bridge: Use consistent NF_DROP returns in nf_pre_routing
[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                <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 struct dst_ops fake_dst_ops = {
108         .family =               AF_INET,
109         .protocol =             cpu_to_be16(ETH_P_IP),
110         .update_pmtu =          fake_update_pmtu,
111 };
112
113 /*
114  * Initialize bogus route table used to keep netfilter happy.
115  * Currently, we fill in the PMTU entry because netfilter
116  * refragmentation needs it, and the rt_flags entry because
117  * ipt_REJECT needs it.  Future netfilter modules might
118  * require us to fill additional fields.
119  */
120 void br_netfilter_rtable_init(struct net_bridge *br)
121 {
122         struct rtable *rt = &br->fake_rtable;
123
124         atomic_set(&rt->dst.__refcnt, 1);
125         rt->dst.dev = br->dev;
126         rt->dst.path = &rt->dst;
127         dst_metric_set(&rt->dst, RTAX_MTU, 1500);
128         rt->dst.flags   = DST_NOXFRM;
129         rt->dst.ops = &fake_dst_ops;
130 }
131
132 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
133 {
134         struct net_bridge_port *port;
135
136         port = br_port_get_rcu(dev);
137         return port ? &port->br->fake_rtable : NULL;
138 }
139
140 static inline struct net_device *bridge_parent(const struct net_device *dev)
141 {
142         struct net_bridge_port *port;
143
144         port = br_port_get_rcu(dev);
145         return port ? port->br->dev : NULL;
146 }
147
148 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
149 {
150         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
151         if (likely(skb->nf_bridge))
152                 atomic_set(&(skb->nf_bridge->use), 1);
153
154         return skb->nf_bridge;
155 }
156
157 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
158 {
159         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
160
161         if (atomic_read(&nf_bridge->use) > 1) {
162                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
163
164                 if (tmp) {
165                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
166                         atomic_set(&tmp->use, 1);
167                 }
168                 nf_bridge_put(nf_bridge);
169                 nf_bridge = tmp;
170         }
171         return nf_bridge;
172 }
173
174 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
175 {
176         unsigned int len = nf_bridge_encap_header_len(skb);
177
178         skb_push(skb, len);
179         skb->network_header -= len;
180 }
181
182 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
183 {
184         unsigned int len = nf_bridge_encap_header_len(skb);
185
186         skb_pull(skb, len);
187         skb->network_header += len;
188 }
189
190 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
191 {
192         unsigned int len = nf_bridge_encap_header_len(skb);
193
194         skb_pull_rcsum(skb, len);
195         skb->network_header += len;
196 }
197
198 static inline void nf_bridge_save_header(struct sk_buff *skb)
199 {
200         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
201
202         skb_copy_from_linear_data_offset(skb, -header_size,
203                                          skb->nf_bridge->data, header_size);
204 }
205
206 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
207 {
208         if (skb->nf_bridge->mask & BRNF_8021Q)
209                 skb->protocol = htons(ETH_P_8021Q);
210         else if (skb->nf_bridge->mask & BRNF_PPPoE)
211                 skb->protocol = htons(ETH_P_PPP_SES);
212 }
213
214 /* When handing a packet over to the IP layer
215  * check whether we have a skb that is in the
216  * expected format
217  */
218
219 static int br_parse_ip_options(struct sk_buff *skb)
220 {
221         struct ip_options *opt;
222         struct iphdr *iph;
223         struct net_device *dev = skb->dev;
224         u32 len;
225
226         iph = ip_hdr(skb);
227         opt = &(IPCB(skb)->opt);
228
229         /* Basic sanity checks */
230         if (iph->ihl < 5 || iph->version != 4)
231                 goto inhdr_error;
232
233         if (!pskb_may_pull(skb, iph->ihl*4))
234                 goto inhdr_error;
235
236         iph = ip_hdr(skb);
237         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
238                 goto inhdr_error;
239
240         len = ntohs(iph->tot_len);
241         if (skb->len < len) {
242                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
243                 goto drop;
244         } else if (len < (iph->ihl*4))
245                 goto inhdr_error;
246
247         if (pskb_trim_rcsum(skb, len)) {
248                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
249                 goto drop;
250         }
251
252         /* Zero out the CB buffer if no options present */
253         if (iph->ihl == 5) {
254                 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
255                 return 0;
256         }
257
258         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
259         if (ip_options_compile(dev_net(dev), opt, skb))
260                 goto inhdr_error;
261
262         /* Check correct handling of SRR option */
263         if (unlikely(opt->srr)) {
264                 struct in_device *in_dev = __in_dev_get_rcu(dev);
265                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
266                         goto drop;
267
268                 if (ip_options_rcv_srr(skb))
269                         goto drop;
270         }
271
272         return 0;
273
274 inhdr_error:
275         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
276 drop:
277         return -1;
278 }
279
280 /* Fill in the header for fragmented IP packets handled by
281  * the IPv4 connection tracking code.
