#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"
-
-static pthread_t helper_thread;
-static pthread_mutex_t helper_lock;
-pthread_cond_t helper_cond;
-int helper_do_stat = 0;
+#include "rate-submit.h"
+#include "helper_thread.h"
static struct fio_mutex *startup_mutex;
static struct flist_head *cgroup_list;
int shm_id = 0;
int temp_stall_ts;
unsigned long done_secs = 0;
-volatile int helper_exit = 0;
#define PAGE_ALIGN(buf) \
(char *) (((uintptr_t) (buf) + page_mask) & ~page_mask)
}
}
-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 (td_io_commit(td))
+ return true;
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;
}
/*
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)
int min_evts = 0;
int ret;
+ if (td->flags & TD_F_REGROW_LOGS) {
+ ret = io_u_quiesce(td);
+ regrow_logs(td);
+ return 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) || !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) ||
if (*ret < 0)
break;
}
+
+ if (td->flags & TD_F_REGROW_LOGS)
+ regrow_logs(td);
+
+ /*
+ * when doing I/O (not when verifying),
+ * check for any errors that are to be ignored
+ */
+ if (!from_verify)
+ break;
+
return 0;
case FIO_Q_QUEUED:
/*
*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;
+}
+/*
+ * Unlinks files from thread data fio_file structure
+ */
+static int unlink_all_files(struct thread_data *td)
+{
+ struct fio_file *f;
+ unsigned int i;
+ int ret = 0;
+
+ for_each_file(td, f, i) {
+ if (f->filetype != FIO_TYPE_FILE)
+ continue;
+ ret = td_io_unlink_file(td, f);
+ if (ret)
+ break;
+ }
+
+ if (ret)
+ td_verror(td, ret, "unlink_all_files");
+
+ return ret;
+}
+
/*
* The main verify engine. Runs over the writes we previously submitted,
* reads the blocks back in, and checks the crc/md5 of the data.
if (td->error)
return;
+ /*
+ * verify_state needs to be reset before verification
+ * proceeds so that expected random seeds match actual
+ * random seeds in headers. The main loop will reset
+ * all random number generators if randrepeat is set.
+ */
+ if (!td->o.rand_repeatable)
+ td_fill_verify_state_seed(td);
+
td_set_runstate(td, TD_VERIFYING);
io_u = NULL;
break;
while ((io_u = get_io_u(td)) != NULL) {
- if (IS_ERR(io_u)) {
+ if (IS_ERR_OR_NULL(io_u)) {
io_u = NULL;
ret = FIO_Q_BUSY;
goto reap;
*/
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.
*
* Returns number of bytes written and trimmed.
*/
-static uint64_t do_io(struct thread_data *td)
+static void do_io(struct thread_data *td, uint64_t *bytes_done)
{
unsigned int i;
int ret = 0;
uint64_t total_bytes, bytes_issued = 0;
+ for (i = 0; i < DDIR_RWDIR_CNT; i++)
+ bytes_done[i] = td->bytes_done[i];
+
if (in_ramp_time(td))
td_set_runstate(td, TD_RAMP);
else
if (flow_threshold_exceeded(td))
continue;
- if (bytes_issued >= total_bytes)
+ /*
+ * Break if we exceeded the bytes. The exception is time
+ * based runs, but we still need to break out of the loop
+ * for those to run verification, if enabled.
