nfsd4: use recall_lock for delegation hashing
authorBenny Halevy <bhalevy@primarydata.com>
Fri, 30 May 2014 13:09:27 +0000 (09:09 -0400)
committerJ. Bruce Fields <bfields@redhat.com>
Wed, 4 Jun 2014 19:41:52 +0000 (15:41 -0400)
This fixes a bug in the handling of the fi_delegations list.

nfs4_setlease does not hold the recall_lock when adding to it. The
client_mutex is held, which prevents against concurrent list changes,
but nfsd_break_deleg_cb does not hold while walking it. New delegations
could theoretically creep onto the list while we're walking it there.

Signed-off-by: Benny Halevy <bhalevy@primarydata.com>
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Cc: stable@vger.kernel.org
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
fs/nfsd/nfs4state.c

index 01a0e43..05ca64c 100644 (file)
@@ -433,12 +433,21 @@ static void unhash_stid(struct nfs4_stid *s)
        s->sc_type = 0;
 }
 
+static void
+hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
+{
+       lockdep_assert_held(&recall_lock);
+
+       list_add(&dp->dl_perfile, &fp->fi_delegations);
+       list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
+}
+
 /* Called under the state lock. */
 static void
 unhash_delegation(struct nfs4_delegation *dp)
 {
-       list_del_init(&dp->dl_perclnt);
        spin_lock(&recall_lock);
+       list_del_init(&dp->dl_perclnt);
        list_del_init(&dp->dl_perfile);
        list_del_init(&dp->dl_recall_lru);
        spin_unlock(&recall_lock);
@@ -3065,11 +3074,12 @@ static int nfs4_setlease(struct nfs4_delegation *dp)
        status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
        if (status)
                goto out_free;
-       list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
        fp->fi_lease = fl;
        fp->fi_deleg_file = get_file(fl->fl_file);
        atomic_set(&fp->fi_delegees, 1);
-       list_add(&dp->dl_perfile, &fp->fi_delegations);
+       spin_lock(&recall_lock);
+       hash_delegation_locked(dp, fp);
+       spin_unlock(&recall_lock);
        return 0;
 out_free:
        locks_free_lock(fl);
@@ -3090,9 +3100,8 @@ static int nfs4_set_delegation(struct nfs4_delegation *dp, struct nfs4_file *fp)
                spin_unlock(&recall_lock);
                return -EAGAIN;
        }
-       list_add(&dp->dl_perfile, &fp->fi_delegations);
+       hash_delegation_locked(dp, fp);
        spin_unlock(&recall_lock);
-       list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
        return 0;
 }
 
@@ -4903,6 +4912,7 @@ static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
        struct nfs4_delegation *dp, *next;
        u64 count = 0;
 
+       lockdep_assert_held(&recall_lock);
        list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
                if (victims)
                        list_move(&dp->dl_recall_lru, victims);