3 * Linux ethernet bridge
6 * Lennert Buytenhek <buytenh@gnu.org>
7 * Bart De Schuymer <bdschuym@pandora.be>
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; either version
12 * 2 of the License, or (at your option) any later version.
14 * Lennert dedicates this file to Kerstin Wurdinger.
17 #include <linux/module.h>
18 #include <linux/kernel.h>
19 #include <linux/slab.h>
21 #include <linux/netdevice.h>
22 #include <linux/skbuff.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <linux/if_pppox.h>
27 #include <linux/ppp_defs.h>
28 #include <linux/netfilter_bridge.h>
29 #include <linux/netfilter_ipv4.h>
30 #include <linux/netfilter_ipv6.h>
31 #include <linux/netfilter_arp.h>
32 #include <linux/in_route.h>
33 #include <linux/inetdevice.h>
37 #include <net/route.h>
38 #include <net/netfilter/br_netfilter.h>
40 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
41 #include <net/netfilter/nf_conntrack.h>
44 #include <asm/uaccess.h>
45 #include "br_private.h"
47 #include <linux/sysctl.h>
51 static struct ctl_table_header *brnf_sysctl_header;
52 static int brnf_call_iptables __read_mostly = 1;
53 static int brnf_call_ip6tables __read_mostly = 1;
54 static int brnf_call_arptables __read_mostly = 1;
55 static int brnf_filter_vlan_tagged __read_mostly = 0;
56 static int brnf_filter_pppoe_tagged __read_mostly = 0;
57 static int brnf_pass_vlan_indev __read_mostly = 0;
59 #define brnf_call_iptables 1
60 #define brnf_call_ip6tables 1
61 #define brnf_call_arptables 1
62 #define brnf_filter_vlan_tagged 0
63 #define brnf_filter_pppoe_tagged 0
64 #define brnf_pass_vlan_indev 0
68 (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
70 #define IS_IPV6(skb) \
71 (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
74 (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
76 static inline __be16 vlan_proto(const struct sk_buff *skb)
78 if (skb_vlan_tag_present(skb))
80 else if (skb->protocol == htons(ETH_P_8021Q))
81 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
86 #define IS_VLAN_IP(skb) \
87 (vlan_proto(skb) == htons(ETH_P_IP) && \
88 brnf_filter_vlan_tagged)
90 #define IS_VLAN_IPV6(skb) \
91 (vlan_proto(skb) == htons(ETH_P_IPV6) && \
92 brnf_filter_vlan_tagged)
94 #define IS_VLAN_ARP(skb) \
95 (vlan_proto(skb) == htons(ETH_P_ARP) && \
96 brnf_filter_vlan_tagged)
98 static inline __be16 pppoe_proto(const struct sk_buff *skb)
100 return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
101 sizeof(struct pppoe_hdr)));
104 #define IS_PPPOE_IP(skb) \
105 (skb->protocol == htons(ETH_P_PPP_SES) && \
106 pppoe_proto(skb) == htons(PPP_IP) && \
107 brnf_filter_pppoe_tagged)
109 #define IS_PPPOE_IPV6(skb) \
110 (skb->protocol == htons(ETH_P_PPP_SES) && \
111 pppoe_proto(skb) == htons(PPP_IPV6) && \
112 brnf_filter_pppoe_tagged)
114 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
116 struct net_bridge_port *port;
118 port = br_port_get_rcu(dev);
119 return port ? &port->br->fake_rtable : NULL;
122 static inline struct net_device *bridge_parent(const struct net_device *dev)
124 struct net_bridge_port *port;
126 port = br_port_get_rcu(dev);
127 return port ? port->br->dev : NULL;
130 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
132 skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
133 if (likely(skb->nf_bridge))
134 atomic_set(&(skb->nf_bridge->use), 1);
136 return skb->nf_bridge;
139 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
