int nr; /* input */
int error; /* output */
- unsigned long bytes_done[2]; /* output */
+ unsigned long bytes_done[DDIR_RWDIR_CNT]; /* output */
struct timeval time; /* output */
};
dprint(FD_RANDOM, "free: b=%llu, idx=%u, bit=%u\n", block, idx, bit);
- return (f->file_map[idx] & (1 << bit)) == 0;
+ return (f->file_map[idx] & (1UL << bit)) == 0;
}
/*
busy_check = !(io_u->flags & IO_U_F_BUSY_OK);
while (nr_blocks) {
- unsigned int this_blocks, mask;
unsigned int idx, bit;
+ unsigned long mask, this_blocks;
/*
* If we have a mixed random workload, we may
do {
if (this_blocks == BLOCKS_PER_MAP)
- mask = -1U;
+ mask = -1UL;
else
- mask = ((1U << this_blocks) - 1) << bit;
-
+ mask = ((1UL << this_blocks) - 1) << bit;
+
if (!(f->file_map[idx] & mask))
break;
if (max_size > f->real_file_size)
max_size = f->real_file_size;
+ if (td->o.zone_range)
+ max_size = td->o.zone_range;
+
max_blocks = max_size / (unsigned long long) td->o.ba[ddir];
if (!max_blocks)
return 0;
static int get_next_free_block(struct thread_data *td, struct fio_file *f,
enum fio_ddir ddir, unsigned long long *b)
{
- unsigned long long min_bs = td->o.rw_min_bs, lastb;
+ unsigned long long block, min_bs = td->o.rw_min_bs, lastb;
int i;
lastb = last_block(td, f, ddir);
return 1;
i = f->last_free_lookup;
- *b = (i * BLOCKS_PER_MAP);
- while ((*b) * min_bs < f->real_file_size &&
- (*b) * min_bs < f->io_size) {
- if (f->file_map[i] != (unsigned int) -1) {
- *b += ffz(f->file_map[i]);
- if (*b > lastb)
+ block = i * BLOCKS_PER_MAP;
+ while (block * min_bs < f->real_file_size &&
+ block * min_bs < f->io_size) {
+ if (f->file_map[i] != -1UL) {
+ block += ffz(f->file_map[i]);
+ if (block > lastb)
break;
f->last_free_lookup = i;
+ *b = block;
return 0;
}
- *b += BLOCKS_PER_MAP;
+ block += BLOCKS_PER_MAP;
i++;
}
static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
enum fio_ddir ddir, unsigned long long *b)
{
- unsigned long long r, lastb;
+ unsigned long long rmax, r, lastb;
int loops = 5;
lastb = last_block(td, f, ddir);
if (!lastb)
return 1;
+ if (f->failed_rands >= 200)
+ goto ffz;
+
+ rmax = td->o.use_os_rand ? OS_RAND_MAX : FRAND_MAX;
do {
- r = os_random_long(&td->random_state);
+ if (td->o.use_os_rand)
+ r = os_random_long(&td->random_state);
+ else
+ r = __rand(&td->__random_state);
+
+ *b = (lastb - 1) * (r / ((unsigned long long) rmax + 1.0));
+
dprint(FD_RANDOM, "off rand %llu\n", r);
- *b = (lastb - 1) * (r / ((unsigned long long) OS_RAND_MAX + 1.0));
+
/*
* if we are not maintaining a random map, we are done.
