Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf
[sfrench/cifs-2.6.git] / net / bridge / br_netfilter_hooks.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/rculist.h>
34 #include <linux/inetdevice.h>
35
36 #include <net/ip.h>
37 #include <net/ipv6.h>
38 #include <net/addrconf.h>
39 #include <net/route.h>
40 #include <net/netfilter/br_netfilter.h>
41 #include <net/netns/generic.h>
42
43 #include <linux/uaccess.h>
44 #include "br_private.h"
45 #ifdef CONFIG_SYSCTL
46 #include <linux/sysctl.h>
47 #endif
48
49 static unsigned int brnf_net_id __read_mostly;
50
51 struct brnf_net {
52         bool enabled;
53 };
54
55 #ifdef CONFIG_SYSCTL
56 static struct ctl_table_header *brnf_sysctl_header;
57 static int brnf_call_iptables __read_mostly = 1;
58 static int brnf_call_ip6tables __read_mostly = 1;
59 static int brnf_call_arptables __read_mostly = 1;
60 static int brnf_filter_vlan_tagged __read_mostly;
61 static int brnf_filter_pppoe_tagged __read_mostly;
62 static int brnf_pass_vlan_indev __read_mostly;
63 #else
64 #define brnf_call_iptables 1
65 #define brnf_call_ip6tables 1
66 #define brnf_call_arptables 1
67 #define brnf_filter_vlan_tagged 0
68 #define brnf_filter_pppoe_tagged 0
69 #define brnf_pass_vlan_indev 0
70 #endif
71
72 #define IS_IP(skb) \
73         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
74
75 #define IS_IPV6(skb) \
76         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
77
78 #define IS_ARP(skb) \
79         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
80
81 static inline __be16 vlan_proto(const struct sk_buff *skb)
82 {
83         if (skb_vlan_tag_present(skb))
84                 return skb->protocol;
85         else if (skb->protocol == htons(ETH_P_8021Q))
86                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
87         else
88                 return 0;
89 }
90
91 #define IS_VLAN_IP(skb) \
92         (vlan_proto(skb) == htons(ETH_P_IP) && \
93          brnf_filter_vlan_tagged)
94
95 #define IS_VLAN_IPV6(skb) \
96         (vlan_proto(skb) == htons(ETH_P_IPV6) && \
97          brnf_filter_vlan_tagged)
98
99 #define IS_VLAN_ARP(skb) \
100         (vlan_proto(skb) == htons(ETH_P_ARP) && \
101          brnf_filter_vlan_tagged)
102
103 static inline __be16 pppoe_proto(const struct sk_buff *skb)
104 {
105         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
106                             sizeof(struct pppoe_hdr)));
107 }
108
109 #define IS_PPPOE_IP(skb) \
110         (skb->protocol == htons(ETH_P_PPP_SES) && \
111          pppoe_proto(skb) == htons(PPP_IP) && \
112          brnf_filter_pppoe_tagged)
113
114 #define IS_PPPOE_IPV6(skb) \
115         (skb->protocol == htons(ETH_P_PPP_SES) && \
116          pppoe_proto(skb) == htons(PPP_IPV6) && \
117          brnf_filter_pppoe_tagged)
118
119 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
120 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
121
122 struct brnf_frag_data {
123         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
124         u8 encap_size;
125         u8 size;
126         u16 vlan_tci;
127         __be16 vlan_proto;
128 };
129
130 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
131
132 static void nf_bridge_info_free(struct sk_buff *skb)
133 {
134         if (skb->nf_bridge) {
135                 nf_bridge_put(skb->nf_bridge);
136                 skb->nf_bridge = NULL;
137         }
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_unshare(struct sk_buff *skb)
149 {
150         struct nf_bridge_info *nf_bridge = skb->nf_bridge;
151
152         if (atomic_read(&nf_bridge->use) > 1) {
153                 struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
154
155                 if (tmp) {
156                         memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
157                         atomic_set(&tmp->use, 1);
158                 }
159                 nf_bridge_put(nf_bridge);
160                 nf_bridge = tmp;
161         }
162         return nf_bridge;
163 }
164
165 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
166 {
167         switch (skb->protocol) {
168         case __cpu_to_be16(ETH_P_8021Q):
169                 return VLAN_HLEN;
170         case __cpu_to_be16(ETH_P_PPP_SES):
171                 return PPPOE_SES_HLEN;
172         default:
173                 return 0;
174         }
175 }
176
177 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
178 {
179         unsigned int len = nf_bridge_encap_header_len(skb);
180
181         skb_pull(skb, len);
182         skb->network_header += len;
183 }
184
185 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
186 {
187         unsigned int len = nf_bridge_encap_header_len(skb);
188
189         skb_pull_rcsum(skb, len);
190         skb->network_header += len;
191 }
192
193 /* When handing a packet over to the IP layer
194  * check whether we have a skb that is in the
195  * expected format
196  */
197
198 static int br_validate_ipv4(struct net *net, struct sk_buff *skb)
