#include "fio.h"
#include "smalloc.h"
#include "filehash.h"
+#include "options.h"
#include "os/os.h"
#include "hash.h"
#include "lib/axmap.h"
static int root_warn;
+static FLIST_HEAD(filename_list);
+
+/*
+ * List entry for filename_list
+ */
+struct file_name {
+ struct flist_head list;
+ char *filename;
+};
+
static inline void clear_error(struct thread_data *td)
{
td->error = 0;
int r, new_layout = 0, unlink_file = 0, flags;
unsigned long long left;
unsigned int bs;
- char *b;
+ char *b = NULL;
if (read_only) {
log_err("fio: refusing extend of file due to read-only\n");
* does that for operations involving reads, or for writes
* where overwrite is set
*/
- if (td_read(td) || (td_write(td) && td->o.overwrite) ||
- (td_write(td) && td->io_ops->flags & FIO_NOEXTEND))
+ if (td_read(td) ||
+ (td_write(td) && td->o.overwrite && !td->o.file_append) ||
+ (td_write(td) && td_ioengine_flagged(td, FIO_NOEXTEND)))
new_layout = 1;
- if (td_write(td) && !td->o.overwrite)
+ if (td_write(td) && !td->o.overwrite && !td->o.file_append)
unlink_file = 1;
if (unlink_file || new_layout) {
+ int ret;
+
dprint(FD_FILE, "layout unlink %s\n", f->file_name);
- if ((unlink(f->file_name) < 0) && (errno != ENOENT)) {
+
+ ret = td_io_unlink_file(td, f);
+ if (ret != 0 && ret != ENOENT) {
td_verror(td, errno, "unlink");
return 1;
}
}
- flags = O_WRONLY | O_CREAT;
+ flags = O_WRONLY;
+ if (td->o.allow_create)
+ flags |= O_CREAT;
if (new_layout)
flags |= O_TRUNC;
+#ifdef WIN32
+ flags |= _O_BINARY;
+#endif
+
dprint(FD_FILE, "open file %s, flags %x\n", f->file_name, flags);
f->fd = open(f->file_name, flags, 0644);
if (f->fd < 0) {
- td_verror(td, errno, "open");
+ int err = errno;
+
+ if (err == ENOENT && !td->o.allow_create)
+ log_err("fio: file creation disallowed by "
+ "allow_file_create=0\n");
+ else
+ td_verror(td, err, "open");
return 1;
}
}
#endif /* CONFIG_POSIX_FALLOCATE */
+ /*
+ * If our jobs don't require regular files initially, we're done.
+ */
if (!new_layout)
goto done;
dprint(FD_FILE, "truncate file %s, size %llu\n", f->file_name,
(unsigned long long) f->real_file_size);
if (ftruncate(f->fd, f->real_file_size) == -1) {
- td_verror(td, errno, "ftruncate");
- goto err;
+ if (errno != EFBIG) {
+ td_verror(td, errno, "ftruncate");
+ goto err;
+ }
}
}
- b = malloc(td->o.max_bs[DDIR_WRITE]);
- memset(b, 0, td->o.max_bs[DDIR_WRITE]);
-
left = f->real_file_size;
+ bs = td->o.max_bs[DDIR_WRITE];
+ if (bs > left)
+ bs = left;
+
+ b = malloc(bs);
+ if (!b) {
+ td_verror(td, errno, "malloc");
+ goto err;
+ }
+
while (left && !td->terminate) {
- bs = td->o.max_bs[DDIR_WRITE];
if (bs > left)
bs = left;
+ fill_io_buffer(td, b, bs, bs);
+
r = write(f->fd, b, bs);
if (r > 0) {
if (td->terminate) {
dprint(FD_FILE, "terminate unlink %s\n", f->file_name);
- unlink(f->file_name);
+ td_io_unlink_file(td, f);
} else if (td->o.create_fsync) {
if (fsync(f->fd) < 0) {
td_verror(td, errno, "fsync");
err:
close(f->fd);
f->fd = -1;
+ if (b)
+ free(b);
return 1;
}
static int pre_read_file(struct thread_data *td, struct fio_file *f)
{
- int r, did_open = 0, old_runstate;
+ int ret = 0, r, did_open = 0, old_runstate;
unsigned long long left;
unsigned int bs;
char *b;
- if (td->io_ops->flags & FIO_PIPEIO)
+ if (td_ioengine_flagged(td, FIO_PIPEIO) ||
+ td_ioengine_flagged(td, FIO_NOIO))
+ return 0;
+
+ if (f->filetype == FIO_TYPE_CHAR)
return 0;
if (!fio_file_open(f)) {
did_open = 1;
}
- old_runstate = td->runstate;
- td_set_runstate(td, TD_PRE_READING);
+ old_runstate = td_bump_runstate(td, TD_PRE_READING);
+ left = f->io_size;
bs = td->o.max_bs[DDIR_READ];
