[BRIDGE]: netfilter dont use __constant_htons
[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/netfilter_bridge.h>
33 #include <linux/netfilter_ipv4.h>
34 #include <linux/netfilter_ipv6.h>
35 #include <linux/netfilter_arp.h>
36 #include <linux/in_route.h>
37
38 #include <net/ip.h>
39 #include <net/ipv6.h>
40 #include <net/route.h>
41
42 #include <asm/uaccess.h>
43 #include <asm/checksum.h>
44 #include "br_private.h"
45 #ifdef CONFIG_SYSCTL
46 #include <linux/sysctl.h>
47 #endif
48
49 #define skb_origaddr(skb)        (((struct bridge_skb_cb *) \
50                                  (skb->nf_bridge->data))->daddr.ipv4)
51 #define store_orig_dstaddr(skb)  (skb_origaddr(skb) = (skb)->nh.iph->daddr)
52 #define dnat_took_place(skb)     (skb_origaddr(skb) != (skb)->nh.iph->daddr)
53
54 #ifdef CONFIG_SYSCTL
55 static struct ctl_table_header *brnf_sysctl_header;
56 static int brnf_call_iptables = 1;
57 static int brnf_call_ip6tables = 1;
58 static int brnf_call_arptables = 1;
59 static int brnf_filter_vlan_tagged = 1;
60 #else
61 #define brnf_filter_vlan_tagged 1
62 #endif
63
64 #define IS_VLAN_IP (skb->protocol == htons(ETH_P_8021Q) &&    \
65         hdr->h_vlan_encapsulated_proto == htons(ETH_P_IP) &&  \
66         brnf_filter_vlan_tagged)
67 #define IS_VLAN_IPV6 (skb->protocol == htons(ETH_P_8021Q) &&    \
68         hdr->h_vlan_encapsulated_proto == htons(ETH_P_IPV6) &&  \
69         brnf_filter_vlan_tagged)
70 #define IS_VLAN_ARP (skb->protocol == htons(ETH_P_8021Q) &&   \
71         hdr->h_vlan_encapsulated_proto == htons(ETH_P_ARP) && \
72         brnf_filter_vlan_tagged)
73
74 /* We need these fake structures to make netfilter happy --
75  * lots of places assume that skb->dst != NULL, which isn't
76  * all that unreasonable.
77  *
78  * Currently, we fill in the PMTU entry because netfilter
79  * refragmentation needs it, and the rt_flags entry because
80  * ipt_REJECT needs it.  Future netfilter modules might
81  * require us to fill additional fields. */
82 static struct net_device __fake_net_device = {
83         .hard_header_len        = ETH_HLEN
84 };
85
86 static struct rtable __fake_rtable = {
87         .u = {
88                 .dst = {
89                         .__refcnt               = ATOMIC_INIT(1),
90                         .dev                    = &__fake_net_device,
91                         .path                   = &__fake_rtable.u.dst,
92                         .metrics                = {[RTAX_MTU - 1] = 1500},
93                         .flags                  = DST_NOXFRM,
94                 }
95         },
96         .rt_flags       = 0,
97 };
98
99 static inline struct net_device *bridge_parent(const struct net_device *dev)
100 {
101         struct net_bridge_port *port = rcu_dereference(dev->br_port);
102
103         return port ? port->br->dev : NULL;
104 }
105
106 /* PF_BRIDGE/PRE_ROUTING *********************************************/
107 /* Undo the changes made for ip6tables PREROUTING and continue the
108  * bridge PRE_ROUTING hook. */
109 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
110 {
111         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
112
113         if (nf_bridge->mask & BRNF_PKT_TYPE) {
114                 skb->pkt_type = PACKET_OTHERHOST;
115                 nf_bridge->mask ^= BRNF_PKT_TYPE;
116         }
117         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
118
119         skb->dst = (struct dst_entry *)&__fake_rtable;
120         dst_hold(skb->dst);
121
122         skb->dev = nf_bridge->physindev;
123         if (skb->protocol == htons(ETH_P_8021Q)) {
124                 skb_push(skb, VLAN_HLEN);
125                 skb->nh.raw -= VLAN_HLEN;
126         }
127         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
128                        br_handle_frame_finish, 1);
129
130         return 0;
131 }
132
133 static void __br_dnat_complain(void)
134 {
135         static unsigned long last_complaint;
136
137         if (jiffies - last_complaint >= 5 * HZ) {
138                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
139                        "forwarding to be enabled\n");
140                 last_complaint = jiffies;
141         }
142 }
143
144 /* This requires some explaining. If DNAT has taken place,
145  * we will need to fix up the destination Ethernet address,
146  * and this is a tricky process.