282  */
283 int nf_bridge_copy_header(struct sk_buff *skb)
284 {
285         int err;
286         unsigned int header_size;
287
288         nf_bridge_update_protocol(skb);
289         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
290         err = skb_cow_head(skb, header_size);
291         if (err)
292                 return err;
293
294         skb_copy_to_linear_data_offset(skb, -header_size,
295                                        skb->nf_bridge->data, header_size);
296         __skb_push(skb, nf_bridge_encap_header_len(skb));
297         return 0;
298 }
299
300 /* PF_BRIDGE/PRE_ROUTING *********************************************/
301 /* Undo the changes made for ip6tables PREROUTING and continue the
302  * bridge PRE_ROUTING hook. */
303 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
304 {
305         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
306         struct rtable *rt;
307
308         if (nf_bridge->mask & BRNF_PKT_TYPE) {
309                 skb->pkt_type = PACKET_OTHERHOST;
310                 nf_bridge->mask ^= BRNF_PKT_TYPE;
311         }
312         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
313
314         rt = bridge_parent_rtable(nf_bridge->physindev);
315         if (!rt) {
316                 kfree_skb(skb);
317                 return 0;
318         }
319         skb_dst_set_noref(skb, &rt->dst);
320
321         skb->dev = nf_bridge->physindev;
322         nf_bridge_update_protocol(skb);
323         nf_bridge_push_encap_header(skb);
324         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
325                        br_handle_frame_finish, 1);
326
327         return 0;
328 }
329
330 /* Obtain the correct destination MAC address, while preserving the original
331  * source MAC address. If we already know this address, we just copy it. If we
332  * don't, we use the neighbour framework to find out. In both cases, we make
333  * sure that br_handle_frame_finish() is called afterwards.
334  */
335 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
336 {
337         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
338         struct dst_entry *dst;
339
340         skb->dev = bridge_parent(skb->dev);
341         if (!skb->dev)
342                 goto free_skb;
343         dst = skb_dst(skb);
344         if (dst->hh) {
345                 neigh_hh_bridge(dst->hh, skb);
346                 skb->dev = nf_bridge->physindev;
347                 return br_handle_frame_finish(skb);
348         } else if (dst->neighbour) {
349                 /* the neighbour function below overwrites the complete
350                  * MAC header, so we save the Ethernet source address and
351                  * protocol number. */
352                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
353                 /* tell br_dev_xmit to continue with forwarding */
354                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
355                 return dst->neighbour->output(skb);
356         }
357 free_skb:
358         kfree_skb(skb);
359         return 0;
360 }
361
362 /* This requires some explaining. If DNAT has taken place,
363  * we will need to fix up the destination Ethernet address.
364  *
365  * There are two cases to consider:
366  * 1. The packet was DNAT'ed to a device in the same bridge
367  *    port group as it was received on. We can still bridge
368  *    the packet.
369  * 2. The packet was DNAT'ed to a different device, either
370  *    a non-bridged device or another bridge port group.
371  *    The packet will need to be routed.
372  *
373  * The correct way of distinguishing between these two cases is to
374  * call ip_route_input() and to look at skb->dst->dev, which is
375  * changed to the destination device if ip_route_input() succeeds.
376  *
377  * Let's first consider the case that ip_route_input() succeeds:
378  *
379  * If the output device equals the logical bridge device the packet
380  * came in on, we can consider this bridging. The corresponding MAC
381  * address will be obtained in br_nf_pre_routing_finish_bridge.
