#include "server.h"
#include "lib/getrusage.h"
#include "idletime.h"
+#include "err.h"
+#include "lib/tp.h"
+
+static pthread_t helper_thread;
+static pthread_mutex_t helper_lock;
+pthread_cond_t helper_cond;
+int helper_do_stat = 0;
-static pthread_t disk_util_thread;
-static struct fio_mutex *disk_thread_mutex;
static struct fio_mutex *startup_mutex;
-static struct fio_mutex *writeout_mutex;
static struct flist_head *cgroup_list;
static char *cgroup_mnt;
static int exit_value;
int shm_id = 0;
int temp_stall_ts;
unsigned long done_secs = 0;
-volatile int disk_util_exit = 0;
+volatile int helper_exit = 0;
#define PAGE_ALIGN(buf) \
(char *) (((uintptr_t) (buf) + page_mask) & ~page_mask)
fio_server_got_signal(sig);
else {
log_info("\nfio: terminating on signal %d\n", sig);
- fflush(stdout);
+ log_info_flush();
exit_value = 128;
}
ratemin);
return 1;
} else {
- rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent;
+ if (spent)
+ rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent;
+ else
+ rate = 0;
+
if (rate < ratemin ||
bytes < td->rate_bytes[ddir]) {
log_err("%s: min rate %u not met, got"
td->o.name, rate_iops);
return 1;
} else {
- rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent;
+ if (spent)
+ rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent;
+ else
+ rate = 0;
+
if (rate < rate_iops_min ||
iops < td->rate_blocks[ddir]) {
log_err("%s: min iops rate %u not met,"
return 0;
}
-static int check_min_rate(struct thread_data *td, struct timeval *now,
- uint64_t *bytes_done)
+static int check_min_rate(struct thread_data *td, struct timeval *now)
{
int ret = 0;
- if (bytes_done[DDIR_READ])
+ if (td->bytes_done[DDIR_READ])
ret |= __check_min_rate(td, now, DDIR_READ);
- if (bytes_done[DDIR_WRITE])
+ if (td->bytes_done[DDIR_WRITE])
ret |= __check_min_rate(td, now, DDIR_WRITE);
- if (bytes_done[DDIR_TRIM])
+ if (td->bytes_done[DDIR_TRIM])
ret |= __check_min_rate(td, now, DDIR_TRIM);
return ret;
/*
* get immediately available events, if any
*/
- r = io_u_queued_complete(td, 0, NULL);
+ r = io_u_queued_complete(td, 0);
if (r < 0)
return;
}
if (td->cur_depth)
- r = io_u_queued_complete(td, td->cur_depth, NULL);
+ r = io_u_queued_complete(td, td->cur_depth);
}
/*
put_io_u(td, io_u);
return 1;
} else if (ret == FIO_Q_QUEUED) {
- if (io_u_queued_complete(td, 1, NULL) < 0)
+ if (io_u_queued_complete(td, 1) < 0)
return 1;
} else if (ret == FIO_Q_COMPLETED) {
if (io_u->error) {
return 1;
}
- if (io_u_sync_complete(td, io_u, NULL) < 0)
+ if (io_u_sync_complete(td, io_u) < 0)
return 1;
} else if (ret == FIO_Q_BUSY) {
if (td_io_commit(td))
return 0;
if (!td->o.timeout)
return 0;
- if (mtime_since(&td->epoch, t) >= td->o.timeout * 1000)
+ if (utime_since(&td->epoch, t) >= td->o.timeout)
return 1;
return 0;
* fill_device option is set.
*/
td_clear_error(td);
- td->terminate = 1;
+ fio_mark_td_terminate(td);
return 1;
} else {
/*
}
}
+static int wait_for_completions(struct thread_data *td, struct timeval *time)
+{
+ const int full = queue_full(td);
+ int min_evts = 0;
+ int ret;
+
+ /*
+ * if the queue is full, we MUST reap at least 1 event
+ */
+ min_evts = min(td->o.iodepth_batch_complete, td->cur_depth);
+ if (full && !min_evts)
+ min_evts = 1;
+
+ if (time && (__should_check_rate(td, DDIR_READ) ||
+ __should_check_rate(td, DDIR_WRITE) ||
+ __should_check_rate(td, DDIR_TRIM)))
+ fio_gettime(time, NULL);
+
+ do {
+ ret = io_u_queued_complete(td, min_evts);
+ if (ret < 0)
+ break;
+ } while (full && (td->cur_depth > td->o.iodepth_low));
+
+ return ret;
+}
+
+int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
+ enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
+ struct timeval *comp_time)
+{
+ int ret2;
+
+ switch (*ret) {
+ case FIO_Q_COMPLETED:
+ if (io_u->error) {
+ *ret = -io_u->error;
+ clear_io_u(td, io_u);
+ } else if (io_u->resid) {
+ int bytes = io_u->xfer_buflen - io_u->resid;
+ struct fio_file *f = io_u->file;
+
+ if (bytes_issued)
+ *bytes_issued += bytes;
+
+ if (!from_verify)
+ trim_io_piece(td, io_u);
+
+ /*
+ * zero read, fail
+ */
+ if (!bytes) {
+ if (!from_verify)
+ unlog_io_piece(td, io_u);
+ td_verror(td, EIO, "full resid");
+ put_io_u(td, io_u);
+ break;
+ }
+
+ io_u->xfer_buflen = io_u->resid;
+ io_u->xfer_buf += bytes;
+ io_u->offset += bytes;
+
+ if (ddir_rw(io_u->ddir))
+ td->ts.short_io_u[io_u->ddir]++;
+
+ f = io_u->file;
+ if (io_u->offset == f->real_file_size)
+ goto sync_done;
+
+ requeue_io_u(td, &io_u);
+ } else {
+sync_done:
+ if (comp_time && (__should_check_rate(td, DDIR_READ) ||
+ __should_check_rate(td, DDIR_WRITE) ||
+ __should_check_rate(td, DDIR_TRIM)))
+ fio_gettime(comp_time, NULL);
+
+ *ret = io_u_sync_complete(td, io_u);
+ if (*ret < 0)
+ break;
+ }
+ return 0;
+ case FIO_Q_QUEUED:
+ /*
+ * if the engine doesn't have a commit hook,
+ * the io_u is really queued. if it does have such
+ * a hook, it has to call io_u_queued() itself.
