netfilter: bridge: replace physindev with physinif in nf_bridge_info
[platform/kernel/linux-rpi.git] / net / bridge / br_netfilter_hooks.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *      Handle firewalling
4  *      Linux ethernet bridge
5  *
6  *      Authors:
7  *      Lennert Buytenhek               <buytenh@gnu.org>
8  *      Bart De Schuymer                <bdschuym@pandora.be>
9  *
10  *      Lennert dedicates this file to Kerstin Wurdinger.
11  */
12
13 #include <linux/module.h>
14 #include <linux/kernel.h>
15 #include <linux/slab.h>
16 #include <linux/ip.h>
17 #include <linux/netdevice.h>
18 #include <linux/skbuff.h>
19 #include <linux/if_arp.h>
20 #include <linux/if_ether.h>
21 #include <linux/if_vlan.h>
22 #include <linux/if_pppox.h>
23 #include <linux/ppp_defs.h>
24 #include <linux/netfilter_bridge.h>
25 #include <uapi/linux/netfilter_bridge.h>
26 #include <linux/netfilter_ipv4.h>
27 #include <linux/netfilter_ipv6.h>
28 #include <linux/netfilter_arp.h>
29 #include <linux/in_route.h>
30 #include <linux/rculist.h>
31 #include <linux/inetdevice.h>
32
33 #include <net/ip.h>
34 #include <net/ipv6.h>
35 #include <net/addrconf.h>
36 #include <net/route.h>
37 #include <net/netfilter/br_netfilter.h>
38 #include <net/netns/generic.h>
39
40 #include <linux/uaccess.h>
41 #include "br_private.h"
42 #ifdef CONFIG_SYSCTL
43 #include <linux/sysctl.h>
44 #endif
45
46 static unsigned int brnf_net_id __read_mostly;
47
48 struct brnf_net {
49         bool enabled;
50
51 #ifdef CONFIG_SYSCTL
52         struct ctl_table_header *ctl_hdr;
53 #endif
54
55         /* default value is 1 */
56         int call_iptables;
57         int call_ip6tables;
58         int call_arptables;
59
60         /* default value is 0 */
61         int filter_vlan_tagged;
62         int filter_pppoe_tagged;
63         int pass_vlan_indev;
64 };
65
66 #define IS_IP(skb) \
67         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
68
69 #define IS_IPV6(skb) \
70         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
71
72 #define IS_ARP(skb) \
73         (!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
74
75 static inline __be16 vlan_proto(const struct sk_buff *skb)
76 {
77         if (skb_vlan_tag_present(skb))
78                 return skb->protocol;
79         else if (skb->protocol == htons(ETH_P_8021Q))
80                 return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
81         else
82                 return 0;
83 }
84
85 static inline bool is_vlan_ip(const struct sk_buff *skb, const struct net *net)
86 {
87         struct brnf_net *brnet = net_generic(net, brnf_net_id);
88
89         return vlan_proto(skb) == htons(ETH_P_IP) && brnet->filter_vlan_tagged;
90 }
91
92 static inline bool is_vlan_ipv6(const struct sk_buff *skb,
93                                 const struct net *net)
94 {
95         struct brnf_net *brnet = net_generic(net, brnf_net_id);
96
97         return vlan_proto(skb) == htons(ETH_P_IPV6) &&
98                brnet->filter_vlan_tagged;
99 }
100
101 static inline bool is_vlan_arp(const struct sk_buff *skb, const struct net *net)
102 {
103         struct brnf_net *brnet = net_generic(net, brnf_net_id);
104
105         return vlan_proto(skb) == htons(ETH_P_ARP) && brnet->filter_vlan_tagged;
106 }
107
108 static inline __be16 pppoe_proto(const struct sk_buff *skb)
109 {
110         return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
111                             sizeof(struct pppoe_hdr)));
112 }
113
114 static inline bool is_pppoe_ip(const struct sk_buff *skb, const struct net *net)
115 {
116         struct brnf_net *brnet = net_generic(net, brnf_net_id);
117
118         return skb->protocol == htons(ETH_P_PPP_SES) &&
119                pppoe_proto(skb) == htons(PPP_IP) && brnet->filter_pppoe_tagged;
120 }
121
122 static inline bool is_pppoe_ipv6(const struct sk_buff *skb,
123                                  const struct net *net)
124 {
125         struct brnf_net *brnet = net_generic(net, brnf_net_id);
126
127         return skb->protocol == htons(ETH_P_PPP_SES) &&
128                pppoe_proto(skb) == htons(PPP_IPV6) &&
