X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=32bc2652bd0ba93793c9b12cc27ca8b4cf81aef0;hp=035ba73b1d0e2c3423a1d008e25aabc5face4354;hb=002fe73409d1e;hpb=5a812f9b9d987d1cacbb6f17a65835cbe0d66361

diff --git a/backend.c b/backend.c
index 035ba73b..32bc2652 100644
--- a/backend.c
+++ b/backend.c
@@ -52,6 +52,7 @@
 #include "server.h"
 #include "lib/getrusage.h"
 #include "idletime.h"
+#include "err.h"
 
 static pthread_t disk_util_thread;
 static struct fio_mutex *disk_thread_mutex;
@@ -238,8 +239,6 @@ static int check_min_rate(struct thread_data *td, struct timeval *now,
  */
 static void cleanup_pending_aio(struct thread_data *td)
 {
-	struct flist_head *entry, *n;
-	struct io_u *io_u;
 	int r;
 
 	/*
@@ -253,18 +252,11 @@ static void cleanup_pending_aio(struct thread_data *td)
 	 * now cancel remaining active events
 	 */
 	if (td->io_ops->cancel) {
-		flist_for_each_safe(entry, n, &td->io_u_busylist) {
-			io_u = flist_entry(entry, struct io_u, list);
+		struct io_u *io_u;
+		int i;
 
-			/*
-			 * 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 {
+		io_u_qiter(&td->io_u_all, io_u, i) {
+			if (io_u->flags & IO_U_F_FLIGHT) {
 				r = td->io_ops->cancel(td, io_u);
 				if (!r)
 					put_io_u(td, io_u);
@@ -487,6 +479,12 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes)
 				break;
 
 			while ((io_u = get_io_u(td)) != NULL) {
+				if (IS_ERR(io_u)) {
+					io_u = NULL;
+					ret = FIO_Q_BUSY;
+					goto reap;
+				}
+
 				/*
 				 * We are only interested in the places where
 				 * we wrote or trimmed IOs. Turn those into
@@ -583,6 +581,7 @@ sync_done:
 		 * completed io_u's first. Note that we can get BUSY even
 		 * without IO queued, if the system is resource starved.
 		 */
+reap:
 		full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
 		if (full || !td->o.iodepth_batch_complete) {
 			min_events = min(td->o.iodepth_batch_complete,
@@ -651,12 +650,25 @@ static uint64_t do_io(struct thread_data *td)
 	uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
 	unsigned int i;
 	int ret = 0;
+	uint64_t total_bytes, bytes_issued = 0;
 
 	if (in_ramp_time(td))
 		td_set_runstate(td, TD_RAMP);
 	else
 		td_set_runstate(td, TD_RUNNING);
 
+	lat_target_init(td);
+
+	/*
+	 * If verify_backlog is enabled, we'll run the verify in this
+	 * handler as well. For that case, we may need up to twice the
+	 * amount of bytes.
+	 */
+	total_bytes = td->o.size;
+	if (td->o.verify != VERIFY_NONE &&
+	   (td_write(td) && td->o.verify_backlog))
+		total_bytes += td->o.size;
+
 	while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
 		(!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) ||
 		td->o.time_based) {
@@ -684,9 +696,22 @@ static uint64_t do_io(struct thread_data *td)
 		if (flow_threshold_exceeded(td))
 			continue;
 
+		if (bytes_issued >= total_bytes)
+			break;
+
 		io_u = get_io_u(td);
-		if (!io_u)
+		if (IS_ERR_OR_NULL(io_u)) {
+			int err = PTR_ERR(io_u);
+
+			io_u = NULL;
+			if (err == -EBUSY) {
+				ret = FIO_Q_BUSY;
+				goto reap;
+			}
+			if (td->o.latency_target)
+				goto reap;
 			break;
+		}
 
 		ddir = io_u->ddir;
 
