Merge tag 'md/3.13' of git://neil.brown.name/md

Pull md update from Neil Brown:
 "Mostly optimisations and obscure bug fixes.
   - raid5 gets less lock contention
   - raid1 gets less contention between normal-io and resync-io during
     resync"

* tag 'md/3.13' of git://neil.brown.name/md:
  md/raid5: Use conf->device_lock protect changing of multi-thread resources.
  md/raid5: Before freeing old multi-thread worker, it should flush them.
  md/raid5: For stripe with R5_ReadNoMerge, we replace REQ_FLUSH with REQ_NOMERGE.
  UAPI: include <asm/byteorder.h> in linux/raid/md_p.h
  raid1: Rewrite the implementation of iobarrier.
  raid1: Add some macros to make code clearly.
  raid1: Replace raise_barrier/lower_barrier with freeze_array/unfreeze_array when reconfiguring the array.
  raid1: Add a field array_frozen to indicate whether raid in freeze state.
  md: Convert use of typedef ctl_table to struct ctl_table
  md/raid5: avoid deadlock when raid5 array has unack badblocks during md_stop_writes.
  md: use MD_RECOVERY_INTR instead of kthread_should_stop in resync thread.
  md: fix some places where mddev_lock return value is not checked.
  raid5: Retry R5_ReadNoMerge flag when hit a read error.
  raid5: relieve lock contention in get_active_stripe()
  raid5: relieve lock contention in get_active_stripe()
  wait: add wait_event_cmd()
  md/raid5.c: add proper locking to error path of raid5_start_reshape.
  md: fix calculation of stacking limits on level change.
  raid5: Use slow_path to release stripe when mddev->thread is null

drivers/md/md.c

@@ -112,7 +112,7 @@ static inline int speed_max(struct mddev *mddev)
 
 static struct ctl_table_header *raid_table_header;
 
-static ctl_table raid_table[] = {
+static struct ctl_table raid_table[] = {
 	{
 		.procname	= "speed_limit_min",
 		.data		= &sysctl_speed_limit_min,
@@ -130,7 +130,7 @@ static ctl_table raid_table[] = {
 	{ }
 };
 
-static ctl_table raid_dir_table[] = {
+static struct ctl_table raid_dir_table[] = {
 	{
 		.procname	= "raid",
 		.maxlen		= 0,
@@ -140,7 +140,7 @@ static ctl_table raid_dir_table[] = {
 	{ }
 };
 
-static ctl_table raid_root_table[] = {
+static struct ctl_table raid_root_table[] = {
 	{
 		.procname	= "dev",
 		.maxlen		= 0,
@@ -562,11 +562,19 @@ static struct mddev * mddev_find(dev_t unit)
 	goto retry;
 }
 
-static inline int mddev_lock(struct mddev * mddev)
+static inline int __must_check mddev_lock(struct mddev * mddev)
 {
 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
 }
 
+/* Sometimes we need to take the lock in a situation where
+ * failure due to interrupts is not acceptable.
+ */
+static inline void mddev_lock_nointr(struct mddev * mddev)
+{
+	mutex_lock(&mddev->reconfig_mutex);
+}
+
 static inline int mddev_is_locked(struct mddev *mddev)
 {
 	return mutex_is_locked(&mddev->reconfig_mutex);
@@ -2978,7 +2986,7 @@ rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
 		for_each_mddev(mddev, tmp) {
 			struct md_rdev *rdev2;
 
-			mddev_lock(mddev);
+			mddev_lock_nointr(mddev);
 			rdev_for_each(rdev2, mddev)
 				if (rdev->bdev == rdev2->bdev &&
 				    rdev != rdev2 &&
@@ -2994,7 +3002,7 @@ rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
 				break;
 			}
 		}
-		mddev_lock(my_mddev);
+		mddev_lock_nointr(my_mddev);
 		if (overlap) {
 			/* Someone else could have slipped in a size
 			 * change here, but doing so is just silly.
@@ -3580,6 +3588,7 @@ level_store(struct mddev *mddev, const char *buf, size_t len)
 		mddev->in_sync = 1;
 		del_timer_sync(&mddev->safemode_timer);
 	}
+	blk_set_stacking_limits(&mddev->queue->limits);
 	pers->run(mddev);
 	set_bit(MD_CHANGE_DEVS, &mddev->flags);
 	mddev_resume(mddev);
@@ -5258,7 +5267,7 @@ static void __md_stop_writes(struct mddev *mddev)
 
 void md_stop_writes(struct mddev *mddev)
 {
-	mddev_lock(mddev);
+	mddev_lock_nointr(mddev);
 	__md_stop_writes(mddev);
 	mddev_unlock(mddev);
 }
@@ -5291,20 +5300,35 @@ EXPORT_SYMBOL_GPL(md_stop);
 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
 {
 	int err = 0;
+	int did_freeze = 0;
+
+	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
+		did_freeze = 1;
+		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+		md_wakeup_thread(mddev->thread);
+	}
+	if (mddev->sync_thread) {
+		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
+		/* Thread might be blocked waiting for metadata update
+		 * which will now never happen */
+		wake_up_process(mddev->sync_thread->tsk);
+	}
+	mddev_unlock(mddev);
+	wait_event(resync_wait, mddev->sync_thread == NULL);
+	mddev_lock_nointr(mddev);
+
 	mutex_lock(&mddev->open_mutex);
-	if (atomic_read(&mddev->openers) > !!bdev) {
+	if (atomic_read(&mddev->openers) > !!bdev ||
+	    mddev->sync_thread ||
+	    (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
 		printk("md: %s still in use.\n",mdname(mddev));
+		if (did_freeze) {
+			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+			md_wakeup_thread(mddev->thread);
+		}
 		err = -EBUSY;
 		goto out;
 	}
-	if (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags)) {
-		/* Someone opened the device since we flushed it
-		 * so page cache could be dirty and it is too late
-		 * to flush.  So abort
-		 */
-		mutex_unlock(&mddev->open_mutex);
-		return -EBUSY;
-	}
 	if (mddev->pers) {
 		__md_stop_writes(mddev);
 
@@ -5315,7 +5339,7 @@ static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
 		set_disk_ro(mddev->gendisk, 1);
 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
 		sysfs_notify_dirent_safe(mddev->sysfs_state);
-		err = 0;	
+		err = 0;
 	}
 out:
 	mutex_unlock(&mddev->open_mutex);
@@ -5331,20 +5355,34 @@ static int do_md_stop(struct mddev * mddev, int mode,
 {
 	struct gendisk *disk = mddev->gendisk;
 	struct md_rdev *rdev;
+	int did_freeze = 0;
+
+	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
+		did_freeze = 1;
+		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+		md_wakeup_thread(mddev->thread);
+	}
+	if (mddev->sync_thread) {
+		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
+		/* Thread might be blocked waiting for metadata update
+		 * which will now never happen */
+		wake_up_process(mddev->sync_thread->tsk);
+	}
+	mddev_unlock(mddev);
+	wait_event(resync_wait, mddev->sync_thread == NULL);
+	mddev_lock_nointr(mddev);
 
 	mutex_lock(&mddev->open_mutex);
 	if (atomic_read(&mddev->openers) > !!bdev ||
-	    mddev->sysfs_active) {
+	    mddev->sysfs_active ||
+	    mddev->sync_thread ||
+	    (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
 		printk("md: %s still in use.\n",mdname(mddev));
 		mutex_unlock(&mddev->open_mutex);
-		return -EBUSY;
-	}
-	if (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags)) {
-		/* Someone opened the device since we flushed it
-		 * so page cache could be dirty and it is too late
-		 * to flush.  So abort
-		 */
-		mutex_unlock(&mddev->open_mutex);
+		if (did_freeze) {
+			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+			md_wakeup_thread(mddev->thread);
+		}
 		return -EBUSY;
 	}
 	if (mddev->pers) {
@@ -6551,7 +6589,7 @@ static int md_ioctl(struct block_device *bdev, fmode_t mode,
 				wait_event(mddev->sb_wait,
 					   !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
 					   !test_bit(MD_CHANGE_PENDING, &mddev->flags));
-				mddev_lock(mddev);
+				mddev_lock_nointr(mddev);
 			}
 		} else {
 			err = -EROFS;
@@ -7361,9 +7399,6 @@ void md_do_sync(struct md_thread *thread)
 		mddev->curr_resync = 2;
 
 	try_again:
-		if (kthread_should_stop())
-			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
-
 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
 			goto skip;
 		for_each_mddev(mddev2, tmp) {
@@ -7388,7 +7423,7 @@ void md_do_sync(struct md_thread *thread)
 				 * be caught by 'softlockup'
 				 */
 				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
-				if (!kthread_should_stop() &&
+				if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
 				    mddev2->curr_resync >= mddev->curr_resync) {
 					printk(KERN_INFO "md: delaying %s of %s"
 					       " until %s has finished (they"
@@ -7464,7 +7499,7 @@ void md_do_sync(struct md_thread *thread)
 	last_check = 0;
 
 	if (j>2) {
-		printk(KERN_INFO 
+		printk(KERN_INFO
 		       "md: resuming %s of %s from checkpoint.\n",
 		       desc, mdname(mddev));
 		mddev->curr_resync = j;
@@ -7501,7 +7536,8 @@ void md_do_sync(struct md_thread *thread)
 			sysfs_notify(&mddev->kobj, NULL, "sync_completed");
 		}
 
