From c39aaf90ed1b23343eba2b341bb8ee6a50a4ea74 Mon Sep 17 00:00:00 2001
From: John Spray <john.spray@redhat.com>
Date: Mon, 27 Mar 2017 12:56:31 +0100
Subject: [PATCH] mds: validate prealloc_inos on sessions after load

Mitigates http://tracker.ceph.com/issues/16842

Signed-off-by: John Spray <john.spray@redhat.com>
---
 src/mds/InoTable.cc | 11 +++++++++++
 src/mds/InoTable.h  |  3 +++
 src/mds/MDSRank.cc  | 34 ++++++++++++++++++++++++++++++++++
 src/mds/MDSRank.h   |  2 ++
 4 files changed, 50 insertions(+)

diff --git a/src/mds/InoTable.cc b/src/mds/InoTable.cc
index f192b12ac05..fc1faa87642 100644
--- a/src/mds/InoTable.cc
+++ b/src/mds/InoTable.cc
@@ -198,6 +198,17 @@ bool InoTable::is_marked_free(inodeno_t id) const
   return free.contains(id) || projected_free.contains(id);
 }
 
+bool InoTable::intersects_free(
+    const interval_set<inodeno_t> &other,
+    interval_set<inodeno_t> *intersection)
+{
+  interval_set<inodeno_t> i;
+  i.intersection_of(free, other);
+  if (intersection != nullptr) {
+    *intersection = i;
+  }
+  return !(i.empty());
+}
 
 bool InoTable::repair(inodeno_t id)
 {
diff --git a/src/mds/InoTable.h b/src/mds/InoTable.h
index a787d348a32..5f6a0b385ac 100644
--- a/src/mds/InoTable.h
+++ b/src/mds/InoTable.h
@@ -43,6 +43,9 @@ class InoTable : public MDSTable {
   void replay_reset();
   bool repair(inodeno_t id);
   bool is_marked_free(inodeno_t id) const;
+  bool intersects_free(
+      const interval_set<inodeno_t> &other,
+      interval_set<inodeno_t> *intersection);
 
   void reset_state() override;
   void encode_state(bufferlist& bl) const override {
diff --git a/src/mds/MDSRank.cc b/src/mds/MDSRank.cc
index eb538f9c849..3eeba37d471 100644
--- a/src/mds/MDSRank.cc
+++ b/src/mds/MDSRank.cc
@@ -1036,11 +1036,45 @@ void MDSRank::boot_start(BootStep step, int r)
       break;
     case MDS_BOOT_REPLAY_DONE:
       assert(is_any_replay());
+
+      // Sessiontable and inotable should be in sync after replay, validate
+      // that they are consistent.
+      validate_sessions();
+
       replay_done();
       break;
   }
 }
 
+void MDSRank::validate_sessions()
+{
+  assert(mds_lock.is_locked_by_me());
+  std::vector<Session*> victims;
+
+  // Identify any sessions which have state inconsistent with other,
+  // after they have been loaded from rados during startup.
+  // Mitigate bugs like: http://tracker.ceph.com/issues/16842
+  const auto &sessions = sessionmap.get_sessions();
+  for (const auto &i : sessions) {
+    Session *session = i.second;
+    interval_set<inodeno_t> badones;
+    if (inotable->intersects_free(session->info.prealloc_inos, &badones)) {
+      clog->error() << "Client session loaded with invalid preallocated "
+                          "inodes, evicting session " << *session;
+
+      // Make the session consistent with inotable so that it can
+      // be cleanly torn down
+      session->info.prealloc_inos.subtract(badones);
+
+      victims.push_back(session);
+    }
+  }
+
+  for (const auto &session: victims) {
+    server->kill_session(session, nullptr);
+  }
+}
+
 void MDSRank::starting_done()
 {
   dout(3) << "starting_done" << dendl;
diff --git a/src/mds/MDSRank.h b/src/mds/MDSRank.h
index 8c0802561b6..5fc6376b6b8 100644
--- a/src/mds/MDSRank.h
+++ b/src/mds/MDSRank.h
@@ -480,6 +480,8 @@ class MDSRank {
     void active_start();
     void stopping_start();
     void stopping_done();
+
+    void validate_sessions();
     // <<<
     
     // >>>
-- 
2.39.5