/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * 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: H5Cprivate.h * 6/3/04 * John Mainzer * * Purpose: Constants and typedefs available to the rest of the * library. * * Modifications: * *------------------------------------------------------------------------- */ #ifndef _H5Cprivate_H #define _H5Cprivate_H #include "H5Cpublic.h" /* public prototypes */ /* Private headers needed by this header */ #include "H5private.h" /* Generic Functions */ #include "H5Fprivate.h" /* File access */ /**************************/ /* Library Private Macros */ /**************************/ #ifndef NDEBUG #define H5C_DO_SANITY_CHECKS 1 #define H5C_DO_SLIST_SANITY_CHECKS 0 #define H5C_DO_TAGGING_SANITY_CHECKS 1 #define H5C_DO_EXTREME_SANITY_CHECKS 0 #else /* NDEBUG */ /* With rare execptions, the following defines should be set * to 0 if NDEBUG is defined */ #define H5C_DO_SANITY_CHECKS 0 #define H5C_DO_SLIST_SANITY_CHECKS 0 #define H5C_DO_TAGGING_SANITY_CHECKS 0 #define H5C_DO_EXTREME_SANITY_CHECKS 0 #endif /* NDEBUG */ /* Note: The memory sanity checks aren't going to work until I/O filters are * changed to call a particular alloc/free routine for their buffers, * because the H5AC__SERIALIZE_RESIZED_FLAG set by the fractal heap * direct block serialize callback calls H5Z_pipeline(). When the I/O * filters are changed, then we should implement "cache image alloc/free" * routines that the fractal heap direct block (and global heap) serialize * calls can use when resizing (and re-allocating) their image in the * cache. -QAK */ #define H5C_DO_MEMORY_SANITY_CHECKS 0 /* This sanity checking constant was picked out of the air. Increase * or decrease it if appropriate. Its purposes is to detect corrupt * object sizes, so it probably doesn't matter if it is a bit big. * * JRM - 5/17/04 */ #define H5C_MAX_ENTRY_SIZE ((size_t)(32 * 1024 * 1024)) /* H5C_COLLECT_CACHE_STATS controls overall collection of statistics * on cache activity. In general, this #define should be set to 1 in * debug mode, and 0 in production mode.. */ #ifndef NDEBUG #define H5C_COLLECT_CACHE_STATS 1 #else /* NDEBUG */ #define H5C_COLLECT_CACHE_STATS 0 #endif /* NDEBUG */ /* H5C_COLLECT_CACHE_ENTRY_STATS controls collection of statistics * in individual cache entries. * * H5C_COLLECT_CACHE_ENTRY_STATS should only be defined to true if * H5C_COLLECT_CACHE_STATS is also defined to true. */ #if H5C_COLLECT_CACHE_STATS #define H5C_COLLECT_CACHE_ENTRY_STATS 1 #else #define H5C_COLLECT_CACHE_ENTRY_STATS 0 #endif /* H5C_COLLECT_CACHE_STATS */ #ifdef H5_HAVE_PARALLEL /* we must maintain the clean and dirty LRU lists when we are compiled * with parallel support. */ #define H5C_MAINTAIN_CLEAN_AND_DIRTY_LRU_LISTS 1 #else /* H5_HAVE_PARALLEL */ /* The clean and dirty LRU lists don't buy us anything here -- we may * want them on for testing on occasion, but in general they should be * off. */ #define H5C_MAINTAIN_CLEAN_AND_DIRTY_LRU_LISTS 0 #endif /* H5_HAVE_PARALLEL */ /* Flags for cache client class behavior */ #define H5C__CLASS_NO_FLAGS_SET ((unsigned)0x0) #define H5C__CLASS_SPECULATIVE_LOAD_FLAG ((unsigned)0x1) #define H5C__CLASS_COMPRESSED_FLAG ((unsigned)0x2) /* The following flags may only appear in test code */ #define H5C__CLASS_NO_IO_FLAG ((unsigned)0x4) #define H5C__CLASS_SKIP_READS ((unsigned)0x8) #define H5C__CLASS_SKIP_WRITES ((unsigned)0x10) /* Flags for pre-serialize callback */ #define H5C__SERIALIZE_NO_FLAGS_SET ((unsigned)0) #define H5C__SERIALIZE_RESIZED_FLAG ((unsigned)0x1) #define H5C__SERIALIZE_MOVED_FLAG ((unsigned)0x2) #define H5C__SERIALIZE_COMPRESSED_FLAG ((unsigned)0x4) /* Upper and lower limits on cache size. These limits are picked * out of a hat -- you should be able to change them as necessary. * * However, if you need a very big cache, you should also increase the * size of the hash table (H5C__HASH_TABLE_LEN in H5Cpkg.h). The current * upper bound on cache size is rather large for the current hash table * size. */ #define H5C__MAX_MAX_CACHE_SIZE ((size_t)(128 * 1024 * 1024)) #define H5C__MIN_MAX_CACHE_SIZE ((size_t)(1024)) /* Default max cache size and min clean size are give here to make * them generally accessable. */ #define H5C__DEFAULT_MAX_CACHE_SIZE ((size_t)(4 * 1024 * 1024)) #define H5C__DEFAULT_MIN_CLEAN_SIZE ((size_t)(2 * 1024 * 1024)) /* Maximum height of flush dependency relationships between entries. This is * currently tuned to the extensible array (H5EA) data structure, which only * requires 6 levels of dependency (i.e. heights 0-6) (actually, the extensible * array needs 4 levels, plus another 2 levels are needed: one for the layer * under the extensible array and one for the layer above it). */ #define H5C__NUM_FLUSH_DEP_HEIGHTS 6 /* Values for cache entry magic field */ #define H5C__H5C_CACHE_ENTRY_T_MAGIC 0x005CAC0A #define H5C__H5C_CACHE_ENTRY_T_BAD_MAGIC 0xDeadBeef /* Cache configuration validation definitions */ #define H5C_RESIZE_CFG__VALIDATE_GENERAL 0x1 #define H5C_RESIZE_CFG__VALIDATE_INCREMENT 0x2 #define H5C_RESIZE_CFG__VALIDATE_DECREMENT 0x4 #define H5C_RESIZE_CFG__VALIDATE_INTERACTIONS 0x8 #define H5C_RESIZE_CFG__VALIDATE_ALL \ ( \ H5C_RESIZE_CFG__VALIDATE_GENERAL | \ H5C_RESIZE_CFG__VALIDATE_INCREMENT | \ H5C_RESIZE_CFG__VALIDATE_DECREMENT | \ H5C_RESIZE_CFG__VALIDATE_INTERACTIONS \ ) /* Cache configuration versions */ #define H5C__CURR_AUTO_SIZE_CTL_VER 1 #define H5C__CURR_AUTO_RESIZE_RPT_FCN_VER 1 /* Number of epoch markers active */ #define H5C__MAX_EPOCH_MARKERS 10 /* Default configuration settings */ #define H5C__DEF_AR_UPPER_THRESHHOLD 0.9999f #define H5C__DEF_AR_LOWER_THRESHHOLD 0.9f #define H5C__DEF_AR_MAX_SIZE ((size_t)(16 * 1024 * 1024)) #define H5C__DEF_AR_INIT_SIZE ((size_t)( 1 * 1024 * 1024)) #define H5C__DEF_AR_MIN_SIZE ((size_t)( 1 * 1024 * 1024)) #define H5C__DEF_AR_MIN_CLEAN_FRAC 0.5f #define H5C__DEF_AR_INCREMENT 2.0f #define H5C__DEF_AR_MAX_INCREMENT ((size_t)( 2 * 1024 * 1024)) #define H5C__DEF_AR_FLASH_MULTIPLE 1.0f #define H5C__DEV_AR_FLASH_THRESHOLD 0.25f #define H5C__DEF_AR_DECREMENT 0.9f #define H5C__DEF_AR_MAX_DECREMENT ((size_t)( 1 * 1024 * 1024)) #define H5C__DEF_AR_EPCHS_B4_EVICT 3 #define H5C__DEF_AR_EMPTY_RESERVE 0.05f #define H5C__MIN_AR_EPOCH_LENGTH 100 #define H5C__DEF_AR_EPOCH_LENGTH 50000 #define H5C__MAX_AR_EPOCH_LENGTH 1000000 /* #defines of flags used in the flags parameters in some of the * cache calls. Note that not all flags are applicable * to all function calls. Flags that don't apply to a particular * function are ignored in that function. * * These flags apply to all function calls: * H5C__NO_FLAGS_SET (generic "no flags set" for all fcn calls) * * * These flags apply to H5C_insert_entry(): * H5C__SET_FLUSH_MARKER_FLAG * H5C__PIN_ENTRY_FLAG * H5C__FLUSH_LAST_FLAG ; super block only * H5C__FLUSH_COLLECTIVELY_FLAG ; super block only * * These flags apply to H5C_protect() * H5C__READ_ONLY_FLAG * H5C__FLUSH_LAST_FLAG ; super block only * H5C__FLUSH_COLLECTIVELY_FLAG ; super block only * * These flags apply to H5C_unprotect(): * H5C__SET_FLUSH_MARKER_FLAG * H5C__DELETED_FLAG * H5C__DIRTIED_FLAG * H5C__PIN_ENTRY_FLAG * H5C__UNPIN_ENTRY_FLAG * H5C__FREE_FILE_SPACE_FLAG * H5C__TAKE_OWNERSHIP_FLAG * * These flags apply to H5C_expunge_entry(): * H5C__FREE_FILE_SPACE_FLAG * * These flags apply to H5C_flush_cache(): * H5C__FLUSH_INVALIDATE_FLAG * H5C__FLUSH_CLEAR_ONLY_FLAG * H5C__FLUSH_MARKED_ENTRIES_FLAG * H5C__FLUSH_IGNORE_PROTECTED_FLAG (can't use this flag in combination * with H5C__FLUSH_INVALIDATE_FLAG) * * These flags apply to H5C_flush_single_entry(): * H5C__FLUSH_INVALIDATE_FLAG * H5C__FLUSH_CLEAR_ONLY_FLAG * H5C__FLUSH_MARKED_ENTRIES_FLAG * H5C__TAKE_OWNERSHIP_FLAG * H5C__DEL_FROM_SLIST_ON_DESTROY_FLAG */ #define H5C__NO_FLAGS_SET 0x0000 #define H5C__SET_FLUSH_MARKER_FLAG 0x0001 #define H5C__DELETED_FLAG 0x0002 #define H5C__DIRTIED_FLAG 0x0004 #define H5C__PIN_ENTRY_FLAG 0x0008 #define H5C__UNPIN_ENTRY_FLAG 0x0010 #define H5C__FLUSH_INVALIDATE_FLAG 0x0020 #define H5C__FLUSH_CLEAR_ONLY_FLAG 0x0040 #define H5C__FLUSH_MARKED_ENTRIES_FLAG 0x0080 #define H5C__FLUSH_IGNORE_PROTECTED_FLAG 0x0100 #define H5C__READ_ONLY_FLAG 0x0200 #define H5C__FREE_FILE_SPACE_FLAG 0x0800 #define H5C__TAKE_OWNERSHIP_FLAG 0x1000 #define H5C__FLUSH_LAST_FLAG 0x2000 #define H5C__FLUSH_COLLECTIVELY_FLAG 0x4000 #define H5C__DEL_FROM_SLIST_ON_DESTROY_FLAG 0x8000 /****************************/ /* Library Private Typedefs */ /****************************/ /* Typedef for the main structure for the cache (defined in H5Cpkg.h) */ typedef struct H5C_t H5C_t; /*************************************************************************** * * Struct H5C_class_t * * Instances of H5C_class_t are used to specify the callback functions * used by the metadata cache for each class of metadata cache entry. * The fields of the structure are discussed below: * * id: Integer field containing the unique ID of the class of metadata * cache entries. * * name: Pointer to a string containing the name of the class of metadata * cache entries. * * mem_type: Instance of H5FD_mem_t, that is used to supply the * mem type passed into H5F_block_read(). * * flags: Flags indicating class-specific behavior. * * Whoever created the flags field neglected to document the meanings * of the flags he created. Hence the following discussion of the * H5C__CLASS_SPECULATIVE_LOAD_FLAG and (to a lesser extent) * H5C__CLASS_COMPRESSED_FLAG should be viewed with suspicion, * as the meanings are divined from