[PKT_SCHED]: Fix dsmark to apply changes consistent
[linux-2.6] / 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_ether.h>
30 #include <linux/if_vlan.h>
31 #include <linux/netfilter_bridge.h>
32 #include <linux/netfilter_ipv4.h>
33 #include <linux/netfilter_ipv6.h>
34 #include <linux/netfilter_arp.h>
35 #include <linux/in_route.h>
36 #include <net/ip.h>
37 #include <net/ipv6.h>
38 #include <asm/uaccess.h>
39 #include <asm/checksum.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) = (skb)->nh.iph->daddr)
48 #define dnat_took_place(skb)     (skb_origaddr(skb) != (skb)->nh.iph->daddr)
49
50 #define has_bridge_parent(device)       ((device)->br_port != NULL)
51 #define bridge_parent(device)           ((device)->br_port->br->dev)
52
53 #ifdef CONFIG_SYSCTL
54 static struct ctl_table_header *brnf_sysctl_header;
55 static int brnf_call_iptables = 1;
56 static int brnf_call_ip6tables = 1;
57 static int brnf_call_arptables = 1;
58 static int brnf_filter_vlan_tagged = 1;
59 #else
60 #define brnf_filter_vlan_tagged 1
61 #endif
62
63 #define IS_VLAN_IP (skb->protocol == __constant_htons(ETH_P_8021Q) &&    \
64         hdr->h_vlan_encapsulated_proto == __constant_htons(ETH_P_IP) &&  \
65         brnf_filter_vlan_tagged)
66 #define IS_VLAN_IPV6 (skb->protocol == __constant_htons(ETH_P_8021Q) &&    \
67         hdr->h_vlan_encapsulated_proto == __constant_htons(ETH_P_IPV6) &&  \
68         brnf_filter_vlan_tagged)
69 #define IS_VLAN_ARP (skb->protocol == __constant_htons(ETH_P_8021Q) &&   \
70         hdr->h_vlan_encapsulated_proto == __constant_htons(ETH_P_ARP) && \
71         brnf_filter_vlan_tagged)
72
73 /* We need these fake structures to make netfilter happy --
74  * lots of places assume that skb->dst != NULL, which isn't
75  * all that unreasonable.
76  *
77  * Currently, we fill in the PMTU entry because netfilter
78  * refragmentation needs it, and the rt_flags entry because
79  * ipt_REJECT needs it.  Future netfilter modules might
80  * require us to fill additional fields. */
81 static struct net_device __fake_net_device = {
82         .hard_header_len        = ETH_HLEN
83 };
84
85 static struct rtable __fake_rtable = {
86         .u = {
87                 .dst = {
88                         .__refcnt               = ATOMIC_INIT(1),
89                         .dev                    = &__fake_net_device,
90                         .path                   = &__fake_rtable.u.dst,
91                         .metrics                = {[RTAX_MTU - 1] = 1500},
92                 }
93         },
94         .rt_flags       = 0,
95 };
96
97
98 /* PF_BRIDGE/PRE_ROUTING *********************************************/
99 /* Undo the changes made for ip6tables PREROUTING and continue the
100  * bridge PRE_ROUTING hook. */
101 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
102 {
103         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
104
105 #ifdef CONFIG_NETFILTER_DEBUG
106         skb->nf_debug ^= (1 << NF_BR_PRE_ROUTING);
107 #endif
108
109         if (nf_bridge->mask & BRNF_PKT_TYPE) {
110                 skb->pkt_type = PACKET_OTHERHOST;
111                 nf_bridge->mask ^= BRNF_PKT_TYPE;
112         }
113         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
114
115         skb->dst = (struct dst_entry *)&__fake_rtable;
116         dst_hold(skb->dst);
117
118         skb->dev = nf_bridge->physindev;
119         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
120                 skb_push(skb, VLAN_HLEN);
121                 skb->nh.raw -= VLAN_HLEN;
122         }
123         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
124                        br_handle_frame_finish, 1);
125
126         return 0;
127 }
128
129 static void __br_dnat_complain(void)
130 {
131         static unsigned long last_complaint;
132
133         if (jiffies - last_complaint >= 5 * HZ) {
134                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
135                         "forwarding to be enabled\n");
136                 last_complaint = jiffies;
137         }
138 }
139
140 /* This requires some explaining. If DNAT has taken place,
141  * we will need to fix up the destination Ethernet address,
142  * and this is a tricky process.
