microblaze: uaccess: fix put_user and get_user macros
[linux-2.6.git] / net / bridge / br_netfilter.c
1 /*
2  *      Handle firewalling
3  *      Linux ethernet bridge
4  *
5  *      Authors:
6  *      Lennert Buytenhek               <buytenh@gnu.org>
7  *      Bart De Schuymer (maintainer)   <bdschuym@pandora.be>
8  *
9  *      Changes:
10  *      Apr 29 2003: physdev module support (bdschuym)
11  *      Jun 19 2003: let arptables see bridged ARP traffic (bdschuym)
12  *      Oct 06 2003: filter encapsulated IP/ARP VLAN traffic on untagged bridge
13  *                   (bdschuym)
14  *      Sep 01 2004: add IPv6 filtering (bdschuym)
15  *
16  *      This program is free software; you can redistribute it and/or
17  *      modify it under the terms of the GNU General Public License
18  *      as published by the Free Software Foundation; either version
19  *      2 of the License, or (at your option) any later version.
20  *
21  *      Lennert dedicates this file to Kerstin Wurdinger.
22  */
23
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/ip.h>
27 #include <linux/netdevice.h>
28 #include <linux/skbuff.h>
29 #include <linux/if_arp.h>
30 #include <linux/if_ether.h>
31 #include <linux/if_vlan.h>
32 #include <linux/if_pppox.h>
33 #include <linux/ppp_defs.h>
34 #include <linux/netfilter_bridge.h>
35 #include <linux/netfilter_ipv4.h>
36 #include <linux/netfilter_ipv6.h>
37 #include <linux/netfilter_arp.h>
38 #include <linux/in_route.h>
39 #include <linux/inetdevice.h>
40
41 #include <net/ip.h>
42 #include <net/ipv6.h>
43 #include <net/route.h>
44
45 #include <asm/uaccess.h>
46 #include "br_private.h"
47 #ifdef CONFIG_SYSCTL
48 #include <linux/sysctl.h>
49 #endif
50
51 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
52                                  (skb->nf_bridge->data))->daddr.ipv4)
53 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = ip_hdr(skb)->daddr)
54 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
55
56 #ifdef CONFIG_SYSCTL
57 static struct ctl_table_header *brnf_sysctl_header;
58 static int brnf_call_iptables __read_mostly = 1;
59 static int brnf_call_ip6tables __read_mostly = 1;
60 static int brnf_call_arptables __read_mostly = 1;
61 static int brnf_filter_vlan_tagged __read_mostly = 0;
62 static int brnf_filter_pppoe_tagged __read_mostly = 0;
63 #else
64 #define brnf_filter_vlan_tagged 0
65 #define brnf_filter_pppoe_tagged 0
66 #endif
67
68 static inline __be16 vlan_proto(const struct sk_buff *skb)
69 {
70         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
71 }
72
73 #define IS_VLAN_IP(skb) \
74         (skb->protocol == htons(ETH_P_8021Q) && \
75          vlan_proto(skb) == htons(ETH_P_IP) &&  \
76          brnf_filter_vlan_tagged)
77
78 #define IS_VLAN_IPV6(skb) \
79         (skb->protocol == htons(ETH_P_8021Q) && \
80          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
81          brnf_filter_vlan_tagged)
82
83 #define IS_VLAN_ARP(skb) \
84         (skb->protocol == htons(ETH_P_8021Q) && \
85          vlan_proto(skb) == htons(ETH_P_ARP) && \
86          brnf_filter_vlan_tagged)
87
88 static inline __be16 pppoe_proto(const struct sk_buff *skb)
89 {
90         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
91                             sizeof(struct pppoe_hdr)));
92 }
93
94 #define IS_PPPOE_IP(skb) \
95         (skb->protocol == htons(ETH_P_PPP_SES) && \
96          pppoe_proto(skb) == htons(PPP_IP) && \
97          brnf_filter_pppoe_tagged)
98
99 #define IS_PPPOE_IPV6(skb) \
100         (skb->protocol == htons(ETH_P_PPP_SES) && \
101          pppoe_proto(skb) == htons(PPP_IPV6) && \
102          brnf_filter_pppoe_tagged)
103
104 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
105 {
106 }
107
108 static struct dst_ops fake_dst_ops = {
109         .family =               AF_INET,
110         .protocol =             cpu_to_be16(ETH_P_IP),
111         .update_pmtu =          fake_update_pmtu,
112         .entries =              ATOMIC_INIT(0),
113 };
114
115 /*
116  * Initialize bogus route table used to keep netfilter happy.