141 struct nf_bridge_info *nf_bridge = skb->nf_bridge;
143 if (atomic_read(&nf_bridge->use) > 1) {
144 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
147 memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
148 atomic_set(&tmp->use, 1);
150 nf_bridge_put(nf_bridge);
156 static unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
158 switch (skb->protocol) {
159 case __cpu_to_be16(ETH_P_8021Q):
161 case __cpu_to_be16(ETH_P_PPP_SES):
162 return PPPOE_SES_HLEN;
168 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
170 unsigned int len = nf_bridge_encap_header_len(skb);
173 skb->network_header -= len;
176 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
178 unsigned int len = nf_bridge_encap_header_len(skb);
181 skb->network_header += len;
184 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
186 unsigned int len = nf_bridge_encap_header_len(skb);
188 skb_pull_rcsum(skb, len);
189 skb->network_header += len;
192 static inline void nf_bridge_save_header(struct sk_buff *skb)
194 int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
196 skb_copy_from_linear_data_offset(skb, -header_size,
197 skb->nf_bridge->data, header_size);
200 /* When handing a packet over to the IP layer
201 * check whether we have a skb that is in the
205 static int br_parse_ip_options(struct sk_buff *skb)
207 const struct iphdr *iph;
208 struct net_device *dev = skb->dev;
211 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
216 /* Basic sanity checks */
217 if (iph->ihl < 5 || iph->version != 4)
220 if (!pskb_may_pull(skb, iph->ihl*4))
224 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
227 len = ntohs(iph->tot_len);
228 if (skb->len < len) {
229 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
231 } else if (len < (iph->ihl*4))
234 if (pskb_trim_rcsum(skb, len)) {
235 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
239 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
240 /* We should really parse IP options here but until
241 * somebody who actually uses IP options complains to
242 * us we'll just silently ignore the options because
248 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
253 static void nf_bridge_update_protocol(struct sk_buff *skb)
255 if (skb->nf_bridge->mask & BRNF_8021Q)
256 skb->protocol = htons(ETH_P_8021Q);
257 else if (skb->nf_bridge->mask & BRNF_PPPoE)
258 skb->protocol = htons(ETH_P_PPP_SES);
261 /* PF_BRIDGE/PRE_ROUTING *********************************************/
262 /* Undo the changes made for ip6tables PREROUTING and continue the
263 * bridge PRE_ROUTING hook. */
264 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
266 struct nf_bridge_info *nf_bridge = skb->nf_bridge;
269 if (nf_bridge->mask & BRNF_PKT_TYPE) {
270 skb->pkt_type = PACKET_OTHERHOST;
271 nf_bridge->mask ^= BRNF_PKT_TYPE;
273 nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
275 rt = bridge_parent_rtable(nf_bridge->physindev);
280 skb_dst_set_noref(skb, &rt->dst);
282 skb->dev = nf_bridge->physindev;
283 nf_bridge_update_protocol(skb);
284 nf_bridge_push_encap_header(skb);
285 NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
286 br_handle_frame_finish, 1);
291 /* Obtain the correct destination MAC address, while preserving the original
292 * source MAC address. If we already know this address, we just copy it. If we
293 * don't, we use the neighbour framework to find out. In both cases, we make
294 * sure that br_handle_frame_finish() is called afterwards.
