From e53af0ed745a7bec1795681c473ba54025814fd5 Mon Sep 17 00:00:00 2001 From: Dana Robinson Date: Tue, 21 Aug 2012 06:47:54 -0500 Subject: [svn-r22703] Description: Propagated Windows and POSIX correctness changes from SEC2 VFD to the MPI-POSIX VFD. Fixes HDFFV-8058. Tested on: jam w/ parallel and fortran enabled --- src/H5FDmpiposix.c | 1088 ++++++++++++++++++++++++++-------------------------- src/H5FDmpiposix.h | 9 +- 2 files changed, 541 insertions(+), 556 deletions(-) diff --git a/src/H5FDmpiposix.c b/src/H5FDmpiposix.c index 4ed3a4d..261f427 100644 --- a/src/H5FDmpiposix.c +++ b/src/H5FDmpiposix.c @@ -37,15 +37,15 @@ #define H5_INTERFACE_INIT_FUNC H5FD_mpiposix_init_interface -#include "H5private.h" /* Generic Functions */ -#include "H5ACprivate.h" /* Metadata cache */ -#include "H5Eprivate.h" /* Error handling */ -#include "H5Fprivate.h" /* File access */ -#include "H5FDprivate.h" /* File drivers */ -#include "H5FDmpi.h" /* MPI-based file drivers */ -#include "H5Iprivate.h" /* IDs */ -#include "H5MMprivate.h" /* Memory management */ -#include "H5Pprivate.h" /* Property lists */ +#include "H5private.h" /* Generic Functions */ +#include "H5ACprivate.h" /* Metadata cache */ +#include "H5Eprivate.h" /* Error handling */ +#include "H5Fprivate.h" /* File access */ +#include "H5FDprivate.h" /* File drivers */ +#include "H5FDmpi.h" /* MPI-based file drivers */ +#include "H5Iprivate.h" /* IDs */ +#include "H5MMprivate.h" /* Memory management */ +#include "H5Pprivate.h" /* Property lists */ /* Features: * H5_HAVE_GPFS -- issue gpfs_fcntl() calls to hopefully improve @@ -71,107 +71,101 @@ */ static hid_t H5FD_MPIPOSIX_g = 0; -/* File operations */ -#define OP_UNKNOWN 0 -#define OP_READ 1 -#define OP_WRITE 2 +/* Since Windows doesn't follow the rest of the world when it comes + * to POSIX I/O types, some typedefs and constants are needed to avoid + * making the code messy with #ifdefs. + */ +#ifdef H5_HAVE_WIN32_API +typedef unsigned int h5_mpiposix_io_t; +typedef int h5_mpiposix_io_ret_t; +static int H5_MPIPOSIX_MAX_IO_BYTES_g = INT_MAX; +#else +/* Unix, everyone else */ +typedef size_t h5_mpiposix_io_t; +typedef ssize_t h5_mpiposix_io_ret_t; +static size_t H5_MPIPOSIX_MAX_IO_BYTES_g = SSIZET_MAX; +#endif /* H5_HAVE_WIN32_API */ /* * The description of a file belonging to this driver. - * The EOF value - * is only used just after the file is opened in order for the library to - * determine whether the file is empty, truncated, or okay. The MPIPOSIX driver - * doesn't bother to keep it updated since it's an expensive operation. + * The EOF value is only used just after the file is opened in order for the + * library to determine whether the file is empty, truncated, or okay. The + * MPIPOSIX driver doesn't bother to keep it updated since it's an expensive + * operation. */ typedef struct H5FD_mpiposix_t { - H5FD_t pub; /*public stuff, must be first */ - int fd; /*the unix file handle */ - MPI_Comm comm; /*communicator */ - int mpi_rank; /* This process's rank */ - int mpi_size; /* Total number of processes */ - haddr_t eof; /*end-of-file marker */ - haddr_t eoa; /*end-of-address marker */ - haddr_t last_eoa; /* Last known end-of-address marker */ - haddr_t pos; /* Current file I/O position */ - int op; /* Last file I/O operation */ - hsize_t naccess; /* Number of (write) accesses to file */ + H5FD_t pub; /* public stuff, must be first */ + int fd; /* the unix file handle */ + MPI_Comm comm; /* communicator */ + int mpi_rank; /* This process's rank */ + int mpi_size; /* Total number of processes */ + haddr_t eof; /* end-of-file marker */ + haddr_t eoa; /* end-of-address marker */ + haddr_t last_eoa; /* Last known end-of-address marker */ + haddr_t pos; /* Current file I/O position */ + H5FD_file_op_t op; /* Last file I/O operation */ + hsize_t naccess; /* Number of (write) accesses to file */ #ifdef H5_HAVE_GPFS - size_t blksize; /* Block size of file system */ + size_t blksize; /* Block size of file system */ #endif - hbool_t use_gpfs; /* Use GPFS to write things */ + hbool_t use_gpfs; /* Use GPFS to write things */ #ifndef H5_HAVE_WIN32_API - /* - * On most systems the combination of device and i-node number uniquely - * identify a file. + /* On most systems the combination of device and i-node number uniquely + * identify a file. Note that Cygwin, MinGW and other Windows POSIX + * environments have the stat function (which fakes inodes) + * and will use the 'device + inodes' scheme as opposed to the + * Windows code further below. */ - dev_t device; /*file device number */ - ino_t inode; /*file i-node number */ + dev_t device; /* file device number */ +#ifdef H5_VMS + ino_t inode[3]; /* file i-node number */ #else - /* - * On H5_HAVE_WIN32_API the low-order word of a unique identifier associated with the - * file and the volume serial number uniquely identify a file. This number - * (which, both? -rpm) may change when the system is restarted or when the - * file is opened. After a process opens a file, the identifier is - * constant until the file is closed. An application can use this - * identifier and the volume serial number to determine whether two - * handles refer to the same file. + ino_t inode; /* file i-node number */ +#endif /* H5_VMS */ +#else + /* Files in windows are uniquely identified by the volume serial + * number and the file index (both low and high parts). + * + * There are caveats where these numbers can change, especially + * on FAT file systems. On NTFS, however, a file should keep + * those numbers the same until renamed or deleted (though you + * can use ReplaceFile() on NTFS to keep the numbers the same + * while renaming). + * + * See the MSDN "BY_HANDLE_FILE_INFORMATION Structure" entry for + * more information. + * + * http://msdn.microsoft.com/en-us/library/aa363788(v=VS.85).aspx */ - int fileindexlo; - int fileindexhi; -#endif + DWORD nFileIndexLow; + DWORD nFileIndexHigh; + DWORD dwVolumeSerialNumber; + + HANDLE hFile; /* Native windows file handle */ +#endif /* H5_HAVE_WIN32_API */ } H5FD_mpiposix_t; /* - * This driver supports systems that have the lseek64() function by defining - * some macros here so we don't have to have conditional compilations later - * throughout the code. - * - * file_offset_t: The datatype for file offsets, the second argument of - * the lseek() or lseek64() call. - * - * file_seek: The function which adjusts the current file position, - * either lseek() or lseek64(). - */ -/* adding for windows NT file system support. */ - -#ifdef