neigh: Pass neighbour entry to output ops.
[linux-3.10.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                <bdschuym@pandora.be>
8  *
9  *      This program is free software; you can redistribute it and/or
10  *      modify it under the terms of the GNU General Public License
11  *      as published by the Free Software Foundation; either version
12  *      2 of the License, or (at your option) any later version.
13  *
14  *      Lennert dedicates this file to Kerstin Wurdinger.
15  */
16
17 #include <linux/module.h>
18 #include <linux/kernel.h>
19 #include <linux/slab.h>
20 #include <linux/ip.h>
21 #include <linux/netdevice.h>
22 #include <linux/skbuff.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <linux/if_pppox.h>
27 #include <linux/ppp_defs.h>
28 #include <linux/netfilter_bridge.h>
29 #include <linux/netfilter_ipv4.h>
30 #include <linux/netfilter_ipv6.h>
31 #include <linux/netfilter_arp.h>
32 #include <linux/in_route.h>
33 #include <linux/inetdevice.h>
34
35 #include <net/ip.h>
36 #include <net/ipv6.h>
37 #include <net/route.h>
38
39 #include <asm/uaccess.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) = ip_hdr(skb)->daddr)
48 #define dnat_took_place(skb)     (skb_origaddr(skb) != ip_hdr(skb)->daddr)
49
50 #ifdef CONFIG_SYSCTL
51 static struct ctl_table_header *brnf_sysctl_header;
52 static int brnf_call_iptables __read_mostly = 1;
53 static int brnf_call_ip6tables __read_mostly = 1;
54 static int brnf_call_arptables __read_mostly = 1;
55 static int brnf_filter_vlan_tagged __read_mostly = 0;
56 static int brnf_filter_pppoe_tagged __read_mostly = 0;
57 #else
58 #define brnf_call_iptables 1
59 #define brnf_call_ip6tables 1
60 #define brnf_call_arptables 1
61 #define brnf_filter_vlan_tagged 0
62 #define brnf_filter_pppoe_tagged 0
63 #endif
64
65 static inline __be16 vlan_proto(const struct sk_buff *skb)
66 {
67         if (vlan_tx_tag_present(skb))
68                 return skb->protocol;
69         else if (skb->protocol == htons(ETH_P_8021Q))
70                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
71         else
72                 return 0;
73 }
74
75 #define IS_VLAN_IP(skb) \
76         (vlan_proto(skb) == htons(ETH_P_IP) && \
77          brnf_filter_vlan_tagged)
78
79 #define IS_VLAN_IPV6(skb) \
80         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
81          brnf_filter_vlan_tagged)
82
83 #define IS_VLAN_ARP(skb) \
84         (vlan_proto(skb) == htons(ETH_P_ARP) && \
85          brnf_filter_vlan_tagged)
86
87 static inline __be16 pppoe_proto(const struct sk_buff *skb)
88 {
89         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
90                             sizeof(struct pppoe_hdr)));
91 }
92
93 #define IS_PPPOE_IP(skb) \
94         (skb->protocol == htons(ETH_P_PPP_SES) && \
95          pppoe_proto(skb) == htons(PPP_IP) && \
96          brnf_filter_pppoe_tagged)
97
98 #define IS_PPPOE_IPV6(skb) \
99         (skb->protocol == htons(ETH_P_PPP_SES) && \
100          pppoe_proto(skb) == htons(PPP_IPV6) && \
101          brnf_filter_pppoe_tagged)
102
103 static void fake_update_pmtu(struct dst_entry *dst, u32 mtu)
104 {
105 }
106
107 static u32 *fake_cow_metrics(struct dst_entry *dst, unsigned long old)
108 {
109         return NULL;
110 }
111
112 static struct dst_ops fake_dst_ops = {
113         .family =               AF_INET,
114         .protocol =             cpu_to_be16(ETH_P_IP),
115         .update_pmtu =          fake_update_pmtu,
116         .cow_metrics =          fake_cow_metrics,
117 };
118
119 /*
120  * Initialize bogus route table used to keep netfilter happy.