+ */
+ if (td->io_ops->commit == NULL)
+ io_u_queued(td, io_u);
+ if (bytes_issued)
+ *bytes_issued += io_u->xfer_buflen;
+ break;
+ case FIO_Q_BUSY:
+ if (!from_verify)
+ unlog_io_piece(td, io_u);
+ requeue_io_u(td, &io_u);
+ ret2 = td_io_commit(td);
+ if (ret2 < 0)
+ *ret = ret2;
+ break;
+ default:
+ assert(ret < 0);
+ td_verror(td, -(*ret), "td_io_queue");
+ break;
+ }
+
+ if (break_on_this_error(td, ddir, ret))
+ return 1;
+
+ return 0;
+}
+
/*
* The main verify engine. Runs over the writes we previously submitted,
* reads the blocks back in, and checks the crc/md5 of the data.
*/
static void do_verify(struct thread_data *td, uint64_t verify_bytes)
{
- uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
struct fio_file *f;
struct io_u *io_u;
int ret, min_events;
io_u = NULL;
while (!td->terminate) {
enum fio_ddir ddir;
- int ret2, full;
+ int full;
update_tv_cache(td);
check_update_rusage(td);
if (runtime_exceeded(td, &td->tv_cache)) {
__update_tv_cache(td);
if (runtime_exceeded(td, &td->tv_cache)) {
- td->terminate = 1;
+ fio_mark_td_terminate(td);
break;
}
}
break;
}
} else {
- if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes)
+ if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes)
break;
while ((io_u = get_io_u(td)) != NULL) {
+ if (IS_ERR(io_u)) {
+ io_u = NULL;
+ ret = FIO_Q_BUSY;
+ goto reap;
+ }
+
/*
* We are only interested in the places where
* we wrote or trimmed IOs. Turn those into
break;
}
+ if (verify_state_should_stop(td, io_u)) {
+ put_io_u(td, io_u);
+ break;
+ }
+
if (td->o.verify_async)
io_u->end_io = verify_io_u_async;
else
io_u->end_io = verify_io_u;
ddir = io_u->ddir;
+ if (!td->o.disable_slat)
+ fio_gettime(&io_u->start_time, NULL);
ret = td_io_queue(td, io_u);
- switch (ret) {
- case FIO_Q_COMPLETED:
- if (io_u->error) {
- ret = -io_u->error;
- clear_io_u(td, io_u);
- } else if (io_u->resid) {
- int bytes = io_u->xfer_buflen - io_u->resid;
-
- /*
- * zero read, fail
- */
- if (!bytes) {
- td_verror(td, EIO, "full resid");
- put_io_u(td, io_u);
- break;
- }
-
- io_u->xfer_buflen = io_u->resid;
- io_u->xfer_buf += bytes;
- io_u->offset += bytes;
-
- if (ddir_rw(io_u->ddir))
- td->ts.short_io_u[io_u->ddir]++;
- f = io_u->file;
- if (io_u->offset == f->real_file_size)
- goto sync_done;
-
- requeue_io_u(td, &io_u);
- } else {
-sync_done:
- ret = io_u_sync_complete(td, io_u, bytes_done);
- if (ret < 0)
- break;
- }
- continue;
- case FIO_Q_QUEUED:
- break;
- case FIO_Q_BUSY:
- requeue_io_u(td, &io_u);
- ret2 = td_io_commit(td);
- if (ret2 < 0)
- ret = ret2;
- break;
- default:
- assert(ret < 0);
- td_verror(td, -ret, "td_io_queue");
- break;
- }
-
- if (break_on_this_error(td, ddir, &ret))
+ if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL))
break;
/*
* completed io_u's first. Note that we can get BUSY even
* without IO queued, if the system is resource starved.
