md-cluster/raid10: resize all the bitmaps before start reshape

To support add disk under grow mode, we need to resize
all the bitmaps of each node before reshape, so that we
can ensure all nodes have the same view of the bitmap of
the clustered raid.

So after the master node resized the bitmap, it broadcast
a message to other slave nodes, and it checks the size of
each bitmap are same or not by compare pages. We can only
continue the reshaping after all nodes update the bitmap
to the same size (by checking the pages), otherwise revert
bitmap size to previous value.

The resize_bitmaps interface and BITMAP_RESIZE message are
introduced in md-cluster.c for the purpose.

Reviewed-by: NeilBrown <neilb@suse.com>
Signed-off-by: Guoqing Jiang <gqjiang@suse.com>
Signed-off-by: Shaohua Li <shli@fb.com>

3 files changed, 120 insertions, 3 deletions

diff --git a/drivers/md/md-cluster.c b/drivers/md/md-cluster.c
index 0b2af6e..e223fae 100644
--- a/drivers/md/md-cluster.c
+++ b/drivers/md/md-cluster.c
@@ -105,6 +105,7 @@ enum msg_type {
 	RE_ADD,
 	BITMAP_NEEDS_SYNC,
 	CHANGE_CAPACITY,
+	BITMAP_RESIZE,
 };
 
 struct cluster_msg {
@@ -612,6 +613,11 @@ static int process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg)
 	case BITMAP_NEEDS_SYNC:
 		__recover_slot(mddev, le32_to_cpu(msg->slot));
 		break;
+	case BITMAP_RESIZE:
+		if (le64_to_cpu(msg->high) != mddev->pers->size(mddev, 0, 0))
+			ret = md_bitmap_resize(mddev->bitmap,
+					    le64_to_cpu(msg->high), 0, 0);
+		break;
 	default:
 		ret = -1;
 		pr_warn("%s:%d Received unknown message from %d\n",
@@ -1102,6 +1108,80 @@ static void metadata_update_cancel(struct mddev *mddev)
 	unlock_comm(cinfo);
 }
 