the source code, and thus may be * inaccurate. Please correct any errors you find. * * Possible flags are: * * H5C__CLASS_NO_FLAGS_SET: No special processing. * * H5C__CLASS_SPECULATIVE_LOAD_FLAG: This flag appears to be used * only in H5C_load_entry(). When it is set, entries are * permitted to change their sizes on the first attempt * to load. * * If the new size is larger than the old, the read buffer * is reallocated to the new size, loaded from file, and the * deserialize routine is called a second time on the * new buffer. The entry returned by the first call to * the deserialize routine is discarded (via the free_icr * call) after the new size is retrieved (via the image_len * call). Note that the new size is used as the size of the * entry in the cache. * * If the new size is smaller than the old, no new loads * or desearializes are performed, but the new size becomes * the size of the entry in the cache. * * When this flag is set, an attempt to read past the * end of file is pemitted. In this case, if the size * returned get_load_size callback would result in a * read past the end of file, the size is trunkated to * avoid this, and processing proceeds as normal. * * H5C__CLASS_COMPRESSED_FLAG: This flags indicates that the entry * may be compressed -- i.e. its on disk image is run through * filters on the way to and from disk. Thus the uncompressed * (or unfiltered) size of the entry may be different from the * size of the entry in file. * * This has the following implications: * * On load, uncompressed size and load size may be different. * Presumably, load size will be smaller than uncompressed * size, but there is no requirement for this in the code * (but note that I have inserted an assertion to this effect, * which has not been triggered to date). * * On insertion, compressed (AKA filtered, AKA on disk) size * is unknown, as the entry has yet to be run through filters. * Compressed size is computed whenever the entry is * written (or the image is updated -- not relevant until * journaling is brought back). * * On dirty (of a clean entry), compressed (AKA filtered, * AKA on disk) size becomes unknown. Thus, compressed size * must be computed by the pre-serialize callback before the * entry may be written. * * Once the compressed size becomes unknown, it remains so * until the on disk image is constructed. * * Observe that the cache needs to know the size of the entry * for space allocation purposes. Since the compressed size * can change or become unknown, it uses the uncompressed * size which may change, but which should always be known. * The compressed size is used only for journaling and disk I/O. * * While there is no logical reason why they could not be * combined, due to absence of need and for simplicity of code, * the cache does not permit both the the * H5C__CLASS_COMPRESSED_FLAG and the * H5C__CLASS_SPECULATIVE_LOAD_FLAG to be set in the same * instance of H5C_class_t. * * The following flags may only appear in test code. * * H5C__CLASS_NO_IO_FLAG: This flag is intended only for use in test * code. When it is set, any attempt to load an entry of * the type with this flag set will trigger an assertion * failure, and any flush of an entry with this flag set * will not result in any write to file. * * H5C__CLASS_SKIP_READS: This flags is intended only for use in test * code. When it is set, reads on load will be skipped, * and an uninitialize buffer will be passed to the * deserialize function. * * H5C__CLASS_SKIP_WRITES: This flags is intended only for use in test * code. When it is set, writes of buffers prepared by the * serialize callback will be skipped. * * GET_LOAD_SIZE: Pointer to the 'get load size' function. * * This function must be able to determine the size of the disk image of * a metadata cache entry, given the 'udata' that will be passed to the * 'deserialize' callback. * * Note that if either the H5C__CLASS_SPECULATIVE_LOAD_FLAG or * the H5C__CLASS_COMPRESSED_FLAG is set, the disk image size * returned by this callback is either a first guess (if the * H5C__CLASS_SPECULATIVE_LOAD_FLAG is set) or (if the * H5C__CLASS_COMPRESSED_FLAG is set), the exact on disk size * of the entry whether it has been run through filters or not. * In all other cases, the value returned should be the correct * uncompressed size of the entry. * * The typedef for the deserialize callback is as follows: * * typedef herr_t (*H5C_get_load_size_func_t)(void *udata_ptr, * size_t *image_len_ptr); * * The parameters of the deserialize callback are as follows: * * udata_ptr: Pointer to user data provided in the protect call, which * will also be passed through to the deserialize callback. * * image_len_ptr: Pointer to the location in which the length in bytes * of the in file image to be deserialized is to be returned. * * This value is used by the cache to determine the size of * the disk image for the metadata, in order to read the disk * image from the file. * * Processing in the get_load_size function should proceed as follows: * * If successful, the function will place the length of the on disk * image associated with supplied user data in *image_len_ptr, and * then return SUCCEED. * * On failure, the function must return FAIL and push error information * onto the error stack with the error API routines, without modifying * the value pointed to by the image_len_ptr. * * * DESERIALIZE: Pointer to the deserialize function. * * This function must be able to read a buffer containing the on disk * image of a metadata cache entry, allocate and load the equivalent * in core representation, and return a pointer to that representation. * * The typedef for the deserialize callback is as follows: * * typedef void *(*H5C_deserialize_func_t)(const void * image_ptr, * size_t len, * void * udata_ptr, * boolean * dirty_ptr); * * The parameters of the deserialize callback are as follows: * * image_ptr: Pointer to a buffer of length len containing the * contents of the file starting at addr and continuing * for len bytes. * * len: Length in bytes of the in file image to be deserialized. * * This parameter is supplied mainly for sanity checking. * Sanity checks should be performed when compiled in debug * mode, but the parameter may be unused when compiled in * production mode. * * udata_ptr: Pointer to user data provided in the protect call, which * must be passed through to the deserialize callback. * * dirty_ptr: Pointer to boolean which the deserialize function * must use to mark the entry dirty if it has to modify * the entry to clean up file corruption left over from * an old bug in the HDF5 library. * * Processing in the deserialize function should proceed as follows: * * If the image contains valid data, and is of the correct length, * the deserialize function must allocate space for an in core * representation of that data, load the contents of the image into * the space allocated for the in core representation, and return * a pointer to the in core representation. Observe that an * instance of H5C_cache_entry_t must be the first item in this * representation. The cache will initialize it after the callback * returns. * * Note that the structure of the in core representation is otherwise * up to the cache client. All that is required is that the pointer * returned be sufficient for the clients purposes when it is returned * on a protect call. * * If the deserialize function has to clean up file corruption * left over from an old bug in the HDF5 library, it must set * *dirty_ptr to TRUE. If it doesn't, no action is needed as * *dirty_ptr will be set to FALSE before the deserialize call. * * If the operation fails for any reason (i.e. bad data in buffer, bad * buffer length, malloc failure, etc.) the function must return NULL and * push error information on the error stack with the error API routines. * * Exceptions to the above: * * If the H5C__CLASS_SPECULATIVE_LOAD_FLAG is set, the buffer supplied * to the function need not be currect on the first invocation of the * callback in any single attempt to load the entry. * * In this case, if the buffer is larger than necessary, the function * should load the entry as described above and not flag an error due * to the oversized buffer. The cache will correct its mis-apprehension * of the entry size with a subsequent call to the image_len callback. * * If the buffer is too small, and this is the first deserialize call * in the entry load operation, the function should not flag an error. * Instead, it must compute the correct size of the entry, allocate an * in core representation and initialize it to the extent that an * immediate call to the image len callback will return the correct * image size. * * In this case, when the deserialize callback returns, the cache will * call the image length callback, realize that the supplied buffer was * too small, discard the returned in core representation, allocate * and load a new buffer of the correct size from file, and then call * the deserialize callback again. * * If the H5C__CLASS_COMPRESSED_FLAG is set, exceptions are as per the * H5C__CLASS_SPECULATIVE_LOAD_FLAG, save that only oversized buffers * are permitted. * * * IMAGE_LEN: Pointer to the image length callback. * * This callback exists primarily to support * H5C__CLASS_SPECULATIVE_LOAD_FLAG and H5C__CLASS_COMPRESSED_FLAG * discussed above, although it is also used to obtain the size of * newly inserted entries. * * In the case of the H5C__CLASS_SPECULATIVE_LOAD_FLAG, it