*/
+reap:
full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
- if (full || !td->o.iodepth_batch_complete) {
- min_events = min(td->o.iodepth_batch_complete,
- td->cur_depth);
- /*
- * if the queue is full, we MUST reap at least 1 event
- */
- if (full && !min_events)
- min_events = 1;
+ if (full || !td->o.iodepth_batch_complete)
+ ret = wait_for_completions(td, NULL);
- do {
- /*
- * Reap required number of io units, if any,
- * and do the verification on them through
- * the callback handler
- */
- if (io_u_queued_complete(td, min_events, bytes_done) < 0) {
- ret = -1;
- break;
- }
- } while (full && (td->cur_depth > td->o.iodepth_low));
- }
if (ret < 0)
break;
}
min_events = td->cur_depth;
if (min_events)
- ret = io_u_queued_complete(td, min_events, NULL);
+ ret = io_u_queued_complete(td, min_events);
} else
cleanup_pending_aio(td);
dprint(FD_VERIFY, "exiting loop\n");
}
-static int io_bytes_exceeded(struct thread_data *td)
+static unsigned int exceeds_number_ios(struct thread_data *td)
+{
+ unsigned long long number_ios;
+
+ if (!td->o.number_ios)
+ return 0;
+
+ number_ios = ddir_rw_sum(td->io_blocks);
+ number_ios += td->io_u_queued + td->io_u_in_flight;
+
+ return number_ios >= (td->o.number_ios * td->loops);
+}
+
+static int io_issue_bytes_exceeded(struct thread_data *td)
+{
+ unsigned long long bytes, limit;
+
+ if (td_rw(td))
+ bytes = td->io_issue_bytes[DDIR_READ] + td->io_issue_bytes[DDIR_WRITE];
+ else if (td_write(td))
+ bytes = td->io_issue_bytes[DDIR_WRITE];
+ else if (td_read(td))
+ bytes = td->io_issue_bytes[DDIR_READ];
+ else
+ bytes = td->io_issue_bytes[DDIR_TRIM];
+
+ if (td->o.io_limit)
+ limit = td->o.io_limit;
+ else
+ limit = td->o.size;
+
+ limit *= td->loops;
+ return bytes >= limit || exceeds_number_ios(td);
+}
+
+static int io_complete_bytes_exceeded(struct thread_data *td)
{
- unsigned long long bytes;
+ unsigned long long bytes, limit;
if (td_rw(td))
bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE];
else
bytes = td->this_io_bytes[DDIR_TRIM];
- return bytes >= td->o.size;
+ if (td->o.io_limit)
+ limit = td->o.io_limit;
+ else
+ limit = td->o.size;
+
+ limit *= td->loops;
+ return bytes >= limit || exceeds_number_ios(td);
}
/*
*/
static uint64_t do_io(struct thread_data *td)
{
- uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
unsigned int i;
int ret = 0;
- uint64_t bytes_issued = 0;
+ uint64_t total_bytes, bytes_issued = 0;
if (in_ramp_time(td))
td_set_runstate(td, TD_RAMP);
lat_target_init(td);
+ total_bytes = td->o.size;
+ /*
+ * Allow random overwrite workloads to write up to io_limit
+ * before starting verification phase as 'size' doesn't apply.
+ */
+ if (td_write(td) && td_random(td) && td->o.norandommap)
+ total_bytes = max(total_bytes, (uint64_t) td->o.io_limit);
+ /*
+ * If verify_backlog is enabled, we'll run the verify in this
+ * handler as well. For that case, we may need up to twice the
+ * amount of bytes.