H5_HAVE_LSEEK64 -# define file_offset_t off64_t -# define file_seek lseek64 -# define file_truncate ftruncate64 -#elif defined (H5_HAVE_WIN32_API) -# /*MSVC*/ -# define file_offset_t __int64 -# define file_seek _lseeki64 -# define file_truncate _ftruncatei64 -#else -# define file_offset_t off_t -# define file_seek HDlseek -# define file_truncate HDftruncate -#endif - -/* * These macros check for overflow of various quantities. These macros - * assume that file_offset_t is signed and haddr_t and size_t are unsigned. + * assume that HDoff_t is signed and haddr_t and size_t are unsigned. * - * ADDR_OVERFLOW: Checks whether a file address of type `haddr_t' - * is too large to be represented by the second argument - * of the file seek function. + * ADDR_OVERFLOW: Checks whether a file address of type `haddr_t' + * is too large to be represented by the second argument + * of the file seek function. * - * SIZE_OVERFLOW: Checks whether a buffer size of type `hsize_t' is too - * large to be represented by the `size_t' type. + * SIZE_OVERFLOW: Checks whether a buffer size of type `hsize_t' is too + * large to be represented by the `size_t' type. * - * REGION_OVERFLOW: Checks whether an address and size pair describe data - * which can be addressed entirely by the second - * argument of the file seek function. + * REGION_OVERFLOW: Checks whether an address and size pair describe data + * which can be addressed entirely by the second + * argument of the file seek function. */ -#define MAXADDR (((haddr_t)1<<(8*sizeof(file_offset_t)-1))-1) -#define ADDR_OVERFLOW(A) (HADDR_UNDEF==(A) || \ - ((A) & ~(haddr_t)MAXADDR)) -#define SIZE_OVERFLOW(Z) ((Z) & ~(hsize_t)MAXADDR) -#define REGION_OVERFLOW(A,Z) (ADDR_OVERFLOW(A) || SIZE_OVERFLOW(Z) || \ - sizeof(file_offset_t)comm, comm))) - HMPI_GOTO_ERROR(FAIL, "MPI_Comm_dup failed", mpi_code) + if (MPI_SUCCESS != (mpi_code = MPI_Comm_dup(fa->comm, comm))) + HMPI_GOTO_ERROR(FAIL, "MPI_Comm_dup failed", mpi_code) } if (use_gpfs) @@ -450,45 +422,39 @@ done: /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_fapl_get - * - * Purpose: Returns a file access property list which could be used to - * create another file the same as this one. + * Function: H5FD_mpiposix_fapl_get * - * Return: Success: Ptr to new file access property list with all - * fields copied from the file pointer. + * Purpose: Returns a file access property list which could be used to + * create another file the same as this one. * - * Failure: NULL + * Return: Success: Ptr to new file access property list with all + * fields copied from the file pointer. + * Failure: NULL * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * Albert Cheng, 2003-04-24 - * Duplicate the communicator object so that the new - * property list is insulated from the old one. - * *------------------------------------------------------------------------- */ static void * H5FD_mpiposix_fapl_get(H5FD_t *_file) { - H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; - H5FD_mpiposix_fapl_t *fa = NULL; - int mpi_code; /* MPI return code */ - void *ret_value; /* Return value */ + H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; + H5FD_mpiposix_fapl_t *fa = NULL; + int mpi_code; /* MPI return code */ + void *ret_value; /* Return value */ FUNC_ENTER_NOAPI_NOINIT - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); - if (NULL==(fa=H5MM_calloc(sizeof(H5FD_mpiposix_fapl_t)))) + if (NULL == (fa = H5MM_calloc(sizeof(H5FD_mpiposix_fapl_t)))) HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, NULL, "memory allocation failed") /* Duplicate the communicator. */ if (MPI_SUCCESS != (mpi_code=MPI_Comm_dup(file->comm, &fa->comm))) - HMPI_GOTO_ERROR(NULL, "MPI_Comm_dup failed", mpi_code) + HMPI_GOTO_ERROR(NULL, "MPI_Comm_dup failed", mpi_code) fa->use_gpfs = file->use_gpfs; @@ -501,19 +467,16 @@ done: /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_fapl_copy + * Function: H5FD_mpiposix_fapl_copy * - * Purpose: Copies the mpiposix-specific file access properties. + * Purpose: Copies the mpiposix-specific file access properties. * - * Return: Success: Ptr to a new property list - * - * Failure: NULL + * Return: Success: Ptr to a new property list + * Failure: NULL * * Programmer: Albert Cheng * Apr 24, 2003 * - * Modifications: - * *------------------------------------------------------------------------- */ static void * @@ -526,24 +489,24 @@ H5FD_mpiposix_fapl_copy(const void *_old_fa) FUNC_ENTER_NOAPI_NOINIT - if (NULL==(new_fa=H5MM_malloc(sizeof(H5FD_mpiposix_fapl_t)))) + if (NULL == (new_fa = H5MM_malloc(sizeof(H5FD_mpiposix_fapl_t)))) HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, NULL, "memory allocation failed") /* Copy the general information */ HDmemcpy(new_fa, old_fa, sizeof(H5FD_mpiposix_fapl_t)); /* Duplicate communicator. */ - if (MPI_SUCCESS != (mpi_code=MPI_Comm_dup(old_fa->comm, &new_fa->comm))) - HMPI_GOTO_ERROR(NULL, "MPI_Comm_dup failed", mpi_code) + if (MPI_SUCCESS != (mpi_code = MPI_Comm_dup(old_fa->comm, &new_fa->comm))) + HMPI_GOTO_ERROR(NULL, "MPI_Comm_dup failed", mpi_code) new_fa->use_gpfs = old_fa->use_gpfs; ret_value = new_fa; done: if (NULL == ret_value){ - /* cleanup */ - if (new_fa) - H5MM_xfree(new_fa); + /* cleanup */ + if (new_fa) + H5MM_xfree(new_fa); } FUNC_LEAVE_NOAPI(ret_value) @@ -551,19 +514,15 @@ done: /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_fapl_free + * Function: H5FD_mpiposix_fapl_free * - * Purpose: Frees the mpiposix-specific file access properties. + * Purpose: Frees the mpiposix-specific file access properties. * - * Return: Success: 0 - * - * Failure: -1 + * Return: SUCCEED (can't fail) * * Programmer: Albert Cheng * Apr 24, 2003 * - * Modifications: - * *------------------------------------------------------------------------- */ static herr_t @@ -573,10 +532,10 @@ H5FD_mpiposix_fapl_free(void *_fa) FUNC_ENTER_NOAPI_NOINIT_NOERR - assert(fa); + HDassert(fa); /* Free the internal communicator */ - assert(MPI_COMM_NULL!