147  *
148  * There are two cases to consider:
149  * 1. The packet was DNAT'ed to a device in the same bridge
150  *    port group as it was received on. We can still bridge
151  *    the packet.
152  * 2. The packet was DNAT'ed to a different device, either
153  *    a non-bridged device or another bridge port group.
154  *    The packet will need to be routed.
155  *
156  * The correct way of distinguishing between these two cases is to
157  * call ip_route_input() and to look at skb->dst->dev, which is
158  * changed to the destination device if ip_route_input() succeeds.
159  *
160  * Let us first consider the case that ip_route_input() succeeds:
161  *
162  * If skb->dst->dev equals the logical bridge device the packet
163  * came in on, we can consider this bridging. We then call
164  * skb->dst->output() which will make the packet enter br_nf_local_out()
165  * not much later. In that function it is assured that the iptables
166  * FORWARD chain is traversed for the packet.
167  *
168  * Otherwise, the packet is considered to be routed and we just
169  * change the destination MAC address so that the packet will
170  * later be passed up to the IP stack to be routed.
171  *
172  * Let us now consider the case that ip_route_input() fails:
173  *
174  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
175  * will fail, while __ip_route_output_key() will return success. The source
176  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
177  * thinks we're handling a locally generated packet and won't care
178  * if IP forwarding is allowed. We send a warning message to the users's
179  * log telling her to put IP forwarding on.
180  *
181  * ip_route_input() will also fail if there is no route available.
182  * In that case we just drop the packet.
183  *
184  * --Lennert, 20020411
185  * --Bart, 20020416 (updated)
186  * --Bart, 20021007 (updated) */
187 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
188 {
189         if (skb->pkt_type == PACKET_OTHERHOST) {
190                 skb->pkt_type = PACKET_HOST;
191                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
192         }
193         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
194
195         skb->dev = bridge_parent(skb->dev);
196         if (!skb->dev)
197                 kfree_skb(skb);
198         else {
199                 if (skb->protocol == htons(ETH_P_8021Q)) {
200                         skb_pull(skb, VLAN_HLEN);
201                         skb->nh.raw += VLAN_HLEN;
202                 }
203                 skb->dst->output(skb);
204         }
205         return 0;
206 }
207
208 static int br_nf_pre_routing_finish(struct sk_buff *skb)
209 {
210         struct net_device *dev = skb->dev;
211         struct iphdr *iph = skb->nh.iph;
212         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
213
214         if (nf_bridge->mask & BRNF_PKT_TYPE) {
215                 skb->pkt_type = PACKET_OTHERHOST;
216                 nf_bridge->mask ^= BRNF_PKT_TYPE;
217         }
218         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
219
220         if (dnat_took_place(skb)) {
221                 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev)) {
222                         struct rtable *rt;
223                         struct flowi fl = {
224                                 .nl_u = {
225                                         .ip4_u = {
226                                                  .daddr = iph->daddr,
227                                                  .saddr = 0,
228                                                  .tos = RT_TOS(iph->tos) },
229                                 },
230                                 .proto = 0,
231                         };
232
233                         if (!ip_route_output_key(&rt, &fl)) {
234                                 /* - Bridged-and-DNAT'ed traffic doesn't
235                                  *   require ip_forwarding.