+ */
+ if (td->o.verify != VERIFY_NONE &&
+ (td_write(td) && td->o.verify_backlog))
+ total_bytes += td->o.size;
+
+ /* In trimwrite mode, each byte is trimmed and then written, so
+ * allow total_bytes to be twice as big */
+ if (td_trimwrite(td))
+ total_bytes += td->total_io_size;
+
while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
- (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) ||
+ (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) ||
td->o.time_based) {
struct timeval comp_time;
- int min_evts = 0;
struct io_u *io_u;
- int ret2, full;
+ int full;
enum fio_ddir ddir;
check_update_rusage(td);
if (runtime_exceeded(td, &td->tv_cache)) {
__update_tv_cache(td);
if (runtime_exceeded(td, &td->tv_cache)) {
- td->terminate = 1;
+ fio_mark_td_terminate(td);
break;
}
}
if (flow_threshold_exceeded(td))
continue;
- if (bytes_issued >= (uint64_t) td->o.size)
+ if (bytes_issued >= total_bytes)
break;
io_u = get_io_u(td);
- if (!io_u) {
+ if (IS_ERR_OR_NULL(io_u)) {
+ int err = PTR_ERR(io_u);
+
+ io_u = NULL;
+ if (err == -EBUSY) {
+ ret = FIO_Q_BUSY;
+ goto reap;
+ }
if (td->o.latency_target)
goto reap;
break;
*/
if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ &&
((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) {
+
+ if (!td->o.verify_pattern_bytes) {
+ io_u->rand_seed = __rand(&td->verify_state);
+ if (sizeof(int) != sizeof(long *))
+ io_u->rand_seed *= __rand(&td->verify_state);
+ }
+
+ if (verify_state_should_stop(td, io_u)) {
+ put_io_u(td, io_u);
+ break;
+ }
+
if (td->o.verify_async)
io_u->end_io = verify_io_u_async;
else
else
td_set_runstate(td, TD_RUNNING);
- ret = td_io_queue(td, io_u);
- switch (ret) {
- case FIO_Q_COMPLETED:
- if (io_u->error) {
- ret = -io_u->error;
- clear_io_u(td, io_u);
- } else if (io_u->resid) {
- int bytes = io_u->xfer_buflen - io_u->resid;
- struct fio_file *f = io_u->file;
-
- bytes_issued += bytes;
- /*
- * zero read, fail
- */
- if (!bytes) {
- td_verror(td, EIO, "full resid");
- put_io_u(td, io_u);
- break;
- }
-
- io_u->xfer_buflen = io_u->resid;
- io_u->xfer_buf += bytes;
- io_u->offset += bytes;
-
- if (ddir_rw(io_u->ddir))
- td->ts.short_io_u[io_u->ddir]++;
-
- if (io_u->offset == f->real_file_size)
- goto sync_done;
-
- requeue_io_u(td, &io_u);
- } else {
-sync_done:
- if (__should_check_rate(td, DDIR_READ) ||
- __should_check_rate(td, DDIR_WRITE) ||
- __should_check_rate(td, DDIR_TRIM))
- fio_gettime(&comp_time, NULL);
+ /*
+ * Always log IO before it's issued, so we know the specific
+ * order of it. The logged unit will track when the IO has
+ * completed.
+ */
+ if (td_write(td) && io_u->ddir == DDIR_WRITE &&
+ td->o.do_verify &&
+ td->o.verify != VERIFY_NONE &&
+ !td->o.experimental_verify)
+ log_io_piece(td, io_u);
- ret = io_u_sync_complete(td, io_u, bytes_done);
- if (ret < 0)
- break;
- bytes_issued += io_u->xfer_buflen;
- }
- break;
- case FIO_Q_QUEUED:
- /*
- * if the engine doesn't have a commit hook,
- * the io_u is really queued. if it does have such
- * a hook, it has to call io_u_queued() itself.
- */
- if (td->io_ops->commit == NULL)
- io_u_queued(td, io_u);
- bytes_issued += io_u->xfer_buflen;
- break;
- case FIO_Q_BUSY:
- requeue_io_u(td, &io_u);
- ret2 = td_io_commit(td);
- if (ret2 < 0)
- ret = ret2;
- break;
- default:
- assert(ret < 0);
- put_io_u(td, io_u);
- break;
- }
+ ret = td_io_queue(td, io_u);
- if (break_on_this_error(td, ddir, &ret))
+ if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 1, &comp_time))
break;
/*
* resource starved.
*/
reap:
- full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
- if (full || !td->o.iodepth_batch_complete) {
- min_evts = min(td->o.iodepth_batch_complete,
- td->cur_depth);
- /*
- * if the queue is full, we MUST reap at least 1 event
- */
- if (full && !min_evts)
- min_evts = 1;
-
- if (__should_check_rate(td, DDIR_READ) ||
- __should_check_rate(td, DDIR_WRITE) ||
- __should_check_rate(td, DDIR_TRIM))
- fio_gettime(&comp_time, NULL);
-
- do {
- ret = io_u_queued_complete(td, min_evts, bytes_done);
- if (ret < 0)
- break;
-
- } while (full && (td->cur_depth > td->o.iodepth_low));
- }
-
+ full = queue_full(td) ||
+ (ret == FIO_Q_BUSY && td->cur_depth);
+ if (full || !td->o.iodepth_batch_complete)
+ ret = wait_for_completions(td, &comp_time);
if (ret < 0)
break;
- if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO))
+ if (!ddir_rw_sum(td->bytes_done) &&
+ !(td->io_ops->flags & FIO_NOIO))
continue;
- if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) {
- if (check_min_rate(td, &comp_time, bytes_done)) {
+ if (!in_ramp_time(td) && should_check_rate(td)) {
+ if (check_min_rate(td, &comp_time)) {
if (exitall_on_terminate)
fio_terminate_threads(td->groupid);
td_verror(td, EIO, "check_min_rate");
if (td->o.fill_device && td->error == ENOSPC) {
td->error = 0;
- td->terminate = 1;
+ fio_mark_td_terminate(td);
}
if (!td->error) {
struct fio_file *f;
i = td->cur_depth;
if (i) {
- ret = io_u_queued_complete(td, i, bytes_done);
+ ret = io_u_queued_complete(td, i);
if (td->o.fill_device && td->error == ENOSPC)
td->error = 0;
}
if (!ddir_rw_sum(td->this_io_bytes))
td->done = 1;
- return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM];
+ return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
}
static void cleanup_io_u(struct thread_data *td)
io_u_rexit(&td->io_u_requeues);
io_u_qexit(&td->io_u_freelist);
io_u_qexit(&td->io_u_all);
+
+ if (td->last_write_comp)
+ sfree(td->last_write_comp);
}
static int init_io_u(struct thread_data *td)
p += max_bs;
}
+ if (td->o.verify != VERIFY_NONE) {
+ td->last_write_comp = scalloc(max_units, sizeof(uint64_t));
+ if (!td->last_write_comp) {
+ log_err("fio: failed to alloc write comp data\n");
+ return 1;
+ }
+ }
+
return 0;
}
/*
* Read back and check that the selected scheduler is now the default.
