X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=fio.c;h=04a06bac1bee010b82866de1ade9ffd5178f3d7f;hp=6a3a87b0feb9257c5e38b3f7826fc2ba96d08d31;hb=bd6f78b2f0b6375383948f126368804612ec0035;hpb=bb5d7d0b5bde867690590aaa61d77307d773107f

diff --git a/fio.c b/fio.c
index 6a3a87b0..04a06bac 100644
--- a/fio.c
+++ b/fio.c
@@ -35,7 +35,7 @@
 #include <sys/mman.h>
 
 #include "fio.h"
-#include "os.h"
+#include "hash.h"
 
 unsigned long page_mask;
 unsigned long page_size;
@@ -60,6 +60,8 @@ struct io_log *agg_io_log[2];
 
 static inline void td_set_runstate(struct thread_data *td, int runstate)
 {
+	dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", td->pid, td->runstate,
+								runstate);
 	td->runstate = runstate;
 }
 
@@ -70,6 +72,7 @@ static void terminate_threads(int group_id)
 
 	for_each_td(td, i) {
 		if (group_id == TERMINATE_ALL || groupid == td->groupid) {
+			dprint(FD_PROCESS, "setting terminate on %d\n",td->pid);
 			/*
 			 * if the thread is running, just let it exit
 			 */
@@ -311,6 +314,7 @@ static void do_verify(struct thread_data *td)
 
 		if (runtime_exceeded(td, &io_u->start_time)) {
 			put_io_u(td, io_u);
+			td->terminate = 1;
 			break;
 		}
 
@@ -333,6 +337,7 @@ static void do_verify(struct thread_data *td)
 				ret = -io_u->error;
 			else if (io_u->resid) {
 				int bytes = io_u->xfer_buflen - io_u->resid;
+				struct fio_file *f = io_u->file;
 
 				/*
 				 * zero read, fail
@@ -342,10 +347,20 @@ static void do_verify(struct thread_data *td)
 					put_io_u(td, io_u);
 					break;
 				}
+
 				io_u->xfer_buflen = io_u->resid;
 				io_u->xfer_buf += bytes;
+				io_u->offset += bytes;
+				f->last_completed_pos = io_u->offset;
+
+				td->ts.short_io_u[io_u->ddir]++;
+
+				if (io_u->offset == f->real_file_size)
+					goto sync_done;
+
 				requeue_io_u(td, &io_u);
 			} else {
+sync_done:
 				ret = io_u_sync_complete(td, io_u);
 				if (ret < 0)
 					break;
@@ -412,7 +427,7 @@ static void do_io(struct thread_data *td)
 
 	td_set_runstate(td, TD_RUNNING);
 
-	while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->io_size) {
+	while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) {
 		struct timeval comp_time;
 		long bytes_done = 0;
 		int min_evts = 0;
@@ -430,9 +445,17 @@ static void do_io(struct thread_data *td)
 
 		if (runtime_exceeded(td, &s)) {
 			put_io_u(td, io_u);
+			td->terminate = 1;
 			break;
 		}
 
+		/*
+		 * Add verification end_io handler, if asked to verify
+		 * a previously written file.
+		 */
+		if (td->o.verify != VERIFY_NONE)
+			io_u->end_io = verify_io_u;
+
 		ret = td_io_queue(td, io_u);
 		switch (ret) {
 		case FIO_Q_COMPLETED:
@@ -440,6 +463,7 @@ static void do_io(struct thread_data *td)
 				ret = -io_u->error;
 			else if (io_u->resid) {
 				int bytes = io_u->xfer_buflen - io_u->resid;
+				struct fio_file *f = io_u->file;
 
 				/*
 				 * zero read, fail
@@ -452,8 +476,17 @@ static void do_io(struct thread_data *td)
 
 				io_u->xfer_buflen = io_u->resid;
 				io_u->xfer_buf += bytes;
+				io_u->offset += bytes;
+				f->last_completed_pos = io_u->offset;
+
+				td->ts.short_io_u[io_u->ddir]++;
+
+				if (io_u->offset == f->real_file_size)
+					goto sync_done;
+
 				requeue_io_u(td, &io_u);
 			} else {
+sync_done:
 				fio_gettime(&comp_time, NULL);
 				bytes_done = io_u_sync_complete(td, io_u);
 				if (bytes_done < 0)
@@ -538,6 +571,10 @@ static void do_io(struct thread_data *td)
 		}
 	}
 