117  * Currently, we fill in the PMTU entry because netfilter
118  * refragmentation needs it, and the rt_flags entry because
119  * ipt_REJECT needs it.  Future netfilter modules might
120  * require us to fill additional fields.
121  */
122 void br_netfilter_rtable_init(struct net_bridge *br)
123 {
124         struct rtable *rt = &br->fake_rtable;
125
126         atomic_set(&rt->u.dst.__refcnt, 1);
127         rt->u.dst.dev = br->dev;
128         rt->u.dst.path = &rt->u.dst;
129         rt->u.dst.metrics[RTAX_MTU - 1] = 1500;
130         rt->u.dst.flags = DST_NOXFRM;
131         rt->u.dst.ops = &fake_dst_ops;
132 }
133
134 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
135 {
136         struct net_bridge_port *port = rcu_dereference(dev->br_port);
137
138         return port ? &port->br->fake_rtable : NULL;
139 }
140
141 static inline struct net_device *bridge_parent(const struct net_device *dev)
142 {
143         struct net_bridge_port *port = rcu_dereference(dev->br_port);
144
145         return port ? port->br->dev : NULL;
146 }
147
148 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
149 {
150         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
151         if (likely(skb->nf_bridge))
152                 atomic_set(&(skb->nf_bridge->use), 1);
153
154         return skb->nf_bridge;
155 }
156
157 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
158 {
159         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
160
161         if (atomic_read(&nf_bridge->use) > 1) {
162                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
163
164                 if (tmp) {
165                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
166                         atomic_set(&tmp->use, 1);
167                         nf_bridge_put(nf_bridge);
168                 }
169                 nf_bridge = tmp;
170         }
171         return nf_bridge;
172 }
173
174 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
175 {
176         unsigned int len = nf_bridge_encap_header_len(skb);
177
178         skb_push(skb, len);
179         skb->network_header -= len;
180 }
181
182 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
183 {
184         unsigned int len = nf_bridge_encap_header_len(skb);
185
186         skb_pull(skb, len);
187         skb->network_header += len;
188 }
189
190 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
191 {
192         unsigned int len = nf_bridge_encap_header_len(skb);
193
194         skb_pull_rcsum(skb, len);
195         skb->network_header += len;
196 }
197
198 static inline void nf_bridge_save_header(struct sk_buff *skb)
199 {
200         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
201
202         skb_copy_from_linear_data_offset(skb, -header_size,
203                                          skb->nf_bridge->data, header_size);
204 }
205
206 /*
207  * When forwarding bridge frames, we save a copy of the original
208  * header before processing.
209  */
210 int nf_bridge_copy_header(struct sk_buff *skb)
211 {
212         int err;
213         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
214
215         err = skb_cow_head(skb, header_size);
216         if (err)
217                 return err;
218
219         skb_copy_to_linear_data_offset(skb, -header_size,
220                                        skb->nf_bridge->data, header_size);
221         __skb_push(skb, nf_bridge_encap_header_len(skb));
222         return 0;
223 }
224
225 /* PF_BRIDGE/PRE_ROUTING *********************************************/
226 /* Undo the changes made for ip6tables PREROUTING and continue the
227  * bridge PRE_ROUTING hook. */
228 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
229 {
230         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
231         struct rtable *rt;
232
233         if (nf_bridge->mask & BRNF_PKT_TYPE) {
234                 skb->pkt_type = PACKET_OTHERHOST;
235                 nf_bridge->mask ^= BRNF_PKT_TYPE;
236         }
237         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
238
239         rt = bridge_parent_rtable(nf_bridge->physindev);
240         if (!rt) {
241                 kfree_skb(skb);
242                 return 0;
243         }
244         dst_hold(&rt->u.dst);
245         skb_dst_set(skb, &rt->u.dst);
246
247         skb->dev = nf_bridge->physindev;
248         nf_bridge_push_encap_header(skb);
249         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
250                        br_handle_frame_finish, 1);
251
252         return 0;
253 }
254
255 static void __br_dnat_complain(void)
256 {
257         static unsigned long last_complaint;
258
259         if (jiffies - last_complaint >= 5 * HZ) {
260                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
261                        "forwarding to be enabled\n");
262                 last_complaint = jiffies;
263         }
264 }
265
266 /* This requires some explaining. If DNAT has taken place,
267  * we will need to fix up the destination Ethernet address,
268  * and this is a tricky process.
