05dc6304992c6c5c2efb058c720da76fe6034b42
[linux-2.6.git] / net / bridge / br_netfilter.c
1 /*
2  *      Handle firewalling
3  *      Linux ethernet bridge
4  *
5  *      Authors:
6  *      Lennert Buytenhek               <buytenh@gnu.org>
7  *      Bart De Schuymer                <bdschuym@pandora.be>
8  *
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.
13  *
14  *      Lennert dedicates this file to Kerstin Wurdinger.
15  */
16
17 #include <linux/module.h>
18 #include <linux/kernel.h>
19 #include <linux/ip.h>
20 #include <linux/netdevice.h>
21 #include <linux/skbuff.h>
22 #include <linux/if_arp.h>
23 #include <linux/if_ether.h>
24 #include <linux/if_vlan.h>
25 #include <linux/if_pppox.h>
26 #include <linux/ppp_defs.h>
27 #include <linux/netfilter_bridge.h>
28 #include <linux/netfilter_ipv4.h>
29 #include <linux/netfilter_ipv6.h>
30 #include <linux/netfilter_arp.h>
31 #include <linux/in_route.h>
32 #include <linux/inetdevice.h>
33
34 #include <net/ip.h>
35 #include <net/ipv6.h>
36 #include <net/route.h>
37
38 #include <asm/uaccess.h>
39 #include "br_private.h"
40 #ifdef CONFIG_SYSCTL
41 #include <linux/sysctl.h>
42 #endif
43
44 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
45                                  (skb->nf_bridge->data))->daddr.ipv4)
46 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
47 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
48
49 #ifdef CONFIG_SYSCTL
50 static struct ctl_table_header *brnf_sysctl_header;
51 static int brnf_call_iptables __read_mostly = 1;
52 static int brnf_call_ip6tables __read_mostly = 1;
53 static int brnf_call_arptables __read_mostly = 1;
54 static int brnf_filter_vlan_tagged __read_mostly = 0;
55 static int brnf_filter_pppoe_tagged __read_mostly = 0;
56 #else
57 #define brnf_filter_vlan_tagged 0
58 #define brnf_filter_pppoe_tagged 0
59 #endif
60
61 static inline __be16 vlan_proto(const struct sk_buff *skb)
62 {
63         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
64 }
65
66 #define IS_VLAN_IP(skb) \
67         (skb->protocol == htons(ETH_P_8021Q) && \
68          vlan_proto(skb) == htons(ETH_P_IP) &&  \
69          brnf_filter_vlan_tagged)
70
71 #define IS_VLAN_IPV6(skb) \
72         (skb->protocol == htons(ETH_P_8021Q) && \
73          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
74          brnf_filter_vlan_tagged)
75
76 #define IS_VLAN_ARP(skb) \
77         (skb->protocol == htons(ETH_P_8021Q) && \
78          vlan_proto(skb) == htons(ETH_P_ARP) && \
79          brnf_filter_vlan_tagged)
80
81 static inline __be16 pppoe_proto(const struct sk_buff *skb)
82 {
83         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
84                             sizeof(struct pppoe_hdr)));
85 }
86
87 #define IS_PPPOE_IP(skb) \
88         (skb->protocol == htons(ETH_P_PPP_SES) && \
89          pppoe_proto(skb) == htons(PPP_IP) && \
90          brnf_filter_pppoe_tagged)
91
92 #define IS_PPPOE_IPV6(skb) \
93         (skb->protocol == htons(ETH_P_PPP_SES) && \
94          pppoe_proto(skb) == htons(PPP_IPV6) && \
95          brnf_filter_pppoe_tagged)
96
97 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
98 {
99 }
100
101 static struct dst_ops fake_dst_ops = {
102         .family =               AF_INET,
103         .protocol =             cpu_to_be16(ETH_P_IP),
104         .update_pmtu =          fake_update_pmtu,
105         .entries =              ATOMIC_INIT(0),
106 };
107
108 /*
109  * Initialize bogus route table used to keep netfilter happy.
110  * Currently, we fill in the PMTU entry because netfilter
111  * refragmentation needs it, and the rt_flags entry because
112  * ipt_REJECT needs it.  Future netfilter modules might
113  * require us to fill additional fields.
