]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
md-cluster: update comments for MD_CLUSTER_SEND_LOCKED_ALREADY
authorGuoqing Jiang <gqjiang@suse.com>
Sun, 20 Dec 2015 23:51:01 +0000 (10:51 +1100)
committerNeilBrown <neilb@suse.com>
Wed, 6 Jan 2016 00:39:21 +0000 (11:39 +1100)
1. fix unbalanced parentheses.
2. add more description about that MD_CLUSTER_SEND_LOCKED_ALREADY
   will be cleared after set it in add_new_disk.

Signed-off-by: Guoqing Jiang <gqjiang@suse.com>
Signed-off-by: NeilBrown <neilb@suse.com>
drivers/md/md-cluster.c

index ad3ec7df15471322172183664e80cceea12449c8..0ded8e97751d270dbfdae0aac73a792043f10621 100644 (file)
@@ -53,11 +53,12 @@ struct resync_info {
  * accomodate lock and hold. See next comment.
  */
 #define                MD_CLUSTER_SEND_LOCK                    4
-/* If cluster operations must lock the communication channel,
- * so as to perform extra operations (and no other operation
- * is allowed on the MD, such as adding a disk. Token needs
- * to be locked and held until the operation completes with
- * a md_update_sb(), which would eventually release the lock.
+/* If cluster operations (such as adding a disk) must lock the
+ * communication channel, so as to perform extra operations
+ * (update metadata) and no other operation is allowed on the
+ * MD. Token needs to be locked and held until the operation
+ * completes witha md_update_sb(), which would eventually release
+ * the lock.
  */
 #define                MD_CLUSTER_SEND_LOCKED_ALREADY          5
 
@@ -1021,6 +1022,18 @@ static int add_new_disk(struct mddev *mddev, struct md_rdev *rdev)
                unlock_comm(cinfo);
        else {
                dlm_lock_sync(cinfo->no_new_dev_lockres, DLM_LOCK_CR);
+               /* Since MD_CHANGE_DEVS will be set in add_bound_rdev which
+                * will run soon after add_new_disk, the below path will be
+                * invoked:
+                *   md_wakeup_thread(mddev->thread)
+                *      -> conf->thread (raid1d)
+                *      -> md_check_recovery -> md_update_sb
+                *      -> metadata_update_start/finish
+                * MD_CLUSTER_SEND_LOCKED_ALREADY will be cleared eventually.
+                *
+                * For other failure cases, metadata_update_cancel and
+                * add_new_disk_cancel also clear below bit as well.
+                * */
                set_bit(MD_CLUSTER_SEND_LOCKED_ALREADY, &cinfo->state);
                wake_up(&cinfo->wait);
        }