+ if (bs > left)
+ bs = left;
+
b = malloc(bs);
+ if (!b) {
+ td_verror(td, errno, "malloc");
+ ret = 1;
+ goto error;
+ }
memset(b, 0, bs);
- lseek(f->fd, f->file_offset, SEEK_SET);
- left = f->io_size;
+ if (lseek(f->fd, f->file_offset, SEEK_SET) < 0) {
+ td_verror(td, errno, "lseek");
+ log_err("fio: failed to lseek pre-read file\n");
+ ret = 1;
+ goto error;
+ }
while (left && !td->terminate) {
if (bs > left)
}
}
- td_set_runstate(td, old_runstate);
+error:
+ td_restore_runstate(td, old_runstate);
if (did_open)
td->io_ops->close_file(td, f);
+
free(b);
- return 0;
+ return ret;
}
-static unsigned long long get_rand_file_size(struct thread_data *td)
+unsigned long long get_rand_file_size(struct thread_data *td)
{
unsigned long long ret, sized;
+ uint64_t frand_max;
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)));
- }
-
+ frand_max = rand_max(&td->file_size_state);
+ 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)));
ret += td->o.file_size_low;
ret -= (ret % td->o.rw_min_bs);
return ret;
int r;
if (td->io_ops->open_file(td, f)) {
- log_err("fio: failed opening blockdev %s for size check\n",
+ log_err("fio: failed opening chardev %s for size check\n",
f->file_name);
return 1;
}
if (f->filetype == FIO_TYPE_FILE)
ret = file_size(td, f);
- else if (f->filetype == FIO_TYPE_BD)
+ else if (f->filetype == FIO_TYPE_BLOCK)
ret = bdev_size(td, f);
else if (f->filetype == FIO_TYPE_CHAR)
ret = char_size(td, f);
else
- f->real_file_size = -1;
+ f->real_file_size = -1ULL;
+ /*
+ * Leave ->real_file_size with 0 since it could be expectation
+ * of initial setup for regular files.
+ */
if (ret)
return ret;
+ /*
+ * If ->real_file_size is -1, a conditional for the message
+ * "offset extends end" is always true, but it makes no sense,
+ * so just return the same value here.
+ */
+ if (f->real_file_size == -1ULL) {
+ log_info("%s: failed to get file size of %s\n", td->o.name,
+ f->file_name);
+ return 1;
+ }
+
+ if (td->o.start_offset && f->file_offset == 0)
+ dprint(FD_FILE, "offset of file %s not initialized yet\n",
+ f->file_name);
+ /*
+ * ->file_offset normally hasn't been initialized yet, so this
+ * is basically always false.
+ */
if (f->file_offset > f->real_file_size) {
log_err("%s: offset extends end (%llu > %llu)\n", td->o.name,
(unsigned long long) f->file_offset,
unsigned long long off,
unsigned long long len)
{
- int ret = 0;
+ int errval = 0, ret = 0;
+
+#ifdef CONFIG_ESX
+ return 0;
+#endif
if (len == -1ULL)
len = f->io_size;
if (len == -1ULL || off == -1ULL)
return 0;
- dprint(FD_IO, "invalidate cache %s: %llu/%llu\n", f->file_name, off,
- len);
-
- /*
- * FIXME: add blockdev flushing too
- */
- 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
+ if (td->io_ops->invalidate) {
+ dprint(FD_IO, "invalidate %s cache %s\n", td->io_ops->name,
+ f->file_name);
+ ret = td->io_ops->invalidate(td, f);
+ if (ret < 0)
+ errval = -ret;
} else if (f->filetype == FIO_TYPE_FILE) {
+ dprint(FD_IO, "declare unneeded cache %s: %llu/%llu\n",
+ f->file_name, off, len);
ret = posix_fadvise(f->fd, off, len, POSIX_FADV_DONTNEED);
- } else if (f->filetype == FIO_TYPE_BD) {
+ if (ret)
+ errval = ret;
+ } else if (f->filetype == FIO_TYPE_BLOCK) {
+ int retry_count = 0;
+
+ dprint(FD_IO, "drop page cache %s\n", f->file_name);
ret = blockdev_invalidate_cache(f);
+ while (ret < 0 && errno == EAGAIN && retry_count++ < 25) {
+ /*
+ * Linux multipath devices reject ioctl while
+ * the maps are being updated. That window can
+ * last tens of milliseconds; we'll try up to
+ * a quarter of a second.