382  * Otherwise, the packet is considered to be routed and we just
383  * change the destination MAC address so that the packet will
384  * later be passed up to the IP stack to be routed. For a redirected
385  * packet, ip_route_input() will give back the localhost as output device,
386  * which differs from the bridge device.
387  *
388  * Let's now consider the case that ip_route_input() fails:
389  *
390  * This can be because the destination address is martian, in which case
391  * the packet will be dropped.
392  * If IP forwarding is disabled, ip_route_input() will fail, while
393  * ip_route_output_key() can return success. The source
394  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
395  * thinks we're handling a locally generated packet and won't care
396  * if IP forwarding is enabled. If the output device equals the logical bridge
397  * device, we proceed as if ip_route_input() succeeded. If it differs from the
398  * logical bridge port or if ip_route_output_key() fails we drop the packet.
399  */
400 static int br_nf_pre_routing_finish(struct sk_buff *skb)
401 {
402         struct net_device *dev = skb->dev;
403         struct iphdr *iph = ip_hdr(skb);
404         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
405         struct rtable *rt;
406         int err;
407
408         if (nf_bridge->mask & BRNF_PKT_TYPE) {
409                 skb->pkt_type = PACKET_OTHERHOST;
410                 nf_bridge->mask ^= BRNF_PKT_TYPE;
411         }
412         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
413         if (dnat_took_place(skb)) {
414                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
415                         struct flowi fl = {
416                                 .fl4_dst = iph->daddr,
417                                 .fl4_tos = RT_TOS(iph->tos),
418                         };
419                         struct in_device *in_dev = __in_dev_get_rcu(dev);
420
421                         /* If err equals -EHOSTUNREACH the error is due to a
422                          * martian destination or due to the fact that
423                          * forwarding is disabled. For most martian packets,
424                          * ip_route_output_key() will fail. It won't fail for 2 types of
425                          * martian destinations: loopback destinations and destination
426                          * 0.0.0.0. In both cases the packet will be dropped because the
427                          * destination is the loopback device and not the bridge. */
428                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
429                                 goto free_skb;
430
431                         if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
432                                 /* - Bridged-and-DNAT'ed traffic doesn't
433                                  *   require ip_forwarding. */
434                                 if (((struct dst_entry *)rt)->dev == dev) {
435                                         skb_dst_set(skb, (struct dst_entry *)rt);
436                                         goto bridged_dnat;
437                                 }
438                                 dst_release((struct dst_entry *)rt);
439                         }
440 free_skb:
441                         kfree_skb(skb);
442                         return 0;
443                 } else {
444                         if (skb_dst(skb)->dev == dev) {
445 bridged_dnat:
446                                 skb->dev = nf_bridge->physindev;
447                                 nf_bridge_update_protocol(skb);
448                                 nf_bridge_push_encap_header(skb);
449                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
450                                                NF_BR_PRE_ROUTING,
451                                                skb, skb->dev, NULL,
452                                                br_nf_pre_routing_finish_bridge,
453                                                1);
454                                 return 0;
455                         }
456                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
457                         skb->pkt_type = PACKET_HOST;
458                 }
459         } else {
460                 rt = bridge_parent_rtable(nf_bridge->physindev);
461                 if (!rt) {
462                         kfree_skb(skb);
463                         return 0;
464                 }
465                 skb_dst_set_noref(skb, &rt->dst);
466         }
467
468         skb->dev = nf_bridge->physindev;
469         nf_bridge_update_protocol(skb);
470         nf_bridge_push_encap_header(skb);
