X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=io_u.c;h=f451d1a756141e3a29eaad8f3f2fb74d113ae82e;hp=34ab58a11ddc83066d410cc6db71fe7ff8bcff8a;hb=32c17adf7f75da1f0957d4691633fea60259910f;hpb=b23b6a2f66637e326c5825bf438059ecc62008d4

diff --git a/io_u.c b/io_u.c
index 34ab58a1..f451d1a7 100644
--- a/io_u.c
+++ b/io_u.c
@@ -8,6 +8,7 @@
 #include "fio.h"
 #include "hash.h"
 #include "verify.h"
+#include "lib/rand.h"
 
 struct io_completion_data {
 	int nr;				/* input */
@@ -53,11 +54,8 @@ static void mark_random_map(struct thread_data *td, struct io_u *io_u)
 		 * If we have a mixed random workload, we may
 		 * encounter blocks we already did IO to.
 		 */
-		if ((td->o.ddir_nr == 1) && !random_map_free(f, block)) {
-			if (!blocks)
-				blocks = 1;
+		if ((td->o.ddir_nr == 1) && !random_map_free(f, block))
 			break;
-		}
 
 		idx = RAND_MAP_IDX(f, block);
 		bit = RAND_MAP_BIT(f, block);
@@ -68,10 +66,20 @@ static void mark_random_map(struct thread_data *td, struct io_u *io_u)
 		if (this_blocks + bit > BLOCKS_PER_MAP)
 			this_blocks = BLOCKS_PER_MAP - bit;
 
-		if (this_blocks == BLOCKS_PER_MAP)
-			mask = -1U;
-		else
-			mask = ((1U << this_blocks) - 1) << bit;
+		do {
+			if (this_blocks == BLOCKS_PER_MAP)
+				mask = -1U;
+			else
+				mask = ((1U << this_blocks) - 1) << bit;
+	
+			if (!(f->file_map[idx] & mask))
+				break;
+
+			this_blocks--;
+		} while (this_blocks);
+
+		if (!this_blocks)
+			break;
 
 		f->file_map[idx] |= mask;
 		nr_blocks -= this_blocks;
@@ -187,7 +195,7 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
  * until we find a free one. For sequential io, just return the end of
  * the last io issued.
  */
-static int get_next_offset(struct thread_data *td, struct io_u *io_u)
+static int __get_next_offset(struct thread_data *td, struct io_u *io_u)
 {
 	struct fio_file *f = io_u->file;
 	unsigned long long b;
@@ -231,12 +239,17 @@ static int get_next_offset(struct thread_data *td, struct io_u *io_u)
 	return 0;
 }
 
-static inline int is_power_of_2(unsigned int val)
+static int get_next_offset(struct thread_data *td, struct io_u *io_u)
 {
-	return (val != 0 && ((val & (val - 1)) == 0));
+	struct prof_io_ops *ops = &td->prof_io_ops;
+
+	if (ops->fill_io_u_off)
+		return ops->fill_io_u_off(td, io_u);
+
+	return __get_next_offset(td, io_u);
 }
 
-static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
+static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u)
 {
 	const int ddir = io_u->ddir;
 	unsigned int uninitialized_var(buflen);
@@ -281,6 +294,16 @@ static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
 	return buflen;
 }
 
+static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
+{
+	struct prof_io_ops *ops = &td->prof_io_ops;
+
+	if (ops->fill_io_u_size)
+		return ops->fill_io_u_size(td, io_u);
+
+	return __get_next_buflen(td, io_u);
+}
+
 static void set_rwmix_bytes(struct thread_data *td)
 {
 	unsigned int diff;
@@ -350,7 +373,7 @@ static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
 	odir = ddir ^ 1;
 	if (td_rw(td) && __should_check_rate(td, odir))
 		td->rate_pending_usleep[odir] -= usec;
-	
+
 	return ddir;
 }
 
@@ -363,6 +386,30 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 {
 	enum fio_ddir ddir;
 