269  *
270  * There are two cases to consider:
271  * 1. The packet was DNAT'ed to a device in the same bridge
272  *    port group as it was received on. We can still bridge
273  *    the packet.
274  * 2. The packet was DNAT'ed to a different device, either
275  *    a non-bridged device or another bridge port group.
276  *    The packet will need to be routed.
277  *
278  * The correct way of distinguishing between these two cases is to
279  * call ip_route_input() and to look at skb->dst->dev, which is
280  * changed to the destination device if ip_route_input() succeeds.
281  *
282  * Let us first consider the case that ip_route_input() succeeds:
283  *
284  * If skb->dst->dev equals the logical bridge device the packet
285  * came in on, we can consider this bridging. The packet is passed
286  * through the neighbour output function to build a new destination
287  * MAC address, which will make the packet enter br_nf_local_out()
288  * not much later. In that function it is assured that the iptables
289  * FORWARD chain is traversed for the packet.
290  *
291  * Otherwise, the packet is considered to be routed and we just
292  * change the destination MAC address so that the packet will
293  * later be passed up to the IP stack to be routed. For a redirected
294  * packet, ip_route_input() will give back the localhost as output device,
295  * which differs from the bridge device.
296  *
297  * Let us now consider the case that ip_route_input() fails:
298  *
299  * This can be because the destination address is martian, in which case
300  * the packet will be dropped.
301  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
302  * will fail, while __ip_route_output_key() will return success. The source
303  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
304  * thinks we're handling a locally generated packet and won't care
305  * if IP forwarding is allowed. We send a warning message to the users's
306  * log telling her to put IP forwarding on.
307  *
308  * ip_route_input() will also fail if there is no route available.
309  * In that case we just drop the packet.
310  *
311  * --Lennert, 20020411
312  * --Bart, 20020416 (updated)
313  * --Bart, 20021007 (updated)
314  * --Bart, 20062711 (updated) */
315 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
316 {
317         if (skb->pkt_type == PACKET_OTHERHOST) {
318                 skb->pkt_type = PACKET_HOST;
319                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
320         }
321         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
322
323         skb->dev = bridge_parent(skb->dev);
324         if (skb->dev) {
325                 struct dst_entry *dst = skb_dst(skb);
326
327                 nf_bridge_pull_encap_header(skb);
328
329                 if (dst->hh)
330                         return neigh_hh_output(dst->hh, skb);
331                 else if (dst->neighbour)
332                         return dst->neighbour->output(skb);
333         }
334         kfree_skb(skb);
335         return 0;
336 }
337
338 static int br_nf_pre_routing_finish(struct sk_buff *skb)
339 {
340         struct net_device *dev = skb->dev;
341         struct iphdr *iph = ip_hdr(skb);
342         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
343         struct rtable *rt;
344         int err;
345
346         if (nf_bridge->mask & BRNF_PKT_TYPE) {
347                 skb->pkt_type = PACKET_OTHERHOST;
348                 nf_bridge->mask ^= BRNF_PKT_TYPE;
349         }
350         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
351         if (dnat_took_place(skb)) {
352                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
353                         struct flowi fl = {
354                                 .nl_u = {
355                                         .ip4_u = {
356                                                  .daddr = iph->daddr,
357                                                  .saddr = 0,
358                                                  .tos = RT_TOS(iph->tos) },
359                                 },
360                                 .proto = 0,
361                         };
362                         struct in_device *in_dev = __in_dev_get_rcu(dev);
363
364                         /* If err equals -EHOSTUNREACH the error is due to a
365                          * martian destination or due to the fact that
366                          * forwarding is disabled. For most martian packets,
367                          * ip_route_output_key() will fail. It won't fail for 2 types of
368                          * martian destinations: loopback destinations and destination
369                          * 0.0.0.0. In both cases the packet will be dropped because the
370                          * destination is the loopback device and not the bridge. */
371                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
372                                 goto free_skb;
373
374                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
375                                 /* - Bridged-and-DNAT'ed traffic doesn't
376                                  *   require ip_forwarding. */
377                                 if (((struct dst_entry *)rt)->dev == dev) {
378                                         skb_dst_set(skb, (struct dst_entry *)rt);
379                                         goto bridged_dnat;
380                                 }
381                                 /* we are sure that forwarding is disabled, so printing
382                                  * this message is no problem. Note that the packet could
383                                  * still have a martian destination address, in which case
384                                  * the packet could be dropped even if forwarding were enabled */
385                                 __br_dnat_complain();
386                                 dst_release((struct dst_entry *)rt);
387                         }
388 free_skb:
389                         kfree_skb(skb);
390                         return 0;
391                 } else {
392                         if (skb_dst(skb)->dev == dev) {
393 bridged_dnat:
394                                 /* Tell br_nf_local_out this is a
395                                  * bridged frame */
396                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
397                                 skb->dev = nf_bridge->physindev;
398                                 nf_bridge_push_encap_header(skb);
399                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
400                                                skb, skb->dev, NULL,
401                                                br_nf_pre_routing_finish_bridge,
402                                                1);
403                                 return 0;
404                         }
405                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
406                         skb->pkt_type = PACKET_HOST;
407                 }
408         } else {
409                 rt = bridge_parent_rtable(nf_bridge->physindev);
410                 if (!rt) {
411                         kfree_skb(skb);
412                         return 0;
413                 }
414                 dst_hold(&rt->u.dst);
415                 skb_dst_set(skb, &rt->u.dst);
416         }
417
418         skb->dev = nf_bridge->physindev;
419         nf_bridge_push_encap_header(skb);
420         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
421                        br_handle_frame_finish, 1);
422
423         return 0;
424 }
425
426 /* Some common code for IPv4/IPv6 */
427 static struct net_device *setup_pre_routing(struct sk_buff *skb)
428 {
429         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
430
431         if (skb->pkt_type == PACKET_OTHERHOST) {
432                 skb->pkt_type = PACKET_HOST;
433                 nf_bridge->mask |= BRNF_PKT_TYPE;
434         }
435
436         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
437         nf_bridge->physindev = skb->dev;
438         skb->dev = bridge_parent(skb->dev);
439
440         return skb->dev;
441 }
442
443 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
444 static int check_hbh_len(struct sk_buff *skb)
445 {
446         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
447         u32 pkt_len;
448         const unsigned char *nh = skb_network_header(skb);
449         int off = raw - nh;
450         int len = (raw[1] + 1) << 3;
451
452         if ((raw + len) - skb->data > skb_headlen(skb))
453                 goto bad;
454
455         off += 2;
456         len -= 2;
457
458         while (len > 0) {
459                 int optlen = nh[off + 1] + 2;
460
461                 switch (nh[off]) {
462                 case IPV6_TLV_PAD0:
463                         optlen = 1;
464                         break;
465
466                 case IPV6_TLV_PADN:
467                         break;
468
469                 case IPV6_TLV_JUMBO:
470                         if (nh[off + 1] != 4 || (off & 3) != 2)
471                                 goto bad;
472                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
473                         if (pkt_len <= IPV6_MAXPLEN ||
474                             ipv6_hdr(skb)->payload_len)
475                                 goto bad;
476                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
477                                 goto bad;
478                         if (pskb_trim_rcsum(skb,
479                                             pkt_len + sizeof(struct ipv6hdr)))
480                                 goto bad;
481                         nh = skb_network_header(skb);
482                         break;
483                 default:
484                         if (optlen > len)
485                                 goto bad;
486                         break;
487                 }
488                 off += optlen;
489                 len -= optlen;
490         }
491         if (len == 0)
492                 return 0;
493 bad:
494         return -1;
495
496 }
497
498 /* Replicate the checks that IPv6 does on packet reception and pass the packet
499  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
500 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
501                                            struct sk_buff *skb,
502                                            const struct net_device *in,
503                                            const struct net_device *out,
504                                            int (*okfn)(struct sk_buff *))
505 {
506         struct ipv6hdr *hdr;
507         u32 pkt_len;
508
509         if (skb->len < sizeof(struct ipv6hdr))
510                 goto inhdr_error;
511
512         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
513                 goto inhdr_error;
514
515         hdr = ipv6_hdr(skb);
516
517         if (hdr->version != 6)
518                 goto inhdr_error;
519
520         pkt_len = ntohs(hdr->payload_len);
521
522         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
523                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
524                         goto inhdr_error;
525                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
526                         goto inhdr_error;
527         }
528         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
529                 goto inhdr_error;
530
531         nf_bridge_put(skb->nf_bridge);
532         if (!nf_bridge_alloc(skb))
533                 return NF_DROP;
534         if (!setup_pre_routing(skb))
535                 return NF_DROP;
536
537         NF_HOOK(PF_INET6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
538                 br_nf_pre_routing_finish_ipv6);
539
540         return NF_STOLEN;
541
542 inhdr_error:
543         return NF_DROP;
544 }
545
546 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
547  * Replicate the checks that IPv4 does on packet reception.