199 {
200         const struct iphdr *iph;
201         u32 len;
202
203         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
204                 goto inhdr_error;
205
206         iph = ip_hdr(skb);
207
208         /* Basic sanity checks */
209         if (iph->ihl < 5 || iph->version != 4)
210                 goto inhdr_error;
211
212         if (!pskb_may_pull(skb, iph->ihl*4))
213                 goto inhdr_error;
214
215         iph = ip_hdr(skb);
216         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
217                 goto inhdr_error;
218
219         len = ntohs(iph->tot_len);
220         if (skb->len < len) {
221                 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
222                 goto drop;
223         } else if (len < (iph->ihl*4))
224                 goto inhdr_error;
225
226         if (pskb_trim_rcsum(skb, len)) {
227                 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
228                 goto drop;
229         }
230
231         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
232         /* We should really parse IP options here but until
233          * somebody who actually uses IP options complains to
234          * us we'll just silently ignore the options because
235          * we're lazy!
236          */
237         return 0;
238
239 inhdr_error:
240         __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
241 drop:
242         return -1;
243 }
244
245 void nf_bridge_update_protocol(struct sk_buff *skb)
246 {
247         switch (skb->nf_bridge->orig_proto) {
248         case BRNF_PROTO_8021Q:
249                 skb->protocol = htons(ETH_P_8021Q);
250                 break;
251         case BRNF_PROTO_PPPOE:
252                 skb->protocol = htons(ETH_P_PPP_SES);
253                 break;
254         case BRNF_PROTO_UNCHANGED:
255                 break;
256         }
257 }
258
259 /* Obtain the correct destination MAC address, while preserving the original
260  * source MAC address. If we already know this address, we just copy it. If we
261  * don't, we use the neighbour framework to find out. In both cases, we make
262  * sure that br_handle_frame_finish() is called afterwards.
263  */
264 int br_nf_pre_routing_finish_bridge(struct net *net, struct sock *sk, struct sk_buff *skb)
265 {
266         struct neighbour *neigh;
267         struct dst_entry *dst;
268
269         skb->dev = bridge_parent(skb->dev);
270         if (!skb->dev)
271                 goto free_skb;
272         dst = skb_dst(skb);
273         neigh = dst_neigh_lookup_skb(dst, skb);
274         if (neigh) {
275                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
276                 int ret;
277
278                 if (neigh->hh.hh_len) {
279                         neigh_hh_bridge(&neigh->hh, skb);
280                         skb->dev = nf_bridge->physindev;
281                         ret = br_handle_frame_finish(net, sk, skb);
282                 } else {
283                         /* the neighbour function below overwrites the complete
284                          * MAC header, so we save the Ethernet source address and
285                          * protocol number.
286                          */
287                         skb_copy_from_linear_data_offset(skb,
288                                                          -(ETH_HLEN-ETH_ALEN),
289                                                          nf_bridge->neigh_header,
290                                                          ETH_HLEN-ETH_ALEN);
291                         /* tell br_dev_xmit to continue with forwarding */
292                         nf_bridge->bridged_dnat = 1;
293                         /* FIXME Need to refragment */
294                         ret = neigh->output(neigh, skb);
295                 }
296                 neigh_release(neigh);
297                 return ret;
298         }
299 free_skb:
300         kfree_skb(skb);
301         return 0;
302 }
303
304 static inline bool
305 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
306                              const struct nf_bridge_info *nf_bridge)
307 {
308         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
309 }
310
311 /* This requires some explaining. If DNAT has taken place,
312  * we will need to fix up the destination Ethernet address.
313  * This is also true when SNAT takes place (for the reply direction).
314  *
315  * There are two cases to consider:
316  * 1. The packet was DNAT'ed to a device in the same bridge
317  *    port group as it was received on. We can still bridge
318  *    the packet.
319  * 2. The packet was DNAT'ed to a different device, either
320  *    a non-bridged device or another bridge port group.
321  *    The packet will need to be routed.
322  *
323  * The correct way of distinguishing between these two cases is to
324  * call ip_route_input() and to look at skb->dst->dev, which is
325  * changed to the destination device if ip_route_input() succeeds.