114  */
115 void br_netfilter_rtable_init(struct net_bridge *br)
116 {
117         struct rtable *rt = &br->fake_rtable;
118
119         atomic_set(&rt->u.dst.__refcnt, 1);
120         rt->u.dst.dev = br->dev;
121         rt->u.dst.path = &rt->u.dst;
122         rt->u.dst.metrics[RTAX_MTU - 1] = 1500;
123         rt->u.dst.flags = DST_NOXFRM;
124         rt->u.dst.ops = &fake_dst_ops;
125 }
126
127 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
128 {
129         struct net_bridge_port *port = rcu_dereference(dev->br_port);
130
131         return port ? &port->br->fake_rtable : NULL;
132 }
133
134 static inline struct net_device *bridge_parent(const struct net_device *dev)
135 {
136         struct net_bridge_port *port = rcu_dereference(dev->br_port);
137
138         return port ? port->br->dev : NULL;
139 }
140
141 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
142 {
143         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
144         if (likely(skb->nf_bridge))
145                 atomic_set(&(skb->nf_bridge->use), 1);
146
147         return skb->nf_bridge;
148 }
149
150 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
151 {
152         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
153
154         if (atomic_read(&nf_bridge->use) > 1) {
155                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
156
157                 if (tmp) {
158                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
159                         atomic_set(&tmp->use, 1);
160                         nf_bridge_put(nf_bridge);
161                 }
162                 nf_bridge = tmp;
163         }
164         return nf_bridge;
165 }
166
167 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
168 {
169         unsigned int len = nf_bridge_encap_header_len(skb);
170
171         skb_push(skb, len);
172         skb->network_header -= len;
173 }
174
175 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
176 {
177         unsigned int len = nf_bridge_encap_header_len(skb);
178
179         skb_pull(skb, len);
180         skb->network_header += len;
181 }
182
183 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
184 {
185         unsigned int len = nf_bridge_encap_header_len(skb);
186
187         skb_pull_rcsum(skb, len);
188         skb->network_header += len;
189 }
190
191 static inline void nf_bridge_save_header(struct sk_buff *skb)
192 {
193         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
194
195         skb_copy_from_linear_data_offset(skb, -header_size,
196                                          skb->nf_bridge->data, header_size);
197 }
198
199 /*
200  * When forwarding bridge frames, we save a copy of the original
201  * header before processing.
202  */
203 int nf_bridge_copy_header(struct sk_buff *skb)
204 {
205         int err;
206         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
207
208         err = skb_cow_head(skb, header_size);
209         if (err)
210                 return err;
211
212         skb_copy_to_linear_data_offset(skb, -header_size,
213                                        skb->nf_bridge->data, header_size);
214         __skb_push(skb, nf_bridge_encap_header_len(skb));
215         return 0;
216 }
217
218 /* PF_BRIDGE/PRE_ROUTING *********************************************/
219 /* Undo the changes made for ip6tables PREROUTING and continue the
220  * bridge PRE_ROUTING hook. */
221 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
222 {
223         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
224         struct rtable *rt;
225
226         if (nf_bridge->mask & BRNF_PKT_TYPE) {
227                 skb->pkt_type = PACKET_OTHERHOST;
228                 nf_bridge->mask ^= BRNF_PKT_TYPE;
229         }
230         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
231
232         rt = bridge_parent_rtable(nf_bridge->physindev);
233         if (!rt) {
234                 kfree_skb(skb);
235                 return 0;
236         }
237         dst_hold(&rt->u.dst);
238         skb_dst_set(skb, &rt->u.dst);
239
240         skb->dev = nf_bridge->physindev;
241         nf_bridge_push_encap_header(skb);
242         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
243                        br_handle_frame_finish, 1);
244
245         return 0;
246 }
247
248 /* This requires some explaining. If DNAT has taken place,
249  * we will need to fix up the destination Ethernet address.
250  *
251  * There are two cases to consider:
252  * 1. The packet was DNAT'ed to a device in the same bridge
253  *    port group as it was received on. We can still bridge
254  *    the packet.
255  * 2. The packet was DNAT'ed to a different device, either
256  *    a non-bridged device or another bridge port group.
257  *    The packet will need to be routed.