is used to * allow the client to change the size of an entry in the deserialize * callback. * * For the H5C__CLASS_COMPRESSED_FLAG, it is used to allow the client * to indicate whether the entry is compressed (i.e. whether entries * are run through filters) and if so, to report both the uncompressed * and the compressed entry size (i.e. the actual on disk size after * the entry has been run through filters) if that value is known. * * The callback is also used in H5C_insert_entry() to obtain the * size of the newly inserted entry. * * The typedef for the image_len callback is as follows: * * typedef herr_t (*H5C_image_len_func_t)(void *thing, * size_t *image_len_ptr, * hbool_t *compressed_ptr, * size_t *compressed_image_len_ptr); * * The parameters of the image_len callback are as follows: * * thing: Pointer to the in core representation of the entry. * * image_len_ptr: Pointer to size_t in which the callback will return * the length (in bytes) of the cache entry. * * If the H5C__CLASS_COMPRESSED_FLAG is not set in the * associated instance of H5C_class_t, or if the flag is * set, and the callback sets *compressed_ptr to FALSE, * this size is the actual size of the entry on disk. * * Otherwise, this size is the uncompressed size of the * entry -- which the cache will use for all purposes OTHER * than journal writes and disk I/O. * * compressed_ptr: Pointer to a boolean flag indicating whether * the cache entry will be compressed / uncompressed on * disk writes / reads. * * If the H5C__CLASS_COMPRESSED_FLAG is not set in the * associated instance of H5C_class_t, *compressed_ptr * must be set to FALSE. * * If the H5C__CLASS_COMPRESSED_FLAG is set in the * associated instance of H5C_class_t, and filters are * not enabled, *compressed_ptr must be set to FALSE. * * If the H5C__CLASS_COMPRESSED_FLAG is set in the * associated instance of H5C_class_t, and filters are * enabled, the callback must set *compressed_ptr to TRUE. * * Note that *compressed_ptr will always be set to FALSE * by the caller prior to invocation of the callback. Thus * callbacks for clients that don't set the * H5C__CLASS_COMPRESSED_FLAG can ignore this parameter. * * compressed_image_len_ptr: Pointer to size_t in which the callback * may return the length (in bytes) of the compressed on * disk image of the entry, or the uncompressed size if the * compressed size has not yet been calculated. * * Since computing the compressed image len is expensive, * the callback should only report the most recently computed * value -- which will typically be incorrect if the entry * is dirty. * * If *compressed_ptr is set to FALSE, *compressed_image_len_ptr * should be set to zero. However, as *compressed_image_len_ptr * will be initialize to zero prior to the call, the callback * need not modify it if the H5C__CLASS_COMPRESSED_FLAG is * not set. * * If the H5C__CLASS_COMPRESSED_FLAG is not set in the associated * instance of H5C_class_t, processing in the image_len function * should proceed as follows: * * If successful, the function will place the length of the on disk * image associated with the in core representation provided in the * thing parameter in *image_len_ptr, and then return SUCCEED. Since * *compressed_ptr and *compressed_image_len_ptr will be initialized to * FALSE and zero respectively before the call, the callback need not * modify these values, and may declare the associated parameters as * UNUSED. * * If the H5C__CLASS_COMPRESSED_FLAG is set in the associated * instance of H5C_class_t, processing in the image_len function * should proceed as follows: * * If successful, the function will place the uncompressed length of * the on disk image associated with the in core representation * provided in the thing parameter in *image_len_ptr. If filters * are not enabled for the entry, it will set *compressed_ptr to FALSE, * and *compressed_image_len_ptr to zero. If filters are enabled, * it will set *compressed_ptr to TRUE. In this case, it must set * *compressed_image_len_ptr equal to the last computed compressed * if the compressed size, or to the uncompressed size if that value * is yet to be computed. In all cases, it will return SUCCEED if * successful. * * In either case, if the function fails, it must return FAIL and * push error information onto the error stack with the error API * routines, and return without modifying the values pointed to by * the image_len_ptr, compressed_ptr, and compressed_image_len_ptr * parameters. * * PRE_SERIALIZE: Pointer to the pre-serialize callback. * * The pre-serialize callback is invoked by the metadata cache before * it needs a current on-disk image of the metadata entry for purposes * either constructing a journal or flushing the entry to disk. * * If the client needs to change the address or compressed or * uncompressed length of the entry prior to flush, the pre-serialize * callback is responsible for these actions, so that the actual * serialize callback (described below) is only responsible for * serializing the data structure, not moving it on disk or resizing it. * * In addition, the client may use the pre-serialize callback to * ensure that the entry is ready to be flushed -- in particular, * if the entry contains references to other entries that are in * temporary file space, the pre-serialize callback must move those * entries into real file space so that the serialzed entry will * contain no invalid data. * * One would think that the base address and length of * the length of the entry's image on disk would be well known. * However, that need not be the case as free space section info * entries will change size (and possibly location) depending on the * number of blocks of free space being manages, and fractal heap * direct blocks can change compressed size (and possibly location) * on serialization if compression is enabled. Similarly, it may * be necessary to move entries from temporary to real file space. * * The pre-serialize callback must report any such changes to the * cache, which must then update its internal structures as needed. * * The typedef for the pre-serialize callback is as follows: * * typedef herr_t (*H5C_pre_serialize_func_t)(const H5F_t *f, * hid_t dxpl_id, * void * thing, * haddr_t addr, * size_t len, * size_t compressed_len, * haddr_t * new_addr_ptr, * size_t * new_len_ptr, * size_t * new_compressed_len_ptr, * unsigned * flags_ptr); * * The parameters of the pre-serialize callback are as follows: * * f: File pointer -- needed if other metadata cache entries * must be modified in the process of serializing the * target entry. * * dxpl_id: dxpl_id passed with the file pointer to the cache, and * passed on to the callback. Necessary as some callbacks * revise the size and location of the target entry, or * possibly other entries on pre-serialize. * * thing: Pointer to void containing the address of the in core * representation of the target metadata cache entry. * This is the same pointer returned by a protect of the * addr and len given above. * * addr: Base address in file of the entry to be serialized. * * This parameter is supplied mainly for sanity checking. * Sanity checks should be performed when compiled in debug * mode, but the parameter may be unused when compiled in * production mode. * * len: Length in bytes of the in file image of the entry to be * serialized. Also the size the image passed to the * serialize callback (discussed below) unless either that * value is altered by this function, or the entry will be * compressed. In the latter case, the compressed size * of the entry will be reported in *new_compressed_len_ptr. * * This parameter is supplied mainly for sanity checking. * Sanity checks should be performed when compiled in debug * mode, but the parameter may be unused when compiled in * production mode. * * compressed_len: If the entry is to be compressed (i.e. run through * filters) prior to flush, Length in bytes of the last know * compressed size of the entry -- or the uncompressed size * if no such value exists (i.e. the entry has been inserted, * but never flushed). This parameter should be set to zero * in all other cases. * * This parameter is supplied mainly for sanity checking. * Sanity checks should be performed when compiled in debug * mode, but the parameter may be unused when compiled in * production mode. * * new_addr_ptr: Pointer to haddr_t. If the entry is moved by * the serialize function, the new on disk base address must * be stored in *new_addr_ptr, and the appropriate flag set * in *flags_ptr. * * If the entry is not moved by the serialize function, * *new_addr_ptr is undefined on pre-serialize callback * return. * * new_len_ptr: Pointer to size_t. If the entry is resized by the * serialize function, the new length of the on disk image * must be stored in *new_len_ptr, and the appropriate flag set * in *flags_ptr. * * If the entry is not resized by the pre-serialize function, * *new_len_ptr is undefined on pre-serialize callback * return. * * new_compressed_len_ptr: Pointer to size_t. If the image will be * compressed (i.e. run through filters) prior to being * written to disk, the compressed size (in bytes) of the * on disk image must be stored in *new_compressed_len_ptr, * and the appropriate flag set in *flags_ptr. * * flags_ptr: Pointer to an unsigned integer used to return flags * indicating whether the preserialize function resized or moved * the entry, or computed its compressed size. If the entry was * neither resized or moved, nor will