121  * Currently, we fill in the PMTU entry because netfilter
122  * refragmentation needs it, and the rt_flags entry because
123  * ipt_REJECT needs it.  Future netfilter modules might
124  * require us to fill additional fields.
125  */
126 static const u32 br_dst_default_metrics[RTAX_MAX] = {
127         [RTAX_MTU - 1] = 1500,
128 };
129
130 void br_netfilter_rtable_init(struct net_bridge *br)
131 {
132         struct rtable *rt = &br->fake_rtable;
133
134         atomic_set(&rt->dst.__refcnt, 1);
135         rt->dst.dev = br->dev;
136         rt->dst.path = &rt->dst;
137         dst_init_metrics(&rt->dst, br_dst_default_metrics, true);
138         rt->dst.flags   = DST_NOXFRM;
139         rt->dst.ops = &fake_dst_ops;
140 }
141
142 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
143 {
144         struct net_bridge_port *port;
145
146         port = br_port_get_rcu(dev);
147         return port ? &port->br->fake_rtable : NULL;
148 }
149
150 static inline struct net_device *bridge_parent(const struct net_device *dev)
151 {
152         struct net_bridge_port *port;
153
154         port = br_port_get_rcu(dev);
155         return port ? port->br->dev : NULL;
156 }
157
158 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
159 {
160         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
161         if (likely(skb->nf_bridge))
162                 atomic_set(&(skb->nf_bridge->use), 1);
163
164         return skb->nf_bridge;
165 }
166
167 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
168 {
169         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
170
171         if (atomic_read(&nf_bridge->use) > 1) {
172                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
173
174                 if (tmp) {
175                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
176                         atomic_set(&tmp->use, 1);
177                 }
178                 nf_bridge_put(nf_bridge);
179                 nf_bridge = tmp;
180         }
181         return nf_bridge;
182 }
183
184 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
185 {
186         unsigned int len = nf_bridge_encap_header_len(skb);
187
188         skb_push(skb, len);
189         skb->network_header -= len;
190 }
191
192 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
193 {
194         unsigned int len = nf_bridge_encap_header_len(skb);
195
196         skb_pull(skb, len);
197         skb->network_header += len;
198 }
199
200 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
201 {
202         unsigned int len = nf_bridge_encap_header_len(skb);
203
204         skb_pull_rcsum(skb, len);
205         skb->network_header += len;
206 }
207
208 static inline void nf_bridge_save_header(struct sk_buff *skb)
209 {
210         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
211
212         skb_copy_from_linear_data_offset(skb, -header_size,
213                                          skb->nf_bridge->data, header_size);
214 }
215
216 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
217 {
218         if (skb->nf_bridge->mask & BRNF_8021Q)
219                 skb->protocol = htons(ETH_P_8021Q);
220         else if (skb->nf_bridge->mask & BRNF_PPPoE)
221                 skb->protocol = htons(ETH_P_PPP_SES);
222 }
223
224 /* When handing a packet over to the IP layer
225  * check whether we have a skb that is in the
226  * expected format
227  */
228
229 static int br_parse_ip_options(struct sk_buff *skb)
230 {
231         struct ip_options *opt;
232         const struct iphdr *iph;
233         struct net_device *dev = skb->dev;
234         u32 len;
235
236         iph = ip_hdr(skb);
237         opt = &(IPCB(skb)->opt);
238
239         /* Basic sanity checks */
240         if (iph->ihl < 5 || iph->version != 4)
241                 goto inhdr_error;
242
243         if (!pskb_may_pull(skb, iph->ihl*4))
244                 goto inhdr_error;
245
246         iph = ip_hdr(skb);
247         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
248                 goto inhdr_error;
249
250         len = ntohs(iph->tot_len);
251         if (skb->len < len) {
252                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
253                 goto drop;
254         } else if (len < (iph->ihl*4))
255                 goto inhdr_error;
256
257         if (pskb_trim_rcsum(skb, len)) {
258                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
259                 goto drop;
260         }
261
262         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
263         if (iph->ihl == 5)
264                 return 0;
265
266         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
267         if (ip_options_compile(dev_net(dev), opt, skb))
268                 goto inhdr_error;
269
270         /* Check correct handling of SRR option */
271         if (unlikely(opt->srr)) {
272                 struct in_device *in_dev = __in_dev_get_rcu(dev);
273                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
274                         goto drop;
275
276                 if (ip_options_rcv_srr(skb))
277                         goto drop;
278         }
279
280         return 0;
281
282 inhdr_error:
283         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
284 drop:
285         return -1;
286 }
287
288 /* Fill in the header for fragmented IP packets handled by
289  * the IPv4 connection tracking code.
