dd6f538ba0b0b4cb56238bd60c7ba36476b9f522
[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  * 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(skb);
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         struct rtable *rt;
326         int err;
327
328         if (nf_bridge->mask & BRNF_PKT_TYPE) {
329                 skb->pkt_type = PACKET_OTHERHOST;
330                 nf_bridge->mask ^= BRNF_PKT_TYPE;
331         }
332         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
333         if (dnat_took_place(skb)) {
334                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
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_rcu(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(dev_net(dev), &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_set(skb, (struct dst_entry *)rt);
361                                         goto bridged_dnat;
362                                 }
363                                 dst_release((struct dst_entry *)rt);
364                         }
365 free_skb:
366                         kfree_skb(skb);
367                         return 0;
368                 } else {
369                         if (skb_dst(skb)->dev == dev) {
370 bridged_dnat:
371                                 /* Tell br_nf_local_out this is a
372                                  * bridged frame */
373                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
374                                 skb->dev = nf_bridge->physindev;
375                                 nf_bridge_push_encap_header(skb);
376                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
377                                                NF_BR_PRE_ROUTING,
378                                                skb, skb->dev, NULL,
379                                                br_nf_pre_routing_finish_bridge,
380                                                1);
381                                 return 0;
382                         }
383                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
384                         skb->pkt_type = PACKET_HOST;
385                 }
386         } else {
387                 rt = bridge_parent_rtable(nf_bridge->physindev);
388                 if (!rt) {
389                         kfree_skb(skb);
390                         return 0;
391                 }
392                 dst_hold(&rt->u.dst);
393                 skb_dst_set(skb, &rt->u.dst);
394         }
395
396         skb->dev = nf_bridge->physindev;
397         nf_bridge_push_encap_header(skb);
398         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
399                        br_handle_frame_finish, 1);
400
401         return 0;
402 }
403
404 /* Some common code for IPv4/IPv6 */
405 static struct net_device *setup_pre_routing(struct sk_buff *skb)
406 {
407         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
408
409         if (skb->pkt_type == PACKET_OTHERHOST) {
410                 skb->pkt_type = PACKET_HOST;
411                 nf_bridge->mask |= BRNF_PKT_TYPE;
412         }
413
414         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
415         nf_bridge->physindev = skb->dev;
416         skb->dev = bridge_parent(skb->dev);
417
418         return skb->dev;
419 }
420
421 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
422 static int check_hbh_len(struct sk_buff *skb)
423 {
424         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
425         u32 pkt_len;
426         const unsigned char *nh = skb_network_header(skb);
427         int off = raw - nh;
428         int len = (raw[1] + 1) << 3;
429
430         if ((raw + len) - skb->data > skb_headlen(skb))
431                 goto bad;
432
433         off += 2;
434         len -= 2;
435
436         while (len > 0) {
437                 int optlen = nh[off + 1] + 2;
438
439                 switch (nh[off]) {
440                 case IPV6_TLV_PAD0:
441                         optlen = 1;
442                         break;
443
444                 case IPV6_TLV_PADN:
445                         break;
446
447                 case IPV6_TLV_JUMBO:
448                         if (nh[off + 1] != 4 || (off & 3) != 2)
449                                 goto bad;
450                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
451                         if (pkt_len <= IPV6_MAXPLEN ||
452                             ipv6_hdr(skb)->payload_len)
453                                 goto bad;
454                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
455                                 goto bad;
456                         if (pskb_trim_rcsum(skb,
457                                             pkt_len + sizeof(struct ipv6hdr)))
458                                 goto bad;
459                         nh = skb_network_header(skb);
460                         break;
461                 default:
462                         if (optlen > len)
463                                 goto bad;
464                         break;
465                 }
466                 off += optlen;
467                 len -= optlen;
468         }
469         if (len == 0)
470                 return 0;
471 bad:
472         return -1;
473
474 }
475
476 /* Replicate the checks that IPv6 does on packet reception and pass the packet
477  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
478 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
479                                            struct sk_buff *skb,
480                                            const struct net_device *in,
481                                            const struct net_device *out,
482                                            int (*okfn)(struct sk_buff *))
483 {
484         struct ipv6hdr *hdr;
485         u32 pkt_len;
486
487         if (skb->len < sizeof(struct ipv6hdr))
488                 goto inhdr_error;
489
490         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
491                 goto inhdr_error;
492
493         hdr = ipv6_hdr(skb);
494
495         if (hdr->version != 6)
496                 goto inhdr_error;
497
498         pkt_len = ntohs(hdr->payload_len);
499
500         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
501                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
502                         goto inhdr_error;
503                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
504                         goto inhdr_error;
505         }
506         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
507                 goto inhdr_error;
508
509         nf_bridge_put(skb->nf_bridge);
510         if (!nf_bridge_alloc(skb))
511                 return NF_DROP;
512         if (!setup_pre_routing(skb))
513                 return NF_DROP;
514
515         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
516                 br_nf_pre_routing_finish_ipv6);
517
518         return NF_STOLEN;
519
520 inhdr_error:
521         return NF_DROP;
522 }
523
524 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
525  * Replicate the checks that IPv4 does on packet reception.