*/
if (!file_randommap(td, f))
- return 0;
+ goto ret_good;
/*
* calculate map offset and check if it's free
*/
if (random_map_free(f, *b))
- return 0;
+ goto ret_good;
dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n",
*b);
} while (--loops);
+ if (!f->failed_rands++)
+ f->last_free_lookup = 0;
+
/*
* we get here, if we didn't suceed in looking up a block. generate
* a random start offset into the filemap, and find the first free
loops = 10;
do {
f->last_free_lookup = (f->num_maps - 1) *
- (r / (OS_RAND_MAX + 1.0));
+ (r / ((unsigned long long) rmax + 1.0));
if (!get_next_free_block(td, f, ddir, b))
- return 0;
+ goto ret;
- r = os_random_long(&td->random_state);
+ if (td->o.use_os_rand)
+ r = os_random_long(&td->random_state);
+ else
+ r = __rand(&td->__random_state);
} while (--loops);
/*
* that didn't work either, try exhaustive search from the start
*/
f->last_free_lookup = 0;
+ffz:
+ if (!get_next_free_block(td, f, ddir, b))
+ return 0;
+ f->last_free_lookup = 0;
return get_next_free_block(td, f, ddir, b);
+ret_good:
+ f->failed_rands = 0;
+ret:
+ return 0;
}
static int get_next_rand_block(struct thread_data *td, struct fio_file *f,
enum fio_ddir ddir, unsigned long long *b)
{
- if (get_next_rand_offset(td, f, ddir, b)) {
- dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n",
- f->file_name, f->last_pos, f->real_file_size);
- return 1;
+ if (!get_next_rand_offset(td, f, ddir, b))
+ return 0;
+
+ if (td->o.time_based) {
+ fio_file_reset(f);
+ if (!get_next_rand_offset(td, f, ddir, b))
+ return 0;
}
- return 0;
+ dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n",
+ f->file_name, f->last_pos, f->real_file_size);
+ return 1;
}
-static int get_next_seq_block(struct thread_data *td, struct fio_file *f,
- enum fio_ddir ddir, unsigned long long *b)
+static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
+ enum fio_ddir ddir, unsigned long long *offset)
{
assert(ddir_rw(ddir));
+ if (f->last_pos >= f->io_size + get_start_offset(td) && td->o.time_based)
+ f->last_pos = f->last_pos - f->io_size;
+
if (f->last_pos < f->real_file_size) {
- *b = (f->last_pos - f->file_offset) / td->o.min_bs[ddir];
+ unsigned long long pos;
+
+ if (f->last_pos == f->file_offset && td->o.ddir_seq_add < 0)
+ f->last_pos = f->real_file_size;
+
+ pos = f->last_pos - f->file_offset;
+ if (pos)
+ pos += td->o.ddir_seq_add;
+
+ *offset = pos;
return 0;
}
}
static int get_next_block(struct thread_data *td, struct io_u *io_u,
- enum fio_ddir ddir, int rw_seq, unsigned long long *b)
+ enum fio_ddir ddir, int rw_seq)
{
struct fio_file *f = io_u->file;
+ unsigned long long b, offset;
int ret;
assert(ddir_rw(ddir));
+ b = offset = -1ULL;
+
if (rw_seq) {
if (td_random(td))
- ret = get_next_rand_block(td, f, ddir, b);
+ ret = get_next_rand_block(td, f, ddir, &b);
else
- ret = get_next_seq_block(td, f, ddir, b);
+ ret = get_next_seq_offset(td, f, ddir, &offset);
} else {
io_u->flags |= IO_U_F_BUSY_OK;
if (td->o.rw_seq == RW_SEQ_SEQ) {
- ret = get_next_seq_block(td, f, ddir, b);
+ ret = get_next_seq_offset(td, f, ddir, &offset);
if (ret)
- ret = get_next_rand_block(td, f, ddir, b);
+ ret = get_next_rand_block(td, f, ddir, &b);
} else if (td->o.rw_seq == RW_SEQ_IDENT) {
if (f->last_start != -1ULL)
- *b = (f->last_start - f->file_offset)
- / td->o.min_bs[ddir];
+ offset = f->last_start - f->file_offset;