+ */
+ if (bytes_issued >= total_bytes &&
+ (!td->o.time_based ||
+ (td->o.time_based && td->o.verify != VERIFY_NONE)))
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;
if (!ddir_rw_sum(td->this_io_bytes))
td->done = 1;
- return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
+ for (i = 0; i < DDIR_RWDIR_CNT; i++)
+ bytes_done[i] = td->bytes_done[i] - bytes_done[i];
+}
+
+static void free_file_completion_logging(struct thread_data *td)
+{
+ struct fio_file *f;
+ unsigned int i;
+
+ for_each_file(td, f, i) {
+ if (!f->last_write_comp)
+ break;
+ sfree(f->last_write_comp);
+ }
+}
+
+static int init_file_completion_logging(struct thread_data *td,
+ unsigned int depth)
+{
+ struct fio_file *f;
+ unsigned int i;
+
+ if (td->o.verify == VERIFY_NONE || !td->o.verify_state_save)
+ return 0;
+
+ for_each_file(td, f, i) {
+ f->last_write_comp = scalloc(depth, sizeof(uint64_t));
+ if (!f->last_write_comp)
+ goto cleanup;
+ }
+
+ return 0;
+
+cleanup:
+ free_file_completion_logging(td);
+ log_err("fio: failed to alloc write comp data\n");
+ return 1;
}
static void cleanup_io_u(struct thread_data *td)
io_u_qexit(&td->io_u_freelist);
io_u_qexit(&td->io_u_all);
- if (td->last_write_comp)
- sfree(td->last_write_comp);
+ free_file_completion_logging(td);
}
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;
- }
- }
+ if (init_file_completion_logging(td, max_units))
+ return 1;
return 0;
}
static int switch_ioscheduler(struct thread_data *td)
{
+#ifdef FIO_HAVE_IOSCHED_SWITCH
char tmp[256], tmp2[128];
FILE *f;
int ret;
*/
tmp[strlen(tmp) - 1] = '\0';
+ /*
+ * Write to "none" entry doesn't fail, so check the result here.
+ */
+ if (!strcmp(tmp, "none")) {
+ log_err("fio: io scheduler is not tunable\n");
+ fclose(f);
+ return 0;
+ }
sprintf(tmp2, "[%s]", td->o.ioscheduler);
if (!strstr(tmp, tmp2)) {
fclose(f);
return 0;
+#else
+ return 0;
+#endif
}
-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;
+ if (fio_files_done(td) && !td->o.io_limit)
+ return false;
- return 1;
+ return true;
}
- return 0;
+ return false;
}
static int exec_string(struct thread_options *o, const char *string, const char *mode)
{
size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1;
- int ret; /* newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1; */
+ int ret;
char *str;
str = malloc(newlen);
break;
io_u = get_io_u(td);
- if (!io_u)
+ if (IS_ERR_OR_NULL(io_u))
break;
io_u_set(io_u, IO_U_F_FLIGHT);
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--;
-}
+struct fork_data {
+ struct thread_data *td;
+ struct sk_out *sk_out;
+};
/*
* Entry point for the thread based jobs. The process based jobs end up
*/
static void *thread_main(void *data)
{
+ struct fork_data *fd = data;
unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, };
- struct thread_data *td = data;
+ struct thread_data *td = fd->td;
struct thread_options *o = &td->o;
- pthread_condattr_t attr;
+ struct sk_out *sk_out = fd->sk_out;
int clear_state;
int ret;
+ sk_out_assign(sk_out);
+ free(fd);
+
if (!o->use_thread) {
setsid();
td->pid = getpid();
INIT_FLIST_HEAD(&td->verify_list);
INIT_FLIST_HEAD(&td->trim_list);
INIT_FLIST_HEAD(&td->next_rand_list);
- pthread_mutex_init(&td->io_u_lock, NULL);
td->io_hist_tree = RB_ROOT;
- pthread_condattr_init(&attr);
- pthread_cond_init(&td->verify_cond, &attr);
- pthread_cond_init(&td->free_cond, &attr);
+ ret = mutex_cond_init_pshared(&td->io_u_lock, &td->free_cond);
+ if (ret) {
+ td_verror(td, ret, "mutex_cond_init_pshared");
+ goto err;
+ }
+ ret = cond_init_pshared(&td->verify_cond);
+ if (ret) {
+ td_verror(td, ret, "mutex_cond_pshared");
+ goto err;
+ }
td_set_runstate(td, TD_INITIALIZED);
dprint(FD_MUTEX, "up startup_mutex\n");
goto err;
}
+ /*
+ * Do this early, we don't want the compress threads to be limited
+ * to the same CPUs as the IO workers. So do this before we set
+ * any potential CPU affinity
