16 #ifdef FIO_HAVE_LINUX_FALLOCATE
17 #include <linux/falloc.h>
22 static inline void clear_error(struct thread_data *td)
29 * Leaves f->fd open on success, caller must close
31 static int extend_file(struct thread_data *td, struct fio_file *f)
33 int r, new_layout = 0, unlink_file = 0, flags;
34 unsigned long long left;
39 log_err("fio: refusing extend of file due to read-only\n");
44 * check if we need to lay the file out complete again. fio
45 * does that for operations involving reads, or for writes
46 * where overwrite is set
48 if (td_read(td) || (td_write(td) && td->o.overwrite) ||
49 (td_write(td) && td->io_ops->flags & FIO_NOEXTEND))
51 if (td_write(td) && !td->o.overwrite)
54 if (unlink_file || new_layout) {
55 dprint(FD_FILE, "layout unlink %s\n", f->file_name);
56 if ((unlink(f->file_name) < 0) && (errno != ENOENT)) {
57 td_verror(td, errno, "unlink");
62 flags = O_WRONLY | O_CREAT;
66 dprint(FD_FILE, "open file %s, flags %x\n", f->file_name, flags);
67 f->fd = open(f->file_name, flags, 0644);
69 td_verror(td, errno, "open");
73 #ifdef FIO_HAVE_FALLOCATE
74 if (!td->o.fill_device) {
75 switch (td->o.fallocate_mode) {
76 case FIO_FALLOCATE_NONE:
78 case FIO_FALLOCATE_POSIX:
79 dprint(FD_FILE, "posix_fallocate file %s size %llu\n",
80 f->file_name, f->real_file_size);
82 r = posix_fallocate(f->fd, 0, f->real_file_size);
84 log_err("fio: posix_fallocate fails: %s\n",
88 #ifdef FIO_HAVE_LINUX_FALLOCATE
89 case FIO_FALLOCATE_KEEP_SIZE:
91 "fallocate(FALLOC_FL_KEEP_SIZE) "
92 "file %s size %llu\n",
93 f->file_name, f->real_file_size);
95 r = fallocate(f->fd, FALLOC_FL_KEEP_SIZE, 0,
98 td_verror(td, errno, "fallocate");
101 #endif /* FIO_HAVE_LINUX_FALLOCATE */
103 log_err("fio: unknown fallocate mode: %d\n",
104 td->o.fallocate_mode);
108 #endif /* FIO_HAVE_FALLOCATE */
114 * The size will be -1ULL when fill_device is used, so don't truncate
115 * or fallocate this file, just write it
117 if (!td->o.fill_device) {
118 dprint(FD_FILE, "truncate file %s, size %llu\n", f->file_name,
120 if (ftruncate(f->fd, f->real_file_size) == -1) {
121 td_verror(td, errno, "ftruncate");
126 b = malloc(td->o.max_bs[DDIR_WRITE]);
127 memset(b, 0, td->o.max_bs[DDIR_WRITE]);
129 left = f->real_file_size;
130 while (left && !td->terminate) {
131 bs = td->o.max_bs[DDIR_WRITE];
135 r = write(f->fd, b, bs);
145 if (td->o.fill_device)
147 log_info("fio: ENOSPC on laying out "
151 td_verror(td, errno, "write");
153 td_verror(td, EIO, "write");
160 dprint(FD_FILE, "terminate unlink %s\n", f->file_name);
161 unlink(f->file_name);
162 } else if (td->o.create_fsync) {
163 if (fsync(f->fd) < 0) {
164 td_verror(td, errno, "fsync");
168 if (td->o.fill_device && !td_write(td)) {
169 fio_file_clear_size_known(f);
170 if (td_io_get_file_size(td, f))
172 if (f->io_size > f->real_file_size)
173 f->io_size = f->real_file_size;
185 static int pre_read_file(struct thread_data *td, struct fio_file *f)
187 int r, did_open = 0, old_runstate;
188 unsigned long long left;
192 if (td->io_ops->flags & FIO_PIPEIO)
195 if (!fio_file_open(f)) {
196 if (td->io_ops->open_file(td, f)) {
197 log_err("fio: cannot pre-read, failed to open file\n");
203 old_runstate = td->runstate;
204 td_set_runstate(td, TD_PRE_READING);
206 bs = td->o.max_bs[DDIR_READ];
210 lseek(f->fd, f->file_offset, SEEK_SET);
213 while (left && !td->terminate) {