+	/*
+	 * see if it's time to fsync
+	 */
+	if (td->o.fsync_blocks &&
+	   !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) &&
+	     td->io_issues[DDIR_WRITE] && should_fsync(td))
+		return DDIR_SYNC;
+
+	/*
+	 * see if it's time to fdatasync
+	 */
+	if (td->o.fdatasync_blocks &&
+	   !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) &&
+	     td->io_issues[DDIR_WRITE] && should_fsync(td))
+		return DDIR_DATASYNC;
+
+	/*
+	 * see if it's time to sync_file_range
+	 */
+	if (td->sync_file_range_nr &&
+	   !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) &&
+	     td->io_issues[DDIR_WRITE] && should_fsync(td))
+		return DDIR_SYNC_FILE_RANGE;
+
 	if (td_rw(td)) {
 		/*
 		 * Check if it's time to seed a new data direction.
@@ -390,7 +437,7 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 	return td->rwmix_ddir;
 }
 
-static void put_file_log(struct thread_data *td, struct fio_file *f)
+void put_file_log(struct thread_data *td, struct fio_file *f)
 {
 	int ret = put_file(td, f);
 
@@ -400,16 +447,27 @@ static void put_file_log(struct thread_data *td, struct fio_file *f)
 
 void put_io_u(struct thread_data *td, struct io_u *io_u)
 {
-	assert((io_u->flags & IO_U_F_FREE) == 0);
+	td_io_u_lock(td);
+
 	io_u->flags |= IO_U_F_FREE;
+	io_u->flags &= ~IO_U_F_FREE_DEF;
 
 	if (io_u->file)
 		put_file_log(td, io_u->file);
 
 	io_u->file = NULL;
-	flist_del(&io_u->list);
+	if (io_u->flags & IO_U_F_IN_CUR_DEPTH)
+		td->cur_depth--;
+	flist_del_init(&io_u->list);
 	flist_add(&io_u->list, &td->io_u_freelist);
-	td->cur_depth--;
+	td_io_u_unlock(td);
+	td_io_u_free_notify(td);
+}
+
+void clear_io_u(struct thread_data *td, struct io_u *io_u)
+{
+	io_u->flags &= ~IO_U_F_FLIGHT;
+	put_io_u(td, io_u);
 }
 
 void requeue_io_u(struct thread_data *td, struct io_u **io_u)
@@ -418,15 +476,18 @@ void requeue_io_u(struct thread_data *td, struct io_u **io_u)
 
 	dprint(FD_IO, "requeue %p\n", __io_u);
 
+	td_io_u_lock(td);
+
 	__io_u->flags |= IO_U_F_FREE;
-	if ((__io_u->flags & IO_U_F_FLIGHT) && (__io_u->ddir != DDIR_SYNC))
+	if ((__io_u->flags & IO_U_F_FLIGHT) && !ddir_sync(__io_u->ddir))
 		td->io_issues[__io_u->ddir]--;
 
 	__io_u->flags &= ~IO_U_F_FLIGHT;
-
+	if (__io_u->flags & IO_U_F_IN_CUR_DEPTH)
+		td->cur_depth--;
 	flist_del(&__io_u->list);
 	flist_add_tail(&__io_u->list, &td->io_u_requeues);
-	td->cur_depth--;
+	td_io_u_unlock(td);
 	*io_u = NULL;
 }
 
@@ -435,17 +496,13 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 	if (td->io_ops->flags & FIO_NOIO)
 		goto out;
 
+	io_u->ddir = get_rw_ddir(td);
+
 	/*
-	 * see if it's time to sync
+	 * fsync() or fdatasync(), we are done
 	 */
-	if (td->o.fsync_blocks &&
-	   !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) &&
-	     td->io_issues[DDIR_WRITE] && should_fsync(td)) {
-		io_u->ddir = DDIR_SYNC;
+	if (ddir_sync(io_u->ddir))
 		goto out;
-	}
-
-	io_u->ddir = get_rw_ddir(td);
 
 	/*
 	 * See if it's time to switch to a new zone
@@ -733,7 +790,8 @@ static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf,
 			opened = 1;
 		}
 
-		dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf, f->flags);
+		dprint(FD_FILE, "goodf=%x, badf=%x, ff=%x\n", goodf, badf,
+								f->flags);
 		if ((!goodf || (f->flags & goodf)) && !(f->flags & badf))
 			break;
 
@@ -747,7 +805,7 @@ static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf,
 	return f;
 }
 
-static struct fio_file *get_next_file(struct thread_data *td)
+static struct fio_file *__get_next_file(struct thread_data *td)
 {
 	struct fio_file *f;
 