-		while (j >= mddev->resync_max && !kthread_should_stop()) {
+		while (j >= mddev->resync_max &&
+		       !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
 			/* As this condition is controlled by user-space,
 			 * we can block indefinitely, so use '_interruptible'
 			 * to avoid triggering warnings.
@@ -7509,17 +7545,18 @@ void md_do_sync(struct md_thread *thread)
 			flush_signals(current); /* just in case */
 			wait_event_interruptible(mddev->recovery_wait,
 						 mddev->resync_max > j
-						 || kthread_should_stop());
+						 || test_bit(MD_RECOVERY_INTR,
+							     &mddev->recovery));
 		}
 
-		if (kthread_should_stop())
-			goto interrupted;
+		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
+			break;
 
 		sectors = mddev->pers->sync_request(mddev, j, &skipped,
 						  currspeed < speed_min(mddev));
 		if (sectors == 0) {
 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
-			goto out;
+			break;
 		}
 
 		if (!skipped) { /* actual IO requested */
@@ -7556,10 +7593,8 @@ void md_do_sync(struct md_thread *thread)
 			last_mark = next;
 		}
 
-
-		if (kthread_should_stop())
-			goto interrupted;
-
+		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
+			break;
 
 		/*
 		 * this loop exits only if either when we are slower than
@@ -7582,11 +7617,12 @@ void md_do_sync(struct md_thread *thread)
 			}
 		}
 	}
-	printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
+	printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
+	       test_bit(MD_RECOVERY_INTR, &mddev->recovery)
+	       ? "interrupted" : "done");
 	/*
 	 * this also signals 'finished resyncing' to md_stop
 	 */
- out:
 	blk_finish_plug(&plug);
 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
 
@@ -7640,16 +7676,6 @@ void md_do_sync(struct md_thread *thread)
 	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
 	md_wakeup_thread(mddev->thread);
 	return;
-
- interrupted:
-	/*
-	 * got a signal, exit.
-	 */
-	printk(KERN_INFO
-	       "md: md_do_sync() got signal ... exiting\n");
-	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
-	goto out;
-
 }
 EXPORT_SYMBOL_GPL(md_do_sync);
 
@@ -7894,6 +7920,7 @@ void md_reap_sync_thread(struct mddev *mddev)
 
 	/* resync has finished, collect result */
 	md_unregister_thread(&mddev->sync_thread);
+	wake_up(&resync_wait);
 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
 	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
 		/* success...*/

drivers/md/raid1.c

@@ -66,7 +66,8 @@
  */
 static int max_queued_requests = 1024;
 
-static void allow_barrier(struct r1conf *conf);
+static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
+			  sector_t bi_sector);
 static void lower_barrier(struct r1conf *conf);
 
 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
@@ -84,10 +85,12 @@ static void r1bio_pool_free(void *r1_bio, void *data)
 }
 
 #define RESYNC_BLOCK_SIZE (64*1024)
-//#define RESYNC_BLOCK_SIZE PAGE_SIZE
+#define RESYNC_DEPTH 32
 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
-#define RESYNC_WINDOW (2048*1024)
+#define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
+#define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
+#define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
 
 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
 {
@@ -225,6 +228,8 @@ static void call_bio_endio(struct r1bio *r1_bio)
 	struct bio *bio = r1_bio->master_bio;
 	int done;
 	struct r1conf *conf = r1_bio->mddev->private;
+	sector_t start_next_window = r1_bio->start_next_window;
+	sector_t bi_sector = bio->bi_sector;
 
 	if (bio->bi_phys_segments) {
 		unsigned long flags;
@@ -232,6 +237,11 @@ static void call_bio_endio(struct r1bio *r1_bio)
 		bio->bi_phys_segments--;
 		done = (bio->bi_phys_segments == 0);
 		spin_unlock_irqrestore(&conf->device_lock, flags);
+		/*
+		 * make_request() might be waiting for
+		 * bi_phys_segments to decrease
+		 */
+		wake_up(&conf->wait_barrier);
 	} else
 		done = 1;
 
@@ -243,7 +253,7 @@ static void call_bio_endio(struct r1bio *r1_bio)
 		 * Wake up any possible resync thread that waits for the device
 		 * to go idle.
 		 */
-		allow_barrier(conf);
+		allow_barrier(conf, start_next_window, bi_sector);
 	}
 }
 
@@ -814,8 +824,6 @@ static void flush_pending_writes(struct r1conf *conf)
  *    there is no normal IO happeing.  It must arrange to call
  *    lower_barrier when the particular background IO completes.
  */
-#define RESYNC_DEPTH 32
-
 static void raise_barrier(struct r1conf *conf)
 {
 	spin_lock_irq(&conf->resync_lock);
@@ -827,9 +835,19 @@ static void raise_barrier(struct r1conf *conf)
 	/* block any new IO from starting */
 	conf->barrier++;
 
-	/* Now wait for all pending IO to complete */
+	/* For these conditions we must wait:
+	 * A: while the array is in frozen state
+	 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
+	 *    the max count which allowed.
+	 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
+	 *    next resync will reach to the window which normal bios are
+	 *    handling.
+	 */
 	wait_event_lock_irq(conf->wait_barrier,
-			    !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
+			    !conf->array_frozen &&
+			    conf->barrier < RESYNC_DEPTH &&
+			    (conf->start_next_window >=
+			     conf->next_resync + RESYNC_SECTORS),
 			    conf->resync_lock);
 
 	spin_unlock_irq(&conf->resync_lock);
@@ -845,10 +863,33 @@ static void lower_barrier(struct r1conf *conf)
 	wake_up(&conf->wait_barrier);
 }
 
-static void wait_barrier(struct r1conf *conf)
+static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
 {
+	bool wait = false;
+
+	if (conf->array_frozen || !bio)
+		wait = true;
+	else if (conf->barrier && bio_data_dir(bio) == WRITE) {
+		if (conf->next_resync < RESYNC_WINDOW_SECTORS)
+			wait = true;
+		else if ((conf->next_resync - RESYNC_WINDOW_SECTORS
+				>= bio_end_sector(bio)) ||
+			 (conf->next_resync + NEXT_NORMALIO_DISTANCE
+				<= bio->bi_sector))
+			wait = false;
+		else
+			wait = true;
+	}
+
+	return wait;
+}
+
+static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
+{
+	sector_t sector = 0;
+
 	spin_lock_irq(&conf->resync_lock);
-	if (conf->barrier) {
+	if (need_to_wait_for_sync(conf, bio)) {
 		conf->nr_waiting++;
 		/* Wait for the barrier to drop.
 		 * However if there are already pending
@@ -860,22 +901,67 @@ static void wait_barrier(struct r1conf *conf)
 		 * count down.
 		 */
 		wait_event_lock_irq(conf->wait_barrier,
-				    !conf->barrier ||
-				    (conf->nr_pending &&
+				    !conf->array_frozen &&
+				    (!conf->barrier ||
+				    ((conf->start_next_window <
+				      conf->next_resync + RESYNC_SECTORS) &&
 				     current->bio_list &&
-				     !bio_list_empty(current->bio_list)),
+				     !bio_list_empty(current->bio_list))),
 				    conf->resync_lock);
 		conf->nr_waiting--;
 	}
+
+	if (bio && bio_data_dir(bio) == WRITE) {
+		if (conf->next_resync + NEXT_NORMALIO_DISTANCE
+		    <= bio->bi_sector) {
+			if (conf->start_next_window == MaxSector)
+				conf->start_next_window =
+					conf->next_resync +
+					NEXT_NORMALIO_DISTANCE;
+
+			if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
+			    <= bio->bi_sector)
+				conf->next_window_requests++;
+			else
+				conf->current_window_requests++;
+		}
+		if (bio->bi_sector >= conf->start_next_window)
+			sector = conf->start_next_window;
+	}
+
 	conf->nr_pending++;
 	spin_unlock_irq(&conf->resync_lock);
+	return sector;
 }
 
