X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=backend.c;h=b180196e4c4ec5b16d61120efaf14c9a2abdbf1f;hb=01cfefcc6255420a9e19d5774053ed2fb705b150;hp=65a3e184a6d77d7d5439f23c9a4c17813a405326;hpb=36870c47801ca85d1a98c7ec47950b0aa360f865;p=fio.git

diff --git a/backend.c b/backend.c
index 65a3e184..b180196e 100644
--- a/backend.c
+++ b/backend.c
@@ -55,6 +55,7 @@
 #include "err.h"
 #include "lib/tp.h"
 #include "workqueue.h"
+#include "lib/mountcheck.h"
 
 static pthread_t helper_thread;
 static pthread_mutex_t helper_lock;
@@ -99,7 +100,7 @@ static void sig_int(int sig)
 	}
 }
 
-static void sig_show_status(int sig)
+void sig_show_status(int sig)
 {
 	show_running_run_stats();
 }
@@ -178,7 +179,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
 		if (spent < td->o.ratecycle)
 			return 0;
 
-		if (td->o.rate[ddir]) {
+		if (td->o.rate[ddir] || td->o.ratemin[ddir]) {
 			/*
 			 * check bandwidth specified rate
 			 */
@@ -219,6 +220,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
 					log_err("%s: min iops rate %u not met,"
 						" got %lu\n", td->o.name,
 							rate_iops_min, rate);
+					return 1;
 				}
 			}
 		}
@@ -363,6 +365,23 @@ static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
 	return 0;
 }
 
+/*
+ * We need to update the runtime consistently in ms, but keep a running
+ * tally of the current elapsed time in microseconds for sub millisecond
+ * updates.
+ */
+static inline void update_runtime(struct thread_data *td,
+				  unsigned long long *elapsed_us,
+				  const enum fio_ddir ddir)
+{
+	if (ddir == DDIR_WRITE && td_write(td) && td->o.verify_only)
+		return;
+
+	td->ts.runtime[ddir] -= (elapsed_us[ddir] + 999) / 1000;
+	elapsed_us[ddir] += utime_since_now(&td->start);
+	td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000;
+}
+
 static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
 			       int *retptr)
 {
@@ -427,8 +446,8 @@ static int wait_for_completions(struct thread_data *td, struct timeval *time)
 	/*
 	 * if the queue is full, we MUST reap at least 1 event
 	 */
-	min_evts = min(td->o.iodepth_batch_complete, td->cur_depth);
-	if (full && !min_evts)
+	min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth);
+    if ((full && !min_evts) || !td->o.iodepth_batch_complete_min)
 		min_evts = 1;
 
 	if (time && (__should_check_rate(td, DDIR_READ) ||
@@ -521,7 +540,7 @@ sync_done:
 			*ret = ret2;
 		break;
 	default:
-		assert(ret < 0);
+		assert(*ret < 0);
 		td_verror(td, -(*ret), "td_io_queue");
 		break;
 	}
@@ -532,6 +551,12 @@ sync_done:
 	return 0;
 }
 
+static inline int io_in_polling(struct thread_data *td)
+{
+	return !td->o.iodepth_batch_complete_min &&
+		   !td->o.iodepth_batch_complete_max;
+}
+
 /*
  * The main verify engine. Runs over the writes we previously submitted,
  * reads the blocks back in, and checks the crc/md5 of the data.
@@ -665,7 +690,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes)
 		 */
 reap:
 		full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
-		if (full || !td->o.iodepth_batch_complete)
+		if (full || io_in_polling(td))
 			ret = wait_for_completions(td, NULL);
 
 		if (ret < 0)
@@ -744,6 +769,25 @@ static int io_complete_bytes_exceeded(struct thread_data *td)
 	return bytes >= limit || exceeds_number_ios(td);
 }
 
