141f069e77a0cfc04408d6a5e0781d69fb043611
[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 (maintainer)   <bdschuym@pandora.be>
8  *
9  *      Changes:
10  *      Apr 29 2003: physdev module support (bdschuym)
11  *      Jun 19 2003: let arptables see bridged ARP traffic (bdschuym)
12  *      Oct 06 2003: filter encapsulated IP/ARP VLAN traffic on untagged bridge
13  *                   (bdschuym)
14  *      Sep 01 2004: add IPv6 filtering (bdschuym)
15  *
16  *      This program is free software; you can redistribute it and/or
17  *      modify it under the terms of the GNU General Public License
18  *      as published by the Free Software Foundation; either version
19  *      2 of the License, or (at your option) any later version.
20  *
21  *      Lennert dedicates this file to Kerstin Wurdinger.
22  */
23
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/ip.h>
27 #include <linux/netdevice.h>
28 #include <linux/skbuff.h>
29 #include <linux/if_arp.h>
30 #include <linux/if_ether.h>
31 #include <linux/if_vlan.h>
32 #include <linux/if_pppox.h>
33 #include <linux/ppp_defs.h>
34 #include <linux/netfilter_bridge.h>
35 #include <linux/netfilter_ipv4.h>
36 #include <linux/netfilter_ipv6.h>
37 #include <linux/netfilter_arp.h>
38 #include <linux/in_route.h>
39 #include <linux/inetdevice.h>
40
41 #include <net/ip.h>
42 #include <net/ipv6.h>
43 #include <net/route.h>
44
45 #include <asm/uaccess.h>
46 #include "br_private.h"
47 #ifdef CONFIG_SYSCTL
48 #include <linux/sysctl.h>
49 #endif
50
51 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
52                                  (skb->nf_bridge->data))->daddr.ipv4)
53 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
54 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
55
56 #ifdef CONFIG_SYSCTL
57 static struct ctl_table_header *brnf_sysctl_header;
58 static int brnf_call_iptables __read_mostly = 1;
59 static int brnf_call_ip6tables __read_mostly = 1;
60 static int brnf_call_arptables __read_mostly = 1;
61 static int brnf_filter_vlan_tagged __read_mostly = 1;
62 static int brnf_filter_pppoe_tagged __read_mostly = 1;
63 #else
64 #define brnf_filter_vlan_tagged 1
65 #define brnf_filter_pppoe_tagged 1
66 #endif
67
68 static inline __be16 vlan_proto(const struct sk_buff *skb)
69 {
70         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
71 }
72
73 #define IS_VLAN_IP(skb) \
74         (skb->protocol == htons(ETH_P_8021Q) && \
75          vlan_proto(skb) == htons(ETH_P_IP) &&  \
76          brnf_filter_vlan_tagged)
77
78 #define IS_VLAN_IPV6(skb) \
79         (skb->protocol == htons(ETH_P_8021Q) && \
80          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
81          brnf_filter_vlan_tagged)
82
83 #define IS_VLAN_ARP(skb) \
84         (skb->protocol == htons(ETH_P_8021Q) && \
85          vlan_proto(skb) == htons(ETH_P_ARP) && \
86          brnf_filter_vlan_tagged)
87
88 static inline __be16 pppoe_proto(const struct sk_buff *skb)
89 {
90         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
91                             sizeof(struct pppoe_hdr)));
92 }
93
94 #define IS_PPPOE_IP(skb) \
95         (skb->protocol == htons(ETH_P_PPP_SES) && \
96          pppoe_proto(skb) == htons(PPP_IP) && \
97          brnf_filter_pppoe_tagged)
98
99 #define IS_PPPOE_IPV6(skb) \
100         (skb->protocol == htons(ETH_P_PPP_SES) && \
101          pppoe_proto(skb) == htons(PPP_IPV6) && \
102          brnf_filter_pppoe_tagged)
103
104 /* We need these fake structures to make netfilter happy --
105  * lots of places assume that skb->dst != NULL, which isn't
106  * all that unreasonable.
