X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=fio.c;h=3cf2a9b13220a2e6eece254d98547667f04963e5;hp=dbea661d961ca5b911b0a35161510c238d6f0b2a;hb=07739b57f09886b41323c605b0dbda7d2c12522b;hpb=0904200b6f113233a28c4c1ddf68f216bd922b78

diff --git a/fio.c b/fio.c
index dbea661d..3cf2a9b1 100644
--- a/fio.c
+++ b/fio.c
@@ -46,7 +46,7 @@ int thread_number = 0;
 int shm_id = 0;
 int temp_stall_ts;
 
-static volatile int startup_sem;
+static struct fio_sem *startup_sem;
 static volatile int fio_abort;
 static int exit_value;
 
@@ -60,17 +60,20 @@ static inline void td_set_runstate(struct thread_data *td, int runstate)
 	td->runstate = runstate;
 }
 
-static void terminate_threads(int group_id, int forced_kill)
+static void terminate_threads(int group_id)
 {
 	struct thread_data *td;
 	int i;
 
 	for_each_td(td, i) {
 		if (group_id == TERMINATE_ALL || groupid == td->groupid) {
+			/*
+			 * if the thread is running, just let it exit
+			 */
+			if (td->runstate < TD_RUNNING)
+				kill(td->pid, SIGQUIT);
 			td->terminate = 1;
 			td->start_delay = 0;
-			if (forced_kill)
-				td_set_runstate(td, TD_EXITED);
 		}
 	}
 }
@@ -86,7 +89,7 @@ static void sig_handler(int sig)
 		default:
 			printf("\nfio: terminating on signal %d\n", sig);
 			fflush(stdout);
-			terminate_threads(TERMINATE_ALL, 0);
+			terminate_threads(TERMINATE_ALL);
 			break;
 	}
 }
@@ -100,6 +103,12 @@ static int check_min_rate(struct thread_data *td, struct timeval *now)
 	unsigned long spent;
 	unsigned long rate;
 
+	/*
+	 * No minimum rate set, always ok
+	 */
+	if (!td->ratemin)
+		return 0;
+
 	/*
 	 * allow a 2 second settle period in the beginning
 	 */
@@ -159,7 +168,7 @@ static void cleanup_pending_aio(struct thread_data *td)
 	/*
 	 * 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;
 
@@ -187,7 +196,7 @@ static void cleanup_pending_aio(struct thread_data *td)
 	}
 
 	if (td->cur_depth)
-		r = io_u_queued_complete(td, td->cur_depth, NULL);
+		r = io_u_queued_complete(td, td->cur_depth);
 }
 
 /*
@@ -217,7 +226,7 @@ requeue:
 		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) {
@@ -225,7 +234,7 @@ requeue:
 			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))
@@ -237,7 +246,7 @@ requeue:
 }
 
 /*
- * The main verify engine. Runs over the writes we previusly submitted,
+ * 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)
@@ -264,6 +273,8 @@ static void do_verify(struct thread_data *td)
 
 	io_u = NULL;
 	while (!td->terminate) {
+		int ret2;
+
 		io_u = __get_io_u(td);
 		if (!io_u)
 			break;
@@ -282,29 +293,33 @@ static void do_verify(struct thread_data *td)
 			put_io_u(td, io_u);
 			break;
 		}
-requeue:
-		ret = td_io_queue(td, io_u);
 
+		io_u->end_io = verify_io_u;
+
+		ret = td_io_queue(td, io_u);
 		switch (ret) {
 		case FIO_Q_COMPLETED:
 			if (io_u->error)
 				ret = -io_u->error;
-			if (io_u->xfer_buflen != io_u->resid && io_u->resid) {
+			else if (io_u->xfer_buflen != io_u->resid && io_u->resid) {
 				int bytes = io_u->xfer_buflen - io_u->resid;
 
 				io_u->xfer_buflen = io_u->resid;
 				io_u->xfer_buf += bytes;
-				goto requeue;
+				requeue_io_u(td, &io_u);
+			} else {
+				ret = io_u_sync_complete(td, io_u);
+				if (ret < 0)
+					break;
 			}
-			ret = io_u_sync_complete(td, io_u, verify_io_u);
-			if (ret < 0)
-				break;
 			continue;
 		case FIO_Q_QUEUED:
 			break;
 		case FIO_Q_BUSY:
 			requeue_io_u(td, &io_u);
-			ret = td_io_commit(td);
+			ret2 = td_io_commit(td);
+			if (ret2 < 0)
+				ret = ret2;
 			break;
 		default:
 			assert(ret < 0);
@@ -331,11 +346,16 @@ requeue:
 		 * 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, verify_io_u) < 0)
+		if (io_u_queued_complete(td, min_events) < 0)
 			break;
 	}
 
