X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=io_u.c;h=7a23d2cf7fdd4a9127ec09298b8548cbca60588c;hp=8201ccde751147c85c4b7605b7ef811fae0d6b49;hb=303032ae4e13c7ac10b8198c40df98d95a4524a6;hpb=d9d91e39df9cdc02d17c7b6831e2408ed1faa432;ds=sidebyside

diff --git a/io_u.c b/io_u.c
index 8201ccde..7a23d2cf 100644
--- a/io_u.c
+++ b/io_u.c
@@ -25,11 +25,13 @@ struct io_completion_data {
  * to yet. Used to make sure we cover the entire range in a fair fashion.
  */
 static int random_map_free(struct thread_data *td, struct fio_file *f,
-			   unsigned long long block)
+			   const unsigned long long block)
 {
 	unsigned int idx = RAND_MAP_IDX(td, f, block);
 	unsigned int bit = RAND_MAP_BIT(td, f, block);
 
+	dprint(FD_RANDOM, "free: b=%llu, idx=%u, bit=%u\n", block, idx, bit);
+
 	return (f->file_map[idx] & (1UL << bit)) == 0;
 }
 
@@ -44,7 +46,7 @@ static void mark_random_map(struct thread_data *td, struct io_u *io_u)
 	unsigned int blocks;
 	unsigned int nr_blocks;
 
-	block = io_u->offset / (unsigned long long) min_bs;
+	block = (io_u->offset - f->file_offset) / (unsigned long long) min_bs;
 	blocks = 0;
 	nr_blocks = (io_u->buflen + min_bs - 1) / min_bs;
 
@@ -55,7 +57,7 @@ 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(td, f, block))
+		if ((td->o.ddir_nr == 1) && !random_map_free(td, f, block))
 			break;
 
 		idx = RAND_MAP_IDX(td, f, block);
@@ -72,19 +74,35 @@ static void mark_random_map(struct thread_data *td, struct io_u *io_u)
 		io_u->buflen = blocks * min_bs;
 }
 
+static inline unsigned long long last_block(struct thread_data *td,
+					    struct fio_file *f,
+					    enum fio_ddir ddir)
+{
+	unsigned long long max_blocks;
+
+	max_blocks = f->io_size / (unsigned long long) td->o.min_bs[ddir];
+	if (!max_blocks)
+		return 0;
+
+	return max_blocks - 1;
+}
+
 /*
  * Return the next free block in the map.
  */
 static int get_next_free_block(struct thread_data *td, struct fio_file *f,
-			       unsigned long long *b)
+			       enum fio_ddir ddir, unsigned long long *b)
 {
+	unsigned long long min_bs = td->o.rw_min_bs;
 	int i;
 
 	i = f->last_free_lookup;
 	*b = (i * BLOCKS_PER_MAP);
-	while ((*b) * td->o.rw_min_bs < f->real_file_size) {
+	while ((*b) * min_bs < f->real_file_size) {
 		if (f->file_map[i] != -1UL) {
 			*b += fio_ffz(f->file_map[i]);
+			if (*b > last_block(td, f, ddir))
+				break;
 			f->last_free_lookup = i;
 			return 0;
 		}
@@ -93,35 +111,36 @@ static int get_next_free_block(struct thread_data *td, struct fio_file *f,
 		i++;
 	}
 
+	dprint(FD_IO, "failed finding a free block\n");
 	return 1;
 }
 
 static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
-				int ddir, unsigned long long *b)
+				enum fio_ddir ddir, unsigned long long *b)
 {
-	unsigned long long max_blocks = f->io_size / td->o.min_bs[ddir];
-	unsigned long long r, rb;
+	unsigned long long r;
 	int loops = 5;
 
 	do {
 		r = os_random_long(&td->random_state);
-		if (!max_blocks)
-			*b = 0;
-		else
-			*b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0));
+		dprint(FD_RANDOM, "off rand %llu\n", r);
+		*b = (last_block(td, f, ddir) - 1)
+			* (r / ((unsigned long long) RAND_MAX + 1.0));
+
 		/*
 		 * if we are not maintaining a random map, we are done.
 		 */
-		if (td->o.norandommap)
+		if (!file_randommap(td, f))
 			return 0;
 
 		/*
-		 * calculate map offset and chec if it's free
+		 * calculate map offset and check if it's free
 		 */
-		rb = *b + (f->file_offset / td->o.min_bs[ddir]);
-		if (random_map_free(td, f, rb))
+		if (random_map_free(td, f, *b))
 			return 0;
 
+		dprint(FD_RANDOM, "get_next_rand_offset: offset %llu busy\n",
+									*b);
 	} while (--loops);
 