236                                  * - Deal with redirected traffic. */
237                                 if (((struct dst_entry *)rt)->dev == dev ||
238                                     rt->rt_type == RTN_LOCAL) {
239                                         skb->dst = (struct dst_entry *)rt;
240                                         goto bridged_dnat;
241                                 }
242                                 __br_dnat_complain();
243                                 dst_release((struct dst_entry *)rt);
244                         }
245                         kfree_skb(skb);
246                         return 0;
247                 } else {
248                         if (skb->dst->dev == dev) {
249 bridged_dnat:
250                                 /* Tell br_nf_local_out this is a
251                                  * bridged frame */
252                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
253                                 skb->dev = nf_bridge->physindev;
254                                 if (skb->protocol ==
255                                     htons(ETH_P_8021Q)) {
256                                         skb_push(skb, VLAN_HLEN);
257                                         skb->nh.raw -= VLAN_HLEN;
258                                 }
259                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
260                                                skb, skb->dev, NULL,
261                                                br_nf_pre_routing_finish_bridge,
262                                                1);
263                                 return 0;
264                         }
265                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
266                         skb->pkt_type = PACKET_HOST;
267                 }
268         } else {
269                 skb->dst = (struct dst_entry *)&__fake_rtable;
270                 dst_hold(skb->dst);
271         }
272
273         skb->dev = nf_bridge->physindev;
274         if (skb->protocol == htons(ETH_P_8021Q)) {
275                 skb_push(skb, VLAN_HLEN);
276                 skb->nh.raw -= VLAN_HLEN;
277         }
278         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
279                        br_handle_frame_finish, 1);
280
281         return 0;
282 }
283
284 /* Some common code for IPv4/IPv6 */
285 static struct net_device *setup_pre_routing(struct sk_buff *skb)
286 {
287         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
288
289         if (skb->pkt_type == PACKET_OTHERHOST) {
290                 skb->pkt_type = PACKET_HOST;
291                 nf_bridge->mask |= BRNF_PKT_TYPE;
292         }
293
294         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
295         nf_bridge->physindev = skb->dev;
296         skb->dev = bridge_parent(skb->dev);
297
298         return skb->dev;
299 }
300
301 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
302 static int check_hbh_len(struct sk_buff *skb)
303 {
304         unsigned char *raw = (u8 *) (skb->nh.ipv6h + 1);
305         u32 pkt_len;
306         int off = raw - skb->nh.raw;
307         int len = (raw[1] + 1) << 3;
308
309         if ((raw + len) - skb->data > skb_headlen(skb))
310                 goto bad;
311
312         off += 2;
313         len -= 2;
314
315         while (len > 0) {
316                 int optlen = skb->nh.raw[off + 1] + 2;
317
318                 switch (skb->nh.raw[off]) {
319                 case IPV6_TLV_PAD0:
320                         optlen = 1;
321                         break;
322
323                 case IPV6_TLV_PADN:
324                         break;
325
326                 case IPV6_TLV_JUMBO:
327                         if (skb->nh.raw[off + 1] != 4 || (off & 3) != 2)
328                                 goto bad;
329                         pkt_len = ntohl(*(u32 *) (skb->nh.raw + off + 2));
330                         if (pkt_len <= IPV6_MAXPLEN ||
331                             skb->nh.ipv6h->payload_len)
332                                 goto bad;
333                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
334                                 goto bad;
335                         if (pskb_trim_rcsum(skb,
336                                             pkt_len + sizeof(struct ipv6hdr)))
337                                 goto bad;
338                         break;
339                 default:
340                         if (optlen > len)
341                                 goto bad;
342                         break;
343                 }
344                 off += optlen;
345                 len -= optlen;
346         }
347         if (len == 0)
348                 return 0;
349 bad:
350         return -1;
351
352 }
353
354 /* Replicate the checks that IPv6 does on packet reception and pass the packet
355  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
356 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
357                                            struct sk_buff *skb,
358                                            const struct net_device *in,
359                                            const struct net_device *out,
360                                            int (*okfn)(struct sk_buff *))
361 {
362         struct ipv6hdr *hdr;
363         u32 pkt_len;
364         struct nf_bridge_info *nf_bridge;
365
366         if (skb->len < sizeof(struct ipv6hdr))
367                 goto inhdr_error;
368
369         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
370                 goto inhdr_error;
371
372         hdr = skb->nh.ipv6h;
373
374         if (hdr->version != 6)
375                 goto inhdr_error;
376
377         pkt_len = ntohs(hdr->payload_len);
378
379         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
380                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
381                         goto inhdr_error;
382                 if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
383                         if (__pskb_trim(skb, pkt_len + sizeof(struct ipv6hdr)))
384                                 goto inhdr_error;
385                         if (skb->ip_summed == CHECKSUM_HW)
386                                 skb->ip_summed = CHECKSUM_NONE;
387                 }
388         }
389         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
390                 goto inhdr_error;
391
392         nf_bridge_put(skb->nf_bridge);
393         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
394                 return NF_DROP;
395         if (!setup_pre_routing(skb))
396                 return NF_DROP;
397
398         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
399                 br_nf_pre_routing_finish_ipv6);
400
401         return NF_STOLEN;
402
403 inhdr_error:
404         return NF_DROP;
405 }
406
407 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
408  * Replicate the checks that IPv4 does on packet reception.