217 r = read(f->fd, b, bs);
223 td_verror(td, EIO, "pre_read");
228 td_set_runstate(td, old_runstate);
231 td->io_ops->close_file(td, f);
236 static unsigned long long get_rand_file_size(struct thread_data *td)
238 unsigned long long ret, sized;
241 if (td->o.use_os_rand) {
242 r = os_random_long(&td->file_size_state);
243 sized = td->o.file_size_high - td->o.file_size_low;
244 ret = (unsigned long long) ((double) sized * (r / (OS_RAND_MAX + 1.0)));
246 r = __rand(&td->__file_size_state);
247 sized = td->o.file_size_high - td->o.file_size_low;
248 ret = (unsigned long long) ((double) sized * (r / (FRAND_MAX + 1.0)));
251 ret += td->o.file_size_low;
252 ret -= (ret % td->o.rw_min_bs);
256 static int file_size(struct thread_data *td, struct fio_file *f)
260 if (stat(f->file_name, &st) == -1) {
261 td_verror(td, errno, "fstat");
265 f->real_file_size = st.st_size;
269 static int bdev_size(struct thread_data *td, struct fio_file *f)
271 unsigned long long bytes = 0;
274 if (td->io_ops->open_file(td, f)) {
275 log_err("fio: failed opening blockdev %s for size check\n",
280 r = blockdev_size(f, &bytes);
282 td_verror(td, r, "blockdev_size");
283 printf("fd is %d\n", f->fd);
288 log_err("%s: zero sized block device?\n", f->file_name);
292 f->real_file_size = bytes;
293 td->io_ops->close_file(td, f);
296 td->io_ops->close_file(td, f);
300 static int char_size(struct thread_data *td, struct fio_file *f)
302 #ifdef FIO_HAVE_CHARDEV_SIZE
303 unsigned long long bytes = 0;
306 if (td->io_ops->open_file(td, f)) {
307 log_err("fio: failed opening blockdev %s for size check\n",
312 r = chardev_size(f, &bytes);
314 td_verror(td, r, "chardev_size");
319 log_err("%s: zero sized char device?\n", f->file_name);
323 f->real_file_size = bytes;
324 td->io_ops->close_file(td, f);
327 td->io_ops->close_file(td, f);
330 f->real_file_size = -1ULL;
335 static int get_file_size(struct thread_data *td, struct fio_file *f)
339 if (fio_file_size_known(f))
342 if (f->filetype == FIO_TYPE_FILE)
343 ret = file_size(td, f);
344 else if (f->filetype == FIO_TYPE_BD)
345 ret = bdev_size(td, f);
346 else if (f->filetype == FIO_TYPE_CHAR)
347 ret = char_size(td, f);
349 f->real_file_size = -1;
354 if (f->file_offset > f->real_file_size) {
355 log_err("%s: offset extends end (%llu > %llu)\n", td->o.name,
356 f->file_offset, f->real_file_size);
360 fio_file_set_size_known(f);
364 static int __file_invalidate_cache(struct thread_data *td, struct fio_file *f,
365 unsigned long long off,
366 unsigned long long len)
373 off = f->file_offset;
375 if (len == -1ULL || off == -1ULL)
378 dprint(FD_IO, "invalidate cache %s: %llu/%llu\n", f->file_name, off,
382 * FIXME: add blockdev flushing too
385 ret = posix_madvise(f->mmap_ptr, f->mmap_sz, POSIX_MADV_DONTNEED);
387 (void) posix_madvise(f->mmap_ptr, f->mmap_sz, FIO_MADV_FREE);
389 } else if (f->filetype == FIO_TYPE_FILE) {
390 ret = posix_fadvise(f->fd, off, len, POSIX_FADV_DONTNEED);
391 } else if (f->filetype == FIO_TYPE_BD) {
392 ret = blockdev_invalidate_cache(f);
393 if (ret < 0 && errno == EACCES && geteuid()) {
395 log_err("fio: only root may flush block "
396 "devices. Cache flush bypassed!\n");
401 } else if (f->filetype == FIO_TYPE_CHAR || f->filetype == FIO_TYPE_PIPE)
405 td_verror(td, errno, "invalidate_cache");
407 } else if (ret > 0) {
408 td_verror(td, ret, "invalidate_cache");