-static void allow_barrier(struct r1conf *conf)
+static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
+			  sector_t bi_sector)
 {
 	unsigned long flags;
+
 	spin_lock_irqsave(&conf->resync_lock, flags);
 	conf->nr_pending--;
+	if (start_next_window) {
+		if (start_next_window == conf->start_next_window) {
+			if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
+			    <= bi_sector)
+				conf->next_window_requests--;
+			else
+				conf->current_window_requests--;
+		} else
+			conf->current_window_requests--;
+
+		if (!conf->current_window_requests) {
+			if (conf->next_window_requests) {
+				conf->current_window_requests =
+					conf->next_window_requests;
+				conf->next_window_requests = 0;
+				conf->start_next_window +=
+					NEXT_NORMALIO_DISTANCE;
+			} else
+				conf->start_next_window = MaxSector;
+		}
+	}
 	spin_unlock_irqrestore(&conf->resync_lock, flags);
 	wake_up(&conf->wait_barrier);
 }
@@ -884,8 +970,7 @@ static void freeze_array(struct r1conf *conf, int extra)
 {
 	/* stop syncio and normal IO and wait for everything to
 	 * go quite.
-	 * We increment barrier and nr_waiting, and then
-	 * wait until nr_pending match nr_queued+extra
+	 * We wait until nr_pending match nr_queued+extra
 	 * This is called in the context of one normal IO request
 	 * that has failed. Thus any sync request that might be pending
 	 * will be blocked by nr_pending, and we need to wait for
@@ -895,8 +980,7 @@ static void freeze_array(struct r1conf *conf, int extra)
 	 * we continue.
 	 */
 	spin_lock_irq(&conf->resync_lock);
-	conf->barrier++;
-	conf->nr_waiting++;
+	conf->array_frozen = 1;
 	wait_event_lock_irq_cmd(conf->wait_barrier,
 				conf->nr_pending == conf->nr_queued+extra,
 				conf->resync_lock,
@@ -907,8 +991,7 @@ static void unfreeze_array(struct r1conf *conf)
 {
 	/* reverse the effect of the freeze */
 	spin_lock_irq(&conf->resync_lock);
-	conf->barrier--;
-	conf->nr_waiting--;
+	conf->array_frozen = 0;
 	wake_up(&conf->wait_barrier);
 	spin_unlock_irq(&conf->resync_lock);
 }
@@ -1013,6 +1096,7 @@ static void make_request(struct mddev *mddev, struct bio * bio)
 	int first_clone;
 	int sectors_handled;
 	int max_sectors;
+	sector_t start_next_window;
 
 	/*
 	 * Register the new request and wait if the reconstruction
@@ -1042,7 +1126,7 @@ static void make_request(struct mddev *mddev, struct bio * bio)
 		finish_wait(&conf->wait_barrier, &w);
 	}
 
-	wait_barrier(conf);
+	start_next_window = wait_barrier(conf, bio);
 
 	bitmap = mddev->bitmap;
 
@@ -1163,6 +1247,7 @@ static void make_request(struct mddev *mddev, struct bio * bio)
 
 	disks = conf->raid_disks * 2;
  retry_write:
+	r1_bio->start_next_window = start_next_window;
 	blocked_rdev = NULL;
 	rcu_read_lock();
 	max_sectors = r1_bio->sectors;
@@ -1231,14 +1316,24 @@ static void make_request(struct mddev *mddev, struct bio * bio)
 	if (unlikely(blocked_rdev)) {
 		/* Wait for this device to become unblocked */
 		int j;
+		sector_t old = start_next_window;
 
 		for (j = 0; j < i; j++)
 			if (r1_bio->bios[j])
 				rdev_dec_pending(conf->mirrors[j].rdev, mddev);
 		r1_bio->state = 0;
-		allow_barrier(conf);
+		allow_barrier(conf, start_next_window, bio->bi_sector);
 		md_wait_for_blocked_rdev(blocked_rdev, mddev);
-		wait_barrier(conf);
+		start_next_window = wait_barrier(conf, bio);
+		/*
+		 * We must make sure the multi r1bios of bio have
+		 * the same value of bi_phys_segments
+		 */
+		if (bio->bi_phys_segments && old &&
+		    old != start_next_window)
+			/* Wait for the former r1bio(s) to complete */
+			wait_event(conf->wait_barrier,
+				   bio->bi_phys_segments == 1);
 		goto retry_write;
 	}
 
@@ -1438,11 +1533,14 @@ static void print_conf(struct r1conf *conf)
 
 static void close_sync(struct r1conf *conf)
 {
-	wait_barrier(conf);
-	allow_barrier(conf);
+	wait_barrier(conf, NULL);
+	allow_barrier(conf, 0, 0);
 
 	mempool_destroy(conf->r1buf_pool);
 	conf->r1buf_pool = NULL;
+
+	conf->next_resync = 0;
+	conf->start_next_window = MaxSector;
 }
 
 static int raid1_spare_active(struct mddev *mddev)
@@ -2714,6 +2812,9 @@ static struct r1conf *setup_conf(struct mddev *mddev)
 	conf->pending_count = 0;
 	conf->recovery_disabled = mddev->recovery_disabled - 1;
 
+	conf->start_next_window = MaxSector;
+	conf->current_window_requests = conf->next_window_requests = 0;
+
 	err = -EIO;
 	for (i = 0; i < conf->raid_disks * 2; i++) {
 
@@ -2871,8 +2972,8 @@ static int stop(struct mddev *mddev)
 			   atomic_read(&bitmap->behind_writes) == 0);
 	}
 
-	raise_barrier(conf);
-	lower_barrier(conf);
+	freeze_array(conf, 0);
+	unfreeze_array(conf);
 
 	md_unregister_thread(&mddev->thread);
 	if (conf->r1bio_pool)
@@ -3031,10 +3132,10 @@ static void raid1_quiesce(struct mddev *mddev, int state)
 		wake_up(&conf->wait_barrier);
 		break;
 	case 1:
-		raise_barrier(conf);
+		freeze_array(conf, 0);
 		break;
 	case 0:
-		lower_barrier(conf);
+		unfreeze_array(conf);
 		break;
 	}
 }
@@ -3051,7 +3152,8 @@ static void *raid1_takeover(struct mddev *mddev)
 		mddev->new_chunk_sectors = 0;
 		conf = setup_conf(mddev);
 		if (!IS_ERR(conf))
-			conf->barrier = 1;
+			/* Array must appear to be quiesced */
+			conf->array_frozen = 1;
 		return conf;
 	}
 	return ERR_PTR(-EINVAL);

drivers/md/raid1.h

@@ -41,6 +41,19 @@ struct r1conf {
 	 */
 	sector_t		next_resync;
 
+	/* When raid1 starts resync, we divide array into four partitions
+	 * |---------|--------------|---------------------|-------------|
+	 *        next_resync   start_next_window       end_window
+	 * start_next_window = next_resync + NEXT_NORMALIO_DISTANCE
+	 * end_window = start_next_window + NEXT_NORMALIO_DISTANCE
+	 * current_window_requests means the count of normalIO between
+	 *   start_next_window and end_window.
+	 * next_window_requests means the count of normalIO after end_window.
+	 * */
+	sector_t		start_next_window;
+	int			current_window_requests;
+	int			next_window_requests;
+
 	spinlock_t		device_lock;
 
 	/* list of 'struct r1bio' that need to be processed by raid1d,
@@ -65,6 +78,7 @@ struct r1conf {
 	int			nr_waiting;
 	int			nr_queued;
 	int			barrier;
+	int			array_frozen;
 