+ */
+ usleep(10000);
+ ret = blockdev_invalidate_cache(f);
+ }
if (ret < 0 && errno == EACCES && geteuid()) {
if (!root_warn) {
log_err("fio: only root may flush block "
}
ret = 0;
}
- } else if (f->filetype == FIO_TYPE_CHAR || f->filetype == FIO_TYPE_PIPE)
+ if (ret < 0)
+ errval = errno;
+ else if (ret) /* probably not supported */
+ errval = ret;
+ } else if (f->filetype == FIO_TYPE_CHAR ||
+ f->filetype == FIO_TYPE_PIPE) {
+ dprint(FD_IO, "invalidate not supported %s\n", f->file_name);
ret = 0;
-
- if (ret < 0) {
- td_verror(td, errno, "invalidate_cache");
- return 1;
- } else if (ret > 0) {
- td_verror(td, ret, "invalidate_cache");
- return 1;
}
- return ret;
+ /*
+ * Cache flushing isn't a fatal condition, and we know it will
+ * happen on some platforms where we don't have the proper
+ * function to flush eg block device caches. So just warn and
+ * continue on our way.
+ */
+ if (errval)
+ log_info("fio: cache invalidation of %s failed: %s\n",
+ f->file_name, strerror(errval));
+
+ return 0;
}
f->shadow_fd = -1;
}
+ f->engine_pos = 0;
return ret;
}
__f = lookup_file_hash(f->file_name);
if (__f) {
dprint(FD_FILE, "found file in hash %s\n", f->file_name);
- /*
- * racy, need the __f->lock locked
- */
f->lock = __f->lock;
from_hash = 1;
} else {
from_hash = 0;
}
+#ifdef WIN32
+ flags |= _O_BINARY;
+#endif
+
f->fd = open(f->file_name, flags, 0600);
return from_hash;
}
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");
goto skip_flags;
if (td->o.odirect)
flags |= OS_O_DIRECT;
+ if (td->o.oatomic) {
+ if (!FIO_O_ATOMIC) {
+ td_verror(td, EINVAL, "OS does not support atomic IO");
+ return 1;
+ }
+ flags |= OS_O_DIRECT | FIO_O_ATOMIC;
+ }
if (td->o.sync_io)
flags |= O_SYNC;
- if (td->o.create_on_open)
+ if (td->o.create_on_open && td->o.allow_create)
flags |= O_CREAT;
skip_flags:
if (f->filetype != FIO_TYPE_FILE)
if (!read_only)
flags |= O_RDWR;
- if (f->filetype == FIO_TYPE_FILE)
+ if (f->filetype == FIO_TYPE_FILE && td->o.allow_create)
flags |= O_CREAT;
if (is_std)
f->fd = dup(STDIN_FILENO);
else
from_hash = file_lookup_open(f, flags);
- } else { //td trim
+ } else if (td_trim(td)) {
+ assert(!td_rw(td)); /* should have matched above */
flags |= O_RDWR;
from_hash = file_lookup_open(f, flags);
}
}
td_verror(td, __e, buf);
+ return 1;
}
if (!from_hash && f->fd != -1) {
* 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
+ * That pollutes the device cache, which can slow down
* unbuffered accesses.
*/
if (f->shadow_fd == -1)
return 0;
}
+/*
+ * This function i.e. get_file_size() is the default .get_file_size
+ * implementation of majority of I/O engines.
+ */
int generic_get_file_size(struct thread_data *td, struct fio_file *f)
{
return get_file_size(td, f);
int err = 0;
for_each_file(td, f, i) {
- dprint(FD_FILE, "get file size for %p/%d/%p\n", f, i,
+ dprint(FD_FILE, "get file size for %p/%d/%s\n", f, i,
f->file_name);
if (td_io_get_file_size(td, f)) {
if (td->error != ENOENT) {
log_err("%s\n", td->verror);
err = 1;
+ break;
}
clear_error(td);
}
+ /*
+ * There are corner cases where we end up with -1 for
+ * ->real_file_size due to unsupported file type, etc.
+ * We then just set to size option value divided by number
+ * of files, similar to the way file ->io_size is set.
+ * stat(2) failure doesn't set ->real_file_size to -1.