409  * Set skb->dev to the bridge device (i.e. parent of the
410  * receiving device) to make netfilter happy, the REDIRECT
411  * target in particular.  Save the original destination IP
412  * address to be able to detect DNAT afterwards. */
413 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
414                                       const struct net_device *in,
415                                       const struct net_device *out,
416                                       int (*okfn)(struct sk_buff *))
417 {
418         struct iphdr *iph;
419         __u32 len;
420         struct sk_buff *skb = *pskb;
421         struct nf_bridge_info *nf_bridge;
422         struct vlan_ethhdr *hdr = vlan_eth_hdr(*pskb);
423
424         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6) {
425 #ifdef CONFIG_SYSCTL
426                 if (!brnf_call_ip6tables)
427                         return NF_ACCEPT;
428 #endif
429                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
430                         goto out;
431
432                 if (skb->protocol == htons(ETH_P_8021Q)) {
433                         skb_pull_rcsum(skb, VLAN_HLEN);
434                         skb->nh.raw += VLAN_HLEN;
435                 }
436                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
437         }
438 #ifdef CONFIG_SYSCTL
439         if (!brnf_call_iptables)
440                 return NF_ACCEPT;
441 #endif
442
443         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP)
444                 return NF_ACCEPT;
445
446         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
447                 goto out;
448
449         if (skb->protocol == htons(ETH_P_8021Q)) {
450                 skb_pull_rcsum(skb, VLAN_HLEN);
451                 skb->nh.raw += VLAN_HLEN;
452         }
453
454         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
455                 goto inhdr_error;
456
457         iph = skb->nh.iph;
458         if (iph->ihl < 5 || iph->version != 4)
459                 goto inhdr_error;
460
461         if (!pskb_may_pull(skb, 4 * iph->ihl))
462                 goto inhdr_error;
463
464         iph = skb->nh.iph;
465         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
466                 goto inhdr_error;
467
468         len = ntohs(iph->tot_len);
469         if (skb->len < len || len < 4 * iph->ihl)
470                 goto inhdr_error;
471
472         if (skb->len > len) {
473                 __pskb_trim(skb, len);
474                 if (skb->ip_summed == CHECKSUM_HW)
475                         skb->ip_summed = CHECKSUM_NONE;
476         }
477
478         nf_bridge_put(skb->nf_bridge);
479         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
480                 return NF_DROP;
481         if (!setup_pre_routing(skb))
482                 return NF_DROP;
483         store_orig_dstaddr(skb);
484
485         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
486                 br_nf_pre_routing_finish);
487
488         return NF_STOLEN;
489
490 inhdr_error:
491 //      IP_INC_STATS_BH(IpInHdrErrors);
492 out:
493         return NF_DROP;
494 }
495
496
497 /* PF_BRIDGE/LOCAL_IN ************************************************/
498 /* The packet is locally destined, which requires a real
499  * dst_entry, so detach the fake one.  On the way up, the
500  * packet would pass through PRE_ROUTING again (which already
501  * took place when the packet entered the bridge), but we
502  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
503  * prevent this from happening. */
504 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
505                                    const struct net_device *in,
506                                    const struct net_device *out,
507                                    int (*okfn)(struct sk_buff *))
508 {
509         struct sk_buff *skb = *pskb;
510
511         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
512                 dst_release(skb->dst);
513                 skb->dst = NULL;
514         }
515
516         return NF_ACCEPT;
517 }
518
519 /* PF_BRIDGE/FORWARD *************************************************/
520 static int br_nf_forward_finish(struct sk_buff *skb)
521 {
522         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
523         struct net_device *in;
524         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
525
526         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP) {
527                 in = nf_bridge->physindev;
528                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
529                         skb->pkt_type = PACKET_OTHERHOST;
530                         nf_bridge->mask ^= BRNF_PKT_TYPE;
531                 }
532         } else {
533                 in = *((struct net_device **)(skb->cb));
534         }
535         if (skb->protocol == htons(ETH_P_8021Q)) {
536                 skb_push(skb, VLAN_HLEN);
537                 skb->nh.raw -= VLAN_HLEN;
538         }
539         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
540                        skb->dev, br_forward_finish, 1);
541         return 0;
542 }
543
544 /* This is the 'purely bridged' case.  For IP, we pass the packet to
545  * netfilter with indev and outdev set to the bridge device,
546  * but we are still able to filter on the 'real' indev/outdev
547  * because of the physdev module. For ARP, indev and outdev are the
548  * bridge ports. */
549 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
550                                      const struct net_device *in,
551                                      const struct net_device *out,