@@ -697,6 +722,13 @@ static uint64_t do_io(struct thread_data *td)
 		 */
 		if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ &&
 		    ((io_u->flags & IO_U_F_VER_LIST) || !td_rw(td))) {
+
+			if (!td->o.verify_pattern_bytes) {
+				io_u->rand_seed = __rand(&td->__verify_state);
+				if (sizeof(int) != sizeof(long *))
+					io_u->rand_seed *= __rand(&td->__verify_state);
+			}
+
 			if (td->o.verify_async)
 				io_u->end_io = verify_io_u_async;
 			else
@@ -707,6 +739,17 @@ static uint64_t do_io(struct thread_data *td)
 		else
 			td_set_runstate(td, TD_RUNNING);
 
+		/*
+		 * Always log IO before it's issued, so we know the specific
+		 * order of it. The logged unit will track when the IO has
+		 * completed.
+		 */
+		if (td_write(td) && io_u->ddir == DDIR_WRITE &&
+		    td->o.do_verify &&
+		    td->o.verify != VERIFY_NONE &&
+		    !td->o.experimental_verify)
+			log_io_piece(td, io_u);
+
 		ret = td_io_queue(td, io_u);
 		switch (ret) {
 		case FIO_Q_COMPLETED:
@@ -717,6 +760,7 @@ static uint64_t do_io(struct thread_data *td)
 				int bytes = io_u->xfer_buflen - io_u->resid;
 				struct fio_file *f = io_u->file;
 
+				bytes_issued += bytes;
 				/*
 				 * zero read, fail
 				 */
@@ -747,6 +791,7 @@ sync_done:
 				ret = io_u_sync_complete(td, io_u, bytes_done);
 				if (ret < 0)
 					break;
+				bytes_issued += io_u->xfer_buflen;
 			}
 			break;
 		case FIO_Q_QUEUED:
@@ -757,6 +802,7 @@ sync_done:
 			 */
 			if (td->io_ops->commit == NULL)
 				io_u_queued(td, io_u);
+			bytes_issued += io_u->xfer_buflen;
 			break;
 		case FIO_Q_BUSY:
 			requeue_io_u(td, &io_u);
@@ -778,6 +824,7 @@ sync_done:
 		 * can get BUSY even without IO queued, if the system is
 		 * resource starved.
 		 */
+reap:
 		full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
 		if (full || !td->o.iodepth_batch_complete) {
 			min_evts = min(td->o.iodepth_batch_complete,
@@ -814,6 +861,8 @@ sync_done:
 				break;
 			}
 		}
+		if (!in_ramp_time(td) && td->o.latency_target)
+			lat_target_check(td);
 
 		if (td->o.thinktime) {
 			unsigned long long b;
@@ -822,6 +871,8 @@ sync_done:
 			if (!(b % td->o.thinktime_blocks)) {
 				int left;
 
+				io_u_quiesce(td);
+
 				if (td->o.thinktime_spin)
 					usec_spin(td->o.thinktime_spin);
 
@@ -835,7 +886,7 @@ sync_done:
 	check_update_rusage(td);
 
 	if (td->trim_entries)
-		log_err("fio: %d trim entries leaked?\n", td->trim_entries);
+		log_err("fio: %lu trim entries leaked?\n", td->trim_entries);
 
 	if (td->o.fill_device && td->error == ENOSPC) {
 		td->error = 0;
@@ -876,13 +927,9 @@ sync_done:
 
 static void cleanup_io_u(struct thread_data *td)
 {
-	struct flist_head *entry, *n;
 	struct io_u *io_u;
 
