netns: ebtable: allow unprivileged users to operate ebtables
[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 static int brnf_pass_vlan_indev __read_mostly = 0;
58 #else
59 #define brnf_call_iptables 1
60 #define brnf_call_ip6tables 1
61 #define brnf_call_arptables 1
62 #define brnf_filter_vlan_tagged 0
63 #define brnf_filter_pppoe_tagged 0
64 #define brnf_pass_vlan_indev 0
65 #endif
66
67 #define IS_IP(skb) \
68         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
69
70 #define IS_IPV6(skb) \
71         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
72
73 #define IS_ARP(skb) \
74         (!vlan_tx_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
75
76 static inline __be16 vlan_proto(const struct sk_buff *skb)
77 {
78         if (vlan_tx_tag_present(skb))
79                 return skb->protocol;
80         else if (skb->protocol == htons(ETH_P_8021Q))
81                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
82         else
83                 return 0;
84 }
85
86 #define IS_VLAN_IP(skb) \
87         (vlan_proto(skb) == htons(ETH_P_IP) && \
88          brnf_filter_vlan_tagged)
89
90 #define IS_VLAN_IPV6(skb) \
91         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
92          brnf_filter_vlan_tagged)
93
94 #define IS_VLAN_ARP(skb) \
95         (vlan_proto(skb) == htons(ETH_P_ARP) && \
96          brnf_filter_vlan_tagged)
97
98 static inline __be16 pppoe_proto(const struct sk_buff *skb)
99 {
100         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
101                             sizeof(struct pppoe_hdr)));
102 }
103
104 #define IS_PPPOE_IP(skb) \
105         (skb->protocol == htons(ETH_P_PPP_SES) && \
106          pppoe_proto(skb) == htons(PPP_IP) && \
107          brnf_filter_pppoe_tagged)
108
109 #define IS_PPPOE_IPV6(skb) \
110         (skb->protocol == htons(ETH_P_PPP_SES) && \
111          pppoe_proto(skb) == htons(PPP_IPV6) && \
112          brnf_filter_pppoe_tagged)
113
114 static void fake_update_pmtu(struct dst_entry *dst, struct sock *sk,
115                              struct sk_buff *skb, u32 mtu)
116 {
117 }
118
119 static void fake_redirect(struct dst_entry *dst, struct sock *sk,
120                           struct sk_buff *skb)
121 {
122 }
123
124 static u32 *fake_cow_metrics(struct dst_entry *dst, unsigned long old)
125 {
126         return NULL;
127 }
128
129 static struct neighbour *fake_neigh_lookup(const struct dst_entry *dst,
130                                            struct sk_buff *skb,
131                                            const void *daddr)
132 {
133         return NULL;
134 }
135
136 static unsigned int fake_mtu(const struct dst_entry *dst)
137 {
138         return dst->dev->mtu;
139 }
140
141 static struct dst_ops fake_dst_ops = {
142         .family =               AF_INET,
143         .protocol =             cpu_to_be16(ETH_P_IP),
144         .update_pmtu =          fake_update_pmtu,
145         .redirect =             fake_redirect,
146         .cow_metrics =          fake_cow_metrics,
147         .neigh_lookup =         fake_neigh_lookup,
148         .mtu =                  fake_mtu,
149 };
150
151 /*
152  * Initialize bogus route table used to keep netfilter happy.
153  * Currently, we fill in the PMTU entry because netfilter
154  * refragmentation needs it, and the rt_flags entry because
155  * ipt_REJECT needs it.  Future netfilter modules might
156  * require us to fill additional fields.