be compressed, * the serialize function must set *flags_ptr to zero. * H5C__SERIALIZE_RESIZED_FLAG, H5C__SERIALIZE_MOVED_FLAG * and H5C__SERIALIZE_COMPRESSED_FLAG must be set to indicate * a resize, a move, or compression respectively. * * If the H5C__SERIALIZE_RESIZED_FLAG is set, the new length * must be stored in *new_len_ptr. * * If the H5C__SERIALIZE_MOVED_FLAG flag is set, the * new image base address must be stored in *new_addr_ptr. * * If the H5C__SERIALIZE_COMPRESSED_FLAG is set, the * compressed size of the new image must be stored in * *new_compressed_len_ptr. * * Processing in the pre-serialize function should proceed as follows: * * The pre-serialize function must examine the in core representation * indicated by the thing parameter, if the pre-serialize function does * not need to change the size or location of the on-disk image, or * compute its compress size, it must set *flags_ptr to zero. * * If the (uncompressed) size of the on-disk image must be changed, * the pre-serialize function must load the length of the new image * into *new_len_ptr, and set the H5C__SERIALIZE_RESIZED_FLAG in * *flags_ptr. * * If the base address of the on disk image must be changed, the * pre-serialize function must set *new_addr_ptr to the new base address, * and set the H5C__SERIALIZE_MOVED_FLAG in *flags_ptr. * * If the H5C__CLASS_COMPRESSED_FLAG is set in the assocated instance * of H5C_class_t, and filters (i.e. compression) are enabled, the * pre-serialize function must compute the compressed size of the * on disk image, and if it has changed, load this value into * *new_compressed_len_ptr, and set H5C__SERIALIZE_COMPRESSED_FLAG in * *flags_ptr. * * Note that to do this, the preserialize function will typically have * to serialize the entry, and run it through the filters to obtain * the compressed size. For efficiency, the compressed image may * be stored to be copied into the supplied buffer by the * serialize callback. Needless to say this is awkward. We may * want to re-work the API for cache clients to simplify this. * * In addition, the pre-serialize callback may perform any other * processing required before the entry is written to disk * * If it is successful, the function must return SUCCEED. * * If it fails for any reason, the function must return FAIL and * push error information on the error stack with the error API * routines. * * * SERIALIZE: Pointer to the serialize callback. * * The serialize callback is invoked by the metadata cache whenever * it needs a current on disk image of the metadata entry for purposes * either constructing a journal entry or flushing the entry to disk. * * At this point, the base address and length of the entry's image on * disk must be well known and not change during the serialization * process. * * While any size and/or location changes must have been handled * by a pre-serialize call, the client may elect to handle any other * changes to the entry required to place it in correct form for * writing to disk in this call. * * The typedef for the serialize callback is as follows: * * typedef herr_t (*H5C_serialize_func_t)(const H5F_t *f, * void * image_ptr, * size_t len, * void * thing); * * The parameters of the serialize callback are as follows: * * f: File pointer -- needed if other metadata cache entries * must be modified in the process of serializing the * target entry. * * image_ptr: Pointer to a buffer of length len bytes into which a * serialized image of the target metadata cache entry is * to be written. * * Note that this buffer will not in general be initialized * to any particular value. Thus the serialize function may * not assume any initial value and must set each byte in * the buffer. * * len: Length in bytes of the in file image of the entry to be * serialized. Also the size of *image_ptr (below). If * compression is not enabled, this value is simply the * uncompressed size of the entry's image on disk. If * compression is enabled, this value is the size of the * compressed image. * * This parameter is supplied mainly for sanity checking. * Sanity checks should be performed when compiled in debug * mode, but the parameter may be unused when compiled in * production mode. * * thing: Pointer to void containing the address of the in core * representation of the target metadata cache entry. * This is the same pointer returned by a protect of the * addr and len given above. * * Processing in the serialize function should proceed as follows: * * If there are any remaining changes to the entry required before * write to disk, they must be dealt with first. * * The serialize function must then examine the in core * representation indicated by the thing parameter, and write a * serialized (and possibly compressed) image of its contents into * the provided buffer. * * If it is successful, the function must return SUCCEED. * * If it fails for any reason, the function must return FAIL and * push error information on the error stack with the error API * routines. * * * NOTIFY: Pointer to the notify callback. * * The notify callback is invoked by the metadata cache when a cache * action on an entry has taken/will take place and the client indicates * it wishes to be notified about the action. * * The typedef for the notify callback is as follows: * * typedef herr_t (*H5C_notify_func_t)(H5C_notify_action_t action, * void *thing); * * The parameters of the notify callback are as follows: * * action: An enum indicating the metadata cache action that has taken/ * will take place. * * thing: Pointer to void containing the address of the in core * representation of the target metadata cache entry. This * is the same pointer that would be returned by a protect * of the addr and len of the entry. * * Processing in the notify function should proceed as follows: * * The notify function may perform any action it would like, including * metadata cache calls. * * If the function is successful, it must return SUCCEED. * * If it fails for any reason, the function must return FAIL and * push error information on the error stack with the error API * routines. * * * FREE_ICR: Pointer to the free ICR callback. * * The free ICR callback is invoked by the metadata cache when it * wishes to evict an entry, and needs the client to free the memory * allocated for the in core representation. * * The typedef for the free ICR callback is as follows: * * typedef herr_t (*H5C_free_icr_func_t)(void * thing)); * * The parameters of the free ICR callback are as follows: * * thing: Pointer to void containing the address of the in core * representation of the target metadata cache entry. This * is the same pointer that would be returned by a protect * of the addr and len of the entry. * * Processing in the free ICR function should proceed as follows: * * The free ICR function must free all memory allocated to the * in core representation. * * If the function is successful, it must return SUCCEED. * * If it fails for any reason, the function must return FAIL and * push error information on the error stack with the error API * routines. * * At least when compiled with debug, it would be useful if the * free ICR call would fail if the in core representation has been * modified since the last serialize of clear callback. * * CLEAR: Pointer to the clear callback. * * In principle, there should be no need for the clear callback, * as the dirty flag should be maintained by the metadata cache. *. * However, some clients maintain dirty bits on internal data, * and we need some way of keeping these dirty bits in sync with * those maintained by the metadata cache. This callback exists * to serve this purpose. If defined, it is called whenever the * cache marks dirty entry clean, or when the cache is about to * discard a dirty entry without writing it to disk (This * happens as the result of an unprotect call with the * H5AC__DELETED_FLAG set, and the H5C__TAKE_OWNERSHIP_FLAG not * set.) * * Arguably, this functionality should be in the NOTIFY callback. * However, this callback is specific to only a few clients, and * it will be called relatively frequently. Hence it is made its * own callback to minimize overhead. * * The typedef for the clear callback is as follows: * * typedef herr_t (*H5C_clear_func_t)(const H5F_t *f, * void * thing, * hbool_t about_to_destroy); * * The parameters of the clear callback are as follows: * * f: File pointer. * * thing: Pointer to void containing the address of the in core * representation of the target metadata cache entry. This * is the same pointer that would be returned by a protect() * call of the associated addr and len. * * about_to_destroy: Boolean flag used to indicate whether the * metadata cache is about to destroy the target metadata * cache entry. The callback may use this flag to omit * operations that are irrelevant it the entry is about * to be destroyed. * * Processing in the clear function should proceed as follows: * * Reset all internal dirty bits in the target metadata cache entry. * * If the about_to_destroy flag is TRUE, the clear function may * ommit any dirty bit that will not trigger a sanity check failure * or otherwise cause problems in the subsequent free icr call. * In particular, the call must ensure that the free icr call will * not fail due to changes prior to this call, and after the * last serialize or clear call. * * If the function is successful, it must return SUCCEED. * * If it fails for any reason, the function must return FAIL and * push error