#include <sys/wait.h>
#include <sys/ipc.h>
#include <sys/mman.h>
+#include <math.h>
#include "fio.h"
#ifndef FIO_NO_HAVE_SHM_H
#include "cgroup.h"
#include "profile.h"
#include "lib/rand.h"
-#include "memalign.h"
+#include "lib/memalign.h"
#include "server.h"
#include "lib/getrusage.h"
#include "idletime.h"
#include "err.h"
-#include "lib/tp.h"
#include "workqueue.h"
+#include "lib/mountcheck.h"
+#include "rate-submit.h"
static pthread_t helper_thread;
static pthread_mutex_t helper_lock;
}
}
-static void sig_show_status(int sig)
+void sig_show_status(int sig)
{
show_running_run_stats();
}
/*
* Check if we are above the minimum rate given.
*/
-static int __check_min_rate(struct thread_data *td, struct timeval *now,
- enum fio_ddir ddir)
+static bool __check_min_rate(struct thread_data *td, struct timeval *now,
+ enum fio_ddir ddir)
{
unsigned long long bytes = 0;
unsigned long iops = 0;
assert(ddir_rw(ddir));
if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir])
- return 0;
+ return false;
/*
* allow a 2 second settle period in the beginning
*/
if (mtime_since(&td->start, now) < 2000)
- return 0;
+ return false;
iops += td->this_io_blocks[ddir];
bytes += td->this_io_bytes[ddir];
if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) {
spent = mtime_since(&td->lastrate[ddir], now);
if (spent < td->o.ratecycle)
- return 0;
+ return false;
- if (td->o.rate[ddir]) {
+ if (td->o.rate[ddir] || td->o.ratemin[ddir]) {
/*
* check bandwidth specified rate
*/
if (bytes < td->rate_bytes[ddir]) {
log_err("%s: min rate %u not met\n", td->o.name,
ratemin);
- return 1;
+ return true;
} else {
if (spent)
rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent;
log_err("%s: min rate %u not met, got"
" %luKB/sec\n", td->o.name,
ratemin, rate);
- return 1;
+ return true;
}
}
} else {
if (iops < rate_iops) {
log_err("%s: min iops rate %u not met\n",
td->o.name, rate_iops);
- return 1;
+ return true;
} else {
if (spent)
rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent;
log_err("%s: min iops rate %u not met,"
" got %lu\n", td->o.name,
rate_iops_min, rate);
+ return true;
}
}
}
td->rate_bytes[ddir] = bytes;
td->rate_blocks[ddir] = iops;
memcpy(&td->lastrate[ddir], now, sizeof(*now));
- return 0;
+ return false;
}
-static int check_min_rate(struct thread_data *td, struct timeval *now)
+static bool check_min_rate(struct thread_data *td, struct timeval *now)
{
- int ret = 0;
+ bool ret = false;
if (td->bytes_done[DDIR_READ])
ret |= __check_min_rate(td, now, DDIR_READ);
* Helper to handle the final sync of a file. Works just like the normal
* io path, just does everything sync.
*/
-static int fio_io_sync(struct thread_data *td, struct fio_file *f)
+static bool fio_io_sync(struct thread_data *td, struct fio_file *f)
{
struct io_u *io_u = __get_io_u(td);
int ret;
if (!io_u)
- return 1;
+ return true;
io_u->ddir = DDIR_SYNC;
io_u->file = f;
if (td_io_prep(td, io_u)) {
put_io_u(td, io_u);
- return 1;
+ return true;
}
requeue:
if (ret < 0) {
td_verror(td, io_u->error, "td_io_queue");
put_io_u(td, io_u);
- return 1;
+ return true;
} else if (ret == FIO_Q_QUEUED) {
if (io_u_queued_complete(td, 1) < 0)
- return 1;
+ return true;
} else if (ret == FIO_Q_COMPLETED) {
if (io_u->error) {
td_verror(td, io_u->error, "td_io_queue");
- return 1;
+ return true;
}
if (io_u_sync_complete(td, io_u) < 0)
- return 1;
+ return true;
} else if (ret == FIO_Q_BUSY) {
if (td_io_commit(td))
- return 1;
+ return true;
goto requeue;
}
- return 0;
+ return false;
}
static int fio_file_fsync(struct thread_data *td, struct fio_file *f)
__update_tv_cache(td);
}
-static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
+static inline bool runtime_exceeded(struct thread_data *td, struct timeval *t)
{
if (in_ramp_time(td))
- return 0;
+ return false;
if (!td->o.timeout)
- return 0;
+ return false;
if (utime_since(&td->epoch, t) >= td->o.timeout)
- return 1;
+ return true;
- return 0;
+ return false;
}
/*
unsigned long long *elapsed_us,
const enum fio_ddir ddir)
{
+ if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only)
+ return;
+
td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000;
elapsed_us[ddir] += utime_since_now(&td->start);
td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000;
}
-static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
- int *retptr)
+static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
+ int *retptr)
{
int ret = *retptr;
eb = td_error_type(ddir, err);
if (!(td->o.continue_on_error & (1 << eb)))
- return 1;
+ return true;
if (td_non_fatal_error(td, eb, err)) {
/*
update_error_count(td, err);
td_clear_error(td);
*retptr = 0;
- return 0;
+ return false;
} else if (td->o.fill_device && err == ENOSPC) {
/*
* We expect to hit this error if
*/
td_clear_error(td);
fio_mark_td_terminate(td);
- return 1;
+ return true;
} else {
/*
* Stop the I/O in case of a fatal
* error.
