ipv4: flush route cache after change accept_local
[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 u32 *fake_cow_metrics(struct dst_entry *dst, unsigned long old)
108 {
109         return NULL;
110 }
111
112 static struct neighbour *fake_neigh_lookup(const struct dst_entry *dst, const void *daddr)
113 {
114         return NULL;
115 }
116
117 static struct dst_ops fake_dst_ops = {
118         .family =               AF_INET,
119         .protocol =             cpu_to_be16(ETH_P_IP),
120         .update_pmtu =          fake_update_pmtu,
121         .cow_metrics =          fake_cow_metrics,
122         .neigh_lookup =         fake_neigh_lookup,
123 };
124
125 /*
126  * Initialize bogus route table used to keep netfilter happy.
127  * Currently, we fill in the PMTU entry because netfilter
128  * refragmentation needs it, and the rt_flags entry because
129  * ipt_REJECT needs it.  Future netfilter modules might
130  * require us to fill additional fields.
131  */
132 static const u32 br_dst_default_metrics[RTAX_MAX] = {
133         [RTAX_MTU - 1] = 1500,
134 };
135
136 void br_netfilter_rtable_init(struct net_bridge *br)
137 {
138         struct rtable *rt = &br->fake_rtable;
139
140         atomic_set(&rt->dst.__refcnt, 1);
141         rt->dst.dev = br->dev;
142         rt->dst.path = &rt->dst;
143         dst_init_metrics(&rt->dst, br_dst_default_metrics, true);
144         rt->dst.flags   = DST_NOXFRM;
145         rt->dst.ops = &fake_dst_ops;
146 }
147
148 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
149 {
150         struct net_bridge_port *port;
151
152         port = br_port_get_rcu(dev);
153         return port ? &port->br->fake_rtable : NULL;
154 }
155
156 static inline struct net_device *bridge_parent(const struct net_device *dev)
157 {
158         struct net_bridge_port *port;
159
160         port = br_port_get_rcu(dev);
161         return port ? port->br->dev : NULL;
162 }
163
164 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
165 {
166         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
167         if (likely(skb->nf_bridge))
168                 atomic_set(&(skb->nf_bridge->use), 1);
169
170         return skb->nf_bridge;
171 }
172
173 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
174 {
175         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
176
177         if (atomic_read(&nf_bridge->use) > 1) {
178                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
179
180                 if (tmp) {
181                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
182                         atomic_set(&tmp->use, 1);
183                 }
184                 nf_bridge_put(nf_bridge);
185                 nf_bridge = tmp;
186         }
187         return nf_bridge;
188 }
189
190 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
191 {
192         unsigned int len = nf_bridge_encap_header_len(skb);
193
194         skb_push(skb, len);
195         skb->network_header -= len;
196 }
197
198 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
199 {
200         unsigned int len = nf_bridge_encap_header_len(skb);
201
202         skb_pull(skb, len);
203         skb->network_header += len;
204 }
205
206 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
207 {
208         unsigned int len = nf_bridge_encap_header_len(skb);
209
210         skb_pull_rcsum(skb, len);
211         skb->network_header += len;
212 }
213
214 static inline void nf_bridge_save_header(struct sk_buff *skb)
215 {
216         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
217
218         skb_copy_from_linear_data_offset(skb, -header_size,
219                                          skb->nf_bridge->data, header_size);
220 }
221
222 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
223 {
224         if (skb->nf_bridge->mask & BRNF_8021Q)
225                 skb->protocol = htons(ETH_P_8021Q);
226         else if (skb->nf_bridge->mask & BRNF_PPPoE)
227                 skb->protocol = htons(ETH_P_PPP_SES);
228 }
229
230 /* When handing a packet over to the IP layer
231  * check whether we have a skb that is in the
232  * expected format
233  */
234
235 static int br_parse_ip_options(struct sk_buff *skb)
236 {
237         struct ip_options *opt;
238         const struct iphdr *iph;
239         struct net_device *dev = skb->dev;
240         u32 len;
241
242         iph = ip_hdr(skb);
243         opt = &(IPCB(skb)->opt);
244
245         /* Basic sanity checks */
246         if (iph->ihl < 5 || iph->version != 4)
247                 goto inhdr_error;
248
249         if (!pskb_may_pull(skb, iph->ihl*4))
250                 goto inhdr_error;
251
252         iph = ip_hdr(skb);
253         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
254                 goto inhdr_error;
255
256         len = ntohs(iph->tot_len);
257         if (skb->len < len) {
258                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
259                 goto drop;
260         } else if (len < (iph->ihl*4))
261                 goto inhdr_error;
262
263         if (pskb_trim_rcsum(skb, len)) {
264                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
265                 goto drop;
266         }
267
268         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
269         if (iph->ihl == 5)
270                 return 0;
271
272         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
273         if (ip_options_compile(dev_net(dev), opt, skb))
274                 goto inhdr_error;
275
276         /* Check correct handling of SRR option */
277         if (unlikely(opt->srr)) {
278                 struct in_device *in_dev = __in_dev_get_rcu(dev);
279                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
280                         goto drop;
281
282                 if (ip_options_rcv_srr(skb))
283                         goto drop;
284         }
285
286         return 0;
287
288 inhdr_error:
289         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
290 drop:
291         return -1;
292 }
293
294 /* Fill in the header for fragmented IP packets handled by
295  * the IPv4 connection tracking code.