information on the error stack with the error API * routines. * * GET_FSF_SIZE: Pointer to the get file space free size callback. * * In principle, there is no need for the get file space free size * callback. However, as an optimization, it is sometimes convenient * to allocate and free file space for a number of cache entries * simultaneously in a single contiguous block of file space. * * File space allocation is done by the client, so the metadata cache * need not be involved. However, since the metadata cache typically * handles file space release when an entry is destroyed, some * adjustment on the part of the metadata cache is required for this * operation. * * The get file space free size callback exists to support this * operation. * * If a group of cache entries that were allocated as a group are to * be discarded and their file space released, the type of the first * (i.e. lowest address) entry in the group must implement the * get free file space size callback. * * To free the file space of all entries in the group in a single * operation, first expunge all entries other than the first without * the free file space flag. * * Then, to complete the operation, unprotect or expunge the first * entry in the block with the free file space flag set. Since * the get free file space callback is implemented, the metadata * cache will use this callback to get the size of the block to be * freed, instead of using the size of the entry as is done otherwise. * * At present this callback is used only by the H5FA and H5EA dblock * and dblock page client classes. * * The typedef for the clear callback is as follows: * * typedef herr_t (*H5C_get_fsf_size_t)(const void * thing, * size_t *fsf_size_ptr); * * The parameters of the clear callback are as follows: * * thing: Pointer to void containing the address of the in core * representation of the target metadata cache entry. This * is the same pointer that would be returned by a protect() * call of the associated addr and len. * * fs_size_ptr: Pointer to size_t in which the callback will return * the size of the piece of file space to be freed. Note * that the space to be freed is presumed to have the same * base address as the cache entry. * * The function simply returns the size of the block of file space * to be freed in *fsf_size_ptr. * * If the function is successful, it must return SUCCEED. * * If it fails for any reason, the function must return FAIL and * push error information on the error stack with the error API * routines. * ***************************************************************************/ /* Actions that can be reported to 'notify' client callback */ typedef enum H5C_notify_action_t { H5C_NOTIFY_ACTION_AFTER_INSERT, /* Entry has been added to the cache * via the insert call */ H5C_NOTIFY_ACTION_AFTER_LOAD, /* Entry has been loaded into the * from file via the protect call */ H5C_NOTIFY_ACTION_AFTER_FLUSH, /* Entry has just been flushed to * file. */ H5C_NOTIFY_ACTION_BEFORE_EVICT /* Entry is about to be evicted * from cache. */ } H5C_notify_action_t; /* Cache client callback function pointers */ typedef herr_t (*H5C_get_load_size_func_t)(const void *udata_ptr, size_t *image_len_ptr); typedef void *(*H5C_deserialize_func_t)(const void *image_ptr, size_t len, void *udata_ptr, hbool_t *dirty_ptr); typedef herr_t (*H5C_image_len_func_t)(const void *thing, size_t *image_len_ptr, hbool_t *compressed_ptr, size_t *compressed_image_len_ptr); typedef herr_t (*H5C_pre_serialize_func_t)(const H5F_t *f, hid_t dxpl_id, void *thing, haddr_t addr, size_t len, size_t compressed_len, haddr_t *new_addr_ptr, size_t *new_len_ptr, size_t *new_compressed_len_ptr, unsigned *flags_ptr); typedef herr_t (*H5C_serialize_func_t)(const H5F_t *f, void *image_ptr, size_t len, void *thing); typedef herr_t (*H5C_notify_func_t)(H5C_notify_action_t action, void *thing); typedef herr_t (*H5C_free_icr_func_t)(void *thing); typedef herr_t (*H5C_clear_func_t)(const H5F_t *f, void * thing, hbool_t about_to_destroy); typedef herr_t (*H5C_get_fsf_size_t)(const void * thing, size_t *fsf_size_ptr); /* Metadata cache client class definition */ typedef struct H5C_class_t { int id; const char * name; H5FD_mem_t mem_type; unsigned flags; H5C_get_load_size_func_t get_load_size; H5C_deserialize_func_t deserialize; H5C_image_len_func_t image_len; H5C_pre_serialize_func_t pre_serialize; H5C_serialize_func_t serialize; H5C_notify_func_t notify; H5C_free_icr_func_t free_icr; H5C_clear_func_t clear; H5C_get_fsf_size_t fsf_size; } H5C_class_t; /* Type defintions of callback functions used by the cache as a whole */ typedef herr_t (*H5C_write_permitted_func_t)(const H5F_t *f, hbool_t *write_permitted_ptr); typedef herr_t (*H5C_log_flush_func_t)(H5C_t *cache_ptr, haddr_t addr, hbool_t was_dirty, unsigned flags); /**************************************************************************** * * structure H5C_cache_entry_t * * Instances of the H5C_cache_entry_t structure are used to store cache * entries in a hash table and sometimes in a skip list. * See H5SL.c for the particulars of the skip list. * * In typical application, this structure is the first field in a * structure to be cached. For historical reasons, the external module * is responsible for managing the is_dirty field (this is no longer * completely true. See the comment on the is_dirty field for details). * All other fields are managed by the cache. * * The fields of this structure are discussed individually below: * * JRM - 4/26/04 * * magic: Unsigned 32 bit integer that must always be set to * H5C__H5C_CACHE_ENTRY_T_MAGIC when the entry is valid. * The field must be set to H5C__H5C_CACHE_ENTRY_T_BAD_MAGIC * just before the entry is freed. * * This is necessary, as the LRU list can be changed out * from under H5C_make_space_in_cache() by the serialize * callback which may change the size of an existing entry, * and/or load a new entry while serializing the target entry. * * This in turn can cause a recursive call to * H5C_make_space_in_cache() which may either flush or evict * the next entry that the first invocation of that function * was about to examine. * * The magic field allows H5C_make_space_in_cache() to * detect this case, and re-start its scan from the bottom * of the LRU when this situation occurs. * * cache_ptr: Pointer to the cache that this entry is contained within. * * addr: Base address of the cache entry on disk. * * size: Length of the cache entry on disk in bytes(exception: if * the entry is compressed on disk, this field contains the * uncompressed size of the entry -- see discussion of * compressed entries below). Note that unlike normal * caches, the entries in this cache are of arbitrary size. * * With the exception of compressed entries, the file space * allocations for cache entries implied by the addr and size * fields must be disjoint. For compressed entries, * the size field contains the uncompressed size -- thus in * in this case, substitution of compressed size for size * must result in disjoint file space allocations. However, * as discussed below, the compressed size may not be know. * * Any entry whose associated instance of H5C_class_t has the * H5C__CLASS_COMPRESSED_FLAG set may be compressed. When * an entry is compressed (that is, when filters are enabled * on it), the compressed flag (see below) must be set, and * the compressed size (if known), must be stored in * the compressed_size field. * * Since the compressed size will be unknown unless the * entry is clean, or has an up to date image (see the * image_ptr and image_up_to_date fields below), we use the * uncompressed size for all purposes other than disk I/O. * * compressed: Boolean flag that is set iff the instance of H5C_class_t * associated with the entry has the H5C__CLASS_COMPRESSED_FLAG * set, and filters are enabled on the entry. * * compressed_size: If compressed is TRUE, this field contains the actual * compressed size of the entry in bytes, which is also its * true size on disk -- or the uncompressed size if the * compressed size is unknown (i.e. the entry has been * inserted in the cache, but it has not been compressed yet). * Note that this value will usually be incorrect if the * entry is dirty. * * Since this value is frequently out of date and expensive to * compute, it is used only for disk I/O. The uncompressed * size of the entry (stored in the size field above) is used * for all other purposes (i.e. computing the sum of the sizes * of all entries in the cache, etc.). * * If compressed is FALSE, this field should contain 0. * * image_ptr: Pointer to void. When not NULL, this field points to a * dynamically allocated block of size bytes in which the * on disk image of the metadata cache entry is stored. * * If the entry is dirty, the pre-serialize and serialize * callbacks must be used to update this image before it is * written to disk * * image_up_to_date: Boolean flag that is set to TRUE when *image_ptr * is up to date, and set to false when the entry is dirtied. * * type: Pointer to the instance of H5C_class_t containing pointers * to the methods for cache entries of the current type. This * field should be NULL when the instance of H5C_cache_entry_t * is not in use. * * The name is not particularly descriptive, but is retained * to avoid changes in existing code. * * is_dirty: Boolean flag indicating whether the contents of the cache * entry has been modified since the last time it was written * to disk. * * NOTE: For historical