From: Frank Zago <fzago@cray.com>
Date: Mon, 11 Aug 2014 23:24:07 +0000 (-0500)
Subject: LU-5195: hsm: delete HSM records not found on disk
X-Git-Tag: 2.6.52~37
X-Git-Url: https://git.whamcloud.com/?a=commitdiff_plain;ds=sidebyside;h=3a83b4b93373db46a4ee60a7388775fa0be2eb9a;p=fs%2Flustre-release.git

LU-5195: hsm: delete HSM records not found on disk

After a MDS crash, it is possible the file containing an hsm record is
not present anymore. When the MDS restarts, we get traces like this:

(mdt_hsm_cdt_actions.c:104:cdt_llog_process()) tas01-MDT0000:
    failed to process HSM_ACTIONS llog (rc=-2)
(mdt_hsm_cdt_actions.c:104:cdt_llog_process())
    Skipped 600 previous similar messages
(llog_cat.c:192:llog_cat_id2handle()) tas01-MDD0000:
    error opening log id 0x1c:1:0: rc = -2
(llog_cat.c:192:llog_cat_id2handle()) Skipped 600 previous similar messages
(llog_cat.c:556:llog_cat_process_cb()) tas01-MDD0000:
    cannot find handle for llog 0x1c:1: -2
(llog_cat.c:556:llog_cat_process_cb()) Skipped 600 previous similar messages
(mdt_hsm_cdt_actions.c:104:cdt_llog_process()) tas01-MDT0000:
    failed to process HSM_ACTIONS llog (rc=-2)
(mdt_hsm_cdt_actions.c:104:cdt_llog_process())
    Skipped 600 previous similar messages
(llog_cat.c:192:llog_cat_id2handle()) tas01-MDD0000:
    error opening log id 0x1c:1:0: rc = -2
(llog_cat.c:192:llog_cat_id2handle()) Skipped 600 previous similar messages

No HSM operation can happen, and the only way to clean is to unmount
the MDT and delete hsm_actions.

If the record can't be found, let the MDS delete it instead.

Change-Id: I61f625d4c18750c8044909ff56d53042cf0b6d86
Signed-off-by: frank zago <fzago@cray.com>
Reviewed-on: http://review.whamcloud.com/11419
Reviewed-by: Niu Yawei <yawei.niu@intel.com>
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Ryan Haasken <haasken@cray.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
---

diff --git a/lustre/obdclass/llog_cat.c b/lustre/obdclass/llog_cat.c
index b34d5d3..05d2f26 100644
--- a/lustre/obdclass/llog_cat.c
+++ b/lustre/obdclass/llog_cat.c
@@ -535,6 +535,17 @@ int llog_cat_process_cb(const struct lu_env *env, struct llog_handle *cat_llh,
 		CERROR("%s: cannot find handle for llog "DOSTID": %d\n",
 		       cat_llh->lgh_ctxt->loc_obd->obd_name,
 		       POSTID(&lir->lid_id.lgl_oi), rc);
+		if (rc == -ENOENT || rc == -ESTALE) {
+			/* After a server crash, a stub of index
+			 * record in catlog could be kept, because
+			 * plain log destroy + catlog index record
+			 * deletion are not atomic. So we end up with
+			 * an index but no actual record. Destroy the
+			 * index and move on. */
+			rc = llog_cat_cleanup(env, cat_llh, NULL,
+					      rec->lrh_index);
+		}
+
 		RETURN(rc);
 	}
 
@@ -650,6 +661,17 @@ static int llog_cat_reverse_process_cb(const struct lu_env *env,
 		CERROR("%s: cannot find handle for llog "DOSTID": %d\n",
 		       cat_llh->lgh_ctxt->loc_obd->obd_name,
 		       POSTID(&lir->lid_id.lgl_oi), rc);
+		if (rc == -ENOENT || rc == -ESTALE) {
+			/* After a server crash, a stub of index
+			 * record in catlog could be kept, because
+			 * plain log destroy + catlog index record
+			 * deletion are not atomic. So we end up with
+			 * an index but no actual record. Destroy the
+			 * index and move on. */
+			rc = llog_cat_cleanup(env, cat_llh, NULL,
+					      rec->lrh_index);
+		}
+
 		RETURN(rc);
 	}