=fa->comm); + HDassert(MPI_COMM_NULL != fa->comm); MPI_Comm_free(&fa->comm); H5MM_xfree(fa); @@ -588,53 +547,46 @@ H5FD_mpiposix_fapl_free(void *_fa) * Function: H5FD_mpiposix_open * * Purpose: Opens a file with name NAME. The FLAGS are a bit field with - * purpose similar to the second argument of open(2) and which - * are defined in H5Fpublic.h. The file access property list - * FAPL_ID contains the properties driver properties and MAXADDR - * is the largest address which this file will be expected to - * access. This is collective. + * purpose similar to the second argument of open(2) and which + * are defined in H5Fpublic.h. The file access property list + * FAPL_ID contains the properties driver properties and MAXADDR + * is the largest address which this file will be expected to + * access. This is collective. * - * Return: Success: A new file pointer. - * Failure: NULL + * Return: Success: A new file pointer. + * Failure: NULL * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * Albert Cheng, 2003-04-24 - * Duplicate the communicator so that file is insulated from the - * old one. - * *------------------------------------------------------------------------- */ static H5FD_t * H5FD_mpiposix_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxaddr) { - H5FD_mpiposix_t *file=NULL; /* New MPIPOSIX file struct */ - int o_flags; /* Flags for file open call */ - int fd=(-1); /* File handle for file opened */ - int mpi_rank; /* MPI rank of this process */ - int mpi_size; /* Total number of MPI processes */ - int mpi_code; /* mpi return code */ - const H5FD_mpiposix_fapl_t *fa=NULL; /* MPIPOSIX file access property list information */ - H5FD_mpiposix_fapl_t _fa; /* Private copy of default file access property list information */ - H5P_genplist_t *plist; /* Property list pointer */ - h5_stat_t sb; /* Portable 'stat' struct */ + H5FD_mpiposix_t *file = NULL; /* New MPIPOSIX file struct */ + int o_flags; /* Flags for file open call */ + int fd = -1; /* File handle for file opened */ + int mpi_rank; /* MPI rank of this process */ + int mpi_size; /* Total number of MPI processes */ + int mpi_code; /* mpi return code */ + const H5FD_mpiposix_fapl_t *fa = NULL; /* MPIPOSIX file access property list information */ + H5FD_mpiposix_fapl_t _fa; /* Private copy of default file access property list information */ + H5P_genplist_t *plist; /* Property list pointer */ + h5_stat_t sb; /* Portable 'stat' struct */ #ifdef H5_HAVE_WIN32_API - HFILE filehandle; struct _BY_HANDLE_FILE_INFORMATION fileinfo; - int results; #endif - H5FD_t *ret_value=NULL; /* Return value */ - MPI_Comm comm_dup=MPI_COMM_NULL; + H5FD_t *ret_value = NULL; /* Return value */ + MPI_Comm comm_dup = MPI_COMM_NULL; FUNC_ENTER_NOAPI_NOINIT /* Check arguments */ if (!name || !*name) HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, NULL, "invalid file name") - if (0==maxaddr || HADDR_UNDEF==maxaddr) + if (0 == maxaddr || HADDR_UNDEF == maxaddr) HGOTO_ERROR(H5E_ARGS, H5E_BADRANGE, NULL, "bogus maxaddr") if (ADDR_OVERFLOW(maxaddr)) HGOTO_ERROR(H5E_ARGS, H5E_OVERFLOW, NULL, "bogus maxaddr") @@ -642,31 +594,31 @@ H5FD_mpiposix_open(const char *name, unsigned flags, hid_t fapl_id, /* Obtain a pointer to mpiposix-specific file access properties */ if(NULL == (plist = H5P_object_verify(fapl_id,H5P_FILE_ACCESS))) HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, NULL, "not a file access property list") - if (H5P_FILE_ACCESS_DEFAULT==fapl_id || H5FD_MPIPOSIX!=H5P_get_driver(plist)) { - _fa.comm = MPI_COMM_SELF; /*default*/ + if (H5P_FILE_ACCESS_DEFAULT == fapl_id || H5FD_MPIPOSIX != H5P_get_driver(plist)) { + _fa.comm = MPI_COMM_SELF; /*default*/ _fa.use_gpfs = FALSE; - fa = &_fa; + fa = &_fa; } /* end if */ else { - fa = H5P_get_driver_info(plist); - assert(fa); + fa = H5P_get_driver_info(plist); + HDassert(fa); } /* end else */ /* Duplicate the communicator for use by this file. */ - if (MPI_SUCCESS != (mpi_code=MPI_Comm_dup(fa->comm, &comm_dup))) - HMPI_GOTO_ERROR(NULL, "MPI_Comm_dup failed", mpi_code) + if (MPI_SUCCESS != (mpi_code = MPI_Comm_dup(fa->comm, &comm_dup))) + HMPI_GOTO_ERROR(NULL, "MPI_Comm_dup failed", mpi_code) /* Get the MPI rank of this process and the total number of processes */ - if (MPI_SUCCESS != (mpi_code=MPI_Comm_rank (comm_dup, &mpi_rank))) + if (MPI_SUCCESS != (mpi_code = MPI_Comm_rank (comm_dup, &mpi_rank))) HMPI_GOTO_ERROR(NULL, "MPI_Comm_rank failed", mpi_code) - if (MPI_SUCCESS != (mpi_code=MPI_Comm_size (comm_dup, &mpi_size))) + if (MPI_SUCCESS != (mpi_code = MPI_Comm_size (comm_dup, &mpi_size))) HMPI_GOTO_ERROR(NULL, "MPI_Comm_size failed", mpi_code) /* Build the open flags */ o_flags = (H5F_ACC_RDWR & flags) ? O_RDWR : O_RDONLY; /* Only set the creation flag(s) for process 0 */ - if(mpi_rank==0) { + if(0 == mpi_rank) { if (H5F_ACC_TRUNC & flags) o_flags |= O_TRUNC; if (H5F_ACC_CREAT & flags) @@ -680,13 +632,14 @@ H5FD_mpiposix_open(const char *name, unsigned flags, hid_t fapl_id, * open (never create) the file and all processes proceed. */ /* Process 0 opens (or creates) file and broadcasts result to other processes */ - if(mpi_rank==0) { + if(0 == mpi_rank) { /* Open the file */ - fd=HDopen(name, o_flags, 0666); + fd = HDopen(name, o_flags, 0666); } /* end if */ - /* Broadcast the results of the open() from process 0 */ - /* This is necessary because of the "tentative open" code in H5F_open() + /* Broadcast the results of the open() from process 0 + * + * This is necessary because of the "tentative open" code in H5F_open() * where the file is attempted to be opened with different flags from the * user's, in order to check for the file's existence, etc. Here, process 0 * gets different flags from the other processes (since it is in charge of @@ -698,34 +651,36 @@ H5FD_mpiposix_open(const char *name, unsigned flags, hid_t fapl_id, HMPI_GOTO_ERROR(NULL, "MPI_Bcast failed", mpi_code) /* If the file open on process 0 failed, bail out on all processes now */ - if(fd<0) + if(fd < 0) HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to open file") /* Other processes (non 0) wait for broadcast result from process 0 and then open file */ - if(mpi_rank!