_decreased, swap_loc, (uint16_t)(depth - 1), - new_cache_info, new_cache_info_flags_ptr, new_node_ptr, next_pos, n, op, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree internal node") - } /* end if */ - else { - if(H5B2__remove_leaf_by_idx(hdr, dxpl_id, new_node_ptr, next_pos, (unsigned)n, op, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree leaf node") - } /* end else */ - - /* Update record count for node pointer to child node */ - if(!collapsed_root) - new_node_ptr->all_nrec--; - - /* Mark node as dirty */ - internal_flags |= H5AC__DIRTIED_FLAG; - -#ifdef H5B2_DEBUG - H5B2__assert_internal((!collapsed_root ? (curr_node_ptr->all_nrec - 1) : new_node_ptr->all_nrec), hdr, internal); -#endif /* H5B2_DEBUG */ - -done: - /* Release the B-tree internal node */ - if(internal && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, internal_addr, internal, internal_flags) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__remove_internal_by_idx() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__neighbor_leaf - * - * Purpose: Locate a record relative to the specified information in a - * B-tree leaf node and return that information by filling in - * fields of the - * caller-supplied UDATA pointer depending on the type of leaf node - * requested. The UDATA can point to additional data passed - * to the key comparison function. - * - * The 'OP' routine is called with the record found and the - * OP_DATA pointer, to allow caller to return information about - * the record. - * - * The RANGE indicates whether to search for records less than or - * equal to, or greater than or equal to the information passed - * in with UDATA. - * - * Return: Non-negative on success, negative on failure. - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Mar 9 2005 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__neighbor_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, - void *neighbor_loc, H5B2_compare_t comp, void *udata, H5B2_found_t op, - void *op_data) -{ - H5B2_leaf_t *leaf; /* Pointer to leaf node */ - unsigned idx; /* Location of record which matches key */ - int cmp = 0; /* Comparison value of records */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - HDassert(op); - - /* Lock current B-tree node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node_ptr->addr, curr_node_ptr->node_nrec, H5AC__READ_ONLY_FLAG))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") - - /* Locate node pointer for child */ - if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - if(cmp > 0) - idx++; - else - if(cmp == 0 && comp == H5B2_COMPARE_GREATER) - idx++; - - /* Set the neighbor location, if appropriate */ - if(comp == H5B2_COMPARE_LESS) { - if(idx > 0) - neighbor_loc = H5B2_LEAF_NREC(leaf, hdr, idx - 1); - } /* end if */ - else { - HDassert(comp == H5B2_COMPARE_GREATER); - - if(idx < leaf->nrec) - neighbor_loc = H5B2_LEAF_NREC(leaf, hdr, idx); - } /* end else */ - - /* Make callback if neighbor record has been found */ - if(neighbor_loc) { - /* Make callback for current record */ - if((op)(neighbor_loc, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "'found' callback failed for B-tree neighbor operation") - } /* end if */ - else - HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "unable to find neighbor record in B-tree") - -done: - /* Release the B-tree internal node */ - if(leaf && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, curr_node_ptr->addr, leaf, H5AC__NO_FLAGS_SET) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree leaf node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__neighbor_leaf() */ - - -/*------------------------------------------------------------------------- - * Function: H5B2__neighbor_internal - * - * Purpose: Locate a record relative to the specified information in a - * B-tree internal node and return that information by filling in - * fields of the - * caller-supplied UDATA pointer depending on the type of leaf node - * requested. The UDATA can point to additional data passed - * to the key comparison function. - * - * The 'OP' routine is called with the record found and the - * OP_DATA pointer, to allow caller to return information about - * the record. - * - * The RANGE indicates whether to search for records less than or - * equal to, or greater than or equal to the information passed - * in with UDATA. - * - * Return: Non-negative on success, negative on failure. - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Mar 9 2005 - * - *------------------------------------------------------------------------- - */ -herr_t -H5B2__neighbor_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - H5B2_node_ptr_t *curr_node_ptr, void *neighbor_loc, H5B2_compare_t comp, - void *udata, H5B2_found_t op, void *op_data) -{ - H5B2_internal_t *internal; /* Pointer to internal node */ - unsigned idx; /* Location of record which matches key */ - int cmp = 0; /* Comparison value of records */ - herr_t ret_value = SUCCEED; /* Return value */ - - FUNC_ENTER_PACKAGE - - /* Check arguments. */ - HDassert(hdr); - HDassert(depth > 0); - HDassert(curr_node_ptr); - HDassert(H5F_addr_defined(curr_node_ptr->addr)); - HDassert(op); - - /* Lock current B-tree node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node_ptr->addr, curr_node_ptr->node_nrec, depth, H5AC__READ_ONLY_FLAG))) - HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") - - /* Locate node pointer for child */ - if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, - udata, &idx, &cmp) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") - if(cmp > 0) - idx++; - - /* Set the neighbor location, if appropriate */ - if(comp == H5B2_COMPARE_LESS) { - if(idx > 0) - neighbor_loc = H5B2_INT_NREC(internal, hdr, idx - 1); - } /* end if */ - else { - HDassert(comp == H5B2_COMPARE_GREATER); - - if(idx < internal->nrec) - neighbor_loc = H5B2_INT_NREC(internal, hdr, idx); - } /* end else */ - - /* Attempt to find neighboring record */ - if(depth > 1) { - if(H5B2__neighbor_internal(hdr, dxpl_id, (uint16_t)(depth - 1), &internal->node_ptrs[idx], neighbor_loc, comp, udata, op, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "unable to find neighbor record in B-tree internal node") - } /* end if */ - else { - if(H5B2__neighbor_leaf(hdr, dxpl_id, &internal->node_ptrs[idx], neighbor_loc, comp, udata, op, op_data) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "unable to find neighbor record in B-tree leaf node") - } /* end else */ - -done: - /* Release the B-tree internal node */ - if(internal && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, H5AC__NO_FLAGS_SET) < 0) - HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") - - FUNC_LEAVE_NOAPI(ret_value) -} /* H5B2__neighbor_internal() */ - - -/*------------------------------------------------------------------------- * Function: H5B2__delete_node * * Purpose: Iterate over all the nodes in a B-tree node deleting them @@ -3392,7 +1624,8 @@ done: */ herr_t H5B2__delete_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - const H5B2_node_ptr_t *curr_node, H5B2_remove_t op, void *op_data) + const H5B2_node_ptr_t *curr_node, void *parent, H5B2_remove_t op, + void *op_data) { const H5AC_class_t *curr_node_class = NULL; /* Pointer to current node's class info */ void *node = NULL; /* Pointers to current node */ @@ -3410,7 +1643,7 @@ H5B2__delete_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, unsigned u; /* Local index */ /* Lock the current B-tree node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node->addr, curr_node->node_nrec, depth, H5AC__NO_FLAGS_SET))) + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, (H5B2_node_ptr_t *)curr_node, depth, FALSE, H5AC__NO_FLAGS_SET))) /* Casting away const OK -QAK */ HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") /* Set up information about current node */ @@ -3420,14 +1653,14 @@ H5B2__delete_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Descend into children */ for(u = 0; u < internal->nrec + (unsigned)1; u++) - if(H5B2__delete_node(hdr, dxpl_id, (uint16_t)(depth - 1), &(internal->node_ptrs[u]), op, op_data) < 0) + if(H5B2__delete_node(hdr, dxpl_id, (uint16_t)(depth - 1), &(internal->node_ptrs[u]), internal, op, op_data) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTLIST, FAIL, "node descent failed") } /* end if */ else { H5B2_leaf_t *leaf; /* Pointer to leaf node */ /* Lock the current B-tree node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node->addr, curr_node->node_nrec, H5AC__NO_FLAGS_SET))) + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, (H5B2_node_ptr_t *)curr_node, FALSE, H5AC__NO_FLAGS_SET))) /* Casting away const OK -QAK */ HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") /* Set up information about current node */ @@ -3450,7 +1683,7 @@ H5B2__delete_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, done: /* Unlock & delete current node */ - if(node && H5AC_unprotect(hdr->f, dxpl_id, curr_node_class, curr_node->addr, node, H5AC__DELETED_FLAG | H5AC__FREE_FILE_SPACE_FLAG) < 0) + if(node && H5AC_unprotect(hdr->f, dxpl_id, curr_node_class, curr_node->addr, node, (unsigned)(H5AC__DELETED_FLAG | (hdr->swmr_write ? 0 : H5AC__FREE_FILE_SPACE_FLAG))) < 0) HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree node") FUNC_LEAVE_NOAPI(ret_value) @@ -3472,7 +1705,7 @@ done: */ herr_t H5B2__node_size(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - const H5B2_node_ptr_t *curr_node, hsize_t *btree_size) + const H5B2_node_ptr_t *curr_node, void *parent, hsize_t *btree_size) { H5B2_internal_t *internal = NULL; /* Pointer to internal node */ herr_t ret_value = SUCCEED; /* Iterator return value */ @@ -3486,7 +1719,7 @@ H5B2__node_size(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, HDassert(depth > 0); /* Lock the current B-tree node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node->addr, curr_node->node_nrec, depth, H5AC__READ_ONLY_FLAG))) + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, (H5B2_node_ptr_t *)curr_node, depth, FALSE, H5AC__READ_ONLY_FLAG))) /* Casting away const OK -QAK */ HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") /* Recursively descend into child nodes, if we are above the "twig" level in the B-tree */ @@ -3495,7 +1728,7 @@ H5B2__node_size(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, /* Descend into children */ for(u = 0; u < internal->nrec + (unsigned)1; u++) - if(H5B2__node_size(hdr, dxpl_id, (uint16_t)(depth - 1), &(internal->node_ptrs[u]), btree_size) < 0) + if(H5B2__node_size(hdr, dxpl_id, (uint16_t)(depth - 1), &(internal->node_ptrs[u]), internal, btree_size) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTLIST, FAIL, "node iteration failed") } /* end if */ else /* depth is 1: count all the leaf nodes from this node */ @@ -3513,219 +1746,205 @@ done: /*------------------------------------------------------------------------- - * Function: H5B2__internal_free + * Function: H5B2__create_flush_depend * - * Purpose: Destroys a B-tree internal node in memory. + * Purpose: Create a flush dependency between two data structure components * - * Return: Non-negative on success/Negative on failure + * Return: SUCCEED/FAIL * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Feb 2 2005 + * Programmer: Dana Robinson + * Fall 2012 * *------------------------------------------------------------------------- */ herr_t -H5B2__internal_free(H5B2_internal_t *internal) +H5B2__create_flush_depend(H5AC_info_t *parent_entry, H5AC_info_t *child_entry) { herr_t ret_value = SUCCEED; /* Return value */ - FUNC_ENTER_PACKAGE - - /* - * Check arguments. - */ - HDassert(internal); - - /* Release internal node's native key buffer */ - if(internal->int_native) - internal->int_native = (uint8_t *)H5FL_FAC_FREE(internal->hdr->node_info[internal->depth].nat_rec_fac, internal->int_native); - - /* Release internal node's node pointer buffer */ - if(internal->node_ptrs) - internal->node_ptrs = (H5B2_node_ptr_t *)H5FL_FAC_FREE(internal->hdr->node_info[internal->depth].node_ptr_fac, internal->node_ptrs); - - /* Decrement ref. count on B-tree header */ - if(H5B2__hdr_decr(internal->hdr) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTDEC, FAIL, "can't decrement ref. count on B-tree header") + FUNC_ENTER_NOAPI_NOINIT + + /* Sanity check */ + HDassert(parent_entry); + HDassert(child_entry); - /* Free B-tree internal node info */ - internal = H5FL_FREE(H5B2_internal_t, internal); + /* Create a flush dependency between parent and child entry */ + if(H5AC_create_flush_dependency(parent_entry, child_entry) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDEPEND, FAIL, "unable to create flush dependency") done: FUNC_LEAVE_NOAPI(ret_value) -} /* end H5B2__internal_free() */ +} /* end H5B2__create_flush_depend() */ /*------------------------------------------------------------------------- - * Function: H5B2__leaf_free + * Function: H5B2__update_flush_depend * - * Purpose: Destroys a B-tree leaf node in memory. + * Purpose: Update flush dependencies for children of a node * - * Return: Non-negative on success/Negative on failure + * Return: SUCCEED/FAIL * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Feb 2 2005 + * Programmer: Quincey Koziol + * koziol@lbl.gov + * Dec 1 2016 * *------------------------------------------------------------------------- */ herr_t -H5B2__leaf_free(H5B2_leaf_t *leaf) +H5B2__update_flush_depend(H5B2_hdr_t *hdr, hid_t dxpl_id, unsigned depth, + const H5B2_node_ptr_t *node_ptr, void *old_parent, void *new_parent) { + const H5AC_class_t *child_class; /* Pointer to child node's class info */ + void *child = NULL; /* Pointer to child node */ + unsigned node_status = 0; /* Node's status in the metadata cache */ herr_t ret_value = SUCCEED; /* Return value */ FUNC_ENTER_PACKAGE + + /* Sanity checks */ + HDassert(hdr); + HDassert(depth > 0); + HDassert(node_ptr); + HDassert(old_parent); + HDassert(new_parent); - /* - * Check arguments. - */ - HDassert(leaf); + /* Check the node's entry status in the metadata cache */ + if(H5AC_get_entry_status(hdr->f, node_ptr->addr, &node_status) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTGET, FAIL, "unable to check status of B-tree node") - /* Release leaf's native key buffer */ - if(leaf->leaf_native) - leaf->leaf_native = (uint8_t *)H5FL_FAC_FREE(leaf->hdr->node_info[0].nat_rec_fac, leaf->leaf_native); + /* If the node is in the cache, check for retargeting its parent */ + if(node_status & H5AC_ES__IN_CACHE) { + void **parent_ptr; /* Pointer to child node's parent */ + hbool_t update_deps = FALSE; /* Whether to update flush dependencies */ - /* Decrement ref. count on B-tree header */ - if(H5B2__hdr_decr(leaf->hdr) < 0) - HGOTO_ERROR(H5E_BTREE, H5E_CANTDEC, FAIL, "can't decrement ref. count on B-tree header") + /* Get child node pointer */ + if(depth > 1) { + H5B2_internal_t *child_int; - /* Free B-tree leaf node info */ - leaf = H5FL_FREE(H5B2_leaf_t, leaf); + /* Protect child */ + if(NULL == (child_int = H5B2__protect_internal(hdr, dxpl_id, new_parent, (H5B2_node_ptr_t *)node_ptr, (uint16_t)(depth - 1), FALSE, H5AC__NO_FLAGS_SET))) /* Casting away const OK -QAK */ + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + child_class = H5AC_BT2_INT; + child = child_int; -done: - FUNC_LEAVE_NOAPI(ret_value) -} /* end H5B2__leaf_free() */ + if(child_int->parent == old_parent) { + parent_ptr = &child_int->parent; + update_deps = TRUE; + } /* end if */ + else + HDassert(child_int->parent == new_parent); + } /* end if */ + else { + H5B2_leaf_t *child_leaf; -#ifdef H5B2_DEBUG - -/*------------------------------------------------------------------------- - * Function: H5B2__assert_leaf - * - * Purpose: Verify than a leaf node is mostly sane - * - * Return: Non-negative on success, negative on failure - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Feb 19 2005 - * - *------------------------------------------------------------------------- - */ -H5_ATTR_PURE static herr_t -H5B2__assert_leaf(const H5B2_hdr_t *hdr, const H5B2_leaf_t *leaf) -{ - /* General sanity checking on node */ - HDassert(leaf->nrec <= hdr->node_info->split_nrec); + /* Protect child */ + if(NULL == (child_leaf = H5B2__protect_leaf(hdr, dxpl_id, new_parent, (H5B2_node_ptr_t *)node_ptr, FALSE, H5AC__NO_FLAGS_SET))) /* Casting away const OK -QAK */ + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + child_class = H5AC_BT2_LEAF; + child = child_leaf; - return(0); -} /* end H5B2__assert_leaf() */ + if(child_leaf->parent == old_parent) { + parent_ptr = &child_leaf->parent; + update_deps = TRUE; + } /* end if */ + else + HDassert(child_leaf->parent == new_parent); + } /* end else */ - -/*------------------------------------------------------------------------- - * Function: H5B2__assert_leaf2 - * - * Purpose: Verify than a leaf node is mostly sane - * - * Return: Non-negative on success, negative on failure - * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Feb 19 2005 - * - *------------------------------------------------------------------------- - */ -H5_ATTR_PURE static herr_t -H5B2__assert_leaf2(const H5B2_hdr_t *hdr, const H5B2_leaf_t *leaf, const H5B2_leaf_t H5_ATTR_UNUSED *leaf2) -{ - /* General sanity checking on node */ - HDassert(leaf->nrec <= hdr->node_info->split_nrec); + /* Update flush dependencies if necessary */ + if(update_deps) { + /* Sanity check */ + HDassert(parent_ptr); + + /* Switch the flush dependency for the node */ + if(H5B2__destroy_flush_depend((H5AC_info_t *)old_parent, (H5AC_info_t *)child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency") + *parent_ptr = new_parent; + if(H5B2__create_flush_depend((H5AC_info_t *)new_parent, (H5AC_info_t *)child) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDEPEND, FAIL, "unable to create flush dependency") + } /* end if */ + } /* end if */ - return(0); -} /* end H5B2__assert_leaf2() */ +done: + /* Unprotect the child */ + if(child) + if(H5AC_unprotect(hdr->f, dxpl_id, child_class, node_ptr->addr, child, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree node") + + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__update_flush_depend() */ /*------------------------------------------------------------------------- - * Function: H5B2__assert_internal + * Function: H5B2__update_child_flush_depends * - * Purpose: Verify than an internal node is mostly sane + * Purpose: Update flush dependencies for children of a node * - * Return: Non-negative on success, negative on failure + * Return: SUCCEED/FAIL * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Feb 19 2005 + * Programmer: Quincey Koziol + * koziol@lbl.gov + * Dec 1 2016 * *------------------------------------------------------------------------- */ -H5_ATTR_PURE static herr_t -H5B2__assert_internal(hsize_t parent_all_nrec, const H5B2_hdr_t *hdr, const H5B2_internal_t *internal) +static herr_t +H5B2__update_child_flush_depends(H5B2_hdr_t *hdr, hid_t dxpl_id, unsigned depth, + const H5B2_node_ptr_t *node_ptrs, unsigned start_idx, unsigned end_idx, + void *old_parent, void *new_parent) { - hsize_t tot_all_nrec; /* Total number of records at or below this node */ - uint16_t u, v; /* Local index variables */ - - /* General sanity checking on node */ - HDassert(internal->nrec <= hdr->node_info->split_nrec); - - /* Sanity checking on node pointers */ - tot_all_nrec = internal->nrec; - for(u = 0; u < internal->nrec + 1; u++) { - tot_all_nrec += internal->node_ptrs[u].all_nrec; - - HDassert(H5F_addr_defined(internal->node_ptrs[u].addr)); - HDassert(internal->node_ptrs[u].addr > 0); - for(v = 0; v < u; v++) - HDassert(internal->node_ptrs[u].addr != internal->node_ptrs[v].addr); - } /* end for */ + unsigned u; /* Local index variable */ + herr_t ret_value = SUCCEED; /* Return value */ - /* Sanity check all_nrec total in parent */ - if(parent_all_nrec > 0) - HDassert(tot_all_nrec == parent_all_nrec); + FUNC_ENTER_STATIC + + /* Sanity checks */ + HDassert(hdr); + HDassert(depth > 1); + HDassert(node_ptrs); + HDassert(start_idx <= end_idx); + HDassert(old_parent); + HDassert(new_parent); + + /* Loop over children */ + for(u = start_idx; u < end_idx; u++) + /* Update parent for children */ + if(H5B2__update_flush_depend(hdr, dxpl_id, depth - 1, &node_ptrs[u], old_parent, new_parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child node to new parent") - return(0); -} /* end H5B2__assert_internal() */ +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__update_child_flush_depends() */ /*------------------------------------------------------------------------- - * Function: H5B2__assert_internal2 + * Function: H5B2__destroy_flush_depend * - * Purpose: Verify than internal nodes are mostly sane + * Purpose: Destroy a flush dependency between two data structure components * - * Return: Non-negative on success, negative on failure + * Return: SUCCEED/FAIL * - * Programmer: Quincey Koziol - * koziol@ncsa.uiuc.edu - * Feb 19 2005 + * Programmer: Dana Robinson + * Fall 2012 * *------------------------------------------------------------------------- */ -H5_ATTR_PURE static herr_t -H5B2__assert_internal2(hsize_t parent_all_nrec, const H5B2_hdr_t *hdr, const H5B2_internal_t *internal, const H5B2_internal_t *internal2) +herr_t +H5B2__destroy_flush_depend(H5AC_info_t *parent_entry, H5AC_info_t *child_entry) { - hsize_t tot_all_nrec; /* Total number of records at or below this node */ - uint16_t u, v; /* Local index variables */ - - /* General sanity checking on node */ - HDassert(internal->nrec <= hdr->node_info->split_nrec); - - /* Sanity checking on node pointers */ - tot_all_nrec =internal->nrec; - for(u =0; u < internal->nrec + 1; u++) { - tot_all_nrec += internal->node_ptrs[u].all_nrec; - - HDassert(H5F_addr_defined(internal->node_ptrs[u].addr)); - HDassert(internal->node_ptrs[u].addr > 0); - for(v = 0; v < u; v++) - HDassert(internal->node_ptrs[u].addr != internal->node_ptrs[v].addr); - for(v = 0; v < internal2->nrec + 1; v++) - HDassert(internal->node_ptrs[u].addr != internal2->node_ptrs[v].addr); - } /* end for */ + herr_t ret_value = SUCCEED; /* Return value */ - /* Sanity check all_nrec total in parent */ - if(parent_all_nrec > 0) - HDassert(tot_all_nrec == parent_all_nrec); + FUNC_ENTER_NOAPI_NOINIT + + /* Sanity check */ + HDassert(parent_entry); + HDassert(child_entry); - return(0); -} /* end H5B2__assert_internal2() */ -#endif /* H5B2_DEBUG */ + /* Destroy a flush dependency between parent and child entry */ + if(H5AC_destroy_flush_dependency(parent_entry, child_entry) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__destroy_flush_depend() */ diff --git a/src/H5B2internal.c b/src/H5B2internal.c new file mode 100644 index 0000000..e74ae59 --- /dev/null +++ b/src/H5B2internal.c @@ -0,0 +1,1443 @@ +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * + * Copyright by The HDF Group. * + * Copyright by the Board of Trustees of the University of Illinois. * + * All rights reserved. * + * * + * This file is part of HDF5. The full HDF5 copyright notice, including * + * terms governing use, modification, and redistribution, is contained in * + * the files COPYING and Copyright.html. COPYING can be found at the root * + * of the source code distribution tree; Copyright.html can be found at the * + * root level of an installed copy of the electronic HDF5 document set and * + * is linked from the top-level documents page. It can also be found at * + * http://hdfgroup.org/HDF5/doc/Copyright.html. If you do not have * + * access to either file, you may request a copy from help@hdfgroup.org. * + * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ + +/*------------------------------------------------------------------------- + * + * Created: H5B2internal.c + * Dec 01 2016 + * Quincey Koziol <koziol@lbl.gov> + * + * Purpose: Routines for managing v2 B-tree internal ndoes. + * + *------------------------------------------------------------------------- + */ + +/****************/ +/* Module Setup */ +/****************/ + +#include "H5B2module.h" /* This source code file is part of the H5B2 module */ + + +/***********/ +/* Headers */ +/***********/ +#include "H5private.h" /* Generic Functions */ +#include "H5B2pkg.h" /* v2 B-trees */ +#include "H5Eprivate.h" /* Error handling */ +#include "H5MFprivate.h" /* File memory management */ + + +/****************/ +/* Local Macros */ +/****************/ + + +/******************/ +/* Local Typedefs */ +/******************/ + + +/********************/ +/* Package Typedefs */ +/********************/ + + +/********************/ +/* Local Prototypes */ +/********************/ +static herr_t H5B2__shadow_internal(H5B2_internal_t *internal, hid_t dxpl_id, + H5B2_node_ptr_t *curr_node_ptr); + + +/*********************/ +/* Package Variables */ +/*********************/ + +/* Declare a free list to manage the H5B2_internal_t struct */ +H5FL_DEFINE(H5B2_internal_t); + + +/*****************************/ +/* Library Private Variables */ +/*****************************/ + + +/*******************/ +/* Local Variables */ +/*******************/ + + + +/*------------------------------------------------------------------------- + * Function: H5B2__create_internal + * + * Purpose: Creates empty internal node of a B-tree and update node pointer + * to point to it. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Feb 3 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__create_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, void *parent, + H5B2_node_ptr_t *node_ptr, uint16_t depth) +{ + H5B2_internal_t *internal = NULL; /* Pointer to new internal node created */ + hbool_t inserted = FALSE; /* Whether the internal node was inserted into cache */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(node_ptr); + HDassert(depth > 0); + + /* Allocate memory for internal node information */ + if(NULL == (internal = H5FL_CALLOC(H5B2_internal_t))) + HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree internal info") + + /* Increment ref. count on B-tree header */ + if(H5B2__hdr_incr(hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINC, FAIL, "can't increment ref. count on B-tree header") + + /* Share B-tree header information */ + internal->hdr = hdr; + + /* Allocate space for the native keys in memory */ + if(NULL == (internal->int_native = (uint8_t *)H5FL_FAC_MALLOC(hdr->node_info[depth].nat_rec_fac))) + HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree internal native keys") + HDmemset(internal->int_native, 0, hdr->cls->nrec_size * hdr->node_info[depth].max_nrec); + + /* Allocate space for the node pointers in memory */ + if(NULL == (internal->node_ptrs = (H5B2_node_ptr_t *)H5FL_FAC_MALLOC(hdr->node_info[depth].node_ptr_fac))) + HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree internal node pointers") + HDmemset(internal->node_ptrs, 0, sizeof(H5B2_node_ptr_t) * (hdr->node_info[depth].max_nrec + 1)); + + /* Set depth of the node */ + internal->depth = depth; + + /* Set parent */ + internal->parent = parent; + + /* Set shadowed epoch to header's epoch */ + internal->shadow_epoch = hdr->shadow_epoch; + + /* Allocate space on disk for the internal node */ + if(HADDR_UNDEF == (node_ptr->addr = H5MF_alloc(hdr->f, H5FD_MEM_BTREE, dxpl_id, (hsize_t)hdr->node_size))) + HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "file allocation failed for B-tree internal node") + + /* Cache the new B-tree node */ + if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_BT2_INT, node_ptr->addr, internal, H5AC__NO_FLAGS_SET) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "can't add B-tree internal node to cache") + inserted = TRUE; + + /* Add internal node as child of 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, internal) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSET, FAIL, "unable to add v2 B-tree node as child of proxy") + internal->top_proxy = hdr->top_proxy; + } /* end if */ + +done: + if(ret_value < 0) { + if(internal) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(internal) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTREMOVE, FAIL, "unable to remove v2 B-tree internal node from cache") + + /* Release internal node's disk space */ + if(H5F_addr_defined(node_ptr->addr) && H5MF_xfree(hdr->f, H5FD_MEM_BTREE, dxpl_id, node_ptr->addr, (hsize_t)hdr->node_size) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTFREE, FAIL, "unable to release file space for v2 B-tree internal node") + + /* Destroy internal node */ + if(H5B2__internal_free(internal) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTFREE, FAIL, "unable to release v2 B-tree internal node") + } /* end if */ + } /* end if */ + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__create_internal() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__protect_internal + * + * Purpose: "Protect" an internal node in the metadata cache + * + * Return: Pointer to internal node on success/NULL on failure + * + * Programmer: Quincey Koziol + * koziol@hdfgroup.org + * Aug 25 2006 + * + *------------------------------------------------------------------------- + */ +H5B2_internal_t * +H5B2__protect_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, void *parent, + H5B2_node_ptr_t *node_ptr, uint16_t depth, hbool_t shadow, unsigned flags) +{ + H5B2_internal_cache_ud_t udata; /* User data to pass through to cache 'deserialize' callback */ + H5B2_internal_t *internal = NULL; /* v2 B-tree internal node */ + H5B2_internal_t *ret_value = NULL; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(node_ptr); + HDassert(H5F_addr_defined(node_ptr->addr)); + HDassert(depth > 0); + + /* only H5AC__READ_ONLY_FLAG may appear in flags */ + HDassert((flags & (unsigned)(~H5AC__READ_ONLY_FLAG)) == 0); + + /* Set up user data for callback */ + udata.f = hdr->f; + udata.hdr = hdr; + udata.parent = parent; + udata.nrec = node_ptr->node_nrec; + udata.depth = depth; + + /* Protect the internal node */ + if(NULL == (internal = (H5B2_internal_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_BT2_INT, node_ptr->addr, &udata, flags))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, NULL, "unable to protect B-tree internal node") + + /* Create top proxy, if it doesn't exist */ + if(hdr->top_proxy && NULL == internal->top_proxy) { + /* Add internal node as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, internal) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSET, NULL, "unable to add v2 B-tree internal node as child of proxy") + internal->top_proxy = hdr->top_proxy; + } /* end if */ + + /* Shadow the node, if requested */ + if(shadow) + if(H5B2__shadow_internal(internal, dxpl_id, node_ptr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOPY, NULL, "unable to shadow internal node") + + /* Set return value */ + ret_value = internal; + +done: + /* Clean up on error */ + if(!ret_value) { + /* Release the internal node, if it was protected */ + if(internal) { + /* Remove from v2 B-tree's proxy, if added */ + if(internal->top_proxy) { + if(H5AC_proxy_entry_remove_child(internal->top_proxy, internal) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, NULL, "unable to destroy flush dependency between internal node and v2 B-tree 'top' proxy") + internal->top_proxy = NULL; + } /* end if */ + + /* Unprotect internal node */ + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, node_ptr->addr, internal, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, NULL, "unable to unprotect v2 B-tree internal node, address = %llu", (unsigned long long)node_ptr->addr) + } /* end if */ + } /* end if */ + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__protect_internal() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__neighbor_internal + * + * Purpose: Locate a record relative to the specified information in a + * B-tree internal node and return that information by filling in + * fields of the + * caller-supplied UDATA pointer depending on the type of leaf node + * requested. The UDATA can point to additional data passed + * to the key comparison function. + * + * The 'OP' routine is called with the record found and the + * OP_DATA pointer, to allow caller to return information about + * the record. + * + * The RANGE indicates whether to search for records less than or + * equal to, or greater than or equal to the information passed + * in with UDATA. + * + * Return: Non-negative on success, negative on failure. + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Mar 9 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__neighbor_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + H5B2_node_ptr_t *curr_node_ptr, void *neighbor_loc, H5B2_compare_t comp, + void *parent, void *udata, H5B2_found_t op, void *op_data) +{ + H5B2_internal_t *internal; /* Pointer to internal node */ + unsigned idx = 0; /* Location of record which matches key */ + int cmp = 0; /* Comparison value of records */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(depth > 0); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + HDassert(op); + + /* Lock current B-tree node */ + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, curr_node_ptr, depth, FALSE, H5AC__READ_ONLY_FLAG))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + + /* Locate node pointer for child */ + if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + if(cmp > 0) + idx++; + + /* Set the neighbor location, if appropriate */ + if(comp == H5B2_COMPARE_LESS) { + if(idx > 0) + neighbor_loc = H5B2_INT_NREC(internal, hdr, idx - 1); + } /* end if */ + else { + HDassert(comp == H5B2_COMPARE_GREATER); + + if(idx < internal->nrec) + neighbor_loc = H5B2_INT_NREC(internal, hdr, idx); + } /* end else */ + + /* Attempt to find neighboring record */ + if(depth > 1) { + if(H5B2__neighbor_internal(hdr, dxpl_id, (uint16_t)(depth - 1), &internal->node_ptrs[idx], neighbor_loc, comp, internal, udata, op, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "unable to find neighbor record in B-tree internal node") + } /* end if */ + else { + if(H5B2__neighbor_leaf(hdr, dxpl_id, &internal->node_ptrs[idx], neighbor_loc, comp, internal, udata, op, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "unable to find neighbor record in B-tree leaf node") + } /* end else */ + +done: + /* Release the B-tree internal node */ + if(internal && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__neighbor_internal() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__insert_internal + * + * Purpose: Adds a new record to a B-tree node. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Mar 2 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__insert_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + unsigned *parent_cache_info_flags_ptr, H5B2_node_ptr_t *curr_node_ptr, + H5B2_nodepos_t curr_pos, void *parent, void *udata) +{ + H5B2_internal_t *internal = NULL; /* Pointer to internal node */ + unsigned internal_flags = H5AC__NO_FLAGS_SET; + unsigned idx = 0; /* Location of record which matches key */ + H5B2_nodepos_t next_pos = H5B2_POS_MIDDLE; /* Position of node */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(depth > 0); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + + /* Lock current B-tree node */ + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, curr_node_ptr, depth, FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + + /* Sanity check number of records */ + HDassert(internal->nrec == curr_node_ptr->node_nrec); + + /* Split or redistribute child node pointers, if necessary */ + { + int cmp; /* Comparison value of records */ + unsigned retries; /* Number of times to attempt redistribution */ + size_t split_nrec; /* Number of records to split node at */ + + /* Locate node pointer for child */ + if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, + udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + if(cmp == 0) + HGOTO_ERROR(H5E_BTREE, H5E_EXISTS, FAIL, "record is already in B-tree") + if(cmp > 0) + idx++; + + /* Set the number of redistribution retries */ + /* This takes care of the case where a B-tree node needs to be + * redistributed, but redistributing the node causes the index + * for insertion to move to another node, which also needs to be + * redistributed. Now, we loop trying to redistribute and then + * eventually force a split */ + retries = 2; + + /* Determine the correct number of records to split child node at */ + split_nrec = hdr->node_info[depth - 1].split_nrec; + + /* Preemptively split/redistribute a node we will enter */ + while(internal->node_ptrs[idx].node_nrec == split_nrec) { + /* Attempt to redistribute records among children */ + if(idx == 0) { /* Left-most child */ + if(retries > 0 && (internal->node_ptrs[idx + 1].node_nrec < split_nrec)) { + if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__split1(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to split child node") + } /* end else */ + } /* end if */ + else if(idx == internal->nrec) { /* Right-most child */ + if(retries > 0 && (internal->node_ptrs[idx - 1].node_nrec < split_nrec)) { + if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, (idx - 1)) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__split1(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to split child node") + } /* end else */ + } /* end if */ + else { /* Middle child */ + if(retries > 0 && ((internal->node_ptrs[idx + 1].node_nrec < split_nrec) || + (internal->node_ptrs[idx - 1].node_nrec < split_nrec))) { + if(H5B2__redistribute3(hdr, dxpl_id, depth, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__split1(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to split child node") + } /* end else */ + } /* end else */ + + /* Locate node pointer for child (after split/redistribute) */ + /* Actually, this can be easily updated (for 2-node redistrib.) and shouldn't require re-searching */ + if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, + udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + if(cmp == 0) + HGOTO_ERROR(H5E_BTREE, H5E_EXISTS, FAIL, "record is already in B-tree") + if(cmp > 0) + idx++; + + /* Decrement the number of redistribution retries left */ + retries--; + } /* end while */ + } /* end block */ + + /* Check if this node is left/right-most */ + if(H5B2_POS_MIDDLE != curr_pos) { + if(idx == 0) { + if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) + next_pos = H5B2_POS_LEFT; + } /* end if */ + else if(idx == internal->nrec) { + if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) + next_pos = H5B2_POS_RIGHT; + } /* end else */ + } /* end if */ + + /* Attempt to insert node */ + if(depth > 1) { + if(H5B2__insert_internal(hdr, dxpl_id, (uint16_t)(depth - 1), &internal_flags, &internal->node_ptrs[idx], next_pos, internal, udata) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into B-tree internal node") + } /* end if */ + else { + if(H5B2__insert_leaf(hdr, dxpl_id, &internal->node_ptrs[idx], next_pos, internal, udata) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into B-tree leaf node") + } /* end else */ + + /* Update record count for node pointer to current node */ + curr_node_ptr->all_nrec++; + + /* Mark node as dirty */ + internal_flags |= H5AC__DIRTIED_FLAG; + +done: + /* Release the B-tree internal node */ + if(internal) { + /* Shadow the node if doing SWMR writes */ + if(hdr->swmr_write && (internal_flags & H5AC__DIRTIED_FLAG)) + if(H5B2__shadow_internal(internal, dxpl_id, curr_node_ptr) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTCOPY, FAIL, "unable to shadow internal B-tree node") + + /* Unprotect node */ + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, internal_flags) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") + } /* end if */ + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__insert_internal() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__update_internal + * + * Purpose: Insert or modify a record in a B-tree internal node. + * If the record exists already, it is modified as if H5B2_modify + * was called). If it doesn't exist, it is inserted as if + * H5B2_insert was called. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@hdfgroup.org + * Dec 24 2015 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__update_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + unsigned *parent_cache_info_flags_ptr, H5B2_node_ptr_t *curr_node_ptr, + H5B2_update_status_t *status, H5B2_nodepos_t curr_pos, void *parent, + void *udata, H5B2_modify_t op, void *op_data) +{ + H5B2_internal_t *internal = NULL; /* Pointer to internal node */ + unsigned internal_flags = H5AC__NO_FLAGS_SET; + int cmp; /* Comparison value of records */ + unsigned idx = 0; /* Location of record which matches key */ + H5B2_nodepos_t next_pos = H5B2_POS_MIDDLE; /* Position of node */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(depth > 0); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + + /* Lock current B-tree node */ + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, curr_node_ptr, depth, FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + + /* Sanity check number of records */ + HDassert(internal->nrec == curr_node_ptr->node_nrec); + + /* Locate node pointer for child */ + if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + + /* Check for modifying existing record */ + if(0 == cmp) { + hbool_t changed = FALSE; /* Whether the 'modify' callback changed the record */ + + /* Make callback for current record */ + if((op)(H5B2_INT_NREC(internal, hdr, idx), op_data, &changed) < 0) { + /* Make certain that the callback didn't modify the value if it failed */ + HDassert(changed == FALSE); + + HGOTO_ERROR(H5E_BTREE, H5E_CANTMODIFY, FAIL, "'modify' callback failed for B-tree update operation") + } /* end if */ + + /* Mark the node as dirty if it changed */ + internal_flags |= (changed ? H5AC__DIRTIED_FLAG : 0); + + /* Indicate that the record was modified */ + *status = H5B2_UPDATE_MODIFY_DONE; + } /* end if */ + else { + /* Adjust index to leave room for node to insert */ + if(cmp > 0) + idx++; + + /* Check if this node is left/right-most */ + if(H5B2_POS_MIDDLE != curr_pos) { + if(idx == 0) { + if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) + next_pos = H5B2_POS_LEFT; + } /* end if */ + else if(idx == internal->nrec) { + if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) + next_pos = H5B2_POS_RIGHT; + } /* end else */ + } /* end if */ + + /* Attempt to update record in child */ + if(depth > 1) { + if(H5B2__update_internal(hdr, dxpl_id, (uint16_t)(depth - 1), &internal_flags, &internal->node_ptrs[idx], status, next_pos, internal, udata, op, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update record in internal B-tree node") + } /* end if */ + else { + if(H5B2__update_leaf(hdr, dxpl_id, &internal->node_ptrs[idx], status, next_pos, internal, udata, op, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update record in leaf B-tree node") + } /* end else */ + + /* Take actions based on child's status report */ + switch(*status) { + case H5B2_UPDATE_MODIFY_DONE: + /* No action */ + break; + + case H5B2_UPDATE_SHADOW_DONE: + /* If child node was shadowed (if SWMR is enabled), mark this node dirty */ + if(hdr->swmr_write) + internal_flags |= H5AC__DIRTIED_FLAG; + + /* No further modifications up the tree are necessary though, so downgrade to merely "modified" */ + *status = H5B2_UPDATE_MODIFY_DONE; + break; + + case H5B2_UPDATE_INSERT_DONE: + /* Mark node as dirty */ + internal_flags |= H5AC__DIRTIED_FLAG; + + /* Update total record count for node pointer to current node */ + curr_node_ptr->all_nrec++; + break; + + case H5B2_UPDATE_INSERT_CHILD_FULL: + /* Split/redistribute this node */ + if(internal->nrec == hdr->node_info[depth].split_nrec) { + hbool_t could_split = FALSE; /* Whether the child node could split */ + + if(idx == 0) { /* Left-most child */ + /* Check for left-most child and its neighbor being close to full */ + if((internal->node_ptrs[idx].node_nrec + internal->node_ptrs[idx + 1].node_nrec) + >= ((hdr->node_info[depth - 1].split_nrec * 2) - 1)) + could_split = TRUE; + } /* end if */ + else if(idx == internal->nrec) { /* Right-most child */ + /* Check for right-most child and its neighbor being close to full */ + if((internal->node_ptrs[idx - 1].node_nrec + internal->node_ptrs[idx].node_nrec) + >= ((hdr->node_info[depth - 1].split_nrec * 2) - 1)) + could_split = TRUE; + } /* end else-if */ + else { /* Middle child */ + /* Check for middle child and its left neighbor being close to full */ + if((internal->node_ptrs[idx - 1].node_nrec + internal->node_ptrs[idx].node_nrec) + >= ((hdr->node_info[depth - 1].split_nrec * 2) - 1)) + could_split = TRUE; + /* Check for middle child and its right neighbor being close to full */ + else if((internal->node_ptrs[idx].node_nrec + internal->node_ptrs[idx + 1].node_nrec) + >= ((hdr->node_info[depth - 1].split_nrec * 2) - 1)) + could_split = TRUE; + } /* end if */ + + /* If this node is full and the child node insertion could + * cause a split, punt back up to caller, leaving the + * "insert child full" status. + */ + if(could_split) { + /* Release the internal B-tree node */ + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, internal_flags) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") + internal = NULL; + + /* Punt back to caller */ + HGOTO_DONE(SUCCEED); + } /* end if */ + } /* end if */ + + /* Release the internal B-tree node */ + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, internal_flags) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") + internal = NULL; + + /* Indicate that the record was inserted */ + *status = H5B2_UPDATE_INSERT_DONE; + + /* Dodge sideways into inserting a record into this node */ + if(H5B2__insert_internal(hdr, dxpl_id, depth, parent_cache_info_flags_ptr, curr_node_ptr, curr_pos, parent, udata) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into internal B-tree node") + break; + + case H5B2_UPDATE_UNKNOWN: + default: + HDassert(0 && "Invalid update status"); + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "invalid update status") + } /* end switch */ + } /* end else */ + +done: + /* Release the internal B-tree node */ + if(internal) { + /* Check if we should shadow this node */ + if(hdr->swmr_write && (internal_flags & H5AC__DIRTIED_FLAG)) { + /* Attempt to shadow the node if doing SWMR writes */ + if(H5B2__shadow_internal(internal, dxpl_id, curr_node_ptr) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTCOPY, FAIL, "unable to shadow internal B-tree node") + + /* Change the state to "shadowed" if only modified currently */ + /* (Triggers parent to be marked dirty) */ + if(*status == H5B2_UPDATE_MODIFY_DONE) + *status = H5B2_UPDATE_SHADOW_DONE; + } /* end if */ + + /* Unprotect node */ + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr->addr, internal, internal_flags) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") + } /* end if */ + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__update_internal() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__shadow_internal + * + * Purpose: "Shadow" an internal node - copy it to a new location, + * leaving the data in the old location intact (for now). + * This is done when writing in SWMR mode to ensure that + * readers do not see nodes that are out of date with + * respect to each other and thereby inconsistent. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Neil Fortner + * Apr 27 2012 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5B2__shadow_internal(H5B2_internal_t *internal, hid_t dxpl_id, + H5B2_node_ptr_t *curr_node_ptr) +{ + H5B2_hdr_t *hdr; /* B-tree header */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_STATIC + + /* + * Check arguments. + */ + HDassert(internal); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + hdr = internal->hdr; + HDassert(hdr); + HDassert(hdr->swmr_write); + + /* We only need to shadow the node if it has not been shadowed since the + * last time the header was flushed, as otherwise it will be unreachable by + * the readers so there will be no need to shadow. To check if it has been + * shadowed, compare the epoch of this node and the header. If this node's + * epoch is <= to the header's, it hasn't been shadowed yet. */ + if(internal->shadow_epoch <= hdr->shadow_epoch) { + haddr_t new_node_addr; /* Address to move node to */ + + /* + * We must clone the old node so readers with an out-of-date version of + * the parent can still see the correct number of children, via the + * shadowed node. Remove it from cache but do not mark it free on disk. + */ + /* Allocate space for the cloned node */ + if(HADDR_UNDEF == (new_node_addr = H5MF_alloc(hdr->f, H5FD_MEM_BTREE, dxpl_id, (hsize_t)hdr->node_size))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "unable to allocate file space to move B-tree node") + + /* Move the location of the node on the disk */ + if(H5AC_move_entry(hdr->f, H5AC_BT2_INT, curr_node_ptr->addr, new_node_addr, dxpl_id) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTMOVE, FAIL, "unable to move B-tree node") + curr_node_ptr->addr = new_node_addr; + + /* Should free the space in the file, but this is not supported by + * SWMR_WRITE code yet - QAK, 2016/12/01 + */ + + /* Set shadow epoch for node ahead of header */ + internal->shadow_epoch = hdr->shadow_epoch + 1; + } /* end if */ + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__shadow_internal() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__remove_internal + * + * Purpose: Removes a record from a B-tree node. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Mar 3 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__remove_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, hbool_t *depth_decreased, + void *swap_loc, void *swap_parent, uint16_t depth, H5AC_info_t *parent_cache_info, + unsigned *parent_cache_info_flags_ptr, H5B2_nodepos_t curr_pos, + H5B2_node_ptr_t *curr_node_ptr, void *udata, H5B2_remove_t op, void *op_data) +{ + H5AC_info_t *new_cache_info; /* Pointer to new cache info */ + unsigned *new_cache_info_flags_ptr = NULL; + H5B2_node_ptr_t *new_node_ptr; /* Pointer to new node pointer */ + H5B2_internal_t *internal; /* Pointer to internal node */ + H5B2_nodepos_t next_pos = H5B2_POS_MIDDLE; /* Position of next node */ + unsigned internal_flags = H5AC__NO_FLAGS_SET; + haddr_t internal_addr = HADDR_UNDEF; /* Address of internal node */ + size_t merge_nrec; /* Number of records to merge node at */ + hbool_t collapsed_root = FALSE; /* Whether the root was collapsed */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(depth > 0); + HDassert(parent_cache_info); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + + /* Lock current B-tree node */ + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent_cache_info, curr_node_ptr, depth, FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + internal_addr = curr_node_ptr->addr; + + /* Determine the correct number of records to merge at */ + merge_nrec = hdr->node_info[depth - 1].merge_nrec; + + /* Check for needing to collapse the root node */ + /* (The root node is the only internal node allowed to have 1 record) */ + if(internal->nrec == 1 && + ((internal->node_ptrs[0].node_nrec + internal->node_ptrs[1].node_nrec) <= ((merge_nrec * 2) + 1))) { + + /* Merge children of root node */ + if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, 0) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") + + /* Let the cache know that the object is deleted */ + internal_flags |= H5AC__DELETED_FLAG; + if(!hdr->swmr_write) + internal_flags |= H5AC__FREE_FILE_SPACE_FLAG; + + /* Reset information in header's root node pointer */ + curr_node_ptr->addr = internal->node_ptrs[0].addr; + curr_node_ptr->node_nrec = internal->node_ptrs[0].node_nrec; + + /* Update flush dependency for child, if using SWMR */ + if(hdr->swmr_write) + if(H5B2__update_flush_depend(hdr, dxpl_id, depth, curr_node_ptr, internal, hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child node to new parent") + + /* Indicate that the level of the B-tree decreased */ + *depth_decreased = TRUE; + + /* Set pointers for advancing to child node */ + new_cache_info = parent_cache_info; + new_cache_info_flags_ptr = parent_cache_info_flags_ptr; + new_node_ptr = curr_node_ptr; + + /* Set flag to indicate root was collapsed */ + collapsed_root = TRUE; + + /* Indicate position of next node */ + next_pos = H5B2_POS_ROOT; + } /* end if */ + /* Merge or redistribute child node pointers, if necessary */ + else { + unsigned idx = 0; /* Location of record which matches key */ + int cmp = 0; /* Comparison value of records */ + unsigned retries; /* Number of times to attempt redistribution */ + + /* Shadow the node if doing SWMR writes */ + if(hdr->swmr_write) { + if(H5B2__shadow_internal(internal, dxpl_id, curr_node_ptr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOPY, FAIL, "unable to shadow internal node") + internal_addr = curr_node_ptr->addr; + } /* end if */ + + /* Locate node pointer for child */ + if(swap_loc) + idx = 0; + else { + if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + if(cmp >= 0) + idx++; + } /* end else */ + + /* Set the number of redistribution retries */ + /* This takes care of the case where a B-tree node needs to be + * redistributed, but redistributing the node causes the index + * for removal to move to another node, which also needs to be + * redistributed. Now, we loop trying to redistribute and then + * eventually force a merge */ + retries = 2; + + /* Preemptively merge/redistribute a node we will enter */ + while(internal->node_ptrs[idx].node_nrec == merge_nrec) { + /* Attempt to redistribute records among children */ + /* (NOTE: These 2-node redistributions should actually get the + * record to promote from the node with more records. - QAK) + */ + /* (NOTE: This code is the same in both H5B2__remove_internal() and + * H5B2__remove_internal_by_idx(), fix bugs in both places! - QAK) + */ + if(idx == 0) { /* Left-most child */ + if(retries > 0 && (internal->node_ptrs[idx + 1].node_nrec > merge_nrec)) { + if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") + } /* end else */ + } /* end if */ + else if(idx == internal->nrec) { /* Right-most child */ + if(retries > 0 && (internal->node_ptrs[idx - 1].node_nrec > merge_nrec)) { + if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, (idx - 1)) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, (idx - 1)) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") + } /* end else */ + } /* end if */ + else { /* Middle child */ + if(retries > 0 && ((internal->node_ptrs[idx + 1].node_nrec > merge_nrec) || + (internal->node_ptrs[idx - 1].node_nrec > merge_nrec))) { + if(H5B2__redistribute3(hdr, dxpl_id, depth, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__merge3(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") + } /* end else */ + } /* end else */ + + /* Locate node pointer for child (after merge/redistribute) */ + if(swap_loc) + idx = 0; + else { +/* Actually, this can be easily updated (for 2-node redistrib.) and shouldn't require re-searching */ + if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + if(cmp >= 0) + idx++; + } /* end else */ + + /* Decrement the number of redistribution retries left */ + retries--; + } /* end while */ + + /* Handle deleting a record from an internal node */ + if(!swap_loc && cmp == 0) { + swap_loc = H5B2_INT_NREC(internal, hdr, idx - 1); + swap_parent = internal; + } /* end if */ + + /* Swap record to delete with record from leaf, if we are the last internal node */ + if(swap_loc && depth == 1) + if(H5B2__swap_leaf(hdr, dxpl_id, depth, internal, &internal_flags, idx, swap_loc) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSWAP, FAIL, "Can't swap records in B-tree") + + /* Set pointers for advancing to child node */ + new_cache_info_flags_ptr = &internal_flags; + new_cache_info = &internal->cache_info; + new_node_ptr = &internal->node_ptrs[idx]; + + /* Indicate position of next node */ + if(H5B2_POS_MIDDLE != curr_pos) { + if(idx == 0) { + if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) + next_pos = H5B2_POS_LEFT; + } /* end if */ + else if(idx == internal->nrec) { + if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) + next_pos = H5B2_POS_RIGHT; + } /* end if */ + } /* end if */ + } /* end else */ + + /* Attempt to remove record from child node */ + if(depth > 1) { + if(H5B2__remove_internal(hdr, dxpl_id, depth_decreased, swap_loc, swap_parent, (uint16_t)(depth - 1), + new_cache_info, new_cache_info_flags_ptr, next_pos, new_node_ptr, udata, op, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree internal node") + } /* end if */ + else { + if(H5B2__remove_leaf(hdr, dxpl_id, new_node_ptr, next_pos, new_cache_info, udata, op, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree leaf node") + } /* end else */ + + /* Update record count for node pointer to current node */ + if(!collapsed_root) + new_node_ptr->all_nrec--; + + /* Mark node as dirty */ + if(!(hdr->swmr_write && collapsed_root)) + internal_flags |= H5AC__DIRTIED_FLAG; + +#ifdef H5B2_DEBUG + H5B2__assert_internal((!collapsed_root ? (curr_node_ptr->all_nrec - 1) : new_node_ptr->all_nrec), hdr, internal); +#endif /* H5B2_DEBUG */ + +done: + /* Release the B-tree internal node */ + if(internal && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, internal_addr, internal, internal_flags) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__remove_internal() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__remove_internal_by_idx + * + * Purpose: Removes a record from a B-tree node, according to the offset + * in the B-tree records + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@hdfgroup.org + * Nov 14 2006 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__remove_internal_by_idx(H5B2_hdr_t *hdr, hid_t dxpl_id, + hbool_t *depth_decreased, void *swap_loc, void *swap_parent, uint16_t depth, + H5AC_info_t *parent_cache_info, unsigned *parent_cache_info_flags_ptr, + H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, hsize_t n, + H5B2_remove_t op, void *op_data) +{ + H5AC_info_t *new_cache_info; /* Pointer to new cache info */ + unsigned *new_cache_info_flags_ptr = NULL; + H5B2_node_ptr_t *new_node_ptr; /* Pointer to new node pointer */ + H5B2_internal_t *internal; /* Pointer to internal node */ + H5B2_nodepos_t next_pos = H5B2_POS_MIDDLE; /* Position of next node */ + unsigned internal_flags = H5AC__NO_FLAGS_SET; + haddr_t internal_addr = HADDR_UNDEF; /* Address of internal node */ + size_t merge_nrec; /* Number of records to merge node at */ + hbool_t collapsed_root = FALSE; /* Whether the root was collapsed */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(depth > 0); + HDassert(parent_cache_info); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + + /* Lock current B-tree node */ + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent_cache_info, curr_node_ptr, depth, FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + internal_addr = curr_node_ptr->addr; + HDassert(internal->nrec == curr_node_ptr->node_nrec); + HDassert(depth == hdr->depth || internal->nrec > 1); + + /* Determine the correct number of records to merge at */ + merge_nrec = hdr->node_info[depth - 1].merge_nrec; + + /* Check for needing to collapse the root node */ + /* (The root node is the only internal node allowed to have 1 record) */ + if(internal->nrec == 1 && + ((internal->node_ptrs[0].node_nrec + internal->node_ptrs[1].node_nrec) <= ((merge_nrec * 2) + 1))) { + HDassert(depth == hdr->depth); + + /* Merge children of root node */ + if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, 0) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") + + /* Let the cache know that the object is deleted */ + internal_flags |= H5AC__DELETED_FLAG; + if(!hdr->swmr_write) + internal_flags |= H5AC__FREE_FILE_SPACE_FLAG; + + /* Reset information in header's root node pointer */ + curr_node_ptr->addr = internal->node_ptrs[0].addr; + curr_node_ptr->node_nrec = internal->node_ptrs[0].node_nrec; + + /* Update flush dependency for child, if using SWMR */ + if(hdr->swmr_write) + if(H5B2__update_flush_depend(hdr, dxpl_id, depth, curr_node_ptr, internal, hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUPDATE, FAIL, "unable to update child node to new parent") + + /* Indicate that the level of the B-tree decreased */ + *depth_decreased = TRUE; + + /* Set pointers for advancing to child node */ + new_cache_info = parent_cache_info; + new_cache_info_flags_ptr = parent_cache_info_flags_ptr; + new_node_ptr = curr_node_ptr; + + /* Set flag to indicate root was collapsed */ + collapsed_root = TRUE; + + /* Indicate position of next node */ + next_pos = H5B2_POS_ROOT; + } /* end if */ + /* Merge or redistribute child node pointers, if necessary */ + else { + hsize_t orig_n = n; /* Original index looked for */ + unsigned idx; /* Location of record which matches key */ + hbool_t found = FALSE; /* Comparison value of records */ + unsigned retries; /* Number of times to attempt redistribution */ + + /* Shadow the node if doing SWMR writes */ + if(hdr->swmr_write) { + if(H5B2__shadow_internal(internal, dxpl_id, curr_node_ptr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOPY, FAIL, "unable to shadow internal node") + internal_addr = curr_node_ptr->addr; + } /* end if */ + + /* Locate node pointer for child */ + if(swap_loc) + idx = 0; + else { + /* Search for record with correct index */ + for(idx = 0; idx < internal->nrec; idx++) { + /* Check which child node contains indexed record */ + if(internal->node_ptrs[idx].all_nrec >= n) { + /* Check if record is in this node */ + if(internal->node_ptrs[idx].all_nrec == n) { + /* Indicate the record was found and that the index + * in child nodes is zero from now on + */ + found = TRUE; + n = 0; + + /* Increment to next record */ + idx++; + } /* end if */ + + /* Break out of loop early */ + break; + } /* end if */ + + /* Decrement index we are looking for to account for the node we + * just advanced past. + */ + n -= (internal->node_ptrs[idx].all_nrec + 1); + } /* end for */ + } /* end else */ + + /* Set the number of redistribution retries */ + /* This takes care of the case where a B-tree node needs to be + * redistributed, but redistributing the node causes the index + * for removal to move to another node, which also needs to be + * redistributed. Now, we loop trying to redistribute and then + * eventually force a merge */ + retries = 2; + + /* Preemptively merge/redistribute a node we will enter */ + while(internal->node_ptrs[idx].node_nrec == merge_nrec) { + /* Attempt to redistribute records among children */ + /* (NOTE: These 2-node redistributions should actually get the + * record to promote from the node with more records. - QAK) + */ + /* (NOTE: This code is the same in both H5B2__remove_internal() and + * H5B2__remove_internal_by_idx(), fix bugs in both places! - QAK) + */ + if(idx == 0) { /* Left-most child */ + if(retries > 0 && (internal->node_ptrs[idx + 1].node_nrec > merge_nrec)) { + if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") + } /* end else */ + } /* end if */ + else if(idx == internal->nrec) { /* Right-most child */ + if(retries > 0 && (internal->node_ptrs[idx - 1].node_nrec > merge_nrec)) { + if(H5B2__redistribute2(hdr, dxpl_id, depth, internal, (idx - 1)) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__merge2(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, (idx - 1)) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") + } /* end else */ + } /* end if */ + else { /* Middle child */ + if(retries > 0 && ((internal->node_ptrs[idx + 1].node_nrec > merge_nrec) || + (internal->node_ptrs[idx - 1].node_nrec > merge_nrec))) { + if(H5B2__redistribute3(hdr, dxpl_id, depth, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTREDISTRIBUTE, FAIL, "unable to redistribute child node records") + } /* end if */ + else { + if(H5B2__merge3(hdr, dxpl_id, depth, curr_node_ptr, + parent_cache_info_flags_ptr, internal, &internal_flags, idx) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSPLIT, FAIL, "unable to merge child node") + } /* end else */ + } /* end else */ + + /* Locate node pointer for child (after merge/redistribute) */ + if(swap_loc) + idx = 0; + else { + /* Count from the orginal index value again */ + n = orig_n; + + /* Reset "found" flag - the record may have shifted during the + * redistribute/merge + */ + found = FALSE; + + /* Search for record with correct index */ + for(idx = 0; idx < internal->nrec; idx++) { + /* Check which child node contains indexed record */ + if(internal->node_ptrs[idx].all_nrec >= n) { + /* Check if record is in this node */ + if(internal->node_ptrs[idx].all_nrec == n) { + /* Indicate the record was found and that the index + * in child nodes is zero from now on + */ + found = TRUE; + n = 0; + + /* Increment to next record */ + idx++; + } /* end if */ + + /* Break out of loop early */ + break; + } /* end if */ + + /* Decrement index we are looking for to account for the node we + * just advanced past. + */ + n -= (internal->node_ptrs[idx].all_nrec + 1); + } /* end for */ + } /* end else */ + + /* Decrement the number of redistribution retries left */ + retries--; + } /* end while */ + + /* Handle deleting a record from an internal node */ + if(!swap_loc && found) { + swap_loc = H5B2_INT_NREC(internal, hdr, idx - 1); + swap_parent = internal; + } /* end if */ + + /* Swap record to delete with record from leaf, if we are the last internal node */ + if(swap_loc && depth == 1) + if(H5B2__swap_leaf(hdr, dxpl_id, depth, internal, &internal_flags, idx, swap_loc) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSWAP, FAIL, "can't swap records in B-tree") + + /* Set pointers for advancing to child node */ + new_cache_info_flags_ptr = &internal_flags; + new_cache_info = &internal->cache_info; + new_node_ptr = &internal->node_ptrs[idx]; + + /* Indicate position of next node */ + if(H5B2_POS_MIDDLE != curr_pos) { + if(idx == 0) { + if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) + next_pos = H5B2_POS_LEFT; + } /* end if */ + else if(idx == internal->nrec) { + if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) + next_pos = H5B2_POS_RIGHT; + } /* end if */ + } /* end if */ + } /* end else */ + + /* Attempt to remove record from child node */ + if(depth > 1) { + if(H5B2__remove_internal_by_idx(hdr, dxpl_id, depth_decreased, swap_loc, swap_parent, (uint16_t)(depth - 1), + new_cache_info, new_cache_info_flags_ptr, new_node_ptr, next_pos, n, op, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree internal node") + } /* end if */ + else { + if(H5B2__remove_leaf_by_idx(hdr, dxpl_id, new_node_ptr, next_pos, new_cache_info, (unsigned)n, op, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record from B-tree leaf node") + } /* end else */ + + /* Update record count for node pointer to child node */ + if(!collapsed_root) + new_node_ptr->all_nrec--; + + /* Mark node as dirty */ + if(!(hdr->swmr_write && collapsed_root)) + internal_flags |= H5AC__DIRTIED_FLAG; + +#ifdef H5B2_DEBUG + H5B2__assert_internal((!collapsed_root ? (curr_node_ptr->all_nrec - 1) : new_node_ptr->all_nrec), hdr, internal); +#endif /* H5B2_DEBUG */ + +done: + /* Release the B-tree internal node */ + if(internal && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, internal_addr, internal, internal_flags) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release internal B-tree node") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__remove_internal_by_idx() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__internal_free + * + * Purpose: Destroys a B-tree internal node in memory. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Feb 2 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__internal_free(H5B2_internal_t *internal) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* + * Check arguments. + */ + HDassert(internal); + + /* Release internal node's native key buffer */ + if(internal->int_native) + internal->int_native = (uint8_t *)H5FL_FAC_FREE(internal->hdr->node_info[internal->depth].nat_rec_fac, internal->int_native); + + /* Release internal node's node pointer buffer */ + if(internal->node_ptrs) + internal->node_ptrs = (H5B2_node_ptr_t *)H5FL_FAC_FREE(internal->hdr->node_info[internal->depth].node_ptr_fac, internal->node_ptrs); + + /* Decrement ref. count on B-tree header */ + if(H5B2__hdr_decr(internal->hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDEC, FAIL, "can't decrement ref. count on B-tree header") + + /* Sanity check */ + HDassert(NULL == internal->top_proxy); + + /* Free B-tree internal node info */ + internal = H5FL_FREE(H5B2_internal_t, internal); + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__internal_free() */ + +#ifdef H5B2_DEBUG + +/*------------------------------------------------------------------------- + * Function: H5B2__assert_internal + * + * Purpose: Verify than an internal node is mostly sane + * + * Return: Non-negative on success, negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Feb 19 2005 + * + *------------------------------------------------------------------------- + */ +H5_ATTR_PURE herr_t +H5B2__assert_internal(hsize_t parent_all_nrec, const H5B2_hdr_t *hdr, const H5B2_internal_t *internal) +{ + hsize_t tot_all_nrec; /* Total number of records at or below this node */ + uint16_t u, v; /* Local index variables */ + + /* General sanity checking on node */ + HDassert(internal->nrec <= hdr->node_info->split_nrec); + + /* Sanity checking on node pointers */ + tot_all_nrec = internal->nrec; + for(u = 0; u < internal->nrec + 1; u++) { + tot_all_nrec += internal->node_ptrs[u].all_nrec; + + HDassert(H5F_addr_defined(internal->node_ptrs[u].addr)); + HDassert(internal->node_ptrs[u].addr > 0); + for(v = 0; v < u; v++) + HDassert(internal->node_ptrs[u].addr != internal->node_ptrs[v].addr); + } /* end for */ + + /* Sanity check all_nrec total in parent */ + if(parent_all_nrec > 0) + HDassert(tot_all_nrec == parent_all_nrec); + + return(0); +} /* end H5B2__assert_internal() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__assert_internal2 + * + * Purpose: Verify than internal nodes are mostly sane + * + * Return: Non-negative on success, negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Feb 19 2005 + * + *------------------------------------------------------------------------- + */ +H5_ATTR_PURE herr_t +H5B2__assert_internal2(hsize_t parent_all_nrec, const H5B2_hdr_t *hdr, const H5B2_internal_t *internal, const H5B2_internal_t *internal2) +{ + hsize_t tot_all_nrec; /* Total number of records at or below this node */ + uint16_t u, v; /* Local index variables */ + + /* General sanity checking on node */ + HDassert(internal->nrec <= hdr->node_info->split_nrec); + + /* Sanity checking on node pointers */ + tot_all_nrec =internal->nrec; + for(u =0; u < internal->nrec + 1; u++) { + tot_all_nrec += internal->node_ptrs[u].all_nrec; + + HDassert(H5F_addr_defined(internal->node_ptrs[u].addr)); + HDassert(internal->node_ptrs[u].addr > 0); + for(v = 0; v < u; v++) + HDassert(internal->node_ptrs[u].addr != internal->node_ptrs[v].addr); + for(v = 0; v < internal2->nrec + 1; v++) + HDassert(internal->node_ptrs[u].addr != internal2->node_ptrs[v].addr); + } /* end for */ + + /* Sanity check all_nrec total in parent */ + if(parent_all_nrec > 0) + HDassert(tot_all_nrec == parent_all_nrec); + + return(0); +} /* end H5B2__assert_internal2() */ +#endif /* H5B2_DEBUG */ + diff --git a/src/H5B2leaf.c b/src/H5B2leaf.c new file mode 100644 index 0000000..4f8b8e6 --- /dev/null +++ b/src/H5B2leaf.c @@ -0,0 +1,1065 @@ +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * + * Copyright by The HDF Group. * + * Copyright by the Board of Trustees of the University of Illinois. * + * All rights reserved. * + * * + * This file is part of HDF5. The full HDF5 copyright notice, including * + * terms governing use, modification, and redistribution, is contained in * + * the files COPYING and Copyright.html. COPYING can be found at the root * + * of the source code distribution tree; Copyright.html can be found at the * + * root level of an installed copy of the electronic HDF5 document set and * + * is linked from the top-level documents page. It can also be found at * + * http://hdfgroup.org/HDF5/doc/Copyright.html. If you do not have * + * access to either file, you may request a copy from help@hdfgroup.org. * + * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ + +/*------------------------------------------------------------------------- + * + * Created: H5B2leaf.c + * Dec 01 2016 + * Quincey Koziol <koziol@lbl.gov> + * + * Purpose: Routines for managing v2 B-tree leaf ndoes. + * + *------------------------------------------------------------------------- + */ + +/****************/ +/* Module Setup */ +/****************/ + +#include "H5B2module.h" /* This source code file is part of the H5B2 module */ + + +/***********/ +/* Headers */ +/***********/ +#include "H5private.h" /* Generic Functions */ +#include "H5B2pkg.h" /* v2 B-trees */ +#include "H5Eprivate.h" /* Error handling */ +#include "H5MFprivate.h" /* File memory management */ +#include "H5MMprivate.h" /* Memory management */ + + +/****************/ +/* Local Macros */ +/****************/ + + +/******************/ +/* Local Typedefs */ +/******************/ + + +/********************/ +/* Package Typedefs */ +/********************/ + + +/********************/ +/* Local Prototypes */ +/********************/ +static herr_t H5B2__shadow_leaf(H5B2_leaf_t *leaf, hid_t dxpl_id, + H5B2_node_ptr_t *curr_node_ptr); + + +/*********************/ +/* Package Variables */ +/*********************/ + +/* Declare a free list to manage the H5B2_leaf_t struct */ +H5FL_DEFINE(H5B2_leaf_t); + + +/*****************************/ +/* Library Private Variables */ +/*****************************/ + + +/*******************/ +/* Local Variables */ +/*******************/ + + + +/*------------------------------------------------------------------------- + * Function: H5B2__create_leaf + * + * Purpose: Creates empty leaf node of a B-tree and update node pointer + * to point to it. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Feb 2 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__create_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, void *parent, H5B2_node_ptr_t *node_ptr) +{ + H5B2_leaf_t *leaf = NULL; /* Pointer to new leaf node created */ + hbool_t inserted = FALSE; /* Whether the leaf node was inserted into cache */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(node_ptr); + + /* Allocate memory for leaf information */ + if(NULL == (leaf = H5FL_CALLOC(H5B2_leaf_t))) + HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree leaf info") + + /* Increment ref. count on B-tree header */ + if(H5B2__hdr_incr(hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINC, FAIL, "can't increment ref. count on B-tree header") + + /* Share B-tree header information */ + leaf->hdr = hdr; + + /* Allocate space for the native keys in memory */ + if(NULL == (leaf->leaf_native = (uint8_t *)H5FL_FAC_MALLOC(hdr->node_info[0].nat_rec_fac))) + HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed for B-tree leaf native keys") + HDmemset(leaf->leaf_native, 0, hdr->cls->nrec_size * hdr->node_info[0].max_nrec); + + /* Set parent */ + leaf->parent = parent; + + /* Set shadowed epoch to header's epoch */ + leaf->shadow_epoch = hdr->shadow_epoch; + + /* Allocate space on disk for the leaf */ + if(HADDR_UNDEF == (node_ptr->addr = H5MF_alloc(hdr->f, H5FD_MEM_BTREE, dxpl_id, (hsize_t)hdr->node_size))) + HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "file allocation failed for B-tree leaf node") + + /* Cache the new B-tree node */ + if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_BT2_LEAF, node_ptr->addr, leaf, H5AC__NO_FLAGS_SET) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINIT, FAIL, "can't add B-tree leaf to cache") + inserted = TRUE; + + /* Add leaf node as child of 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, leaf) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSET, FAIL, "unable to add v2 B-tree node as child of proxy") + leaf->top_proxy = hdr->top_proxy; + } /* end if */ + +done: + if(ret_value < 0) { + if(leaf) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(leaf) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTREMOVE, FAIL, "unable to remove v2 B-tree leaf node from cache") + + /* Release leaf node's disk space */ + if(H5F_addr_defined(node_ptr->addr) && H5MF_xfree(hdr->f, H5FD_MEM_BTREE, dxpl_id, node_ptr->addr, (hsize_t)hdr->node_size) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTFREE, FAIL, "unable to release file space for v2 B-tree leaf node") + + /* Destroy leaf node */ + if(H5B2__leaf_free(leaf) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTFREE, FAIL, "unable to release v2 B-tree leaf node") + } /* end if */ + } /* end if */ + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__create_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__protect_leaf + * + * Purpose: "Protect" an leaf node in the metadata cache + * + * Return: Pointer to leaf node on success/NULL on failure + * + * Programmer: Quincey Koziol + * koziol@hdfgroup.org + * May 5 2010 + * + *------------------------------------------------------------------------- + */ +H5B2_leaf_t * +H5B2__protect_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, void *parent, + H5B2_node_ptr_t *node_ptr, hbool_t shadow, unsigned flags) +{ + H5B2_leaf_cache_ud_t udata; /* User-data for callback */ + H5B2_leaf_t *leaf; /* v2 B-tree leaf node */ + H5B2_leaf_t *ret_value = NULL; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(node_ptr); + HDassert(H5F_addr_defined(node_ptr->addr)); + + /* only H5AC__READ_ONLY_FLAG may appear in flags */ + HDassert((flags & (unsigned)(~H5AC__READ_ONLY_FLAG)) == 0); + + /* Set up user data for callback */ + udata.f = hdr->f; + udata.hdr = hdr; + udata.parent = parent; + udata.nrec = node_ptr->node_nrec; + + /* Protect the leaf node */ + if(NULL == (leaf = (H5B2_leaf_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_BT2_LEAF, node_ptr->addr, &udata, flags))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, NULL, "unable to protect B-tree leaf node") + + /* Create top proxy, if it doesn't exist */ + if(hdr->top_proxy && NULL == leaf->top_proxy) { + /* Add leaf node as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, leaf) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTSET, NULL, "unable to add v2 B-tree leaf node as child of proxy") + leaf->top_proxy = hdr->top_proxy; + } /* end if */ + + /* Shadow the node, if requested */ + if(shadow) + if(H5B2__shadow_leaf(leaf, dxpl_id, node_ptr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOPY, NULL, "unable to shadow leaf node") + + /* Set return value */ + ret_value = leaf; + +done: + /* Clean up on error */ + if(!ret_value) { + /* Release the leaf node, if it was protected */ + if(leaf) { + /* Remove from v2 B-tree's proxy, if added */ + if(leaf->top_proxy) { + if(H5AC_proxy_entry_remove_child(leaf->top_proxy, leaf) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNDEPEND, NULL, "unable to destroy flush dependency between leaf node and v2 B-tree 'top' proxy") + leaf->top_proxy = NULL; + } /* end if */ + + /* Unprotect leaf node */ + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, node_ptr->addr, leaf, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, NULL, "unable to unprotect v2 B-tree leaf node, address = %llu", (unsigned long long)node_ptr->addr) + } /* end if */ + } /* end if */ + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__protect_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__neighbor_leaf + * + * Purpose: Locate a record relative to the specified information in a + * B-tree leaf node and return that information by filling in + * fields of the + * caller-supplied UDATA pointer depending on the type of leaf node + * requested. The UDATA can point to additional data passed + * to the key comparison function. + * + * The 'OP' routine is called with the record found and the + * OP_DATA pointer, to allow caller to return information about + * the record. + * + * The RANGE indicates whether to search for records less than or + * equal to, or greater than or equal to the information passed + * in with UDATA. + * + * Return: Non-negative on success, negative on failure. + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Mar 9 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__neighbor_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, + void *neighbor_loc, H5B2_compare_t comp, void *parent, void *udata, H5B2_found_t op, + void *op_data) +{ + H5B2_leaf_t *leaf; /* Pointer to leaf node */ + unsigned idx = 0; /* Location of record which matches key */ + int cmp = 0; /* Comparison value of records */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + HDassert(op); + + /* Lock current B-tree node */ + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, curr_node_ptr, FALSE, H5AC__READ_ONLY_FLAG))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + + /* Locate node pointer for child */ + if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + if(cmp > 0) + idx++; + else + if(cmp == 0 && comp == H5B2_COMPARE_GREATER) + idx++; + + /* Set the neighbor location, if appropriate */ + if(comp == H5B2_COMPARE_LESS) { + if(idx > 0) + neighbor_loc = H5B2_LEAF_NREC(leaf, hdr, idx - 1); + } /* end if */ + else { + HDassert(comp == H5B2_COMPARE_GREATER); + + if(idx < leaf->nrec) + neighbor_loc = H5B2_LEAF_NREC(leaf, hdr, idx); + } /* end else */ + + /* Make callback if neighbor record has been found */ + if(neighbor_loc) { + /* Make callback for current record */ + if((op)(neighbor_loc, op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "'found' callback failed for B-tree neighbor operation") + } /* end if */ + else + HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "unable to find neighbor record in B-tree") + +done: + /* Release the B-tree leaf node */ + if(leaf && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, curr_node_ptr->addr, leaf, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree leaf node") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__neighbor_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__insert_leaf + * + * Purpose: Adds a new record to a B-tree leaf node. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Mar 3 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__insert_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, + H5B2_nodepos_t curr_pos, void *parent, void *udata) +{ + H5B2_leaf_t *leaf; /* Pointer to leaf node */ + unsigned leaf_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting the leaf node */ + int cmp; /* Comparison value of records */ + unsigned idx = 0; /* Location of record which matches key */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + + /* Lock current B-tree node */ + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, curr_node_ptr, FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + + /* Must have a leaf node with enough space to insert a record now */ + HDassert(curr_node_ptr->node_nrec < hdr->node_info[0].max_nrec); + + /* Sanity check number of records */ + HDassert(curr_node_ptr->all_nrec == curr_node_ptr->node_nrec); + HDassert(leaf->nrec == curr_node_ptr->node_nrec); + + /* Check for inserting into empty leaf */ + if(leaf->nrec == 0) + idx = 0; + else { + /* Find correct location to insert this record */ + if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + if(cmp == 0) + HGOTO_ERROR(H5E_BTREE, H5E_EXISTS, FAIL, "record is already in B-tree") + if(cmp > 0) + idx++; + + /* Make room for new record */ + if(idx < leaf->nrec) + HDmemmove(H5B2_LEAF_NREC(leaf, hdr, idx + 1), H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size * (leaf->nrec - idx)); + } /* end else */ + + /* Make callback to store record in native form */ + if((hdr->cls->store)(H5B2_LEAF_NREC(leaf, hdr, idx), udata) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into leaf node") + + /* Mark the node as dirty */ + leaf_flags |= H5AC__DIRTIED_FLAG; + + /* Update record count for node pointer to current node */ + curr_node_ptr->all_nrec++; + curr_node_ptr->node_nrec++; + + /* Update record count for current node */ + leaf->nrec++; + + /* Check for new record being the min or max for the tree */ + /* (Don't use 'else' for the idx check, to allow for root leaf node) */ + if(H5B2_POS_MIDDLE != curr_pos) { + if(idx == 0) { + if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) { + if(hdr->min_native_rec == NULL) + if(NULL == (hdr->min_native_rec = H5MM_malloc(hdr->cls->nrec_size))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "memory allocation failed for v2 B-tree min record info") + HDmemcpy(hdr->min_native_rec, H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size); + } /* end if */ + } /* end if */ + if(idx == (unsigned)(leaf->nrec - 1)) { + if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) { + if(hdr->max_native_rec == NULL) + if(NULL == (hdr->max_native_rec = H5MM_malloc(hdr->cls->nrec_size))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "memory allocation failed for v2 B-tree max record info") + HDmemcpy(hdr->max_native_rec, H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size); + } /* end if */ + } /* end if */ + } /* end if */ + +done: + /* Release the B-tree leaf node (marked as dirty) */ + if(leaf) { + /* Shadow the node if doing SWMR writes */ + if(hdr->swmr_write && (leaf_flags & H5AC__DIRTIED_FLAG)) + if(H5B2__shadow_leaf(leaf, dxpl_id, curr_node_ptr) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTCOPY, FAIL, "unable to shadow leaf B-tree node") + + /* Unprotect leaf node */ + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, curr_node_ptr->addr, leaf, leaf_flags) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release leaf B-tree node") + } /* end if */ + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__insert_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__update_leaf + * + * Purpose: Insert or modify a record in a B-tree leaf node. + * If the record exists already, it is modified as if H5B2_modify + * was called). If it doesn't exist, it is inserted as if + * H5B2_insert was called. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@hdfgroup.org + * Dec 23 2015 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__update_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, + H5B2_update_status_t *status, H5B2_nodepos_t curr_pos, void *parent, + void *udata, H5B2_modify_t op, void *op_data) +{ + H5B2_leaf_t *leaf; /* Pointer to leaf node */ + unsigned leaf_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting the leaf node */ + int cmp = -1; /* Comparison value of records */ + unsigned idx = 0; /* Location of record which matches key */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + + /* Lock current B-tree node */ + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, curr_node_ptr, FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + + /* Sanity check number of records */ + HDassert(curr_node_ptr->all_nrec == curr_node_ptr->node_nrec); + HDassert(leaf->nrec == curr_node_ptr->node_nrec); + + /* Check for inserting into empty leaf */ + if(leaf->nrec == 0) + idx = 0; + else { + /* Find correct location to insert this record */ + if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + + /* Check for inserting a record */ + if(0 != cmp) { + /* Check if the leaf node is full */ + if(curr_node_ptr->node_nrec == hdr->node_info[0].split_nrec) { + /* Indicate that the leaf is full, but we need to insert */ + *status = H5B2_UPDATE_INSERT_CHILD_FULL; + + /* Let calling routine handle insertion */ + HGOTO_DONE(SUCCEED) + } /* end if */ + + /* Adjust index to leave room for record to insert */ + if(cmp > 0) + idx++; + + /* Make room for new record */ + if(idx < leaf->nrec) + HDmemmove(H5B2_LEAF_NREC(leaf, hdr, idx + 1), H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size * (leaf->nrec - idx)); + } /* end if */ + } /* end else */ + + /* Check for modifying existing record */ + if(0 == cmp) { + hbool_t changed = FALSE; /* Whether the 'modify' callback changed the record */ + + /* Make callback for current record */ + if((op)(H5B2_LEAF_NREC(leaf, hdr, idx), op_data, &changed) < 0) { + /* Make certain that the callback didn't modify the value if it failed */ + HDassert(changed == FALSE); + + HGOTO_ERROR(H5E_BTREE, H5E_CANTMODIFY, FAIL, "'modify' callback failed for B-tree update operation") + } /* end if */ + + /* Mark the node as dirty if it changed */ + leaf_flags |= (changed ? H5AC__DIRTIED_FLAG : 0); + + /* Indicate that the record was modified */ + *status = H5B2_UPDATE_MODIFY_DONE; + } /* end if */ + else { + /* Must have a leaf node with enough space to insert a record now */ + HDassert(curr_node_ptr->node_nrec < hdr->node_info[0].max_nrec); + + /* Make callback to store record in native form */ + if((hdr->cls->store)(H5B2_LEAF_NREC(leaf, hdr, idx), udata) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTINSERT, FAIL, "unable to insert record into leaf node") + + /* Mark the node as dirty */ + leaf_flags |= H5AC__DIRTIED_FLAG; + + /* Indicate that the record was inserted */ + *status = H5B2_UPDATE_INSERT_DONE; + + /* Update record count for node pointer to current node */ + curr_node_ptr->all_nrec++; + curr_node_ptr->node_nrec++; + + /* Update record count for current node */ + leaf->nrec++; + } /* end else */ + + /* Check for new record being the min or max for the tree */ + /* (Don't use 'else' for the idx check, to allow for root leaf node) */ + if(H5B2_POS_MIDDLE != curr_pos) { + if(idx == 0) { + if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) { + if(hdr->min_native_rec == NULL) + if(NULL == (hdr->min_native_rec = H5MM_malloc(hdr->cls->nrec_size))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "memory allocation failed for v2 B-tree min record info") + HDmemcpy(hdr->min_native_rec, H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size); + } /* end if */ + } /* end if */ + if(idx == (unsigned)(leaf->nrec - 1)) { + if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) { + if(hdr->max_native_rec == NULL) + if(NULL == (hdr->max_native_rec = H5MM_malloc(hdr->cls->nrec_size))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "memory allocation failed for v2 B-tree max record info") + HDmemcpy(hdr->max_native_rec, H5B2_LEAF_NREC(leaf, hdr, idx), hdr->cls->nrec_size); + } /* end if */ + } /* end if */ + } /* end if */ + +done: + /* Release the B-tree leaf node */ + if(leaf) { + /* Check if we should shadow this node */ + if(hdr->swmr_write && (leaf_flags & H5AC__DIRTIED_FLAG)) { + /* Attempt to shadow the node if doing SWMR writes */ + if(H5B2__shadow_leaf(leaf, dxpl_id, curr_node_ptr) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTCOPY, FAIL, "unable to shadow leaf B-tree node") + + /* Change the state to "shadowed" if only modified currently */ + /* (Triggers parent to be marked dirty) */ + if(*status == H5B2_UPDATE_MODIFY_DONE) + *status = H5B2_UPDATE_SHADOW_DONE; + } /* end if */ + + /* Unprotect leaf node */ + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, curr_node_ptr->addr, leaf, leaf_flags) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release leaf B-tree node") + } /* end if */ + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__update_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__swap_leaf + * + * Purpose: Swap a record in a node with a record in a leaf node + * + * Return: Success: Non-negative + * + * Failure: Negative + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Mar 4 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__swap_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + H5B2_internal_t *internal, unsigned *internal_flags_ptr, + unsigned idx, void *swap_loc) +{ + const H5AC_class_t *child_class; /* Pointer to child node's class info */ + haddr_t child_addr; /* Address of child node */ + void *child = NULL; /* Pointer to child node */ + uint8_t *child_native; /* Pointer to child's native records */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(internal); + HDassert(internal_flags_ptr); + HDassert(idx <= internal->nrec); + + /* Check for the kind of B-tree node to swap */ + if(depth > 1) { + H5B2_internal_t *child_internal; /* Pointer to internal node */ + + /* Setup information for unlocking child node */ + child_class = H5AC_BT2_INT; + + /* Lock B-tree child nodes */ + if(NULL == (child_internal = H5B2__protect_internal(hdr, dxpl_id, internal, &internal->node_ptrs[idx], (uint16_t)(depth - 1), FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree internal node") + child_addr = internal->node_ptrs[idx].addr; + + /* More setup for accessing child node information */ + child = child_internal; + child_native = child_internal->int_native; + } /* end if */ + else { + H5B2_leaf_t *child_leaf; /* Pointer to leaf node */ + + /* Setup information for unlocking child nodes */ + child_class = H5AC_BT2_LEAF; + + /* Lock B-tree child node */ + if(NULL == (child_leaf = H5B2__protect_leaf(hdr, dxpl_id, internal, &internal->node_ptrs[idx], FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + child_addr = internal->node_ptrs[idx].addr; + + /* More setup for accessing child node information */ + child = child_leaf; + child_native = child_leaf->leaf_native; + } /* end else */ + + /* Swap records (use disk page as temporary buffer) */ + HDmemcpy(hdr->page, H5B2_NAT_NREC(child_native, hdr, 0), hdr->cls->nrec_size); + HDmemcpy(H5B2_NAT_NREC(child_native, hdr, 0), swap_loc, hdr->cls->nrec_size); + HDmemcpy(swap_loc, hdr->page, hdr->cls->nrec_size); + + /* Mark parent as dirty */ + *internal_flags_ptr |= H5AC__DIRTIED_FLAG; + +#ifdef H5B2_DEBUG + H5B2__assert_internal((hsize_t)0, hdr, internal); + if(depth > 1) + H5B2__assert_internal(internal->node_ptrs[idx].all_nrec, hdr, (H5B2_internal_t *)child); + else + H5B2__assert_leaf(hdr, (H5B2_leaf_t *)child); +#endif /* H5B2_DEBUG */ + +done: + /* Unlock child node */ + if(child && H5AC_unprotect(hdr->f, dxpl_id, child_class, child_addr, child, H5AC__DIRTIED_FLAG) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree child node") + + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__swap_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__shadow_leaf + * + * Purpose: "Shadow" a leaf node - copy it to a new location, leaving + * the data in the old location intact (for now). This is + * done when writing in SWMR mode to ensure that readers do + * not see nodes that are out of date with respect to each + * other and thereby inconsistent. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Neil Fortner + * Apr 27 2012 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5B2__shadow_leaf(H5B2_leaf_t *leaf, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr) +{ + H5B2_hdr_t *hdr; /* B-tree header */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_STATIC + + /* + * Check arguments. + */ + HDassert(leaf); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + hdr = leaf->hdr; + HDassert(hdr); + HDassert(hdr->swmr_write); + + /* We only need to shadow the node if it has not been shadowed since the + * last time the header was flushed, as otherwise it will be unreachable by + * the readers so there will be no need to shadow. To check if it has been + * shadowed, compare the epoch of this node and the header. If this node's + * epoch is <= to the header's, it hasn't been shadowed yet. */ + if(leaf->shadow_epoch <= hdr->shadow_epoch) { + haddr_t new_node_addr; /* Address to move node to */ + + /* + * We must clone the old node so readers with an out-of-date version of + * the parent can still see the correct number of children, via the + * shadowed node. Remove it from cache but do not mark it free on disk. + */ + /* Allocate space for the cloned node */ + if(HADDR_UNDEF == (new_node_addr = H5MF_alloc(hdr->f, H5FD_MEM_BTREE, dxpl_id, (hsize_t)hdr->node_size))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTALLOC, FAIL, "unable to allocate file space to move B-tree node") + + /* Move the location of the old child on the disk */ + if(H5AC_move_entry(hdr->f, H5AC_BT2_LEAF, curr_node_ptr->addr, new_node_addr, dxpl_id) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTMOVE, FAIL, "unable to move B-tree node") + curr_node_ptr->addr = new_node_addr; + + /* Should free the space in the file, but this is not supported by + * SWMR_WRITE code yet - QAK, 2016/12/01 + */ + + /* Set shadow epoch for node ahead of header */ + leaf->shadow_epoch = hdr->shadow_epoch + 1; + } /* end if */ + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__shadow_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__remove_leaf + * + * Purpose: Removes a record from a B-tree leaf node. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Mar 3 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__remove_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, + H5B2_nodepos_t curr_pos, void *parent, void *udata, H5B2_remove_t op, void *op_data) +{ + H5B2_leaf_t *leaf; /* Pointer to leaf node */ + haddr_t leaf_addr = HADDR_UNDEF; /* Leaf address on disk */ + unsigned leaf_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting leaf node */ + unsigned idx = 0; /* Location of record which matches key */ + int cmp; /* Comparison value of records */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + + /* Lock current B-tree node */ + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, curr_node_ptr, FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + leaf_addr = curr_node_ptr->addr; + + /* Sanity check number of records */ + HDassert(curr_node_ptr->all_nrec == curr_node_ptr->node_nrec); + HDassert(leaf->nrec == curr_node_ptr->node_nrec); + + /* Find correct location to remove this record */ + if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOMPARE, FAIL, "can't compare btree2 records") + if(cmp != 0) + HGOTO_ERROR(H5E_BTREE, H5E_NOTFOUND, FAIL, "record is not in B-tree") + + /* Check for invalidating the min/max record for the tree */ + if(H5B2_POS_MIDDLE != curr_pos) { + /* (Don't use 'else' for the idx check, to allow for root leaf node) */ + if(idx == 0) { + if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) { + if(hdr->min_native_rec) + hdr->min_native_rec = H5MM_xfree(hdr->min_native_rec); + } /* end if */ + } /* end if */ + if(idx == (unsigned)(leaf->nrec - 1)) { + if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) { + if(hdr->max_native_rec) + hdr->max_native_rec = H5MM_xfree(hdr->max_native_rec); + } /* end if */ + } /* end if */ + } /* end if */ + + /* Make 'remove' callback if there is one */ + if(op) + if((op)(H5B2_LEAF_NREC(leaf, hdr, idx), op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record into leaf node") + + /* Update number of records in node */ + leaf->nrec--; + + if(leaf->nrec > 0) { + /* Shadow the node if doing SWMR writes */ + if(hdr->swmr_write) { + if(H5B2__shadow_leaf(leaf, dxpl_id, curr_node_ptr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOPY, FAIL, "unable to shadow leaf node") + leaf_addr = curr_node_ptr->addr; + } /* end if */ + + /* Pack record out of leaf */ + if(idx < leaf->nrec) + HDmemmove(H5B2_LEAF_NREC(leaf, hdr, idx), H5B2_LEAF_NREC(leaf, hdr, (idx + 1)), hdr->cls->nrec_size * (leaf->nrec - idx)); + + /* Mark leaf node as dirty also */ + leaf_flags |= H5AC__DIRTIED_FLAG; + } /* end if */ + else { + /* Let the cache know that the object is deleted */ + leaf_flags |= H5AC__DELETED_FLAG; + if(!hdr->swmr_write) + leaf_flags |= H5AC__DIRTIED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; + + /* Reset address of parent node pointer */ + curr_node_ptr->addr = HADDR_UNDEF; + } /* end else */ + + /* Update record count for parent of leaf node */ + curr_node_ptr->node_nrec--; + +done: + /* Release the B-tree leaf node */ + if(leaf && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, leaf_addr, leaf, leaf_flags) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release leaf B-tree node") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__remove_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__remove_leaf_by_idx + * + * Purpose: Removes a record from a B-tree leaf node, according to the + * offset in the B-tree records. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@hdfgroup.org + * Nov 14 2006 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__remove_leaf_by_idx(H5B2_hdr_t *hdr, hid_t dxpl_id, + H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, void *parent, + unsigned idx, H5B2_remove_t op, void *op_data) +{ + H5B2_leaf_t *leaf; /* Pointer to leaf node */ + haddr_t leaf_addr = HADDR_UNDEF; /* Leaf address on disk */ + unsigned leaf_flags = H5AC__NO_FLAGS_SET; /* Flags for unprotecting leaf node */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* Check arguments. */ + HDassert(hdr); + HDassert(curr_node_ptr); + HDassert(H5F_addr_defined(curr_node_ptr->addr)); + + /* Lock B-tree leaf node */ + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, curr_node_ptr, FALSE, H5AC__NO_FLAGS_SET))) + HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + leaf_addr = curr_node_ptr->addr; + + /* Sanity check number of records */ + HDassert(curr_node_ptr->all_nrec == curr_node_ptr->node_nrec); + HDassert(leaf->nrec == curr_node_ptr->node_nrec); + HDassert(idx < leaf->nrec); + + /* Check for invalidating the min/max record for the tree */ + if(H5B2_POS_MIDDLE != curr_pos) { + /* (Don't use 'else' for the idx check, to allow for root leaf node) */ + if(idx == 0) { + if(H5B2_POS_LEFT == curr_pos || H5B2_POS_ROOT == curr_pos) { + if(hdr->min_native_rec) + hdr->min_native_rec = H5MM_xfree(hdr->min_native_rec); + } /* end if */ + } /* end if */ + if(idx == (unsigned)(leaf->nrec - 1)) { + if(H5B2_POS_RIGHT == curr_pos || H5B2_POS_ROOT == curr_pos) { + if(hdr->max_native_rec) + hdr->max_native_rec = H5MM_xfree(hdr->max_native_rec); + } /* end if */ + } /* end if */ + } /* end if */ + + /* Make 'remove' callback if there is one */ + if(op) + if((op)(H5B2_LEAF_NREC(leaf, hdr, idx), op_data) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDELETE, FAIL, "unable to remove record into leaf node") + + /* Update number of records in node */ + leaf->nrec--; + + if(leaf->nrec > 0) { + /* Shadow the node if doing SWMR writes */ + if(hdr->swmr_write) { + if(H5B2__shadow_leaf(leaf, dxpl_id, curr_node_ptr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTCOPY, FAIL, "unable to shadow leaf node") + leaf_addr = curr_node_ptr->addr; + } /* end if */ + + /* Pack record out of leaf */ + if(idx < leaf->nrec) + HDmemmove(H5B2_LEAF_NREC(leaf, hdr, idx), H5B2_LEAF_NREC(leaf, hdr, (idx + 1)), hdr->cls->nrec_size * (leaf->nrec - idx)); + + /* Mark leaf node as dirty also */ + leaf_flags |= H5AC__DIRTIED_FLAG; + } /* end if */ + else { + /* Let the cache know that the object is deleted */ + leaf_flags |= H5AC__DELETED_FLAG; + if(!hdr->swmr_write) + leaf_flags |= H5AC__DIRTIED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; + + /* Reset address of parent node pointer */ + curr_node_ptr->addr = HADDR_UNDEF; + } /* end else */ + + /* Update record count for parent of leaf node */ + curr_node_ptr->node_nrec--; + +done: + /* Release the B-tree leaf node */ + if(leaf && H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_LEAF, leaf_addr, leaf, leaf_flags) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release leaf B-tree node") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5B2__remove_leaf_by_idx() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__leaf_free + * + * Purpose: Destroys a B-tree leaf node in memory. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Feb 2 2005 + * + *------------------------------------------------------------------------- + */ +herr_t +H5B2__leaf_free(H5B2_leaf_t *leaf) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_PACKAGE + + /* + * Check arguments. + */ + HDassert(leaf); + + /* Release leaf's native key buffer */ + if(leaf->leaf_native) + leaf->leaf_native = (uint8_t *)H5FL_FAC_FREE(leaf->hdr->node_info[0].nat_rec_fac, leaf->leaf_native); + + /* Decrement ref. count on B-tree header */ + if(H5B2__hdr_decr(leaf->hdr) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTDEC, FAIL, "can't decrement ref. count on B-tree header") + + /* Sanity check */ + HDassert(NULL == leaf->top_proxy); + + /* Free B-tree leaf node info */ + leaf = H5FL_FREE(H5B2_leaf_t, leaf); + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5B2__leaf_free() */ + +#ifdef H5B2_DEBUG + +/*------------------------------------------------------------------------- + * Function: H5B2__assert_leaf + * + * Purpose: Verify than a leaf node is mostly sane + * + * Return: Non-negative on success, negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Feb 19 2005 + * + *------------------------------------------------------------------------- + */ +H5_ATTR_PURE herr_t +H5B2__assert_leaf(const H5B2_hdr_t *hdr, const H5B2_leaf_t *leaf) +{ + /* General sanity checking on node */ + HDassert(leaf->nrec <= hdr->node_info->split_nrec); + + return(0); +} /* end H5B2__assert_leaf() */ + + +/*------------------------------------------------------------------------- + * Function: H5B2__assert_leaf2 + * + * Purpose: Verify than a leaf node is mostly sane + * + * Return: Non-negative on success, negative on failure + * + * Programmer: Quincey Koziol + * koziol@ncsa.uiuc.edu + * Feb 19 2005 + * + *------------------------------------------------------------------------- + */ +H5_ATTR_PURE herr_t +H5B2__assert_leaf2(const H5B2_hdr_t *hdr, const H5B2_leaf_t *leaf, const H5B2_leaf_t H5_ATTR_UNUSED *leaf2) +{ + /* General sanity checking on node */ + HDassert(leaf->nrec <= hdr->node_info->split_nrec); + + return(0); +} /* end H5B2__assert_leaf2() */ +#endif /* H5B2_DEBUG */ + diff --git a/src/H5B2pkg.h b/src/H5B2pkg.h index 71700fe..7b1ec4d 100644 --- a/src/H5B2pkg.h +++ b/src/H5B2pkg.h @@ -125,6 +125,9 @@ #define H5B2_NUM_INT_REC(h, d) \ (((h)->node_size - (H5B2_INT_PREFIX_SIZE + H5B2_INT_POINTER_SIZE(h, d))) / ((h)->rrec_size + H5B2_INT_POINTER_SIZE(h, d))) +/* Uncomment this macro to enable extra sanity checking */ +/* #define H5B2_DEBUG */ + /****************************/ /* Package Private Typedefs */ @@ -185,6 +188,26 @@ typedef struct H5B2_hdr_t { void *min_native_rec; /* Pointer to minimum native record */ void *max_native_rec; /* Pointer to maximum native record */ + /* SWMR / Flush dependency information (not stored) */ + hbool_t swmr_write; /* Whether we are doing SWMR writes */ + H5AC_proxy_entry_t *top_proxy; /* 'Top' proxy cache entry for all B-tree entries */ + void *parent; /* Pointer to 'top' proxy flush dependency + * parent, if it exists, otherwise NULL. + * If the v2 B-tree is being used to index a + * chunked dataset and the dataset metadata is + * modified by a SWMR writer, this field will + * be set equal to the object header proxy + * that is the flush dependency parent + * of the v2 B-tree header. + * + * The field is used to avoid duplicate setups + * of the flush dependency relationship, and to + * allow the v2 B-tree header to destroy the + * flush dependency on receipt of an eviction + * notification from the metadata cache. + */ + uint64_t shadow_epoch; /* Epoch of header, for making shadow copies */ + /* Client information (not stored) */ const H5B2_class_t *cls; /* Class of B-tree client */ void *cb_ctx; /* Client callback context */ @@ -199,6 +222,11 @@ typedef struct H5B2_leaf_t { H5B2_hdr_t *hdr; /* Pointer to the [pinned] v2 B-tree header */ uint8_t *leaf_native; /* Pointer to native records */ uint16_t nrec; /* Number of records in node */ + + /* SWMR / Flush dependency information (not stored) */ + H5AC_proxy_entry_t *top_proxy; /* 'Top' proxy cache entry for all B-tree entries */ + void *parent; /* Flush dependency parent for leaf */ + uint64_t shadow_epoch; /* Epoch of node, for making shadow copies */ } H5B2_leaf_t; /* B-tree internal node information */ @@ -212,6 +240,11 @@ typedef struct H5B2_internal_t { H5B2_node_ptr_t *node_ptrs; /* Pointer to node pointers */ uint16_t nrec; /* Number of records in node */ uint16_t depth; /* Depth of this node in the B-tree */ + + /* SWMR / Flush dependency information (not stored) */ + H5AC_proxy_entry_t *top_proxy; /* 'Top' proxy cache entry for all B-tree entries */ + void *parent; /* Flush dependency parent for internal node */ + uint64_t shadow_epoch; /* Epoch of node, for making shadow copies */ } H5B2_internal_t; /* v2 B-tree */ @@ -232,6 +265,7 @@ typedef enum H5B2_nodepos_t { typedef enum H5B2_update_status_t { H5B2_UPDATE_UNKNOWN, /* Unknown update status (initial state) */ H5B2_UPDATE_MODIFY_DONE, /* Update successfully modified existing record */ + H5B2_UPDATE_SHADOW_DONE, /* Update modified existing record and modified node was shadowed */ H5B2_UPDATE_INSERT_DONE, /* Update inserted record successfully */ H5B2_UPDATE_INSERT_CHILD_FULL /* Update will insert record, but child is full */ } H5B2_update_status_t; @@ -247,6 +281,7 @@ typedef struct H5B2_hdr_cache_ud_t { typedef struct H5B2_internal_cache_ud_t { H5F_t *f; /* File that v2 b-tree header is within */ H5B2_hdr_t *hdr; /* v2 B-tree header */ + void *parent; /* Flush dependency parent */ uint16_t nrec; /* Number of records in node to load */ uint16_t depth; /* Depth of node to load */ } H5B2_internal_cache_ud_t; @@ -255,6 +290,7 @@ typedef struct H5B2_internal_cache_ud_t { typedef struct H5B2_leaf_cache_ud_t { H5F_t *f; /* File that v2 b-tree header is within */ H5B2_hdr_t *hdr; /* v2 B-tree header */ + void *parent; /* Flush dependency parent */ uint16_t nrec; /* Number of records in node to load */ } H5B2_leaf_cache_ud_t; @@ -306,6 +342,30 @@ extern const H5B2_class_t *const H5B2_client_class_g[H5B2_NUM_BTREE_ID]; /* Package Private Prototypes */ /******************************/ +/* Generic routines */ +H5_DLL herr_t H5B2__create_flush_depend(H5AC_info_t *parent_entry, + H5AC_info_t *child_entry); +H5_DLL herr_t H5B2__update_flush_depend(H5B2_hdr_t *hdr, hid_t dxpl_id, + unsigned depth, const H5B2_node_ptr_t *node_ptr, void *old_parent, + void *new_parent); +H5_DLL herr_t H5B2__destroy_flush_depend(H5AC_info_t *parent_entry, + H5AC_info_t *child_entry); + +/* Internal node management routines */ +H5_DLL herr_t H5B2__split1(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, + H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx); +H5_DLL herr_t H5B2__redistribute2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + H5B2_internal_t *internal, unsigned idx); +H5_DLL herr_t H5B2__redistribute3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx); +H5_DLL herr_t H5B2__merge2(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, + H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx); +H5_DLL herr_t H5B2__merge3(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + H5B2_node_ptr_t *curr_node_ptr, unsigned *parent_cache_info_flags_ptr, + H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx); + /* Routines for managing B-tree header info */ H5_DLL H5B2_hdr_t *H5B2__hdr_alloc(H5F_t *f); H5_DLL haddr_t H5B2__hdr_create(H5F_t *f, hid_t dxpl_id, @@ -324,17 +384,23 @@ H5_DLL herr_t H5B2__hdr_unprotect(H5B2_hdr_t *hdr, hid_t dxpl_id, H5_DLL herr_t H5B2__hdr_delete(H5B2_hdr_t *hdr, hid_t dxpl_id); /* Routines for operating on leaf nodes */ -H5B2_leaf_t *H5B2__protect_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, haddr_t addr, - uint16_t nrec, unsigned flags); +H5_DLL H5B2_leaf_t * H5B2__protect_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, + void *parent, H5B2_node_ptr_t *node_ptr, hbool_t shadow, unsigned flags); +H5_DLL herr_t H5B2__swap_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, + H5B2_internal_t *internal, unsigned *internal_flags_ptr, unsigned idx, + void *swap_loc); /* Routines for operating on internal nodes */ H5_DLL H5B2_internal_t *H5B2__protect_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, - haddr_t addr, uint16_t nrec, uint16_t depth, unsigned flags); + void *parent, H5B2_node_ptr_t *node_ptr, uint16_t depth, hbool_t shadow, + unsigned flags); /* Routines for allocating nodes */ H5_DLL herr_t H5B2__split_root(H5B2_hdr_t *hdr, hid_t dxpl_id); -H5_DLL herr_t H5B2__create_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, +H5_DLL herr_t H5B2__create_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, void *parent, H5B2_node_ptr_t *node_ptr); +H5_DLL herr_t H5B2__create_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, void *parent, + H5B2_node_ptr_t *node_ptr, uint16_t depth); /* Routines for releasing structures */ H5_DLL herr_t H5B2__hdr_free(H5B2_hdr_t *hdr); @@ -342,59 +408,64 @@ H5_DLL herr_t H5B2__leaf_free(H5B2_leaf_t *l); H5_DLL herr_t H5B2__internal_free(H5B2_internal_t *i); /* Routines for inserting records */ -H5_DLL herr_t H5B2__insert_hdr(H5B2_hdr_t *hdr, hid_t dxpl_id, void *udata); +H5_DLL herr_t H5B2__insert(H5B2_hdr_t *hdr, hid_t dxpl_id, void *udata); H5_DLL herr_t H5B2__insert_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, unsigned *parent_cache_info_flags_ptr, - H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, void *udata); + H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, void *parent, void *udata); H5_DLL herr_t H5B2__insert_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, - H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, void *udata); + H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, void *parent, void *udata); /* Routines for update records */ H5_DLL herr_t H5B2__update_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, unsigned *parent_cache_info_flags_ptr, H5B2_node_ptr_t *curr_node_ptr, H5B2_update_status_t *status, - H5B2_nodepos_t curr_pos, void *udata, H5B2_modify_t op, void *op_data); + H5B2_nodepos_t curr_pos, void *parent, void *udata, H5B2_modify_t op, + void *op_data); H5_DLL herr_t H5B2__update_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, H5B2_node_ptr_t *curr_node_ptr, H5B2_update_status_t *status, - H5B2_nodepos_t curr_pos, void *udata, H5B2_modify_t op, void *op_data); + H5B2_nodepos_t curr_pos, void *parent, void *udata, H5B2_modify_t op, + void *op_data); /* Routines for iterating over nodes/records */ H5_DLL herr_t H5B2__iterate_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - const H5B2_node_ptr_t *curr_node, H5B2_operator_t op, void *op_data); + const H5B2_node_ptr_t *curr_node, void *parent, H5B2_operator_t op, void *op_data); H5_DLL herr_t H5B2__node_size(H5B2_hdr_t *hdr, hid_t dxpl_id, - uint16_t depth, const H5B2_node_ptr_t *curr_node, hsize_t *op_data); + uint16_t depth, const H5B2_node_ptr_t *curr_node, void *parent, + hsize_t *op_data); /* Routines for locating records */ H5_DLL herr_t H5B2__locate_record(const H5B2_class_t *type, unsigned nrec, size_t *rec_off, const uint8_t *native, const void *udata, unsigned *idx, int *result); H5_DLL herr_t H5B2__neighbor_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, H5B2_node_ptr_t *curr_node_ptr, void *neighbor_loc, - H5B2_compare_t comp, void *udata, H5B2_found_t op, void *op_data); + H5B2_compare_t comp, void *parent, void *udata, H5B2_found_t op, + void *op_data); H5_DLL herr_t H5B2__neighbor_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, - H5B2_node_ptr_t *curr_node_ptr, void *neighbor_loc, - H5B2_compare_t comp, void *udata, H5B2_found_t op, void *op_data); + H5B2_node_ptr_t *curr_node_ptr, void *neighbor_loc, H5B2_compare_t comp, + void *parent, void *udata, H5B2_found_t op, void *op_data); /* Routines for removing records */ H5_DLL herr_t H5B2__remove_internal(H5B2_hdr_t *hdr, hid_t dxpl_id, - hbool_t *depth_decreased, void *swap_loc, uint16_t depth, + hbool_t *depth_decreased, void *swap_loc, void *swap_parent, uint16_t depth, H5AC_info_t *parent_cache_info, unsigned *parent_cache_info_flags_ptr, H5B2_nodepos_t curr_pos, H5B2_node_ptr_t *curr_node_ptr, void *udata, H5B2_remove_t op, void *op_data); H5_DLL herr_t H5B2__remove_leaf(H5B2_hdr_t *hdr, hid_t dxpl_id, - H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, + H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, void *parent, void *udata, H5B2_remove_t op, void *op_data); H5_DLL herr_t H5B2__remove_internal_by_idx(H5B2_hdr_t *hdr, hid_t dxpl_id, - hbool_t *depth_decreased, void *swap_loc, uint16_t depth, + hbool_t *depth_decreased, void *swap_loc, void *swap_parent, uint16_t depth, H5AC_info_t *parent_cache_info, unsigned *parent_cache_info_flags_ptr, H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, hsize_t n, H5B2_remove_t op, void *op_data); H5_DLL herr_t H5B2__remove_leaf_by_idx(H5B2_hdr_t *hdr, hid_t dxpl_id, - H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, + H5B2_node_ptr_t *curr_node_ptr, H5B2_nodepos_t curr_pos, void *parent, unsigned idx, H5B2_remove_t op, void *op_data); /* Routines for deleting nodes */ H5_DLL herr_t H5B2__delete_node(H5B2_hdr_t *hdr, hid_t dxpl_id, uint16_t depth, - const H5B2_node_ptr_t *curr_node, H5B2_remove_t op, void *op_data); + const H5B2_node_ptr_t *curr_node, void *parent, H5B2_remove_t op, + void *op_data); /* Debugging routines for dumping file structures */ H5_DLL herr_t H5B2__hdr_debug(H5F_t *f, hid_t dxpl_id, haddr_t addr, @@ -406,6 +477,15 @@ H5_DLL herr_t H5B2__leaf_debug(H5F_t *f, hid_t dxpl_id, haddr_t addr, FILE *stream, int indent, int fwidth, const H5B2_class_t *type, haddr_t hdr_addr, unsigned nrec, haddr_t obj_addr); +/* Sanity checking routines */ +#ifdef H5B2_DEBUG +/* Don't label these with H5_ATTR_PURE or you'll get even more warnings... */ +H5_DLL herr_t H5B2__assert_internal(hsize_t parent_all_nrec, const H5B2_hdr_t *hdr, const H5B2_internal_t *internal); +H5_DLL herr_t H5B2__assert_internal2(hsize_t parent_all_nrec, const H5B2_hdr_t *hdr, const H5B2_internal_t *internal, const H5B2_internal_t *internal2); +H5_DLL herr_t H5B2__assert_leaf(const H5B2_hdr_t *hdr, const H5B2_leaf_t *leaf); +H5_DLL herr_t H5B2__assert_leaf2(const H5B2_hdr_t *hdr, const H5B2_leaf_t *leaf, const H5B2_leaf_t *leaf2); +#endif /* H5B2_DEBUG */ + /* Testing routines */ #ifdef H5B2_TESTING H5_DLL herr_t H5B2_get_root_addr_test(H5B2_t *bt2, haddr_t *root_addr); diff --git a/src/H5B2private.h b/src/H5B2private.h index 3caf41f..161e25e 100644 --- a/src/H5B2private.h +++ b/src/H5B2private.h @@ -31,7 +31,8 @@ #include "H5B2public.h" /* Private headers needed by this file */ -#include "H5Fprivate.h" /* File access */ +#include "H5ACprivate.h" /* Metadata cache */ +#include "H5Fprivate.h" /* File access */ /**************************/ /* Library Private Macros */ @@ -153,6 +154,7 @@ H5_DLL herr_t H5B2_size(H5B2_t *bt2, hid_t dxpl_id, H5_DLL herr_t H5B2_close(H5B2_t *bt2, hid_t dxpl_id); H5_DLL herr_t H5B2_delete(H5F_t *f, hid_t dxpl_id, haddr_t addr, void *ctx_udata, H5B2_remove_t op, void *op_data); +H5_DLL herr_t H5B2_depend(H5B2_t *bt2, hid_t dxpl_id, H5AC_proxy_entry_t *parent); H5_DLL herr_t H5B2_patch_file(H5B2_t *fa, H5F_t *f); /* Statistics routines */ diff --git a/src/H5B2stat.c b/src/H5B2stat.c index 10c692e..da721c6 100644 --- a/src/H5B2stat.c +++ b/src/H5B2stat.c @@ -139,7 +139,7 @@ H5B2_size(H5B2_t *bt2, hid_t dxpl_id, hsize_t *btree_size) *btree_size += hdr->node_size; else /* Iterate through nodes */ - if(H5B2__node_size(hdr, dxpl_id, hdr->depth, &hdr->root, btree_size) < 0) + if(H5B2__node_size(hdr, dxpl_id, hdr->depth, &hdr->root, hdr, btree_size) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTLIST, FAIL, "node iteration failed") } /* end if */ diff --git a/src/H5B2test.c b/src/H5B2test.c index 24163e0..aec2aba 100644 --- a/src/H5B2test.c +++ b/src/H5B2test.c @@ -518,6 +518,7 @@ H5B2_get_node_info_test(H5B2_t *bt2, hid_t dxpl_id, void *udata, { H5B2_hdr_t *hdr; /* Pointer to the B-tree header */ H5B2_node_ptr_t curr_node_ptr; /* Node pointer info for current node */ + void *parent = NULL; /* Parent of current node */ uint16_t depth; /* Current depth of the tree */ int cmp; /* Comparison value of records */ unsigned idx; /* Location of record which matches key */ @@ -537,6 +538,10 @@ H5B2_get_node_info_test(H5B2_t *bt2, hid_t dxpl_id, void *udata, /* Make copy of the root node pointer to start search with */ curr_node_ptr = hdr->root; + /* Set initial parent, if doing swmr writes */ + if(hdr->swmr_write) + parent = hdr; + /* Current depth of the tree */ depth = hdr->depth; @@ -551,9 +556,16 @@ H5B2_get_node_info_test(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_node_ptr_t next_node_ptr; /* Node pointer info for next node */ /* Lock B-tree current node */ - if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, curr_node_ptr.addr, curr_node_ptr.node_nrec, depth, H5AC__READ_ONLY_FLAG))) + if(NULL == (internal = H5B2__protect_internal(hdr, dxpl_id, parent, &curr_node_ptr, depth, FALSE, H5AC__READ_ONLY_FLAG))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to load B-tree internal node") + /* Unpin parent if necessary */ + if(parent) { + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + parent = NULL; + } /* end if */ + /* Locate node pointer for child */ if(H5B2__locate_record(hdr->cls, internal->nrec, hdr->nat_off, internal->int_native, udata, &idx, &cmp) < 0) @@ -567,9 +579,13 @@ H5B2_get_node_info_test(H5B2_t *bt2, hid_t dxpl_id, void *udata, next_node_ptr = internal->node_ptrs[idx]; /* Unlock current node */ - if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, H5AC__NO_FLAGS_SET) < 0) + if(H5AC_unprotect(hdr->f, dxpl_id, H5AC_BT2_INT, curr_node_ptr.addr, internal, (unsigned)(hdr->swmr_write ? H5AC__PIN_ENTRY_FLAG : H5AC__NO_FLAGS_SET)) < 0) HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPROTECT, FAIL, "unable to release B-tree node") + /* Keep track of parent if necessary */ + if(hdr->swmr_write) + parent = internal; + /* Set pointer to next node to load */ curr_node_ptr = next_node_ptr; } /* end if */ @@ -594,9 +610,16 @@ H5B2_get_node_info_test(H5B2_t *bt2, hid_t dxpl_id, void *udata, H5B2_leaf_t *leaf; /* Pointer to leaf node in B-tree */ /* Lock B-tree leaf node */ - if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, curr_node_ptr.addr, curr_node_ptr.node_nrec, H5AC__READ_ONLY_FLAG))) + if(NULL == (leaf = H5B2__protect_leaf(hdr, dxpl_id, parent, &curr_node_ptr, FALSE, H5AC__READ_ONLY_FLAG))) HGOTO_ERROR(H5E_BTREE, H5E_CANTPROTECT, FAIL, "unable to protect B-tree leaf node") + /* Unpin parent if necessary */ + if(parent) { + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HGOTO_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + parent = NULL; + } /* end if */ + /* Locate record */ if(H5B2__locate_record(hdr->cls, leaf->nrec, hdr->nat_off, leaf->leaf_native, udata, &idx, &cmp) < 0) @@ -616,6 +639,12 @@ H5B2_get_node_info_test(H5B2_t *bt2, hid_t dxpl_id, void *udata, ninfo->nrec = curr_node_ptr.node_nrec; done: + if(parent) { + HDassert(ret_value < 0); + if(parent != hdr && H5AC_unpin_entry(parent) < 0) + HDONE_ERROR(H5E_BTREE, H5E_CANTUNPIN, FAIL, "unable to unpin parent entry") + } /* end if */ + FUNC_LEAVE_NOAPI(ret_value) } /* H5B2_get_node_info_test() */ diff --git a/src/H5Bdbg.c b/src/H5Bdbg.c index d92a24b..b22264d 100644 --- a/src/H5Bdbg.c +++ b/src/H5Bdbg.c @@ -77,6 +77,9 @@ H5B_debug(H5F_t *f, hid_t dxpl_id, haddr_t addr, FILE *stream, int indent, int f HDassert(fwidth >= 0); HDassert(type); + /* Currently does not support SWMR access */ + HDassert(!