X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=fio.c;h=3e55f92b31f49199f444ac62c5ce87890c7d85f6;hp=84819d9f51e996b2a25c35f1da29a62772b99e8b;hb=22819ec237297fc39435ed566bee01a4225bfb39;hpb=6ce15a32b28f359b3753ef65d883ab2d824e19f2

diff --git a/fio.c b/fio.c
index 84819d9f..3e55f92b 100644
--- a/fio.c
+++ b/fio.c
@@ -26,20 +26,20 @@
 #include <string.h>
 #include <signal.h>
 #include <time.h>
+#include <locale.h>
 #include <assert.h>
 #include <sys/stat.h>
 #include <sys/wait.h>
 #include <sys/ipc.h>
 #include <sys/shm.h>
-#include <sys/ioctl.h>
 #include <sys/mman.h>
 
 #include "fio.h"
 #include "os.h"
 
-#define MASK	(4095)
-
-#define ALIGN(buf)	(char *) (((unsigned long) (buf) + MASK) & ~(MASK))
+static unsigned long page_mask;
+#define ALIGN(buf)	\
+	(char *) (((unsigned long) (buf) + page_mask) & ~page_mask)
 
 int groupid = 0;
 int thread_number = 0;
@@ -48,6 +48,9 @@ int temp_stall_ts;
 
 static volatile int startup_sem;
 static volatile int fio_abort;
+static int exit_value;
+
+struct io_log *agg_io_log[2];
 
 #define TERMINATE_ALL		(-1)
 #define JOB_START_TIMEOUT	(5 * 1000)
@@ -80,11 +83,6 @@ static void sig_handler(int sig)
 			disk_util_timer_arm();
 			print_thread_status();
 			break;
-		case SIGSEGV:
-			fprintf(stderr, "fio: got segfault, aborting\n");
-			terminate_threads(TERMINATE_ALL, 1);
-			fio_abort = 1;
-			exit(0);
 		default:
 			printf("\nfio: terminating on signal %d\n", sig);
 			fflush(stdout);
@@ -176,7 +174,7 @@ static void cleanup_pending_aio(struct thread_data *td)
 	 */
 	r = td_io_getevents(td, 0, td->cur_depth, &ts);
 	if (r > 0) {
-		icd.nr = r;
+		init_icd(&icd, NULL, r);
 		ios_completed(td, &icd);
 	}
 
@@ -196,7 +194,7 @@ static void cleanup_pending_aio(struct thread_data *td)
 	if (td->cur_depth) {
 		r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL);
 		if (r > 0) {
-			icd.nr = r;
+			init_icd(&icd, NULL, r);
 			ios_completed(td, &icd);
 		}
 	}
@@ -224,23 +222,32 @@ static int fio_io_sync(struct thread_data *td, struct fio_file *f)
 	}
 
 	ret = td_io_queue(td, io_u);
-	if (ret) {
+	if (ret < 0) {
 		td_verror(td, io_u->error);
 		put_io_u(td, io_u);
 		return 1;
-	}
+	} else if (ret == FIO_Q_QUEUED) {
+		ret = td_io_getevents(td, 1, td->cur_depth, NULL);
+		if (ret < 0) {
+			td_verror(td, -ret);
+			return 1;
+		}
 
-	ret = td_io_getevents(td, 1, td->cur_depth, NULL);
-	if (ret < 0) {
-		td_verror(td, ret);
-		return 1;
-	}
+		init_icd(&icd, NULL, ret);
+		ios_completed(td, &icd);
+		if (icd.error) {
+			td_verror(td, icd.error);
+			return 1;
+		}
+	} else if (ret == FIO_Q_COMPLETED) {
+		if (io_u->error) {
+			td_verror(td, io_u->error);
+			return 1;
+		}
 
-	icd.nr = ret;
-	ios_completed(td, &icd);
-	if (icd.error) {
-		td_verror(td, icd.error);
-		return 1;
+		init_icd(&icd, NULL, 1);
+		io_completed(td, io_u, &icd);
+		put_io_u(td, io_u);
 	}
 
 	return 0;
@@ -252,10 +259,9 @@ static int fio_io_sync(struct thread_data *td, struct fio_file *f)
  */
 static void do_verify(struct thread_data *td)
 {
-	struct io_u *io_u, *v_io_u = NULL;
-	struct io_completion_data icd;
 	struct fio_file *f;
-	int ret, i;
+	struct io_u *io_u;
+	int ret, i, min_events;
 
