netfilter pull request 24-02-29

-----BEGIN PGP SIGNATURE-----
 
 iQIzBAABCgAdFiEEN9lkrMBJgcdVAPub1V2XiooUIOQFAmXfx4MACgkQ1V2XiooU
 IOSX+g//UHBfqYJASMMQJpWdMwWe7tB2m1LRzLYI+WUdUenK/MEylS7rNp/bwGkW
 42eDeGA0eov7kYNOY0rLB7lQBdUHwCpNZkdetTWFV9eHcEKA8cQ6OqcD1G8i41qg
 sCvObS+K/hq3f7fX9bJ9RvS5RvYoeuS1trw4mezhHwPS+1sj80v4FdqDOFCUqiT3
 65BfeoV65pVVteCRmJQxeeZ4Bepd4LRXW+VVyr3uXli/H87jqQOFxsOTqyXNEXIq
 jMYL0jnbYs0ARbNYXRYySLYQCWmbVXpfnt4JIBRP0S1e6Prby2hqUwJBeyNcXBAu
 CwBTjCEdLIV5G25EWTZWBYQdihct58s0GDRX078Sj/AozQJAWTxBEn0QLhKy2gvH
 2uspA0S2z1PS69hUvHfgGjDiBKw41T2O6D/12NBxI1DOYDLsk7ApE5tKqynUnUIj
 pOLUiolFnJd4JKnGZ/CTATpGi8KX/iSWdX8OElCpGOvKQgZyU8IXrydjcHnJz7b4
 AdsIfpjjZSdz2VU6ZmzLYJrWf6ukAchO5kYL2FIJt/eFEyGqDfwGL36FIO7YGcnu
 NPHtIF23Ldl+GIesc9UT08k+IOsfR9LMbUduJC6Dg63FDrEkFfOv+wXA1eURW3kS
 tq+eWs+QjlCeWG9FgW2NHj3+rGyjQbGOe+v1yTgl1x/BhXNV1cM=
 =2BRo
 -----END PGP SIGNATURE-----

Merge tag 'nf-24-02-29' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf

Pablo Neira Ayuso says:

====================
Netfilter fixes for net

Patch #1 restores NFPROTO_INET with nft_compat, from Ignat Korchagin.

Patch #2 fixes an issue with bridge netfilter and broadcast/multicast
packets.

There is a day 0 bug in br_netfilter when used with connection tracking.

Conntrack assumes that an nf_conn structure that is not yet added to
hash table ("unconfirmed"), is only visible by the current cpu that is
processing the sk_buff.

For bridge this isn't true, sk_buff can get cloned in between, and
clones can be processed in parallel on different cpu.

This patch disables NAT and conntrack helpers for multicast packets.

Patch #3 adds a selftest to cover for the br_netfilter bug.

netfilter pull request 24-02-29

* tag 'nf-24-02-29' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
  selftests: netfilter: add bridge conntrack + multicast test case
  netfilter: bridge: confirm multicast packets before passing them up the stack
  netfilter: nf_tables: allow NFPROTO_INET in nft_(match/target)_validate()
====================

Link: https://lore.kernel.org/r/20240229000135.8780-1-pablo@netfilter.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

include/linux/netfilter.h

@@ -474,6 +474,7 @@ struct nf_ct_hook {
 			      const struct sk_buff *);
 	void (*attach)(struct sk_buff *nskb, const struct sk_buff *skb);
 	void (*set_closing)(struct nf_conntrack *nfct);
+	int (*confirm)(struct sk_buff *skb);
 };
 extern const struct nf_ct_hook __rcu *nf_ct_hook;
 

net/bridge/br_netfilter_hooks.c

@@ -43,6 +43,10 @@
 #include <linux/sysctl.h>
 #endif
 
+#if IS_ENABLED(CONFIG_NF_CONNTRACK)
+#include <net/netfilter/nf_conntrack_core.h>
+#endif
+
 static unsigned int brnf_net_id __read_mostly;
 
 struct brnf_net {
@@ -553,6 +557,90 @@ static unsigned int br_nf_pre_routing(void *priv,
 	return NF_STOLEN;
 }
 