290  */
291 int nf_bridge_copy_header(struct sk_buff *skb)
292 {
293         int err;
294         unsigned int header_size;
295
296         nf_bridge_update_protocol(skb);
297         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
298         err = skb_cow_head(skb, header_size);
299         if (err)
300                 return err;
301
302         skb_copy_to_linear_data_offset(skb, -header_size,
303                                        skb->nf_bridge->data, header_size);
304         __skb_push(skb, nf_bridge_encap_header_len(skb));
305         return 0;
306 }
307
308 /* PF_BRIDGE/PRE_ROUTING *********************************************/
309 /* Undo the changes made for ip6tables PREROUTING and continue the
310  * bridge PRE_ROUTING hook. */
311 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
312 {
313         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
314         struct rtable *rt;
315
316         if (nf_bridge->mask & BRNF_PKT_TYPE) {
317                 skb->pkt_type = PACKET_OTHERHOST;
318                 nf_bridge->mask ^= BRNF_PKT_TYPE;
319         }
320         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
321
322         rt = bridge_parent_rtable(nf_bridge->physindev);
323         if (!rt) {
324                 kfree_skb(skb);
325                 return 0;
326         }
327         skb_dst_set_noref(skb, &rt->dst);
328
329         skb->dev = nf_bridge->physindev;
330         nf_bridge_update_protocol(skb);
331         nf_bridge_push_encap_header(skb);
332         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
333                        br_handle_frame_finish, 1);
334
335         return 0;
336 }
337
338 /* Obtain the correct destination MAC address, while preserving the original
339  * source MAC address. If we already know this address, we just copy it. If we
340  * don't, we use the neighbour framework to find out. In both cases, we make
341  * sure that br_handle_frame_finish() is called afterwards.
342  */
343 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
344 {
345         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
346         struct neighbour *neigh;
347         struct dst_entry *dst;
348
349         skb->dev = bridge_parent(skb->dev);
350         if (!skb->dev)
351                 goto free_skb;
352         dst = skb_dst(skb);
353         neigh = dst->neighbour;
354         if (neigh->hh.hh_len) {
355                 neigh_hh_bridge(&neigh->hh, skb);
356                 skb->dev = nf_bridge->physindev;
357                 return br_handle_frame_finish(skb);
358         } else {
359                 /* the neighbour function below overwrites the complete
360                  * MAC header, so we save the Ethernet source address and
361                  * protocol number. */
362                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
363                 /* tell br_dev_xmit to continue with forwarding */
364                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
365                 return neigh->output(neigh, skb);
366         }
367 free_skb:
368         kfree_skb(skb);
369         return 0;
370 }
371
372 /* This requires some explaining. If DNAT has taken place,
373  * we will need to fix up the destination Ethernet address.
374  *
375  * There are two cases to consider:
376  * 1. The packet was DNAT'ed to a device in the same bridge
377  *    port group as it was received on. We can still bridge
378  *    the packet.
379  * 2. The packet was DNAT'ed to a different device, either
380  *    a non-bridged device or another bridge port group.
381  *    The packet will need to be routed.
382  *
383  * The correct way of distinguishing between these two cases is to
384  * call ip_route_input() and to look at skb->dst->dev, which is
385  * changed to the destination device if ip_route_input() succeeds.
386  *
387  * Let's first consider the case that ip_route_input() succeeds:
388  *
389  * If the output device equals the logical bridge device the packet
390  * came in on, we can consider this bridging. The corresponding MAC
391  * address will be obtained in br_nf_pre_routing_finish_bridge.
