Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/cmarinas...
[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/slab.h>
20 #include <linux/ip.h>
21 #include <linux/netdevice.h>
22 #include <linux/skbuff.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <linux/if_pppox.h>
27 #include <linux/ppp_defs.h>
28 #include <linux/netfilter_bridge.h>
29 #include <linux/netfilter_ipv4.h>
30 #include <linux/netfilter_ipv6.h>
31 #include <linux/netfilter_arp.h>
32 #include <linux/in_route.h>
33 #include <linux/inetdevice.h>
34
35 #include <net/ip.h>
36 #include <net/ipv6.h>
37 #include <net/route.h>
38
39 #include <asm/uaccess.h>
40 #include "br_private.h"
41 #ifdef CONFIG_SYSCTL
42 #include <linux/sysctl.h>
43 #endif
44
45 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
46                                  (skb->nf_bridge->data))->daddr.ipv4)
47 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
48 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
49
50 #ifdef CONFIG_SYSCTL
51 static struct ctl_table_header *brnf_sysctl_header;
52 static int brnf_call_iptables __read_mostly = 1;
53 static int brnf_call_ip6tables __read_mostly = 1;
54 static int brnf_call_arptables __read_mostly = 1;
55 static int brnf_filter_vlan_tagged __read_mostly = 0;
56 static int brnf_filter_pppoe_tagged __read_mostly = 0;
57 #else
58 #define brnf_call_iptables 1
59 #define brnf_call_ip6tables 1
60 #define brnf_call_arptables 1
61 #define brnf_filter_vlan_tagged 0
62 #define brnf_filter_pppoe_tagged 0
63 #endif
64
65 static inline __be16 vlan_proto(const struct sk_buff *skb)
66 {
67         if (vlan_tx_tag_present(skb))
68                 return skb->protocol;
69         else if (skb->protocol == htons(ETH_P_8021Q))
70                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
71         else
72                 return 0;
73 }
74
75 #define IS_VLAN_IP(skb) \
76         (vlan_proto(skb) == htons(ETH_P_IP) && \
77          brnf_filter_vlan_tagged)
78
79 #define IS_VLAN_IPV6(skb) \
80         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
81          brnf_filter_vlan_tagged)
82
83 #define IS_VLAN_ARP(skb) \
84         (vlan_proto(skb) == htons(ETH_P_ARP) && \
85          brnf_filter_vlan_tagged)
86
87 static inline __be16 pppoe_proto(const struct sk_buff *skb)
88 {
89         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
90                             sizeof(struct pppoe_hdr)));
91 }
92
93 #define IS_PPPOE_IP(skb) \
94         (skb->protocol == htons(ETH_P_PPP_SES) && \
95          pppoe_proto(skb) == htons(PPP_IP) && \
96          brnf_filter_pppoe_tagged)
97
98 #define IS_PPPOE_IPV6(skb) \
99         (skb->protocol == htons(ETH_P_PPP_SES) && \
100          pppoe_proto(skb) == htons(PPP_IPV6) && \
101          brnf_filter_pppoe_tagged)
102
103 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
104 {
105 }
106
107 static struct dst_ops fake_dst_ops = {
108         .family =               AF_INET,
109         .protocol =             cpu_to_be16(ETH_P_IP),
110         .update_pmtu =          fake_update_pmtu,
111 };
112
113 /*
114  * Initialize bogus route table used to keep netfilter happy.
115  * Currently, we fill in the PMTU entry because netfilter
116  * refragmentation needs it, and the rt_flags entry because
117  * ipt_REJECT needs it.  Future netfilter modules might
118  * require us to fill additional fields.