else
- *b = 0;
+ offset = 0;
ret = 0;
} else {
log_err("fio: unknown rw_seq=%d\n", td->o.rw_seq);
ret = 1;
}
}
-
+
+ if (!ret) {
+ if (offset != -1ULL)
+ io_u->offset = offset;
+ else if (b != -1ULL)
+ io_u->offset = b * td->o.ba[ddir];
+ else {
+ log_err("fio: bug in offset generation\n");
+ ret = 1;
+ }
+ }
+
return ret;
}
static int __get_next_offset(struct thread_data *td, struct io_u *io_u)
{
struct fio_file *f = io_u->file;
- unsigned long long b;
enum fio_ddir ddir = io_u->ddir;
int rw_seq_hit = 0;
td->ddir_seq_nr = td->o.ddir_seq_nr;
}
- if (get_next_block(td, io_u, ddir, rw_seq_hit, &b))
+ if (get_next_block(td, io_u, ddir, rw_seq_hit))
return 1;
- io_u->offset = b * td->o.ba[ddir];
if (io_u->offset >= f->io_size) {
dprint(FD_IO, "get_next_offset: offset %llu >= io_size %llu\n",
io_u->offset, f->io_size);
return __get_next_offset(td, io_u);
}
+static inline int io_u_fits(struct thread_data *td, struct io_u *io_u,
+ unsigned int buflen)
+{
+ struct fio_file *f = io_u->file;
+
+ return io_u->offset + buflen <= f->io_size + get_start_offset(td);
+}
+
static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u)
{
const int ddir = io_u->ddir;
unsigned int uninitialized_var(buflen);
unsigned int minbs, maxbs;
- long r;
+ unsigned long r, rand_max;
assert(ddir_rw(ddir));
maxbs = td->o.max_bs[ddir];
if (minbs == maxbs)
- buflen = minbs;
- else {
- r = os_random_long(&td->bsrange_state);
+ return minbs;
+
+ /*
+ * If we can't satisfy the min block size from here, then fail
+ */
+ if (!io_u_fits(td, io_u, minbs))
+ return 0;
+
+ if (td->o.use_os_rand)
+ rand_max = OS_RAND_MAX;
+ else
+ rand_max = FRAND_MAX;
+
+ do {
+ if (td->o.use_os_rand)
+ r = os_random_long(&td->bsrange_state);
+ else
+ r = __rand(&td->__bsrange_state);
+
if (!td->o.bssplit_nr[ddir]) {
buflen = 1 + (unsigned int) ((double) maxbs *
- (r / (OS_RAND_MAX + 1.0)));
+ (r / (rand_max + 1.0)));
if (buflen < minbs)
buflen = minbs;
} else {
buflen = bsp->bs;
perc += bsp->perc;
- if (r <= ((OS_RAND_MAX / 100L) * perc))
+ if ((r <= ((rand_max / 100L) * perc)) &&
+ io_u_fits(td, io_u, buflen))
break;
}
}
+
if (!td->o.bs_unaligned && is_power_of_2(minbs))
buflen = (buflen + minbs - 1) & ~(minbs - 1);
- }
- if (io_u->offset + buflen > io_u->file->real_file_size) {
- dprint(FD_IO, "lower buflen %u -> %u (ddir=%d)\n", buflen,
- minbs, ddir);
- buflen = minbs;
- }
+ } while (!io_u_fits(td, io_u, buflen));
return buflen;
}
static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
{
unsigned int v;
- long r;
+ unsigned long r;
+
+ if (td->o.use_os_rand) {
+ r = os_random_long(&td->rwmix_state);
+ v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0)));
+ } else {
+ r = __rand(&td->__rwmix_state);
+ v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0)));
+ }
- r = os_random_long(&td->rwmix_state);
- v = 1 + (int) (100.0 * (r / (OS_RAND_MAX + 1.0)));
if (v <= td->o.rwmix[DDIR_READ])
return DDIR_READ;
} else
usec = td->rate_pending_usleep[ddir];
+ /*
+ * We are going to sleep, ensure that we flush anything pending as
+ * not to skew our latency numbers.
+ *
+ * Changed to only monitor 'in flight' requests here instead of the
+ * td->cur_depth, b/c td->cur_depth does not accurately represent
+ * io's that have been actually submitted to an async engine,
+ * and cur_depth is meaningless for sync engines.