-	flist_for_each_safe(entry, n, &td->io_u_freelist) {
-		io_u = flist_entry(entry, struct io_u, list);
-
-		flist_del(&io_u->list);
+	while ((io_u = io_u_qpop(&td->io_u_freelist)) != NULL) {
 
 		if (td->io_ops->io_u_free)
 			td->io_ops->io_u_free(td, io_u);
@@ -891,6 +938,10 @@ static void cleanup_io_u(struct thread_data *td)
 	}
 
 	free_io_mem(td);
+
+	io_u_rexit(&td->io_u_requeues);
+	io_u_qexit(&td->io_u_freelist);
+	io_u_qexit(&td->io_u_all);
 }
 
 static int init_io_u(struct thread_data *td)
@@ -898,7 +949,7 @@ static int init_io_u(struct thread_data *td)
 	struct io_u *io_u;
 	unsigned int max_bs, min_write;
 	int cl_align, i, max_units;
-	int data_xfer = 1;
+	int data_xfer = 1, err;
 	char *p;
 
 	max_units = td->o.iodepth;
@@ -910,6 +961,26 @@ static int init_io_u(struct thread_data *td)
 	if ((td->io_ops->flags & FIO_NOIO) || !(td_read(td) || td_write(td)))
 		data_xfer = 0;
 
+	err = 0;
+	err += io_u_rinit(&td->io_u_requeues, td->o.iodepth);
+	err += io_u_qinit(&td->io_u_freelist, td->o.iodepth);
+	err += io_u_qinit(&td->io_u_all, td->o.iodepth);
+
+	if (err) {
+		log_err("fio: failed setting up IO queues\n");
+		return 1;
+	}
+
+	/*
+	 * if we may later need to do address alignment, then add any
+	 * possible adjustment here so that we don't cause a buffer
+	 * overflow later. this adjustment may be too much if we get
+	 * lucky and the allocator gives us an aligned address.
+	 */
+	if (td->o.odirect || td->o.mem_align || td->o.oatomic ||
+	    (td->io_ops->flags & FIO_RAWIO))
+		td->orig_buffer_size += page_mask + td->o.mem_align;
+
 	if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) {
 		unsigned long bs;
 
@@ -925,7 +996,7 @@ static int init_io_u(struct thread_data *td)
 	if (data_xfer && allocate_io_mem(td))
 		return 1;
 
-	if (td->o.odirect || td->o.mem_align ||
+	if (td->o.odirect || td->o.mem_align || td->o.oatomic ||
 	    (td->io_ops->flags & FIO_RAWIO))
 		p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align;
 	else
@@ -947,7 +1018,7 @@ static int init_io_u(struct thread_data *td)
 
 		io_u = ptr;
 		memset(io_u, 0, sizeof(*io_u));
-		INIT_FLIST_HEAD(&io_u->list);
+		INIT_FLIST_HEAD(&io_u->verify_list);
 		dprint(FD_MEM, "io_u alloc %p, index %u\n", io_u, i);
 
 		if (data_xfer) {
@@ -961,13 +1032,19 @@ static int init_io_u(struct thread_data *td)
 				 * Fill the buffer with the pattern if we are
 				 * going to be doing writes.
 				 */
-				fill_pattern(td, io_u->buf, max_bs, io_u, 0, 0);
+				fill_verify_pattern(td, io_u->buf, max_bs, io_u, 0, 0);
 			}
 		}
 
 		io_u->index = i;
 		io_u->flags = IO_U_F_FREE;
-		flist_add(&io_u->list, &td->io_u_freelist);
+		io_u_qpush(&td->io_u_freelist, io_u);
+
+		/*
+		 * io_u never leaves this stack, used for iteration of all
+		 * io_u buffers.
+		 */
+		io_u_qpush(&td->io_u_all, io_u);
 
 		if (td->io_ops->io_u_init) {
 			int ret = td->io_ops->io_u_init(td, io_u);
@@ -1062,20 +1139,24 @@ static int keep_running(struct thread_data *td)
 		if (diff < td_max_bs(td))
 			return 0;
 