@@ -782,6 +840,16 @@ out:
 	return f;
 }
 
+static struct fio_file *get_next_file(struct thread_data *td)
+{
+	struct prof_io_ops *ops = &td->prof_io_ops;
+
+	if (ops->get_next_file)
+		return ops->get_next_file(td);
+
+	return __get_next_file(td);
+}
+
 static int set_io_u_file(struct thread_data *td, struct io_u *io_u)
 {
 	struct fio_file *f;
@@ -802,7 +870,8 @@ static int set_io_u_file(struct thread_data *td, struct io_u *io_u)
 		io_u->file = NULL;
 		fio_file_set_done(f);
 		td->nr_done_files++;
-		dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name, td->nr_done_files, td->o.nr_files);
+		dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name,
+					td->nr_done_files, td->o.nr_files);
 	} while (1);
 
 	return 0;
@@ -813,6 +882,9 @@ struct io_u *__get_io_u(struct thread_data *td)
 {
 	struct io_u *io_u = NULL;
 
+	td_io_u_lock(td);
+
+again:
 	if (!flist_empty(&td->io_u_requeues))
 		io_u = flist_entry(td->io_u_requeues.next, struct io_u, list);
 	else if (!queue_full(td)) {
@@ -826,14 +898,23 @@ struct io_u *__get_io_u(struct thread_data *td)
 
 	if (io_u) {
 		assert(io_u->flags & IO_U_F_FREE);
-		io_u->flags &= ~IO_U_F_FREE;
+		io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF);
 
 		io_u->error = 0;
 		flist_del(&io_u->list);
 		flist_add(&io_u->list, &td->io_u_busylist);
 		td->cur_depth++;
+		io_u->flags |= IO_U_F_IN_CUR_DEPTH;
+	} else if (td->o.verify_async) {
+		/*
+		 * We ran out, wait for async verify threads to finish and
+		 * return one
+		 */
+		pthread_cond_wait(&td->free_cond, &td->io_u_lock);
+		goto again;
 	}
 
+	td_io_u_unlock(td);
 	return io_u;
 }
 
@@ -852,6 +933,24 @@ struct io_u *get_io_u(struct thread_data *td)
 		return NULL;
 	}
 
+	if (td->o.verify_backlog && td->io_hist_len) {
+		int get_verify = 0;
+
+		if (td->verify_batch) {
+			td->verify_batch--;
+			get_verify = 1;
+		} else if (!(td->io_hist_len % td->o.verify_backlog) &&
+			 td->last_ddir != DDIR_READ) {
+			td->verify_batch = td->o.verify_batch;
+			if (!td->verify_batch)
+				td->verify_batch = td->o.verify_backlog;
+			get_verify = 1;
+		}
+
+		if (get_verify && !get_next_verify(td, io_u))
+			goto out;
+	}
+
 	/*
 	 * from a requeue, io_u already setup
 	 */
@@ -872,7 +971,7 @@ struct io_u *get_io_u(struct thread_data *td)
 	f = io_u->file;
 	assert(fio_file_open(f));
 
-	if (io_u->ddir != DDIR_SYNC) {
+	if (!ddir_sync(io_u->ddir)) {
 		if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) {
 			dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u);
 			goto err_put;
@@ -930,52 +1029,74 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 	 * initialized, silence that warning.
 	 */
 	unsigned long uninitialized_var(usec);
+	struct fio_file *f;
 
 	dprint_io_u(io_u, "io complete");
 
+	td_io_u_lock(td);
 	assert(io_u->flags & IO_U_F_FLIGHT);
 	io_u->flags &= ~IO_U_F_FLIGHT;
+	td_io_u_unlock(td);
 
-	if (io_u->ddir == DDIR_SYNC) {
+	if (ddir_sync(io_u->ddir)) {
 		td->last_was_sync = 1;
+		f = io_u->file;
+		if (f) {
+			f->first_write = -1ULL;
+			f->last_write = -1ULL;
+		}
 		return;
 	}
 
 	td->last_was_sync = 0;
+	td->last_ddir = io_u->ddir;
 
 	if (!io_u->error) {
 		unsigned int bytes = io_u->buflen - io_u->resid;
 		const enum fio_ddir idx = io_u->ddir;
+		const enum fio_ddir odx = io_u->ddir ^ 1;
 		int ret;
 
 		td->io_blocks[idx]++;
 		td->io_bytes[idx] += bytes;
 		td->this_io_bytes[idx] += bytes;
 