548  * Set skb->dev to the bridge device (i.e. parent of the
549  * receiving device) to make netfilter happy, the REDIRECT
550  * target in particular.  Save the original destination IP
551  * address to be able to detect DNAT afterwards. */
552 static unsigned int br_nf_pre_routing(unsigned int hook, 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 iphdr *iph;
558         __u32 len = nf_bridge_encap_header_len(skb);
559
560         if (unlikely(!pskb_may_pull(skb, len)))
561                 goto out;
562
563         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
564             IS_PPPOE_IPV6(skb)) {
565 #ifdef CONFIG_SYSCTL
566                 if (!brnf_call_ip6tables)
567                         return NF_ACCEPT;
568 #endif
569                 nf_bridge_pull_encap_header_rcsum(skb);
570                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
571         }
572 #ifdef CONFIG_SYSCTL
573         if (!brnf_call_iptables)
574                 return NF_ACCEPT;
575 #endif
576
577         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
578             !IS_PPPOE_IP(skb))
579                 return NF_ACCEPT;
580
581         nf_bridge_pull_encap_header_rcsum(skb);
582
583         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
584                 goto inhdr_error;
585
586         iph = ip_hdr(skb);
587         if (iph->ihl < 5 || iph->version != 4)
588                 goto inhdr_error;
589
590         if (!pskb_may_pull(skb, 4 * iph->ihl))
591                 goto inhdr_error;
592
593         iph = ip_hdr(skb);
594         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
595                 goto inhdr_error;
596
597         len = ntohs(iph->tot_len);
598         if (skb->len < len || len < 4 * iph->ihl)
599                 goto inhdr_error;
600
601         pskb_trim_rcsum(skb, len);
602
603         nf_bridge_put(skb->nf_bridge);
604         if (!nf_bridge_alloc(skb))
605                 return NF_DROP;
606         if (!setup_pre_routing(skb))
607                 return NF_DROP;
608         store_orig_dstaddr(skb);
609
610         NF_HOOK(PF_INET, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
611                 br_nf_pre_routing_finish);
612
613         return NF_STOLEN;
614
615 inhdr_error:
616 //      IP_INC_STATS_BH(IpInHdrErrors);
617 out:
618         return NF_DROP;
619 }
620
621
622 /* PF_BRIDGE/LOCAL_IN ************************************************/
623 /* The packet is locally destined, which requires a real
624  * dst_entry, so detach the fake one.  On the way up, the
625  * packet would pass through PRE_ROUTING again (which already
626  * took place when the packet entered the bridge), but we
627  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
628  * prevent this from happening. */
629 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
630                                    const struct net_device *in,
631                                    const struct net_device *out,
632                                    int (*okfn)(struct sk_buff *))
633 {
634         struct rtable *rt = skb_rtable(skb);
635
636         if (rt && rt == bridge_parent_rtable(in))
637                 skb_dst_drop(skb);
638
639         return NF_ACCEPT;
640 }
641
642 /* PF_BRIDGE/FORWARD *************************************************/
643 static int br_nf_forward_finish(struct sk_buff *skb)
644 {
645         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
646         struct net_device *in;
647
648         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
649                 in = nf_bridge->physindev;
650                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
651                         skb->pkt_type = PACKET_OTHERHOST;
652                         nf_bridge->mask ^= BRNF_PKT_TYPE;
653                 }
654         } else {
655                 in = *((struct net_device **)(skb->cb));
656         }
657         nf_bridge_push_encap_header(skb);
658         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
659                        skb->dev, br_forward_finish, 1);
660         return 0;
661 }
662
663 /* This is the 'purely bridged' case.  For IP, we pass the packet to
664  * netfilter with indev and outdev set to the bridge device,