=0) { + if(mpi_rank != 0) { /* Open the file */ - if ((fd=HDopen(name, o_flags, 0666))<0) + if ((fd = HDopen(name, o_flags, 0666)) < 0) HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to open file") } /* end if */ /* Process 0 fstat()s the file and broadcasts the results to the other processes */ - if(mpi_rank==0) { + if(0 == mpi_rank) { /* Get the stat information */ - if (HDfstat(fd, &sb)<0) + if (HDfstat(fd, &sb) < 0) HGOTO_ERROR(H5E_FILE, H5E_BADFILE, NULL, "unable to fstat file") } /* end if */ /* Broadcast the results of the fstat() from process 0 */ - if (MPI_SUCCESS != (mpi_code= MPI_Bcast(&sb, sizeof(h5_stat_t), MPI_BYTE, 0, comm_dup))) + if (MPI_SUCCESS != (mpi_code = MPI_Bcast(&sb, sizeof(h5_stat_t), MPI_BYTE, 0, comm_dup))) HMPI_GOTO_ERROR(NULL, "MPI_Bcast failed", mpi_code) #ifdef H5_HAVE_GPFS if (fa->use_gpfs) { /* - * Free all byte range tokens. This is a good thing to do if raw data is aligned on 256kB boundaries (a GPFS page is - * 256kB). Care should be taken that there aren't too many sub-page writes, or the mmfsd may become overwhelmed. This - * should probably eventually be passed down here as a property. The gpfs_fcntl() will most likely fail if `fd' isn't - * on a GPFS file system. */ + * Free all byte range tokens. This is a good thing to do if raw data + * is aligned on 256kB boundaries (a GPFS page is 256kB). Care should + * be taken that there aren't too many sub-page writes, or the mmfsd + * may become overwhelmed. This should probably eventually be passed + * down here as a property. The gpfs_fcntl() will most likely fail if + * 'fd' isn't on a GPFS file system. */ struct { gpfsFcntlHeader_t hdr; gpfsFreeRange_t fr; @@ -738,13 +693,13 @@ H5FD_mpiposix_open(const char *name, unsigned flags, hid_t fapl_id, hint.fr.start = 0; hint.fr.length = 0; - if (gpfs_fcntl(fd, &hint)<0) + if (gpfs_fcntl(fd, &hint) < 0) HGOTO_ERROR(H5E_FILE, H5E_FCNTL, NULL, "failed to send hints to GPFS") } #endif /* H5_HAVE_GPFS */ /* Build the file struct and initialize it */ - if (NULL==(file=H5MM_calloc(sizeof(H5FD_mpiposix_t)))) + if (NULL == (file=H5MM_calloc(sizeof(H5FD_mpiposix_t)))) HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, NULL, "memory allocation failed") #ifdef REPORT_IO @@ -755,8 +710,9 @@ H5FD_mpiposix_open(const char *name, unsigned flags, hid_t fapl_id, file->fd = fd; file->eof = sb.st_size; - /* for H5_HAVE_WIN32_API support. H5_HAVE_WIN32_API 'stat' does not have st_blksize and st_blksize - is only used for the H5_HAVE_GPFS case */ + /* for H5_HAVE_WIN32_API support. H5_HAVE_WIN32_API 'stat' does not have + * st_blksize and st_blksize is only used for the H5_HAVE_GPFS case. + */ #ifdef H5_HAVE_GPFS file->blksize = sb.st_blksize; #endif @@ -775,26 +731,38 @@ H5FD_mpiposix_open(const char *name, unsigned flags, hid_t fapl_id, /* Set the information for the file's device and inode */ #ifdef H5_HAVE_WIN32_API - filehandle = _get_osfhandle(fd); - results = GetFileInformationByHandle((HANDLE)filehandle, &fileinfo); - file->fileindexhi = fileinfo.nFileIndexHigh; - file->fileindexlo = fileinfo.nFileIndexLow; -#else + file->hFile = (HANDLE)_get_osfhandle(fd); + if(INVALID_HANDLE_VALUE == file->hFile) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to get Windows file handle") + + if(!GetFileInformationByHandle((HANDLE)file->hFile, &fileinfo)) + HGOTO_ERROR(H5E_FILE, H5E_CANTOPENFILE, NULL, "unable to get Windows file information") + + file->nFileIndexHigh = fileinfo.nFileIndexHigh; + file->nFileIndexLow = fileinfo.nFileIndexLow; + file->dwVolumeSerialNumber = fileinfo.dwVolumeSerialNumber; +#else /* H5_HAVE_WIN32_API */ file->device = sb.st_dev; +#ifdef H5_VMS + file->inode[0] = sb.st_ino[0]; + file->inode[1] = sb.st_ino[1]; + file->inode[2] = sb.st_ino[2]; +#else /* H5_VMS */ file->inode = sb.st_ino; -#endif +#endif /* H5_VMS */ +#endif /* H5_HAVE_WIN32_API */ /* Indicate success */ - ret_value=(H5FD_t *)file; + ret_value = (H5FD_t *)file; done: /* Error cleanup */ - if(ret_value==NULL) { + if(NULL == ret_value) { /* Close the file if it was left open */ - if(fd!=(-1)) + if(fd != -1) HDclose(fd); - if (MPI_COMM_NULL != comm_dup) - MPI_Comm_free(&comm_dup); + if (MPI_COMM_NULL != comm_dup) + MPI_Comm_free(&comm_dup); } /* end if */ FUNC_LEAVE_NOAPI(ret_value) @@ -806,30 +774,26 @@ done: * * Purpose: Closes a file. * - * Return: Success: Non-negative - * Failure: Negative + * Return: SUCCEED/FAIL * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * Albert Cheng, 2003-04-24 - * Free the communicator stored. *------------------------------------------------------------------------- */ static herr_t H5FD_mpiposix_close(H5FD_t *_file) { - H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; - herr_t ret_value=SUCCEED; /* Return value */ + H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; + herr_t ret_value = SUCCEED; /* Return value */ FUNC_ENTER_NOAPI_NOINIT - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); /* Close the unix file */ - if (HDclose(file->fd)<0) + if(HDclose(file->fd) < 0) HGOTO_ERROR(H5E_IO, H5E_CANTCLOSEFILE, FAIL, "unable to close file") /* make sure all processes have closed the file before returning. */ @@ -844,20 +808,17 @@ done: /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_cmp + * Function: H5FD_mpiposix_cmp * - * Purpose: Compares two files belonging to this driver using an - * arbitrary (but consistent) ordering. + * Purpose: Compares two files belonging to this driver using an + * arbitrary (but consistent) ordering. * - * Return: Success: A value like strcmp() - * Failure: never fails (arguments were checked by the - * caller). + * Return: Success: A value like strcmp() + * Failure: never fails (arguments were checked by the caller). * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * *------------------------------------------------------------------------- */ static int @@ -865,33 +826,39 @@ H5FD_mpiposix_cmp(const H5FD_t *_f1, const H5FD_t *_f2) { const H5FD_mpiposix_t *f1 = (const H5FD_mpiposix_t*)_f1; const H5FD_mpiposix_t *f2 = (const H5FD_mpiposix_t*)_f2; - int ret_value=0; + int ret_value = 0; FUNC_ENTER_NOAPI_NOINIT_NOERR #ifdef H5_HAVE_WIN32_API - if (f1->fileindexhi < f2->fileindexhi) HGOTO_DONE(-1) - if (f1->fileindexhi > f2->fileindexhi) HGOTO_DONE(1) + if(f1->dwVolumeSerialNumber < f2->dwVolumeSerialNumber) HGOTO_DONE(-1) + if(f1->dwVolumeSerialNumber > f2->dwVolumeSerialNumber) HGOTO_DONE(1) - if (f1->fileindexlo < f2->fileindexlo) HGOTO_DONE(-1) - if (f1->fileindexlo > f2->fileindexlo) HGOTO_DONE(1) + if(f1->nFileIndexHigh < f2->nFileIndexHigh) HGOTO_DONE(-1) + if(f1->nFileIndexHigh > f2->nFileIndexHigh) HGOTO_DONE(1) -#else + if(f1->nFileIndexLow < f2->nFileIndexLow) HGOTO_DONE(-1) + if(f1->nFileIndexLow > f2->nFileIndexLow) HGOTO_DONE(1) +#else /* H5_HAVE_WIN32_API */ #ifdef H5_DEV_T_IS_SCALAR - if (f1->device < f2->device) HGOTO_DONE(-1) - if (f1->device > f2->device) HGOTO_DONE(1) + if(f1->device < f2->device) HGOTO_DONE(-1) + if(f1->device > f2->device) HGOTO_DONE(1) #else /* H5_DEV_T_IS_SCALAR */ /* If dev_t isn't a scalar value on this system, just use memcmp to * determine if the values are the same or not. The actual return value * shouldn't really matter... */ - if(HDmemcmp(&(f1->device),&(f2->device),sizeof(dev_t))<0) HGOTO_DONE(-1) - if(HDmemcmp(&(f1->device),&(f2->device),sizeof(dev_t))>0) HGOTO_DONE(1) + if(HDmemcmp(&(f1->device),&(f2->device),sizeof(dev_t)) < 