326  *
327  * Let's first consider the case that ip_route_input() succeeds:
328  *
329  * If the output device equals the logical bridge device the packet
330  * came in on, we can consider this bridging. The corresponding MAC
331  * address will be obtained in br_nf_pre_routing_finish_bridge.
332  * Otherwise, the packet is considered to be routed and we just
333  * change the destination MAC address so that the packet will
334  * later be passed up to the IP stack to be routed. For a redirected
335  * packet, ip_route_input() will give back the localhost as output device,
336  * which differs from the bridge device.
337  *
338  * Let's now consider the case that ip_route_input() fails:
339  *
340  * This can be because the destination address is martian, in which case
341  * the packet will be dropped.
342  * If IP forwarding is disabled, ip_route_input() will fail, while
343  * ip_route_output_key() can return success. The source
344  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
345  * thinks we're handling a locally generated packet and won't care
346  * if IP forwarding is enabled. If the output device equals the logical bridge
347  * device, we proceed as if ip_route_input() succeeded. If it differs from the
348  * logical bridge port or if ip_route_output_key() fails we drop the packet.
349  */
350 static int br_nf_pre_routing_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
351 {
352         struct net_device *dev = skb->dev;
353         struct iphdr *iph = ip_hdr(skb);
354         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
355         struct rtable *rt;
356         int err;
357
358         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
359
360         if (nf_bridge->pkt_otherhost) {
361                 skb->pkt_type = PACKET_OTHERHOST;
362                 nf_bridge->pkt_otherhost = false;
363         }
364         nf_bridge->in_prerouting = 0;
365         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
366                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
367                         struct in_device *in_dev = __in_dev_get_rcu(dev);
368
369                         /* If err equals -EHOSTUNREACH the error is due to a
370                          * martian destination or due to the fact that
371                          * forwarding is disabled. For most martian packets,
372                          * ip_route_output_key() will fail. It won't fail for 2 types of
373                          * martian destinations: loopback destinations and destination
374                          * 0.0.0.0. In both cases the packet will be dropped because the
375                          * destination is the loopback device and not the bridge. */
376                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
377                                 goto free_skb;
378
379                         rt = ip_route_output(net, iph->daddr, 0,
380                                              RT_TOS(iph->tos), 0);
381                         if (!IS_ERR(rt)) {
382                                 /* - Bridged-and-DNAT'ed traffic doesn't
383                                  *   require ip_forwarding. */
384                                 if (rt->dst.dev == dev) {
385                                         skb_dst_set(skb, &rt->dst);
386                                         goto bridged_dnat;
387                                 }
388                                 ip_rt_put(rt);
389                         }
390 free_skb:
391                         kfree_skb(skb);
392                         return 0;
393                 } else {
394                         if (skb_dst(skb)->dev == dev) {
395 bridged_dnat:
396                                 skb->dev = nf_bridge->physindev;
397                                 nf_bridge_update_protocol(skb);
398                                 nf_bridge_push_encap_header(skb);
399                                 br_nf_hook_thresh(NF_BR_PRE_ROUTING,
400                                                   net, sk, skb, skb->dev,
401                                                   NULL,
402                                                   br_nf_pre_routing_finish_bridge);
403                                 return 0;
404                         }
405                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
406                         skb->pkt_type = PACKET_HOST;
407                 }
408         } else {
409                 rt = bridge_parent_rtable(nf_bridge->physindev);
410                 if (!rt) {
411                         kfree_skb(skb);
412                         return 0;
413                 }
414                 skb_dst_set_noref(skb, &rt->dst);
415         }
416
417         skb->dev = nf_bridge->physindev;
418         nf_bridge_update_protocol(skb);
419         nf_bridge_push_encap_header(skb);
420         br_nf_hook_thresh(NF_BR_PRE_ROUTING, net, sk, skb, skb->dev, NULL,
421                           br_handle_frame_finish);
422         return 0;
423 }
424
425 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
426 {
427         struct net_device *vlan, *br;
428
429         br = bridge_parent(dev);
430         if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
431                 return br;
432
433         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
434                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
435
436         return vlan ? vlan : br;
437 }
438
439 /* Some common code for IPv4/IPv6 */
440 struct net_device *setup_pre_routing(struct sk_buff *skb)
441 {
442         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
443
444         if (skb->pkt_type == PACKET_OTHERHOST) {
445                 skb->pkt_type = PACKET_HOST;
446                 nf_bridge->pkt_otherhost = true;
447         }
448
449         nf_bridge->in_prerouting = 1;
450         nf_bridge->physindev = skb->dev;
451         skb->dev = brnf_get_logical_dev(skb, skb->dev);
452
453         if (skb->protocol == htons(ETH_P_8021Q))
454                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
455         else if (skb->protocol == htons(ETH_P_PPP_SES))
456                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
457
458         /* Must drop socket now because of tproxy. */
459         skb_orphan(skb);
460         return skb->dev;
461 }
462
463 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
464  * Replicate the checks that IPv4 does on packet reception.