 	/* Set to 1 if a full sync is needed, (fresh device added).
 	 * Cleared when a sync completes.
@@ -111,6 +125,7 @@ struct r1bio {
 						 * in this BehindIO request
 						 */
 	sector_t		sector;
+	sector_t		start_next_window;
 	int			sectors;
 	unsigned long		state;
 	struct mddev		*mddev;

drivers/md/raid10.c

@@ -4384,7 +4384,11 @@ static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr,
 		set_bit(MD_CHANGE_DEVS, &mddev->flags);
 		md_wakeup_thread(mddev->thread);
 		wait_event(mddev->sb_wait, mddev->flags == 0 ||
-			   kthread_should_stop());
+			   test_bit(MD_RECOVERY_INTR, &mddev->recovery));
+		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
+			allow_barrier(conf);
+			return sectors_done;
+		}
 		conf->reshape_safe = mddev->reshape_position;
 		allow_barrier(conf);
 	}

drivers/md/raid5.c

@@ -85,6 +85,42 @@ static inline struct hlist_head *stripe_hash(struct r5conf *conf, sector_t sect)
 	return &conf->stripe_hashtbl[hash];
 }
 
+static inline int stripe_hash_locks_hash(sector_t sect)
+{
+	return (sect >> STRIPE_SHIFT) & STRIPE_HASH_LOCKS_MASK;
+}
+
+static inline void lock_device_hash_lock(struct r5conf *conf, int hash)
+{
+	spin_lock_irq(conf->hash_locks + hash);
+	spin_lock(&conf->device_lock);
+}
+
+static inline void unlock_device_hash_lock(struct r5conf *conf, int hash)
+{
+	spin_unlock(&conf->device_lock);
+	spin_unlock_irq(conf->hash_locks + hash);
+}
+
+static inline void lock_all_device_hash_locks_irq(struct r5conf *conf)
+{
+	int i;
+	local_irq_disable();
+	spin_lock(conf->hash_locks);
+	for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
+		spin_lock_nest_lock(conf->hash_locks + i, conf->hash_locks);
+	spin_lock(&conf->device_lock);
+}
+
+static inline void unlock_all_device_hash_locks_irq(struct r5conf *conf)
+{
+	int i;
+	spin_unlock(&conf->device_lock);
+	for (i = NR_STRIPE_HASH_LOCKS; i; i--)
+		spin_unlock(conf->hash_locks + i - 1);
+	local_irq_enable();
+}
+
 /* bio's attached to a stripe+device for I/O are linked together in bi_sector
  * order without overlap.  There may be several bio's per stripe+device, and
  * a bio could span several devices.
@@ -249,7 +285,8 @@ static void raid5_wakeup_stripe_thread(struct stripe_head *sh)
 	}
 }
 
-static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh)
+static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh,
+			      struct list_head *temp_inactive_list)
 {
 	BUG_ON(!list_empty(&sh->lru));
 	BUG_ON(atomic_read(&conf->active_stripes)==0);
@@ -278,23 +315,68 @@ static void do_release_stripe(struct r5conf *conf, struct stripe_head *sh)
 			    < IO_THRESHOLD)
 				md_wakeup_thread(conf->mddev->thread);
 		atomic_dec(&conf->active_stripes);
-		if (!test_bit(STRIPE_EXPANDING, &sh->state)) {
-			list_add_tail(&sh->lru, &conf->inactive_list);
-			wake_up(&conf->wait_for_stripe);
-			if (conf->retry_read_aligned)
-				md_wakeup_thread(conf->mddev->thread);
-		}
+		if (!test_bit(STRIPE_EXPANDING, &sh->state))
+			list_add_tail(&sh->lru, temp_inactive_list);
 	}
 }
 
-static void __release_stripe(struct r5conf *conf, struct stripe_head *sh)
+static void __release_stripe(struct r5conf *conf, struct stripe_head *sh,
+			     struct list_head *temp_inactive_list)
 {
 	if (atomic_dec_and_test(&sh->count))
-		do_release_stripe(conf, sh);
+		do_release_stripe(conf, sh, temp_inactive_list);
+}
+
+/*
+ * @hash could be NR_STRIPE_HASH_LOCKS, then we have a list of inactive_list
+ *
+ * Be careful: Only one task can add/delete stripes from temp_inactive_list at
+ * given time. Adding stripes only takes device lock, while deleting stripes
+ * only takes hash lock.
+ */
+static void release_inactive_stripe_list(struct r5conf *conf,
+					 struct list_head *temp_inactive_list,
+					 int hash)
+{
+	int size;
+	bool do_wakeup = false;
+	unsigned long flags;
+
+	if (hash == NR_STRIPE_HASH_LOCKS) {
+		size = NR_STRIPE_HASH_LOCKS;
+		hash = NR_STRIPE_HASH_LOCKS - 1;
+	} else
+		size = 1;
+	while (size) {
+		struct list_head *list = &temp_inactive_list[size - 1];
+
+		/*
+		 * We don't hold any lock here yet, get_active_stripe() might
+		 * remove stripes from the list
+		 */
+		if (!list_empty_careful(list)) {
+			spin_lock_irqsave(conf->hash_locks + hash, flags);
+			if (list_empty(conf->inactive_list + hash) &&
+			    !list_empty(list))
+				atomic_dec(&conf->empty_inactive_list_nr);
+			list_splice_tail_init(list, conf->inactive_list + hash);
+			do_wakeup = true;
+			spin_unlock_irqrestore(conf->hash_locks + hash, flags);
+		}
+		size--;
+		hash--;
+	}
+
+	if (do_wakeup) {
+		wake_up(&conf->wait_for_stripe);
+		if (conf->retry_read_aligned)
+			md_wakeup_thread(conf->mddev->thread);
+	}
 }
 
 /* should hold conf->device_lock already */
-static int release_stripe_list(struct r5conf *conf)
+static int release_stripe_list(struct r5conf *conf,
+			       struct list_head *temp_inactive_list)
 {
 	struct stripe_head *sh;
 	int count = 0;
@@ -303,6 +385,8 @@ static int release_stripe_list(struct r5conf *conf)
 	head = llist_del_all(&conf->released_stripes);
 	head = llist_reverse_order(head);
 	while (head) {
+		int hash;
+
 		sh = llist_entry(head, struct stripe_head, release_list);
 		head = llist_next(head);
 		/* sh could be readded after STRIPE_ON_RELEASE_LIST is cleard */
@@ -313,7 +397,8 @@ static int release_stripe_list(struct r5conf *conf)
 		 * again, the count is always > 1. This is true for
 		 * STRIPE_ON_UNPLUG_LIST bit too.
 		 */
-		__release_stripe(conf, sh);
+		hash = sh->hash_lock_index;
+		__release_stripe(conf, sh, &temp_inactive_list[hash]);
 		count++;
 	}
 
@@ -324,9 +409,12 @@ static void release_stripe(struct stripe_head *sh)
 {
 	struct r5conf *conf = sh->raid_conf;
 	unsigned long flags;
+	struct list_head list;
+	int hash;
 	bool wakeup;
 
-	if (test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
+	if (unlikely(!conf->mddev->thread) ||
+		test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state))
 		goto slow_path;
 	wakeup = llist_add(&sh->release_list, &conf->released_stripes);
 	if (wakeup)
@@ -336,8 +424,11 @@ static void release_stripe(struct stripe_head *sh)
 	local_irq_save(flags);
 	/* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */
 	if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) {
-		do_release_stripe(conf, sh);
+		INIT_LIST_HEAD(&list);
+		hash = sh->hash_lock_index;
+		do_release_stripe(conf, sh, &list);
 		spin_unlock(&conf->device_lock);
+		release_inactive_stripe_list(conf, &list, hash);
 	}
 	local_irq_restore(flags);
 }
@@ -362,18 +453,21 @@ static inline void insert_hash(struct r5conf *conf, struct stripe_head *sh)
 
 
 /* find an idle stripe, make sure it is unhashed, and return it. */
-static struct stripe_head *get_free_stripe(struct r5conf *conf)
+static struct stripe_head *get_free_stripe(struct r5conf *conf, int hash)
 {
 	struct stripe_head *sh = NULL;
 	struct list_head *first;
 
-	if (list_empty(&conf->inactive_list))
+	if (list_empty(conf->inactive_list + hash))
 		goto out;
-	first = conf->inactive_list.next;
+	first = (conf->inactive_list + hash)->next;
 	sh = list_entry(first, struct stripe_head, lru);
 	list_del_init(first);
 	remove_hash(sh);
 	atomic_inc(&conf->active_stripes);
+	BUG_ON(hash != sh->hash_lock_index);
+	if (list_empty(conf->inactive_list + hash))
+		atomic_inc(&conf->empty_inactive_list_nr);
 out:
 	return sh;
 }
@@ -416,7 +510,7 @@ static void stripe_set_idx(sector_t stripe, struct r5conf *conf, int previous,
 static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
 {
 	struct r5conf *conf = sh->raid_conf;
-	int i;
+	int i, seq;
 