+#if IS_ENABLED(CONFIG_NF_CONNTRACK)
+/* conntracks' nf_confirm logic cannot handle cloned skbs referencing
+ * the same nf_conn entry, which will happen for multicast (broadcast)
+ * Frames on bridges.
+ *
+ * Example:
+ *      macvlan0
+ *      br0
+ *  ethX  ethY
+ *
+ * ethX (or Y) receives multicast or broadcast packet containing
+ * an IP packet, not yet in conntrack table.
+ *
+ * 1. skb passes through bridge and fake-ip (br_netfilter)Prerouting.
+ *    -> skb->_nfct now references a unconfirmed entry
+ * 2. skb is broad/mcast packet. bridge now passes clones out on each bridge
+ *    interface.
+ * 3. skb gets passed up the stack.
+ * 4. In macvlan case, macvlan driver retains clone(s) of the mcast skb
+ *    and schedules a work queue to send them out on the lower devices.
+ *
+ *    The clone skb->_nfct is not a copy, it is the same entry as the
+ *    original skb.  The macvlan rx handler then returns RX_HANDLER_PASS.
+ * 5. Normal conntrack hooks (in NF_INET_LOCAL_IN) confirm the orig skb.
+ *
+ * The Macvlan broadcast worker and normal confirm path will race.
+ *
+ * This race will not happen if step 2 already confirmed a clone. In that
+ * case later steps perform skb_clone() with skb->_nfct already confirmed (in
+ * hash table).  This works fine.
+ *
+ * But such confirmation won't happen when eb/ip/nftables rules dropped the
+ * packets before they reached the nf_confirm step in postrouting.
+ *
+ * Work around this problem by explicit confirmation of the entry at
+ * LOCAL_IN time, before upper layer has a chance to clone the unconfirmed
+ * entry.
+ *
+ */
+static unsigned int br_nf_local_in(void *priv,
+				   struct sk_buff *skb,
+				   const struct nf_hook_state *state)
+{
+	struct nf_conntrack *nfct = skb_nfct(skb);
+	const struct nf_ct_hook *ct_hook;
+	struct nf_conn *ct;
+	int ret;
+
+	if (!nfct || skb->pkt_type == PACKET_HOST)
+		return NF_ACCEPT;
+
+	ct = container_of(nfct, struct nf_conn, ct_general);
+	if (likely(nf_ct_is_confirmed(ct)))
+		return NF_ACCEPT;
+
+	WARN_ON_ONCE(skb_shared(skb));
+	WARN_ON_ONCE(refcount_read(&nfct->use) != 1);
+
+	/* We can't call nf_confirm here, it would create a dependency
+	 * on nf_conntrack module.
+	 */
+	ct_hook = rcu_dereference(nf_ct_hook);
+	if (!ct_hook) {
+		skb->_nfct = 0ul;
+		nf_conntrack_put(nfct);
+		return NF_ACCEPT;
+	}
+
+	nf_bridge_pull_encap_header(skb);
+	ret = ct_hook->confirm(skb);
+	switch (ret & NF_VERDICT_MASK) {
+	case NF_STOLEN:
+		return NF_STOLEN;
+	default:
+		nf_bridge_push_encap_header(skb);
+		break;
+	}
+
+	ct = container_of(nfct, struct nf_conn, ct_general);
+	WARN_ON_ONCE(!nf_ct_is_confirmed(ct));
+
+	return ret;
+}
+#endif
 
 /* PF_BRIDGE/FORWARD *************************************************/
 static int br_nf_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
@@ -964,6 +1052,14 @@ static const struct nf_hook_ops br_nf_ops[] = {
 		.hooknum = NF_BR_PRE_ROUTING,
 		.priority = NF_BR_PRI_BRNF,
 	},
+#if IS_ENABLED(CONFIG_NF_CONNTRACK)
+	{
+		.hook = br_nf_local_in,
+		.pf = NFPROTO_BRIDGE,
+		.hooknum = NF_BR_LOCAL_IN,
+		.priority = NF_BR_PRI_LAST,
+	},
+#endif
 	{
 		.hook = br_nf_forward,
 		.pf = NFPROTO_BRIDGE,

net/bridge/netfilter/nf_conntrack_bridge.c

@@ -291,6 +291,30 @@ static unsigned int nf_ct_bridge_pre(void *priv, struct sk_buff *skb,
 	return nf_conntrack_in(skb, &bridge_state);
 }
 