258  *
259  * The correct way of distinguishing between these two cases is to
260  * call ip_route_input() and to look at skb->dst->dev, which is
261  * changed to the destination device if ip_route_input() succeeds.
262  *
263  * Let's first consider the case that ip_route_input() succeeds:
264  *
265  * If the output device equals the logical bridge device the packet
266  * came in on, we can consider this bridging. The corresponding MAC
267  * address will be obtained in br_nf_pre_routing_finish_bridge.
268  * Otherwise, the packet is considered to be routed and we just
269  * change the destination MAC address so that the packet will
270  * later be passed up to the IP stack to be routed. For a redirected
271  * packet, ip_route_input() will give back the localhost as output device,
272  * which differs from the bridge device.
273  *
274  * Let's now consider the case that ip_route_input() fails:
275  *
276  * This can be because the destination address is martian, in which case
277  * the packet will be dropped.
278  * If IP forwarding is disabled, ip_route_input() will fail, while
279  * ip_route_output_key() can return success. The source
280  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
281  * thinks we're handling a locally generated packet and won't care
282  * if IP forwarding is enabled. If the output device equals the logical bridge
283  * device, we proceed as if ip_route_input() succeeded. If it differs from the
284  * logical bridge port or if ip_route_output_key() fails we drop the packet.
285  */
286
287 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
288 {
289         skb->dev = bridge_parent(skb->dev);
290         if (skb->dev) {
291                 struct dst_entry *dst = skb_dst(skb);
292
293                 nf_bridge_pull_encap_header(skb);
294                 skb->nf_bridge->mask |= BRNF_BRIDGED_DNAT;
295
296                 if (dst->hh)
297                         return neigh_hh_output(dst->hh, skb);
298                 else if (dst->neighbour)
299                         return dst->neighbour->output(skb);
300         }
301         kfree_skb(skb);
302         return 0;
303 }
304
305 static int br_nf_pre_routing_finish(struct sk_buff *skb)
306 {
307         struct net_device *dev = skb->dev;
308         struct iphdr *iph = ip_hdr(skb);
309         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
310         struct rtable *rt;
311         int err;
312
313         if (nf_bridge->mask & BRNF_PKT_TYPE) {
314                 skb->pkt_type = PACKET_OTHERHOST;
315                 nf_bridge->mask ^= BRNF_PKT_TYPE;
316         }
317         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
318         if (dnat_took_place(skb)) {
319                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
320                         struct flowi fl = {
321                                 .nl_u = {
322                                         .ip4_u = {
323                                                  .daddr = iph->daddr,
324                                                  .saddr = 0,
325                                                  .tos = RT_TOS(iph->tos) },
326                                 },
327                                 .proto = 0,
328                         };
329                         struct in_device *in_dev = __in_dev_get_rcu(dev);
330
331                         /* If err equals -EHOSTUNREACH the error is due to a
332                          * martian destination or due to the fact that
333                          * forwarding is disabled. For most martian packets,
334                          * ip_route_output_key() will fail. It won't fail for 2 types of
335                          * martian destinations: loopback destinations and destination
336                          * 0.0.0.0. In both cases the packet will be dropped because the
337                          * destination is the loopback device and not the bridge. */
338                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
339                                 goto free_skb;
340
341                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
342                                 /* - Bridged-and-DNAT'ed traffic doesn't
343                                  *   require ip_forwarding. */
344                                 if (((struct dst_entry *)rt)->dev == dev) {
345                                         skb_dst_set(skb, (struct dst_entry *)rt);
346                                         goto bridged_dnat;
347                                 }
348                                 dst_release((struct dst_entry *)rt);
349                         }
350 free_skb:
351                         kfree_skb(skb);
352                         return 0;
353                 } else {
354                         if (skb_dst(skb)->dev == dev) {
355 bridged_dnat:
356                                 skb->dev = nf_bridge->physindev;
357                                 nf_bridge_push_encap_header(skb);
358                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
359                                                NF_BR_PRE_ROUTING,
360                                                skb, skb->dev, NULL,
361                                                br_nf_pre_routing_finish_bridge,