129                brnet->filter_pppoe_tagged;
130 }
131
132 /* largest possible L2 header, see br_nf_dev_queue_xmit() */
133 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
134
135 struct brnf_frag_data {
136         char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
137         u8 encap_size;
138         u8 size;
139         u16 vlan_tci;
140         __be16 vlan_proto;
141 };
142
143 static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
144
145 static void nf_bridge_info_free(struct sk_buff *skb)
146 {
147         skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
148 }
149
150 static inline struct net_device *bridge_parent(const struct net_device *dev)
151 {
152         struct net_bridge_port *port;
153
154         port = br_port_get_rcu(dev);
155         return port ? port->br->dev : NULL;
156 }
157
158 static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
159 {
160         return skb_ext_add(skb, SKB_EXT_BRIDGE_NF);
161 }
162
163 unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
164 {
165         switch (skb->protocol) {
166         case __cpu_to_be16(ETH_P_8021Q):
167                 return VLAN_HLEN;
168         case __cpu_to_be16(ETH_P_PPP_SES):
169                 return PPPOE_SES_HLEN;
170         default:
171                 return 0;
172         }
173 }
174
175 static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
176 {
177         unsigned int len = nf_bridge_encap_header_len(skb);
178
179         skb_pull(skb, len);
180         skb->network_header += len;
181 }
182
183 static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
184 {
185         unsigned int len = nf_bridge_encap_header_len(skb);
186
187         skb_pull_rcsum(skb, len);
188         skb->network_header += len;
189 }
190
191 /* When handing a packet over to the IP layer
192  * check whether we have a skb that is in the
193  * expected format
194  */
195
196 static int br_validate_ipv4(struct net *net, struct sk_buff *skb)
197 {
198         const struct iphdr *iph;
199         u32 len;
200
201         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
202                 goto inhdr_error;
203
204         iph = ip_hdr(skb);
205
206         /* Basic sanity checks */
207         if (iph->ihl < 5 || iph->version != 4)
208                 goto inhdr_error;
209
210         if (!pskb_may_pull(skb, iph->ihl*4))
211                 goto inhdr_error;
212
213         iph = ip_hdr(skb);
214         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
215                 goto csum_error;
216
217         len = skb_ip_totlen(skb);
218         if (skb->len < len) {
219                 __IP_INC_STATS(net, IPSTATS_MIB_INTRUNCATEDPKTS);
220                 goto drop;
221         } else if (len < (iph->ihl*4))
222                 goto inhdr_error;
223
224         if (pskb_trim_rcsum(skb, len)) {
225                 __IP_INC_STATS(net, IPSTATS_MIB_INDISCARDS);
226                 goto drop;
227         }
228
229         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
230         /* We should really parse IP options here but until
231          * somebody who actually uses IP options complains to
232          * us we'll just silently ignore the options because
233          * we're lazy!
234          */
235         return 0;
236
237 csum_error:
238         __IP_INC_STATS(net, IPSTATS_MIB_CSUMERRORS);
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         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
248
249         switch (nf_bridge->orig_proto) {
250         case BRNF_PROTO_8021Q:
251                 skb->protocol = htons(ETH_P_8021Q);
252                 break;
253         case BRNF_PROTO_PPPOE:
254                 skb->protocol = htons(ETH_P_PPP_SES);
255                 break;
256         case BRNF_PROTO_UNCHANGED:
257                 break;
258         }
259 }
260
261 /* Obtain the correct destination MAC address, while preserving the original
262  * source MAC address. If we already know this address, we just copy it. If we
263  * don't, we use the neighbour framework to find out. In both cases, we make
264  * sure that br_handle_frame_finish() is called afterwards.