+static unsigned int nf_ct_bridge_in(void *priv, struct sk_buff *skb,
+				    const struct nf_hook_state *state)
+{
+	enum ip_conntrack_info ctinfo;
+	struct nf_conn *ct;
+
+	if (skb->pkt_type == PACKET_HOST)
+		return NF_ACCEPT;
+
+	/* nf_conntrack_confirm() cannot handle concurrent clones,
+	 * this happens for broad/multicast frames with e.g. macvlan on top
+	 * of the bridge device.
+	 */
+	ct = nf_ct_get(skb, &ctinfo);
+	if (!ct || nf_ct_is_confirmed(ct) || nf_ct_is_template(ct))
+		return NF_ACCEPT;
+
+	/* let inet prerouting call conntrack again */
+	skb->_nfct = 0;
+	nf_ct_put(ct);
+
+	return NF_ACCEPT;
+}
+
 static void nf_ct_bridge_frag_save(struct sk_buff *skb,
 				   struct nf_bridge_frag_data *data)
 {
@@ -385,6 +409,12 @@ static struct nf_hook_ops nf_ct_bridge_hook_ops[] __read_mostly = {
 		.hooknum	= NF_BR_PRE_ROUTING,
 		.priority	= NF_IP_PRI_CONNTRACK,
 	},
+	{
+		.hook		= nf_ct_bridge_in,
+		.pf		= NFPROTO_BRIDGE,
+		.hooknum	= NF_BR_LOCAL_IN,
+		.priority	= NF_IP_PRI_CONNTRACK_CONFIRM,
+	},
 	{
 		.hook		= nf_ct_bridge_post,
 		.pf		= NFPROTO_BRIDGE,

net/netfilter/nf_conntrack_core.c

@@ -2756,6 +2756,7 @@ static const struct nf_ct_hook nf_conntrack_hook = {
 	.get_tuple_skb  = nf_conntrack_get_tuple_skb,
 	.attach		= nf_conntrack_attach,
 	.set_closing	= nf_conntrack_set_closing,
+	.confirm	= __nf_conntrack_confirm,
 };
 
 void nf_conntrack_init_end(void)

net/netfilter/nft_compat.c

@@ -359,10 +359,20 @@ static int nft_target_validate(const struct nft_ctx *ctx,
 
 	if (ctx->family != NFPROTO_IPV4 &&
 	    ctx->family != NFPROTO_IPV6 &&
+	    ctx->family != NFPROTO_INET &&
 	    ctx->family != NFPROTO_BRIDGE &&
 	    ctx->family != NFPROTO_ARP)
 		return -EOPNOTSUPP;
 
+	ret = nft_chain_validate_hooks(ctx->chain,
+				       (1 << NF_INET_PRE_ROUTING) |
+				       (1 << NF_INET_LOCAL_IN) |
+				       (1 << NF_INET_FORWARD) |
+				       (1 << NF_INET_LOCAL_OUT) |
+				       (1 << NF_INET_POST_ROUTING));
+	if (ret)
+		return ret;
+
 	if (nft_is_base_chain(ctx->chain)) {
 		const struct nft_base_chain *basechain =
 						nft_base_chain(ctx->chain);
@@ -610,10 +620,20 @@ static int nft_match_validate(const struct nft_ctx *ctx,
 
 	if (ctx->family != NFPROTO_IPV4 &&
 	    ctx->family != NFPROTO_IPV6 &&
+	    ctx->family != NFPROTO_INET &&
 	    ctx->family != NFPROTO_BRIDGE &&
 	    ctx->family != NFPROTO_ARP)
 		return -EOPNOTSUPP;
 