+		if (idx == DDIR_WRITE) {
+			f = io_u->file;
+			if (f) {
+				if (f->first_write == -1ULL ||
+				    io_u->offset < f->first_write)
+					f->first_write = io_u->offset;
+				if (f->last_write == -1ULL ||
+				    ((io_u->offset + bytes) > f->last_write))
+					f->last_write = io_u->offset + bytes;
+			}
+		}
+
 		if (ramp_time_over(td)) {
 			unsigned long uninitialized_var(lusec);
-			unsigned long uninitialized_var(rusec);
 
 			if (!td->o.disable_clat || !td->o.disable_bw)
 				lusec = utime_since(&io_u->issue_time,
 							&icd->time);
-			if (__should_check_rate(td, idx) ||
-			    __should_check_rate(td, idx ^ 1))
-				rusec = utime_since(&io_u->start_time,
-							&icd->time);
 
 			if (!td->o.disable_clat) {
-				add_clat_sample(td, idx, usec, bytes);
+				add_clat_sample(td, idx, lusec, bytes);
 				io_u_mark_latency(td, lusec);
 			}
 			if (!td->o.disable_bw)
 				add_bw_sample(td, idx, bytes, &icd->time);
 			if (__should_check_rate(td, idx)) {
-				td->rate_pending_usleep[idx] +=
-					(long) td->rate_usec_cycle[idx] - rusec;
+				td->rate_pending_usleep[idx] =
+					((td->this_io_bytes[idx] *
+					  td->rate_nsec_cycle[idx]) / 1000 -
+					 utime_since_now(&td->start));
 			}
 			if (__should_check_rate(td, idx ^ 1))
-				td->rate_pending_usleep[idx ^ 1] -= rusec;
+				td->rate_pending_usleep[odx] =
+					((td->this_io_bytes[odx] *
+					  td->rate_nsec_cycle[odx]) / 1000 -
+					 utime_since_now(&td->start));
 		}
 
 		if (td_write(td) && idx == DDIR_WRITE &&
@@ -994,6 +1115,17 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 		icd->error = io_u->error;
 		io_u_log_error(td, io_u);
 	}
+	if (td->o.continue_on_error && icd->error &&
+	    td_non_fatal_error(icd->error)) {
+		/*
+		 * If there is a non_fatal error, then add to the error count
+		 * and clear all the errors.
+		 */
+		update_error_count(td, icd->error);
+		td_clear_error(td);
+		icd->error = 0;
+		io_u->error = 0;
+	}
 }
 
 static void init_icd(struct thread_data *td, struct io_completion_data *icd,
@@ -1018,7 +1150,9 @@ static void ios_completed(struct thread_data *td,
 		io_u = td->io_ops->event(td, i);
 
 		io_completed(td, io_u, icd);
-		put_io_u(td, io_u);
+
+		if (!(io_u->flags & IO_U_F_FREE_DEF))
+			put_io_u(td, io_u);
 	}
 }
 
@@ -1032,7 +1166,9 @@ int io_u_sync_complete(struct thread_data *td, struct io_u *io_u,
 
 	init_icd(td, &icd, 1);
 	io_completed(td, io_u, &icd);
-	put_io_u(td, io_u);
+
+	if (!(io_u->flags & IO_U_F_FREE_DEF))
+		put_io_u(td, io_u);
 
 	if (icd.error) {
 		td_verror(td, icd.error, "io_u_sync_complete");
@@ -1107,9 +1243,16 @@ void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u,
 	long *ptr = io_u->buf;
 
 	if (!td->o.zero_buffers) {
+		unsigned long r = __rand(&__fio_rand_state);
+
+		if (sizeof(int) != sizeof(*ptr))
+			r *= (unsigned long) __rand(&__fio_rand_state);
+
 		while ((void *) ptr - io_u->buf < max_bs) {
-			*ptr = rand() * GOLDEN_RATIO_PRIME;
+			*ptr = r;
 			ptr++;
+			r *= GOLDEN_RATIO_PRIME;
+			r >>= 3;
 		}
 	} else
 		memset(ptr, 0, max_bs);