0) HGOTO_DONE(-1) + if(HDmemcmp(&(f1->device),&(f2->device),sizeof(dev_t)) > 0) HGOTO_DONE(1) #endif /* H5_DEV_T_IS_SCALAR */ - - if (f1->inode < f2->inode) HGOTO_DONE(-1) - if (f1->inode > f2->inode) HGOTO_DONE(1) -#endif +#ifdef H5_VMS + if(HDmemcmp(&(f1->inode), &(f2->inode), 3 * sizeof(ino_t)) < 0) HGOTO_DONE(-1) + if(HDmemcmp(&(f1->inode), &(f2->inode), 3 * sizeof(ino_t)) > 0) HGOTO_DONE(1) +#else /* H5_VMS */ + if(f1->inode < f2->inode) HGOTO_DONE(-1) + if(f1->inode > f2->inode) HGOTO_DONE(1) +#endif /* H5_VMS */ +#endif /* H5_HAVE_WIN32_API */ done: FUNC_LEAVE_NOAPI(ret_value) @@ -899,25 +866,16 @@ done: /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_query + * Function: H5FD_mpiposix_query * - * Purpose: Set the flags that this VFL driver is capable of supporting. + * Purpose: Set the flags that this VFL driver is capable of supporting. * (listed in H5FDpublic.h) * - * Return: Success: non-negative - * Failure: negative + * Return: SUCCEED (can't fail) * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * - * John Mainzer -- 9/21/05 - * Modified code to turn off the - * H5FD_FEAT_ACCUMULATE_METADATA_WRITE flag. - * With the movement of all cache writes to process 0, - * this flag has become problematic in PHDF5. - * *------------------------------------------------------------------------- */ static herr_t @@ -928,10 +886,10 @@ H5FD_mpiposix_query(const H5FD_t UNUSED *_file, unsigned long *flags /* out */) /* Set the VFL feature flags that this driver supports */ if(flags) { *flags=0; - *flags|=H5FD_FEAT_AGGREGATE_METADATA; /* OK to aggregate metadata allocations */ - *flags|=H5FD_FEAT_AGGREGATE_SMALLDATA; /* OK to aggregate "small" raw data allocations */ - *flags|=H5FD_FEAT_HAS_MPI; /* This driver uses MPI */ - *flags|=H5FD_FEAT_ALLOCATE_EARLY; /* Allocate space early instead of late */ + *flags |= H5FD_FEAT_AGGREGATE_METADATA; /* OK to aggregate metadata allocations */ + *flags |= H5FD_FEAT_AGGREGATE_SMALLDATA; /* OK to aggregate "small" raw data allocations */ + *flags |= H5FD_FEAT_HAS_MPI; /* This driver uses MPI */ + *flags |= H5FD_FEAT_ALLOCATE_EARLY; /* Allocate space early instead of late */ } /* end if */ FUNC_LEAVE_NOAPI(SUCCEED) @@ -939,23 +897,18 @@ H5FD_mpiposix_query(const H5FD_t UNUSED *_file, unsigned long *flags /* out */) /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_get_eoa + * Function: H5FD_mpiposix_get_eoa * - * Purpose: Gets the end-of-address marker for the file. The EOA marker - * is the first address past the last byte allocated in the - * format address space. + * Purpose: Gets the end-of-address marker for the file. The EOA marker + * is the first address past the last byte allocated in the + * format address space. * - * Return: Success: The end-of-address marker. - * Failure: HADDR_UNDEF + * Return: Success: The end-of-address marker. + * Failure: HADDR_UNDEF * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * Raymond Lu - * 21 Dec. 2006 - * Added the parameter TYPE. It's only used for MULTI driver. - * *------------------------------------------------------------------------- */ static haddr_t @@ -965,31 +918,25 @@ H5FD_mpiposix_get_eoa(const H5FD_t *_file, H5FD_mem_t UNUSED type) FUNC_ENTER_NOAPI_NOINIT_NOERR - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); FUNC_LEAVE_NOAPI(file->eoa) } /* end H5FD_mpiposix_get_eoa() */ /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_set_eoa + * Function: H5FD_mpiposix_set_eoa * - * Purpose: Set the end-of-address marker for the file. This function is - * called shortly after an existing HDF5 file is opened in order - * to tell the driver where the end of the HDF5 data is located. + * Purpose: Set the end-of-address marker for the file. This function is + * called shortly after an existing HDF5 file is opened in order + * to tell the driver where the end of the HDF5 data is located. * - * Return: Success: non-negative - * Failure: negative + * Return: SUCCEED (can't fail) * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * Raymond Lu - * 21 Dec. 2006 - * Added the parameter TYPE. It's only used for MULTI driver. - * *------------------------------------------------------------------------- */ static herr_t @@ -999,8 +946,8 @@ H5FD_mpiposix_set_eoa(H5FD_t *_file, H5FD_mem_t UNUSED type, haddr_t addr) FUNC_ENTER_NOAPI_NOINIT_NOERR - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); file->eoa = addr; @@ -1009,25 +956,23 @@ H5FD_mpiposix_set_eoa(H5FD_t *_file, H5FD_mem_t UNUSED type, haddr_t addr) /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_get_eof + * Function: H5FD_mpiposix_get_eof * - * Purpose: Gets the end-of-file marker for the file. The EOF marker - * is the real size of the file. + * Purpose: Gets the end-of-file marker for the file. The EOF marker + * is the real size of the file. * - * The MPIPOSIX driver doesn't bother keeping this field updated - * since that's a relatively expensive operation. Fortunately - * the library only needs the EOF just after the file is opened - * in order to determine whether the file is empty, truncated, - * or okay. + * The MPIPOSIX driver doesn't bother keeping this field updated + * since that's a relatively expensive operation. Fortunately + * the library only needs the EOF just after the file is opened + * in order to determine whether the file is empty, truncated, + * or okay. * - * Return: Success: The end-of-address marker. - * Failure: HADDR_UNDEF + * Return: Success: The end-of-address marker. + * Failure: HADDR_UNDEF * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * *------------------------------------------------------------------------- */ static haddr_t @@ -1037,10 +982,10 @@ H5FD_mpiposix_get_eof(const H5FD_t *_file) FUNC_ENTER_NOAPI_NOINIT_NOERR - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); - FUNC_LEAVE_NOAPI(MAX(file->eof,file->eoa)) + FUNC_LEAVE_NOAPI(MAX(file->eof, file->eoa)) } /* end H5FD_mpiposix_get_eof() */ @@ -1049,13 +994,11 @@ H5FD_mpiposix_get_eof(const H5FD_t *_file) * * Purpose: Returns the file handle of MPI-POSIX file driver. * - * Returns: Non-negative if succeed or negative if fails. + * Returns: SUCCEED/FAIL * * Programmer: Raymond Lu * Sept. 16, 2002 * - * Modifications: - * *------------------------------------------------------------------------- */ static herr_t @@ -1077,43 +1020,41 @@ done: /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_read + * Function: H5FD_mpiposix_read * - * Purpose: Reads SIZE bytes of data from FILE beginning at address ADDR - * into buffer BUF according