+	ret = nft_chain_validate_hooks(ctx->chain,
+				       (1 << NF_INET_PRE_ROUTING) |
+				       (1 << NF_INET_LOCAL_IN) |
+				       (1 << NF_INET_FORWARD) |
+				       (1 << NF_INET_LOCAL_OUT) |
+				       (1 << NF_INET_POST_ROUTING));
+	if (ret)
+		return ret;
+
 	if (nft_is_base_chain(ctx->chain)) {
 		const struct nft_base_chain *basechain =
 						nft_base_chain(ctx->chain);

tools/testing/selftests/netfilter/Makefile

@@ -7,7 +7,8 @@ TEST_PROGS := nft_trans_stress.sh nft_fib.sh nft_nat.sh bridge_brouter.sh \
 	nft_queue.sh nft_meta.sh nf_nat_edemux.sh \
 	ipip-conntrack-mtu.sh conntrack_tcp_unreplied.sh \
 	conntrack_vrf.sh nft_synproxy.sh rpath.sh nft_audit.sh \
-	conntrack_sctp_collision.sh xt_string.sh
+	conntrack_sctp_collision.sh xt_string.sh \
+	bridge_netfilter.sh
 
 HOSTPKG_CONFIG := pkg-config
 

tools/testing/selftests/netfilter/bridge_netfilter.sh

@@ -0,0 +1,188 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+#
+# Test bridge netfilter + conntrack, a combination that doesn't really work,
+# with multicast/broadcast packets racing for hash table insertion.
+
+#           eth0    br0     eth0
+# setup is: ns1 <->,ns0 <-> ns3
+#           ns2 <-'    `'-> ns4
+
+# Kselftest framework requirement - SKIP code is 4.
+ksft_skip=4
+ret=0
+
+sfx=$(mktemp -u "XXXXXXXX")
+ns0="ns0-$sfx"
+ns1="ns1-$sfx"
+ns2="ns2-$sfx"
+ns3="ns3-$sfx"
+ns4="ns4-$sfx"
+
+ebtables -V > /dev/null 2>&1
+if [ $? -ne 0 ];then
+	echo "SKIP: Could not run test without ebtables"
+	exit $ksft_skip
+fi
+
+ip -Version > /dev/null 2>&1
+if [ $? -ne 0 ];then
+	echo "SKIP: Could not run test without ip tool"
+	exit $ksft_skip
+fi
+
+for i in $(seq 0 4); do
+  eval ip netns add \$ns$i
+done
+
+cleanup() {
+  for i in $(seq 0 4); do eval ip netns del \$ns$i;done
+}
+
+trap cleanup EXIT
+
+do_ping()
+{
+	fromns="$1"
+	dstip="$2"
+
+	ip netns exec $fromns ping -c 1 -q $dstip > /dev/null
+	if [ $? -ne 0 ]; then
+		echo "ERROR: ping from $fromns to $dstip"
+		ip netns exec ${ns0} nft list ruleset
+		ret=1
+	fi
+}
+
+bcast_ping()
+{
+	fromns="$1"
+	dstip="$2"
+
+	for i in $(seq 1 1000); do
+		ip netns exec $fromns ping -q -f -b -c 1 -q $dstip > /dev/null 2>&1
+		if [ $? -ne 0 ]; then
+			echo "ERROR: ping -b from $fromns to $dstip"
+			ip netns exec ${ns0} nft list ruleset
+			fi
+	done
+}
+
+ip link add veth1 netns ${ns0} type veth peer name eth0 netns ${ns1}
+if [ $? -ne 0 ]; then
+	echo "SKIP: Can't create veth device"
+	exit $ksft_skip
+fi
+
+ip link add veth2 netns ${ns0} type veth peer name eth0 netns $ns2
+ip link add veth3 netns ${ns0} type veth peer name eth0 netns $ns3
+ip link add veth4 netns ${ns0} type veth peer name eth0 netns $ns4
+
+ip -net ${ns0} link set lo up
+
+for i in $(seq 1 4); do
+  ip -net ${ns0} link set veth$i up
+done
+
+ip -net ${ns0} link add br0 type bridge stp_state 0 forward_delay 0 nf_call_iptables 1 nf_call_ip6tables 1 nf_call_arptables 1
