Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[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 u32 *fake_cow_metrics(struct dst_entry *dst, unsigned long old)
108 {
109         return NULL;
110 }
111
112 static struct neighbour *fake_neigh_lookup(const struct dst_entry *dst, const void *daddr)
113 {
114         return NULL;
115 }
116
117 static unsigned int fake_mtu(const struct dst_entry *dst)
118 {
119         return dst->dev->mtu;
120 }
121
122 static struct dst_ops fake_dst_ops = {
123         .family =               AF_INET,
124         .protocol =             cpu_to_be16(ETH_P_IP),
125         .update_pmtu =          fake_update_pmtu,
126         .cow_metrics =          fake_cow_metrics,
127         .neigh_lookup =         fake_neigh_lookup,
128         .mtu =                  fake_mtu,
129 };
130
131 /*
132  * Initialize bogus route table used to keep netfilter happy.
133  * Currently, we fill in the PMTU entry because netfilter
134  * refragmentation needs it, and the rt_flags entry because
135  * ipt_REJECT needs it.  Future netfilter modules might
136  * require us to fill additional fields.
137  */
138 static const u32 br_dst_default_metrics[RTAX_MAX] = {
139         [RTAX_MTU - 1] = 1500,
140 };
141
142 void br_netfilter_rtable_init(struct net_bridge *br)
143 {
144         struct rtable *rt = &br->fake_rtable;
145
146         atomic_set(&rt->dst.__refcnt, 1);
147         rt->dst.dev = br->dev;
148         rt->dst.path = &rt->dst;
149         dst_init_metrics(&rt->dst, br_dst_default_metrics, true);
150         rt->dst.flags   = DST_NOXFRM | DST_NOPEER;
151         rt->dst.ops = &fake_dst_ops;
152 }
153
154 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
155 {
156         struct net_bridge_port *port;
157
158         port = br_port_get_rcu(dev);
159         return port ? &port->br->fake_rtable : NULL;
160 }
161
162 static inline struct net_device *bridge_parent(const struct net_device *dev)
163 {
164         struct net_bridge_port *port;
165
166         port = br_port_get_rcu(dev);
167         return port ? port->br->dev : NULL;
168 }
169
170 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
171 {
172         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
173         if (likely(skb->nf_bridge))
174                 atomic_set(&(skb->nf_bridge->use), 1);
175
176         return skb->nf_bridge;
177 }
178
179 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
180 {
181         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
182
183         if (atomic_read(&nf_bridge->use) > 1) {
184                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
185
186                 if (tmp) {
187                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
188                         atomic_set(&tmp->use, 1);
189                 }
190                 nf_bridge_put(nf_bridge);
191                 nf_bridge = tmp;
192         }
193         return nf_bridge;
194 }
195
196 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
197 {
198         unsigned int len = nf_bridge_encap_header_len(skb);
199
200         skb_push(skb, len);
201         skb->network_header -= len;
202 }
203
204 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
205 {
206         unsigned int len = nf_bridge_encap_header_len(skb);
207
208         skb_pull(skb, len);
209         skb->network_header += len;
210 }
211
212 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
213 {
214         unsigned int len = nf_bridge_encap_header_len(skb);
215
216         skb_pull_rcsum(skb, len);
217         skb->network_header += len;
218 }
219
220 static inline void nf_bridge_save_header(struct sk_buff *skb)
221 {
222         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
223
224         skb_copy_from_linear_data_offset(skb, -header_size,
225                                          skb->nf_bridge->data, header_size);
226 }
227
228 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
229 {
230         if (skb->nf_bridge->mask & BRNF_8021Q)
231                 skb->protocol = htons(ETH_P_8021Q);
232         else if (skb->nf_bridge->mask & BRNF_PPPoE)
233                 skb->protocol = htons(ETH_P_PPP_SES);
234 }
235
236 /* When handing a packet over to the IP layer
237  * check whether we have a skb that is in the
238  * expected format
239  */
240
241 static int br_parse_ip_options(struct sk_buff *skb)
242 {
243         struct ip_options *opt;
244         const struct iphdr *iph;
245         struct net_device *dev = skb->dev;
246         u32 len;
247
248         iph = ip_hdr(skb);
249         opt = &(IPCB(skb)->opt);
250
251         /* Basic sanity checks */
252         if (iph->ihl < 5 || iph->version != 4)
253                 goto inhdr_error;
254
255         if (!pskb_may_pull(skb, iph->ihl*4))
256                 goto inhdr_error;
257
258         iph = ip_hdr(skb);
259         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
260                 goto inhdr_error;
261
262         len = ntohs(iph->tot_len);
263         if (skb->len < len) {
264                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
265                 goto drop;
266         } else if (len < (iph->ihl*4))
267                 goto inhdr_error;
268
269         if (pskb_trim_rcsum(skb, len)) {
270                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
271                 goto drop;
272         }
273
274         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
275         if (iph->ihl == 5)
276                 return 0;
277
278         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
279         if (ip_options_compile(dev_net(dev), opt, skb))
280                 goto inhdr_error;
281
282         /* Check correct handling of SRR option */
283         if (unlikely(opt->srr)) {
284                 struct in_device *in_dev = __in_dev_get_rcu(dev);
285                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
286                         goto drop;
287
288                 if (ip_options_rcv_srr(skb))
289                         goto drop;
290         }
291
292         return 0;
293
294 inhdr_error:
295         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
296 drop:
297         return -1;
298 }
299
300 /* Fill in the header for fragmented IP packets handled by
301  * the IPv4 connection tracking code.
