X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=filesetup.c;h=ab5cc48da71ddb317bab84109eb1845b9a7818e5;hp=15c431603718291d04b3ab50cf623ea6c83b80c7;hb=d537c08b0eacdeb05835ee8c5f269b872239713a;hpb=22a57ba81de2c3f458797e9158da760c9e0ea435 diff --git a/filesetup.c b/filesetup.c index 15c43160..ab5cc48d 100644 --- a/filesetup.c +++ b/filesetup.c @@ -3,6 +3,7 @@ #include #include #include +#include #include #include #include @@ -10,6 +11,13 @@ #include "fio.h" #include "smalloc.h" #include "filehash.h" +#include "os/os.h" +#include "hash.h" +#include "lib/axmap.h" + +#ifdef FIO_HAVE_LINUX_FALLOCATE +#include +#endif static int root_warn; @@ -64,6 +72,43 @@ static int extend_file(struct thread_data *td, struct fio_file *f) return 1; } +#ifdef FIO_HAVE_FALLOCATE + if (!td->o.fill_device) { + switch (td->o.fallocate_mode) { + case FIO_FALLOCATE_NONE: + break; + case FIO_FALLOCATE_POSIX: + dprint(FD_FILE, "posix_fallocate file %s size %llu\n", + f->file_name, f->real_file_size); + + r = posix_fallocate(f->fd, 0, f->real_file_size); + if (r > 0) { + log_err("fio: posix_fallocate fails: %s\n", + strerror(r)); + } + break; +#ifdef FIO_HAVE_LINUX_FALLOCATE + case FIO_FALLOCATE_KEEP_SIZE: + dprint(FD_FILE, + "fallocate(FALLOC_FL_KEEP_SIZE) " + "file %s size %llu\n", + f->file_name, f->real_file_size); + + r = fallocate(f->fd, FALLOC_FL_KEEP_SIZE, 0, + f->real_file_size); + if (r != 0) { + td_verror(td, errno, "fallocate"); + } + break; +#endif /* FIO_HAVE_LINUX_FALLOCATE */ + default: + log_err("fio: unknown fallocate mode: %d\n", + td->o.fallocate_mode); + assert(0); + } + } +#endif /* FIO_HAVE_FALLOCATE */ + if (!new_layout) goto done; @@ -78,16 +123,6 @@ static int extend_file(struct thread_data *td, struct fio_file *f) td_verror(td, errno, "ftruncate"); goto err; } - -#ifdef FIO_HAVE_FALLOCATE - dprint(FD_FILE, "fallocate file %s, size %llu\n", f->file_name, - f->real_file_size); - r = posix_fallocate(f->fd, 0, f->real_file_size); - if (r < 0) { - log_err("fio: posix_fallocate fails: %s\n", - strerror(-r)); - } -#endif } b = malloc(td->o.max_bs[DDIR_WRITE]); @@ -156,6 +191,9 @@ static int pre_read_file(struct thread_data *td, struct fio_file *f) unsigned int bs; char *b; + if (td->io_ops->flags & FIO_PIPEIO) + return 0; + if (!fio_file_open(f)) { if (td->io_ops->open_file(td, f)) { log_err("fio: cannot pre-read, failed to open file\n"); @@ -200,11 +238,18 @@ static int pre_read_file(struct thread_data *td, struct fio_file *f) static unsigned long long get_rand_file_size(struct thread_data *td) { unsigned long long ret, sized; - long r; + unsigned long r; + + if (td->o.use_os_rand) { + r = os_random_long(&td->file_size_state); + sized = td->o.file_size_high - td->o.file_size_low; + ret = (unsigned long long) ((double) sized * (r / (OS_RAND_MAX + 1.0))); + } else { + r = __rand(&td->__file_size_state); + sized = td->o.file_size_high - td->o.file_size_low; + ret = (unsigned long long) ((double) sized * (r / (FRAND_MAX + 1.0))); + } - r = os_random_long(&td->file_size_state); - sized = td->o.file_size_high - td->o.file_size_low; - ret = (unsigned long long) ((double) sized * (r / (OS_RAND_MAX + 1.0))); ret += td->o.file_size_low; ret -= (ret % td->o.rw_min_bs); return ret; @@ -225,7 +270,7 @@ static int file_size(struct thread_data *td, struct fio_file *f) static int bdev_size(struct thread_data *td, struct fio_file *f) { - unsigned long long bytes; + unsigned long long bytes = 0; int r; if (td->io_ops->open_file(td, f)) { @@ -234,7 +279,7 @@ static int bdev_size(struct thread_data *td, struct fio_file *f) return 1; } - r = blockdev_size(f->fd, &bytes); + r = blockdev_size(f, &bytes); if (r) { td_verror(td, r, "blockdev_size"); goto err; @@ -253,6 +298,41 @@ err: return 1; } +static int char_size(struct thread_data *td, struct fio_file *f) +{ +#ifdef FIO_HAVE_CHARDEV_SIZE + unsigned long long bytes = 0; + int r; + + if (td->io_ops->open_file(td, f)) { + log_err("fio: failed opening blockdev %s for size check\n", + f->file_name); + return 1; + } + + r = chardev_size(f, &bytes); + if (r) { + td_verror(td, r, "chardev_size"); + goto err; + } + + if (!bytes) { + log_err("%s: zero sized char device?