 	/*
 	 * sync io first and invalidate cache, to make sure we really
@@ -268,78 +274,95 @@ static void do_verify(struct thread_data *td)
 
 	td_set_runstate(td, TD_VERIFYING);
 
-	do {
-		if (td->terminate)
-			break;
+	io_u = NULL;
+	while (!td->terminate) {
+		struct io_completion_data icd;
+		struct timespec *timeout;
 
 		io_u = __get_io_u(td);
 		if (!io_u)
 			break;
 
-		if (runtime_exceeded(td, &io_u->start_time)) {
-			put_io_u(td, io_u);
+		if (runtime_exceeded(td, &io_u->start_time))
 			break;
-		}
 
-		if (get_next_verify(td, io_u)) {
-			put_io_u(td, io_u);
+		if (get_next_verify(td, io_u))
 			break;
-		}
 
-		f = get_next_file(td);
-		if (!f)
+		if (td_io_prep(td, io_u))
 			break;
 
-		io_u->file = f;
+requeue:
+		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) {
+				int bytes = io_u->xfer_buflen - io_u->resid;
 
-		if (td_io_prep(td, io_u)) {
+				io_u->xfer_buflen = io_u->resid;
+				io_u->xfer_buf += bytes;
+				goto requeue;
+			}
+			init_icd(&icd, verify_io_u, 1);
+			io_completed(td, io_u, &icd);
+			if (icd.error) {
+				ret = icd.error;
+				break;
+			}
 			put_io_u(td, io_u);
+			continue;
+		case FIO_Q_QUEUED:
+			break;
+		default:
+			assert(ret < 0);
+			td_verror(td, -ret);
 			break;
 		}
 
-		ret = td_io_queue(td, io_u);
-		if (ret) {
-			td_verror(td, io_u->error);
-			put_io_u(td, io_u);
+		if (ret < 0)
 			break;
-		}
 
 		/*
-		 * we have one pending to verify, do that while
-		 * we are doing io on the next one
+		 * if we can queue more, do so. but check if there are
+		 * completed io_u's first.
 		 */
-		if (do_io_u_verify(td, &v_io_u))
-			break;
+		if (queue_full(td)) {
+			timeout = NULL;
+			min_events = 1;
+		} else {
+			struct timespec ts;
 
-		ret = td_io_getevents(td, 1, 1, NULL);
-		if (ret != 1) {
-			if (ret < 0)
-				td_verror(td, ret);
-			break;
+			ts.tv_sec = 0;
+			ts.tv_nsec = 0;
+			timeout = &ts;
+			min_events = 0;
 		}
 
-		v_io_u = td->io_ops->event(td, 0);
-		icd.nr = 1;
-		icd.error = 0;
-		fio_gettime(&icd.time, NULL);
-		io_completed(td, v_io_u, &icd);
+		/*
+		 * Reap required number of io units, if any, and do the
+		 * verification on them through the callback handler
+		 */
+		ret = td_io_getevents(td, min_events, td->cur_depth, timeout);
+		if (ret < 0) {
+			td_verror(td, -ret);
+			break;
+		} else if (!ret)
+			continue;
+
+		init_icd(&icd, verify_io_u, ret);
+		ios_completed(td, &icd);
 
 		if (icd.error) {
 			td_verror(td, icd.error);
-			put_io_u(td, v_io_u);
-			v_io_u = NULL;
 			break;
 		}
+	}
 
-		/*
-		 * if we can't submit more io, we need to verify now
-		 */
-		if (queue_full(td) && do_io_u_verify(td, &v_io_u))
-			break;
-
-	} while (1);
-
-	do_io_u_verify(td, &v_io_u);
+	if (io_u)
+		put_io_u(td, io_u);
 
 	if (td->cur_depth)
 		cleanup_pending_aio(td);
@@ -386,7 +409,7 @@ static void do_io(struct thread_data *td)
 
 	td_set_runstate(td, TD_RUNNING);
 