265  */
266 int br_nf_pre_routing_finish_bridge(struct net *net, struct sock *sk, struct sk_buff *skb)
267 {
268         struct neighbour *neigh;
269         struct dst_entry *dst;
270
271         skb->dev = bridge_parent(skb->dev);
272         if (!skb->dev)
273                 goto free_skb;
274         dst = skb_dst(skb);
275         neigh = dst_neigh_lookup_skb(dst, skb);
276         if (neigh) {
277                 struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
278                 int ret;
279
280                 if ((READ_ONCE(neigh->nud_state) & NUD_CONNECTED) &&
281                     READ_ONCE(neigh->hh.hh_len)) {
282                         struct net_device *br_indev;
283
284                         br_indev = nf_bridge_get_physindev(skb, net);
285                         if (!br_indev) {
286                                 neigh_release(neigh);
287                                 goto free_skb;
288                         }
289
290                         neigh_hh_bridge(&neigh->hh, skb);
291                         skb->dev = br_indev;
292
293                         ret = br_handle_frame_finish(net, sk, skb);
294                 } else {
295                         /* the neighbour function below overwrites the complete
296                          * MAC header, so we save the Ethernet source address and
297                          * protocol number.
298                          */
299                         skb_copy_from_linear_data_offset(skb,
300                                                          -(ETH_HLEN-ETH_ALEN),
301                                                          nf_bridge->neigh_header,
302                                                          ETH_HLEN-ETH_ALEN);
303                         /* tell br_dev_xmit to continue with forwarding */
304                         nf_bridge->bridged_dnat = 1;
305                         /* FIXME Need to refragment */
306                         ret = READ_ONCE(neigh->output)(neigh, skb);
307                 }
308                 neigh_release(neigh);
309                 return ret;
310         }
311 free_skb:
312         kfree_skb(skb);
313         return 0;
314 }
315
316 static inline bool
317 br_nf_ipv4_daddr_was_changed(const struct sk_buff *skb,
318                              const struct nf_bridge_info *nf_bridge)
319 {
320         return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
321 }
322
323 /* This requires some explaining. If DNAT has taken place,
324  * we will need to fix up the destination Ethernet address.
325  * This is also true when SNAT takes place (for the reply direction).
326  *
327  * There are two cases to consider:
328  * 1. The packet was DNAT'ed to a device in the same bridge
329  *    port group as it was received on. We can still bridge
330  *    the packet.
331  * 2. The packet was DNAT'ed to a different device, either
332  *    a non-bridged device or another bridge port group.
333  *    The packet will need to be routed.
334  *
335  * The correct way of distinguishing between these two cases is to
336  * call ip_route_input() and to look at skb->dst->dev, which is
337  * changed to the destination device if ip_route_input() succeeds.
338  *
339  * Let's first consider the case that ip_route_input() succeeds:
340  *
341  * If the output device equals the logical bridge device the packet
342  * came in on, we can consider this bridging. The corresponding MAC
343  * address will be obtained in br_nf_pre_routing_finish_bridge.
344  * Otherwise, the packet is considered to be routed and we just
345  * change the destination MAC address so that the packet will
346  * later be passed up to the IP stack to be routed. For a redirected
347  * packet, ip_route_input() will give back the localhost as output device,
348  * which differs from the bridge device.
349  *
350  * Let's now consider the case that ip_route_input() fails:
351  *
352  * This can be because the destination address is martian, in which case
353  * the packet will be dropped.
354  * If IP forwarding is disabled, ip_route_input() will fail, while
355  * ip_route_output_key() can return success. The source
356  * address for ip_route_output_key() is set to zero, so ip_route_output_key()
357  * thinks we're handling a locally generated packet and won't care
358  * if IP forwarding is enabled. If the output device equals the logical bridge
359  * device, we proceed as if ip_route_input() succeeded. If it differs from the
360  * logical bridge port or if ip_route_output_key() fails we drop the packet.
