md: use separate bio pool for each md device.

bio_clone and bio_alloc allocate from a common bio pool.
If an md device is stacked with other devices that use this pool, or under
something like swap which uses the pool, then the multiple calls on
the pool can cause deadlocks.

So allocate a local bio pool for each md array and use that rather
than the common pool.

This pool is used both for regular IO and metadata updates.

Signed-off-by: NeilBrown <neilb@suse.de>

6 files changed, 95 insertions, 12 deletions

diff --git a/drivers/md/faulty.c b/drivers/md/faulty.c
index 1a8987884614..339fdc670751 100644
--- a/drivers/md/faulty.c
+++ b/drivers/md/faulty.c
@@ -210,7 +210,7 @@ static int make_request(mddev_t *mddev, struct bio *bio)
 		}
 	}
 	if (failit) {
-		struct bio *b = bio_clone(bio, GFP_NOIO);
+		struct bio *b = bio_clone_mddev(bio, GFP_NOIO, mddev);
 		b->bi_bdev = conf->rdev->bdev;
 		b->bi_private = bio;
 		b->bi_end_io = faulty_fail;
diff --git a/drivers/md/md.c b/drivers/md/md.c
index 3149bf33f1c5..4e957f3140a8 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -149,6 +149,72 @@ static const struct block_device_operations md_fops;
 
 static int start_readonly;
 
+/* bio_clone_mddev
+ * like bio_clone, but with a local bio set
+ */
+
+static void mddev_bio_destructor(struct bio *bio)
+{
+	mddev_t *mddev, **mddevp;
+
+	mddevp = (void*)bio;
+	mddev = mddevp[-1];
+
+	bio_free(bio, mddev->bio_set);
+}
+
+struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
+			    mddev_t *mddev)
+{
+	struct bio *b;
+	mddev_t **mddevp;
+
+	if (!mddev || !mddev->bio_set)
+		return bio_alloc(gfp_mask, nr_iovecs);
+
+	b = bio_alloc_bioset(gfp_mask, nr_iovecs,
+			     mddev->bio_set);
+	if (!b)
+		return NULL;
+	mddevp = (void*)b;
+	mddevp[-1] = mddev;
+	b->bi_destructor = mddev_bio_destructor;
+	return b;
+}
+EXPORT_SYMBOL_GPL(bio_alloc_mddev);
+
+struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
+			    mddev_t *mddev)
+{
+	struct bio *b;
+	mddev_t **mddevp;
+
+	if (!mddev || !mddev->bio_set)
+		return bio_clone(bio, gfp_mask);
+
+	b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
+			     mddev->bio_set);
+	if (!b)
+		return NULL;
+	mddevp = (void*)b;
+	mddevp[-1] = mddev;
+	b->bi_destructor = mddev_bio_destructor;
+	__bio_clone(b, bio);
+	if (bio_integrity(bio)) {
+		int ret;
+
+		ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
+
+		if (ret < 0) {
+			bio_put(b);
+			return NULL;
+		}
+	}
+
+	return b;
+}
+EXPORT_SYMBOL_GPL(bio_clone_mddev);
+
 /*
  * We have a system wide 'event count' that is incremented
  * on any 'interesting' event, and readers of /proc/mdstat
@@ -321,7 +387,7 @@ static void submit_flushes(mddev_t *mddev)
 			atomic_inc(&rdev->nr_pending);
 			atomic_inc(&rdev->nr_pending);
 			rcu_read_unlock();
-			bi = bio_alloc(GFP_KERNEL, 0);
+			bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
 			bi->bi_end_io = md_end_flush;
 			bi->bi_private = rdev;
 			bi->bi_bdev = rdev->bdev;
@@ -428,6 +494,8 @@ static void mddev_delayed_delete(struct work_struct *ws);
 
 static void mddev_put(mddev_t *mddev)
 {
+	struct bio_set *bs = NULL;
+
 	if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 		return;
 	if (!mddev->raid_disks && list_empty(&mddev->disks) &&
@@ -435,6 +503,8 @@ static void mddev_put(mddev_t *mddev)
 		/* Array is not configured at all, and not held active,
 		 * so destroy it */
 		list_del(&mddev->all_mddevs);
+		bs = mddev->bio_set;
+		mddev->bio_set = NULL;
 		if (mddev->gendisk) {
 			/* We did a probe so need to clean up.  Call
 			 * queue_work inside the spinlock so that
@@ -447,6 +517,8 @@ static void mddev_put(mddev_t *mddev)
 			kfree(mddev);
 	}
 	spin_unlock(&all_mddevs_lock);
+	if (bs)
+		bioset_free(bs);
 }
 
 void mddev_init(mddev_t *mddev)
@@ -690,7 +762,7 @@ void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 	 * if zero is reached.
 	 * If an error occurred, call md_error
 	 */
-	struct bio *bio = bio_alloc(GFP_NOIO, 1);
+	struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
 
 	bio->bi_bdev = rdev->bdev;
 	bio->bi_sector = sector;
@@ -724,7 +796,7 @@ static void bi_complete(struct bio *bio, int error)
 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
 		 struct page *page, int rw)
 {
-	struct bio *bio = bio_alloc(GFP_NOIO, 1);
+	struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
 	struct completion event;
 	int ret;
 