*/
- ret = fread(tmp, 1, sizeof(tmp), f);
+ ret = fread(tmp, sizeof(tmp), 1, f);
if (ferror(f) || ret < 0) {
td_verror(td, errno, "fread");
fclose(f);
return 1;
}
+ tmp[sizeof(tmp) - 1] = '\0';
+
sprintf(tmp2, "[%s]", td->o.ioscheduler);
if (!strstr(tmp, tmp2)) {
static int keep_running(struct thread_data *td)
{
+ unsigned long long limit;
+
if (td->done)
return 0;
if (td->o.time_based)
td->o.loops--;
return 1;
}
+ if (exceeds_number_ios(td))
+ return 0;
- if (td->o.size != -1ULL && ddir_rw_sum(td->io_bytes) < td->o.size) {
+ if (td->o.io_limit)
+ limit = td->o.io_limit;
+ else
+ limit = td->o.size;
+
+ if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) {
uint64_t diff;
/*
* If the difference is less than the minimum IO size, we
* are done.
*/
- diff = td->o.size - ddir_rw_sum(td->io_bytes);
+ diff = limit - ddir_rw_sum(td->io_bytes);
if (diff < td_max_bs(td))
return 0;
+ if (fio_files_done(td))
+ return 0;
+
return 1;
}
return ret;
}
+/*
+ * Dry run to compute correct state of numberio for verification.
+ */
+static uint64_t do_dry_run(struct thread_data *td)
+{
+ td_set_runstate(td, TD_RUNNING);
+
+ while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
+ (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) {
+ struct io_u *io_u;
+ int ret;
+
+ if (td->terminate || td->done)
+ break;
+
+ io_u = get_io_u(td);
+ if (!io_u)
+ break;
+
+ io_u->flags |= IO_U_F_FLIGHT;
+ io_u->error = 0;
+ io_u->resid = 0;
+ if (ddir_rw(acct_ddir(io_u)))
+ td->io_issues[acct_ddir(io_u)]++;
+ if (ddir_rw(io_u->ddir)) {
+ io_u_mark_depth(td, 1);
+ td->ts.total_io_u[io_u->ddir]++;
+ }
+
+ if (td_write(td) && io_u->ddir == DDIR_WRITE &&
+ td->o.do_verify &&
+ td->o.verify != VERIFY_NONE &&
+ !td->o.experimental_verify)
+ log_io_piece(td, io_u);
+
+ ret = io_u_sync_complete(td, io_u);
+ (void) ret;
+ }
+
+ return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
+}
+
/*
* Entry point for the thread based jobs. The process based jobs end up
* here as well, after a little setup.
} else
td->pid = gettid();
- /*
- * fio_time_init() may not have been called yet if running as a server
- */
- fio_time_init();
-
fio_local_clock_init(o->use_thread);
dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
fio_mutex_down(td->mutex);
dprint(FD_MUTEX, "done waiting on td->mutex\n");
- /*
- * the ->mutex mutex is now no longer used, close it to avoid
- * eating a file descriptor
- */
- fio_mutex_remove(td->mutex);
- td->mutex = NULL;
-
/*
* A new gid requires privilege, so we need to do this before setting
* the uid.
* Set affinity first, in case it has an impact on the memory
* allocations.