+ */
+ if (td->io_u_in_flight) {
+ int fio_unused ret;
+
+ ret = io_u_queued_complete(td, td->io_u_in_flight, NULL);
+ }
+
fio_gettime(&t, NULL);
usec_sleep(td, usec);
usec = utime_since_now(&t);
if (td_rw(td) && __should_check_rate(td, odir))
td->rate_pending_usleep[odir] -= usec;
+ if (ddir_trim(ddir))
+ return ddir;
return ddir;
}
ddir = td->rwmix_ddir;
} else if (td_read(td))
ddir = DDIR_READ;
- else
+ else if (td_write(td))
ddir = DDIR_WRITE;
+ else
+ ddir = DDIR_TRIM;
td->rwmix_ddir = rate_ddir(td, ddir);
return td->rwmix_ddir;
{
td_io_u_lock(td);
- io_u->flags |= IO_U_F_FREE;
- io_u->flags &= ~IO_U_F_FREE_DEF;
-
- if (io_u->file)
+ if (io_u->file && !(io_u->flags & IO_U_F_FREE_DEF))
put_file_log(td, io_u->file);
-
io_u->file = NULL;
+ io_u->flags &= ~IO_U_F_FREE_DEF;
+ io_u->flags |= IO_U_F_FREE;
+
if (io_u->flags & IO_U_F_IN_CUR_DEPTH)
td->cur_depth--;
flist_del_init(&io_u->list);
/*
* See if it's time to switch to a new zone
*/
- if (td->zone_bytes >= td->o.zone_size) {
+ if (td->zone_bytes >= td->o.zone_size && td->o.zone_skip) {
td->zone_bytes = 0;
- io_u->file->last_pos += td->o.zone_skip;
+ io_u->file->file_offset += td->o.zone_range + td->o.zone_skip;
+ io_u->file->last_pos = io_u->file->file_offset;
td->io_skip_bytes += td->o.zone_skip;
}
int fno;
do {
- long r = os_random_long(&td->next_file_state);
int opened = 0;
+ unsigned long r;
+
+ if (td->o.use_os_rand) {
+ r = os_random_long(&td->next_file_state);
+ fno = (unsigned int) ((double) td->o.nr_files
+ * (r / (OS_RAND_MAX + 1.0)));
+ } else {
+ r = __rand(&td->__next_file_state);
+ fno = (unsigned int) ((double) td->o.nr_files
+ * (r / (FRAND_MAX + 1.0)));
+ }
- fno = (unsigned int) ((double) td->o.nr_files
- * (r / (OS_RAND_MAX + 1.0)));
f = td->files[fno];
if (fio_file_done(f))
continue;
assert(io_u->flags & IO_U_F_FREE);
io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF);
io_u->flags &= ~(IO_U_F_TRIMMED | IO_U_F_BARRIER);
+ io_u->flags &= ~IO_U_F_VER_LIST;
io_u->error = 0;
flist_del(&io_u->list);
if (td->o.verify_backlog && td->io_hist_len) {
int get_verify = 0;
- if (td->verify_batch) {
- td->verify_batch--;
+ if (td->verify_batch)
get_verify = 1;
- } else if (!(td->io_hist_len % td->o.verify_backlog) &&
+ else if (!(td->io_hist_len % td->o.verify_backlog) &&
td->last_ddir != DDIR_READ) {
td->verify_batch = td->o.verify_batch;
if (!td->verify_batch)
get_verify = 1;
}
- if (get_verify && !get_next_verify(td, io_u))
+ if (get_verify && !get_next_verify(td, io_u)) {
+ td->verify_batch--;
return 1;
+ }
}
return 0;
}
+/*
+ * Fill offset and start time into the buffer content, to prevent too
+ * easy compressible data for simple de-dupe attempts. Do this for every
+ * 512b block in the range, since that should be the smallest block size
+ * we can expect from a device.
+ */
+static void small_content_scramble(struct io_u *io_u)
+{
+ unsigned int i, nr_blocks = io_u->buflen / 512;
+ unsigned long long boffset;
+ unsigned int offset;
+ void *p, *end;
+
+ if (!nr_blocks)
+ return;
+
+ p = io_u->xfer_buf;
+ boffset = io_u->offset;
+ io_u->buf_filled_len = 0;
+
+ for (i = 0; i < nr_blocks; i++) {
+ /*
+ * Fill the byte offset into a "random" start offset of
+ * the buffer, given by the product of the usec time
+ * and the actual offset.