143  *
144  * There are two cases to consider:
145  * 1. The packet was DNAT'ed to a device in the same bridge
146  *    port group as it was received on. We can still bridge
147  *    the packet.
148  * 2. The packet was DNAT'ed to a different device, either
149  *    a non-bridged device or another bridge port group.
150  *    The packet will need to be routed.
151  *
152  * The correct way of distinguishing between these two cases is to
153  * call ip_route_input() and to look at skb->dst->dev, which is
154  * changed to the destination device if ip_route_input() succeeds.
155  *
156  * Let us first consider the case that ip_route_input() succeeds:
157  *
158  * If skb->dst->dev equals the logical bridge device the packet
159  * came in on, we can consider this bridging. We then call
160  * skb->dst->output() which will make the packet enter br_nf_local_out()
161  * not much later. In that function it is assured that the iptables
162  * FORWARD chain is traversed for the packet.
163  *
164  * Otherwise, the packet is considered to be routed and we just
165  * change the destination MAC address so that the packet will
166  * later be passed up to the IP stack to be routed.
167  *
168  * Let us now consider the case that ip_route_input() fails:
169  *
170  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
171  * will fail, while __ip_route_output_key() will return success. The source
172  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
173  * thinks we're handling a locally generated packet and won't care
174  * if IP forwarding is allowed. We send a warning message to the users's
175  * log telling her to put IP forwarding on.
176  *
177  * ip_route_input() will also fail if there is no route available.
178  * In that case we just drop the packet.
179  *
180  * --Lennert, 20020411
181  * --Bart, 20020416 (updated)
182  * --Bart, 20021007 (updated) */
183 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
184 {
185 #ifdef CONFIG_NETFILTER_DEBUG
186         skb->nf_debug |= (1 << NF_BR_PRE_ROUTING) | (1 << NF_BR_FORWARD);
187 #endif
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->protocol == __constant_htons(ETH_P_8021Q)) {
197                 skb_pull(skb, VLAN_HLEN);
198                 skb->nh.raw += VLAN_HLEN;
199         }
200         skb->dst->output(skb);
201         return 0;
202 }
203
204 static int br_nf_pre_routing_finish(struct sk_buff *skb)
205 {
206         struct net_device *dev = skb->dev;
207         struct iphdr *iph = skb->nh.iph;
208         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
209
210 #ifdef CONFIG_NETFILTER_DEBUG
211         skb->nf_debug ^= (1 << NF_BR_PRE_ROUTING);
212 #endif
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,
222                     dev)) {
223                         struct rtable *rt;
224                         struct flowi fl = { .nl_u = 
225                         { .ip4_u = { .daddr = iph->daddr, .saddr = 0 ,
226                                      .tos = RT_TOS(iph->tos)} }, .proto = 0};
227
228                         if (!ip_route_output_key(&rt, &fl)) {
229                                 /* Bridged-and-DNAT'ed traffic doesn't
230                                  * require ip_forwarding. */
231                                 if (((struct dst_entry *)rt)->dev == dev) {
232                                         skb->dst = (struct dst_entry *)rt;
233                                         goto bridged_dnat;
234                                 }
235                                 __br_dnat_complain();
236                                 dst_release((struct dst_entry *)rt);
237                         }
238                         kfree_skb(skb);
239                         return 0;
240                 } else {
241                         if (skb->dst->dev == dev) {
242 bridged_dnat:
243                                 /* Tell br_nf_local_out this is a
244                                  * bridged frame */
245                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
246                                 skb->dev = nf_bridge->physindev;
247                                 if (skb->protocol ==
248                                     __constant_htons(ETH_P_8021Q)) {
249                                         skb_push(skb, VLAN_HLEN);
250                                         skb->nh.raw -= VLAN_HLEN;
251                                 }
252                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
253                                                skb, skb->dev, NULL,
254                                                br_nf_pre_routing_finish_bridge,
255                                                1);
256                                 return 0;
257                         }
258                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr,
259                                ETH_ALEN);
260                         skb->pkt_type = PACKET_HOST;
261                 }
262         } else {