552                                      int (*okfn)(struct sk_buff *))
553 {
554         struct sk_buff *skb = *pskb;
555         struct nf_bridge_info *nf_bridge;
556         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
557         struct net_device *parent;
558         int pf;
559
560         if (!skb->nf_bridge)
561                 return NF_ACCEPT;
562
563         parent = bridge_parent(out);
564         if (!parent)
565                 return NF_DROP;
566
567         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP)
568                 pf = PF_INET;
569         else
570                 pf = PF_INET6;
571
572         if (skb->protocol == htons(ETH_P_8021Q)) {
573                 skb_pull(*pskb, VLAN_HLEN);
574                 (*pskb)->nh.raw += VLAN_HLEN;
575         }
576
577         nf_bridge = skb->nf_bridge;
578         if (skb->pkt_type == PACKET_OTHERHOST) {
579                 skb->pkt_type = PACKET_HOST;
580                 nf_bridge->mask |= BRNF_PKT_TYPE;
581         }
582
583         /* The physdev module checks on this */
584         nf_bridge->mask |= BRNF_BRIDGED;
585         nf_bridge->physoutdev = skb->dev;
586
587         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
588                 br_nf_forward_finish);
589
590         return NF_STOLEN;
591 }
592
593 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
594                                       const struct net_device *in,
595                                       const struct net_device *out,
596                                       int (*okfn)(struct sk_buff *))
597 {
598         struct sk_buff *skb = *pskb;
599         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
600         struct net_device **d = (struct net_device **)(skb->cb);
601
602 #ifdef CONFIG_SYSCTL
603         if (!brnf_call_arptables)
604                 return NF_ACCEPT;
605 #endif
606
607         if (skb->protocol != htons(ETH_P_ARP)) {
608                 if (!IS_VLAN_ARP)
609                         return NF_ACCEPT;
610                 skb_pull(*pskb, VLAN_HLEN);
611                 (*pskb)->nh.raw += VLAN_HLEN;
612         }
613
614         if (skb->nh.arph->ar_pln != 4) {
615                 if (IS_VLAN_ARP) {
616                         skb_push(*pskb, VLAN_HLEN);
617                         (*pskb)->nh.raw -= VLAN_HLEN;
618                 }
619                 return NF_ACCEPT;
620         }
621         *d = (struct net_device *)in;
622         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
623                 (struct net_device *)out, br_nf_forward_finish);
624
625         return NF_STOLEN;
626 }
627
628 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
629 static int br_nf_local_out_finish(struct sk_buff *skb)
630 {
631         if (skb->protocol == htons(ETH_P_8021Q)) {
632                 skb_push(skb, VLAN_HLEN);
633                 skb->nh.raw -= VLAN_HLEN;
634         }
635
636         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
637                        br_forward_finish, NF_BR_PRI_FIRST + 1);
638
639         return 0;
640 }
641
642 /* This function sees both locally originated IP packets and forwarded
643  * IP packets (in both cases the destination device is a bridge
644  * device). It also sees bridged-and-DNAT'ed packets.
645  * To be able to filter on the physical bridge devices (with the physdev
646  * module), we steal packets destined to a bridge device away from the
647  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
648  * when we have determined the real output device. This is done in here.
649  *
650  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
651  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
652  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
653  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
654  * will be executed.
655  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
656  * this packet before, and so the packet was locally originated. We fake
657  * the PF_INET/LOCAL_OUT hook.
658  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
659  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
660  * even routed packets that didn't arrive on a bridge interface have their
661  * nf_bridge->physindev set. */
662 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
663                                     const struct net_device *in,
664                                     const struct net_device *out,
665                                     int (*okfn)(struct sk_buff *))
666 {
667         struct net_device *realindev, *realoutdev;
668         struct sk_buff *skb = *pskb;
669         struct nf_bridge_info *nf_bridge;
670         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
671         int pf;
672
673         if (!skb->nf_bridge)
674                 return NF_ACCEPT;
675
676         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP)
677                 pf = PF_INET;
678         else
679                 pf = PF_INET6;
680
681 #ifdef CONFIG_NETFILTER_DEBUG
682         /* Sometimes we get packets with NULL ->dst here (for example,
683          * running a dhcp client daemon triggers this). This should now
684          * be fixed, but let's keep the check around. */
685         if (skb->dst == NULL) {
686                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
687                 return NF_ACCEPT;
688         }
689 #endif
690
691         nf_bridge = skb->nf_bridge;
692         nf_bridge->physoutdev = skb->dev;
693         realindev = nf_bridge->physindev;
694
695         /* Bridged, take PF_BRIDGE/FORWARD.