reasons, this field is not maintained * by the cache. Instead, the module using the cache * sets this flag when it modifies the entry, and the * flush and clear functions supplied by that module * reset the dirty when appropriate. * * This is a bit quirky, so we may want to change this * someday. However it will require a change in the * cache interface. * * Update: Management of the is_dirty field has been largely * moved into the cache. The only remaining exceptions * are the flush and clear functions supplied by the * modules using the cache. These still clear the * is_dirty field as before. -- JRM 7/5/05 * * Update: Management of the is_dirty field is now entirely * in the cache. -- JRM 7/5/07 * * dirtied: Boolean flag used to indicate that the entry has been * dirtied while protected. * * This field is set to FALSE in the protect call, and may * be set to TRUE by the H5C_mark_entry_dirty() call at any * time prior to the unprotect call. * * The H5C_mark_entry_dirty() call exists as a convenience * function for the fractal heap code which may not know if * an entry is protected or pinned, but knows that is either * protected or pinned. The dirtied field was added as in * the parallel case, it is necessary to know whether a * protected entry is dirty prior to the protect call. * * is_protected: Boolean flag indicating whether this entry is protected * (or locked, to use more conventional terms). When it is * protected, the entry cannot be flushed or accessed until * it is unprotected (or unlocked -- again to use more * conventional terms). * * Note that protected entries are removed from the LRU lists * and inserted on the protected list. * * is_read_only: Boolean flag that is only meaningful if is_protected is * TRUE. In this circumstance, it indicates whether the * entry has been protected read only, or read/write. * * If the entry has been protected read only (i.e. is_protected * and is_read_only are both TRUE), we allow the entry to be * protected more than once. * * In this case, the number of readers is maintained in the * ro_ref_count field (see below), and unprotect calls simply * decrement that field until it drops to zero, at which point * the entry is actually unprotected. * * ro_ref_count: Integer field used to maintain a count of the number of * outstanding read only protects on this entry. This field * must be zero whenever either is_protected or is_read_only * are TRUE. * * is_pinned: Boolean flag indicating whether the entry has been pinned * in the cache. * * For very hot entries, the protect / unprotect overhead * can become excessive. Thus the cache has been extended * to allow an entry to be "pinned" in the cache. * * Pinning an entry in the cache has several implications: * * 1) A pinned entry cannot be evicted. Thus unprotected * pinned entries must be stored in the pinned entry * list, instead of being managed by the replacement * policy code (LRU at present). * * 2) A pinned entry can be accessed or modified at any time. * This places an extra burden on the pre-serialize and * serialize callbacks, which must ensure that a pinned * entry is consistant and ready to write to disk before * generating an image. * * 3) A pinned entry can be marked as dirty (and possibly * change size) while it is unprotected. * * 4) The flush-destroy code must allow pinned entries to * be unpinned (and possibly unprotected) during the * flush. * * JRM -- 3/16/06 * * in_slist: Boolean flag indicating whether the entry is in the skip list * As a general rule, entries are placed in the list when they * are marked dirty. However they may remain in the list after * being flushed. * * Update: Dirty entries are now removed from the skip list * when they are flushed. * * flush_marker: Boolean flag indicating that the entry is to be flushed * the next time H5C_flush_cache() is called with the * H5C__FLUSH_MARKED_ENTRIES_FLAG. The flag is reset when * the entry is flushed for whatever reason. * * flush_me_last: Boolean flag indicating that this entry should not be * flushed from the cache until all other entries without * the flush_me_last flag set have been flushed. * * flush_me_collectively: Boolean flag indicating that this entry needs * to be flushed collectively when in a parallel situation. * * Note: * * At this time, the flush_me_last and flush_me_collectively * flags will only be applied to one entry, the superblock, * and the code utilizing these flags is protected with HDasserts * to enforce this. This restraint can certainly be relaxed in * the future if the the need for multiple entries getting flushed * last or collectively arises, though the code allowing for that * will need to be expanded and tested appropriately if that * functionality is desired. * * Update: There are now two possible last entries * (superblock and file driver info message). This * number will probably increase as we add superblock * messages. JRM -- 11/18/14 * * clear_on_unprotect: Boolean flag used only in PHDF5. When H5C is used * to implement the metadata cache In the parallel case, only * the cache with mpi rank 0 is allowed to actually write to * file -- all other caches must retain dirty entries until they * are advised that the entry is clean. * * This flag is used in the case that such an advisory is * received when the entry is protected. If it is set when an * entry is unprotected, and the dirtied flag is not set in * the unprotect, the entry's is_dirty flag is reset by flushing * it with the H5C__FLUSH_CLEAR_ONLY_FLAG. * * flush_immediately: Boolean flag used only in Phdf5 -- and then only * for H5AC_METADATA_WRITE_STRATEGY__DISTRIBUTED. * * When a destributed metadata write is triggered at a * sync point, this field is used to mark entries that * must be flushed before leaving the sync point. At all * other times, this field should be set to FALSE. * * flush_in_progress: Boolean flag that is set to true iff the entry * is in the process of being flushed. This allows the cache * to detect when a call is the result of a flush callback. * * destroy_in_progress: Boolean flag that is set to true iff the entry * is in the process of being flushed and destroyed. * * * Fields supporting the 'flush dependency' feature: * * Entries in the cache may have a 'flush dependency' on another entry in the * cache. A flush dependency requires that all dirty child entries be flushed * to the file before a dirty parent entry (of those child entries) can be * flushed to the file. This can be used by cache clients to create data * structures that allow Single-Writer/Multiple-Reader (SWMR) access for the * data structure. * * The leaf child entry will have a "height" of 0, with any parent entries * having a height of 1 greater than the maximum height of any of their child * entries (flush dependencies are allowed to create asymmetric trees of * relationships). * * flush_dep_parent: Pointer to the parent entry for an entry in a flush * dependency relationship. * * child_flush_dep_height_rc: An array of reference counts for child entries, * where the number of children of each height is tracked. * * flush_dep_height: The height of the entry, which is one greater than the * maximum height of any of its child entries.. * * pinned_from_client: Whether the entry was pinned by an explicit pin request * from a cache client. * * pinned_from_cache: Whether the entry was pinned implicitly as a * request of being a parent entry in a flush dependency * relationship. * * * Fields supporting the hash table: * * Fields in the cache are indexed by a more or less conventional hash table. * If there are multiple entries in any hash bin, they are stored in a doubly * linked list. * * ht_next: Next pointer used by the hash table to store multiple * entries in a single hash bin. This field points to the * next entry in the doubly linked list of entries in the * hash bin, or NULL if there is no next entry. * * ht_prev: Prev pointer used by the hash table to store multiple * entries in a single hash bin. This field points to the * previous entry in the doubly linked list of entries in * the hash bin, or NULL if there is no previuos entry. * * * Fields supporting replacement policies: * * The cache must have a replacement policy, and it will usually be * necessary for this structure to contain fields supporting that policy. * * While there has been interest in several replacement policies for * this cache, the initial development schedule is tight. Thus I have * elected to support only a modified LRU policy for the first cut. * * When additional replacement policies are added, the fields in this * section will be used in different ways or not at all. Thus the * documentation of these fields is repeated for each replacement policy. * * Modified LRU: * * When operating in parallel mode, we must ensure that a read does not * cause a write. If it does, the process will hang, as the write will * be collective and the other processes will not know to participate. * * To deal with this issue, I have modified the usual LRU policy by adding * clean and dirty LRU lists to the usual LRU list. When reading in * parallel mode, we evict from the clean LRU list only. This implies * that we must try to ensure that the clean LRU list is reasonably well * stocked. See the comments on H5C_t in H5Cpkg.h for more details. * * Note that even if we start with a completely clean cache, a sequence * of protects without unprotects can empty the clean LRU list. In this * case, the cache must grow temporarily. At