NFSv4: Don't exit the state manager without clearing NFS4CLNT_MANAGER_RUNNING

If we exit the NFSv4 state manager due to a umount, then we can end up
leaving the NFS4CLNT_MANAGER_RUNNING flag set. If another mount causes
the nfs4_client to be rereferenced before it is destroyed, then we end
up never being able to recover state.

Fixes: 47c2199b6eb5 ("NFSv4.1: Ensure state manager thread dies on last ...")
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Cc: stable@vger.kernel.org # v4.15+

diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 62ae0fd345ad6751d5dbbf1ab8aefca85eff9e68..98d1b6a6646a041fcc4dc4652d22f65a12efc794 100644 (file)
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -2601,11 +2601,12 @@ static void nfs4_state_manager(struct nfs_client *clp)
                nfs4_clear_state_manager_bit(clp);
                /* Did we race with an attempt to give us more work? */
                if (clp->cl_state == 0)
-                       break;
+                       return;
                if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
-                       break;
+                       return;
        } while (refcount_read(&clp->cl_count) > 1);
-       return;
+       goto out_drain;
+
 out_error:
        if (strlen(section))
                section_sep = ": ";
@@ -2613,6 +2614,7 @@ out_error:
                        " with error %d\n", section_sep, section,
                        clp->cl_hostname, -status);
        ssleep(1);
+out_drain:
        nfs4_end_drain_session(clp);
        nfs4_clear_state_manager_bit(clp);
 }