struct io_completion_data {
int nr; /* input */
+ int account; /* input */
int error; /* output */
unsigned long bytes_done[2]; /* 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;
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,
assert(ddir_rw(ddir));
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;
+
+ *b = pos / td->o.min_bs[ddir];
return 0;
}
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 + td->o.start_offset;
+}
+
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 (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
+ */
+ if (td->cur_depth) {
+ int fio_unused ret;
+
+ ret = io_u_queued_complete(td, td->cur_depth, NULL);
+ }
+
fio_gettime(&t, NULL);
usec_sleep(td, usec);
usec = utime_since_now(&t);
{
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);
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;
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;
+
+ 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.verify != VERIFY_NONE)
+ populate_verify_io_u(td, io_u);
+ else if (td->o.refill_buffers)
+ io_u_fill_buffer(td, io_u, io_u->xfer_buflen);
+ else if (td->o.scramble_buffers)
+ do_scramble = 1;
+ } 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:
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 (!icd->account)
+ return;
+
+ 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 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 (ramp_time_over(td)) {
- unsigned long uninitialized_var(lusec);
+ if (ramp_time_over(td) && td->runstate == TD_RUNNING) {
+ account_io_completion(td, io_u, icd, idx, bytes);
- 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);
if (__should_check_rate(td, idx)) {
td->rate_pending_usleep[idx] =
((td->this_io_bytes[idx] *
fio_gettime(&icd->time, NULL);
icd->nr = nr;
+ icd->account = 1;
icd->error = 0;
icd->bytes_done[0] = icd->bytes_done[1] = 0;
if (!(io_u->flags & IO_U_F_FREE_DEF))
put_io_u(td, io_u);
+
+ icd->account = 0;
}
}
io_u->buf_filled_len = 0;
if (!td->o.zero_buffers)
- fill_random_buf(io_u->buf, max_bs);
+ fill_random_buf(&td->buf_state, io_u->buf, max_bs);
else
memset(io_u->buf, 0, max_bs);
}
projects / fio.git / blobdiff