362                                                1);
363                                 return 0;
364                         }
365                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
366                         skb->pkt_type = PACKET_HOST;
367                 }
368         } else {
369                 rt = bridge_parent_rtable(nf_bridge->physindev);
370                 if (!rt) {
371                         kfree_skb(skb);
372                         return 0;
373                 }
374                 dst_hold(&rt->u.dst);
375                 skb_dst_set(skb, &rt->u.dst);
376         }
377
378         skb->dev = nf_bridge->physindev;
379         nf_bridge_push_encap_header(skb);
380         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
381                        br_handle_frame_finish, 1);
382
383         return 0;
384 }
385
386 /* Some common code for IPv4/IPv6 */
387 static struct net_device *setup_pre_routing(struct sk_buff *skb)
388 {
389         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
390
391         if (skb->pkt_type == PACKET_OTHERHOST) {
392                 skb->pkt_type = PACKET_HOST;
393                 nf_bridge->mask |= BRNF_PKT_TYPE;
394         }
395
396         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
397         nf_bridge->physindev = skb->dev;
398         skb->dev = bridge_parent(skb->dev);
399
400         return skb->dev;
401 }
402
403 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
404 static int check_hbh_len(struct sk_buff *skb)
405 {
406         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
407         u32 pkt_len;
408         const unsigned char *nh = skb_network_header(skb);
409         int off = raw - nh;
410         int len = (raw[1] + 1) << 3;
411
412         if ((raw + len) - skb->data > skb_headlen(skb))
413                 goto bad;
414
415         off += 2;
416         len -= 2;
417
418         while (len > 0) {
419                 int optlen = nh[off + 1] + 2;
420
421                 switch (nh[off]) {
422                 case IPV6_TLV_PAD0:
423                         optlen = 1;
424                         break;
425
426                 case IPV6_TLV_PADN:
427                         break;
428
429                 case IPV6_TLV_JUMBO:
430                         if (nh[off + 1] != 4 || (off & 3) != 2)
431                                 goto bad;
432                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
433                         if (pkt_len <= IPV6_MAXPLEN ||
434                             ipv6_hdr(skb)->payload_len)
435                                 goto bad;
436                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
437                                 goto bad;
438                         if (pskb_trim_rcsum(skb,
439                                             pkt_len + sizeof(struct ipv6hdr)))
440                                 goto bad;
441                         nh = skb_network_header(skb);
442                         break;
443                 default:
444                         if (optlen > len)
445                                 goto bad;
446                         break;
447                 }
448                 off += optlen;
449                 len -= optlen;
450         }
451         if (len == 0)
452                 return 0;
453 bad:
454         return -1;
455
456 }
457
458 /* Replicate the checks that IPv6 does on packet reception and pass the packet
459  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
460 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
461                                            struct sk_buff *skb,
462                                            const struct net_device *in,
463                                            const struct net_device *out,
464                                            int (*okfn)(struct sk_buff *))
465 {
466         struct ipv6hdr *hdr;
467         u32 pkt_len;
468
469         if (skb->len < sizeof(struct ipv6hdr))
470                 goto inhdr_error;
471
472         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
473                 goto inhdr_error;
474
475         hdr = ipv6_hdr(skb);
476
477         if (hdr->version != 6)
478                 goto inhdr_error;
479
480         pkt_len = ntohs(hdr->payload_len);
481
482         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
483                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
484                         goto inhdr_error;
485                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
486                         goto inhdr_error;
487         }
488         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
489                 goto inhdr_error;
490
491         nf_bridge_put(skb->nf_bridge);
492         if (!nf_bridge_alloc(skb))
493                 return NF_DROP;
494         if (!setup_pre_routing(skb))
495                 return NF_DROP;
496
497         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
498                 br_nf_pre_routing_finish_ipv6);
499
500         return NF_STOLEN;
501
502 inhdr_error:
503         return NF_DROP;
504 }
505
506 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
507  * Replicate the checks that IPv4 does on packet reception.