-	while (td->this_io_bytes[td->ddir] < td->io_size) {
+	while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) {
 		struct timespec *timeout;
 		int min_evts = 0;
 		struct io_u *io_u;
@@ -403,38 +426,65 @@ static void do_io(struct thread_data *td)
 			break;
 
 		memcpy(&s, &io_u->start_time, sizeof(s));
-
+requeue:
 		ret = td_io_queue(td, io_u);
-		if (ret) {
-			td_verror(td, io_u->error);
+
+		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) {
+				int bytes = io_u->xfer_buflen - io_u->resid;
+
+				io_u->xfer_buflen = io_u->resid;
+				io_u->xfer_buf += bytes;
+				goto requeue;
+			}
+			init_icd(&icd, NULL, 1);
+			io_completed(td, io_u, &icd);
+			put_io_u(td, io_u);
+			break;
+		case FIO_Q_QUEUED:
+			break;
+		default:
+			assert(ret < 0);
 			put_io_u(td, io_u);
 			break;
 		}
 
+		if (ret < 0)
+			break;
+
 		add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time));
 
-		if (td->cur_depth < td->iodepth) {
-			struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
+		if (ret == FIO_Q_QUEUED) {
+			if (td->cur_depth < td->iodepth) {
+				struct timespec ts;
 
-			timeout = &ts;
-			min_evts = 0;
-		} else {
-			timeout = NULL;
-			min_evts = 1;
-		}
+				ts.tv_sec = 0;
+				ts.tv_nsec = 0;
+				timeout = &ts;
+				min_evts = 0;
+			} else {
+				timeout = NULL;
+				min_evts = 1;
+			}
 
-		ret = td_io_getevents(td, min_evts, td->cur_depth, timeout);
-		if (ret < 0) {
-			td_verror(td, ret);
-			break;
-		} else if (!ret)
-			continue;
+			ret = td_io_getevents(td, min_evts, td->cur_depth, timeout);
+			if (ret < 0) {
+				td_verror(td, -ret);
+				break;
+			} else if (!ret)
+				continue;
 
-		icd.nr = ret;
-		ios_completed(td, &icd);
-		if (icd.error) {
-			td_verror(td, icd.error);
-			break;
+			init_icd(&icd, NULL, ret);
+			ios_completed(td, &icd);
+			if (icd.error) {
+				td_verror(td, icd.error);
+				break;
+			}
 		}
 
 		/*
@@ -449,7 +499,7 @@ static void do_io(struct thread_data *td)
 		if (check_min_rate(td, &icd.time)) {
 			if (exitall_on_terminate)
 				terminate_threads(td->groupid, 0);
-			td_verror(td, ENOMEM);
+			td_verror(td, ENODATA);
 			break;
 		}
 
@@ -460,8 +510,16 @@ static void do_io(struct thread_data *td)
 			unsigned long long b;
 
 			b = td->io_blocks[0] + td->io_blocks[1];
-			if (!(b % td->thinktime_blocks))
-				usec_sleep(td, td->thinktime);
+			if (!(b % td->thinktime_blocks)) {
+				int left;
+
+				if (td->thinktime_spin)
+					__usec_sleep(td->thinktime_spin);
+
+				left = td->thinktime - td->thinktime_spin;
+				if (left)
+					usec_sleep(td, left);
+			}
 		}
 	}
 
@@ -526,7 +584,7 @@ static int init_io_u(struct thread_data *td)
 	if (td->mem_type == MEM_SHMHUGE || td->mem_type == MEM_MMAPHUGE)
 		td->orig_buffer_size = (td->orig_buffer_size + td->hugepage_size - 1) & ~(td->hugepage_size - 1);
 	else
-		td->orig_buffer_size += MASK;
+		td->orig_buffer_size += page_mask;
 
 	if (allocate_io_mem(td))
 		return 1;
@@ -645,9 +703,6 @@ static void *thread_main(void *data)
 		goto err;
 	}
 
-	if (td_io_init(td))
-		goto err;
-
 	if (init_iolog(td))
 		goto err;
 
@@ -678,8 +733,17 @@ static void *thread_main(void *data)
 	if (open_files(td))
 		goto err;
 
-	if (td->exec_prerun)
-		system(td->exec_prerun);
+	/*
+	 * 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 (td->exec_prerun) {
+		if (system(td->exec_prerun) < 0)
+			goto err;
+	}
 
 	fio_gettime(&td->epoch, NULL);
 	getrusage(RUSAGE_SELF, &td->ru_start);
@@ -734,40 +798,45 @@ static void *thread_main(void *data)
 		finish_log(td, td->clat_log, "clat");
 	if (td->write_iolog_file)
 		write_iolog_close(td);
-	if (td->exec_postrun)
-		system(td->exec_postrun);
+	if (td->exec_postrun) {
+		if (system(td->exec_postrun) < 0)
+			log_err("fio: postrun %s failed\n", td->exec_postrun);
+	}
 
 	if (exitall_on_terminate)
 		terminate_threads(td->groupid, 0);
 
 err:
+	if (td->error)
+		printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
 	close_files(td);
 	close_ioengine(td);
 	cleanup_io_u(td);
 	td_set_runstate(td, TD_EXITED);
-	return NULL;
-
+	return (void *) td->error;
 }
 