296  */
297 int nf_bridge_copy_header(struct sk_buff *skb)
298 {
299         int err;
300         unsigned int header_size;
301
302         nf_bridge_update_protocol(skb);
303         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
304         err = skb_cow_head(skb, header_size);
305         if (err)
306                 return err;
307
308         skb_copy_to_linear_data_offset(skb, -header_size,
309                                        skb->nf_bridge->data, header_size);
310         __skb_push(skb, nf_bridge_encap_header_len(skb));
311         return 0;
312 }
313
314 /* PF_BRIDGE/PRE_ROUTING *********************************************/
315 /* Undo the changes made for ip6tables PREROUTING and continue the
316  * bridge PRE_ROUTING hook. */
317 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
318 {
319         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
320         struct rtable *rt;
321
322         if (nf_bridge->mask & BRNF_PKT_TYPE) {
323                 skb->pkt_type = PACKET_OTHERHOST;
324                 nf_bridge->mask ^= BRNF_PKT_TYPE;
325         }
326         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
327
328         rt = bridge_parent_rtable(nf_bridge->physindev);
329         if (!rt) {
330                 kfree_skb(skb);
331                 return 0;
332         }
333         skb_dst_set_noref(skb, &rt->dst);
334
335         skb->dev = nf_bridge->physindev;
336         nf_bridge_update_protocol(skb);
337         nf_bridge_push_encap_header(skb);
338         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
339                        br_handle_frame_finish, 1);
340
341         return 0;
342 }
343
344 /* Obtain the correct destination MAC address, while preserving the original
345  * source MAC address. If we already know this address, we just copy it. If we
346  * don't, we use the neighbour framework to find out. In both cases, we make
347  * sure that br_handle_frame_finish() is called afterwards.
348  */
349 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
350 {
351         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
352         struct neighbour *neigh;
353         struct dst_entry *dst;
354
355         skb->dev = bridge_parent(skb->dev);
356         if (!skb->dev)
357                 goto free_skb;
358         dst = skb_dst(skb);
359         neigh = dst_get_neighbour(dst);
360         if (neigh->hh.hh_len) {
361                 neigh_hh_bridge(&neigh->hh, skb);
362                 skb->dev = nf_bridge->physindev;
363                 return br_handle_frame_finish(skb);
364         } else {
365                 /* the neighbour function below overwrites the complete
366                  * MAC header, so we save the Ethernet source address and
367                  * protocol number. */
368                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
369                 /* tell br_dev_xmit to continue with forwarding */
370                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
371                 return neigh->output(neigh, skb);
372         }
373 free_skb:
374         kfree_skb(skb);
375         return 0;
376 }
377
378 /* This requires some explaining. If DNAT has taken place,
379  * we will need to fix up the destination Ethernet address.
380  *
381  * There are two cases to consider:
382  * 1. The packet was DNAT'ed to a device in the same bridge
383  *    port group as it was received on. We can still bridge
384  *    the packet.