+		if (fio_files_done(td))
+			return 0;
+
 		return 1;
 	}
 
 	return 0;
 }
 
-static int exec_string(const char *string)
+static int exec_string(struct thread_options *o, const char *string, const char *mode)
 {
-	int ret, newlen = strlen(string) + 1 + 8;
+	int ret, newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1;
 	char *str;
 
 	str = malloc(newlen);
-	sprintf(str, "sh -c %s", string);
+	sprintf(str, "%s &> %s.%s.txt", string, o->name, mode);
 
+	log_info("%s : Saving output of %s in %s.%s.txt\n",o->name, mode, o->name, mode);
 	ret = system(str);
 	if (ret == -1)
 		log_err("fio: exec of cmd <%s> failed\n", str);
@@ -1084,6 +1165,44 @@ static int exec_string(const char *string)
 	return ret;
 }
 
+/*
+ * Dry run to compute correct state of numberio for verification.
+ */
+static uint64_t do_dry_run(struct thread_data *td)
+{
+	uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
+
+	td_set_runstate(td, TD_RUNNING);
+
+	while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
+		(!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td)) {
+		struct io_u *io_u;
+		int ret;
+
+		if (td->terminate || td->done)
+			break;
+
+		io_u = get_io_u(td);
+		if (!io_u)
+			break;
+
+		io_u->flags |= IO_U_F_FLIGHT;
+		io_u->error = 0;
+		io_u->resid = 0;
+		if (ddir_rw(acct_ddir(io_u)))
+			td->io_issues[acct_ddir(io_u)]++;
+		if (ddir_rw(io_u->ddir)) {
+			io_u_mark_depth(td, 1);
+			td->ts.total_io_u[io_u->ddir]++;
+		}
+
+		ret = io_u_sync_complete(td, io_u, bytes_done);
+		(void) ret;
+	}
+
+	return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM];
+}
+
 /*
  * Entry point for the thread based jobs. The process based jobs end up
  * here as well, after a little setup.
@@ -1095,6 +1214,7 @@ static void *thread_main(void *data)
 	struct thread_options *o = &td->o;
 	pthread_condattr_t attr;
 	int clear_state;
+	int ret;
 
 	if (!o->use_thread) {
 		setsid();
@@ -1102,13 +1222,18 @@ static void *thread_main(void *data)
 	} else
 		td->pid = gettid();
 
+	/*
+	 * fio_time_init() may not have been called yet if running as a server
+	 */
+	fio_time_init();
+
 	fio_local_clock_init(o->use_thread);
 
 	dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
 
-	INIT_FLIST_HEAD(&td->io_u_freelist);
-	INIT_FLIST_HEAD(&td->io_u_busylist);
-	INIT_FLIST_HEAD(&td->io_u_requeues);
+	if (is_backend)
+		fio_server_send_start(td);
+
 	INIT_FLIST_HEAD(&td->io_log_list);
 	INIT_FLIST_HEAD(&td->io_hist_list);
 	INIT_FLIST_HEAD(&td->verify_list);
@@ -1128,12 +1253,6 @@ static void *thread_main(void *data)
 	fio_mutex_down(td->mutex);
 	dprint(FD_MUTEX, "done waiting on td->mutex\n");
 
-	/*
-	 * the ->mutex mutex is now no longer used, close it to avoid
-	 * eating a file descriptor
-	 */
-	fio_mutex_remove(td->mutex);
-
 	/*
 	 * A new gid requires privilege, so we need to do this before setting
 	 * the uid.
@@ -1159,7 +1278,8 @@ static void *thread_main(void *data)
 	 * allocations.
 	 */
 	if (o->cpumask_set) {
-		if (fio_setaffinity(td->pid, o->cpumask) == -1) {
+		ret = fio_setaffinity(td->pid, o->cpumask);
+		if (ret == -1) {
 			td_verror(td, errno, "cpu_set_affinity");
 			goto err;
 		}
@@ -1224,8 +1344,9 @@ static void *thread_main(void *data)
 	if (o->verify_async && verify_async_init(td))
 		goto err;
 