526  * Set skb->dev to the bridge device (i.e. parent of the
527  * receiving device) to make netfilter happy, the REDIRECT
528  * target in particular.  Save the original destination IP
529  * address to be able to detect DNAT afterwards. */
530 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
531                                       const struct net_device *in,
532                                       const struct net_device *out,
533                                       int (*okfn)(struct sk_buff *))
534 {
535         struct iphdr *iph;
536         __u32 len = nf_bridge_encap_header_len(skb);
537
538         if (unlikely(!pskb_may_pull(skb, len)))
539                 goto out;
540
541         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
542             IS_PPPOE_IPV6(skb)) {
543 #ifdef CONFIG_SYSCTL
544                 if (!brnf_call_ip6tables)
545                         return NF_ACCEPT;
546 #endif
547                 nf_bridge_pull_encap_header_rcsum(skb);
548                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
549         }
550 #ifdef CONFIG_SYSCTL
551         if (!brnf_call_iptables)
552                 return NF_ACCEPT;
553 #endif
554
555         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
556             !IS_PPPOE_IP(skb))
557                 return NF_ACCEPT;
558
559         nf_bridge_pull_encap_header_rcsum(skb);
560
561         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
562                 goto inhdr_error;
563
564         iph = ip_hdr(skb);
565         if (iph->ihl < 5 || iph->version != 4)
566                 goto inhdr_error;
567
568         if (!pskb_may_pull(skb, 4 * iph->ihl))
569                 goto inhdr_error;
570
571         iph = ip_hdr(skb);
572         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
573                 goto inhdr_error;
574
575         len = ntohs(iph->tot_len);
576         if (skb->len < len || len < 4 * iph->ihl)
577                 goto inhdr_error;
578
579         pskb_trim_rcsum(skb, len);
580
581         nf_bridge_put(skb->nf_bridge);
582         if (!nf_bridge_alloc(skb))
583                 return NF_DROP;
584         if (!setup_pre_routing(skb))
585                 return NF_DROP;
586         store_orig_dstaddr(skb);
587
588         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
589                 br_nf_pre_routing_finish);
590
591         return NF_STOLEN;
592
593 inhdr_error:
594 //      IP_INC_STATS_BH(IpInHdrErrors);
595 out:
596         return NF_DROP;
597 }
598
599
600 /* PF_BRIDGE/LOCAL_IN ************************************************/
601 /* The packet is locally destined, which requires a real
602  * dst_entry, so detach the fake one.  On the way up, the
603  * packet would pass through PRE_ROUTING again (which already
604  * took place when the packet entered the bridge), but we
605  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
606  * prevent this from happening. */
607 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
608                                    const struct net_device *in,
609                                    const struct net_device *out,
610                                    int (*okfn)(struct sk_buff *))
611 {
612         struct rtable *rt = skb_rtable(skb);
613
614         if (rt && rt == bridge_parent_rtable(in))
615                 skb_dst_drop(skb);
616
617         return NF_ACCEPT;
618 }
619
620 /* PF_BRIDGE/FORWARD *************************************************/
621 static int br_nf_forward_finish(struct sk_buff *skb)
622 {
623         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
624         struct net_device *in;
625
626         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
627                 in = nf_bridge->physindev;
628                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
629                         skb->pkt_type = PACKET_OTHERHOST;
630                         nf_bridge->mask ^= BRNF_PKT_TYPE;