(H5F_INTENT(f) & H5F_ACC_SWMR_WRITE)); + /* Get shared info for B-tree */ if(NULL == (rc_shared = (type->get_shared)(f, udata))) HGOTO_ERROR(H5E_BTREE, H5E_CANTGET, FAIL, "can't retrieve B-tree's shared ref. count object") @@ -742,7 +742,39 @@ done: /*------------------------------------------------------------------------- + * Function: H5C_evict * + * Purpose: Evict all except pinned entries in the cache + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Vailin Choi + * Dec 2013 + * + *------------------------------------------------------------------------- + */ +herr_t +H5C_evict(H5F_t * f, hid_t dxpl_id) +{ + H5C_t *cache_ptr = f->shared->cache; + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity check */ + HDassert(cache_ptr); + HDassert(cache_ptr->magic == H5C__H5C_T_MAGIC); + + /* Flush and invalidate all cache entries except the pinned entries */ + if(H5C_flush_invalidate_cache(f, dxpl_id, H5C__EVICT_ALLOW_LAST_PINS_FLAG) < 0 ) + HGOTO_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "unable to evict entries in the cache") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* H5C_evict() */ + + +/*------------------------------------------------------------------------- * Function: H5C_expunge_entry * * Purpose: Use this function to tell the cache to expunge an entry @@ -1543,6 +1575,85 @@ done: /*------------------------------------------------------------------------- + * Function: H5C_mark_entry_clean + * + * Purpose: Mark a pinned entry as clean. The target entry MUST be pinned. + * + * If the entry is not + * already clean, the function places function marks the entry + * clean and removes it from the skip list. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * 7/23/16 + * + *------------------------------------------------------------------------- + */ +herr_t +H5C_mark_entry_clean(void *_thing) +{ + H5C_t * cache_ptr; + H5C_cache_entry_t * entry_ptr = (H5C_cache_entry_t *)_thing; + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(entry_ptr); + HDassert(H5F_addr_defined(entry_ptr->addr)); + cache_ptr = entry_ptr->cache_ptr; + HDassert(cache_ptr); + HDassert(cache_ptr->magic == H5C__H5C_T_MAGIC); + + /* Operate on pinned entry */ + if(entry_ptr->is_protected) + HGOTO_ERROR(H5E_CACHE, H5E_CANTMARKCLEAN, FAIL, "entry is protected") + else if(entry_ptr->is_pinned) { + hbool_t was_dirty; /* Whether the entry was previously dirty */ + + /* Remember previous dirty status */ + was_dirty = entry_ptr->is_dirty; + + /* Mark the entry as clean if it isn't already */ + entry_ptr->is_dirty = FALSE; + + /* Also reset the 'flush_marker' flag, since the entry shouldn't be flushed now */ + entry_ptr->flush_marker = FALSE; + + /* Modify cache data structures */ + if(was_dirty) + H5C__UPDATE_INDEX_FOR_ENTRY_CLEAN(cache_ptr, entry_ptr) + if(entry_ptr->in_slist) + H5C__REMOVE_ENTRY_FROM_SLIST(cache_ptr, entry_ptr, FALSE) + + /* Update stats for entry being marked clean */ + H5C__UPDATE_STATS_FOR_CLEAR(cache_ptr, entry_ptr) + + /* Check for entry changing status and do notifications, etc. */ + if(was_dirty) { + /* If the entry's type has a 'notify' callback send a 'entry cleaned' + * notice now that the entry is fully integrated into the cache. + */ + if(entry_ptr->type->notify && + (entry_ptr->type->notify)(H5C_NOTIFY_ACTION_ENTRY_CLEANED, entry_ptr) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTNOTIFY, FAIL, "can't notify client about entry dirty flag cleared") + + /* Propagate the clean up the flush dependency chain, if appropriate */ + if(entry_ptr->flush_dep_nparents > 0) + if(H5C__mark_flush_dep_clean(entry_ptr) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTMARKCLEAN, FAIL, "Can't propagate flush dep clean") + } /* end if */ + } /* end if */ + else + HGOTO_ERROR(H5E_CACHE, H5E_CANTMARKCLEAN, FAIL, "Entry is not pinned??") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* H5C_mark_entry_clean() */ + + +/*------------------------------------------------------------------------- * * Function: H5C_move_entry * @@ -1591,9 +1702,6 @@ H5C_move_entry(H5C_t * cache_ptr, HDassert(entry_ptr->addr == old_addr); HDassert(entry_ptr->type == type); - if(entry_ptr->is_protected) - HGOTO_ERROR(H5E_CACHE, H5E_CANTMOVE, FAIL, "target entry is protected.") - H5C__SEARCH_INDEX(cache_ptr, new_addr, test_entry_ptr, FAIL) if(test_entry_ptr != NULL) { /* we are hosed */ @@ -4955,17 +5063,19 @@ H5C_flush_invalidate_cache(const H5F_t * f, hid_t dxpl_id, unsigned flags) } /* end while */ /* Invariants, after destroying all entries in the hash table */ - HDassert(cache_ptr->index_size == 0); - HDassert(cache_ptr->clean_index_size == 0); - HDassert(cache_ptr->dirty_index_size == 0); - HDassert(cache_ptr->slist_len == 0); - HDassert(cache_ptr->slist_size == 0); - HDassert(cache_ptr->pel_len == 0); - HDassert(cache_ptr->pel_size == 0); - HDassert(cache_ptr->pl_len == 0); - HDassert(cache_ptr->pl_size == 0); - HDassert(cache_ptr->LRU_list_len == 0); - HDassert(cache_ptr->LRU_list_size == 0); + if(!(flags & H5C__EVICT_ALLOW_LAST_PINS_FLAG)) { + HDassert(cache_ptr->index_size == 0); + HDassert(cache_ptr->clean_index_size == 0); + HDassert(cache_ptr->dirty_index_size == 0); + HDassert(cache_ptr->slist_len == 0); + HDassert(cache_ptr->slist_size == 0); + HDassert(cache_ptr->pel_len == 0); + HDassert(cache_ptr->pel_size == 0); + HDassert(cache_ptr->pl_len == 0); + HDassert(cache_ptr->pl_size == 0); + HDassert(cache_ptr->LRU_list_len == 0); + HDassert(cache_ptr->LRU_list_size == 0); + } /* end if */ done: FUNC_LEAVE_NOAPI(ret_value) @@ -5307,12 +5417,16 @@ H5C_flush_invalidate_ring(const H5F_t * f, hid_t dxpl_id, H5C_ring_t ring, * into the cache, or moved to a new location * in the file as a side effect of the flush. * - * If this happens, and one of the target - * entries happens to be the next entry in - * the hash bucket, we could find ourselves - * either find ourselves either scanning a - * non-existant entry, scanning through a - * different bucket, or skipping an entry. + * It's also possible that removing a clean + * entry will remove the last child of a proxy + * entry, allowing it to be removed also and + * invalidating the next_entry_ptr. + * + * If either of these happen, and one of the target + * or proxy entries happens to be the next entry in + * the hash bucket, we could either find ourselves + * either scanning a non-existant entry, scanning + * through a different bucket, or skipping an entry. * * Neither of these are good, so restart the * the scan at the head of the hash bucket @@ -5889,6 +6003,11 @@ H5C__flush_single_entry(const H5F_t *f, hid_t dxpl_id, H5C_cache_entry_t *entry_ H5C__UPDATE_STATS_FOR_FLUSH(cache_ptr, entry_ptr) } /* end else if */ + /* Note that the algorithm below is (very) similar to the set of operations + * in H5C_remove_entry() and should be kept in sync with changes + * to that code. - QAK, 2016/11/30 + */ + /* Update the cache internal data structures. */ if(destroy) { /* Sanity checks */ @@ -7724,3 +7843,120 @@ done: FUNC_LEAVE_NOAPI(ret_value) } /* H5C__generate_image */ + +/*------------------------------------------------------------------------- + * + * Function: H5C_remove_entry + * + * Purpose: Remove an entry from the cache. Must be not protected, pinned, + * dirty, involved in flush dependencies, etc. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * September 17, 2016 + * + *------------------------------------------------------------------------- + */ +herr_t +H5C_remove_entry(void *_entry) +{ + H5C_cache_entry_t *entry = (H5C_cache_entry_t *)_entry; /* Entry to remove */ + H5C_t *cache; /* Cache for file */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Sanity checks */ + HDassert(entry); + HDassert(entry->ring != H5C_RING_UNDEFINED); + cache = entry->cache_ptr; + HDassert(cache); + HDassert(cache->magic == H5C__H5C_T_MAGIC); + + /* Check for error conditions */ + if(entry->is_dirty) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "can't remove dirty entry from cache") + if(entry->is_protected) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "can't remove protected entry from cache") + if(entry->is_pinned) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "can't remove pinned entry from cache") + if(entry->flush_dep_nparents > 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "can't remove entry with flush dependency parents from cache") + if(entry->flush_dep_nchildren > 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "can't remove entry with flush dependency children from cache") + + /* Additional internal cache consistency checks */ + HDassert(!entry->in_slist); + HDassert(!entry->flush_marker); + HDassert(!entry->flush_in_progress); + + /* Note that the algorithm below is (very) similar to the set of operations + * in H5C__flush_single_entry() and should be kept in sync with changes + * to that code. - QAK, 2016/11/30 + */ + + /* Update stats, as if we are "destroying" and taking ownership of the entry */ + H5C__UPDATE_STATS_FOR_EVICTION(cache, entry, TRUE) + + /* If the entry's type has a 'notify' callback, send a 'before eviction' + * notice while the entry is still fully integrated in the cache. + */ + if(entry->type->notify && (entry->type->notify)(H5C_NOTIFY_ACTION_BEFORE_EVICT, entry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTNOTIFY, FAIL, "can't notify client about entry to evict") + + /* Update the cache internal data structures as appropriate for a destroy. + * Specifically: + * 1) Delete it from the index + * 2) Update the replacement policy for eviction + * 3) Remove it from the tag list for this object + */ + + H5C__DELETE_FROM_INDEX(cache, entry) + + H5C__UPDATE_RP_FOR_EVICTION(cache, entry, FAIL) + + /* Remove entry from tag list */ + if(H5C__untag_entry(cache, entry) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTREMOVE, FAIL, "can't remove entry from tag list") + + /* Increment entries_removed_counter and set last_entry_removed_ptr. + * As we me be about to free the entry, recall that last_entry_removed_ptr + * must NEVER be dereferenced. + * + * Recall that these fields are maintained to allow functions that perform + * scans of lists of entries to detect the unexpected removal of entries + * (via expunge, eviction, or take ownership at present), so that they can + * re-start their scans if necessary. + * + * Also check if the entry we are watching for removal is being + * removed (usually the 'next' entry for an iteration) and reset + * it to indicate that it was removed. + */ + cache->entries_removed_counter++; + cache->last_entry_removed_ptr = entry; + if(entry == cache->entry_watched_for_removal) + cache->entry_watched_for_removal = NULL; + + /* Internal cache data structures should now be up to date, and + * consistant with the status of the entry. + * + * Now clean up internal cache fields if appropriate. + */ + + /* Free the buffer for the on disk image */ + if(entry->image_ptr != NULL) + entry->image_ptr = H5MM_xfree(entry->image_ptr); + + /* Reset the pointer to the cache the entry is within */ + entry->cache_ptr = NULL; + + /* Client is taking ownership of the entry. Set bad magic here so the + * cache will choke unless the entry is re-inserted properly + */ + entry->magic = H5C__H5C_CACHE_ENTRY_T_BAD_MAGIC; + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* H5C__remove_entry() */ + diff --git a/src/H5Cpkg.h b/src/H5Cpkg.h index 16efb5c..fae4d74 100644 --- a/src/H5Cpkg.h +++ b/src/H5Cpkg.h @@ -1508,7 +1508,6 @@ if ( ( (cache_ptr)->index_size != \ HDassert( (cache_ptr) ); \ HDassert( (cache_ptr)->magic == H5C__H5C_T_MAGIC ); \ HDassert( (entry_ptr) ); \ - HDassert( !((entry_ptr)->is_protected) ); \ HDassert( !((entry_ptr)->is_read_only) ); \ HDassert( ((entry_ptr)->ro_ref_count) == 0 ); \ HDassert( (entry_ptr)->size > 0 ); \ @@ -1548,7 +1547,6 @@ if ( ( (cache_ptr)->index_size != \ HDassert( (cache_ptr) ); \ HDassert( (cache_ptr)->magic == H5C__H5C_T_MAGIC ); \ HDassert( (entry_ptr) ); \ - HDassert( !((entry_ptr)->is_protected) ); \ HDassert( !((entry_ptr)->is_read_only) ); \ HDassert( ((entry_ptr)->ro_ref_count) == 0 ); \ HDassert( (entry_ptr)->in_slist ); \ @@ -2398,12 +2396,11 @@ if ( ( (cache_ptr)->index_size != \ HDassert( (cache_ptr) ); \ HDassert( (cache_ptr)->magic == H5C__H5C_T_MAGIC ); \ HDassert( (entry_ptr) ); \ - HDassert( !((entry_ptr)->is_protected) ); \ HDassert( !((entry_ptr)->is_read_only) ); \ HDassert( ((entry_ptr)->ro_ref_count) == 0 ); \ HDassert( (entry_ptr)->size > 0 ); \ \ - if ( ! ( (entry_ptr)->is_pinned ) ) { \ + if ( ! ( (entry_ptr)->is_pinned ) && ! ( (entry_ptr->is_protected ) ) ) { \ \ /* modified LRU specific code */ \ \ @@ -2476,12 +2473,11 @@ if ( ( (cache_ptr)->index_size != \ HDassert( (cache_ptr) ); \ HDassert( (cache_ptr)->magic == H5C__H5C_T_MAGIC ); \ HDassert( (entry_ptr) ); \ - HDassert( !((entry_ptr)->is_protected) ); \ HDassert( !((entry_ptr)->is_read_only) ); \ HDassert( ((entry_ptr)->ro_ref_count) == 0 ); \ HDassert( (entry_ptr)->size > 0 ); \ \ - if ( ! ( (entry_ptr)->is_pinned ) ) { \ + if ( ! ( (entry_ptr)->is_pinned ) && ! ( (entry_ptr->is_protected ) ) ) { \ \ /* modified LRU specific code */ \ \ diff --git a/src/H5Cprivate.h b/src/H5Cprivate.h index 923083f..baa80a2 100644 --- a/src/H5Cprivate.h +++ b/src/H5Cprivate.h @@ -41,7 +41,7 @@ /**************************/ /* Cache configuration settings */ -#define H5C__MAX_NUM_TYPE_IDS 28 +#define H5C__MAX_NUM_TYPE_IDS 29 #define H5C__PREFIX_LEN 32 /* This sanity checking constant was picked out of the air. Increase @@ -169,6 +169,9 @@ * These flags apply to H5C_expunge_entry(): * H5C__FREE_FILE_SPACE_FLAG * + * These flags apply to H5C_evict(): + * H5C__EVICT_ALLOW_LAST_PINS_FLAG + * * These flags apply to H5C_flush_cache(): * H5C__FLUSH_INVALIDATE_FLAG * H5C__FLUSH_CLEAR_ONLY_FLAG @@ -1716,6 +1719,7 @@ H5_DLL void H5C_def_auto_resize_rpt_fcn(H5C_t *cache_ptr, int32_t version, size_t old_max_cache_size, size_t new_max_cache_size, size_t old_min_clean_size, size_t new_min_clean_size); H5_DLL herr_t H5C_dest(H5F_t *f, hid_t dxpl_id); +H5_DLL herr_t H5C_evict(H5F_t *f, hid_t dxpl_id); H5_DLL herr_t H5C_expunge_entry(H5F_t *f, hid_t dxpl_id, const H5C_class_t *type, haddr_t addr, unsigned flags); H5_DLL herr_t H5C_flush_cache(H5F_t *f, hid_t dxpl_id, unsigned flags); @@ -1743,6 +1747,7 @@ H5_DLL FILE *H5C_get_trace_file_ptr_from_entry(const H5C_cache_entry_t *entry_pt H5_DLL herr_t H5C_insert_entry(H5F_t *f, hid_t dxpl_id, const H5C_class_t *type, haddr_t addr, void *thing, unsigned int flags); H5_DLL herr_t H5C_mark_entry_dirty(void *thing); +H5_DLL herr_t H5C_mark_entry_clean(void *thing); H5_DLL herr_t H5C_move_entry(H5C_t *cache_ptr, const H5C_class_t *type, haddr_t old_addr, haddr_t new_addr); H5_DLL herr_t H5C_pin_protected_entry(void *thing); @@ -1770,6 +1775,7 @@ H5_DLL hbool_t H5C_get_ignore_tags(const H5C_t *cache_ptr); H5_DLL herr_t H5C_retag_entries(H5C_t * cache_ptr, haddr_t src_tag, haddr_t dest_tag); H5_DLL herr_t H5C_cork(H5C_t *cache_ptr, haddr_t obj_addr, unsigned action, hbool_t *corked); H5_DLL herr_t H5C_get_entry_ring(const H5F_t *f, haddr_t addr, H5C_ring_t *ring); +H5_DLL herr_t H5C_remove_entry(void * thing); #ifdef H5_HAVE_PARALLEL H5_DLL herr_t H5C_apply_candidate_list(H5F_t *f, hid_t dxpl_id, diff --git a/src/H5Ctag.c b/src/H5Ctag.c index 6d5b454..6b47cc7 100644 --- a/src/H5Ctag.c +++ b/src/H5Ctag.c @@ -627,8 +627,10 @@ H5C_verify_tag(int id, haddr_t tag) * constraints are met. */ if(tag == H5AC__IGNORE_TAG) HGOTO_ERROR(H5E_CACHE, H5E_CANTTAG, FAIL, "cannot ignore a tag while doing verification.") - else if(tag == H5AC__INVALID_TAG) - HGOTO_ERROR(H5E_CACHE, H5E_CANTTAG, FAIL, "no metadata tag provided") + else if(tag == H5AC__INVALID_TAG) { + if(id != H5AC_PROXY_ENTRY_ID) + HGOTO_ERROR(H5E_CACHE, H5E_CANTTAG, FAIL, "no metadata tag provided") + } /* end else-if */ else { /* Perform some sanity checks on tag value. Certain entry * types require certain tag values, so check that these diff --git a/src/H5Dbtree.c b/src/H5Dbtree.c index 85238da..8ef14b9 100644 --- a/src/H5Dbtree.c +++ b/src/H5Dbtree.c @@ -153,6 +153,7 @@ static herr_t H5D__btree_idx_dest(const H5D_chk_idx_info_t *idx_info); /* v1 B-tree indexed chunk I/O ops */ const H5D_chunk_ops_t H5D_COPS_BTREE[1] = {{ + FALSE, /* v1 B-tree indices does not support SWMR access */ H5D__btree_idx_init, /* insert */ H5D__btree_idx_create, /* create */ H5D__btree_idx_is_space_alloc, /* is_space_alloc */ diff --git a/src/H5Dbtree2.c b/src/H5Dbtree2.c index 3b9b803..f687a5d 100644 --- a/src/H5Dbtree2.c +++ b/src/H5Dbtree2.c @@ -105,6 +105,7 @@ static herr_t H5D__bt2_filt_debug(FILE *stream, int indent, int fwidth, /* Helper routine */ static herr_t H5D__bt2_idx_open(const H5D_chk_idx_info_t *idx_info); +static herr_t H5D__btree2_idx_depend(const H5D_chk_idx_info_t *idx_info); /* Callback for H5B2_iterate() which is called in H5D__bt2_idx_iterate() */ static int H5D__bt2_idx_iterate_cb(const void *_record, void *_udata); @@ -152,6 +153,7 @@ static herr_t H5D__bt2_idx_dest(const H5D_chk_idx_info_t *idx_info); /* Chunked dataset I/O ops for v2 B-tree indexing */ const H5D_chunk_ops_t H5D_COPS_BT2[1] = {{ + TRUE, /* Fixed array indices support SWMR access */ H5D__bt2_idx_init, /* init */ H5D__bt2_idx_create, /* create */ H5D__bt2_idx_is_space_alloc, /* is_space_alloc */ @@ -623,6 +625,68 @@ H5D__bt2_idx_init(const H5D_chk_idx_info_t H5_ATTR_UNUSED *idx_info, /*------------------------------------------------------------------------- + * Function: H5D__btree2_idx_depend + * + * Purpose: Create flush dependency between v2 B-tree and dataset's + * object header. + * + * Return: Success: non-negative + * Failure: negative + * + * Programmer: Quincey Koziol + * Friday, December 18, 2015 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5D__btree2_idx_depend(const H5D_chk_idx_info_t *idx_info) +{ + H5O_t *oh = NULL; /* Object header */ + H5O_loc_t oloc; /* Temporary object header location for dataset */ + H5AC_proxy_entry_t *oh_proxy; /* Dataset's object header proxy */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_STATIC + + /* Check args */ + HDassert(idx_info); + HDassert(idx_info->f); + HDassert(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE); + HDassert(idx_info->pline); + HDassert(idx_info->layout); + HDassert(H5D_CHUNK_IDX_BT2 == idx_info->layout->idx_type); + HDassert(idx_info->storage); + HDassert(H5D_CHUNK_IDX_BT2 == idx_info->storage->idx_type); + HDassert(H5F_addr_defined(idx_info->storage->idx_addr)); + HDassert(idx_info->storage->u.btree2.bt2); + + /* Set up object header location for dataset */ + H5O_loc_reset(&oloc); + oloc.file = idx_info->f; + oloc.addr = idx_info->storage->u.btree.dset_ohdr_addr; + + /* Get header */ + if(NULL == (oh = H5O_protect(&oloc, idx_info->dxpl_id, H5AC__READ_ONLY_FLAG, TRUE))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTPROTECT, FAIL, "unable to protect object header") + + /* Retrieve the dataset's object header proxy */ + if(NULL == (oh_proxy = H5O_get_proxy(oh))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "unable to get dataset object header proxy") + + /* Make the v2 B-tree a child flush dependency of the dataset's object header proxy */ + if(H5B2_depend(idx_info->storage->u.btree2.bt2, idx_info->dxpl_id, oh_proxy) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header proxy") + +done: + /* Release the object header from the cache */ + if(oh && H5O_unprotect(&oloc, idx_info->dxpl_id, oh, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_DATASET, H5E_CANTUNPROTECT, FAIL, "unable to release object header") + + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5D__btree2_idx_depend() */ + + +/*------------------------------------------------------------------------- * Function: H5D__bt2_idx_open() * * Purpose: Opens an existing v2 B-tree. @@ -667,6 +731,11 @@ H5D__bt2_idx_open(const H5D_chk_idx_info_t *idx_info) if(NULL == (idx_info->storage->u.btree2.bt2 = H5B2_open(idx_info->f, idx_info->dxpl_id, idx_info->storage->idx_addr, &u_ctx))) HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't open v2 B-tree for tracking chunked dataset") + /* Check for SWMR writes to the file */ + if(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE) + if(H5D__btree2_idx_depend(idx_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header") + done: FUNC_LEAVE_NOAPI(ret_value) } /* end H5D__bt2_idx_open() */ @@ -738,6 +807,11 @@ H5D__bt2_idx_create(const H5D_chk_idx_info_t *idx_info) if(H5B2_get_addr(idx_info->storage->u.btree2.bt2, &(idx_info->storage->idx_addr)) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get v2 B-tree address for tracking chunked dataset") + /* Check for SWMR writes to the file */ + if(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE) + if(H5D__btree2_idx_depend(idx_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header") + done: FUNC_LEAVE_NOAPI(ret_value) } /* end H5D__bt2_idx_create() */ @@ -1186,7 +1260,7 @@ H5D__bt2_idx_remove(const H5D_chk_idx_info_t *idx_info, H5D_chunk_common_ud_t *u /* Remove the record for the "dataset chunk" object from the v2 B-tree */ /* (space in the file for the object is freed in the 'remove' callback) */ - if(H5B2_remove(bt2, idx_info->dxpl_id, &bt2_udata, H5D__bt2_remove_cb, &remove_udata) < 0) + if(H5B2_remove(bt2, idx_info->dxpl_id, &bt2_udata, (H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE) ? NULL : H5D__bt2_remove_cb, &remove_udata) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTREMOVE, FAIL, "can't remove object from B-tree") done: @@ -1243,8 +1317,11 @@ H5D__bt2_idx_delete(const H5D_chk_idx_info_t *idx_info) remove_udata.f = idx_info->f; remove_udata.dxpl_id = idx_info->dxpl_id; - /* Set remove operation. */ - remove_op = H5D__bt2_remove_cb; + /* Set remove operation. Do not remove chunks in SWMR_WRITE mode */ + if(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE) + remove_op = NULL; + else + remove_op = H5D__bt2_remove_cb; /* Delete the v2 B-tree */ /*(space in the file for each object is freed in the 'remove' callback) */ diff --git a/src/H5Dchunk.c b/src/H5Dchunk.c index 7637339..7a646af 100644 --- a/src/H5Dchunk.c +++ b/src/H5Dchunk.c @@ -6292,8 +6292,13 @@ H5D__chunk_file_alloc(const H5D_chk_idx_info_t *idx_info, const H5F_block_t *old /* Check for chunk being same size */ if(new_chunk->length != old_chunk->length) { /* Release previous chunk */ - if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, old_chunk->offset, old_chunk->length) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + /* Only free the old location if not doing SWMR writes - otherwise + * we must keep the old chunk around in case a reader has an + * outdated version of the B-tree node + */ + if(!(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE)) + if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, old_chunk->offset, old_chunk->length) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") alloc_chunk = TRUE; } /* end if */ else { diff --git a/src/H5Dearray.c b/src/H5Dearray.c index 3f12e97..e9dbd0d 100644 --- a/src/H5Dearray.c +++ b/src/H5Dearray.c @@ -148,6 +148,7 @@ static herr_t H5D__earray_idx_dest(const H5D_chk_idx_info_t *idx_info); /* Generic extensible array routines */ static herr_t H5D__earray_idx_open(const H5D_chk_idx_info_t *idx_info); +static herr_t H5D__earray_idx_depend(const H5D_chk_idx_info_t *idx_info); /*********************/ @@ -156,6 +157,7 @@ static herr_t H5D__earray_idx_open(const H5D_chk_idx_info_t *idx_info); /* Extensible array indexed chunk I/O ops */ const H5D_chunk_ops_t H5D_COPS_EARRAY[1] = {{ + TRUE, /* Extensible array indices support SWMR access */ H5D__earray_idx_init, /* init */ H5D__earray_idx_create, /* create */ H5D__earray_idx_is_space_alloc, /* is_space_alloc */ @@ -716,6 +718,68 @@ H5D__earray_dst_dbg_context(void *_dbg_ctx) /*------------------------------------------------------------------------- + * Function: H5D__earray_idx_depend + * + * Purpose: Create flush dependency between extensible array and dataset's + * object header. + * + * Return: Success: non-negative + * Failure: negative + * + * Programmer: Quincey Koziol + * Tuesday, June 2, 2009 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5D__earray_idx_depend(const H5D_chk_idx_info_t *idx_info) +{ + H5O_t *oh = NULL; /* Object header */ + H5O_loc_t oloc; /* Temporary object header location for dataset */ + H5AC_proxy_entry_t *oh_proxy; /* Dataset's object header proxy */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_STATIC + + /* Check args */ + HDassert(idx_info); + HDassert(idx_info->f); + HDassert(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE); + HDassert(idx_info->pline); + HDassert(idx_info->layout); + HDassert(H5D_CHUNK_IDX_EARRAY == idx_info->layout->idx_type); + HDassert(idx_info->storage); + HDassert(H5D_CHUNK_IDX_EARRAY == idx_info->storage->idx_type); + HDassert(H5F_addr_defined(idx_info->storage->idx_addr)); + HDassert(idx_info->storage->u.earray.ea); + + /* Set up object header location for dataset */ + H5O_loc_reset(&oloc); + oloc.file = idx_info->f; + oloc.addr = idx_info->storage->u.earray.dset_ohdr_addr; + + /* Get header */ + if(NULL == (oh = H5O_protect(&oloc, idx_info->dxpl_id, H5AC__READ_ONLY_FLAG, TRUE))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTPROTECT, FAIL, "unable to protect object header") + + /* Retrieve the dataset's object header proxy */ + if(NULL == (oh_proxy = H5O_get_proxy(oh))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "unable to get dataset object header proxy") + + /* Make the extensible array a child flush dependency of the dataset's object header */ + if(H5EA_depend(idx_info->storage->u.earray.ea, idx_info->dxpl_id, oh_proxy) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header proxy") + +done: + /* Release the object header from the cache */ + if(oh && H5O_unprotect(&oloc, idx_info->dxpl_id, oh, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_DATASET, H5E_CANTUNPROTECT, FAIL, "unable to release object header") + + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5D__earray_idx_depend() */ + + +/*------------------------------------------------------------------------- * Function: H5D__earray_idx_open * * Purpose: Opens an existing extensible array. @@ -760,6 +824,11 @@ H5D__earray_idx_open(const H5D_chk_idx_info_t *idx_info) if(NULL == (idx_info->storage->u.earray.ea = H5EA_open(idx_info->f, idx_info->dxpl_id, idx_info->storage->idx_addr, &udata))) HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't open extensible array") + /* Check for SWMR writes to the file */ + if(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE) + if(H5D__earray_idx_depend(idx_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header") + done: FUNC_LEAVE_NOAPI(ret_value) } /* end H5D__earray_idx_open() */ @@ -910,6 +979,11 @@ H5D__earray_idx_create(const H5D_chk_idx_info_t *idx_info) if(H5EA_get_addr(idx_info->storage->u.earray.ea, &(idx_info->storage->idx_addr)) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't query extensible array address") + /* Check for SWMR writes to the file */ + if(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE) + if(H5D__earray_idx_depend(idx_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header") + done: FUNC_LEAVE_NOAPI(ret_value) } /* end H5D__earray_idx_create() */ @@ -1355,11 +1429,13 @@ H5D__earray_idx_remove(const H5D_chk_idx_info_t *idx_info, H5D_chunk_common_ud_t if(H5EA_get(ea, idx_info->dxpl_id, idx, &elmt) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk info") - /* Remove raw data chunk from file */ + /* Remove raw data chunk from file if not doing SWMR writes */ HDassert(H5F_addr_defined(elmt.addr)); - H5_CHECK_OVERFLOW(elmt.nbytes, /*From: */uint32_t, /*To: */hsize_t); - if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, elmt.addr, (hsize_t)elmt.nbytes) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + if(!(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE)) { + H5_CHECK_OVERFLOW(elmt.nbytes, /*From: */uint32_t, /*To: */hsize_t); + if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, elmt.addr, (hsize_t)elmt.nbytes) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + } /* end if */ /* Reset the info about the chunk for the index */ elmt.addr = HADDR_UNDEF; @@ -1375,11 +1451,13 @@ H5D__earray_idx_remove(const H5D_chk_idx_info_t *idx_info, H5D_chunk_common_ud_t if(H5EA_get(ea, idx_info->dxpl_id, idx, &addr) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk address") - /* Remove raw data chunk from file */ + /* Remove raw data chunk from file if not doing SWMR writes */ HDassert(H5F_addr_defined(addr)); - H5_CHECK_OVERFLOW(idx_info->layout->size, /*From: */uint32_t, /*To: */hsize_t); - if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, addr, (hsize_t)idx_info->layout->size) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + if(!(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE)) { + H5_CHECK_OVERFLOW(idx_info->layout->size, /*From: */uint32_t, /*To: */hsize_t); + if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, addr, (hsize_t)idx_info->layout->size) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + } /* end if */ /* Reset the address of the chunk for the index */ addr = HADDR_UNDEF; diff --git a/src/H5Dfarray.c b/src/H5Dfarray.c index 212f1f1..6b95e12 100644 --- a/src/H5Dfarray.c +++ b/src/H5Dfarray.c @@ -147,6 +147,7 @@ static herr_t H5D__farray_idx_dest(const H5D_chk_idx_info_t *idx_info); /* Generic fixed array routines */ static herr_t H5D__farray_idx_open(const H5D_chk_idx_info_t *idx_info); +static herr_t H5D__farray_idx_depend(const H5D_chk_idx_info_t *idx_info); /*********************/ /* Package Variables */ @@ -154,6 +155,7 @@ static herr_t H5D__farray_idx_open(const H5D_chk_idx_info_t *idx_info); /* Fixed array indexed chunk I/O ops */ const H5D_chunk_ops_t H5D_COPS_FARRAY[1] = {{ + TRUE, /* Fixed array indices support SWMR access */ H5D__farray_idx_init, /* init */ H5D__farray_idx_create, /* create */ H5D__farray_idx_is_space_alloc, /* is_space_alloc */ @@ -716,6 +718,68 @@ H5D__farray_filt_debug(FILE *stream, int indent, int fwidth, hsize_t idx, /*------------------------------------------------------------------------- + * Function: H5D__farray_idx_depend + * + * Purpose: Create flush dependency between fixed array and dataset's + * object header. + * + * Return: Success: non-negative + * Failure: negative + * + * Programmer: Vailin Choi + * Thursday, April 30, 2009 + * + *------------------------------------------------------------------------- + */ +static herr_t +H5D__farray_idx_depend(const H5D_chk_idx_info_t *idx_info) +{ + H5O_t *oh = NULL; /* Object header */ + H5O_loc_t oloc; /* Temporary object header location for dataset */ + H5AC_proxy_entry_t *oh_proxy; /* Dataset's object header proxy */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_STATIC + + /* Check args */ + HDassert(idx_info); + HDassert(idx_info->f); + HDassert(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE); + HDassert(idx_info->pline); + HDassert(idx_info->layout); + HDassert(H5D_CHUNK_IDX_FARRAY == idx_info->layout->idx_type); + HDassert(idx_info->storage); + HDassert(H5D_CHUNK_IDX_FARRAY == idx_info->storage->idx_type); + HDassert(H5F_addr_defined(idx_info->storage->idx_addr)); + HDassert(idx_info->storage->u.farray.fa); + + /* Set up object header location for dataset */ + H5O_loc_reset(&oloc); + oloc.file = idx_info->f; + oloc.addr = idx_info->storage->u.farray.dset_ohdr_addr; + + /* Get header */ + if(NULL == (oh = H5O_protect(&oloc, idx_info->dxpl_id, H5AC__READ_ONLY_FLAG, TRUE))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTPROTECT, FAIL, "unable to protect object header") + + /* Retrieve the dataset's object header proxy */ + if(NULL == (oh_proxy = H5O_get_proxy(oh))) + HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "unable to get dataset object header proxy") + + /* Make the fixed array a child flush dependency of the dataset's object header proxy */ + if(H5FA_depend(idx_info->storage->u.farray.fa, idx_info->dxpl_id, oh_proxy) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header proxy") + +done: + /* Release the object header from the cache */ + if(oh && H5O_unprotect(&oloc, idx_info->dxpl_id, oh, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_DATASET, H5E_CANTUNPROTECT, FAIL, "unable to release object header") + + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5D__farray_idx_depend() */ + + +/*------------------------------------------------------------------------- * Function: H5D__farray_idx_init * * Purpose: Initialize the indexing information for a dataset. @@ -784,6 +848,11 @@ H5D__farray_idx_open(const H5D_chk_idx_info_t *idx_info) if(NULL == (idx_info->storage->u.farray.fa = H5FA_open(idx_info->f, idx_info->dxpl_id, idx_info->storage->idx_addr, &udata))) HGOTO_ERROR(H5E_DATASET, H5E_CANTINIT, FAIL, "can't open fixed array") + /* Check for SWMR writes to the file */ + if(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE) + if(H5D__farray_idx_depend(idx_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header") + done: FUNC_LEAVE_NOAPI(ret_value) } /* end H5D__farray_idx_open() */ @@ -860,6 +929,11 @@ H5D__farray_idx_create(const H5D_chk_idx_info_t *idx_info) if(H5FA_get_addr(idx_info->storage->u.farray.fa, &(idx_info->storage->idx_addr)) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't query fixed array address") + /* Check for SWMR writes to the file */ + if(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE) + if(H5D__farray_idx_depend(idx_info) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTDEPEND, FAIL, "unable to create flush dependency on object header") + done: FUNC_LEAVE_NOAPI(ret_value) } /* end H5D__farray_idx_create() */ @@ -1224,11 +1298,13 @@ H5D__farray_idx_remove(const H5D_chk_idx_info_t *idx_info, H5D_chunk_common_ud_t if(H5FA_get(fa, idx_info->dxpl_id, idx, &elmt) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk info") - /* Remove raw data chunk from file */ + /* Remove raw data chunk from file if not doing SWMR writes */ HDassert(H5F_addr_defined(elmt.addr)); - H5_CHECK_OVERFLOW(elmt.nbytes, /*From: */uint32_t, /*To: */hsize_t); - if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, elmt.addr, (hsize_t)elmt.nbytes) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + if(!(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE)) { + H5_CHECK_OVERFLOW(elmt.nbytes, /*From: */uint32_t, /*To: */hsize_t); + if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, elmt.addr, (hsize_t)elmt.nbytes) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + } /* end if */ /* Reset the info about the chunk for the index */ elmt.addr = HADDR_UNDEF; @@ -1244,11 +1320,13 @@ H5D__farray_idx_remove(const H5D_chk_idx_info_t *idx_info, H5D_chunk_common_ud_t if(H5FA_get(fa, idx_info->dxpl_id, idx, &addr) < 0) HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk address") - /* Remove raw data chunk from file */ + /* Remove raw data chunk from file if not doing SWMR writes */ HDassert(H5F_addr_defined(addr)); - H5_CHECK_OVERFLOW(idx_info->layout->size, /*From: */uint32_t, /*To: */hsize_t); - if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, addr, (hsize_t)idx_info->layout->size) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + if(!