+/*
+ * used to calculate the next io time for rate control
+ *
+ */
+static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir)
+{
+	uint64_t secs, remainder, bps, bytes;
+
+	assert(!(td->flags & TD_F_CHILD));
+	bytes = td->rate_io_issue_bytes[ddir];
+	bps = td->rate_bps[ddir];
+	if (bps) {
+		secs = bytes / bps;
+		remainder = bytes % bps;
+		return remainder * 1000000 / bps + secs * 1000000;
+	} else
+		return 0;
+}
+
 /*
  * Main IO worker function. It retrieves io_u's to process and queues
  * and reaps them, checking for rate and errors along the way.
@@ -873,10 +917,17 @@ static uint64_t do_io(struct thread_data *td)
 			if (td->error)
 				break;
 			ret = workqueue_enqueue(&td->io_wq, io_u);
+
+			if (should_check_rate(td))
+				td->rate_next_io_time[ddir] = usec_for_io(td, ddir);
+
 		} else {
 			ret = td_io_queue(td, io_u);
 
-			if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 1, &comp_time))
+			if (should_check_rate(td))
+				td->rate_next_io_time[ddir] = usec_for_io(td, ddir);
+
+			if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 0, &comp_time))
 				break;
 
 			/*
@@ -887,7 +938,7 @@ static uint64_t do_io(struct thread_data *td)
 reap:
 			full = queue_full(td) ||
 				(ret == FIO_Q_BUSY && td->cur_depth);
-			if (full || !td->o.iodepth_batch_complete)
+			if (full || io_in_polling(td))
 				ret = wait_for_completions(td, &comp_time);
 		}
 		if (ret < 0)
@@ -1157,13 +1208,17 @@ static int switch_ioscheduler(struct thread_data *td)
 	/*
 	 * Read back and check that the selected scheduler is now the default.
 	 */
+	memset(tmp, 0, sizeof(tmp));
 	ret = fread(tmp, sizeof(tmp), 1, f);
 	if (ferror(f) || ret < 0) {
 		td_verror(td, errno, "fread");
 		fclose(f);
 		return 1;
 	}
-	tmp[sizeof(tmp) - 1] = '\0';
+	/*
+	 * either a list of io schedulers or "none\n" is expected.
+	 */
+	tmp[strlen(tmp) - 1] = '\0';
 
 
 	sprintf(tmp2, "[%s]", td->o.ioscheduler);
@@ -1220,7 +1275,8 @@ static int keep_running(struct thread_data *td)
 
 static int exec_string(struct thread_options *o, const char *string, const char *mode)
 {
-	int ret, newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1;
+	size_t newlen = strlen(string) + strlen(o->name) + strlen(mode) + 9 + 1;
+	int ret;
 	char *str;
 
 	str = malloc(newlen);
@@ -1306,7 +1362,7 @@ static void io_workqueue_fn(struct thread_data *td, struct io_u *io_u)
  */
 static void *thread_main(void *data)
 {
-	unsigned long long elapsed;
+	unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, };
 	struct thread_data *td = data;
 	struct thread_options *o = &td->o;
 	pthread_condattr_t attr;
@@ -1544,18 +1600,12 @@ static void *thread_main(void *data)
 		check_update_rusage(td);
 
 		fio_mutex_down(stat_mutex);
-		if (td_read(td) && td->io_bytes[DDIR_READ]) {
-			elapsed = mtime_since_now(&td->start);
-			td->ts.runtime[DDIR_READ] += elapsed;
-		}
-		if (td_write(td) && td->io_bytes[DDIR_WRITE]) {
-			elapsed = mtime_since_now(&td->start);
-			td->ts.runtime[DDIR_WRITE] += elapsed;
-		}
-		if (td_trim(td) && td->io_bytes[DDIR_TRIM]) {
-			elapsed = mtime_since_now(&td->start);
-			td->ts.runtime[DDIR_TRIM] += elapsed;
-		}
+		if (td_read(td) && td->io_bytes[DDIR_READ])
+			update_runtime(td, elapsed_us, DDIR_READ);
+		if (td_write(td) && td->io_bytes[DDIR_WRITE])
+			update_runtime(td, elapsed_us, DDIR_WRITE);
+		if (td_trim(td) && td->io_bytes[DDIR_TRIM])
+			update_runtime(td, elapsed_us, DDIR_TRIM);
 		fio_gettime(&td->start, NULL);
 		fio_mutex_up(stat_mutex);
 