+	if (td->o.fill_device && td->error == ENOSPC) {
+		td->error = 0;
+		td->terminate = 1;
+	}
 	if (!td->error) {
 		struct fio_file *f;
 
@@ -556,6 +593,12 @@ static void do_io(struct thread_data *td)
 		}
 	} else
 		cleanup_pending_aio(td);
+
+	/*
+	 * stop job if we failed doing any IO
+	 */
+	if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0)
+		td->done = 1;
 }
 
 static void cleanup_io_u(struct thread_data *td)
@@ -576,19 +619,21 @@ static void cleanup_io_u(struct thread_data *td)
 /*
  * "randomly" fill the buffer contents
  */
-static void fill_rand_buf(struct io_u *io_u, int max_bs)
+static void fill_io_buf(struct thread_data *td, struct io_u *io_u, int max_bs)
 {
-	int *ptr = io_u->buf;
+	long *ptr = io_u->buf;
 
-	while ((void *) ptr - io_u->buf < max_bs) {
-		*ptr = rand() * 0x9e370001;
-		ptr++;
-	}
+	if (!td->o.zero_buffers) {
+		while ((void *) ptr - io_u->buf < max_bs) {
+			*ptr = rand() * GOLDEN_RATIO_PRIME;
+			ptr++;
+		}
+	} else
+		memset(ptr, 0, max_bs);
 }
 
 static int init_io_u(struct thread_data *td)
 {
-	unsigned long long buf_size;
 	struct io_u *io_u;
 	unsigned int max_bs;
 	int i, max_units;
@@ -600,32 +645,37 @@ static int init_io_u(struct thread_data *td)
 		max_units = td->o.iodepth;
 
 	max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
-	buf_size = (unsigned long long) max_bs * (unsigned long long) max_units;
-	buf_size += page_mask;
-	if (buf_size != (size_t) buf_size) {
-		log_err("fio: IO memory too large. Reduce max_bs or iodepth\n");
-		return 1;
-	}
-
-	td->orig_buffer_size = buf_size;
+	td->orig_buffer_size = (unsigned long long) max_bs * (unsigned long long) max_units;
 
 	if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE)
 		td->orig_buffer_size = (td->orig_buffer_size + td->o.hugepage_size - 1) & ~(td->o.hugepage_size - 1);
-	else if (td->orig_buffer_size & page_mask)
-		td->orig_buffer_size = (td->orig_buffer_size + page_mask) & ~page_mask;
+
+	if (td->orig_buffer_size != (size_t) td->orig_buffer_size) {
+		log_err("fio: IO memory too large. Reduce max_bs or iodepth\n");
+		return 1;
+	}
 
 	if (allocate_io_mem(td))
 		return 1;
 
-	p = ALIGN(td->orig_buffer);
+	if (td->o.odirect)
+		p = ALIGN(td->orig_buffer);
+	else
+		p = td->orig_buffer;
+
 	for (i = 0; i < max_units; i++) {
+		if (td->terminate)
+			return 1;
 		io_u = malloc(sizeof(*io_u));
 		memset(io_u, 0, sizeof(*io_u));
 		INIT_LIST_HEAD(&io_u->list);
 
-		io_u->buf = p + max_bs * i;
-		if (td_write(td) || td_rw(td))
-			fill_rand_buf(io_u, max_bs);
+		if (!(td->io_ops->flags & FIO_NOIO)) {
+			io_u->buf = p + max_bs * i;
+
+			if (td_write(td))
+				fill_io_buf(td, io_u, max_bs);
+		}
 
 		io_u->index = i;
 		io_u->flags = IO_U_F_FREE;
@@ -650,7 +700,11 @@ static int switch_ioscheduler(struct thread_data *td)
 
 	f = fopen(tmp, "r+");
 	if (!f) {
-		td_verror(td, errno, "fopen");
+		if (errno == ENOENT) {
+			log_err("fio: os or kernel doesn't support IO scheduler switching\n");
+			return 0;
+		}
+		td_verror(td, errno, "fopen iosched");
 		return 1;
 	}
 
