ceph: queue MDS requests to REJECTED sessions when CLEANRECOVER is set
authorJeff Layton <jlayton@kernel.org>
Mon, 21 Sep 2020 17:12:53 +0000 (13:12 -0400)
committerIlya Dryomov <idryomov@gmail.com>
Mon, 14 Dec 2020 22:21:47 +0000 (23:21 +0100)
Ilya noticed that the first access to a blacklisted mount would often
get back -EACCES, but then subsequent calls would be OK. The problem is
in __do_request. If the session is marked as REJECTED, a hard error is
returned instead of waiting for a new session to come into being.

When the session is REJECTED and the mount was done with
recover_session=clean, queue the request to the waiting_for_map queue,
which will be awoken after tearing down the old session. We can only
do this for sync requests though, so check for async ones first and
just let the callers redrive a sync request.

URL: https://tracker.ceph.com/issues/47385
Reported-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: "Yan, Zheng" <zyan@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
fs/ceph/mds_client.c

index de95e08..aab8485 100644 (file)
@@ -2818,10 +2818,6 @@ static void __do_request(struct ceph_mds_client *mdsc,
             ceph_session_state_name(session->s_state));
        if (session->s_state != CEPH_MDS_SESSION_OPEN &&
            session->s_state != CEPH_MDS_SESSION_HUNG) {
-               if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
-                       err = -EACCES;
-                       goto out_session;
-               }
                /*
                 * We cannot queue async requests since the caps and delegated
                 * inodes are bound to the session. Just return -EJUKEBOX and
@@ -2831,6 +2827,20 @@ static void __do_request(struct ceph_mds_client *mdsc,
                        err = -EJUKEBOX;
                        goto out_session;
                }
+
+               /*
+                * If the session has been REJECTED, then return a hard error,
+                * unless it's a CLEANRECOVER mount, in which case we'll queue
+                * it to the mdsc queue.
+                */
+               if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
+                       if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
+                               list_add(&req->r_wait, &mdsc->waiting_for_map);
+                       else
+                               err = -EACCES;
+                       goto out_session;
+               }
+
                if (session->s_state == CEPH_MDS_SESSION_NEW ||
                    session->s_state == CEPH_MDS_SESSION_CLOSING) {
                        err = __open_session(mdsc, session);