 /*
  * We cannot pass the td data into a forked process, so attach the td and
  * pass it to the thread worker.
  */
-static void *fork_main(int shmid, int offset)
+static int fork_main(int shmid, int offset)
 {
 	struct thread_data *td;
-	void *data;
+	void *data, *ret;
 
 	data = shmat(shmid, NULL, 0);
 	if (data == (void *) -1) {
+		int __err = errno;
+
 		perror("shmat");
-		return NULL;
+		return __err;
 	}
 
 	td = data + offset * sizeof(struct thread_data);
-	thread_main(td);
+	ret = thread_main(td);
 	shmdt(data);
-	return NULL;
+	return (int) ret;
 }
 
 /*
@@ -776,7 +845,7 @@ static void *fork_main(int shmid, int offset)
 static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 {
 	struct thread_data *td;
-	int i, cputhreads, pending;
+	int i, cputhreads, pending, status, ret;
 
 	/*
 	 * reap exited threads (TD_EXITED -> TD_REAPED)
@@ -790,6 +859,22 @@ 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);
+				td_set_runstate(td, TD_REAPED);
+				goto reaped;
+			}
+		}
+
 		if (td->runstate != TD_EXITED) {
 			if (td->runstate < TD_RUNNING)
 				pending++;
@@ -797,6 +882,9 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 			continue;
 		}
 
+		if (td->error)
+			exit_value++;
+
 		td_set_runstate(td, TD_REAPED);
 
 		if (td->use_thread) {
@@ -804,9 +892,19 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 
 			if (pthread_join(td->thread, (void *) &ret))
 				perror("thread_join");
-		} else
-			waitpid(td->pid, NULL, 0);
+		} 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++;
+			}
+		}
+
+reaped:
 		(*nr_running)--;
 		(*m_rate) -= td->ratemin;
 		(*t_rate) -= td->rate;
@@ -835,7 +933,6 @@ static void run_threads(void)
 
 	signal(SIGINT, sig_handler);
 	signal(SIGALRM, sig_handler);
-	signal(SIGSEGV, sig_handler);
 
 	todo = thread_number;
 	nr_running = 0;
@@ -856,6 +953,9 @@ static void run_threads(void)
 		 * client data interspersed on disk
 		 */
 		if (setup_files(td)) {
+			exit_value++;
+			if (td->error)
+				log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
 			td_set_runstate(td, TD_REAPED);
 			todo--;
 		}
@@ -912,8 +1012,9 @@ static void run_threads(void)
 				if (fork())
 					fio_sem_down(&startup_sem);
 				else {
-					fork_main(shm_id, i);
-					exit(0);
+					int ret = fork_main(shm_id, i);
+
+					exit(ret);
 				}
 			}
 		}
@@ -990,6 +1091,15 @@ static void run_threads(void)
 
 int main(int argc, char *argv[])
 {
+	long ps;
+
+	/*
+	 * We need locale for number printing, if it isn't set then just
+	 * go with the US format.
+	 */
+	if (!getenv("LC_NUMERIC"))
+		setlocale(LC_NUMERIC, "en_US");
+
 	if (parse_options(argc, argv))
 		return 1;
 
@@ -998,12 +1108,30 @@ int main(int argc, char *argv[])
 		return 1;
 	}
 
+	ps = sysconf(_SC_PAGESIZE);
+	if (ps < 0) {
+		log_err("Failed to get page size\n");
+		return 1;
+	}
+
+	page_mask = ps - 1;
+
+	if (write_bw_log) {
+		setup_log(&agg_io_log[DDIR_READ]);
+		setup_log(&agg_io_log[DDIR_WRITE]);
+	}
+
 	disk_util_timer_arm();
 
 	run_threads();
 
-	if (!fio_abort)
+	if (!fio_abort) {
 		show_run_stats();
+		if (write_bw_log) {
+			__finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log");
+			__finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log");
+		}
+	}
 
-	return 0;
+	return exit_value;
 }