ceph: flush mdlog before umounting

Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 9269718221748c793f57f3966fd5344b4846671b..d98a3eda0d4c4917fdb74d4e9c568873d26ae197 100644 (file)
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -4693,6 +4693,30 @@ static void wait_requests(struct ceph_mds_client *mdsc)
        dout("wait_requests done\n");
 }
 
+void send_flush_mdlog(struct ceph_mds_session *s)
+{
+       struct ceph_msg *msg;
+
+       /*
+        * Pre-luminous MDS crashes when it sees an unknown session request
+        */
+       if (!CEPH_HAVE_FEATURE(s->s_con.peer_features, SERVER_LUMINOUS))
+               return;
+
+       mutex_lock(&s->s_mutex);
+       dout("request mdlog flush to mds%d (%s)s seq %lld\n", s->s_mds,
+            ceph_session_state_name(s->s_state), s->s_seq);
+       msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_FLUSH_MDLOG,
+                                     s->s_seq);
+       if (!msg) {
+               pr_err("failed to request mdlog flush to mds%d (%s) seq %lld\n",
+                      s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
+       } else {
+               ceph_con_send(&s->s_con, msg);
+       }
+       mutex_unlock(&s->s_mutex);
+}
+
 /*
  * called before mount is ro, and before dentries are torn down.
  * (hmm, does this still race with new lookups?)
@@ -4702,6 +4726,7 @@ void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
        dout("pre_umount\n");
        mdsc->stopping = 1;
 
+       ceph_mdsc_iterate_sessions(mdsc, send_flush_mdlog, true);
        ceph_mdsc_iterate_sessions(mdsc, lock_unlock_session, false);
        ceph_flush_dirty_caps(mdsc);
        wait_requests(mdsc);

diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
index 4a75a14c2a884520d9fef53756b3d3217bf39096..97c7f7bfa55f39ba67ae1f733791148142406241 100644 (file)
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -522,6 +522,7 @@ static inline void ceph_mdsc_put_request(struct ceph_mds_request *req)
        kref_put(&req->r_kref, ceph_mdsc_release_request);
 }
 
+extern void send_flush_mdlog(struct ceph_mds_session *s);
 extern void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
                                       void (*cb)(struct ceph_mds_session *),
                                       bool check_state);

diff --git a/fs/ceph/strings.c b/fs/ceph/strings.c
index 4a79f3632260e83a9f3169aea4c11da0119f8117..573bb9556fb56cbf16802d7770fcf22305bc5469 100644 (file)
--- a/fs/ceph/strings.c
+++ b/fs/ceph/strings.c
@@ -46,6 +46,7 @@ const char *ceph_session_op_name(int op)
        case CEPH_SESSION_FLUSHMSG_ACK: return "flushmsg_ack";
        case CEPH_SESSION_FORCE_RO: return "force_ro";
        case CEPH_SESSION_REJECT: return "reject";
+       case CEPH_SESSION_REQUEST_FLUSH_MDLOG: return "flush_mdlog";
        }
        return "???";
 }

diff --git a/include/linux/ceph/ceph_fs.h b/include/linux/ceph/ceph_fs.h
index e41a811026f669bd937abd817891abd29b998898..bc2699feddbeb7d1b29000dd7b51c4888276495d 100644 (file)
--- a/include/linux/ceph/ceph_fs.h
+++ b/include/linux/ceph/ceph_fs.h
@@ -299,6 +299,7 @@ enum {
        CEPH_SESSION_FLUSHMSG_ACK,
        CEPH_SESSION_FORCE_RO,
        CEPH_SESSION_REJECT,
+       CEPH_SESSION_REQUEST_FLUSH_MDLOG,
 };
 
 extern const char *ceph_session_op_name(int op);