Received-SPF: pass (google.com: domain of linux-nfs+bounces-773-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk
MIME-Version: 1.0
From: "NeilBrown" <neilb@suse.de>
To: "Chuck Lever" <chuck.lever@oracle.com>, "Jeff Layton" <jlayton@kernel.org>,
 "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Olga Kornievskaia" <kolga@netapp.com>, "Dai Ngo" <Dai.Ngo@oracle.com>,
 "Tom Talpey" <tom@talpey.com>, linux-nfs@vger.kernel.org
Subject: [PATCH v2] nfsd: drop st_mutex and rp_mutex before calling
 move_to_close_lru()
In-reply-to: <170320926037.11005.9834662167645370066@noble.neil.brown.name>
References: <170320926037.11005.9834662167645370066@noble.neil.brown.name>
Date: Fri, 22 Dec 2023 13:12:10 +1100
Message-id: <170321113026.11005.17173312563294650530@noble.neil.brown.name>


move_to_close_lru() is currently called with ->st_mutex and .rp_mutex held.
This can lead to a deadlock as move_to_close_lru() waits for sc_count to
drop to 2, and some threads holding a reference might be waiting for either
mutex.  These references will never be dropped so sc_count will never
reach 2.

There can be no harm in dropping ->st_mutex to before
move_to_close_lru() because the only place that takes the mutex is
nfsd4_lock_ol_stateid(), and it quickly aborts if sc_type is
NFS4_CLOSED_STID, which it will be before move_to_close_lru() is called.

Similarly dropping .rp_mutex is safe after the state is closed and so
no longer usable.  Another way to look at this is that nothing
significant happens between when nfsd4_close() now calls
nfsd4_cstate_clear_replay(), and where nfsd4_proc_compound calls
nfsd4_cstate_clear_replay() a little later.

See also
 https://lore.kernel.org/lkml/4dd1fe21e11344e5969bb112e954affb@jd.com/T/
where this problem was raised but not successfully resolved.

Signed-off-by: NeilBrown <neilb@suse.de>
---

Sorry - I posted v1 a little hastily.  I need to drop rp_mutex as well
to avoid the deadlock.  This also is safe.

 fs/nfsd/nfs4state.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index 40415929e2ae..453714fbcd66 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -7055,7 +7055,7 @@ nfsd4_open_downgrade(struct svc_rqst *rqstp,
 	return status;
 }
=20
-static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
+static bool nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
 {
 	struct nfs4_client *clp =3D s->st_stid.sc_client;
 	bool unhashed;
@@ -7072,11 +7072,11 @@ static void nfsd4_close_open_stateid(struct nfs4_ol_s=
tateid *s)
 		list_for_each_entry(stp, &reaplist, st_locks)
 			nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
 		free_ol_stateid_reaplist(&reaplist);
+		return false;
 	} else {
 		spin_unlock(&clp->cl_lock);
 		free_ol_stateid_reaplist(&reaplist);
-		if (unhashed)
-			move_to_close_lru(s, clp->net);
+		return unhashed;
 	}
 }
=20
@@ -7092,6 +7092,7 @@ nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compou=
nd_state *cstate,
 	struct nfs4_ol_stateid *stp;
 	struct net *net =3D SVC_NET(rqstp);
 	struct nfsd_net *nn =3D net_generic(net, nfsd_net_id);
+	bool need_move_to_close_list;
=20
 	dprintk("NFSD: nfsd4_close on file %pd\n",=20
 			cstate->current_fh.fh_dentry);
@@ -7114,8 +7115,11 @@ nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compo=
und_state *cstate,
 	 */
 	nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
=20
-	nfsd4_close_open_stateid(stp);
+	need_move_to_close_list =3D nfsd4_close_open_stateid(stp);
 	mutex_unlock(&stp->st_mutex);
+	nfsd4_cstate_clear_replay(cstate);
+	if (need_move_to_close_list)
+		move_to_close_lru(stp, net);
=20
 	/* v4.1+ suggests that we send a special stateid in here, since the
 	 * clients should just ignore this anyway. Since this is not useful
--=20
2.43.0