mds: ensure next replay is queued on req drop

Not all client replay requests are queued at once since [1]. We require
the next request by queued when completed (unsafely) or during cleanup.
Not all code paths seem to handle this [2] so move it to a generic
location, MDCache::request_cleanup. Even so, this doesn't handle all
errors (so we must still be careful) as sometimes we must queue the next
replay request before an MDRequest is constructed [3] during some error
conditions.

Additionally, preserve the behavior of Server::journal_and_reply
queueing the next replay op. Otherwise, must wait for the request to be
durable before moving onto the next one, unnecessarily.

For reproducing, two specific cases are highlighted (thanks to @Mer1997 on
Github for locating these):

- The request is killed by a session close / eviction while a replayed request
  is queued and waiting for a journal flush (e.g. dirty inest locks).

- The request construction fails because the request is already in the
  active_requests. This could happen theoretically if a client resends the same
  request (same reqid) twice.

The first case is most probable but very difficult to reproduce for testing
purposes. The replayed op would need to wait on a journal flush (to be
restarted by C_MDS_RetryRequest).  Then, the request would need killed by a
session close.

[1] ed6a18d
[2] https://github.com/ceph/ceph/blob/a6f1a1c6c09d74f5918c715b05789f34f2ea0e90/src/mds/Server.cc#L2253-L2262
[3] https://github.com/ceph/ceph/blob/a6f1a1c6c09d74f5918c715b05789f34f2ea0e90/src/mds/Server.cc#L2380

Fixes: https://tracker.ceph.com/issues/56577
Signed-off-by: Patrick Donnelly <pdonnell@redhat.com>
(cherry picked from commit 078ecaa)

Conflicts:
	src/mds/Mutation.h: lock dump changes not backported
	src/mds/Server.cc: minor code change

src/mds/MDCache.cc

@@ -9724,6 +9724,12 @@ void MDCache::request_cleanup(MDRequestRef& mdr)
   // remove from map
   active_requests.erase(mdr->reqid);
 
+  // queue next replay op?
+  if (mdr->is_queued_for_replay() && !mdr->get_queued_next_replay_op()) {
+    mdr->set_queued_next_replay_op();
+    mds->queue_one_replay();
+  }
+
   if (mds->logger)
     log_stat();
 

src/mds/MDSRank.cc

@@ -2055,13 +2055,15 @@ bool MDSRank::queue_one_replay()
   if (!replay_queue.empty()) {
     queue_waiter(replay_queue.front());
     replay_queue.pop_front();
+    dout(10) << " queued next replay op" << dendl;
     return true;
   }
   if (!replaying_requests_done) {
     replaying_requests_done = true;
     mdlog->flush();
   }
   maybe_clientreplay_done();
+  dout(10) << " journaled last replay op" << dendl;
   return false;
 }
 

src/mds/Mutation.h

@@ -387,6 +387,12 @@ struct MDRequestImpl : public MutationImpl {
   void set_filepath(const filepath& fp);
   void set_filepath2(const filepath& fp);
   bool is_queued_for_replay() const;
+  bool get_queued_next_replay_op() const {
+    return queued_next_replay_op;
+  }
+  void set_queued_next_replay_op() {
+    queued_next_replay_op = true;
+  }
   int compare_paths();
 
   bool can_batch();
@@ -456,6 +462,7 @@ struct MDRequestImpl : public MutationImpl {
   void _dump_op_descriptor_unlocked(std::ostream& stream) const override;
 private:
   mutable ceph::spinlock msg_lock;
+  bool queued_next_replay_op = false;
 };
 
 struct MDPeerUpdate {

src/mds/Server.cc

@@ -299,6 +299,7 @@ void Server::dispatch(const cref_t<Message> &m)
 	return;
       }
       bool queue_replay = false;
+      dout(5) << "dispatch request in up:reconnect: " << *req << dendl;
       if (req->is_replay() || req->is_async()) {
 	dout(3) << "queuing replayed op" << dendl;
 	queue_replay = true;
@@ -317,10 +318,13 @@ void Server::dispatch(const cref_t<Message> &m)
 	// process completed request in clientreplay stage. The completed request
 	// might have created new file/directorie. This guarantees MDS sends a reply
 	// to client before other request modifies the new file/directorie.
-	if (session->have_completed_request(req->get_reqid().tid, NULL)) {
-	  dout(3) << "queuing completed op" << dendl;
+        bool r = session->have_completed_request(req->get_reqid().tid, NULL);
+	if (r) {
+	  dout(3) << __func__ << ": queuing completed op" << dendl;
 	  queue_replay = true;
-	}
+	} else {
+          dout(20) << __func__  << ": request not complete" << dendl;
+        }
 	// this request was created before the cap reconnect message, drop any embedded
 	// cap releases.
 	req->releases.clear();
@@ -1975,13 +1979,16 @@ void Server::journal_and_reply(MDRequestRef& mdr, CInode *in, CDentry *dn, LogEv
 
   mdr->committing = true;
   submit_mdlog_entry(le, fin, mdr, __func__);
-
-  if (mdr->client_request && mdr->client_request->is_queued_for_replay()) {
-    if (mds->queue_one_replay()) {
-      dout(10) << " queued next replay op" << dendl;
-    } else {
-      dout(10) << " journaled last replay op" << dendl;
-    }
+
+  if (mdr->is_queued_for_replay()) {
+
+    /* We want to queue the next replay op while waiting for the journaling, so
+     * do it now when the early (unsafe) replay is dispatched. Don't wait until
+     * this request is cleaned up in MDCache.cc.
+     */
+
+    mdr->set_queued_next_replay_op();
+    mds->queue_one_replay();
   } else if (mdr->did_early_reply)
     mds->locker->drop_rdlocks_for_early_reply(mdr.get());
   else
@@ -2282,15 +2289,12 @@ void Server::reply_client_request(MDRequestRef& mdr, const ref_t<MClientReply> &
     mds->send_message_client(reply, session);
   }
 
-  if (req->is_queued_for_replay() &&
-      (mdr->has_completed || reply->get_result() < 0)) {
-    if (reply->get_result() < 0) {
-      int r = reply->get_result();
+  if (req->is_queued_for_replay()) {
+    if (int r = reply->get_result(); r < 0) {
       derr << "reply_client_request: failed to replay " << *req
-	   << " error " << r << " (" << cpp_strerror(r)  << ")" << dendl;
+           << " error " << r << " (" << cpp_strerror(r)  << ")" << dendl;
       mds->clog->warn() << "failed to replay " << req->get_reqid() << " error " << r;
     }
-    mds->queue_one_replay();
   }
 
   // clean up request
@@ -2488,8 +2492,12 @@ void Server::handle_client_request(const cref_t<MClientRequest> &req)
 
   // register + dispatch
   MDRequestRef mdr = mdcache->request_start(req);
-  if (!mdr.get())
+  if (!mdr.get()) {
+    dout(5) << __func__ << ": possibly duplicate op " << *req << dendl;
+    if (req->is_queued_for_replay())
+      mds->queue_one_replay();
     return;
+  }
 
   if (session) {
     mdr->session = session;