-	if (td->ioprio_set) {
-		if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
+	if (o->ioprio) {
+		ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
+		if (ret == -1) {
 			td_verror(td, errno, "ioprio_set");
 			goto err;
 		}
@@ -1252,7 +1373,7 @@ static void *thread_main(void *data)
 	if (init_random_map(td))
 		goto err;
 
-	if (o->exec_prerun && exec_string(o->exec_prerun))
+	if (o->exec_prerun && exec_string(o, o->exec_prerun, (const char *)"prerun"))
 		goto err;
 
 	if (o->pre_read) {
@@ -1288,7 +1409,10 @@ static void *thread_main(void *data)
 
 		prune_io_piece_log(td);
 
-		verify_bytes = do_io(td);
+		if (td->o.verify_only && (td_write(td) || td_rw(td)))
+			verify_bytes = do_dry_run(td);
+		else
+			verify_bytes = do_io(td);
 
 		clear_state = 1;
 
@@ -1375,7 +1499,7 @@ static void *thread_main(void *data)
 
 	fio_mutex_up(writeout_mutex);
 	if (o->exec_postrun)
-		exec_string(o->exec_postrun);
+		exec_string(o, o->exec_postrun, (const char *)"postrun");
 
 	if (exitall_on_terminate)
 		fio_terminate_threads(td->groupid);
@@ -1408,6 +1532,9 @@ err:
 	fio_mutex_remove(td->rusage_sem);
 	td->rusage_sem = NULL;
 
+	fio_mutex_remove(td->mutex);
+	td->mutex = NULL;
+
 	td_set_runstate(td, TD_EXITED);
 	return (void *) (uintptr_t) td->error;
 }
@@ -1536,12 +1663,19 @@ reaped:
 			exit_value++;
 
 		done_secs += mtime_since_now(&td->epoch) / 1000;
+		profile_td_exit(td);
 	}
 
 	if (*nr_running == cputhreads && !pending && realthreads)
 		fio_terminate_threads(TERMINATE_ALL);
 }
 
+static void do_usleep(unsigned int usecs)
+{
+	check_for_running_stats();
+	usleep(usecs);
+}
+
 /*
  * Main function for kicking off and reaping jobs, as needed.
  */
@@ -1553,7 +1687,7 @@ static void run_threads(void)
 
 	if (fio_gtod_offload && fio_start_gtod_thread())
 		return;
-	
+
 	fio_idle_prof_init();
 
 	set_sig_handlers();
@@ -1720,7 +1854,7 @@ static void run_threads(void)
 			if (mtime_since_now(&this_start) > JOB_START_TIMEOUT)
 				break;
 
-			usleep(100000);
+			do_usleep(100000);
 
 			for (i = 0; i < this_jobs; i++) {
 				td = map[i];
@@ -1771,21 +1905,13 @@ static void run_threads(void)
 
 		reap_threads(&nr_running, &t_rate, &m_rate);
 
-		if (todo) {
-			if (is_backend)
-				fio_server_idle_loop();
-			else
-				usleep(100000);
-		}
+		if (todo)
+			do_usleep(100000);
 	}
 
 	while (nr_running) {
 		reap_threads(&nr_running, &t_rate, &m_rate);
-
-		if (is_backend)
-			fio_server_idle_loop();
-		else
-			usleep(10000);
+		do_usleep(10000);
 	}
 
 	fio_idle_prof_stop();
@@ -1881,6 +2007,7 @@ int fio_backend(void)
 		return 1;
 
 	set_genesis_time();
+	stat_init();
 	create_disk_util_thread();
 
 	cgroup_list = smalloc(sizeof(*cgroup_list));
@@ -1910,5 +2037,6 @@ int fio_backend(void)
 	fio_mutex_remove(startup_mutex);
 	fio_mutex_remove(writeout_mutex);
 	fio_mutex_remove(disk_thread_mutex);
+	stat_exit();
 	return exit_value;
 }