107  *
108  * Currently, we fill in the PMTU entry because netfilter
109  * refragmentation needs it, and the rt_flags entry because
110  * ipt_REJECT needs it.  Future netfilter modules might
111  * require us to fill additional fields. */
112 static struct net_device __fake_net_device = {
113         .hard_header_len        = ETH_HLEN
114 };
115
116 static struct rtable __fake_rtable = {
117         .u = {
118                 .dst = {
119                         .__refcnt               = ATOMIC_INIT(1),
120                         .dev                    = &__fake_net_device,
121                         .path                   = &__fake_rtable.u.dst,
122                         .metrics                = {[RTAX_MTU - 1] = 1500},
123                         .flags                  = DST_NOXFRM,
124                 }
125         },
126         .rt_flags       = 0,
127 };
128
129 static inline struct net_device *bridge_parent(const struct net_device *dev)
130 {
131         struct net_bridge_port *port = rcu_dereference(dev->br_port);
132
133         return port ? port->br->dev : NULL;
134 }
135
136 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
137 {
138         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
139         if (likely(skb->nf_bridge))
140                 atomic_set(&(skb->nf_bridge->use), 1);
141
142         return skb->nf_bridge;
143 }
144
145 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
146 {
147         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
148
149         if (atomic_read(&nf_bridge->use) > 1) {
150                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
151
152                 if (tmp) {
153                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
154                         atomic_set(&tmp->use, 1);
155                         nf_bridge_put(nf_bridge);
156                 }
157                 nf_bridge = tmp;
158         }
159         return nf_bridge;
160 }
161
162 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
163 {
164         unsigned int len = nf_bridge_encap_header_len(skb);
165
166         skb_push(skb, len);
167         skb->network_header -= len;
168 }
169
170 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
171 {
172         unsigned int len = nf_bridge_encap_header_len(skb);
173
174         skb_pull(skb, len);
175         skb->network_header += len;
176 }
177
178 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
179 {
180         unsigned int len = nf_bridge_encap_header_len(skb);
181
182         skb_pull_rcsum(skb, len);
183         skb->network_header += len;
184 }
185
186 static inline void nf_bridge_save_header(struct sk_buff *skb)
187 {
188         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
189
190         skb_copy_from_linear_data_offset(skb, -header_size,
191                                          skb->nf_bridge->data, header_size);
192 }
193
194 /*
195  * When forwarding bridge frames, we save a copy of the original
196  * header before processing.
197  */
198 int nf_bridge_copy_header(struct sk_buff *skb)
199 {
200         int err;
201         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
202
203         err = skb_cow_head(skb, header_size);
204         if (err)
205                 return err;
206
207         skb_copy_to_linear_data_offset(skb, -header_size,
208                                        skb->nf_bridge->data, header_size);
209         __skb_push(skb, nf_bridge_encap_header_len(skb));
210         return 0;
211 }
212
213 /* PF_BRIDGE/PRE_ROUTING *********************************************/
214 /* Undo the changes made for ip6tables PREROUTING and continue the
215  * bridge PRE_ROUTING hook. */
216 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
217 {
218         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
219
220         if (nf_bridge->mask & BRNF_PKT_TYPE) {
221                 skb->pkt_type = PACKET_OTHERHOST;
222                 nf_bridge->mask ^= BRNF_PKT_TYPE;
223         }
224         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
225
226         skb->dst = (struct dst_entry *)&__fake_rtable;
227         dst_hold(skb->dst);
228
229         skb->dev = nf_bridge->physindev;
230         nf_bridge_push_encap_header(skb);
231         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
232                        br_handle_frame_finish, 1);
233
234         return 0;
235 }
236
237 static void __br_dnat_complain(void)
238 {
239         static unsigned long last_complaint;
240
241         if (jiffies - last_complaint >= 5 * HZ) {
242                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
243                        "forwarding to be enabled\n");
244                 last_complaint = jiffies;
245         }
246 }
247
248 /* This requires some explaining. If DNAT has taken place,
249  * we will need to fix up the destination Ethernet address,
250  * and this is a tricky process.
