1 files changed, 118 insertions, 2 deletions

diff --git a/net/netfilter/nf_nat_core.c b/net/netfilter/nf_nat_core.c
index 6d8da6dddf99..d9ea2c26f309 100644
--- a/net/netfilter/nf_nat_core.c
+++ b/net/netfilter/nf_nat_core.c
@@ -183,7 +183,35 @@ hash_by_src(const struct net *net,
 	return reciprocal_scale(hash, nf_nat_htable_size);
 }
 
-/* Is this tuple already taken? (not by us) */
+/**
+ * nf_nat_used_tuple - check if proposed nat tuple clashes with existing entry
+ * @tuple: proposed NAT binding
+ * @ignored_conntrack: our (unconfirmed) conntrack entry
+ *
+ * A conntrack entry can be inserted to the connection tracking table
+ * if there is no existing entry with an identical tuple in either direction.
+ *
+ * Example:
+ * INITIATOR -> NAT/PAT -> RESPONDER
+ *
+ * INITIATOR passes through NAT/PAT ("us") and SNAT is done (saddr rewrite).
+ * Then, later, NAT/PAT itself also connects to RESPONDER.
+ *
+ * This will not work if the SNAT done earlier has same IP:PORT source pair.
+ *
+ * Conntrack table has:
+ * ORIGINAL: $IP_INITIATOR:$SPORT -> $IP_RESPONDER:$DPORT
+ * REPLY:    $IP_RESPONDER:$DPORT -> $IP_NAT:$SPORT
+ *
+ * and new locally originating connection wants:
+ * ORIGINAL: $IP_NAT:$SPORT -> $IP_RESPONDER:$DPORT
+ * REPLY:    $IP_RESPONDER:$DPORT -> $IP_NAT:$SPORT
+ *
+ * ... which would mean incoming packets cannot be distinguished between
+ * the existing and the newly added entry (identical IP_CT_DIR_REPLY tuple).
+ *
+ * @return: true if the proposed NAT mapping collides with an existing entry.
+ */
 static int
 nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 		  const struct nf_conn *ignored_conntrack)
@@ -200,6 +228,94 @@ nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
 	return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
 }
 
+static bool nf_nat_allow_clash(const struct nf_conn *ct)
+{
+	return nf_ct_l4proto_find(nf_ct_protonum(ct))->allow_clash;
+}
+
+/**
+ * nf_nat_used_tuple_new - check if to-be-inserted conntrack collides with existing entry
+ * @tuple: proposed NAT binding
+ * @ignored_ct: our (unconfirmed) conntrack entry
+ *
+ * Same as nf_nat_used_tuple, but also check for rare clash in reverse
+ * direction. Should be called only when @tuple has not been altered, i.e.
+ * @ignored_conntrack will not be subject to NAT.
+ *
+ * @return: true if the proposed NAT mapping collides with existing entry.
+ */
+static noinline bool
+nf_nat_used_tuple_new(const struct nf_conntrack_tuple *tuple,
+		      const struct nf_conn *ignored_ct)
+{
+	static const unsigned long uses_nat = IPS_NAT_MASK | IPS_SEQ_ADJUST_BIT;
+	const struct nf_conntrack_tuple_hash *thash;
+	const struct nf_conntrack_zone *zone;
+	struct nf_conn *ct;
+	bool taken = true;
+	struct net *net;
+
+	if (!nf_nat_used_tuple(tuple, ignored_ct))
+		return false;
+
+	if (!nf_nat_allow_clash(ignored_ct))
+		return true;
+
+	/* Initial choice clashes with existing conntrack.
+	 * Check for (rare) reverse collision.
+	 *
+	 * This can happen when new packets are received in both directions
+	 * at the exact same time on different CPUs.
+	 *
+	 * Without SMP, first packet creates new conntrack entry and second
+	 * packet is resolved as established reply packet.
+	 *
+	 * With parallel processing, both packets could be picked up as
+	 * new and both get their own ct entry allocated.
+	 *
+	 * If ignored_conntrack and colliding ct are not subject to NAT then
+	 * pretend the tuple is available and let later clash resolution
+	 * handle this at insertion time.
+	 *
+	 * Without it, the 'reply' packet has its source port rewritten
+	 * by nat engine.
+	 */
+	if (READ_ONCE(ignored_ct->status) & uses_nat)
+		return true;
+
+	net = nf_ct_net(ignored_ct);
+	zone = nf_ct_zone(ignored_ct);
+
+	thash = nf_conntrack_find_get(net, zone, tuple);
+	if (unlikely(!thash)) /* clashing entry went away */
+		return false;
+
+	ct = nf_ct_tuplehash_to_ctrack(thash);
+
+	/* NB: IP_CT_DIR_ORIGINAL should be impossible because
+	 * nf_nat_used_tuple() handles origin collisions.
+	 *
+	 * Handle remote chance other CPU confirmed its ct right after.
+	 */
+	if (thash->tuple.dst.dir != IP_CT_DIR_REPLY)
+		goto out;
+
+	/* clashing connection subject to NAT? Retry with new tuple. */
+	if (READ_ONCE(ct->status) & uses_nat)
+		goto out;
+
+	if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
+			      &ignored_ct->tuplehash[IP_CT_DIR_REPLY].tuple) &&
+	    nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
+			      &ignored_ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple)) {
+		taken = false;
+		goto out;
+	}
+out:
+	nf_ct_put(ct);
+	return taken;
+}
+
 static bool nf_nat_may_kill(struct nf_conn *ct, unsigned long flags)
 {
 	static const unsigned long flags_refuse = IPS_FIXED_TIMEOUT |
@@ -611,7 +727,7 @@ get_unique_tuple(struct nf_conntrack_tuple *tuple,
 	    !(range->flags & NF_NAT_RANGE_PROTO_RANDOM_ALL)) {
 		/* try the original tuple first */
 		if (nf_in_range(orig_tuple, range)) {
-			if (!nf_nat_used_tuple(orig_tuple, ct)) {
+			if (!nf_nat_used_tuple_new(orig_tuple, ct)) {
 				*tuple = *orig_tuple;
 				return;
 			}