 	BUG_ON(atomic_read(&sh->count) != 0);
 	BUG_ON(test_bit(STRIPE_HANDLE, &sh->state));
@@ -426,7 +520,8 @@ static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
 		(unsigned long long)sh->sector);
 
 	remove_hash(sh);
-
+retry:
+	seq = read_seqcount_begin(&conf->gen_lock);
 	sh->generation = conf->generation - previous;
 	sh->disks = previous ? conf->previous_raid_disks : conf->raid_disks;
 	sh->sector = sector;
@@ -448,6 +543,8 @@ static void init_stripe(struct stripe_head *sh, sector_t sector, int previous)
 		dev->flags = 0;
 		raid5_build_block(sh, i, previous);
 	}
+	if (read_seqcount_retry(&conf->gen_lock, seq))
+		goto retry;
 	insert_hash(conf, sh);
 	sh->cpu = smp_processor_id();
 }
@@ -552,29 +649,31 @@ get_active_stripe(struct r5conf *conf, sector_t sector,
 		  int previous, int noblock, int noquiesce)
 {
 	struct stripe_head *sh;
+	int hash = stripe_hash_locks_hash(sector);
 
 	pr_debug("get_stripe, sector %llu\n", (unsigned long long)sector);
 
-	spin_lock_irq(&conf->device_lock);
+	spin_lock_irq(conf->hash_locks + hash);
 
 	do {
 		wait_event_lock_irq(conf->wait_for_stripe,
 				    conf->quiesce == 0 || noquiesce,
-				    conf->device_lock);
+				    *(conf->hash_locks + hash));
 		sh = __find_stripe(conf, sector, conf->generation - previous);
 		if (!sh) {
 			if (!conf->inactive_blocked)
-				sh = get_free_stripe(conf);
+				sh = get_free_stripe(conf, hash);
 			if (noblock && sh == NULL)
 				break;
 			if (!sh) {
 				conf->inactive_blocked = 1;
-				wait_event_lock_irq(conf->wait_for_stripe,
-						    !list_empty(&conf->inactive_list) &&
-						    (atomic_read(&conf->active_stripes)
-						     < (conf->max_nr_stripes *3/4)
-						     || !conf->inactive_blocked),
-						    conf->device_lock);
+				wait_event_lock_irq(
+					conf->wait_for_stripe,
+					!list_empty(conf->inactive_list + hash) &&
+					(atomic_read(&conf->active_stripes)
+					 < (conf->max_nr_stripes * 3 / 4)
+					 || !conf->inactive_blocked),
+					*(conf->hash_locks + hash));
 				conf->inactive_blocked = 0;
 			} else
 				init_stripe(sh, sector, previous);
@@ -585,9 +684,11 @@ get_active_stripe(struct r5conf *conf, sector_t sector,
 				    && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)
 				    && !test_bit(STRIPE_ON_RELEASE_LIST, &sh->state));
 			} else {
+				spin_lock(&conf->device_lock);
 				if (!test_bit(STRIPE_HANDLE, &sh->state))
 					atomic_inc(&conf->active_stripes);
 				if (list_empty(&sh->lru) &&
+				    !test_bit(STRIPE_ON_RELEASE_LIST, &sh->state) &&
 				    !test_bit(STRIPE_EXPANDING, &sh->state))
 					BUG();
 				list_del_init(&sh->lru);
@@ -595,6 +696,7 @@ get_active_stripe(struct r5conf *conf, sector_t sector,
 					sh->group->stripes_cnt--;
 					sh->group = NULL;
 				}
+				spin_unlock(&conf->device_lock);
 			}
 		}
 	} while (sh == NULL);
@@ -602,7 +704,7 @@ get_active_stripe(struct r5conf *conf, sector_t sector,
 	if (sh)
 		atomic_inc(&sh->count);
 
-	spin_unlock_irq(&conf->device_lock);
+	spin_unlock_irq(conf->hash_locks + hash);
 	return sh;
 }
 
@@ -758,7 +860,7 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
 				bi->bi_sector = (sh->sector
 						 + rdev->data_offset);
 			if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
-				bi->bi_rw |= REQ_FLUSH;
+				bi->bi_rw |= REQ_NOMERGE;
 
 			bi->bi_vcnt = 1;
 			bi->bi_io_vec[0].bv_len = STRIPE_SIZE;
@@ -1582,7 +1684,7 @@ static void raid_run_ops(struct stripe_head *sh, unsigned long ops_request)
 	put_cpu();
 }
 
-static int grow_one_stripe(struct r5conf *conf)
+static int grow_one_stripe(struct r5conf *conf, int hash)
 {
 	struct stripe_head *sh;
 	sh = kmem_cache_zalloc(conf->slab_cache, GFP_KERNEL);
@@ -1598,6 +1700,7 @@ static int grow_one_stripe(struct r5conf *conf)
 		kmem_cache_free(conf->slab_cache, sh);
 		return 0;
 	}
+	sh->hash_lock_index = hash;
 	/* we just created an active stripe so... */
 	atomic_set(&sh->count, 1);
 	atomic_inc(&conf->active_stripes);
@@ -1610,6 +1713,7 @@ static int grow_stripes(struct r5conf *conf, int num)
 {
 	struct kmem_cache *sc;
 	int devs = max(conf->raid_disks, conf->previous_raid_disks);
+	int hash;
 
 	if (conf->mddev->gendisk)
 		sprintf(conf->cache_name[0],
@@ -1627,9 +1731,13 @@ static int grow_stripes(struct r5conf *conf, int num)
 		return 1;
 	conf->slab_cache = sc;
 	conf->pool_size = devs;
-	while (num--)
-		if (!grow_one_stripe(conf))
+	hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
+	while (num--) {
+		if (!grow_one_stripe(conf, hash))
 			return 1;
+		conf->max_nr_stripes++;
+		hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
+	}
 	return 0;
 }
 
@@ -1687,6 +1795,7 @@ static int resize_stripes(struct r5conf *conf, int newsize)
 	int err;
 	struct kmem_cache *sc;
 	int i;
+	int hash, cnt;
 
 	if (newsize <= conf->pool_size)
 		return 0; /* never bother to shrink */
@@ -1726,19 +1835,29 @@ static int resize_stripes(struct r5conf *conf, int newsize)
 	 * OK, we have enough stripes, start collecting inactive
 	 * stripes and copying them over
 	 */
+	hash = 0;
+	cnt = 0;
 	list_for_each_entry(nsh, &newstripes, lru) {
-		spin_lock_irq(&conf->device_lock);
-		wait_event_lock_irq(conf->wait_for_stripe,
-				    !list_empty(&conf->inactive_list),
-				    conf->device_lock);
-		osh = get_free_stripe(conf);
-		spin_unlock_irq(&conf->device_lock);
+		lock_device_hash_lock(conf, hash);
+		wait_event_cmd(conf->wait_for_stripe,
+				    !list_empty(conf->inactive_list + hash),
+				    unlock_device_hash_lock(conf, hash),
+				    lock_device_hash_lock(conf, hash));
+		osh = get_free_stripe(conf, hash);
+		unlock_device_hash_lock(conf, hash);
 		atomic_set(&nsh->count, 1);
 		for(i=0; i<conf->pool_size; i++)
 			nsh->dev[i].page = osh->dev[i].page;
 		for( ; i<newsize; i++)
 			nsh->dev[i].page = NULL;
+		nsh->hash_lock_index = hash;
 		kmem_cache_free(conf->slab_cache, osh);
+		cnt++;
+		if (cnt >= conf->max_nr_stripes / NR_STRIPE_HASH_LOCKS +
+		    !!((conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS) > hash)) {
+			hash++;
+			cnt = 0;
+		}
 	}
 	kmem_cache_destroy(conf->slab_cache);
 
@@ -1797,13 +1916,13 @@ static int resize_stripes(struct r5conf *conf, int newsize)
 	return err;
 }
 
-static int drop_one_stripe(struct r5conf *conf)
+static int drop_one_stripe(struct r5conf *conf, int hash)
 {
 	struct stripe_head *sh;
 