251  *
252  * There are two cases to consider:
253  * 1. The packet was DNAT'ed to a device in the same bridge
254  *    port group as it was received on. We can still bridge
255  *    the packet.
256  * 2. The packet was DNAT'ed to a different device, either
257  *    a non-bridged device or another bridge port group.
258  *    The packet will need to be routed.
259  *
260  * The correct way of distinguishing between these two cases is to
261  * call ip_route_input() and to look at skb->dst->dev, which is
262  * changed to the destination device if ip_route_input() succeeds.
263  *
264  * Let us first consider the case that ip_route_input() succeeds:
265  *
266  * If skb->dst->dev equals the logical bridge device the packet
267  * came in on, we can consider this bridging. The packet is passed
268  * through the neighbour output function to build a new destination
269  * MAC address, which will make the packet enter br_nf_local_out()
270  * not much later. In that function it is assured that the iptables
271  * FORWARD chain is traversed for the packet.
272  *
273  * Otherwise, the packet is considered to be routed and we just
274  * change the destination MAC address so that the packet will
275  * later be passed up to the IP stack to be routed. For a redirected
276  * packet, ip_route_input() will give back the localhost as output device,
277  * which differs from the bridge device.
278  *
279  * Let us now consider the case that ip_route_input() fails:
280  *
281  * This can be because the destination address is martian, in which case
282  * the packet will be dropped.
283  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
284  * will fail, while __ip_route_output_key() will return success. The source
285  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
286  * thinks we're handling a locally generated packet and won't care
287  * if IP forwarding is allowed. We send a warning message to the users's
288  * log telling her to put IP forwarding on.
289  *
290  * ip_route_input() will also fail if there is no route available.
291  * In that case we just drop the packet.
292  *
293  * --Lennert, 20020411
294  * --Bart, 20020416 (updated)
295  * --Bart, 20021007 (updated)
296  * --Bart, 20062711 (updated) */
297 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
298 {
299         if (skb->pkt_type == PACKET_OTHERHOST) {
300                 skb->pkt_type = PACKET_HOST;
301                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
302         }
303         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
304
305         skb->dev = bridge_parent(skb->dev);
306         if (skb->dev) {
307                 struct dst_entry *dst = skb->dst;
308
309                 nf_bridge_pull_encap_header(skb);
310
311                 if (dst->hh)
312                         return neigh_hh_output(dst->hh, skb);
313                 else if (dst->neighbour)
314                         return dst->neighbour->output(skb);
315         }
316         kfree_skb(skb);
317         return 0;
318 }
319
320 static int br_nf_pre_routing_finish(struct sk_buff *skb)
321 {
322         struct net_device *dev = skb->dev;
323         struct iphdr *iph = ip_hdr(skb);
324         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
325         int err;
326
327         if (nf_bridge->mask & BRNF_PKT_TYPE) {
328                 skb->pkt_type = PACKET_OTHERHOST;
329                 nf_bridge->mask ^= BRNF_PKT_TYPE;
330         }
331         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
332         if (dnat_took_place(skb)) {
333                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
334                         struct rtable *rt;
335                         struct flowi fl = {
336                                 .nl_u = {
337                                         .ip4_u = {
338                                                  .daddr = iph->daddr,
339                                                  .saddr = 0,
340                                                  .tos = RT_TOS(iph->tos) },
341                                 },
342                                 .proto = 0,
343                         };
344                         struct in_device *in_dev = in_dev_get(dev);
345
346                         /* If err equals -EHOSTUNREACH the error is due to a
347                          * martian destination or due to the fact that
348                          * forwarding is disabled. For most martian packets,
349                          * ip_route_output_key() will fail. It won't fail for 2 types of
350                          * martian destinations: loopback destinations and destination
351                          * 0.0.0.0. In both cases the packet will be dropped because the
352                          * destination is the loopback device and not the bridge. */
353                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
354                                 goto free_skb;
355
356                         if (!ip_route_output_key(&rt, &fl)) {
357                                 /* - Bridged-and-DNAT'ed traffic doesn't
358                                  *   require ip_forwarding. */
359                                 if (((struct dst_entry *)rt)->dev == dev) {
360                                         skb->dst = (struct dst_entry *)rt;
361                                         goto bridged_dnat;
362                                 }
363                                 /* we are sure that forwarding is disabled, so printing
364                                  * this message is no problem. Note that the packet could
365                                  * still have a martian destination address, in which case
366                                  * the packet could be dropped even if forwarding were enabled */
367                                 __br_dnat_complain();