*/
- if (o->cpumask_set) {
+ if (fio_option_is_set(o, cpumask)) {
+ if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) {
+ ret = fio_cpus_split(&o->cpumask, td->thread_number - 1);
+ if (!ret) {
+ log_err("fio: no CPUs set\n");
+ log_err("fio: Try increasing number of available CPUs\n");
+ td_verror(td, EINVAL, "cpus_split");
+ goto err;
+ }
+ }
ret = fio_setaffinity(td->pid, o->cpumask);
if (ret == -1) {
td_verror(td, errno, "cpu_set_affinity");
#ifdef CONFIG_LIBNUMA
/* numa node setup */
- if (o->numa_cpumask_set || o->numa_memmask_set) {
- int ret;
+ if (fio_option_is_set(o, numa_cpunodes) ||
+ fio_option_is_set(o, numa_memnodes)) {
+ struct bitmask *mask;
if (numa_available() < 0) {
td_verror(td, errno, "Does not support NUMA API\n");
goto err;
}
- if (o->numa_cpumask_set) {
- ret = numa_run_on_node_mask(o->numa_cpunodesmask);
+ if (fio_option_is_set(o, numa_cpunodes)) {
+ mask = numa_parse_nodestring(o->numa_cpunodes);
+ ret = numa_run_on_node_mask(mask);
+ numa_free_nodemask(mask);
if (ret == -1) {
td_verror(td, errno, \
"numa_run_on_node_mask failed\n");
}
}
- if (o->numa_memmask_set) {
+ if (fio_option_is_set(o, numa_memnodes)) {
+ mask = NULL;
+ if (o->numa_memnodes)
+ mask = numa_parse_nodestring(o->numa_memnodes);
switch (o->numa_mem_mode) {
case MPOL_INTERLEAVE:
- numa_set_interleave_mask(o->numa_memnodesmask);
+ numa_set_interleave_mask(mask);
break;
case MPOL_BIND:
- numa_set_membind(o->numa_memnodesmask);
+ numa_set_membind(mask);
break;
case MPOL_LOCAL:
numa_set_localalloc();
break;
}
+ if (mask)
+ numa_free_nodemask(mask);
+
}
}
#endif
if (o->verify_async && verify_async_init(td))
goto err;
- if (o->ioprio) {
+ if (fio_option_is_set(o, ioprio) ||
+ fio_option_is_set(o, ioprio_class)) {
ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
if (ret == -1) {
td_verror(td, errno, "ioprio_set");
goto err;
}
+ if (td->flags & TD_F_COMPRESS_LOG)
+ tp_init(&td->tp_data);
+
fio_verify_init(td);
fio_gettime(&td->epoch, NULL);
prune_io_piece_log(td);
- verify_bytes = do_io(td);
+ if (td->o.verify_only && (td_write(td) || td_rw(td)))
+ verify_bytes = do_dry_run(td);
+ else
+ verify_bytes = do_io(td);
clear_state = 1;
+ /*
+ * Make sure we've successfully updated the rusage stats
+ * before waiting on the stat mutex. Otherwise we could have
+ * the stat thread holding stat mutex and waiting for
+ * the rusage_sem, which would never get upped because
+ * this thread is waiting for the stat mutex.
+ */
+ check_update_rusage(td);
+
+ fio_mutex_down(stat_mutex);
if (td_read(td) && td->io_bytes[DDIR_READ]) {
- elapsed = utime_since_now(&td->start);
+ elapsed = mtime_since_now(&td->start);
td->ts.runtime[DDIR_READ] += elapsed;
}
if (td_write(td) && td->io_bytes[DDIR_WRITE]) {
- elapsed = utime_since_now(&td->start);
+ elapsed = mtime_since_now(&td->start);
td->ts.runtime[DDIR_WRITE] += elapsed;
}
if (td_trim(td) && td->io_bytes[DDIR_TRIM]) {
- elapsed = utime_since_now(&td->start);
+ elapsed = mtime_since_now(&td->start);
td->ts.runtime[DDIR_TRIM] += elapsed;
}
+ fio_gettime(&td->start, NULL);
+ fio_mutex_up(stat_mutex);
if (td->error || td->terminate)
break;
do_verify(td, verify_bytes);
- td->ts.runtime[DDIR_READ] += utime_since_now(&td->start);
+ /*
+ * See comment further up for why this is done here.
+ */
+ check_update_rusage(td);
+
+ fio_mutex_down(stat_mutex);
+ td->ts.runtime[DDIR_READ] += mtime_since_now(&td->start);
+ fio_gettime(&td->start, NULL);
+ fio_mutex_up(stat_mutex);
if (td->error || td->terminate)
break;
}
update_rusage_stat(td);
- td->ts.runtime[DDIR_READ] = (td->ts.runtime[DDIR_READ] + 999) / 1000;
- td->ts.runtime[DDIR_WRITE] = (td->ts.runtime[DDIR_WRITE] + 999) / 1000;
- td->ts.runtime[DDIR_TRIM] = (td->ts.runtime[DDIR_TRIM] + 999) / 1000;
td->ts.total_run_time = mtime_since_now(&td->epoch);
td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ];
td->ts.io_bytes[DDIR_WRITE] = td->io_bytes[DDIR_WRITE];
td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM];
- fio_unpin_memory(td);
+ if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) &&
+ (td->o.verify != VERIFY_NONE && td_write(td))) {
+ struct all_io_list *state;
+ size_t sz;
- fio_mutex_down(writeout_mutex);
- if (td->bw_log) {
- if (o->bw_log_file) {
