1 files changed, 199 insertions, 55 deletions

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 4e7c9f398bc6..4824d50526fa 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -125,74 +125,165 @@ static inline int speed_max(struct mddev *mddev)
 		mddev->sync_speed_max : sysctl_speed_limit_max;
 }
 
-static int rdev_init_wb(struct md_rdev *rdev)
+static void rdev_uninit_serial(struct md_rdev *rdev)
 {
-	if (rdev->bdev->bd_queue->nr_hw_queues == 1)
+	if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
+		return;
+
+	kvfree(rdev->serial);
+	rdev->serial = NULL;
+}
+
+static void rdevs_uninit_serial(struct mddev *mddev)
+{
+	struct md_rdev *rdev;
+
+	rdev_for_each(rdev, mddev)
+		rdev_uninit_serial(rdev);
+}
+
+static int rdev_init_serial(struct md_rdev *rdev)
+{
+	/* serial_nums equals with BARRIER_BUCKETS_NR */
+	int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
+	struct serial_in_rdev *serial = NULL;
+
+	if (test_bit(CollisionCheck, &rdev->flags))
 		return 0;
 
-	spin_lock_init(&rdev->wb_list_lock);
-	INIT_LIST_HEAD(&rdev->wb_list);
-	init_waitqueue_head(&rdev->wb_io_wait);
-	set_bit(WBCollisionCheck, &rdev->flags);
+	serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
+			  GFP_KERNEL);
+	if (!serial)
+		return -ENOMEM;
 
-	return 1;
+	for (i = 0; i < serial_nums; i++) {
+		struct serial_in_rdev *serial_tmp = &serial[i];
+
+		spin_lock_init(&serial_tmp->serial_lock);
+		serial_tmp->serial_rb = RB_ROOT_CACHED;
+		init_waitqueue_head(&serial_tmp->serial_io_wait);
+	}
+
+	rdev->serial = serial;
+	set_bit(CollisionCheck, &rdev->flags);
+
+	return 0;
+}
+
+static int rdevs_init_serial(struct mddev *mddev)
+{
+	struct md_rdev *rdev;
+	int ret = 0;
+
+	rdev_for_each(rdev, mddev) {
+		ret = rdev_init_serial(rdev);
+		if (ret)
+			break;
+	}
+
+	/* Free all resources if pool is not existed */
+	if (ret && !mddev->serial_info_pool)
+		rdevs_uninit_serial(mddev);
+
+	return ret;
 }
 
 /*
- * Create wb_info_pool if rdev is the first multi-queue device flaged
- * with writemostly, also write-behind mode is enabled.
+ * rdev needs to enable serial stuffs if it meets the conditions:
+ * 1. it is multi-queue device flaged with writemostly.
+ * 2. the write-behind mode is enabled.
  */
-void mddev_create_wb_pool(struct mddev *mddev, struct md_rdev *rdev,
-			  bool is_suspend)
+static int rdev_need_serial(struct md_rdev *rdev)
 {
-	if (mddev->bitmap_info.max_write_behind == 0)
-		return;
+	return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
+		rdev->bdev->bd_queue->nr_hw_queues != 1 &&
+		test_bit(WriteMostly, &rdev->flags));
+}
+
+/*
+ * Init resource for rdev(s), then create serial_info_pool if:
+ * 1. rdev is the first device which return true from rdev_enable_serial.
+ * 2. rdev is NULL, means we want to enable serialization for all rdevs.
+ */
+void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
+			      bool is_suspend)
+{
+	int ret = 0;
 
-	if (!test_bit(WriteMostly, &rdev->flags) || !rdev_init_wb(rdev))
+	if (rdev && !rdev_need_serial(rdev) &&
+	    !test_bit(CollisionCheck, &rdev->flags))
 		return;
 
-	if (mddev->wb_info_pool == NULL) {
+	if (!is_suspend)
+		mddev_suspend(mddev);
+
+	if (!rdev)
+		ret = rdevs_init_serial(mddev);
+	else
+		ret = rdev_init_serial(rdev);
+	if (ret)
+		goto abort;
+
+	if (mddev->serial_info_pool == NULL) {
 		unsigned int noio_flag;
 