302  */
303 int nf_bridge_copy_header(struct sk_buff *skb)
304 {
305         int err;
306         unsigned int header_size;
307
308         nf_bridge_update_protocol(skb);
309         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
310         err = skb_cow_head(skb, header_size);
311         if (err)
312                 return err;
313
314         skb_copy_to_linear_data_offset(skb, -header_size,
315                                        skb->nf_bridge->data, header_size);
316         __skb_push(skb, nf_bridge_encap_header_len(skb));
317         return 0;
318 }
319
320 /* PF_BRIDGE/PRE_ROUTING *********************************************/
321 /* Undo the changes made for ip6tables PREROUTING and continue the
322  * bridge PRE_ROUTING hook. */
323 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
324 {
325         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
326         struct rtable *rt;
327
328         if (nf_bridge->mask & BRNF_PKT_TYPE) {
329                 skb->pkt_type = PACKET_OTHERHOST;
330                 nf_bridge->mask ^= BRNF_PKT_TYPE;
331         }
332         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
333
334         rt = bridge_parent_rtable(nf_bridge->physindev);
335         if (!rt) {
336                 kfree_skb(skb);
337                 return 0;
338         }
339         skb_dst_set_noref(skb, &rt->dst);
340
341         skb->dev = nf_bridge->physindev;
342         nf_bridge_update_protocol(skb);
343         nf_bridge_push_encap_header(skb);
344         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
345                        br_handle_frame_finish, 1);
346
347         return 0;
348 }
349
350 /* Obtain the correct destination MAC address, while preserving the original
351  * source MAC address. If we already know this address, we just copy it. If we
352  * don't, we use the neighbour framework to find out. In both cases, we make
353  * sure that br_handle_frame_finish() is called afterwards.
354  */
355 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
356 {
357         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
358         struct neighbour *neigh;
359         struct dst_entry *dst;
360
361         skb->dev = bridge_parent(skb->dev);
362         if (!skb->dev)
363                 goto free_skb;
364         dst = skb_dst(skb);
365         neigh = dst_get_neighbour_noref(dst);
366         if (neigh->hh.hh_len) {
367                 neigh_hh_bridge(&neigh->hh, skb);
368                 skb->dev = nf_bridge->physindev;
369                 return br_handle_frame_finish(skb);
370         } else {
371                 /* the neighbour function below overwrites the complete
372                  * MAC header, so we save the Ethernet source address and
373                  * protocol number. */
374                 skb_copy_from_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN), skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
375                 /* tell br_dev_xmit to continue with forwarding */
376                 nf_bridge->mask |= BRNF_BRIDGED_DNAT;
377                 return neigh->output(neigh, skb);
378         }
379 free_skb:
380         kfree_skb(skb);
381         return 0;
382 }
383
384 /* This requires some explaining. If DNAT has taken place,
385  * we will need to fix up the destination Ethernet address.
386  *
387  * There are two cases to consider:
388  * 1. The packet was DNAT'ed to a device in the same bridge
389  *    port group as it was received on. We can still bridge
390  *    the packet.
391  * 2. The packet was DNAT'ed to a different device, either
392  *    a non-bridged device or another bridge port group.