157  */
158 static const u32 br_dst_default_metrics[RTAX_MAX] = {
159         [RTAX_MTU - 1] = 1500,
160 };
161
162 void br_netfilter_rtable_init(struct net_bridge *br)
163 {
164         struct rtable *rt = &br->fake_rtable;
165
166         atomic_set(&rt->dst.__refcnt, 1);
167         rt->dst.dev = br->dev;
168         rt->dst.path = &rt->dst;
169         dst_init_metrics(&rt->dst, br_dst_default_metrics, true);
170         rt->dst.flags   = DST_NOXFRM | DST_NOPEER | DST_FAKE_RTABLE;
171         rt->dst.ops = &fake_dst_ops;
172 }
173
174 static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
175 {
176         struct net_bridge_port *port;
177
178         port = br_port_get_rcu(dev);
179         return port ? &port->br->fake_rtable : NULL;
180 }
181
182 static inline struct net_device *bridge_parent(const struct net_device *dev)
183 {
184         struct net_bridge_port *port;
185
186         port = br_port_get_rcu(dev);
187         return port ? port->br->dev : NULL;
188 }
189
190 static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
191 {
192         skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
193         if (likely(skb->nf_bridge))
194                 atomic_set(&(skb->nf_bridge->use), 1);
195
196         return skb->nf_bridge;
197 }
198
199 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
200 {
201         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
202
203         if (atomic_read(&nf_bridge->use) > 1) {
204                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
205
206                 if (tmp) {
207                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
208                         atomic_set(&tmp->use, 1);
209                 }
210                 nf_bridge_put(nf_bridge);
211                 nf_bridge = tmp;
212         }
213         return nf_bridge;
214 }
215
216 static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
217 {
218         unsigned int len = nf_bridge_encap_header_len(skb);
219
220         skb_push(skb, len);
221         skb->network_header -= len;
222 }
223
224 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
225 {
226         unsigned int len = nf_bridge_encap_header_len(skb);
227
228         skb_pull(skb, len);
229         skb->network_header += len;
230 }
231
232 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
233 {
234         unsigned int len = nf_bridge_encap_header_len(skb);
235
236         skb_pull_rcsum(skb, len);
237         skb->network_header += len;
238 }
239
240 static inline void nf_bridge_save_header(struct sk_buff *skb)
241 {
242         int header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
243
244         skb_copy_from_linear_data_offset(skb, -header_size,
245                                          skb->nf_bridge->data, header_size);
246 }
247
248 static inline void nf_bridge_update_protocol(struct sk_buff *skb)
249 {
250         if (skb->nf_bridge->mask & BRNF_8021Q)
251                 skb->protocol = htons(ETH_P_8021Q);
252         else if (skb->nf_bridge->mask & BRNF_PPPoE)
253                 skb->protocol = htons(ETH_P_PPP_SES);
254 }
255
256 /* When handing a packet over to the IP layer
257  * check whether we have a skb that is in the
258  * expected format
259  */
260
261 static int br_parse_ip_options(struct sk_buff *skb)
262 {
263         struct ip_options *opt;
264         const struct iphdr *iph;
265         struct net_device *dev = skb->dev;
266         u32 len;
267
268         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
269                 goto inhdr_error;
270
271         iph = ip_hdr(skb);
272         opt = &(IPCB(skb)->opt);
273
274         /* Basic sanity checks */
275         if (iph->ihl < 5 || iph->version != 4)
276                 goto inhdr_error;
277
278         if (!pskb_may_pull(skb, iph->ihl*4))
279                 goto inhdr_error;
280
281         iph = ip_hdr(skb);
282         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
283                 goto inhdr_error;
284
285         len = ntohs(iph->tot_len);
286         if (skb->len < len) {
287                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
288                 goto drop;
289         } else if (len < (iph->ihl*4))
290                 goto inhdr_error;
291
292         if (pskb_trim_rcsum(skb, len)) {
293                 IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
294                 goto drop;
295         }
296
297         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
298         if (iph->ihl == 5)
299                 return 0;
300
301         opt->optlen = iph->ihl*4 - sizeof(struct iphdr);
302         if (ip_options_compile(dev_net(dev), opt, skb))
303                 goto inhdr_error;
304
305         /* Check correct handling of SRR option */
306         if (unlikely(opt->srr)) {
307                 struct in_device *in_dev = __in_dev_get_rcu(dev);
308                 if (in_dev && !IN_DEV_SOURCE_ROUTE(in_dev))
309                         goto drop;
310
311                 if (ip_options_rcv_srr(skb))
312                         goto drop;
313         }
314
315         return 0;
316
317 inhdr_error:
318         IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
319 drop:
320         return -1;
321 }
322
323 /* Fill in the header for fragmented IP packets handled by
324  * the IPv4 connection tracking code.