+ */
+ if (iolog_compress_init(td, sk_out))
+ goto err;
+
/*
* If we have a gettimeofday() thread, make sure we exclude that
* thread from this job
goto err;
}
- if (td->flags & TD_F_COMPRESS_LOG)
- tp_init(&td->tp_data);
-
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, sk_out))
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, 0);
- if (clear_state)
- clear_io_state(td);
+ if (o->unlink_each_loop && unlink_all_files(td))
+ break;
+ }
prune_io_piece_log(td);
if (td->o.verify_only && (td_write(td) || td_rw(td)))
verify_bytes = do_dry_run(td);
- else
- verify_bytes = do_io(td);
+ else {
+ uint64_t bytes_done[DDIR_RWDIR_CNT];
+
+ do_io(td, bytes_done);
+
+ if (!ddir_rw_sum(bytes_done)) {
+ fio_mark_td_terminate(td);
+ verify_bytes = 0;
+ } else {
+ verify_bytes = bytes_done[DDIR_WRITE] +
+ bytes_done[DDIR_TRIM];
+ }
+ }
clear_state = 1;
(td->io_ops->flags & FIO_UNIDIR))
continue;
- clear_io_state(td);
+ clear_io_state(td, 0);
fio_gettime(&td->start, NULL);
break;
}
+ td_set_runstate(td, TD_FINISHING);
+
update_rusage_stat(td);
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_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);
+ td_writeout_logs(td, true);
- 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:
cgroup_shutdown(td, &cgroup_mnt);
verify_free_state(td);
+ if (td->zone_state_index) {
+ int i;
+
+ for (i = 0; i < DDIR_RWDIR_CNT; i++)
+ free(td->zone_state_index[i]);
+ free(td->zone_state_index);
+ td->zone_state_index = NULL;
+ }
+
if (fio_option_is_set(o, cpumask)) {
ret = fio_cpuset_exit(&o->cpumask);
if (ret)
*/
check_update_rusage(td);
+ sk_out_drop();
return (void *) (uintptr_t) td->error;
}
-
-/*
- * We cannot pass the td data into a forked process, so attach the td and
- * pass it to the thread worker.
- */
-static int fork_main(int shmid, int offset)
-{
- struct thread_data *td;
- void *data, *ret;
-
-#if !defined(__hpux) && !defined(CONFIG_NO_SHM)
- data = shmat(shmid, NULL, 0);
- if (data == (void *) -1) {
- int __err = errno;
-
- perror("shmat");
- return __err;
- }
-#else
- /*
- * HP-UX inherits shm mappings?
- */
- data = threads;
-#endif
-
- td = data + offset * sizeof(struct thread_data);
- ret = thread_main(td);
- shmdt(data);
- 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,
+ log_err("fio: job '%s' (state=%d) hasn't exited in %lu seconds, it "
+ "appears to be stuck. Doing forceful exit of this job.\n",
+ td->o.name, td->runstate,
(unsigned long) time_since_now(&td->terminate_time));
}
* move on.
*/
if (td->terminate &&
+ td->runstate < TD_FSYNCING &&
time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) {
dump_td_info(td);
td_set_runstate(td, TD_REAPED);
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, &ver);
+ td->thread_number - 1, &data);
if (!ret)
- verify_convert_assign_state(td, data, ver);
+ verify_assign_state(td, data);
} else
ret = verify_load_state(td, "local");
usleep(usecs);
}
-static int check_mount_writes(struct thread_data *td)
+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 0;
+ return false;
for_each_file(td, f, i) {
if (f->filetype != FIO_TYPE_BD)
goto mounted;
}
- return 0;
+ return false;
mounted:
log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.", f->file_name);
- return 1;
+ return true;
+}
+
+static bool waitee_running(struct thread_data *me)
+{
+ const char *waitee = me->o.wait_for;
+ const char *self = me->o.name;
+ struct thread_data *td;
+ int i;
+
+ if (!waitee)
+ return false;
+
+ for_each_td(td, i) {
+ if (!strcmp(td->o.name, self) || strcmp(td->o.name, waitee))
+ continue;
+
+ if (td->runstate < TD_EXITED) {
+ dprint(FD_PROCESS, "%s fenced by %s(%s)\n",
+ self, td->o.name,
+ runstate_to_name(td->runstate));
+ return true;
+ }
+ }
+
+ dprint(FD_PROCESS, "%s: %s completed, can run\n", self, waitee);
+ return false;
}
/*
* Main function for kicking off and reaping jobs, as needed.