 	/*
@@ -132,7 +151,7 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
 	loops = 10;
 	do {
 		f->last_free_lookup = (f->num_maps - 1) * (r / (RAND_MAX+1.0));
-		if (!get_next_free_block(td, f, b))
+		if (!get_next_free_block(td, f, ddir, b))
 			return 0;
 
 		r = os_random_long(&td->random_state);
@@ -142,7 +161,7 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
 	 * that didn't work either, try exhaustive search from the start
 	 */
 	f->last_free_lookup = 0;
-	return get_next_free_block(td, f, b);
+	return get_next_free_block(td, f, ddir, b);
 }
 
 /*
@@ -153,8 +172,8 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
 static int get_next_offset(struct thread_data *td, struct io_u *io_u)
 {
 	struct fio_file *f = io_u->file;
-	const int ddir = io_u->ddir;
 	unsigned long long b;
+	enum fio_ddir ddir = io_u->ddir;
 
 	if (td_random(td) && (td->o.ddir_nr && !--td->ddir_nr)) {
 		td->ddir_nr = td->o.ddir_nr;
@@ -163,15 +182,19 @@ static int get_next_offset(struct thread_data *td, struct io_u *io_u)
 			return 1;
 	} else {
 		if (f->last_pos >= f->real_file_size) {
-			if (!td_random(td) || get_next_rand_offset(td, f, ddir, &b))
+			if (!td_random(td) ||
+			     get_next_rand_offset(td, f, ddir, &b))
 				return 1;
 		} else
 			b = (f->last_pos - f->file_offset) / td->o.min_bs[ddir];
 	}
 
 	io_u->offset = (b * td->o.min_bs[ddir]) + f->file_offset;
-	if (io_u->offset >= f->real_file_size)
+	if (io_u->offset >= f->real_file_size) {
+		dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n",
+					io_u->offset, f->real_file_size);
 		return 1;
+	}
 
 	return 0;
 }
@@ -186,9 +209,11 @@ static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
 		buflen = td->o.min_bs[ddir];
 	else {
 		r = os_random_long(&td->bsrange_state);
-		if (!td->o.bssplit_nr)
-			buflen = (unsigned int) (1 + (double) (td->o.max_bs[ddir] - 1) * r / (RAND_MAX + 1.0));
-		else {
+		if (!td->o.bssplit_nr) {
+			buflen = (unsigned int)
+					(1 + (double) (td->o.max_bs[ddir] - 1)
+					* r / (RAND_MAX + 1.0));
+		} else {
 			long perc = 0;
 			unsigned int i;
 
@@ -201,12 +226,17 @@ static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
 					break;
 			}
 		}
-		if (!td->o.bs_unaligned)
-			buflen = (buflen + td->o.min_bs[ddir] - 1) & ~(td->o.min_bs[ddir] - 1);
+		if (!td->o.bs_unaligned) {
+			buflen = (buflen + td->o.min_bs[ddir] - 1)
+					& ~(td->o.min_bs[ddir] - 1);
+		}
 	}
 
-	if (io_u->offset + buflen > io_u->file->real_file_size)
+	if (io_u->offset + buflen > io_u->file->real_file_size) {
+		dprint(FD_IO, "lower buflen %u -> %u (ddir=%d)\n", buflen,
+						td->o.min_bs[ddir], ddir);
 		buflen = td->o.min_bs[ddir];
+	}
 
 	return buflen;
 }
@@ -224,7 +254,8 @@ static void set_rwmix_bytes(struct thread_data *td)
 	rbytes = td->io_bytes[td->rwmix_ddir] - td->rwmix_bytes;
 	diff = td->o.rwmix[td->rwmix_ddir ^ 1];
 
-	td->rwmix_bytes = td->io_bytes[td->rwmix_ddir] + (rbytes * ((100 - diff)) / diff);
+	td->rwmix_bytes = td->io_bytes[td->rwmix_ddir]
+				+ (rbytes * ((100 - diff)) / diff);
 }
 
 static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
@@ -253,7 +284,7 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 		unsigned int cycle;
 
 		fio_gettime(&now, NULL);
-	 	elapsed = mtime_since_now(&td->rwmix_switch);
+		elapsed = mtime_since_now(&td->rwmix_switch);
 
 		/*
 		 * if this is the first cycle, make it shorter
@@ -268,7 +299,7 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 		if (elapsed >= cycle ||
 		    td->io_bytes[td->rwmix_ddir] >= td->rwmix_bytes) {
 			unsigned long long max_bytes;
-			enum fio_ddir ddir;			
+			enum fio_ddir ddir;
 