325  */
326 int nf_bridge_copy_header(struct sk_buff *skb)
327 {
328         int err;
329         unsigned int header_size;
330
331         nf_bridge_update_protocol(skb);
332         header_size = ETH_HLEN + nf_bridge_encap_header_len(skb);
333         err = skb_cow_head(skb, header_size);
334         if (err)
335                 return err;
336
337         skb_copy_to_linear_data_offset(skb, -header_size,
338                                        skb->nf_bridge->data, header_size);
339         __skb_push(skb, nf_bridge_encap_header_len(skb));
340         return 0;
341 }
342
343 /* PF_BRIDGE/PRE_ROUTING *********************************************/
344 /* Undo the changes made for ip6tables PREROUTING and continue the
345  * bridge PRE_ROUTING hook. */
346 static int br_nf_pre_routing_finish_ipv6(struct sk_buff *skb)
347 {
348         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
349         struct rtable *rt;
350
351         if (nf_bridge->mask & BRNF_PKT_TYPE) {
352                 skb->pkt_type = PACKET_OTHERHOST;
353                 nf_bridge->mask ^= BRNF_PKT_TYPE;
354         }
355         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
356
357         rt = bridge_parent_rtable(nf_bridge->physindev);
358         if (!rt) {
359                 kfree_skb(skb);
360                 return 0;
361         }
362         skb_dst_set_noref(skb, &rt->dst);
363
364         skb->dev = nf_bridge->physindev;
365         nf_bridge_update_protocol(skb);
366         nf_bridge_push_encap_header(skb);
367         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
368                        br_handle_frame_finish, 1);
369
370         return 0;
371 }
372
373 /* Obtain the correct destination MAC address, while preserving the original
374  * source MAC address. If we already know this address, we just copy it. If we
375  * don't, we use the neighbour framework to find out. In both cases, we make
376  * sure that br_handle_frame_finish() is called afterwards.
377  */
378 static int br_nf_pre_routing_finish_bridge(struct sk_buff *skb)
379 {
380         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
381         struct neighbour *neigh;
382         struct dst_entry *dst;
383
384         skb->dev = bridge_parent(skb->dev);
385         if (!skb->dev)
386                 goto free_skb;
387         dst = skb_dst(skb);
388         neigh = dst_neigh_lookup_skb(dst, skb);
389         if (neigh) {
390                 int ret;
391
392                 if (neigh->hh.hh_len) {
393                         neigh_hh_bridge(&neigh->hh, skb);
394                         skb->dev = nf_bridge->physindev;
395                         ret = br_handle_frame_finish(skb);
396                 } else {
397                         /* the neighbour function below overwrites the complete
398                          * MAC header, so we save the Ethernet source address and
399                          * protocol number.
400                          */
401                         skb_copy_from_linear_data_offset(skb,
402                                                          -(ETH_HLEN-ETH_ALEN),
403                                                          skb->nf_bridge->data,
404                                                          ETH_HLEN-ETH_ALEN);
405                         /* tell br_dev_xmit to continue with forwarding */
406                         nf_bridge->mask |= BRNF_BRIDGED_DNAT;
407                         ret = neigh->output(neigh, skb);
408                 }
409                 neigh_release(neigh);
410                 return ret;
411         }
412 free_skb:
413         kfree_skb(skb);
414         return 0;
415 }
416
417 /* This requires some explaining. If DNAT has taken place,
418  * we will need to fix up the destination Ethernet address.
419  *
420  * There are two cases to consider:
421  * 1. The packet was DNAT'ed to a device in the same bridge
422  *    port group as it was received on. We can still bridge
423  *    the packet.
424  * 2. The packet was DNAT'ed to a different device, either
425  *    a non-bridged device or another bridge port group.
426  *    The packet will need to be routed.
427  *
428  * The correct way of distinguishing between these two cases is to
429  * call ip_route_input() and to look at skb->dst->dev, which is
430  * changed to the destination device if ip_route_input() succeeds.
431  *
432  * Let's first consider the case that ip_route_input() succeeds:
433  *
434  * If the output device equals the logical bridge device the packet
435  * came in on, we can consider this bridging. The corresponding MAC
436  * address will be obtained in br_nf_pre_routing_finish_bridge.
437  * Otherwise, the packet is considered to be routed and we just
438  * change the destination MAC address so that the packet will
439  * later be passed up to the IP stack to be routed. For a redirected
440  * packet, ip_route_input() will give back the localhost as output device,
441  * which differs from the bridge device.
442  *
443  * Let's now consider the case that ip_route_input() fails:
444  *
445  * This can be because the destination address is martian, in which case
446  * the packet will be dropped.
447  * If IP forwarding is disabled, ip_route_input() will fail, while
448  * ip_route_output_key() can return success. The source
449  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
450  * thinks we're handling a locally generated packet and won't care
451  * if IP forwarding is enabled. If the output device equals the logical bridge
452  * device, we proceed as if ip_route_input() succeeded. If it differs from the
453  * logical bridge port or if ip_route_output_key() fails we drop the packet.