119  */
120 void br_netfilter_rtable_init(struct net_bridge *br)
121 {
122         struct rtable *rt = &br->fake_rtable;
123
124         atomic_set(&rt->dst.__refcnt, 1);
125         rt->dst.dev = br->dev;
126         rt->dst.path = &rt->dst;
127         dst_metric_set(&rt->dst, RTAX_MTU, 1500);
128         rt->dst.flags   = DST_NOXFRM;
129         rt->dst.ops = &fake_dst_ops;
130 }
131
132 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
133 {
134         struct net_bridge_port *port;
135
136         port = br_port_get_rcu(dev);
137         return port ? &port->br->fake_rtable : NULL;
138 }
139
140 static inline struct net_device *bridge_parent(const struct net_device *dev)
141 {
142         struct net_bridge_port *port;
143
144         port = br_port_get_rcu(dev);
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                 }
168                 nf_bridge_put(nf_bridge);
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 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
207 {
208         if (skb->nf_bridge->mask & BRNF_8021Q)
209                 skb->protocol = htons(ETH_P_8021Q);
210         else if (skb->nf_bridge->mask & BRNF_PPPoE)
211                 skb->protocol = htons(ETH_P_PPP_SES);
212 }
213
214 /* When handing a packet over to the IP layer
215  * check whether we have a skb that is in the
216  * expected format
217  */
218
219 static int br_parse_ip_options(struct sk_buff *skb)
220 {
221         struct ip_options *opt;
222         struct iphdr *iph;
223         struct net_device *dev = skb->dev;
224         u32 len;
225
226         iph = ip_hdr(skb);
227         opt = &(IPCB(skb)->opt);
228
229         /* Basic sanity checks */
230         if (iph->ihl < 5 || iph->version != 4)
231                 goto inhdr_error;
232
233         if (!pskb_may_pull(skb, iph->ihl*4))
234                 goto inhdr_error;
235
236         iph = ip_hdr(skb);
237         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
238                 goto inhdr_error;
239
240         len = ntohs(iph->tot_len);
241         if (skb->len < len) {
242                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
243                 goto drop;
244         } else if (len < (iph->ihl*4))
245                 goto inhdr_error;
246
247         if (pskb_trim_rcsum(skb, len)) {
248                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
249                 goto drop;
250         }
251
252         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
253         if (iph->ihl == 5)
254                 return 0;
255
256         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
257         if (ip_options_compile(dev_net(dev), opt, skb))
258                 goto inhdr_error;
259
260         /* Check correct handling of SRR option */
261         if (unlikely(opt->srr)) {
262                 struct in_device *in_dev = __in_dev_get_rcu(dev);
263                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
264                         goto drop;
265
266                 if (ip_options_rcv_srr(skb))
267                         goto drop;
268         }
269
270         return 0;
271
272 inhdr_error:
273         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
274 drop:
275         return -1;
276 }
277
278 /* Fill in the header for fragmented IP packets handled by
279  * the IPv4 connection tracking code.
280  */
281 int nf_bridge_copy_header(struct sk_buff *skb)
282 {
283         int err;
284         unsigned int header_size;
285
286         nf_bridge_update_protocol(skb);
287         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
288         err = skb_cow_head(skb, header_size);
289         if (err)
290                 return err;
291
292         skb_copy_to_linear_data_offset(skb, -header_size,
293                                        skb->nf_bridge->data, header_size);
294         __skb_push(skb, nf_bridge_encap_header_len(skb));
295         return 0;
296 }
297
298 /* PF_BRIDGE/PRE_ROUTING *********************************************/
299 /* Undo the changes made for ip6tables PREROUTING and continue the
300  * bridge PRE_ROUTING hook. */
301 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
302 {
303         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
304         struct rtable *rt;
305
306         if (nf_bridge->mask & BRNF_PKT_TYPE) {
307                 skb->pkt_type = PACKET_OTHERHOST;
308                 nf_bridge->mask ^= BRNF_PKT_TYPE;
309         }
310         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
311
312         rt = bridge_parent_rtable(nf_bridge->physindev);
313         if (!rt) {
314                 kfree_skb(skb);
315                 return 0;
316         }
317         skb_dst_set_noref(skb, &rt->dst);
318
319         skb->dev = nf_bridge->physindev;
320         nf_bridge_update_protocol(skb);
321         nf_bridge_push_encap_header(skb);
322         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
323                        br_handle_frame_finish, 1);
324
325         return 0;
326 }
327
328 /* Obtain the correct destination MAC address, while preserving the original
329  * source MAC address. If we already know this address, we just copy it. If we
330  * don't, we use the neighbour framework to find out. In both cases, we make
331  * sure that br_handle_frame_finish() is called afterwards.
