int groupid = 0;
int thread_number = 0;
-static char run_str[MAX_JOBS + 1];
int shm_id = 0;
-static struct timeval genesis;
int temp_stall_ts;
char *fio_inst_prefix = _INST_PREFIX;
-static void print_thread_status(void);
-
extern unsigned long long mlock_size;
-/*
- * Thread life cycle. Once a thread has a runstate beyond TD_INITIALIZED, it
- * will never back again. It may cycle between running/verififying/fsyncing.
- * Once the thread reaches TD_EXITED, it is just waiting for the core to
- * reap it.
- */
-enum {
- TD_NOT_CREATED = 0,
- TD_CREATED,
- TD_INITIALIZED,
- TD_RUNNING,
- TD_VERIFYING,
- TD_FSYNCING,
- TD_EXITED,
- TD_REAPED,
-};
-
#define should_fsync(td) ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync))
static volatile int startup_sem;
buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1);
}
- if (buflen > td->io_size - td->this_io_bytes[td->ddir])
+ if (buflen > td->io_size - td->this_io_bytes[td->ddir]) {
+ /*
+ * if using direct/raw io, we may not be able to
+ * shrink the size. so just fail it.
+ */
+ if (td->io_ops->flags & FIO_RAWIO)
+ return 0;
+
buflen = td->io_size - td->this_io_bytes[td->ddir];
+ }
return buflen;
}
return NULL;
}
- if (io_u->buflen + io_u->offset > f->file_size)
+ if (io_u->buflen + io_u->offset > f->file_size) {
+ if (td->io_ops->flags & FIO_RAWIO) {
+ put_io_u(td, io_u);
+ return NULL;
+ }
+
io_u->buflen = f->file_size - io_u->offset;
+ }
if (!io_u->buflen) {
put_io_u(td, io_u);
static struct fio_file *get_next_file(struct thread_data *td)
{
- struct fio_file *f = &td->files[td->next_file];
+ int old_next_file = td->next_file;
+ struct fio_file *f;
+
+ do {
+ f = &td->files[td->next_file];
- td->next_file++;
- if (td->next_file >= td->nr_files)
- td->next_file = 0;
+ td->next_file++;
+ if (td->next_file >= td->nr_files)
+ td->next_file = 0;
+
+ if (f->fd != -1)
+ break;
+
+ f = NULL;
+ } while (td->next_file != old_next_file);
return f;
}
return 0;
}
-static int io_u_getevents(struct thread_data *td, int min, int max,
+static int td_io_getevents(struct thread_data *td, int min, int max,
struct timespec *t)
{
return td->io_ops->getevents(td, min, max, t);
}
-static int io_u_queue(struct thread_data *td, struct io_u *io_u)
+static int td_io_queue(struct thread_data *td, struct io_u *io_u)
{
gettimeofday(&io_u->issue_time, NULL);
/*
* get immediately available events, if any
*/
- r = io_u_getevents(td, 0, td->cur_depth, &ts);
+ r = td_io_getevents(td, 0, td->cur_depth, &ts);
if (r > 0) {
icd.nr = r;
ios_completed(td, &icd);
}
if (td->cur_depth) {
- r = io_u_getevents(td, td->cur_depth, td->cur_depth, NULL);
+ r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL);
if (r > 0) {
icd.nr = r;
ios_completed(td, &icd);
break;
}
- ret = io_u_queue(td, io_u);
+ ret = td_io_queue(td, io_u);
if (ret) {
put_io_u(td, io_u);
td_verror(td, ret);
if (do_io_u_verify(td, &v_io_u))
break;
- ret = io_u_getevents(td, 1, 1, NULL);
+ ret = td_io_getevents(td, 1, 1, NULL);
if (ret != 1) {
if (ret < 0)
td_verror(td, ret);
memcpy(&s, &io_u->start_time, sizeof(s));
- ret = io_u_queue(td, io_u);
+ ret = td_io_queue(td, io_u);
if (ret) {
put_io_u(td, io_u);
td_verror(td, ret);
min_evts = 1;
}
- ret = io_u_getevents(td, min_evts, td->cur_depth, timeout);
+ ret = td_io_getevents(td, min_evts, td->cur_depth, timeout);
if (ret < 0) {
td_verror(td, ret);
break;
}
}
-static int init_io(struct thread_data *td)
+static int td_io_init(struct thread_data *td)
{
if (td->io_ops->init)
return td->io_ops->init(td);
goto err;
}
- if (init_io(td))
+ if (td_io_init(td))
goto err;
if (init_iolog(td))
return NULL;
}
-/*
- * Sets the status of the 'td' in the printed status map.
- */
-static void check_str_update(struct thread_data *td)
-{
- char c = run_str[td->thread_number - 1];
-
- switch (td->runstate) {
- case TD_REAPED:
- c = '_';
- break;
- case TD_EXITED:
- c = 'E';
- break;
- case TD_RUNNING:
- if (td_rw(td)) {
- if (td->sequential)
- c = 'M';
- else
- c = 'm';
- } else if (td_read(td)) {
- if (td->sequential)
- c = 'R';
- else
- c = 'r';
- } else {
- if (td->sequential)
- c = 'W';
- else
- c = 'w';
- }
- break;
- case TD_VERIFYING:
- c = 'V';
- break;
- case TD_FSYNCING:
- c = 'F';
- break;
- case TD_CREATED:
- c = 'C';
- break;
- case TD_INITIALIZED:
- c = 'I';
- break;
- case TD_NOT_CREATED:
- c = 'P';
- break;
- default:
- log_err("state %d\n", td->runstate);
- }
-
- run_str[td->thread_number - 1] = c;
-}
-
-/*
- * Convert seconds to a printable string.