454  */
455 static int br_nf_pre_routing_finish(struct sk_buff *skb)
456 {
457         struct net_device *dev = skb->dev;
458         struct iphdr *iph = ip_hdr(skb);
459         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
460         struct rtable *rt;
461         int err;
462
463         if (nf_bridge->mask & BRNF_PKT_TYPE) {
464                 skb->pkt_type = PACKET_OTHERHOST;
465                 nf_bridge->mask ^= BRNF_PKT_TYPE;
466         }
467         nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
468         if (dnat_took_place(skb)) {
469                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
470                         struct in_device *in_dev = __in_dev_get_rcu(dev);
471
472                         /* If err equals -EHOSTUNREACH the error is due to a
473                          * martian destination or due to the fact that
474                          * forwarding is disabled. For most martian packets,
475                          * ip_route_output_key() will fail. It won't fail for 2 types of
476                          * martian destinations: loopback destinations and destination
477                          * 0.0.0.0. In both cases the packet will be dropped because the
478                          * destination is the loopback device and not the bridge. */
479                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
480                                 goto free_skb;
481
482                         rt = ip_route_output(dev_net(dev), iph->daddr, 0,
483                                              RT_TOS(iph->tos), 0);
484                         if (!IS_ERR(rt)) {
485                                 /* - Bridged-and-DNAT'ed traffic doesn't
486                                  *   require ip_forwarding. */
487                                 if (rt->dst.dev == dev) {
488                                         skb_dst_set(skb, &rt->dst);
489                                         goto bridged_dnat;
490                                 }
491                                 ip_rt_put(rt);
492                         }
493 free_skb:
494                         kfree_skb(skb);
495                         return 0;
496                 } else {
497                         if (skb_dst(skb)->dev == dev) {
498 bridged_dnat:
499                                 skb->dev = nf_bridge->physindev;
500                                 nf_bridge_update_protocol(skb);
501                                 nf_bridge_push_encap_header(skb);
502                                 NF_HOOK_THRESH(NFPROTO_BRIDGE,
503                                                NF_BR_PRE_ROUTING,
504                                                skb, skb->dev, NULL,
505                                                br_nf_pre_routing_finish_bridge,
506                                                1);
507                                 return 0;
508                         }
509                         memcpy(eth_hdr(skb)->h_dest, dev->dev_addr, ETH_ALEN);
510                         skb->pkt_type = PACKET_HOST;
511                 }
512         } else {
513                 rt = bridge_parent_rtable(nf_bridge->physindev);
514                 if (!rt) {
515                         kfree_skb(skb);
516                         return 0;
517                 }
518                 skb_dst_set_noref(skb, &rt->dst);
519         }
520
521         skb->dev = nf_bridge->physindev;
522         nf_bridge_update_protocol(skb);
523         nf_bridge_push_encap_header(skb);
524         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, skb, skb->dev, NULL,
525                        br_handle_frame_finish, 1);
526
527         return 0;
528 }
529
530 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
531 {
532         struct net_device *vlan, *br;
533
534         br = bridge_parent(dev);
535         if (brnf_pass_vlan_indev == 0 || !vlan_tx_tag_present(skb))
536                 return br;
537
538         vlan = __vlan_find_dev_deep(br, vlan_tx_tag_get(skb) & VLAN_VID_MASK);
539
540         return vlan ? vlan : br;
541 }
542
543 /* Some common code for IPv4/IPv6 */
544 static struct net_device *setup_pre_routing(struct sk_buff *skb)
545 {
546         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
547
548         if (skb->pkt_type == PACKET_OTHERHOST) {
549                 skb->pkt_type = PACKET_HOST;
550                 nf_bridge->mask |= BRNF_PKT_TYPE;
551         }
552
553         nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
554         nf_bridge->physindev = skb->dev;
555         skb->dev = brnf_get_logical_dev(skb, skb->dev);
556         if (skb->protocol == htons(ETH_P_8021Q))
557                 nf_bridge->mask |= BRNF_8021Q;
558         else if (skb->protocol == htons(ETH_P_PPP_SES))
559                 nf_bridge->mask |= BRNF_PPPoE;
560
561         return skb->dev;
562 }
563
564 /* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