+ */
if (f->real_file_size == -1ULL && td->o.size)
f->real_file_size = td->o.size / td->o.nr_files;
}
struct stat sb;
char buf[256];
- if (f->filetype == FIO_TYPE_BD || f->filetype == FIO_TYPE_CHAR) {
+ if (f->filetype == FIO_TYPE_BLOCK || 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);
+ buf[255] = '\0';
+ strncpy(buf, f->file_name, 255);
if (stat(buf, &sb) < 0) {
if (errno != ENOENT)
if (fm)
continue;
- fm = malloc(sizeof(*fm));
- strcpy(fm->__base, buf);
+ fm = calloc(1, sizeof(*fm));
+ strncpy(fm->__base, buf, sizeof(fm->__base) - 1);
fm->base = basename(fm->__base);
fm->key = sb.st_dev;
flist_add(&fm->list, &list);
fm = flist_entry(n, struct fio_mount, list);
flist_del(&fm->list);
- sz = get_fs_size(fm->base);
+ sz = get_fs_free_size(fm->base);
if (sz && sz != -1ULL)
ret += sz;
return ret;
}
-uint64_t get_start_offset(struct thread_data *td)
+uint64_t get_start_offset(struct thread_data *td, struct fio_file *f)
{
- return td->o.start_offset +
- (td->thread_number - 1) * td->o.offset_increment;
+ struct thread_options *o = &td->o;
+ unsigned long long align_bs;
+ unsigned long long offset;
+
+ if (o->file_append && f->filetype == FIO_TYPE_FILE)
+ return f->real_file_size;
+
+ if (o->start_offset_percent > 0) {
+ /*
+ * if blockalign is provided, find the min across read, write,
+ * and trim
+ */
+ if (fio_option_is_set(o, ba)) {
+ align_bs = (unsigned long long) min(o->ba[DDIR_READ], o->ba[DDIR_WRITE]);
+ align_bs = min((unsigned long long) o->ba[DDIR_TRIM], align_bs);
+ } else {
+ /* else take the minimum block size */
+ align_bs = td_min_bs(td);
+ }
+
+ /* calculate the raw offset */
+ offset = (f->real_file_size * o->start_offset_percent / 100) +
+ (td->subjob_number * o->offset_increment);
+
+ /*
+ * block align the offset at the next available boundary at
+ * ceiling(offset / align_bs) * align_bs
+ */
+ offset = (offset / align_bs + (offset % align_bs != 0)) * align_bs;
+
+ } else {
+ /* start_offset_percent not set */
+ offset = o->start_offset +
+ td->subjob_number * o->offset_increment;
+ }
+
+ return offset;
}
/*
unsigned long long total_size, extend_size;
struct thread_options *o = &td->o;
struct fio_file *f;
- unsigned int i;
+ unsigned int i, nr_fs_extra = 0;
int err = 0, need_extend;
int old_state;
+ const unsigned int bs = td_min_bs(td);
+ uint64_t fs = 0;
dprint(FD_FILE, "setup files\n");
- old_state = td->runstate;
- td_set_runstate(td, TD_SETTING_UP);
+ old_state = td_bump_runstate(td, TD_SETTING_UP);
if (o->read_iolog_file)
goto done;
/*
- * if ioengine defines a setup() method, it's responsible for
+ * Find out physical size of files or devices for this thread,
+ * before we determine I/O size and range of our targets.
+ * If ioengine defines a setup() method, it's responsible for
* opening the files and setting f->real_file_size to indicate
* the valid range for that file.
*/
*/
total_size = 0;
for_each_file(td, f, i) {
+ f->fileno = i;
if (f->real_file_size == -1ULL)
total_size = -1ULL;
else
* device/file sizes are zero and no size given, punt
*/
if ((!total_size || total_size == -1ULL) && !o->size &&
- !(td->io_ops->flags & FIO_NOIO) && !o->fill_device &&
+ !td_ioengine_flagged(td, FIO_NOIO) && !o->fill_device &&
!(o->nr_files && (o->file_size_low || o->file_size_high))) {
log_err("%s: you need to specify size=\n", o->name);
td_verror(td, EINVAL, "total_file_size");
goto err_out;
}
+ /*
+ * Calculate per-file size and potential extra size for the
+ * first files, if needed (i.e. if we don't have a fixed size).
+ */
+ if (!o->file_size_low && o->nr_files) {
+ uint64_t all_fs;
+
+ fs = o->size / o->nr_files;
+ all_fs = fs * o->nr_files;
+
+ if (all_fs < o->size)
+ nr_fs_extra = (o->size - all_fs) / bs;
+ }
+
/*
* now file sizes are known, so we can set ->io_size. if size= is
* not given, ->io_size is just equal to ->real_file_size. if size
extend_size = total_size = 0;
need_extend = 0;
for_each_file(td, f, i) {
- f->file_offset = get_start_offset(td);
+ f->file_offset = get_start_offset(td, f);
+ /*
+ * Update ->io_size depending on options specified.
+ * ->file_size_low being 0 means filesize option isn't set.
+ * Non zero ->file_size_low equals ->file_size_high means
+ * filesize option is set in a fixed size format.
