From: Samuel Just <sjust@redhat.com>
Date: Sat, 29 Oct 2016 22:00:57 +0000 (-0700)
Subject: ReplicatedPG: we might actually recover an object past crt on repair
X-Git-Tag: v11.1.0~245^2~2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=d8e0965cc657586650da2e338044ac041715a839;p=ceph.git

ReplicatedPG: we might actually recover an object past crt on repair

Signed-off-by: Samuel Just <sjust@redhat.com>
---

diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index 773dd8e7deb2..418a15875d9c 100644
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -362,6 +362,15 @@ void ReplicatedPG::on_local_recover(
   // keep track of active pushes for scrub
   ++active_pushes;
 
+  if (recovery_info.version > pg_log.get_can_rollback_to()) {
+    /* This can only happen during a repair, and even then, it would
+     * be one heck of a race.  If we are repairing the object, the
+     * write in question must be fully committed, so it's not valid
+     * to roll it back anyway (and we'll be rolled forward shortly
+     * anyway) */
+    PGLogEntryHandler h{this, t};
+    pg_log.roll_forward_to(recovery_info.version, &h);
+  }
   recover_got(recovery_info.soid, recovery_info.version);
 
   if (is_primary()) {