565 static int check_hbh_len(struct sk_buff *skb)
566 {
567         unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
568         u32 pkt_len;
569         const unsigned char *nh = skb_network_header(skb);
570         int off = raw - nh;
571         int len = (raw[1] + 1) << 3;
572
573         if ((raw + len) - skb->data > skb_headlen(skb))
574                 goto bad;
575
576         off += 2;
577         len -= 2;
578
579         while (len > 0) {
580                 int optlen = nh[off + 1] + 2;
581
582                 switch (nh[off]) {
583                 case IPV6_TLV_PAD1:
584                         optlen = 1;
585                         break;
586
587                 case IPV6_TLV_PADN:
588                         break;
589
590                 case IPV6_TLV_JUMBO:
591                         if (nh[off + 1] != 4 || (off & 3) != 2)
592                                 goto bad;
593                         pkt_len = ntohl(*(__be32 *) (nh + off + 2));
594                         if (pkt_len <= IPV6_MAXPLEN ||
595                             ipv6_hdr(skb)->payload_len)
596                                 goto bad;
597                         if (pkt_len > skb->len - sizeof(struct ipv6hdr))
598                                 goto bad;
599                         if (pskb_trim_rcsum(skb,
600                                             pkt_len + sizeof(struct ipv6hdr)))
601                                 goto bad;
602                         nh = skb_network_header(skb);
603                         break;
604                 default:
605                         if (optlen > len)
606                                 goto bad;
607                         break;
608                 }
609                 off += optlen;
610                 len -= optlen;
611         }
612         if (len == 0)
613                 return 0;
614 bad:
615         return -1;
616
617 }
618
619 /* Replicate the checks that IPv6 does on packet reception and pass the packet
620  * to ip6tables, which doesn't support NAT, so things are fairly simple. */
621 static unsigned int br_nf_pre_routing_ipv6(unsigned int hook,
622                                            struct sk_buff *skb,
623                                            const struct net_device *in,
624                                            const struct net_device *out,
625                                            int (*okfn)(struct sk_buff *))
626 {
627         const struct ipv6hdr *hdr;
628         u32 pkt_len;
629
630         if (skb->len < sizeof(struct ipv6hdr))
631                 return NF_DROP;
632
633         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
634                 return NF_DROP;
635
636         hdr = ipv6_hdr(skb);
637
638         if (hdr->version != 6)
639                 return NF_DROP;
640
641         pkt_len = ntohs(hdr->payload_len);
642
643         if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
644                 if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
645                         return NF_DROP;
646                 if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
647                         return NF_DROP;
648         }
649         if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
650                 return NF_DROP;
651
652         nf_bridge_put(skb->nf_bridge);
653         if (!nf_bridge_alloc(skb))
654                 return NF_DROP;
655         if (!setup_pre_routing(skb))
656                 return NF_DROP;
657
658         skb->protocol = htons(ETH_P_IPV6);
659         NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
660                 br_nf_pre_routing_finish_ipv6);
661
662         return NF_STOLEN;
663 }
664
665 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
666  * Replicate the checks that IPv4 does on packet reception.
667  * Set skb->dev to the bridge device (i.e. parent of the
668  * receiving device) to make netfilter happy, the REDIRECT
669  * target in particular.  Save the original destination IP
670  * address to be able to detect DNAT afterwards. */
671 static unsigned int br_nf_pre_routing(unsigned int hook, struct sk_buff *skb,
672                                       const struct net_device *in,
673                                       const struct net_device *out,
674                                       int (*okfn)(struct sk_buff *))
675 {
676         struct net_bridge_port *p;
677         struct net_bridge *br;
678         __u32 len = nf_bridge_encap_header_len(skb);
679
680         if (unlikely(!pskb_may_pull(skb, len)))
681                 return NF_DROP;
682
683         p = br_port_get_rcu(in);
684         if (p == NULL)
685                 return NF_DROP;
686         br = p->br;
687
688         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
689                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
690                         return NF_ACCEPT;
691
692                 nf_bridge_pull_encap_header_rcsum(skb);