(H5F_INTENT(idx_info->f) & H5F_ACC_SWMR_WRITE)) { + H5_CHECK_OVERFLOW(idx_info->layout->size, /*From: */uint32_t, /*To: */hsize_t); + if(H5MF_xfree(idx_info->f, H5FD_MEM_DRAW, idx_info->dxpl_id, addr, (hsize_t)idx_info->layout->size) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTFREE, FAIL, "unable to free chunk") + } /* end if */ /* Reset the address of the chunk for the index */ addr = HADDR_UNDEF; diff --git a/src/H5Dnone.c b/src/H5Dnone.c index 677ca67..0cadac2 100644 --- a/src/H5Dnone.c +++ b/src/H5Dnone.c @@ -82,6 +82,7 @@ static herr_t H5D__none_idx_dump(const H5O_storage_chunk_t *storage, FILE *strea /* Non Index chunk I/O ops */ const H5D_chunk_ops_t H5D_COPS_NONE[1] = {{ + FALSE, /* Non-indexed chunking don't current support SWMR access */ NULL, /* init */ H5D__none_idx_create, /* create */ H5D__none_idx_is_space_alloc, /* is_space_alloc */ diff --git a/src/H5Dpkg.h b/src/H5Dpkg.h index 12966f2..f54a9f2 100644 --- a/src/H5Dpkg.h +++ b/src/H5Dpkg.h @@ -325,6 +325,7 @@ typedef herr_t (*H5D_chunk_dest_func_t)(const H5D_chk_idx_info_t *idx_info); /* Typedef for grouping chunk I/O routines */ typedef struct H5D_chunk_ops_t { + hbool_t can_swim; /* Flag to indicate that the index supports SWMR access */ H5D_chunk_init_func_t init; /* Routine to initialize indexing information in memory */ H5D_chunk_create_func_t create; /* Routine to create chunk index */ H5D_chunk_is_space_alloc_func_t is_space_alloc; /* Query routine to determine if storage/index is allocated */ diff --git a/src/H5Dsingle.c b/src/H5Dsingle.c index e09a5cf..04b8971 100644 --- a/src/H5Dsingle.c +++ b/src/H5Dsingle.c @@ -84,6 +84,7 @@ static herr_t H5D__single_idx_dump(const H5O_storage_chunk_t *storage, FILE *str /* Non Index chunk I/O ops */ const H5D_chunk_ops_t H5D_COPS_SINGLE[1] = {{ + FALSE, /* Single Chunk indexing doesn't current support SWMR access */ H5D__single_idx_init, /* init */ H5D__single_idx_create, /* create */ H5D__single_idx_is_space_alloc, /* is_space_alloc */ diff --git a/src/H5Dvirtual.c b/src/H5Dvirtual.c index 3780b88..c85b1e9 100644 --- a/src/H5Dvirtual.c +++ b/src/H5Dvirtual.c @@ -768,7 +768,7 @@ H5D__virtual_open_source_dset(const H5D_t *vdset, /* Check if we need to open the source file */ if(HDstrcmp(source_dset->file_name, ".")) { /* Open the source file */ - if(NULL == (src_file = H5F_open(source_dset->file_name, H5F_INTENT(vdset->oloc.file) & H5F_ACC_RDWR, H5P_FILE_CREATE_DEFAULT, vdset->shared->layout.storage.u.virt.source_fapl, dxpl_id))) + if(NULL == (src_file = H5F_open(source_dset->file_name, H5F_INTENT(vdset->oloc.file) & (H5F_ACC_RDWR | H5F_ACC_SWMR_WRITE | H5F_ACC_SWMR_READ), H5P_FILE_CREATE_DEFAULT, vdset->shared->layout.storage.u.virt.source_fapl, dxpl_id))) H5E_clear_stack(NULL); /* Quick hack until proper support for H5Fopen with missing file is implemented */ else src_file_open = TRUE; @@ -795,7 +795,7 @@ END_FUNC(PRIV) /* end H5EA_get() */ */ BEGIN_FUNC(PRIV, ERR, herr_t, SUCCEED, FAIL, -H5EA_depend(H5AC_info_t *parent_entry, H5EA_t *ea)) +H5EA_depend(H5EA_t *ea, hid_t dxpl_id, H5AC_proxy_entry_t *parent)) /* Local variables */ H5EA_hdr_t *hdr = ea->hdr; /* Header for EA */ @@ -805,13 +805,25 @@ H5EA_depend(H5AC_info_t *parent_entry, H5EA_t *ea)) */ HDassert(ea); HDassert(hdr); + HDassert(parent); - /* Set the shared array header's file context for this operation */ - hdr->f = ea->f; + /* + * Check to see if a flush dependency between the extensible array + * and another data structure in the file has already been set up. + * If it hasn't, do so now. + */ + if(NULL == hdr->parent) { + /* Sanity check */ + HDassert(hdr->top_proxy); - /* Set up flush dependency between parent entry and extensible array header */ - if(H5EA__create_flush_depend(parent_entry, (H5AC_info_t *)hdr) < 0) - H5E_THROW(H5E_CANTDEPEND, "unable to create flush dependency on file metadata") + /* Set the shared array header's file context for this operation */ + hdr->f = ea->f; + + /* Add the extensible array as a child of the parent (proxy) */ + if(H5AC_proxy_entry_add_child(parent, hdr->f, dxpl_id, hdr->top_proxy) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array as child of proxy") + hdr->parent = parent; + } /* end if */ CATCH diff --git a/src/H5EAcache.c b/src/H5EAcache.c index b7846e9..1119e39 100644 --- a/src/H5EAcache.c +++ b/src/H5EAcache.c @@ -80,6 +80,7 @@ static void *H5EA__cache_hdr_deserialize(const void *image, size_t len, static herr_t H5EA__cache_hdr_image_len(const void *thing, size_t *image_len); static herr_t H5EA__cache_hdr_serialize(const H5F_t *f, void *image, size_t len, void *thing); +static herr_t H5EA__cache_hdr_notify(H5AC_notify_action_t action, void *thing); static herr_t H5EA__cache_hdr_free_icr(void *thing); static herr_t H5EA__cache_iblock_get_initial_load_size(void *udata, size_t *image_len); @@ -142,7 +143,7 @@ const H5AC_class_t H5AC_EARRAY_HDR[1] = {{ H5EA__cache_hdr_image_len, /* 'image_len' callback */ NULL, /* 'pre_serialize' callback */ H5EA__cache_hdr_serialize, /* 'serialize' callback */ - NULL, /* 'notify' callback */ + H5EA__cache_hdr_notify, /* 'notify' callback */ H5EA__cache_hdr_free_icr, /* 'free_icr' callback */ NULL, /* 'fsf_size' callback */ }}; @@ -521,6 +522,80 @@ END_FUNC(STATIC) /* end H5EA__cache_hdr_serialize() */ /*------------------------------------------------------------------------- + * Function: H5EA__cache_hdr_notify + * + * Purpose: Handle cache action notifications + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: John Mainzer + * 11/30/15 + * + *------------------------------------------------------------------------- + */ +BEGIN_FUNC(STATIC, ERR, +herr_t, SUCCEED, FAIL, +H5EA__cache_hdr_notify(H5AC_notify_action_t action, void *_thing)) + + /* Local variables */ + H5EA_hdr_t *hdr = (H5EA_hdr_t *)_thing; /* Pointer to the object */ + + /* Sanity check */ + HDassert(hdr); + + /* Check if the file was opened with SWMR-write access */ + if(hdr->swmr_write) { + /* Determine which action to take */ + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + case H5AC_NOTIFY_ACTION_AFTER_LOAD: + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* If hdr->parent != NULL, hdr->parent is used to destroy + * the flush dependency before the header is evicted. + */ + if(hdr->parent) { + /* Sanity check */ + HDassert(hdr->top_proxy); + + /* Destroy flush dependency on object header proxy */ + if(H5AC_proxy_entry_remove_child((H5AC_proxy_entry_t *)hdr->parent, (void *)hdr->top_proxy) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between extensible array and proxy") + hdr->parent = NULL; + } /* end if */ + + /* Detach from 'top' proxy for extensible array */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_remove_child(hdr->top_proxy, hdr) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between header and extensible array 'top' proxy") + /* Don't reset hdr->top_proxy here, it's destroyed when the header is freed -QAK */ + } /* end if */ + break; + + default: +#ifdef NDEBUG + H5E_THROW(H5E_BADVALUE, "unknown action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + } /* end if */ + else + HDassert(NULL == hdr->parent); + +CATCH + +END_FUNC(STATIC) /* end H5EA__cache_hdr_notify() */ + + +/*------------------------------------------------------------------------- * Function: H5EA__cache_hdr_free_icr * * Purpose: Destroy/release an "in core representation" of a data @@ -890,6 +965,13 @@ H5EA__cache_iblock_notify(H5AC_notify_action_t action, void *_thing)) /* Destroy flush dependency on extensible array header */ if(H5EA__destroy_flush_depend((H5AC_info_t *)iblock->hdr, (H5AC_info_t *)iblock) < 0) H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between index block and header, address = %llu", (unsigned long long)iblock->addr) + + /* Detach from 'top' proxy for extensible array */ + if(iblock->top_proxy) { + if(H5AC_proxy_entry_remove_child(iblock->top_proxy, iblock) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between index block and extensible array 'top' proxy") + iblock->top_proxy = NULL; + } /* end if */ break; default: @@ -1286,6 +1368,13 @@ H5EA__cache_sblock_notify(H5AC_notify_action_t action, void *_thing)) H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between super block and header, address = %llu", (unsigned long long)sblock->addr) sblock->has_hdr_depend = FALSE; } /* end if */ + + /* Detach from 'top' proxy for extensible array */ + if(sblock->top_proxy) { + if(H5AC_proxy_entry_remove_child(sblock->top_proxy, sblock) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between super block and extensible array 'top' proxy") + sblock->top_proxy = NULL; + } /* end if */ break; case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: @@ -1690,6 +1779,13 @@ H5EA__cache_dblock_notify(H5AC_notify_action_t action, void *_thing)) H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between data block and header, address = %llu", (unsigned long long)dblock->addr) dblock->has_hdr_depend = FALSE; } /* end if */ + + /* Detach from 'top' proxy for extensible array */ + if(dblock->top_proxy) { + if(H5AC_proxy_entry_remove_child(dblock->top_proxy, dblock) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between data block and extensible array 'top' proxy") + dblock->top_proxy = NULL; + } /* end if */ break; case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: @@ -2062,6 +2158,13 @@ H5EA__cache_dblk_page_notify(H5AC_notify_action_t action, void *_thing)) H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between data block page and header, address = %llu", (unsigned long long)dblk_page->addr) dblk_page->has_hdr_depend = FALSE; } /* end if */ + + /* Detach from 'top' proxy for extensible array */ + if(dblk_page->top_proxy) { + if(H5AC_proxy_entry_remove_child(dblk_page->top_proxy, dblk_page) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between data block page and extensible array 'top' proxy") + dblk_page->top_proxy = NULL; + } /* end if */ break; case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: diff --git a/src/H5EAdblkpage.c b/src/H5EAdblkpage.c index b903a8a..927dcb6 100644 --- a/src/H5EAdblkpage.c +++ b/src/H5EAdblkpage.c @@ -155,6 +155,7 @@ H5EA__dblk_page_create(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_sblock_t *parent, /* Local variables */ H5EA_dblk_page_t *dblk_page = NULL; /* Extensible array data block page */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ /* Sanity check */ HDassert(hdr); @@ -174,10 +175,23 @@ H5EA__dblk_page_create(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_sblock_t *parent, /* Cache the new extensible array data block page */ if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_EARRAY_DBLK_PAGE, dblk_page->addr, dblk_page, H5AC__NO_FLAGS_SET) < 0) H5E_THROW(H5E_CANTINSERT, "can't add extensible array data block page to cache") + inserted = TRUE; + + /* Add data block page as child of 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, dblk_page) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + dblk_page->top_proxy = hdr->top_proxy; + } /* end if */ CATCH if(ret_value < 0) if(dblk_page) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(dblk_page) < 0) + H5E_THROW(H5E_CANTREMOVE, "unable to remove extensible array data block page from cache") + /* Destroy data block page */ if(H5EA__dblk_page_dest(dblk_page) < 0) H5E_THROW(H5E_CANTFREE, "unable to destroy extensible array data block page") @@ -206,6 +220,7 @@ H5EA__dblk_page_protect(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_sblock_t *parent, haddr_t dblk_page_addr, unsigned flags)) /* Local variables */ + H5EA_dblk_page_t *dblk_page = NULL; /* Extensible array data block page */ H5EA_dblk_page_cache_ud_t udata; /* Information needed for loading data block page */ /* Sanity check */ @@ -221,10 +236,27 @@ H5EA__dblk_page_protect(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_sblock_t *parent, udata.dblk_page_addr = dblk_page_addr; /* Protect the data block page */ - if(NULL == (ret_value = (H5EA_dblk_page_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_EARRAY_DBLK_PAGE, dblk_page_addr, &udata, flags))) + if(NULL == (dblk_page = (H5EA_dblk_page_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_EARRAY_DBLK_PAGE, dblk_page_addr, &udata, flags))) H5E_THROW(H5E_CANTPROTECT, "unable to protect extensible array data block page, address = %llu", (unsigned long long)dblk_page_addr) + /* Create top proxy, if it doesn't exist */ + if(hdr->top_proxy && NULL == dblk_page->top_proxy) { + /* Add data block page as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, dblk_page) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + dblk_page->top_proxy = hdr->top_proxy; + } /* end if */ + + /* Set return value */ + ret_value = dblk_page; + CATCH + /* Clean up on error */ + if(!ret_value) { + /* Release the data block page, if it was protected */ + if(dblk_page && H5AC_unprotect(hdr->f, dxpl_id, H5AC_EARRAY_DBLK_PAGE, dblk_page->addr, dblk_page, H5AC__NO_FLAGS_SET) < 0) + H5E_THROW(H5E_CANTUNPROTECT, "unable to unprotect extensible array data block page, address = %llu", (unsigned long long)dblk_page->addr) + } /* end if */ END_FUNC(PKG) /* end H5EA__dblk_page_protect() */ @@ -299,6 +331,9 @@ H5EA__dblk_page_dest(H5EA_dblk_page_t *dblk_page)) dblk_page->hdr = NULL; } /* end if */ + /* Sanity check */ + HDassert(NULL == dblk_page->top_proxy); + /* Free the data block page itself */ dblk_page = H5FL_FREE(H5EA_dblk_page_t, dblk_page); diff --git a/src/H5EAdblock.c b/src/H5EAdblock.c index 9511a9d..c288c69 100644 --- a/src/H5EAdblock.c +++ b/src/H5EAdblock.c @@ -168,6 +168,7 @@ H5EA__dblock_create(H5EA_hdr_t *hdr, hid_t dxpl_id, void *parent, /* Local variables */ H5EA_dblock_t *dblock = NULL; /* Extensible array data block */ haddr_t dblock_addr; /* Extensible array data block address */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ /* Sanity check */ HDassert(hdr); @@ -198,6 +199,14 @@ H5EA__dblock_create(H5EA_hdr_t *hdr, hid_t dxpl_id, void *parent, /* Cache the new extensible array data block */ if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_EARRAY_DBLOCK, dblock_addr, dblock, H5AC__NO_FLAGS_SET) < 0) H5E_THROW(H5E_CANTINSERT, "can't add extensible array data block to cache") + inserted = TRUE; + + /* Add data block as child of 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, dblock) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + dblock->top_proxy = hdr->top_proxy; + } /* end if */ /* Update extensible array data block statistics */ hdr->stats.stored.ndata_blks++; @@ -215,6 +224,11 @@ H5EA__dblock_create(H5EA_hdr_t *hdr, hid_t dxpl_id, void *parent, CATCH if(!H5F_addr_defined(ret_value)) if(dblock) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(dblock) < 0) + H5E_THROW(H5E_CANTREMOVE, "unable to remove extensible array data block from cache") + /* Release data block's disk space */ if(H5F_addr_defined(dblock->addr) && H5MF_xfree(hdr->f, H5FD_MEM_EARRAY_DBLOCK, dxpl_id, dblock->addr, (hsize_t)dblock->size) < 0) H5E_THROW(H5E_CANTFREE, "unable to release extensible array data block") @@ -284,7 +298,8 @@ H5EA__dblock_protect(H5EA_hdr_t *hdr, hid_t dxpl_id, void *parent, haddr_t dblk_addr, size_t dblk_nelmts, unsigned flags)) /* Local variables */ - H5EA_dblock_cache_ud_t udata; /* Information needed for loading data block */ + H5EA_dblock_t *dblock; /* Extensible array data block */ + H5EA_dblock_cache_ud_t udata; /* Information needed for loading data block */ /* Sanity check */ HDassert(hdr); @@ -301,11 +316,29 @@ H5EA__dblock_protect(H5EA_hdr_t *hdr, hid_t dxpl_id, void *parent, udata.dblk_addr = dblk_addr; /* Protect the data block */ - if(NULL == (ret_value = (H5EA_dblock_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_EARRAY_DBLOCK, dblk_addr, &udata, flags))) + if(NULL == (dblock = (H5EA_dblock_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_EARRAY_DBLOCK, dblk_addr, &udata, flags))) H5E_THROW(H5E_CANTPROTECT, "unable to protect extensible array data block, address = %llu", (unsigned long long)dblk_addr) + /* Create top proxy, if it doesn't exist */ + if(hdr->top_proxy && NULL == dblock->top_proxy) { + /* Add data block as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, dblock) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + dblock->top_proxy = hdr->top_proxy; + } /* end if */ + + /* Set return value */ + ret_value = dblock; + CATCH + /* Clean up on error */ + if(!ret_value) { + /* Release the data block, if it was protected */ + if(dblock && H5AC_unprotect(hdr->f, dxpl_id, H5AC_EARRAY_DBLOCK, dblock->addr, dblock, H5AC__NO_FLAGS_SET) < 0) + H5E_THROW(H5E_CANTUNPROTECT, "unable to unprotect extensible array data block, address = %llu", (unsigned long long)dblock->addr) + } /* end if */ + END_FUNC(PKG) /* end H5EA__dblock_protect() */ @@ -442,6 +475,9 @@ H5EA__dblock_dest(H5EA_dblock_t *dblock)) dblock->hdr = NULL; } /* end if */ + /* Sanity check */ + HDassert(NULL == dblock->top_proxy); + /* Free the data block itself */ dblock = H5FL_FREE(H5EA_dblock_t, dblock); diff --git a/src/H5EAhdr.c b/src/H5EAhdr.c index d8be3cb..76d8733 100644 --- a/src/H5EAhdr.c +++ b/src/H5EAhdr.c @@ -135,6 +135,7 @@ H5EA__hdr_alloc(H5F_t *f)) /* Set the internal parameters for the array */ hdr->f = f; + hdr->swmr_write = (H5F_INTENT(f) & H5F_ACC_SWMR_WRITE) > 0; hdr->sizeof_addr = H5F_SIZEOF_ADDR(f); hdr->sizeof_size = H5F_SIZEOF_SIZE(f); @@ -361,6 +362,7 @@ H5EA__hdr_create(H5F_t *f, hid_t dxpl_id, const H5EA_create_t *cparam, /* Local variables */ H5EA_hdr_t *hdr = NULL; /* Extensible array header */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ /* Check arguments */ HDassert(f); @@ -418,9 +420,20 @@ H5EA__hdr_create(H5F_t *f, hid_t dxpl_id, const H5EA_create_t *cparam, if(HADDR_UNDEF == (hdr->addr = H5MF_alloc(f, H5FD_MEM_EARRAY_HDR, dxpl_id, (hsize_t)hdr->size))) H5E_THROW(H5E_CANTALLOC, "file allocation failed for extensible array header") + /* Create 'top' proxy for extensible array entries */ + if(hdr->swmr_write) + if(NULL == (hdr->top_proxy = H5AC_proxy_entry_create())) + H5E_THROW(H5E_CANTCREATE, "can't create extensible array entry proxy") + /* Cache the new extensible array header */ if(H5AC_insert_entry(f, dxpl_id, H5AC_EARRAY_HDR, hdr->addr, hdr, H5AC__NO_FLAGS_SET) < 0) H5E_THROW(H5E_CANTINSERT, "can't add extensible array header to cache") + inserted = TRUE; + + /* Add header as child of 'top' proxy */ + if(hdr->top_proxy) + if(H5AC_proxy_entry_add_child(hdr->top_proxy, f, dxpl_id, hdr) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") /* Set address of array header to return */ ret_value = hdr->addr; @@ -428,6 +441,11 @@ H5EA__hdr_create(H5F_t *f, hid_t dxpl_id, const H5EA_create_t *cparam, CATCH if(!H5F_addr_defined(ret_value)) if(hdr) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(hdr) < 0) + H5E_THROW(H5E_CANTREMOVE, "unable to remove extensible array header from cache") + /* Release header's disk space */ if(H5F_addr_defined(hdr->addr) && H5MF_xfree(f, H5FD_MEM_EARRAY_HDR, dxpl_id, hdr->addr, (hsize_t)hdr->size) < 0) H5E_THROW(H5E_CANTFREE, "unable to free extensible array header") @@ -614,6 +632,7 @@ H5EA__hdr_protect(H5F_t *f, hid_t dxpl_id, haddr_t ea_addr, void *ctx_udata, unsigned flags)) /* Local variables */ + H5EA_hdr_t *hdr; /* Extensible array header */ H5EA_hdr_cache_ud_t udata; /* User data for cache callbacks */ /* Sanity check */ @@ -629,9 +648,23 @@ H5EA__hdr_protect(H5F_t *f, hid_t dxpl_id, haddr_t ea_addr, void *ctx_udata, udata.ctx_udata = ctx_udata; /* Protect the header */ - if(NULL == (ret_value = (H5EA_hdr_t *)H5AC_protect(f, dxpl_id, H5AC_EARRAY_HDR, ea_addr, &udata, flags))) + if(NULL == (hdr = (H5EA_hdr_t *)H5AC_protect(f, dxpl_id, H5AC_EARRAY_HDR, ea_addr, &udata, flags))) H5E_THROW(H5E_CANTPROTECT, "unable to protect extensible array header, address = %llu", (unsigned long long)ea_addr) - ret_value->f = f; /* (Must be set again here, in case the header was already in the cache -QAK) */ + hdr->f = f; /* (Must be set again here, in case the header was already in the cache -QAK) */ + + /* Create top proxy, if it doesn't exist */ + if(hdr->swmr_write && NULL == hdr->top_proxy) { + /* Create 'top' proxy for extensible array entries */ + if(NULL == (hdr->top_proxy = H5AC_proxy_entry_create())) + H5E_THROW(H5E_CANTCREATE, "can't create extensible array entry proxy") + + /* Add header as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, f, dxpl_id, hdr) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + } /* end if */ + + /* Set return value */ + ret_value = hdr; CATCH @@ -779,6 +812,13 @@ H5EA__hdr_dest(H5EA_hdr_t *hdr)) if(hdr->sblk_info) hdr->sblk_info = (H5EA_sblk_info_t *)H5FL_SEQ_FREE(H5EA_sblk_info_t, hdr->sblk_info); + /* Destroy the 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_dest(hdr->top_proxy) < 0) + H5E_THROW(H5E_CANTRELEASE, "unable to destroy extensible array 'top' proxy") + hdr->top_proxy = NULL; + } /* end if */ + /* Free the shared info itself */ hdr = H5FL_FREE(H5EA_hdr_t, hdr); diff --git a/src/H5EAiblock.c b/src/H5EAiblock.c index 48b52a4..a3723c5 100644 --- a/src/H5EAiblock.c +++ b/src/H5EAiblock.c @@ -183,6 +183,7 @@ H5EA__iblock_create(H5EA_hdr_t *hdr, hid_t dxpl_id, hbool_t *stats_changed)) /* Local variables */ H5EA_iblock_t *iblock = NULL; /* Extensible array index block */ haddr_t iblock_addr; /* Extensible array index block address */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ #ifdef QAK HDfprintf(stderr, "%s: Called\n", FUNC); @@ -233,6 +234,14 @@ HDfprintf(stderr, "%s: iblock->size = %Zu\n", FUNC, iblock->size); /* Cache the new extensible array index block */ if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_EARRAY_IBLOCK, iblock_addr, iblock, H5AC__NO_FLAGS_SET) < 0) H5E_THROW(H5E_CANTINSERT, "can't add extensible array index block to cache") + inserted = TRUE; + + /* Add index block as child of 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, iblock) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + iblock->top_proxy = hdr->top_proxy; + } /* end if */ /* Update extensible array index block statistics */ HDassert(0 == hdr->stats.computed.nindex_blks); @@ -252,6 +261,11 @@ HDfprintf(stderr, "%s: iblock->size = %Zu\n", FUNC, iblock->size); CATCH if(!H5F_addr_defined(ret_value)) if(iblock) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(iblock) < 0) + H5E_THROW(H5E_CANTREMOVE, "unable to remove extensible array index block from cache") + /* Release index block's disk space */ if(H5F_addr_defined(iblock->addr) && H5MF_xfree(hdr->f, H5FD_MEM_EARRAY_IBLOCK, dxpl_id, iblock->addr, (hsize_t)iblock->size) < 0) H5E_THROW(H5E_CANTFREE, "unable to release file space for extensible array index block") @@ -281,6 +295,9 @@ BEGIN_FUNC(PKG, ERR, H5EA_iblock_t *, NULL, NULL, H5EA__iblock_protect(H5EA_hdr_t *hdr, hid_t dxpl_id, unsigned flags)) + /* Local variables */ + H5EA_iblock_t *iblock = NULL; /* Pointer to index block */ + #ifdef QAK HDfprintf(stderr, "%s: Called\n", FUNC); #endif /* QAK */ @@ -292,10 +309,27 @@ HDfprintf(stderr, "%s: Called\n", FUNC); HDassert((flags & (unsigned)(~H5AC__READ_ONLY_FLAG)) == 0); /* Protect the index block */ - if(NULL == (ret_value = (H5EA_iblock_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_EARRAY_IBLOCK, hdr->idx_blk_addr, hdr, flags))) + if(NULL == (iblock = (H5EA_iblock_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_EARRAY_IBLOCK, hdr->idx_blk_addr, hdr, flags))) H5E_THROW(H5E_CANTPROTECT, "unable to protect extensible array index block, address = %llu", (unsigned long long)hdr->idx_blk_addr) + /* Create top proxy, if it doesn't exist */ + if(hdr->top_proxy && NULL == iblock->top_proxy) { + /* Add index block as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, iblock) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + iblock->top_proxy = hdr->top_proxy; + } /* end if */ + + /* Set return value */ + ret_value = iblock; + CATCH + /* Clean up on error */ + if(!ret_value) { + /* Release the index block, if it was protected */ + if(iblock && H5AC_unprotect(hdr->f, dxpl_id, H5AC_EARRAY_IBLOCK, iblock->addr, iblock, H5AC__NO_FLAGS_SET) < 0) + H5E_THROW(H5E_CANTUNPROTECT, "unable to unprotect extensible array index block, address = %llu", (unsigned long long)iblock->addr) + } /* end if */ END_FUNC(PKG) /* end H5EA__iblock_protect() */ @@ -470,6 +504,9 @@ H5EA__iblock_dest(H5EA_iblock_t *iblock)) iblock->hdr = NULL; } /* end if */ + /* Sanity check */ + HDassert(NULL == iblock->top_proxy); + /* Free the index block itself */ iblock = H5FL_FREE(H5EA_iblock_t, iblock); diff --git a/src/H5EApkg.h b/src/H5EApkg.h index 70e6a55..093403c 100644 --- a/src/H5EApkg.h +++ b/src/H5EApkg.h @@ -206,6 +206,28 @@ typedef struct H5EA_hdr_t { /* Client information (not stored) */ void *cb_ctx; /* Callback context */ + + /* SWMR / Flush dependency information (not stored) */ + hbool_t swmr_write; /* Flag indicating the file is opened with SWMR-write access */ + H5AC_proxy_entry_t *top_proxy; /* 'Top' proxy cache entry for all array entries */ + void *parent; /* Pointer to 'top' proxy flush dependency + * parent, if it exists, otherwise NULL. + * If the extensible array is being used + * to index a chunked dataset and the + * dataset metadata is modified by a + * SWMR writer, this field will be set + * equal to the object header proxy + * that is the flush dependency parent + * of the extensible array header. + * + * The field is used to avoid duplicate + * setups of the flush dependency + * relationship, and to allow the + * extensible array header to destroy + * the flush dependency on receipt of + * an eviction notification from the + * metadata cache. + */ } H5EA_hdr_t; /* The extensible array index block information */ @@ -223,6 +245,9 @@ typedef struct H5EA_iblock_t { haddr_t addr; /* Address of this index block on disk */ size_t size; /* Size of index block on disk */ + /* SWMR / Flush dependency information (not stored) */ + H5AC_proxy_entry_t *top_proxy; /* "Top" proxy cache entry for all array entries */ + /* Computed/cached values (not stored) */ size_t nsblks; /* # of super blocks whose data block addresses are in index block */ size_t ndblk_addrs; /* Number of pointers to data blocks in index block */ @@ -244,8 +269,9 @@ typedef struct H5EA_sblock_t { haddr_t addr; /* Address of this index block on disk */ size_t size; /* Size of index block on disk */ - /* Flush dependency information (not stored) */ + /* SWMR / Flush dependency information (not stored) */ hbool_t has_hdr_depend; /* Whether this object has a flush dependency on the header */ + H5AC_proxy_entry_t *top_proxy; /* "Top" proxy cache entry for all array entries */ H5EA_iblock_t *parent; /* Parent object for super block (index block) */ /* Computed/cached values (not stored) */ @@ -271,8 +297,9 @@ typedef struct H5EA_dblock_t { haddr_t addr; /* Address of this data block on disk */ size_t size; /* Size of data block on disk */ - /* Flush dependency information (not stored) */ + /* SWMR / Flush dependency information (not stored) */ hbool_t has_hdr_depend; /* Whether this object has a flush dependency on the header */ + H5AC_proxy_entry_t *top_proxy; /* 'Top' proxy cache entry for all array entries */ void *parent; /* Parent object for data block (index or super block) */ /* Computed/cached values (not stored) */ @@ -293,8 +320,9 @@ typedef struct H5EA_dbk_page_t { haddr_t addr; /* Address of this data block page on disk */ size_t size; /* Size of data block page on disk */ - /* Flush dependency information (not stored) */ + /* SWMR / Flush dependency information (not stored) */ hbool_t has_hdr_depend; /* Whether this object has a flush dependency on the header */ + H5AC_proxy_entry_t *top_proxy; /* "Top" proxy cache entry for all array entries */ H5EA_sblock_t *parent; /* Parent object for data block page (super block) */ /* Computed/cached values (not stored) */ diff --git a/src/H5EAprivate.h b/src/H5EAprivate.h index 0a1b945..66d88e0 100644 --- a/src/H5EAprivate.h +++ b/src/H5EAprivate.h @@ -142,7 +142,7 @@ H5_DLL herr_t H5EA_get_nelmts(const H5EA_t *ea, hsize_t *nelmts); H5_DLL herr_t H5EA_get_addr(const H5EA_t *ea, haddr_t *addr); H5_DLL herr_t H5EA_set(const H5EA_t *ea, hid_t dxpl_id, hsize_t idx, const void *elmt); H5_DLL herr_t H5EA_get(const H5EA_t *ea, hid_t dxpl_id, hsize_t idx, void *elmt); -H5_DLL herr_t H5EA_depend(H5AC_info_t *parent_entry, H5EA_t *ea); +H5_DLL herr_t H5EA_depend(H5EA_t *ea, hid_t dxpl_id, H5AC_proxy_entry_t *parent); H5_DLL herr_t H5EA_iterate(H5EA_t *fa, hid_t dxpl_id, H5EA_operator_t op, void *udata); H5_DLL herr_t H5EA_close(H5EA_t *ea, hid_t dxpl_id); H5_DLL herr_t H5EA_delete(H5F_t *f, hid_t dxpl_id, haddr_t ea_addr, void *ctx_udata); diff --git a/src/H5EAsblock.c b/src/H5EAsblock.c index 4f153cd..2fa87e0 100644 --- a/src/H5EAsblock.c +++ b/src/H5EAsblock.c @@ -195,6 +195,7 @@ H5EA__sblock_create(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_iblock_t *parent, H5EA_sblock_t *sblock = NULL; /* Extensible array super block */ haddr_t sblock_addr; /* Extensible array super block address */ haddr_t tmp_addr = HADDR_UNDEF; /* Address value to fill data block addresses with */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ /* Sanity check */ HDassert(hdr); @@ -221,6 +222,14 @@ H5EA__sblock_create(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_iblock_t *parent, /* Cache the new extensible array super block */ if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_EARRAY_SBLOCK, sblock_addr, sblock, H5AC__NO_FLAGS_SET) < 0) H5E_THROW(H5E_CANTINSERT, "can't add extensible array super block to cache") + inserted = TRUE; + + /* Add super block as child of 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, sblock) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + sblock->top_proxy = hdr->top_proxy; + } /* end if */ /* Update extensible array super block statistics */ hdr->stats.stored.nsuper_blks++; @@ -235,6 +244,11 @@ H5EA__sblock_create(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_iblock_t *parent, CATCH if(!H5F_addr_defined(ret_value)) if(sblock) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(sblock) < 0) + H5E_THROW(H5E_CANTREMOVE, "unable to remove extensible array super block from cache") + /* Release super block's disk space */ if(H5F_addr_defined(sblock->addr) && H5MF_xfree(hdr->f, H5FD_MEM_EARRAY_SBLOCK, dxpl_id, sblock->addr, (hsize_t)sblock->size) < 0) H5E_THROW(H5E_CANTFREE, "unable to release extensible array super block") @@ -266,6 +280,7 @@ H5EA__sblock_protect(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_iblock_t *parent, haddr_t sblk_addr, unsigned sblk_idx, unsigned flags)) /* Local variables */ + H5EA_sblock_t *sblock = NULL; /* Pointer to super block */ H5EA_sblock_cache_ud_t udata; /* Information needed for loading super block */ /* Sanity check */ @@ -282,10 +297,27 @@ H5EA__sblock_protect(H5EA_hdr_t *hdr, hid_t dxpl_id, H5EA_iblock_t *parent, udata.sblk_addr = sblk_addr; /* Protect the super block */ - if(NULL == (ret_value = (H5EA_sblock_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_EARRAY_SBLOCK, sblk_addr, &udata, flags))) + if(NULL == (sblock = (H5EA_sblock_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_EARRAY_SBLOCK, sblk_addr, &udata, flags))) H5E_THROW(H5E_CANTPROTECT, "unable to protect extensible array super block, address = %llu", (unsigned long long)sblk_addr) + /* Create top proxy, if it doesn't exist */ + if(hdr->top_proxy && NULL == sblock->top_proxy) { + /* Add super block as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, sblock) < 0) + H5E_THROW(H5E_CANTSET, "unable to add extensible array entry as child of array proxy") + sblock->top_proxy = hdr->top_proxy; + } /* end if */ + + /* Set return value */ + ret_value = sblock; + CATCH + /* Clean up on error */ + if(!ret_value) { + /* Release the super block, if it was protected */ + if(sblock && H5AC_unprotect(hdr->f, dxpl_id, H5AC_EARRAY_SBLOCK, sblock->addr, sblock, H5AC__NO_FLAGS_SET) < 0) + H5E_THROW(H5E_CANTUNPROTECT, "unable to unprotect extensible array super block, address = %llu", (unsigned long long)sblock->addr) + } /* end if */ END_FUNC(PKG) /* end H5EA__sblock_protect() */ @@ -409,6 +441,9 @@ H5EA__sblock_dest(H5EA_sblock_t *sblock)) sblock->hdr = NULL; } /* end if */ + /* Sanity check */ + HDassert(NULL == sblock->top_proxy); + /* Free the super block itself */ sblock = H5FL_FREE(H5EA_sblock_t, sblock); @@ -456,9 +456,9 @@ H5Fcreate(const char *filename, unsigned flags, hid_t fcpl_id, hid_t fapl_id) if(!filename || !*filename) HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "invalid file name") /* In this routine, we only accept the following flags: - * H5F_ACC_EXCL and H5F_ACC_TRUNC + * H5F_ACC_EXCL, H5F_ACC_TRUNC and H5F_ACC_SWMR_WRITE */ - if(flags & ~(H5F_ACC_EXCL | H5F_ACC_TRUNC)) + if(flags & ~(H5F_ACC_EXCL | H5F_ACC_TRUNC | H5F_ACC_SWMR_WRITE)) HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "invalid flags") /* The H5F_ACC_EXCL and H5F_ACC_TRUNC flags are mutually exclusive */ if((flags & H5F_ACC_EXCL) && (flags & H5F_ACC_TRUNC)) @@ -563,6 +563,12 @@ H5Fopen(const char *filename, unsigned flags, hid_t fapl_id) if((flags & ~H5F_ACC_PUBLIC_FLAGS) || (flags & H5F_ACC_TRUNC) || (flags & H5F_ACC_EXCL)) HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "invalid file open flags") + /* Asking for SWMR write access on a read-only file is invalid */ + if((flags & H5F_ACC_SWMR_WRITE) && 0 == (flags & H5F_ACC_RDWR)) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, FAIL, "SWMR write access on a file open for read-only access is not allowed") + /* Asking for SWMR read access on a non-read-only file is invalid */ + if((flags & H5F_ACC_SWMR_READ) && (flags & H5F_ACC_RDWR)) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, FAIL, "SWMR read access on a file open for read-write access is not allowed") /* Verify access property list and get correct dxpl */ if(H5P_verify_apl_and_dxpl(&fapl_id, H5P_CLS_FACC, &dxpl_id, H5I_INVALID_HID, TRUE) < 0) @@ -857,10 +863,20 @@ H5Fget_intent(hid_t file_id, unsigned *intent_flags) * Simplify things for them so that they only get either H5F_ACC_RDWR * or H5F_ACC_RDONLY. */ - if(H5F_INTENT(file) & H5F_ACC_RDWR) + if(H5F_INTENT(file) & H5F_ACC_RDWR) { *intent_flags = H5F_ACC_RDWR; - else + + /* Check for SWMR write access on the file */ + if(H5F_INTENT(file) & H5F_ACC_SWMR_WRITE) + *intent_flags |= H5F_ACC_SWMR_WRITE; + } /* end if */ + else { *intent_flags = H5F_ACC_RDONLY; + + /* Check for SWMR read access on the file */ + if(H5F_INTENT(file) & H5F_ACC_SWMR_READ) + *intent_flags |= H5F_ACC_SWMR_READ; + } /* end else */ } /* end if */ done: @@ -1541,6 +1557,222 @@ done: /*------------------------------------------------------------------------- + * Function: H5Fstart_swmr_write + * + * Purpose: To enable SWMR writing mode for the file + * 1) Refresh opened objects: part 1 + * 2) Flush & reset accumulator + * 3) Mark the file in SWMR writing mode + * 4) Set metadata read attempts and retries info + * 5) Disable accumulator + * 6) Evict all cache entries except the superblock + * 7) Refresh opened objects (part 2) + * 8) Unlock the file + * + * Pre-conditions: + * 1) The file being opened has v3 superblock + * 2) The file is opened with H5F_ACC_RDWR + * 3) The file is not already marked for SWMR writing + * 4) Current implementaion for opened objects: + * --only allow datasets and groups without attributes + * --disallow named datatype with/without attributes + * --disallow opened attributes attached to objects + * NOTE: Currently, only opened groups and datasets are allowed + * when enabling SWMR via H5Fstart_swmr_write(). + * Will later implement a different approach-- + * set up flush dependency/proxy even for file opened without + * SWMR to resolve issues with opened objects. + * + * Return: Non-negative on success/negative on failure + * + * Programmer: + * Vailin Choi; Feb 2014 + * + *------------------------------------------------------------------------- + */ +herr_t +H5Fstart_swmr_write(hid_t file_id) +{ + H5F_t *file = NULL; /* File info */ + size_t grp_dset_count=0; /* # of open objects: groups & datasets */ + size_t nt_attr_count=0; /* # of opened named datatypes + opened attributes */ + hid_t *obj_ids=NULL; /* List of ids */ + H5G_loc_t *obj_glocs=NULL; /* Group location of the object */ + H5O_loc_t *obj_olocs=NULL; /* Object location */ + H5G_name_t *obj_paths=NULL; /* Group hierarchy path */ + size_t u; /* Local index variable */ + hbool_t setup = FALSE; /* Boolean flag to indicate whether SWMR setting is enabled */ + H5F_io_info_t fio_info; /* I/O info for operation */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_API(FAIL) + H5TRACE1("e", "i", file_id); + + /* check args */ + if(NULL == (file = (H5F_t *)H5I_object_verify(file_id, H5I_FILE))) + HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file") + + /* Should have write permission */ + if((H5F_INTENT(file) & H5F_ACC_RDWR) == 0) + HGOTO_ERROR(H5E_FILE, H5E_BADVALUE, FAIL, "no write intent on file") + + if(file->shared->sblock->super_vers < HDF5_SUPERBLOCK_VERSION_3) + HGOTO_ERROR(H5E_FILE, H5E_BADVALUE, FAIL, "file superblock version should be at least 3") + HDassert(file->shared->latest_flags == H5F_LATEST_ALL_FLAGS); + + /* Should not be marked for SWMR writing mode already */ + if(file->shared->sblock->status_flags & H5F_SUPER_SWMR_WRITE_ACCESS) + HGOTO_ERROR(H5E_FILE, H5E_BADVALUE, FAIL, "file already in SWMR writing mode") + + HDassert(file->shared->sblock->status_flags & H5F_SUPER_WRITE_ACCESS); + + /* Flush data buffers */ + if(H5F_flush(file, H5AC_ind_read_dxpl_id, FALSE) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTFLUSH, FAIL, "unable to flush file's cached information") + + /* Get the # of opened named datatypes and attributes */ + if(H5F_get_obj_count(file, H5F_OBJ_DATATYPE|H5F_OBJ_ATTR, FALSE, &nt_attr_count) < 0) + HGOTO_ERROR(H5E_INTERNAL, H5E_BADITER, FAIL, "H5F_get_obj_count failed") + if(nt_attr_count) + HGOTO_ERROR(H5E_FILE, H5E_BADVALUE, FAIL, "named datatypes and/or attributes opened in the file") + + /* Get the # of opened datasets and groups */ + if(H5F_get_obj_count(file, H5F_OBJ_GROUP|H5F_OBJ_DATASET, FALSE, &grp_dset_count) < 0) + HGOTO_ERROR(H5E_INTERNAL, H5E_BADITER, FAIL, "H5F_get_obj_count failed") + + if(grp_dset_count) { + /* Allocate space for group and object locations */ + if((obj_ids = (hid_t *) H5MM_malloc(grp_dset_count * sizeof(hid_t))) == NULL) + HGOTO_ERROR(H5E_FILE, H5E_NOSPACE, FAIL, "can't allocate buffer for hid_t") + if((obj_glocs = (H5G_loc_t *) H5MM_malloc(grp_dset_count * sizeof(H5G_loc_t))) == NULL) + HGOTO_ERROR(H5E_FILE, H5E_NOSPACE, FAIL, "can't allocate buffer for H5G_loc_t") + if((obj_olocs = (H5O_loc_t *) H5MM_malloc(grp_dset_count * sizeof(H5O_loc_t))) == NULL) + HGOTO_ERROR(H5E_FILE, H5E_NOSPACE, FAIL, "can't allocate buffer for H5O_loc_t") + if((obj_paths = (H5G_name_t *) H5MM_malloc(grp_dset_count * sizeof(H5G_name_t))) == NULL) + HGOTO_ERROR(H5E_FILE, H5E_NOSPACE, FAIL, "can't allocate buffer for H5G_name_t") + + /* Get the list of opened object ids (groups & datasets) */ + if(H5F_get_obj_ids(file, H5F_OBJ_GROUP|H5F_OBJ_DATASET, grp_dset_count, obj_ids, FALSE, &grp_dset_count) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTGET, FAIL, "H5F_get_obj_ids failed") + + /* Refresh opened objects (groups, datasets) in the file */ + for(u = 0; u < grp_dset_count; u++) { + H5O_loc_t *oloc; /* object location */ + H5G_loc_t tmp_loc; + + /* Set up the id's group location */ + obj_glocs[u].oloc = &obj_olocs[u]; + obj_glocs[u].path = &obj_paths[u]; + H5G_loc_reset(&obj_glocs[u]); + + /* get the id's object location */ + if((oloc = H5O_get_loc(obj_ids[u])) == NULL) + HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not an object") + + /* Make deep local copy of object's location information */ + H5G_loc(obj_ids[u], &tmp_loc); + H5G_loc_copy(&obj_glocs[u], &tmp_loc, H5_COPY_DEEP); + + /* Close the object */ + if(H5I_dec_ref(obj_ids[u]) < 0) + HGOTO_ERROR(H5E_ATOM, H5E_CANTCLOSEOBJ, FAIL, "decrementing object ID failed") + } /* end for */ + } /* end if */ + + /* Set up I/O info for operation */ + fio_info.f = file; + if(NULL == (fio_info.dxpl = (H5P_genplist_t *)H5I_object(H5AC_ind_read_dxpl_id))) + HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "can't get property list") + + /* Flush and reset the accumulator */ + if(H5F__accum_reset(&fio_info, TRUE) < 0) + HGOTO_ERROR(H5E_IO, H5E_CANTRESET, FAIL, "can't reset accumulator") + + /* Turn on SWMR write in shared file open flags */ + file->shared->flags |= H5F_ACC_SWMR_WRITE; + + /* Mark the file in SWMR writing mode */ + file->shared->sblock->status_flags |= H5F_SUPER_SWMR_WRITE_ACCESS; + + /* Set up metadata read attempts */ + file->shared->read_attempts = H5F_SWMR_METADATA_READ_ATTEMPTS; + + /* Initialize "retries" and "retries_nbins" */ + if(H5F_set_retries(file) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTINIT, FAIL, "can't set retries and retries_nbins") + + /* Turn off usage of accumulator */ + file->shared->feature_flags &= ~(unsigned)H5FD_FEAT_ACCUMULATE_METADATA; + if(H5FD_set_feature_flags(file->shared->lf, file->shared->feature_flags) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTSET, FAIL, "can't set feature_flags in VFD") + + setup = TRUE; + + /* Mark superblock as dirty */ + if(H5F_super_dirty(file) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTMARKDIRTY, FAIL, "unable to mark superblock as dirty") + + /* Flush the superblock */ + if(H5F_flush_tagged_metadata(file, H5AC__SUPERBLOCK_TAG, H5AC_ind_read_dxpl_id) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTFLUSH, FAIL, "unable to flush superblock") + + /* Evict all flushed entries in the cache except the pinned superblock */ + if(H5F__evict_cache_entries(file, H5AC_ind_read_dxpl_id) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTFLUSH, FAIL, "unable to evict file's cached information") + + /* Refresh (reopen) the objects (groups & datasets) in the file */ + for(u = 0; u < grp_dset_count; u++) + if(H5O_refresh_metadata_reopen(obj_ids[u], &obj_glocs[u], H5AC_ind_read_dxpl_id, TRUE) < 0) + HGOTO_ERROR(H5E_ATOM, H5E_CLOSEERROR, FAIL, "can't refresh-close object") + + /* Unlock the file */ + if(H5FD_unlock(file->shared->lf) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, FAIL, "unable to unlock the file") + +done: + if(ret_value < 0 && setup) { + HDassert(file); + + /* Re-enable accumulator */ + file->shared->feature_flags |= (unsigned)H5FD_FEAT_ACCUMULATE_METADATA; + if(H5FD_set_feature_flags(file->shared->lf, file->shared->feature_flags) < 0) + HDONE_ERROR(H5E_FILE, H5E_CANTSET, FAIL, "can't set feature_flags in VFD") + + /* Reset the # of read attempts */ + file->shared->read_attempts = H5F_METADATA_READ_ATTEMPTS; + if(H5F_set_retries(file) < 0) + HDONE_ERROR(H5E_FILE, H5E_CANTINIT, FAIL, "can't set retries and retries_nbins") + + /* Un-set H5F_ACC_SWMR_WRITE in shared open flags */ + file->shared->flags &= ~H5F_ACC_SWMR_WRITE; + + /* Unmark the file: not in SWMR writing mode */ + file->shared->sblock->status_flags &= (uint8_t)(~H5F_SUPER_SWMR_WRITE_ACCESS); + + /* Mark superblock as dirty */ + if(H5F_super_dirty(file) < 0) + HDONE_ERROR(H5E_FILE, H5E_CANTMARKDIRTY, FAIL, "unable to mark superblock as dirty") + + /* Flush the superblock */ + if(H5F_flush_tagged_metadata(file, H5AC__SUPERBLOCK_TAG, H5AC_ind_read_dxpl_id) < 0) + HDONE_ERROR(H5E_FILE, H5E_CANTFLUSH, FAIL, "unable to flush superblock") + } /* end if */ + + /* Free memory */ + if(obj_ids) + H5MM_xfree(obj_ids); + if(obj_glocs) + H5MM_xfree(obj_glocs); + if(obj_olocs) + H5MM_xfree(obj_olocs); + if(obj_paths) + H5MM_xfree(obj_paths); + + FUNC_LEAVE_API(ret_value) +} /* end H5Fstart_swmr_write() */ + + +/*------------------------------------------------------------------------- * Function: H5Fstart_mdc_logging * * Purpose: Start metadata cache logging operations for a file. @@ -737,6 +737,56 @@ END_FUNC(PRIV) /* end H5FA_iterate() */ /*------------------------------------------------------------------------- + * Function: H5FA_depend + * + * Purpose: Make a child flush dependency between the fixed array + * and another piece of metadata in the file. + * + * Return: SUCCEED/FAIL + * + * Programmer: Dana Robinson + * Fall 2012 + * + *------------------------------------------------------------------------- + */ +BEGIN_FUNC(PRIV, ERR, +herr_t, SUCCEED, FAIL, +H5FA_depend(H5FA_t *fa, hid_t dxpl_id, H5AC_proxy_entry_t *parent)) + + /* Local variables */ + H5FA_hdr_t *hdr = fa->hdr; /* Header for FA */ + + /* + * Check arguments. + */ + HDassert(fa); + HDassert(hdr); + HDassert(parent); + + /* + * Check to see if a flush dependency between the fixed array + * and another data structure in the file has already been set up. + * If it hasn't, do so now. + */ + if(NULL == hdr->parent) { + /* Sanity check */ + HDassert(hdr->top_proxy); + + /* Set the shared array header's file context for this operation */ + hdr->f = fa->f; + + /* Add the fixed array as a child of the parent (proxy) */ + if(H5AC_proxy_entry_add_child(parent, hdr->f, dxpl_id, hdr->top_proxy) < 0) + H5E_THROW(H5E_CANTSET, "unable to add fixed array as child of proxy") + hdr->parent = parent; + } /* end if */ + +CATCH + +END_FUNC(PRIV) /* end H5FA_depend() */ + + +/*------------------------------------------------------------------------- * Function: H5FA_patch_file * * Purpose: Patch the top-level file pointer contained in fa diff --git a/src/H5FAcache.c b/src/H5FAcache.c index 82b87dd..11e8571 100644 --- a/src/H5FAcache.c +++ b/src/H5FAcache.c @@ -78,6 +78,7 @@ static void *H5FA__cache_hdr_deserialize(const void *image, size_t len, static herr_t H5FA__cache_hdr_image_len(const void *thing, size_t *image_len); static herr_t H5FA__cache_hdr_serialize(const H5F_t *f, void *image, size_t len, void *thing); +static herr_t H5FA__cache_hdr_notify(H5AC_notify_action_t action, void *thing); static herr_t H5FA__cache_hdr_free_icr(void *thing); static herr_t H5FA__cache_dblock_get_initial_load_size(void *udata, size_t *image_len); @@ -87,6 +88,7 @@ static void *H5FA__cache_dblock_deserialize(const void *image, size_t len, static herr_t H5FA__cache_dblock_image_len(const void *thing, size_t *image_len); static herr_t H5FA__cache_dblock_serialize(const H5F_t *f, void *image, size_t len, void *thing); +static herr_t H5FA__cache_dblock_notify(H5AC_notify_action_t action, void *thing); static herr_t H5FA__cache_dblock_free_icr(void *thing); static herr_t H5FA__cache_dblock_fsf_size(const void *thing, size_t *fsf_size); @@ -97,6 +99,7 @@ static void *H5FA__cache_dblk_page_deserialize(const void *image, size_t len, static herr_t H5FA__cache_dblk_page_image_len(const void *thing, size_t *image_len); static herr_t H5FA__cache_dblk_page_serialize(const H5F_t *f, void *image, size_t len, void *thing); +static herr_t H5FA__cache_dblk_page_notify(H5AC_notify_action_t action, void *thing); static herr_t H5FA__cache_dblk_page_free_icr(void *thing); @@ -117,7 +120,7 @@ const H5AC_class_t H5AC_FARRAY_HDR[1] = {{ H5FA__cache_hdr_image_len, /* 'image_len' callback */ NULL, /* 'pre_serialize' callback */ H5FA__cache_hdr_serialize, /* 'serialize' callback */ - NULL, /* 'notify' callback */ + H5FA__cache_hdr_notify, /* 'notify' callback */ H5FA__cache_hdr_free_icr, /* 'free_icr' callback */ NULL, /* 'fsf_size' callback */ }}; @@ -135,7 +138,7 @@ const H5AC_class_t H5AC_FARRAY_DBLOCK[1] = {{ H5FA__cache_dblock_image_len, /* 'image_len' callback */ NULL, /* 'pre_serialize' callback */ H5FA__cache_dblock_serialize, /* 'serialize' callback */ - NULL, /* 'notify' callback */ + H5FA__cache_dblock_notify, /* 'notify' callback */ H5FA__cache_dblock_free_icr, /* 'free_icr' callback */ H5FA__cache_dblock_fsf_size, /* 'fsf_size' callback */ }}; @@ -153,7 +156,7 @@ const H5AC_class_t H5AC_FARRAY_DBLK_PAGE[1] = {{ H5FA__cache_dblk_page_image_len, /* 'image_len' callback */ NULL, /* 'pre_serialize' callback */ H5FA__cache_dblk_page_serialize, /* 'serialize' callback */ - NULL, /* 'notify' callback */ + H5FA__cache_dblk_page_notify, /* 'notify' callback */ H5FA__cache_dblk_page_free_icr, /* 'free_icr' callback */ NULL, /* 'fsf_size' callback */ }}; @@ -440,6 +443,80 @@ END_FUNC(STATIC) /* end H5FA__cache_hdr_serialize() */ /*------------------------------------------------------------------------- + * Function: H5FA__cache_hdr_notify + * + * Purpose: Handle cache action notifications + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Dana Robinson + * December 2015 + * + *------------------------------------------------------------------------- + */ +BEGIN_FUNC(STATIC, ERR, +herr_t, SUCCEED, FAIL, +H5FA__cache_hdr_notify(H5AC_notify_action_t action, void *_thing)) + + /* Local variables */ + H5FA_hdr_t *hdr = (H5FA_hdr_t *)_thing; /* Pointer to the object */ + + /* Sanity check */ + HDassert(hdr); + + /* Check if the file was opened with SWMR-write access */ + if(hdr->swmr_write) { + /* Determine which action to take */ + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + case H5AC_NOTIFY_ACTION_AFTER_LOAD: + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* If hdr->parent != NULL, hdr->parent is used to destroy + * the flush dependency before the header is evicted. + */ + if(hdr->parent) { + /* Sanity check */ + HDassert(hdr->top_proxy); + + /* Destroy flush dependency on object header proxy */ + if(H5AC_proxy_entry_remove_child((H5AC_proxy_entry_t *)hdr->parent, (void *)hdr->top_proxy) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between fixed array and proxy") + hdr->parent = NULL; + } /* end if */ + + /* Detach from 'top' proxy for fixed array */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_remove_child(hdr->top_proxy, hdr) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between header and fixed array 'top' proxy") + /* Don't reset hdr->top_proxy here, it's destroyed when the header is freed -QAK */ + } /* end if */ + break; + + default: +#ifdef NDEBUG + H5E_THROW(H5E_BADVALUE, "unknown action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + } /* end if */ + else + HDassert(NULL == hdr->parent); + +CATCH + +END_FUNC(STATIC) /* end H5FA__cache_hdr_notify() */ + + +/*------------------------------------------------------------------------- * Function: H5FA__cache_hdr_free_icr * * Purpose: Destroy/release an "in core representation" of a data @@ -765,6 +842,75 @@ END_FUNC(STATIC) /* end H5FA__cache_dblock_serialize() */ /*------------------------------------------------------------------------- + * Function: H5FA__cache_dblock_notify + * + * Purpose: Handle cache action notifications + * + * Return: SUCCEED/FAIL + * + * Programmer: Dana Robinson + * Fall 2012 + * + *------------------------------------------------------------------------- + */ +BEGIN_FUNC(STATIC, ERR, +herr_t, SUCCEED, FAIL, +H5FA__cache_dblock_notify(H5AC_notify_action_t action, void *_thing)) + + /* Local variables */ + H5FA_dblock_t *dblock = (H5FA_dblock_t *)_thing; + + /* Sanity check */ + HDassert(dblock); + + /* Check if the file was opened with SWMR-write access */ + if(dblock->hdr->swmr_write) { + /* Determine which action to take */ + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + case H5AC_NOTIFY_ACTION_AFTER_LOAD: + /* Create flush dependency on parent */ + if(H5FA__create_flush_depend((H5AC_info_t *)dblock->hdr, (H5AC_info_t *)dblock) < 0) + H5E_THROW(H5E_CANTDEPEND, "unable to create flush dependency between data block and header, address = %llu", (unsigned long long)dblock->addr) + break; + + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* Destroy flush dependency on parent */ + if(H5FA__destroy_flush_depend((H5AC_info_t *)dblock->hdr, (H5AC_info_t *)dblock) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency") + + /* Detach from 'top' proxy for fixed array */ + if(dblock->top_proxy) { + if(H5AC_proxy_entry_remove_child(dblock->top_proxy, dblock) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between data block and fixed array 'top' proxy") + dblock->top_proxy = NULL; + } /* end if */ + break; + + default: +#ifdef NDEBUG + H5E_THROW(H5E_BADVALUE, "unknown action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + } /* end if */ + +CATCH + +END_FUNC(STATIC) /* end H5FA__cache_dblock_notify() */ + + + +/*------------------------------------------------------------------------- * Function: H5FA__cache_dblock_free_icr * * Purpose: Destroy/release an "in core representation" of a data @@ -1065,6 +1211,66 @@ END_FUNC(STATIC) /* end H5FA__cache_dblk_page_serialize() */ /*------------------------------------------------------------------------- + * Function: H5FA__cache_dblk_page_notify + * + * Purpose: Handle cache action notifications + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * koziol@lbl.gov + * Oct 17 2016 + * + *------------------------------------------------------------------------- + */ +BEGIN_FUNC(STATIC, ERR, +herr_t, SUCCEED, FAIL, +H5FA__cache_dblk_page_notify(H5AC_notify_action_t action, void *_thing)) + + /* Local variables */ + H5FA_dblk_page_t *dblk_page = (H5FA_dblk_page_t *)_thing; /* Pointer to the object */ + + /* Sanity check */ + HDassert(dblk_page); + + /* Determine which action to take */ + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + case H5AC_NOTIFY_ACTION_AFTER_LOAD: + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* Detach from 'top' proxy for fixed array */ + if(dblk_page->top_proxy) { + if(H5AC_proxy_entry_remove_child(dblk_page->top_proxy, dblk_page) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency between data block page and fixed array 'top' proxy") + dblk_page->top_proxy = NULL; + } /* end if */ + break; + + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* do nothing */ + break; + + default: +#ifdef NDEBUG + H5E_THROW(H5E_BADVALUE, "unknown action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + +CATCH + +END_FUNC(STATIC) /* end H5FA__cache_dblk_page_notify() */ + + +/*------------------------------------------------------------------------- * Function: H5FA__cache_dblk_page_free_icr * * Purpose: Destroy/release an "in core representation" of a data diff --git a/src/H5FAdblkpage.c b/src/H5FAdblkpage.c index 054ffb9..09278f8 100644 --- a/src/H5FAdblkpage.c +++ b/src/H5FAdblkpage.c @@ -153,6 +153,7 @@ H5FA__dblk_page_create(H5FA_hdr_t *hdr, hid_t dxpl_id, haddr_t addr, size_t nelm /* Local variables */ H5FA_dblk_page_t *dblk_page = NULL; /* Fixed array data block page */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ #ifdef H5FA_DEBUG HDfprintf(stderr, "%s: Called, addr = %a\n", FUNC, addr); @@ -179,10 +180,23 @@ HDfprintf(stderr, "%s: dblk_page->size = %Zu\n", FUNC, dblk_page->size); /* Cache the new fixed array data block page */ if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_FARRAY_DBLK_PAGE, dblk_page->addr, dblk_page, H5AC__NO_FLAGS_SET) < 0) H5E_THROW(H5E_CANTINSERT, "can't add fixed array data block page to cache") + inserted = TRUE; + + /* Add data block page as child of 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, dblk_page) < 0) + H5E_THROW(H5E_CANTSET, "unable to add fixed array entry as child of array proxy") + dblk_page->top_proxy = hdr->top_proxy; + } /* end if */ CATCH if(ret_value < 0) if(dblk_page) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(dblk_page) < 0) + H5E_THROW(H5E_CANTREMOVE, "unable to remove fixed array data block page from cache") + /* Destroy data block page */ if(H5FA__dblk_page_dest(dblk_page) < 0) H5E_THROW(H5E_CANTFREE, "unable to destroy fixed array data block page") @@ -210,6 +224,7 @@ H5FA__dblk_page_protect(H5FA_hdr_t *hdr, hid_t dxpl_id, haddr_t dblk_page_addr, size_t dblk_page_nelmts, unsigned flags)) /* Local variables */ + H5FA_dblk_page_t *dblk_page = NULL; /* Fixed array data block page */ H5FA_dblk_page_cache_ud_t udata; /* Information needed for loading data block page */ #ifdef H5FA_DEBUG @@ -229,11 +244,29 @@ HDfprintf(stderr, "%s: Called\n", FUNC); udata.dblk_page_addr = dblk_page_addr; /* Protect the data block page */ - if(NULL == (ret_value = (H5FA_dblk_page_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_FARRAY_DBLK_PAGE, dblk_page_addr, &udata, flags))) + if(NULL == (dblk_page = (H5FA_dblk_page_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_FARRAY_DBLK_PAGE, dblk_page_addr, &udata, flags))) H5E_THROW(H5E_CANTPROTECT, "unable to protect fixed array data block page, address = %llu", (unsigned long long)dblk_page_addr) + /* Create top proxy, if it doesn't exist */ + if(hdr->top_proxy && NULL == dblk_page->top_proxy) { + /* Add data block page as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, dblk_page) < 0) + H5E_THROW(H5E_CANTSET, "unable to add fixed array entry as child of array proxy") + dblk_page->top_proxy = hdr->top_proxy; + } /* end if */ + + /* Set return value */ + ret_value = dblk_page; + CATCH + /* Clean up on error */ + if(!ret_value) { + /* Release the data block page, if it was protected */ + if(dblk_page && H5AC_unprotect(hdr->f, dxpl_id, H5AC_FARRAY_DBLK_PAGE, dblk_page->addr, dblk_page, H5AC__NO_FLAGS_SET) < 0) + H5E_THROW(H5E_CANTUNPROTECT, "unable to unprotect fixed array data block page, address = %llu", (unsigned long long)dblk_page->addr) + } /* end if */ + END_FUNC(PKG) /* end H5FA__dblk_page_protect() */ @@ -306,6 +339,9 @@ H5FA__dblk_page_dest(H5FA_dblk_page_t *dblk_page)) dblk_page->hdr = NULL; } /* end if */ + /* Sanity check */ + HDassert(NULL == dblk_page->top_proxy); + /* Free the data block page itself */ dblk_page = H5FL_FREE(H5FA_dblk_page_t, dblk_page); diff --git a/src/H5FAdblock.c b/src/H5FAdblock.c index f597c8e..440447d 100644 --- a/src/H5FAdblock.c +++ b/src/H5FAdblock.c @@ -191,6 +191,7 @@ H5FA__dblock_create(H5FA_hdr_t *hdr, hid_t dxpl_id, hbool_t *hdr_dirty)) /* Local variables */ H5FA_dblock_t *dblock = NULL; /* Fixed array data block */ haddr_t dblock_addr; /* Fixed array data block address */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ /* Sanity check */ HDassert(hdr); @@ -217,6 +218,14 @@ H5FA__dblock_create(H5FA_hdr_t *hdr, hid_t dxpl_id, hbool_t *hdr_dirty)) /* Cache the new fixed array data block */ if(H5AC_insert_entry(hdr->f, dxpl_id, H5AC_FARRAY_DBLOCK, dblock_addr, dblock, H5AC__NO_FLAGS_SET) < 0) H5E_THROW(H5E_CANTINSERT, "can't add fixed array data block to cache") + inserted = TRUE; + + /* Add data block as child of 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, dblock) < 0) + H5E_THROW(H5E_CANTSET, "unable to add fixed array entry as child of array proxy") + dblock->top_proxy = hdr->top_proxy; + } /* end if */ /* Mark the header dirty (for updating statistics) */ *hdr_dirty = TRUE; @@ -228,6 +237,11 @@ CATCH if(!H5F_addr_defined(ret_value)) if(dblock) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(dblock) < 0) + H5E_THROW(H5E_CANTREMOVE, "unable to remove fixed array data block from cache") + /* Release data block's disk space */ if(H5F_addr_defined(dblock->addr) && H5MF_xfree(hdr->f, H5FD_MEM_FARRAY_DBLOCK, dxpl_id, dblock->addr, (hsize_t)dblock->size) < 0) H5E_THROW(H5E_CANTFREE, "unable to release fixed array data block") @@ -258,6 +272,7 @@ H5FA__dblock_protect(H5FA_hdr_t *hdr, hid_t dxpl_id, haddr_t dblk_addr, unsigned flags)) /* Local variables */ + H5FA_dblock_t *dblock; /* Fixed array data block */ H5FA_dblock_cache_ud_t udata; /* Information needed for loading data block */ /* Sanity check */ @@ -272,11 +287,28 @@ H5FA__dblock_protect(H5FA_hdr_t *hdr, hid_t dxpl_id, haddr_t dblk_addr, udata.dblk_addr = dblk_addr; /* Protect the data block */ - if(NULL == (ret_value = (H5FA_dblock_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_FARRAY_DBLOCK, dblk_addr, &udata, flags))) + if(NULL == (dblock = (H5FA_dblock_t *)H5AC_protect(hdr->f, dxpl_id, H5AC_FARRAY_DBLOCK, dblk_addr, &udata, flags))) H5E_THROW(H5E_CANTPROTECT, "unable to protect fixed array data block, address = %llu", (unsigned long long)dblk_addr) + /* Create top proxy, if it doesn't exist */ + if(hdr->top_proxy && NULL == dblock->top_proxy) { + /* Add data block as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, hdr->f, dxpl_id, dblock) < 0) + H5E_THROW(H5E_CANTSET, "unable to add fixed array entry as child of array proxy") + dblock->top_proxy = hdr->top_proxy; + } /* end if */ + + /* Set return value */ + ret_value = dblock; + CATCH + /* Clean up on error */ + if(!ret_value) + /* Release the data block, if it was protected */ + if(dblock && H5AC_unprotect(hdr->f, dxpl_id, H5AC_FARRAY_DBLOCK, dblock->addr, dblock, H5AC__NO_FLAGS_SET) < 0) + H5E_THROW(H5E_CANTUNPROTECT, "unable to unprotect fixed array data block, address = %llu", (unsigned long long)dblock->addr) + END_FUNC(PKG) /* end H5FA__dblock_protect() */ @@ -408,6 +440,9 @@ H5FA__dblock_dest(H5FA_dblock_t *dblock)) dblock->hdr = NULL; } /* end if */ + /* Sanity check */ + HDassert(NULL == dblock->top_proxy); + /* Free the data block itself */ dblock = H5FL_FREE(H5FA_dblock_t, dblock); diff --git a/src/H5FAhdr.c b/src/H5FAhdr.c index d6e32dc..52b90d1 100644 --- a/src/H5FAhdr.c +++ b/src/H5FAhdr.c @@ -112,6 +112,7 @@ H5FA__hdr_alloc(H5F_t *f)) /* Set the internal parameters for the array */ hdr->f = f; + hdr->swmr_write = (H5F_INTENT(f) & H5F_ACC_SWMR_WRITE) > 0; hdr->sizeof_addr = H5F_SIZEOF_ADDR(f); hdr->sizeof_size = H5F_SIZEOF_SIZE(f); @@ -184,6 +185,7 @@ H5FA__hdr_create(H5F_t *f, hid_t dxpl_id, const H5FA_create_t *cparam, /* Local variables */ H5FA_hdr_t *hdr = NULL; /* Fixed array header */ + hbool_t inserted = FALSE; /* Whether the header was inserted into cache */ /* Check arguments */ HDassert(f); @@ -218,9 +220,20 @@ H5FA__hdr_create(H5F_t *f, hid_t dxpl_id, const H5FA_create_t *cparam, if(HADDR_UNDEF == (hdr->addr = H5MF_alloc(f, H5FD_MEM_FARRAY_HDR, dxpl_id, (hsize_t)hdr->size))) H5E_THROW(H5E_CANTALLOC, "file allocation failed for Fixed Array header") + /* Create 'top' proxy for extensible array entries */ + if(hdr->swmr_write) + if(NULL == (hdr->top_proxy = H5AC_proxy_entry_create())) + H5E_THROW(H5E_CANTCREATE, "can't create fixed array entry proxy") + /* Cache the new Fixed Array header */ if(H5AC_insert_entry(f, dxpl_id, H5AC_FARRAY_HDR, hdr->addr, hdr, H5AC__NO_FLAGS_SET) < 0) H5E_THROW(H5E_CANTINSERT, "can't add fixed array header to cache") + inserted = TRUE; + + /* Add header as child of 'top' proxy */ + if(hdr->top_proxy) + if(H5AC_proxy_entry_add_child(hdr->top_proxy, f, dxpl_id, hdr) < 0) + H5E_THROW(H5E_CANTSET, "unable to add fixed array entry as child of array proxy") /* Set address of array header to return */ ret_value = hdr->addr; @@ -229,6 +242,11 @@ CATCH if(!H5F_addr_defined(ret_value)) if(hdr) { + /* Remove from cache, if inserted */ + if(inserted) + if(H5AC_remove_entry(hdr) < 0) + H5E_THROW(H5E_CANTREMOVE, "unable to remove fixed array header from cache") + /* Release header's disk space */ if(H5F_addr_defined(hdr->addr) && H5MF_xfree(f, H5FD_MEM_FARRAY_HDR, dxpl_id, hdr->addr, (hsize_t)hdr->size) < 0) H5E_THROW(H5E_CANTFREE, "unable to free Fixed Array header") @@ -409,6 +427,7 @@ H5FA__hdr_protect(H5F_t *f, hid_t dxpl_id, haddr_t fa_addr, void *ctx_udata, unsigned flags)) /* Local variables */ + H5FA_hdr_t *hdr; /* Fixed array header */ H5FA_hdr_cache_ud_t udata; /* User data for cache callbacks */ /* Sanity check */ @@ -424,9 +443,23 @@ H5FA__hdr_protect(H5F_t *f, hid_t dxpl_id, haddr_t fa_addr, void *ctx_udata, udata.ctx_udata = ctx_udata; /* Protect the header */ - if(NULL == (ret_value = (H5FA_hdr_t *)H5AC_protect(f, dxpl_id, H5AC_FARRAY_HDR, fa_addr, &udata, flags))) + if(NULL == (hdr = (H5FA_hdr_t *)H5AC_protect(f, dxpl_id, H5AC_FARRAY_HDR, fa_addr, &udata, flags))) H5E_THROW(H5E_CANTPROTECT, "unable to protect fixed array header, address = %llu", (unsigned long long)fa_addr) - ret_value->f = f; /* (Must be set again here, in case the header was already in the cache -QAK) */ + hdr->f = f; /* (Must be set again here, in case the header was already in the cache -QAK) */ + + /* Create top proxy, if it doesn't exist */ + if(hdr->swmr_write && NULL == hdr->top_proxy) { + /* Create 'top' proxy for fixed array entries */ + if(NULL == (hdr->top_proxy = H5AC_proxy_entry_create())) + H5E_THROW(H5E_CANTCREATE, "can't create fixed array entry proxy") + + /* Add header as child of 'top' proxy */ + if(H5AC_proxy_entry_add_child(hdr->top_proxy, f, dxpl_id, hdr) < 0) + H5E_THROW(H5E_CANTSET, "unable to add fixed array entry as child of array proxy") + } /* end if */ + + /* Set return value */ + ret_value = hdr; CATCH @@ -547,6 +580,13 @@ H5FA__hdr_dest(H5FA_hdr_t *hdr)) } /* end if */ hdr->cb_ctx = NULL; + /* Destroy the 'top' proxy */ + if(hdr->top_proxy) { + if(H5AC_proxy_entry_dest(hdr->top_proxy) < 0) + H5E_THROW(H5E_CANTRELEASE, "unable to destroy fixed array 'top' proxy") + hdr->top_proxy = NULL; + } /* end if */ + /* Free the shared info itself */ hdr = H5FL_FREE(H5FA_hdr_t, hdr); diff --git a/src/H5FAint.c b/src/H5FAint.c new file mode 100644 index 0000000..331227b --- /dev/null +++ b/src/H5FAint.c @@ -0,0 +1,139 @@ +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * + * Copyright by The HDF Group. * + * Copyright by the Board of Trustees of the University of Illinois. * + * All rights reserved. * + * * + * This file is part of HDF5. The full HDF5 copyright notice, including * + * terms governing use, modification, and redistribution, is contained in * + * the files COPYING and Copyright.html. COPYING can be found at the root * + * of the source code distribution tree; Copyright.html can be found at the * + * root level of an installed copy of the electronic HDF5 document set and * + * is linked from the top-level documents page. It can also be found at * + * http://hdfgroup.org/HDF5/doc/Copyright.html. If you do not have * + * access to either file, you may request a copy from help@hdfgroup.org. * + * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ + +/*------------------------------------------------------------------------- + * + * Created: H5FAint.c + * Fall 2012 + * Dana Robinson <derobins@hdfgroup.org> + * + * Purpose: Internal routines for fixed arrays. + * + *------------------------------------------------------------------------- + */ + +/**********************/ +/* Module Declaration */ +/**********************/ + +#include "H5FAmodule.h" /* This source code file is part of the H5FA module */ + + +/***********************/ +/* Other Packages Used */ +/***********************/ + + +/***********/ +/* Headers */ +/***********/ +#include "H5private.h" /* Generic Functions */ +#include "H5Eprivate.h" /* Error Handling */ +#include "H5FApkg.h" /* Fixed Arrays */ + + +/****************/ +/* Local Macros */ +/****************/ + + +/******************/ +/* Local Typedefs */ +/******************/ + + +/********************/ +/* Package Typedefs */ +/********************/ + + +/********************/ +/* Local Prototypes */ +/********************/ + + +/*********************/ +/* Package Variables */ +/*********************/ + + +/*****************************/ +/* Library Private Variables */ +/*****************************/ + + +/*******************/ +/* Local Variables */ +/*******************/ + + + +/*------------------------------------------------------------------------- + * Function: H5FA__create_flush_depend + * + * Purpose: Create a flush dependency between two data structure components + * + * Return: SUCCEED/FAIL + * + * Programmer: Dana Robinson + * Fall 2012 + * + *------------------------------------------------------------------------- + */ +BEGIN_FUNC(PKG, ERR, +herr_t, SUCCEED, FAIL, +H5FA__create_flush_depend(H5AC_info_t *parent_entry, H5AC_info_t *child_entry)) + + /* Sanity check */ + HDassert(parent_entry); + HDassert(child_entry); + + /* Create a flush dependency between parent and child entry */ + if(H5AC_create_flush_dependency(parent_entry, child_entry) < 0) + H5E_THROW(H5E_CANTDEPEND, "unable to create flush dependency") + +CATCH + +END_FUNC(PKG) /* end H5FA__create_flush_depend() */ + + +/*------------------------------------------------------------------------- + * Function: H5FA__destroy_flush_depend + * + * Purpose: Destroy a flush dependency between two data structure components + * + * Return: SUCCEED/FAIL + * + * Programmer: Dana Robinson + * Fall 2012 + * + *------------------------------------------------------------------------- + */ +BEGIN_FUNC(PKG, ERR, +herr_t, SUCCEED, FAIL, +H5FA__destroy_flush_depend(H5AC_info_t *parent_entry, H5AC_info_t *child_entry)) + + /* Sanity check */ + HDassert(parent_entry); + HDassert(child_entry); + + /* Destroy a flush dependency between parent and child entry */ + if(H5AC_destroy_flush_dependency(parent_entry, child_entry) < 0) + H5E_THROW(H5E_CANTUNDEPEND, "unable to destroy flush dependency") + +CATCH + +END_FUNC(PKG) /* end H5FA__destroy_flush_depend() */ + diff --git a/src/H5FApkg.h b/src/H5FApkg.h index c29322a..ccef562 100644 --- a/src/H5FApkg.h +++ b/src/H5FApkg.h @@ -144,6 +144,28 @@ typedef struct H5FA_hdr_t { /* Client information (not stored) */ void *cb_ctx; /* Callback context */ + + /* SWMR / Flush dependency information (not stored) */ + hbool_t swmr_write; /* Flag indicating the file is opened with SWMR-write access */ + H5AC_proxy_entry_t *top_proxy; /* 'Top' proxy cache entry for all array entries */ + void *parent; /* Pointer to 'top' proxy flush dependency + * parent, if it exists, otherwise NULL. + * If the fixed array is being used + * to index a chunked dataset and the + * dataset metadata is modified by a + * SWMR writer, this field will be set + * equal to the object header proxy + * that is the flush dependency parent + * of the fixed array header. + * + * The field is used to avoid duplicate + * setups of the flush dependency + * relationship, and to allow the + * fixed array header to destroy + * the flush dependency on receipt of + * an eviction notification from the + * metadata cache. + */ } H5FA_hdr_t; /* The fixed array data block information */ @@ -158,6 +180,9 @@ typedef struct H5FA_dblock_t { /* Internal array information (not stored) */ H5FA_hdr_t *hdr; /* Shared array header info */ + /* SWMR / Flush dependency information (not stored) */ + H5AC_proxy_entry_t *top_proxy; /* 'Top' proxy cache entry for all array entries */ + /* Computed/cached values (not stored) */ haddr_t addr; /* Address of this data block on disk */ hsize_t size; /* Size of data block on disk */ @@ -181,6 +206,9 @@ typedef struct H5FA_dbk_page_t { /* Internal array information (not stored) */ H5FA_hdr_t *hdr; /* Shared array header info */ + /* SWMR / Flush dependency information (not stored) */ + H5AC_proxy_entry_t *top_proxy; /* 'Top' proxy cache entry for all array entries */ + /* Computed/cached values (not stored) */ haddr_t addr; /* Address of this data block page on disk */ size_t size; /* Size of data block page on disk */ @@ -241,6 +269,12 @@ H5_DLLVAR const H5FA_class_t *const H5FA_client_class_g[H5FA_NUM_CLS_ID]; /* Package Private Prototypes */ /******************************/ +/* Generic routines */ +H5_DLL herr_t H5FA__create_flush_depend(H5AC_info_t *parent_entry, + H5AC_info_t *child_entry); +H5_DLL herr_t H5FA__destroy_flush_depend(H5AC_info_t *parent_entry, + H5AC_info_t *child_entry); + /* Header routines */ H5_DLL H5FA_hdr_t *H5FA__hdr_alloc(H5F_t *f); H5_DLL herr_t H5FA__hdr_init(H5FA_hdr_t *hdr, void *ctx_udata); diff --git a/src/H5FAprivate.h b/src/H5FAprivate.h index 1e76468..612f3a2 100644 --- a/src/H5FAprivate.h +++ b/src/H5FAprivate.h @@ -32,6 +32,7 @@ #endif /* NOT_YET */ /* Private headers needed by this file */ +#include "H5ACprivate.h" /* Metadata cache */ #include "H5Fprivate.h" /* File access */ @@ -126,6 +127,7 @@ H5_DLL herr_t H5FA_get_nelmts(const H5FA_t *fa, hsize_t *nelmts); H5_DLL herr_t H5FA_get_addr(const H5FA_t *fa, haddr_t *addr); H5_DLL herr_t H5FA_set(const H5FA_t *fa, hid_t dxpl_id, hsize_t idx, const void *elmt); H5_DLL herr_t H5FA_get(const H5FA_t *fa, hid_t dxpl_id, hsize_t idx, void *elmt); +H5_DLL herr_t H5FA_depend(H5FA_t *fa, hid_t dxpl_id, H5AC_proxy_entry_t *parent); H5_DLL herr_t H5FA_iterate(H5FA_t *fa, hid_t dxpl_id, H5FA_operator_t op, void *udata); H5_DLL herr_t H5FA_close(H5FA_t *fa, hid_t dxpl_id); H5_DLL herr_t H5FA_delete(H5F_t *f, hid_t dxpl_id, haddr_t fa_addr, void *ctx_udata); @@ -811,6 +811,9 @@ H5FD_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxaddr) if(NULL == (file = (driver->open)(name, flags, fapl_id, maxaddr))) HGOTO_ERROR(H5E_VFL, H5E_CANTINIT, NULL, "open failed") + /* Set the file access flags */ + file->access_flags = flags; + /* * Fill in public fields. We must increment the reference count on the * driver ID to prevent it from being freed while this file is open. @@ -1441,6 +1444,32 @@ H5FD_get_feature_flags(const H5FD_t *file, unsigned long *feature_flags) /*------------------------------------------------------------------------- + * Function: H5FD_set_feature_flags + * + * Purpose: Set the feature flags for the VFD + * + * Return: Success: Non-negative + * Failure: Negative + * + * Programmer: Vailin Choi; Oct 2013 + * + *------------------------------------------------------------------------- + */ +herr_t +H5FD_set_feature_flags(H5FD_t *file, unsigned long feature_flags) +{ + FUNC_ENTER_NOAPI_NOINIT_NOERR + + HDassert(file); + + /* Set the file's feature flags */ + file->feature_flags = feature_flags; + + FUNC_LEAVE_NOAPI(SUCCEED) +} /* end H5FD_set_feature_flags() */ + + +/*------------------------------------------------------------------------- * Function: H5FD_get_fs_type_map * * Purpose: Retrieve the free space type mapping for the VFD diff --git a/src/H5FDint.c b/src/H5FDint.c index cd48197..744c3d1 100644 --- a/src/H5FDint.c +++ b/src/H5FDint.c @@ -207,9 +207,16 @@ H5P_genplist_t *dxpl, H5FD_mem_t type, haddr_t addr, if(HADDR_UNDEF == (eoa = (file->cls->get_eoa)(file, type))) HGOTO_ERROR(H5E_VFL, H5E_CANTINIT, FAIL, "driver get_eoa request failed") - if((addr + file->base_addr + size) > eoa) - HGOTO_ERROR(H5E_ARGS, H5E_OVERFLOW, FAIL, "addr overflow, addr = %llu, size=%llu, eoa=%llu", - (unsigned long long)(addr+ file->base_addr), (unsigned long long)size, (unsigned long long)eoa) + + /* + * If the file is open for SWMR read access, allow access to data past + * the end of the allocated space (the 'eoa'). This is done because the + * eoa stored in the file's superblock might be out of sync with the + * objects being written within the file by the application performing + * SWMR write operations. + */ + if(!(file->access_flags & H5F_ACC_SWMR_READ) && ((addr + file->base_addr + size) > eoa)) + HGOTO_ERROR(H5E_ARGS, H5E_OVERFLOW, FAIL, "addr overflow, addr = %llu, size = %llu, eoa = %llu", (unsigned long long)(addr + file->base_addr), (unsigned long long)size, (unsigned long long)eoa) /* Dispatch to driver */ if((file->cls->read)(file, type, H5P_PLIST_ID(dxpl), addr + file->base_addr, size, buf) < 0) diff --git a/src/H5FDlog.c b/src/H5FDlog.c index a2891cb..03228d2 100644 --- a/src/H5FDlog.c +++ b/src/H5FDlog.c @@ -898,6 +898,7 @@ H5FD_log_query(const H5FD_t *_file, unsigned long *flags /* out */) *flags |= H5FD_FEAT_DATA_SIEVE; /* OK to perform data sieving for faster raw data reads & writes */ *flags |= H5FD_FEAT_AGGREGATE_SMALLDATA; /* OK to aggregate "small" raw data allocations */ *flags |= H5FD_FEAT_POSIX_COMPAT_HANDLE; /* VFD handle is POSIX I/O call compatible */ + *flags |= H5FD_FEAT_SUPPORTS_SWMR_IO; /* VFD supports the single-writer/multiple-readers (SWMR) pattern */ /* Check for flags that are set by h5repart */ if(file && file->fam_to_sec2) diff --git a/src/H5FDpkg.h b/src/H5FDpkg.h index a0c1b3a..45bcfd8 100644 --- a/src/H5FDpkg.h +++ b/src/H5FDpkg.h @@ -59,6 +59,7 @@ H5_DLL herr_t H5FD_free_real(H5FD_t *file, hid_t dxpl_id, H5FD_mem_t type, /* Testing functions */ #ifdef H5FD_TESTING +H5_DLL hbool_t H5FD_supports_swmr_test(const char *vfd_name); #endif /* H5FD_TESTING */ #endif /* _H5FDpkg_H */ diff --git a/src/H5FDprivate.h b/src/H5FDprivate.h index c40dc54..427b42c 100644 --- a/src/H5FDprivate.h +++ b/src/H5FDprivate.h @@ -157,6 +157,7 @@ H5_DLL herr_t H5FD_set_eoa(H5FD_t *file, H5FD_mem_t type, haddr_t addr); H5_DLL haddr_t H5FD_get_eof(const H5FD_t *file, H5FD_mem_t type); H5_DLL haddr_t H5FD_get_maxaddr(const H5FD_t *file); H5_DLL herr_t H5FD_get_feature_flags(const H5FD_t *file, unsigned long *feature_flags); +H5_DLL herr_t H5FD_set_feature_flags(H5FD_t *file, unsigned long feature_flags); H5_DLL herr_t H5FD_get_fs_type_map(const H5FD_t *file, H5FD_mem_t *type_map); H5_DLL herr_t H5FD_read(H5FD_t *file, #ifndef H5_DEBUG_BUILD diff --git a/src/H5FDpublic.h b/src/H5FDpublic.h index 4931e0f..436be10 100644 --- a/src/H5FDpublic.h +++ b/src/H5FDpublic.h @@ -234,6 +234,11 @@ typedef enum H5F_mem_t H5FD_mem_t; * image to store in memory. */ #define H5FD_FEAT_CAN_USE_FILE_IMAGE_CALLBACKS 0x00000800 + /* + * Defining H5FD_FEAT_SUPPORTS_SWMR_IO for a VFL driver means that the + * driver supports the single-writer/multiple-readers I/O pattern. + */ +#define H5FD_FEAT_SUPPORTS_SWMR_IO 0x00001000 /* Forward declaration */ typedef struct H5FD_t H5FD_t; @@ -294,6 +299,7 @@ struct H5FD_t { hid_t driver_id; /*driver ID for this file */ const H5FD_class_t *cls; /*constant class info */ unsigned long fileno; /* File 'serial' number */ + unsigned access_flags; /* File access flags (from create or open) */ unsigned long feature_flags; /* VFL Driver feature Flags */ haddr_t maxaddr; /* For this file, overrides class */ haddr_t base_addr; /* Base address for HDF5 data w/in file */ diff --git a/src/H5FDsec2.c b/src/H5FDsec2.c index 04a0790..0ca5efb 100644 --- a/src/H5FDsec2.c +++ b/src/H5FDsec2.c @@ -530,6 +530,7 @@ H5FD_sec2_query(const H5FD_t *_file, unsigned long *flags /* out */) *flags |= H5FD_FEAT_DATA_SIEVE; /* OK to perform data sieving for faster raw data reads & writes */ *flags |= H5FD_FEAT_AGGREGATE_SMALLDATA; /* OK to aggregate "small" raw data allocations */ *flags |= H5FD_FEAT_POSIX_COMPAT_HANDLE; /* VFD handle is POSIX I/O call compatible */ + *flags |= H5FD_FEAT_SUPPORTS_SWMR_IO; /* VFD supports the single-writer/multiple-readers (SWMR) pattern */ /* Check for flags that are set by h5repart */ if(file && file->fam_to_sec2) diff --git a/src/H5FDstdio.c b/src/H5FDstdio.c index 7056f7f..4c62bcc 100644 --- a/src/H5FDstdio.c +++ b/src/H5FDstdio.c @@ -550,7 +550,11 @@ H5FD_stdio_query(const H5FD_t *_f, unsigned long /*OUT*/ *flags) /* Quiet the compiler */ _f=_f; - /* Set the VFL feature flags that this driver supports */ + /* Set the VFL feature flags that this driver supports. + * + * Note that this VFD does not support SWMR due to the unpredictable + * nature of the buffering layer. + */ if(flags) { *flags = 0; *flags|=H5FD_FEAT_AGGREGATE_METADATA; /* OK to aggregate metadata allocations */ diff --git a/src/H5FDtest.c b/src/H5FDtest.c new file mode 100644 index 0000000..9d2f2e9 --- /dev/null +++ b/src/H5FDtest.c @@ -0,0 +1,119 @@ +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * + * Copyright by The HDF Group. * + * Copyright by the Board of Trustees of the University of Illinois. * + * All rights reserved. * + * * + * This file is part of HDF5. The full HDF5 copyright notice, including * + * terms governing use, modification, and redistribution, is contained in * + * the files COPYING and Copyright.html. COPYING can be found at the root * + * of the source code distribution tree; Copyright.html can be found at the * + * root level of an installed copy of the electronic HDF5 document set and * + * is linked from the top-level documents page. It can also be found at * + * http://hdfgroup.org/HDF5/doc/Copyright.html. If you do not have * + * access to either file, you may request a copy from help@hdfgroup.org. * + * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ + +/*------------------------------------------------------------------------- + * + * Created: H5FDtest.c + * Fall 2014 + * + * Purpose: File driver testing routines. + * + *------------------------------------------------------------------------- + */ + +/****************/ +/* Module Setup */ +/****************/ + +#include "H5FDmodule.h" /* This source code file is part of the H5FD module */ +#define H5FD_TESTING /* Suppress warning about H5FD testing funcs */ + + +/***********/ +/* Headers */ +/***********/ +#include "H5private.h" /* Generic Functions */ +#include "H5FDpkg.h" /* File Drivers */ + +/****************/ +/* Local Macros */ +/****************/ + + +/******************/ +/* Local Typedefs */ +/******************/ + + +/********************/ +/* Package Typedefs */ +/********************/ + + +/********************/ +/* Local Prototypes */ +/********************/ + + +/*********************/ +/* Package Variables */ +/*********************/ + + +/*****************************/ +/* Library Private Variables */ +/*****************************/ + + +/*******************/ +/* Local Variables */ +/*******************/ + + +/*------------------------------------------------------------------------- + * Function: H5FD_supports_swmr_test() + * + * Purpose: Determines if a VFD supports SWMR. + * + * The function determines SWMR support by inspecting the + * HDF5_DRIVER environment variable, not by checking the + * VFD feature flags (which do not exist until the driver + * is instantiated). + * + * See test/Makefile.am for a list of the VFD strings. + * + * This function is only intended for use in the test code. + * + * Return: TRUE (1) if the VFD supports SWMR I/O or vfd_name is + * NULL or the empty string (which implies the default VFD). + * + * FALSE (0) if it does not + * + * This function cannot fail at this time so there is no + * error return value. + * + * Programmer: Dana Robinson + * Fall 2014 + * + *------------------------------------------------------------------------- + */ +hbool_t +H5FD_supports_swmr_test(const char *vfd_name) +{ + hbool_t ret_value = FALSE; + + FUNC_ENTER_NOAPI_NOINIT_NOERR + + if(!vfd_name || !HDstrcmp(vfd_name, "")) + ret_value = TRUE; + else + ret_value = !HDstrcmp(vfd_name, "direct") + || !HDstrcmp(vfd_name, "log") + || !HDstrcmp(vfd_name, "sec2"); + + FUNC_LEAVE_NOAPI(ret_value) + +} /* end H5FD_supports_swmr_test() */ + @@ -132,6 +132,7 @@ HDfprintf(stderr, "%s: Creating free space manager, nclasses = %Zu\n", FUNC, ncl fspace->expand_percent = fs_create->expand_percent; fspace->max_sect_addr = fs_create->max_sect_addr; fspace->max_sect_size = fs_create->max_sect_size; + fspace->swmr_write = (H5F_INTENT(f) & H5F_ACC_SWMR_WRITE) > 0; fspace->alignment = alignment; fspace->threshold = threshold; diff --git a/src/H5FScache.c b/src/H5FScache.c index c1a9e0a..f43676d 100644 --- a/src/H5FScache.c +++ b/src/H5FScache.c @@ -100,6 +100,7 @@ static herr_t H5FS__cache_sinfo_pre_serialize(const H5F_t *f, hid_t dxpl_id, unsigned *flags); static herr_t H5FS__cache_sinfo_serialize(const H5F_t *f, void *image, size_t len, void *thing); +static herr_t H5FS__cache_sinfo_notify(H5AC_notify_action_t action, void *thing); static herr_t H5FS__cache_sinfo_free_icr(void *thing); @@ -138,7 +139,7 @@ const H5AC_class_t H5AC_FSPACE_SINFO[1] = {{ H5FS__cache_sinfo_image_len, /* 'image_len' callback */ H5FS__cache_sinfo_pre_serialize, /* 'pre_serialize' callback */ H5FS__cache_sinfo_serialize, /* 'serialize' callback */ - NULL, /* 'notify' callback */ + H5FS__cache_sinfo_notify, /* 'notify' callback */ H5FS__cache_sinfo_free_icr, /* 'free_icr' callback */ NULL, /* 'fsf_size' callback */ }}; @@ -1235,6 +1236,68 @@ done: /*------------------------------------------------------------------------- + * Function: H5FS__cache_sinfo_notify + * + * Purpose: Handle cache action notifications + * + * Return: SUCCEED/FAIL + * + * Programmer: Dana Robinson + * Fall 2012 + * + *------------------------------------------------------------------------- + */ +herr_t +H5FS__cache_sinfo_notify(H5AC_notify_action_t action, void *_thing) +{ + H5FS_sinfo_t *sinfo = (H5FS_sinfo_t *)_thing; + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI_NOINIT + + /* Sanity check */ + HDassert(sinfo); + + /* Check if the file was opened with SWMR-write access */ + if(sinfo->fspace->swmr_write) { + /* Determine which action to take */ + switch(action) { + case H5AC_NOTIFY_ACTION_AFTER_INSERT: + case H5AC_NOTIFY_ACTION_AFTER_LOAD: + /* Create flush dependency on parent */ + if(H5FS__create_flush_depend((H5AC_info_t *)sinfo->fspace, (H5AC_info_t *)sinfo) < 0) + HGOTO_ERROR(H5E_FSPACE, H5E_CANTDEPEND, FAIL, "unable to create flush dependency between data block and header, address = %llu", (unsigned long long)sinfo->fspace->sect_addr) + break; + + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: + case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: + case H5AC_NOTIFY_ACTION_ENTRY_CLEANED: + case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: + case H5AC_NOTIFY_ACTION_CHILD_CLEANED: + /* do nothing */ + break; + + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + /* Destroy flush dependency on parent */ + if(H5FS__destroy_flush_depend((H5AC_info_t *)sinfo->fspace, (H5AC_info_t *)sinfo) < 0) + HGOTO_ERROR(H5E_FSPACE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency") + break; + + default: +#ifdef NDEBUG + HGOTO_ERROR(H5E_FSPACE, H5E_BADVALUE, FAIL, "unknown action from metadata cache") +#else /* NDEBUG */ + HDassert(0 && "Unknown action?!?"); +#endif /* NDEBUG */ + } /* end switch */ + } /* end if */ + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5FS__cache_sinfo_notify() */ + + +/*------------------------------------------------------------------------- * Function: H5FS__cache_sinfo_free_icr * * Purpose: Free the memory used for the in core representation of the diff --git a/src/H5FSint.c b/src/H5FSint.c new file mode 100644 index 0000000..60cedd5 --- /dev/null +++ b/src/H5FSint.c @@ -0,0 +1,145 @@ +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * + * Copyright by The HDF Group. * + * Copyright by the Board of Trustees of the University of Illinois. * + * All rights reserved. * + * * + * This file is part of HDF5. The full HDF5 copyright notice, including * + * terms governing use, modification, and redistribution, is contained in * + * the files COPYING and Copyright.html. COPYING can be found at the root * + * of the source code distribution tree; Copyright.html can be found at the * + * root level of an installed copy of the electronic HDF5 document set and * + * is linked from the top-level documents page. It can also be found at * + * http://hdfgroup.org/HDF5/doc/Copyright.html. If you do not have * + * access to either file, you may request a copy from help@hdfgroup.org. * + * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ + +/*------------------------------------------------------------------------- + * + * Created: H5FSint.c + * Fall 2012 + * Dana Robinson <derobins@hdfgroup.org> + * + * Purpose: Internal routines for free space managers. + * + *------------------------------------------------------------------------- + */ + +/**********************/ +/* Module Declaration */ +/**********************/ + +#include "H5FSmodule.h" /* This source code file is part of the H5FS module */ + + +/***********************/ +/* Other Packages Used */ +/***********************/ + + +/***********/ +/* Headers */ +/***********/ +#include "H5private.h" /* Generic Functions */ +#include "H5Eprivate.h" /* Error Handling */ +#include "H5FSpkg.h" /* Free Space Managers */ + + +/****************/ +/* Local Macros */ +/****************/ + + +/******************/ +/* Local Typedefs */ +/******************/ + + +/********************/ +/* Package Typedefs */ +/********************/ + + +/********************/ +/* Local Prototypes */ +/********************/ + + +/*********************/ +/* Package Variables */ +/*********************/ + + +/*****************************/ +/* Library Private Variables */ +/*****************************/ + + +/*******************/ +/* Local Variables */ +/*******************/ + + + +/*------------------------------------------------------------------------- + * Function: H5FS__create_flush_depend + * + * Purpose: Create a flush dependency between two data structure components + * + * Return: SUCCEED/FAIL + * + * Programmer: Dana Robinson + * Fall 2012 + * + *------------------------------------------------------------------------- + */ +herr_t +H5FS__create_flush_depend(H5AC_info_t *parent_entry, H5AC_info_t *child_entry) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI_NOINIT + + /* Sanity check */ + HDassert(parent_entry); + HDassert(child_entry); + + /* Create a flush dependency between parent and child entry */ + if(H5AC_create_flush_dependency(parent_entry, child_entry) < 0) + HGOTO_ERROR(H5E_FSPACE, H5E_CANTDEPEND, FAIL, "unable to create flush dependency") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5FS__create_flush_depend() */ + + +/*------------------------------------------------------------------------- + * Function: H5FS__destroy_flush_depend + * + * Purpose: Destroy a flush dependency between two data structure components + * + * Return: SUCCEED/FAIL + * + * Programmer: Dana Robinson + * Fall 2012 + * + *------------------------------------------------------------------------- + */ +herr_t +H5FS__destroy_flush_depend(H5AC_info_t *parent_entry, H5AC_info_t *child_entry) +{ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI_NOINIT + + /* Sanity check */ + HDassert(parent_entry); + HDassert(child_entry); + + /* Destroy a flush dependency between parent and child entry */ + if(H5AC_destroy_flush_dependency(parent_entry, child_entry) < 0) + HGOTO_ERROR(H5E_FSPACE, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency") + +done: + FUNC_LEAVE_NOAPI(ret_value) +} /* end H5FS__destroy_flush_depend() */ + diff --git a/src/H5FSpkg.h b/src/H5FSpkg.h index 7e52b5f..b3b1548 100644 --- a/src/H5FSpkg.h +++ b/src/H5FSpkg.h @@ -179,6 +179,7 @@ struct H5FS_t { haddr_t addr; /* Address of free space header on disk */ size_t hdr_size; /* Size of free space header on disk */ H5FS_sinfo_t *sinfo; /* Section information */ + hbool_t swmr_write; /* Flag indicating the file is opened with SWMR-write access */ unsigned sinfo_lock_count; /* # of times the section info has been locked */ hbool_t sinfo_protected; /* Whether the section info was protected when locked */ hbool_t sinfo_modified; /* Whether the section info has been modified while locked */ @@ -222,6 +223,12 @@ H5FL_EXTERN(H5FS_t); /* Package Private Prototypes */ /******************************/ +/* Generic routines */ +H5_DLL herr_t H5FS__create_flush_depend(H5AC_info_t *parent_entry, + H5AC_info_t *child_entry); +H5_DLL herr_t H5FS__destroy_flush_depend(H5AC_info_t *parent_entry, + H5AC_info_t *child_entry); + /* Free space manager header routines */ H5_DLL H5FS_t *H5FS__new(const H5F_t *f, uint16_t nclasses, const H5FS_section_class_t *classes[], void *cls_init_udata); diff --git a/src/H5Faccum.c b/src/H5Faccum.c index 3fac184..48f9bdd 100644 --- a/src/H5Faccum.c +++ b/src/H5Faccum.c @@ -726,6 +726,12 @@ H5F__accum_write(const H5F_io_info_t *fio_info, H5FD_mem_t map_type, haddr_t add } /* end else */ } /* end if */ else { + /* Make certain that data in accumulator is visible before new write */ + if((H5F_INTENT(fio_info->f) & H5F_ACC_SWMR_WRITE) > 0) + /* Flush if dirty and reset accumulator */ + if(H5F__accum_reset(fio_info, TRUE) < 0) + HGOTO_ERROR(H5E_IO, H5E_CANTRESET, FAIL, "can't reset accumulator") + /* Write the data */ if(H5FD_write(fio_info->f->shared->lf, fio_info->dxpl, map_type, addr, size, buf) < 0) HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "file write failed") diff --git a/src/H5Fint.c b/src/H5Fint.c index 9cf129c..363812d 100644 --- a/src/H5Fint.c +++ b/src/H5Fint.c @@ -641,8 +641,8 @@ H5F_new(H5F_file_t *shared, unsigned flags, hid_t fcpl_id, hid_t fapl_id, H5FD_t HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, NULL, "can't get sieve buffer size") if(H5P_get(plist, H5F_ACS_LATEST_FORMAT_NAME, &latest_format) < 0) HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, NULL, "can't get 'latest format' flag") - /* For latest format, activate all latest version support */ - if(latest_format) + /* For latest format or SWMR_WRITE, activate all latest version support */ + if(latest_format || (H5F_INTENT(f) & H5F_ACC_SWMR_WRITE)) f->shared->latest_flags |= H5F_LATEST_ALL_FLAGS; if(H5P_get(plist, H5F_ACS_USE_MDC_LOGGING_NAME, &(f->shared->use_mdc_logging)) < 0) HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, NULL, "can't get 'use mdc logging' flag") @@ -672,6 +672,15 @@ H5F_new(H5F_file_t *shared, unsigned flags, hid_t fcpl_id, hid_t fapl_id, H5FD_t HGOTO_ERROR(H5E_FILE, H5E_BADVALUE, NULL, "bad maximum address from VFD") if(H5FD_get_feature_flags(lf, &f->shared->feature_flags) < 0) HGOTO_ERROR(H5E_FILE, H5E_CANTGET, NULL, "can't get feature flags from VFD") + + /* Require the SWMR feature flag if SWMR I/O is desired */ + if(!H5F_HAS_FEATURE(f, H5FD_FEAT_SUPPORTS_SWMR_IO) && (H5F_INTENT(f) & (H5F_ACC_SWMR_WRITE | H5F_ACC_SWMR_READ))) + HGOTO_ERROR(H5E_FILE, H5E_BADVALUE, NULL, "must use a SWMR-compatible VFD when SWMR is specified") + + /* Require a POSIX compatible VFD to use SWMR feature */ + /* (It's reasonable to try to expand this to other VFDs eventually -QAK) */ + if(!H5F_HAS_FEATURE(f, H5FD_FEAT_POSIX_COMPAT_HANDLE) && (H5F_INTENT(f) & (H5F_ACC_SWMR_WRITE | H5F_ACC_SWMR_READ))) + HGOTO_ERROR(H5E_FILE, H5E_BADVALUE, NULL, "must use POSIX compatible VFD with SWMR write access") if(H5FD_get_fs_type_map(lf, f->shared->fs_type_map) < 0) HGOTO_ERROR(H5E_FILE, H5E_CANTGET, NULL, "can't get free space type mapping from VFD") if(H5MF_init_merge_flags(f) < 0) @@ -692,9 +701,23 @@ H5F_new(H5F_file_t *shared, unsigned flags, hid_t fcpl_id, hid_t fapl_id, H5FD_t if(H5P_get(plist, H5F_ACS_METADATA_READ_ATTEMPTS_NAME, &f->shared->read_attempts) < 0) HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, NULL, "can't get the # of read attempts") - /* If no value for read attempts has been set, use the default */ - if(!f->shared->read_attempts) - f->shared->read_attempts = H5F_METADATA_READ_ATTEMPTS; + /* When opening file with SWMR access, the # of read attempts is H5F_SWMR_METADATA_READ_ATTEMPTS if not set */ + /* When opening file without SWMR access, the # of read attempts is always H5F_METADATA_READ_ATTEMPTS (set or not set) */ + if(H5F_INTENT(f) & (H5F_ACC_SWMR_READ | H5F_ACC_SWMR_WRITE)) { + /* If no value for read attempts has been set, use the default */ + if(!f->shared->read_attempts) + f->shared->read_attempts = H5F_SWMR_METADATA_READ_ATTEMPTS; + + /* Turn off accumulator with SWMR */ + f->shared->feature_flags &= ~(unsigned)H5FD_FEAT_ACCUMULATE_METADATA; + if(H5FD_set_feature_flags(f->shared->lf, f->shared->feature_flags) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTSET, NULL, "can't set feature_flags in VFD") + } /* end if */ + else { + /* If no value for read attempts has been set, use the default */ + if(!f->shared->read_attempts) + f->shared->read_attempts = H5F_METADATA_READ_ATTEMPTS; + } /* end else */ /* Determine the # of bins for metdata read retries */ if(H5F_set_retries(f) < 0) @@ -831,10 +854,19 @@ H5F_dest(H5F_t *f, hid_t dxpl_id, hbool_t flush) /* Flush the file again (if requested), as shutting down the * free space manager may dirty some data structures again. */ - if(flush) + if(flush) { + /* Clear status_flags */ + f->shared->sblock->status_flags &= (uint8_t)(~H5F_SUPER_WRITE_ACCESS); + f->shared->sblock->status_flags &= (uint8_t)(~H5F_SUPER_SWMR_WRITE_ACCESS); + /* Mark superblock dirty in cache, so change will get encoded */ + /* Push error, but keep going*/ + if(H5F_super_dirty(f) < 0) + HDONE_ERROR(H5E_FILE, H5E_CANTMARKDIRTY, FAIL, "unable to mark superblock as dirty") + if(H5F_flush(f, dxpl_id, TRUE) < 0) /* Push error, but keep going*/ HDONE_ERROR(H5E_CACHE, H5E_CANTFLUSH, FAIL, "unable to flush cache") + } /* end if */ } /* end if */ /* if it exists, unpin the driver information block cache entry, @@ -979,6 +1011,37 @@ H5F_dest(H5F_t *f, hid_t dxpl_id, hbool_t flush) * The ACCESS_PARMS argument is optional. A null pointer will * cause the default file access parameters to be used. * + * The following two tables show results of file opens for single and concurrent access: + * + * SINGLE PROCESS ACCESS CONCURRENT ACCESS + * + * #1st open# #1st open# + * -- SR SR -- -- SR SR -- -- SR SR -- -- SR SR -- + * -- -- SW SW SW SW -- -- -- -- SW SW SW SW -- -- + * W W W W R R R R W W W W R R R R + * #2nd open# #2nd open# + * -------------------------- -------------------------- + * -- -- W | s x x s x x f f | -- -- W | f x x f x x f f | + * SR -- W | x x x x x x x x | SR -- W | x x x x x x x x | + * SR SW W | x x x x x x x x | SR SW W | x x x x x x x x | + * -- SW W | f x x s x x f f | -- SW W | f x x f x x f f | + * -- SW R | x x x x x x x x | -- SW R | x x x x x x x x | + * SR SW R | x x x x x x x x | SR SW R | x x x x x x x x | + * SR -- R | s x x s x x s f | SR -- R | f x x s x x s s | + * -- -- R | s x x s x x s s | -- -- R | f x x f x x s s | + * -------------------------- -------------------------- + * + * Notations: + * W: H5F_ACC_RDWR + * R: H5F_ACC_RDONLY + * SW: H5F_ACC_SWMR_WRITE + * SR: H5F_ACC_SWMR_READ + * + * x: the first open or second open itself fails due to invalid flags combination + * f: the open fails with flags combination from both the first and second opens + * s: the open succeeds with flags combination from both the first and second opens + * + * * Return: Success: A new file pointer. * Failure: NULL * @@ -998,6 +1061,8 @@ H5F_open(const char *name, unsigned flags, hid_t fcpl_id, hid_t fapl_id, H5FD_class_t *drvr; /*file driver class info */ H5P_genplist_t *a_plist; /*file access property list */ H5F_close_degree_t fc_degree; /*file close degree */ + hbool_t set_flag = FALSE; /*set the status_flags in the superblock */ + hbool_t clear = FALSE; /*clear the status_flags */ hbool_t evict_on_close; /* evict on close value from plist */ H5F_t *ret_value = NULL; /*actual return value */ char *lock_env_var = NULL;/*env var pointer */ @@ -1074,7 +1139,8 @@ H5F_open(const char *name, unsigned flags, hid_t fcpl_id, hid_t fapl_id, * file (since we can't do that while the file is open), or if the * request was to create a non-existent file (since the file already * exists), or if the new request adds write access (since the - * readers don't expect the file to change under them). + * readers don't expect the file to change under them), or if the + * SWMR write/read access flags don't agree. */ if(H5FD_close(lf) < 0) HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to close low-level file info") @@ -1085,6 +1151,11 @@ H5F_open(const char *name, unsigned flags, hid_t fcpl_id, hid_t fapl_id, if((flags & H5F_ACC_RDWR) && 0 == (shared->flags & H5F_ACC_RDWR)) HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "file is already open for read-only") + if((flags & H5F_ACC_SWMR_WRITE) && 0 == (shared->flags & H5F_ACC_SWMR_WRITE)) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "SWMR write access flag not the same for file that is already open") + if((flags & H5F_ACC_SWMR_READ) && !((shared->flags & H5F_ACC_SWMR_WRITE) || (shared->flags & H5F_ACC_SWMR_READ) || (shared->flags & H5F_ACC_RDWR))) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "SWMR read access flag not the same for file that is already open") + /* Allocate new "high-level" file struct */ if((file = H5F_new(shared, flags, fcpl_id, fapl_id, NULL)) == NULL) HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to create new file object") @@ -1116,6 +1187,10 @@ H5F_open(const char *name, unsigned flags, hid_t fcpl_id, hid_t fapl_id, /* Create the 'top' file structure */ if(NULL == (file = H5F_new(NULL, flags, fcpl_id, fapl_id, lf))) HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to initialize file structure") + + /* Need to set status_flags in the superblock if the driver has a 'lock' method */ + if(drvr->lock) + set_flag = TRUE; } /* end else */ /* Retain the name the file was opened with */ @@ -1171,6 +1246,15 @@ H5F_open(const char *name, unsigned flags, hid_t fcpl_id, hid_t fapl_id, if(H5P_get(a_plist, H5F_ACS_CLOSE_DEGREE_NAME, &fc_degree) < 0) HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, NULL, "can't get file close degree") + /* This is a private property to clear the status_flags in the super block */ + /* Use by h5clear and a routine in test/flush2.c to clear the test file's status_flags */ + if(H5P_exist_plist(a_plist, H5F_ACS_CLEAR_STATUS_FLAGS_NAME) > 0) { + if(H5P_get(a_plist, H5F_ACS_CLEAR_STATUS_FLAGS_NAME, &clear) < 0) + HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, NULL, "can't get clearance for status_flags") + else if(clear) + file->shared->sblock->status_flags = 0; + } /* end if */ + if(shared->nrefs == 1) { if(fc_degree == H5F_CLOSE_DEFAULT) shared->fc_degree = lf->cls->fc_degree; @@ -1207,6 +1291,53 @@ H5F_open(const char *name, unsigned flags, hid_t fcpl_id, hid_t fapl_id, if(H5F_build_actual_name(file, a_plist, name, &file->actual_name) < 0) HGOTO_ERROR(H5E_FILE, H5E_CANTINIT, NULL, "unable to build actual name") + if(set_flag) { + if(H5F_INTENT(file) & H5F_ACC_RDWR) { /* Set and check consistency of status_flags */ + /* Skip check of status_flags for file with < superblock version 3 */ + if(file->shared->sblock->super_vers >= HDF5_SUPERBLOCK_VERSION_3) { + + if(file->shared->sblock->status_flags & H5F_SUPER_WRITE_ACCESS || + file->shared->sblock->status_flags & H5F_SUPER_SWMR_WRITE_ACCESS) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "file is already open for write/SWMR write (may use <h5clear file> to clear file consistency flags)") + } /* version 3 superblock */ + + file->shared->sblock->status_flags |= H5F_SUPER_WRITE_ACCESS; + if(H5F_INTENT(file) & H5F_ACC_SWMR_WRITE) + file->shared->sblock->status_flags |= H5F_SUPER_SWMR_WRITE_ACCESS; + + /* Flush the superblock */ + if(H5F_super_dirty(file) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTMARKDIRTY, NULL, "unable to mark superblock as dirty") + if(H5F_flush_tagged_metadata(file, H5AC__SUPERBLOCK_TAG, H5AC_ind_read_dxpl_id) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTFLUSH, NULL, "unable to flush superblock") + + /* Remove the file lock for SWMR_WRITE */ + if(use_file_locking && (H5F_INTENT(file) & H5F_ACC_SWMR_WRITE)) { + if(H5FD_unlock(file->shared->lf) < 0) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to unlock the file") + } /* end if */ + } /* end if */ + else { /* H5F_ACC_RDONLY: check consistency of status_flags */ + /* Skip check of status_flags for file with < superblock version 3 */ + if(file->shared->sblock->super_vers >= HDF5_SUPERBLOCK_VERSION_3) { + + if(H5F_INTENT(file) & H5F_ACC_SWMR_READ) { + if((file->shared->sblock->status_flags & H5F_SUPER_WRITE_ACCESS && + !(file->shared->sblock->status_flags & H5F_SUPER_SWMR_WRITE_ACCESS)) + || + (!(file->shared->sblock->status_flags & H5F_SUPER_WRITE_ACCESS) && + file->shared->sblock->status_flags & H5F_SUPER_SWMR_WRITE_ACCESS)) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "file is not already open for SWMR writing") + + } /* end if */ + else if((file->shared->sblock->status_flags & H5F_SUPER_WRITE_ACCESS) || + (file->shared->sblock->status_flags & H5F_SUPER_SWMR_WRITE_ACCESS)) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "file is already open for write (may use <h5clear file> to clear file consistency flags)") + + } /* version 3 superblock */ + } /* end else */ + } /* end if set_flag */ + /* Success */ ret_value = file; @@ -2149,6 +2280,8 @@ H5F_get_file_image(H5F_t *file, void *buf_ptr, size_t buf_len, hid_t dxpl_id) /* test to see if a buffer was provided -- if not, we are done */ if(buf_ptr != NULL) { size_t space_needed; /* size of file image */ + hsize_t tmp; + size_t tmp_size; H5P_genplist_t *xfer_plist= NULL; /* Dataset transfer property list object */ /* Check for buffer too small */ @@ -2165,6 +2298,15 @@ H5F_get_file_image(H5F_t *file, void *buf_ptr, size_t buf_len, hid_t dxpl_id) /* (Note compensation for base address addition in internal routine) */ if(H5FD_read(fd_ptr, xfer_plist, H5FD_MEM_DEFAULT, 0, space_needed, buf_ptr) < 0) HGOTO_ERROR(H5E_FILE, H5E_READERROR, FAIL, "file image read request failed") + + /* Offset to "status_flags" in the superblock */ + tmp = H5F_SUPER_STATUS_FLAGS_OFF(file->shared->sblock->super_vers); + /* Size of "status_flags" depends on the superblock version */ + tmp_size = H5F_SUPER_STATUS_FLAGS_SIZE(file->shared->sblock->super_vers); + + /* Clear "status_flags" */ + HDmemset((uint8_t *)(buf_ptr) + tmp, 0, tmp_size); + } /* end if */ done: diff --git a/src/H5Fio.c b/src/H5Fio.c index 273ee74..e215666 100644 --- a/src/H5Fio.c +++ b/src/H5Fio.c @@ -264,26 +264,70 @@ H5F_evict_tagged_metadata(H5F_t * f, haddr_t tag, hid_t dxpl_id) FUNC_ENTER_NOAPI(FAIL) - /* Unpin the superblock, as this will be marked for eviction and it can't - be pinned. */ - if(H5AC_unpin_entry(f->shared->sblock) < 0) - HGOTO_ERROR(H5E_CACHE, H5E_CANTUNPIN, FAIL, "unable to unpin superblock") - f->shared->sblock = NULL; - /* Evict the object's metadata */ if(H5AC_evict_tagged_metadata(f, tag, TRUE, dxpl_id) < 0) HGOTO_ERROR(H5E_CACHE, H5E_CANTEXPUNGE, FAIL, "unable to evict tagged metadata") - /* Re-read the superblock. */ - if(H5F__super_read(f, dxpl_id, FALSE) < 0) - HGOTO_ERROR(H5E_FILE, H5E_READERROR, FAIL, "unable to read superblock") - done: FUNC_LEAVE_NOAPI(ret_value); } /* end H5F_evict_tagged_metadata */ /*------------------------------------------------------------------------- + * Function: H5F__evict_cache_entries + * + * Purpose: To evict all cache entries except the pinned superblock entry + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Vailin Choi + * Dec 2013 + * + *------------------------------------------------------------------------- + */ +herr_t +H5F__evict_cache_entries(H5F_t *f, hid_t dxpl_id) +{ + herr_t ret_value = SUCCEED; + + FUNC_ENTER_PACKAGE + + HDassert(f); + HDassert(f->shared); + + /* Evict all except pinned entries in the cache */ + if(H5AC_evict(f, dxpl_id) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_CANTEXPUNGE, FAIL, "unable to evict all except pinned entries") + +#ifndef NDEBUG +{ + unsigned status = 0; + int32_t cur_num_entries; + + /* Retrieve status of the superblock */ + if(H5AC_get_entry_status(f, (haddr_t)0, &status) < 0) + HGOTO_ERROR(H5E_HEAP, H5E_CANTGET, FAIL, "unable to get entry status") + + /* Verify status of the superblock entry in the cache */ + if(!(status & H5AC_ES__IN_CACHE) || !(status & H5AC_ES__IS_PINNED)) + HGOTO_ERROR(H5E_HEAP, H5E_CANTGET, FAIL, "unable to get entry status") + + /* Get the number of cache entries */ + if(H5AC_get_cache_size(f->shared->cache, NULL, NULL, NULL, &cur_num_entries) < 0) + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, "H5AC_get_cache_size() failed.") + + /* Should be the only one left in the cache (the superblock) */ + if(cur_num_entries != 1) + HGOTO_ERROR(H5E_CACHE, H5E_SYSTEM, FAIL, "number of cache entries is not correct") +} +#endif /* NDEBUG */ + +done: + FUNC_LEAVE_NOAPI(ret_value); +} /* end H5F__evict_cache_entries() */ + + +/*------------------------------------------------------------------------- * Function: H5F_get_checksums * * Purpose: Decode checksum stored in the buffer diff --git a/src/H5Fpkg.h b/src/H5Fpkg.h index e9c45f3..1adf74b 100644 --- a/src/H5Fpkg.h +++ b/src/H5Fpkg.h @@ -56,10 +56,11 @@ /* Superblock status flags */ #define H5F_SUPER_WRITE_ACCESS 0x01 #define H5F_SUPER_FILE_OK 0x02 -#define H5F_SUPER_ALL_FLAGS (H5F_SUPER_WRITE_ACCESS | H5F_SUPER_FILE_OK) +#define H5F_SUPER_SWMR_WRITE_ACCESS 0x04 +#define H5F_SUPER_ALL_FLAGS (H5F_SUPER_WRITE_ACCESS | H5F_SUPER_FILE_OK | H5F_SUPER_SWMR_WRITE_ACCESS) /* Mask for removing private file access flags */ -#define H5F_ACC_PUBLIC_FLAGS 0x001fu +#define H5F_ACC_PUBLIC_FLAGS 0x007fu /* Free space section+aggregator merge flags */ #define H5F_FS_MERGE_METADATA 0x01 /* Section can merge with metadata aggregator */ @@ -131,12 +132,28 @@ #define H5F_SUPERBLOCK_VARLEN_SIZE(v, sizeof_addr, sizeof_size) ( \ (v == 0 ? H5F_SUPERBLOCK_VARLEN_SIZE_V0(sizeof_addr, sizeof_size) : 0) \ + (v == 1 ? H5F_SUPERBLOCK_VARLEN_SIZE_V1(sizeof_addr, sizeof_size) : 0) \ - + (v == 2 ? H5F_SUPERBLOCK_VARLEN_SIZE_V2(sizeof_addr) : 0)) + + (v >= 2 ? H5F_SUPERBLOCK_VARLEN_SIZE_V2(sizeof_addr) : 0)) /* Total size of superblock, depends on superblock version */ #define H5F_SUPERBLOCK_SIZE(s) ( H5F_SUPERBLOCK_FIXED_SIZE \ + H5F_SUPERBLOCK_VARLEN_SIZE((s)->super_vers, (s)->sizeof_addr, (s)->sizeof_size)) +/* For superblock version 0 & 1: + Offset to the file consistency flags (status_flags) in the superblock (excluding H5F_SUPERBLOCK_FIXED_SIZE) */ +#define H5F_SUPER_STATUS_OFF_V01 \ + (2 /* freespace, and root group versions */ \ + + 1 /* reserved */ \ + + 3 /* shared header vers, size of address, size of lengths */ \ + + 1 /* reserved */ \ + + 4) /* group leaf k, group internal k */ + +#define H5F_SUPER_STATUS_OFF(v) (v >= 2 ? 2 : H5F_SUPER_STATUS_OFF_V01) + +/* Offset to the file consistency flags (status_flags) in the superblock */ +#define H5F_SUPER_STATUS_FLAGS_OFF(v) (H5F_SUPERBLOCK_FIXED_SIZE + H5F_SUPER_STATUS_OFF(v)) + +/* Size of file consistency flags (status_flags) in the superblock */ +#define H5F_SUPER_STATUS_FLAGS_SIZE(v) (v >= 2 ? 1 : 4) /* Forward declaration external file cache struct used below (defined in * H5Fefc.c) */ @@ -406,6 +423,9 @@ H5_DLL herr_t H5F_efc_try_close(H5F_t *f); H5_DLL herr_t H5F__set_eoa(const H5F_t *f, H5F_mem_t type, haddr_t addr); H5_DLL herr_t H5F__set_base_addr(const H5F_t *f, haddr_t addr); +/* Functions that flush or evict */ +H5_DLL herr_t H5F__evict_cache_entries(H5F_t *f, hid_t dxpl_id); + /* Testing functions */ #ifdef H5F_TESTING H5_DLL herr_t H5F_get_sohm_mesg_count_test(hid_t fid, unsigned type_id, diff --git a/src/H5Fprivate.h b/src/H5Fprivate.h index 50cd763..a6d1c4a 100644 --- a/src/H5Fprivate.h +++ b/src/H5Fprivate.h @@ -473,6 +473,7 @@ #define H5F_ACS_OBJECT_FLUSH_CB_NAME "object_flush_cb" /* Object flush callback */ #define H5F_ACS_EFC_SIZE_NAME "efc_size" /* Size of external file cache */ #define H5F_ACS_FILE_IMAGE_INFO_NAME "file_image_info" /* struct containing initial file image and callback info */ +#define H5F_ACS_CLEAR_STATUS_FLAGS_NAME "clear_status_flags" /* Whether to clear superblock status_flags (private property only used by h5clear) */ #define H5F_ACS_USE_MDC_LOGGING_NAME "use_mdc_logging" /* Whether to use metadata cache logging */ #define H5F_ACS_MDC_LOG_LOCATION_NAME "mdc_log_location" /* Name of metadata cache log location */ #define H5F_ACS_START_MDC_LOG_ON_ACCESS_NAME "start_mdc_log_on_access" /* Whether logging starts on file create/open */ @@ -498,7 +499,8 @@ #define HDF5_SUPERBLOCK_VERSION_DEF 0 /* The default super block format */ #define HDF5_SUPERBLOCK_VERSION_1 1 /* Version with non-default B-tree 'K' value */ #define HDF5_SUPERBLOCK_VERSION_2 2 /* Revised version with superblock extension and checksum */ -#define HDF5_SUPERBLOCK_VERSION_LATEST HDF5_SUPERBLOCK_VERSION_2 /* The maximum super block format */ +#define HDF5_SUPERBLOCK_VERSION_3 3 /* With file locking and consistency flags (at least this version for SWMR support) */ +#define HDF5_SUPERBLOCK_VERSION_LATEST HDF5_SUPERBLOCK_VERSION_3 /* The maximum super block format */ #define HDF5_SUPERBLOCK_VERSION_V18_LATEST HDF5_SUPERBLOCK_VERSION_2 /* The latest superblock version for v18 */ #define HDF5_FREESPACE_VERSION 0 /* of the Free-Space Info */ #define HDF5_OBJECTDIR_VERSION 0 /* of the Object Directory format */ @@ -523,6 +525,7 @@ /* Metadata read attempt values */ #define H5F_METADATA_READ_ATTEMPTS 1 /* Default # of read attempts for non-SWMR access */ +#define H5F_SWMR_METADATA_READ_ATTEMPTS 100 /* Default # of read attempts for SWMR access */ /* Macros to define signatures of all objects in the file */ diff --git a/src/H5Fpublic.h b/src/H5Fpublic.h index 018618e..a79da75 100644 --- a/src/H5Fpublic.h +++ b/src/H5Fpublic.h @@ -55,6 +55,23 @@ #define H5F_ACC_EXCL (H5CHECK H5OPEN 0x0004u) /*fail if file already exists*/ /* NOTE: 0x0008u was H5F_ACC_DEBUG, now deprecated */ #define H5F_ACC_CREAT (H5CHECK H5OPEN 0x0010u) /*create non-existing files */ +#define H5F_ACC_SWMR_WRITE (H5CHECK 0x0020u) /*indicate that this file is + * open for writing in a + * single-writer/multi-reader (SWMR) + * scenario. Note that the + * process(es) opening the file + * for reading must open the file + * with RDONLY access, and use + * the special "SWMR_READ" access + * flag. */ +#define H5F_ACC_SWMR_READ (H5CHECK 0x0040u) /*indicate that this file is + * open for reading in a + * single-writer/multi-reader (SWMR) + * scenario. Note that the + * process(es) opening the file + * for SWMR reading must also + * open the file with the RDONLY + * flag. */ /* Value passed to H5Pset_elink_acc_flags to cause flags to be taken from the * parent file. */ @@ -225,6 +242,7 @@ H5_DLL herr_t H5Freset_mdc_hit_rate_stats(hid_t file_id); H5_DLL ssize_t H5Fget_name(hid_t obj_id, char *name, size_t size); H5_DLL herr_t H5Fget_info2(hid_t obj_id, H5F_info2_t *finfo); H5_DLL herr_t H5Fget_metadata_read_retry_info(hid_t file_id, H5F_retry_info_t *info); +H5_DLL herr_t H5Fstart_swmr_write(hid_t file_id); H5_DLL ssize_t H5Fget_free_sections(hid_t file_id, H5F_mem_t type, size_t nsects, H5F_sect_info_t *sect_info/*out*/); H5_DLL herr_t H5Fclear_elink_file_cache(hid_t file_id); diff --git a/src/H5Fsuper.c b/src/H5Fsuper.c index 489a149..2a82618 100644 --- a/src/H5Fsuper.c +++ b/src/H5Fsuper.c @@ -250,6 +250,7 @@ H5F__super_read(H5F_t *f, hid_t dxpl_id, hbool_t initial_read) haddr_t super_addr; /* Absolute address of superblock */ haddr_t eof; /* End of file address */ unsigned rw_flags; /* Read/write permissions for file */ + hbool_t skip_eof_check = FALSE; /* Whether to skip checking the EOF value */ herr_t ret_value = SUCCEED; /* Return value */ FUNC_ENTER_PACKAGE_TAG(dxpl_id, H5AC__SUPERBLOCK_TAG, FAIL) @@ -317,6 +318,14 @@ H5F__super_read(H5F_t *f, hid_t dxpl_id, hbool_t initial_read) if(NULL == (sblock = (H5F_super_t *)H5AC_protect(f, dxpl_id, H5AC_SUPERBLOCK, (haddr_t)0, &udata, rw_flags))) HGOTO_ERROR(H5E_FILE, H5E_CANTPROTECT, FAIL, "unable to load superblock") + if(H5F_INTENT(f) & H5F_ACC_SWMR_WRITE) + if(sblock->super_vers < HDF5_SUPERBLOCK_VERSION_3) + HGOTO_ERROR(H5E_FILE, H5E_CANTPROTECT, FAIL, "invalid superblock version for SWMR_WRITE") + + /* Enable all latest version support when file has v3 superblock */ + if(sblock->super_vers >= HDF5_SUPERBLOCK_VERSION_3) + f->shared->latest_flags |= H5F_LATEST_ALL_FLAGS; + /* Pin the superblock in the cache */ if(H5AC_pin_protected_entry(sblock) < 0) HGOTO_ERROR(H5E_FILE, H5E_CANTPIN, FAIL, "unable to pin superblock") @@ -410,7 +419,22 @@ H5F__super_read(H5F_t *f, hid_t dxpl_id, hbool_t initial_read) * Note: the aggregator is changed again after being reset * earlier before H5AC_flush due to allocation of tmp addresses. */ - if(initial_read) { + /* The EOF check must be skipped when the file is opened for SWMR read, + * as the file can appear truncated if only part of it has been + * been flushed to disk by the SWMR writer process. + */ + if(H5F_INTENT(f) & H5F_ACC_SWMR_READ) { + /* + * When the file is opened for SWMR read access, skip the check if: + * --the file is already marked for SWMR writing and + * --the file has version 3 superblock for SWMR support + */ + if((sblock->status_flags & H5F_SUPER_SWMR_WRITE_ACCESS) && + (sblock->status_flags & H5F_SUPER_WRITE_ACCESS) && + sblock->super_vers >= HDF5_SUPERBLOCK_VERSION_3) + skip_eof_check = TRUE; + } /* end if */ + if(!skip_eof_check && initial_read) { if(HADDR_UNDEF == (eof = H5FD_get_eof(f->shared->lf, H5FD_MEM_DEFAULT))) HGOTO_ERROR(H5E_FILE, H5E_CANTGET, FAIL, "unable to determine file size") diff --git a/src/H5Gent.c b/src/H5Gent.c index 6020028..f64eaf0 100644 --- a/src/H5Gent.c +++ b/src/H5Gent.c @@ -460,7 +460,7 @@ H5G__ent_convert(H5F_t *f, hid_t dxpl_id, H5HL_t *heap, const char *name, targ_oloc.addr = lnk->u.hard.addr; /* Get the object header */ - if(NULL == (oh = H5O_protect(&targ_oloc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(&targ_oloc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_SYM, H5E_CANTPROTECT, FAIL, "unable to protect target object header") /* Check if a symbol table message exists */ diff --git a/src/H5Gtest.c b/src/H5Gtest.c index d69a804..0e0a897 100644 --- a/src/H5Gtest.c +++ b/src/H5Gtest.c @@ -701,7 +701,7 @@ H5G_verify_cached_stabs_test_cb(H5F_t *f, hid_t dxpl_id, targ_oloc.addr = sn->entry[i].header; /* Load target object header */ - if(NULL == (targ_oh = H5O_protect(&targ_oloc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (targ_oh = H5O_protect(&targ_oloc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_SYM, H5E_CANTPROTECT, H5_ITER_ERROR, "unable to protect target object header") /* Check if a symbol table message exists */ diff --git a/src/H5Lexternal.c b/src/H5Lexternal.c index 12cce73..7f59e50 100644 --- a/src/H5Lexternal.c +++ b/src/H5Lexternal.c @@ -323,7 +323,7 @@ H5L_extern_traverse(const char H5_ATTR_UNUSED *link_name, hid_t cur_group, */ /* Simplify intent flags for open calls */ - intent = ((intent & H5F_ACC_RDWR) ? H5F_ACC_RDWR : H5F_ACC_RDONLY); + intent &= (H5F_ACC_RDWR | H5F_ACC_SWMR_WRITE | H5F_ACC_SWMR_READ); /* Copy the file name to use */ if(NULL == (temp_file_name = H5MM_strdup(file_name))) @@ -1281,6 +1281,7 @@ H5O_create(H5F_t *f, hid_t dxpl_id, size_t size_hint, size_t initial_rc, oh->version = H5O_VERSION_1; oh->sizeof_size = H5F_SIZEOF_SIZE(f); oh->sizeof_addr = H5F_SIZEOF_ADDR(f); + oh->swmr_write = !!(H5F_INTENT(f) & H5F_ACC_SWMR_WRITE); #ifdef H5O_ENABLE_BAD_MESG_COUNT /* Check whether the "bad message count" property is set */ if(H5P_exist_plist(oc_plist, H5O_BAD_MESG_COUNT_NAME) > 0) { @@ -1290,6 +1291,15 @@ H5O_create(H5F_t *f, hid_t dxpl_id, size_t size_hint, size_t initial_rc, } /* end if */ #endif /* H5O_ENABLE_BAD_MESG_COUNT */ + /* Create object header proxy if doing SWMR writes */ + if(oh->swmr_write) { + /* Create virtual entry, for use as proxy */ + if(NULL == (oh->proxy = H5AC_proxy_entry_create())) + HGOTO_ERROR(H5E_OHDR, H5E_CANTCREATE, FAIL, "can't create object header proxy") + } /* end if */ + else + oh->proxy = NULL; + /* Set initial status flags */ oh->flags = oh_flags; @@ -1356,6 +1366,7 @@ H5O_create(H5F_t *f, hid_t dxpl_id, size_t size_hint, size_t initial_rc, /* (including space for serializing the object header prefix */ if(NULL == (oh->chunk[0].image = H5FL_BLK_CALLOC(chunk_image, oh_size))) HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL, "memory allocation failed") + oh->chunk[0].chunk_proxy = NULL; /* Put magic # for object header in first chunk */ if(oh->version > H5O_VERSION_1) @@ -1791,7 +1802,8 @@ done: *------------------------------------------------------------------------- */ H5O_t * -H5O_protect(const H5O_loc_t *loc, hid_t dxpl_id, unsigned prot_flags) +H5O_protect(const H5O_loc_t *loc, hid_t dxpl_id, unsigned prot_flags, + hbool_t pin_all_chunks) { H5O_t *oh = NULL; /* Object header protected */ H5O_cache_ud_t udata; /* User data for protecting object header */ @@ -1915,6 +1927,39 @@ H5O_protect(const H5O_loc_t *loc, hid_t dxpl_id, unsigned prot_flags) H5O_assert(oh); #endif /* H5O_DEBUG */ + /* Pin the other chunks also when requested, so that the object header + * proxy can be set up. + */ + if(pin_all_chunks && oh->nchunks > 1) { + unsigned u; /* Local index variable */ + + /* Sanity check */ + HDassert(oh->swmr_write); + + /* Iterate over chunks > 0 */ + for(u = 1; u < oh->nchunks; u++) { + H5O_chunk_proxy_t *chk_proxy; /* Chunk proxy */ + + /* Protect chunk */ + if(NULL == (chk_proxy = H5O_chunk_protect(loc->file, dxpl_id, oh, u))) + HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, NULL, "unable to protect object header chunk") + + /* Pin chunk proxy*/ + if(H5AC_pin_protected_entry(chk_proxy) < 0 ) + HGOTO_ERROR(H5E_OHDR, H5E_CANTPIN, NULL, "unable to pin object header chunk") + + /* Unprotect chunk */ + if(H5O_chunk_unprotect(loc->file, dxpl_id, chk_proxy, FALSE) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTUNPROTECT, NULL, "unable to unprotect object header chunk") + + /* Preserve chunk proxy pointer for later */ + oh->chunk[u].chunk_proxy = chk_proxy; + } /* end for */ + + /* Set the flag for the unprotect */ + oh->chunks_pinned = TRUE; + } /* end if */ + /* Set return value */ ret_value = oh; @@ -1956,7 +2001,7 @@ H5O_pin(const H5O_loc_t *loc, hid_t dxpl_id) HDassert(loc); /* Get header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__NO_FLAGS_SET))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__NO_FLAGS_SET, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, NULL, "unable to protect object header") /* Increment the reference count on the object header */ @@ -2038,6 +2083,27 @@ H5O_unprotect(const H5O_loc_t *loc, hid_t dxpl_id, H5O_t *oh, unsigned oh_flags) HDassert(loc); HDassert(oh); + /* Unpin the other chunks */ + if(oh->chunks_pinned && oh->nchunks > 1) { + unsigned u; /* Local index variable */ + + /* Sanity check */ + HDassert(oh->swmr_write); + + /* Iterate over chunks > 0 */ + for(u = 1; u < oh->nchunks; u++) { + if(NULL != oh->chunk[u].chunk_proxy) { + /* Release chunk proxy */ + if(H5AC_unpin_entry(oh->chunk[u].chunk_proxy) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTUNPIN, FAIL, "unable to unpin object header chunk") + oh->chunk[u].chunk_proxy = NULL; + } /* end if */ + } /* end for */ + + /* Reet the flag from the unprotect */ + oh->chunks_pinned = FALSE; + } /* end if */ + /* Unprotect the object header */ if(H5AC_unprotect(loc->file, dxpl_id, H5AC_OHDR, oh->chunk[0].addr, oh, oh_flags) < 0) HGOTO_ERROR(H5E_OHDR, H5E_CANTUNPROTECT, FAIL, "unable to release object header") @@ -2169,7 +2235,7 @@ H5O_touch(const H5O_loc_t *loc, hbool_t force, hid_t dxpl_id) HDassert(loc); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__NO_FLAGS_SET))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__NO_FLAGS_SET, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Create/Update the modification time message */ @@ -2295,7 +2361,7 @@ H5O_delete(H5F_t *f, hid_t dxpl_id, haddr_t addr) loc.holding_file = FALSE; /* Get the object header information */ - if(NULL == (oh = H5O_protect(&loc, dxpl_id, H5AC__NO_FLAGS_SET))) + if(NULL == (oh = H5O_protect(&loc, dxpl_id, H5AC__NO_FLAGS_SET, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Delete object */ @@ -2386,7 +2452,7 @@ H5O_obj_type(const H5O_loc_t *loc, H5O_type_t *obj_type, hid_t dxpl_id) FUNC_ENTER_NOAPI_TAG(dxpl_id, loc->addr, FAIL) /* Load the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Retrieve the type of the object */ @@ -2464,7 +2530,7 @@ H5O_obj_class(const H5O_loc_t *loc, hid_t dxpl_id) FUNC_ENTER_NOAPI_NOINIT_TAG(dxpl_id, loc->addr, NULL) /* Load the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, NULL, "unable to load object header") /* Test whether entry qualifies as a particular type of object */ @@ -2761,7 +2827,7 @@ H5O_get_hdr_info(const H5O_loc_t *loc, hid_t dxpl_id, H5O_hdr_info_t *hdr) HDmemset(hdr, 0, sizeof(*hdr)); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTLOAD, FAIL, "unable to load object header") /* Get the information for the object header */ @@ -2885,7 +2951,7 @@ H5O_get_info(const H5O_loc_t *loc, hid_t dxpl_id, hbool_t want_ih_info, HDassert(oinfo); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Reset the object info structure */ @@ -3006,7 +3072,7 @@ H5O_get_create_plist(const H5O_loc_t *loc, hid_t dxpl_id, H5P_genplist_t *oc_pli HDassert(oc_plist); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Set property values, if they were used for the object */ @@ -3061,7 +3127,7 @@ H5O_get_nlinks(const H5O_loc_t *loc, hid_t dxpl_id, hsize_t *nlinks) HDassert(nlinks); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Retrieve the # of link messages seen when the object header was loaded */ @@ -3178,7 +3244,7 @@ H5O_get_rc_and_type(const H5O_loc_t *loc, hid_t dxpl_id, unsigned *rc, H5O_type_ HDassert(loc); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Set the object's reference count */ @@ -3551,7 +3617,7 @@ H5O_dec_rc_by_loc(const H5O_loc_t *loc, hid_t dxpl_id) HDassert(loc); /* Get header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to protect object header") /* Decrement the reference count on the object header */ @@ -3569,6 +3635,30 @@ done: /*------------------------------------------------------------------------- + * Function: H5O_get_proxy + * + * Purpose: Retrieve the proxy for the object header. + * + * Return: Non-negative on success/Negative on failure + * + * Programmer: Quincey Koziol + * July 24 2016 + * + *------------------------------------------------------------------------- + */ +H5AC_proxy_entry_t * +H5O_get_proxy(const H5O_t *oh) +{ + FUNC_ENTER_NOAPI_NOINIT_NOERR + + /* Check args */ + HDassert(oh); + + FUNC_LEAVE_NOAPI(oh->proxy) +} /* end H5O_get_proxy() */ + + +/*------------------------------------------------------------------------- * Function: H5O__free * * Purpose: Destroys an object header. @@ -3585,8 +3675,9 @@ herr_t H5O__free(H5O_t *oh) { unsigned u; /* Local index variable */ + herr_t ret_value = SUCCEED; /* Return value */ - FUNC_ENTER_PACKAGE_NOERR + FUNC_ENTER_PACKAGE /* check args */ HDassert(oh); @@ -3621,9 +3712,15 @@ H5O__free(H5O_t *oh) oh->mesg = (H5O_mesg_t *)H5FL_SEQ_FREE(H5O_mesg_t, oh->mesg); } /* end if */ + /* Destroy the proxy */ + if(oh->proxy) + if(H5AC_proxy_entry_dest(oh->proxy) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTFREE, FAIL, "unable to destroy virtual entry used for proxy") + /* destroy object header */ oh = H5FL_FREE(H5O_t, oh); - FUNC_LEAVE_NOAPI(SUCCEED) +done: + FUNC_LEAVE_NOAPI(ret_value) } /* end