368                                 dst_release((struct dst_entry *)rt);
369                         }
370 free_skb:
371                         kfree_skb(skb);
372                         return 0;
373                 } else {
374                         if (skb->dst->dev == dev) {
375 bridged_dnat:
376                                 /* Tell br_nf_local_out this is a
377                                  * bridged frame */
378                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
379                                 skb->dev = nf_bridge->physindev;
380                                 nf_bridge_push_encap_header(skb);
381                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
382                                                skb, skb->dev, NULL,
383                                                br_nf_pre_routing_finish_bridge,
384                                                1);
385                                 return 0;
386                         }
387                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
388                         skb->pkt_type = PACKET_HOST;
389                 }
390         } else {
391                 skb->dst = (struct dst_entry *)&__fake_rtable;
392                 dst_hold(skb->dst);
393         }
394
395         skb->dev = nf_bridge->physindev;
396         nf_bridge_push_encap_header(skb);
397         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
398                        br_handle_frame_finish, 1);
399
400         return 0;
401 }
402
403 /* Some common code for IPv4/IPv6 */
404 static struct net_device *setup_pre_routing(struct sk_buff *skb)
405 {
406         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
407
408         if (skb->pkt_type == PACKET_OTHERHOST) {
409                 skb->pkt_type = PACKET_HOST;
410                 nf_bridge->mask |= BRNF_PKT_TYPE;
411         }
412
413         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
414         nf_bridge->physindev = skb->dev;
415         skb->dev = bridge_parent(skb->dev);
416
417         return skb->dev;
418 }
419
420 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
421 static int check_hbh_len(struct sk_buff *skb)
422 {
423         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
424         u32 pkt_len;
425         const unsigned char *nh = skb_network_header(skb);
426         int off = raw - nh;
427         int len = (raw[1] + 1) << 3;
428
429         if ((raw + len) - skb->data > skb_headlen(skb))
430                 goto bad;
431
432         off += 2;
433         len -= 2;
434
435         while (len > 0) {
436                 int optlen = nh[off + 1] + 2;
437
438                 switch (nh[off]) {
439                 case IPV6_TLV_PAD0:
440                         optlen = 1;
441                         break;
442
443                 case IPV6_TLV_PADN:
444                         break;
445
446                 case IPV6_TLV_JUMBO:
447                         if (nh[off + 1] != 4 || (off & 3) != 2)
448                                 goto bad;
449                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
450                         if (pkt_len <= IPV6_MAXPLEN ||
451                             ipv6_hdr(skb)->payload_len)
452                                 goto bad;
453                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
454                                 goto bad;
455                         if (pskb_trim_rcsum(skb,
456                                             pkt_len + sizeof(struct ipv6hdr)))
457                                 goto bad;
458                         nh = skb_network_header(skb);
459                         break;
460                 default:
461                         if (optlen > len)
462                                 goto bad;
463                         break;
464                 }
465                 off += optlen;
466                 len -= optlen;
467         }
468         if (len == 0)
469                 return 0;
470 bad:
471         return -1;
472
473 }
474
475 /* Replicate the checks that IPv6 does on packet reception and pass the packet
476  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
477 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
478                                            struct sk_buff *skb,
479                                            const struct net_device *in,
480                                            const struct net_device *out,
481                                            int (*okfn)(struct sk_buff *))
482 {
483         struct ipv6hdr *hdr;
484         u32 pkt_len;
485
486         if (skb->len < sizeof(struct ipv6hdr))
487                 goto inhdr_error;
488
489         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
490                 goto inhdr_error;
491
492         hdr = ipv6_hdr(skb);
493
494         if (hdr->version != 6)
495                 goto inhdr_error;
496
497         pkt_len = ntohs(hdr->payload_len);
498
499         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
500                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
501                         goto inhdr_error;
502                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
503                         goto inhdr_error;
504         }
505         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
506                 goto inhdr_error;
507
508         nf_bridge_put(skb->nf_bridge);
509         if (!nf_bridge_alloc(skb))
510                 return NF_DROP;
511         if (!setup_pre_routing(skb))
512                 return NF_DROP;
513
514         NF_HOOK(PF_INET6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
515                 br_nf_pre_routing_finish_ipv6);
516
517         return NF_STOLEN;
518
519 inhdr_error:
520         return NF_DROP;
521 }
522
523 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
524  * Replicate the checks that IPv4 does on packet reception.