465  * Set skb->dev to the bridge device (i.e. parent of the
466  * receiving device) to make netfilter happy, the REDIRECT
467  * target in particular.  Save the original destination IP
468  * address to be able to detect DNAT afterwards. */
469 static unsigned int br_nf_pre_routing(void *priv,
470                                       struct sk_buff *skb,
471                                       const struct nf_hook_state *state)
472 {
473         struct nf_bridge_info *nf_bridge;
474         struct net_bridge_port *p;
475         struct net_bridge *br;
476         __u32 len = nf_bridge_encap_header_len(skb);
477
478         if (unlikely(!pskb_may_pull(skb, len)))
479                 return NF_DROP;
480
481         p = br_port_get_rcu(state->in);
482         if (p == NULL)
483                 return NF_DROP;
484         br = p->br;
485
486         if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
487                 if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
488                         return NF_ACCEPT;
489
490                 nf_bridge_pull_encap_header_rcsum(skb);
491                 return br_nf_pre_routing_ipv6(priv, skb, state);
492         }
493
494         if (!brnf_call_iptables && !br->nf_call_iptables)
495                 return NF_ACCEPT;
496
497         if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
498                 return NF_ACCEPT;
499
500         nf_bridge_pull_encap_header_rcsum(skb);
501
502         if (br_validate_ipv4(state->net, skb))
503                 return NF_DROP;
504
505         nf_bridge_put(skb->nf_bridge);
506         if (!nf_bridge_alloc(skb))
507                 return NF_DROP;
508         if (!setup_pre_routing(skb))
509                 return NF_DROP;
510
511         nf_bridge = nf_bridge_info_get(skb);
512         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
513
514         skb->protocol = htons(ETH_P_IP);
515
516         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->net, state->sk, skb,
517                 skb->dev, NULL,
518                 br_nf_pre_routing_finish);
519
520         return NF_STOLEN;
521 }
522
523
524 /* PF_BRIDGE/LOCAL_IN ************************************************/
525 /* The packet is locally destined, which requires a real
526  * dst_entry, so detach the fake one.  On the way up, the
527  * packet would pass through PRE_ROUTING again (which already
528  * took place when the packet entered the bridge), but we
529  * register an IPv4 PRE_ROUTING 'sabotage' hook that will
530  * prevent this from happening. */
531 static unsigned int br_nf_local_in(void *priv,
532                                    struct sk_buff *skb,
533                                    const struct nf_hook_state *state)
534 {
535         br_drop_fake_rtable(skb);
536         return NF_ACCEPT;
537 }
538
539 /* PF_BRIDGE/FORWARD *************************************************/
540 static int br_nf_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
541 {
542         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
543         struct net_device *in;
544
545         if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
546
547                 if (skb->protocol == htons(ETH_P_IP))
548                         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
549
550                 if (skb->protocol == htons(ETH_P_IPV6))
551                         nf_bridge->frag_max_size = IP6CB(skb)->frag_max_size;
552
553                 in = nf_bridge->physindev;
554                 if (nf_bridge->pkt_otherhost) {
555                         skb->pkt_type = PACKET_OTHERHOST;
556                         nf_bridge->pkt_otherhost = false;
557                 }
558                 nf_bridge_update_protocol(skb);
559         } else {
560                 in = *((struct net_device **)(skb->cb));
561         }
562         nf_bridge_push_encap_header(skb);
563
564         br_nf_hook_thresh(NF_BR_FORWARD, net, sk, skb, in, skb->dev,
565                           br_forward_finish);
566         return 0;
567 }
568
569
570 /* This is the 'purely bridged' case.  For IP, we pass the packet to
571  * netfilter with indev and outdev set to the bridge device,
572  * but we are still able to filter on the 'real' indev/outdev
573  * because of the physdev module. For ARP, indev and outdev are the
574  * bridge ports. */
575 static unsigned int br_nf_forward_ip(void *priv,
576                                      struct sk_buff *skb,
577                                      const struct nf_hook_state *state)
578 {
579         struct nf_bridge_info *nf_bridge;
580         struct net_device *parent;
581         u_int8_t pf;
582
583         if (!skb->nf_bridge)
584                 return NF_ACCEPT;
585
586         /* Need exclusive nf_bridge_info since we might have multiple
587          * different physoutdevs. */
588         if (!nf_bridge_unshare(skb))
589                 return NF_DROP;
590
591         nf_bridge = nf_bridge_info_get(skb);
592         if (!nf_bridge)