\n", f->file_name); + goto err; + } + + f->real_file_size = bytes; + td->io_ops->close_file(td, f); + return 0; +err: + td->io_ops->close_file(td, f); + return 1; +#else + f->real_file_size = -1ULL; + return 0; +#endif +} + static int get_file_size(struct thread_data *td, struct fio_file *f) { int ret = 0; @@ -264,6 +344,8 @@ static int get_file_size(struct thread_data *td, struct fio_file *f) ret = file_size(td, f); else if (f->filetype == FIO_TYPE_BD) ret = bdev_size(td, f); + else if (f->filetype == FIO_TYPE_CHAR) + ret = char_size(td, f); else f->real_file_size = -1; @@ -271,7 +353,7 @@ static int get_file_size(struct thread_data *td, struct fio_file *f) return ret; if (f->file_offset > f->real_file_size) { - log_err("%s: offset extends end (%Lu > %Lu)\n", td->o.name, + log_err("%s: offset extends end (%llu > %llu)\n", td->o.name, f->file_offset, f->real_file_size); return 1; } @@ -300,12 +382,15 @@ static int __file_invalidate_cache(struct thread_data *td, struct fio_file *f, /* * FIXME: add blockdev flushing too */ - if (f->mmap_ptr) - ret = madvise(f->mmap_ptr, f->mmap_sz, MADV_DONTNEED); - else if (f->filetype == FIO_TYPE_FILE) { - ret = fadvise(f->fd, off, len, POSIX_FADV_DONTNEED); + if (f->mmap_ptr) { + ret = posix_madvise(f->mmap_ptr, f->mmap_sz, POSIX_MADV_DONTNEED); +#ifdef FIO_MADV_FREE + (void) posix_madvise(f->mmap_ptr, f->mmap_sz, FIO_MADV_FREE); +#endif + } else if (f->filetype == FIO_TYPE_FILE) { + ret = posix_fadvise(f->fd, off, len, POSIX_FADV_DONTNEED); } else if (f->filetype == FIO_TYPE_BD) { - ret = blockdev_invalidate_cache(f->fd); + ret = blockdev_invalidate_cache(f); if (ret < 0 && errno == EACCES && geteuid()) { if (!root_warn) { log_err("fio: only root may flush block " @@ -349,10 +434,16 @@ int generic_close_file(struct thread_data fio_unused *td, struct fio_file *f) ret = errno; f->fd = -1; + + if (f->shadow_fd != -1) { + close(f->shadow_fd); + f->shadow_fd = -1; + } + return ret; } -static int file_lookup_open(struct fio_file *f, int flags) +int file_lookup_open(struct fio_file *f, int flags) { struct fio_file *__f; int from_hash; @@ -377,6 +468,24 @@ static int file_lookup_open(struct fio_file *f, int flags) return from_hash; } +static int file_close_shadow_fds(struct thread_data *td) +{ + struct fio_file *f; + int num_closed = 0; + unsigned int i; + + for_each_file(td, f, i) { + if (f->shadow_fd == -1) + continue; + + close(f->shadow_fd); + f->shadow_fd = -1; + num_closed++; + } + + return num_closed; +} + int generic_open_file(struct thread_data *td, struct fio_file *f) { int is_std = 0; @@ -385,6 +494,11 @@ int generic_open_file(struct thread_data *td, struct fio_file *f) dprint(FD_FILE, "fd open %s\n", f->file_name); + if (td_trim(td) && f->filetype != FIO_TYPE_BD) { + log_err("fio: trim only applies to block device\n"); + return 1; + } + if (!strcmp(f->file_name, "-")) { if (td_rw(td)) { log_err("fio: can't read/write to stdin/out\n"); @@ -399,14 +513,17 @@ int generic_open_file(struct thread_data *td, struct fio_file *f) f_out = stderr; } + if (td_trim(td)) + goto skip_flags; if (td->o.odirect) flags |= OS_O_DIRECT; if (td->o.sync_io) flags |= O_SYNC; - if (f->filetype != FIO_TYPE_FILE) - flags |= FIO_O_NOATIME; if (td->o.create_on_open) flags |= O_CREAT; +skip_flags: + if (f->filetype != FIO_TYPE_FILE) + flags |= FIO_O_NOATIME; open_again: if (td_write(td)) { @@ -420,7 +537,7 @@ open_again: f->fd = dup(STDOUT_FILENO); else from_hash = file_lookup_open(f, flags); - } else { + } else if (td_read(td)) { if (f->filetype == FIO_TYPE_CHAR && !read_only) flags |= O_RDWR; else @@ -430,30 +547,53 @@ open_again: f->fd = dup(STDIN_FILENO); else from_hash = file_lookup_open(f, flags); + } else { //td trim + flags |= O_RDWR; + from_hash = file_lookup_open(f, flags); } if (f->fd == -1) { char buf[FIO_VERROR_SIZE]; int __e = errno; - if (errno == EPERM && (flags & FIO_O_NOATIME)) { + if (__e == EPERM && (flags & FIO_O_NOATIME)) { flags &= ~FIO_O_NOATIME; goto open_again; } + if (__e == EMFILE && file_close_shadow_fds(td)) + goto open_again; snprintf(buf, sizeof(buf) - 1, "open(%s)", f->file_name); + if (__e == EINVAL && (flags & OS_O_DIRECT)) { + log_err("fio: looks like your file system does not " \ + "support direct=1/buffered=0\n"); + } + td_verror(td, __e, buf); } if (!from_hash && f->fd != -1) { if (add_file_hash(f)) { - int ret; + int fio_unused ret; /* - * OK to ignore, we haven't done anything with it + * Stash away descriptor for later close. This is to + * work-around a "feature" on Linux, where a close of + * an fd that has been opened for write will trigger + * udev to call blkid to check partitions, fs id, etc. + * That polutes the device cache, which can slow down + * unbuffered accesses. */ - ret = generic_close_file(td, f); + if (f->shadow_fd == -1) + f->shadow_fd = f->fd; + else { + /* + * OK to ignore, we haven't done anything + * with it + */ + ret = generic_close_file(td, f); + } goto open_again; } } @@ -494,6 +634,87 @@ static int get_file_sizes(struct thread_data *td) return err; } +struct fio_mount { + struct flist_head list; + const char *base; + char __base[256]; + unsigned int key; +}; + +/* + * Get free number of bytes for each file on each unique mount. + */ +static unsigned long long get_fs_free_counts(struct thread_data *td) +{ + struct flist_head *n, *tmp; + unsigned long long ret = 0; + struct fio_mount *fm; + FLIST_HEAD(list); + struct fio_file *f; + unsigned int i; + + for_each_file(td, f, i) { + struct stat sb; + char buf[256]; + + if (f->filetype == FIO_TYPE_BD || f->filetype == FIO_TYPE_CHAR) { + if (f->real_file_size != -1ULL) + ret += f->real_file_size; + continue; + } else if (f->filetype != FIO_TYPE_FILE) + continue; + + strcpy(buf, f->file_name); + + if (stat(buf, &sb) < 0) { + if (errno != ENOENT) + break; + strcpy(buf, "."); + if (stat(buf, &sb) < 0) + break; + } + + fm = NULL; + flist_for_each(n, &list) { + fm = flist_entry(n, struct fio_mount, list); + if (fm->key == sb.st_dev) + break; + + fm = NULL; + } + + if (fm) + continue; + + fm = malloc(sizeof(*fm)); + strcpy(fm->__base, buf); + fm->base = basename(fm->__base); + fm->key = sb.st_dev; + flist_add(&fm->list, &list); + } + + flist_for_each_safe(n, tmp, &list) { + unsigned long long sz; + + fm = flist_entry(n, struct fio_mount, list); + flist_del(&fm->list); + + sz = get_fs_size(fm->base); + if (sz && sz != -1ULL) + ret += sz; + + free(fm); + } + + return ret; +} + +uint64_t get_start_offset(struct thread_data *td) +{ + return td->o.start_offset + + (td->thread_number - 1) * td->o.offset_increment; +} + /* * Open the files and setup files sizes, creating files if necessary. */ @@ -507,7 +728,7 @@ int setup_files(struct thread_data *td) dprint(FD_FILE, "setup files\n"); if (td->o.read_iolog_file) - return 0; + goto done; /* * if