525  * Set skb->dev to the bridge device (i.e. parent of the
526  * receiving device) to make netfilter happy, the REDIRECT
527  * target in particular.  Save the original destination IP
528  * address to be able to detect DNAT afterwards. */
529 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
530                                       const struct net_device *in,
531                                       const struct net_device *out,
532                                       int (*okfn)(struct sk_buff *))
533 {
534         struct iphdr *iph;
535         __u32 len = nf_bridge_encap_header_len(skb);
536
537         if (unlikely(!pskb_may_pull(skb, len)))
538                 goto out;
539
540         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
541             IS_PPPOE_IPV6(skb)) {
542 #ifdef CONFIG_SYSCTL
543                 if (!brnf_call_ip6tables)
544                         return NF_ACCEPT;
545 #endif
546                 nf_bridge_pull_encap_header_rcsum(skb);
547                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
548         }
549 #ifdef CONFIG_SYSCTL
550         if (!brnf_call_iptables)
551                 return NF_ACCEPT;
552 #endif
553
554         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
555             !IS_PPPOE_IP(skb))
556                 return NF_ACCEPT;
557
558         nf_bridge_pull_encap_header_rcsum(skb);
559
560         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
561                 goto inhdr_error;
562
563         iph = ip_hdr(skb);
564         if (iph->ihl < 5 || iph->version != 4)
565                 goto inhdr_error;
566
567         if (!pskb_may_pull(skb, 4 * iph->ihl))
568                 goto inhdr_error;
569
570         iph = ip_hdr(skb);
571         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
572                 goto inhdr_error;
573
574         len = ntohs(iph->tot_len);
575         if (skb->len < len || len < 4 * iph->ihl)
576                 goto inhdr_error;
577
578         pskb_trim_rcsum(skb, len);
579
580         nf_bridge_put(skb->nf_bridge);
581         if (!nf_bridge_alloc(skb))
582                 return NF_DROP;
583         if (!setup_pre_routing(skb))
584                 return NF_DROP;
585         store_orig_dstaddr(skb);
586
587         NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
588                 br_nf_pre_routing_finish);
589
590         return NF_STOLEN;
591
592 inhdr_error:
593 //      IP_INC_STATS_BH(IpInHdrErrors);
594 out:
595         return NF_DROP;
596 }
597
598
599 /* PF_BRIDGE/LOCAL_IN ************************************************/
600 /* The packet is locally destined, which requires a real
601  * dst_entry, so detach the fake one.  On the way up, the
602  * packet would pass through PRE_ROUTING again (which already
603  * took place when the packet entered the bridge), but we
604  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
605  * prevent this from happening. */
606 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
607                                    const struct net_device *in,
608                                    const struct net_device *out,
609                                    int (*okfn)(struct sk_buff *))
610 {
611         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
612                 dst_release(skb->dst);
613                 skb->dst = NULL;
614         }
615
616         return NF_ACCEPT;
617 }
618
619 /* PF_BRIDGE/FORWARD *************************************************/
620 static int br_nf_forward_finish(struct sk_buff *skb)
621 {
622         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
623         struct net_device *in;
624
625         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
626                 in = nf_bridge->physindev;
627                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
628                         skb->pkt_type = PACKET_OTHERHOST;
629                         nf_bridge->mask ^= BRNF_PKT_TYPE;
630                 }
631         } else {
632                 in = *((struct net_device **)(skb->cb));
633         }
634         nf_bridge_push_encap_header(skb);
635         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
636                        skb->dev, br_forward_finish, 1);
637         return 0;
638 }
639
