client: more precise CEPH_CLIENT_CAPS_PENDING_CAPSNAP

Client uses this flag to tell mds if there is more cap snap need to
flush. It's mainly for the case that client needs to re-send cap/snap
flushes after mds failover, but CEPH_CAP_ANY_FILE_WR on corresponding
inodes are all released before mds failover

Fixes: https://tracker.ceph.com/issues/40474
Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit 8628340e28208ad63019a6da61642f48fbde6313)

diff --git a/src/client/Client.cc b/src/client/Client.cc
index 946006730c58e720f781c15d0e1bc7b46386206a..89a00bc9c62420978421b5b347741161618185d4 100644 (file)
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -3309,7 +3309,7 @@ void Client::cap_delay_requeue(Inode *in)
 }
 
 void Client::send_cap(Inode *in, MetaSession *session, Cap *cap,
-                     bool sync, int used, int want, int retain,
+                     int flags, int used, int want, int retain,
                      int flush, ceph_tid_t flush_tid)
 {
   int held = cap->issued | cap->implemented;
@@ -3320,7 +3320,6 @@ void Client::send_cap(Inode *in, MetaSession *session, Cap *cap,
 
   ldout(cct, 10) << __func__ << " " << *in
           << " mds." << session->mds_num << " seq " << cap->seq
-          << (sync ? " sync " : " async ")
           << " used " << ccap_string(used)
           << " want " << ccap_string(want)
           << " flush " << ccap_string(flush)
@@ -3395,11 +3394,13 @@ void Client::send_cap(Inode *in, MetaSession *session, Cap *cap,
   m->btime = in->btime;
   m->time_warp_seq = in->time_warp_seq;
   m->change_attr = in->change_attr;
-  if (sync)
-    m->flags |= MClientCaps::FLAG_SYNC;
-  if (!in->cap_snaps.empty())
-    m->flags |= MClientCaps::FLAG_PENDING_CAPSNAP;
-    
+
+  if (!(flags & MClientCaps::FLAG_PENDING_CAPSNAP) &&
+      !in->cap_snaps.empty() &&
+      in->cap_snaps.rbegin()->second.flush_tid == 0)
+    flags |= MClientCaps::FLAG_PENDING_CAPSNAP;
+  m->flags = flags;
+
   if (flush & CEPH_CAP_FILE_WR) {
     m->inline_version = in->inline_version;
     m->inline_data = in->inline_data;
@@ -3603,8 +3604,9 @@ void Client::check_caps(Inode *in, unsigned flags)
       flush_tid = 0;
     }
 
-    send_cap(in, session, &cap, flags & CHECK_CAPS_SYNCHRONOUS, cap_used, wanted,
-            retain, flushing, flush_tid);
+    int msg_flags = (flags & CHECK_CAPS_SYNCHRONOUS) ? MClientCaps::FLAG_SYNC : 0;
+    send_cap(in, session, &cap, msg_flags, cap_used, wanted, retain,
+            flushing, flush_tid);
   }
 }
 
@@ -4427,13 +4429,23 @@ void Client::kick_flushing_caps(Inode *in, MetaSession *session)
   Cap *cap = in->auth_cap;
   ceph_assert(cap->session == session);
 
+  ceph_tid_t last_snap_flush = 0;
+  for (auto p = in->flushing_cap_tids.rbegin();
+       p != in->flushing_cap_tids.rend();
+       ++p) {
+    if (!p->second) {
+      last_snap_flush = p->first;
+      break;
+    }
+  }
+
   int wanted = in->caps_wanted();
   int used = get_caps_used(in) | in->caps_dirty();
   auto it = in->cap_snaps.begin();
   for (auto& p : in->flushing_cap_tids) {
     if (p.second) {
-      send_cap(in, session, cap, false,
-              used, wanted, (cap->issued | cap->implemented),
+      int msg_flags = p.first < last_snap_flush ? MClientCaps::FLAG_PENDING_CAPSNAP : 0;
+      send_cap(in, session, cap, msg_flags, used, wanted, (cap->issued | cap->implemented),
               p.second, p.first);
     } else {
       ceph_assert(it != in->cap_snaps.end());

diff --git a/src/client/Client.h b/src/client/Client.h
index 608288abcd6d36d11525deb7814c363f6425c47e..742cb6812fc7e0d0a5fbc09ecf0d0b7f87d026d0 100644 (file)
--- a/src/client/Client.h
+++ b/src/client/Client.h
@@ -671,7 +671,7 @@ public:
   void handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m);
   void cap_delay_requeue(Inode *in);
 
-  void send_cap(Inode *in, MetaSession *session, Cap *cap, bool sync,
+  void send_cap(Inode *in, MetaSession *session, Cap *cap, int flags,
                int used, int want, int retain, int flush,
                ceph_tid_t flush_tid);