-	if (td->cur_depth)
+	if (!td->error) {
+		min_events = td->cur_depth;
+
+		if (min_events)
+			ret = io_u_queued_complete(td, min_events);
+	} else
 		cleanup_pending_aio(td);
 
 	td_set_runstate(td, TD_RUNNING);
@@ -383,6 +403,7 @@ static void do_io(struct thread_data *td)
 		long bytes_done = 0;
 		int min_evts = 0;
 		struct io_u *io_u;
+		int ret2;
 
 		if (td->terminate)
 			break;
@@ -397,26 +418,24 @@ static void do_io(struct thread_data *td)
 			put_io_u(td, io_u);
 			break;
 		}
-requeue:
-		ret = td_io_queue(td, io_u);
 
+		ret = td_io_queue(td, io_u);
 		switch (ret) {
 		case FIO_Q_COMPLETED:
-			if (io_u->error) {
-				ret = io_u->error;
-				break;
-			}
-			if (io_u->xfer_buflen != io_u->resid && io_u->resid) {
+			if (io_u->error)
+				ret = -io_u->error;
+			else if (io_u->xfer_buflen != io_u->resid && io_u->resid) {
 				int bytes = io_u->xfer_buflen - io_u->resid;
 
 				io_u->xfer_buflen = io_u->resid;
 				io_u->xfer_buf += bytes;
-				goto requeue;
+				requeue_io_u(td, &io_u);
+			} else {
+				fio_gettime(&comp_time, NULL);
+				bytes_done = io_u_sync_complete(td, io_u);
+				if (bytes_done < 0)
+					ret = bytes_done;
 			}
-			fio_gettime(&comp_time, NULL);
-			bytes_done = io_u_sync_complete(td, io_u, NULL);
-			if (bytes_done < 0)
-				ret = bytes_done;
 			break;
 		case FIO_Q_QUEUED:
 			/*
@@ -429,7 +448,9 @@ requeue:
 			break;
 		case FIO_Q_BUSY:
 			requeue_io_u(td, &io_u);
-			ret = td_io_commit(td);
+			ret2 = td_io_commit(td);
+			if (ret2 < 0)
+				ret = ret2;
 			break;
 		default:
 			assert(ret < 0);
@@ -453,7 +474,7 @@ requeue:
 			}
 
 			fio_gettime(&comp_time, NULL);
-			bytes_done = io_u_queued_complete(td, min_evts, NULL);
+			bytes_done = io_u_queued_complete(td, min_evts);
 			if (bytes_done < 0)
 				break;
 		}
@@ -472,7 +493,7 @@ requeue:
 
 		if (check_min_rate(td, &comp_time)) {
 			if (exitall_on_terminate)
-				terminate_threads(td->groupid, 0);
+				terminate_threads(td->groupid);
 			td_verror(td, ENODATA, "check_min_rate");
 			break;
 		}
@@ -497,15 +518,17 @@ requeue:
 	if (!td->error) {
 		struct fio_file *f;
 
-		if (td->cur_depth)
-			cleanup_pending_aio(td);
+		i = td->cur_depth;
+		if (i)
+			ret = io_u_queued_complete(td, i);
 
 		if (should_fsync(td) && td->end_fsync) {
 			td_set_runstate(td, TD_FSYNCING);
 			for_each_file(td, f, i)
 				fio_io_sync(td, f);
 		}
-	}
+	} else
+		cleanup_pending_aio(td);
 }
 
 static void cleanup_io_u(struct thread_data *td)
@@ -633,10 +656,10 @@ static int switch_ioscheduler(struct thread_data *td)
 	return 0;
 }
 
-static void clear_io_state(struct thread_data *td)
+static int clear_io_state(struct thread_data *td)
 {
 	struct fio_file *f;
-	int i;
+	int i, ret;
 
 	td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0;
 	td->this_io_bytes[0] = td->this_io_bytes[1] = 0;
@@ -644,16 +667,17 @@ static void clear_io_state(struct thread_data *td)
 
 	td->last_was_sync = 0;
 
-	for_each_file(td, f, i) {
-		f->last_completed_pos = 0;
-
-		f->last_pos = 0;
-		if (td->io_ops->flags & FIO_SYNCIO)
-			lseek(f->fd, SEEK_SET, 0);
+	for_each_file(td, f, i)
+		td_io_close_file(td, f);
 
-		if (f->file_map)
-			memset(f->file_map, 0, f->num_maps * sizeof(long));
+	ret = 0;
+	for_each_file(td, f, i) {
+		ret = td_io_open_file(td, f);
+		if (ret)
+			break;
 	}
+
+	return ret;
 }
 
 /*
@@ -664,6 +688,7 @@ static void *thread_main(void *data)
 {
 	unsigned long long runtime[2];
 	struct thread_data *td = data;
+	int clear_state;
 
 	if (!td->use_thread)
 		setsid();
@@ -706,21 +731,18 @@ static void *thread_main(void *data)
 		goto err;
 
 	td_set_runstate(td, TD_INITIALIZED);
-	fio_sem_up(&startup_sem);
-	fio_sem_down(&td->mutex);
+	fio_sem_up(startup_sem);
+	fio_sem_down(td->mutex);
 
 	if (!td->create_serialize && setup_files(td))
 		goto err;
-	if (open_files(td))
-		goto err;
 