640 /* This is the 'purely bridged' case.  For IP, we pass the packet to
641  * netfilter with indev and outdev set to the bridge device,
642  * but we are still able to filter on the 'real' indev/outdev
643  * because of the physdev module. For ARP, indev and outdev are the
644  * bridge ports. */
645 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
646                                      const struct net_device *in,
647                                      const struct net_device *out,
648                                      int (*okfn)(struct sk_buff *))
649 {
650         struct nf_bridge_info *nf_bridge;
651         struct net_device *parent;
652         int pf;
653
654         if (!skb->nf_bridge)
655                 return NF_ACCEPT;
656
657         /* Need exclusive nf_bridge_info since we might have multiple
658          * different physoutdevs. */
659         if (!nf_bridge_unshare(skb))
660                 return NF_DROP;
661
662         parent = bridge_parent(out);
663         if (!parent)
664                 return NF_DROP;
665
666         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
667             IS_PPPOE_IP(skb))
668                 pf = PF_INET;
669         else
670                 pf = PF_INET6;
671
672         nf_bridge_pull_encap_header(skb);
673
674         nf_bridge = skb->nf_bridge;
675         if (skb->pkt_type == PACKET_OTHERHOST) {
676                 skb->pkt_type = PACKET_HOST;
677                 nf_bridge->mask |= BRNF_PKT_TYPE;
678         }
679
680         /* The physdev module checks on this */
681         nf_bridge->mask |= BRNF_BRIDGED;
682         nf_bridge->physoutdev = skb->dev;
683
684         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
685                 br_nf_forward_finish);
686
687         return NF_STOLEN;
688 }
689
690 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
691                                       const struct net_device *in,
692                                       const struct net_device *out,
693                                       int (*okfn)(struct sk_buff *))
694 {
695         struct net_device **d = (struct net_device **)(skb->cb);
696
697 #ifdef CONFIG_SYSCTL
698         if (!brnf_call_arptables)
699                 return NF_ACCEPT;
700 #endif
701
702         if (skb->protocol != htons(ETH_P_ARP)) {
703                 if (!IS_VLAN_ARP(skb))
704                         return NF_ACCEPT;
705                 nf_bridge_pull_encap_header(skb);
706         }
707
708         if (arp_hdr(skb)->ar_pln != 4) {
709                 if (IS_VLAN_ARP(skb))
710                         nf_bridge_push_encap_header(skb);
711                 return NF_ACCEPT;
712         }
713         *d = (struct net_device *)in;
714         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
715                 (struct net_device *)out, br_nf_forward_finish);
716
717         return NF_STOLEN;
718 }
719
720 /* PF_BRIDGE/LOCAL_OUT ***********************************************
721  *
722  * This function sees both locally originated IP packets and forwarded
723  * IP packets (in both cases the destination device is a bridge
724  * device). It also sees bridged-and-DNAT'ed packets.
725  *
726  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
727  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
728  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
729  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
730  * will be executed.
731  */
732 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff *skb,
733                                     const struct net_device *in,
734                                     const struct net_device *out,
735                                     int (*okfn)(struct sk_buff *))
736 {
737         struct net_device *realindev;
738         struct nf_bridge_info *nf_bridge;
739
740         if (!skb->nf_bridge)
741                 return NF_ACCEPT;
742
743         /* Need exclusive nf_bridge_info since we might have multiple
744          * different physoutdevs. */
745         if (!nf_bridge_unshare(skb))
746                 return NF_DROP;
747
748         nf_bridge = skb->nf_bridge;
749         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
750                 return NF_ACCEPT;
751
752         /* Bridged, take PF_BRIDGE/FORWARD.