416 int file_invalidate_cache(struct thread_data *td, struct fio_file *f)
418 if (!fio_file_open(f))
421 return __file_invalidate_cache(td, f, -1ULL, -1ULL);
424 int generic_close_file(struct thread_data fio_unused *td, struct fio_file *f)
428 dprint(FD_FILE, "fd close %s\n", f->file_name);
432 if (close(f->fd) < 0)
439 static int file_lookup_open(struct fio_file *f, int flags)
441 struct fio_file *__f;
444 __f = lookup_file_hash(f->file_name);
446 dprint(FD_FILE, "found file in hash %s\n", f->file_name);
448 * racy, need the __f->lock locked
451 f->lock_owner = __f->lock_owner;
452 f->lock_batch = __f->lock_batch;
453 f->lock_ddir = __f->lock_ddir;
456 dprint(FD_FILE, "file not found in hash %s\n", f->file_name);
460 f->fd = open(f->file_name, flags, 0600);
464 int generic_open_file(struct thread_data *td, struct fio_file *f)
470 dprint(FD_FILE, "fd open %s\n", f->file_name);
472 if (!strcmp(f->file_name, "-")) {
474 log_err("fio: can't read/write to stdin/out\n");
480 * move output logging to stderr, if we are writing to stdout
487 flags |= OS_O_DIRECT;
490 if (f->filetype != FIO_TYPE_FILE)
491 flags |= FIO_O_NOATIME;
492 if (td->o.create_on_open)
500 if (f->filetype == FIO_TYPE_FILE)
504 f->fd = dup(STDOUT_FILENO);
506 from_hash = file_lookup_open(f, flags);
508 if (f->filetype == FIO_TYPE_CHAR && !read_only)
514 f->fd = dup(STDIN_FILENO);
516 from_hash = file_lookup_open(f, flags);
520 char buf[FIO_VERROR_SIZE];
523 if (__e == EPERM && (flags & FIO_O_NOATIME)) {
524 flags &= ~FIO_O_NOATIME;
528 snprintf(buf, sizeof(buf) - 1, "open(%s)", f->file_name);
530 td_verror(td, __e, buf);
533 if (!from_hash && f->fd != -1) {
534 if (add_file_hash(f)) {
538 * OK to ignore, we haven't done anything with it
540 ret = generic_close_file(td, f);
548 int generic_get_file_size(struct thread_data *td, struct fio_file *f)
550 return get_file_size(td, f);
554 * open/close all files, so that ->real_file_size gets set
556 static int get_file_sizes(struct thread_data *td)
562 for_each_file(td, f, i) {
563 dprint(FD_FILE, "get file size for %p/%d/%p\n", f, i,
566 if (td_io_get_file_size(td, f)) {
567 if (td->error != ENOENT) {
568 log_err("%s\n", td->verror);
574 if (f->real_file_size == -1ULL && td->o.size)
575 f->real_file_size = td->o.size / td->o.nr_files;
582 struct flist_head list;
589 * Get free number of bytes for each file on each unique mount.
591 static unsigned long long get_fs_free_counts(struct thread_data *td)
593 struct flist_head *n, *tmp;
594 unsigned long long ret = 0;
595 struct fio_mount *fm;
600 for_each_file(td, f, i) {
604 if (f->filetype == FIO_TYPE_BD || f->filetype == FIO_TYPE_CHAR) {
605 if (f->real_file_size != -1ULL)
606 ret += f->real_file_size;
608 } else if (f->filetype != FIO_TYPE_FILE)
611 strcpy(buf, f->file_name);
613 if (stat(buf, &sb) < 0) {
617 if (stat(buf, &sb) < 0)
622 flist_for_each(n, &list) {
623 fm = flist_entry(n, struct fio_mount, list);
624 if (fm->key == sb.st_dev)
633 fm = malloc(sizeof(*fm));
634 strcpy(fm->__base, buf);
635 fm->base = basename(fm->__base);
637 flist_add(&fm->list, &list);
640 flist_for_each_safe(n, tmp, &list) {
641 unsigned long long sz;
643 fm = flist_entry(n, struct fio_mount, list);
644 flist_del(&fm->list);
646 sz = get_fs_size(fm->base);
647 if (sz && sz != -1ULL)
657 * Open the files and setup files sizes, creating files if necessary.