- finish_log_named(td, td->bw_log,
- o->bw_log_file, "bw");
- } else
- finish_log(td, td->bw_log, "bw");
- }
- if (td->lat_log) {
- if (o->lat_log_file) {
- finish_log_named(td, td->lat_log,
- o->lat_log_file, "lat");
- } else
- finish_log(td, td->lat_log, "lat");
- }
- if (td->slat_log) {
- if (o->lat_log_file) {
- finish_log_named(td, td->slat_log,
- o->lat_log_file, "slat");
- } else
- finish_log(td, td->slat_log, "slat");
- }
- if (td->clat_log) {
- if (o->lat_log_file) {
- finish_log_named(td, td->clat_log,
- o->lat_log_file, "clat");
- } else
- finish_log(td, td->clat_log, "clat");
- }
- if (td->iops_log) {
- if (o->iops_log_file) {
- finish_log_named(td, td->iops_log,
- o->iops_log_file, "iops");
- } else
- finish_log(td, td->iops_log, "iops");
+ state = get_all_io_list(td->thread_number, &sz);
+ if (state) {
+ __verify_save_state(state, "local");
+ free(state);
+ }
}
- fio_mutex_up(writeout_mutex);
+ fio_unpin_memory(td);
+
+ fio_writeout_logs(td);
+
+ if (td->flags & TD_F_COMPRESS_LOG)
+ tp_exit(&td->tp_data);
+
if (o->exec_postrun)
exec_string(o, o->exec_postrun, (const char *)"postrun");
cleanup_io_u(td);
close_ioengine(td);
cgroup_shutdown(td, &cgroup_mnt);
+ verify_free_state(td);
- if (o->cpumask_set) {
- int ret = fio_cpuset_exit(&o->cpumask);
-
- td_verror(td, ret, "fio_cpuset_exit");
+ if (fio_option_is_set(o, cpumask)) {
+ ret = fio_cpuset_exit(&o->cpumask);
+ if (ret)
+ td_verror(td, ret, "fio_cpuset_exit");
}
/*
if (o->write_iolog_file)
write_iolog_close(td);
- fio_mutex_remove(td->rusage_sem);
- td->rusage_sem = NULL;
+ fio_mutex_remove(td->mutex);
+ td->mutex = NULL;
td_set_runstate(td, TD_EXITED);
+
+ /*
+ * Do this last after setting our runstate to exited, so we
+ * know that the stat thread is signaled.
+ */
+ check_update_rusage(td);
+
return (void *) (uintptr_t) td->error;
}
struct thread_data *td;
void *data, *ret;
-#ifndef __hpux
+#if !defined(__hpux) && !defined(CONFIG_NO_SHM)
data = shmat(shmid, NULL, 0);
if (data == (void *) -1) {
int __err = errno;
return (int) (uintptr_t) ret;
}
+static void dump_td_info(struct thread_data *td)
+{
+ log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to "
+ "be stuck. Doing forceful exit of this job.\n", td->o.name,
+ (unsigned long) time_since_now(&td->terminate_time));
+}
+
/*
* Run over the job map and reap the threads that have exited, if any.
*/
}
}
+ /*
+ * If the job is stuck, do a forceful timeout of it and
+ * move on.
+ */
+ if (td->terminate &&
+ time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) {
+ dump_td_info(td);
+ td_set_runstate(td, TD_REAPED);
+ goto reaped;
+ }
+
/*
* thread is not dead, continue
*/
fio_terminate_threads(TERMINATE_ALL);
}
+static int __check_trigger_file(void)
+{
+ struct stat sb;
+
+ if (!trigger_file)
+ return 0;
+
+ if (stat(trigger_file, &sb))
+ return 0;
+
+ if (unlink(trigger_file) < 0)
+ log_err("fio: failed to unlink %s: %s\n", trigger_file,
+ strerror(errno));
+
+ return 1;
+}
+
+static int trigger_timedout(void)
+{
+ if (trigger_timeout)
+ return time_since_genesis() >= trigger_timeout;
+
+ return 0;
+}
+
+void exec_trigger(const char *cmd)
+{
+ int ret;
+
+ if (!cmd)
+ return;
+
+ ret = system(cmd);
+ if (ret == -1)
+ log_err("fio: failed executing %s trigger\n", cmd);
+}
+
+void check_trigger_file(void)
+{
+ if (__check_trigger_file() || trigger_timedout()) {
+ if (nr_clients)
+ fio_clients_send_trigger(trigger_remote_cmd);
+ else {
+ verify_save_state();
+ fio_terminate_threads(TERMINATE_ALL);
+ exec_trigger(trigger_cmd);
+ }
+ }
+}
+
+static int fio_verify_load_state(struct thread_data *td)
+{
+ int ret;
+
+ if (!td->o.verify_state)
+ return 0;
+
+ if (is_backend) {
+ void *data;
+
+ ret = fio_server_get_verify_state(td->o.name,
+ td->thread_number - 1, &data);
+ if (!ret)
+ verify_convert_assign_state(td, data);
+ } else
+ ret = verify_load_state(td, "local");
+
+ return ret;
+}
+
static void do_usleep(unsigned int usecs)
{
check_for_running_stats();
+ check_trigger_file();
usleep(usecs);
}
static void run_threads(void)
{
struct thread_data *td;
- unsigned long spent;
unsigned int i, todo, nr_running, m_rate, t_rate, nr_started;
+ uint64_t spent;
if (fio_gtod_offload && fio_start_gtod_thread())
return;
nr_process > 1 ? "es" : "");