332  */
333 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
334 {
335         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
336         struct dst_entry *dst;
337
338         skb->dev = bridge_parent(skb->dev);
339         if (!skb->dev)
340                 goto free_skb;
341         dst = skb_dst(skb);
342         if (dst->hh) {
343                 neigh_hh_bridge(dst->hh, skb);
344                 skb->dev = nf_bridge->physindev;
345                 return br_handle_frame_finish(skb);
346         } else if (dst->neighbour) {
347                 /* the neighbour function below overwrites the complete
348                  * MAC header, so we save the Ethernet source address and
349                  * protocol number. */
350                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
351                 /* tell br_dev_xmit to continue with forwarding */
352                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
353                 return dst->neighbour->output(skb);
354         }
355 free_skb:
356         kfree_skb(skb);
357         return 0;
358 }
359
360 /* This requires some explaining. If DNAT has taken place,
361  * we will need to fix up the destination Ethernet address.
362  *
363  * There are two cases to consider:
364  * 1. The packet was DNAT'ed to a device in the same bridge
365  *    port group as it was received on. We can still bridge
366  *    the packet.
367  * 2. The packet was DNAT'ed to a different device, either
368  *    a non-bridged device or another bridge port group.
369  *    The packet will need to be routed.
370  *
371  * The correct way of distinguishing between these two cases is to
372  * call ip_route_input() and to look at skb->dst->dev, which is
373  * changed to the destination device if ip_route_input() succeeds.
374  *
375  * Let's first consider the case that ip_route_input() succeeds:
376  *
377  * If the output device equals the logical bridge device the packet
378  * came in on, we can consider this bridging. The corresponding MAC
379  * address will be obtained in br_nf_pre_routing_finish_bridge.
380  * Otherwise, the packet is considered to be routed and we just
381  * change the destination MAC address so that the packet will
382  * later be passed up to the IP stack to be routed. For a redirected
383  * packet, ip_route_input() will give back the localhost as output device,
384  * which differs from the bridge device.
385  *
386  * Let's now consider the case that ip_route_input() fails:
387  *
388  * This can be because the destination address is martian, in which case
389  * the packet will be dropped.
390  * If IP forwarding is disabled, ip_route_input() will fail, while
391  * ip_route_output_key() can return success. The source
392  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
393  * thinks we're handling a locally generated packet and won't care
394  * if IP forwarding is enabled. If the output device equals the logical bridge
395  * device, we proceed as if ip_route_input() succeeded. If it differs from the
396  * logical bridge port or if ip_route_output_key() fails we drop the packet.
397  */
398 static int br_nf_pre_routing_finish(struct sk_buff *skb)
399 {
400         struct net_device *dev = skb->dev;
401         struct iphdr *iph = ip_hdr(skb);
402         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
403         struct rtable *rt;
404         int err;
405
406         if (nf_bridge->mask & BRNF_PKT_TYPE) {
407                 skb->pkt_type = PACKET_OTHERHOST;
408                 nf_bridge->mask ^= BRNF_PKT_TYPE;
409         }
410         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
411         if (dnat_took_place(skb)) {
412                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
413                         struct in_device *in_dev = __in_dev_get_rcu(dev);
414
415                         /* If err equals -EHOSTUNREACH the error is due to a
416                          * martian destination or due to the fact that
417                          * forwarding is disabled. For most martian packets,
418                          * ip_route_output_key() will fail. It won't fail for 2 types of
419                          * martian destinations: loopback destinations and destination
420                          * 0.0.0.0. In both cases the packet will be dropped because the
421                          * destination is the loopback device and not the bridge. */
422                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
423                                 goto free_skb;
424
425                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
426                                              RT_TOS(iph->tos), 0);
427                         if (!IS_ERR(rt)) {
428                                 /* - Bridged-and-DNAT'ed traffic doesn't
429                                  *   require ip_forwarding. */
430                                 if (rt->dst.dev == dev) {
431                                         skb_dst_set(skb, &rt->dst);
432                                         goto bridged_dnat;
433                                 }
434                                 ip_rt_put(rt);
435                         }
436 free_skb:
437                         kfree_skb(skb);
438                         return 0;
439                 } else {
440                         if (skb_dst(skb)->dev == dev) {
441 bridged_dnat:
442                                 skb->dev = nf_bridge->physindev;
443                                 nf_bridge_update_protocol(skb);
444                                 nf_bridge_push_encap_header(skb);
445                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