 			/*
 			 * Put a top limit on how many bytes we do for
@@ -277,10 +308,12 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 			 */
 			ddir = get_rand_ddir(td);
 			max_bytes = td->this_io_bytes[ddir];
-			if (max_bytes >= (td->o.size * td->o.rwmix[ddir] / 100)) {
+			if (max_bytes >=
+			    (td->o.size * td->o.rwmix[ddir] / 100)) {
 				if (!td->rw_end_set[ddir]) {
 					td->rw_end_set[ddir] = 1;
-					memcpy(&td->rw_end[ddir], &now, sizeof(now));
+					memcpy(&td->rw_end[ddir], &now,
+						sizeof(now));
 				}
 				ddir ^= 1;
 			}
@@ -303,8 +336,12 @@ void put_io_u(struct thread_data *td, struct io_u *io_u)
 	assert((io_u->flags & IO_U_F_FREE) == 0);
 	io_u->flags |= IO_U_F_FREE;
 
-	if (io_u->file)
-		put_file(td, io_u->file);
+	if (io_u->file) {
+		int ret = put_file(td, io_u->file);
+
+		if (ret)
+			td_verror(td, ret, "file close");
+	}
 
 	io_u->file = NULL;
 	list_del(&io_u->list);
@@ -317,6 +354,9 @@ void requeue_io_u(struct thread_data *td, struct io_u **io_u)
 	struct io_u *__io_u = *io_u;
 
 	__io_u->flags |= IO_U_F_FREE;
+	if ((__io_u->flags & IO_U_F_FLIGHT) && (__io_u->ddir != DDIR_SYNC))
+		td->io_issues[__io_u->ddir]--;
+
 	__io_u->flags &= ~IO_U_F_FLIGHT;
 
 	list_del(&__io_u->list);
@@ -355,26 +395,35 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 	 * No log, let the seq/rand engine retrieve the next buflen and
 	 * position.
 	 */
-	if (get_next_offset(td, io_u))
+	if (get_next_offset(td, io_u)) {
+		dprint(FD_IO, "io_u %p, failed getting offset\n", io_u);
 		return 1;
+	}
 
 	io_u->buflen = get_next_buflen(td, io_u);
-	if (!io_u->buflen)
+	if (!io_u->buflen) {
+		dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u);
 		return 1;
+	}
 
-	if (io_u->offset + io_u->buflen > io_u->file->real_file_size)
+	if (io_u->offset + io_u->buflen > io_u->file->real_file_size) {
+		dprint(FD_IO, "io_u %p, offset too large\n", io_u);
+		dprint(FD_IO, "  off=%llu/%lu > %llu\n", io_u->offset,
+				io_u->buflen, io_u->file->real_file_size);
 		return 1;
+	}
 
 	/*
 	 * mark entry before potentially trimming io_u
 	 */
-	if (td_random(td) && !td->o.norandommap)
+	if (td_random(td) && file_randommap(td, io_u->file))
 		mark_random_map(td, io_u);
 
 	/*
 	 * If using a write iolog, store this entry.
 	 */
 out:
+	dprint_io_u(io_u, "fill_io_u");
 	td->zone_bytes += io_u->buflen;
 	log_io_u(td, io_u);
 	return 0;
@@ -519,13 +568,16 @@ static struct fio_file *get_next_file_rand(struct thread_data *td, int goodf,
 	do {
 		long r = os_random_long(&td->next_file_state);
 
-		fno = (unsigned int) ((double) td->o.nr_files * (r / (RAND_MAX + 1.0)));
-		f = &td->files[fno];
+		fno = (unsigned int) ((double) td->o.nr_files
+			* (r / (RAND_MAX + 1.0)));
+		f = td->files[fno];
 		if (f->flags & FIO_FILE_DONE)
 			continue;
 
-		if ((!goodf || (f->flags & goodf)) && !(f->flags & badf))
+		if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) {
+			dprint(FD_FILE, "get_next_file_rand: %p\n", f);
 			return f;
+		}
 	} while (1);
 }
 
@@ -539,7 +591,7 @@ static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf,
 	struct fio_file *f;
 
 	do {
-		f = &td->files[td->next_file];
+		f = td->files[td->next_file];
 
 		td->next_file++;
 		if (td->next_file >= td->o.nr_files)
@@ -556,6 +608,7 @@ static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf,
 		f = NULL;
 	} while (td->next_file != old_next_file);
 
+	dprint(FD_FILE, "get_next_file_rr: %p\n", f);
 	return f;
 }
 
@@ -565,12 +618,17 @@ static struct fio_file *get_next_file(struct thread_data *td)
 
 	assert(td->o.nr_files <= td->files_index);
 