-	spin_lock_irq(&conf->device_lock);
-	sh = get_free_stripe(conf);
-	spin_unlock_irq(&conf->device_lock);
+	spin_lock_irq(conf->hash_locks + hash);
+	sh = get_free_stripe(conf, hash);
+	spin_unlock_irq(conf->hash_locks + hash);
 	if (!sh)
 		return 0;
 	BUG_ON(atomic_read(&sh->count));
@@ -1815,8 +1934,10 @@ static int drop_one_stripe(struct r5conf *conf)
 
 static void shrink_stripes(struct r5conf *conf)
 {
-	while (drop_one_stripe(conf))
-		;
+	int hash;
+	for (hash = 0; hash < NR_STRIPE_HASH_LOCKS; hash++)
+		while (drop_one_stripe(conf, hash))
+			;
 
 	if (conf->slab_cache)
 		kmem_cache_destroy(conf->slab_cache);
@@ -1921,6 +2042,9 @@ static void raid5_end_read_request(struct bio * bi, int error)
 			       mdname(conf->mddev), bdn);
 		else
 			retry = 1;
+		if (set_bad && test_bit(In_sync, &rdev->flags)
+		    && !test_bit(R5_ReadNoMerge, &sh->dev[i].flags))
+			retry = 1;
 		if (retry)
 			if (test_bit(R5_ReadNoMerge, &sh->dev[i].flags)) {
 				set_bit(R5_ReadError, &sh->dev[i].flags);
@@ -3900,7 +4024,8 @@ static void raid5_activate_delayed(struct r5conf *conf)
 	}
 }
 
-static void activate_bit_delay(struct r5conf *conf)
+static void activate_bit_delay(struct r5conf *conf,
+	struct list_head *temp_inactive_list)
 {
 	/* device_lock is held */
 	struct list_head head;
@@ -3908,9 +4033,11 @@ static void activate_bit_delay(struct r5conf *conf)
 	list_del_init(&conf->bitmap_list);
 	while (!list_empty(&head)) {
 		struct stripe_head *sh = list_entry(head.next, struct stripe_head, lru);
+		int hash;
 		list_del_init(&sh->lru);
 		atomic_inc(&sh->count);
-		__release_stripe(conf, sh);
+		hash = sh->hash_lock_index;
+		__release_stripe(conf, sh, &temp_inactive_list[hash]);
 	}
 }
 
@@ -3926,7 +4053,7 @@ int md_raid5_congested(struct mddev *mddev, int bits)
 		return 1;
 	if (conf->quiesce)
 		return 1;
-	if (list_empty_careful(&conf->inactive_list))
+	if (atomic_read(&conf->empty_inactive_list_nr))
 		return 1;
 
 	return 0;
@@ -4256,6 +4383,7 @@ static struct stripe_head *__get_priority_stripe(struct r5conf *conf, int group)
 struct raid5_plug_cb {
 	struct blk_plug_cb	cb;
 	struct list_head	list;
+	struct list_head	temp_inactive_list[NR_STRIPE_HASH_LOCKS];
 };
 
 static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
@@ -4266,6 +4394,7 @@ static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
 	struct mddev *mddev = cb->cb.data;
 	struct r5conf *conf = mddev->private;
 	int cnt = 0;
+	int hash;
 
 	if (cb->list.next && !list_empty(&cb->list)) {
 		spin_lock_irq(&conf->device_lock);
@@ -4283,11 +4412,14 @@ static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule)
 			 * STRIPE_ON_RELEASE_LIST could be set here. In that
 			 * case, the count is always > 1 here
 			 */
-			__release_stripe(conf, sh);
+			hash = sh->hash_lock_index;
+			__release_stripe(conf, sh, &cb->temp_inactive_list[hash]);
 			cnt++;
 		}
 		spin_unlock_irq(&conf->device_lock);
 	}
+	release_inactive_stripe_list(conf, cb->temp_inactive_list,
+				     NR_STRIPE_HASH_LOCKS);
 	if (mddev->queue)
 		trace_block_unplug(mddev->queue, cnt, !from_schedule);
 	kfree(cb);
@@ -4308,8 +4440,12 @@ static void release_stripe_plug(struct mddev *mddev,
 
 	cb = container_of(blk_cb, struct raid5_plug_cb, cb);
 
-	if (cb->list.next == NULL)
+	if (cb->list.next == NULL) {
+		int i;
 		INIT_LIST_HEAD(&cb->list);
+		for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
+			INIT_LIST_HEAD(cb->temp_inactive_list + i);
+	}
 
 	if (!test_and_set_bit(STRIPE_ON_UNPLUG_LIST, &sh->state))
 		list_add_tail(&sh->lru, &cb->list);
@@ -4692,14 +4828,19 @@ static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *sk
 	    time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) {
 		/* Cannot proceed until we've updated the superblock... */
 		wait_event(conf->wait_for_overlap,
-			   atomic_read(&conf->reshape_stripes)==0);
+			   atomic_read(&conf->reshape_stripes)==0
+			   || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
+		if (atomic_read(&conf->reshape_stripes) != 0)
+			return 0;
 		mddev->reshape_position = conf->reshape_progress;
 		mddev->curr_resync_completed = sector_nr;
 		conf->reshape_checkpoint = jiffies;
 		set_bit(MD_CHANGE_DEVS, &mddev->flags);
 		md_wakeup_thread(mddev->thread);
 		wait_event(mddev->sb_wait, mddev->flags == 0 ||
-			   kthread_should_stop());
+			   test_bit(MD_RECOVERY_INTR, &mddev->recovery));
+		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
+			return 0;
 		spin_lock_irq(&conf->device_lock);
 		conf->reshape_safe = mddev->reshape_position;
 		spin_unlock_irq(&conf->device_lock);
@@ -4782,7 +4923,10 @@ static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *sk
 	    >= mddev->resync_max - mddev->curr_resync_completed) {
 		/* Cannot proceed until we've updated the superblock... */
 		wait_event(conf->wait_for_overlap,
-			   atomic_read(&conf->reshape_stripes) == 0);
+			   atomic_read(&conf->reshape_stripes) == 0
+			   || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
+		if (atomic_read(&conf->reshape_stripes) != 0)
+			goto ret;
 		mddev->reshape_position = conf->reshape_progress;
 		mddev->curr_resync_completed = sector_nr;
 		conf->reshape_checkpoint = jiffies;
@@ -4790,13 +4934,16 @@ static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, int *sk
 		md_wakeup_thread(mddev->thread);
 		wait_event(mddev->sb_wait,
 			   !test_bit(MD_CHANGE_DEVS, &mddev->flags)
-			   || kthread_should_stop());
+			   || test_bit(MD_RECOVERY_INTR, &mddev->recovery));
+		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
+			goto ret;
 		spin_lock_irq(&conf->device_lock);
 		conf->reshape_safe = mddev->reshape_position;
 		spin_unlock_irq(&conf->device_lock);
 		wake_up(&conf->wait_for_overlap);
 		sysfs_notify(&mddev->kobj, NULL, "sync_completed");
 	}
+ret:
 	return reshape_sectors;
 }
 
@@ -4954,27 +5101,45 @@ static int  retry_aligned_read(struct r5conf *conf, struct bio *raid_bio)
 }
 
 static int handle_active_stripes(struct r5conf *conf, int group,
-				 struct r5worker *worker)
+				 struct r5worker *worker,
+				 struct list_head *temp_inactive_list)
 {
 	struct stripe_head *batch[MAX_STRIPE_BATCH], *sh;
-	int i, batch_size = 0;
+	int i, batch_size = 0, hash;
+	bool release_inactive = false;
 
 	while (batch_size < MAX_STRIPE_BATCH &&
 			(sh = __get_priority_stripe(conf, group)) != NULL)
 		batch[batch_size++] = sh;
 
-	if (batch_size == 0)
-		return batch_size;
+	if (batch_size == 0) {
+		for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
+			if (!list_empty(temp_inactive_list + i))
+				break;
+		if (i == NR_STRIPE_HASH_LOCKS)
+			return batch_size;
+		release_inactive = true;
+	}
 	spin_unlock_irq(&conf->device_lock);
 
+	release_inactive_stripe_list(conf, temp_inactive_list,
+				     NR_STRIPE_HASH_LOCKS);
+
+	if (release_inactive) {
+		spin_lock_irq(&conf->device_lock);
+		return 0;
+	}
+
 	for (i = 0; i < batch_size; i++)
 		handle_stripe(batch[i]);
 
 	cond_resched();
 
 	spin_lock_irq(&conf->device_lock);
-	for (i = 0; i < batch_size; i++)
-		__release_stripe(conf, batch[i]);
+	for (i = 0; i < batch_size; i++) {
+		hash = batch[i]->hash_lock_index;
+		__release_stripe(conf, batch[i], &temp_inactive_list[hash]);
+	}
 	return batch_size;
 }
 
@@ -4995,9 +5160,10 @@ static void raid5_do_work(struct work_struct *work)
 	while (1) {
 		int batch_size, released;
 