385  * 2. The packet was DNAT'ed to a different device, either
386  *    a non-bridged device or another bridge port group.
387  *    The packet will need to be routed.
388  *
389  * The correct way of distinguishing between these two cases is to
390  * call ip_route_input() and to look at skb->dst->dev, which is
391  * changed to the destination device if ip_route_input() succeeds.
392  *
393  * Let's first consider the case that ip_route_input() succeeds:
394  *
395  * If the output device equals the logical bridge device the packet
396  * came in on, we can consider this bridging. The corresponding MAC
397  * address will be obtained in br_nf_pre_routing_finish_bridge.
398  * Otherwise, the packet is considered to be routed and we just
399  * change the destination MAC address so that the packet will
400  * later be passed up to the IP stack to be routed. For a redirected
401  * packet, ip_route_input() will give back the localhost as output device,
402  * which differs from the bridge device.
403  *
404  * Let's now consider the case that ip_route_input() fails:
405  *
406  * This can be because the destination address is martian, in which case
407  * the packet will be dropped.
408  * If IP forwarding is disabled, ip_route_input() will fail, while
409  * ip_route_output_key() can return success. The source
410  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
411  * thinks we're handling a locally generated packet and won't care
412  * if IP forwarding is enabled. If the output device equals the logical bridge
413  * device, we proceed as if ip_route_input() succeeded. If it differs from the
414  * logical bridge port or if ip_route_output_key() fails we drop the packet.
415  */
416 static int br_nf_pre_routing_finish(struct sk_buff *skb)
417 {
418         struct net_device *dev = skb->dev;
419         struct iphdr *iph = ip_hdr(skb);
420         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
421         struct rtable *rt;
422         int err;
423
424         if (nf_bridge->mask & BRNF_PKT_TYPE) {
425                 skb->pkt_type = PACKET_OTHERHOST;
426                 nf_bridge->mask ^= BRNF_PKT_TYPE;
427         }
428         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
429         if (dnat_took_place(skb)) {
430                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
431                         struct in_device *in_dev = __in_dev_get_rcu(dev);
432
433                         /* If err equals -EHOSTUNREACH the error is due to a
434                          * martian destination or due to the fact that
435                          * forwarding is disabled. For most martian packets,
436                          * ip_route_output_key() will fail. It won't fail for 2 types of
437                          * martian destinations: loopback destinations and destination
438                          * 0.0.0.0. In both cases the packet will be dropped because the
439                          * destination is the loopback device and not the bridge. */
440                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
441                                 goto free_skb;
442
443                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
444                                              RT_TOS(iph->tos), 0);
445                         if (!IS_ERR(rt)) {
446                                 /* - Bridged-and-DNAT'ed traffic doesn't
447                                  *   require ip_forwarding. */
448                                 if (rt->dst.dev == dev) {
449                                         skb_dst_set(skb, &rt->dst);
450                                         goto bridged_dnat;
451                                 }
452                                 ip_rt_put(rt);
453                         }
454 free_skb:
455                         kfree_skb(skb);
456                         return 0;
457                 } else {
458                         if (skb_dst(skb)->dev == dev) {
459 bridged_dnat:
460                                 skb->dev = nf_bridge->physindev;
461                                 nf_bridge_update_protocol(skb);
462                                 nf_bridge_push_encap_header(skb);
463                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
464                                                NF_BR_PRE_ROUTING,
465                                                skb, skb->dev, NULL,
466                                                br_nf_pre_routing_finish_bridge,
467                                                1);
468                                 return 0;
469                         }
470                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
471                         skb->pkt_type = PACKET_HOST;
472                 }
473         } else {
474                 rt = bridge_parent_rtable(nf_bridge->physindev);
475                 if (!rt) {