-	if (!td->nr_open_files || td->nr_done_files >= td->o.nr_files)
+	if (!td->nr_open_files || td->nr_done_files >= td->o.nr_files) {
+		dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d,"
+				" nr_files=%d\n", td->nr_open_files,
+						  td->nr_done_files,
+						  td->o.nr_files);
 		return NULL;
+	}
 
 	f = td->file_service_file;
 	if (f && (f->flags & FIO_FILE_OPEN) && td->file_service_left--)
-		return f;
+		goto out;
 
 	if (td->o.file_service_type == FIO_FSERVICE_RR)
 		f = get_next_file_rr(td, FIO_FILE_OPEN, FIO_FILE_CLOSING);
@@ -579,6 +637,8 @@ static struct fio_file *get_next_file(struct thread_data *td)
 
 	td->file_service_file = f;
 	td->file_service_left = td->file_service_nr - 1;
+out:
+	dprint(FD_FILE, "get_next_file: %p\n", f);
 	return f;
 }
 
@@ -679,8 +739,10 @@ struct io_u *get_io_u(struct thread_data *td)
 	struct io_u *io_u;
 
 	io_u = __get_io_u(td);
-	if (!io_u)
+	if (!io_u) {
+		dprint(FD_IO, "__get_io_u failed\n");
 		return NULL;
+	}
 
 	/*
 	 * from a requeue, io_u already setup
@@ -694,15 +756,19 @@ struct io_u *get_io_u(struct thread_data *td)
 	if (td->o.read_iolog_file) {
 		if (read_iolog_get(td, io_u))
 			goto err_put;
-	} else if (set_io_u_file(td, io_u))
+	} else if (set_io_u_file(td, io_u)) {
+		dprint(FD_IO, "io_u %p, setting file failed\n", io_u);
 		goto err_put;
-	
+	}
+
 	f = io_u->file;
 	assert(f->flags & FIO_FILE_OPEN);
 
 	if (io_u->ddir != DDIR_SYNC) {
-		if (!io_u->buflen)
+		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;
+		}
 
 		f->last_pos = io_u->offset + io_u->buflen;
 
@@ -714,15 +780,15 @@ struct io_u *get_io_u(struct thread_data *td)
 	 * Set io data pointers.
 	 */
 	io_u->endpos = io_u->offset + io_u->buflen;
-out:
 	io_u->xfer_buf = io_u->buf;
 	io_u->xfer_buflen = io_u->buflen;
-
+out:
 	if (!td_io_prep(td, io_u)) {
 		fio_gettime(&io_u->start_time, NULL);
 		return io_u;
 	}
 err_put:
+	dprint(FD_IO, "get_io_u failed\n");
 	put_io_u(td, io_u);
 	return NULL;
 }
@@ -738,7 +804,8 @@ void io_u_log_error(struct thread_data *td, struct io_u *io_u)
 
 	log_err(": %s\n", strerror(io_u->error));
 
-	log_err("     %s offset=%llu, buflen=%lu\n", msg[io_u->ddir], io_u->offset, io_u->xfer_buflen);
+	log_err("     %s offset=%llu, buflen=%lu\n", msg[io_u->ddir],
+					io_u->offset, io_u->xfer_buflen);
 
 	if (!td->error)
 		td_verror(td, io_u->error, "io_u error");
@@ -749,6 +816,8 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 {
 	unsigned long usec;
 
+	dprint_io_u(io_u, "io complete");
+
 	assert(io_u->flags & IO_U_F_FLIGHT);
 	io_u->flags &= ~IO_U_F_FLIGHT;
 
@@ -768,8 +837,6 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 		td->io_bytes[idx] += bytes;
 		td->this_io_bytes[idx] += bytes;
 
-		io_u->file->last_completed_pos = io_u->endpos;
-
 		usec = utime_since(&io_u->issue_time, &icd->time);
 
 		add_clat_sample(td, idx, usec);
@@ -846,6 +913,8 @@ long io_u_queued_complete(struct thread_data *td, int min_events)
 	int ret;
 	struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, };
 
+	dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_events);
+
 	if (!min_events)
 		tvp = &ts;
 
@@ -895,7 +964,10 @@ static void io_u_dump(struct io_u *io_u)
 	unsigned long t_issue = mtime_since_now(&io_u->issue_time);
 
 	log_err("io_u=%p, t_start=%lu, t_issue=%lu\n", io_u, t_start, t_issue);
-	log_err("  buf=%p/%p, len=%lu/%lu, offset=%llu\n", io_u->buf, io_u->xfer_buf, io_u->buflen, io_u->xfer_buflen, io_u->offset);
+	log_err("  buf=%p/%p, len=%lu/%lu, offset=%llu\n", io_u->buf,
+						io_u->xfer_buf, io_u->buflen,
+						io_u->xfer_buflen,
+						io_u->offset);
 	log_err("  ddir=%d, fname=%s\n", io_u->ddir, io_u->file->file_name);
 }
 #else