-		released = release_stripe_list(conf);
+		released = release_stripe_list(conf, worker->temp_inactive_list);
 
-		batch_size = handle_active_stripes(conf, group_id, worker);
+		batch_size = handle_active_stripes(conf, group_id, worker,
+						   worker->temp_inactive_list);
 		worker->working = false;
 		if (!batch_size && !released)
 			break;
@@ -5036,7 +5202,7 @@ static void raid5d(struct md_thread *thread)
 		struct bio *bio;
 		int batch_size, released;
 
-		released = release_stripe_list(conf);
+		released = release_stripe_list(conf, conf->temp_inactive_list);
 
 		if (
 		    !list_empty(&conf->bitmap_list)) {
@@ -5046,7 +5212,7 @@ static void raid5d(struct md_thread *thread)
 			bitmap_unplug(mddev->bitmap);
 			spin_lock_irq(&conf->device_lock);
 			conf->seq_write = conf->seq_flush;
-			activate_bit_delay(conf);
+			activate_bit_delay(conf, conf->temp_inactive_list);
 		}
 		raid5_activate_delayed(conf);
 
@@ -5060,7 +5226,8 @@ static void raid5d(struct md_thread *thread)
 			handled++;
 		}
 
-		batch_size = handle_active_stripes(conf, ANY_GROUP, NULL);
+		batch_size = handle_active_stripes(conf, ANY_GROUP, NULL,
+						   conf->temp_inactive_list);
 		if (!batch_size && !released)
 			break;
 		handled += batch_size;
@@ -5096,22 +5263,29 @@ raid5_set_cache_size(struct mddev *mddev, int size)
 {
 	struct r5conf *conf = mddev->private;
 	int err;
+	int hash;
 
 	if (size <= 16 || size > 32768)
 		return -EINVAL;
+	hash = (conf->max_nr_stripes - 1) % NR_STRIPE_HASH_LOCKS;
 	while (size < conf->max_nr_stripes) {
-		if (drop_one_stripe(conf))
+		if (drop_one_stripe(conf, hash))
 			conf->max_nr_stripes--;
 		else
 			break;
+		hash--;
+		if (hash < 0)
+			hash = NR_STRIPE_HASH_LOCKS - 1;
 	}
 	err = md_allow_write(mddev);
 	if (err)
 		return err;
+	hash = conf->max_nr_stripes % NR_STRIPE_HASH_LOCKS;
 	while (size > conf->max_nr_stripes) {
-		if (grow_one_stripe(conf))
+		if (grow_one_stripe(conf, hash))
 			conf->max_nr_stripes++;
 		else break;
+		hash = (hash + 1) % NR_STRIPE_HASH_LOCKS;
 	}
 	return 0;
 }
@@ -5199,15 +5373,18 @@ raid5_show_group_thread_cnt(struct mddev *mddev, char *page)
 		return 0;
 }
 
-static int alloc_thread_groups(struct r5conf *conf, int cnt);
+static int alloc_thread_groups(struct r5conf *conf, int cnt,
+			       int *group_cnt,
+			       int *worker_cnt_per_group,
+			       struct r5worker_group **worker_groups);
 static ssize_t
 raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
 {
 	struct r5conf *conf = mddev->private;
 	unsigned long new;
 	int err;
-	struct r5worker_group *old_groups;
-	int old_group_cnt;
+	struct r5worker_group *new_groups, *old_groups;
+	int group_cnt, worker_cnt_per_group;
 
 	if (len >= PAGE_SIZE)
 		return -EINVAL;
@@ -5223,14 +5400,19 @@ raid5_store_group_thread_cnt(struct mddev *mddev, const char *page, size_t len)
 	mddev_suspend(mddev);
 
 	old_groups = conf->worker_groups;
-	old_group_cnt = conf->worker_cnt_per_group;
+	if (old_groups)
+		flush_workqueue(raid5_wq);
+
+	err = alloc_thread_groups(conf, new,
+				  &group_cnt, &worker_cnt_per_group,
+				  &new_groups);
+	if (!err) {
+		spin_lock_irq(&conf->device_lock);
+		conf->group_cnt = group_cnt;
+		conf->worker_cnt_per_group = worker_cnt_per_group;
+		conf->worker_groups = new_groups;
+		spin_unlock_irq(&conf->device_lock);
 
-	conf->worker_groups = NULL;
-	err = alloc_thread_groups(conf, new);
-	if (err) {
-		conf->worker_groups = old_groups;
-		conf->worker_cnt_per_group = old_group_cnt;
-	} else {
 		if (old_groups)
 			kfree(old_groups[0].workers);
 		kfree(old_groups);
@@ -5260,40 +5442,47 @@ static struct attribute_group raid5_attrs_group = {
 	.attrs = raid5_attrs,
 };
 
-static int alloc_thread_groups(struct r5conf *conf, int cnt)
+static int alloc_thread_groups(struct r5conf *conf, int cnt,
+			       int *group_cnt,
+			       int *worker_cnt_per_group,
+			       struct r5worker_group **worker_groups)
 {
-	int i, j;
+	int i, j, k;
 	ssize_t size;
 	struct r5worker *workers;
 
-	conf->worker_cnt_per_group = cnt;
+	*worker_cnt_per_group = cnt;
 	if (cnt == 0) {
-		conf->worker_groups = NULL;
+		*group_cnt = 0;
+		*worker_groups = NULL;
 		return 0;
 	}
-	conf->group_cnt = num_possible_nodes();
+	*group_cnt = num_possible_nodes();
 	size = sizeof(struct r5worker) * cnt;
-	workers = kzalloc(size * conf->group_cnt, GFP_NOIO);
-	conf->worker_groups = kzalloc(sizeof(struct r5worker_group) *
-				conf->group_cnt, GFP_NOIO);
-	if (!conf->worker_groups || !workers) {
+	workers = kzalloc(size * *group_cnt, GFP_NOIO);
+	*worker_groups = kzalloc(sizeof(struct r5worker_group) *
+				*group_cnt, GFP_NOIO);
+	if (!*worker_groups || !workers) {
 		kfree(workers);
-		kfree(conf->worker_groups);
-		conf->worker_groups = NULL;
+		kfree(*worker_groups);
 		return -ENOMEM;
 	}
 
-	for (i = 0; i < conf->group_cnt; i++) {
+	for (i = 0; i < *group_cnt; i++) {
 		struct r5worker_group *group;
 
-		group = &conf->worker_groups[i];
+		group = worker_groups[i];
 		INIT_LIST_HEAD(&group->handle_list);
 		group->conf = conf;
 		group->workers = workers + i * cnt;
 
 		for (j = 0; j < cnt; j++) {
-			group->workers[j].group = group;
-			INIT_WORK(&group->workers[j].work, raid5_do_work);
+			struct r5worker *worker = group->workers + j;
+			worker->group = group;
+			INIT_WORK(&worker->work, raid5_do_work);
+
+			for (k = 0; k < NR_STRIPE_HASH_LOCKS; k++)
+				INIT_LIST_HEAD(worker->temp_inactive_list + k);
 		}
 	}
 
@@ -5444,6 +5633,9 @@ static struct r5conf *setup_conf(struct mddev *mddev)
 	struct md_rdev *rdev;
 	struct disk_info *disk;
 	char pers_name[6];
+	int i;
+	int group_cnt, worker_cnt_per_group;
+	struct r5worker_group *new_group;
 
 	if (mddev->new_level != 5
 	    && mddev->new_level != 4
@@ -5478,7 +5670,12 @@ static struct r5conf *setup_conf(struct mddev *mddev)
 	if (conf == NULL)
 		goto abort;
 	/* Don't enable multi-threading by default*/
-	if (alloc_thread_groups(conf, 0))
+	if (!alloc_thread_groups(conf, 0, &group_cnt, &worker_cnt_per_group,
+				 &new_group)) {
+		conf->group_cnt = group_cnt;
+		conf->worker_cnt_per_group = worker_cnt_per_group;
+		conf->worker_groups = new_group;
+	} else
 		goto abort;
 	spin_lock_init(&conf->device_lock);
 	seqcount_init(&conf->gen_lock);
@@ -5488,7 +5685,6 @@ static struct r5conf *setup_conf(struct mddev *mddev)
 	INIT_LIST_HEAD(&conf->hold_list);
 	INIT_LIST_HEAD(&conf->delayed_list);
 	INIT_LIST_HEAD(&conf->bitmap_list);
-	INIT_LIST_HEAD(&conf->inactive_list);
 	init_llist_head(&conf->released_stripes);
 	atomic_set(&conf->active_stripes, 0);
 	atomic_set(&conf->preread_active_stripes, 0);
@@ -5514,6 +5710,21 @@ static struct r5conf *setup_conf(struct mddev *mddev)
 	if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
 		goto abort;
 