+ * Non zero ->file_size_low not equals ->file_size_high means
+ * filesize option is set in a range format.
+ */
if (!o->file_size_low) {
/*
- * no file size range given, file size is equal to
- * total size divided by number of files. if that is
- * zero, set it to the real file size.
+ * no file size or range given, file size is equal to
+ * total size divided by number of files. If the size
+ * doesn't divide nicely with the min blocksize,
+ * make the first files bigger.
+ */
+ f->io_size = fs;
+ if (nr_fs_extra) {
+ nr_fs_extra--;
+ f->io_size += bs;
+ }
+
+ /*
+ * We normally don't come here for regular files, but
+ * if the result is 0 for a regular file, set it to the
+ * real file size. This could be size of the existing
+ * one if it already exists, but otherwise will be set
+ * to 0. A new file won't be created because
+ * ->io_size + ->file_offset equals ->real_file_size.
*/
- f->io_size = o->size / o->nr_files;
- if (!f->io_size)
+ if (!f->io_size) {
+ if (f->file_offset > f->real_file_size)
+ goto err_offset;
f->io_size = f->real_file_size - f->file_offset;
+ if (!f->io_size)
+ log_info("fio: file %s may be ignored\n",
+ f->file_name);
+ }
} else if (f->real_file_size < o->file_size_low ||
f->real_file_size > o->file_size_high) {
if (f->file_offset > o->file_size_low)
if (f->io_size == -1ULL)
total_size = -1ULL;
else {
- if (o->size_percent)
- f->io_size = (f->io_size * o->size_percent) / 100;
+ if (o->size_percent) {
+ uint64_t file_size;
+
+ file_size = f->io_size + f->file_offset;
+ f->io_size = (file_size *
+ o->size_percent) / 100;
+ if (f->io_size > (file_size - f->file_offset))
+ f->io_size = file_size - f->file_offset;
+
+ f->io_size -= (f->io_size % td_min_bs(td));
+ }
total_size += f->io_size;
}
if (f->filetype == FIO_TYPE_FILE &&
(f->io_size + f->file_offset) > f->real_file_size &&
- !(td->io_ops->flags & FIO_DISKLESSIO)) {
+ !td_ioengine_flagged(td, FIO_DISKLESSIO)) {
if (!o->create_on_open) {
need_extend++;
extend_size += (f->io_size + f->file_offset);
+ fio_file_set_extend(f);
} else
f->real_file_size = f->io_size + f->file_offset;
- fio_file_set_extend(f);
}
}
- if (!o->size || o->size > total_size)
+ if (td->o.block_error_hist) {
+ int len;
+
+ assert(td->o.nr_files == 1); /* checked in fixup_options */
+ f = td->files[0];
+ len = f->io_size / td->o.bs[DDIR_TRIM];
+ if (len > MAX_NR_BLOCK_INFOS || len <= 0) {
+ log_err("fio: cannot calculate block histogram with "
+ "%d trim blocks, maximum %d\n",
+ len, MAX_NR_BLOCK_INFOS);
+ td_verror(td, EINVAL, "block_error_hist");
+ goto err_out;
+ }
+
+ td->ts.nr_block_infos = len;
+ for (i = 0; i < len; i++)
+ td->ts.block_infos[i] =
+ BLOCK_INFO(0, BLOCK_STATE_UNINIT);
+ } else
+ td->ts.nr_block_infos = 0;
+
+ if (!o->size || (total_size && o->size > total_size))
o->size = total_size;
+ if (o->size < td_min_bs(td)) {
+ log_err("fio: blocksize too large for data set\n");
+ goto err_out;
+ }
+
/*
- * See if we need to extend some files
+ * See if we need to extend some files, typically needed when our
+ * target regular files don't exist yet, but our jobs require them
+ * initially due to read I/Os.
*/
if (need_extend) {
temp_stall_ts = 1;
- if (output_format == FIO_OUTPUT_NORMAL)
- log_info("%s: Laying out IO file(s) (%u file(s) /"
- " %lluMB)\n", o->name, need_extend,
- extend_size >> 20);
+ if (output_format & FIO_OUTPUT_NORMAL) {
+ log_info("%s: Laying out IO file%s (%u file%s / %s%lluMiB)\n",
+ o->name,
+ need_extend > 1 ? "s" : "",
+ need_extend,
+ need_extend > 1 ? "s" : "",
+ need_extend > 1 ? "total " : "",
+ extend_size >> 20);
+ }
for_each_file(td, f, i) {
unsigned long long old_len = -1ULL, extend_len = -1ULL;
err = __file_invalidate_cache(td, f, old_len,
extend_len);
- close(f->fd);
+
+ /*
+ * Shut up static checker
+ */
+ if (f->fd != -1)
+ close(f->fd);
+
f->fd = -1;
if (err)
break;
* iolog already set the total io size, if we read back
* stored entries.