-		if (!is_suspend)
-			mddev_suspend(mddev);
 		noio_flag = memalloc_noio_save();
-		mddev->wb_info_pool = mempool_create_kmalloc_pool(NR_WB_INFOS,
-							sizeof(struct wb_info));
+		mddev->serial_info_pool =
+			mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
+						sizeof(struct serial_info));
 		memalloc_noio_restore(noio_flag);
-		if (!mddev->wb_info_pool)
-			pr_err("can't alloc memory pool for writemostly\n");
-		if (!is_suspend)
-			mddev_resume(mddev);
+		if (!mddev->serial_info_pool) {
+			rdevs_uninit_serial(mddev);
+			pr_err("can't alloc memory pool for serialization\n");
+		}
 	}
+
+abort:
+	if (!is_suspend)
+		mddev_resume(mddev);
 }
-EXPORT_SYMBOL_GPL(mddev_create_wb_pool);
 
 /*
- * destroy wb_info_pool if rdev is the last device flaged with WBCollisionCheck.
+ * Free resource from rdev(s), and destroy serial_info_pool under conditions:
+ * 1. rdev is the last device flaged with CollisionCheck.
+ * 2. when bitmap is destroyed while policy is not enabled.
+ * 3. for disable policy, the pool is destroyed only when no rdev needs it.
  */
-static void mddev_destroy_wb_pool(struct mddev *mddev, struct md_rdev *rdev)
+void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
+			       bool is_suspend)
 {
-	if (!test_and_clear_bit(WBCollisionCheck, &rdev->flags))
+	if (rdev && !test_bit(CollisionCheck, &rdev->flags))
 		return;
 
-	if (mddev->wb_info_pool) {
+	if (mddev->serial_info_pool) {
 		struct md_rdev *temp;
-		int num = 0;
+		int num = 0; /* used to track if other rdevs need the pool */
 
-		/*
-		 * Check if other rdevs need wb_info_pool.
-		 */
-		rdev_for_each(temp, mddev)
-			if (temp != rdev &&
-			    test_bit(WBCollisionCheck, &temp->flags))
+		if (!is_suspend)
+			mddev_suspend(mddev);
+		rdev_for_each(temp, mddev) {
+			if (!rdev) {
+				if (!mddev->serialize_policy ||
+				    !rdev_need_serial(temp))
+					rdev_uninit_serial(temp);
+				else
+					num++;
+			} else if (temp != rdev &&
+				   test_bit(CollisionCheck, &temp->flags))
 				num++;
-		if (!num) {
-			mddev_suspend(rdev->mddev);
-			mempool_destroy(mddev->wb_info_pool);
-			mddev->wb_info_pool = NULL;
-			mddev_resume(rdev->mddev);
 		}
+
+		if (rdev)
+			rdev_uninit_serial(rdev);
+
+		if (num)
+			pr_info("The mempool could be used by other devices\n");
+		else {
+			mempool_destroy(mddev->serial_info_pool);
+			mddev->serial_info_pool = NULL;
+		}
+		if (!is_suspend)
+			mddev_resume(mddev);
 	}
 }
 
@@ -2337,7 +2428,7 @@ static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
 	pr_debug("md: bind<%s>\n", b);
 
 	if (mddev->raid_disks)
-		mddev_create_wb_pool(mddev, rdev, false);
+		mddev_create_serial_pool(mddev, rdev, false);
 