*/
-static void run_threads(void)
+static void run_threads(struct sk_out *sk_out)
{
struct thread_data *td;
unsigned int i, todo, nr_running, m_rate, t_rate, nr_started;
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,
struct thread_data *map[REAL_MAX_JOBS];
struct timeval this_start;
int this_jobs = 0, left;
+ struct fork_data *fd;
/*
* create threads (TD_NOT_CREATED -> TD_CREATED)
break;
}
+ if (waitee_running(td)) {
+ dprint(FD_PROCESS, "%s: waiting for %s\n",
+ td->o.name, td->o.wait_for);
+ continue;
+ }
+
init_disk_util(td);
td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED);
map[this_jobs++] = td;
nr_started++;
+ fd = calloc(1, sizeof(*fd));
+ fd->td = td;
+ fd->sk_out = sk_out;
+
if (td->o.use_thread) {
int ret;
dprint(FD_PROCESS, "will pthread_create\n");
ret = pthread_create(&td->thread, NULL,
- thread_main, td);
+ thread_main, fd);
if (ret) {
log_err("pthread_create: %s\n",
strerror(ret));
+ free(fd);
nr_started--;
break;
}
dprint(FD_PROCESS, "will fork\n");
pid = fork();
if (!pid) {
- int ret = fork_main(shm_id, i);
+ int ret;
+ ret = (int)(uintptr_t)thread_main(fd);
_exit(ret);
} else if (i == fio_debug_jobno)
*fio_debug_jobp = pid;
}
dprint(FD_MUTEX, "wait on startup_mutex\n");
- if (fio_mutex_down_timeout(startup_mutex, 10)) {
+ if (fio_mutex_down_timeout(startup_mutex, 10000)) {
log_err("fio: job startup hung? exiting.\n");
fio_terminate_threads(TERMINATE_ALL);
fio_abort = 1;
update_io_ticks();
}
-static void wait_for_helper_thread_exit(void)
-{
- void *ret;
-
- helper_exit = 1;
- pthread_cond_signal(&helper_cond);
- pthread_join(helper_thread, &ret);
-}
-
static void free_disk_util(void)
{
disk_util_prune_entries();
-
- pthread_cond_destroy(&helper_cond);
-}
-
-static void *helper_thread_main(void *data)
-{
- int ret = 0;
-
- fio_mutex_up(startup_mutex);
-
- 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();
- }
-
- return NULL;
-}
-
-static int create_helper_thread(void)
-{
- int ret;
-
- setup_disk_util();
-
- pthread_cond_init(&helper_cond, NULL);
- pthread_mutex_init(&helper_lock, NULL);
-
- ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL);
- if (ret) {
- log_err("Can't create helper thread: %s\n", strerror(ret));
- return 1;
- }
-
- dprint(FD_MUTEX, "wait on startup_mutex\n");
- fio_mutex_down(startup_mutex);
- dprint(FD_MUTEX, "done waiting on startup_mutex\n");
- return 0;
+ helper_thread_destroy();
}
-int fio_backend(void)
+int fio_backend(struct sk_out *sk_out)
{
struct thread_data *td;
int i;
set_genesis_time();
stat_init();
- create_helper_thread();
+ helper_thread_create(startup_mutex, sk_out);
cgroup_list = smalloc(sizeof(*cgroup_list));
INIT_FLIST_HEAD(cgroup_list);
- run_threads();
+ run_threads(sk_out);
- wait_for_helper_thread_exit();
+ helper_thread_exit();
if (!fio_abort) {
__show_run_stats();
for (i = 0; i < DDIR_RWDIR_CNT; i++) {
struct io_log *log = agg_io_log[i];
- flush_log(log, 0);
+ flush_log(log, false);
free_log(log);
}
}
projects / fio.git / blobdiff