446                                                NF_BR_PRE_ROUTING,
447                                                skb, skb->dev, NULL,
448                                                br_nf_pre_routing_finish_bridge,
449                                                1);
450                                 return 0;
451                         }
452                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
453                         skb->pkt_type = PACKET_HOST;
454                 }
455         } else {
456                 rt = bridge_parent_rtable(nf_bridge->physindev);
457                 if (!rt) {
458                         kfree_skb(skb);
459                         return 0;
460                 }
461                 skb_dst_set_noref(skb, &rt->dst);
462         }
463
464         skb->dev = nf_bridge->physindev;
465         nf_bridge_update_protocol(skb);
466         nf_bridge_push_encap_header(skb);
467         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
468                        br_handle_frame_finish, 1);
469
470         return 0;
471 }
472
473 /* Some common code for IPv4/IPv6 */
474 static struct net_device *setup_pre_routing(struct sk_buff *skb)
475 {
476         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
477
478         if (skb->pkt_type == PACKET_OTHERHOST) {
479                 skb->pkt_type = PACKET_HOST;
480                 nf_bridge->mask |= BRNF_PKT_TYPE;
481         }
482
483         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
484         nf_bridge->physindev = skb->dev;
485         skb->dev = bridge_parent(skb->dev);
486         if (skb->protocol == htons(ETH_P_8021Q))
487                 nf_bridge->mask |= BRNF_8021Q;
488         else if (skb->protocol == htons(ETH_P_PPP_SES))
489                 nf_bridge->mask |= BRNF_PPPoE;
490
491         return skb->dev;
492 }
493
494 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
495 static int check_hbh_len(struct sk_buff *skb)
496 {
497         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
498         u32 pkt_len;
499         const unsigned char *nh = skb_network_header(skb);
500         int off = raw - nh;
501         int len = (raw[1] + 1) << 3;
502
503         if ((raw + len) - skb->data > skb_headlen(skb))
504                 goto bad;
505
506         off += 2;
507         len -= 2;
508
509         while (len > 0) {
510                 int optlen = nh[off + 1] + 2;
511
512                 switch (nh[off]) {
513                 case IPV6_TLV_PAD0:
514                         optlen = 1;
515                         break;
516
517                 case IPV6_TLV_PADN:
518                         break;
519
520                 case IPV6_TLV_JUMBO:
521                         if (nh[off + 1] != 4 || (off & 3) != 2)
522                                 goto bad;
523                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
524                         if (pkt_len <= IPV6_MAXPLEN ||
525                             ipv6_hdr(skb)->payload_len)
526                                 goto bad;
527                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
528                                 goto bad;
529                         if (pskb_trim_rcsum(skb,
530                                             pkt_len + sizeof(struct ipv6hdr)))
531                                 goto bad;
532                         nh = skb_network_header(skb);
533                         break;
534                 default:
535                         if (optlen > len)
536                                 goto bad;
537                         break;
538                 }
539                 off += optlen;
540                 len -= optlen;
541         }
542         if (len == 0)
543                 return 0;
544 bad:
545         return -1;
546
547 }
548
549 /* Replicate the checks that IPv6 does on packet reception and pass the packet
550  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
551 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
552                                            struct sk_buff *skb,
553                                            const struct net_device *in,
554                                            const struct net_device *out,
555                                            int (*okfn)(struct sk_buff *))
556 {
557         struct ipv6hdr *hdr;
558         u32 pkt_len;
559
560         if (skb->len < sizeof(struct ipv6hdr))
561                 return NF_DROP;
562
563         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
564                 return NF_DROP;
565
566         hdr = ipv6_hdr(skb);
567
568         if (hdr->version != 6)
569                 return NF_DROP;
570
571         pkt_len = ntohs(hdr->payload_len);
572
573         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
574                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
575                         return NF_DROP;
576                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
577                         return NF_DROP;
578         }
579         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
580                 return NF_DROP;
581
582         nf_bridge_put(skb->nf_bridge);
583         if (!nf_bridge_alloc(skb))
584                 return NF_DROP;
585         if (!setup_pre_routing(skb))
586                 return NF_DROP;
587
588         skb->protocol = htons(ETH_P_IPV6);
589         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
590                 br_nf_pre_routing_finish_ipv6);
591
592         return NF_STOLEN;
593 }
594
595 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
596  * Replicate the checks that IPv4 does on packet reception.