693                 return br_nf_pre_routing_ipv6(hook, skb, in, out, okfn);
694         }
695
696         if (!brnf_call_iptables && !br->nf_call_iptables)
697                 return NF_ACCEPT;
698
699         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
700                 return NF_ACCEPT;
701
702         nf_bridge_pull_encap_header_rcsum(skb);
703
704         if (br_parse_ip_options(skb))
705                 return NF_DROP;
706
707         nf_bridge_put(skb->nf_bridge);
708         if (!nf_bridge_alloc(skb))
709                 return NF_DROP;
710         if (!setup_pre_routing(skb))
711                 return NF_DROP;
712         store_orig_dstaddr(skb);
713         skb->protocol = htons(ETH_P_IP);
714
715         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, skb, skb->dev, NULL,
716                 br_nf_pre_routing_finish);
717
718         return NF_STOLEN;
719 }
720
721
722 /* PF_BRIDGE/LOCAL_IN ************************************************/
723 /* The packet is locally destined, which requires a real
724  * dst_entry, so detach the fake one.  On the way up, the
725  * packet would pass through PRE_ROUTING again (which already
726  * took place when the packet entered the bridge), but we
727  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
728  * prevent this from happening. */
729 static unsigned int br_nf_local_in(unsigned int hook, struct sk_buff *skb,
730                                    const struct net_device *in,
731                                    const struct net_device *out,
732                                    int (*okfn)(struct sk_buff *))
733 {
734         br_drop_fake_rtable(skb);
735         return NF_ACCEPT;
736 }
737
738 /* PF_BRIDGE/FORWARD *************************************************/
739 static int br_nf_forward_finish(struct sk_buff *skb)
740 {
741         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
742         struct net_device *in;
743
744         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
745                 in = nf_bridge->physindev;
746                 if (nf_bridge->mask & BRNF_PKT_TYPE) {
747                         skb->pkt_type = PACKET_OTHERHOST;
748                         nf_bridge->mask ^= BRNF_PKT_TYPE;
749                 }
750                 nf_bridge_update_protocol(skb);
751         } else {
752                 in = *((struct net_device **)(skb->cb));
753         }
754         nf_bridge_push_encap_header(skb);
755
756         NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, skb, in,
757                        skb->dev, br_forward_finish, 1);
758         return 0;
759 }
760
761
762 /* This is the 'purely bridged' case.  For IP, we pass the packet to
763  * netfilter with indev and outdev set to the bridge device,
764  * but we are still able to filter on the 'real' indev/outdev
765  * because of the physdev module. For ARP, indev and outdev are the
766  * bridge ports. */
767 static unsigned int br_nf_forward_ip(unsigned int hook, struct sk_buff *skb,
768                                      const struct net_device *in,
769                                      const struct net_device *out,
770                                      int (*okfn)(struct sk_buff *))
771 {
772         struct nf_bridge_info *nf_bridge;
773         struct net_device *parent;
774         u_int8_t pf;
775
776         if (!skb->nf_bridge)
777                 return NF_ACCEPT;
778
779         /* Need exclusive nf_bridge_info since we might have multiple
780          * different physoutdevs. */
781         if (!nf_bridge_unshare(skb))
782                 return NF_DROP;
783
784         parent = bridge_parent(out);
785         if (!parent)
786                 return NF_DROP;
787
788         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
789                 pf = NFPROTO_IPV4;
790         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
791                 pf = NFPROTO_IPV6;
792         else
793                 return NF_ACCEPT;
794
795         nf_bridge_pull_encap_header(skb);
796
797         nf_bridge = skb->nf_bridge;
798         if (skb->pkt_type == PACKET_OTHERHOST) {
799                 skb->pkt_type = PACKET_HOST;
800                 nf_bridge->mask |= BRNF_PKT_TYPE;
801         }
802
803         if (pf == NFPROTO_IPV4 && br_parse_ip_options(skb))
804                 return NF_DROP;
805
806         /* The physdev module checks on this */
807         nf_bridge->mask |= BRNF_BRIDGED;
808         nf_bridge->physoutdev = skb->dev;
809         if (pf == NFPROTO_IPV4)
810                 skb->protocol = htons(ETH_P_IP);
811         else
812                 skb->protocol = htons(ETH_P_IPV6);
813
814         NF_HOOK(pf, NF_INET_FORWARD, skb, brnf_get_logical_dev(skb, in), parent,
815                 br_nf_forward_finish);
816
817         return NF_STOLEN;
818 }
819
820 static unsigned int br_nf_forward_arp(unsigned int hook, struct sk_buff *skb,
821                                       const struct net_device *in,
822                                       const struct net_device *out,
823                                       int (*okfn)(struct sk_buff *))
824 {
825         struct net_bridge_port *p;
826         struct net_bridge *br;
827         struct net_device **d = (struct net_device **)(skb->cb);
828
829         p = br_port_get_rcu(out);
830         if (p == NULL)
831                 return NF_ACCEPT;
832         br = p->br;
833
834         if (!brnf_call_arptables && !br->nf_call_arptables)
835                 return NF_ACCEPT;
836
837         if (!IS_ARP(skb)) {
838                 if (!IS_VLAN_ARP(skb))
839                         return NF_ACCEPT;
840                 nf_bridge_pull_encap_header(skb);
841         }
842
843         if (arp_hdr(skb)->ar_pln != 4) {
844                 if (IS_VLAN_ARP(skb))
845                         nf_bridge_push_encap_header(skb);
846                 return NF_ACCEPT;
847         }
848         *d = (struct net_device *)in;
849         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, skb, (struct net_device *)in,
850                 (struct net_device *)out, br_nf_forward_finish);
851
852         return NF_STOLEN;
853 }
854
855 #if IS_ENABLED(CONFIG_NF_CONNTRACK_IPV4)
856 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
857 {
858         int ret;
859
860         if (skb->nfct != NULL && skb->protocol == htons(ETH_P_IP) &&
861             skb->len + nf_bridge_mtu_reduction(skb) > skb->dev->mtu &&
862             !skb_is_gso(skb)) {
863                 if (br_parse_ip_options(skb))
864                         /* Drop invalid packet */
865                         return NF_DROP;
866                 ret = ip_fragment(skb, br_dev_queue_push_xmit);
867         } else
868                 ret = br_dev_queue_push_xmit(skb);
869
870         return ret;
871 }
872 #else
873 static int br_nf_dev_queue_xmit(struct sk_buff *skb)
874 {
875         return br_dev_queue_push_xmit(skb);
876 }
877 #endif
878
879 /* PF_BRIDGE/POST_ROUTING ********************************************/
880 static unsigned int br_nf_post_routing(unsigned int hook, struct sk_buff *skb,
881                                        const struct net_device *in,
882                                        const struct net_device *out,
883                                        int (*okfn)(struct sk_buff *))
884 {
885         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
886         struct net_device *realoutdev = bridge_parent(skb->dev);
887         u_int8_t pf;
888
889         if (!nf_bridge || !(nf_bridge->mask & BRNF_BRIDGED))
890                 return NF_ACCEPT;
891
892         if (!realoutdev)
893                 return NF_DROP;
894
895         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
896                 pf = NFPROTO_IPV4;
897         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
898                 pf = NFPROTO_IPV6;
899         else
900                 return NF_ACCEPT;
901
902         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
903          * about the value of skb->pkt_type. */
904         if (skb->pkt_type == PACKET_OTHERHOST) {
905                 skb->pkt_type = PACKET_HOST;
906                 nf_bridge->mask |= BRNF_PKT_TYPE;
907         }
908
909         nf_bridge_pull_encap_header(skb);
910         nf_bridge_save_header(skb);
911         if (pf == NFPROTO_IPV4)
912                 skb->protocol = htons(ETH_P_IP);
913         else
914                 skb->protocol = htons(ETH_P_IPV6);
915
916         NF_HOOK(pf, NF_INET_POST_ROUTING, skb, NULL, realoutdev,
917                 br_nf_dev_queue_xmit);
918
919         return NF_STOLEN;
920 }
921
922 /* IP/SABOTAGE *****************************************************/
923 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
924  * for the second time. */
925 static unsigned int ip_sabotage_in(unsigned int hook, struct sk_buff *skb,
926                                    const struct net_device *in,
927                                    const struct net_device *out,
928                                    int (*okfn)(struct sk_buff *))
929 {
930         if (skb->nf_bridge &&
931             !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
932                 return NF_STOP;
933         }
934
935         return NF_ACCEPT;
936 }
937
938 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
939  * br_dev_queue_push_xmit is called afterwards */
940 static struct nf_hook_ops br_nf_ops[] __read_mostly = {
941         {
942                 .hook = br_nf_pre_routing,
943                 .owner = THIS_MODULE,
944                 .pf = NFPROTO_BRIDGE,
945                 .hooknum = NF_BR_PRE_ROUTING,
946                 .priority = NF_BR_PRI_BRNF,
947         },
948         {
949                 .hook = br_nf_local_in,
950                 .owner = THIS_MODULE,
951                 .pf = NFPROTO_BRIDGE,
952                 .hooknum = NF_BR_LOCAL_IN,
953                 .priority = NF_BR_PRI_BRNF,