the next write, we will * attempt to evict enough entries to get the cache down to its nominal * maximum size. * * The use of the replacement policy fields under the Modified LRU policy * is discussed below: * * next: Next pointer in either the LRU, the protected list, or * the pinned list depending on the current values of * is_protected and is_pinned. If there is no next entry * on the list, this field should be set to NULL. * * prev: Prev pointer in either the LRU, the protected list, * or the pinned list depending on the current values of * is_protected and is_pinned. If there is no previous * entry on the list, this field should be set to NULL. * * aux_next: Next pointer on either the clean or dirty LRU lists. * This entry should be NULL when either is_protected or * is_pinned is true. * * When is_protected and is_pinned are false, and is_dirty is * true, it should point to the next item on the dirty LRU * list. * * When is_protected and is_pinned are false, and is_dirty is * false, it should point to the next item on the clean LRU * list. In either case, when there is no next item, it * should be NULL. * * aux_prev: Previous pointer on either the clean or dirty LRU lists. * This entry should be NULL when either is_protected or * is_pinned is true. * * When is_protected and is_pinned are false, and is_dirty is * true, it should point to the previous item on the dirty * LRU list. * * When is_protected and is_pinned are false, and is_dirty * is false, it should point to the previous item on the * clean LRU list. * * In either case, when there is no previous item, it should * be NULL. * * Cache entry stats collection fields: * * These fields should only be compiled in when both H5C_COLLECT_CACHE_STATS * and H5C_COLLECT_CACHE_ENTRY_STATS are true. When present, they allow * collection of statistics on individual cache entries. * * accesses: int32_t containing the number of times this cache entry has * been referenced in its lifetime. * * clears: int32_t containing the number of times this cache entry has * been cleared in its life time. * * flushes: int32_t containing the number of times this cache entry has * been flushed to file in its life time. * * pins: int32_t containing the number of times this cache entry has * been pinned in cache in its life time. * ****************************************************************************/ typedef struct H5C_cache_entry_t { uint32_t magic; H5C_t * cache_ptr; haddr_t addr; size_t size; hbool_t compressed; size_t compressed_size; void * image_ptr; hbool_t image_up_to_date; const H5C_class_t * type; haddr_t tag; hbool_t is_dirty; hbool_t dirtied; hbool_t is_protected; hbool_t is_read_only; int ro_ref_count; hbool_t is_pinned; hbool_t in_slist; hbool_t flush_marker; hbool_t flush_me_last; #ifdef H5_HAVE_PARALLEL hbool_t flush_me_collectively; hbool_t clear_on_unprotect; hbool_t flush_immediately; #endif /* H5_HAVE_PARALLEL */ hbool_t flush_in_progress; hbool_t destroy_in_progress; /* fields supporting the 'flush dependency' feature: */ struct H5C_cache_entry_t * flush_dep_parent; uint64_t child_flush_dep_height_rc[H5C__NUM_FLUSH_DEP_HEIGHTS]; unsigned flush_dep_height; hbool_t pinned_from_client; hbool_t pinned_from_cache; /* fields supporting the hash table: */ struct H5C_cache_entry_t * ht_next; struct H5C_cache_entry_t * ht_prev; /* fields supporting replacement policies: */ struct H5C_cache_entry_t * next; struct H5C_cache_entry_t * prev; struct H5C_cache_entry_t * aux_next; struct H5C_cache_entry_t * aux_prev; #if H5C_COLLECT_CACHE_ENTRY_STATS /* cache entry stats fields */ int32_t accesses; int32_t clears; int32_t flushes; int32_t pins; #endif /* H5C_COLLECT_CACHE_ENTRY_STATS */ } H5C_cache_entry_t; /**************************************************************************** * * structure H5C_auto_size_ctl_t * * Instances of H5C_auto_size_ctl_t are used to get and set the control * fields for automatic cache re-sizing. * * The fields of the structure are discussed individually below: * * version: Integer field containing the version number of this version * of the H5C_auto_size_ctl_t structure. Any instance of * H5C_auto_size_ctl_t passed to the cache must have a known * version number, or an error will be flagged. * * report_fcn: Pointer to the function that is to be called to report * activities each time the auto cache resize code is executed. If the * field is NULL, no call is made. * * If the field is not NULL, it must contain the address of a function * of type H5C_auto_resize_report_fcn. * * set_initial_size: Boolean flag indicating whether the size of the * initial size of the cache is to be set to the value given in * the initial_size field. If set_initial_size is FALSE, the * initial_size field is ignored. * * initial_size: If enabled, this field contain the size the cache is * to be set to upon receipt of this structure. Needless to say, * initial_size must lie in the closed interval [min_size, max_size]. * * min_clean_fraction: double in the range 0 to 1 indicating the fraction * of the cache that is to be kept clean. This field is only used * in parallel mode. Typical values are 0.1 to 0.5. * * max_size: Maximum size to which the cache can be adjusted. The * supplied value must fall in the closed interval * [MIN_MAX_CACHE_SIZE, MAX_MAX_CACHE_SIZE]. Also, max_size must * be greater than or equal to min_size. * * min_size: Minimum size to which the cache can be adjusted. The * supplied value must fall in the closed interval * [MIN_MAX_CACHE_SIZE, MAX_MAX_CACHE_SIZE]. Also, min_size must * be less than or equal to max_size. * * epoch_length: Number of accesses on the cache over which to collect * hit rate stats before running the automatic cache resize code, * if it is enabled. * * At the end of an epoch, we discard prior hit rate data and start * collecting afresh. The epoch_length must lie in the closed * interval [H5C__MIN_AR_EPOCH_LENGTH, H5C__MAX_AR_EPOCH_LENGTH]. * * * Cache size increase control fields: * * incr_mode: Instance of the H5C_cache_incr_mode enumerated type whose * value indicates how we determine whether the cache size should be * increased. At present there are two possible values: * * H5C_incr__off: Don't attempt to increase the size of the cache * automatically. * * When this increment mode is selected, the remaining fields * in the cache size increase section ar ignored. * * H5C_incr__threshold: Attempt to increase the size of the cache * whenever the average hit rate over the last epoch drops * below the value supplied in the lower_hr_threshold * field. * * Note that this attempt will fail if the cache is already * at its maximum size, or if the cache is not already using * all available space. * * lower_hr_threshold: Lower hit rate threshold. If the increment mode * (incr_mode) is H5C_incr__threshold and the hit rate drops below the * value supplied in this field in an epoch, increment the cache size by * size_increment. Note that cache size may not be incremented above * max_size, and that the increment may be further restricted by the * max_increment field if it is enabled. * * When enabled, this field must contain a value in the range [0.0, 1.0]. * Depending on the incr_mode selected, it may also have to be less than * upper_hr_threshold. * * increment: Double containing the multiplier used to derive the new * cache size from the old if a cache size increment is triggered. * The increment must be greater than 1.0, and should not exceed 2.0. * * The new cache size is obtained by multiplying the current max cache * size by the increment, and then clamping to max_size and to stay * within the max_increment as necessary. * * apply_max_increment: Boolean flag indicating whether the max_increment * field should be used to limit the maximum cache size increment. * * max_increment: If enabled by the apply_max_increment field described * above, this field contains the maximum number of bytes by which the * cache size can be increased in a single re-size. * * flash_incr_mode: Instance of the H5C_cache_flash_incr_mode enumerated * type whose value indicates whether and by what algorithm we should * make flash increases in the size of the cache to accomodate insertion * of large entries and large increases in the size of a single entry. * * The addition of the flash increment mode was occasioned by performance * problems that appear when a local heap is increased to a size in excess * of the current cache size. While the existing re-size code dealt with * this eventually, performance was very bad for the remainder of the * epoch. * * At present, there are two possible values for the flash_incr_mode: * * H5C_flash_incr__off: Don't perform flash increases in the size of * the cache. * * H5C_flash_incr__add_space: Let x be either the size of a newly * newly inserted entry, or the number of bytes by which the * size of an existing entry has been increased. * * If * x > flash_threshold * current max cache size, * * increase the current maximum cache size by x * flash_multiple * less any free space in the cache, and start a new epoch. For * now at least, pay no attention to the maximum increment. * * * With a little thought, it should be obvious that the above flash * cache size increase algorithm is not sufficient for all * circumstances -- for example, suppose the user round robins through * (1/flash_threshold) +1 groups, adding one data set to each on each * pass. Then all will increase in size at about the same time, requiring * the max