[NETFILTER] bridge: code rearrangement for clarity
[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 int brnf_deferred_hooks;
65 EXPORT_SYMBOL_GPL(brnf_deferred_hooks);
66
67 static __be16 inline vlan_proto(const struct sk_buff *skb)
68 {
69         return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
70 }
71
72 #define IS_VLAN_IP(skb) \
73         (skb->protocol == htons(ETH_P_8021Q) && \
74          vlan_proto(skb) == htons(ETH_P_IP) &&  \
75          brnf_filter_vlan_tagged)
76
77 #define IS_VLAN_IPV6(skb) \
78         (skb->protocol == htons(ETH_P_8021Q) && \
79          vlan_proto(skb) == htons(ETH_P_IPV6) &&\
80          brnf_filter_vlan_tagged)
81
82 #define IS_VLAN_ARP(skb) \
83         (skb->protocol == htons(ETH_P_8021Q) && \
84          vlan_proto(skb) == htons(ETH_P_ARP) && \
85          brnf_filter_vlan_tagged)
86
87 /* We need these fake structures to make netfilter happy --
88  * lots of places assume that skb->dst != NULL, which isn't
89  * all that unreasonable.
90  *
91  * Currently, we fill in the PMTU entry because netfilter
92  * refragmentation needs it, and the rt_flags entry because
93  * ipt_REJECT needs it.  Future netfilter modules might
94  * require us to fill additional fields. */
95 static struct net_device __fake_net_device = {
96         .hard_header_len        = ETH_HLEN
97 };
98
99 static struct rtable __fake_rtable = {
100         .u = {
101                 .dst = {
102                         .__refcnt               = ATOMIC_INIT(1),
103                         .dev                    = &__fake_net_device,
104                         .path                   = &__fake_rtable.u.dst,
105                         .metrics                = {[RTAX_MTU - 1] = 1500},
106                         .flags                  = DST_NOXFRM,
107                 }
108         },
109         .rt_flags       = 0,
110 };
111
112 static inline struct net_device *bridge_parent(const struct net_device *dev)
113 {
114         struct net_bridge_port *port = rcu_dereference(dev->br_port);
115
116         return port ? port->br->dev : NULL;
117 }
118
119 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
120 {
121         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
122         if (likely(skb->nf_bridge))
123                 atomic_set(&(skb->nf_bridge->use), 1);
124
125         return skb->nf_bridge;
126 }
127
128 static inline void nf_bridge_save_header(struct sk_buff *skb)
129 {
130         int header_size = ETH_HLEN;
131
132         if (skb->protocol == htons(ETH_P_8021Q))
133                 header_size += VLAN_HLEN;
134
135         memcpy(skb->nf_bridge->data, skb->data - header_size, header_size);
136 }
137
138 /*
139  * When forwarding bridge frames, we save a copy of the original
140  * header before processing.
141  */
142 int nf_bridge_copy_header(struct sk_buff *skb)
143 {
144         int err;
145         int header_size = ETH_HLEN;
146
147         if (skb->protocol == htons(ETH_P_8021Q))
148                 header_size += VLAN_HLEN;
149
150         err = skb_cow(skb, header_size);
151         if (err)
152                 return err;
153
154         memcpy(skb->data - header_size, skb->nf_bridge->data, header_size);
155
156         if (skb->protocol == htons(ETH_P_8021Q))
157                 __skb_push(skb, VLAN_HLEN);
158         return 0;
159 }
160
161 /* PF_BRIDGE/PRE_ROUTING *********************************************/
162 /* Undo the changes made for ip6tables PREROUTING and continue the
163  * bridge PRE_ROUTING hook. */
164 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
165 {
166         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
167
168         if (nf_bridge->mask & BRNF_PKT_TYPE) {
169                 skb->pkt_type = PACKET_OTHERHOST;
170                 nf_bridge->mask ^= BRNF_PKT_TYPE;
171         }
172         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
173
174         skb->dst = (struct dst_entry *)&__fake_rtable;
175         dst_hold(skb->dst);
176
177         skb->dev = nf_bridge->physindev;
178         if (skb->protocol == htons(ETH_P_8021Q)) {
179                 skb_push(skb, VLAN_HLEN);
180                 skb->nh.raw -= VLAN_HLEN;
181         }
182         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
183                        br_handle_frame_finish, 1);
184
185         return 0;
186 }
187
188 static void __br_dnat_complain(void)
189 {
190         static unsigned long last_complaint;
191
192         if (jiffies - last_complaint >= 5 * HZ) {
193                 printk(KERN_WARNING "Performing cross-bridge DNAT requires IP "
194                        "forwarding to be enabled\n");
195                 last_complaint = jiffies;
196         }
197 }
198
199 /* This requires some explaining. If DNAT has taken place,
200  * we will need to fix up the destination Ethernet address,
201  * and this is a tricky process.
202  *
203  * There are two cases to consider:
204  * 1. The packet was DNAT'ed to a device in the same bridge
205  *    port group as it was received on. We can still bridge
206  *    the packet.
207  * 2. The packet was DNAT'ed to a different device, either
208  *    a non-bridged device or another bridge port group.
209  *    The packet will need to be routed.
210  *
211  * The correct way of distinguishing between these two cases is to
212  * call ip_route_input() and to look at skb->dst->dev, which is
213  * changed to the destination device if ip_route_input() succeeds.
214  *
215  * Let us first consider the case that ip_route_input() succeeds:
216  *
217  * If skb->dst->dev equals the logical bridge device the packet
218  * came in on, we can consider this bridging. We then call
219  * skb->dst->output() which will make the packet enter br_nf_local_out()
220  * not much later. In that function it is assured that the iptables
221  * FORWARD chain is traversed for the packet.
222  *
223  * Otherwise, the packet is considered to be routed and we just
224  * change the destination MAC address so that the packet will
225  * later be passed up to the IP stack to be routed.
226  *
227  * Let us now consider the case that ip_route_input() fails:
228  *
229  * After a "echo '0' > /proc/sys/net/ipv4/ip_forward" ip_route_input()
230  * will fail, while __ip_route_output_key() will return success. The source
231  * address for __ip_route_output_key() is set to zero, so __ip_route_output_key
232  * thinks we're handling a locally generated packet and won't care
233  * if IP forwarding is allowed. We send a warning message to the users's
234  * log telling her to put IP forwarding on.
235  *
236  * ip_route_input() will also fail if there is no route available.
237  * In that case we just drop the packet.
238  *
239  * --Lennert, 20020411
240  * --Bart, 20020416 (updated)
241  * --Bart, 20021007 (updated) */
242 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
243 {
244         if (skb->pkt_type == PACKET_OTHERHOST) {
245                 skb->pkt_type = PACKET_HOST;
246                 skb->nf_bridge->mask |= BRNF_PKT_TYPE;
247         }
248         skb->nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
249
250         skb->dev = bridge_parent(skb->dev);
251         if (!skb->dev)
252                 kfree_skb(skb);
253         else {
254                 if (skb->protocol == htons(ETH_P_8021Q)) {
255                         skb_pull(skb, VLAN_HLEN);
256                         skb->nh.raw += VLAN_HLEN;
257                 }
258                 skb->dst->output(skb);
259         }
260         return 0;
261 }
262
263 static int br_nf_pre_routing_finish(struct sk_buff *skb)
264 {
265         struct net_device *dev = skb->dev;
266         struct iphdr *iph = skb->nh.iph;
267         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
268
269         if (nf_bridge->mask & BRNF_PKT_TYPE) {
270                 skb->pkt_type = PACKET_OTHERHOST;
271                 nf_bridge->mask ^= BRNF_PKT_TYPE;
272         }
273         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
274
275         if (dnat_took_place(skb)) {
276                 if (ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev)) {
277                         struct rtable *rt;
278                         struct flowi fl = {
279                                 .nl_u = {
280                                         .ip4_u = {
281                                                  .daddr = iph->daddr,
282                                                  .saddr = 0,
283                                                  .tos = RT_TOS(iph->tos) },
284                                 },
285                                 .proto = 0,
286                         };
287
288                         if (!ip_route_output_key(&rt, &fl)) {
289                                 /* - Bridged-and-DNAT'ed traffic doesn't
290                                  *   require ip_forwarding.
291                                  * - Deal with redirected traffic. */
292                                 if (((struct dst_entry *)rt)->dev == dev ||
293                                     rt->rt_type == RTN_LOCAL) {
294                                         skb->dst = (struct dst_entry *)rt;
295                                         goto bridged_dnat;
296                                 }
297                                 __br_dnat_complain();
298                                 dst_release((struct dst_entry *)rt);
299                         }
300                         kfree_skb(skb);
301                         return 0;
302                 } else {
303                         if (skb->dst->dev == dev) {
304 bridged_dnat:
305                                 /* Tell br_nf_local_out this is a
306                                  * bridged frame */
307                                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
308                                 skb->dev = nf_bridge->physindev;
309                                 if (skb->protocol ==
310                                     htons(ETH_P_8021Q)) {
311                                         skb_push(skb, VLAN_HLEN);
312                                         skb->nh.raw -= VLAN_HLEN;
313                                 }
314                                 NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING,
315                                                skb, skb->dev, NULL,
316                                                br_nf_pre_routing_finish_bridge,
317                                                1);
318                                 return 0;
319                         }
320                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
321                         skb->pkt_type = PACKET_HOST;
322                 }
323         } else {
324                 skb->dst = (struct dst_entry *)&__fake_rtable;
325                 dst_hold(skb->dst);
326         }
327
328         skb->dev = nf_bridge->physindev;
329         if (skb->protocol == htons(ETH_P_8021Q)) {
330                 skb_push(skb, VLAN_HLEN);
331                 skb->nh.raw -= VLAN_HLEN;
332         }
333         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
334                        br_handle_frame_finish, 1);
335
336         return 0;
337 }
338
339 /* Some common code for IPv4/IPv6 */
340 static struct net_device *setup_pre_routing(struct sk_buff *skb)
341 {
342         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
343
344         if (skb->pkt_type == PACKET_OTHERHOST) {
345                 skb->pkt_type = PACKET_HOST;
346                 nf_bridge->mask |= BRNF_PKT_TYPE;
347         }
348
349         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
350         nf_bridge->physindev = skb->dev;
351         skb->dev = bridge_parent(skb->dev);
352
353         return skb->dev;
354 }
355
356 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
357 static int check_hbh_len(struct sk_buff *skb)
358 {
359         unsigned char *raw = (u8 *) (skb->nh.ipv6h + 1);
360         u32 pkt_len;
361         int off = raw - skb->nh.raw;
362         int len = (raw[1] + 1) << 3;
363
364         if ((raw + len) - skb->data > skb_headlen(skb))
365                 goto bad;
366
367         off += 2;
368         len -= 2;
369
370         while (len > 0) {
371                 int optlen = skb->nh.raw[off + 1] + 2;
372
373                 switch (skb->nh.raw[off]) {
374                 case IPV6_TLV_PAD0:
375                         optlen = 1;
376                         break;
377
378                 case IPV6_TLV_PADN:
379                         break;
380
381                 case IPV6_TLV_JUMBO:
382                         if (skb->nh.raw[off + 1] != 4 || (off & 3) != 2)
383                                 goto bad;
384                         pkt_len = ntohl(*(u32 *) (skb->nh.raw + off + 2));
385                         if (pkt_len <= IPV6_MAXPLEN ||
386                             skb->nh.ipv6h->payload_len)
387                                 goto bad;
388                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
389                                 goto bad;
390                         if (pskb_trim_rcsum(skb,
391                                             pkt_len + sizeof(struct ipv6hdr)))
392                                 goto bad;
393                         break;
394                 default:
395                         if (optlen > len)
396                                 goto bad;
397                         break;
398                 }
399                 off += optlen;
400                 len -= optlen;
401         }
402         if (len == 0)
403                 return 0;
404 bad:
405         return -1;
406
407 }
408
409 /* Replicate the checks that IPv6 does on packet reception and pass the packet
410  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
411 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
412                                            struct sk_buff *skb,
413                                            const struct net_device *in,
414                                            const struct net_device *out,
415                                            int (*okfn)(struct sk_buff *))
416 {
417         struct ipv6hdr *hdr;
418         u32 pkt_len;
419
420         if (skb->len < sizeof(struct ipv6hdr))
421                 goto inhdr_error;
422
423         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
424                 goto inhdr_error;
425
426         hdr = skb->nh.ipv6h;
427
428         if (hdr->version != 6)
429                 goto inhdr_error;
430
431         pkt_len = ntohs(hdr->payload_len);
432
433         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
434                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
435                         goto inhdr_error;
436                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
437                         goto inhdr_error;
438         }
439         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
440                 goto inhdr_error;
441
442         nf_bridge_put(skb->nf_bridge);
443         if (!nf_bridge_alloc(skb))
444                 return NF_DROP;
445         if (!setup_pre_routing(skb))
446                 return NF_DROP;
447
448         NF_HOOK(PF_INET6, NF_IP6_PRE_ROUTING, skb, skb->dev, NULL,
449                 br_nf_pre_routing_finish_ipv6);
450
451         return NF_STOLEN;
452
453 inhdr_error:
454         return NF_DROP;
455 }
456
457 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
458  * Replicate the checks that IPv4 does on packet reception.
459  * Set skb->dev to the bridge device (i.e. parent of the
460  * receiving device) to make netfilter happy, the REDIRECT
461  * target in particular.  Save the original destination IP
462  * address to be able to detect DNAT afterwards. */
463 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff **pskb,
464                                       const struct net_device *in,
465                                       const struct net_device *out,
466                                       int (*okfn)(struct sk_buff *))
467 {
468         struct iphdr *iph;
469         __u32 len;
470         struct sk_buff *skb = *pskb;
471
472         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb)) {
473 #ifdef CONFIG_SYSCTL
474                 if (!brnf_call_ip6tables)
475                         return NF_ACCEPT;
476 #endif
477                 if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
478                         goto out;
479
480                 if (skb->protocol == htons(ETH_P_8021Q)) {
481                         skb_pull_rcsum(skb, VLAN_HLEN);
482                         skb->nh.raw += VLAN_HLEN;
483                 }
484                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
485         }
486 #ifdef CONFIG_SYSCTL
487         if (!brnf_call_iptables)
488                 return NF_ACCEPT;
489 #endif
490
491         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb))
492                 return NF_ACCEPT;
493
494         if ((skb = skb_share_check(*pskb, GFP_ATOMIC)) == NULL)
495                 goto out;
496
497         if (skb->protocol == htons(ETH_P_8021Q)) {
498                 skb_pull_rcsum(skb, VLAN_HLEN);
499                 skb->nh.raw += VLAN_HLEN;
500         }
501
502         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
503                 goto inhdr_error;
504
505         iph = skb->nh.iph;
506         if (iph->ihl < 5 || iph->version != 4)
507                 goto inhdr_error;
508
509         if (!pskb_may_pull(skb, 4 * iph->ihl))
510                 goto inhdr_error;
511
512         iph = skb->nh.iph;
513         if (ip_fast_csum((__u8 *) iph, iph->ihl) != 0)
514                 goto inhdr_error;
515
516         len = ntohs(iph->tot_len);
517         if (skb->len < len || len < 4 * iph->ihl)
518                 goto inhdr_error;
519
520         pskb_trim_rcsum(skb, len);
521
522         nf_bridge_put(skb->nf_bridge);
523         if (!nf_bridge_alloc(skb))
524                 return NF_DROP;
525         if (!setup_pre_routing(skb))
526                 return NF_DROP;
527         store_orig_dstaddr(skb);
528
529         NF_HOOK(PF_INET, NF_IP_PRE_ROUTING, skb, skb->dev, NULL,
530                 br_nf_pre_routing_finish);
531
532         return NF_STOLEN;
533
534 inhdr_error:
535 //      IP_INC_STATS_BH(IpInHdrErrors);
536 out:
537         return NF_DROP;
538 }
539
540
541 /* PF_BRIDGE/LOCAL_IN ************************************************/
542 /* The packet is locally destined, which requires a real
543  * dst_entry, so detach the fake one.  On the way up, the
544  * packet would pass through PRE_ROUTING again (which already
545  * took place when the packet entered the bridge), but we
546  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
547  * prevent this from happening. */
548 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff **pskb,
549                                    const struct net_device *in,
550                                    const struct net_device *out,
551                                    int (*okfn)(struct sk_buff *))
552 {
553         struct sk_buff *skb = *pskb;
554
555         if (skb->dst == (struct dst_entry *)&__fake_rtable) {
556                 dst_release(skb->dst);
557                 skb->dst = NULL;
558         }
559
560         return NF_ACCEPT;
561 }
562
563 /* PF_BRIDGE/FORWARD *************************************************/
564 static int br_nf_forward_finish(struct sk_buff *skb)
565 {
566         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
567         struct net_device *in;
568
569         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
570                 in = nf_bridge->physindev;
571                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
572                         skb->pkt_type = PACKET_OTHERHOST;
573                         nf_bridge->mask ^= BRNF_PKT_TYPE;
574                 }
575         } else {
576                 in = *((struct net_device **)(skb->cb));
577         }
578         if (skb->protocol == htons(ETH_P_8021Q)) {
579                 skb_push(skb, VLAN_HLEN);
580                 skb->nh.raw -= VLAN_HLEN;
581         }
582         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_FORWARD, skb, in,
583                        skb->dev, br_forward_finish, 1);
584         return 0;
585 }
586
587 /* This is the 'purely bridged' case.  For IP, we pass the packet to
588  * netfilter with indev and outdev set to the bridge device,
589  * but we are still able to filter on the 'real' indev/outdev
590  * because of the physdev module. For ARP, indev and outdev are the
591  * bridge ports. */
592 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff **pskb,
593                                      const struct net_device *in,
594                                      const struct net_device *out,
595                                      int (*okfn)(struct sk_buff *))
596 {
597         struct sk_buff *skb = *pskb;
598         struct nf_bridge_info *nf_bridge;
599         struct net_device *parent;
600         int pf;
601
602         if (!skb->nf_bridge)
603                 return NF_ACCEPT;
604
605         parent = bridge_parent(out);
606         if (!parent)
607                 return NF_DROP;
608
609         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
610                 pf = PF_INET;
611         else
612                 pf = PF_INET6;
613
614         if (skb->protocol == htons(ETH_P_8021Q)) {
615                 skb_pull(*pskb, VLAN_HLEN);
616                 (*pskb)->nh.raw += VLAN_HLEN;
617         }
618
619         nf_bridge = skb->nf_bridge;
620         if (skb->pkt_type == PACKET_OTHERHOST) {
621                 skb->pkt_type = PACKET_HOST;
622                 nf_bridge->mask |= BRNF_PKT_TYPE;
623         }
624
625         /* The physdev module checks on this */
626         nf_bridge->mask |= BRNF_BRIDGED;
627         nf_bridge->physoutdev = skb->dev;
628
629         NF_HOOK(pf, NF_IP_FORWARD, skb, bridge_parent(in), parent,
630                 br_nf_forward_finish);
631
632         return NF_STOLEN;
633 }
634
635 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff **pskb,
636                                       const struct net_device *in,
637                                       const struct net_device *out,
638                                       int (*okfn)(struct sk_buff *))
639 {
640         struct sk_buff *skb = *pskb;
641         struct net_device **d = (struct net_device **)(skb->cb);
642
643 #ifdef CONFIG_SYSCTL
644         if (!brnf_call_arptables)
645                 return NF_ACCEPT;
646 #endif
647
648         if (skb->protocol != htons(ETH_P_ARP)) {
649                 if (!IS_VLAN_ARP(skb))
650                         return NF_ACCEPT;
651                 skb_pull(*pskb, VLAN_HLEN);
652                 (*pskb)->nh.raw += VLAN_HLEN;
653         }
654
655         if (skb->nh.arph->ar_pln != 4) {
656                 if (IS_VLAN_ARP(skb)) {
657                         skb_push(*pskb, VLAN_HLEN);
658                         (*pskb)->nh.raw -= VLAN_HLEN;
659                 }
660                 return NF_ACCEPT;
661         }
662         *d = (struct net_device *)in;
663         NF_HOOK(NF_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
664                 (struct net_device *)out, br_nf_forward_finish);
665
666         return NF_STOLEN;
667 }
668
669 /* PF_BRIDGE/LOCAL_OUT ***********************************************/
670 static int br_nf_local_out_finish(struct sk_buff *skb)
671 {
672         if (skb->protocol == htons(ETH_P_8021Q)) {
673                 skb_push(skb, VLAN_HLEN);
674                 skb->nh.raw -= VLAN_HLEN;
675         }
676
677         NF_HOOK_THRESH(PF_BRIDGE, NF_BR_LOCAL_OUT, skb, NULL, skb->dev,
678                        br_forward_finish, NF_BR_PRI_FIRST + 1);
679
680         return 0;
681 }
682
683 /* This function sees both locally originated IP packets and forwarded
684  * IP packets (in both cases the destination device is a bridge
685  * device). It also sees bridged-and-DNAT'ed packets.
686  * To be able to filter on the physical bridge devices (with the physdev
687  * module), we steal packets destined to a bridge device away from the
688  * PF_INET/FORWARD and PF_INET/OUTPUT hook functions, and give them back later,
689  * when we have determined the real output device. This is done in here.
690  *
691  * If (nf_bridge->mask & BRNF_BRIDGED_DNAT) then the packet is bridged
692  * and we fake the PF_BRIDGE/FORWARD hook. The function br_nf_forward()
693  * will then fake the PF_INET/FORWARD hook. br_nf_local_out() has priority
694  * NF_BR_PRI_FIRST, so no relevant PF_BRIDGE/INPUT functions have been nor
695  * will be executed.
696  * Otherwise, if nf_bridge->physindev is NULL, the bridge-nf code never touched
697  * this packet before, and so the packet was locally originated. We fake
698  * the PF_INET/LOCAL_OUT hook.
699  * Finally, if nf_bridge->physindev isn't NULL, then the packet was IP routed,
700  * so we fake the PF_INET/FORWARD hook. ip_sabotage_out() makes sure
701  * even routed packets that didn't arrive on a bridge interface have their
702  * nf_bridge->physindev set. */
703 static unsigned int br_nf_local_out(unsigned int hook, struct sk_buff **pskb,
704                                     const struct net_device *in,
705                                     const struct net_device *out,
706                                     int (*okfn)(struct sk_buff *))
707 {
708         struct net_device *realindev, *realoutdev;
709         struct sk_buff *skb = *pskb;
710         struct nf_bridge_info *nf_bridge;
711         int pf;
712
713         if (!skb->nf_bridge)
714                 return NF_ACCEPT;
715
716         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
717                 pf = PF_INET;
718         else
719                 pf = PF_INET6;
720
721 #ifdef CONFIG_NETFILTER_DEBUG
722         /* Sometimes we get packets with NULL ->dst here (for example,
723          * running a dhcp client daemon triggers this). This should now
724          * be fixed, but let's keep the check around. */
725         if (skb->dst == NULL) {
726                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
727                 return NF_ACCEPT;
728         }
729 #endif
730
731         nf_bridge = skb->nf_bridge;
732         nf_bridge->physoutdev = skb->dev;
733         realindev = nf_bridge->physindev;
734
735         /* Bridged, take PF_BRIDGE/FORWARD.
736          * (see big note in front of br_nf_pre_routing_finish) */
737         if (nf_bridge->mask & BRNF_BRIDGED_DNAT) {
738                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
739                         skb->pkt_type = PACKET_OTHERHOST;
740                         nf_bridge->mask ^= BRNF_PKT_TYPE;
741                 }
742                 if (skb->protocol == htons(ETH_P_8021Q)) {
743                         skb_push(skb, VLAN_HLEN);
744                         skb->nh.raw -= VLAN_HLEN;
745                 }
746
747                 NF_HOOK(PF_BRIDGE, NF_BR_FORWARD, skb, realindev,
748                         skb->dev, br_forward_finish);
749                 goto out;
750         }
751         realoutdev = bridge_parent(skb->dev);
752         if (!realoutdev)
753                 return NF_DROP;
754
755 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
756         /* iptables should match -o br0.x */
757         if (nf_bridge->netoutdev)
758                 realoutdev = nf_bridge->netoutdev;
759 #endif
760         if (skb->protocol == htons(ETH_P_8021Q)) {
761                 skb_pull(skb, VLAN_HLEN);
762                 (*pskb)->nh.raw += VLAN_HLEN;
763         }
764         /* IP forwarded traffic has a physindev, locally
765          * generated traffic hasn't. */
766         if (realindev != NULL) {
767                 if (!(nf_bridge->mask & BRNF_DONT_TAKE_PARENT)) {
768                         struct net_device *parent = bridge_parent(realindev);
769                         if (parent)
770                                 realindev = parent;
771                 }
772
773                 NF_HOOK_THRESH(pf, NF_IP_FORWARD, skb, realindev,
774                                realoutdev, br_nf_local_out_finish,
775                                NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD + 1);
776         } else {
777                 NF_HOOK_THRESH(pf, NF_IP_LOCAL_OUT, skb, realindev,
778                                realoutdev, br_nf_local_out_finish,
779                                NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT + 1);
780         }
781
782 out:
783         return NF_STOLEN;
784 }
785
786 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
787 {
788         if (skb->protocol == htons(ETH_P_IP) &&
789             skb->len > skb->dev->mtu &&
790             !skb_is_gso(skb))
791                 return ip_fragment(skb, br_dev_queue_push_xmit);
792         else
793                 return br_dev_queue_push_xmit(skb);
794 }
795
796 /* PF_BRIDGE/POST_ROUTING ********************************************/
797 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff **pskb,
798                                        const struct net_device *in,
799                                        const struct net_device *out,
800                                        int (*okfn)(struct sk_buff *))
801 {
802         struct sk_buff *skb = *pskb;
803         struct nf_bridge_info *nf_bridge = (*pskb)->nf_bridge;
804         struct net_device *realoutdev = bridge_parent(skb->dev);
805         int pf;
806
807 #ifdef CONFIG_NETFILTER_DEBUG
808         /* Be very paranoid. This probably won't happen anymore, but let's
809          * keep the check just to be sure... */
810         if (skb->mac.raw < skb->head || skb->mac.raw + ETH_HLEN > skb->data) {
811                 printk(KERN_CRIT "br_netfilter: Argh!! br_nf_post_routing: "
812                        "bad mac.raw pointer.");
813                 goto print_error;
814         }
815 #endif
816
817         if (!nf_bridge)
818                 return NF_ACCEPT;
819
820         if (!realoutdev)
821                 return NF_DROP;
822
823         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb))
824                 pf = PF_INET;
825         else
826                 pf = PF_INET6;
827
828 #ifdef CONFIG_NETFILTER_DEBUG
829         if (skb->dst == NULL) {
830                 printk(KERN_CRIT "br_netfilter: skb->dst == NULL.");
831                 goto print_error;
832         }
833 #endif
834
835         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
836          * about the value of skb->pkt_type. */
837         if (skb->pkt_type == PACKET_OTHERHOST) {
838                 skb->pkt_type = PACKET_HOST;
839                 nf_bridge->mask |= BRNF_PKT_TYPE;
840         }
841
842         if (skb->protocol == htons(ETH_P_8021Q)) {
843                 skb_pull(skb, VLAN_HLEN);
844                 skb->nh.raw += VLAN_HLEN;
845         }
846
847         nf_bridge_save_header(skb);
848
849 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
850         if (nf_bridge->netoutdev)
851                 realoutdev = nf_bridge->netoutdev;
852 #endif
853         NF_HOOK(pf, NF_IP_POST_ROUTING, skb, NULL, realoutdev,
854                 br_nf_dev_queue_xmit);
855
856         return NF_STOLEN;
857
858 #ifdef CONFIG_NETFILTER_DEBUG
859 print_error:
860         if (skb->dev != NULL) {
861                 printk("[%s]", skb->dev->name);
862                 if (realoutdev)
863                         printk("[%s]", realoutdev->name);
864         }
865         printk(" head:%p, raw:%p, data:%p\n", skb->head, skb->mac.raw,
866                skb->data);
867         return NF_ACCEPT;
868 #endif
869 }
870
871 /* IP/SABOTAGE *****************************************************/
872 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
873  * for the second time. */
874 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff **pskb,
875                                    const struct net_device *in,
876                                    const struct net_device *out,
877                                    int (*okfn)(struct sk_buff *))
878 {
879         if ((*pskb)->nf_bridge &&
880             !((*pskb)->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
881                 return NF_STOP;
882         }
883
884         return NF_ACCEPT;
885 }
886
887 /* Postpone execution of PF_INET(6)/FORWARD, PF_INET(6)/LOCAL_OUT
888  * and PF_INET(6)/POST_ROUTING until we have done the forwarding
889  * decision in the bridge code and have determined nf_bridge->physoutdev. */
890 static unsigned int ip_sabotage_out(unsigned int hook, struct sk_buff **pskb,
891                                     const struct net_device *in,
892                                     const struct net_device *out,
893                                     int (*okfn)(struct sk_buff *))
894 {
895         struct sk_buff *skb = *pskb;
896
897         if ((out->hard_start_xmit == br_dev_xmit &&
898              okfn != br_nf_forward_finish &&
899              okfn != br_nf_local_out_finish && okfn != br_nf_dev_queue_xmit)
900 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
901             || ((out->priv_flags & IFF_802_1Q_VLAN) &&
902                 VLAN_DEV_INFO(out)->real_dev->hard_start_xmit == br_dev_xmit)
903 #endif
904             ) {
905                 struct nf_bridge_info *nf_bridge;
906
907                 if (!skb->nf_bridge) {
908 #ifdef CONFIG_SYSCTL
909                         /* This code is executed while in the IP(v6) stack,
910                            the version should be 4 or 6. We can't use
911                            skb->protocol because that isn't set on
912                            PF_INET(6)/LOCAL_OUT. */
913                         struct iphdr *ip = skb->nh.iph;
914
915                         if (ip->version == 4 && !brnf_call_iptables)
916                                 return NF_ACCEPT;
917                         else if (ip->version == 6 && !brnf_call_ip6tables)
918                                 return NF_ACCEPT;
919                         else if (!brnf_deferred_hooks)
920                                 return NF_ACCEPT;
921 #endif
922                         if (hook == NF_IP_POST_ROUTING)
923                                 return NF_ACCEPT;
924                         if (!nf_bridge_alloc(skb))
925                                 return NF_DROP;
926                 }
927
928                 nf_bridge = skb->nf_bridge;
929
930                 /* This frame will arrive on PF_BRIDGE/LOCAL_OUT and we
931                  * will need the indev then. For a brouter, the real indev
932                  * can be a bridge port, so we make sure br_nf_local_out()
933                  * doesn't use the bridge parent of the indev by using
934                  * the BRNF_DONT_TAKE_PARENT mask. */
935                 if (hook == NF_IP_FORWARD && nf_bridge->physindev == NULL) {
936                         nf_bridge->mask |= BRNF_DONT_TAKE_PARENT;
937                         nf_bridge->physindev = (struct net_device *)in;
938                 }
939 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
940                 /* the iptables outdev is br0.x, not br0 */
941                 if (out->priv_flags & IFF_802_1Q_VLAN)
942                         nf_bridge->netoutdev = (struct net_device *)out;
943 #endif
944                 return NF_STOP;
945         }
946
947         return NF_ACCEPT;
948 }
949
950 /* For br_nf_local_out we need (prio = NF_BR_PRI_FIRST), to insure that innocent
951  * PF_BRIDGE/NF_BR_LOCAL_OUT functions don't get bridged traffic as input.
952  * For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
953  * ip_refrag() can return NF_STOLEN. */
954 static struct nf_hook_ops br_nf_ops[] = {
955         { .hook = br_nf_pre_routing, 
956           .owner = THIS_MODULE, 
957           .pf = PF_BRIDGE, 
958           .hooknum = NF_BR_PRE_ROUTING, 
959           .priority = NF_BR_PRI_BRNF, },
960         { .hook = br_nf_local_in,
961           .owner = THIS_MODULE,
962           .pf = PF_BRIDGE,
963           .hooknum = NF_BR_LOCAL_IN,
964           .priority = NF_BR_PRI_BRNF, },
965         { .hook = br_nf_forward_ip,
966           .owner = THIS_MODULE,
967           .pf = PF_BRIDGE,
968           .hooknum = NF_BR_FORWARD,
969           .priority = NF_BR_PRI_BRNF - 1, },
970         { .hook = br_nf_forward_arp,
971           .owner = THIS_MODULE,
972           .pf = PF_BRIDGE,
973           .hooknum = NF_BR_FORWARD,
974           .priority = NF_BR_PRI_BRNF, },
975         { .hook = br_nf_local_out,
976           .owner = THIS_MODULE,
977           .pf = PF_BRIDGE,
978           .hooknum = NF_BR_LOCAL_OUT,
979           .priority = NF_BR_PRI_FIRST, },
980         { .hook = br_nf_post_routing,
981           .owner = THIS_MODULE,
982           .pf = PF_BRIDGE,
983           .hooknum = NF_BR_POST_ROUTING,
984           .priority = NF_BR_PRI_LAST, },
985         { .hook = ip_sabotage_in,
986           .owner = THIS_MODULE,
987           .pf = PF_INET,
988           .hooknum = NF_IP_PRE_ROUTING,
989           .priority = NF_IP_PRI_FIRST, },
990         { .hook = ip_sabotage_in,
991           .owner = THIS_MODULE,
992           .pf = PF_INET6,
993           .hooknum = NF_IP6_PRE_ROUTING,