597  * Set skb->dev to the bridge device (i.e. parent of the
598  * receiving device) to make netfilter happy, the REDIRECT
599  * target in particular.  Save the original destination IP
600  * address to be able to detect DNAT afterwards. */
601 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
602                                       const struct net_device *in,
603                                       const struct net_device *out,
604                                       int (*okfn)(struct sk_buff *))
605 {
606         struct net_bridge_port *p;
607         struct net_bridge *br;
608         __u32 len = nf_bridge_encap_header_len(skb);
609
610         if (unlikely(!pskb_may_pull(skb, len)))
611                 return NF_DROP;
612
613         p = br_port_get_rcu(in);
614         if (p == NULL)
615                 return NF_DROP;
616         br = p->br;
617
618         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
619             IS_PPPOE_IPV6(skb)) {
620                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
621                         return NF_ACCEPT;
622
623                 nf_bridge_pull_encap_header_rcsum(skb);
624                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
625         }
626
627         if (!brnf_call_iptables && !br->nf_call_iptables)
628                 return NF_ACCEPT;
629
630         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
631             !IS_PPPOE_IP(skb))
632                 return NF_ACCEPT;
633
634         nf_bridge_pull_encap_header_rcsum(skb);
635
636         if (br_parse_ip_options(skb))
637                 return NF_DROP;
638
639         nf_bridge_put(skb->nf_bridge);
640         if (!nf_bridge_alloc(skb))
641                 return NF_DROP;
642         if (!setup_pre_routing(skb))
643                 return NF_DROP;
644         store_orig_dstaddr(skb);
645         skb->protocol = htons(ETH_P_IP);
646
647         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
648                 br_nf_pre_routing_finish);
649
650         return NF_STOLEN;
651 }
652
653
654 /* PF_BRIDGE/LOCAL_IN ************************************************/
655 /* The packet is locally destined, which requires a real
656  * dst_entry, so detach the fake one.  On the way up, the
657  * packet would pass through PRE_ROUTING again (which already
658  * took place when the packet entered the bridge), but we
659  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
660  * prevent this from happening. */
661 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
662                                    const struct net_device *in,
663                                    const struct net_device *out,
664                                    int (*okfn)(struct sk_buff *))
665 {
666         struct rtable *rt = skb_rtable(skb);
667
668         if (rt && rt == bridge_parent_rtable(in))
669                 skb_dst_drop(skb);
670
671         return NF_ACCEPT;
672 }
673
674 /* PF_BRIDGE/FORWARD *************************************************/
675 static int br_nf_forward_finish(struct sk_buff *skb)
676 {
677         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
678         struct net_device *in;
679
680         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
681                 in = nf_bridge->physindev;
682                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
683                         skb->pkt_type = PACKET_OTHERHOST;
684                         nf_bridge->mask ^= BRNF_PKT_TYPE;
685                 }
686                 nf_bridge_update_protocol(skb);
687         } else {
688                 in = *((struct net_device **)(skb->cb));
689         }
690         nf_bridge_push_encap_header(skb);
691
692         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
693                        skb->dev, br_forward_finish, 1);
694         return 0;
695 }
696
697 /* This is the 'purely bridged' case.  For IP, we pass the packet to
698  * netfilter with indev and outdev set to the bridge device,
699  * but we are still able to filter on the 'real' indev/outdev
700  * because of the physdev module. For ARP, indev and outdev are the
701  * bridge ports. */
702 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
703                                      const struct net_device *in,
704                                      const struct net_device *out,
705                                      int (*okfn)(struct sk_buff *))
706 {
707         struct nf_bridge_info *nf_bridge;
708         struct net_device *parent;
709         u_int8_t pf;
710
711         if (!skb->nf_bridge)
712                 return NF_ACCEPT;
713
714         /* Need exclusive nf_bridge_info since we might have multiple
715          * different physoutdevs. */
716         if (!nf_bridge_unshare(skb))
717                 return NF_DROP;
718
719         parent = bridge_parent(out);
720         if (!parent)
721                 return NF_DROP;
722
723         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
724             IS_PPPOE_IP(skb))
725                 pf = PF_INET;
726         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
727                  IS_PPPOE_IPV6(skb))
728                 pf = PF_INET6;
729         else
730                 return NF_ACCEPT;
731