471         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
472                        br_handle_frame_finish, 1);
473
474         return 0;
475 }
476
477 /* Some common code for IPv4/IPv6 */
478 static struct net_device *setup_pre_routing(struct sk_buff *skb)
479 {
480         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
481
482         if (skb->pkt_type == PACKET_OTHERHOST) {
483                 skb->pkt_type = PACKET_HOST;
484                 nf_bridge->mask |= BRNF_PKT_TYPE;
485         }
486
487         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
488         nf_bridge->physindev = skb->dev;
489         skb->dev = bridge_parent(skb->dev);
490         if (skb->protocol == htons(ETH_P_8021Q))
491                 nf_bridge->mask |= BRNF_8021Q;
492         else if (skb->protocol == htons(ETH_P_PPP_SES))
493                 nf_bridge->mask |= BRNF_PPPoE;
494
495         return skb->dev;
496 }
497
498 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
499 static int check_hbh_len(struct sk_buff *skb)
500 {
501         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
502         u32 pkt_len;
503         const unsigned char *nh = skb_network_header(skb);
504         int off = raw - nh;
505         int len = (raw[1] + 1) << 3;
506
507         if ((raw + len) - skb->data > skb_headlen(skb))
508                 goto bad;
509
510         off += 2;
511         len -= 2;
512
513         while (len > 0) {
514                 int optlen = nh[off + 1] + 2;
515
516                 switch (nh[off]) {
517                 case IPV6_TLV_PAD0:
518                         optlen = 1;
519                         break;
520
521                 case IPV6_TLV_PADN:
522                         break;
523
524                 case IPV6_TLV_JUMBO:
525                         if (nh[off + 1] != 4 || (off & 3) != 2)
526                                 goto bad;
527                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
528                         if (pkt_len <= IPV6_MAXPLEN ||
529                             ipv6_hdr(skb)->payload_len)
530                                 goto bad;
531                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
532                                 goto bad;
533                         if (pskb_trim_rcsum(skb,
534                                             pkt_len + sizeof(struct ipv6hdr)))
535                                 goto bad;
536                         nh = skb_network_header(skb);
537                         break;
538                 default:
539                         if (optlen > len)
540                                 goto bad;
541                         break;
542                 }
543                 off += optlen;
544                 len -= optlen;
545         }
546         if (len == 0)
547                 return 0;
548 bad:
549         return -1;
550
551 }
552
553 /* Replicate the checks that IPv6 does on packet reception and pass the packet
554  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
555 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
556                                            struct sk_buff *skb,
557                                            const struct net_device *in,
558                                            const struct net_device *out,
559                                            int (*okfn)(struct sk_buff *))
560 {
561         struct ipv6hdr *hdr;
562         u32 pkt_len;
563
564         if (skb->len < sizeof(struct ipv6hdr))
565                 return NF_DROP;
566
567         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
568                 return NF_DROP;
569
570         hdr = ipv6_hdr(skb);
571
572         if (hdr->version != 6)
573                 return NF_DROP;
574
575         pkt_len = ntohs(hdr->payload_len);
576
577         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
578                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
579                         return NF_DROP;
580                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
581                         return NF_DROP;
582         }
583         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
584                 return NF_DROP;
585
586         nf_bridge_put(skb->nf_bridge);
587         if (!nf_bridge_alloc(skb))
588                 return NF_DROP;
589         if (!setup_pre_routing(skb))
590                 return NF_DROP;
591
592         skb->protocol = htons(ETH_P_IPV6);
593         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
594                 br_nf_pre_routing_finish_ipv6);
595
596         return NF_STOLEN;
597 }
598
599 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
600  * Replicate the checks that IPv4 does on packet reception.
