Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6] / 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 = 0;
62 static int brnf_filter_pppoe_tagged __read_mostly = 0;
63 #else
64 #define brnf_filter_vlan_tagged 0
65 #define brnf_filter_pppoe_tagged 0
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 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
105 {
106 }
107
108 static struct dst_ops fake_dst_ops = {
109         .family =               AF_INET,
110         .protocol =             cpu_to_be16(ETH_P_IP),
111         .update_pmtu =          fake_update_pmtu,
112         .entries =              ATOMIC_INIT(0),
113 };
114
115 /*
116  * Initialize bogus route table used to keep netfilter happy.
117  * Currently, we fill in the PMTU entry because netfilter
118  * refragmentation needs it, and the rt_flags entry because
119  * ipt_REJECT needs it.  Future netfilter modules might
120  * require us to fill additional fields.
121  */
122 void br_netfilter_rtable_init(struct net_bridge *br)
123 {
124         struct rtable *rt = &br->fake_rtable;
125
126         atomic_set(&rt->u.dst.__refcnt, 1);
127         rt->u.dst.dev = br->dev;
128         rt->u.dst.path = &rt->u.dst;
129         rt->u.dst.metrics[RTAX_MTU - 1] = 1500;
130         rt->u.dst.flags = DST_NOXFRM;
131         rt->u.dst.ops = &fake_dst_ops;
132 }
133
134 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
135 {
136         struct net_bridge_port *port = rcu_dereference(dev->br_port);
137
138         return port ? &port->br->fake_rtable : NULL;
139 }
140
141 static inline struct net_device *bridge_parent(const struct net_device *dev)
142 {
143         struct net_bridge_port *port = rcu_dereference(dev->br_port);
144
145         return port ? port->br->dev : NULL;
146 }
147
148 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
149 {
150         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
151         if (likely(skb->nf_bridge))
152                 atomic_set(&(skb->nf_bridge->use), 1);
153
154         return skb->nf_bridge;
155 }
156
157 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
158 {
159         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
160
161         if (atomic_read(&nf_bridge->use) > 1) {
162                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
163
164                 if (tmp) {
165                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
166                         atomic_set(&tmp->use, 1);
167                         nf_bridge_put(nf_bridge);
168                 }
169                 nf_bridge = tmp;
170         }
171         return nf_bridge;
172 }
173
174 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
175 {
176         unsigned int len = nf_bridge_encap_header_len(skb);
177
178         skb_push(skb, len);
179         skb->network_header -= len;
180 }
181
182 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
183 {
184         unsigned int len = nf_bridge_encap_header_len(skb);
185
186         skb_pull(skb, len);
187         skb->network_header += len;
188 }
189
190 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
191 {
192         unsigned int len = nf_bridge_encap_header_len(skb);
193
194         skb_pull_rcsum(skb, len);
195         skb->network_header += len;
196 }
197
198 static inline void nf_bridge_save_header(struct sk_buff *skb)
199 {
200         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
201
202         skb_copy_from_linear_data_offset(skb, -header_size,
203                                          skb->nf_bridge->data, header_size);
204 }
205
206 /*
207  * When forwarding bridge frames, we save a copy of the original
208  * header before processing.
209  */
210 int nf_bridge_copy_header(struct sk_buff *skb)
211 {
212         int err;
213         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
214
215         err = skb_cow_head(skb, header_size);
216         if (err)
217                 return err;
218
219         skb_copy_to_linear_data_offset(skb, -header_size,
220                                        skb->nf_bridge->data, header_size);
221         __skb_push(skb, nf_bridge_encap_header_len(skb));
222         return 0;
223 }
224
225 /* PF_BRIDGE/PRE_ROUTING *********************************************/
226 /* Undo the changes made for ip6tables PREROUTING and continue the
227  * bridge PRE_ROUTING hook. */
228 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
229 {
230         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
231
232         if (nf_bridge->mask & BRNF_PKT_TYPE) {
233                 skb->pkt_type = PACKET_OTHERHOST;
234                 nf_bridge->mask ^= BRNF_PKT_TYPE;
235         }
236         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
237
238         skb->rtable = bridge_parent_rtable(nf_bridge->physindev);
239         if (!skb->rtable) {
240                 kfree_skb(skb);
241                 return 0;
242         }
243         dst_hold(&skb->rtable->u.dst);
244
245         skb->dev = nf_bridge->physindev;
246         nf_bridge_push_encap_header(skb);
247         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
248                        br_handle_frame_finish, 1);
249
250         return 0;
251 }
252
253 static void __br_dnat_complain(void)
254 {
255         static unsigned long last_complaint;
256
257         if (jiffies - last_complaint >= 5 * HZ) {
258                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
259                        "forwarding to be enabled\n");
260                 last_complaint = jiffies;
261         }
262 }
263
264 /* This requires some explaining. If DNAT has taken place,
265  * we will need to fix up the destination Ethernet address,
266  * and this is a tricky process.