296 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
298 struct nf_bridge_info *nf_bridge = skb->nf_bridge;
299 struct neighbour *neigh;
300 struct dst_entry *dst;
302 skb->dev = bridge_parent(skb->dev);
306 neigh = dst_neigh_lookup_skb(dst, skb);
310 if (neigh->hh.hh_len) {
311 neigh_hh_bridge(&neigh->hh, skb);
312 skb->dev = nf_bridge->physindev;
313 ret = br_handle_frame_finish(skb);
315 /* the neighbour function below overwrites the complete
316 * MAC header, so we save the Ethernet source address and
319 skb_copy_from_linear_data_offset(skb,
320 -(ETH_HLEN-ETH_ALEN),
321 skb->nf_bridge->data,
323 /* tell br_dev_xmit to continue with forwarding */
324 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
325 /* FIXME Need to refragment */
326 ret = neigh->output(neigh, skb);
328 neigh_release(neigh);
336 static bool dnat_took_place(const struct sk_buff *skb)
338 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
339 enum ip_conntrack_info ctinfo;
342 ct = nf_ct_get(skb, &ctinfo);
343 if (!ct || nf_ct_is_untracked(ct))
346 return test_bit(IPS_DST_NAT_BIT, &ct->status);
352 /* This requires some explaining. If DNAT has taken place,
353 * we will need to fix up the destination Ethernet address.
355 * There are two cases to consider:
356 * 1. The packet was DNAT'ed to a device in the same bridge
357 * port group as it was received on. We can still bridge
359 * 2. The packet was DNAT'ed to a different device, either
360 * a non-bridged device or another bridge port group.
361 * The packet will need to be routed.
363 * The correct way of distinguishing between these two cases is to
364 * call ip_route_input() and to look at skb->dst->dev, which is
365 * changed to the destination device if ip_route_input() succeeds.
367 * Let's first consider the case that ip_route_input() succeeds:
369 * If the output device equals the logical bridge device the packet
370 * came in on, we can consider this bridging. The corresponding MAC
371 * address will be obtained in br_nf_pre_routing_finish_bridge.
372 * Otherwise, the packet is considered to be routed and we just
373 * change the destination MAC address so that the packet will
374 * later be passed up to the IP stack to be routed. For a redirected
375 * packet, ip_route_input() will give back the localhost as output device,
376 * which differs from the bridge device.
378 * Let's now consider the case that ip_route_input() fails:
380 * This can be because the destination address is martian, in which case
381 * the packet will be dropped.
382 * If IP forwarding is disabled, ip_route_input() will fail, while
383 * ip_route_output_key() can return success. The source
384 * address for ip_route_output_key() is set to zero, so ip_route_output_key()
385 * thinks we're handling a locally generated packet and won't care
386 * if IP forwarding is enabled. If the output device equals the logical bridge
387 * device, we proceed as if ip_route_input() succeeded. If it differs from the
388 * logical bridge port or if ip_route_output_key() fails we drop the packet.
390 static int br_nf_pre_routing_finish(struct sk_buff *skb)
392 struct net_device *dev = skb->dev;
393 struct iphdr *iph = ip_hdr(skb);
394 struct nf_bridge_info *nf_bridge = skb->nf_bridge;
399 frag_max_size = IPCB(skb)->frag_max_size;
400 BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
402 if (nf_bridge->mask & BRNF_PKT_TYPE) {
403 skb->pkt_type = PACKET_OTHERHOST;
404 nf_bridge->mask ^= BRNF_PKT_TYPE;
406 nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
407 if (dnat_took_place(skb)) {
408 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
409 struct in_device *in_dev = __in_dev_get_rcu(dev);
411 /* If err equals -EHOSTUNREACH the error is due to a
412 * martian destination or due to the fact that
413 * forwarding is disabled. For most martian packets,