659 int setup_files(struct thread_data *td)
661 unsigned long long total_size, extend_size;
664 int err = 0, need_extend;
666 dprint(FD_FILE, "setup files\n");
668 if (td->o.read_iolog_file)
672 * if ioengine defines a setup() method, it's responsible for
673 * opening the files and setting f->real_file_size to indicate
674 * the valid range for that file.
676 if (td->io_ops->setup)
677 err = td->io_ops->setup(td);
679 err = get_file_sizes(td);
685 * check sizes. if the files/devices do not exist and the size
686 * isn't passed to fio, abort.
689 for_each_file(td, f, i) {
690 if (f->real_file_size == -1ULL)
693 total_size += f->real_file_size;
696 if (td->o.fill_device)
697 td->fill_device_size = get_fs_free_counts(td);
700 * device/file sizes are zero and no size given, punt
702 if ((!total_size || total_size == -1ULL) && !td->o.size &&
703 !(td->io_ops->flags & FIO_NOIO) && !td->o.fill_device) {
704 log_err("%s: you need to specify size=\n", td->o.name);
705 td_verror(td, EINVAL, "total_file_size");
710 * now file sizes are known, so we can set ->io_size. if size= is
711 * not given, ->io_size is just equal to ->real_file_size. if size
712 * is given, ->io_size is size / nr_files.
714 extend_size = total_size = 0;
716 for_each_file(td, f, i) {
717 f->file_offset = td->o.start_offset;
719 if (!td->o.file_size_low) {
721 * no file size range given, file size is equal to
722 * total size divided by number of files. if that is
723 * zero, set it to the real file size.
725 f->io_size = td->o.size / td->o.nr_files;
727 f->io_size = f->real_file_size - f->file_offset;
728 } else if (f->real_file_size < td->o.file_size_low ||
729 f->real_file_size > td->o.file_size_high) {
730 if (f->file_offset > td->o.file_size_low)
733 * file size given. if it's fixed, use that. if it's a
734 * range, generate a random size in-between.
736 if (td->o.file_size_low == td->o.file_size_high) {
737 f->io_size = td->o.file_size_low
740 f->io_size = get_rand_file_size(td)
744 f->io_size = f->real_file_size - f->file_offset;
746 if (f->io_size == -1ULL)
749 total_size += f->io_size;
751 if (f->filetype == FIO_TYPE_FILE &&
752 (f->io_size + f->file_offset) > f->real_file_size &&
753 !(td->io_ops->flags & FIO_DISKLESSIO)) {
754 if (!td->o.create_on_open) {
756 extend_size += (f->io_size + f->file_offset);
758 f->real_file_size = f->io_size + f->file_offset;
759 fio_file_set_extend(f);
763 if (!td->o.size || td->o.size > total_size)
764 td->o.size = total_size;
767 * See if we need to extend some files
772 log_info("%s: Laying out IO file(s) (%u file(s) /"
773 " %lluMB)\n", td->o.name, need_extend,
776 for_each_file(td, f, i) {
777 unsigned long long old_len = -1ULL, extend_len = -1ULL;
779 if (!fio_file_extend(f))
782 assert(f->filetype == FIO_TYPE_FILE);
783 fio_file_clear_extend(f);
784 if (!td->o.fill_device) {
785 old_len = f->real_file_size;
786 extend_len = f->io_size + f->file_offset -
789 f->real_file_size = (f->io_size + f->file_offset);
790 err = extend_file(td, f);
794 err = __file_invalidate_cache(td, f, old_len,
807 if (!td->o.zone_size)
808 td->o.zone_size = td->o.size;
811 * iolog already set the total io size, if we read back
814 if (!td->o.read_iolog_file)
815 td->total_io_size = td->o.size * td->o.loops;
818 log_err("%s: you need to specify valid offset=\n", td->o.name);
822 int pre_read_files(struct thread_data *td)