508  * Set skb->dev to the bridge device (i.e. parent of the
509  * receiving device) to make netfilter happy, the REDIRECT
510  * target in particular.  Save the original destination IP
511  * address to be able to detect DNAT afterwards. */
512 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
513                                       const struct net_device *in,
514                                       const struct net_device *out,
515                                       int (*okfn)(struct sk_buff *))
516 {
517         struct iphdr *iph;
518         __u32 len = nf_bridge_encap_header_len(skb);
519
520         if (unlikely(!pskb_may_pull(skb, len)))
521                 goto out;
522
523         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
524             IS_PPPOE_IPV6(skb)) {
525 #ifdef CONFIG_SYSCTL
526                 if (!brnf_call_ip6tables)
527                         return NF_ACCEPT;
528 #endif
529                 nf_bridge_pull_encap_header_rcsum(skb);
530                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
531         }
532 #ifdef CONFIG_SYSCTL
533         if (!brnf_call_iptables)
534                 return NF_ACCEPT;
535 #endif
536
537         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
538             !IS_PPPOE_IP(skb))
539                 return NF_ACCEPT;
540
541         nf_bridge_pull_encap_header_rcsum(skb);
542
543         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
544                 goto inhdr_error;
545
546         iph = ip_hdr(skb);
547         if (iph->ihl < 5 || iph->version != 4)
548                 goto inhdr_error;
549
550         if (!pskb_may_pull(skb, 4 * iph->ihl))
551                 goto inhdr_error;
552
553         iph = ip_hdr(skb);
554         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
555                 goto inhdr_error;
556
557         len = ntohs(iph->tot_len);
558         if (skb->len < len || len < 4 * iph->ihl)
559                 goto inhdr_error;
560
561         pskb_trim_rcsum(skb, len);
562
563         nf_bridge_put(skb->nf_bridge);
564         if (!nf_bridge_alloc(skb))
565                 return NF_DROP;
566         if (!setup_pre_routing(skb))
567                 return NF_DROP;
568         store_orig_dstaddr(skb);
569
570         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
571                 br_nf_pre_routing_finish);
572
573         return NF_STOLEN;
574
575 inhdr_error:
576 //      IP_INC_STATS_BH(IpInHdrErrors);
577 out:
578         return NF_DROP;
579 }
580
581
582 /* PF_BRIDGE/LOCAL_IN ************************************************/
583 /* The packet is locally destined, which requires a real
584  * dst_entry, so detach the fake one.  On the way up, the
585  * packet would pass through PRE_ROUTING again (which already
586  * took place when the packet entered the bridge), but we
587  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
588  * prevent this from happening. */
589 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
590                                    const struct net_device *in,
591                                    const struct net_device *out,
592                                    int (*okfn)(struct sk_buff *))
593 {
594         struct rtable *rt = skb_rtable(skb);
595
596         if (rt && rt == bridge_parent_rtable(in))
597                 skb_dst_drop(skb);
598
599         return NF_ACCEPT;
600 }
601
602 /* PF_BRIDGE/FORWARD *************************************************/
603 static int br_nf_forward_finish(struct sk_buff *skb)
604 {
605         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
606         struct net_device *in;
607
608         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
609                 in = nf_bridge->physindev;
610                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
611                         skb->pkt_type = PACKET_OTHERHOST;
612                         nf_bridge->mask ^= BRNF_PKT_TYPE;
613                 }
614         } else {
615                 in = *((struct net_device **)(skb->cb));
616         }
617         nf_bridge_push_encap_header(skb);
618         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
619                        skb->dev, br_forward_finish, 1);
620         return 0;
621 }
622
623 /* This is the 'purely bridged' case.  For IP, we pass the packet to
624  * netfilter with indev and outdev set to the bridge device,
625  * but we are still able to filter on the 'real' indev/outdev
626  * because of the physdev module. For ARP, indev and outdev are the
627  * bridge ports. */
628 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
629                                      const struct net_device *in,
630                                      const struct net_device *out,
631                                      int (*okfn)(struct sk_buff *))
632 {
633         struct nf_bridge_info *nf_bridge;
634         struct net_device *parent;
635         u_int8_t pf;
636
637         if (!skb->nf_bridge)
638                 return NF_ACCEPT;
639