361  */
362 static int br_nf_pre_routing_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
363 {
364         struct net_device *dev = skb->dev, *br_indev;
365         struct iphdr *iph = ip_hdr(skb);
366         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
367         struct rtable *rt;
368         int err;
369
370         br_indev = nf_bridge_get_physindev(skb, net);
371         if (!br_indev) {
372                 kfree_skb(skb);
373                 return 0;
374         }
375
376         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
377
378         if (nf_bridge->pkt_otherhost) {
379                 skb->pkt_type = PACKET_OTHERHOST;
380                 nf_bridge->pkt_otherhost = false;
381         }
382         nf_bridge->in_prerouting = 0;
383         if (br_nf_ipv4_daddr_was_changed(skb, nf_bridge)) {
384                 if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
385                         struct in_device *in_dev = __in_dev_get_rcu(dev);
386
387                         /* If err equals -EHOSTUNREACH the error is due to a
388                          * martian destination or due to the fact that
389                          * forwarding is disabled. For most martian packets,
390                          * ip_route_output_key() will fail. It won't fail for 2 types of
391                          * martian destinations: loopback destinations and destination
392                          * 0.0.0.0. In both cases the packet will be dropped because the
393                          * destination is the loopback device and not the bridge. */
394                         if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
395                                 goto free_skb;
396
397                         rt = ip_route_output(net, iph->daddr, 0,
398                                              RT_TOS(iph->tos), 0);
399                         if (!IS_ERR(rt)) {
400                                 /* - Bridged-and-DNAT'ed traffic doesn't
401                                  *   require ip_forwarding. */
402                                 if (rt->dst.dev == dev) {
403                                         skb_dst_drop(skb);
404                                         skb_dst_set(skb, &rt->dst);
405                                         goto bridged_dnat;
406                                 }
407                                 ip_rt_put(rt);
408                         }
409 free_skb:
410                         kfree_skb(skb);
411                         return 0;
412                 } else {
413                         if (skb_dst(skb)->dev == dev) {
414 bridged_dnat:
415                                 skb->dev = br_indev;
416                                 nf_bridge_update_protocol(skb);
417                                 nf_bridge_push_encap_header(skb);
418                                 br_nf_hook_thresh(NF_BR_PRE_ROUTING,
419                                                   net, sk, skb, skb->dev,
420                                                   NULL,
421                                                   br_nf_pre_routing_finish_bridge);
422                                 return 0;
423                         }
424                         ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
425                         skb->pkt_type = PACKET_HOST;
426                 }
427         } else {
428                 rt = bridge_parent_rtable(br_indev);
429                 if (!rt) {
430                         kfree_skb(skb);
431                         return 0;
432                 }
433                 skb_dst_drop(skb);
434                 skb_dst_set_noref(skb, &rt->dst);
435         }
436
437         skb->dev = br_indev;
438         nf_bridge_update_protocol(skb);
439         nf_bridge_push_encap_header(skb);
440         br_nf_hook_thresh(NF_BR_PRE_ROUTING, net, sk, skb, skb->dev, NULL,
441                           br_handle_frame_finish);
442         return 0;
443 }
444
445 static struct net_device *brnf_get_logical_dev(struct sk_buff *skb,
446                                                const struct net_device *dev,
447                                                const struct net *net)
448 {
449         struct net_device *vlan, *br;
450         struct brnf_net *brnet = net_generic(net, brnf_net_id);
451
452         br = bridge_parent(dev);
453
454         if (brnet->pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
455                 return br;
456
457         vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
458                                     skb_vlan_tag_get(skb) & VLAN_VID_MASK);
459
460         return vlan ? vlan : br;
461 }
462
463 /* Some common code for IPv4/IPv6 */
464 struct net_device *setup_pre_routing(struct sk_buff *skb, const struct net *net)
465 {
466         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
467
468         if (skb->pkt_type == PACKET_OTHERHOST) {
469                 skb->pkt_type = PACKET_HOST;
470                 nf_bridge->pkt_otherhost = true;
471         }
472
473         nf_bridge->in_prerouting = 1;
474         nf_bridge->physinif = skb->dev->ifindex;
475         skb->dev = brnf_get_logical_dev(skb, skb->dev, net);
476
477         if (skb->protocol == htons(ETH_P_8021Q))
478                 nf_bridge->orig_proto = BRNF_PROTO_8021Q;
479         else if (skb->protocol == htons(ETH_P_PPP_SES))
480                 nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
481
482         /* Must drop socket now because of tproxy. */
483         skb_orphan(skb);
484         return skb->dev;
485 }
486
487 /* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
488  * Replicate the checks that IPv4 does on packet reception.