393  *    The packet will need to be routed.
394  *
395  * The correct way of distinguishing between these two cases is to
396  * call ip_route_input() and to look at skb->dst->dev, which is
397  * changed to the destination device if ip_route_input() succeeds.
398  *
399  * Let's first consider the case that ip_route_input() succeeds:
400  *
401  * If the output device equals the logical bridge device the packet
402  * came in on, we can consider this bridging. The corresponding MAC
403  * address will be obtained in br_nf_pre_routing_finish_bridge.
404  * Otherwise, the packet is considered to be routed and we just
405  * change the destination MAC address so that the packet will
406  * later be passed up to the IP stack to be routed. For a redirected
407  * packet, ip_route_input() will give back the localhost as output device,
408  * which differs from the bridge device.
409  *
410  * Let's now consider the case that ip_route_input() fails:
411  *
412  * This can be because the destination address is martian, in which case
413  * the packet will be dropped.
414  * If IP forwarding is disabled, ip_route_input() will fail, while
415  * ip_route_output_key() can return success. The source
416  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
417  * thinks we're handling a locally generated packet and won't care
418  * if IP forwarding is enabled. If the output device equals the logical bridge
419  * device, we proceed as if ip_route_input() succeeded. If it differs from the
420  * logical bridge port or if ip_route_output_key() fails we drop the packet.
421  */
422 static int br_nf_pre_routing_finish(struct sk_buff *skb)
423 {
424         struct net_device *dev = skb->dev;
425         struct iphdr *iph = ip_hdr(skb);
426         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
427         struct rtable *rt;
428         int err;
429
430         if (nf_bridge->mask & BRNF_PKT_TYPE) {
431                 skb->pkt_type = PACKET_OTHERHOST;
432                 nf_bridge->mask ^= BRNF_PKT_TYPE;
433         }
434         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
435         if (dnat_took_place(skb)) {
436                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
437                         struct in_device *in_dev = __in_dev_get_rcu(dev);
438
439                         /* If err equals -EHOSTUNREACH the error is due to a
440                          * martian destination or due to the fact that
441                          * forwarding is disabled. For most martian packets,
442                          * ip_route_output_key() will fail. It won't fail for 2 types of
443                          * martian destinations: loopback destinations and destination
444                          * 0.0.0.0. In both cases the packet will be dropped because the
445                          * destination is the loopback device and not the bridge. */
446                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
447                                 goto free_skb;
448
449                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
450                                              RT_TOS(iph->tos), 0);
451                         if (!IS_ERR(rt)) {
452                                 /* - Bridged-and-DNAT'ed traffic doesn't
453                                  *   require ip_forwarding. */
454                                 if (rt->dst.dev == dev) {
455                                         skb_dst_set(skb, &rt->dst);
456                                         goto bridged_dnat;
457                                 }
458                                 ip_rt_put(rt);
459                         }
460 free_skb:
461                         kfree_skb(skb);
462                         return 0;
463                 } else {
464                         if (skb_dst(skb)->dev == dev) {
465 bridged_dnat:
466                                 skb->dev = nf_bridge->physindev;
467                                 nf_bridge_update_protocol(skb);
468                                 nf_bridge_push_encap_header(skb);
469                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
470                                                NF_BR_PRE_ROUTING,
471                                                skb, skb->dev, NULL,
472                                                br_nf_pre_routing_finish_bridge,
473                                                1);
474                                 return 0;
475                         }
476                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
477                         skb->pkt_type = PACKET_HOST;
478                 }
479         } else {
480                 rt = bridge_parent_rtable(nf_bridge->physindev);
481                 if (!rt) {
482                         kfree_skb(skb);
483                         return 0;
484                 }
485                 skb_dst_set_noref(skb, &rt->dst);