696          * (see big note in front of br_nf_pre_routing_finish) */
697         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
698                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
699                         skb->pkt_type = PACKET_OTHERHOST;
700                         nf_bridge->mask ^= BRNF_PKT_TYPE;
701                 }
702                 if (skb->protocol == htons(ETH_P_8021Q)) {
703                         skb_push(skb, VLAN_HLEN);
704                         skb->nh.raw -= VLAN_HLEN;
705                 }
706
707                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
708                         skb->dev, br_forward_finish);
709                 goto out;
710         }
711         realoutdev = bridge_parent(skb->dev);
712         if (!realoutdev)
713                 return NF_DROP;
714
715 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
716         /* iptables should match -o br0.x */
717         if (nf_bridge->netoutdev)
718                 realoutdev = nf_bridge->netoutdev;
719 #endif
720         if (skb->protocol == htons(ETH_P_8021Q)) {
721                 skb_pull(skb, VLAN_HLEN);
722                 (*pskb)->nh.raw += VLAN_HLEN;
723         }
724         /* IP forwarded traffic has a physindev, locally
725          * generated traffic hasn't. */
726         if (realindev != NULL) {
727                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT)) {
728                         struct net_device *parent = bridge_parent(realindev);
729                         if (parent)
730                                 realindev = parent;
731                 }
732
733                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
734                                realoutdev, br_nf_local_out_finish,
735                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
736         } else {
737                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
738                                realoutdev, br_nf_local_out_finish,
739                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
740         }
741
742 out:
743         return NF_STOLEN;
744 }
745
746
747 /* PF_BRIDGE/POST_ROUTING ********************************************/
748 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
749                                        const struct net_device *in,
750                                        const struct net_device *out,
751                                        int (*okfn)(struct sk_buff *))
752 {
753         struct sk_buff *skb = *pskb;
754         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
755         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
756         struct net_device *realoutdev = bridge_parent(skb->dev);
757         int pf;
758
759 #ifdef CONFIG_NETFILTER_DEBUG
760         /* Be very paranoid. This probably won't happen anymore, but let's
761          * keep the check just to be sure... */
762         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
763                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
764                        "bad mac.raw pointer.");
765                 goto print_error;
766         }
767 #endif
768
769         if (!nf_bridge)
770                 return NF_ACCEPT;
771
772         if (!realoutdev)
773                 return NF_DROP;
774
775         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP)
776                 pf = PF_INET;
777         else
778                 pf = PF_INET6;
779
780 #ifdef CONFIG_NETFILTER_DEBUG
781         if (skb->dst == NULL) {
782                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
783                 goto print_error;
784         }
785 #endif
786
787         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
788          * about the value of skb->pkt_type. */
789         if (skb->pkt_type == PACKET_OTHERHOST) {
790                 skb->pkt_type = PACKET_HOST;
791                 nf_bridge->mask |= BRNF_PKT_TYPE;
792         }
793
794         if (skb->protocol == htons(ETH_P_8021Q)) {
795                 skb_pull(skb, VLAN_HLEN);
796                 skb->nh.raw += VLAN_HLEN;
797         }
798
799         nf_bridge_save_header(skb);
800
801 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
802         if (nf_bridge->netoutdev)
803                 realoutdev = nf_bridge->netoutdev;
804 #endif
805         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
806                 br_dev_queue_push_xmit);
807
808         return NF_STOLEN;
809
810 #ifdef CONFIG_NETFILTER_DEBUG
811 print_error:
812         if (skb->dev != NULL) {
813                 printk("[%s]", skb->dev->name);
814                 if (realoutdev)
815                         printk("[%s]", realoutdev->name);
816         }
817         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
818                skb->data);
819         return NF_ACCEPT;
820 #endif
821 }
822
823 /* IP/SABOTAGE *****************************************************/
824 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
825  * for the second time. */
826 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