+	/* We init hash_locks[0] separately to that it can be used
+	 * as the reference lock in the spin_lock_nest_lock() call
+	 * in lock_all_device_hash_locks_irq in order to convince
+	 * lockdep that we know what we are doing.
+	 */
+	spin_lock_init(conf->hash_locks);
+	for (i = 1; i < NR_STRIPE_HASH_LOCKS; i++)
+		spin_lock_init(conf->hash_locks + i);
+
+	for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
+		INIT_LIST_HEAD(conf->inactive_list + i);
+
+	for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
+		INIT_LIST_HEAD(conf->temp_inactive_list + i);
+
 	conf->level = mddev->new_level;
 	if (raid5_alloc_percpu(conf) != 0)
 		goto abort;
@@ -5554,7 +5765,6 @@ static struct r5conf *setup_conf(struct mddev *mddev)
 	else
 		conf->max_degraded = 1;
 	conf->algorithm = mddev->new_layout;
-	conf->max_nr_stripes = NR_STRIPES;
 	conf->reshape_progress = mddev->reshape_position;
 	if (conf->reshape_progress != MaxSector) {
 		conf->prev_chunk_sectors = mddev->chunk_sectors;
@@ -5563,7 +5773,8 @@ static struct r5conf *setup_conf(struct mddev *mddev)
 
 	memory = conf->max_nr_stripes * (sizeof(struct stripe_head) +
 		 max_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024;
-	if (grow_stripes(conf, conf->max_nr_stripes)) {
+	atomic_set(&conf->empty_inactive_list_nr, NR_STRIPE_HASH_LOCKS);
+	if (grow_stripes(conf, NR_STRIPES)) {
 		printk(KERN_ERR
 		       "md/raid:%s: couldn't allocate %dkB for buffers\n",
 		       mdname(mddev), memory);
@@ -6369,12 +6580,18 @@ static int raid5_start_reshape(struct mddev *mddev)
 	if (!mddev->sync_thread) {
 		mddev->recovery = 0;
 		spin_lock_irq(&conf->device_lock);
+		write_seqcount_begin(&conf->gen_lock);
 		mddev->raid_disks = conf->raid_disks = conf->previous_raid_disks;
+		mddev->new_chunk_sectors =
+			conf->chunk_sectors = conf->prev_chunk_sectors;
+		mddev->new_layout = conf->algorithm = conf->prev_algo;
 		rdev_for_each(rdev, mddev)
 			rdev->new_data_offset = rdev->data_offset;
 		smp_wmb();
+		conf->generation --;
 		conf->reshape_progress = MaxSector;
 		mddev->reshape_position = MaxSector;
+		write_seqcount_end(&conf->gen_lock);
 		spin_unlock_irq(&conf->device_lock);
 		return -EAGAIN;
 	}
@@ -6462,27 +6679,28 @@ static void raid5_quiesce(struct mddev *mddev, int state)
 		break;
 
 	case 1: /* stop all writes */
-		spin_lock_irq(&conf->device_lock);
+		lock_all_device_hash_locks_irq(conf);
 		/* '2' tells resync/reshape to pause so that all
 		 * active stripes can drain
 		 */
 		conf->quiesce = 2;
-		wait_event_lock_irq(conf->wait_for_stripe,
+		wait_event_cmd(conf->wait_for_stripe,
 				    atomic_read(&conf->active_stripes) == 0 &&
 				    atomic_read(&conf->active_aligned_reads) == 0,
-				    conf->device_lock);
+				    unlock_all_device_hash_locks_irq(conf),
+				    lock_all_device_hash_locks_irq(conf));
 		conf->quiesce = 1;
-		spin_unlock_irq(&conf->device_lock);
+		unlock_all_device_hash_locks_irq(conf);
 		/* allow reshape to continue */
 		wake_up(&conf->wait_for_overlap);
 		break;
 
 	case 0: /* re-enable writes */
-		spin_lock_irq(&conf->device_lock);
+		lock_all_device_hash_locks_irq(conf);
 		conf->quiesce = 0;
 		wake_up(&conf->wait_for_stripe);
 		wake_up(&conf->wait_for_overlap);
-		spin_unlock_irq(&conf->device_lock);
+		unlock_all_device_hash_locks_irq(conf);
 		break;
 	}
 }

drivers/md/raid5.h

@@ -205,6 +205,7 @@ struct stripe_head {
 	short			pd_idx;		/* parity disk index */
 	short			qd_idx;		/* 'Q' disk index for raid6 */
 	short			ddf_layout;/* use DDF ordering to calculate Q */
+	short			hash_lock_index;
 	unsigned long		state;		/* state flags */
 	atomic_t		count;	      /* nr of active thread/requests */
 	int			bm_seq;	/* sequence number for bitmap flushes */
@@ -367,9 +368,18 @@ struct disk_info {
 	struct md_rdev	*rdev, *replacement;
 };
 
+/* NOTE NR_STRIPE_HASH_LOCKS must remain below 64.
+ * This is because we sometimes take all the spinlocks
+ * and creating that much locking depth can cause
+ * problems.
+ */
+#define NR_STRIPE_HASH_LOCKS 8
+#define STRIPE_HASH_LOCKS_MASK (NR_STRIPE_HASH_LOCKS - 1)
+
 struct r5worker {
 	struct work_struct work;
 	struct r5worker_group *group;
+	struct list_head temp_inactive_list[NR_STRIPE_HASH_LOCKS];
 	bool working;
 };
 
@@ -382,6 +392,8 @@ struct r5worker_group {
 
 struct r5conf {
 	struct hlist_head	*stripe_hashtbl;
+	/* only protect corresponding hash list and inactive_list */
+	spinlock_t		hash_locks[NR_STRIPE_HASH_LOCKS];
 	struct mddev		*mddev;
 	int			chunk_sectors;
 	int			level, algorithm;
@@ -462,7 +474,8 @@ struct r5conf {
 	 * Free stripes pool
 	 */
 	atomic_t		active_stripes;
-	struct list_head	inactive_list;
+	struct list_head	inactive_list[NR_STRIPE_HASH_LOCKS];
+	atomic_t		empty_inactive_list_nr;
 	struct llist_head	released_stripes;
 	wait_queue_head_t	wait_for_stripe;
 	wait_queue_head_t	wait_for_overlap;
@@ -477,6 +490,7 @@ struct r5conf {
 	 * the new thread here until we fully activate the array.
 	 */
 	struct md_thread	*thread;
+	struct list_head	temp_inactive_list[NR_STRIPE_HASH_LOCKS];
 	struct r5worker_group	*worker_groups;
 	int			group_cnt;
 	int			worker_cnt_per_group;

include/linux/wait.h

@@ -278,6 +278,31 @@ do {									\
 	__ret;								\
 })
 
+#define __wait_event_cmd(wq, condition, cmd1, cmd2)			\
+	(void)___wait_event(wq, condition, TASK_UNINTERRUPTIBLE, 0, 0,	\
+			    cmd1; schedule(); cmd2)
+
+/**
+ * wait_event_cmd - sleep until a condition gets true
+ * @wq: the waitqueue to wait on
+ * @condition: a C expression for the event to wait for
+ * cmd1: the command will be executed before sleep
+ * cmd2: the command will be executed after sleep
+ *
+ * The process is put to sleep (TASK_UNINTERRUPTIBLE) until the
+ * @condition evaluates to true. The @condition is checked each time
+ * the waitqueue @wq is woken up.
+ *
+ * wake_up() has to be called after changing any variable that could
+ * change the result of the wait condition.
+ */
+#define wait_event_cmd(wq, condition, cmd1, cmd2)			\
+do {									\
+	if (condition)							\
+		break;							\
+	__wait_event_cmd(wq, condition, cmd1, cmd2);			\
+} while (0)
+
 #define __wait_event_interruptible(wq, condition)			\
 	___wait_event(wq, condition, TASK_INTERRUPTIBLE, 0, 0,		\
 		      schedule())

include/uapi/linux/raid/md_p.h

@@ -16,6 +16,7 @@
 #define _MD_P_H
 
 #include <linux/types.h>
+#include <asm/byteorder.h>
 
 /*
  * RAID superblock.