476                         kfree_skb(skb);
477                         return 0;
478                 }
479                 skb_dst_set_noref(skb, &rt->dst);
480         }
481
482         skb->dev = nf_bridge->physindev;
483         nf_bridge_update_protocol(skb);
484         nf_bridge_push_encap_header(skb);
485         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
486                        br_handle_frame_finish, 1);
487
488         return 0;
489 }
490
491 /* Some common code for IPv4/IPv6 */
492 static struct net_device *setup_pre_routing(struct sk_buff *skb)
493 {
494         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
495
496         if (skb->pkt_type == PACKET_OTHERHOST) {
497                 skb->pkt_type = PACKET_HOST;
498                 nf_bridge->mask |= BRNF_PKT_TYPE;
499         }
500
501         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
502         nf_bridge->physindev = skb->dev;
503         skb->dev = bridge_parent(skb->dev);
504         if (skb->protocol == htons(ETH_P_8021Q))
505                 nf_bridge->mask |= BRNF_8021Q;
506         else if (skb->protocol == htons(ETH_P_PPP_SES))
507                 nf_bridge->mask |= BRNF_PPPoE;
508
509         return skb->dev;
510 }
511
512 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
513 static int check_hbh_len(struct sk_buff *skb)
514 {
515         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
516         u32 pkt_len;
517         const unsigned char *nh = skb_network_header(skb);
518         int off = raw - nh;
519         int len = (raw[1] + 1) << 3;
520
521         if ((raw + len) - skb->data > skb_headlen(skb))
522                 goto bad;
523
524         off += 2;
525         len -= 2;
526
527         while (len > 0) {
528                 int optlen = nh[off + 1] + 2;
529
530                 switch (nh[off]) {
531                 case IPV6_TLV_PAD0:
532                         optlen = 1;
533                         break;
534
535                 case IPV6_TLV_PADN:
536                         break;
537
538                 case IPV6_TLV_JUMBO:
539                         if (nh[off + 1] != 4 || (off & 3) != 2)
540                                 goto bad;
541                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
542                         if (pkt_len <= IPV6_MAXPLEN ||
543                             ipv6_hdr(skb)->payload_len)
544                                 goto bad;
545                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
546                                 goto bad;
547                         if (pskb_trim_rcsum(skb,
548                                             pkt_len + sizeof(struct ipv6hdr)))
549                                 goto bad;
550                         nh = skb_network_header(skb);
551                         break;
552                 default:
553                         if (optlen > len)
554                                 goto bad;
555                         break;
556                 }
557                 off += optlen;
558                 len -= optlen;
559         }
560         if (len == 0)
561                 return 0;
562 bad:
563         return -1;
564
565 }
566
567 /* Replicate the checks that IPv6 does on packet reception and pass the packet
568  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
569 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
570                                            struct sk_buff *skb,
571                                            const struct net_device *in,
572                                            const struct net_device *out,
573                                            int (*okfn)(struct sk_buff *))
574 {
575         const struct ipv6hdr *hdr;
576         u32 pkt_len;
577
578         if (skb->len < sizeof(struct ipv6hdr))
579                 return NF_DROP;
580
581         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
582                 return NF_DROP;
583
584         hdr = ipv6_hdr(skb);
585
586         if (hdr->version != 6)
587                 return NF_DROP;
588
589         pkt_len = ntohs(hdr->payload_len);
590
591         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
592                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
593                         return NF_DROP;
594                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
595                         return NF_DROP;
596         }
597         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
598                 return NF_DROP;
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
606         skb->protocol = htons(ETH_P_IPV6);
607         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
608                 br_nf_pre_routing_finish_ipv6);
609
610         return NF_STOLEN;
611 }
612
613 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
614  * Replicate the checks that IPv4 does on packet reception.