ioengine defines a setup() method, it's responsible for @@ -534,6 +755,9 @@ int setup_files(struct thread_data *td) total_size += f->real_file_size; } + if (td->o.fill_device) + td->fill_device_size = get_fs_free_counts(td); + /* * device/file sizes are zero and no size given, punt */ @@ -552,7 +776,7 @@ int setup_files(struct thread_data *td) extend_size = total_size = 0; need_extend = 0; for_each_file(td, f, i) { - f->file_offset = td->o.start_offset; + f->file_offset = get_start_offset(td); if (!td->o.file_size_low) { /* @@ -583,8 +807,11 @@ int setup_files(struct thread_data *td) if (f->io_size == -1ULL) total_size = -1ULL; - else + else { + if (td->o.size_percent) + f->io_size = (f->io_size * td->o.size_percent) / 100; total_size += f->io_size; + } if (f->filetype == FIO_TYPE_FILE && (f->io_size + f->file_offset) > f->real_file_size && @@ -606,9 +833,9 @@ int setup_files(struct thread_data *td) */ if (need_extend) { temp_stall_ts = 1; - if (!terse_output) + if (output_format == FIO_OUTPUT_NORMAL) log_info("%s: Laying out IO file(s) (%u file(s) /" - " %LuMiB)\n", td->o.name, need_extend, + " %lluMB)\n", td->o.name, need_extend, extend_size >> 20); for_each_file(td, f, i) { @@ -621,7 +848,8 @@ int setup_files(struct thread_data *td) fio_file_clear_extend(f); if (!td->o.fill_device) { old_len = f->real_file_size; - extend_len = f->io_size + f->file_offset - old_len; + extend_len = f->io_size + f->file_offset - + old_len; } f->real_file_size = (f->io_size + f->file_offset); err = extend_file(td, f); @@ -650,6 +878,11 @@ int setup_files(struct thread_data *td) */ if (!td->o.read_iolog_file) td->total_io_size = td->o.size * td->o.loops; + +done: + if (td->o.create_only) + td->done = 1; + return 0; err_offset: log_err("%s: you need to specify valid offset=\n", td->o.name); @@ -670,25 +903,75 @@ int pre_read_files(struct thread_data *td) return 1; } +static int __init_rand_distribution(struct thread_data *td, struct fio_file *f) +{ + unsigned int range_size, seed; + unsigned long nranges; + uint64_t file_size; + + range_size = min(td->o.min_bs[DDIR_READ], td->o.min_bs[DDIR_WRITE]); + file_size = min(f->real_file_size, f->io_size); + + nranges = (file_size + range_size - 1) / range_size; + + seed = jhash(f->file_name, strlen(f->file_name), 0) * td->thread_number; + if (!td->o.rand_repeatable) + seed = td->rand_seeds[4]; + + if (td->o.random_distribution == FIO_RAND_DIST_ZIPF) + zipf_init(&f->zipf, nranges, td->o.zipf_theta, seed); + else + pareto_init(&f->zipf, nranges, td->o.pareto_h, seed); + + return 1; +} + +static int init_rand_distribution(struct thread_data *td) +{ + struct fio_file *f; + unsigned int i; + int state; + + if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) + return 0; + + state = td->runstate; + td_set_runstate(td, TD_SETTING_UP); + for_each_file(td, f, i) + __init_rand_distribution(td, f); + td_set_runstate(td, state); + + return 1; +} + int init_random_map(struct thread_data *td) { - unsigned long long blocks, num_maps; + unsigned long long blocks; struct fio_file *f; unsigned int i; - if (td->o.norandommap || !td_random(td)) + if (init_rand_distribution(td)) + return 0; + if (!td_random(td)) return 0; for_each_file(td, f, i) { blocks = (f->real_file_size + td->o.rw_min_bs - 1) / (unsigned long long) td->o.rw_min_bs; - num_maps = (blocks + BLOCKS_PER_MAP - 1) / - (unsigned long long) BLOCKS_PER_MAP; - f->file_map = smalloc(num_maps * sizeof(int)); - if (f->file_map) { - f->num_maps = num_maps; + if (td->o.random_generator == FIO_RAND_GEN_LFSR) { + unsigned long seed; + + seed = td->rand_seeds[FIO_RAND_BLOCK_OFF]; + + if (!lfsr_init(&f->lfsr, blocks, seed)) + continue; + } else if (!td->o.norandommap) { + f->io_axmap = axmap_new(blocks); + if (f->io_axmap) + continue; + } else if (td->o.norandommap) continue; - } + if (!td->o.softrandommap) { log_err("fio: failed allocating random map. If running" " a large number of jobs, try the 'norandommap'" @@ -699,7 +982,6 @@ int init_random_map(struct thread_data *td) log_info("fio: file %s failed allocating random map. Running " "job without.