+ */
+ offset = (io_u->start_time.tv_usec ^ boffset) & 511;
+ offset &= ~(sizeof(unsigned long long) - 1);
+ if (offset >= 512 - sizeof(unsigned long long))
+ offset -= sizeof(unsigned long long);
+ memcpy(p + offset, &boffset, sizeof(boffset));
+
+ end = p + 512 - sizeof(io_u->start_time);
+ memcpy(end, &io_u->start_time, sizeof(io_u->start_time));
+ p += 512;
+ boffset += 512;
+ }
+}
+
/*
* Return an io_u to be processed. Gets a buflen and offset, sets direction,
* etc. The returned io_u is fully ready to be prepped and submitted.
{
struct fio_file *f;
struct io_u *io_u;
+ int do_scramble = 0;
io_u = __get_io_u(td);
if (!io_u) {
f->last_start = io_u->offset;
f->last_pos = io_u->offset + io_u->buflen;
- if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_WRITE)
- populate_verify_io_u(td, io_u);
- else if (td->o.refill_buffers && io_u->ddir == DDIR_WRITE)
- io_u_fill_buffer(td, io_u, io_u->xfer_buflen);
- else if (io_u->ddir == DDIR_READ) {
+ if (io_u->ddir == DDIR_WRITE) {
+ if (td->o.refill_buffers) {
+ io_u_fill_buffer(td, io_u,
+ io_u->xfer_buflen, io_u->xfer_buflen);
+ } else if (td->o.scramble_buffers)
+ do_scramble = 1;
+ if (td->o.verify != VERIFY_NONE) {
+ populate_verify_io_u(td, io_u);
+ do_scramble = 0;
+ }
+ } else if (io_u->ddir == DDIR_READ) {
/*
* Reset the buf_filled parameters so next time if the
* buffer is used for writes it is refilled.
if (!td_io_prep(td, io_u)) {
if (!td->o.disable_slat)
fio_gettime(&io_u->start_time, NULL);
+ if (do_scramble)
+ small_content_scramble(io_u);
return io_u;
}
err_put:
void io_u_log_error(struct thread_data *td, struct io_u *io_u)
{
+ enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error);
const char *msg[] = { "read", "write", "sync", "datasync",
"sync_file_range", "wait", "trim" };
-
+ if (td_non_fatal_error(td, eb, io_u->error) && !td->o.error_dump)
+ return;
log_err("fio: io_u error");
td_verror(td, io_u->error, "io_u error");
}
+static void account_io_completion(struct thread_data *td, struct io_u *io_u,
+ struct io_completion_data *icd,
+ const enum fio_ddir idx, unsigned int bytes)
+{
+ unsigned long uninitialized_var(lusec);
+
+ if (!td->o.disable_clat || !td->o.disable_bw)
+ lusec = utime_since(&io_u->issue_time, &icd->time);
+
+ if (!td->o.disable_lat) {
+ unsigned long tusec;
+
+ tusec = utime_since(&io_u->start_time, &icd->time);
+ add_lat_sample(td, idx, tusec, bytes);
+ }
+
+ if (!td->o.disable_clat) {
+ add_clat_sample(td, idx, lusec, bytes);
+ io_u_mark_latency(td, lusec);
+ }
+
+ if (!td->o.disable_bw)
+ add_bw_sample(td, idx, bytes, &icd->time);
+
+ add_iops_sample(td, idx, &icd->time);
+}
+
+static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir)
+{
+ unsigned long long secs, remainder, bps, bytes;
+ bytes = td->this_io_bytes[ddir];
+ bps = td->rate_bps[ddir];
+ secs = bytes / bps;
+ remainder = bytes % bps;
+ return remainder * 1000000 / bps + secs * 1000000;
+}
+
static void io_completed(struct thread_data *td, struct io_u *io_u,
struct io_completion_data *icd)
{
int ret;
td->io_blocks[idx]++;
+ td->this_io_blocks[idx]++;