+static int update_bitmap_size(struct mddev *mddev, sector_t size)
+{
+	struct md_cluster_info *cinfo = mddev->cluster_info;
+	struct cluster_msg cmsg = {0};
+	int ret;
+
+	cmsg.type = cpu_to_le32(BITMAP_RESIZE);
+	cmsg.high = cpu_to_le64(size);
+	ret = sendmsg(cinfo, &cmsg, 0);
+	if (ret)
+		pr_err("%s:%d: failed to send BITMAP_RESIZE message (%d)\n",
+			__func__, __LINE__, ret);
+	return ret;
+}
+
+static int resize_bitmaps(struct mddev *mddev, sector_t newsize, sector_t oldsize)
+{
+	struct bitmap_counts *counts;
+	char str[64];
+	struct dlm_lock_resource *bm_lockres;
+	struct bitmap *bitmap = mddev->bitmap;
+	unsigned long my_pages = bitmap->counts.pages;
+	int i, rv;
+
+	/*
+	 * We need to ensure all the nodes can grow to a larger
+	 * bitmap size before make the reshaping.
+	 */
+	rv = update_bitmap_size(mddev, newsize);
+	if (rv)
+		return rv;
+
+	for (i = 0; i < mddev->bitmap_info.nodes; i++) {
+		if (i == md_cluster_ops->slot_number(mddev))
+			continue;
+
+		bitmap = get_bitmap_from_slot(mddev, i);
+		if (IS_ERR(bitmap)) {
+			pr_err("can't get bitmap from slot %d\n", i);
+			goto out;
+		}
+		counts = &bitmap->counts;
+
+		/*
+		 * If we can hold the bitmap lock of one node then
+		 * the slot is not occupied, update the pages.
+		 */
+		snprintf(str, 64, "bitmap%04d", i);
+		bm_lockres = lockres_init(mddev, str, NULL, 1);
+		if (!bm_lockres) {
+			pr_err("Cannot initialize %s lock\n", str);
+			goto out;
+		}
+		bm_lockres->flags |= DLM_LKF_NOQUEUE;
+		rv = dlm_lock_sync(bm_lockres, DLM_LOCK_PW);
+		if (!rv)
+			counts->pages = my_pages;
+		lockres_free(bm_lockres);
+
+		if (my_pages != counts->pages)
+			/*
+			 * Let's revert the bitmap size if one node
+			 * can't resize bitmap
+			 */
+			goto out;
+	}
+
+	return 0;
+out:
+	md_bitmap_free(bitmap);
+	update_bitmap_size(mddev, oldsize);
+	return -1;
+}
+
 /*
  * return 0 if all the bitmaps have the same sync_size
  */
@@ -1492,6 +1572,7 @@ static struct md_cluster_operations cluster_ops = {
 	.remove_disk = remove_disk,
 	.load_bitmaps = load_bitmaps,
 	.gather_bitmaps = gather_bitmaps,
+	.resize_bitmaps = resize_bitmaps,
 	.lock_all_bitmaps = lock_all_bitmaps,
 	.unlock_all_bitmaps = unlock_all_bitmaps,
 	.update_size = update_size,
diff --git a/drivers/md/md-cluster.h b/drivers/md/md-cluster.h
index c024070..9bd753a 100644
--- a/drivers/md/md-cluster.h
+++ b/drivers/md/md-cluster.h
@@ -26,6 +26,7 @@ struct md_cluster_operations {
 	int (*remove_disk)(struct mddev *mddev, struct md_rdev *rdev);
 	void (*load_bitmaps)(struct mddev *mddev, int total_slots);
 	int (*gather_bitmaps)(struct md_rdev *rdev);
+	int (*resize_bitmaps)(struct mddev *mddev, sector_t newsize, sector_t oldsize);
 	int (*lock_all_bitmaps)(struct mddev *mddev);
 	void (*unlock_all_bitmaps)(struct mddev *mddev);
 	void (*update_size)(struct mddev *mddev, sector_t old_dev_sectors);
diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index a52f43b..72e5292 100644
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c
@@ -25,6 +25,7 @@
 #include <linux/seq_file.h>
 #include <linux/ratelimit.h>
 #include <linux/kthread.h>
+#include <linux/raid/md_p.h>
 #include <trace/events/block.h>
 #include "md.h"
 #include "raid10.h"
@@ -4288,12 +4289,46 @@ static int raid10_start_reshape(struct mddev *mddev)
 	spin_unlock_irq(&conf->device_lock);
 
 	if (mddev->delta_disks && mddev->bitmap) {
-		ret = md_bitmap_resize(mddev->bitmap,
-				       raid10_size(mddev, 0, conf->geo.raid_disks),
-				       0, 0);
+		struct mdp_superblock_1 *sb = NULL;
+		sector_t oldsize, newsize;
+
+		oldsize = raid10_size(mddev, 0, 0);
+		newsize = raid10_size(mddev, 0, conf->geo.raid_disks);
+
+		if (!mddev_is_clustered(mddev)) {
+			ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0);
+			if (ret)
+				goto abort;
+			else
+				goto out;
+		}
+
+		rdev_for_each(rdev, mddev) {
+			if (rdev->raid_disk > -1 &&
+			    !test_bit(Faulty, &rdev->flags))
+				sb = page_address(rdev->sb_page);
+		}
+
+		/*
+		 * some node is already performing reshape, and no need to
+		 * call md_bitmap_resize again since it should be called when
+		 * receiving BITMAP_RESIZE msg
+		 */
+		if ((sb && (le32_to_cpu(sb->feature_map) &
+			    MD_FEATURE_RESHAPE_ACTIVE)) || (oldsize == newsize))
+			goto out;
+
+		ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0);
 		if (ret)
 			goto abort;
+
+		ret = md_cluster_ops->resize_bitmaps(mddev, newsize, oldsize);
+		if (ret) {
+			md_bitmap_resize(mddev->bitmap, oldsize, 0, 0);
+			goto abort;
+		}
 	}
+out:
 	if (mddev->delta_disks > 0) {
 		rdev_for_each(rdev, mddev)
 			if (rdev->raid_disk < 0 &&