267  *
268  * There are two cases to consider:
269  * 1. The packet was DNAT'ed to a device in the same bridge
270  *    port group as it was received on. We can still bridge
271  *    the packet.
272  * 2. The packet was DNAT'ed to a different device, either
273  *    a non-bridged device or another bridge port group.
274  *    The packet will need to be routed.
275  *
276  * The correct way of distinguishing between these two cases is to
277  * call ip_route_input() and to look at skb->dst->dev, which is
278  * changed to the destination device if ip_route_input() succeeds.
279  *
280  * Let us first consider the case that ip_route_input() succeeds:
281  *
282  * If skb->dst->dev equals the logical bridge device the packet
283  * came in on, we can consider this bridging. The packet is passed
284  * through the neighbour output function to build a new destination
285  * MAC address, which will make the packet enter br_nf_local_out()
286  * not much later. In that function it is assured that the iptables
287  * FORWARD chain is traversed for the packet.
288  *
289  * Otherwise, the packet is considered to be routed and we just
290  * change the destination MAC address so that the packet will
291  * later be passed up to the IP stack to be routed. For a redirected
292  * packet, ip_route_input() will give back the localhost as output device,
293  * which differs from the bridge device.
294  *
295  * Let us now consider the case that ip_route_input() fails:
296  *
297  * This can be because the destination address is martian, in which case
298  * the packet will be dropped.
299  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
300  * will fail, while __ip_route_output_key() will return success. The source
301  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
302  * thinks we're handling a locally generated packet and won't care
303  * if IP forwarding is allowed. We send a warning message to the users's
304  * log telling her to put IP forwarding on.
305  *
306  * ip_route_input() will also fail if there is no route available.
307  * In that case we just drop the packet.
308  *
309  * --Lennert, 20020411
310  * --Bart, 20020416 (updated)
311  * --Bart, 20021007 (updated)
312  * --Bart, 20062711 (updated) */
313 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
314 {
315         if (skb->pkt_type == PACKET_OTHERHOST) {
316                 skb->pkt_type = PACKET_HOST;
317                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
318         }
319         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
320
321         skb->dev = bridge_parent(skb->dev);
322         if (skb->dev) {
323                 struct dst_entry *dst = skb->dst;
324
325                 nf_bridge_pull_encap_header(skb);
326
327                 if (dst->hh)
328                         return neigh_hh_output(dst->hh, skb);
329                 else if (dst->neighbour)
330                         return dst->neighbour->output(skb);
331         }
332         kfree_skb(skb);
333         return 0;
334 }
335
336 static int br_nf_pre_routing_finish(struct sk_buff *skb)
337 {
338         struct net_device *dev = skb->dev;
339         struct iphdr *iph = ip_hdr(skb);
340         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
341         int err;
342
343         if (nf_bridge->mask & BRNF_PKT_TYPE) {
344                 skb->pkt_type = PACKET_OTHERHOST;
345                 nf_bridge->mask ^= BRNF_PKT_TYPE;
346         }
347         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
348         if (dnat_took_place(skb)) {
349                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
350                         struct rtable *rt;
351                         struct flowi fl = {
352                                 .nl_u = {
353                                         .ip4_u = {
354                                                  .daddr = iph->daddr,
355                                                  .saddr = 0,
356                                                  .tos = RT_TOS(iph->tos) },
357                                 },
358                                 .proto = 0,
359                         };
360                         struct in_device *in_dev = in_dev_get(dev);
361
362                         /* If err equals -EHOSTUNREACH the error is due to a
363                          * martian destination or due to the fact that
364                          * forwarding is disabled. For most martian packets,
365                          * ip_route_output_key() will fail. It won't fail for 2 types of
366                          * martian destinations: loopback destinations and destination
367                          * 0.0.0.0. In both cases the packet will be dropped because the
368                          * destination is the loopback device and not the bridge. */
369                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
370                                 goto free_skb;
371
372                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
373                                 /* - Bridged-and-DNAT'ed traffic doesn't
374                                  *   require ip_forwarding. */
375                                 if (((struct dst_entry *)rt)->dev == dev) {
376                                         skb->dst = (struct dst_entry *)rt;
377                                         goto bridged_dnat;
378                                 }
379                                 /* we are sure that forwarding is disabled, so printing
380                                  * this message is no problem. Note that the packet could
381                                  * still have a martian destination address, in which case
382                                  * the packet could be dropped even if forwarding were enabled */
383                                 __br_dnat_complain();
384                                 dst_release((struct dst_entry *)rt);
385                         }
386 free_skb:
387                         kfree_skb(skb);
388                         return 0;
389                 } else {
390                         if (skb->dst->dev == dev) {
391 bridged_dnat:
392                                 /* Tell br_nf_local_out this is a
393                                  * bridged frame */
394                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
395                                 skb->dev = nf_bridge->physindev;
396                                 nf_bridge_push_encap_header(skb);
397                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
398                                                skb, skb->dev, NULL,
399                                                br_nf_pre_routing_finish_bridge,
400                                                1);
401                                 return 0;
402                         }
403                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
404                         skb->pkt_type = PACKET_HOST;
405                 }
406         } else {
407                 skb->rtable = bridge_parent_rtable(nf_bridge->physindev);
408                 if (!skb->rtable) {
409                         kfree_skb(skb);
410                         return 0;
411                 }
412                 dst_hold(&skb->rtable->u.dst);
413         }
414
415         skb->dev = nf_bridge->physindev;
416         nf_bridge_push_encap_header(skb);
417         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
418                        br_handle_frame_finish, 1);
419
420         return 0;
421 }
422
423 /* Some common code for IPv4/IPv6 */
424 static struct net_device *setup_pre_routing(struct sk_buff *skb)
425 {
426         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
427
428         if (skb->pkt_type == PACKET_OTHERHOST) {
429                 skb->pkt_type = PACKET_HOST;
430                 nf_bridge->mask |= BRNF_PKT_TYPE;
431         }
432
433         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
434         nf_bridge->physindev = skb->dev;
435         skb->dev = bridge_parent(skb->dev);
436
437         return skb->dev;
438 }
439
440 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
441 static int check_hbh_len(struct sk_buff *skb)
442 {
443         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
444         u32 pkt_len;
445         const unsigned char *nh = skb_network_header(skb);
446         int off = raw - nh;
447         int len = (raw[1] + 1) << 3;
448
449         if ((raw + len) - skb->data > skb_headlen(skb))
450                 goto bad;
451
452         off += 2;
453         len -= 2;
454
455         while (len > 0) {
456                 int optlen = nh[off + 1] + 2;
457
458                 switch (nh[off]) {
459                 case IPV6_TLV_PAD0:
460                         optlen = 1;
461                         break;
462
463                 case IPV6_TLV_PADN:
464                         break;
465
466                 case IPV6_TLV_JUMBO:
467                         if (nh[off + 1] != 4 || (off & 3) != 2)
468                                 goto bad;
469                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
470                         if (pkt_len <= IPV6_MAXPLEN ||
471                             ipv6_hdr(skb)->payload_len)
472                                 goto bad;
473                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
474                                 goto bad;
475                         if (pskb_trim_rcsum(skb,
476                                             pkt_len + sizeof(struct ipv6hdr)))
477                                 goto bad;
478                         nh = skb_network_header(skb);
479                         break;
480                 default:
481                         if (optlen > len)
482                                 goto bad;
483                         break;
484                 }
485                 off += optlen;
486                 len -= optlen;
487         }
488         if (len == 0)
489                 return 0;
490 bad:
491         return -1;
492
493 }
494
495 /* Replicate the checks that IPv6 does on packet reception and pass the packet
496  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
497 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
498                                            struct sk_buff *skb,
499                                            const struct net_device *in,
500                                            const struct net_device *out,
501                                            int (*okfn)(struct sk_buff *))
502 {
503         struct ipv6hdr *hdr;
504         u32 pkt_len;
505
506         if (skb->len < sizeof(struct ipv6hdr))
507                 goto inhdr_error;
508
509         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
510                 goto inhdr_error;
511
512         hdr = ipv6_hdr(skb);
513
514         if (hdr->version != 6)
515                 goto inhdr_error;
516
517         pkt_len = ntohs(hdr->payload_len);
518
519         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
520                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
521                         goto inhdr_error;
522                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
523                         goto inhdr_error;
524         }
525         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
526                 goto inhdr_error;
527
528         nf_bridge_put(skb->nf_bridge);
529         if (!nf_bridge_alloc(skb))
530                 return NF_DROP;
531         if (!setup_pre_routing(skb))
532                 return NF_DROP;
533
534         NF_HOOK(PF_INET6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
535                 br_nf_pre_routing_finish_ipv6);
536
537         return NF_STOLEN;
538
539 inhdr_error:
540         return NF_DROP;
541 }
542
543 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
544  * Replicate the checks that IPv4 does on packet reception.