954         },
955         {
956                 .hook = br_nf_forward_ip,
957                 .owner = THIS_MODULE,
958                 .pf = NFPROTO_BRIDGE,
959                 .hooknum = NF_BR_FORWARD,
960                 .priority = NF_BR_PRI_BRNF - 1,
961         },
962         {
963                 .hook = br_nf_forward_arp,
964                 .owner = THIS_MODULE,
965                 .pf = NFPROTO_BRIDGE,
966                 .hooknum = NF_BR_FORWARD,
967                 .priority = NF_BR_PRI_BRNF,
968         },
969         {
970                 .hook = br_nf_post_routing,
971                 .owner = THIS_MODULE,
972                 .pf = NFPROTO_BRIDGE,
973                 .hooknum = NF_BR_POST_ROUTING,
974                 .priority = NF_BR_PRI_LAST,
975         },
976         {
977                 .hook = ip_sabotage_in,
978                 .owner = THIS_MODULE,
979                 .pf = NFPROTO_IPV4,
980                 .hooknum = NF_INET_PRE_ROUTING,
981                 .priority = NF_IP_PRI_FIRST,
982         },
983         {
984                 .hook = ip_sabotage_in,
985                 .owner = THIS_MODULE,
986                 .pf = NFPROTO_IPV6,
987                 .hooknum = NF_INET_PRE_ROUTING,
988                 .priority = NF_IP6_PRI_FIRST,
989         },
990 };
991
992 #ifdef CONFIG_SYSCTL
993 static
994 int brnf_sysctl_call_tables(ctl_table * ctl, int write,
995                             void __user * buffer, size_t * lenp, loff_t * ppos)
996 {
997         int ret;
998
999         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1000
1001         if (write && *(int *)(ctl->data))
1002                 *(int *)(ctl->data) = 1;
1003         return ret;
1004 }
1005
1006 static ctl_table brnf_table[] = {
1007         {
1008                 .procname       = "bridge-nf-call-arptables",
1009                 .data           = &brnf_call_arptables,
1010                 .maxlen         = sizeof(int),
1011                 .mode           = 0644,
1012                 .proc_handler   = brnf_sysctl_call_tables,
1013         },
1014         {
1015                 .procname       = "bridge-nf-call-iptables",
1016                 .data           = &brnf_call_iptables,
1017                 .maxlen         = sizeof(int),
1018                 .mode           = 0644,
1019                 .proc_handler   = brnf_sysctl_call_tables,
1020         },
1021         {
1022                 .procname       = "bridge-nf-call-ip6tables",
1023                 .data           = &brnf_call_ip6tables,
1024                 .maxlen         = sizeof(int),
1025                 .mode           = 0644,
1026                 .proc_handler   = brnf_sysctl_call_tables,
1027         },
1028         {
1029                 .procname       = "bridge-nf-filter-vlan-tagged",
1030                 .data           = &brnf_filter_vlan_tagged,
1031                 .maxlen         = sizeof(int),
1032                 .mode           = 0644,
1033                 .proc_handler   = brnf_sysctl_call_tables,
1034         },
1035         {
1036                 .procname       = "bridge-nf-filter-pppoe-tagged",
1037                 .data           = &brnf_filter_pppoe_tagged,
1038                 .maxlen         = sizeof(int),
1039                 .mode           = 0644,
1040                 .proc_handler   = brnf_sysctl_call_tables,
1041         },
1042         {
1043                 .procname       = "bridge-nf-pass-vlan-input-dev",
1044                 .data           = &brnf_pass_vlan_indev,
1045                 .maxlen         = sizeof(int),
1046                 .mode           = 0644,
1047                 .proc_handler   = brnf_sysctl_call_tables,
1048         },
1049         { }
1050 };
1051 #endif
1052
1053 int __init br_netfilter_init(void)
1054 {
1055         int ret;
1056
1057         ret = dst_entries_init(&fake_dst_ops);
1058         if (ret < 0)
1059                 return ret;
1060
1061         ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1062         if (ret < 0) {
1063                 dst_entries_destroy(&fake_dst_ops);
1064                 return ret;
1065         }
1066 #ifdef CONFIG_SYSCTL
1067         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1068         if (brnf_sysctl_header == NULL) {
1069                 printk(KERN_WARNING
1070                        "br_netfilter: can't register to sysctl.\n");
1071                 nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1072                 dst_entries_destroy(&fake_dst_ops);
1073                 return -ENOMEM;
1074         }
1075 #endif
1076         printk(KERN_NOTICE "Bridge firewalling registered\n");
1077         return 0;
1078 }
1079
1080 void br_netfilter_fini(void)
1081 {
1082         nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1083 #ifdef CONFIG_SYSCTL
1084         unregister_net_sysctl_table(brnf_sysctl_header);
1085 #endif
1086         dst_entries_destroy(&fake_dst_ops);
1087 }