640         /* Need exclusive nf_bridge_info since we might have multiple
641          * different physoutdevs. */
642         if (!nf_bridge_unshare(skb))
643                 return NF_DROP;
644
645         parent = bridge_parent(out);
646         if (!parent)
647                 return NF_DROP;
648
649         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
650             IS_PPPOE_IP(skb))
651                 pf = PF_INET;
652         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
653                  IS_PPPOE_IPV6(skb))
654                 pf = PF_INET6;
655         else
656                 return NF_ACCEPT;
657
658         nf_bridge_pull_encap_header(skb);
659
660         nf_bridge = skb->nf_bridge;
661         if (skb->pkt_type == PACKET_OTHERHOST) {
662                 skb->pkt_type = PACKET_HOST;
663                 nf_bridge->mask |= BRNF_PKT_TYPE;
664         }
665
666         /* The physdev module checks on this */
667         nf_bridge->mask |= BRNF_BRIDGED;
668         nf_bridge->physoutdev = skb->dev;
669
670         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
671                 br_nf_forward_finish);
672
673         return NF_STOLEN;
674 }
675
676 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
677                                       const struct net_device *in,
678                                       const struct net_device *out,
679                                       int (*okfn)(struct sk_buff *))
680 {
681         struct net_device **d = (struct net_device **)(skb->cb);
682
683 #ifdef CONFIG_SYSCTL
684         if (!brnf_call_arptables)
685                 return NF_ACCEPT;
686 #endif
687
688         if (skb->protocol != htons(ETH_P_ARP)) {
689                 if (!IS_VLAN_ARP(skb))
690                         return NF_ACCEPT;
691                 nf_bridge_pull_encap_header(skb);
692         }
693
694         if (arp_hdr(skb)->ar_pln != 4) {
695                 if (IS_VLAN_ARP(skb))
696                         nf_bridge_push_encap_header(skb);
697                 return NF_ACCEPT;
698         }
699         *d = (struct net_device *)in;
700         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
701                 (struct net_device *)out, br_nf_forward_finish);
702
703         return NF_STOLEN;
704 }
705
706 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
707 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
708 {
709         if (skb->nfct != NULL &&
710             (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb)) &&
711             skb->len > skb->dev->mtu &&
712             !skb_is_gso(skb))
713                 return ip_fragment(skb, br_dev_queue_push_xmit);
714         else
715                 return br_dev_queue_push_xmit(skb);
716 }
717 #else
718 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
719 {
720         return br_dev_queue_push_xmit(skb);
721 }
722 #endif
723
724 /* PF_BRIDGE/POST_ROUTING ********************************************/
725 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
726                                        const struct net_device *in,
727                                        const struct net_device *out,
728                                        int (*okfn)(struct sk_buff *))
729 {
730         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
731         struct net_device *realoutdev = bridge_parent(skb->dev);
732         u_int8_t pf;
733
734         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
735                 return NF_ACCEPT;
736
737         if (!realoutdev)
738                 return NF_DROP;
739
740         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
741             IS_PPPOE_IP(skb))
742                 pf = PF_INET;
743         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
744                  IS_PPPOE_IPV6(skb))
745                 pf = PF_INET6;
746         else
747                 return NF_ACCEPT;
748
749         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
750          * about the value of skb->pkt_type. */
751         if (skb->pkt_type == PACKET_OTHERHOST) {
752                 skb->pkt_type = PACKET_HOST;
753                 nf_bridge->mask |= BRNF_PKT_TYPE;
754         }
755
756         nf_bridge_pull_encap_header(skb);
757         nf_bridge_save_header(skb);
758
759         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
760                 br_nf_dev_queue_xmit);
761
762         return NF_STOLEN;
763 }
764
765 /* IP/SABOTAGE *****************************************************/
766 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
767  * for the second time. */
768 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
769                                    const struct net_device *in,
770                                    const struct net_device *out,
771                                    int (*okfn)(struct sk_buff *))
772 {
773         if (skb->nf_bridge &&
774             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
775                 return NF_STOP;
776         }
777