414 * ip_route_output_key() will fail. It won't fail for 2 types of
415 * martian destinations: loopback destinations and destination
416 * 0.0.0.0. In both cases the packet will be dropped because the
417 * destination is the loopback device and not the bridge. */
418 if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
421 rt = ip_route_output(dev_net(dev), iph->daddr, 0,
422 RT_TOS(iph->tos), 0);
424 /* - Bridged-and-DNAT'ed traffic doesn't
425 * require ip_forwarding. */
426 if (rt->dst.dev == dev) {
427 skb_dst_set(skb, &rt->dst);
436 if (skb_dst(skb)->dev == dev) {
438 skb->dev = nf_bridge->physindev;
439 nf_bridge_update_protocol(skb);
440 nf_bridge_push_encap_header(skb);
441 NF_HOOK_THRESH(NFPROTO_BRIDGE,
444 br_nf_pre_routing_finish_bridge,
448 ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
449 skb->pkt_type = PACKET_HOST;
452 rt = bridge_parent_rtable(nf_bridge->physindev);
457 skb_dst_set_noref(skb, &rt->dst);
460 skb->dev = nf_bridge->physindev;
461 nf_bridge_update_protocol(skb);
462 nf_bridge_push_encap_header(skb);
463 NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
464 br_handle_frame_finish, 1);
469 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
471 struct net_device *vlan, *br;
473 br = bridge_parent(dev);
474 if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
477 vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
478 skb_vlan_tag_get(skb) & VLAN_VID_MASK);
480 return vlan ? vlan : br;
483 /* Some common code for IPv4/IPv6 */
484 static struct net_device *setup_pre_routing(struct sk_buff *skb)
486 struct nf_bridge_info *nf_bridge = skb->nf_bridge;
488 if (skb->pkt_type == PACKET_OTHERHOST) {
489 skb->pkt_type = PACKET_HOST;
490 nf_bridge->mask |= BRNF_PKT_TYPE;
493 nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
494 nf_bridge->physindev = skb->dev;
495 skb->dev = brnf_get_logical_dev(skb, skb->dev);
496 if (skb->protocol == htons(ETH_P_8021Q))
497 nf_bridge->mask |= BRNF_8021Q;
498 else if (skb->protocol == htons(ETH_P_PPP_SES))
499 nf_bridge->mask |= BRNF_PPPoE;
501 /* Must drop socket now because of tproxy. */
506 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
507 static int check_hbh_len(struct sk_buff *skb)
509 unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
511 const unsigned char *nh = skb_network_header(skb);
513 int len = (raw[1] + 1) << 3;
515 if ((raw + len) - skb->data > skb_headlen(skb))
522 int optlen = nh[off + 1] + 2;
533 if (nh[off + 1] != 4 || (off & 3) != 2)
535 pkt_len = ntohl(*(__be32 *) (nh + off + 2));
536 if (pkt_len <= IPV6_MAXPLEN ||
537 ipv6_hdr(skb)->payload_len)
539 if (pkt_len > skb->len - sizeof(struct ipv6hdr))
541 if (pskb_trim_rcsum(skb,
542 pkt_len + sizeof(struct ipv6hdr)))
544 nh = skb_network_header(skb);
561 /* Replicate the checks that IPv6 does on packet reception and pass the packet
562 * to ip6tables, which doesn't support NAT, so things are fairly simple. */
563 static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
565 const struct nf_hook_state *state)
567 const struct ipv6hdr *hdr;
570 if (skb->len < sizeof(struct ipv6hdr))
573 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
578 if (hdr->version != 6)
581 pkt_len = ntohs(hdr->payload_len);
583 if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
584 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
586 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
589 if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
592 nf_bridge_put(skb->nf_bridge);
593 if (!nf_bridge_alloc(skb))
595 if (!setup_pre_routing(skb))
598 skb->protocol = htons(ETH_P_IPV6);
599 NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
600 br_nf_pre_routing_finish_ipv6);
605 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
606 * Replicate the checks that IPv4 does on packet reception.