486         }
487
488         skb->dev = nf_bridge->physindev;
489         nf_bridge_update_protocol(skb);
490         nf_bridge_push_encap_header(skb);
491         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
492                        br_handle_frame_finish, 1);
493
494         return 0;
495 }
496
497 /* Some common code for IPv4/IPv6 */
498 static struct net_device *setup_pre_routing(struct sk_buff *skb)
499 {
500         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
501
502         if (skb->pkt_type == PACKET_OTHERHOST) {
503                 skb->pkt_type = PACKET_HOST;
504                 nf_bridge->mask |= BRNF_PKT_TYPE;
505         }
506
507         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
508         nf_bridge->physindev = skb->dev;
509         skb->dev = bridge_parent(skb->dev);
510         if (skb->protocol == htons(ETH_P_8021Q))
511                 nf_bridge->mask |= BRNF_8021Q;
512         else if (skb->protocol == htons(ETH_P_PPP_SES))
513                 nf_bridge->mask |= BRNF_PPPoE;
514
515         return skb->dev;
516 }
517
518 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
519 static int check_hbh_len(struct sk_buff *skb)
520 {
521         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
522         u32 pkt_len;
523         const unsigned char *nh = skb_network_header(skb);
524         int off = raw - nh;
525         int len = (raw[1] + 1) << 3;
526
527         if ((raw + len) - skb->data > skb_headlen(skb))
528                 goto bad;
529
530         off += 2;
531         len -= 2;
532
533         while (len > 0) {
534                 int optlen = nh[off + 1] + 2;
535
536                 switch (nh[off]) {
537                 case IPV6_TLV_PAD0:
538                         optlen = 1;
539                         break;
540
541                 case IPV6_TLV_PADN:
542                         break;
543
544                 case IPV6_TLV_JUMBO:
545                         if (nh[off + 1] != 4 || (off & 3) != 2)
546                                 goto bad;
547                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
548                         if (pkt_len <= IPV6_MAXPLEN ||
549                             ipv6_hdr(skb)->payload_len)
550                                 goto bad;
551                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
552                                 goto bad;
553                         if (pskb_trim_rcsum(skb,
554                                             pkt_len + sizeof(struct ipv6hdr)))
555                                 goto bad;
556                         nh = skb_network_header(skb);
557                         break;
558                 default:
559                         if (optlen > len)
560                                 goto bad;
561                         break;
562                 }
563                 off += optlen;
564                 len -= optlen;
565         }
566         if (len == 0)
567                 return 0;
568 bad:
569         return -1;
570
571 }
572
573 /* Replicate the checks that IPv6 does on packet reception and pass the packet
574  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
575 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
576                                            struct sk_buff *skb,
577                                            const struct net_device *in,
578                                            const struct net_device *out,
579                                            int (*okfn)(struct sk_buff *))
580 {
581         const struct ipv6hdr *hdr;
582         u32 pkt_len;
583
584         if (skb->len < sizeof(struct ipv6hdr))
585                 return NF_DROP;
586
587         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
588                 return NF_DROP;
589
590         hdr = ipv6_hdr(skb);
591
592         if (hdr->version != 6)
593                 return NF_DROP;
594
595         pkt_len = ntohs(hdr->payload_len);
596
597         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
598                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
599                         return NF_DROP;
600                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
601                         return NF_DROP;
602         }
603         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
604                 return NF_DROP;
605
606         nf_bridge_put(skb->nf_bridge);
607         if (!nf_bridge_alloc(skb))
608                 return NF_DROP;
609         if (!setup_pre_routing(skb))
610                 return NF_DROP;
611
612         skb->protocol = htons(ETH_P_IPV6);
613         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
614                 br_nf_pre_routing_finish_ipv6);
615
616         return NF_STOLEN;
617 }
618
619 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
620  * Replicate the checks that IPv4 does on packet reception.
