From 33e5e6080b0e8e5bac31ddf3167f5db27e97c34f Mon Sep 17 00:00:00 2001
From: Xiubo Li <xiubli@redhat.com>
Date: Fri, 15 Sep 2023 08:41:35 +0800
Subject: [PATCH] mds: record the internal client request and receive client
 reply

This will be used to avoid possible multiple reintegration issue
later.

Fixes: https://tracker.ceph.com/issues/62702
Signed-off-by: Xiubo Li <xiubli@redhat.com>
(cherry picked from commit 723c1b7c9e773a68b6c2586155092ca95ab4f3e7)

Conflicts: with commit 7243b680526 ("mds: ensure next replay is
		queued on req drop")
---
 src/mds/MDSMetaRequest.h | 33 +++++++++++++++++++++++++++++++++
 src/mds/MDSRank.cc       |  2 ++
 src/mds/MDSRank.h        |  3 +++
 src/mds/Server.cc        | 30 ++++++++++++++++++++++++++++++
 src/mds/Server.h         |  1 +
 5 files changed, 69 insertions(+)
 create mode 100644 src/mds/MDSMetaRequest.h

diff --git a/src/mds/MDSMetaRequest.h b/src/mds/MDSMetaRequest.h
new file mode 100644
index 0000000000000..ad47204106869
--- /dev/null
+++ b/src/mds/MDSMetaRequest.h
@@ -0,0 +1,33 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2023 Red Hat, Inc.
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation.  See file COPYING.
+ *
+ */
+
+#ifndef CEPH_MDS_META_REQUEST_H
+#define CEPH_MDS_META_REQUEST_H
+
+#include "include/types.h"
+
+struct MDSMetaRequest {
+private:
+  int op;
+  ceph_tid_t tid;
+public:
+  explicit MDSMetaRequest(int o, ceph_tid_t t) :
+    op(o), tid(t) { }
+  virtual ~MDSMetaRequest() { }
+
+  int get_op() { return op; }
+  ceph_tid_t get_tid() { return tid; }
+};
+
+#endif // !CEPH_MDS_META_REQUEST_H
diff --git a/src/mds/MDSRank.cc b/src/mds/MDSRank.cc
index c5eaf9873ec8d..4f2e13b90be7f 100644
--- a/src/mds/MDSRank.cc
+++ b/src/mds/MDSRank.cc
@@ -1179,6 +1179,7 @@ bool MDSRank::is_valid_message(const cref_t<Message> &m) {
       type == CEPH_MSG_CLIENT_RECONNECT ||
       type == CEPH_MSG_CLIENT_RECLAIM ||
       type == CEPH_MSG_CLIENT_REQUEST ||
+      type == CEPH_MSG_CLIENT_REPLY ||
       type == MSG_MDS_PEER_REQUEST ||
       type == MSG_MDS_HEARTBEAT ||
       type == MSG_MDS_TABLE_REQUEST ||
@@ -1232,6 +1233,7 @@ void MDSRank::handle_message(const cref_t<Message> &m)
       ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_CLIENT);
       // fall-thru
     case CEPH_MSG_CLIENT_REQUEST:
+    case CEPH_MSG_CLIENT_REPLY:
       server->dispatch(m);
       break;
     case MSG_MDS_PEER_REQUEST:
diff --git a/src/mds/MDSRank.h b/src/mds/MDSRank.h
index 1744e8268dd3b..31effdaae3bd6 100644
--- a/src/mds/MDSRank.h
+++ b/src/mds/MDSRank.h
@@ -43,6 +43,7 @@
 #include "Server.h"
 #include "MetricsHandler.h"
 #include "osdc/Journaler.h"
+#include "MDSMetaRequest.h"
 
 // Full .h import instead of forward declaration for PerfCounter, for the
 // benefit of those including this header and using MDSRank::logger
@@ -423,6 +424,8 @@ class MDSRank {
     PerfCounters *logger = nullptr, *mlogger = nullptr;
     OpTracker op_tracker;
 
+    std::map<ceph_tid_t, std::unique_ptr<MDSMetaRequest>> internal_client_requests;
+
     // The last different state I held before current
     MDSMap::DaemonState last_state = MDSMap::STATE_BOOT;
     // The state assigned to me by the MDSMap
diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index e6fe0f8c9a1c1..dbb0067020d4e 100644
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -31,6 +31,7 @@
 #include "Mutation.h"
 #include "MetricsHandler.h"
 #include "cephfs_features.h"
+#include "MDSContext.h"
 
 #include "msg/Messenger.h"
 
@@ -358,6 +359,9 @@ void Server::dispatch(const cref_t<Message> &m)
   case CEPH_MSG_CLIENT_REQUEST:
     handle_client_request(ref_cast<MClientRequest>(m));
     return;
+  case CEPH_MSG_CLIENT_REPLY:
+    handle_client_reply(ref_cast<MClientReply>(m));
+    return;
   case CEPH_MSG_CLIENT_RECLAIM:
     handle_client_reclaim(ref_cast<MClientReclaim>(m));
     return;
@@ -2289,6 +2293,10 @@ void Server::reply_client_request(MDRequestRef& mdr, const ref_t<MClientReply> &
     mds->send_message_client(reply, session);
   }
 
+  if (client_inst.name.is_mds() && reply->get_op() == CEPH_MDS_OP_RENAME) {
+    mds->send_message(reply, mdr->client_request->get_connection());
+  }
+
   if (req->is_queued_for_replay()) {
     if (int r = reply->get_result(); r < 0) {
       derr << "reply_client_request: failed to replay " << *req
@@ -2521,6 +2529,28 @@ void Server::handle_client_request(const cref_t<MClientRequest> &req)
   return;
 }
 
+void Server::handle_client_reply(const cref_t<MClientReply> &reply)
+{
+  dout(4) << "handle_client_reply " << *reply << dendl;
+
+  ceph_assert(reply->is_safe());
+  ceph_tid_t tid = reply->get_tid();
+
+  if (mds->internal_client_requests.count(tid) == 0) {
+    dout(1) << " no pending request on tid " << tid << dendl;
+    return;
+  }
+
+  switch (reply->get_op()) {
+  case CEPH_MDS_OP_RENAME:
+    break;
+  default:
+    dout(5) << " unknown client op " << reply->get_op() << dendl;
+  }
+
+  mds->internal_client_requests.erase(tid);
+}
+
 void Server::handle_osd_map()
 {
   /* Note that we check the OSDMAP_FULL flag directly rather than
diff --git a/src/mds/Server.h b/src/mds/Server.h
index efe9e0636d575..a3da04cae9834 100644
--- a/src/mds/Server.h
+++ b/src/mds/Server.h
@@ -158,6 +158,7 @@ public:
 
   // -- requests --
   void handle_client_request(const cref_t<MClientRequest> &m);
+  void handle_client_reply(const cref_t<MClientReply> &m);
 
   void journal_and_reply(MDRequestRef& mdr, CInode *tracei, CDentry *tracedn,
 			 LogEvent *le, MDSLogContextBase *fin);
-- 
2.39.5