593                 return NF_DROP;
594
595         parent = bridge_parent(state->out);
596         if (!parent)
597                 return NF_DROP;
598
599         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
600                 pf = NFPROTO_IPV4;
601         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
602                 pf = NFPROTO_IPV6;
603         else
604                 return NF_ACCEPT;
605
606         nf_bridge_pull_encap_header(skb);
607
608         if (skb->pkt_type == PACKET_OTHERHOST) {
609                 skb->pkt_type = PACKET_HOST;
610                 nf_bridge->pkt_otherhost = true;
611         }
612
613         if (pf == NFPROTO_IPV4) {
614                 if (br_validate_ipv4(state->net, skb))
615                         return NF_DROP;
616                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
617         }
618
619         if (pf == NFPROTO_IPV6) {
620                 if (br_validate_ipv6(state->net, skb))
621                         return NF_DROP;
622                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
623         }
624
625         nf_bridge->physoutdev = skb->dev;
626         if (pf == NFPROTO_IPV4)
627                 skb->protocol = htons(ETH_P_IP);
628         else
629                 skb->protocol = htons(ETH_P_IPV6);
630
631         NF_HOOK(pf, NF_INET_FORWARD, state->net, NULL, skb,
632                 brnf_get_logical_dev(skb, state->in),
633                 parent, br_nf_forward_finish);
634
635         return NF_STOLEN;
636 }
637
638 static unsigned int br_nf_forward_arp(void *priv,
639                                       struct sk_buff *skb,
640                                       const struct nf_hook_state *state)
641 {
642         struct net_bridge_port *p;
643         struct net_bridge *br;
644         struct net_device **d = (struct net_device **)(skb->cb);
645
646         p = br_port_get_rcu(state->out);
647         if (p == NULL)
648                 return NF_ACCEPT;
649         br = p->br;
650
651         if (!brnf_call_arptables && !br->nf_call_arptables)
652                 return NF_ACCEPT;
653
654         if (!IS_ARP(skb)) {
655                 if (!IS_VLAN_ARP(skb))
656                         return NF_ACCEPT;
657                 nf_bridge_pull_encap_header(skb);
658         }
659
660         if (arp_hdr(skb)->ar_pln != 4) {
661                 if (IS_VLAN_ARP(skb))
662                         nf_bridge_push_encap_header(skb);
663                 return NF_ACCEPT;
664         }
665         *d = state->in;
666         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->net, state->sk, skb,
667                 state->in, state->out, br_nf_forward_finish);
668
669         return NF_STOLEN;
670 }
671
672 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
673 {
674         struct brnf_frag_data *data;
675         int err;
676
677         data = this_cpu_ptr(&brnf_frag_data_storage);
678         err = skb_cow_head(skb, data->size);
679
680         if (err) {
681                 kfree_skb(skb);
682                 return 0;
683         }
684
685         if (data->vlan_tci) {
686                 skb->vlan_tci = data->vlan_tci;
687                 skb->vlan_proto = data->vlan_proto;
688         }
689
690         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
691         __skb_push(skb, data->encap_size);
692
693         nf_bridge_info_free(skb);
694         return br_dev_queue_push_xmit(net, sk, skb);
695 }
696
697 static int
698 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
699                   int (*output)(struct net *, struct sock *, struct sk_buff *))
700 {
701         unsigned int mtu = ip_skb_dst_mtu(sk, skb);
702         struct iphdr *iph = ip_hdr(skb);
703
704         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
705                      (IPCB(skb)->frag_max_size &&
706                       IPCB(skb)->frag_max_size > mtu))) {
707                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
708                 kfree_skb(skb);
709                 return -EMSGSIZE;
710         }
711
712         return ip_do_fragment(net, sk, skb, output);
713 }
714
715 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
716 {
717         if (skb->nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
718                 return PPPOE_SES_HLEN;
719         return 0;
720 }
721
722 static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
723 {
724         struct nf_bridge_info *nf_bridge;
725         unsigned int mtu_reserved;
726
727         mtu_reserved = nf_bridge_mtu_reduction(skb);
728
729         if (skb_is_gso(skb) || skb->len + mtu_reserved <= skb->dev->mtu) {
730                 nf_bridge_info_free(skb);
731                 return br_dev_queue_push_xmit(net, sk, skb);
732         }
733
734         nf_bridge = nf_bridge_info_get(skb);
735
736         /* This is wrong! We should preserve the original fragment
737          * boundaries by preserving frag_list rather than refragmenting.