994           .priority = NF_IP6_PRI_FIRST, },
995         { .hook = ip_sabotage_out,
996           .owner = THIS_MODULE,
997           .pf = PF_INET,
998           .hooknum = NF_IP_FORWARD,
999           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_FORWARD, },
1000         { .hook = ip_sabotage_out,
1001           .owner = THIS_MODULE,
1002           .pf = PF_INET6,
1003           .hooknum = NF_IP6_FORWARD,
1004           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_FORWARD, },
1005         { .hook = ip_sabotage_out,
1006           .owner = THIS_MODULE,
1007           .pf = PF_INET,
1008           .hooknum = NF_IP_LOCAL_OUT,
1009           .priority = NF_IP_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
1010         { .hook = ip_sabotage_out,
1011           .owner = THIS_MODULE,
1012           .pf = PF_INET6,
1013           .hooknum = NF_IP6_LOCAL_OUT,
1014           .priority = NF_IP6_PRI_BRIDGE_SABOTAGE_LOCAL_OUT, },
1015         { .hook = ip_sabotage_out,
1016           .owner = THIS_MODULE,
1017           .pf = PF_INET,
1018           .hooknum = NF_IP_POST_ROUTING,
1019           .priority = NF_IP_PRI_FIRST, },
1020         { .hook = ip_sabotage_out,
1021           .owner = THIS_MODULE,
1022           .pf = PF_INET6,
1023           .hooknum = NF_IP6_POST_ROUTING,
1024           .priority = NF_IP6_PRI_FIRST, },
1025 };
1026
1027 #ifdef CONFIG_SYSCTL
1028 static
1029 int brnf_sysctl_call_tables(ctl_table * ctl, int write, struct file *filp,
1030                             void __user * buffer, size_t * lenp, loff_t * ppos)
1031 {
1032         int ret;
1033
1034         ret = proc_dointvec(ctl, write, filp, buffer, lenp, ppos);
1035
1036         if (write && *(int *)(ctl->data))
1037                 *(int *)(ctl->data) = 1;
1038         return ret;
1039 }
1040
1041 static ctl_table brnf_table[] = {
1042         {
1043                 .ctl_name       = NET_BRIDGE_NF_CALL_ARPTABLES,
1044                 .procname       = "bridge-nf-call-arptables",
1045                 .data           = &brnf_call_arptables,
1046                 .maxlen         = sizeof(int),
1047                 .mode           = 0644,
1048                 .proc_handler   = &brnf_sysctl_call_tables,
1049         },
1050         {
1051                 .ctl_name       = NET_BRIDGE_NF_CALL_IPTABLES,
1052                 .procname       = "bridge-nf-call-iptables",
1053                 .data           = &brnf_call_iptables,
1054                 .maxlen         = sizeof(int),
1055                 .mode           = 0644,
1056                 .proc_handler   = &brnf_sysctl_call_tables,
1057         },
1058         {
1059                 .ctl_name       = NET_BRIDGE_NF_CALL_IP6TABLES,
1060                 .procname       = "bridge-nf-call-ip6tables",
1061                 .data           = &brnf_call_ip6tables,
1062                 .maxlen         = sizeof(int),
1063                 .mode           = 0644,
1064                 .proc_handler   = &brnf_sysctl_call_tables,
1065         },
1066         {
1067                 .ctl_name       = NET_BRIDGE_NF_FILTER_VLAN_TAGGED,
1068                 .procname       = "bridge-nf-filter-vlan-tagged",
1069                 .data           = &brnf_filter_vlan_tagged,
1070                 .maxlen         = sizeof(int),
1071                 .mode           = 0644,
1072                 .proc_handler   = &brnf_sysctl_call_tables,
1073         },
1074         { .ctl_name = 0 }
1075 };
1076
1077 static ctl_table brnf_bridge_table[] = {
1078         {
1079                 .ctl_name       = NET_BRIDGE,
1080                 .procname       = "bridge",
1081                 .mode           = 0555,
1082                 .child          = brnf_table,
1083         },
1084         { .ctl_name = 0 }
1085 };
1086
1087 static ctl_table brnf_net_table[] = {
1088         {
1089                 .ctl_name       = CTL_NET,
1090                 .procname       = "net",
1091                 .mode           = 0555,
1092                 .child          = brnf_bridge_table,
1093         },
1094         { .ctl_name = 0 }
1095 };
1096 #endif
1097
1098 int br_netfilter_init(void)
1099 {
1100         int i;
1101
1102         for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++) {
1103                 int ret;
1104
1105                 if ((ret = nf_register_hook(&br_nf_ops[i])) >= 0)
1106                         continue;
1107
1108                 while (i--)
1109                         nf_unregister_hook(&br_nf_ops[i]);
1110
1111                 return ret;
1112         }
1113
1114 #ifdef CONFIG_SYSCTL
1115         brnf_sysctl_header = register_sysctl_table(brnf_net_table, 0);
1116         if (brnf_sysctl_header == NULL) {
1117                 printk(KERN_WARNING
1118                        "br_netfilter: can't register to sysctl.\n");
1119                 for (i = 0; i < ARRAY_SIZE(br_nf_ops); i++)
1120                         nf_unregister_hook(&br_nf_ops[i]);
1121                 return -EFAULT;
1122         }
1123 #endif
1124
1125         printk(KERN_NOTICE "Bridge firewalling registered\n");
1126
1127         return 0;
1128 }
1129
1130 void br_netfilter_fini(void)
1131 {
1132         int i;
1133
1134         for (i = ARRAY_SIZE(br_nf_ops) - 1; i >= 0; i--)
1135                 nf_unregister_hook(&br_nf_ops[i]);
1136 #ifdef CONFIG_SYSCTL
1137         unregister_sysctl_table(brnf_sysctl_header);
1138 #endif
1139 }