392  * Otherwise, the packet is considered to be routed and we just
393  * change the destination MAC address so that the packet will
394  * later be passed up to the IP stack to be routed. For a redirected
395  * packet, ip_route_input() will give back the localhost as output device,
396  * which differs from the bridge device.
397  *
398  * Let's now consider the case that ip_route_input() fails:
399  *
400  * This can be because the destination address is martian, in which case
401  * the packet will be dropped.
402  * If IP forwarding is disabled, ip_route_input() will fail, while
403  * ip_route_output_key() can return success. The source
404  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
405  * thinks we're handling a locally generated packet and won't care
406  * if IP forwarding is enabled. If the output device equals the logical bridge
407  * device, we proceed as if ip_route_input() succeeded. If it differs from the
408  * logical bridge port or if ip_route_output_key() fails we drop the packet.
409  */
410 static int br_nf_pre_routing_finish(struct sk_buff *skb)
411 {
412         struct net_device *dev = skb->dev;
413         struct iphdr *iph = ip_hdr(skb);
414         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
415         struct rtable *rt;
416         int err;
417
418         if (nf_bridge->mask & BRNF_PKT_TYPE) {
419                 skb->pkt_type = PACKET_OTHERHOST;
420                 nf_bridge->mask ^= BRNF_PKT_TYPE;
421         }
422         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
423         if (dnat_took_place(skb)) {
424                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
425                         struct in_device *in_dev = __in_dev_get_rcu(dev);
426
427                         /* If err equals -EHOSTUNREACH the error is due to a
428                          * martian destination or due to the fact that
429                          * forwarding is disabled. For most martian packets,
430                          * ip_route_output_key() will fail. It won't fail for 2 types of
431                          * martian destinations: loopback destinations and destination
432                          * 0.0.0.0. In both cases the packet will be dropped because the
433                          * destination is the loopback device and not the bridge. */
434                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
435                                 goto free_skb;
436
437                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
438                                              RT_TOS(iph->tos), 0);
439                         if (!IS_ERR(rt)) {
440                                 /* - Bridged-and-DNAT'ed traffic doesn't
441                                  *   require ip_forwarding. */
442                                 if (rt->dst.dev == dev) {
443                                         skb_dst_set(skb, &rt->dst);
444                                         goto bridged_dnat;
445                                 }
446                                 ip_rt_put(rt);
447                         }
448 free_skb:
449                         kfree_skb(skb);
450                         return 0;
451                 } else {
452                         if (skb_dst(skb)->dev == dev) {
453 bridged_dnat:
454                                 skb->dev = nf_bridge->physindev;
455                                 nf_bridge_update_protocol(skb);
456                                 nf_bridge_push_encap_header(skb);
457                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
458                                                NF_BR_PRE_ROUTING,
459                                                skb, skb->dev, NULL,
460                                                br_nf_pre_routing_finish_bridge,
461                                                1);
462                                 return 0;
463                         }
464                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
465                         skb->pkt_type = PACKET_HOST;
466                 }
467         } else {
468                 rt = bridge_parent_rtable(nf_bridge->physindev);
469                 if (!rt) {
470                         kfree_skb(skb);
471                         return 0;
472                 }
473                 skb_dst_set_noref(skb, &rt->dst);
474         }
475
476         skb->dev = nf_bridge->physindev;
477         nf_bridge_update_protocol(skb);
478         nf_bridge_push_encap_header(skb);
479         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
480                        br_handle_frame_finish, 1);
481
482         return 0;
483 }
484
485 /* Some common code for IPv4/IPv6 */
486 static struct net_device *setup_pre_routing(struct sk_buff *skb)
487 {
488         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
489
490         if (skb->pkt_type == PACKET_OTHERHOST) {
491                 skb->pkt_type = PACKET_HOST;
492                 nf_bridge->mask |= BRNF_PKT_TYPE;
493         }
494
495         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
496         nf_bridge->physindev = skb->dev;
497         skb->dev = bridge_parent(skb->dev);
498         if (skb->protocol == htons(ETH_P_8021Q))
499                 nf_bridge->mask |= BRNF_8021Q;
500         else if (skb->protocol == htons(ETH_P_PPP_SES))
501                 nf_bridge->mask |= BRNF_PPPoE;
502
503         return skb->dev;
504 }
505
506 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
507 static int check_hbh_len(struct sk_buff *skb)
508 {
509         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
510         u32 pkt_len;
511         const unsigned char *nh = skb_network_header(skb);
512         int off = raw - nh;
513         int len = (raw[1] + 1) << 3;
514
515         if ((raw + len) - skb->data > skb_headlen(skb))
516                 goto bad;
517
518         off += 2;
519         len -= 2;
520
521         while (len > 0) {
522                 int optlen = nh[off + 1] + 2;
523
524                 switch (nh[off]) {
525                 case IPV6_TLV_PAD0:
526                         optlen = 1;
527                         break;
528
529                 case IPV6_TLV_PADN:
530                         break;
531
532                 case IPV6_TLV_JUMBO:
533                         if (nh[off + 1] != 4 || (off & 3) != 2)
534                                 goto bad;
535                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
536                         if (pkt_len <= IPV6_MAXPLEN ||
537                             ipv6_hdr(skb)->payload_len)
538                                 goto bad;
539                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
540                                 goto bad;
541                         if (pskb_trim_rcsum(skb,
542                                             pkt_len + sizeof(struct ipv6hdr)))
543                                 goto bad;
544                         nh = skb_network_header(skb);
545                         break;
546                 default:
547                         if (optlen > len)
548                                 goto bad;
549                         break;
550                 }
551                 off += optlen;
552                 len -= optlen;
553         }
554         if (len == 0)
555                 return 0;
556 bad:
557         return -1;
558
559 }
560
561 /* Replicate the checks that IPv6 does on packet reception and pass the packet
562  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
563 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
564                                            struct sk_buff *skb,
565                                            const struct net_device *in,
566                                            const struct net_device *out,
567                                            int (*okfn)(struct sk_buff *))
568 {
569         const struct ipv6hdr *hdr;
570         u32 pkt_len;
571
572         if (skb->len < sizeof(struct ipv6hdr))
573                 return NF_DROP;
574
575         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
576                 return NF_DROP;
577
578         hdr = ipv6_hdr(skb);
579
580         if (hdr->version != 6)
581                 return NF_DROP;
582
583         pkt_len = ntohs(hdr->payload_len);
584
585         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
586                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
587                         return NF_DROP;
588                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
589                         return NF_DROP;
590         }
591         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
592                 return NF_DROP;
593
594         nf_bridge_put(skb->nf_bridge);
595         if (!nf_bridge_alloc(skb))
596                 return NF_DROP;
597         if (!setup_pre_routing(skb))
598                 return NF_DROP;
599
600         skb->protocol = htons(ETH_P_IPV6);
601         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
602                 br_nf_pre_routing_finish_ipv6);
603
604         return NF_STOLEN;
605 }
606
607 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
608  * Replicate the checks that IPv4 does on packet reception.
609  * Set skb->dev to the bridge device (i.e. parent of the
610  * receiving device) to make netfilter happy, the REDIRECT
611  * target in particular.  Save the original destination IP
612  * address to be able to detect DNAT afterwards. */
613 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
614                                       const struct net_device *in,
615                                       const struct net_device *out,
616                                       int (*okfn)(struct sk_buff *))
617 {
618         struct net_bridge_port *p;
619         struct net_bridge *br;
620         __u32 len = nf_bridge_encap_header_len(skb);
621
622         if (unlikely(!pskb_may_pull(skb, len)))
623                 return NF_DROP;
624
625         p = br_port_get_rcu(in);
626         if (p == NULL)
627                 return NF_DROP;
628         br = p->br;
629
630         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
631             IS_PPPOE_IPV6(skb)) {
632                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
633                         return NF_ACCEPT;
634
635                 nf_bridge_pull_encap_header_rcsum(skb);
636                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
637         }
638
639         if (!brnf_call_iptables && !br->nf_call_iptables)
640                 return NF_ACCEPT;
641
642         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
643             !IS_PPPOE_IP(skb))
644                 return NF_ACCEPT;
645
646         nf_bridge_pull_encap_header_rcsum(skb);
647
648         if (br_parse_ip_options(skb))
649                 return NF_DROP;
650
651         nf_bridge_put(skb->nf_bridge);
652         if (!nf_bridge_alloc(skb))
653                 return NF_DROP;
654         if (!setup_pre_routing(skb))
655                 return NF_DROP;
656         store_orig_dstaddr(skb);
657         skb->protocol = htons(ETH_P_IP);
658
659         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
660                 br_nf_pre_routing_finish);
661
662         return NF_STOLEN;
663 }
664
665
666 /* PF_BRIDGE/LOCAL_IN ************************************************/
667 /* The packet is locally destined, which requires a real
668  * dst_entry, so detach the fake one.  On the way up, the
669  * packet would pass through PRE_ROUTING again (which already
670  * took place when the packet entered the bridge), but we
671  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
672  * prevent this from happening. */
673 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
674                                    const struct net_device *in,
675                                    const struct net_device *out,
676                                    int (*okfn)(struct sk_buff *))
677 {
678         struct rtable *rt = skb_rtable(skb);
679
680         if (rt && rt == bridge_parent_rtable(in))
681                 skb_dst_drop(skb);
682
683         return NF_ACCEPT;
684 }
685
686 /* PF_BRIDGE/FORWARD *************************************************/
687 static int br_nf_forward_finish(struct sk_buff *skb)
688 {
689         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
690         struct net_device *in;
691
692         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
693                 in = nf_bridge->physindev;
694                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
695                         skb->pkt_type = PACKET_OTHERHOST;
696                         nf_bridge->mask ^= BRNF_PKT_TYPE;
697                 }
698                 nf_bridge_update_protocol(skb);
699         } else {
700                 in = *((struct net_device **)(skb->cb));
701         }
702         nf_bridge_push_encap_header(skb);
703
704         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
705                        skb->dev, br_forward_finish, 1);
706         return 0;
707 }
708
709 /* This is the 'purely bridged' case.  For IP, we pass the packet to
710  * netfilter with indev and outdev set to the bridge device,
711  * but we are still able to filter on the 'real' indev/outdev
712  * because of the physdev module. For ARP, indev and outdev are the
713  * bridge ports. */
714 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
715                                      const struct net_device *in,
716                                      const struct net_device *out,
717                                      int (*okfn)(struct sk_buff *))
718 {
719         struct nf_bridge_info *nf_bridge;
720         struct net_device *parent;
721         u_int8_t pf;
722
723         if (!skb->nf_bridge)
724                 return NF_ACCEPT;
725
726         /* Need exclusive nf_bridge_info since we might have multiple
727          * different physoutdevs. */
728         if (!nf_bridge_unshare(skb))
729                 return NF_DROP;
730
731         parent = bridge_parent(out);
732         if (!parent)
733                 return NF_DROP;
734
735         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
736             IS_PPPOE_IP(skb))
737                 pf = PF_INET;
738         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
739                  IS_PPPOE_IPV6(skb))
740                 pf = PF_INET6;
741         else
742                 return NF_ACCEPT;
743