738          */
739         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) &&
740             skb->protocol == htons(ETH_P_IP)) {
741                 struct brnf_frag_data *data;
742
743                 if (br_validate_ipv4(net, skb))
744                         goto drop;
745
746                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
747
748                 nf_bridge_update_protocol(skb);
749
750                 data = this_cpu_ptr(&brnf_frag_data_storage);
751
752                 data->vlan_tci = skb->vlan_tci;
753                 data->vlan_proto = skb->vlan_proto;
754                 data->encap_size = nf_bridge_encap_header_len(skb);
755                 data->size = ETH_HLEN + data->encap_size;
756
757                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
758                                                  data->size);
759
760                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
761         }
762         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
763             skb->protocol == htons(ETH_P_IPV6)) {
764                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
765                 struct brnf_frag_data *data;
766
767                 if (br_validate_ipv6(net, skb))
768                         goto drop;
769
770                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
771
772                 nf_bridge_update_protocol(skb);
773
774                 data = this_cpu_ptr(&brnf_frag_data_storage);
775                 data->encap_size = nf_bridge_encap_header_len(skb);
776                 data->size = ETH_HLEN + data->encap_size;
777
778                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
779                                                  data->size);
780
781                 if (v6ops)
782                         return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
783
784                 kfree_skb(skb);
785                 return -EMSGSIZE;
786         }
787         nf_bridge_info_free(skb);
788         return br_dev_queue_push_xmit(net, sk, skb);
789  drop:
790         kfree_skb(skb);
791         return 0;
792 }
793
794 /* PF_BRIDGE/POST_ROUTING ********************************************/
795 static unsigned int br_nf_post_routing(void *priv,
796                                        struct sk_buff *skb,
797                                        const struct nf_hook_state *state)
798 {
799         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
800         struct net_device *realoutdev = bridge_parent(skb->dev);
801         u_int8_t pf;
802
803         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
804          * on a bridge, but was delivered locally and is now being routed:
805          *
806          * POST_ROUTING was already invoked from the ip stack.
807          */
808         if (!nf_bridge || !nf_bridge->physoutdev)
809                 return NF_ACCEPT;
810
811         if (!realoutdev)
812                 return NF_DROP;
813
814         if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
815                 pf = NFPROTO_IPV4;
816         else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
817                 pf = NFPROTO_IPV6;
818         else
819                 return NF_ACCEPT;
820
821         /* We assume any code from br_dev_queue_push_xmit onwards doesn't care
822          * about the value of skb->pkt_type. */
823         if (skb->pkt_type == PACKET_OTHERHOST) {
824                 skb->pkt_type = PACKET_HOST;
825                 nf_bridge->pkt_otherhost = true;
826         }
827
828         nf_bridge_pull_encap_header(skb);
829         if (pf == NFPROTO_IPV4)
830                 skb->protocol = htons(ETH_P_IP);
831         else
832                 skb->protocol = htons(ETH_P_IPV6);
833
834         NF_HOOK(pf, NF_INET_POST_ROUTING, state->net, state->sk, skb,
835                 NULL, realoutdev,
836                 br_nf_dev_queue_xmit);
837
838         return NF_STOLEN;
839 }
840
841 /* IP/SABOTAGE *****************************************************/
842 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
843  * for the second time. */
844 static unsigned int ip_sabotage_in(void *priv,
845                                    struct sk_buff *skb,
846                                    const struct nf_hook_state *state)
847 {
848         if (skb->nf_bridge && !skb->nf_bridge->in_prerouting) {
849                 state->okfn(state->net, state->sk, skb);
850                 return NF_STOLEN;
851         }
852
853         return NF_ACCEPT;
854 }
855
856 /* This is called when br_netfilter has called into iptables/netfilter,
857  * and DNAT has taken place on a bridge-forwarded packet.
858  *
859  * neigh->output has created a new MAC header, with local br0 MAC
860  * as saddr.
861  *
862  * This restores the original MAC saddr of the bridged packet
863  * before invoking bridge forward logic to transmit the packet.