@@ -4379,6 +4451,9 @@ int md_run(mddev_t *mddev)
 		sysfs_notify_dirent_safe(rdev->sysfs_state);
 	}
 
+	if (mddev->bio_set == NULL)
+		mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
+
 	spin_lock(&pers_lock);
 	pers = find_pers(mddev->level, mddev->clevel);
 	if (!pers || !try_module_get(pers->owner)) {
diff --git a/drivers/md/md.h b/drivers/md/md.h
index 5ee537135553..d05bab55df4e 100644
--- a/drivers/md/md.h
+++ b/drivers/md/md.h
@@ -331,6 +331,8 @@ struct mddev_s
 	struct attribute_group		*to_remove;
 	struct plug_handle		*plug; /* if used by personality */
 
+	struct bio_set			*bio_set;
+
 	/* Generic flush handling.
 	 * The last to finish preflush schedules a worker to submit
 	 * the rest of the request (without the REQ_FLUSH flag).
@@ -517,4 +519,8 @@ extern void md_rdev_init(mdk_rdev_t *rdev);
 
 extern void mddev_suspend(mddev_t *mddev);
 extern void mddev_resume(mddev_t *mddev);
+extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
+				   mddev_t *mddev);
+extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
+				   mddev_t *mddev);
 #endif /* _MD_MD_H */
diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
index 08c66afea494..54f60c9e5f85 100644
--- a/drivers/md/raid1.c
+++ b/drivers/md/raid1.c
@@ -861,7 +861,7 @@ static int make_request(mddev_t *mddev, struct bio * bio)
 		}
 		r1_bio->read_disk = rdisk;
 
-		read_bio = bio_clone(bio, GFP_NOIO);
+		read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
 
 		r1_bio->bios[rdisk] = read_bio;
 
@@ -950,7 +950,7 @@ static int make_request(mddev_t *mddev, struct bio * bio)
 		if (!r1_bio->bios[i])
 			continue;
 
-		mbio = bio_clone(bio, GFP_NOIO);
+		mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
 		r1_bio->bios[i] = mbio;
 
 		mbio->bi_sector	= r1_bio->sector + conf->mirrors[i].rdev->data_offset;
@@ -1640,7 +1640,8 @@ static void raid1d(mddev_t *mddev)
 					mddev->ro ? IO_BLOCKED : NULL;
 				r1_bio->read_disk = disk;
 				bio_put(bio);
-				bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
+				bio = bio_clone_mddev(r1_bio->master_bio,
+						      GFP_NOIO, mddev);
 				r1_bio->bios[r1_bio->read_disk] = bio;
 				rdev = conf->mirrors[disk].rdev;
 				if (printk_ratelimit())
diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index d54122b9927e..c67aa54694ae 100644
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c
@@ -889,7 +889,7 @@ static int make_request(mddev_t *mddev, struct bio * bio)
 		}
 		mirror = conf->mirrors + disk;
 
-		read_bio = bio_clone(bio, GFP_NOIO);
+		read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
 
 		r10_bio->devs[slot].bio = read_bio;
 
@@ -958,7 +958,7 @@ static int make_request(mddev_t *mddev, struct bio * bio)
 		if (!r10_bio->devs[i].bio)
 			continue;
 
-		mbio = bio_clone(bio, GFP_NOIO);
+		mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
 		r10_bio->devs[i].bio = mbio;
 
 		mbio->bi_sector	= r10_bio->devs[i].addr+
@@ -1746,7 +1746,8 @@ static void raid10d(mddev_t *mddev)
 					       mdname(mddev),
 					       bdevname(rdev->bdev,b),
 					       (unsigned long long)r10_bio->sector);
-				bio = bio_clone(r10_bio->master_bio, GFP_NOIO);
+				bio = bio_clone_mddev(r10_bio->master_bio,
+						      GFP_NOIO, mddev);
 				r10_bio->devs[r10_bio->read_slot].bio = bio;
 				bio->bi_sector = r10_bio->devs[r10_bio->read_slot].addr
 					+ rdev->data_offset;
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 8abc159b377a..dc574f303f8b 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -3876,9 +3876,9 @@ static int chunk_aligned_read(mddev_t *mddev, struct bio * raid_bio)
 		return 0;
 	}
 	/*
-	 * use bio_clone to make a copy of the bio
+	 * use bio_clone_mddev to make a copy of the bio
 	 */
-	align_bi = bio_clone(raid_bio, GFP_NOIO);
+	align_bi = bio_clone_mddev(raid_bio, GFP_NOIO, mddev);
 	if (!align_bi)
 		return 0;
 	/*