*/
update_error_count(td, err);
- return 1;
+ return true;
}
}
- return 0;
+ return false;
}
static void check_update_rusage(struct thread_data *td)
/*
* 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) || !td->o.iodepth_batch_complete)
+ min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth);
+ if ((full && !min_evts) || !td->o.iodepth_batch_complete_min)
min_evts = 1;
if (time && (__should_check_rate(td, DDIR_READ) ||
*ret = ret2;
break;
default:
- assert(ret < 0);
+ assert(*ret < 0);
td_verror(td, -(*ret), "td_io_queue");
break;
}
return 0;
}
+static inline bool io_in_polling(struct thread_data *td)
+{
+ return !td->o.iodepth_batch_complete_min &&
+ !td->o.iodepth_batch_complete_max;
+}
+
/*
* The main verify engine. Runs over the writes we previously submitted,
* reads the blocks back in, and checks the crc/md5 of the data.
*/
reap:
full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
- if (full || !td->o.iodepth_batch_complete)
+ if (full || io_in_polling(td))
ret = wait_for_completions(td, NULL);
if (ret < 0)
dprint(FD_VERIFY, "exiting loop\n");
}
-static unsigned int exceeds_number_ios(struct thread_data *td)
+static bool exceeds_number_ios(struct thread_data *td)
{
unsigned long long number_ios;
if (!td->o.number_ios)
- return 0;
+ return false;
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)
+static bool io_issue_bytes_exceeded(struct thread_data *td)
{
unsigned long long bytes, limit;
return bytes >= limit || exceeds_number_ios(td);
}
-static int io_complete_bytes_exceeded(struct thread_data *td)
+static bool io_complete_bytes_exceeded(struct thread_data *td)
{
unsigned long long bytes, limit;
return bytes >= limit || exceeds_number_ios(td);
}
+/*
+ * used to calculate the next io time for rate control
+ *
+ */
+static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir)
+{
+ uint64_t secs, remainder, bps, bytes, iops;
+
+ assert(!(td->flags & TD_F_CHILD));
+ bytes = td->rate_io_issue_bytes[ddir];
+ bps = td->rate_bps[ddir];
+
+ if (td->o.rate_process == RATE_PROCESS_POISSON) {
+ uint64_t val;
+ iops = bps / td->o.bs[ddir];
+ val = (int64_t) (1000000 / iops) *
+ -logf(__rand_0_1(&td->poisson_state));
+ if (val) {
+ dprint(FD_RATE, "poisson rate iops=%llu\n",
+ (unsigned long long) 1000000 / val);
+ }
+ td->last_usec += val;
+ return td->last_usec;
+ } else if (bps) {
+ secs = bytes / bps;
+ remainder = bytes % bps;
+ return remainder * 1000000 / bps + secs * 1000000;
+ }
+
+ return 0;
+}
+
/*
* Main IO worker function. It retrieves io_u's to process and queues
* and reaps them, checking for rate and errors along the way.