263                 skb->dst = (struct dst_entry *)&__fake_rtable;
264                 dst_hold(skb->dst);
265         }
266
267         skb->dev = nf_bridge->physindev;
268         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
269                 skb_push(skb, VLAN_HLEN);
270                 skb->nh.raw -= VLAN_HLEN;
271         }
272         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
273                        br_handle_frame_finish, 1);
274
275         return 0;
276 }
277
278 /* Some common code for IPv4/IPv6 */
279 static void setup_pre_routing(struct sk_buff *skb)
280 {
281         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
282
283         if (skb->pkt_type == PACKET_OTHERHOST) {
284                 skb->pkt_type = PACKET_HOST;
285                 nf_bridge->mask |= BRNF_PKT_TYPE;
286         }
287
288         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
289         nf_bridge->physindev = skb->dev;
290         skb->dev = bridge_parent(skb->dev);
291 }
292
293 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
294 static int check_hbh_len(struct sk_buff *skb)
295 {
296         unsigned char *raw = (u8*)(skb->nh.ipv6h+1);
297         u32 pkt_len;
298         int off = raw - skb->nh.raw;
299         int len = (raw[1]+1)<<3;
300
301         if ((raw + len) - skb->data > skb_headlen(skb))
302                 goto bad;
303
304         off += 2;
305         len -= 2;
306
307         while (len > 0) {
308                 int optlen = raw[off+1]+2;
309
310                 switch (skb->nh.raw[off]) {
311                 case IPV6_TLV_PAD0:
312                         optlen = 1;
313                         break;
314
315                 case IPV6_TLV_PADN:
316                         break;
317
318                 case IPV6_TLV_JUMBO:
319                         if (skb->nh.raw[off+1] != 4 || (off&3) != 2)
320                                 goto bad;
321
322                         pkt_len = ntohl(*(u32*)(skb->nh.raw+off+2));
323
324                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
325                                 goto bad;
326                         if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
327                                 if (__pskb_trim(skb,
328                                     pkt_len + sizeof(struct ipv6hdr)))
329                                         goto bad;
330                                 if (skb->ip_summed == CHECKSUM_HW)
331                                         skb->ip_summed = CHECKSUM_NONE;
332                         }
333                         break;
334                 default:
335                         if (optlen > len)
336                                 goto bad;
337                         break;
338                 }
339                 off += optlen;
340                 len -= optlen;
341         }
342         if (len == 0)
343                 return 0;
344 bad:
345         return -1;
346
347 }
348
349 /* Replicate the checks that IPv6 does on packet reception and pass the packet
350  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
351 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
352    struct sk_buff *skb, const struct net_device *in,
353    const struct net_device *out, int (*okfn)(struct sk_buff *))
354 {
355         struct ipv6hdr *hdr;
356         u32 pkt_len;
357         struct nf_bridge_info *nf_bridge;
358
359         if (skb->len < sizeof(struct ipv6hdr))
360                 goto inhdr_error;
361
362         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
363                 goto inhdr_error;
364
365         hdr = skb->nh.ipv6h;
366
367         if (hdr->version != 6)
368                 goto inhdr_error;
369
370         pkt_len = ntohs(hdr->payload_len);
371
372         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
373                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
374                         goto inhdr_error;
375                 if (pkt_len + sizeof(struct ipv6hdr) < skb->len) {
376                         if (__pskb_trim(skb, pkt_len + sizeof(struct ipv6hdr)))
377                                 goto inhdr_error;
378                         if (skb->ip_summed == CHECKSUM_HW)
379                                 skb->ip_summed = CHECKSUM_NONE;
380                 }
381         }
382         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
383                         goto inhdr_error;
384
385 #ifdef CONFIG_NETFILTER_DEBUG
386         skb->nf_debug ^= (1 << NF_IP6_PRE_ROUTING);
387 #endif
388         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
389                 return NF_DROP;
390         setup_pre_routing(skb);
391
392         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
393                 br_nf_pre_routing_finish_ipv6);
394
395         return NF_STOLEN;
396
397 inhdr_error:
398         return NF_DROP;
399 }
400
401 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
402  * Replicate the checks that IPv4 does on packet reception.
