Reinstate zvol_taskq to fix aio on zvol

Commit 37f9dac removed the zvol_taskq for processing zvol requests.
This was removed as part of switching to make_request_fn and was
motivated by a concern at the time over dispatch latency.

However, this also made all bio request synchronous, and caused
serious performance issues as the bio submitter would wait for
every bio it submitted, effectively making the IO depth 1.

This patch reinstate zvol_taskq, and to make sure overlapped I/Os
are ordered properly, we take range lock in zvol_request, and pass
it along with bio to the I/O functions zvol_{write,discard,read}.

In order to facilitate benchmarks a zvol_request_sync module
option was added to switch between sync and async request handling.
For the moment, the default behavior is synchronous but this is
likely to change pending additional testing.

Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Chunwei Chen <[email protected]>
Closes #5824

3 files changed, 199 insertions, 82 deletions

diff --git a/include/linux/blkdev_compat.h b/include/linux/blkdev_compat.h
index ad095cdf1..15824c0b8 100644
--- a/include/linux/blkdev_compat.h
+++ b/include/linux/blkdev_compat.h
@@ -498,8 +498,15 @@ blk_queue_discard_granularity(struct request_queue *q, unsigned int dg)
 #define	VDEV_HOLDER			((void *)0x2401de7)
 
 #ifndef HAVE_GENERIC_IO_ACCT
-#define	generic_start_io_acct(rw, slen, part)		((void)0)
-#define	generic_end_io_acct(rw, part, start_jiffies)	((void)0)
+static inline void
+generic_start_io_acct(int rw, unsigned long sectors, struct hd_struct *part)
+{
+}
+
+static inline void
+generic_end_io_acct(int rw, struct hd_struct *part, unsigned long start_time)
+{
+}
 #endif
 
 #endif /* _ZFS_BLKDEV_H */
diff --git a/man/man5/zfs-module-parameters.5 b/man/man5/zfs-module-parameters.5
index f85c28bf8..77f1af6d5 100644
--- a/man/man5/zfs-module-parameters.5
+++ b/man/man5/zfs-module-parameters.5
@@ -2034,6 +2034,31 @@ Default value: \fB131,072\fR.
 .sp
 .ne 2
 .na
+\fBzvol_request_sync\fR (uint)
+.ad
+.RS 12n
+When processing I/O requests for a zvol submit them synchronously.  This
+effectively limits the queue depth to 1 for each I/O submitter.  When set
+to 0 requests are handled asynchronously by a thread pool.  The number of
+requests which can be handled concurrently is controller by \fBzvol_threads\fR.
+.sp
+Default value: \fB1\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fBzvol_threads\fR (uint)
+.ad
+.RS 12n
+Max number of threads which can handle zvol I/O requests concurrently.
+.sp
+Default value: \fB32\fR.
+.RE
+
+.sp
+.ne 2
+.na
 \fBzfs_qat_disable\fR (int)
 .ad
 .RS 12n
diff --git a/module/zfs/zvol.c b/module/zfs/zvol.c
index 420282c4a..7590ed316 100644
--- a/module/zfs/zvol.c
+++ b/module/zfs/zvol.c
@@ -56,9 +56,12 @@
 
 unsigned int zvol_inhibit_dev = 0;
 unsigned int zvol_major = ZVOL_MAJOR;
+unsigned int zvol_threads = 32;
+unsigned int zvol_request_sync = 1;
 unsigned int zvol_prefetch_bytes = (128 * 1024);
 unsigned long zvol_max_discard_blocks = 16384;
 
+static taskq_t *zvol_taskq;
 static kmutex_t zvol_state_lock;
 static list_t zvol_state_list;
 
@@ -636,21 +639,48 @@ zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
 	}
 }
 
-static int
-zvol_write(zvol_state_t *zv, uio_t *uio, boolean_t sync)
+typedef struct zv_request {
+	zvol_state_t	*zv;
+	struct bio	*bio;
+	rl_t		*rl;
+} zv_request_t;
+
+static void
+uio_from_bio(uio_t *uio, struct bio *bio)
 {
+	uio->uio_bvec = &bio->bi_io_vec[BIO_BI_IDX(bio)];
+	uio->uio_skip = BIO_BI_SKIP(bio);
+	uio->uio_resid = BIO_BI_SIZE(bio);
+	uio->uio_iovcnt = bio->bi_vcnt - BIO_BI_IDX(bio);
+	uio->uio_loffset = BIO_BI_SECTOR(bio) << 9;
+	uio->uio_limit = MAXOFFSET_T;
+	uio->uio_segflg = UIO_BVEC;
+}
+
+static void
+zvol_write(void *arg)
+{
+	zv_request_t *zvr = arg;
+	struct bio *bio = zvr->bio;
+	uio_t uio;
+	zvol_state_t *zv = zvr->zv;
 	uint64_t volsize = zv->zv_volsize;
-	rl_t *rl;
+	boolean_t sync;
 	int error = 0;
+	unsigned long start_jif;
+
+	uio_from_bio(&uio, bio);
 