778         return NF_ACCEPT;
779 }
780
781 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
782  * br_dev_queue_push_xmit is called afterwards */
783 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
784         {
785                 .hook = br_nf_pre_routing,
786                 .owner = THIS_MODULE,
787                 .pf = PF_BRIDGE,
788                 .hooknum = NF_BR_PRE_ROUTING,
789                 .priority = NF_BR_PRI_BRNF,
790         },
791         {
792                 .hook = br_nf_local_in,
793                 .owner = THIS_MODULE,
794                 .pf = PF_BRIDGE,
795                 .hooknum = NF_BR_LOCAL_IN,
796                 .priority = NF_BR_PRI_BRNF,
797         },
798         {
799                 .hook = br_nf_forward_ip,
800                 .owner = THIS_MODULE,
801                 .pf = PF_BRIDGE,
802                 .hooknum = NF_BR_FORWARD,
803                 .priority = NF_BR_PRI_BRNF - 1,
804         },
805         {
806                 .hook = br_nf_forward_arp,
807                 .owner = THIS_MODULE,
808                 .pf = PF_BRIDGE,
809                 .hooknum = NF_BR_FORWARD,
810                 .priority = NF_BR_PRI_BRNF,
811         },
812         {
813                 .hook = br_nf_post_routing,
814                 .owner = THIS_MODULE,
815                 .pf = PF_BRIDGE,
816                 .hooknum = NF_BR_POST_ROUTING,
817                 .priority = NF_BR_PRI_LAST,
818         },
819         {
820                 .hook = ip_sabotage_in,
821                 .owner = THIS_MODULE,
822                 .pf = PF_INET,
823                 .hooknum = NF_INET_PRE_ROUTING,
824                 .priority = NF_IP_PRI_FIRST,
825         },
826         {
827                 .hook = ip_sabotage_in,
828                 .owner = THIS_MODULE,
829                 .pf = PF_INET6,
830                 .hooknum = NF_INET_PRE_ROUTING,
831                 .priority = NF_IP6_PRI_FIRST,
832         },
833 };
834
835 #ifdef CONFIG_SYSCTL
836 static
837 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
838                             void __user * buffer, size_t * lenp, loff_t * ppos)
839 {
840         int ret;
841
842         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
843
844         if (write && *(int *)(ctl->data))
845                 *(int *)(ctl->data) = 1;
846         return ret;
847 }
848
849 static ctl_table brnf_table[] = {
850         {
851                 .procname       = "bridge-nf-call-arptables",
852                 .data           = &brnf_call_arptables,
853                 .maxlen         = sizeof(int),
854                 .mode           = 0644,
855                 .proc_handler   = brnf_sysctl_call_tables,
856         },
857         {
858                 .procname       = "bridge-nf-call-iptables",
859                 .data           = &brnf_call_iptables,
860                 .maxlen         = sizeof(int),
861                 .mode           = 0644,
862                 .proc_handler   = brnf_sysctl_call_tables,
863         },
864         {
865                 .procname       = "bridge-nf-call-ip6tables",
866                 .data           = &brnf_call_ip6tables,
867                 .maxlen         = sizeof(int),
868                 .mode           = 0644,
869                 .proc_handler   = brnf_sysctl_call_tables,
870         },
871         {
872                 .procname       = "bridge-nf-filter-vlan-tagged",
873                 .data           = &brnf_filter_vlan_tagged,
874                 .maxlen         = sizeof(int),
875                 .mode           = 0644,
876                 .proc_handler   = brnf_sysctl_call_tables,
877         },
878         {
879                 .procname       = "bridge-nf-filter-pppoe-tagged",
880                 .data           = &brnf_filter_pppoe_tagged,
881                 .maxlen         = sizeof(int),
882                 .mode           = 0644,
883                 .proc_handler   = brnf_sysctl_call_tables,
884         },
885         { }
886 };
887
888 static struct ctl_path brnf_path[] = {
889         { .procname = "net", },
890         { .procname = "bridge", },
891         { }
892 };
893 #endif
894
895 int __init br_netfilter_init(void)
896 {
897         int ret;
898
899         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
900         if (ret < 0)
901                 return ret;
902 #ifdef CONFIG_SYSCTL
903         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
904         if (brnf_sysctl_header == NULL) {
905                 printk(KERN_WARNING
906                        "br_netfilter: can't register to sysctl.\n");
907                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
908                 return -ENOMEM;
909         }
910 #endif
911         printk(KERN_NOTICE "Bridge firewalling registered\n");
912         return 0;
913 }
914
915 void br_netfilter_fini(void)
916 {
917         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
918 #ifdef CONFIG_SYSCTL
919         unregister_sysctl_table(brnf_sysctl_header);
920 #endif
921 }