-	/*
-	 * Do this late, as some IO engines would like to have the
-	 * files setup prior to initializing structures.
-	 */
 	if (td_io_init(td))
 		goto err;
 
+	if (open_files(td))
+		goto err;
+
 	if (td->exec_prerun) {
 		if (system(td->exec_prerun) < 0)
 			goto err;
@@ -731,6 +753,7 @@ static void *thread_main(void *data)
 	getrusage(RUSAGE_SELF, &td->ts.ru_start);
 
 	runtime[0] = runtime[1] = 0;
+	clear_state = 0;
 	while (td->loops--) {
 		fio_gettime(&td->start, NULL);
 		memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
@@ -738,7 +761,9 @@ static void *thread_main(void *data)
 		if (td->ratemin)
 			memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate));
 
-		clear_io_state(td);
+		if (clear_state && clear_io_state(td))
+			break;
+
 		prune_io_piece_log(td);
 
 		if (td->io_ops->flags & FIO_CPUIO)
@@ -746,6 +771,8 @@ static void *thread_main(void *data)
 		else
 			do_io(td);
 
+		clear_state = 1;
+
 		if (td_read(td) && td->io_bytes[DDIR_READ])
 			runtime[DDIR_READ] += utime_since_now(&td->start);
 		if (td_write(td) && td->io_bytes[DDIR_WRITE])
@@ -757,7 +784,9 @@ static void *thread_main(void *data)
 		if (td->verify == VERIFY_NONE)
 			continue;
 
-		clear_io_state(td);
+		if (clear_io_state(td))
+			break;
+
 		fio_gettime(&td->start, NULL);
 
 		do_verify(td);
@@ -769,9 +798,11 @@ static void *thread_main(void *data)
 	}
 
 	update_rusage_stat(td);
-	fio_gettime(&td->end_time, NULL);
-	td->runtime[0] = runtime[0] / 1000;
-	td->runtime[1] = runtime[1] / 1000;
+	td->ts.runtime[0] = runtime[0] / 1000;
+	td->ts.runtime[1] = runtime[1] / 1000;
+	td->ts.total_run_time = mtime_since_now(&td->epoch);
+	td->ts.io_bytes[0] = td->io_bytes[0];
+	td->ts.io_bytes[1] = td->io_bytes[1];
 
 	if (td->ts.bw_log)
 		finish_log(td, td->ts.bw_log, "bw");
@@ -787,7 +818,7 @@ static void *thread_main(void *data)
 	}
 
 	if (exitall_on_terminate)
-		terminate_threads(td->groupid, 0);
+		terminate_threads(td->groupid);
 
 err:
 	if (td->error)
@@ -873,7 +904,8 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 			if (WIFSIGNALED(status)) {
 				int sig = WTERMSIG(status);
 
-				log_err("fio: pid=%d, got signal=%d\n", td->pid, sig);
+				if (sig != SIGQUIT)
+					log_err("fio: pid=%d, got signal=%d\n", td->pid, sig);
 				td_set_runstate(td, TD_REAPED);
 				goto reaped;
 			}
@@ -898,6 +930,8 @@ reaped:
 				perror("pthread_join");
 		}
 
+		fio_sem_remove(td->mutex);
+
 		(*nr_running)--;
 		(*m_rate) -= td->ratemin;
 		(*t_rate) -= td->rate;
@@ -907,7 +941,7 @@ reaped:
 	}
 
 	if (*nr_running == cputhreads && !pending)
-		terminate_threads(TERMINATE_ALL, 0);
+		terminate_threads(TERMINATE_ALL);
 }
 
 /*
@@ -998,7 +1032,6 @@ static void run_threads(void)
 			 */
 			td_set_runstate(td, TD_CREATED);
 			map[this_jobs++] = td;
-			fio_sem_init(&startup_sem, 1);
 			nr_started++;
 
 			if (td->use_thread) {
@@ -1007,14 +1040,13 @@ static void run_threads(void)
 					nr_started--;
 				}
 			} else {
-				if (fork())
-					fio_sem_down(&startup_sem);
-				else {
+				if (!fork()) {
 					int ret = fork_main(shm_id, i);
 
 					exit(ret);
 				}
 			}
+			fio_sem_down(startup_sem);
 		}
 
 		/*
@@ -1069,7 +1101,7 @@ static void run_threads(void)
 			m_rate += td->ratemin;
 			t_rate += td->rate;
 			todo--;
-			fio_sem_up(&td->mutex);
+			fio_sem_up(td->mutex);
 		}
 
 		reap_threads(&nr_running, &t_rate, &m_rate);
@@ -1119,6 +1151,8 @@ int main(int argc, char *argv[])
 		setup_log(&agg_io_log[DDIR_WRITE]);
 	}
 
+	startup_sem = fio_sem_init(0);
+
 	set_genesis_time();
 
 	disk_util_timer_arm();
@@ -1133,5 +1167,6 @@ int main(int argc, char *argv[])
 		}
 	}
 
+	fio_sem_remove(startup_sem);
 	return exit_value;
 }