601  * Set skb->dev to the bridge device (i.e. parent of the
602  * receiving device) to make netfilter happy, the REDIRECT
603  * target in particular.  Save the original destination IP
604  * address to be able to detect DNAT afterwards. */
605 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
606                                       const struct net_device *in,
607                                       const struct net_device *out,
608                                       int (*okfn)(struct sk_buff *))
609 {
610         struct net_bridge_port *p;
611         struct net_bridge *br;
612         __u32 len = nf_bridge_encap_header_len(skb);
613
614         if (unlikely(!pskb_may_pull(skb, len)))
615                 return NF_DROP;
616
617         p = br_port_get_rcu(in);
618         if (p == NULL)
619                 return NF_DROP;
620         br = p->br;
621
622         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
623             IS_PPPOE_IPV6(skb)) {
624                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
625                         return NF_ACCEPT;
626
627                 nf_bridge_pull_encap_header_rcsum(skb);
628                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
629         }
630
631         if (!brnf_call_iptables && !br->nf_call_iptables)
632                 return NF_ACCEPT;
633
634         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
635             !IS_PPPOE_IP(skb))
636                 return NF_ACCEPT;
637
638         nf_bridge_pull_encap_header_rcsum(skb);
639
640         if (br_parse_ip_options(skb))
641                 return NF_DROP;
642
643         nf_bridge_put(skb->nf_bridge);
644         if (!nf_bridge_alloc(skb))
645                 return NF_DROP;
646         if (!setup_pre_routing(skb))
647                 return NF_DROP;
648         store_orig_dstaddr(skb);
649         skb->protocol = htons(ETH_P_IP);
650
651         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
652                 br_nf_pre_routing_finish);
653
654         return NF_STOLEN;
655 }
656
657
658 /* PF_BRIDGE/LOCAL_IN ************************************************/
659 /* The packet is locally destined, which requires a real
660  * dst_entry, so detach the fake one.  On the way up, the
661  * packet would pass through PRE_ROUTING again (which already
662  * took place when the packet entered the bridge), but we
663  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
664  * prevent this from happening. */
665 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
666                                    const struct net_device *in,
667                                    const struct net_device *out,
668                                    int (*okfn)(struct sk_buff *))
669 {
670         struct rtable *rt = skb_rtable(skb);
671
672         if (rt && rt == bridge_parent_rtable(in))
673                 skb_dst_drop(skb);
674
675         return NF_ACCEPT;
676 }
677
678 /* PF_BRIDGE/FORWARD *************************************************/
679 static int br_nf_forward_finish(struct sk_buff *skb)
680 {
681         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
682         struct net_device *in;
683
684         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
685                 in = nf_bridge->physindev;
686                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
687                         skb->pkt_type = PACKET_OTHERHOST;
688                         nf_bridge->mask ^= BRNF_PKT_TYPE;
689                 }
690                 nf_bridge_update_protocol(skb);
691         } else {
692                 in = *((struct net_device **)(skb->cb));
693         }
694         nf_bridge_push_encap_header(skb);
695
696         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
697                        skb->dev, br_forward_finish, 1);
698         return 0;
699 }
700
701 /* This is the 'purely bridged' case.  For IP, we pass the packet to
702  * netfilter with indev and outdev set to the bridge device,
703  * but we are still able to filter on the 'real' indev/outdev
704  * because of the physdev module. For ARP, indev and outdev are the
705  * bridge ports. */
706 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
707                                      const struct net_device *in,
708                                      const struct net_device *out,
709                                      int (*okfn)(struct sk_buff *))
710 {
711         struct nf_bridge_info *nf_bridge;
712         struct net_device *parent;
713         u_int8_t pf;
714
715         if (!skb->nf_bridge)
716                 return NF_ACCEPT;
717
718         /* Need exclusive nf_bridge_info since we might have multiple
719          * different physoutdevs. */
720         if (!nf_bridge_unshare(skb))
721                 return NF_DROP;
722
723         parent = bridge_parent(out);
724         if (!parent)
725                 return NF_DROP;
726
727         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
728             IS_PPPOE_IP(skb))
729                 pf = PF_INET;
730         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
731                  IS_PPPOE_IPV6(skb))
732                 pf = PF_INET6;
733         else