827 dprint(FD_FILE, "pre_read files\n");
829 for_each_file(td, f, i) {
830 pre_read_file(td, f);
836 int init_random_map(struct thread_data *td)
838 unsigned long long blocks, num_maps;
842 if (td->o.norandommap || !td_random(td))
845 for_each_file(td, f, i) {
846 blocks = (f->real_file_size + td->o.rw_min_bs - 1) /
847 (unsigned long long) td->o.rw_min_bs;
848 num_maps = (blocks + BLOCKS_PER_MAP - 1) /
849 (unsigned long long) BLOCKS_PER_MAP;
850 f->file_map = smalloc(num_maps * sizeof(unsigned long));
852 f->num_maps = num_maps;
855 if (!td->o.softrandommap) {
856 log_err("fio: failed allocating random map. If running"
857 " a large number of jobs, try the 'norandommap'"
858 " option or set 'softrandommap'. Or give"
859 " a larger --alloc-size to fio.\n");
863 log_info("fio: file %s failed allocating random map. Running "
864 "job without.\n", f->file_name);
871 void close_files(struct thread_data *td)
876 for_each_file(td, f, i) {
877 if (fio_file_open(f))
878 td_io_close_file(td, f);
882 void close_and_free_files(struct thread_data *td)
887 dprint(FD_FILE, "close files\n");
889 for_each_file(td, f, i) {
890 if (td->o.unlink && f->filetype == FIO_TYPE_FILE) {
891 dprint(FD_FILE, "free unlink %s\n", f->file_name);
892 unlink(f->file_name);
895 if (fio_file_open(f))
896 td_io_close_file(td, f);
907 td->o.filename = NULL;
914 static void get_file_type(struct fio_file *f)
918 if (!strcmp(f->file_name, "-"))
919 f->filetype = FIO_TYPE_PIPE;
921 f->filetype = FIO_TYPE_FILE;
923 if (!stat(f->file_name, &sb)) {
924 /* \\.\ is the device namespace in Windows, where every file is
926 if (S_ISBLK(sb.st_mode) || strncmp(f->file_name, "\\\\.\\", 4) == 0)
927 f->filetype = FIO_TYPE_BD;
928 else if (S_ISCHR(sb.st_mode))
929 f->filetype = FIO_TYPE_CHAR;
930 else if (S_ISFIFO(sb.st_mode))
931 f->filetype = FIO_TYPE_PIPE;
935 int add_file(struct thread_data *td, const char *fname)
937 int cur_files = td->files_index;
938 char file_name[PATH_MAX];
942 dprint(FD_FILE, "add file %s\n", fname);
944 f = smalloc(sizeof(*f));
946 log_err("fio: smalloc OOM\n");
953 if (td->files_size <= td->files_index) {
954 int new_size = td->o.nr_files + 1;
956 dprint(FD_FILE, "resize file array to %d files\n", new_size);
958 td->files = realloc(td->files, new_size * sizeof(f));
959 td->files_size = new_size;
961 td->files[cur_files] = f;
964 * init function, io engine may not be loaded yet
966 if (td->io_ops && (td->io_ops->flags & FIO_DISKLESSIO))
967 f->real_file_size = -1ULL;
970 len = sprintf(file_name, "%s/", td->o.directory);
972 sprintf(file_name + len, "%s", fname);
973 f->file_name = smalloc_strdup(file_name);
975 log_err("fio: smalloc OOM\n");
981 switch (td->o.file_lock_mode) {
984 case FILE_LOCK_READWRITE:
985 f->lock = fio_mutex_rw_init();
987 case FILE_LOCK_EXCLUSIVE:
988 f->lock = fio_mutex_init(1);
991 log_err("fio: unknown lock mode: %d\n", td->o.file_lock_mode);
996 if (f->filetype == FIO_TYPE_FILE)
997 td->nr_normal_files++;
999 dprint(FD_FILE, "file %p \"%s\" added at %d\n", f, f->file_name,
1005 int add_file_exclusive(struct thread_data *td, const char *fname)
1010 for_each_file(td, f, i) {
1011 if (!strcmp(f->file_name, fname))
1015 return add_file(td, fname);
1018 void get_file(struct fio_file *f)
1020 dprint(FD_FILE, "get file %s, ref=%d\n", f->file_name, f->references);
1021 assert(fio_file_open(f));