631                 }
632         } else {
633                 in = *((struct net_device **)(skb->cb));
634         }
635         nf_bridge_push_encap_header(skb);
636         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
637                        skb->dev, br_forward_finish, 1);
638         return 0;
639 }
640
641 /* This is the 'purely bridged' case.  For IP, we pass the packet to
642  * netfilter with indev and outdev set to the bridge device,
643  * but we are still able to filter on the 'real' indev/outdev
644  * because of the physdev module. For ARP, indev and outdev are the
645  * bridge ports. */
646 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
647                                      const struct net_device *in,
648                                      const struct net_device *out,
649                                      int (*okfn)(struct sk_buff *))
650 {
651         struct nf_bridge_info *nf_bridge;
652         struct net_device *parent;
653         u_int8_t pf;
654
655         if (!skb->nf_bridge)
656                 return NF_ACCEPT;
657
658         /* Need exclusive nf_bridge_info since we might have multiple
659          * different physoutdevs. */
660         if (!nf_bridge_unshare(skb))
661                 return NF_DROP;
662
663         parent = bridge_parent(out);
664         if (!parent)
665                 return NF_DROP;
666
667         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
668             IS_PPPOE_IP(skb))
669                 pf = PF_INET;
670         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
671                  IS_PPPOE_IPV6(skb))
672                 pf = PF_INET6;
673         else
674                 return NF_ACCEPT;
675
676         nf_bridge_pull_encap_header(skb);
677
678         nf_bridge = skb->nf_bridge;
679         if (skb->pkt_type == PACKET_OTHERHOST) {
680                 skb->pkt_type = PACKET_HOST;
681                 nf_bridge->mask |= BRNF_PKT_TYPE;
682         }
683
684         /* The physdev module checks on this */
685         nf_bridge->mask |= BRNF_BRIDGED;
686         nf_bridge->physoutdev = skb->dev;
687
688         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
689                 br_nf_forward_finish);
690
691         return NF_STOLEN;
692 }
693
694 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
695                                       const struct net_device *in,
696                                       const struct net_device *out,
697                                       int (*okfn)(struct sk_buff *))
698 {
699         struct net_device **d = (struct net_device **)(skb->cb);
700
701 #ifdef CONFIG_SYSCTL
702         if (!brnf_call_arptables)
703                 return NF_ACCEPT;
704 #endif
705
706         if (skb->protocol != htons(ETH_P_ARP)) {
707                 if (!IS_VLAN_ARP(skb))
708                         return NF_ACCEPT;
709                 nf_bridge_pull_encap_header(skb);
710         }
711
712         if (arp_hdr(skb)->ar_pln != 4) {
713                 if (IS_VLAN_ARP(skb))
714                         nf_bridge_push_encap_header(skb);
715                 return NF_ACCEPT;
716         }
717         *d = (struct net_device *)in;
718         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
719                 (struct net_device *)out, br_nf_forward_finish);
720
721         return NF_STOLEN;
722 }
723
724 /* PF_BRIDGE/LOCAL_OUT ***********************************************
725  *
726  * This function sees both locally originated IP packets and forwarded
727  * IP packets (in both cases the destination device is a bridge
728  * device). It also sees bridged-and-DNAT'ed packets.
729  *
730  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
731  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
732  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
733  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
734  * will be executed.
