[PATCH v3 06/10] nfsd: revoke copy-notify stateids before dropping their reference

[Jeff Layton <jlayton@kernel.org>]

Copy-notify stateids live in the s2s_cp_stateids IDR and on their parent
stid's sc_cp_list, pinned by a single membership reference.
_free_cpntf_state_locked() only unlinks an entry once its refcount reaches
zero, so any revoke path that runs while a concurrent
find_cpntf_state()/manage_cpntf_state() holder has elevated cs_count drops
the reference without unlinking, leaving the entry discoverable with its
membership reference already consumed. A second revoke or a laundromat tick
then frees it while the reader still holds the pointer -- a
KASAN-detectable use-after-free at the reader's nfs4_put_cpntf_state().

This affected all three revoke paths:

  - The parent-stid drain (nfs4_free_cpntf_statelist()) repeatedly called
    _free_cpntf_state_locked() on the first list entry; a holder that had
    bumped cs_count made it return early, so the next iteration
    re-decremented and burned the holder's reference.

  - OFFLOAD_CANCEL (manage_cpntf_state()) and laundromat expiry likewise
    used _free_cpntf_state_locked() and could drop 2->1 without unlinking.

Add revoke_cpntf_state_locked(), which unhashes the entry from the IDR and
sc_cp_list first (deferring the final free to any holder), and use it from
all three revoke paths. The drain now walks with list_for_each_entry_safe()
and revokes each entry unconditionally, so it terminates in one pass per
entry regardless of cs_count. The unhash is gated on
!list_empty(&cps->cp_list); the idr_remove() gate matters because
idr_alloc_cyclic() may have recycled the so_id by then. Keep
_free_cpntf_state_locked() for the reference-holder put path only, where a
concurrent revoke may already have unlinked the entry (its list_del_init()
then a no-op).

Fixes: 624322f1adc5 ("NFSD add COPY_NOTIFY operation")
Assisted-by: Claude:claude-opus-4-7
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 fs/nfsd/nfs4state.c | 78 ++++++++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 62 insertions(+), 16 deletions(-)

diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index b3fe163a148d..99f450f292a0 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -1026,18 +1026,66 @@ void nfs4_free_copy_state(struct nfsd4_copy *copy)
 	spin_unlock(&nn->s2s_cp_lock);
 }
 
+/*
+ * Drop the parent's reference on an already-unlinked cpntf entry. If a
+ * concurrent holder still owns a reference, its nfs4_put_cpntf_state() does
+ * the final free.
+ *
+ * nn->s2s_cp_lock must be held.
+ */
+static void put_cpntf_state_unlinked_locked(struct nfs4_cpntf_state *cps)
+{
+	WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
+	WARN_ON_ONCE(!list_empty(&cps->cp_list));
+
+	if (refcount_dec_and_test(&cps->cp_stateid.cs_count))
+		kfree(cps);
+}
+
+/*
+ * Unhash from the IDR and sc_cp_list. Gated on list_empty() to avoid
+ * evicting a recycled so_id.
+ */
+static void nfsd4_unhash_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
+{
+	lockdep_assert_held(&nn->s2s_cp_lock);
+
+	if (!list_empty(&cps->cp_list)) {
+		list_del_init(&cps->cp_list);
+		idr_remove(&nn->s2s_cp_stateids, cps->cp_stateid.cs_stid.si_opaque.so_id);
+	}
+}
+
+/*
+ * Revoke a copy-notify stateid: unlink it from the IDR and sc_cp_list first
+ * so no new finder can discover it, then drop the membership reference. Every
+ * revoke path (cancel, laundromat, drain) must use this rather than
+ * _free_cpntf_state_locked(), which unlinks only at refcount zero and so could
+ * let a second revoke free the entry under a concurrent reader.
+ *
+ * nn->s2s_cp_lock must be held.
+ */
+static void revoke_cpntf_state_locked(struct nfsd_net *nn,
+				      struct nfs4_cpntf_state *cps)
+{
+	nfsd4_unhash_cpntf_state(nn, cps);
+	put_cpntf_state_unlinked_locked(cps);
+}
+
 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
 {
-	struct nfs4_cpntf_state *cps;
+	struct nfs4_cpntf_state *cps, *tmp;
 	struct nfsd_net *nn;
 
 	nn = net_generic(net, nfsd_net_id);
 	spin_lock(&nn->s2s_cp_lock);
-	while (!list_empty(&stid->sc_cp_list)) {
-		cps = list_first_entry(&stid->sc_cp_list,
-				       struct nfs4_cpntf_state, cp_list);
-		_free_cpntf_state_locked(nn, cps);
-	}
+	/*
+	 * Revoke unlinks each entry before dropping the parent's reference, so
+	 * the drain terminates in one pass per entry regardless of cs_count; a
+	 * concurrent holder does the final kfree via nfs4_put_cpntf_state().
+	 */
+	list_for_each_entry_safe(cps, tmp, &stid->sc_cp_list, cp_list)
+		revoke_cpntf_state_locked(nn, cps);
 	spin_unlock(&nn->s2s_cp_lock);
 }
 
@@ -7484,7 +7532,7 @@ nfs4_laundromat(struct nfsd_net *nn)
 		cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
 		if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
 				state_expired(&lt, cps->cpntf_time))
-			_free_cpntf_state_locked(nn, cps);
+			revoke_cpntf_state_locked(nn, cps);
 	}
 	spin_unlock(&nn->s2s_cp_lock);
 	nfsd4_async_copy_reaper(nn);
@@ -7871,16 +7919,14 @@ nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
 out:
 	return status;
 }
-static void
-_free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
+
+static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
 {
 	WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
-	if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
-		return;
-	list_del_init(&cps->cp_list);
-	idr_remove(&nn->s2s_cp_stateids,
-		   cps->cp_stateid.cs_stid.si_opaque.so_id);
-	kfree(cps);
+	if (refcount_dec_and_test(&cps->cp_stateid.cs_count)) {
+		nfsd4_unhash_cpntf_state(nn, cps);
+		kfree(cps);
+	}
 }
 /*
  * A READ from an inter server to server COPY will have a
@@ -7917,7 +7963,7 @@ __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
 			state = NULL;
 			goto unlock;
 		} else {
-			_free_cpntf_state_locked(nn, state);
+			revoke_cpntf_state_locked(nn, state);
 		}
 	}
 unlock:

-- 
2.55.0