732         nf_bridge_pull_encap_header(skb);
733
734         nf_bridge = skb->nf_bridge;
735         if (skb->pkt_type == PACKET_OTHERHOST) {
736                 skb->pkt_type = PACKET_HOST;
737                 nf_bridge->mask |= BRNF_PKT_TYPE;
738         }
739
740         if (pf == PF_INET && br_parse_ip_options(skb))
741                 return NF_DROP;
742
743         /* The physdev module checks on this */
744         nf_bridge->mask |= BRNF_BRIDGED;
745         nf_bridge->physoutdev = skb->dev;
746         if (pf == PF_INET)
747                 skb->protocol = htons(ETH_P_IP);
748         else
749                 skb->protocol = htons(ETH_P_IPV6);
750
751         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
752                 br_nf_forward_finish);
753
754         return NF_STOLEN;
755 }
756
757 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
758                                       const struct net_device *in,
759                                       const struct net_device *out,
760                                       int (*okfn)(struct sk_buff *))
761 {
762         struct net_bridge_port *p;
763         struct net_bridge *br;
764         struct net_device **d = (struct net_device **)(skb->cb);
765
766         p = br_port_get_rcu(out);
767         if (p == NULL)
768                 return NF_ACCEPT;
769         br = p->br;
770
771         if (!brnf_call_arptables && !br->nf_call_arptables)
772                 return NF_ACCEPT;
773
774         if (skb->protocol != htons(ETH_P_ARP)) {
775                 if (!IS_VLAN_ARP(skb))
776                         return NF_ACCEPT;
777                 nf_bridge_pull_encap_header(skb);
778         }
779
780         if (arp_hdr(skb)->ar_pln != 4) {
781                 if (IS_VLAN_ARP(skb))
782                         nf_bridge_push_encap_header(skb);
783                 return NF_ACCEPT;
784         }
785         *d = (struct net_device *)in;
786         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
787                 (struct net_device *)out, br_nf_forward_finish);
788
789         return NF_STOLEN;
790 }
791
792 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
793 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
794 {
795         int ret;
796
797         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
798             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
799             !skb_is_gso(skb)) {
800                 if (br_parse_ip_options(skb))
801                         /* Drop invalid packet */
802                         return NF_DROP;
803                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
804         } else
805                 ret = br_dev_queue_push_xmit(skb);
806
807         return ret;
808 }
809 #else
810 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
811 {
812         return br_dev_queue_push_xmit(skb);
813 }
814 #endif
815
816 /* PF_BRIDGE/POST_ROUTING ********************************************/
817 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
818                                        const struct net_device *in,
819                                        const struct net_device *out,
820                                        int (*okfn)(struct sk_buff *))
821 {
822         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
823         struct net_device *realoutdev = bridge_parent(skb->dev);
824         u_int8_t pf;
825
826         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
827                 return NF_ACCEPT;
828
829         if (!realoutdev)
830                 return NF_DROP;
831
832         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
833             IS_PPPOE_IP(skb))
834                 pf = PF_INET;
835         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
836                  IS_PPPOE_IPV6(skb))
837                 pf = PF_INET6;
838         else
839                 return NF_ACCEPT;
840
841         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
842          * about the value of skb->pkt_type. */
843         if (skb->pkt_type == PACKET_OTHERHOST) {
844                 skb->pkt_type = PACKET_HOST;
845                 nf_bridge->mask |= BRNF_PKT_TYPE;
846         }
847
848         nf_bridge_pull_encap_header(skb);
849         nf_bridge_save_header(skb);
850         if (pf == PF_INET)
851                 skb->protocol = htons(ETH_P_IP);
852         else
853                 skb->protocol = htons(ETH_P_IPV6);
854
855         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
856                 br_nf_dev_queue_xmit);
857
858         return NF_STOLEN;
859 }
860
861 /* IP/SABOTAGE *****************************************************/
862 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
863  * for the second time. */
864 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
865                                    const struct net_device *in,
866                                    const struct net_device *out,
867                                    int (*okfn)(struct sk_buff *))
868 {
869         if (skb->nf_bridge &&
870             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
871                 return NF_STOP;
872         }
873
874         return NF_ACCEPT;
875 }
876
877 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