@@ -688,6 +742,26 @@ static int switch_ioscheduler(struct thread_data *td)
 	return 0;
 }
 
+static int keep_running(struct thread_data *td)
+{
+	unsigned long long io_done;
+
+	if (td->done)
+		return 0;
+	if (td->o.time_based)
+		return 1;
+	if (td->o.loops) {
+		td->o.loops--;
+		return 1;
+	}
+
+	io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] + td->io_skip_bytes;
+	if (io_done < td->o.size)
+		return 1;
+
+	return 0;
+}
+
 static int clear_io_state(struct thread_data *td)
 {
 	struct fio_file *f;
@@ -703,11 +777,18 @@ static int clear_io_state(struct thread_data *td)
 
 	td->last_was_sync = 0;
 
+	/*
+	 * reset file done count if we are to start over
+	 */
+	if (td->o.time_based || td->o.loops)
+		td->nr_done_files = 0;
+
 	for_each_file(td, f, i)
 		td_io_close_file(td, f);
 
 	ret = 0;
 	for_each_file(td, f, i) {
+		f->flags &= ~FIO_FILE_DONE;
 		ret = td_io_open_file(td, f);
 		if (ret)
 			break;
@@ -722,9 +803,8 @@ static int clear_io_state(struct thread_data *td)
  */
 static void *thread_main(void *data)
 {
-	unsigned long long runtime[2];
+	unsigned long long runtime[2], elapsed;
 	struct thread_data *td = data;
-	unsigned long elapsed;
 	int clear_state;
 
 	if (!td->o.use_thread)
@@ -732,47 +812,54 @@ static void *thread_main(void *data)
 
 	td->pid = getpid();
 
+	dprint(FD_PROCESS, "jobs pid=%d started\n", td->pid);
+
 	INIT_LIST_HEAD(&td->io_u_freelist);
 	INIT_LIST_HEAD(&td->io_u_busylist);
 	INIT_LIST_HEAD(&td->io_u_requeues);
 	INIT_LIST_HEAD(&td->io_log_list);
+	INIT_LIST_HEAD(&td->io_hist_list);
 	td->io_hist_tree = RB_ROOT;
 
+	td_set_runstate(td, TD_INITIALIZED);
+	fio_sem_up(startup_sem);
+	fio_sem_down(td->mutex);
+
+	/*
+	 * the ->mutex semaphore is now no longer used, close it to avoid
+	 * eating a file descriptor
+	 */
+	fio_sem_remove(td->mutex);
+
+	/*
+	 * May alter parameters that init_io_u() will use, so we need to
+	 * do this first.
+	 */
+	if (init_iolog(td))
+		goto err;
+
 	if (init_io_u(td))
-		goto err_sem;
+		goto err;
 
-	if (fio_setaffinity(td) == -1) {
+	if (td->o.cpumask_set && fio_setaffinity(td) == -1) {
 		td_verror(td, errno, "cpu_set_affinity");
-		goto err_sem;
+		goto err;
 	}
 
-	if (init_iolog(td))
-		goto err_sem;
-
-	if (td->ioprio) {
+	if (td->ioprio_set) {
 		if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
 			td_verror(td, errno, "ioprio_set");
-			goto err_sem;
+			goto err;
 		}
 	}
 
 	if (nice(td->o.nice) == -1) {
 		td_verror(td, errno, "nice");
-		goto err_sem;
+		goto err;
 	}
 
 	if (td->o.ioscheduler && switch_ioscheduler(td))
-		goto err_sem;
-
-	td_set_runstate(td, TD_INITIALIZED);
-	fio_sem_up(startup_sem);
-	fio_sem_down(td->mutex);
-
-	/*
-	 * the ->mutex semaphore is now no longer used, close it to avoid
-	 * eating a file descriptor
-	 */
-	fio_sem_remove(td->mutex);
+		goto err;
 
 	if (!td->o.create_serialize && setup_files(td))
 		goto err;
@@ -797,7 +884,7 @@ static void *thread_main(void *data)
 
 	runtime[0] = runtime[1] = 0;
 	clear_state = 0;
-	while (td->o.loops--) {
+	while (keep_running(td)) {
 		fio_gettime(&td->start, NULL);
 		memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
 