753          * (see big note in front of br_nf_pre_routing_finish) */
754         nf_bridge->physoutdev = skb->dev;
755         realindev = nf_bridge->physindev;
756
757         if (nf_bridge->mask & BRNF_PKT_TYPE) {
758                 skb->pkt_type = PACKET_OTHERHOST;
759                 nf_bridge->mask ^= BRNF_PKT_TYPE;
760         }
761         nf_bridge_push_encap_header(skb);
762
763         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
764                 br_forward_finish);
765         return NF_STOLEN;
766 }
767
768 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
769 {
770         if (skb->protocol == htons(ETH_P_IP) &&
771             skb->len > skb->dev->mtu &&
772             !skb_is_gso(skb))
773                 return ip_fragment(skb, br_dev_queue_push_xmit);
774         else
775                 return br_dev_queue_push_xmit(skb);
776 }
777
778 /* PF_BRIDGE/POST_ROUTING ********************************************/
779 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
780                                        const struct net_device *in,
781                                        const struct net_device *out,
782                                        int (*okfn)(struct sk_buff *))
783 {
784         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
785         struct net_device *realoutdev = bridge_parent(skb->dev);
786         int pf;
787
788 #ifdef CONFIG_NETFILTER_DEBUG
789         /* Be very paranoid. This probably won't happen anymore, but let's
790          * keep the check just to be sure... */
791         if (skb_mac_header(skb) < skb->head ||
792             skb_mac_header(skb) + ETH_HLEN > skb->data) {
793                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
794                        "bad mac.raw pointer.\n");
795                 goto print_error;
796         }
797 #endif
798
799         if (!nf_bridge)
800                 return NF_ACCEPT;
801
802         if (!(nf_bridge->mask & (BRNF_BRIDGED | BRNF_BRIDGED_DNAT)))
803                 return NF_ACCEPT;
804
805         if (!realoutdev)
806                 return NF_DROP;
807
808         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
809             IS_PPPOE_IP(skb))
810                 pf = PF_INET;
811         else
812                 pf = PF_INET6;
813
814 #ifdef CONFIG_NETFILTER_DEBUG
815         if (skb->dst == NULL) {
816                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
817                 goto print_error;
818         }
819 #endif
820
821         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
822          * about the value of skb->pkt_type. */
823         if (skb->pkt_type == PACKET_OTHERHOST) {
824                 skb->pkt_type = PACKET_HOST;
825                 nf_bridge->mask |= BRNF_PKT_TYPE;
826         }
827
828         nf_bridge_pull_encap_header(skb);
829         nf_bridge_save_header(skb);
830
831 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
832         if (nf_bridge->netoutdev)
833                 realoutdev = nf_bridge->netoutdev;
834 #endif
835         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
836                 br_nf_dev_queue_xmit);
837
838         return NF_STOLEN;
839
840 #ifdef CONFIG_NETFILTER_DEBUG
841 print_error:
842         if (skb->dev != NULL) {
843                 printk("[%s]", skb->dev->name);
844                 if (realoutdev)
845                         printk("[%s]", realoutdev->name);
846         }
847         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
848                skb->data);
849         dump_stack();
850         return NF_ACCEPT;
851 #endif
852 }
853
854 /* IP/SABOTAGE *****************************************************/
855 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
856  * for the second time. */
857 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
858                                    const struct net_device *in,
859                                    const struct net_device *out,
860                                    int (*okfn)(struct sk_buff *))
861 {
862         if (skb->nf_bridge &&
863             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
864                 return NF_STOP;
865         }
866
867         return NF_ACCEPT;
868 }
869
870 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
871  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
872  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
873  * ip_refrag() can return NF_STOLEN. */
874 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
875         { .hook = br_nf_pre_routing,
876           .owner = THIS_MODULE,