864  */
865 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
866 {
867         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
868
869         skb_pull(skb, ETH_HLEN);
870         nf_bridge->bridged_dnat = 0;
871
872         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
873
874         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
875                                        nf_bridge->neigh_header,
876                                        ETH_HLEN - ETH_ALEN);
877         skb->dev = nf_bridge->physindev;
878
879         nf_bridge->physoutdev = NULL;
880         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
881 }
882
883 static int br_nf_dev_xmit(struct sk_buff *skb)
884 {
885         if (skb->nf_bridge && skb->nf_bridge->bridged_dnat) {
886                 br_nf_pre_routing_finish_bridge_slow(skb);
887                 return 1;
888         }
889         return 0;
890 }
891
892 static const struct nf_br_ops br_ops = {
893         .br_dev_xmit_hook =     br_nf_dev_xmit,
894 };
895
896 void br_netfilter_enable(void)
897 {
898 }
899 EXPORT_SYMBOL_GPL(br_netfilter_enable);
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                 .pf = NFPROTO_BRIDGE,
907                 .hooknum = NF_BR_PRE_ROUTING,
908                 .priority = NF_BR_PRI_BRNF,
909         },
910         {
911                 .hook = br_nf_local_in,
912                 .pf = NFPROTO_BRIDGE,
913                 .hooknum = NF_BR_LOCAL_IN,
914                 .priority = NF_BR_PRI_BRNF,
915         },
916         {
917                 .hook = br_nf_forward_ip,
918                 .pf = NFPROTO_BRIDGE,
919                 .hooknum = NF_BR_FORWARD,
920                 .priority = NF_BR_PRI_BRNF - 1,
921         },
922         {
923                 .hook = br_nf_forward_arp,
924                 .pf = NFPROTO_BRIDGE,
925                 .hooknum = NF_BR_FORWARD,
926                 .priority = NF_BR_PRI_BRNF,
927         },
928         {
929                 .hook = br_nf_post_routing,
930                 .pf = NFPROTO_BRIDGE,
931                 .hooknum = NF_BR_POST_ROUTING,
932                 .priority = NF_BR_PRI_LAST,
933         },
934         {
935                 .hook = ip_sabotage_in,
936                 .pf = NFPROTO_IPV4,
937                 .hooknum = NF_INET_PRE_ROUTING,
938                 .priority = NF_IP_PRI_FIRST,
939         },
940         {
941                 .hook = ip_sabotage_in,
942                 .pf = NFPROTO_IPV6,
943                 .hooknum = NF_INET_PRE_ROUTING,
944                 .priority = NF_IP6_PRI_FIRST,
945         },
946 };
947
948 static int brnf_device_event(struct notifier_block *unused, unsigned long event,
949                              void *ptr)
950 {
951         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
952         struct brnf_net *brnet;
953         struct net *net;
954         int ret;
955
956         if (event != NETDEV_REGISTER || !(dev->priv_flags & IFF_EBRIDGE))
957                 return NOTIFY_DONE;
958
959         ASSERT_RTNL();
960
961         net = dev_net(dev);
962         brnet = net_generic(net, brnf_net_id);
963         if (brnet->enabled)
964                 return NOTIFY_OK;
965
966         ret = nf_register_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
967         if (ret)
968                 return NOTIFY_BAD;
969
970         brnet->enabled = true;
971         return NOTIFY_OK;
972 }
973
974 static void __net_exit brnf_exit_net(struct net *net)
975 {
976         struct brnf_net *brnet = net_generic(net, brnf_net_id);
977
978         if (!brnet->enabled)
979                 return;
980
981         nf_unregister_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
982         brnet->enabled = false;
983 }
984
985 static struct pernet_operations brnf_net_ops __read_mostly = {
986         .exit = brnf_exit_net,
987         .id   = &brnf_net_id,
988         .size = sizeof(struct brnf_net),
989 };
990
991 static struct notifier_block brnf_notifier __read_mostly = {
992         .notifier_call = brnf_device_event,
993 };
994
995 /* recursively invokes nf_hook_slow (again), skipping already-called
996  * hooks (< NF_BR_PRI_BRNF).
997  *
998  * Called with rcu read lock held.
999  */
1000 int br_nf_hook_thresh(unsigned int hook, struct net *net,
1001                       struct sock *sk, struct sk_buff *skb,
1002                       struct net_device *indev,
1003                       struct net_device *outdev,
1004                       int (*okfn)(struct net *, struct sock *,
1005                                   struct sk_buff *))
1006 {
1007         struct nf_hook_entry *elem;
1008         struct nf_hook_state state;
1009         int ret;
1010
1011         for (elem = rcu_dereference(net->nf.hooks[NFPROTO_BRIDGE][hook]);
1012              elem && nf_hook_entry_priority(elem) <= NF_BR_PRI_BRNF;
1013              elem = rcu_dereference(elem->next))
1014                 ;
1015
1016         if (!elem)
1017                 return okfn(net, sk, skb);
1018
1019         /* We may already have this, but read-locks nest anyway */
1020         rcu_read_lock();
1021         nf_hook_state_init(&state, hook, NFPROTO_BRIDGE, indev, outdev,
1022                            sk, net, okfn);
1023
1024         ret = nf_hook_slow(skb, &state, elem);
1025         rcu_read_unlock();
1026         if (ret == 1)
1027                 ret = okfn(net, sk, skb);
1028
1029         return ret;
1030 }
1031
1032 #ifdef CONFIG_SYSCTL
1033 static
1034 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1035                             void __user *buffer, size_t *lenp, loff_t *ppos)
1036 {
1037         int ret;
1038
1039         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1040
1041         if (write && *(int *)(ctl->data))
1042                 *(int *)(ctl->data) = 1;
1043         return ret;
1044 }
1045
1046 static struct ctl_table brnf_table[] = {
1047         {
1048                 .procname       = "bridge-nf-call-arptables",
1049                 .data           = &brnf_call_arptables,
1050                 .maxlen         = sizeof(int),
1051                 .mode           = 0644,
1052                 .proc_handler   = brnf_sysctl_call_tables,
1053         },
1054         {
1055                 .procname       = "bridge-nf-call-iptables",
1056                 .data           = &brnf_call_iptables,
1057                 .maxlen         = sizeof(int),
1058                 .mode           = 0644,
1059                 .proc_handler   = brnf_sysctl_call_tables,
1060         },
1061         {
1062                 .procname       = "bridge-nf-call-ip6tables",
1063                 .data           = &brnf_call_ip6tables,
1064                 .maxlen         = sizeof(int),
1065                 .mode           = 0644,
1066                 .proc_handler   = brnf_sysctl_call_tables,
1067         },
1068         {
1069                 .procname       = "bridge-nf-filter-vlan-tagged",
1070                 .data           = &brnf_filter_vlan_tagged,
1071                 .maxlen         = sizeof(int),
1072                 .mode           = 0644,
1073                 .proc_handler   = brnf_sysctl_call_tables,
1074         },
1075         {
1076                 .procname       = "bridge-nf-filter-pppoe-tagged",
1077                 .data           = &brnf_filter_pppoe_tagged,
1078                 .maxlen         = sizeof(int),
1079                 .mode           = 0644,
1080                 .proc_handler   = brnf_sysctl_call_tables,
1081         },
1082         {
1083                 .procname       = "bridge-nf-pass-vlan-input-dev",
1084                 .data           = &brnf_pass_vlan_indev,
1085                 .maxlen         = sizeof(int),
1086                 .mode           = 0644,
1087                 .proc_handler   = brnf_sysctl_call_tables,
1088         },
1089         { }
1090 };
1091 #endif
1092
1093 static int __init br_netfilter_init(void)
1094 {
1095         int ret;
1096
1097         ret = register_pernet_subsys(&brnf_net_ops);
1098         if (ret < 0)
1099                 return ret;
1100
1101         ret = register_netdevice_notifier(&brnf_notifier);
1102         if (ret < 0) {
1103                 unregister_pernet_subsys(&brnf_net_ops);
1104                 return ret;
1105         }
1106
1107 #ifdef CONFIG_SYSCTL
1108         brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1109         if (brnf_sysctl_header == NULL) {
1110                 printk(KERN_WARNING
1111                        "br_netfilter: can't register to sysctl.\n");
1112                 unregister_netdevice_notifier(&brnf_notifier);
1113                 unregister_pernet_subsys(&brnf_net_ops);
1114                 return -ENOMEM;
1115         }
1116 #endif
1117         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1118         printk(KERN_NOTICE "Bridge firewalling registered\n");
1119         return 0;
1120 }
1121
1122 static void __exit br_netfilter_fini(void)
1123 {
1124         RCU_INIT_POINTER(nf_br_ops, NULL);
1125         unregister_netdevice_notifier(&brnf_notifier);
1126         unregister_pernet_subsys(&brnf_net_ops);
1127 #ifdef CONFIG_SYSCTL
1128         unregister_net_sysctl_table(brnf_sysctl_header);
1129 #endif
1130 }
1131
1132 module_init(br_netfilter_init);
1133 module_exit(br_netfilter_fini);
1134
1135 MODULE_LICENSE("GPL");
1136 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1137 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1138 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");