827                                    const struct net_device *in,
828                                    const struct net_device *out,
829                                    int (*okfn)(struct sk_buff *))
830 {
831         if ((*pskb)->nf_bridge &&
832             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
833                 return NF_STOP;
834         }
835
836         return NF_ACCEPT;
837 }
838
839 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
840  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
841  * decision in the bridge code and have determined nf_bridge->physoutdev. */
842 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
843                                     const struct net_device *in,
844                                     const struct net_device *out,
845                                     int (*okfn)(struct sk_buff *))
846 {
847         struct sk_buff *skb = *pskb;
848
849         if ((out->hard_start_xmit == br_dev_xmit &&
850              okfn != br_nf_forward_finish &&
851              okfn != br_nf_local_out_finish && okfn != br_dev_queue_push_xmit)
852 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
853             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
854                 VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
855 #endif
856             ) {
857                 struct nf_bridge_info *nf_bridge;
858
859                 if (!skb->nf_bridge) {
860 #ifdef CONFIG_SYSCTL
861                         /* This code is executed while in the IP(v6) stack,
862                            the version should be 4 or 6. We can't use
863                            skb->protocol because that isn't set on
864                            PF_INET(6)/LOCAL_OUT. */
865                         struct iphdr *ip = skb->nh.iph;
866
867                         if (ip->version == 4 && !brnf_call_iptables)
868                                 return NF_ACCEPT;
869                         else if (ip->version == 6 && !brnf_call_ip6tables)
870                                 return NF_ACCEPT;
871 #endif
872                         if (hook == NF_IP_POST_ROUTING)
873                                 return NF_ACCEPT;
874                         if (!nf_bridge_alloc(skb))
875                                 return NF_DROP;
876                 }
877
878                 nf_bridge = skb->nf_bridge;
879
880                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
881                  * will need the indev then. For a brouter, the real indev
882                  * can be a bridge port, so we make sure br_nf_local_out()
883                  * doesn't use the bridge parent of the indev by using
884                  * the BRNF_DONT_TAKE_PARENT mask. */
885                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
886                         nf_bridge->mask |= BRNF_DONT_TAKE_PARENT;
887                         nf_bridge->physindev = (struct net_device *)in;
888                 }
889 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
890                 /* the iptables outdev is br0.x, not br0 */
891                 if (out->priv_flags & IFF_802_1Q_VLAN)
892                         nf_bridge->netoutdev = (struct net_device *)out;
893 #endif
894                 return NF_STOP;
895         }
896
897         return NF_ACCEPT;
898 }
899
900 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
901  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
902  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
903  * ip_refrag() can return NF_STOLEN. */
904 static struct nf_hook_ops br_nf_ops[] = {
905         { .hook = br_nf_pre_routing, 
906           .owner = THIS_MODULE, 
907           .pf = PF_BRIDGE, 
908           .hooknum = NF_BR_PRE_ROUTING, 
909           .priority = NF_BR_PRI_BRNF, },
910         { .hook = br_nf_local_in,
911           .owner = THIS_MODULE,
912           .pf = PF_BRIDGE,
913           .hooknum = NF_BR_LOCAL_IN,
914           .priority = NF_BR_PRI_BRNF, },
915         { .hook = br_nf_forward_ip,
916           .owner = THIS_MODULE,
917           .pf = PF_BRIDGE,
918           .hooknum = NF_BR_FORWARD,
919           .priority = NF_BR_PRI_BRNF - 1, },
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         { .hook = br_nf_local_out,
926           .owner = THIS_MODULE,
927           .pf = PF_BRIDGE,
928           .hooknum = NF_BR_LOCAL_OUT,
929           .priority = NF_BR_PRI_FIRST, },
930         { .hook = br_nf_post_routing,
931           .owner = THIS_MODULE,
932           .pf = PF_BRIDGE,
933           .hooknum = NF_BR_POST_ROUTING,
934           .priority = NF_BR_PRI_LAST, },
935         { .hook = ip_sabotage_in,
936           .owner = THIS_MODULE,
937           .pf = PF_INET,
938           .hooknum = NF_IP_PRE_ROUTING,
939           .priority = NF_IP_PRI_FIRST, },
940         { .hook = ip_sabotage_in,
941           .owner = THIS_MODULE,
942           .pf = PF_INET6,
943           .hooknum = NF_IP6_PRE_ROUTING,
944           .priority = NF_IP6_PRI_FIRST, },
945         { .hook = ip_sabotage_out,
946           .owner = THIS_MODULE,
947           .pf = PF_INET,