403  * Set skb->dev to the bridge device (i.e. parent of the
404  * receiving device) to make netfilter happy, the REDIRECT
405  * target in particular.  Save the original destination IP
406  * address to be able to detect DNAT afterwards. */
407 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
408    const struct net_device *in, const struct net_device *out,
409    int (*okfn)(struct sk_buff *))
410 {
411         struct iphdr *iph;
412         __u32 len;
413         struct sk_buff *skb = *pskb;
414         struct nf_bridge_info *nf_bridge;
415         struct vlan_ethhdr *hdr = vlan_eth_hdr(*pskb);
416
417         if (skb->protocol == __constant_htons(ETH_P_IPV6) || IS_VLAN_IPV6) {
418 #ifdef CONFIG_SYSCTL
419                 if (!brnf_call_ip6tables)
420                         return NF_ACCEPT;
421 #endif
422                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
423                         goto out;
424
425                 if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
426                         skb_pull(skb, VLAN_HLEN);
427                         (skb)->nh.raw += VLAN_HLEN;
428                 }
429                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
430         }
431 #ifdef CONFIG_SYSCTL
432         if (!brnf_call_iptables)
433                 return NF_ACCEPT;
434 #endif
435
436         if (skb->protocol != __constant_htons(ETH_P_IP) && !IS_VLAN_IP)
437                 return NF_ACCEPT;
438
439         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
440                 goto out;
441
442         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
443                 skb_pull(skb, VLAN_HLEN);
444                 (skb)->nh.raw += VLAN_HLEN;
445         }
446
447         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
448                 goto inhdr_error;
449
450         iph = skb->nh.iph;
451         if (iph->ihl < 5 || iph->version != 4)
452                 goto inhdr_error;
453
454         if (!pskb_may_pull(skb, 4*iph->ihl))
455                 goto inhdr_error;
456
457         iph = skb->nh.iph;
458         if (ip_fast_csum((__u8 *)iph, iph->ihl) != 0)
459                 goto inhdr_error;
460
461         len = ntohs(iph->tot_len);
462         if (skb->len < len || len < 4*iph->ihl)
463                 goto inhdr_error;
464
465         if (skb->len > len) {
466                 __pskb_trim(skb, len);
467                 if (skb->ip_summed == CHECKSUM_HW)
468                         skb->ip_summed = CHECKSUM_NONE;
469         }
470
471 #ifdef CONFIG_NETFILTER_DEBUG
472         skb->nf_debug ^= (1 << NF_IP_PRE_ROUTING);
473 #endif
474         if ((nf_bridge = nf_bridge_alloc(skb)) == NULL)
475                 return NF_DROP;
476         setup_pre_routing(skb);
477         store_orig_dstaddr(skb);
478
479         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
480                 br_nf_pre_routing_finish);
481
482         return NF_STOLEN;
483
484 inhdr_error:
485 //      IP_INC_STATS_BH(IpInHdrErrors);
486 out:
487         return NF_DROP;
488 }
489
490
491 /* PF_BRIDGE/LOCAL_IN ************************************************/
492 /* The packet is locally destined, which requires a real
493  * dst_entry, so detach the fake one.  On the way up, the
494  * packet would pass through PRE_ROUTING again (which already
495  * took place when the packet entered the bridge), but we
496  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
497  * prevent this from happening. */
498 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
499    const struct net_device *in, const struct net_device *out,
500    int (*okfn)(struct sk_buff *))
501 {
502         struct sk_buff *skb = *pskb;
503
504         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
505                 dst_release(skb->dst);
506                 skb->dst = NULL;
507         }
508
509         return NF_ACCEPT;
510 }
511
512
513 /* PF_BRIDGE/FORWARD *************************************************/
514 static int br_nf_forward_finish(struct sk_buff *skb)
515 {
516         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
517         struct net_device *in;
518         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
519
520 #ifdef CONFIG_NETFILTER_DEBUG
521         skb->nf_debug ^= (1 << NF_BR_FORWARD);
522 #endif
523
524         if (skb->protocol != __constant_htons(ETH_P_ARP) && !IS_VLAN_ARP) {
525                 in = nf_bridge->physindev;
526                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
527                         skb->pkt_type = PACKET_OTHERHOST;
528                         nf_bridge->mask ^= BRNF_PKT_TYPE;
529                 }
530         } else {
531                 in = *((struct net_device **)(skb->cb));
532         }
533         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
534                 skb_push(skb, VLAN_HLEN);
535                 skb->nh.raw -= VLAN_HLEN;
536         }
537         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
538                         skb->dev, br_forward_finish, 1);
539         return 0;
540 }
541
542 /* This is the 'purely bridged' case.  For IP, we pass the packet to
543  * netfilter with indev and outdev set to the bridge device,
544  * but we are still able to filter on the 'real' indev/outdev
545  * because of the physdev module. For ARP, indev and outdev are the
546  * bridge ports. */
547 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
548    const struct net_device *in, const struct net_device *out,
