diff --git a/src/api-copy.c b/src/api-copy.c index 07588baf..26005f84 100644 --- a/src/api-copy.c +++ b/src/api-copy.c @@ -479,7 +479,7 @@ __cold static int copy_with_compacting(MDBX_env *env, MDBX_txn *txn, mdbx_fileha if (meta->geometry.now != meta->geometry.first_unallocated) { const size_t whole_size = pgno2bytes(env, meta->geometry.now); if (!dest_is_pipe) - return osal_fallocate(fd, whole_size); + return osal_fsetsize(fd, whole_size); const size_t used_size = pgno2bytes(env, meta->geometry.first_unallocated); memset(data_buffer, 0, (size_t)MDBX_ENVCOPY_WRITEBUF); @@ -648,7 +648,7 @@ retry_snap_meta: /* Extend file if required */ if (likely(rc == MDBX_SUCCESS) && whole_size != used_size) { if (!dest_is_pipe) - rc = osal_fallocate(fd, whole_size); + rc = osal_fsetsize(fd, whole_size); else { memset(data_buffer, 0, (size_t)MDBX_ENVCOPY_WRITEBUF); for (size_t offset = used_size; rc == MDBX_SUCCESS && offset < whole_size;) { diff --git a/src/dxb.c b/src/dxb.c index 198b1805..b431547c 100644 --- a/src/dxb.c +++ b/src/dxb.c @@ -532,7 +532,7 @@ __cold int dxb_setup(MDBX_env *env, const int lck_rc, const mdbx_mode_t mode_bit if (unlikely(err != MDBX_SUCCESS)) return err; - err = osal_fallocate(env->lazy_fd, env->dxb_mmap.filesize = env->dxb_mmap.current = env->geo_in_bytes.now); + err = osal_fsetsize(env->lazy_fd, env->dxb_mmap.filesize = env->dxb_mmap.current = env->geo_in_bytes.now); if (unlikely(err != MDBX_SUCCESS)) return err; diff --git a/src/lck-windows.c b/src/lck-windows.c index cc1e6328..0eeb14cc 100644 --- a/src/lck-windows.c +++ b/src/lck-windows.c @@ -546,7 +546,7 @@ int lck_destroy(MDBX_env *env, MDBX_env *inprocess_neighbor, const uint32_t curr if (synced && !inprocess_neighbor && env->lck_mmap.fd != INVALID_HANDLE_VALUE && lck_upgrade(env, true) == MDBX_SUCCESS) /* this will fail if LCK is used/mmapped by other process(es) */ - osal_ftruncate(env->lck_mmap.fd, 0); + osal_fsetsize(env->lck_mmap.fd, 0); } lck_unlock(env); return MDBX_SUCCESS; diff --git a/src/options.h b/src/options.h index 7ea12ac8..732cdeab 100644 --- a/src/options.h +++ b/src/options.h @@ -350,6 +350,19 @@ #error MDBX_USE_COPYFILERANGE must be defined as 0 or 1 #endif /* MDBX_USE_COPYFILERANGE */ +/** Advanced: Using posix_fallocate() or fcntl(F_PREALLOCATE) (autodetection by default). */ +#ifndef MDBX_USE_FALLOCATE +#if defined(__APPLE__) +#define MDBX_USE_FALLOCATE 0 /* Too slow and unclean, but not required to prevent SIGBUS */ +#elif (defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 200112L) || (__GLIBC_PREREQ(2, 10) && defined(_GNU_SOURCE)) +#define MDBX_USE_FALLOCATE 1 +#else +#define MDBX_USE_FALLOCATE 0 +#endif +#elif !(MDBX_USE_FALLOCATE == 0 || MDBX_USE_FALLOCATE == 1) +#error MDBX_USE_FALLOCATE must be defined as 0 or 1 +#endif /* MDBX_USE_FALLOCATE */ + //------------------------------------------------------------------------------ #ifndef MDBX_CPU_WRITEBACK_INCOHERENT diff --git a/src/osal.c b/src/osal.c index e27a35bb..6f1fdff3 100644 --- a/src/osal.c +++ b/src/osal.c @@ -1592,8 +1592,7 @@ int osal_is_pipe(mdbx_filehandle_t fd) { #endif } -/* truncate file: just set the length of a file */ -int osal_ftruncate(mdbx_filehandle_t fd, uint64_t length) { +int osal_fsetsize(mdbx_filehandle_t fd, const uint64_t length) { #if defined(_WIN32) || defined(_WIN64) if (imports.SetFileInformationByHandle) { FILE_END_OF_FILE_INFO EndOfFileInfo; @@ -1608,31 +1607,48 @@ int osal_ftruncate(mdbx_filehandle_t fd, uint64_t length) { } #else STATIC_ASSERT_MSG(sizeof(off_t) >= sizeof(size_t), "libmdbx requires 64-bit file I/O on 64-bit systems"); - return ftruncate(fd, length) == 0 ? MDBX_SUCCESS : errno; -#endif -} -/* extend file: set the length of a file AND ensure the space has been allocated */ -int osal_fallocate(mdbx_filehandle_t fd, uint64_t length) { - assert(length > 0); - int err = MDBX_RESULT_TRUE; -#if (defined(__linux__) || defined(__gnu_linux__)) && \ - ((defined(_GNU_SOURCE) && __GLIBC_PREREQ(2, 10)) || (defined(__ANDROID_API__) && __ANDROID_API__ >= 21)) - err = fallocate(fd, 0, 0, length) ? ignore_enosys_and_eremote(errno) : MDBX_SUCCESS; -#elif defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 200112L && !defined(__APPLE__) - err = posix_fallocate(fd, 0, length) ? ignore_enosys_and_eremote(errno) : MDBX_SUCCESS; -#elif defined(__APPLE__) - fstore_t store = {F_ALLOCATEALL, F_PEOFPOSMODE, 0, length, 0}; - if (fcntl(fd, F_PREALLOCATE, &store)) - err = ignore_enosys_and_eremote(errno); -#endif /* Apple */ -#if !defined(_WIN32) && !defined(_WIN64) - /* Workaround for testing: ignore ENOSPC for TMPFS/RAMFS. - * This is insignificant for production, but it helps in some tests using /dev/shm inside docker/containers. */ - if (err == ENOSPC && osal_check_fs_incore(fd) == MDBX_RESULT_TRUE) - err = MDBX_RESULT_TRUE; +#if MDBX_USE_FALLOCATE + struct stat info; + if (unlikely(fstat(fd, &info))) + return errno; + + const uint64_t allocated = UINT64_C(512) * info.st_blocks; + if (length > allocated) { +#if defined(__APPLE__) + fstore_t store = { + .fst_flags = F_ALLOCATECONTIG, .fst_posmode = F_PEOFPOSMODE, .fst_offset = 0, .fst_length = length}; + int err = MDBX_SUCCESS; + if (fcntl(fd, F_PREALLOCATE, &store)) { + /* TODO: implement step-by-step allocation in chunks of 16384, 8192, 4094, 2048, 1024 Kb */ + store.fst_flags = F_ALLOCATEALL; + if (fcntl(fd, F_PREALLOCATE, &store)) + err = errno; + } +#elif defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 200112L + const int err = posix_fallocate(fd, 0, length); + if (!err && length > (uint64_t)info.st_size) + info.st_size = length /* posix_fallocate() extends the file */; +#else + const int err = fallocate(fd, 0, 0, length) ? errno : MDBX_SUCCESS; + if (!err && length > (uint64_t)info.st_size) + info.st_size = length /* fallocate() extends the file */; +#endif + if (unlikely(err) && ignore_enosys_and_eremote(err) != MDBX_RESULT_TRUE) { + /* Workaround for testing: ignore ENOSPC for TMPFS/RAMFS. + * This is insignificant for production, but it helps in some tests using /dev/shm inside docker/containers. */ + if (err != ENOSPC || osal_check_fs_incore(fd) != MDBX_RESULT_TRUE) + return err; + } + } + + if (length == (uint64_t)info.st_size) + return MDBX_SUCCESS; +#endif + + return unlikely(ftruncate(fd, length)) ? errno : MDBX_SUCCESS; + #endif /* !Windows */ - return (err == MDBX_RESULT_TRUE) ? osal_ftruncate(fd, length) : err; } int osal_fseek(mdbx_filehandle_t fd, uint64_t pos) { @@ -2085,8 +2101,8 @@ int osal_mmap(const int flags, osal_mmap_t *map, size_t size, const size_t limit return err; if ((flags & MDBX_RDONLY) == 0 && (options & MMAP_OPTION_SETLENGTH) != 0) { - err = osal_fallocate(map->fd, size); - VERBOSE("ftruncate %zu, err %d", size, err); + err = osal_fsetsize(map->fd, size); + VERBOSE("osal_fsetsize %zu, err %d", size, err); if (err != MDBX_SUCCESS) return err; map->filesize = size; @@ -2331,7 +2347,7 @@ retry_file_and_section: } if ((flags & MDBX_RDONLY) == 0 && map->filesize != size) { - err = osal_fallocate(map->fd, size); + err = osal_fsetsize(map->fd, size); if (err == MDBX_SUCCESS) map->filesize = size; /* ignore error, because Windows unable shrink file @@ -2411,11 +2427,11 @@ retry_mapview:; } else { if (map->filesize != size) { if (size > map->filesize) { - rc = osal_fallocate(map->fd, size); - VERBOSE("f%s-%s %zu, err %d", "allocate", "extend", size, rc); + rc = osal_fsetsize(map->fd, size); + VERBOSE("osal_fsetsize-%s %zu, err %d", "extend", size, rc); } else if (flags & txn_shrink_allowed) { - rc = osal_ftruncate(map->fd, size); - VERBOSE("f%s-%s %zu, err %d", "truncate", "shrink", size, rc); + rc = osal_fsetsize(map->fd, size); + VERBOSE("osal_fsetsize-%s %zu, err %d", "shrink", size, rc); } if (unlikely(rc != MDBX_SUCCESS)) return rc; diff --git a/src/osal.h b/src/osal.h index ea734856..0d91d3d4 100644 --- a/src/osal.h +++ b/src/osal.h @@ -433,8 +433,7 @@ enum osal_syncmode_bits { }; MDBX_INTERNAL int osal_fsync(mdbx_filehandle_t fd, const enum osal_syncmode_bits mode_bits); -MDBX_INTERNAL int osal_ftruncate(mdbx_filehandle_t fd, uint64_t length); -MDBX_INTERNAL int osal_fallocate(mdbx_filehandle_t fd, uint64_t length); +MDBX_INTERNAL int osal_fsetsize(mdbx_filehandle_t fd, const uint64_t length); MDBX_INTERNAL int osal_fseek(mdbx_filehandle_t fd, uint64_t pos); MDBX_INTERNAL int osal_filesize(mdbx_filehandle_t fd, uint64_t *length);