cache size to at least double to maintain acceptable * performance, however the above flash increment algorithm will not be * triggered. * * Hopefully, the add space algorithm detailed above will be sufficient * for the performance problems encountered to date. However, we should * expect to revisit the issue. * * flash_multiple: Double containing the multiple described above in the * H5C_flash_incr__add_space section of the discussion of the * flash_incr_mode section. This field is ignored unless flash_incr_mode * is H5C_flash_incr__add_space. * * flash_threshold: Double containing the factor by which current max cache * size is multiplied to obtain the size threshold for the add_space * flash increment algorithm. The field is ignored unless * flash_incr_mode is H5C_flash_incr__add_space. * * * Cache size decrease control fields: * * decr_mode: Instance of the H5C_cache_decr_mode enumerated type whose * value indicates how we determine whether the cache size should be * decreased. At present there are four possibilities. * * H5C_decr__off: Don't attempt to decrease the size of the cache * automatically. * * When this increment mode is selected, the remaining fields * in the cache size decrease section are ignored. * * H5C_decr__threshold: Attempt to decrease the size of the cache * whenever the average hit rate over the last epoch rises * above the value supplied in the upper_hr_threshold * field. * * H5C_decr__age_out: At the end of each epoch, search the cache for * entries that have not been accessed for at least the number * of epochs specified in the epochs_before_eviction field, and * evict these entries. Conceptually, the maximum cache size * is then decreased to match the new actual cache size. However, * this reduction may be modified by the min_size, the * max_decrement, and/or the empty_reserve. * * H5C_decr__age_out_with_threshold: Same as age_out, but we only * attempt to reduce the cache size when the hit rate observed * over the last epoch exceeds the value provided in the * upper_hr_threshold field. * * upper_hr_threshold: Upper hit rate threshold. The use of this field * varies according to the current decr_mode: * * H5C_decr__off or H5C_decr__age_out: The value of this field is * ignored. * * H5C_decr__threshold: If the hit rate exceeds this threshold in any * epoch, attempt to decrement the cache size by size_decrement. * * Note that cache size may not be decremented below min_size. * * Note also that if the upper_threshold is 1.0, the cache size * will never be reduced. * * H5C_decr__age_out_with_threshold: If the hit rate exceeds this * threshold in any epoch, attempt to reduce the cache size * by evicting entries that have not been accessed for more * than the specified number of epochs. * * decrement: This field is only used when the decr_mode is * H5C_decr__threshold. * * The field is a double containing the multiplier used to derive the * new cache size from the old if a cache size decrement is triggered. * The decrement must be in the range 0.0 (in which case the cache will * try to contract to its minimum size) to 1.0 (in which case the * cache will never shrink). * * apply_max_decrement: Boolean flag used to determine whether decrements * in cache size are to be limited by the max_decrement field. * * max_decrement: Maximum number of bytes by which the cache size can be * decreased in a single re-size. Note that decrements may also be * restricted by the min_size of the cache, and (in age out modes) by * the empty_reserve field. * * epochs_before_eviction: Integer field used in H5C_decr__age_out and * H5C_decr__age_out_with_threshold decrement modes. * * This field contains the number of epochs an entry must remain * unaccessed before it is evicted in an attempt to reduce the * cache size. If applicable, this field must lie in the range * [1, H5C__MAX_EPOCH_MARKERS]. * * apply_empty_reserve: Boolean field controlling whether the empty_reserve * field is to be used in computing the new cache size when the * decr_mode is H5C_decr__age_out or H5C_decr__age_out_with_threshold. * * empty_reserve: To avoid a constant racheting down of cache size by small * amounts in the H5C_decr__age_out and H5C_decr__age_out_with_threshold * modes, this field allows one to require that any cache size * reductions leave the specified fraction of unused space in the cache. * * The value of this field must be in the range [0.0, 1.0]. I would * expect typical values to be in the range of 0.01 to 0.1. * ****************************************************************************/ enum H5C_resize_status { in_spec, increase, flash_increase, decrease, at_max_size, at_min_size, increase_disabled, decrease_disabled, not_full }; /* enum H5C_resize_conditions */ typedef void (*H5C_auto_resize_rpt_fcn)(H5C_t * cache_ptr, int32_t version, double hit_rate, enum H5C_resize_status status, size_t old_max_cache_size, size_t new_max_cache_size, size_t old_min_clean_size, size_t new_min_clean_size); typedef struct H5C_auto_size_ctl_t { /* general configuration fields: */ int32_t version; H5C_auto_resize_rpt_fcn rpt_fcn; hbool_t set_initial_size; size_t initial_size; double min_clean_fraction; size_t max_size; size_t min_size; int64_t epoch_length; /* size increase control fields: */ enum H5C_cache_incr_mode incr_mode; double lower_hr_threshold; double increment; hbool_t apply_max_increment; size_t max_increment; enum H5C_cache_flash_incr_mode flash_incr_mode; double flash_multiple; double flash_threshold; /* size decrease control fields: */ enum H5C_cache_decr_mode decr_mode; double upper_hr_threshold; double decrement; hbool_t apply_max_decrement; size_t max_decrement; int32_t epochs_before_eviction; hbool_t apply_empty_reserve; double empty_reserve; } H5C_auto_size_ctl_t; /***************************************/ /* Library-private Function Prototypes */ /***************************************/ H5_DLL H5C_t *H5C_create(size_t max_cache_size, size_t min_clean_size, int max_type_id, const char *(*type_name_table_ptr), H5C_write_permitted_func_t check_write_permitted, hbool_t write_permitted, H5C_log_flush_func_t log_flush, void *aux_ptr); H5_DLL void H5C_def_auto_resize_rpt_fcn(H5C_t *cache_ptr, int32_t version, double hit_rate, enum H5C_resize_status status, 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_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); H5_DLL herr_t H5C_flush_to_min_clean(H5F_t *f, hid_t dxpl_id); H5_DLL herr_t H5C_get_cache_auto_resize_config(const H5C_t *cache_ptr, H5C_auto_size_ctl_t *config_ptr); H5_DLL herr_t H5C_get_cache_size(H5C_t *cache_ptr, size_t *max_size_ptr, size_t *min_clean_size_ptr, size_t *cur_size_ptr, int32_t *cur_num_entries_ptr); H5_DLL herr_t H5C_get_cache_hit_rate(H5C_t *cache_ptr, double *hit_rate_ptr); H5_DLL herr_t H5C_get_entry_status(const H5F_t *f, haddr_t addr, size_t *size_ptr, hbool_t *in_cache_ptr, hbool_t *is_dirty_ptr, hbool_t *is_protected_ptr, hbool_t *is_pinned_ptr, hbool_t *is_flush_dep_parent_ptr, hbool_t *is_flush_dep_child_ptr); H5_DLL herr_t H5C_get_evictions_enabled(const H5C_t *cache_ptr, hbool_t *evictions_enabled_ptr); H5_DLL FILE *H5C_get_trace_file_ptr(const H5C_t *cache_ptr); H5_DLL FILE *H5C_get_trace_file_ptr_from_entry(const H5C_cache_entry_t *entry_ptr); 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_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); H5_DLL herr_t H5C_create_flush_dependency(void *parent_thing, void *child_thing); H5_DLL void * H5C_protect(H5F_t *f, hid_t dxpl_id, const H5C_class_t *type, haddr_t addr, void *udata, unsigned flags); H5_DLL herr_t H5C_reset_cache_hit_rate_stats(H5C_t *cache_ptr); H5_DLL herr_t H5C_resize_entry(void *thing, size_t new_size); H5_DLL herr_t H5C_set_cache_auto_resize_config(H5C_t *cache_ptr, H5C_auto_size_ctl_t *config_ptr); H5_DLL herr_t H5C_set_evictions_enabled(H5C_t *cache_ptr, hbool_t evictions_enabled); H5_DLL herr_t H5C_set_prefix(H5C_t *cache_ptr, char *prefix); H5_DLL herr_t H5C_set_trace_file_ptr(H5C_t *cache_ptr, FILE *trace_file_ptr); H5_DLL herr_t H5C_stats(H5C_t *cache_ptr, const char *cache_name, hbool_t display_detailed_stats); H5_DLL void H5C_stats__reset(H5C_t *cache_ptr); H5_DLL herr_t H5C_dump_cache(H5C_t *cache_ptr, const char *cache_name); H5_DLL herr_t H5C_unpin_entry(void *thing); H5_DLL herr_t H5C_destroy_flush_dependency(void *parent_thing, void *child_thing); H5_DLL herr_t H5C_unprotect(H5F_t *f, hid_t dxpl_id, haddr_t addr, void *thing, unsigned int flags); H5_DLL herr_t H5C_validate_resize_config(H5C_auto_size_ctl_t *config_ptr, unsigned int tests); H5_DLL herr_t H5C_ignore_tags(H5C_t *cache_ptr); H5_DLL void H5C_retag_copied_metadata(H5C_t *cache_ptr, haddr_t metadata_tag); #ifdef H5_HAVE_PARALLEL H5_DLL herr_t H5C_apply_candidate_list(H5F_t *f, hid_t dxpl_id, H5C_t *cache_ptr, int num_candidates, haddr_t *candidates_list_ptr, int mpi_rank, int mpi_size); H5_DLL herr_t H5C_construct_candidate_list__clean_cache(H5C_t *cache_ptr); H5_DLL herr_t H5C_construct_candidate_list__min_clean(H5C_t *cache_ptr); H5_DLL herr_t H5C_mark_entries_as_clean(H5F_t *f, hid_t dxpl_id, int32_t ce_array_len, haddr_t *ce_array_ptr); #endif /* H5_HAVE_PARALLEL */ #ifndef NDEBUG /* debugging functions */ H5_DLL herr_t H5C_get_entry_ptr_from_addr(const H5F_t *f, haddr_t addr, void **entry_ptr_ptr); H5_DLL herr_t H5C_verify_entry_type(const H5F_t *f, haddr_t addr, const H5C_class_t *expected_type, hbool_t *in_cache_ptr, hbool_t *type_ok_ptr); #endif /* NDEBUG */ #endif /* !_H5Cprivate_H */