607 * Set skb->dev to the bridge device (i.e. parent of the
608 * receiving device) to make netfilter happy, the REDIRECT
609 * target in particular. Save the original destination IP
610 * address to be able to detect DNAT afterwards. */
611 static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
613 const struct nf_hook_state *state)
615 struct net_bridge_port *p;
616 struct net_bridge *br;
617 __u32 len = nf_bridge_encap_header_len(skb);
619 if (unlikely(!pskb_may_pull(skb, len)))
622 p = br_port_get_rcu(state->in);
627 if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
628 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
631 nf_bridge_pull_encap_header_rcsum(skb);
632 return br_nf_pre_routing_ipv6(ops, skb, state);
635 if (!brnf_call_iptables && !br->nf_call_iptables)
638 if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
641 nf_bridge_pull_encap_header_rcsum(skb);
643 if (br_parse_ip_options(skb))
646 nf_bridge_put(skb->nf_bridge);
647 if (!nf_bridge_alloc(skb))
649 if (!setup_pre_routing(skb))
652 skb->protocol = htons(ETH_P_IP);
654 NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
655 br_nf_pre_routing_finish);
661 /* PF_BRIDGE/LOCAL_IN ************************************************/
662 /* The packet is locally destined, which requires a real
663 * dst_entry, so detach the fake one. On the way up, the
664 * packet would pass through PRE_ROUTING again (which already
665 * took place when the packet entered the bridge), but we
666 * register an IPv4 PRE_ROUTING 'sabotage' hook that will
667 * prevent this from happening. */
668 static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
670 const struct nf_hook_state *state)
672 br_drop_fake_rtable(skb);
676 /* PF_BRIDGE/FORWARD *************************************************/
677 static int br_nf_forward_finish(struct sk_buff *skb)
679 struct nf_bridge_info *nf_bridge = skb->nf_bridge;
680 struct net_device *in;
682 if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
683 in = nf_bridge->physindev;
684 if (nf_bridge->mask & BRNF_PKT_TYPE) {
685 skb->pkt_type = PACKET_OTHERHOST;
686 nf_bridge->mask ^= BRNF_PKT_TYPE;
688 nf_bridge_update_protocol(skb);
690 in = *((struct net_device **)(skb->cb));
692 nf_bridge_push_encap_header(skb);
694 NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
695 skb->dev, br_forward_finish, 1);
700 /* This is the 'purely bridged' case. For IP, we pass the packet to
701 * netfilter with indev and outdev set to the bridge device,
702 * but we are still able to filter on the 'real' indev/outdev
703 * because of the physdev module. For ARP, indev and outdev are the
705 static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
707 const struct nf_hook_state *state)
709 struct nf_bridge_info *nf_bridge;
710 struct net_device *parent;
716 /* Need exclusive nf_bridge_info since we might have multiple
717 * different physoutdevs. */
718 if (!nf_bridge_unshare(skb))
721 parent = bridge_parent(state->out);
725 if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
727 else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
732 nf_bridge_pull_encap_header(skb);
734 nf_bridge = skb->nf_bridge;
735 if (skb->pkt_type == PACKET_OTHERHOST) {
736 skb->pkt_type = PACKET_HOST;
737 nf_bridge->mask |= BRNF_PKT_TYPE;
740 if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
743 nf_bridge->physoutdev = skb->dev;
744 if (pf == NFPROTO_IPV4)
745 skb->protocol = htons(ETH_P_IP);
747 skb->protocol = htons(ETH_P_IPV6);
749 NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, state->in),
750 parent, br_nf_forward_finish);
755 static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
757 const struct nf_hook_state *state)
759 struct net_bridge_port *p;
760 struct net_bridge *br;
761 struct net_device **d = (struct net_device **)(skb->cb);
763 p = br_port_get_rcu(state->out);
768 if (!brnf_call_arptables && !br->nf_call_arptables)
772 if (!IS_VLAN_ARP(skb))
774 nf_bridge_pull_encap_header(skb);
777 if (arp_hdr(skb)->ar_pln != 4) {
778 if (IS_VLAN_ARP(skb))
779 nf_bridge_push_encap_header(skb);
783 NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, state->in,
784 state->out, br_nf_forward_finish);
789 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
790 static bool nf_bridge_copy_header(struct sk_buff *skb)
793 unsigned int header_size;
795 nf_bridge_update_protocol(skb);
796 header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
797 err = skb_cow_head(skb, header_size);
801 skb_copy_to_linear_data_offset(skb, -header_size,
802 skb->nf_bridge->data, header_size);
803 __skb_push(skb, nf_bridge_encap_header_len(skb));
807 static int br_nf_push_frag_xmit(struct sk_buff *skb)
809 if (!nf_bridge_copy_header(skb)) {
814 return br_dev_queue_push_xmit(skb);
817 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
821 unsigned int mtu_reserved;
823 if (skb_is_gso(skb) || skb->protocol != htons(ETH_P_IP))
824 return br_dev_queue_push_xmit(skb);
826 mtu_reserved = nf_bridge_mtu_reduction(skb);
827 /* This is wrong! We should preserve the original fragment
828 * boundaries by preserving frag_list rather than refragmenting.