 	ASSERT(zv && zv->zv_open_count > 0);
 
-	rl = zfs_range_lock(&zv->zv_range_lock, uio->uio_loffset,
-	    uio->uio_resid, RL_WRITER);
+	start_jif = jiffies;
+	generic_start_io_acct(WRITE, bio_sectors(bio), &zv->zv_disk->part0);
 
-	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
-		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
-		uint64_t off = uio->uio_loffset;
+	sync = bio_is_fua(bio) || zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS;
+
+	while (uio.uio_resid > 0 && uio.uio_loffset < volsize) {
+		uint64_t bytes = MIN(uio.uio_resid, DMU_MAX_ACCESS >> 1);
+		uint64_t off = uio.uio_loffset;
 		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
 
 		if (bytes > volsize - off)	/* don't write past the end */
@@ -664,7 +694,7 @@ zvol_write(zvol_state_t *zv, uio_t *uio, boolean_t sync)
 			dmu_tx_abort(tx);
 			break;
 		}
-		error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
+		error = dmu_write_uio_dbuf(zv->zv_dbuf, &uio, bytes, tx);
 		if (error == 0)
 			zvol_log_write(zv, tx, off, bytes, sync);
 		dmu_tx_commit(tx);
@@ -672,10 +702,14 @@ zvol_write(zvol_state_t *zv, uio_t *uio, boolean_t sync)
 		if (error)
 			break;
 	}
-	zfs_range_unlock(rl);
+	zfs_range_unlock(zvr->rl);
 	if (sync)
 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
-	return (error);
+
+	rw_exit(&zv->zv_suspend_lock);
+	generic_end_io_acct(WRITE, &zv->zv_disk->part0, start_jif);
+	BIO_END_IO(bio, -error);
+	kmem_free(zvr, sizeof (zv_request_t));
 }
 
 /*
@@ -702,21 +736,28 @@ zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
 	zil_itx_assign(zilog, itx, tx);
 }
 
-static int
-zvol_discard(struct bio *bio)
+static void
+zvol_discard(void *arg)
 {
-	zvol_state_t *zv = bio->bi_bdev->bd_disk->private_data;
+	zv_request_t *zvr = arg;
+	struct bio *bio = zvr->bio;
+	zvol_state_t *zv = zvr->zv;
 	uint64_t start = BIO_BI_SECTOR(bio) << 9;
 	uint64_t size = BIO_BI_SIZE(bio);
 	uint64_t end = start + size;
-	int error;
-	rl_t *rl;
+	int error = 0;
 	dmu_tx_t *tx;
+	unsigned long start_jif;
 
 	ASSERT(zv && zv->zv_open_count > 0);
 
-	if (end > zv->zv_volsize)
-		return (SET_ERROR(EIO));
+	start_jif = jiffies;
+	generic_start_io_acct(WRITE, bio_sectors(bio), &zv->zv_disk->part0);
+
+	if (end > zv->zv_volsize) {
+		error = SET_ERROR(EIO);
+		goto out;
+	}
 
 	/*
 	 * Align the request to volume block boundaries when a secure erase is
@@ -731,9 +772,8 @@ zvol_discard(struct bio *bio)
 	}
 
 	if (start >= end)
-		return (0);
+		goto out;
 
-	rl = zfs_range_lock(&zv->zv_range_lock, start, size, RL_WRITER);
 	tx = dmu_tx_create(zv->zv_objset);
 	dmu_tx_mark_netfree(tx);
 	error = dmu_tx_assign(tx, TXG_WAIT);
@@ -746,30 +786,40 @@ zvol_discard(struct bio *bio)
 		    ZVOL_OBJ, start, size);
 	}
 
-	zfs_range_unlock(rl);
-
-	return (error);
+out:
+	zfs_range_unlock(zvr->rl);
+	rw_exit(&zv->zv_suspend_lock);
+	generic_end_io_acct(WRITE, &zv->zv_disk->part0, start_jif);
+	BIO_END_IO(bio, -error);
+	kmem_free(zvr, sizeof (zv_request_t));
 }
 
-static int
-zvol_read(zvol_state_t *zv, uio_t *uio)
+static void
+zvol_read(void *arg)
 {
+	zv_request_t *zvr = arg;
+	struct bio *bio = zvr->bio;
+	uio_t uio;
+	zvol_state_t *zv = zvr->zv;
 	uint64_t volsize = zv->zv_volsize;
-	rl_t *rl;
 	int error = 0;
+	unsigned long start_jif;
+
+	uio_from_bio(&uio, bio);
 
