cluster: RHEL6 - gfs_controld: fix first recovery case

[David Teigland]

Gitweb:       
http://git.fedorahosted.org/git/?p=cluster.git;a=commitdiff;h=72619738e20...
Commit:        72619738e20e2627a7e5fc3268b003d33ce699b2
Parent:        e3f8a987f0108b0f5c1c76e8750c35f23fca2191
Author:        David Teigland <teigland(a)redhat.com&gt;
AuthorDate:    Tue Jul 9 11:35:25 2013 -0500
Committer:     David Teigland <teigland(a)redhat.com&gt;
CommitterDate: Fri Jan 10 14:28:38 2014 -0600

gfs_controld: fix first recovery case

- node A is doing first recovery
- node B joins the mount group and is waiting for A to finish
- node A sets some journals X and Y as needing recovery based
  on start message from A (it's not clear how/why A has journals
  X,Y marked as needing recovery if it's doing first recovery.)
- node A fails
- node B marks A's journal as needing recovery
- node B takes over doing first recovery
- node B successfully finishes first recovery
- node B still has X,Y,A journals as needing recovery, which
  prevents the mount group recovery from completing

First mount recovery allows the first mounter to recover all
journals without any other nodes present.  This is meant to
guarantee that all journals are clean when first mount recovery
is done.  So, after B completes first mount recovery it should
assume all journals are clean and it should clear any needs
recovery indication on journals.

bz 982305

Signed-off-by: David Teigland <teigland(a)redhat.com&gt;
---
 group/gfs_controld/cpg-new.c |   12 ++++++++++--
 1 files changed, 10 insertions(+), 2 deletions(-)

diff --git a/group/gfs_controld/cpg-new.c b/group/gfs_controld/cpg-new.c
index 537624d..8943f62 100644
--- a/group/gfs_controld/cpg-new.c
+++ b/group/gfs_controld/cpg-new.c
@@ -2304,8 +2304,8 @@ void process_recovery_uevent(char *name, int jid, int
recover_status,
 		   to check below that we've seen uevents for all jids
 		   during first recovery before sending first_recovery_done. */
 
-		log_group(mg, "recovery_uevent jid %d first recovery done %d",
-			  jid, mg->first_done_uevent);
+		log_group(mg, "recovery_uevent jid %d status %d first recovery done %d",
+			  jid, recover_status, mg->first_done_uevent);
 
 		/* ignore extraneous uevent from others_may_mount */
 		if (mg->first_done_uevent)
@@ -2323,6 +2323,14 @@ void process_recovery_uevent(char *name, int jid, int
recover_status,
 		if (first_done) {
 			log_group(mg, "recovery_uevent first_done");
 			mg->first_done_uevent = 1;
+
+			list_for_each_entry(j, &mg->journals, list) {
+				if (!j->needs_recovery)
+					continue;
+				j->needs_recovery = 0;
+				log_debug("recovery_uevent first_done clear %d needs_recovery",
j->jid);
+			}
+
 			send_first_recovery_done(mg);
 		}
 	}