877           .pf = PF_BRIDGE,
878           .hooknum = NF_BR_PRE_ROUTING,
879           .priority = NF_BR_PRI_BRNF, },
880         { .hook = br_nf_local_in,
881           .owner = THIS_MODULE,
882           .pf = PF_BRIDGE,
883           .hooknum = NF_BR_LOCAL_IN,
884           .priority = NF_BR_PRI_BRNF, },
885         { .hook = br_nf_forward_ip,
886           .owner = THIS_MODULE,
887           .pf = PF_BRIDGE,
888           .hooknum = NF_BR_FORWARD,
889           .priority = NF_BR_PRI_BRNF - 1, },
890         { .hook = br_nf_forward_arp,
891           .owner = THIS_MODULE,
892           .pf = PF_BRIDGE,
893           .hooknum = NF_BR_FORWARD,
894           .priority = NF_BR_PRI_BRNF, },
895         { .hook = br_nf_local_out,
896           .owner = THIS_MODULE,
897           .pf = PF_BRIDGE,
898           .hooknum = NF_BR_LOCAL_OUT,
899           .priority = NF_BR_PRI_FIRST, },
900         { .hook = br_nf_post_routing,
901           .owner = THIS_MODULE,
902           .pf = PF_BRIDGE,
903           .hooknum = NF_BR_POST_ROUTING,
904           .priority = NF_BR_PRI_LAST, },
905         { .hook = ip_sabotage_in,
906           .owner = THIS_MODULE,
907           .pf = PF_INET,
908           .hooknum = NF_INET_PRE_ROUTING,
909           .priority = NF_IP_PRI_FIRST, },
910         { .hook = ip_sabotage_in,
911           .owner = THIS_MODULE,
912           .pf = PF_INET6,
913           .hooknum = NF_INET_PRE_ROUTING,
914           .priority = NF_IP6_PRI_FIRST, },
915 };
916
917 #ifdef CONFIG_SYSCTL
918 static
919 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
920                             void __user * buffer, size_t * lenp, loff_t * ppos)
921 {
922         int ret;
923
924         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
925
926         if (write && *(int *)(ctl->data))
927                 *(int *)(ctl->data) = 1;
928         return ret;
929 }
930
931 static ctl_table brnf_table[] = {
932         {
933                 .procname       = "bridge-nf-call-arptables",
934                 .data           = &brnf_call_arptables,
935                 .maxlen         = sizeof(int),
936                 .mode           = 0644,
937                 .proc_handler   = &brnf_sysctl_call_tables,
938         },
939         {
940                 .procname       = "bridge-nf-call-iptables",
941                 .data           = &brnf_call_iptables,
942                 .maxlen         = sizeof(int),
943                 .mode           = 0644,
944                 .proc_handler   = &brnf_sysctl_call_tables,
945         },
946         {
947                 .procname       = "bridge-nf-call-ip6tables",
948                 .data           = &brnf_call_ip6tables,
949                 .maxlen         = sizeof(int),
950                 .mode           = 0644,
951                 .proc_handler   = &brnf_sysctl_call_tables,
952         },
953         {
954                 .procname       = "bridge-nf-filter-vlan-tagged",
955                 .data           = &brnf_filter_vlan_tagged,
956                 .maxlen         = sizeof(int),
957                 .mode           = 0644,
958                 .proc_handler   = &brnf_sysctl_call_tables,
959         },
960         {
961                 .procname       = "bridge-nf-filter-pppoe-tagged",
962                 .data           = &brnf_filter_pppoe_tagged,
963                 .maxlen         = sizeof(int),
964                 .mode           = 0644,
965                 .proc_handler   = &brnf_sysctl_call_tables,
966         },
967         { .ctl_name = 0 }
968 };
969
970 static struct ctl_path brnf_path[] = {
971         { .procname = "net", .ctl_name = CTL_NET, },
972         { .procname = "bridge", .ctl_name = NET_BRIDGE, },
973         { }
974 };
975 #endif
976
977 int __init br_netfilter_init(void)
978 {
979         int ret;
980
981         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
982         if (ret < 0)
983                 return ret;
984 #ifdef CONFIG_SYSCTL
985         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
986         if (brnf_sysctl_header == NULL) {
987                 printk(KERN_WARNING
988                        "br_netfilter: can't register to sysctl.\n");
989                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
990                 return -ENOMEM;
991         }
992 #endif
993         printk(KERN_NOTICE "Bridge firewalling registered\n");
994         return 0;
995 }
996
997 void br_netfilter_fini(void)
998 {
999         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1000 #ifdef CONFIG_SYSCTL
1001         unregister_sysctl_table(brnf_sysctl_header);
1002 #endif
1003 }