 	ASSERT(zv && zv->zv_open_count > 0);
 
-	rl = zfs_range_lock(&zv->zv_range_lock, uio->uio_loffset,
-	    uio->uio_resid, RL_READER);
-	while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
-		uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
+	start_jif = jiffies;
+	generic_start_io_acct(READ, bio_sectors(bio), &zv->zv_disk->part0);
+
+	while (uio.uio_resid > 0 && uio.uio_loffset < volsize) {
+		uint64_t bytes = MIN(uio.uio_resid, DMU_MAX_ACCESS >> 1);
 
 		/* don't read past the end */
-		if (bytes > volsize - uio->uio_loffset)
-			bytes = volsize - uio->uio_loffset;
+		if (bytes > volsize - uio.uio_loffset)
+			bytes = volsize - uio.uio_loffset;
 
-		error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes);
+		error = dmu_read_uio_dbuf(zv->zv_dbuf, &uio, bytes);
 		if (error) {
 			/* convert checksum errors into IO errors */
 			if (error == ECKSUM)
@@ -777,76 +827,93 @@ zvol_read(zvol_state_t *zv, uio_t *uio)
 			break;
 		}
 	}
-	zfs_range_unlock(rl);
-	return (error);
+	zfs_range_unlock(zvr->rl);
+
+	rw_exit(&zv->zv_suspend_lock);
+	generic_end_io_acct(READ, &zv->zv_disk->part0, start_jif);
+	BIO_END_IO(bio, -error);
+	kmem_free(zvr, sizeof (zv_request_t));
 }
 
 static MAKE_REQUEST_FN_RET
 zvol_request(struct request_queue *q, struct bio *bio)
 {
-	uio_t uio;
 	zvol_state_t *zv = q->queuedata;
 	fstrans_cookie_t cookie = spl_fstrans_mark();
+	uint64_t offset = BIO_BI_SECTOR(bio) << 9;
+	uint64_t size = BIO_BI_SIZE(bio);
 	int rw = bio_data_dir(bio);
-#ifdef HAVE_GENERIC_IO_ACCT
-	unsigned long start = jiffies;
-#endif
-	int error = 0;
-
-	rw_enter(&zv->zv_suspend_lock, RW_READER);
-
-	uio.uio_bvec = &bio->bi_io_vec[BIO_BI_IDX(bio)];
-	uio.uio_skip = BIO_BI_SKIP(bio);
-	uio.uio_resid = BIO_BI_SIZE(bio);
-	uio.uio_iovcnt = bio->bi_vcnt - BIO_BI_IDX(bio);
-	uio.uio_loffset = BIO_BI_SECTOR(bio) << 9;
-	uio.uio_limit = MAXOFFSET_T;
-	uio.uio_segflg = UIO_BVEC;
+	zv_request_t *zvr;
 
-	if (bio_has_data(bio) && uio.uio_loffset + uio.uio_resid >
-	    zv->zv_volsize) {
+	if (bio_has_data(bio) && offset + size > zv->zv_volsize) {
 		printk(KERN_INFO
 		    "%s: bad access: offset=%llu, size=%lu\n",
 		    zv->zv_disk->disk_name,
-		    (long long unsigned)uio.uio_loffset,
-		    (long unsigned)uio.uio_resid);
-		error = SET_ERROR(EIO);
-		goto out1;
-	}
+		    (long long unsigned)offset,
+		    (long unsigned)size);
 
-	generic_start_io_acct(rw, bio_sectors(bio), &zv->zv_disk->part0);
+		BIO_END_IO(bio, -SET_ERROR(EIO));
+		goto out;
+	}
 
 	if (rw == WRITE) {
 		if (unlikely(zv->zv_flags & ZVOL_RDONLY)) {
-			error = SET_ERROR(EROFS);
-			goto out2;
+			BIO_END_IO(bio, -SET_ERROR(EROFS));
+			goto out;
 		}
 
-		if (bio_is_discard(bio) || bio_is_secure_erase(bio)) {
-			error = zvol_discard(bio);
-			goto out2;
+		/*
+		 * To be released in the I/O function. See the comment on
+		 * zfs_range_lock below.
+		 */
+		rw_enter(&zv->zv_suspend_lock, RW_READER);
+
+		/* bio marked as FLUSH need to flush before write */
+		if (bio_is_flush(bio))
+			zil_commit(zv->zv_zilog, ZVOL_OBJ);
+
+		/* Some requests are just for flush and nothing else. */
+		if (size == 0) {
+			rw_exit(&zv->zv_suspend_lock);
+			BIO_END_IO(bio, 0);
+			goto out;
 		}
 