*/
- if (!o->read_iolog_file)
- td->total_io_size = o->size * o->loops;
+ if (!o->read_iolog_file) {
+ if (o->io_size)
+ td->total_io_size = o->io_size * o->loops;
+ else
+ td->total_io_size = o->size * o->loops;
+ }
done:
if (o->create_only)
td->done = 1;
- td_set_runstate(td, old_state);
+ td_restore_runstate(td, old_state);
return 0;
err_offset:
log_err("%s: you need to specify valid offset=\n", o->name);
err_out:
- td_set_runstate(td, old_state);
+ td_restore_runstate(td, old_state);
return 1;
}
dprint(FD_FILE, "pre_read files\n");
for_each_file(td, f, i) {
- pre_read_file(td, f);
+ if (pre_read_file(td, f))
+ return -1;
}
- return 1;
+ return 0;
}
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;
+ uint64_t fsize;
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);
+ fsize = min(f->real_file_size, f->io_size);
- nranges = (file_size + range_size - 1) / range_size;
+ nranges = (fsize + range_size - 1) / range_size;
seed = jhash(f->file_name, strlen(f->file_name), 0) * td->thread_number;
if (!td->o.rand_repeatable)
if (td->o.random_distribution == FIO_RAND_DIST_ZIPF)
zipf_init(&f->zipf, nranges, td->o.zipf_theta.u.f, seed);
- else
+ else if (td->o.random_distribution == FIO_RAND_DIST_PARETO)
pareto_init(&f->zipf, nranges, td->o.pareto_h.u.f, seed);
+ else if (td->o.random_distribution == FIO_RAND_DIST_GAUSS)
+ gauss_init(&f->gauss, nranges, td->o.gauss_dev.u.f, seed);
return 1;
}
if (td->o.random_distribution == FIO_RAND_DIST_RANDOM)
return 0;
- state = td->runstate;
- td_set_runstate(td, TD_SETTING_UP);
+ state = td_bump_runstate(td, TD_SETTING_UP);
+
for_each_file(td, f, i)
__init_rand_distribution(td, f);
- td_set_runstate(td, state);
+
+ td_restore_runstate(td, state);
return 1;
}
+/*
+ * Check if the number of blocks exceeds the randomness capability of
+ * the selected generator. Tausworthe is 32-bit, the others are fullly
+ * 64-bit capable.
+ */
+static int check_rand_gen_limits(struct thread_data *td, struct fio_file *f,
+ uint64_t blocks)
+{
+ if (blocks <= FRAND32_MAX)
+ return 0;
+ if (td->o.random_generator != FIO_RAND_GEN_TAUSWORTHE)
+ return 0;
+
+ /*
+ * If the user hasn't specified a random generator, switch
+ * to tausworthe64 with informational warning. If the user did
+ * specify one, just warn.
+ */
+ log_info("fio: file %s exceeds 32-bit tausworthe random generator.\n",
+ f->file_name);
+
+ if (!fio_option_is_set(&td->o, random_generator)) {
+ log_info("fio: Switching to tausworthe64. Use the "
+ "random_generator= option to get rid of this "
+ "warning.\n");
+ td->o.random_generator = FIO_RAND_GEN_TAUSWORTHE64;
+ return 0;
+ }
+
+ /*
+ * Just make this information to avoid breaking scripts.
+ */
+ log_info("fio: Use the random_generator= option to switch to lfsr or "
+ "tausworthe64.\n");
+ return 0;
+}
+
int init_random_map(struct thread_data *td)
{
unsigned long long blocks;
return 0;
for_each_file(td, f, i) {
- uint64_t file_size = min(f->real_file_size, f->io_size);
+ uint64_t fsize = min(f->real_file_size, f->io_size);
- blocks = file_size / (unsigned long long) td->o.rw_min_bs;
+ blocks = fsize / (unsigned long long) td->o.rw_min_bs;
+
+ if (check_rand_gen_limits(td, f, blocks))
+ return 1;
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, seed & 0xF))
+
+ if (!lfsr_init(&f->lfsr, blocks, seed, 0)) {
+ fio_file_set_lfsr(f);
continue;
+ }
} else if (!td->o.norandommap) {
f->io_axmap = axmap_new(blocks);
- if (f->io_axmap)
+ if (f->io_axmap) {
+ fio_file_set_axmap(f);
continue;
+ }
} else if (td->o.norandommap)
continue;
for_each_file(td, f, i) {
if (td->o.unlink && f->filetype == FIO_TYPE_FILE) {
dprint(FD_FILE, "free unlink %s\n", f->file_name);
- unlink(f->file_name);
+ td_io_unlink_file(td, f);
}
if (fio_file_open(f))
remove_file_hash(f);
+ if (td->o.unlink && f->filetype == FIO_TYPE_FILE) {
+ dprint(FD_FILE, "free unlink %s\n", f->file_name);
+ td_io_unlink_file(td, f);
+ }
+
sfree(f->file_name);
f->file_name = NULL;
- axmap_free(f->io_axmap);
- f->io_axmap = NULL;
+ if (fio_file_axmap(f)) {
+ axmap_free(f->io_axmap);
+ f->io_axmap = NULL;
+ }
sfree(f);
}
td->files_index = 0;
td->files = NULL;
td->file_locks = NULL;
+ td->o.file_lock_mode = FILE_LOCK_NONE;
td->o.nr_files = 0;
}
else
f->filetype = FIO_TYPE_FILE;
+#ifdef WIN32
/* \\.\ 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;
+ f->filetype = FIO_TYPE_BLOCK;
+#endif
if (!stat(f->file_name, &sb)) {
if (S_ISBLK(sb.st_mode))
- f->filetype = FIO_TYPE_BD;
+ f->filetype = FIO_TYPE_BLOCK;
else if (S_ISCHR(sb.st_mode))
f->filetype = FIO_TYPE_CHAR;
else if (S_ISFIFO(sb.st_mode))
}
}
-int add_file(struct thread_data *td, const char *fname)
+static bool __is_already_allocated(const char *fname, bool set)
{
- int cur_files = td->files_index;
- char file_name[PATH_MAX];
- struct fio_file *f;
- int len = 0;
+ struct flist_head *entry;
+ bool ret;
- dprint(FD_FILE, "add file %s\n", fname);
+ ret = file_bloom_exists(fname, set);
+ if (!ret)
+ return ret;
+
+ flist_for_each(entry, &filename_list) {
+ struct file_name *fn;
+
+ fn = flist_entry(entry, struct file_name, list);
+
+ if (!strcmp(fn->filename, fname))
+ return true;
+ }
+
+ return false;
+}
+
+static bool is_already_allocated(const char *fname)
+{
+ bool ret;
+
+ fio_file_hash_lock();
+ ret = __is_already_allocated(fname, false);
+ fio_file_hash_unlock();
+
+ return ret;
+}
+
+static void set_already_allocated(const char *fname)
+{
+ struct file_name *fn;
+
+ fn = malloc(sizeof(struct file_name));
+ fn->filename = strdup(fname);
+
+ fio_file_hash_lock();
+ if (!__is_already_allocated(fname, true)) {
+ flist_add_tail(&fn->list, &filename_list);
+ fn = NULL;
+ }
+ fio_file_hash_unlock();
+
+ if (fn) {
+ free(fn->filename);
+ free(fn);
+ }
+}
+
+static void free_already_allocated(void)
+{
+ struct flist_head *entry, *tmp;
+ struct file_name *fn;
+
+ if (flist_empty(&filename_list))
+ return;
+
+ fio_file_hash_lock();
+ flist_for_each_safe(entry, tmp, &filename_list) {
+ fn = flist_entry(entry, struct file_name, list);
+ free(fn->filename);
+ flist_del(&fn->list);
+ free(fn);
+ }
+
+ fio_file_hash_unlock();
+}
+
+static struct fio_file *alloc_new_file(struct thread_data *td)
+{
+ struct fio_file *f;
f = smalloc(sizeof(*f));
if (!f) {
- log_err("fio: smalloc OOM\n");
assert(0);
+ return NULL;
}
f->fd = -1;
f->shadow_fd = -1;
fio_file_reset(td, f);
+ return f;
+}
+
+bool exists_and_not_regfile(const char *filename)
+{
+ struct stat sb;
+
+ if (lstat(filename, &sb) == -1)
+ return false;
+
+#ifndef WIN32 /* NOT Windows */
+ if (S_ISREG(sb.st_mode))
+ return false;
+#else
+ /* \\.\ is the device namespace in Windows, where every file
+ * is a device node */
+ if (S_ISREG(sb.st_mode) && strncmp(filename, "\\\\.\\", 4) != 0)
+ return false;
+#endif
+
+ return true;
+}
+
+int add_file(struct thread_data *td, const char *fname, int numjob, int inc)
+{
+ int cur_files = td->files_index;
+ char file_name[PATH_MAX];
+ struct fio_file *f;
+ int len = 0;
+
+ dprint(FD_FILE, "add file %s\n", fname);
+
+ if (td->o.directory)
+ len = set_name_idx(file_name, PATH_MAX, td->o.directory, numjob,
+ td->o.unique_filename);
+
+ sprintf(file_name + len, "%s", fname);
+
+ /* clean cloned siblings using existing files */
+ if (numjob && is_already_allocated(file_name) &&
+ !exists_and_not_regfile(fname))
+ return 0;
+
+ f = alloc_new_file(td);
if (td->files_size <= td->files_index) {
unsigned int new_size = td->o.nr_files + 1;
/*
* init function, io engine may not be loaded yet
*/
- if (td->io_ops && (td->io_ops->flags & FIO_DISKLESSIO))
+ if (td->io_ops && td_ioengine_flagged(td, FIO_DISKLESSIO))
f->real_file_size = -1ULL;
- if (td->o.directory)
- len = sprintf(file_name, "%s/", td->o.directory);
-
- sprintf(file_name + len, "%s", fname);
f->file_name = smalloc_strdup(file_name);
- if (!f->file_name) {
- log_err("fio: smalloc OOM\n");
+ if (!f->file_name)
assert(0);
- }
get_file_type(f);
if (f->filetype == FIO_TYPE_FILE)
td->nr_normal_files++;
+ set_already_allocated(file_name);
+
+ if (inc)
+ td->o.nr_files++;
+
dprint(FD_FILE, "file %p \"%s\" added at %d\n", f, f->file_name,
cur_files);
return i;
}
- return add_file(td, fname);
+ return add_file(td, fname, 0, 1);
}
void get_file(struct fio_file *f)
if (--f->references)
return 0;
- if (should_fsync(td) && td->o.fsync_on_close)
+ if (should_fsync(td) && td->o.fsync_on_close) {
f_ret = fsync(f->fd);
+ if (f_ret < 0)
+ f_ret = errno;
+ }
if (td->io_ops->close_file)
ret = td->io_ops->close_file(td, f);
void unlock_file_all(struct thread_data *td, struct fio_file *f)
{
+ if (td->o.file_lock_mode == FILE_LOCK_NONE || !td->file_locks)
+ return;
if (td->file_locks[f->fileno] != FILE_LOCK_NONE)
unlock_file(td, f);
}
if (lstat(full_path, &sb) == -1) {
if (errno != ENOENT) {
td_verror(td, errno, "stat");
- return 1;
+ ret = 1;
+ break;
}
}
if (S_ISREG(sb.st_mode)) {
- add_file(td, full_path);
- td->o.nr_files++;
+ add_file(td, full_path, 0, 1);
continue;
}
if (!S_ISDIR(sb.st_mode))
for_each_file(org, f, i) {
struct fio_file *__f;
- __f = smalloc(sizeof(*__f));
- if (!__f) {
- log_err("fio: smalloc OOM\n");
- assert(0);
- }
- __f->fd = -1;
- fio_file_reset(td, __f);
+ __f = alloc_new_file(td);
if (f->file_name) {
__f->file_name = smalloc_strdup(f->file_name);
- if (!__f->file_name) {
- log_err("fio: smalloc OOM\n");
+ if (!__f->file_name)
assert(0);
- }
__f->filetype = f->filetype;
}
+ if (td->o.file_lock_mode == FILE_LOCK_EXCLUSIVE)
+ __f->lock = f->lock;
+ else if (td->o.file_lock_mode == FILE_LOCK_READWRITE)
+ __f->rwlock = f->rwlock;
+
td->files[i] = __f;
}
}
void free_release_files(struct thread_data *td)
{
close_files(td);
+ td->o.nr_files = 0;
+ td->o.open_files = 0;
td->files_index = 0;
td->nr_normal_files = 0;
}
void fio_file_reset(struct thread_data *td, struct fio_file *f)
{
- f->last_pos = f->file_offset;
- f->last_start = -1ULL;
- if (f->io_axmap)
+ int i;
+
+ for (i = 0; i < DDIR_RWDIR_CNT; i++) {
+ f->last_pos[i] = f->file_offset;
+ f->last_start[i] = -1ULL;
+ }
+
+ if (fio_file_axmap(f))
axmap_reset(f->io_axmap);
- if (td->o.random_generator == FIO_RAND_GEN_LFSR)
+ else if (fio_file_lfsr(f))
lfsr_reset(&f->lfsr, td->rand_seeds[FIO_RAND_BLOCK_OFF]);
}
+
+bool fio_files_done(struct thread_data *td)
+{
+ struct fio_file *f;
+ unsigned int i;
+
+ for_each_file(td, f, i)
+ if (!fio_file_done(f))
+ return false;
+
+ return true;
+}
+
+/* free memory used in initialization phase only */
+void filesetup_mem_free(void)
+{
+ free_already_allocated();
+}