545  * Set skb->dev to the bridge device (i.e. parent of the
546  * receiving device) to make netfilter happy, the REDIRECT
547  * target in particular.  Save the original destination IP
548  * address to be able to detect DNAT afterwards. */
549 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
550                                       const struct net_device *in,
551                                       const struct net_device *out,
552                                       int (*okfn)(struct sk_buff *))
553 {
554         struct iphdr *iph;
555         __u32 len = nf_bridge_encap_header_len(skb);
556
557         if (unlikely(!pskb_may_pull(skb, len)))
558                 goto out;
559
560         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
561             IS_PPPOE_IPV6(skb)) {
562 #ifdef CONFIG_SYSCTL
563                 if (!brnf_call_ip6tables)
564                         return NF_ACCEPT;
565 #endif
566                 nf_bridge_pull_encap_header_rcsum(skb);
567                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
568         }
569 #ifdef CONFIG_SYSCTL
570         if (!brnf_call_iptables)
571                 return NF_ACCEPT;
572 #endif
573
574         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
575             !IS_PPPOE_IP(skb))
576                 return NF_ACCEPT;
577
578         nf_bridge_pull_encap_header_rcsum(skb);
579
580         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
581                 goto inhdr_error;
582
583         iph = ip_hdr(skb);
584         if (iph->ihl < 5 || iph->version != 4)
585                 goto inhdr_error;
586
587         if (!pskb_may_pull(skb, 4 * iph->ihl))
588                 goto inhdr_error;
589
590         iph = ip_hdr(skb);
591         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
592                 goto inhdr_error;
593
594         len = ntohs(iph->tot_len);
595         if (skb->len < len || len < 4 * iph->ihl)
596                 goto inhdr_error;
597
598         pskb_trim_rcsum(skb, len);
599
600         nf_bridge_put(skb->nf_bridge);
601         if (!nf_bridge_alloc(skb))
602                 return NF_DROP;
603         if (!setup_pre_routing(skb))
604                 return NF_DROP;
605         store_orig_dstaddr(skb);
606
607         NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
608                 br_nf_pre_routing_finish);
609
610         return NF_STOLEN;
611
612 inhdr_error:
613 //      IP_INC_STATS_BH(IpInHdrErrors);
614 out:
615         return NF_DROP;
616 }
617
618
619 /* PF_BRIDGE/LOCAL_IN ************************************************/
620 /* The packet is locally destined, which requires a real
621  * dst_entry, so detach the fake one.  On the way up, the
622  * packet would pass through PRE_ROUTING again (which already
623  * took place when the packet entered the bridge), but we
624  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
625  * prevent this from happening. */
626 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
627                                    const struct net_device *in,
628                                    const struct net_device *out,
629                                    int (*okfn)(struct sk_buff *))
630 {
631         if (skb->rtable && skb->rtable == bridge_parent_rtable(in)) {
632                 dst_release(&skb->rtable->u.dst);
633                 skb->rtable = NULL;
634         }
635
636         return NF_ACCEPT;
637 }
638
639 /* PF_BRIDGE/FORWARD *************************************************/
640 static int br_nf_forward_finish(struct sk_buff *skb)
641 {
642         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
643         struct net_device *in;
644
645         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
646                 in = nf_bridge->physindev;
647                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
648                         skb->pkt_type = PACKET_OTHERHOST;
649                         nf_bridge->mask ^= BRNF_PKT_TYPE;
650                 }
651         } else {
652                 in = *((struct net_device **)(skb->cb));
653         }
654         nf_bridge_push_encap_header(skb);
655         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
656                        skb->dev, br_forward_finish, 1);
657         return 0;
658 }
659
660 /* This is the 'purely bridged' case.  For IP, we pass the packet to