615  * Set skb->dev to the bridge device (i.e. parent of the
616  * receiving device) to make netfilter happy, the REDIRECT
617  * target in particular.  Save the original destination IP
618  * address to be able to detect DNAT afterwards. */
619 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
620                                       const struct net_device *in,
621                                       const struct net_device *out,
622                                       int (*okfn)(struct sk_buff *))
623 {
624         struct net_bridge_port *p;
625         struct net_bridge *br;
626         __u32 len = nf_bridge_encap_header_len(skb);
627
628         if (unlikely(!pskb_may_pull(skb, len)))
629                 return NF_DROP;
630
631         p = br_port_get_rcu(in);
632         if (p == NULL)
633                 return NF_DROP;
634         br = p->br;
635
636         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
637             IS_PPPOE_IPV6(skb)) {
638                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
639                         return NF_ACCEPT;
640
641                 nf_bridge_pull_encap_header_rcsum(skb);
642                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
643         }
644
645         if (!brnf_call_iptables && !br->nf_call_iptables)
646                 return NF_ACCEPT;
647
648         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
649             !IS_PPPOE_IP(skb))
650                 return NF_ACCEPT;
651
652         nf_bridge_pull_encap_header_rcsum(skb);
653
654         if (br_parse_ip_options(skb))
655                 return NF_DROP;
656
657         nf_bridge_put(skb->nf_bridge);
658         if (!nf_bridge_alloc(skb))
659                 return NF_DROP;
660         if (!setup_pre_routing(skb))
661                 return NF_DROP;
662         store_orig_dstaddr(skb);
663         skb->protocol = htons(ETH_P_IP);
664
665         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
666                 br_nf_pre_routing_finish);
667
668         return NF_STOLEN;
669 }
670
671
672 /* PF_BRIDGE/LOCAL_IN ************************************************/
673 /* The packet is locally destined, which requires a real
674  * dst_entry, so detach the fake one.  On the way up, the
675  * packet would pass through PRE_ROUTING again (which already
676  * took place when the packet entered the bridge), but we
677  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
678  * prevent this from happening. */
679 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
680                                    const struct net_device *in,
681                                    const struct net_device *out,
682                                    int (*okfn)(struct sk_buff *))
683 {
684         struct rtable *rt = skb_rtable(skb);
685
686         if (rt && rt == bridge_parent_rtable(in))
687                 skb_dst_drop(skb);
688
689         return NF_ACCEPT;
690 }
691
692 /* PF_BRIDGE/FORWARD *************************************************/
693 static int br_nf_forward_finish(struct sk_buff *skb)
694 {
695         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
696         struct net_device *in;
697
698         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
699                 in = nf_bridge->physindev;
700                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
701                         skb->pkt_type = PACKET_OTHERHOST;
702                         nf_bridge->mask ^= BRNF_PKT_TYPE;
703                 }
704                 nf_bridge_update_protocol(skb);
705         } else {
706                 in = *((struct net_device **)(skb->cb));
707         }
708         nf_bridge_push_encap_header(skb);
709
710         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
711                        skb->dev, br_forward_finish, 1);
712         return 0;
713 }
714
715 /* This is the 'purely bridged' case.  For IP, we pass the packet to
716  * netfilter with indev and outdev set to the bridge device,
717  * but we are still able to filter on the 'real' indev/outdev
718  * because of the physdev module. For ARP, indev and outdev are the
719  * bridge ports. */
720 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
721                                      const struct net_device *in,
722                                      const struct net_device *out,
723                                      int (*okfn)(struct sk_buff *))
724 {
725         struct nf_bridge_info *nf_bridge;
726         struct net_device *parent;
727         u_int8_t pf;
728
729         if (!skb->nf_bridge)
730                 return NF_ACCEPT;
731
732         /* Need exclusive nf_bridge_info since we might have multiple
733          * different physoutdevs. */
734         if (!nf_bridge_unshare(skb))
735                 return NF_DROP;
736
737         parent = bridge_parent(out);
738         if (!parent)
739                 return NF_DROP;
740
741         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
742             IS_PPPOE_IP(skb))
743                 pf = PF_INET;
744         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
745                  IS_PPPOE_IPV6(skb))
746                 pf = PF_INET6;
747         else
748                 return NF_ACCEPT;
749