489  * Set skb->dev to the bridge device (i.e. parent of the
490  * receiving device) to make netfilter happy, the REDIRECT
491  * target in particular.  Save the original destination IP
492  * address to be able to detect DNAT afterwards. */
493 static unsigned int br_nf_pre_routing(void *priv,
494                                       struct sk_buff *skb,
495                                       const struct nf_hook_state *state)
496 {
497         struct nf_bridge_info *nf_bridge;
498         struct net_bridge_port *p;
499         struct net_bridge *br;
500         __u32 len = nf_bridge_encap_header_len(skb);
501         struct brnf_net *brnet;
502
503         if (unlikely(!pskb_may_pull(skb, len)))
504                 return NF_DROP;
505
506         p = br_port_get_rcu(state->in);
507         if (p == NULL)
508                 return NF_DROP;
509         br = p->br;
510
511         brnet = net_generic(state->net, brnf_net_id);
512         if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
513             is_pppoe_ipv6(skb, state->net)) {
514                 if (!brnet->call_ip6tables &&
515                     !br_opt_get(br, BROPT_NF_CALL_IP6TABLES))
516                         return NF_ACCEPT;
517                 if (!ipv6_mod_enabled()) {
518                         pr_warn_once("Module ipv6 is disabled, so call_ip6tables is not supported.");
519                         return NF_DROP;
520                 }
521
522                 nf_bridge_pull_encap_header_rcsum(skb);
523                 return br_nf_pre_routing_ipv6(priv, skb, state);
524         }
525
526         if (!brnet->call_iptables && !br_opt_get(br, BROPT_NF_CALL_IPTABLES))
527                 return NF_ACCEPT;
528
529         if (!IS_IP(skb) && !is_vlan_ip(skb, state->net) &&
530             !is_pppoe_ip(skb, state->net))
531                 return NF_ACCEPT;
532
533         nf_bridge_pull_encap_header_rcsum(skb);
534
535         if (br_validate_ipv4(state->net, skb))
536                 return NF_DROP;
537
538         if (!nf_bridge_alloc(skb))
539                 return NF_DROP;
540         if (!setup_pre_routing(skb, state->net))
541                 return NF_DROP;
542
543         nf_bridge = nf_bridge_info_get(skb);
544         nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
545
546         skb->protocol = htons(ETH_P_IP);
547         skb->transport_header = skb->network_header + ip_hdr(skb)->ihl * 4;
548
549         NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->net, state->sk, skb,
550                 skb->dev, NULL,
551                 br_nf_pre_routing_finish);
552
553         return NF_STOLEN;
554 }
555
556
557 /* PF_BRIDGE/FORWARD *************************************************/
558 static int br_nf_forward_finish(struct net *net, struct sock *sk, struct sk_buff *skb)
559 {
560         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
561         struct net_device *in;
562
563         if (!IS_ARP(skb) && !is_vlan_arp(skb, net)) {
564
565                 if (skb->protocol == htons(ETH_P_IP))
566                         nf_bridge->frag_max_size = IPCB(skb)->frag_max_size;
567
568                 if (skb->protocol == htons(ETH_P_IPV6))
569                         nf_bridge->frag_max_size = IP6CB(skb)->frag_max_size;
570
571                 in = nf_bridge_get_physindev(skb, net);
572                 if (!in) {
573                         kfree_skb(skb);
574                         return 0;
575                 }
576                 if (nf_bridge->pkt_otherhost) {
577                         skb->pkt_type = PACKET_OTHERHOST;
578                         nf_bridge->pkt_otherhost = false;
579                 }
580                 nf_bridge_update_protocol(skb);
581         } else {
582                 in = *((struct net_device **)(skb->cb));