to data transfer properties in - * DXPL_ID using potentially complex file and buffer types to - * effect the transfer. + * Purpose: Reads SIZE bytes of data from FILE beginning at address ADDR + * into buffer BUF according to data transfer properties in + * DXPL_ID using potentially complex file and buffer types to + * effect the transfer. * - * Reading past the end of the file returns zeros instead of - * failing. + * Reading past the end of the file returns zeros instead of + * failing. * - * Return: Success: Non-negative. Result is stored in caller-supplied - * buffer BUF. - * Failure: Negative, Contents of buffer BUF are undefined. + * Return: Success: Non-negative. Result is stored in caller-supplied + * buffer BUF. + * Failure: Negative, Contents of buffer BUF are undefined. * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * *------------------------------------------------------------------------- */ static herr_t -H5FD_mpiposix_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t UNUSED dxpl_id, haddr_t addr, size_t size, - void *buf/*out*/) +H5FD_mpiposix_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t UNUSED dxpl_id, + haddr_t addr, size_t size, void *buf/*out*/) { - H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; - ssize_t nbytes; /* Number of bytes read each I/O call */ - herr_t ret_value=SUCCEED; + H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; + ssize_t nbytes; /* Number of bytes read each I/O call */ + herr_t ret_value = SUCCEED; FUNC_ENTER_NOAPI_NOINIT - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); - assert(buf); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); + HDassert(buf); /* Check for overflow conditions */ - if (HADDR_UNDEF==addr) + if (HADDR_UNDEF == addr) HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "addr undefined") if (REGION_OVERFLOW(addr, size)) HGOTO_ERROR(H5E_ARGS, H5E_OVERFLOW, FAIL, "addr overflow") @@ -1130,31 +1071,52 @@ H5FD_mpiposix_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t UNUSED dxpl_id, #endif /* Seek to the correct location */ - if ((addr!=file->pos || OP_READ!=file->op) && - file_seek(file->fd, (file_offset_t)addr, SEEK_SET)<0) - HGOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "unable to seek to proper position") + if(addr != file->pos || OP_READ != file->op) { + if(HDlseek(file->fd, (HDoff_t)addr, SEEK_SET) < 0) + HGOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "unable to seek to proper position") + } /* end if */ - /* - * Read data, being careful of interrupted system calls, partial results, + /* Read data, being careful of interrupted system calls, partial results, * and the end of the file. */ - while (size>0) { + while(size > 0) { + + h5_mpiposix_io_t bytes_in = 0; /* # of bytes to read */ + h5_mpiposix_io_ret_t bytes_read = -1; /* # of bytes actually read */ + + /* Trying to read more bytes than the return type can handle is + * undefined behavior in POSIX. + */ + if(size > H5_MPIPOSIX_MAX_IO_BYTES_g) + bytes_in = H5_MPIPOSIX_MAX_IO_BYTES_g; + else + bytes_in = (h5_mpiposix_io_t)size; + do { - nbytes = HDread(file->fd, buf, size); - } while (-1==nbytes && EINTR==errno); - if (-1==nbytes) - HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed") - if (0==nbytes) { + bytes_read = HDread(file->fd, buf, bytes_in); + } while(-1 == bytes_read && EINTR == errno); + + if(-1 == bytes_read) { /* error */ + int myerrno = errno; + time_t mytime = HDtime(NULL); + HDoff_t myoffset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR); + + HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed: time = %s, file descriptor = %d, errno = %d, error message = '%s', buf = %p, size = %lu, offset = %llu", HDctime(&mytime), file->fd, myerrno, HDstrerror(myerrno), buf, (unsigned long)size, (unsigned long long)myoffset); + } /* end if */ + + if(0 == bytes_read) { /* end of file but not end of format address space */ HDmemset(buf, 0, size); break; } /* end if */ - assert(nbytes>=0); - assert((size_t)nbytes<=size); - size -= nbytes; - addr += (haddr_t)nbytes; - buf = (char*)buf + nbytes; - } + + HDassert(bytes_read >= 0); + HDassert((size_t)bytes_read <= size); + + size -= (size_t)bytes_read; + addr += (haddr_t)bytes_read; + buf = (char *)buf + bytes_read; + } /* end while */ /* Update current position */ file->pos = addr; @@ -1162,7 +1124,7 @@ H5FD_mpiposix_read(H5FD_t *_file, H5FD_mem_t UNUSED type, hid_t UNUSED dxpl_id, done: /* Check for error */ - if(ret_value<0) { + if(ret_value < 0) { /* Reset last file I/O information */ file->pos = HADDR_UNDEF; file->op = OP_UNKNOWN; @@ -1173,55 +1135,46 @@ done: /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_write + * Function: H5FD_mpiposix_write * - * Purpose: Writes SIZE bytes of data to FILE beginning at address ADDR - * from buffer BUF according to data transfer properties in - * DXPL_ID using potentially complex file and buffer types to - * effect the transfer. + * Purpose: Writes SIZE bytes of data to FILE beginning at address ADDR + * from buffer BUF according to data transfer properties in + * DXPL_ID using potentially complex file and buffer types to + * effect the transfer. * - * Return: Success: non-negative - * Failure: negative + * Return: SUCCEED/FAIL * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * - * Quincey Koziol - 2002/07/18 - * Added "block_before_meta_write" dataset transfer flag, which - * is set during writes from a metadata cache flush and indicates - * that all the processes must sync up before (one of them) - * writing metadata. - * *------------------------------------------------------------------------- */ static herr_t H5FD_mpiposix_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr, size_t size, const void *buf) { - H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; + H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; #if 0 /* JRM */ int mpi_code; /* MPI return code */ #endif /* JRM */ - ssize_t nbytes; /* Number of bytes written each I/O call */ - H5P_genplist_t *plist; /* Property list pointer */ - herr_t ret_value=SUCCEED; /* Return value */ + ssize_t nbytes; /* Number of bytes written each I/O call */ + H5P_genplist_t *plist; /* Property list pointer */ + herr_t ret_value = SUCCEED; /* Return value */ FUNC_ENTER_NOAPI_NOINIT - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); - assert(H5I_GENPROP_LST==H5I_get_type(dxpl_id)); - assert(TRUE==H5P_isa_class(dxpl_id,H5P_DATASET_XFER)); - assert(buf); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); + HDassert(H5I_GENPROP_LST == H5I_get_type(dxpl_id)); + HDassert(TRUE == H5P_isa_class(dxpl_id,H5P_DATASET_XFER)); + HDassert(buf); /* Check for overflow conditions */ - if (HADDR_UNDEF==addr) + if (HADDR_UNDEF == addr) HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "addr undefined") if (REGION_OVERFLOW(addr, size)) HGOTO_ERROR(H5E_ARGS, H5E_OVERFLOW, FAIL, "addr overflow") - if (addr+size>file->eoa) + if ((addr + size) > file->eoa) HGOTO_ERROR(H5E_ARGS, H5E_OVERFLOW, FAIL, "addr overflow") /* Obtain the data transfer properties */ @@ -1237,26 +1190,26 @@ H5FD_mpiposix_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr, * metadata. * -- JRM 9/1/05 */ - if(type!