+		zvr = kmem_alloc(sizeof (zv_request_t), KM_SLEEP);
+		zvr->zv = zv;
+		zvr->bio = bio;
+
 		/*
-		 * Some requests are just for flush and nothing else.
+		 * To be released in the I/O function. Since the I/O functions
+		 * are asynchronous, we take it here synchronously to make
+		 * sure overlapped I/Os are properly ordered.
 		 */
-		if (uio.uio_resid == 0) {
-			if (bio_is_flush(bio))
-				zil_commit(zv->zv_zilog, ZVOL_OBJ);
-			goto out2;
+		zvr->rl = zfs_range_lock(&zv->zv_range_lock, offset, size,
+		    RL_WRITER);
+		if (bio_is_discard(bio) || bio_is_secure_erase(bio)) {
+			if (zvol_request_sync || taskq_dispatch(zvol_taskq,
+			    zvol_discard, zvr, TQ_SLEEP) == TASKQID_INVALID)
+				zvol_discard(zvr);
+		} else {
+			if (zvol_request_sync || taskq_dispatch(zvol_taskq,
+			    zvol_write, zvr, TQ_SLEEP) == TASKQID_INVALID)
+				zvol_write(zvr);
 		}
+	} else {
+		zvr = kmem_alloc(sizeof (zv_request_t), KM_SLEEP);
+		zvr->zv = zv;
+		zvr->bio = bio;
 
-		error = zvol_write(zv, &uio,
-		    bio_is_flush(bio) || bio_is_fua(bio) ||
-		    zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
-	} else
-		error = zvol_read(zv, &uio);
+		rw_enter(&zv->zv_suspend_lock, RW_READER);
 
-out2:
-	generic_end_io_acct(rw, &zv->zv_disk->part0, start);
-out1:
-	BIO_END_IO(bio, -error);
-	rw_exit(&zv->zv_suspend_lock);
+		zvr->rl = zfs_range_lock(&zv->zv_range_lock, offset, size,
+		    RL_READER);
+		if (zvol_request_sync || taskq_dispatch(zvol_taskq,
+		    zvol_read, zvr, TQ_SLEEP) == TASKQID_INVALID)
+			zvol_read(zvr);
+	}
+
+out:
 	spl_fstrans_unmark(cookie);
 #ifdef HAVE_MAKE_REQUEST_FN_RET_INT
 	return (0);
@@ -2166,6 +2233,7 @@ zvol_rename_minors(spa_t *spa, const char *name1, const char *name2,
 int
 zvol_init(void)
 {
+	int threads = MIN(MAX(zvol_threads, 1), 1024);
 	int i, error;
 
 	list_create(&zvol_state_list, sizeof (zvol_state_t),
@@ -2173,11 +2241,19 @@ zvol_init(void)
 	mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL);
 	ida_init(&zvol_ida);
 
+	zvol_taskq = taskq_create(ZVOL_DRIVER, threads, maxclsyspri,
+	    threads * 2, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC);
+	if (zvol_taskq == NULL) {
+		printk(KERN_INFO "ZFS: taskq_create() failed\n");
+		error = -ENOMEM;
+		goto out;
+	}
+
 	zvol_htable = kmem_alloc(ZVOL_HT_SIZE * sizeof (struct hlist_head),
 	    KM_SLEEP);
 	if (!zvol_htable) {
-		error = ENOMEM;
-		goto out;
+		error = -ENOMEM;
+		goto out_taskq;
 	}
 	for (i = 0; i < ZVOL_HT_SIZE; i++)
 		INIT_HLIST_HEAD(&zvol_htable[i]);
@@ -2195,6 +2271,8 @@ zvol_init(void)
 
 out_free:
 	kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head));
+out_taskq:
+	taskq_destroy(zvol_taskq);
 out:
 	mutex_destroy(&zvol_state_lock);
 	list_destroy(&zvol_state_list);
@@ -2211,6 +2289,7 @@ zvol_fini(void)
 	unregister_blkdev(zvol_major, ZVOL_DRIVER);
 	kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head));
 
+	taskq_destroy(zvol_taskq);
 	list_destroy(&zvol_state_list);
 	mutex_destroy(&zvol_state_lock);
 
@@ -2224,6 +2303,12 @@ MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes");
 module_param(zvol_major, uint, 0444);
 MODULE_PARM_DESC(zvol_major, "Major number for zvol device");
 
+module_param(zvol_threads, uint, 0444);
+MODULE_PARM_DESC(zvol_threads, "Max number of threads to handle I/O requests");
+
+module_param(zvol_request_sync, uint, 0644);
+MODULE_PARM_DESC(zvol_request_sync, "Synchronously handle bio requests");
+
 module_param(zvol_max_discard_blocks, ulong, 0444);
 MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard");