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