878  * br_dev_queue_push_xmit is called afterwards */
879 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
880         {
881                 .hook = br_nf_pre_routing,
882                 .owner = THIS_MODULE,
883                 .pf = PF_BRIDGE,
884                 .hooknum = NF_BR_PRE_ROUTING,
885                 .priority = NF_BR_PRI_BRNF,
886         },
887         {
888                 .hook = br_nf_local_in,
889                 .owner = THIS_MODULE,
890                 .pf = PF_BRIDGE,
891                 .hooknum = NF_BR_LOCAL_IN,
892                 .priority = NF_BR_PRI_BRNF,
893         },
894         {
895                 .hook = br_nf_forward_ip,
896                 .owner = THIS_MODULE,
897                 .pf = PF_BRIDGE,
898                 .hooknum = NF_BR_FORWARD,
899                 .priority = NF_BR_PRI_BRNF - 1,
900         },
901         {
902                 .hook = br_nf_forward_arp,
903                 .owner = THIS_MODULE,
904                 .pf = PF_BRIDGE,
905                 .hooknum = NF_BR_FORWARD,
906                 .priority = NF_BR_PRI_BRNF,
907         },
908         {
909                 .hook = br_nf_post_routing,
910                 .owner = THIS_MODULE,
911                 .pf = PF_BRIDGE,
912                 .hooknum = NF_BR_POST_ROUTING,
913                 .priority = NF_BR_PRI_LAST,
914         },
915         {
916                 .hook = ip_sabotage_in,
917                 .owner = THIS_MODULE,
918                 .pf = PF_INET,
919                 .hooknum = NF_INET_PRE_ROUTING,
920                 .priority = NF_IP_PRI_FIRST,
921         },
922         {
923                 .hook = ip_sabotage_in,
924                 .owner = THIS_MODULE,
925                 .pf = PF_INET6,
926                 .hooknum = NF_INET_PRE_ROUTING,
927                 .priority = NF_IP6_PRI_FIRST,
928         },
929 };
930
931 #ifdef CONFIG_SYSCTL
932 static
933 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
934                             void __user * buffer, size_t * lenp, loff_t * ppos)
935 {
936         int ret;
937
938         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
939
940         if (write && *(int *)(ctl->data))
941                 *(int *)(ctl->data) = 1;
942         return ret;
943 }
944
945 static ctl_table brnf_table[] = {
946         {
947                 .procname       = "bridge-nf-call-arptables",
948                 .data           = &brnf_call_arptables,
949                 .maxlen         = sizeof(int),
950                 .mode           = 0644,
951                 .proc_handler   = brnf_sysctl_call_tables,
952         },
953         {
954                 .procname       = "bridge-nf-call-iptables",
955                 .data           = &brnf_call_iptables,
956                 .maxlen         = sizeof(int),
957                 .mode           = 0644,
958                 .proc_handler   = brnf_sysctl_call_tables,
959         },
960         {
961                 .procname       = "bridge-nf-call-ip6tables",
962                 .data           = &brnf_call_ip6tables,
963                 .maxlen         = sizeof(int),
964                 .mode           = 0644,
965                 .proc_handler   = brnf_sysctl_call_tables,
966         },
967         {
968                 .procname       = "bridge-nf-filter-vlan-tagged",
969                 .data           = &brnf_filter_vlan_tagged,
970                 .maxlen         = sizeof(int),
971                 .mode           = 0644,
972                 .proc_handler   = brnf_sysctl_call_tables,
973         },
974         {
975                 .procname       = "bridge-nf-filter-pppoe-tagged",
976                 .data           = &brnf_filter_pppoe_tagged,
977                 .maxlen         = sizeof(int),
978                 .mode           = 0644,
979                 .proc_handler   = brnf_sysctl_call_tables,
980         },
981         { }
982 };
983
984 static struct ctl_path brnf_path[] = {
985         { .procname = "net", },
986         { .procname = "bridge", },
987         { }
988 };
989 #endif
990
991 int __init br_netfilter_init(void)
992 {
993         int ret;
994
995         ret = dst_entries_init(&fake_dst_ops);
996         if (ret < 0)
997                 return ret;
998
999         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1000         if (ret < 0) {
1001                 dst_entries_destroy(&fake_dst_ops);
1002                 return ret;
1003         }
1004 #ifdef CONFIG_SYSCTL
1005         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1006         if (brnf_sysctl_header == NULL) {
1007                 printk(KERN_WARNING
1008                        "br_netfilter: can't register to sysctl.\n");
1009                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1010                 dst_entries_destroy(&fake_dst_ops);
1011                 return -ENOMEM;
1012         }
1013 #endif
1014         printk(KERN_NOTICE "Bridge firewalling registered\n");
1015         return 0;
1016 }
1017
1018 void br_netfilter_fini(void)
1019 {
1020         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1021 #ifdef CONFIG_SYSCTL
1022         unregister_sysctl_table(brnf_sysctl_header);
1023 #endif
1024         dst_entries_destroy(&fake_dst_ops);
1025 }