621  * Set skb->dev to the bridge device (i.e. parent of the
622  * receiving device) to make netfilter happy, the REDIRECT
623  * target in particular.  Save the original destination IP
624  * address to be able to detect DNAT afterwards. */
625 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
626                                       const struct net_device *in,
627                                       const struct net_device *out,
628                                       int (*okfn)(struct sk_buff *))
629 {
630         struct net_bridge_port *p;
631         struct net_bridge *br;
632         __u32 len = nf_bridge_encap_header_len(skb);
633
634         if (unlikely(!pskb_may_pull(skb, len)))
635                 return NF_DROP;
636
637         p = br_port_get_rcu(in);
638         if (p == NULL)
639                 return NF_DROP;
640         br = p->br;
641
642         if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
643             IS_PPPOE_IPV6(skb)) {
644                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
645                         return NF_ACCEPT;
646
647                 nf_bridge_pull_encap_header_rcsum(skb);
648                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
649         }
650
651         if (!brnf_call_iptables && !br->nf_call_iptables)
652                 return NF_ACCEPT;
653
654         if (skb->protocol != htons(ETH_P_IP) && !IS_VLAN_IP(skb) &&
655             !IS_PPPOE_IP(skb))
656                 return NF_ACCEPT;
657
658         nf_bridge_pull_encap_header_rcsum(skb);
659
660         if (br_parse_ip_options(skb))
661                 return NF_DROP;
662
663         nf_bridge_put(skb->nf_bridge);
664         if (!nf_bridge_alloc(skb))
665                 return NF_DROP;
666         if (!setup_pre_routing(skb))
667                 return NF_DROP;
668         store_orig_dstaddr(skb);
669         skb->protocol = htons(ETH_P_IP);
670
671         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
672                 br_nf_pre_routing_finish);
673
674         return NF_STOLEN;
675 }
676
677
678 /* PF_BRIDGE/LOCAL_IN ************************************************/
679 /* The packet is locally destined, which requires a real
680  * dst_entry, so detach the fake one.  On the way up, the
681  * packet would pass through PRE_ROUTING again (which already
682  * took place when the packet entered the bridge), but we
683  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
684  * prevent this from happening. */
685 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
686                                    const struct net_device *in,
687                                    const struct net_device *out,
688                                    int (*okfn)(struct sk_buff *))
689 {
690         struct rtable *rt = skb_rtable(skb);
691
692         if (rt && rt == bridge_parent_rtable(in))
693                 skb_dst_drop(skb);
694
695         return NF_ACCEPT;
696 }
697
698 /* PF_BRIDGE/FORWARD *************************************************/
699 static int br_nf_forward_finish(struct sk_buff *skb)
700 {
701         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
702         struct net_device *in;
703
704         if (skb->protocol != htons(ETH_P_ARP) && !IS_VLAN_ARP(skb)) {
705                 in = nf_bridge->physindev;
706                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
707                         skb->pkt_type = PACKET_OTHERHOST;
708                         nf_bridge->mask ^= BRNF_PKT_TYPE;
709                 }
710                 nf_bridge_update_protocol(skb);
711         } else {
712                 in = *((struct net_device **)(skb->cb));
713         }
714         nf_bridge_push_encap_header(skb);
715
716         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
717                        skb->dev, br_forward_finish, 1);
718         return 0;
719 }
720
721 /* This is the 'purely bridged' case.  For IP, we pass the packet to
722  * netfilter with indev and outdev set to the bridge device,
723  * but we are still able to filter on the 'real' indev/outdev
724  * because of the physdev module. For ARP, indev and outdev are the
725  * bridge ports. */
726 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
727                                      const struct net_device *in,
728                                      const struct net_device *out,
729                                      int (*okfn)(struct sk_buff *))
730 {
731         struct nf_bridge_info *nf_bridge;
732         struct net_device *parent;
733         u_int8_t pf;
734
735         if (!skb->nf_bridge)
736                 return NF_ACCEPT;
737
738         /* Need exclusive nf_bridge_info since we might have multiple
739          * different physoutdevs. */
740         if (!nf_bridge_unshare(skb))
741                 return NF_DROP;
742
743         parent = bridge_parent(out);
744         if (!parent)
745                 return NF_DROP;
746
747         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
748             IS_PPPOE_IP(skb))
749                 pf = PF_INET;
750         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
751                  IS_PPPOE_IPV6(skb))
752                 pf = PF_INET6;
753         else
754                 return NF_ACCEPT;
755