H5O__free() */ diff --git a/src/H5Oalloc.c b/src/H5Oalloc.c index d22fd92..4f98cfa 100644 --- a/src/H5Oalloc.c +++ b/src/H5Oalloc.c @@ -745,9 +745,8 @@ H5O__alloc_find_best_nonnull(const H5F_t *f, const H5O_t *oh, size_t *size, size_t cont_size; /* Continuation message size */ size_t multi_size; /* Size of all the messages in the last chunk */ unsigned u; /* Local index variable */ - herr_t ret_value = SUCCEED; /* Return value */ - FUNC_ENTER_STATIC + FUNC_ENTER_STATIC_NOERR /* Check args */ HDassert(f); @@ -865,8 +864,7 @@ H5O__alloc_find_best_nonnull(const H5F_t *f, const H5O_t *oh, size_t *size, else *size += (size_t)H5O_SIZEOF_MSGHDR_OH(oh) + oh->mesg[found_msg->msgno].raw_size; -done: - FUNC_LEAVE_NOAPI(ret_value) + FUNC_LEAVE_NOAPI(SUCCEED) } /* H5O__alloc_find_best_nonnull() */ @@ -949,6 +947,7 @@ H5O__alloc_chunk(H5F_t *f, hid_t dxpl_id, H5O_t *oh, size_t size, oh->chunk[chunkno].gap = 0; if(NULL == (oh->chunk[chunkno].image = p = H5FL_BLK_CALLOC(chunk_image, size))) HGOTO_ERROR(H5E_OHDR, H5E_CANTALLOC, FAIL, "can't allocate image for chunk, size = %zu", size) + oh->chunk[chunkno].chunk_proxy = NULL; /* If this is a later version of the object header format, put the magic * # at the beginning of the chunk image. @@ -1221,10 +1220,9 @@ static herr_t H5O__alloc_find_best_null(const H5O_t *oh, size_t size, size_t *mesg_idx) { size_t idx; /* Index of message which fits allocation */ - int found_null; /* Best fit null message */ - herr_t ret_value = SUCCEED; /* Return value */ + ssize_t found_null; /* Best fit null message */ - FUNC_ENTER_STATIC + FUNC_ENTER_STATIC_NOERR /* check args */ HDassert(oh); @@ -1239,11 +1237,11 @@ H5O__alloc_find_best_null(const H5O_t *oh, size_t size, size_t *mesg_idx) if(oh->mesg[idx].raw_size == size) { /* Keep first exact fit */ if(found_null < 0) - found_null = idx; + found_null = (ssize_t)idx; else /* If we've got more than one exact fit, choose the one in the earliest chunk */ if(oh->mesg[idx].chunkno < oh->mesg[found_null].chunkno) { - found_null = idx; + found_null = (ssize_t)idx; /* If we found an exact fit in object header chunk #0, we can get out */ /* (Could extend this to look for earliest message in @@ -1257,29 +1255,26 @@ H5O__alloc_find_best_null(const H5O_t *oh, size_t size, size_t *mesg_idx) else if(oh->mesg[idx].raw_size > size) { /* Keep first one found */ if(found_null < 0) - found_null = idx; + found_null = (ssize_t)idx; else /* Check for better fit */ if(oh->mesg[idx].raw_size < oh->mesg[found_null].raw_size) - found_null = idx; + found_null = (ssize_t)idx; else { /* If they are the same size, choose the one in the earliest chunk */ if(oh->mesg[idx].raw_size == oh->mesg[found_null].raw_size) { if(oh->mesg[idx].chunkno < oh->mesg[found_null].chunkno) - found_null = idx; + found_null = (ssize_t)idx; } /* end if */ } /* end else */ } /* end else-if */ - /* Ignore too-small null messages */ - else - ; + /* else: Ignore too-small null messages */ } /* end if */ } /* end for */ if(found_null >= 0) - *mesg_idx = found_null; + *mesg_idx = (size_t)found_null; -done: - FUNC_LEAVE_NOAPI(ret_value) + FUNC_LEAVE_NOAPI(SUCCEED) } /* H5O__alloc_find_best_null() */ @@ -1627,6 +1622,7 @@ H5O_move_msgs_forward(H5F_t *f, hid_t dxpl_id, H5O_t *oh) { H5O_chunk_proxy_t *null_chk_proxy = NULL; /* Chunk that null message is in */ H5O_chunk_proxy_t *curr_chk_proxy = NULL; /* Chunk that message is in */ + H5O_chunk_proxy_t *cont_targ_chk_proxy = NULL; /* Chunk that continuation message points to */ hbool_t null_chk_dirtied = FALSE; /* Flags for unprotecting null chunk */ hbool_t curr_chk_dirtied = FALSE; /* Flags for unprotecting curr chunk */ hbool_t packed_msg; /* Flag to indicate that messages were packed */ @@ -1741,6 +1737,60 @@ H5O_move_msgs_forward(H5F_t *f, hid_t dxpl_id, H5O_t *oh) if(NULL == (curr_chk_proxy = H5O_chunk_protect(f, dxpl_id, oh, curr_msg->chunkno))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header chunk") + /* If the message being moved is a continuation + * message and we are doing SWMR writes, we must + * update the flush dependencies */ + if(oh->swmr_write && (H5O_CONT_ID == curr_msg->type->id)) { + void *null_chk_mdc_obj; /* The metadata cache object for the null_msg chunk */ + + /* Point to the metadata cache object for the + * null message chunk, oh if in chunk 0 or the + * proxy otherwise */ + null_chk_mdc_obj = (null_msg->chunkno == 0 + ? (void *)oh + : (void *)null_chk_proxy); + + /* The other chunks involved should never be + * chunk 0 */ + HDassert(curr_msg->chunkno > 0); + HDassert(((H5O_cont_t *)(curr_msg->native))->chunkno > 0); + + /* Protect continuation message target chunk */ + if(NULL == (cont_targ_chk_proxy = H5O_chunk_protect(f, dxpl_id, oh, ((H5O_cont_t *)(curr_msg->native))->chunkno))) + HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header chunk") + + /* Remove flush dependency on old continuation + * message chunk */ + HDassert(cont_targ_chk_proxy); + HDassert(cont_targ_chk_proxy->parent); + HDassert(curr_chk_proxy); + HDassert((void *)curr_chk_proxy == cont_targ_chk_proxy->parent); + + if(H5AC_destroy_flush_dependency(curr_chk_proxy, cont_targ_chk_proxy) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency") + + cont_targ_chk_proxy->parent = NULL; + + /* Create flush dependency on new continuation + * message chunk */ + if(H5AC_create_flush_dependency(null_chk_mdc_obj, cont_targ_chk_proxy) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTDEPEND, FAIL, "unable to create flush dependency") + + HDassert(null_chk_mdc_obj); + HDassert(((H5C_cache_entry_t *)null_chk_mdc_obj)->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC); + HDassert(((H5C_cache_entry_t *)null_chk_mdc_obj)->type); + HDassert((((H5C_cache_entry_t *)null_chk_mdc_obj)->type->id == H5AC_OHDR_ID) || + (((H5C_cache_entry_t *)null_chk_mdc_obj)->type->id == H5AC_OHDR_CHK_ID)); + + cont_targ_chk_proxy->parent = null_chk_mdc_obj; + + /* Unprotect continuation message target chunk + */ + if(H5O_chunk_unprotect(f, dxpl_id, cont_targ_chk_proxy, FALSE) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTUNPROTECT, FAIL, "unable to unprotect object header chunk") + cont_targ_chk_proxy = NULL; + } /* end if */ + /* Copy raw data for non-null message to new chunk */ HDmemcpy(null_msg->raw - H5O_SIZEOF_MSGHDR_OH(oh), curr_msg->raw - H5O_SIZEOF_MSGHDR_OH(oh), curr_msg->raw_size + (size_t)H5O_SIZEOF_MSGHDR_OH(oh)); @@ -1889,9 +1939,11 @@ done: HDONE_ERROR(H5E_OHDR, H5E_CANTUNPROTECT, FAIL, "unable to unprotect null object header chunk") if(curr_chk_proxy && H5O_chunk_unprotect(f, dxpl_id, curr_chk_proxy, curr_chk_dirtied) < 0) HDONE_ERROR(H5E_OHDR, H5E_CANTUNPROTECT, FAIL, "unable to unprotect current object header chunk") + if(cont_targ_chk_proxy && H5O_chunk_unprotect(f, dxpl_id, cont_targ_chk_proxy, FALSE) < 0) + HDONE_ERROR(H5E_OHDR, H5E_CANTUNPROTECT, FAIL, "unable to unprotect continuation message target object header chunk") } /* end if */ else - HDassert(!null_chk_proxy && !curr_chk_proxy); + HDassert(!null_chk_proxy && !curr_chk_proxy && !cont_targ_chk_proxy); FUNC_LEAVE_NOAPI(ret_value) } /* H5O_move_msgs_forward() */ diff --git a/src/H5Oattribute.c b/src/H5Oattribute.c index 9b494a1..71f54eb 100644 --- a/src/H5Oattribute.c +++ b/src/H5Oattribute.c @@ -484,7 +484,7 @@ H5O_attr_open_by_name(const H5O_loc_t *loc, const char *name, hid_t dxpl_id) HDassert(name); /* Protect the object header to iterate over */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_ATTR, H5E_CANTPROTECT, NULL, "unable to load object header") /* Check for attribute info stored */ @@ -633,7 +633,7 @@ H5O_attr_open_by_idx(const H5O_loc_t *loc, H5_index_t idx_type, HGOTO_ERROR(H5E_ATTR, H5E_BADITER, NULL, "can't locate attribute") /* Protect the object header to iterate over */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_ATTR, H5E_CANTPROTECT, NULL, "unable to load object header") /* Find out whether it has already been opened. If it has, close the object @@ -1284,7 +1284,7 @@ H5O_attr_iterate_real(hid_t loc_id, const H5O_loc_t *loc, hid_t dxpl_id, HDassert(attr_op); /* Protect the object header to iterate over */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_ATTR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Check for attribute info stored */ @@ -1847,7 +1847,7 @@ H5O_attr_exists(const H5O_loc_t *loc, const char *name, hid_t dxpl_id) HDassert(name); /* Protect the object header to iterate over */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_ATTR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Check for attribute info stored */ @@ -2001,7 +2001,7 @@ H5O_attr_count(const H5O_loc_t *loc, hid_t dxpl_id) HDassert(loc); /* Protect the object header to iterate over */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_ATTR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Retrieve # of attributes on object */ diff --git a/src/H5Ocache.c b/src/H5Ocache.c index 88a4d85..94498ee 100644 --- a/src/H5Ocache.c +++ b/src/H5Ocache.c @@ -441,6 +441,18 @@ H5O__cache_deserialize(const void *_image, size_t len, void *_udata, /* Retrieve partially deserialized object header from user data */ oh = udata->oh; + /* Set SWMR flag, if appropriate */ + oh->swmr_write = !!(H5F_INTENT(udata->common.f) & H5F_ACC_SWMR_WRITE); + + /* Create object header proxy if doing SWMR writes */ + if(oh->swmr_write) { + /* Create virtual entry, for use as proxy */ + if(NULL == (oh->proxy = H5AC_proxy_entry_create())) + HGOTO_ERROR(H5E_OHDR, H5E_CANTCREATE, NULL, "can't create object header proxy") + } /* end if */ + else + oh->proxy = NULL; + /* Parse the first chunk */ if(H5O__chunk_deserialize(oh, udata->common.addr, udata->chunk0_size, (const uint8_t *)_image, &(udata->common), dirty) < 0) HGOTO_ERROR(H5E_OHDR, H5E_CANTINIT, NULL, "can't deserialize first object header chunk") @@ -667,6 +679,16 @@ H5O__cache_notify(H5AC_notify_action_t action, void *_thing) switch(action) { case H5AC_NOTIFY_ACTION_AFTER_INSERT: case H5AC_NOTIFY_ACTION_AFTER_LOAD: + if(oh->swmr_write) { + /* Sanity check */ + HDassert(oh->proxy); + + /* Register the object header as a parent of the virtual entry */ + if(H5AC_proxy_entry_add_parent(oh->proxy, oh) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTSET, FAIL, "can't add object header as parent of proxy") + } /* end if */ + break; + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: /* do nothing */ @@ -689,10 +711,17 @@ H5O__cache_notify(H5AC_notify_action_t action, void *_thing) case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: case H5AC_NOTIFY_ACTION_CHILD_CLEANED: - case H5AC_NOTIFY_ACTION_BEFORE_EVICT: /* do nothing */ break; + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + if(oh->swmr_write) { + /* Unregister the object header as a parent of the virtual entry */ + if(H5AC_proxy_entry_remove_parent(oh->proxy, oh) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTSET, FAIL, "can't remove object header as parent of proxy") + } /* end if */ + break; + default: #ifdef NDEBUG HGOTO_ERROR(H5E_OHDR, H5E_BADVALUE, FAIL, "unknown action from metadata cache") @@ -998,7 +1027,6 @@ static herr_t H5O__cache_chk_notify(H5AC_notify_action_t action, void *_thing) { H5O_chunk_proxy_t *chk_proxy = (H5O_chunk_proxy_t *)_thing; - void *parent = NULL; /* Chunk containing continuation message that points to this chunk */ H5O_chunk_proxy_t *cont_chk_proxy = NULL; /* Proxy for chunk containing continuation message that points to this chunk, if not chunk 0 */ herr_t ret_value = SUCCEED; /* Return value */ @@ -1013,6 +1041,51 @@ H5O__cache_chk_notify(H5AC_notify_action_t action, void *_thing) switch(action) { case H5AC_NOTIFY_ACTION_AFTER_INSERT: case H5AC_NOTIFY_ACTION_AFTER_LOAD: + if(chk_proxy->oh->swmr_write) { + /* Add flush dependency on chunk parent */ + { + void *parent; /* Chunk containing continuation message that points to this chunk */ + + /* Determine the parent of the chunk */ + if(chk_proxy->cont_chunkno == 0) + parent = chk_proxy->oh; + else { + if(NULL == (cont_chk_proxy = H5O_chunk_protect(chk_proxy->f, H5AC_ind_read_dxpl_id, chk_proxy->oh, chk_proxy->cont_chunkno))) + HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header chunk") + parent = cont_chk_proxy; + } /* end else */ + + /* Sanity checks */ + HDassert(parent); + HDassert(((H5C_cache_entry_t *)parent)->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC); + HDassert(((H5C_cache_entry_t *)parent)->type); + HDassert((((H5C_cache_entry_t *)(parent))->type->id == H5AC_OHDR_ID) + || (((H5C_cache_entry_t *)(parent))->type->id == H5AC_OHDR_CHK_ID)); + + /* Add flush dependency from chunk containing the continuation message + * that points to this chunk (either oh or another chunk proxy object) + */ + if(H5AC_create_flush_dependency(parent, chk_proxy) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTDEPEND, FAIL, "unable to create flush dependency") + + /* Make note of the address and pointer of the flush dependency + * parent so we can take the dependency down on eviction. + */ + chk_proxy->parent = parent; + } + + /* Add flush dependency on object header proxy, if proxy exists */ + { + /* Sanity check */ + HDassert(chk_proxy->oh->proxy); + + /* Register the object header chunk as a parent of the virtual entry */ + if(H5AC_proxy_entry_add_parent(chk_proxy->oh->proxy, chk_proxy) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTSET, FAIL, "can't add object header chunk as parent of proxy") + } + } /* end if */ + break; + case H5AC_NOTIFY_ACTION_AFTER_FLUSH: case H5AC_NOTIFY_ACTION_ENTRY_DIRTIED: /* do nothing */ @@ -1031,10 +1104,29 @@ H5O__cache_chk_notify(H5AC_notify_action_t action, void *_thing) case H5AC_NOTIFY_ACTION_CHILD_DIRTIED: case H5AC_NOTIFY_ACTION_CHILD_CLEANED: - case H5AC_NOTIFY_ACTION_BEFORE_EVICT: /* do nothing */ break; + case H5AC_NOTIFY_ACTION_BEFORE_EVICT: + if(chk_proxy->oh->swmr_write) { + /* Remove flush dependency on parent object header chunk */ + { + /* Sanity checks */ + HDassert(chk_proxy->parent != NULL); + HDassert(((H5C_cache_entry_t *)(chk_proxy->parent))->magic == H5C__H5C_CACHE_ENTRY_T_MAGIC); + HDassert(((H5C_cache_entry_t *)(chk_proxy->parent))->type); + HDassert((((H5C_cache_entry_t *)(chk_proxy->parent))->type->id == H5AC_OHDR_ID) || (((H5C_cache_entry_t *)(chk_proxy->parent))->type->id == H5AC_OHDR_CHK_ID)); + + if(H5AC_destroy_flush_dependency(chk_proxy->parent, chk_proxy) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTUNDEPEND, FAIL, "unable to destroy flush dependency") + } + + /* Unregister the object header chunk as a parent of the virtual entry */ + if(H5AC_proxy_entry_remove_parent(chk_proxy->oh->proxy, chk_proxy) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTSET, FAIL, "can't remove object header chunk as parent of proxy") + } /* end if */ + break; + default: #ifdef NDEBUG HGOTO_ERROR(H5E_OHDR, H5E_BADVALUE, FAIL, "unknown action from metadata cache") @@ -1200,6 +1292,7 @@ H5O__chunk_deserialize(H5O_t *oh, haddr_t addr, size_t len, const uint8_t *image oh->chunk[chunkno].size = len; if(NULL == (oh->chunk[chunkno].image = H5FL_BLK_MALLOC(chunk_image, oh->chunk[chunkno].size))) HGOTO_ERROR(H5E_OHDR, H5E_CANTALLOC, FAIL, "memory allocation failed") + oh->chunk[chunkno].chunk_proxy = NULL; /* Copy disk image into chunk's image */ HDmemcpy(oh->chunk[chunkno].image, image, oh->chunk[chunkno].size); diff --git a/src/H5Ochunk.c b/src/H5Ochunk.c index f6eb135..50be171 100644 --- a/src/H5Ochunk.c +++ b/src/H5Ochunk.c @@ -403,7 +403,9 @@ H5O_chunk_delete(H5F_t *f, hid_t dxpl_id, H5O_t *oh, unsigned idx) HDassert(chk_proxy->oh == oh); HDassert(chk_proxy->chunkno == idx); - cache_flags |= H5AC__DIRTIED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; + /* Only free file space if not doing SWMR writes */ + if(!oh->swmr_write) + cache_flags |= H5AC__DIRTIED_FLAG | H5AC__FREE_FILE_SPACE_FLAG; /* Release the chunk proxy from the cache, marking it deleted */ if(H5AC_unprotect(f, dxpl_id, H5AC_OHDR_CHK, oh->chunk[idx].addr, chk_proxy, cache_flags) < 0) diff --git a/src/H5Ocopy.c b/src/H5Ocopy.c index c1c72cf..ddf375c 100644 --- a/src/H5Ocopy.c +++ b/src/H5Ocopy.c @@ -381,7 +381,7 @@ H5O_copy_header_real(const H5O_loc_t *oloc_src, H5O_loc_t *oloc_dst /*out*/, } /* end if */ /* Get source object header */ - if(NULL == (oh_src = H5O_protect(oloc_src, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh_src = H5O_protect(oloc_src, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Retrieve user data for particular type of object to copy */ @@ -450,6 +450,7 @@ H5O_copy_header_real(const H5O_loc_t *oloc_src, H5O_loc_t *oloc_dst /*out*/, oh_dst->attr_msgs_seen = oh_src->attr_msgs_seen; oh_dst->sizeof_size = H5F_SIZEOF_SIZE(oloc_dst->file); oh_dst->sizeof_addr = H5F_SIZEOF_ADDR(oloc_dst->file); + oh_dst->swmr_write = !!(H5F_INTENT(oloc_dst->file) & H5F_ACC_SWMR_WRITE); /* Copy time fields */ oh_dst->atime = oh_src->atime; @@ -461,6 +462,15 @@ H5O_copy_header_real(const H5O_loc_t *oloc_src, H5O_loc_t *oloc_dst /*out*/, oh_dst->max_compact = oh_src->max_compact; oh_dst->min_dense = oh_src->min_dense; + /* Create object header proxy if doing SWMR writes */ + if(oh_dst->swmr_write) { + /* Create virtual entry, for use as proxy */ + if(NULL == (oh_dst->proxy = H5AC_proxy_entry_create())) + HGOTO_ERROR(H5E_OHDR, H5E_CANTCREATE, FAIL, "can't create object header proxy") + } /* end if */ + else + oh_dst->proxy = NULL; + /* Initialize size of chunk array. Start off with zero chunks so this field * is consistent with the current state of the chunk array. This is * important if an error occurs. diff --git a/src/H5Odbg.c b/src/H5Odbg.c index d9d4b8b..827f40c 100644 --- a/src/H5Odbg.c +++ b/src/H5Odbg.c @@ -578,7 +578,7 @@ H5O_debug(H5F_t *f, hid_t dxpl_id, haddr_t addr, FILE *stream, int indent, int f loc.addr = addr; loc.holding_file = FALSE; - if(NULL == (oh = H5O_protect(&loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(&loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* debug */ diff --git a/src/H5Oflush.c b/src/H5Oflush.c index 4fc9c8e..e399c6c 100644 --- a/src/H5Oflush.c +++ b/src/H5Oflush.c @@ -162,7 +162,7 @@ H5O_oh_tag(const H5O_loc_t *oloc, hid_t dxpl_id, haddr_t *tag) HDassert(oloc); /* Get object header for object */ - if(NULL == (oh = H5O_protect(oloc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(oloc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to protect object's object header") /* Get object header's address (i.e. the tag value for this object) */ @@ -262,7 +262,7 @@ H5O_refresh_metadata(hid_t oid, H5O_loc_t oloc, hid_t dxpl_id) HGOTO_ERROR(H5E_OHDR, H5E_CANTLOAD, FAIL, "unable to refresh object") /* Re-open the object, re-fetching its metadata */ - if((H5O_refresh_metadata_reopen(oid, &obj_loc, dxpl_id)) < 0) + if((H5O_refresh_metadata_reopen(oid, &obj_loc, dxpl_id, FALSE)) < 0) HGOTO_ERROR(H5E_OHDR, H5E_CANTLOAD, FAIL, "unable to refresh object") } /* end if */ @@ -360,7 +360,7 @@ done: *------------------------------------------------------------------------- */ herr_t -H5O_refresh_metadata_reopen(hid_t oid, H5G_loc_t *obj_loc, hid_t dxpl_id) +H5O_refresh_metadata_reopen(hid_t oid, H5G_loc_t *obj_loc, hid_t dxpl_id, hbool_t start_swmr) { void *object = NULL; /* Dataset for this operation */ H5I_type_t type; /* Type of object for the ID */ @@ -388,8 +388,9 @@ H5O_refresh_metadata_reopen(hid_t oid, H5G_loc_t *obj_loc, hid_t dxpl_id) /* Re-open the dataset */ if(NULL == (object = H5D_open(obj_loc, H5P_DATASET_ACCESS_DEFAULT, dxpl_id))) HGOTO_ERROR(H5E_DATASET, H5E_CANTOPENOBJ, FAIL, "unable to open dataset") - if(H5D_mult_refresh_reopen((H5D_t *)object, dxpl_id) < 0) - HGOTO_ERROR(H5E_DATASET, H5E_CANTOPENOBJ, FAIL, "unable to finish refresh for dataset") + if(!start_swmr) /* No need to handle multiple opens when H5Fstart_swmr_write() */ + if(H5D_mult_refresh_reopen((H5D_t *)object, dxpl_id) < 0) + HGOTO_ERROR(H5E_DATASET, H5E_CANTOPENOBJ, FAIL, "unable to finish refresh for dataset") break; case(H5I_UNINIT): diff --git a/src/H5Omessage.c b/src/H5Omessage.c index 7063ef6..ee21e49 100644 --- a/src/H5Omessage.c +++ b/src/H5Omessage.c @@ -477,7 +477,7 @@ H5O_msg_read(const H5O_loc_t *loc, unsigned type_id, void *mesg, HDassert(type_id < NELMTS(H5O_msg_class_g)); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, NULL, "unable to protect object header") /* Call the "real" read routine */ @@ -803,7 +803,7 @@ H5O_msg_count(const H5O_loc_t *loc, unsigned type_id, hid_t dxpl_id) HDassert(type); /* Load the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to protect object header") /* Count the messages of the correct type */ @@ -885,7 +885,7 @@ H5O_msg_exists(const H5O_loc_t *loc, unsigned type_id, hid_t dxpl_id) HDassert(type_id < NELMTS(H5O_msg_class_g)); /* Load the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to protect object header") /* Call the "real" exists routine */ @@ -1225,7 +1225,7 @@ H5O_msg_iterate(const H5O_loc_t *loc, unsigned type_id, HDassert(op); /* Protect the object header to iterate over */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to protect object header") /* Call the "real" iterate routine */ diff --git a/src/H5Opkg.h b/src/H5Opkg.h index b80f736..e778f90 100644 --- a/src/H5Opkg.h +++ b/src/H5Opkg.h @@ -269,6 +269,7 @@ typedef struct H5O_chunk_t { size_t size; /*chunk size */ size_t gap; /*space at end of chunk too small for null message */ uint8_t *image; /*image of file */ + struct H5O_chunk_proxy_t *chunk_proxy; /* Pointer to a chunk's proxy when chunk protected */ } H5O_chunk_t; struct H5O_t { @@ -278,6 +279,7 @@ struct H5O_t { /* File-specific information (not stored) */ size_t sizeof_size; /* Size of file sizes */ size_t sizeof_addr; /* Size of file addresses */ + hbool_t swmr_write; /* Whether we are doing SWMR writes */ /* Debugging information (not stored) */ #ifdef H5O_ENABLE_BAD_MESG_COUNT @@ -320,6 +322,10 @@ struct H5O_t { size_t nchunks; /*number of chunks */ size_t alloc_nchunks; /*chunks allocated */ H5O_chunk_t *chunk; /*array of chunks */ + hbool_t chunks_pinned; /* Whether chunks are pinned from ohdr protect */ + + /* Object header proxy information (not stored) */ + H5AC_proxy_entry_t *proxy; /* Proxy cache entry for all ohdr entries */ }; /* Class for types of objects in file */ @@ -381,6 +387,20 @@ typedef struct H5O_chunk_proxy_t { H5O_t *oh; /* Object header for this chunk */ unsigned chunkno; /* Chunk number for this chunk */ unsigned cont_chunkno; /* Chunk number for the chunk containing the continuation message that points to this chunk */ + + /* Flush depencency parent information (not stored) + * + * The following field is used to store a pointer + * to the in-core representation of the chunk proxy's flush dependency + * parent -- if it exists. If it does not exist, this field will + * contain NULL. + * + * If the file is opened in SWMR write mode, the flush dependency + * parent of the chunk proxy will be either its object header + * (if cont_chunkno == 0) or the chunk proxy indicated by the + * cont_chunkno field (if cont_chunkno > 0). + */ + void *parent; /* Pointer to flush dependency parent */ } H5O_chunk_proxy_t; /* Callback information for loading object header chunk from disk */ @@ -621,9 +641,6 @@ H5_DLL herr_t H5O_attr_link(H5F_t *f, hid_t dxpl_id, H5O_t *open_oh, void *_mesg H5_DLL herr_t H5O_attr_count_real(H5F_t *f, hid_t dxpl_id, H5O_t *oh, hsize_t *nattrs); -/* These functions operate on object locations */ -H5_DLL H5O_loc_t *H5O_get_loc(hid_t id); - /* Testing functions */ #ifdef H5O_TESTING H5_DLL htri_t H5O_is_attr_empty_test(hid_t oid); @@ -633,6 +650,9 @@ H5_DLL herr_t H5O_attr_dense_info_test(hid_t oid, hsize_t *name_count, hsize_t * H5_DLL herr_t H5O_check_msg_marked_test(hid_t oid, hbool_t flag_val); H5_DLL herr_t H5O_expunge_chunks_test(const H5O_loc_t *oloc, hid_t dxpl_id); H5_DLL herr_t H5O_get_rc(const H5O_loc_t *oloc, hid_t dxpl_id, unsigned *rc); +H5_DLL herr_t H5O_msg_get_chunkno_test(hid_t oid, unsigned msg_type, + unsigned *chunk_num); +H5_DLL herr_t H5O_msg_move_to_new_chunk_test(hid_t oid, unsigned msg_type); #endif /* H5O_TESTING */ /* Object header debugging routines */ diff --git a/src/H5Oprivate.h b/src/H5Oprivate.h index 2169906..e430b1f 100644 --- a/src/H5Oprivate.h +++ b/src/H5Oprivate.h @@ -829,7 +829,8 @@ H5_DLL herr_t H5O_create(H5F_t *f, hid_t dxpl_id, size_t size_hint, H5_DLL herr_t H5O_open(H5O_loc_t *loc); H5_DLL herr_t H5O_close(H5O_loc_t *loc, hbool_t *file_closed/*out*/); H5_DLL int H5O_link(const H5O_loc_t *loc, int adjust, hid_t dxpl_id); -H5_DLL H5O_t *H5O_protect(const H5O_loc_t *loc, hid_t dxpl_id, unsigned prot_flags); +H5_DLL H5O_t *H5O_protect(const H5O_loc_t *loc, hid_t dxpl_id, + unsigned prot_flags, hbool_t pin_all_chunks); H5_DLL H5O_t *H5O_pin(const H5O_loc_t *loc, hid_t dxpl_id); H5_DLL herr_t H5O_unpin(H5O_t *oh); H5_DLL herr_t H5O_dec_rc_by_loc(const H5O_loc_t *loc, hid_t dxpl_id); @@ -852,6 +853,7 @@ H5_DLL herr_t H5O_get_nlinks(const H5O_loc_t *loc, hid_t dxpl_id, hsize_t *nlink H5_DLL void *H5O_obj_create(H5F_t *f, H5O_type_t obj_type, void *crt_info, H5G_loc_t *obj_loc, hid_t dxpl_id); H5_DLL haddr_t H5O_get_oh_addr(const H5O_t *oh); H5_DLL herr_t H5O_get_rc_and_type(const H5O_loc_t *oloc, hid_t dxpl_id, unsigned *rc, H5O_type_t *otype); +H5_DLL H5AC_proxy_entry_t *H5O_get_proxy(const H5O_t *oh); /* Object header message routines */ H5_DLL herr_t H5O_msg_create(const H5O_loc_t *loc, unsigned type_id, unsigned mesg_flags, @@ -903,7 +905,7 @@ H5_DLL herr_t H5O_msg_delete(H5F_t *f, hid_t dxpl_id, H5O_t *open_oh, H5_DLL herr_t H5O_flush_common(H5O_loc_t *oloc, hid_t obj_id, hid_t dxpl_id); H5_DLL herr_t H5O_refresh_metadata(hid_t oid, H5O_loc_t oloc, hid_t dxpl_id); H5_DLL herr_t H5O_refresh_metadata_close(hid_t oid, H5O_loc_t oloc, H5G_loc_t *obj_loc, hid_t dxpl_id); -H5_DLL herr_t H5O_refresh_metadata_reopen(hid_t oid, H5G_loc_t *obj_loc, hid_t dxpl_id); +H5_DLL herr_t H5O_refresh_metadata_reopen(hid_t oid, H5G_loc_t *obj_loc, hid_t dxpl_id, hbool_t start_swmr); /* Object copying routines */ H5_DLL herr_t H5O_copy_header_map(const H5O_loc_t *oloc_src, H5O_loc_t *oloc_dst /*out */, @@ -923,6 +925,7 @@ H5_DLL herr_t H5O_loc_reset(H5O_loc_t *loc); H5_DLL herr_t H5O_loc_copy(H5O_loc_t *dst, H5O_loc_t *src, H5_copy_depth_t depth); H5_DLL herr_t H5O_loc_hold_file(H5O_loc_t *loc); H5_DLL herr_t H5O_loc_free(H5O_loc_t *loc); +H5_DLL H5O_loc_t *H5O_get_loc(hid_t id); /* EFL operators */ H5_DLL hsize_t H5O_efl_total_size(H5O_efl_t *efl); diff --git a/src/H5Otest.c b/src/H5Otest.c index c3e487d..8f8980a 100644 --- a/src/H5Otest.c +++ b/src/H5Otest.c @@ -108,7 +108,7 @@ H5O_is_attr_dense_test(hid_t oid) HGOTO_ERROR(H5E_SYM, H5E_NOTFOUND, FAIL, "object not found") /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Check for attribute info stored */ @@ -173,7 +173,7 @@ H5O_is_attr_empty_test(hid_t oid) HGOTO_ERROR(H5E_SYM, H5E_NOTFOUND, FAIL, "object not found") /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Check for attribute info stored */ @@ -266,7 +266,7 @@ H5O_num_attrs_test(hid_t oid, hsize_t *nattrs) HGOTO_ERROR(H5E_SYM, H5E_NOTFOUND, FAIL, "object not found") /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Check for attribute info stored */ @@ -361,7 +361,7 @@ H5O_attr_dense_info_test(hid_t oid, hsize_t *name_count, hsize_t *corder_count) H5_BEGIN_TAG(H5AC_ind_read_dxpl_id, loc->addr, FAIL); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR_TAG(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Check for attribute info stored */ @@ -452,7 +452,7 @@ H5O_check_msg_marked_test(hid_t oid, hbool_t flag_val) HGOTO_ERROR(H5E_SYM, H5E_NOTFOUND, FAIL, "object not found") /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") /* Locate "unknown" message */ @@ -511,7 +511,7 @@ H5O_expunge_chunks_test(const H5O_loc_t *loc, hid_t dxpl_id) FUNC_ENTER_NOAPI(FAIL) /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__NO_FLAGS_SET))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__NO_FLAGS_SET, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to protect object header") /* Safety check */ @@ -571,7 +571,7 @@ H5O_get_rc(const H5O_loc_t *loc, hid_t dxpl_id, unsigned *rc) HDassert(rc); /* Get the object header */ - if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(loc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to protect object header") /* Save the refcount for the object header */ @@ -585,3 +585,165 @@ done: FUNC_LEAVE_NOAPI(ret_value) } /* H5O_expunge_chunks_test() */ + +/*-------------------------------------------------------------------------- + NAME + H5O_msg_get_chunkno_test + PURPOSE + Retrieve the chunk number for an object header message of a given type. + USAGE + herr_t H5O_check_msg_marked_test(oid, chunk_num) + hid_t oid; IN: Object to check + unsigned msg_type; IN: Object header message type to check + unsigned *chunk_num; OUT: Object header chunk that the message is in + RETURNS + Non-negative on success, negative on failure + DESCRIPTION + Retrieves the chunk number for the first object header message of a given + type found in an object's header. + GLOBAL VARIABLES + COMMENTS, BUGS, ASSUMPTIONS + DO NOT USE THIS FUNCTION FOR ANYTHING EXCEPT TESTING + EXAMPLES + REVISION LOG +--------------------------------------------------------------------------*/ +herr_t +H5O_msg_get_chunkno_test(hid_t oid, unsigned msg_type, unsigned *chunk_num) +{ + H5O_t *oh = NULL; /* Object header */ + H5O_loc_t *loc; /* Pointer to object's location */ + H5O_mesg_t *idx_msg; /* Pointer to message */ + unsigned idx; /* Index of message */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Get object location for object */ + if(NULL == (loc = H5O_get_loc(oid))) + HGOTO_ERROR(H5E_SYM, H5E_NOTFOUND, FAIL, "object not found") + + /* Get the object header */ + if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) + HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") + + /* Locate first message of given type */ + for(idx = 0, idx_msg = &oh->mesg[0]; idx < oh->nmesgs; idx++, idx_msg++) + if(idx_msg->type->id == msg_type) { + /* Set the chunk number for the message */ + *chunk_num = idx_msg->chunkno; + + /* Break out of loop, to indicate that the message was found */ + break; + } /* end if */ + + /* Check for not finding a message of the given type*/ + if(idx == oh->nmesgs) + HGOTO_ERROR(H5E_OHDR, H5E_NOTFOUND, FAIL, "message of type not found") + +done: + if(oh && H5O_unprotect(loc, H5AC_ind_read_dxpl_id, oh, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_OHDR, H5E_CANTUNPROTECT, FAIL, "unable to release object header") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5O_msg_get_chunkno_test() */ + + +/*-------------------------------------------------------------------------- + NAME + H5O_msg_move_to_new_chunk_test + PURPOSE + Move a message into a new chunk + USAGE + herr_t H5O_msg_move_to_new_chunk_test(oid, msg_type) + hid_t oid; IN: Object to check + unsigned msg_type; IN: Object header message type to check + RETURNS + Non-negative on success, negative on failure + DESCRIPTION + Moves the first message of the given type to a new object header chunk. + GLOBAL VARIABLES + COMMENTS, BUGS, ASSUMPTIONS + DO NOT USE THIS FUNCTION FOR ANYTHING EXCEPT TESTING + EXAMPLES + REVISION LOG +--------------------------------------------------------------------------*/ +herr_t +H5O_msg_move_to_new_chunk_test(hid_t oid, unsigned msg_type) +{ + H5O_t *oh = NULL; /* Object header */ + H5O_loc_t *loc; /* Pointer to object's location */ + H5O_mesg_t *curr_msg; /* Pointer to current message */ + unsigned idx; /* Index of message */ + herr_t ret_value = SUCCEED; /* Return value */ + + FUNC_ENTER_NOAPI(FAIL) + + /* Get object location for object */ + if(NULL == (loc = H5O_get_loc(oid))) + HGOTO_ERROR(H5E_SYM, H5E_NOTFOUND, FAIL, "object not found") + + /* Get the object header */ + if(NULL == (oh = H5O_protect(loc, H5AC_ind_read_dxpl_id, H5AC__NO_FLAGS_SET, FALSE))) + HGOTO_ERROR(H5E_OHDR, H5E_CANTPROTECT, FAIL, "unable to load object header") + + /* Locate first message of given type */ + for(idx = 0, curr_msg = &oh->mesg[0]; idx < oh->nmesgs; idx++, curr_msg++) + if(curr_msg->type->id == msg_type) { + H5O_msg_alloc_info_t found_msg; /* Information about message to move */ + unsigned msg_chunkno = curr_msg->chunkno; /* Chunk that the message is in */ + uint8_t *end_chunk_data = (oh->chunk[msg_chunkno].image + oh->chunk[msg_chunkno].size) - (H5O_SIZEOF_CHKSUM_OH(oh) + oh->chunk[msg_chunkno].gap); /* End of message data in chunk */ + uint8_t *end_msg = curr_msg->raw + curr_msg->raw_size; /* End of current message */ + size_t gap_size = 0; /* Size of gap after current message */ + size_t null_size = 0; /* Size of NULL message after current message */ + unsigned null_msgno = 0; /* Index of NULL message after current message */ + size_t total_size; /* Total size of available space "around" current message */ + size_t new_idx; /* Index of new null message */ + + /* Check if the message is the last one in the chunk */ + if(end_msg == end_chunk_data) + gap_size = oh->chunk[msg_chunkno].gap; + else { + H5O_mesg_t *tmp_msg; /* Temp. pointer to message to operate on */ + unsigned v; /* Local index variable */ + + /* Check for null message after this message, in same chunk */ + for(v = 0, tmp_msg = &oh->mesg[0]; v < oh->nmesgs; v++, tmp_msg++) { + if(tmp_msg->type->id == H5O_NULL_ID && (tmp_msg->raw - H5O_SIZEOF_MSGHDR_OH(oh)) == end_msg) { + null_msgno = v; + null_size = (size_t)H5O_SIZEOF_MSGHDR_OH(oh) + tmp_msg->raw_size; + break; + } /* end if */ + + /* XXX: Should also check for NULL message in front of current message... */ + + } /* end for */ + } /* end else */ + + /* Add up current message's total available space */ + total_size = curr_msg->raw_size + gap_size + null_size; + + /* Set up "found message" info for moving the message */ + found_msg.msgno = (int)idx; + found_msg.id = curr_msg->type->id; + found_msg.chunkno = msg_chunkno; + found_msg.gap_size = gap_size; + found_msg.null_size = null_size; + found_msg.total_size = total_size; + found_msg.null_msgno = null_msgno; + + /* Allocate and initialize new chunk in the file, moving the found message */ + /* (*new_idx returned from this routine is unused here) */ + if(H5O__alloc_chunk(loc->file, H5AC_ind_read_dxpl_id, oh, 40, oh->nmesgs, &found_msg, &new_idx) < 0) + HGOTO_ERROR(H5E_OHDR, H5E_CANTALLOC, FAIL, "can't allocate new object header chunk") + + /* Break out of loop, the message was found */ + break; + } /* end if */ + +done: + if(oh && H5O_unprotect(loc, H5AC_ind_read_dxpl_id, oh, H5AC__NO_FLAGS_SET) < 0) + HDONE_ERROR(H5E_OHDR, H5E_CANTUNPROTECT, FAIL, "unable to release object header") + + FUNC_LEAVE_NOAPI(ret_value) +} /* H5O_msg_get_chunkno_test() */ + diff --git a/src/H5Pfapl.c b/src/H5Pfapl.c index 81cd260..91a0e03 100644 --- a/src/H5Pfapl.c +++ b/src/H5Pfapl.c @@ -186,6 +186,9 @@ /* Definition for object flush callback */ #define H5F_ACS_OBJECT_FLUSH_CB_SIZE sizeof(H5F_object_flush_t) #define H5F_ACS_OBJECT_FLUSH_CB_DEF {NULL, NULL} +/* Definition for status_flags in the superblock */ +#define H5F_ACS_CLEAR_STATUS_FLAGS_SIZE sizeof(hbool_t) +#define H5F_ACS_CLEAR_STATUS_FLAGS_DEF FALSE /* Definition for 'use metadata cache logging' flag */ #define H5F_ACS_USE_MDC_LOGGING_SIZE sizeof(hbool_t) #define H5F_ACS_USE_MDC_LOGGING_DEF FALSE @@ -334,6 +337,7 @@ static const hbool_t H5F_def_core_write_tracking_flag_g = H5F_ACS_CORE_WRITE_TRA static const size_t H5F_def_core_write_tracking_page_size_g = H5F_ACS_CORE_WRITE_TRACKING_PAGE_SIZE_DEF; /* Default core VFD write tracking page size */ static const unsigned H5F_def_metadata_read_attempts_g = H5F_ACS_METADATA_READ_ATTEMPTS_DEF; /* Default setting for the # of metadata read attempts */ static const H5F_object_flush_t H5F_def_object_flush_cb_g = H5F_ACS_OBJECT_FLUSH_CB_DEF; /* Default setting for object flush callback */ +static const hbool_t H5F_def_clear_status_flags_g = H5F_ACS_CLEAR_STATUS_FLAGS_DEF; /* Default to clear the superblock status_flags */ static const hbool_t H5F_def_use_mdc_logging_g = H5F_ACS_USE_MDC_LOGGING_DEF; /* Default metadata cache logging flag */ static const char *H5F_def_mdc_log_location_g = H5F_ACS_MDC_LOG_LOCATION_DEF; /* Default mdc log location */ static const hbool_t H5F_def_start_mdc_log_on_access_g = H5F_ACS_START_MDC_LOG_ON_ACCESS_DEF; /* Default mdc log start on access flag */ @@ -510,6 +514,11 @@ H5P__facc_reg_prop(H5P_genclass_t *pclass) NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL) < 0) HGOTO_ERROR(H5E_PLIST, H5E_CANTINSERT, FAIL, "can't insert property into class") + /* Register the private property of whether to clear the superblock status_flags. It's used by h5clear only. */ + if(H5P_register_real(pclass, H5F_ACS_CLEAR_STATUS_FLAGS_NAME, H5F_ACS_CLEAR_STATUS_FLAGS_SIZE, &H5F_def_clear_status_flags_g, + NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL) < 0) + HGOTO_ERROR(H5E_PLIST, H5E_CANTINSERT, FAIL, "can't insert property into class") + /* Register the metadata cache logging flag. */ if(H5P_register_real(pclass, H5F_ACS_USE_MDC_LOGGING_NAME, H5F_ACS_USE_MDC_LOGGING_SIZE, &H5F_def_use_mdc_logging_g, NULL, NULL, NULL, H5F_ACS_USE_MDC_LOGGING_ENC, H5F_ACS_USE_MDC_LOGGING_DEC, NULL, NULL, NULL, NULL) < 0) diff --git a/src/H5Plapl.c b/src/H5Plapl.c index 266dc63..a0ec7f1 100644 --- a/src/H5Plapl.c +++ b/src/H5Plapl.c @@ -1168,7 +1168,9 @@ H5Pset_elink_acc_flags(hid_t lapl_id, unsigned flags) H5TRACE2("e", "iIu", lapl_id, flags); /* Check that flags are valid */ - if((flags != H5F_ACC_RDWR) && (flags != H5F_ACC_RDONLY) && (flags != H5F_ACC_DEFAULT)) + if(( flags != H5F_ACC_RDWR) && (flags != (H5F_ACC_RDWR | H5F_ACC_SWMR_WRITE)) + && (flags != H5F_ACC_RDONLY) && (flags != (H5F_ACC_RDONLY | H5F_ACC_SWMR_READ)) + && (flags != H5F_ACC_DEFAULT)) HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "invalid file open flags") /* Get the plist structure */ @@ -2436,7 +2436,7 @@ H5SM_read_mesg(H5F_t *f, const H5SM_sohm_t *mesg, H5HF_t *fheap, HGOTO_ERROR(H5E_SOHM, H5E_CANTLOAD, FAIL, "unable to open object header") /* Load the object header from the cache */ - if(NULL == (oh = H5O_protect(&oloc, dxpl_id, H5AC__READ_ONLY_FLAG))) + if(NULL == (oh = H5O_protect(&oloc, dxpl_id, H5AC__READ_ONLY_FLAG, FALSE))) HGOTO_ERROR(H5E_SOHM, H5E_CANTPROTECT, FAIL, "unable to load object header") } /* end if */ else diff --git a/src/H5err.txt b/src/H5err.txt index a156316..9531df9 100644 --- a/src/H5err.txt +++ b/src/H5err.txt @@ -169,7 +169,9 @@ MINOR, CACHE, H5E_CANTUNPROTECT, Unable to unprotect metadata MINOR, CACHE, H5E_CANTPIN, Unable to pin cache entry MINOR, CACHE, H5E_CANTUNPIN, Unable to un-pin cache entry MINOR, CACHE, H5E_CANTMARKDIRTY, Unable to mark a pinned entry as dirty +MINOR, CACHE, H5E_CANTMARKCLEAN, Unable to mark a pinned entry as clean MINOR, CACHE, H5E_CANTDIRTY, Unable to mark metadata as dirty +MINOR, CACHE, H5E_CANTCLEAN, Unable to mark metadata as clean MINOR, CACHE, H5E_CANTEXPUNGE, Unable to expunge a metadata cache entry MINOR, CACHE, H5E_CANTRESIZE, Unable to resize a metadata cache entry MINOR, CACHE, H5E_CANTDEPEND, Unable to create a flush dependency diff --git a/src/Makefile.am b/src/Makefile.am index ff845d1..939e151 100644 --- a/src/Makefile.am +++ b/src/Makefile.am @@ -42,9 +42,10 @@ DISTCLEANFILES=H5pubconf.h # library sources libhdf5_la_SOURCES= H5.c H5checksum.c H5dbg.c H5system.c H5timer.c H5trace.c \ H5A.c H5Abtree2.c H5Adense.c H5Adeprec.c H5Aint.c H5Atest.c \ - H5AC.c H5ACdbg.c H5AClog.c \ + H5AC.c H5ACdbg.c H5AClog.c H5ACproxy_entry.c \ H5B.c H5Bcache.c H5Bdbg.c \ - H5B2.c H5B2cache.c H5B2dbg.c H5B2hdr.c H5B2int.c H5B2stat.c H5B2test.c \ + H5B2.c H5B2cache.c H5B2dbg.c H5B2hdr.c H5B2int.c H5B2internal.c \ + H5B2leaf.c H5B2stat.c H5B2test.c \ H5C.c H5Cdbg.c H5Cepoch.c H5Clog.c H5Cquery.c H5Ctag.c H5Ctest.c \ H5CS.c \ H5D.c H5Dbtree.c H5Dbtree2.c H5Dchunk.c H5Dcompact.c H5Dcontig.c H5Ddbg.c \ @@ -60,11 +61,12 @@ libhdf5_la_SOURCES= H5.c H5checksum.c H5dbg.c H5system.c H5timer.c H5trace.c \ H5Fmount.c H5Fquery.c \ H5Fsfile.c H5Fsuper.c H5Fsuper_cache.c H5Ftest.c \ H5FA.c H5FAcache.c H5FAdbg.c H5FAdblock.c H5FAdblkpage.c H5FAhdr.c \ - H5FAstat.c H5FAtest.c \ + H5FAint.c H5FAstat.c H5FAtest.c \ H5FD.c H5FDcore.c \ H5FDfamily.c H5FDint.c H5FDlog.c \ - H5FDmulti.c H5FDsec2.c H5FDspace.c H5FDstdio.c \ - H5FL.c H5FO.c H5FS.c H5FScache.c H5FSdbg.c H5FSsection.c H5FSstat.c H5FStest.c \ + H5FDmulti.c H5FDsec2.c H5FDspace.c H5FDstdio.c H5FDtest.c \ + H5FL.c H5FO.c H5FS.c H5FScache.c H5FSdbg.c H5FSint.c H5FSsection.c \ + H5FSstat.c H5FStest.c \ H5G.c H5Gbtree2.c H5Gcache.c \ H5Gcompact.c H5Gdense.c H5Gdeprec.c H5Gent.c \ H5Gint.c H5Glink.c \ |