735  */
736 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff *skb,
737                                     const struct net_device *in,
738                                     const struct net_device *out,
739                                     int (*okfn)(struct sk_buff *))
740 {
741         struct net_device *realindev;
742         struct nf_bridge_info *nf_bridge;
743
744         if (!skb->nf_bridge)
745                 return NF_ACCEPT;
746
747         /* Need exclusive nf_bridge_info since we might have multiple
748          * different physoutdevs. */
749         if (!nf_bridge_unshare(skb))
750                 return NF_DROP;
751
752         nf_bridge = skb->nf_bridge;
753         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
754                 return NF_ACCEPT;
755
756         /* Bridged, take PF_BRIDGE/FORWARD.
757          * (see big note in front of br_nf_pre_routing_finish) */
758         nf_bridge->physoutdev = skb->dev;
759         realindev = nf_bridge->physindev;
760
761         if (nf_bridge->mask & BRNF_PKT_TYPE) {
762                 skb->pkt_type = PACKET_OTHERHOST;
763                 nf_bridge->mask ^= BRNF_PKT_TYPE;
764         }
765         nf_bridge_push_encap_header(skb);
766
767         NF_HOOK(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
768                 br_forward_finish);
769         return NF_STOLEN;
770 }
771
772 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
773 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
774 {
775         if (skb->nfct != NULL &&
776             (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb)) &&
777             skb->len > skb->dev->mtu &&
778             !skb_is_gso(skb))
779                 return ip_fragment(skb, br_dev_queue_push_xmit);
780         else
781                 return br_dev_queue_push_xmit(skb);
782 }
783 #else
784 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
785 {
786         return br_dev_queue_push_xmit(skb);
787 }
788 #endif
789
790 /* PF_BRIDGE/POST_ROUTING ********************************************/
791 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
792                                        const struct net_device *in,
793                                        const struct net_device *out,
794                                        int (*okfn)(struct sk_buff *))
795 {
796         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
797         struct net_device *realoutdev = bridge_parent(skb->dev);
798         u_int8_t pf;
799
800         if (!nf_bridge)
801                 return NF_ACCEPT;
802
803         if (!(nf_bridge->mask & (BRNF_BRIDGED | BRNF_BRIDGED_DNAT)))
804                 return NF_ACCEPT;
805
806         if (!realoutdev)
807                 return NF_DROP;
808
809         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
810             IS_PPPOE_IP(skb))
811                 pf = PF_INET;
812         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
813                  IS_PPPOE_IPV6(skb))
814                 pf = PF_INET6;
815         else
816                 return NF_ACCEPT;
817
818         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
819          * about the value of skb->pkt_type. */
820         if (skb->pkt_type == PACKET_OTHERHOST) {
821                 skb->pkt_type = PACKET_HOST;
822                 nf_bridge->mask |= BRNF_PKT_TYPE;
823         }
824
825         nf_bridge_pull_encap_header(skb);
826         nf_bridge_save_header(skb);
827
828         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
829                 br_nf_dev_queue_xmit);
830
831         return NF_STOLEN;
832 }
833
834 /* IP/SABOTAGE *****************************************************/
835 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
836  * for the second time. */
837 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
838                                    const struct net_device *in,
839                                    const struct net_device *out,
840                                    int (*okfn)(struct sk_buff *))
841 {
842         if (skb->nf_bridge &&
843             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
844                 return NF_STOP;
845         }
846
847         return NF_ACCEPT;
848 }
849
850 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
851  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
852  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
853  * ip_refrag() can return NF_STOLEN. */
854 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
855         {
856                 .hook = br_nf_pre_routing,
857                 .owner = THIS_MODULE,
858                 .pf = PF_BRIDGE,
859                 .hooknum = NF_BR_PRE_ROUTING,
860                 .priority = NF_BR_PRI_BRNF,
861         },
862         {