756         nf_bridge_pull_encap_header(skb);
757
758         nf_bridge = skb->nf_bridge;
759         if (skb->pkt_type == PACKET_OTHERHOST) {
760                 skb->pkt_type = PACKET_HOST;
761                 nf_bridge->mask |= BRNF_PKT_TYPE;
762         }
763
764         if (pf == PF_INET && br_parse_ip_options(skb))
765                 return NF_DROP;
766
767         /* The physdev module checks on this */
768         nf_bridge->mask |= BRNF_BRIDGED;
769         nf_bridge->physoutdev = skb->dev;
770         if (pf == PF_INET)
771                 skb->protocol = htons(ETH_P_IP);
772         else
773                 skb->protocol = htons(ETH_P_IPV6);
774
775         NF_HOOK(pf, NF_INET_FORWARD, skb, bridge_parent(in), parent,
776                 br_nf_forward_finish);
777
778         return NF_STOLEN;
779 }
780
781 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
782                                       const struct net_device *in,
783                                       const struct net_device *out,
784                                       int (*okfn)(struct sk_buff *))
785 {
786         struct net_bridge_port *p;
787         struct net_bridge *br;
788         struct net_device **d = (struct net_device **)(skb->cb);
789
790         p = br_port_get_rcu(out);
791         if (p == NULL)
792                 return NF_ACCEPT;
793         br = p->br;
794
795         if (!brnf_call_arptables && !br->nf_call_arptables)
796                 return NF_ACCEPT;
797
798         if (skb->protocol != htons(ETH_P_ARP)) {
799                 if (!IS_VLAN_ARP(skb))
800                         return NF_ACCEPT;
801                 nf_bridge_pull_encap_header(skb);
802         }
803
804         if (arp_hdr(skb)->ar_pln != 4) {
805                 if (IS_VLAN_ARP(skb))
806                         nf_bridge_push_encap_header(skb);
807                 return NF_ACCEPT;
808         }
809         *d = (struct net_device *)in;
810         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
811                 (struct net_device *)out, br_nf_forward_finish);
812
813         return NF_STOLEN;
814 }
815
816 #if IS_ENABLED(CONFIG_NF_CONNTRACK_IPV4)
817 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
818 {
819         int ret;
820
821         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
822             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
823             !skb_is_gso(skb)) {
824                 if (br_parse_ip_options(skb))
825                         /* Drop invalid packet */
826                         return NF_DROP;
827                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
828         } else
829                 ret = br_dev_queue_push_xmit(skb);
830
831         return ret;
832 }
833 #else
834 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
835 {
836         return br_dev_queue_push_xmit(skb);
837 }
838 #endif
839
840 /* PF_BRIDGE/POST_ROUTING ********************************************/
841 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
842                                        const struct net_device *in,
843                                        const struct net_device *out,
844                                        int (*okfn)(struct sk_buff *))
845 {
846         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
847         struct net_device *realoutdev = bridge_parent(skb->dev);
848         u_int8_t pf;
849
850         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
851                 return NF_ACCEPT;
852
853         if (!realoutdev)
854                 return NF_DROP;
855
856         if (skb->protocol == htons(ETH_P_IP) || IS_VLAN_IP(skb) ||
857             IS_PPPOE_IP(skb))
858                 pf = PF_INET;
859         else if (skb->protocol == htons(ETH_P_IPV6) || IS_VLAN_IPV6(skb) ||
860                  IS_PPPOE_IPV6(skb))
861                 pf = PF_INET6;
862         else
863                 return NF_ACCEPT;
864
865         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
866          * about the value of skb->pkt_type. */
867         if (skb->pkt_type == PACKET_OTHERHOST) {
868                 skb->pkt_type = PACKET_HOST;
869                 nf_bridge->mask |= BRNF_PKT_TYPE;
870         }
871
872         nf_bridge_pull_encap_header(skb);
873         nf_bridge_save_header(skb);
874         if (pf == PF_INET)
875                 skb->protocol = htons(ETH_P_IP);
876         else
877                 skb->protocol = htons(ETH_P_IPV6);
878
879         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
880                 br_nf_dev_queue_xmit);
881
882         return NF_STOLEN;
883 }
884
885 /* IP/SABOTAGE *****************************************************/
886 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
887  * for the second time. */
888 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
889                                    const struct net_device *in,
890                                    const struct net_device *out,
891                                    int (*okfn)(struct sk_buff *))
892 {
893         if (skb->nf_bridge &&
894             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
895                 return NF_STOP;
896         }
897
898         return NF_ACCEPT;
899 }
900
901 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