td->io_bytes[idx] += bytes;
- td->this_io_bytes[idx] += bytes;
+
+ if (!(io_u->flags & IO_U_F_VER_LIST))
+ td->this_io_bytes[idx] += bytes;
if (idx == DDIR_WRITE) {
f = io_u->file;
}
}
- if (ramp_time_over(td)) {
- unsigned long uninitialized_var(lusec);
-
- if (!td->o.disable_clat || !td->o.disable_bw)
- lusec = utime_since(&io_u->issue_time,
- &icd->time);
- if (!td->o.disable_lat) {
- unsigned long tusec;
+ if (ramp_time_over(td) && (td->runstate == TD_RUNNING ||
+ td->runstate == TD_VERIFYING)) {
+ account_io_completion(td, io_u, icd, idx, bytes);
- tusec = utime_since(&io_u->start_time,
- &icd->time);
- add_lat_sample(td, idx, tusec, bytes);
- }
- if (!td->o.disable_clat) {
- add_clat_sample(td, idx, lusec, bytes);
- io_u_mark_latency(td, lusec);
- }
- if (!td->o.disable_bw)
- add_bw_sample(td, idx, bytes, &icd->time);
if (__should_check_rate(td, idx)) {
td->rate_pending_usleep[idx] =
- ((td->this_io_bytes[idx] *
- td->rate_nsec_cycle[idx]) / 1000 -
+ (usec_for_io(td, idx) -
utime_since_now(&td->start));
}
- if (__should_check_rate(td, idx ^ 1))
+ if (idx != DDIR_TRIM && __should_check_rate(td, odx))
td->rate_pending_usleep[odx] =
- ((td->this_io_bytes[odx] *
- td->rate_nsec_cycle[odx]) / 1000 -
+ (usec_for_io(td, odx) -
utime_since_now(&td->start));
}
icd->error = io_u->error;
io_u_log_error(td, io_u);
}
- if (td->o.continue_on_error && icd->error &&
- td_non_fatal_error(icd->error)) {
+ if (icd->error) {
+ enum error_type_bit eb = td_error_type(io_u->ddir, icd->error);
+ if (!td_non_fatal_error(td, eb, icd->error))
+ return;
/*
* If there is a non_fatal error, then add to the error count
* and clear all the errors.
static void init_icd(struct thread_data *td, struct io_completion_data *icd,
int nr)
{
+ int ddir;
if (!td->o.disable_clat || !td->o.disable_bw)
fio_gettime(&icd->time, NULL);
icd->nr = nr;
icd->error = 0;
- icd->bytes_done[0] = icd->bytes_done[1] = 0;
+ for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+ icd->bytes_done[ddir] = 0;
}
static void ios_completed(struct thread_data *td,
}
if (bytes) {
- bytes[0] += icd.bytes_done[0];
- bytes[1] += icd.bytes_done[1];
+ int ddir;
+
+ for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+ bytes[ddir] += icd.bytes_done[ddir];
}
return 0;
}
if (bytes) {
- bytes[0] += icd.bytes_done[0];
- bytes[1] += icd.bytes_done[1];
+ int ddir;
+
+ for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+ bytes[ddir] += icd.bytes_done[ddir];
}
return 0;
* "randomly" fill the buffer contents
*/
void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u,
- unsigned int max_bs)
+ unsigned int min_write, unsigned int max_bs)
{
io_u->buf_filled_len = 0;
- if (!td->o.zero_buffers)
- fill_random_buf(io_u->buf, max_bs);
- else
+ if (!td->o.zero_buffers) {
+ unsigned int perc = td->o.compress_percentage;
+
+ if (perc) {
+ unsigned int seg = min_write;
+
+ seg = min(min_write, td->o.compress_chunk);
+ fill_random_buf_percentage(&td->buf_state, io_u->buf,
+ perc, seg, max_bs);
+ } else
+ fill_random_buf(&td->buf_state, io_u->buf, max_bs);
+ } else
memset(io_u->buf, 0, max_bs);
}