@@ -833,7 +920,9 @@ static void *thread_main(void *data)
 		if (td->error || td->terminate)
 			break;
 
-		if (td->o.verify == VERIFY_NONE)
+		if (!td->o.do_verify ||
+		    td->o.verify == VERIFY_NONE ||
+		    (td->io_ops->flags & FIO_UNIDIR))
 			continue;
 
 		if (clear_io_state(td))
@@ -850,8 +939,8 @@ static void *thread_main(void *data)
 	}
 
 	update_rusage_stat(td);
-	td->ts.runtime[0] = runtime[0] / 1000;
-	td->ts.runtime[1] = runtime[1] / 1000;
+	td->ts.runtime[0] = (runtime[0] + 999) / 1000;
+	td->ts.runtime[1] = (runtime[1] + 999) / 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];
@@ -862,8 +951,6 @@ static void *thread_main(void *data)
 		finish_log(td, td->ts.slat_log, "slat");
 	if (td->ts.clat_log)
 		finish_log(td, td->ts.clat_log, "clat");
-	if (td->o.write_iolog_file)
-		write_iolog_close(td);
 	if (td->o.exec_postrun) {
 		if (system(td->o.exec_postrun) < 0)
 			log_err("fio: postrun %s failed\n", td->o.exec_postrun);
@@ -878,12 +965,16 @@ err:
 	close_files(td);
 	close_ioengine(td);
 	cleanup_io_u(td);
+
+	/*
+	 * do this very late, it will log file closing as well
+	 */
+	if (td->o.write_iolog_file)
+		write_iolog_close(td);
+
 	options_mem_free(td);
 	td_set_runstate(td, TD_EXITED);
 	return (void *) (unsigned long) td->error;
-err_sem:
-	fio_sem_up(startup_sem);
-	goto err;
 }
 
 /*
@@ -915,12 +1006,12 @@ static int 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, status, ret;
+	int i, cputhreads, realthreads, pending, status, ret;
 
 	/*
 	 * reap exited threads (TD_EXITED -> TD_REAPED)
 	 */
-	pending = cputhreads = 0;
+	realthreads = pending = cputhreads = 0;
 	for_each_td(td, i) {
 		int flags = 0;
 
@@ -930,6 +1021,8 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 		 */
 		if (td->io_ops && !strcmp(td->io_ops->name, "cpuio"))
 			cputhreads++;
+		else
+			realthreads++;
 
 		if (!td->pid || td->runstate == TD_REAPED)
 			continue;
@@ -977,24 +1070,29 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 		/*
 		 * thread is not dead, continue
 		 */
+		pending++;
 		continue;
 reaped:
 		if (td->o.use_thread) {
 			long ret;
 
-			if (pthread_join(td->thread, (void *) &ret))
+			dprint(FD_PROCESS, "joining tread %d\n", td->pid);
+			if (pthread_join(td->thread, (void *) &ret)) {
+				dprint(FD_PROCESS, "join failed %ld\n", ret);
 				perror("pthread_join");
+			}
 		}
 
 		(*nr_running)--;
 		(*m_rate) -= td->o.ratemin;
 		(*t_rate) -= td->o.rate;
+		pending--;
 
 		if (td->error)
 			exit_value++;
 	}
 
-	if (*nr_running == cputhreads && !pending)
+	if (*nr_running == cputhreads && !pending && realthreads)
 		terminate_threads(TERMINATE_ALL);
 }
 
@@ -1097,12 +1195,14 @@ static void run_threads(void)
 			nr_started++;
 
 			if (td->o.use_thread) {
+				dprint(FD_PROCESS, "will pthread_create\n");
 				if (pthread_create(&td->thread, NULL, thread_main, td)) {
 					perror("thread_create");
 					nr_started--;
 					break;
 				}
 			} else {
+				dprint(FD_PROCESS, "will fork\n");
 				if (!fork()) {
 					int ret = fork_main(shm_id, i);
 
@@ -1196,10 +1296,8 @@ int main(int argc, char *argv[])
 	if (parse_options(argc, argv))
 		return 1;
 
-	if (!thread_number) {
-		log_err("Nothing to do\n");
-		return 1;
-	}
+	if (!thread_number)
+		return 0;
 
 	ps = sysconf(_SC_PAGESIZE);
 	if (ps < 0) {