583         }
584         nf_bridge_push_encap_header(skb);
585
586         br_nf_hook_thresh(NF_BR_FORWARD, net, sk, skb, in, skb->dev,
587                           br_forward_finish);
588         return 0;
589 }
590
591
592 /* This is the 'purely bridged' case.  For IP, we pass the packet to
593  * netfilter with indev and outdev set to the bridge device,
594  * but we are still able to filter on the 'real' indev/outdev
595  * because of the physdev module. For ARP, indev and outdev are the
596  * bridge ports. */
597 static unsigned int br_nf_forward_ip(void *priv,
598                                      struct sk_buff *skb,
599                                      const struct nf_hook_state *state)
600 {
601         struct nf_bridge_info *nf_bridge;
602         struct net_device *parent;
603         u_int8_t pf;
604
605         nf_bridge = nf_bridge_info_get(skb);
606         if (!nf_bridge)
607                 return NF_ACCEPT;
608
609         /* Need exclusive nf_bridge_info since we might have multiple
610          * different physoutdevs. */
611         if (!nf_bridge_unshare(skb))
612                 return NF_DROP;
613
614         nf_bridge = nf_bridge_info_get(skb);
615         if (!nf_bridge)
616                 return NF_DROP;
617
618         parent = bridge_parent(state->out);
619         if (!parent)
620                 return NF_DROP;
621
622         if (IS_IP(skb) || is_vlan_ip(skb, state->net) ||
623             is_pppoe_ip(skb, state->net))
624                 pf = NFPROTO_IPV4;
625         else if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
626                  is_pppoe_ipv6(skb, state->net))
627                 pf = NFPROTO_IPV6;
628         else
629                 return NF_ACCEPT;
630
631         nf_bridge_pull_encap_header(skb);
632
633         if (skb->pkt_type == PACKET_OTHERHOST) {
634                 skb->pkt_type = PACKET_HOST;
635                 nf_bridge->pkt_otherhost = true;
636         }
637
638         if (pf == NFPROTO_IPV4) {
639                 if (br_validate_ipv4(state->net, skb))
640                         return NF_DROP;
641                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
642         }
643
644         if (pf == NFPROTO_IPV6) {
645                 if (br_validate_ipv6(state->net, skb))
646                         return NF_DROP;
647                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
648         }
649
650         nf_bridge->physoutdev = skb->dev;
651         if (pf == NFPROTO_IPV4)
652                 skb->protocol = htons(ETH_P_IP);
653         else
654                 skb->protocol = htons(ETH_P_IPV6);
655
656         NF_HOOK(pf, NF_INET_FORWARD, state->net, NULL, skb,
657                 brnf_get_logical_dev(skb, state->in, state->net),
658                 parent, br_nf_forward_finish);
659
660         return NF_STOLEN;
661 }
662
663 static unsigned int br_nf_forward_arp(void *priv,
664                                       struct sk_buff *skb,
665                                       const struct nf_hook_state *state)
666 {
667         struct net_bridge_port *p;
668         struct net_bridge *br;
669         struct net_device **d = (struct net_device **)(skb->cb);
670         struct brnf_net *brnet;
671
672         p = br_port_get_rcu(state->out);
673         if (p == NULL)
674                 return NF_ACCEPT;
675         br = p->br;
676
677         brnet = net_generic(state->net, brnf_net_id);
678         if (!brnet->call_arptables && !br_opt_get(br, BROPT_NF_CALL_ARPTABLES))
679                 return NF_ACCEPT;
680
681         if (!IS_ARP(skb)) {
682                 if (!is_vlan_arp(skb, state->net))
683                         return NF_ACCEPT;
684                 nf_bridge_pull_encap_header(skb);
685         }