=H5FD_MEM_DRAW) { - unsigned block_before_meta_write=0; /* Whether to block before a metadata write */ - - /* Check if we need to syncronize all processes before attempting metadata write - * (Prevents race condition where the process writing the metadata goes ahead - * and writes the metadata to the file before all the processes have - * read the data, "transmitting" data from the "future" to the reading - * process. -QAK ) + if(type != H5FD_MEM_DRAW) { + unsigned block_before_meta_write = 0; /* Whether to block before a metadata write */ + + /* Check if we need to synchronize all processes before attempting + * metadata write (Prevents race condition where the process writing + * the metadata goes ahead and writes the metadata to the file before + * all the processes have read the data, "transmitting" data from the + * "future" to the reading process. -QAK ) * * The only time we don't want to block before a metadata write is when * we are flushing out a bunch of metadata. Then, we block before the * first write and don't block for further writes in the sequence. */ - if(H5P_exist_plist(plist,H5AC_BLOCK_BEFORE_META_WRITE_NAME)>0) - if(H5P_get(plist,H5AC_BLOCK_BEFORE_META_WRITE_NAME,&block_before_meta_write)<0) + if(H5P_exist_plist(plist,H5AC_BLOCK_BEFORE_META_WRITE_NAME) > 0) + if(H5P_get(plist,H5AC_BLOCK_BEFORE_META_WRITE_NAME,&block_before_meta_write) < 0) HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, FAIL, "can't get H5AC property") #if 0 /* JRM */ if(block_before_meta_write) - if (MPI_SUCCESS!= (mpi_code=MPI_Barrier(file->comm))) + if (MPI_SUCCESS != (mpi_code = MPI_Barrier(file->comm))) HMPI_GOTO_ERROR(FAIL, "MPI_Barrier failed", mpi_code) #endif /* JRM */ @@ -1273,11 +1226,11 @@ H5FD_mpiposix_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr, MPI_Comm_rank(MPI_COMM_WORLD, &commrank); HDfprintf(stderr, "write: rank=%d file=0x%08lx type=%d, addr=%a size=%Zu %s\n", commrank, (unsigned long)file, (int)type, addr, size, - 0==file->naccess?"(FIRST ACCESS)":""); + 0 == file->naccess ? "(FIRST ACCESS)" : ""); } #endif - if (0==file->naccess++) { + if (0 == file->naccess++) { /* First write access to this file */ #ifdef H5_HAVE_GPFS if (file->use_gpfs) { @@ -1302,50 +1255,76 @@ H5FD_mpiposix_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr, } /* Seek to the correct location */ - if ((addr!=file->pos || OP_WRITE!=file->op) && - file_seek(file->fd, (file_offset_t)addr, SEEK_SET)<0) - HGOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "unable to seek to proper position") + if(addr != file->pos || OP_WRITE != file->op) { + if(HDlseek(file->fd, (HDoff_t)addr, SEEK_SET) < 0) + HSYS_GOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "unable to seek to proper position") + } /* end if */ - /* - * Write the data, being careful of interrupted system calls and partial - * results + /* Write data, being careful of interrupted system calls, partial results, + * and the end of the file. */ - while (size>0) { + while(size > 0) { + + h5_mpiposix_io_t bytes_in = 0; /* # of bytes to write */ + h5_mpiposix_io_ret_t bytes_wrote = -1; /* # of bytes actually written */ + + /* Trying to write more bytes than the return type can handle is + * undefined behavior in POSIX. + */ + if(size > H5_MPIPOSIX_MAX_IO_BYTES_g) + bytes_in = H5_MPIPOSIX_MAX_IO_BYTES_g; + else + bytes_in = (h5_mpiposix_io_t)size; + do { - nbytes = HDwrite(file->fd, buf, size); - } while (-1==nbytes && EINTR==errno); - if (-1==nbytes) - HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "file write failed") - assert(nbytes>0); - assert((size_t)nbytes<=size); - size -= nbytes; - addr += (haddr_t)nbytes; - buf = (const char*)buf + nbytes; + bytes_wrote = HDwrite(file->fd, buf, bytes_in); + } while(-1 == bytes_wrote && EINTR == errno); + + if(-1 == bytes_wrote) { /* error */ + int myerrno = errno; + time_t mytime = HDtime(NULL); + HDoff_t myoffset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR); + + HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file write failed: time = %s, file descriptor = %d, errno = %d, error message = '%s', buf = %p, size = %lu, offset = %llu", HDctime(&mytime), file->fd, myerrno, HDstrerror(myerrno), buf, (unsigned long)size, (unsigned long long)myoffset); + } /* end if */ + + if(0 == bytes_wrote) { + /* end of file but not end of format address space */ + HDmemset(buf, 0, size); + break; + } /* end if */ + + HDassert(bytes_wrote >= 0); + HDassert((size_t)bytes_wrote <= size); + + size -= (size_t)bytes_wrote; + addr += (haddr_t)bytes_wrote; + buf = (char *)buf + bytes_wrote; } /* end while */ - /* Update current last file I/O information */ + /* Update current position */ file->pos = addr; file->op = OP_WRITE; done: /* Check for error */ - if(ret_value<0) { + if(ret_value < 0) { /* Reset last file I/O information */ file->pos = HADDR_UNDEF; file->op = OP_UNKNOWN; } /* end if */ #if 0 /* JRM */ - /* Since metadata writes are now done by process 0 only, this broadcast - * is no longer needed. I leave it in and commented out to remind us - * that we need to re-work this function to reflect this reallity. - * - * -- JRM 9/1/05 - */ + /* Since metadata writes are now done by process 0 only, this broadcast + * is no longer needed. I leave it in and commented out to remind us + * that we need to re-work this function to reflect this reallity. + * + * -- JRM 9/1/05 + */ /* Guard against getting into metadata broadcast in failure cases */ else { /* when only one process writes, need to broadcast the ret_value to other processes */ - if (type!