549    int (*okfn)(struct sk_buff *))
550 {
551         struct sk_buff *skb = *pskb;
552         struct nf_bridge_info *nf_bridge;
553         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
554         int pf;
555
556         if (!skb->nf_bridge)
557                 return NF_ACCEPT;
558
559         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
560                 pf = PF_INET;
561         else
562                 pf = PF_INET6;
563
564         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
565                 skb_pull(*pskb, VLAN_HLEN);
566                 (*pskb)->nh.raw += VLAN_HLEN;
567         }
568
569 #ifdef CONFIG_NETFILTER_DEBUG
570         skb->nf_debug ^= (1 << NF_BR_FORWARD);
571 #endif
572         nf_bridge = skb->nf_bridge;
573         if (skb->pkt_type == PACKET_OTHERHOST) {
574                 skb->pkt_type = PACKET_HOST;
575                 nf_bridge->mask |= BRNF_PKT_TYPE;
576         }
577
578         /* The physdev module checks on this */
579         nf_bridge->mask |= BRNF_BRIDGED;
580         nf_bridge->physoutdev = skb->dev;
581
582         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in),
583                 bridge_parent(out), br_nf_forward_finish);
584
585         return NF_STOLEN;
586 }
587
588 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
589    const struct net_device *in, const struct net_device *out,
590    int (*okfn)(struct sk_buff *))
591 {
592         struct sk_buff *skb = *pskb;
593         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
594         struct net_device **d = (struct net_device **)(skb->cb);
595
596 #ifdef CONFIG_SYSCTL
597         if (!brnf_call_arptables)
598                 return NF_ACCEPT;
599 #endif
600
601         if (skb->protocol != __constant_htons(ETH_P_ARP)) {
602                 if (!IS_VLAN_ARP)
603                         return NF_ACCEPT;
604                 skb_pull(*pskb, VLAN_HLEN);
605                 (*pskb)->nh.raw += VLAN_HLEN;
606         }
607
608 #ifdef CONFIG_NETFILTER_DEBUG
609         skb->nf_debug ^= (1 << NF_BR_FORWARD);
610 #endif
611
612         if (skb->nh.arph->ar_pln != 4) {
613                 if (IS_VLAN_ARP) {
614                         skb_push(*pskb, VLAN_HLEN);
615                         (*pskb)->nh.raw -= VLAN_HLEN;
616                 }
617                 return NF_ACCEPT;
618         }
619         *d = (struct net_device *)in;
620         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
621                 (struct net_device *)out, br_nf_forward_finish);
622
623         return NF_STOLEN;
624 }
625
626
627 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
628 static int br_nf_local_out_finish(struct sk_buff *skb)
629 {
630 #ifdef CONFIG_NETFILTER_DEBUG
631         skb->nf_debug &= ~(1 << NF_BR_LOCAL_OUT);
632 #endif
633         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
634                 skb_push(skb, VLAN_HLEN);
635                 skb->nh.raw -= VLAN_HLEN;
636         }
637
638         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
639                         br_forward_finish, NF_BR_PRI_FIRST + 1);
640
641         return 0;
642 }
643
644 /* This function sees both locally originated IP packets and forwarded
645  * IP packets (in both cases the destination device is a bridge
646  * device). It also sees bridged-and-DNAT'ed packets.
647  * To be able to filter on the physical bridge devices (with the physdev
648  * module), we steal packets destined to a bridge device away from the
649  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
650  * when we have determined the real output device. This is done in here.
651  *
652  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
653  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
654  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
655  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
656  * will be executed.
657  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
658  * this packet before, and so the packet was locally originated. We fake
659  * the PF_INET/LOCAL_OUT hook.
660  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
661  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
662  * even routed packets that didn't arrive on a bridge interface have their
663  * nf_bridge->physindev set. */
664 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
665    const struct net_device *in, const struct net_device *out,
666    int (*okfn)(struct sk_buff *))
667 {
668         struct net_device *realindev, *realoutdev;
669         struct sk_buff *skb = *pskb;
670         struct nf_bridge_info *nf_bridge;
671         struct vlan_ethhdr *hdr = vlan_eth_hdr(skb);
672         int pf;
673
674         if (!skb->nf_bridge)
675                 return NF_ACCEPT;
676
677         if (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
678                 pf = PF_INET;
679         else
680                 pf = PF_INET6;
681
682 #ifdef CONFIG_NETFILTER_DEBUG
683         /* Sometimes we get packets with NULL ->dst here (for example,
684          * running a dhcp client daemon triggers this). This should now
685          * be fixed, but let's keep the check around. */
686         if (skb->dst == NULL) {
687                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
688                 return NF_ACCEPT;
689         }
690 #endif
691
692         nf_bridge = skb->nf_bridge;
693         nf_bridge->physoutdev = skb->dev;
694         realindev = nf_bridge->physindev;
695
696         /* Bridged, take PF_BRIDGE/FORWARD.