661  * netfilter with indev and outdev set to the bridge device,
662  * but we are still able to filter on the 'real' indev/outdev
663  * because of the physdev module. For ARP, indev and outdev are the
664  * bridge ports. */
665 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
666                                      const struct net_device *in,
667                                      const struct net_device *out,
668                                      int (*okfn)(struct sk_buff *))
669 {
670         struct nf_bridge_info *nf_bridge;
671         struct net_device *parent;
672         u_int8_t pf;
673
674         if (!skb->nf_bridge)
675                 return NF_ACCEPT;
676
677         /* Need exclusive nf_bridge_info since we might have multiple
678          * different physoutdevs. */
679         if (!nf_bridge_unshare(skb))
680                 return NF_DROP;
681
682         parent = bridge_parent(out);
683         if (!parent)
684                 return NF_DROP;
685
686         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
687             IS_PPPOE_IP(skb))
688                 pf = PF_INET;
689         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
690                  IS_PPPOE_IPV6(skb))
691                 pf = PF_INET6;
692         else
693                 return NF_ACCEPT;
694
695         nf_bridge_pull_encap_header(skb);
696
697         nf_bridge = skb->nf_bridge;
698         if (skb->pkt_type == PACKET_OTHERHOST) {
699                 skb->pkt_type = PACKET_HOST;
700                 nf_bridge->mask |= BRNF_PKT_TYPE;
701         }
702
703         /* The physdev module checks on this */
704         nf_bridge->mask |= BRNF_BRIDGED;
705         nf_bridge->physoutdev = skb->dev;
706
707         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
708                 br_nf_forward_finish);
709
710         return NF_STOLEN;
711 }
712
713 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
714                                       const struct net_device *in,
715                                       const struct net_device *out,
716                                       int (*okfn)(struct sk_buff *))
717 {
718         struct net_device **d = (struct net_device **)(skb->cb);
719
720 #ifdef CONFIG_SYSCTL
721         if (!brnf_call_arptables)
722                 return NF_ACCEPT;
723 #endif
724
725         if (skb->protocol != htons(ETH_P_ARP)) {
726                 if (!IS_VLAN_ARP(skb))
727                         return NF_ACCEPT;
728                 nf_bridge_pull_encap_header(skb);
729         }
730
731         if (arp_hdr(skb)->ar_pln != 4) {
732                 if (IS_VLAN_ARP(skb))
733                         nf_bridge_push_encap_header(skb);
734                 return NF_ACCEPT;
735         }
736         *d = (struct net_device *)in;
737         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
738                 (struct net_device *)out, br_nf_forward_finish);
739
740         return NF_STOLEN;
741 }
742
743 /* PF_BRIDGE/LOCAL_OUT ***********************************************
744  *
745  * This function sees both locally originated IP packets and forwarded
746  * IP packets (in both cases the destination device is a bridge
747  * device). It also sees bridged-and-DNAT'ed packets.
748  *
749  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
750  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
751  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
752  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
753  * will be executed.
754  */
755 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff *skb,
756                                     const struct net_device *in,
757                                     const struct net_device *out,
758                                     int (*okfn)(struct sk_buff *))
759 {
760         struct net_device *realindev;
761         struct nf_bridge_info *nf_bridge;
762
763         if (!skb->nf_bridge)
764                 return NF_ACCEPT;
765
766         /* Need exclusive nf_bridge_info since we might have multiple
767          * different physoutdevs. */
768         if (!nf_bridge_unshare(skb))
769                 return NF_DROP;
770
771         nf_bridge = skb->nf_bridge;
772         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
773                 return NF_ACCEPT;
774
775         /* Bridged, take PF_BRIDGE/FORWARD.
