X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=fio.c;h=0e1eadb35ce2d11555ee1a1ad3213083d136c5c5;hb=faf5c880f6ddcd44ff896ba1cf3da38b23fe0f39;hp=ae89bb9388d36973c1cad8b36aba1496c52ed206;hpb=f692a7677e025a274ad08ecfa6a3001e61b1232d;p=fio.git

diff --git a/fio.c b/fio.c
index ae89bb93..0e1eadb3 100644
--- a/fio.c
+++ b/fio.c
@@ -160,9 +160,19 @@ static void cleanup_pending_aio(struct thread_data *td)
 		list_for_each_safe(entry, n, &td->io_u_busylist) {
 			io_u = list_entry(entry, struct io_u, list);
 
-			r = td->io_ops->cancel(td, io_u);
-			if (!r)
+			/*
+			 * if the io_u isn't in flight, then that generally
+			 * means someone leaked an io_u. complain but fix
+			 * it up, so we don't stall here.
+			 */
+			if ((io_u->flags & IO_U_F_FLIGHT) == 0) {
+				log_err("fio: non-busy IO on busy list\n");
 				put_io_u(td, io_u);
+			} else {
+				r = td->io_ops->cancel(td, io_u);
+				if (!r)
+					put_io_u(td, io_u);
+			}
 		}
 	}
 
@@ -197,7 +207,7 @@ requeue:
 		put_io_u(td, io_u);
 		return 1;
 	} else if (ret == FIO_Q_QUEUED) {
-		if (io_u_queued_complete(td, 1, NULL))
+		if (io_u_queued_complete(td, 1, NULL) < 0)
 			return 1;
 	} else if (ret == FIO_Q_COMPLETED) {
 		if (io_u->error) {
@@ -248,15 +258,20 @@ static void do_verify(struct thread_data *td)
 		if (!io_u)
 			break;
 
-		if (runtime_exceeded(td, &io_u->start_time))
+		if (runtime_exceeded(td, &io_u->start_time)) {
+			put_io_u(td, io_u);
 			break;
+		}
 
-		if (get_next_verify(td, io_u))
+		if (get_next_verify(td, io_u)) {
+			put_io_u(td, io_u);
 			break;
+		}
 
-		if (td_io_prep(td, io_u))
+		if (td_io_prep(td, io_u)) {
+			put_io_u(td, io_u);
 			break;
-
+		}
 requeue:
 		ret = td_io_queue(td, io_u);
 
@@ -272,7 +287,7 @@ requeue:
 				goto requeue;
 			}
 			ret = io_u_sync_complete(td, io_u, verify_io_u);
-			if (ret)
+			if (ret < 0)
 				break;
 			continue;
 		case FIO_Q_QUEUED:
@@ -306,7 +321,7 @@ 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))
+		if (io_u_queued_complete(td, min_events, verify_io_u) < 0)
 			break;
 	}
 
@@ -548,9 +563,12 @@ static int init_io_u(struct thread_data *td)
 			fill_rand_buf(io_u, max_bs);
 
 		io_u->index = i;
+		io_u->flags = IO_U_F_FREE;
 		list_add(&io_u->list, &td->io_u_freelist);
 	}
 
+	io_u_init_timeout();
+
 	return 0;
 }
 
@@ -699,6 +717,7 @@ static void *thread_main(void *data)
 	}
 
 	fio_gettime(&td->epoch, NULL);
+	memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch));
 	getrusage(RUSAGE_SELF, &td->ts.ru_start);
 
 	runtime[0] = runtime[1] = 0;
@@ -805,6 +824,8 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 	 */
 	pending = cputhreads = 0;
 	for_each_td(td, i) {
+		int flags = 0;
+
 		/*
 		 * ->io_ops is NULL for a thread that has closed its
 		 * io engine
@@ -812,55 +833,66 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 		if (td->io_ops && td->io_ops->flags & FIO_CPUIO)
 			cputhreads++;
 
-		if (td->runstate < TD_EXITED) {
-			/*
-			 * check if someone quit or got killed in an unusual way
-			 */
-			ret = waitpid(td->pid, &status, WNOHANG);
-			if (ret < 0)
-				perror("waitpid");
-			else if ((ret == td->pid) && WIFSIGNALED(status)) {
-				int sig = WTERMSIG(status);
-
-				log_err("fio: pid=%d, got signal=%d\n", td->pid, sig);
+		if (!td->pid || td->runstate == TD_REAPED)
+			continue;
+		if (td->use_thread) {
+			if (td->runstate == TD_EXITED) {
 				td_set_runstate(td, TD_REAPED);
 				goto reaped;
 			}
+			continue;
 		}
 
-		if (td->runstate != TD_EXITED) {
-			if (td->runstate < TD_RUNNING)
-				pending++;
+		flags = WNOHANG;
+		if (td->runstate == TD_EXITED)
+			flags = 0;
 
-			continue;
-		}
+		/*
+		 * check if someone quit or got killed in an unusual way
+		 */
+		ret = waitpid(td->pid, &status, flags);
+		if (ret < 0) {
+			if (errno == ECHILD) {
+				log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate);
+				td_set_runstate(td, TD_REAPED);
+				goto reaped;
+			}
+			perror("waitpid");
+		} else if (ret == td->pid) {
+			if (WIFSIGNALED(status)) {
+				int sig = WTERMSIG(status);
 
-		if (td->error)
-			exit_value++;
+				log_err("fio: pid=%d, got signal=%d\n", td->pid, sig);
+				td_set_runstate(td, TD_REAPED);
+				goto reaped;
+			}
+			if (WIFEXITED(status)) {
+				if (WEXITSTATUS(status) && !td->error)
+					td->error = WEXITSTATUS(status);
 
-		td_set_runstate(td, TD_REAPED);
+				td_set_runstate(td, TD_REAPED);
+				goto reaped;
+			}
+		}
 
+		/*
+		 * thread is not dead, continue
+		 */
+		continue;
+reaped:
 		if (td->use_thread) {
 			long ret;
 
 			if (pthread_join(td->thread, (void *) &ret))
-				perror("thread_join");
-		} else {
-			int status;
-
-			ret = waitpid(td->pid, &status, 0);
-			if (ret < 0)
-				perror("waitpid");
-			else if (WIFEXITED(status) && WEXITSTATUS(status)) {
-				if (!exit_value)
-					exit_value++;
-			}
+				perror("pthread_join");
 		}
 
-reaped:
 		(*nr_running)--;
 		(*m_rate) -= td->ratemin;
 		(*t_rate) -= td->rate;
+
+		if (td->error)
+			exit_value++;
 	}
 
 	if (*nr_running == cputhreads && !pending)
@@ -916,6 +948,8 @@ static void run_threads(void)
 		init_disk_util(td);
 	}
 
+	set_genesis_time();
+
 	while (todo) {
 		struct thread_data *map[MAX_JOBS];
 		struct timeval this_start;
@@ -1074,6 +1108,8 @@ int main(int argc, char *argv[])
 		setup_log(&agg_io_log[DDIR_WRITE]);
 	}
 
+	set_genesis_time();
+
 	disk_util_timer_arm();
 
 	run_threads();