830 if (skb->len + mtu_reserved > skb->dev->mtu) {
831 frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
832 if (br_parse_ip_options(skb))
833 /* Drop invalid packet */
835 IPCB(skb)->frag_max_size = frag_max_size;
836 ret = ip_fragment(skb, br_nf_push_frag_xmit);
838 ret = br_dev_queue_push_xmit(skb);
843 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
845 return br_dev_queue_push_xmit(skb);
849 /* PF_BRIDGE/POST_ROUTING ********************************************/
850 static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
852 const struct nf_hook_state *state)
854 struct nf_bridge_info *nf_bridge = skb->nf_bridge;
855 struct net_device *realoutdev = bridge_parent(skb->dev);
858 /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
859 * on a bridge, but was delivered locally and is now being routed:
861 * POST_ROUTING was already invoked from the ip stack.
863 if (!nf_bridge || !nf_bridge->physoutdev)
869 if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
871 else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
876 /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
877 * about the value of skb->pkt_type. */
878 if (skb->pkt_type == PACKET_OTHERHOST) {
879 skb->pkt_type = PACKET_HOST;
880 nf_bridge->mask |= BRNF_PKT_TYPE;
883 nf_bridge_pull_encap_header(skb);
884 nf_bridge_save_header(skb);
885 if (pf == NFPROTO_IPV4)
886 skb->protocol = htons(ETH_P_IP);
888 skb->protocol = htons(ETH_P_IPV6);
890 NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
891 br_nf_dev_queue_xmit);
896 /* IP/SABOTAGE *****************************************************/
897 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
898 * for the second time. */
899 static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
901 const struct nf_hook_state *state)
903 if (skb->nf_bridge &&
904 !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
911 /* This is called when br_netfilter has called into iptables/netfilter,
912 * and DNAT has taken place on a bridge-forwarded packet.
914 * neigh->output has created a new MAC header, with local br0 MAC
917 * This restores the original MAC saddr of the bridged packet
918 * before invoking bridge forward logic to transmit the packet.