776          * (see big note in front of br_nf_pre_routing_finish) */
777         nf_bridge->physoutdev = skb->dev;
778         realindev = nf_bridge->physindev;
779
780         if (nf_bridge->mask & BRNF_PKT_TYPE) {
781                 skb->pkt_type = PACKET_OTHERHOST;
782                 nf_bridge->mask ^= BRNF_PKT_TYPE;
783         }
784         nf_bridge_push_encap_header(skb);
785
786         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
787                 br_forward_finish);
788         return NF_STOLEN;
789 }
790
791 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
792 {
793         if (skb->protocol == htons(ETH_P_IP) &&
794             skb->len > skb->dev->mtu &&
795             !skb_is_gso(skb))
796                 return ip_fragment(skb, br_dev_queue_push_xmit);
797         else
798                 return br_dev_queue_push_xmit(skb);
799 }
800
801 /* PF_BRIDGE/POST_ROUTING ********************************************/
802 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
803                                        const struct net_device *in,
804                                        const struct net_device *out,
805                                        int (*okfn)(struct sk_buff *))
806 {
807         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
808         struct net_device *realoutdev = bridge_parent(skb->dev);
809         u_int8_t pf;
810
811 #ifdef CONFIG_NETFILTER_DEBUG
812         /* Be very paranoid. This probably won't happen anymore, but let's
813          * keep the check just to be sure... */
814         if (skb_mac_header(skb) < skb->head ||
815             skb_mac_header(skb) + ETH_HLEN > skb->data) {
816                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
817                        "bad mac.raw pointer.\n");
818                 goto print_error;
819         }
820 #endif
821
822         if (!nf_bridge)
823                 return NF_ACCEPT;
824
825         if (!(nf_bridge->mask & (BRNF_BRIDGED | BRNF_BRIDGED_DNAT)))
826                 return NF_ACCEPT;
827
828         if (!realoutdev)
829                 return NF_DROP;
830
831         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
832             IS_PPPOE_IP(skb))
833                 pf = PF_INET;
834         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
835                  IS_PPPOE_IPV6(skb))
836                 pf = PF_INET6;
837         else
838                 return NF_ACCEPT;
839
840 #ifdef CONFIG_NETFILTER_DEBUG
841         if (skb->dst == NULL) {
842                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
843                 goto print_error;
844         }
845 #endif
846
847         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
848          * about the value of skb->pkt_type. */
849         if (skb->pkt_type == PACKET_OTHERHOST) {
850                 skb->pkt_type = PACKET_HOST;
851                 nf_bridge->mask |= BRNF_PKT_TYPE;
852         }
853
854         nf_bridge_pull_encap_header(skb);
855         nf_bridge_save_header(skb);
856
857         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
858                 br_nf_dev_queue_xmit);
859
860         return NF_STOLEN;
861
862 #ifdef CONFIG_NETFILTER_DEBUG
863 print_error:
864         if (skb->dev != NULL) {
865                 printk("[%s]", skb->dev->name);
866                 if (realoutdev)
867                         printk("[%s]", realoutdev->name);
868         }
869         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
870                skb->data);
871         dump_stack();
872         return NF_ACCEPT;
873 #endif
874 }
875
876 /* IP/SABOTAGE *****************************************************/
877 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
878  * for the second time. */
879 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
880                                    const struct net_device *in,
881                                    const struct net_device *out,
882                                    int (*okfn)(struct sk_buff *))
883 {
884         if (skb->nf_bridge &&
885             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
886                 return NF_STOP;
887         }
888
889         return NF_ACCEPT;
890 }
891
892 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
893  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
894  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
895  * ip_refrag() can return NF_STOLEN. */
896 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
897         { .hook = br_nf_pre_routing,
898           .owner = THIS_MODULE,
899           .pf = PF_BRIDGE,