863                 .hook = br_nf_local_in,
864                 .owner = THIS_MODULE,
865                 .pf = PF_BRIDGE,
866                 .hooknum = NF_BR_LOCAL_IN,
867                 .priority = NF_BR_PRI_BRNF,
868         },
869         {
870                 .hook = br_nf_forward_ip,
871                 .owner = THIS_MODULE,
872                 .pf = PF_BRIDGE,
873                 .hooknum = NF_BR_FORWARD,
874                 .priority = NF_BR_PRI_BRNF - 1,
875         },
876         {
877                 .hook = br_nf_forward_arp,
878                 .owner = THIS_MODULE,
879                 .pf = PF_BRIDGE,
880                 .hooknum = NF_BR_FORWARD,
881                 .priority = NF_BR_PRI_BRNF,
882         },
883         {
884                 .hook = br_nf_local_out,
885                 .owner = THIS_MODULE,
886                 .pf = PF_BRIDGE,
887                 .hooknum = NF_BR_LOCAL_OUT,
888                 .priority = NF_BR_PRI_FIRST,
889         },
890         {
891                 .hook = br_nf_post_routing,
892                 .owner = THIS_MODULE,
893                 .pf = PF_BRIDGE,
894                 .hooknum = NF_BR_POST_ROUTING,
895                 .priority = NF_BR_PRI_LAST,
896         },
897         {
898                 .hook = ip_sabotage_in,
899                 .owner = THIS_MODULE,
900                 .pf = PF_INET,
901                 .hooknum = NF_INET_PRE_ROUTING,
902                 .priority = NF_IP_PRI_FIRST,
903         },
904         {
905                 .hook = ip_sabotage_in,
906                 .owner = THIS_MODULE,
907                 .pf = PF_INET6,
908                 .hooknum = NF_INET_PRE_ROUTING,
909                 .priority = NF_IP6_PRI_FIRST,
910         },
911 };
912
913 #ifdef CONFIG_SYSCTL
914 static
915 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
916                             void __user * buffer, size_t * lenp, loff_t * ppos)
917 {
918         int ret;
919
920         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
921
922         if (write && *(int *)(ctl->data))
923                 *(int *)(ctl->data) = 1;
924         return ret;
925 }
926
927 static ctl_table brnf_table[] = {
928         {
929                 .procname       = "bridge-nf-call-arptables",
930                 .data           = &brnf_call_arptables,
931                 .maxlen         = sizeof(int),
932                 .mode           = 0644,
933                 .proc_handler   = brnf_sysctl_call_tables,
934         },
935         {
936                 .procname       = "bridge-nf-call-iptables",
937                 .data           = &brnf_call_iptables,
938                 .maxlen         = sizeof(int),
939                 .mode           = 0644,
940                 .proc_handler   = brnf_sysctl_call_tables,
941         },
942         {
943                 .procname       = "bridge-nf-call-ip6tables",
944                 .data           = &brnf_call_ip6tables,
945                 .maxlen         = sizeof(int),
946                 .mode           = 0644,
947                 .proc_handler   = brnf_sysctl_call_tables,
948         },
949         {
950                 .procname       = "bridge-nf-filter-vlan-tagged",
951                 .data           = &brnf_filter_vlan_tagged,
952                 .maxlen         = sizeof(int),
953                 .mode           = 0644,
954                 .proc_handler   = brnf_sysctl_call_tables,
955         },
956         {
957                 .procname       = "bridge-nf-filter-pppoe-tagged",
958                 .data           = &brnf_filter_pppoe_tagged,
959                 .maxlen         = sizeof(int),
960                 .mode           = 0644,
961                 .proc_handler   = brnf_sysctl_call_tables,
962         },
963         { }
964 };
965
966 static struct ctl_path brnf_path[] = {
967         { .procname = "net", },
968         { .procname = "bridge", },
969         { }
970 };
971 #endif
972
973 int __init br_netfilter_init(void)
974 {
975         int ret;
976
977         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
978         if (ret < 0)
979                 return ret;
980 #ifdef CONFIG_SYSCTL
981         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
982         if (brnf_sysctl_header == NULL) {
983                 printk(KERN_WARNING
984                        "br_netfilter: can't register to sysctl.\n");
985                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
986                 return -ENOMEM;
987         }
988 #endif
989         printk(KERN_NOTICE "Bridge firewalling registered\n");
990         return 0;
991 }
992
993 void br_netfilter_fini(void)
994 {
995         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
996 #ifdef CONFIG_SYSCTL
997         unregister_sysctl_table(brnf_sysctl_header);
998 #endif
999 }