697          * (see big note in front of br_nf_pre_routing_finish) */
698         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
699                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
700                         skb->pkt_type = PACKET_OTHERHOST;
701                         nf_bridge->mask ^= BRNF_PKT_TYPE;
702                 }
703                 if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
704                         skb_push(skb, VLAN_HLEN);
705                         skb->nh.raw -= VLAN_HLEN;
706                 }
707
708                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
709                         skb->dev, br_forward_finish);
710                 goto out;
711         }
712         realoutdev = bridge_parent(skb->dev);
713
714 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
715         /* iptables should match -o br0.x */
716         if (nf_bridge->netoutdev)
717                 realoutdev = nf_bridge->netoutdev;
718 #endif
719         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
720                 skb_pull(skb, VLAN_HLEN);
721                 (*pskb)->nh.raw += VLAN_HLEN;
722         }
723         /* IP forwarded traffic has a physindev, locally
724          * generated traffic hasn't. */
725         if (realindev != NULL) {
726                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT) &&
727                     has_bridge_parent(realindev))
728                         realindev = bridge_parent(realindev);
729
730                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
731                                realoutdev, br_nf_local_out_finish,
732                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
733         } else {
734 #ifdef CONFIG_NETFILTER_DEBUG
735                 skb->nf_debug ^= (1 << NF_IP_LOCAL_OUT);
736 #endif
737
738                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
739                                realoutdev, br_nf_local_out_finish,
740                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
741         }
742
743 out:
744         return NF_STOLEN;
745 }
746
747
748 /* PF_BRIDGE/POST_ROUTING ********************************************/
749 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
750    const struct net_device *in, 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 (skb->protocol == __constant_htons(ETH_P_IP) || IS_VLAN_IP)
773                 pf = PF_INET;
774         else
775                 pf = PF_INET6;
776
777 #ifdef CONFIG_NETFILTER_DEBUG
778         if (skb->dst == NULL) {
779                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
780                 goto print_error;
781         }
782
783         skb->nf_debug ^= (1 << NF_IP_POST_ROUTING);
784 #endif
785
786         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
787          * about the value of skb->pkt_type. */
788         if (skb->pkt_type == PACKET_OTHERHOST) {
789                 skb->pkt_type = PACKET_HOST;
790                 nf_bridge->mask |= BRNF_PKT_TYPE;
791         }
792
793         if (skb->protocol == __constant_htons(ETH_P_8021Q)) {
794                 skb_pull(skb, VLAN_HLEN);
795                 skb->nh.raw += VLAN_HLEN;
796         }
797
798         nf_bridge_save_header(skb);
799
800 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
801         if (nf_bridge->netoutdev)
802                 realoutdev = nf_bridge->netoutdev;
803 #endif
804         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
805                 br_dev_queue_push_xmit);
806
807         return NF_STOLEN;
808
809 #ifdef CONFIG_NETFILTER_DEBUG
810 print_error:
811         if (skb->dev != NULL) {
812                 printk("[%s]", skb->dev->name);
813                 if (has_bridge_parent(skb->dev))
814                         printk("[%s]", bridge_parent(skb->dev)->name);
815         }
816         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
817                                               skb->data);
818         return NF_ACCEPT;
819 #endif
820 }
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, const struct net_device *out,
828    int (*okfn)(struct sk_buff *))
829 {
830         if ((*pskb)->nf_bridge &&
831             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
832                 return NF_STOP;
833         }
834
835         return NF_ACCEPT;
836 }
837
838 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
839  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
840  * decision in the bridge code and have determined nf_bridge->physoutdev. */
841 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
842    const struct net_device *in, const struct net_device *out,
843    int (*okfn)(struct sk_buff *))
844 {
845         struct sk_buff *skb = *pskb;
846
847         if ((out->hard_start_xmit == br_dev_xmit &&
848             okfn != br_nf_forward_finish &&
849             okfn != br_nf_local_out_finish &&
850             okfn != br_dev_queue_push_xmit)
851 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
852             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
853             VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
854 #endif
855             ) {
856                 struct nf_bridge_info *nf_bridge;
857
858                 if (!skb->nf_bridge) {
859 #ifdef CONFIG_SYSCTL
860                         /* This code is executed while in the IP(v6) stack,
861                            the version should be 4 or 6. We can't use
862                            skb->protocol because that isn't set on
863                            PF_INET(6)/LOCAL_OUT. */
864                         struct iphdr *ip = skb->nh.iph;
865
866                         if (ip->version == 4 && !brnf_call_iptables)
867                                 return NF_ACCEPT;
868                         else if (ip->version == 6 && !brnf_call_ip6tables)
869                                 return NF_ACCEPT;
870 #endif
871                         if (hook == NF_IP_POST_ROUTING)
872                                 return NF_ACCEPT;
873                         if (!nf_bridge_alloc(skb))
874                                 return NF_DROP;
875                 }
876
877                 nf_bridge = skb->nf_bridge;
878
879                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
880                  * will need the indev then. For a brouter, the real indev
881                  * can be a bridge port, so we make sure br_nf_local_out()
882                  * doesn't use the bridge parent of the indev by using
883                  * the BRNF_DONT_TAKE_PARENT mask. */
884                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
885                         nf_bridge->mask &= BRNF_DONT_TAKE_PARENT;
886                         nf_bridge->physindev = (struct net_device *)in;
887                 }
888 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
889                 /* the iptables outdev is br0.x, not br0 */
890                 if (out->priv_flags & IFF_802_1Q_VLAN)
891                         nf_bridge->netoutdev = (struct net_device *)out;
892 #endif
893                 return NF_STOP;
894         }
895
896         return NF_ACCEPT;
897 }
898
899 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
900  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
901  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
902  * ip_refrag() can return NF_STOLEN. */
903 static struct nf_hook_ops br_nf_ops[] = {
904         { .hook = br_nf_pre_routing, 
905           .owner = THIS_MODULE, 
906           .pf = PF_BRIDGE, 
907           .hooknum = NF_BR_PRE_ROUTING, 
908           .priority = NF_BR_PRI_BRNF, },
909         { .hook = br_nf_local_in,
910           .owner = THIS_MODULE,
911           .pf = PF_BRIDGE,
912           .hooknum = NF_BR_LOCAL_IN,
913           .priority = NF_BR_PRI_BRNF, },
914         { .hook = br_nf_forward_ip,
915           .owner = THIS_MODULE,
916           .pf = PF_BRIDGE,
917           .hooknum = NF_BR_FORWARD,
918           .priority = NF_BR_PRI_BRNF - 1, },
919         { .hook = br_nf_forward_arp,
920           .owner = THIS_MODULE,
921           .pf = PF_BRIDGE,
922           .hooknum = NF_BR_FORWARD,
923           .priority = NF_BR_PRI_BRNF, },
924         { .hook = br_nf_local_out,
925           .owner = THIS_MODULE,
926           .pf = PF_BRIDGE,
927           .hooknum = NF_BR_LOCAL_OUT,
928           .priority = NF_BR_PRI_FIRST, },
929         { .hook = br_nf_post_routing,
930           .owner = THIS_MODULE,
931           .pf = PF_BRIDGE,
932           .hooknum = NF_BR_POST_ROUTING,
933           .priority = NF_BR_PRI_LAST, },
934         { .hook = ip_sabotage_in,
935           .owner = THIS_MODULE,
936           .pf = PF_INET,
937           .hooknum = NF_IP_PRE_ROUTING,
938           .priority = NF_IP_PRI_FIRST, },
939         { .hook = ip_sabotage_in,
940           .owner = THIS_MODULE,
941           .pf = PF_INET6,
942           .hooknum = NF_IP6_PRE_ROUTING,
943           .priority = NF_IP6_PRI_FIRST, },
944         { .hook = ip_sabotage_out,
945           .owner = THIS_MODULE,
946           .pf = PF_INET,
947           .hooknum = NF_IP_FORWARD,
948           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
949         { .hook = ip_sabotage_out,
950           .owner = THIS_MODULE,
951           .pf = PF_INET6,
952           .hooknum = NF_IP6_FORWARD,
953           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
954         { .hook = ip_sabotage_out,