 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
 		goto fail;
@@ -2375,7 +2466,7 @@ static void unbind_rdev_from_array(struct md_rdev *rdev)
 	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
 	list_del_rcu(&rdev->same_set);
 	pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
-	mddev_destroy_wb_pool(rdev->mddev, rdev);
+	mddev_destroy_serial_pool(rdev->mddev, rdev, false);
 	rdev->mddev = NULL;
 	sysfs_remove_link(&rdev->kobj, "block");
 	sysfs_put(rdev->sysfs_state);
@@ -2888,10 +2979,10 @@ state_store(struct md_rdev *rdev, const char *buf, size_t len)
 		}
 	} else if (cmd_match(buf, "writemostly")) {
 		set_bit(WriteMostly, &rdev->flags);
-		mddev_create_wb_pool(rdev->mddev, rdev, false);
+		mddev_create_serial_pool(rdev->mddev, rdev, false);
 		err = 0;
 	} else if (cmd_match(buf, "-writemostly")) {
-		mddev_destroy_wb_pool(rdev->mddev, rdev);
+		mddev_destroy_serial_pool(rdev->mddev, rdev, false);
 		clear_bit(WriteMostly, &rdev->flags);
 		err = 0;
 	} else if (cmd_match(buf, "blocked")) {
@@ -5277,6 +5368,57 @@ static struct md_sysfs_entry md_fail_last_dev =
 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
        fail_last_dev_store);
 
+static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
+{
+	if (mddev->pers == NULL || (mddev->pers->level != 1))
+		return sprintf(page, "n/a\n");
+	else
+		return sprintf(page, "%d\n", mddev->serialize_policy);
+}
+
+/*
+ * Setting serialize_policy to true to enforce write IO is not reordered
+ * for raid1.
+ */
+static ssize_t
+serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
+{
+	int err;
+	bool value;
+
+	err = kstrtobool(buf, &value);
+	if (err)
+		return err;
+
+	if (value == mddev->serialize_policy)
+		return len;
+
+	err = mddev_lock(mddev);
+	if (err)
+		return err;
+	if (mddev->pers == NULL || (mddev->pers->level != 1)) {
+		pr_err("md: serialize_policy is only effective for raid1\n");
+		err = -EINVAL;
+		goto unlock;
+	}
+
+	mddev_suspend(mddev);
+	if (value)
+		mddev_create_serial_pool(mddev, NULL, true);
+	else
+		mddev_destroy_serial_pool(mddev, NULL, true);
+	mddev->serialize_policy = value;
+	mddev_resume(mddev);
+unlock:
+	mddev_unlock(mddev);
+	return err ?: len;
+}
+
+static struct md_sysfs_entry md_serialize_policy =
+__ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
+       serialize_policy_store);
+
+
 static struct attribute *md_default_attrs[] = {
 	&md_level.attr,
 	&md_layout.attr,
@@ -5294,6 +5436,7 @@ static struct attribute *md_default_attrs[] = {
 	&max_corr_read_errors.attr,
 	&md_consistency_policy.attr,
 	&md_fail_last_dev.attr,
+	&md_serialize_policy.attr,
 	NULL,
 };
 
@@ -5769,18 +5912,18 @@ int md_run(struct mddev *mddev)
 		goto bitmap_abort;
 
 	if (mddev->bitmap_info.max_write_behind > 0) {
-		bool creat_pool = false;
+		bool create_pool = false;
 
 		rdev_for_each(rdev, mddev) {
 			if (test_bit(WriteMostly, &rdev->flags) &&
-			    rdev_init_wb(rdev))
-				creat_pool = true;
-		}
-		if (creat_pool && mddev->wb_info_pool == NULL) {
-			mddev->wb_info_pool =
-				mempool_create_kmalloc_pool(NR_WB_INFOS,
-						    sizeof(struct wb_info));
-			if (!mddev->wb_info_pool) {
+			    rdev_init_serial(rdev))
+				create_pool = true;
+		}
+		if (create_pool && mddev->serial_info_pool == NULL) {
+			mddev->serial_info_pool =
+				mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
+						    sizeof(struct serial_info));
+			if (!mddev->serial_info_pool) {
 				err = -ENOMEM;
 				goto bitmap_abort;
 			}
@@ -6025,8 +6168,9 @@ static void __md_stop_writes(struct mddev *mddev)
 			mddev->in_sync = 1;
 		md_update_sb(mddev, 1);
 	}
-	mempool_destroy(mddev->wb_info_pool);
-	mddev->wb_info_pool = NULL;
+	/* disable policy to guarantee rdevs free resources for serialization */
+	mddev->serialize_policy = 0;
+	mddev_destroy_serial_pool(mddev, NULL, true);
 }
 
 void md_stop_writes(struct mddev *mddev)