1025 int put_file(struct thread_data *td, struct fio_file *f)
1027 int f_ret = 0, ret = 0;
1029 dprint(FD_FILE, "put file %s, ref=%d\n", f->file_name, f->references);
1031 if (!fio_file_open(f)) {
1032 assert(f->fd == -1);
1036 assert(f->references);
1037 if (--f->references)
1040 if (should_fsync(td) && td->o.fsync_on_close)
1041 f_ret = fsync(f->fd);
1043 if (td->io_ops->close_file)
1044 ret = td->io_ops->close_file(td, f);
1049 td->nr_open_files--;
1050 fio_file_clear_open(f);
1051 assert(f->fd == -1);
1055 void lock_file(struct thread_data *td, struct fio_file *f, enum fio_ddir ddir)
1057 if (!f->lock || td->o.file_lock_mode == FILE_LOCK_NONE)
1060 if (f->lock_owner == td && f->lock_batch--)
1063 if (td->o.file_lock_mode == FILE_LOCK_READWRITE) {
1064 if (ddir == DDIR_READ)
1065 fio_mutex_down_read(f->lock);
1067 fio_mutex_down_write(f->lock);
1068 } else if (td->o.file_lock_mode == FILE_LOCK_EXCLUSIVE)
1069 fio_mutex_down(f->lock);
1072 f->lock_batch = td->o.lockfile_batch;
1073 f->lock_ddir = ddir;
1076 void unlock_file(struct thread_data *td, struct fio_file *f)
1078 if (!f->lock || td->o.file_lock_mode == FILE_LOCK_NONE)
1083 if (td->o.file_lock_mode == FILE_LOCK_READWRITE) {
1084 const int is_read = f->lock_ddir == DDIR_READ;
1085 int val = fio_mutex_getval(f->lock);
1087 if ((is_read && val == 1) || (!is_read && val == -1))
1088 f->lock_owner = NULL;
1091 fio_mutex_up_read(f->lock);
1093 fio_mutex_up_write(f->lock);
1094 } else if (td->o.file_lock_mode == FILE_LOCK_EXCLUSIVE) {
1095 int val = fio_mutex_getval(f->lock);
1098 f->lock_owner = NULL;
1100 fio_mutex_up(f->lock);
1104 void unlock_file_all(struct thread_data *td, struct fio_file *f)
1106 if (f->lock_owner != td)
1113 static int recurse_dir(struct thread_data *td, const char *dirname)
1119 D = opendir(dirname);
1121 char buf[FIO_VERROR_SIZE];
1123 snprintf(buf, FIO_VERROR_SIZE - 1, "opendir(%s)", dirname);
1124 td_verror(td, errno, buf);
1128 while ((dir = readdir(D)) != NULL) {
1129 char full_path[PATH_MAX];
1132 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
1135 sprintf(full_path, "%s/%s", dirname, dir->d_name);
1137 if (lstat(full_path, &sb) == -1) {
1138 if (errno != ENOENT) {
1139 td_verror(td, errno, "stat");
1144 if (S_ISREG(sb.st_mode)) {
1145 add_file(td, full_path);
1149 if (!S_ISDIR(sb.st_mode))
1152 ret = recurse_dir(td, full_path);
1161 int add_dir_files(struct thread_data *td, const char *path)
1163 int ret = recurse_dir(td, path);
1166 log_info("fio: opendir added %d files\n", td->o.nr_files);
1171 void dup_files(struct thread_data *td, struct thread_data *org)
1176 dprint(FD_FILE, "dup files: %d\n", org->files_index);
1181 td->files = malloc(org->files_index * sizeof(f));
1183 for_each_file(org, f, i) {
1184 struct fio_file *__f;
1186 __f = smalloc(sizeof(*__f));
1188 log_err("fio: smalloc OOM\n");
1192 fio_file_reset(__f);
1195 __f->file_name = smalloc_strdup(f->file_name);
1196 if (!__f->file_name) {
1197 log_err("fio: smalloc OOM\n");
1201 __f->filetype = f->filetype;
1209 * Returns the index that matches the filename, or -1 if not there
1211 int get_fileno(struct thread_data *td, const char *fname)
1216 for_each_file(td, f, i)
1217 if (!strcmp(f->file_name, fname))
1224 * For log usage, where we add/open/close files automatically
1226 void free_release_files(struct thread_data *td)
1229 td->files_index = 0;
1230 td->nr_normal_files = 0;