From: Sage Weil <sage@newdream.net>
Date: Wed, 18 Aug 2010 20:49:11 +0000 (-0700)
Subject: mds: fix null snapflush logic
X-Git-Tag: v0.22~255
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=c8701f29f0a6f3777c41f8952c054ba4dd41b9d4;p=ceph.git

mds: fix null snapflush logic

We only want to do a null snapflush if we _know_ there isn't another one
coming: that is, there aren't any outstanding issued excl/wr cap bits at
the client.  The old test has the bitwise NOT backwards.  We can also
limit the test to the bits we care about.
---

diff --git a/src/mds/Locker.cc b/src/mds/Locker.cc
index 801f197a636..99bcf80105a 100644
--- a/src/mds/Locker.cc
+++ b/src/mds/Locker.cc
@@ -1862,11 +1862,11 @@ void Locker::handle_client_caps(MClientCaps *m)
 
 
     // missing/skipped snapflush?
-    //  ** only if the client has completed all revocations.  if we
+    //  ** only if the client has completed all WR/EXCL revocations.  if we
     //     are still revoking, it's possible the client is waiting for
     //     data writeback and hasn't sent the flushsnap yet.  **
     if (head_in->client_need_snapflush.size()) {
-      if ((cap->pending() & ~cap->issued()) == 0) {
+      if ((cap->issued() & CEPH_CAP_ANY_WR & ~cap->pending()) == 0) {
 	map<snapid_t, set<client_t> >::iterator p = head_in->client_need_snapflush.begin();
 	while (p != head_in->client_need_snapflush.end()) {
 	  // p->first is the snap inode's ->last