if (flow_threshold_exceeded(td))
continue;
- if (bytes_issued >= total_bytes)
+ if (!td->o.time_based && bytes_issued >= total_bytes)
break;
io_u = get_io_u(td);
log_io_piece(td, io_u);
if (td->o.io_submit_mode == IO_MODE_OFFLOAD) {
+ const unsigned long blen = io_u->xfer_buflen;
+ const enum fio_ddir ddir = acct_ddir(io_u);
+
if (td->error)
break;
- ret = workqueue_enqueue(&td->io_wq, io_u);
+
+ workqueue_enqueue(&td->io_wq, &io_u->work);
+ ret = FIO_Q_QUEUED;
+
+ if (ddir_rw(ddir)) {
+ td->io_issues[ddir]++;
+ td->io_issue_bytes[ddir] += blen;
+ td->rate_io_issue_bytes[ddir] += blen;
+ }
+
+ if (should_check_rate(td))
+ td->rate_next_io_time[ddir] = usec_for_io(td, ddir);
+
} else {
ret = td_io_queue(td, io_u);
- if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 1, &comp_time))
+ if (should_check_rate(td))
+ td->rate_next_io_time[ddir] = usec_for_io(td, ddir);
+
+ if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time))
break;
/*
reap:
full = queue_full(td) ||
(ret == FIO_Q_BUSY && td->cur_depth);
- if (full || !td->o.iodepth_batch_complete)
+ if (full || io_in_polling(td))
ret = wait_for_completions(td, &comp_time);
}
if (ret < 0)
if (!in_ramp_time(td) && should_check_rate(td)) {
if (check_min_rate(td, &comp_time)) {
- if (exitall_on_terminate)
+ if (exitall_on_terminate || td->o.exitall_error)
fio_terminate_threads(td->groupid);
td_verror(td, EIO, "check_min_rate");
break;
/*
* Read back and check that the selected scheduler is now the default.
*/
+ memset(tmp, 0, sizeof(tmp));
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';
+ /*
+ * either a list of io schedulers or "none\n" is expected.
+ */
+ tmp[strlen(tmp) - 1] = '\0';
sprintf(tmp2, "[%s]", td->o.ioscheduler);
return 0;
}
-static int keep_running(struct thread_data *td)
+static bool keep_running(struct thread_data *td)
{
unsigned long long limit;
if (td->done)
- return 0;
+ return false;
if (td->o.time_based)
- return 1;
+ return true;
if (td->o.loops) {
td->o.loops--;
- return 1;
+ return true;
}
if (exceeds_number_ios(td))
- return 0;
+ return false;
if (td->o.io_limit)
limit = td->o.io_limit;
*/
diff = limit - ddir_rw_sum(td->io_bytes);
if (diff < td_max_bs(td))
- return 0;
+ return false;
if (fio_files_done(td))
- return 0;
+ return false;
- return 1;
+ return true;
}
- return 0;
+ return false;
}
static int exec_string(struct thread_options *o, const char *string, const char *mode)
{
- int ret, newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1;
+ size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1;
+ int ret;
char *str;
str = malloc(newlen);
return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
}
-static void io_workqueue_fn(struct thread_data *td, struct io_u *io_u)
-{
- const enum fio_ddir ddir = io_u->ddir;
- int ret;
-
- dprint(FD_RATE, "io_u %p queued by %u\n", io_u, gettid());
-
- io_u_set(io_u, IO_U_F_NO_FILE_PUT);
-
- td->cur_depth++;
-
- ret = td_io_queue(td, io_u);
-
- dprint(FD_RATE, "io_u %p ret %d by %u\n", io_u, ret, gettid());
-
- io_queue_event(td, io_u, &ret, ddir, NULL, 0, NULL);
-
- if (ret == FIO_Q_QUEUED)
- ret = io_u_queued_complete(td, 1);
-
- td->cur_depth--;
-}
-
/*
* Entry point for the thread based jobs. The process based jobs end up
* here as well, after a little setup.