750         nf_bridge_pull_encap_header(skb);
751
752         nf_bridge = skb->nf_bridge;
753         if (skb->pkt_type == PACKET_OTHERHOST) {
754                 skb->pkt_type = PACKET_HOST;
755                 nf_bridge->mask |= BRNF_PKT_TYPE;
756         }
757
758         if (pf == PF_INET && br_parse_ip_options(skb))
759                 return NF_DROP;
760
761         /* The physdev module checks on this */
762         nf_bridge->mask |= BRNF_BRIDGED;
763         nf_bridge->physoutdev = skb->dev;
764         if (pf == PF_INET)
765                 skb->protocol = htons(ETH_P_IP);
766         else
767                 skb->protocol = htons(ETH_P_IPV6);
768
769         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
770                 br_nf_forward_finish);
771
772         return NF_STOLEN;
773 }
774
775 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
776                                       const struct net_device *in,
777                                       const struct net_device *out,
778                                       int (*okfn)(struct sk_buff *))
779 {
780         struct net_bridge_port *p;
781         struct net_bridge *br;
782         struct net_device **d = (struct net_device **)(skb->cb);
783
784         p = br_port_get_rcu(out);
785         if (p == NULL)
786                 return NF_ACCEPT;
787         br = p->br;
788
789         if (!brnf_call_arptables && !br->nf_call_arptables)
790                 return NF_ACCEPT;
791
792         if (skb->protocol != htons(ETH_P_ARP)) {
793                 if (!IS_VLAN_ARP(skb))
794                         return NF_ACCEPT;
795                 nf_bridge_pull_encap_header(skb);
796         }
797
798         if (arp_hdr(skb)->ar_pln != 4) {
799                 if (IS_VLAN_ARP(skb))
800                         nf_bridge_push_encap_header(skb);
801                 return NF_ACCEPT;
802         }
803         *d = (struct net_device *)in;
804         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
805                 (struct net_device *)out, br_nf_forward_finish);
806
807         return NF_STOLEN;
808 }
809
810 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
811 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
812 {
813         int ret;
814
815         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
816             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
817             !skb_is_gso(skb)) {
818                 if (br_parse_ip_options(skb))
819                         /* Drop invalid packet */
820                         return NF_DROP;
821                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
822         } else
823                 ret = br_dev_queue_push_xmit(skb);
824
825         return ret;
826 }
827 #else
828 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
829 {
830         return br_dev_queue_push_xmit(skb);
831 }
832 #endif
833
834 /* PF_BRIDGE/POST_ROUTING ********************************************/
835 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
836                                        const struct net_device *in,
837                                        const struct net_device *out,
838                                        int (*okfn)(struct sk_buff *))
839 {
840         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
841         struct net_device *realoutdev = bridge_parent(skb->dev);
842         u_int8_t pf;
843
844         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
845                 return NF_ACCEPT;
846
847         if (!realoutdev)
848                 return NF_DROP;
849
850         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
851             IS_PPPOE_IP(skb))
852                 pf = PF_INET;
853         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
854                  IS_PPPOE_IPV6(skb))
855                 pf = PF_INET6;
856         else
857                 return NF_ACCEPT;
858
859         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
860          * about the value of skb->pkt_type. */
861         if (skb->pkt_type == PACKET_OTHERHOST) {
862                 skb->pkt_type = PACKET_HOST;
863                 nf_bridge->mask |= BRNF_PKT_TYPE;
864         }
865
866         nf_bridge_pull_encap_header(skb);
867         nf_bridge_save_header(skb);
868         if (pf == PF_INET)
869                 skb->protocol = htons(ETH_P_IP);
870         else
871                 skb->protocol = htons(ETH_P_IPV6);
872
873         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
874                 br_nf_dev_queue_xmit);
875
876         return NF_STOLEN;
877 }
878
879 /* IP/SABOTAGE *****************************************************/
880 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
881  * for the second time. */
882 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
883                                    const struct net_device *in,
884                                    const struct net_device *out,
885                                    int (*okfn)(struct sk_buff *))
886 {
887         if (skb->nf_bridge &&
888             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
889                 return NF_STOP;
890         }
891
892         return NF_ACCEPT;
893 }
894
895 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