948           .hooknum = NF_IP_FORWARD,
949           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
950         { .hook = ip_sabotage_out,
951           .owner = THIS_MODULE,
952           .pf = PF_INET6,
953           .hooknum = NF_IP6_FORWARD,
954           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
955         { .hook = ip_sabotage_out,
956           .owner = THIS_MODULE,
957           .pf = PF_INET,
958           .hooknum = NF_IP_LOCAL_OUT,
959           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
960         { .hook = ip_sabotage_out,
961           .owner = THIS_MODULE,
962           .pf = PF_INET6,
963           .hooknum = NF_IP6_LOCAL_OUT,
964           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
965         { .hook = ip_sabotage_out,
966           .owner = THIS_MODULE,
967           .pf = PF_INET,
968           .hooknum = NF_IP_POST_ROUTING,
969           .priority = NF_IP_PRI_FIRST, },
970         { .hook = ip_sabotage_out,
971           .owner = THIS_MODULE,
972           .pf = PF_INET6,
973           .hooknum = NF_IP6_POST_ROUTING,
974           .priority = NF_IP6_PRI_FIRST, },
975 };
976
977 #ifdef CONFIG_SYSCTL
978 static
979 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
980                             void __user * buffer, size_t * lenp, loff_t * ppos)
981 {
982         int ret;
983
984         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
985
986         if (write && *(int *)(ctl->data))
987                 *(int *)(ctl->data) = 1;
988         return ret;
989 }
990
991 static ctl_table brnf_table[] = {
992         {
993                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
994                 .procname       = "bridge-nf-call-arptables",
995                 .data           = &brnf_call_arptables,
996                 .maxlen         = sizeof(int),
997                 .mode           = 0644,
998                 .proc_handler   = &brnf_sysctl_call_tables,
999         },
1000         {
1001                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
1002                 .procname       = "bridge-nf-call-iptables",
1003                 .data           = &brnf_call_iptables,
1004                 .maxlen         = sizeof(int),
1005                 .mode           = 0644,
1006                 .proc_handler   = &brnf_sysctl_call_tables,
1007         },
1008         {
1009                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
1010                 .procname       = "bridge-nf-call-ip6tables",
1011                 .data           = &brnf_call_ip6tables,
1012                 .maxlen         = sizeof(int),
1013                 .mode           = 0644,
1014                 .proc_handler   = &brnf_sysctl_call_tables,
1015         },
1016         {
1017                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
1018                 .procname       = "bridge-nf-filter-vlan-tagged",
1019                 .data           = &brnf_filter_vlan_tagged,
1020                 .maxlen         = sizeof(int),
1021                 .mode           = 0644,
1022                 .proc_handler   = &brnf_sysctl_call_tables,
1023         },
1024         { .ctl_name = 0 }
1025 };
1026
1027 static ctl_table brnf_bridge_table[] = {
1028         {
1029                 .ctl_name       = NET_BRIDGE,
1030                 .procname       = "bridge",
1031                 .mode           = 0555,
1032                 .child          = brnf_table,
1033         },
1034         { .ctl_name = 0 }
1035 };
1036
1037 static ctl_table brnf_net_table[] = {
1038         {
1039                 .ctl_name       = CTL_NET,
1040                 .procname       = "net",
1041                 .mode           = 0555,
1042                 .child          = brnf_bridge_table,
1043         },
1044         { .ctl_name = 0 }
1045 };
1046 #endif
1047
1048 int br_netfilter_init(void)
1049 {
1050         int i;
1051
1052         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1053                 int ret;
1054
1055                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1056                         continue;
1057
1058                 while (i--)
1059                         nf_unregister_hook(&br_nf_ops[i]);
1060
1061                 return ret;
1062         }
1063
1064 #ifdef CONFIG_SYSCTL
1065         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1066         if (brnf_sysctl_header == NULL) {
1067                 printk(KERN_WARNING
1068                        "br_netfilter: can't register to sysctl.\n");
1069                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1070                         nf_unregister_hook(&br_nf_ops[i]);
1071                 return -EFAULT;
1072         }
1073 #endif
1074
1075         printk(KERN_NOTICE "Bridge firewalling registered\n");
1076
1077         return 0;
1078 }
1079
1080 void br_netfilter_fini(void)
1081 {
1082         int i;
1083
1084         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1085                 nf_unregister_hook(&br_nf_ops[i]);
1086 #ifdef CONFIG_SYSCTL
1087         unregister_sysctl_table(brnf_sysctl_header);
1088 #endif
1089 }