734                 return NF_ACCEPT;
735
736         nf_bridge_pull_encap_header(skb);
737
738         nf_bridge = skb->nf_bridge;
739         if (skb->pkt_type == PACKET_OTHERHOST) {
740                 skb->pkt_type = PACKET_HOST;
741                 nf_bridge->mask |= BRNF_PKT_TYPE;
742         }
743
744         /* The physdev module checks on this */
745         nf_bridge->mask |= BRNF_BRIDGED;
746         nf_bridge->physoutdev = skb->dev;
747         if (pf == PF_INET)
748                 skb->protocol = htons(ETH_P_IP);
749         else
750                 skb->protocol = htons(ETH_P_IPV6);
751
752         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
753                 br_nf_forward_finish);
754
755         return NF_STOLEN;
756 }
757
758 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
759                                       const struct net_device *in,
760                                       const struct net_device *out,
761                                       int (*okfn)(struct sk_buff *))
762 {
763         struct net_bridge_port *p;
764         struct net_bridge *br;
765         struct net_device **d = (struct net_device **)(skb->cb);
766
767         p = br_port_get_rcu(out);
768         if (p == NULL)
769                 return NF_ACCEPT;
770         br = p->br;
771
772         if (!brnf_call_arptables && !br->nf_call_arptables)
773                 return NF_ACCEPT;
774
775         if (skb->protocol != htons(ETH_P_ARP)) {
776                 if (!IS_VLAN_ARP(skb))
777                         return NF_ACCEPT;
778                 nf_bridge_pull_encap_header(skb);
779         }
780
781         if (arp_hdr(skb)->ar_pln != 4) {
782                 if (IS_VLAN_ARP(skb))
783                         nf_bridge_push_encap_header(skb);
784                 return NF_ACCEPT;
785         }
786         *d = (struct net_device *)in;
787         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
788                 (struct net_device *)out, br_nf_forward_finish);
789
790         return NF_STOLEN;
791 }
792
793 #if defined(CONFIG_NF_CONNTRACK_IPV4) || defined(CONFIG_NF_CONNTRACK_IPV4_MODULE)
794 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
795 {
796         int ret;
797
798         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
799             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
800             !skb_is_gso(skb)) {
801                 if (br_parse_ip_options(skb))
802                         /* Drop invalid packet */
803                         return NF_DROP;
804                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
805         } else
806                 ret = br_dev_queue_push_xmit(skb);
807
808         return ret;
809 }
810 #else
811 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
812 {
813         return br_dev_queue_push_xmit(skb);
814 }
815 #endif
816
817 /* PF_BRIDGE/POST_ROUTING ********************************************/
818 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
819                                        const struct net_device *in,
820                                        const struct net_device *out,
821                                        int (*okfn)(struct sk_buff *))
822 {
823         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
824         struct net_device *realoutdev = bridge_parent(skb->dev);
825         u_int8_t pf;
826
827         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
828                 return NF_ACCEPT;
829
830         if (!realoutdev)
831                 return NF_DROP;
832
833         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
834             IS_PPPOE_IP(skb))
835                 pf = PF_INET;
836         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
837                  IS_PPPOE_IPV6(skb))
838                 pf = PF_INET6;
839         else
840                 return NF_ACCEPT;
841
842         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
843          * about the value of skb->pkt_type. */
844         if (skb->pkt_type == PACKET_OTHERHOST) {
845                 skb->pkt_type = PACKET_HOST;
846                 nf_bridge->mask |= BRNF_PKT_TYPE;
847         }
848
849         nf_bridge_pull_encap_header(skb);
850         nf_bridge_save_header(skb);
851         if (pf == PF_INET)
852                 skb->protocol = htons(ETH_P_IP);
853         else
854                 skb->protocol = htons(ETH_P_IPV6);
855
856         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
857                 br_nf_dev_queue_xmit);
858
859         return NF_STOLEN;
860 }
861
862 /* IP/SABOTAGE *****************************************************/
863 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
864  * for the second time. */
865 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
866                                    const struct net_device *in,
867                                    const struct net_device *out,
868                                    int (*okfn)(struct sk_buff *))
869 {
870         if (skb->nf_bridge &&
871             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
872                 return NF_STOP;
873         }
874
875         return NF_ACCEPT;
876 }
877
878 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
879  * br_dev_queue_push_xmit is called afterwards */
880 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
881         {
882                 .hook = br_nf_pre_routing,
883                 .owner = THIS_MODULE,
884                 .pf = PF_BRIDGE,
885                 .hooknum = NF_BR_PRE_ROUTING,
886                 .priority = NF_BR_PRI_BRNF,
887         },
888         {
889                 .hook = br_nf_local_in,
890                 .owner = THIS_MODULE,
891                 .pf = PF_BRIDGE,
892                 .hooknum = NF_BR_LOCAL_IN,
893                 .priority = NF_BR_PRI_BRNF,
894         },
895         {
896                 .hook = br_nf_forward_ip,
897                 .owner = THIS_MODULE,
898                 .pf = PF_BRIDGE,
899                 .hooknum = NF_BR_FORWARD,
900                 .priority = NF_BR_PRI_BRNF - 1,
901         },
902         {
903                 .hook = br_nf_forward_arp,
904                 .owner = THIS_MODULE,
905                 .pf = PF_BRIDGE,
906                 .hooknum = NF_BR_FORWARD,
907                 .priority = NF_BR_PRI_BRNF,
908         },
909         {
910                 .hook = br_nf_post_routing,
911                 .owner = THIS_MODULE,
912                 .pf = PF_BRIDGE,
913                 .hooknum = NF_BR_POST_ROUTING,
914                 .priority = NF_BR_PRI_LAST,
915         },
916         {
917                 .hook = ip_sabotage_in,
918                 .owner = THIS_MODULE,
919                 .pf = PF_INET,
920                 .hooknum = NF_INET_PRE_ROUTING,
921                 .priority = NF_IP_PRI_FIRST,
922         },
923         {
924                 .hook = ip_sabotage_in,
925                 .owner = THIS_MODULE,
926                 .pf = PF_INET6,
927                 .hooknum = NF_INET_PRE_ROUTING,
928                 .priority = NF_IP6_PRI_FIRST,
929         },
930 };
931
932 #ifdef CONFIG_SYSCTL
933 static
934 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
935                             void __user * buffer, size_t * lenp, loff_t * ppos)
936 {
937         int ret;
938
939         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
940
941         if (write && *(int *)(ctl->data))
942                 *(int *)(ctl->data) = 1;
943         return ret;
944 }
945
946 static ctl_table brnf_table[] = {
947         {
948                 .procname       = "bridge-nf-call-arptables",
949                 .data           = &brnf_call_arptables,
950                 .maxlen         = sizeof(int),
951                 .mode           = 0644,
952                 .proc_handler   = brnf_sysctl_call_tables,
953         },
954         {
955                 .procname       = "bridge-nf-call-iptables",
956                 .data           = &brnf_call_iptables,
957                 .maxlen         = sizeof(int),
958                 .mode           = 0644,
959                 .proc_handler   = brnf_sysctl_call_tables,
960         },
961         {
962                 .procname       = "bridge-nf-call-ip6tables",
963                 .data           = &brnf_call_ip6tables,
964                 .maxlen         = sizeof(int),
965                 .mode           = 0644,
966                 .proc_handler   = brnf_sysctl_call_tables,
967         },
968         {
969                 .procname       = "bridge-nf-filter-vlan-tagged",
970                 .data           = &brnf_filter_vlan_tagged,
971                 .maxlen         = sizeof(int),
972                 .mode           = 0644,
973                 .proc_handler   = brnf_sysctl_call_tables,
974         },
975         {
976                 .procname       = "bridge-nf-filter-pppoe-tagged",
977                 .data           = &brnf_filter_pppoe_tagged,
978                 .maxlen         = sizeof(int),
979                 .mode           = 0644,
980                 .proc_handler   = brnf_sysctl_call_tables,
981         },
982         { }
983 };
984
985 static struct ctl_path brnf_path[] = {
986         { .procname = "net", },
987         { .procname = "bridge", },
988         { }
989 };
990 #endif
991
992 int __init br_netfilter_init(void)
993 {
994         int ret;
995
996         ret = dst_entries_init(&fake_dst_ops);
997         if (ret < 0)
998                 return ret;
999
1000         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1001         if (ret < 0) {
1002                 dst_entries_destroy(&fake_dst_ops);
1003                 return ret;
1004         }
1005 #ifdef CONFIG_SYSCTL
1006         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1007         if (brnf_sysctl_header == NULL) {
1008                 printk(KERN_WARNING
1009                        "br_netfilter: can't register to sysctl.\n");
1010                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1011                 dst_entries_destroy(&fake_dst_ops);
1012                 return -ENOMEM;
1013         }
1014 #endif
1015         printk(KERN_NOTICE "Bridge firewalling registered\n");
1016         return 0;
1017 }
1018
1019 void br_netfilter_fini(void)
1020 {
1021         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1022 #ifdef CONFIG_SYSCTL
1023         unregister_sysctl_table(brnf_sysctl_header);
1024 #endif
1025         dst_entries_destroy(&fake_dst_ops);
1026 }