955           .owner = THIS_MODULE,
956           .pf = PF_INET,
957           .hooknum = NF_IP_LOCAL_OUT,
958           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
959         { .hook = ip_sabotage_out,
960           .owner = THIS_MODULE,
961           .pf = PF_INET6,
962           .hooknum = NF_IP6_LOCAL_OUT,
963           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
964         { .hook = ip_sabotage_out,
965           .owner = THIS_MODULE,
966           .pf = PF_INET,
967           .hooknum = NF_IP_POST_ROUTING,
968           .priority = NF_IP_PRI_FIRST, },
969         { .hook = ip_sabotage_out,
970           .owner = THIS_MODULE,
971           .pf = PF_INET6,
972           .hooknum = NF_IP6_POST_ROUTING,
973           .priority = NF_IP6_PRI_FIRST, },
974 };
975
976 #ifdef CONFIG_SYSCTL
977 static
978 int brnf_sysctl_call_tables(ctl_table *ctl, int write, struct file * filp,
979                         void __user *buffer, size_t *lenp, loff_t *ppos)
980 {
981         int ret;
982
983         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
984
985         if (write && *(int *)(ctl->data))
986                 *(int *)(ctl->data) = 1;
987         return ret;
988 }
989
990 static ctl_table brnf_table[] = {
991         {
992                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
993                 .procname       = "bridge-nf-call-arptables",
994                 .data           = &brnf_call_arptables,
995                 .maxlen         = sizeof(int),
996                 .mode           = 0644,
997                 .proc_handler   = &brnf_sysctl_call_tables,
998         },
999         {
1000                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
1001                 .procname       = "bridge-nf-call-iptables",
1002                 .data           = &brnf_call_iptables,
1003                 .maxlen         = sizeof(int),
1004                 .mode           = 0644,
1005                 .proc_handler   = &brnf_sysctl_call_tables,
1006         },
1007         {
1008                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
1009                 .procname       = "bridge-nf-call-ip6tables",
1010                 .data           = &brnf_call_ip6tables,
1011                 .maxlen         = sizeof(int),
1012                 .mode           = 0644,
1013                 .proc_handler   = &brnf_sysctl_call_tables,
1014         },
1015         {
1016                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
1017                 .procname       = "bridge-nf-filter-vlan-tagged",
1018                 .data           = &brnf_filter_vlan_tagged,
1019                 .maxlen         = sizeof(int),
1020                 .mode           = 0644,
1021                 .proc_handler   = &brnf_sysctl_call_tables,
1022         },
1023         { .ctl_name = 0 }
1024 };
1025
1026 static ctl_table brnf_bridge_table[] = {
1027         {
1028                 .ctl_name       = NET_BRIDGE,
1029                 .procname       = "bridge",
1030                 .mode           = 0555,
1031                 .child          = brnf_table,
1032         },
1033         { .ctl_name = 0 }
1034 };
1035
1036 static ctl_table brnf_net_table[] = {
1037         {
1038                 .ctl_name       = CTL_NET,
1039                 .procname       = "net",
1040                 .mode           = 0555,
1041                 .child          = brnf_bridge_table,
1042         },
1043         { .ctl_name = 0 }
1044 };
1045 #endif
1046
1047 int br_netfilter_init(void)
1048 {
1049         int i;
1050
1051         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1052                 int ret;
1053
1054                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1055                         continue;
1056
1057                 while (i--)
1058                         nf_unregister_hook(&br_nf_ops[i]);
1059
1060                 return ret;
1061         }
1062
1063 #ifdef CONFIG_SYSCTL
1064         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1065         if (brnf_sysctl_header == NULL) {
1066                 printk(KERN_WARNING "br_netfilter: can't register to sysctl.\n");
1067                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1068                         nf_unregister_hook(&br_nf_ops[i]);
1069                 return -EFAULT;
1070         }
1071 #endif
1072
1073         printk(KERN_NOTICE "Bridge firewalling registered\n");
1074
1075         return 0;
1076 }
1077
1078 void br_netfilter_fini(void)
1079 {
1080         int i;
1081
1082         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1083                 nf_unregister_hook(&br_nf_ops[i]);
1084 #ifdef CONFIG_SYSCTL
1085         unregister_sysctl_table(brnf_sysctl_header);
1086 #endif
1087 }