- */
-static void eta_to_str(char *str, int eta_sec)
-{
- unsigned int d, h, m, s;
- static int always_d, always_h;
-
- d = h = m = s = 0;
-
- s = eta_sec % 60;
- eta_sec /= 60;
- m = eta_sec % 60;
- eta_sec /= 60;
- h = eta_sec % 24;
- eta_sec /= 24;
- d = eta_sec;
-
- if (d || always_d) {
- always_d = 1;
- str += sprintf(str, "%02dd:", d);
- }
- if (h || always_h) {
- always_h = 1;
- str += sprintf(str, "%02dh:", h);
- }
-
- str += sprintf(str, "%02dm:", m);
- str += sprintf(str, "%02ds", s);
-}
-
-/*
- * Best effort calculation of the estimated pending runtime of a job.
- */
-static int thread_eta(struct thread_data *td, unsigned long elapsed)
-{
- unsigned long long bytes_total, bytes_done;
- unsigned int eta_sec = 0;
-
- bytes_total = td->total_io_size;
-
- /*
- * if writing, bytes_total will be twice the size. If mixing,
- * assume a 50/50 split and thus bytes_total will be 50% larger.
- */
- if (td->verify) {
- if (td_rw(td))
- bytes_total = bytes_total * 3 / 2;
- else
- bytes_total <<= 1;
- }
- if (td->zone_size && td->zone_skip)
- bytes_total /= (td->zone_skip / td->zone_size);
-
- if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
- double perc;
-
- bytes_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE];
- perc = (double) bytes_done / (double) bytes_total;
- if (perc > 1.0)
- perc = 1.0;
-
- eta_sec = (elapsed * (1.0 / perc)) - elapsed;
-
- if (td->timeout && eta_sec > (td->timeout - elapsed))
- eta_sec = td->timeout - elapsed;
- } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
- || td->runstate == TD_INITIALIZED) {
- int t_eta = 0, r_eta = 0;
-
- /*
- * We can only guess - assume it'll run the full timeout
- * if given, otherwise assume it'll run at the specified rate.
- */
- if (td->timeout)
- t_eta = td->timeout + td->start_delay - elapsed;
- if (td->rate) {
- r_eta = (bytes_total / 1024) / td->rate;
- r_eta += td->start_delay - elapsed;
- }
-
- if (r_eta && t_eta)
- eta_sec = min(r_eta, t_eta);
- else if (r_eta)
- eta_sec = r_eta;
- else if (t_eta)
- eta_sec = t_eta;
- else
- eta_sec = 0;
- } else {
- /*
- * thread is already done or waiting for fsync
- */
- eta_sec = 0;
- }
-
- return eta_sec;
-}
-
-/*
- * Print status of the jobs we know about. This includes rate estimates,
- * ETA, thread state, etc.
- */
-static void print_thread_status(void)
-{
- unsigned long elapsed = time_since_now(&genesis);
- int i, nr_running, nr_pending, t_rate, m_rate, *eta_secs, eta_sec;
- char eta_str[32];
- double perc = 0.0;
-
- if (temp_stall_ts || terse_output)
- return;
-
- eta_secs = malloc(thread_number * sizeof(int));
- memset(eta_secs, 0, thread_number * sizeof(int));
-
- nr_pending = nr_running = t_rate = m_rate = 0;
- for (i = 0; i < thread_number; i++) {
- struct thread_data *td = &threads[i];
-
- if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING||
- td->runstate == TD_FSYNCING) {
- nr_running++;
- t_rate += td->rate;
- m_rate += td->ratemin;
- } else if (td->runstate < TD_RUNNING)
- nr_pending++;
-
- if (elapsed >= 3)
- eta_secs[i] = thread_eta(td, elapsed);
- else
- eta_secs[i] = INT_MAX;
-
- check_str_update(td);
- }
-
- if (exitall_on_terminate)
- eta_sec = INT_MAX;
- else
- eta_sec = 0;
-
- for (i = 0; i < thread_number; i++) {
- if (exitall_on_terminate) {
- if (eta_secs[i] < eta_sec)
- eta_sec = eta_secs[i];
- } else {
- if (eta_secs[i] > eta_sec)
- eta_sec = eta_secs[i];
- }
- }
-
- if (eta_sec != INT_MAX && elapsed) {
- perc = (double) elapsed / (double) (elapsed + eta_sec);
- eta_to_str(eta_str, eta_sec);
- }
-
- if (!nr_running && !nr_pending)
- return;
-
- printf("Threads running: %d", nr_running);
- if (m_rate || t_rate)
- printf(", commitrate %d/%dKiB/sec", t_rate, m_rate);
- if (eta_sec != INT_MAX && nr_running) {
- perc *= 100.0;
- printf(": [%s] [%3.2f%% done] [eta %s]", run_str, perc,eta_str);
- }
- printf("\r");
- fflush(stdout);
- free(eta_secs);
-}
-
/*
* Run over the job map and reap the threads that have exited, if any.
*/
for (i = 0; i < thread_number; i++) {
td = &threads[i];
- run_str[td->thread_number - 1] = 'P';
+ print_status_init(td->thread_number - 1);
init_disk_util(td);
}
}
- gettimeofday(&genesis, NULL);
+ time_init();
while (todo) {
struct thread_data *map[MAX_JOBS];
}
if (td->start_delay) {
- spent = mtime_since_now(&genesis);
+ spent = mtime_since_genesis();
if (td->start_delay * 1000 > spent)
continue;