goto err;
}
- if (td->flags & TD_F_COMPRESS_LOG)
- tp_init(&td->tp_data);
+ if (iolog_compress_init(td))
+ goto err;
fio_verify_init(td);
- if ((o->io_submit_mode == IO_MODE_OFFLOAD) &&
- workqueue_init(td, &td->io_wq, io_workqueue_fn, td->o.iodepth))
+ if (rate_submit_init(td))
goto err;
fio_gettime(&td->epoch, NULL);
fio_getrusage(&td->ru_start);
+ memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch));
+ memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch));
+
+ if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] ||
+ o->ratemin[DDIR_TRIM]) {
+ memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time,
+ sizeof(td->bw_sample_time));
+ memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time,
+ sizeof(td->bw_sample_time));
+ memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time,
+ sizeof(td->bw_sample_time));
+ }
+
clear_state = 0;
while (keep_running(td)) {
uint64_t verify_bytes;
fio_gettime(&td->start, NULL);
- memcpy(&td->bw_sample_time, &td->start, sizeof(td->start));
- memcpy(&td->iops_sample_time, &td->start, sizeof(td->start));
memcpy(&td->tv_cache, &td->start, sizeof(td->start));
- if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] ||
- o->ratemin[DDIR_TRIM]) {
- memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time,
- sizeof(td->bw_sample_time));
- memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time,
- sizeof(td->bw_sample_time));
- memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time,
- sizeof(td->bw_sample_time));
- }
-
if (clear_state)
- clear_io_state(td);
+ clear_io_state(td, 0);
prune_io_piece_log(td);
if (td->o.verify_only && (td_write(td) || td_rw(td)))
verify_bytes = do_dry_run(td);
- else
+ else {
verify_bytes = do_io(td);
+ if (!verify_bytes)
+ fio_mark_td_terminate(td);
+ }
clear_state = 1;
(td->io_ops->flags & FIO_UNIDIR))
continue;
- clear_io_state(td);
+ clear_io_state(td, 0);
fio_gettime(&td->start, NULL);
td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM];
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;
-
- state = get_all_io_list(td->thread_number, &sz);
- if (state) {
- __verify_save_state(state, "local");
- free(state);
- }
- }
+ (td->o.verify != VERIFY_NONE && td_write(td)))
+ verify_save_state(td->thread_number);
fio_unpin_memory(td);
fio_writeout_logs(td);
- if (o->io_submit_mode == IO_MODE_OFFLOAD)
- workqueue_exit(&td->io_wq);
-
- if (td->flags & TD_F_COMPRESS_LOG)
- tp_exit(&td->tp_data);
+ iolog_compress_exit(td);
+ rate_submit_exit(td);
if (o->exec_postrun)
exec_string(o, o->exec_postrun, (const char *)"postrun");
- if (exitall_on_terminate)
+ if (exitall_on_terminate || (o->exitall_error && td->error))
fio_terminate_threads(td->groupid);
err:
fio_terminate_threads(TERMINATE_ALL);
}
-static int __check_trigger_file(void)
+static bool __check_trigger_file(void)
{
struct stat sb;
if (!trigger_file)
- return 0;
+ return false;
if (stat(trigger_file, &sb))
- return 0;
+ return false;
if (unlink(trigger_file) < 0)
log_err("fio: failed to unlink %s: %s\n", trigger_file,
strerror(errno));
- return 1;
+ return true;
}
-static int trigger_timedout(void)
+static bool trigger_timedout(void)
{
if (trigger_timeout)
return time_since_genesis() >= trigger_timeout;
- return 0;
+ return false;
}
void exec_trigger(const char *cmd)
if (nr_clients)
fio_clients_send_trigger(trigger_remote_cmd);
else {
- verify_save_state();
+ verify_save_state(IO_LIST_ALL);
fio_terminate_threads(TERMINATE_ALL);
exec_trigger(trigger_cmd);
}
if (is_backend) {
void *data;
+ int ver;
ret = fio_server_get_verify_state(td->o.name,
- td->thread_number - 1, &data);
+ td->thread_number - 1, &data, &ver);
if (!ret)
- verify_convert_assign_state(td, data);
+ verify_convert_assign_state(td, data, ver);
} else
ret = verify_load_state(td, "local");
usleep(usecs);
}
+static bool check_mount_writes(struct thread_data *td)
+{
+ struct fio_file *f;
+ unsigned int i;
+
+ if (!td_write(td) || td->o.allow_mounted_write)
+ return false;
+
+ for_each_file(td, f, i) {
+ if (f->filetype != FIO_TYPE_BD)
+ continue;
+ if (device_is_mounted(f->file_name))
+ goto mounted;
+ }
+
+ return false;
+mounted:
+ log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.", f->file_name);
+ return true;
+}
+
/*
* Main function for kicking off and reaping jobs, as needed.
*/
nr_thread = nr_process = 0;
for_each_td(td, i) {
+ if (check_mount_writes(td))
+ return;
if (td->o.use_thread)
nr_thread++;
else
nr_process++;
}
- if (output_format == FIO_OUTPUT_NORMAL) {
+ if (output_format & FIO_OUTPUT_NORMAL) {
log_info("Starting ");
if (nr_thread)
log_info("%d thread%s", nr_thread,
projects / fio.git / blobdiff