From: Niu Yawei <niu@whamcloud.com>
Date: Sat, 22 Oct 2011 08:16:55 +0000 (+0800)
Subject: LU-753 obdfilter: improper LASSERT in filter_commitrw_write()
X-Git-Tag: 2.1.52~40
X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=commitdiff_plain;h=0935fc8eae15696820f87951d8955a9c8bc66177

LU-753 obdfilter: improper LASSERT in filter_commitrw_write()

In rare cases fsfilt_commit_wait() will wake up and return after the
transaction has finished its work and updated j_commit_sequence but
the commit callbacks have not been run yet. Which will trigger the
LASSERT(oti->oti_transno <= obd->obd_last_committed) improperly.

We should just wait for the commit callback finished instead of put
an improper LASSERT here.

Signed-off-by: Niu Yawei <niu@whamcloud.com>
Change-Id: Ibd5add8d352d2e7598be49b0bf8fa37d40ce6e1f
Reviewed-on: http://review.whamcloud.com/1583
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: Hudson
Reviewed-by: Jinshan Xiong <jay@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
---

diff --git a/lustre/obdfilter/filter_io_26.c b/lustre/obdfilter/filter_io_26.c
index a15884a..cc0d72c 100644
--- a/lustre/obdfilter/filter_io_26.c
+++ b/lustre/obdfilter/filter_io_26.c
@@ -785,10 +785,28 @@ retry:
                         rc = err;
         }
 
-        if (obd->obd_replayable && !rc && wait_handle)
-                LASSERTF(oti->oti_transno <= obd->obd_last_committed,
-                         "oti_transno "LPU64" last_committed "LPU64"\n",
-                         oti->oti_transno, obd->obd_last_committed);
+        /* In rare cases fsfilt_commit_wait() will wake up and return after
+         * the transaction has finished its work and updated j_commit_sequence
+         * but the commit callbacks have not been run yet.  Wait here until
+         * that is finished so that clients requesting sync IO don't see the
+         * reply transno < last_committed.  LU-753 */
+        if (unlikely(obd->obd_replayable && !rc && wait_handle &&
+                     oti->oti_transno > obd->obd_last_committed)) {
+                cfs_waitq_t wq;
+                struct l_wait_info lwi =
+                        LWI_TIMEOUT_INTERVAL(cfs_time_seconds(5),
+                                             (cfs_duration_t)((HZ + 4)/5),
+                                             NULL, NULL);
+                cfs_waitq_init(&wq);
+                l_wait_event(wq,
+                             oti->oti_transno <= obd->obd_last_committed,
+                             &lwi);
+
+                /* commit callback isn't done after waiting for 5 secs ? */
+                if (unlikely(oti->oti_transno > obd->obd_last_committed))
+                        CERROR("transno:"LPU64" > last_committed:"LPU64"\n",
+                               oti->oti_transno, obd->obd_last_committed);
+        }
 
         fsfilt_check_slow(obd, now, "commitrw commit");