902  * br_dev_queue_push_xmit is called afterwards */
903 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
904         {
905                 .hook = br_nf_pre_routing,
906                 .owner = THIS_MODULE,
907                 .pf = PF_BRIDGE,
908                 .hooknum = NF_BR_PRE_ROUTING,
909                 .priority = NF_BR_PRI_BRNF,
910         },
911         {
912                 .hook = br_nf_local_in,
913                 .owner = THIS_MODULE,
914                 .pf = PF_BRIDGE,
915                 .hooknum = NF_BR_LOCAL_IN,
916                 .priority = NF_BR_PRI_BRNF,
917         },
918         {
919                 .hook = br_nf_forward_ip,
920                 .owner = THIS_MODULE,
921                 .pf = PF_BRIDGE,
922                 .hooknum = NF_BR_FORWARD,
923                 .priority = NF_BR_PRI_BRNF - 1,
924         },
925         {
926                 .hook = br_nf_forward_arp,
927                 .owner = THIS_MODULE,
928                 .pf = PF_BRIDGE,
929                 .hooknum = NF_BR_FORWARD,
930                 .priority = NF_BR_PRI_BRNF,
931         },
932         {
933                 .hook = br_nf_post_routing,
934                 .owner = THIS_MODULE,
935                 .pf = PF_BRIDGE,
936                 .hooknum = NF_BR_POST_ROUTING,
937                 .priority = NF_BR_PRI_LAST,
938         },
939         {
940                 .hook = ip_sabotage_in,
941                 .owner = THIS_MODULE,
942                 .pf = PF_INET,
943                 .hooknum = NF_INET_PRE_ROUTING,
944                 .priority = NF_IP_PRI_FIRST,
945         },
946         {
947                 .hook = ip_sabotage_in,
948                 .owner = THIS_MODULE,
949                 .pf = PF_INET6,
950                 .hooknum = NF_INET_PRE_ROUTING,
951                 .priority = NF_IP6_PRI_FIRST,
952         },
953 };
954
955 #ifdef CONFIG_SYSCTL
956 static
957 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
958                             void __user * buffer, size_t * lenp, loff_t * ppos)
959 {
960         int ret;
961
962         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
963
964         if (write && *(int *)(ctl->data))
965                 *(int *)(ctl->data) = 1;
966         return ret;
967 }
968
969 static ctl_table brnf_table[] = {
970         {
971                 .procname       = "bridge-nf-call-arptables",
972                 .data           = &brnf_call_arptables,
973                 .maxlen         = sizeof(int),
974                 .mode           = 0644,
975                 .proc_handler   = brnf_sysctl_call_tables,
976         },
977         {
978                 .procname       = "bridge-nf-call-iptables",
979                 .data           = &brnf_call_iptables,
980                 .maxlen         = sizeof(int),
981                 .mode           = 0644,
982                 .proc_handler   = brnf_sysctl_call_tables,
983         },
984         {
985                 .procname       = "bridge-nf-call-ip6tables",
986                 .data           = &brnf_call_ip6tables,
987                 .maxlen         = sizeof(int),
988                 .mode           = 0644,
989                 .proc_handler   = brnf_sysctl_call_tables,
990         },
991         {
992                 .procname       = "bridge-nf-filter-vlan-tagged",
993                 .data           = &brnf_filter_vlan_tagged,
994                 .maxlen         = sizeof(int),
995                 .mode           = 0644,
996                 .proc_handler   = brnf_sysctl_call_tables,
997         },
998         {
999                 .procname       = "bridge-nf-filter-pppoe-tagged",
1000                 .data           = &brnf_filter_pppoe_tagged,
1001                 .maxlen         = sizeof(int),
1002                 .mode           = 0644,
1003                 .proc_handler   = brnf_sysctl_call_tables,
1004         },
1005         { }
1006 };
1007
1008 static struct ctl_path brnf_path[] = {
1009         { .procname = "net", },
1010         { .procname = "bridge", },
1011         { }
1012 };
1013 #endif
1014
1015 int __init br_netfilter_init(void)
1016 {
1017         int ret;
1018
1019         ret = dst_entries_init(&fake_dst_ops);
1020         if (ret < 0)
1021                 return ret;
1022
1023         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1024         if (ret < 0) {
1025                 dst_entries_destroy(&fake_dst_ops);
1026                 return ret;
1027         }
1028 #ifdef CONFIG_SYSCTL
1029         brnf_sysctl_header = register_sysctl_paths(brnf_path, brnf_table);
1030         if (brnf_sysctl_header == NULL) {
1031                 printk(KERN_WARNING
1032                        "br_netfilter: can't register to sysctl.\n");
1033                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1034                 dst_entries_destroy(&fake_dst_ops);
1035                 return -ENOMEM;
1036         }
1037 #endif
1038         printk(KERN_NOTICE "Bridge firewalling registered\n");
1039         return 0;
1040 }
1041
1042 void br_netfilter_fini(void)
1043 {
1044         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1045 #ifdef CONFIG_SYSCTL
1046         unregister_sysctl_table(brnf_sysctl_header);
1047 #endif
1048         dst_entries_destroy(&fake_dst_ops);
1049 }