744         nf_bridge_pull_encap_header(skb);
745
746         nf_bridge = skb->nf_bridge;
747         if (skb->pkt_type == PACKET_OTHERHOST) {
748                 skb->pkt_type = PACKET_HOST;
749                 nf_bridge->mask |= BRNF_PKT_TYPE;
750         }
751
752         if (pf == PF_INET && br_parse_ip_options(skb))
753                 return NF_DROP;
754
755         /* The physdev module checks on this */
756         nf_bridge->mask |= BRNF_BRIDGED;
757         nf_bridge->physoutdev = skb->dev;
758         if (pf == PF_INET)
759                 skb->protocol = htons(ETH_P_IP);
760         else
761                 skb->protocol = htons(ETH_P_IPV6);
762
763         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
764                 br_nf_forward_finish);
765
766         return NF_STOLEN;
767 }
768
769 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
770                                       const struct net_device *in,
771                                       const struct net_device *out,
772                                       int (*okfn)(struct sk_buff *))
773 {
774         struct net_bridge_port *p;
775         struct net_bridge *br;
776         struct net_device **d = (struct net_device **)(skb->cb);
777
778         p = br_port_get_rcu(out);
779         if (p == NULL)
780                 return NF_ACCEPT;
781         br = p->br;
782
783         if (!brnf_call_arptables && !br->nf_call_arptables)
784                 return NF_ACCEPT;
785
786         if (skb->protocol != htons(ETH_P_ARP)) {
787                 if (!IS_VLAN_ARP(skb))
788                         return NF_ACCEPT;
789                 nf_bridge_pull_encap_header(skb);
790         }
791
792         if (arp_hdr(skb)->ar_pln != 4) {
793                 if (IS_VLAN_ARP(skb))
794                         nf_bridge_push_encap_header(skb);
795                 return NF_ACCEPT;
796         }
797         *d = (struct net_device *)in;
798         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
799                 (struct net_device *)out, br_nf_forward_finish);
800
801         return NF_STOLEN;
802 }
803
804 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
805 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
806 {
807         int ret;
808
809         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
810             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
811             !skb_is_gso(skb)) {
812                 if (br_parse_ip_options(skb))
813                         /* Drop invalid packet */
814                         return NF_DROP;
815                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
816         } else
817                 ret = br_dev_queue_push_xmit(skb);
818
819         return ret;
820 }
821 #else
822 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
823 {
824         return br_dev_queue_push_xmit(skb);
825 }
826 #endif
827
828 /* PF_BRIDGE/POST_ROUTING ********************************************/
829 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
830                                        const struct net_device *in,
831                                        const struct net_device *out,
832                                        int (*okfn)(struct sk_buff *))
833 {
834         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
835         struct net_device *realoutdev = bridge_parent(skb->dev);
836         u_int8_t pf;
837
838         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
839                 return NF_ACCEPT;
840
841         if (!realoutdev)
842                 return NF_DROP;
843
844         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
845             IS_PPPOE_IP(skb))
846                 pf = PF_INET;
847         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
848                  IS_PPPOE_IPV6(skb))
849                 pf = PF_INET6;
850         else
851                 return NF_ACCEPT;
852
853         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
854          * about the value of skb->pkt_type. */
855         if (skb->pkt_type == PACKET_OTHERHOST) {
856                 skb->pkt_type = PACKET_HOST;
857                 nf_bridge->mask |= BRNF_PKT_TYPE;
858         }
859
860         nf_bridge_pull_encap_header(skb);
861         nf_bridge_save_header(skb);
862         if (pf == PF_INET)
863                 skb->protocol = htons(ETH_P_IP);
864         else
865                 skb->protocol = htons(ETH_P_IPV6);
866
867         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
868                 br_nf_dev_queue_xmit);
869
870         return NF_STOLEN;
871 }
872
873 /* IP/SABOTAGE *****************************************************/
874 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
875  * for the second time. */
876 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
877                                    const struct net_device *in,
878                                    const struct net_device *out,
879                                    int (*okfn)(struct sk_buff *))
880 {
881         if (skb->nf_bridge &&
882             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
883                 return NF_STOP;
884         }
885
886         return NF_ACCEPT;
887 }
888
889 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
890  * br_dev_queue_push_xmit is called afterwards */
891 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
892         {
893                 .hook = br_nf_pre_routing,
894                 .owner = THIS_MODULE,
895                 .pf = PF_BRIDGE,
896                 .hooknum = NF_BR_PRE_ROUTING,
897                 .priority = NF_BR_PRI_BRNF,
898         },
899         {
900                 .hook = br_nf_local_in,
901                 .owner = THIS_MODULE,
902                 .pf = PF_BRIDGE,
903                 .hooknum = NF_BR_LOCAL_IN,
904                 .priority = NF_BR_PRI_BRNF,
905         },
906         {
907                 .hook = br_nf_forward_ip,
908                 .owner = THIS_MODULE,
909                 .pf = PF_BRIDGE,
910                 .hooknum = NF_BR_FORWARD,
911                 .priority = NF_BR_PRI_BRNF - 1,
912         },
913         {
914                 .hook = br_nf_forward_arp,
915                 .owner = THIS_MODULE,
916                 .pf = PF_BRIDGE,
917                 .hooknum = NF_BR_FORWARD,
918                 .priority = NF_BR_PRI_BRNF,
919         },
920         {
921                 .hook = br_nf_post_routing,
922                 .owner = THIS_MODULE,
923                 .pf = PF_BRIDGE,
924                 .hooknum = NF_BR_POST_ROUTING,
925                 .priority = NF_BR_PRI_LAST,
926         },
927         {
928                 .hook = ip_sabotage_in,
929                 .owner = THIS_MODULE,
930                 .pf = PF_INET,
931                 .hooknum = NF_INET_PRE_ROUTING,
932                 .priority = NF_IP_PRI_FIRST,
933         },
934         {
935                 .hook = ip_sabotage_in,
936                 .owner = THIS_MODULE,
937                 .pf = PF_INET6,
938                 .hooknum = NF_INET_PRE_ROUTING,
939                 .priority = NF_IP6_PRI_FIRST,
940         },
941 };
942
943 #ifdef CONFIG_SYSCTL
944 static
945 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
946                             void __user * buffer, size_t * lenp, loff_t * ppos)
947 {
948         int ret;
949
950         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
951
952         if (write && *(int *)(ctl->data))
953                 *(int *)(ctl->data) = 1;
954         return ret;
955 }
956
957 static ctl_table brnf_table[] = {
958         {
959                 .procname       = "bridge-nf-call-arptables",
960                 .data           = &brnf_call_arptables,
961                 .maxlen         = sizeof(int),
962                 .mode           = 0644,
963                 .proc_handler   = brnf_sysctl_call_tables,
964         },
965         {
966                 .procname       = "bridge-nf-call-iptables",
967                 .data           = &brnf_call_iptables,
968                 .maxlen         = sizeof(int),
969                 .mode           = 0644,
970                 .proc_handler   = brnf_sysctl_call_tables,
971         },
972         {
973                 .procname       = "bridge-nf-call-ip6tables",
974                 .data           = &brnf_call_ip6tables,
975                 .maxlen         = sizeof(int),
976                 .mode           = 0644,
977                 .proc_handler   = brnf_sysctl_call_tables,
978         },
979         {
980                 .procname       = "bridge-nf-filter-vlan-tagged",
981                 .data           = &brnf_filter_vlan_tagged,
982                 .maxlen         = sizeof(int),
983                 .mode           = 0644,
984                 .proc_handler   = brnf_sysctl_call_tables,
985         },
986         {
987                 .procname       = "bridge-nf-filter-pppoe-tagged",
988                 .data           = &brnf_filter_pppoe_tagged,
989                 .maxlen         = sizeof(int),
990                 .mode           = 0644,
991                 .proc_handler   = brnf_sysctl_call_tables,
992         },
993         { }
994 };
995
996 static struct ctl_path brnf_path[] = {
997         { .procname = "net", },
998         { .procname = "bridge", },
999         { }
1000 };
1001 #endif
1002
1003 int __init br_netfilter_init(void)
1004 {
1005         int ret;
1006
1007         ret = dst_entries_init(&fake_dst_ops);
1008         if (ret < 0)
1009                 return ret;
1010
1011         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1012         if (ret < 0) {
1013                 dst_entries_destroy(&fake_dst_ops);
1014                 return ret;
1015         }
1016 #ifdef CONFIG_SYSCTL
1017         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1018         if (brnf_sysctl_header == NULL) {
1019                 printk(KERN_WARNING
1020                        "br_netfilter: can't register to sysctl.\n");
1021                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1022                 dst_entries_destroy(&fake_dst_ops);
1023                 return -ENOMEM;
1024         }
1025 #endif
1026         printk(KERN_NOTICE "Bridge firewalling registered\n");
1027         return 0;
1028 }
1029
1030 void br_netfilter_fini(void)
1031 {
1032         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1033 #ifdef CONFIG_SYSCTL
1034         unregister_sysctl_table(brnf_sysctl_header);
1035 #endif
1036         dst_entries_destroy(&fake_dst_ops);
1037 }