From: John Spray <john.spray@redhat.com>
Date: Thu, 30 Apr 2015 16:44:23 +0000 (+0100)
Subject: mds: handle missing/corrupt data during boot
X-Git-Tag: v9.0.2~217^2~1
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=6cb1f6cca7d9014d974e27cfd32876d490ded3bb;p=ceph.git

mds: handle missing/corrupt data during boot

These errors trickle up to boot_start from e.g.
CInode::_fetched when something goes wrong.

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

diff --git a/src/mds/MDS.cc b/src/mds/MDS.cc
index 5aea27d9c31..be3d39bc19f 100644
--- a/src/mds/MDS.cc
+++ b/src/mds/MDS.cc
@@ -1939,7 +1939,14 @@ void MDS::boot_start(BootStep step, int r)
       dout(0) << "boot_start encountered an error EAGAIN"
               << ", respawning since we fell behind journal" << dendl;
       respawn();
+    } else if (r == -EINVAL || r == -ENOENT) {
+      // Invalid or absent data, indicates damaged on-disk structures
+      clog->error() << "Error loading MDS rank " << whoami << ": "
+        << cpp_strerror(r);
+      damaged();
+      assert(r == 0);  // Unreachable, damaged() calls respawn()
     } else {
+      // Completely unexpected error, give up and die
       dout(0) << "boot_start encountered an error, failing" << dendl;
       suicide();
       return;