diff --git a/drivers/md/md-cluster.c b/drivers/md/md-cluster.c index f2bd37d72a46..6ce75c82969c 100644 --- a/drivers/md/md-cluster.c +++ b/drivers/md/md-cluster.c @@ -57,6 +57,7 @@ struct resync_info { #define MD_CLUSTER_ALREADY_IN_CLUSTER 6 #define MD_CLUSTER_PENDING_RECV_EVENT 7 #define MD_CLUSTER_HOLDING_MUTEX_FOR_RECVD 8 +#define MD_CLUSTER_WAITING_FOR_SYNC 9 struct md_cluster_info { struct mddev *mddev; /* the md device which md_cluster_info belongs to */ @@ -92,6 +93,7 @@ struct md_cluster_info { sector_t sync_hi; }; +/* For compatibility, add the new msg_type at the end. */ enum msg_type { METADATA_UPDATED = 0, RESYNCING, @@ -101,6 +103,7 @@ enum msg_type { BITMAP_NEEDS_SYNC, CHANGE_CAPACITY, BITMAP_RESIZE, + RESYNCING_START, }; struct cluster_msg { @@ -461,6 +464,7 @@ static void process_suspend_info(struct mddev *mddev, clear_bit(MD_RESYNCING_REMOTE, &mddev->recovery); remove_suspend_info(mddev, slot); set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + clear_bit(MD_CLUSTER_WAITING_FOR_SYNC, &cinfo->state); md_wakeup_thread(mddev->thread); return; } @@ -531,6 +535,7 @@ static int process_add_new_disk(struct mddev *mddev, struct cluster_msg *cmsg) res = -1; } clear_bit(MD_CLUSTER_WAITING_FOR_NEWDISK, &cinfo->state); + set_bit(MD_CLUSTER_WAITING_FOR_SYNC, &cinfo->state); return res; } @@ -599,6 +604,9 @@ static int process_recvd_msg(struct mddev *mddev, struct cluster_msg *msg) case CHANGE_CAPACITY: set_capacity_and_notify(mddev->gendisk, mddev->array_sectors); break; + case RESYNCING_START: + clear_bit(MD_CLUSTER_WAITING_FOR_SYNC, &mddev->cluster_info->state); + break; case RESYNCING: set_bit(MD_RESYNCING_REMOTE, &mddev->recovery); process_suspend_info(mddev, le32_to_cpu(msg->slot), @@ -1345,6 +1353,23 @@ static void resync_info_get(struct mddev *mddev, sector_t *lo, sector_t *hi) spin_unlock_irq(&cinfo->suspend_lock); } +static int resync_status_get(struct mddev *mddev) +{ + struct md_cluster_info *cinfo = mddev->cluster_info; + + return test_bit(MD_CLUSTER_WAITING_FOR_SYNC, &cinfo->state); +} + +static int resync_start_notify(struct mddev *mddev) +{ + struct md_cluster_info *cinfo = mddev->cluster_info; + struct cluster_msg cmsg = {0}; + + cmsg.type = cpu_to_le32(RESYNCING_START); + + return sendmsg(cinfo, &cmsg, 0); +} + static int resync_info_update(struct mddev *mddev, sector_t lo, sector_t hi) { struct md_cluster_info *cinfo = mddev->cluster_info; @@ -1579,6 +1604,8 @@ static const struct md_cluster_operations cluster_ops = { .resync_start = resync_start, .resync_finish = resync_finish, .resync_info_update = resync_info_update, + .resync_start_notify = resync_start_notify, + .resync_status_get = resync_status_get, .resync_info_get = resync_info_get, .metadata_update_start = metadata_update_start, .metadata_update_finish = metadata_update_finish, diff --git a/drivers/md/md-cluster.h b/drivers/md/md-cluster.h index a78e3021775d..470bf18ffde5 100644 --- a/drivers/md/md-cluster.h +++ b/drivers/md/md-cluster.h @@ -14,6 +14,8 @@ struct md_cluster_operations { int (*leave)(struct mddev *mddev); int (*slot_number)(struct mddev *mddev); int (*resync_info_update)(struct mddev *mddev, sector_t lo, sector_t hi); + int (*resync_start_notify)(struct mddev *mddev); + int (*resync_status_get)(struct mddev *mddev); void (*resync_info_get)(struct mddev *mddev, sector_t *lo, sector_t *hi); int (*metadata_update_start)(struct mddev *mddev); int (*metadata_update_finish)(struct mddev *mddev); diff --git a/drivers/md/md.c b/drivers/md/md.c index 64693913ed18..d3a837506a36 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -8978,7 +8978,8 @@ void md_do_sync(struct md_thread *thread) * This will mean we have to start checking from the beginning again. * */ - + if (mddev_is_clustered(mddev)) + md_cluster_ops->resync_start_notify(mddev); do { int mddev2_minor = -1; mddev->curr_resync = MD_RESYNC_DELAYED; @@ -9992,8 +9993,18 @@ static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev) */ if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE && !(le32_to_cpu(sb->feature_map) & - MD_FEATURE_RESHAPE_ACTIVE)) { - rdev2->saved_raid_disk = role; + MD_FEATURE_RESHAPE_ACTIVE) && + !md_cluster_ops->resync_status_get(mddev)) { + /* + * -1 to make raid1_add_disk() set conf->fullsync + * to 1. This could avoid skipping sync when the + * remote node is down during resyncing. + */ + if ((le32_to_cpu(sb->feature_map) + & MD_FEATURE_RECOVERY_OFFSET)) + rdev2->saved_raid_disk = -1; + else + rdev2->saved_raid_disk = role; ret = remove_and_add_spares(mddev, rdev2); pr_info("Activated spare: %pg\n", rdev2->bdev);