665  * but we are still able to filter on the 'real' indev/outdev
666  * because of the physdev module. For ARP, indev and outdev are the
667  * bridge ports. */
668 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
669                                      const struct net_device *in,
670                                      const struct net_device *out,
671                                      int (*okfn)(struct sk_buff *))
672 {
673         struct nf_bridge_info *nf_bridge;
674         struct net_device *parent;
675         u_int8_t pf;
676
677         if (!skb->nf_bridge)
678                 return NF_ACCEPT;
679
680         /* Need exclusive nf_bridge_info since we might have multiple
681          * different physoutdevs. */
682         if (!nf_bridge_unshare(skb))
683                 return NF_DROP;
684
685         parent = bridge_parent(out);
686         if (!parent)
687                 return NF_DROP;
688
689         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
690             IS_PPPOE_IP(skb))
691                 pf = PF_INET;
692         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
693                  IS_PPPOE_IPV6(skb))
694                 pf = PF_INET6;
695         else
696                 return NF_ACCEPT;
697
698         nf_bridge_pull_encap_header(skb);
699
700         nf_bridge = skb->nf_bridge;
701         if (skb->pkt_type == PACKET_OTHERHOST) {
702                 skb->pkt_type = PACKET_HOST;
703                 nf_bridge->mask |= BRNF_PKT_TYPE;
704         }
705
706         /* The physdev module checks on this */
707         nf_bridge->mask |= BRNF_BRIDGED;
708         nf_bridge->physoutdev = skb->dev;
709
710         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
711                 br_nf_forward_finish);
712
713         return NF_STOLEN;
714 }
715
716 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
717                                       const struct net_device *in,
718                                       const struct net_device *out,
719                                       int (*okfn)(struct sk_buff *))
720 {
721         struct net_device **d = (struct net_device **)(skb->cb);
722
723 #ifdef CONFIG_SYSCTL
724         if (!brnf_call_arptables)
725                 return NF_ACCEPT;
726 #endif
727
728         if (skb->protocol != htons(ETH_P_ARP)) {
729                 if (!IS_VLAN_ARP(skb))
730                         return NF_ACCEPT;
731                 nf_bridge_pull_encap_header(skb);
732         }
733
734         if (arp_hdr(skb)->ar_pln != 4) {
735                 if (IS_VLAN_ARP(skb))
736                         nf_bridge_push_encap_header(skb);
737                 return NF_ACCEPT;
738         }
739         *d = (struct net_device *)in;
740         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
741                 (struct net_device *)out, br_nf_forward_finish);
742
743         return NF_STOLEN;
744 }
745
746 /* PF_BRIDGE/LOCAL_OUT ***********************************************
747  *
748  * This function sees both locally originated IP packets and forwarded
749  * IP packets (in both cases the destination device is a bridge
750  * device). It also sees bridged-and-DNAT'ed packets.
751  *
752  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
753  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
754  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
755  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
756  * will be executed.
757  */
758 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff *skb,
759                                     const struct net_device *in,
760                                     const struct net_device *out,
761                                     int (*okfn)(struct sk_buff *))
762 {
763         struct net_device *realindev;
764         struct nf_bridge_info *nf_bridge;
765
766         if (!skb->nf_bridge)
767                 return NF_ACCEPT;
768
769         /* Need exclusive nf_bridge_info since we might have multiple
770          * different physoutdevs. */
771         if (!nf_bridge_unshare(skb))
772                 return NF_DROP;
773
774         nf_bridge = skb->nf_bridge;
775         if (!(nf_bridge->mask & BRNF_BRIDGED_DNAT))
776                 return NF_ACCEPT;
777
778         /* Bridged, take PF_BRIDGE/FORWARD.
