From: John Spray <john.spray@redhat.com>
Date: Tue, 21 Feb 2017 15:51:47 +0000 (+0000)
Subject: mon: fix stale iterator during fs rm
X-Git-Tag: v12.0.1~238^2~1
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=5fd66839bb8be57541bff2c7fd58dcd31a2bdabb;p=ceph.git

mon: fix stale iterator during fs rm

We were failing out any standbys in the map for
the removed filesystem, while looping over the
mds_info -- this could have crashed.

Signed-off-by: John Spray <john.spray@redhat.com>
---

diff --git a/src/mon/FSCommands.cc b/src/mon/FSCommands.cc
index 693bf0d5e36e1..b48ee11a25efa 100644
--- a/src/mon/FSCommands.cc
+++ b/src/mon/FSCommands.cc
@@ -551,13 +551,17 @@ class RemoveFilesystemHandler : public FileSystemCommandHandler
       fsmap.set_legacy_client_fscid(FS_CLUSTER_ID_NONE);
     }
 
+    std::vector<mds_gid_t> to_fail;
     // There may be standby_replay daemons left here
     for (const auto &i : fs->mds_map.get_mds_info()) {
       assert(i.second.state == MDSMap::STATE_STANDBY_REPLAY);
+      to_fail.push_back(i.first);
+    }
 
+    for (const auto &gid : to_fail) {
       // Standby replays don't write, so it isn't important to
       // wait for an osdmap propose here: ignore return value.
-      mon->mdsmon()->fail_mds_gid(i.first);
+      mon->mdsmon()->fail_mds_gid(gid);
     }
 
     fsmap.erase_filesystem(fs->fscid);