}
log_info("\n");
- fflush(stdout);
+ log_info_flush();
}
todo = thread_number;
if (!td->o.create_serialize)
continue;
+ if (fio_verify_load_state(td))
+ goto reap;
+
/*
* do file setup here so it happens sequentially,
* we don't want X number of threads getting their
* client data interspersed on disk
*/
if (setup_files(td)) {
+reap:
exit_value++;
if (td->error)
log_err("fio: pid=%d, err=%d/%s\n",
}
if (td->o.start_delay) {
- spent = mtime_since_genesis();
+ spent = utime_since_genesis();
- if (td->o.start_delay * 1000 > spent)
+ if (td->o.start_delay > spent)
continue;
}
update_io_ticks();
}
-void wait_for_disk_thread_exit(void)
+static void wait_for_helper_thread_exit(void)
{
- fio_mutex_down(disk_thread_mutex);
+ void *ret;
+
+ helper_exit = 1;
+ pthread_cond_signal(&helper_cond);
+ pthread_join(helper_thread, &ret);
}
static void free_disk_util(void)
{
- disk_util_start_exit();
- wait_for_disk_thread_exit();
disk_util_prune_entries();
+
+ pthread_cond_destroy(&helper_cond);
}
-static void *disk_thread_main(void *data)
+static void *helper_thread_main(void *data)
{
int ret = 0;
fio_mutex_up(startup_mutex);
- while (threads && !ret) {
- usleep(DISK_UTIL_MSEC * 1000);
- if (!threads)
- break;
+ while (!ret) {
+ uint64_t sec = DISK_UTIL_MSEC / 1000;
+ uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000;
+ struct timespec ts;
+ struct timeval tv;
+
+ gettimeofday(&tv, NULL);
+ ts.tv_sec = tv.tv_sec + sec;
+ ts.tv_nsec = (tv.tv_usec * 1000) + nsec;
+
+ if (ts.tv_nsec >= 1000000000ULL) {
+ ts.tv_nsec -= 1000000000ULL;
+ ts.tv_sec++;
+ }
+
+ pthread_cond_timedwait(&helper_cond, &helper_lock, &ts);
+
ret = update_io_ticks();
+ if (helper_do_stat) {
+ helper_do_stat = 0;
+ __show_running_run_stats();
+ }
+
if (!is_backend)
print_thread_status();
}
- fio_mutex_up(disk_thread_mutex);
return NULL;
}
-static int create_disk_util_thread(void)
+static int create_helper_thread(void)
{
int ret;
setup_disk_util();
- disk_thread_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
-
- ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL);
- if (ret) {
- fio_mutex_remove(disk_thread_mutex);
- log_err("Can't create disk util thread: %s\n", strerror(ret));
- return 1;
- }
+ pthread_cond_init(&helper_cond, NULL);
+ pthread_mutex_init(&helper_lock, NULL);
- ret = pthread_detach(disk_util_thread);
+ ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL);
if (ret) {
- fio_mutex_remove(disk_thread_mutex);
- log_err("Can't detatch disk util thread: %s\n", strerror(ret));
+ log_err("Can't create helper thread: %s\n", strerror(ret));
return 1;
}
return 0;
if (write_bw_log) {
- setup_log(&agg_io_log[DDIR_READ], 0, IO_LOG_TYPE_BW);
- setup_log(&agg_io_log[DDIR_WRITE], 0, IO_LOG_TYPE_BW);
- setup_log(&agg_io_log[DDIR_TRIM], 0, IO_LOG_TYPE_BW);
+ struct log_params p = {
+ .log_type = IO_LOG_TYPE_BW,
+ };
+
+ setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log");
+ setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log");
+ setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log");
}
startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
if (startup_mutex == NULL)
return 1;
- writeout_mutex = fio_mutex_init(FIO_MUTEX_UNLOCKED);
- if (writeout_mutex == NULL)
- return 1;
set_genesis_time();
stat_init();
- create_disk_util_thread();
+ create_helper_thread();
cgroup_list = smalloc(sizeof(*cgroup_list));
INIT_FLIST_HEAD(cgroup_list);
run_threads();
+ wait_for_helper_thread_exit();
+
if (!fio_abort) {
- show_run_stats();
+ __show_run_stats();
if (write_bw_log) {
- __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log");
- __finish_log(agg_io_log[DDIR_WRITE],
- "agg-write_bw.log");
- __finish_log(agg_io_log[DDIR_TRIM],
- "agg-write_bw.log");
+ for (i = 0; i < DDIR_RWDIR_CNT; i++) {
+ struct io_log *log = agg_io_log[i];
+
+ flush_log(log);
+ free_log(log);
+ }
}
}
- for_each_td(td, i)
+ for_each_td(td, i) {
fio_options_free(td);
+ if (td->rusage_sem) {
+ fio_mutex_remove(td->rusage_sem);
+ td->rusage_sem = NULL;
+ }
+ }
free_disk_util();
cgroup_kill(cgroup_list);
sfree(cgroup_mnt);
fio_mutex_remove(startup_mutex);
- fio_mutex_remove(writeout_mutex);
- fio_mutex_remove(disk_thread_mutex);
stat_exit();
return exit_value;
}
projects / fio.git / blobdiff