900           .hooknum = NF_BR_PRE_ROUTING,
901           .priority = NF_BR_PRI_BRNF, },
902         { .hook = br_nf_local_in,
903           .owner = THIS_MODULE,
904           .pf = PF_BRIDGE,
905           .hooknum = NF_BR_LOCAL_IN,
906           .priority = NF_BR_PRI_BRNF, },
907         { .hook = br_nf_forward_ip,
908           .owner = THIS_MODULE,
909           .pf = PF_BRIDGE,
910           .hooknum = NF_BR_FORWARD,
911           .priority = NF_BR_PRI_BRNF - 1, },
912         { .hook = br_nf_forward_arp,
913           .owner = THIS_MODULE,
914           .pf = PF_BRIDGE,
915           .hooknum = NF_BR_FORWARD,
916           .priority = NF_BR_PRI_BRNF, },
917         { .hook = br_nf_local_out,
918           .owner = THIS_MODULE,
919           .pf = PF_BRIDGE,
920           .hooknum = NF_BR_LOCAL_OUT,
921           .priority = NF_BR_PRI_FIRST, },
922         { .hook = br_nf_post_routing,
923           .owner = THIS_MODULE,
924           .pf = PF_BRIDGE,
925           .hooknum = NF_BR_POST_ROUTING,
926           .priority = NF_BR_PRI_LAST, },
927         { .hook = ip_sabotage_in,
928           .owner = THIS_MODULE,
929           .pf = PF_INET,
930           .hooknum = NF_INET_PRE_ROUTING,
931           .priority = NF_IP_PRI_FIRST, },
932         { .hook = ip_sabotage_in,
933           .owner = THIS_MODULE,
934           .pf = PF_INET6,
935           .hooknum = NF_INET_PRE_ROUTING,
936           .priority = NF_IP6_PRI_FIRST, },
937 };
938
939 #ifdef CONFIG_SYSCTL
940 static
941 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
942                             void __user * buffer, size_t * lenp, loff_t * ppos)
943 {
944         int ret;
945
946         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
947
948         if (write && *(int *)(ctl->data))
949                 *(int *)(ctl->data) = 1;
950         return ret;
951 }
952
953 static ctl_table brnf_table[] = {
954         {
955                 .procname       = "bridge-nf-call-arptables",
956                 .data           = &brnf_call_arptables,
957                 .maxlen         = sizeof(int),
958                 .mode           = 0644,
959                 .proc_handler   = brnf_sysctl_call_tables,
960         },
961         {
962                 .procname       = "bridge-nf-call-iptables",
963                 .data           = &brnf_call_iptables,
964                 .maxlen         = sizeof(int),
965                 .mode           = 0644,
966                 .proc_handler   = brnf_sysctl_call_tables,
967         },
968         {
969                 .procname       = "bridge-nf-call-ip6tables",
970                 .data           = &brnf_call_ip6tables,
971                 .maxlen         = sizeof(int),
972                 .mode           = 0644,
973                 .proc_handler   = brnf_sysctl_call_tables,
974         },
975         {
976                 .procname       = "bridge-nf-filter-vlan-tagged",
977                 .data           = &brnf_filter_vlan_tagged,
978                 .maxlen         = sizeof(int),
979                 .mode           = 0644,
980                 .proc_handler   = brnf_sysctl_call_tables,
981         },
982         {
983                 .procname       = "bridge-nf-filter-pppoe-tagged",
984                 .data           = &brnf_filter_pppoe_tagged,
985                 .maxlen         = sizeof(int),
986                 .mode           = 0644,
987                 .proc_handler   = brnf_sysctl_call_tables,
988         },
989         { .ctl_name = 0 }
990 };
991
992 static struct ctl_path brnf_path[] = {
993         { .procname = "net", .ctl_name = CTL_NET, },
994         { .procname = "bridge", .ctl_name = NET_BRIDGE, },
995         { }
996 };
997 #endif
998
999 int __init br_netfilter_init(void)
1000 {
1001         int ret;
1002
1003         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1004         if (ret < 0)
1005                 return ret;
1006 #ifdef CONFIG_SYSCTL
1007         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1008         if (brnf_sysctl_header == NULL) {
1009                 printk(KERN_WARNING
1010                        "br_netfilter: can't register to sysctl.\n");
1011                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1012                 return -ENOMEM;
1013         }
1014 #endif
1015         printk(KERN_NOTICE "Bridge firewalling registered\n");
1016         return 0;
1017 }
1018
1019 void br_netfilter_fini(void)
1020 {
1021         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1022 #ifdef CONFIG_SYSCTL
1023         unregister_sysctl_table(brnf_sysctl_header);
1024 #endif
1025 }