=H5FD_MEM_DRAW) { + if (type != H5FD_MEM_DRAW) { if (MPI_SUCCESS != (mpi_code= MPI_Bcast(&ret_value, sizeof(ret_value), MPI_BYTE, H5_PAR_META_WRITE, file->comm))) HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", mpi_code) } /* end if */ @@ -1359,11 +1338,10 @@ done: /*------------------------------------------------------------------------- * Function: H5FD_mpiposix_truncate * - * Purpose: Makes sure that the true file size is the same (or larger) - * than the end-of-address. + * Purpose: Makes sure that the true file size is the same (or larger) + * than the end-of-address. * - * Return: Success: Non-negative - * Failure: Negative + * Return: SUCCEED/FAIL * * Programmer: Quincey Koziol * Thursday, July 11, 2002 @@ -1374,12 +1352,8 @@ static herr_t H5FD_mpiposix_truncate(H5FD_t *_file, hid_t UNUSED dxpl_id, hbool_t UNUSED closing) { H5FD_mpiposix_t *file = (H5FD_mpiposix_t*)_file; -#ifdef H5_HAVE_WIN32_API - HFILE filehandle; /* Windows file handle */ - LARGE_INTEGER li; /* 64-bit integer for SetFilePointer() call */ -#endif /* H5_HAVE_WIN32_API */ - int mpi_code; /* MPI return code */ - herr_t ret_value = SUCCEED; + herr_t ret_value = SUCCEED; /* Return value */ + int mpi_code; /* MPI return code */ FUNC_ENTER_NOAPI_NOINIT @@ -1390,19 +1364,42 @@ H5FD_mpiposix_truncate(H5FD_t *_file, hid_t UNUSED dxpl_id, hbool_t UNUSED closi if(file->eoa > file->last_eoa) { /* Use the round-robin process to truncate (extend) the file */ if(file->mpi_rank == H5_PAR_META_WRITE) { + #ifdef H5_HAVE_WIN32_API - /* Map the posix file handle to a Windows file handle */ - filehandle = _get_osfhandle(file->fd); - - /* Translate 64-bit integers into form Windows wants */ - /* [This algorithm is from the Windows documentation for SetFilePointer()] */ - li.QuadPart = file->eoa; - SetFilePointer((HANDLE)filehandle, li.LowPart, &li.HighPart, FILE_BEGIN); - if(SetEndOfFile((HANDLE)filehandle) == 0) + LARGE_INTEGER li; /* 64-bit (union) integer for SetFilePointer() call */ + DWORD dwPtrLow; /* Low-order pointer bits from SetFilePointer() + * Only used as an error code here. + */ + DWORD dwError; /* DWORD error code from GetLastError() */ + BOOL bError; /* Boolean error flag */ + + /* Windows uses this odd QuadPart union for 32/64-bit portability */ + li.QuadPart = (__int64)file->eoa; + + /* Extend the file to make sure it's large enough. + * + * Since INVALID_SET_FILE_POINTER can technically be a valid return value + * from SetFilePointer(), we also need to check GetLastError(). + */ + dwPtrLow = SetFilePointer(file->hFile, li.LowPart, &li.HighPart, FILE_BEGIN); + if(INVALID_SET_FILE_POINTER == dwPtrLow) { + dwError = GetLastError(); + if(dwError != NO_ERROR ) + HGOTO_ERROR(H5E_FILE, H5E_FILEOPEN, FAIL, "unable to set file pointer") + } + + bError = SetEndOfFile(file->hFile); + if(0 == bError) HGOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "unable to extend file properly") #else /* H5_HAVE_WIN32_API */ - if(-1==file_truncate(file->fd, (file_offset_t)file->eoa)) - HGOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "unable to extend file properly") +#ifdef H5_VMS + /* Reset seek offset to the beginning of the file, so that the file isn't + * re-extended later. This may happen on Open VMS. */ + if(-1 == HDlseek(file->fd, (HDoff_t)0, SEEK_SET)) + HSYS_GOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "unable to seek to proper position") +#endif /* H5_VMS */ + if(-1 == HDftruncate(file->fd, (HDoff_t)file->eoa)) + HSYS_GOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "unable to extend file properly") #endif /* H5_HAVE_WIN32_API */ } /* end if */ @@ -1430,18 +1427,15 @@ done: /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_mpi_rank + * Function: H5FD_mpiposix_mpi_rank * - * Purpose: Returns the MPI rank for a process + * Purpose: Returns the MPI rank for a process * - * Return: Success: non-negative - * Failure: negative + * Return: MPI rank. Cannot report failure. * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * *------------------------------------------------------------------------- */ static int @@ -1451,26 +1445,23 @@ H5FD_mpiposix_mpi_rank(const H5FD_t *_file) FUNC_ENTER_NOAPI_NOINIT_NOERR - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); FUNC_LEAVE_NOAPI(file->mpi_rank) } /* end H5FD_mpiposix_mpi_rank() */ /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_mpi_size + * Function: H5FD_mpiposix_mpi_size * - * Purpose: Returns the number of MPI processes + * Purpose: Returns the number of MPI processes * - * Return: Success: non-negative - * Failure: negative + * Return: The number of MPI processes. Cannot report failure. * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * *------------------------------------------------------------------------- */ static int @@ -1480,27 +1471,23 @@ H5FD_mpiposix_mpi_size(const H5FD_t *_file) FUNC_ENTER_NOAPI_NOINIT_NOERR - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); FUNC_LEAVE_NOAPI(file->mpi_size) } /* end H5FD_mpiposix_mpi_size() */ /*------------------------------------------------------------------------- - * Function: H5FD_mpiposix_communicator - * - * Purpose: Returns the MPI communicator for the file. + * Function: H5FD_mpiposix_communicator * - * Return: Success: The communicator + * Purpose: Returns the MPI communicator for the file. * - * Failure: NULL + * Return: The MPI communicator. Cannot report failure. * * Programmer: Quincey Koziol * Thursday, July 11, 2002 * - * Modifications: - * *------------------------------------------------------------------------- */ static MPI_Comm @@ -1510,11 +1497,10 @@ H5FD_mpiposix_communicator(const H5FD_t *_file) FUNC_ENTER_NOAPI_NOINIT_NOERR - assert(file); - assert(H5FD_MPIPOSIX==file->pub.driver_id); + HDassert(file); + HDassert(H5FD_MPIPOSIX == file->pub.driver_id); FUNC_LEAVE_NOAPI(file->comm) } /* end H5FD_mpi_posix_communicator() */ #endif /*H5_HAVE_PARALLEL*/ - diff --git a/src/H5FDmpiposix.h b/src/H5FDmpiposix.h index af13fad..e3764e1 100644 --- a/src/H5FDmpiposix.h +++ b/src/H5FDmpiposix.h @@ -14,7 +14,7 @@ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ /* - * Programmer: Quincey Koziol + * Programmer: Quincey Koziol * Thursday, July 11, 2002 * * Purpose: The public header file for the mpiposix driver. @@ -24,14 +24,14 @@ #define __H5FDmpiposix_H #ifdef H5_HAVE_PARALLEL -# define H5FD_MPIPOSIX (H5FD_mpiposix_init()) +# define H5FD_MPIPOSIX (H5FD_mpiposix_init()) #else -# define H5FD_MPIPOSIX (-1) +# define H5FD_MPIPOSIX (-1) #endif /* Macros */ -#define IS_H5FD_MPIPOSIX(f) /* (H5F_t *f) */ \ +#define IS_H5FD_MPIPOSIX(f) /* (H5F_t *f) */ \ (H5FD_MPIPOSIX==H5F_DRIVER_ID(f)) #ifdef H5_HAVE_PARALLEL @@ -52,4 +52,3 @@ H5_DLL herr_t H5Pget_fapl_mpiposix(hid_t fapl_id, MPI_Comm *comm/*out*/, hbool_t #endif /*H5_HAVE_PARALLEL*/ #endif /* __H5FDmpiposix_H */ - -- cgit v0.12