+if [ $? -ne 0 ]; then
+	echo "SKIP: Can't create bridge br0"
+	exit $ksft_skip
+fi
+
+# make veth0,1,2 part of bridge.
+for i in $(seq 1 3); do
+  ip -net ${ns0} link set veth$i master br0
+done
+
+# add a macvlan on top of the bridge.
+MACVLAN_ADDR=ba:f3:13:37:42:23
+ip -net ${ns0} link add link br0 name macvlan0 type macvlan mode private
+ip -net ${ns0} link set macvlan0 address ${MACVLAN_ADDR}
+ip -net ${ns0} link set macvlan0 up
+ip -net ${ns0} addr add 10.23.0.1/24 dev macvlan0
+
+# add a macvlan on top of veth4.
+MACVLAN_ADDR=ba:f3:13:37:42:24
+ip -net ${ns0} link add link veth4 name macvlan4 type macvlan mode vepa
+ip -net ${ns0} link set macvlan4 address ${MACVLAN_ADDR}
+ip -net ${ns0} link set macvlan4 up
+
+# make the macvlan part of the bridge.
+# veth4 is not a bridge port, only the macvlan on top of it.
+ip -net ${ns0} link set macvlan4 master br0
+
+ip -net ${ns0} link set br0 up
+ip -net ${ns0} addr add 10.0.0.1/24 dev br0
+ip netns exec ${ns0} sysctl -q net.bridge.bridge-nf-call-iptables=1
+ret=$?
+if [ $ret -ne 0 ] ; then
+	echo "SKIP: bridge netfilter not available"
+	ret=$ksft_skip
+fi
+
+# for testing, so namespaces will reply to ping -b probes.
+ip netns exec ${ns0} sysctl -q net.ipv4.icmp_echo_ignore_broadcasts=0
+
+# enable conntrack in ns0 and drop broadcast packets in forward to
+# avoid them from getting confirmed in the postrouting hook before
+# the cloned skb is passed up the stack.
+ip netns exec ${ns0} nft -f - <<EOF
+table ip filter {
+	chain input {
+		type filter hook input priority 1; policy accept
+		iifname br0 counter
+		ct state new accept
+	}
+}
+
+table bridge filter {
+	chain forward {
+		type filter hook forward priority 0; policy accept
+		meta pkttype broadcast ip protocol icmp counter drop
+	}
+}
+EOF
+
+# place 1, 2 & 3 in same subnet, connected via ns0:br0.
+# ns4 is placed in same subnet as well, but its not
+# part of the bridge: the corresponding veth4 is not
+# part of the bridge, only its macvlan interface.
+for i in $(seq 1 4); do
+  eval ip -net \$ns$i link set lo up
+  eval ip -net \$ns$i link set eth0 up
+done
+for i in $(seq 1 2); do
+  eval ip -net \$ns$i addr add 10.0.0.1$i/24 dev eth0
+done
+
+ip -net ${ns3} addr add 10.23.0.13/24 dev eth0
+ip -net ${ns4} addr add 10.23.0.14/24 dev eth0
+
+# test basic connectivity
+do_ping ${ns1} 10.0.0.12
+do_ping ${ns3} 10.23.0.1
+do_ping ${ns4} 10.23.0.1
+
+if [ $ret -eq 0 ];then
+	echo "PASS: netns connectivity: ns1 can reach ns2, ns3 and ns4 can reach ns0"
+fi
+
+bcast_ping ${ns1} 10.0.0.255
+
+# This should deliver broadcast to macvlan0, which is on top of ns0:br0.
+bcast_ping ${ns3} 10.23.0.255
+
+# same, this time via veth4:macvlan4.
+bcast_ping ${ns4} 10.23.0.255
+
+read t < /proc/sys/kernel/tainted
+
+if [ $t -eq 0 ];then
+	echo PASS: kernel not tainted
+else
+	echo ERROR: kernel is tainted
+	ret=1
+fi
+
+exit $ret