779          * (see big note in front of br_nf_pre_routing_finish) */
780         nf_bridge->physoutdev = skb->dev;
781         realindev = nf_bridge->physindev;
782
783         if (nf_bridge->mask & BRNF_PKT_TYPE) {
784                 skb->pkt_type = PACKET_OTHERHOST;
785                 nf_bridge->mask ^= BRNF_PKT_TYPE;
786         }
787         nf_bridge_push_encap_header(skb);
788
789         NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev, skb->dev,
790                 br_forward_finish);
791         return NF_STOLEN;
792 }
793
794 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
795 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
796 {
797         if (skb->nfct != NULL &&
798             (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb)) &&
799             skb->len > skb->dev->mtu &&
800             !skb_is_gso(skb))
801                 return ip_fragment(skb, br_dev_queue_push_xmit);
802         else
803                 return br_dev_queue_push_xmit(skb);
804 }
805 #else
806 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
807 {
808         return br_dev_queue_push_xmit(skb);
809 }
810 #endif
811
812 /* PF_BRIDGE/POST_ROUTING ********************************************/
813 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
814                                        const struct net_device *in,
815                                        const struct net_device *out,
816                                        int (*okfn)(struct sk_buff *))
817 {
818         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
819         struct net_device *realoutdev = bridge_parent(skb->dev);
820         u_int8_t pf;
821
822 #ifdef CONFIG_NETFILTER_DEBUG
823         /* Be very paranoid. This probably won't happen anymore, but let's
824          * keep the check just to be sure... */
825         if (skb_mac_header(skb) < skb->head ||
826             skb_mac_header(skb) + ETH_HLEN > skb->data) {
827                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
828                        "bad mac.raw pointer.\n");
829                 goto print_error;
830         }
831 #endif
832
833         if (!nf_bridge)
834                 return NF_ACCEPT;
835
836         if (!(nf_bridge->mask & (BRNF_BRIDGED | BRNF_BRIDGED_DNAT)))
837                 return NF_ACCEPT;
838
839         if (!realoutdev)
840                 return NF_DROP;
841
842         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
843             IS_PPPOE_IP(skb))
844                 pf = PF_INET;
845         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
846                  IS_PPPOE_IPV6(skb))
847                 pf = PF_INET6;
848         else
849                 return NF_ACCEPT;
850
851 #ifdef CONFIG_NETFILTER_DEBUG
852         if (skb_dst(skb) == NULL) {
853                 printk(KERN_INFO "br_netfilter post_routing: skb->dst == NULL\n");
854                 goto print_error;
855         }
856 #endif
857
858         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
859          * about the value of skb->pkt_type. */
860         if (skb->pkt_type == PACKET_OTHERHOST) {
861                 skb->pkt_type = PACKET_HOST;
862                 nf_bridge->mask |= BRNF_PKT_TYPE;
863         }
864
865         nf_bridge_pull_encap_header(skb);
866         nf_bridge_save_header(skb);
867
868         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
869                 br_nf_dev_queue_xmit);
870
871         return NF_STOLEN;
872
873 #ifdef CONFIG_NETFILTER_DEBUG
874 print_error:
875         if (skb->dev != NULL) {
876                 printk("[%s]", skb->dev->name);
877                 if (realoutdev)
878                         printk("[%s]", realoutdev->name);
879         }
880         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb_mac_header(skb),
881                skb->data);
882         dump_stack();
883         return NF_ACCEPT;
884 #endif
885 }
886
887 /* IP/SABOTAGE *****************************************************/
888 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
889  * for the second time. */
890 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
891                                    const struct net_device *in,
892                                    const struct net_device *out,
893                                    int (*okfn)(struct sk_buff *))
894 {
895         if (skb->nf_bridge &&
896             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
897                 return NF_STOP;
898         }
899
900         return NF_ACCEPT;
901 }
902
903 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
904  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
905  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
906  * ip_refrag() can return NF_STOLEN. */
907 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
908         {
909                 .hook = br_nf_pre_routing,
910                 .owner = THIS_MODULE,
911                 .pf = PF_BRIDGE,