\n", f->file_name); - f->num_maps = 0; } return 0; @@ -710,8 +992,10 @@ void close_files(struct thread_data *td) struct fio_file *f; unsigned int i; - for_each_file(td, f, i) - td_io_close_file(td, f); + for_each_file(td, f, i) { + if (fio_file_open(f)) + td_io_close_file(td, f); + } } void close_and_free_files(struct thread_data *td) @@ -734,11 +1018,8 @@ void close_and_free_files(struct thread_data *td) sfree(f->file_name); f->file_name = NULL; - - if (f->file_map) { - sfree(f->file_map); - f->file_map = NULL; - } + axmap_free(f->io_axmap); + f->io_axmap = NULL; sfree(f); } @@ -758,7 +1039,12 @@ static void get_file_type(struct fio_file *f) else f->filetype = FIO_TYPE_FILE; - if (!lstat(f->file_name, &sb)) { + /* \\.\ is the device namespace in Windows, where every file is + * a block device */ + if (strncmp(f->file_name, "\\\\.\\", 4) == 0) + f->filetype = FIO_TYPE_BD; + + if (!stat(f->file_name, &sb)) { if (S_ISBLK(sb.st_mode)) f->filetype = FIO_TYPE_BD; else if (S_ISCHR(sb.st_mode)) @@ -782,18 +1068,25 @@ int add_file(struct thread_data *td, const char *fname) log_err("fio: smalloc OOM\n"); assert(0); } - + f->fd = -1; + f->shadow_fd = -1; + fio_file_reset(f); if (td->files_size <= td->files_index) { - int new_size = td->o.nr_files + 1; + unsigned int new_size = td->o.nr_files + 1; dprint(FD_FILE, "resize file array to %d files\n", new_size); td->files = realloc(td->files, new_size * sizeof(f)); + if (td->files == NULL) { + log_err("fio: realloc OOM\n"); + assert(0); + } td->files_size = new_size; } td->files[cur_files] = f; + f->fileno = cur_files; /* * init function, io engine may not be loaded yet @@ -810,7 +1103,7 @@ int add_file(struct thread_data *td, const char *fname) log_err("fio: smalloc OOM\n"); assert(0); } - + get_file_type(f); switch (td->o.file_lock_mode) { @@ -820,7 +1113,7 @@ int add_file(struct thread_data *td, const char *fname) f->lock = fio_mutex_rw_init(); break; case FILE_LOCK_EXCLUSIVE: - f->lock = fio_mutex_init(1); + f->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED); break; default: log_err("fio: unknown lock mode: %d\n", td->o.file_lock_mode); @@ -837,6 +1130,19 @@ int add_file(struct thread_data *td, const char *fname) return cur_files; } +int add_file_exclusive(struct thread_data *td, const char *fname) +{ + struct fio_file *f; + unsigned int i; + + for_each_file(td, f, i) { + if (!strcmp(f->file_name, fname)) + return i; + } + + return add_file(td, fname); +} + void get_file(struct fio_file *f) { dprint(FD_FILE, "get file %s, ref=%d\n", f->file_name, f->references); @@ -954,7 +1260,7 @@ static int recurse_dir(struct thread_data *td, const char *dirname) if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, "..")) continue; - sprintf(full_path, "%s/%s", dirname, dir->d_name); + sprintf(full_path, "%s%s%s", dirname, FIO_OS_PATH_SEPARATOR, dir->d_name); if (lstat(full_path, &sb) == -1) { if (errno != ENOENT) { @@ -1011,14 +1317,15 @@ void dup_files(struct thread_data *td, struct thread_data *org) assert(0); } __f->fd = -1; - + fio_file_reset(__f); + if (f->file_name) { __f->file_name = smalloc_strdup(f->file_name); if (!__f->file_name) { log_err("fio: smalloc OOM\n"); assert(0); } - + __f->filetype = f->filetype; }