@@ -1579,7 +1629,7 @@ static void *thread_main(void *data)
 		check_update_rusage(td);
 
 		fio_mutex_down(stat_mutex);
-		td->ts.runtime[DDIR_READ] += mtime_since_now(&td->start);
+		update_runtime(td, elapsed_us, DDIR_READ);
 		fio_gettime(&td->start, NULL);
 		fio_mutex_up(stat_mutex);
 
@@ -1594,16 +1644,8 @@ static void *thread_main(void *data)
 	td->ts.io_bytes[DDIR_TRIM] = td->io_bytes[DDIR_TRIM];
 
 	if (td->o.verify_state_save && !(td->flags & TD_F_VSTATE_SAVED) &&
-	    (td->o.verify != VERIFY_NONE && td_write(td))) {
-		struct all_io_list *state;
-		size_t sz;
-
-		state = get_all_io_list(td->thread_number, &sz);
-		if (state) {
-			__verify_save_state(state, "local");
-			free(state);
-		}
-	}
+	    (td->o.verify != VERIFY_NONE && td_write(td)))
+		verify_save_state(td->thread_number);
 
 	fio_unpin_memory(td);
 
@@ -1853,7 +1895,7 @@ void check_trigger_file(void)
 		if (nr_clients)
 			fio_clients_send_trigger(trigger_remote_cmd);
 		else {
-			verify_save_state();
+			verify_save_state(IO_LIST_ALL);
 			fio_terminate_threads(TERMINATE_ALL);
 			exec_trigger(trigger_cmd);
 		}
@@ -1869,11 +1911,12 @@ static int fio_verify_load_state(struct thread_data *td)
 
 	if (is_backend) {
 		void *data;
+		int ver;
 
 		ret = fio_server_get_verify_state(td->o.name,
-					td->thread_number - 1, &data);
+					td->thread_number - 1, &data, &ver);
 		if (!ret)
-			verify_convert_assign_state(td, data);
+			verify_convert_assign_state(td, data, ver);
 	} else
 		ret = verify_load_state(td, "local");
 
@@ -1887,6 +1930,27 @@ static void do_usleep(unsigned int usecs)
 	usleep(usecs);
 }
 
+static int check_mount_writes(struct thread_data *td)
+{
+	struct fio_file *f;
+	unsigned int i;
+
+	if (!td_write(td) || td->o.allow_mounted_write)
+		return 0;
+
+	for_each_file(td, f, i) {
+		if (f->filetype != FIO_TYPE_BD)
+			continue;
+		if (device_is_mounted(f->file_name))
+			goto mounted;
+	}
+
+	return 0;
+mounted:
+	log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.", f->file_name);
+	return 1;
+}
+
 /*
  * Main function for kicking off and reaping jobs, as needed.
  */
@@ -1905,13 +1969,15 @@ static void run_threads(void)
 
 	nr_thread = nr_process = 0;
 	for_each_td(td, i) {
+		if (check_mount_writes(td))
+			return;
 		if (td->o.use_thread)
 			nr_thread++;
 		else
 			nr_process++;
 	}
 
-	if (output_format == FIO_OUTPUT_NORMAL) {
+	if (output_format & FIO_OUTPUT_NORMAL) {
 		log_info("Starting ");
 		if (nr_thread)
 			log_info("%d thread%s", nr_thread,
@@ -2253,7 +2319,7 @@ int fio_backend(void)
 			for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 				struct io_log *log = agg_io_log[i];
 
-				flush_log(log);
+				flush_log(log, 0);
 				free_log(log);
 			}
 		}