920 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
922 struct nf_bridge_info *nf_bridge = skb->nf_bridge;
924 skb_pull(skb, ETH_HLEN);
925 nf_bridge->mask &= ~BRNF_BRIDGED_DNAT;
927 skb_copy_to_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN),
928 skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
929 skb->dev = nf_bridge->physindev;
930 br_handle_frame_finish(skb);
933 static int br_nf_dev_xmit(struct sk_buff *skb)
935 if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
936 br_nf_pre_routing_finish_bridge_slow(skb);
942 static const struct nf_br_ops br_ops = {
943 .br_dev_xmit_hook = br_nf_dev_xmit,
946 void br_netfilter_enable(void)
949 EXPORT_SYMBOL_GPL(br_netfilter_enable);
951 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
952 * br_dev_queue_push_xmit is called afterwards */
953 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
955 .hook = br_nf_pre_routing,
956 .owner = THIS_MODULE,
957 .pf = NFPROTO_BRIDGE,
958 .hooknum = NF_BR_PRE_ROUTING,
959 .priority = NF_BR_PRI_BRNF,
962 .hook = br_nf_local_in,
963 .owner = THIS_MODULE,
964 .pf = NFPROTO_BRIDGE,
965 .hooknum = NF_BR_LOCAL_IN,
966 .priority = NF_BR_PRI_BRNF,
969 .hook = br_nf_forward_ip,
970 .owner = THIS_MODULE,
971 .pf = NFPROTO_BRIDGE,
972 .hooknum = NF_BR_FORWARD,
973 .priority = NF_BR_PRI_BRNF - 1,
976 .hook = br_nf_forward_arp,
977 .owner = THIS_MODULE,
978 .pf = NFPROTO_BRIDGE,
979 .hooknum = NF_BR_FORWARD,
980 .priority = NF_BR_PRI_BRNF,
983 .hook = br_nf_post_routing,
984 .owner = THIS_MODULE,
985 .pf = NFPROTO_BRIDGE,
986 .hooknum = NF_BR_POST_ROUTING,
987 .priority = NF_BR_PRI_LAST,
990 .hook = ip_sabotage_in,
991 .owner = THIS_MODULE,
993 .hooknum = NF_INET_PRE_ROUTING,
994 .priority = NF_IP_PRI_FIRST,
997 .hook = ip_sabotage_in,
998 .owner = THIS_MODULE,
1000 .hooknum = NF_INET_PRE_ROUTING,
1001 .priority = NF_IP6_PRI_FIRST,
1005 #ifdef CONFIG_SYSCTL
1007 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1008 void __user *buffer, size_t *lenp, loff_t *ppos)
1012 ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1014 if (write && *(int *)(ctl->data))
1015 *(int *)(ctl->data) = 1;
1019 static struct ctl_table brnf_table[] = {
1021 .procname = "bridge-nf-call-arptables",
1022 .data = &brnf_call_arptables,
1023 .maxlen = sizeof(int),
1025 .proc_handler = brnf_sysctl_call_tables,
1028 .procname = "bridge-nf-call-iptables",
1029 .data = &brnf_call_iptables,
1030 .maxlen = sizeof(int),
1032 .proc_handler = brnf_sysctl_call_tables,
1035 .procname = "bridge-nf-call-ip6tables",
1036 .data = &brnf_call_ip6tables,
1037 .maxlen = sizeof(int),
1039 .proc_handler = brnf_sysctl_call_tables,
1042 .procname = "bridge-nf-filter-vlan-tagged",
1043 .data = &brnf_filter_vlan_tagged,
1044 .maxlen = sizeof(int),
1046 .proc_handler = brnf_sysctl_call_tables,
1049 .procname = "bridge-nf-filter-pppoe-tagged",
1050 .data = &brnf_filter_pppoe_tagged,
1051 .maxlen = sizeof(int),
1053 .proc_handler = brnf_sysctl_call_tables,
1056 .procname = "bridge-nf-pass-vlan-input-dev",
1057 .data = &brnf_pass_vlan_indev,
1058 .maxlen = sizeof(int),
1060 .proc_handler = brnf_sysctl_call_tables,
1066 static int __init br_netfilter_init(void)
1070 ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1074 #ifdef CONFIG_SYSCTL
1075 brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1076 if (brnf_sysctl_header == NULL) {
1078 "br_netfilter: can't register to sysctl.\n");
1079 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1083 RCU_INIT_POINTER(nf_br_ops, &br_ops);
1084 printk(KERN_NOTICE "Bridge firewalling registered\n");
1088 static void __exit br_netfilter_fini(void)
1090 RCU_INIT_POINTER(nf_br_ops, NULL);
1091 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1092 #ifdef CONFIG_SYSCTL
1093 unregister_net_sysctl_table(brnf_sysctl_header);
1097 module_init(br_netfilter_init);
1098 module_exit(br_netfilter_fini);
1100 MODULE_LICENSE("GPL");
1101 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1102 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1103 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");