912                 .hooknum = NF_BR_PRE_ROUTING,
913                 .priority = NF_BR_PRI_BRNF,
914         },
915         {
916                 .hook = br_nf_local_in,
917                 .owner = THIS_MODULE,
918                 .pf = PF_BRIDGE,
919                 .hooknum = NF_BR_LOCAL_IN,
920                 .priority = NF_BR_PRI_BRNF,
921         },
922         {
923                 .hook = br_nf_forward_ip,
924                 .owner = THIS_MODULE,
925                 .pf = PF_BRIDGE,
926                 .hooknum = NF_BR_FORWARD,
927                 .priority = NF_BR_PRI_BRNF - 1,
928         },
929         {
930                 .hook = br_nf_forward_arp,
931                 .owner = THIS_MODULE,
932                 .pf = PF_BRIDGE,
933                 .hooknum = NF_BR_FORWARD,
934                 .priority = NF_BR_PRI_BRNF,
935         },
936         {
937                 .hook = br_nf_local_out,
938                 .owner = THIS_MODULE,
939                 .pf = PF_BRIDGE,
940                 .hooknum = NF_BR_LOCAL_OUT,
941                 .priority = NF_BR_PRI_FIRST,
942         },
943         {
944                 .hook = br_nf_post_routing,
945                 .owner = THIS_MODULE,
946                 .pf = PF_BRIDGE,
947                 .hooknum = NF_BR_POST_ROUTING,
948                 .priority = NF_BR_PRI_LAST,
949         },
950         {
951                 .hook = ip_sabotage_in,
952                 .owner = THIS_MODULE,
953                 .pf = PF_INET,
954                 .hooknum = NF_INET_PRE_ROUTING,
955                 .priority = NF_IP_PRI_FIRST,
956         },
957         {
958                 .hook = ip_sabotage_in,
959                 .owner = THIS_MODULE,
960                 .pf = PF_INET6,
961                 .hooknum = NF_INET_PRE_ROUTING,
962                 .priority = NF_IP6_PRI_FIRST,
963         },
964 };
965
966 #ifdef CONFIG_SYSCTL
967 static
968 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
969                             void __user * buffer, size_t * lenp, loff_t * ppos)
970 {
971         int ret;
972
973         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
974
975         if (write && *(int *)(ctl->data))
976                 *(int *)(ctl->data) = 1;
977         return ret;
978 }
979
980 static ctl_table brnf_table[] = {
981         {
982                 .procname       = "bridge-nf-call-arptables",
983                 .data           = &brnf_call_arptables,
984                 .maxlen         = sizeof(int),
985                 .mode           = 0644,
986                 .proc_handler   = brnf_sysctl_call_tables,
987         },
988         {
989                 .procname       = "bridge-nf-call-iptables",
990                 .data           = &brnf_call_iptables,
991                 .maxlen         = sizeof(int),
992                 .mode           = 0644,
993                 .proc_handler   = brnf_sysctl_call_tables,
994         },
995         {
996                 .procname       = "bridge-nf-call-ip6tables",
997                 .data           = &brnf_call_ip6tables,
998                 .maxlen         = sizeof(int),
999                 .mode           = 0644,
1000                 .proc_handler   = brnf_sysctl_call_tables,
1001         },
1002         {
1003                 .procname       = "bridge-nf-filter-vlan-tagged",
1004                 .data           = &brnf_filter_vlan_tagged,
1005                 .maxlen         = sizeof(int),
1006                 .mode           = 0644,
1007                 .proc_handler   = brnf_sysctl_call_tables,
1008         },
1009         {
1010                 .procname       = "bridge-nf-filter-pppoe-tagged",
1011                 .data           = &brnf_filter_pppoe_tagged,
1012                 .maxlen         = sizeof(int),
1013                 .mode           = 0644,
1014                 .proc_handler   = brnf_sysctl_call_tables,
1015         },
1016         { }
1017 };
1018
1019 static struct ctl_path brnf_path[] = {
1020         { .procname = "net", },
1021         { .procname = "bridge", },
1022         { }
1023 };
1024 #endif
1025
1026 int __init br_netfilter_init(void)
1027 {
1028         int ret;
1029
1030         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1031         if (ret < 0)
1032                 return ret;
1033 #ifdef CONFIG_SYSCTL
1034         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1035         if (brnf_sysctl_header == NULL) {
1036                 printk(KERN_WARNING
1037                        "br_netfilter: can't register to sysctl.\n");
1038                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1039                 return -ENOMEM;
1040         }
1041 #endif
1042         printk(KERN_NOTICE "Bridge firewalling registered\n");
1043         return 0;
1044 }
1045
1046 void br_netfilter_fini(void)
1047 {
1048         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1049 #ifdef CONFIG_SYSCTL
1050         unregister_sysctl_table(brnf_sysctl_header);
1051 #endif
1052 }