896  * br_dev_queue_push_xmit is called afterwards */
897 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
898         {
899                 .hook = br_nf_pre_routing,
900                 .owner = THIS_MODULE,
901                 .pf = PF_BRIDGE,
902                 .hooknum = NF_BR_PRE_ROUTING,
903                 .priority = NF_BR_PRI_BRNF,
904         },
905         {
906                 .hook = br_nf_local_in,
907                 .owner = THIS_MODULE,
908                 .pf = PF_BRIDGE,
909                 .hooknum = NF_BR_LOCAL_IN,
910                 .priority = NF_BR_PRI_BRNF,
911         },
912         {
913                 .hook = br_nf_forward_ip,
914                 .owner = THIS_MODULE,
915                 .pf = PF_BRIDGE,
916                 .hooknum = NF_BR_FORWARD,
917                 .priority = NF_BR_PRI_BRNF - 1,
918         },
919         {
920                 .hook = br_nf_forward_arp,
921                 .owner = THIS_MODULE,
922                 .pf = PF_BRIDGE,
923                 .hooknum = NF_BR_FORWARD,
924                 .priority = NF_BR_PRI_BRNF,
925         },
926         {
927                 .hook = br_nf_post_routing,
928                 .owner = THIS_MODULE,
929                 .pf = PF_BRIDGE,
930                 .hooknum = NF_BR_POST_ROUTING,
931                 .priority = NF_BR_PRI_LAST,
932         },
933         {
934                 .hook = ip_sabotage_in,
935                 .owner = THIS_MODULE,
936                 .pf = PF_INET,
937                 .hooknum = NF_INET_PRE_ROUTING,
938                 .priority = NF_IP_PRI_FIRST,
939         },
940         {
941                 .hook = ip_sabotage_in,
942                 .owner = THIS_MODULE,
943                 .pf = PF_INET6,
944                 .hooknum = NF_INET_PRE_ROUTING,
945                 .priority = NF_IP6_PRI_FIRST,
946         },
947 };
948
949 #ifdef CONFIG_SYSCTL
950 static
951 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
952                             void __user * buffer, size_t * lenp, loff_t * ppos)
953 {
954         int ret;
955
956         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
957
958         if (write && *(int *)(ctl->data))
959                 *(int *)(ctl->data) = 1;
960         return ret;
961 }
962
963 static ctl_table brnf_table[] = {
964         {
965                 .procname       = "bridge-nf-call-arptables",
966                 .data           = &brnf_call_arptables,
967                 .maxlen         = sizeof(int),
968                 .mode           = 0644,
969                 .proc_handler   = brnf_sysctl_call_tables,
970         },
971         {
972                 .procname       = "bridge-nf-call-iptables",
973                 .data           = &brnf_call_iptables,
974                 .maxlen         = sizeof(int),
975                 .mode           = 0644,
976                 .proc_handler   = brnf_sysctl_call_tables,
977         },
978         {
979                 .procname       = "bridge-nf-call-ip6tables",
980                 .data           = &brnf_call_ip6tables,
981                 .maxlen         = sizeof(int),
982                 .mode           = 0644,
983                 .proc_handler   = brnf_sysctl_call_tables,
984         },
985         {
986                 .procname       = "bridge-nf-filter-vlan-tagged",
987                 .data           = &brnf_filter_vlan_tagged,
988                 .maxlen         = sizeof(int),
989                 .mode           = 0644,
990                 .proc_handler   = brnf_sysctl_call_tables,
991         },
992         {
993                 .procname       = "bridge-nf-filter-pppoe-tagged",
994                 .data           = &brnf_filter_pppoe_tagged,
995                 .maxlen         = sizeof(int),
996                 .mode           = 0644,
997                 .proc_handler   = brnf_sysctl_call_tables,
998         },
999         { }
1000 };
1001
1002 static struct ctl_path brnf_path[] = {
1003         { .procname = "net", },
1004         { .procname = "bridge", },
1005         { }
1006 };
1007 #endif
1008
1009 int __init br_netfilter_init(void)
1010 {
1011         int ret;
1012
1013         ret = dst_entries_init(&fake_dst_ops);
1014         if (ret < 0)
1015                 return ret;
1016
1017         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1018         if (ret < 0) {
1019                 dst_entries_destroy(&fake_dst_ops);
1020                 return ret;
1021         }
1022 #ifdef CONFIG_SYSCTL
1023         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1024         if (brnf_sysctl_header == NULL) {
1025                 printk(KERN_WARNING
1026                        "br_netfilter: can't register to sysctl.\n");
1027                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1028                 dst_entries_destroy(&fake_dst_ops);
1029                 return -ENOMEM;
1030         }
1031 #endif
1032         printk(KERN_NOTICE "Bridge firewalling registered\n");
1033         return 0;
1034 }
1035
1036 void br_netfilter_fini(void)
1037 {
1038         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1039 #ifdef CONFIG_SYSCTL
1040         unregister_sysctl_table(brnf_sysctl_header);
1041 #endif
1042         dst_entries_destroy(&fake_dst_ops);
1043 }