686
687         if (unlikely(!pskb_may_pull(skb, sizeof(struct arphdr))))
688                 return NF_DROP;
689
690         if (arp_hdr(skb)->ar_pln != 4) {
691                 if (is_vlan_arp(skb, state->net))
692                         nf_bridge_push_encap_header(skb);
693                 return NF_ACCEPT;
694         }
695         *d = state->in;
696         NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->net, state->sk, skb,
697                 state->in, state->out, br_nf_forward_finish);
698
699         return NF_STOLEN;
700 }
701
702 static int br_nf_push_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
703 {
704         struct brnf_frag_data *data;
705         int err;
706
707         data = this_cpu_ptr(&brnf_frag_data_storage);
708         err = skb_cow_head(skb, data->size);
709
710         if (err) {
711                 kfree_skb(skb);
712                 return 0;
713         }
714
715         if (data->vlan_proto)
716                 __vlan_hwaccel_put_tag(skb, data->vlan_proto, data->vlan_tci);
717
718         skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
719         __skb_push(skb, data->encap_size);
720
721         nf_bridge_info_free(skb);
722         return br_dev_queue_push_xmit(net, sk, skb);
723 }
724
725 static int
726 br_nf_ip_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
727                   int (*output)(struct net *, struct sock *, struct sk_buff *))
728 {
729         unsigned int mtu = ip_skb_dst_mtu(sk, skb);
730         struct iphdr *iph = ip_hdr(skb);
731
732         if (unlikely(((iph->frag_off & htons(IP_DF)) && !skb->ignore_df) ||
733                      (IPCB(skb)->frag_max_size &&
734                       IPCB(skb)->frag_max_size > mtu))) {
735                 IP_INC_STATS(net, IPSTATS_MIB_FRAGFAILS);
736                 kfree_skb(skb);
737                 return -EMSGSIZE;
738         }
739
740         return ip_do_fragment(net, sk, skb, output);
741 }
742
743 static unsigned int nf_bridge_mtu_reduction(const struct sk_buff *skb)
744 {
745         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
746
747         if (nf_bridge->orig_proto == BRNF_PROTO_PPPOE)
748                 return PPPOE_SES_HLEN;
749         return 0;
750 }
751
752 static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
753 {
754         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
755         unsigned int mtu, mtu_reserved;
756
757         mtu_reserved = nf_bridge_mtu_reduction(skb);
758         mtu = skb->dev->mtu;
759
760         if (nf_bridge->pkt_otherhost) {
761                 skb->pkt_type = PACKET_OTHERHOST;
762                 nf_bridge->pkt_otherhost = false;
763         }
764
765         if (nf_bridge->frag_max_size && nf_bridge->frag_max_size < mtu)
766                 mtu = nf_bridge->frag_max_size;
767
768         nf_bridge_update_protocol(skb);
769         nf_bridge_push_encap_header(skb);
770
771         if (skb_is_gso(skb) || skb->len + mtu_reserved <= mtu) {
772                 nf_bridge_info_free(skb);
773                 return br_dev_queue_push_xmit(net, sk, skb);
774         }
775
776         /* This is wrong! We should preserve the original fragment
777          * boundaries by preserving frag_list rather than refragmenting.
778          */
779         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) &&
780             skb->protocol == htons(ETH_P_IP)) {
781                 struct brnf_frag_data *data;
782
783                 if (br_validate_ipv4(net, skb))
784                         goto drop;
785
786                 IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
787
788                 data = this_cpu_ptr(&brnf_frag_data_storage);
789
790                 if (skb_vlan_tag_present(skb)) {
791                         data->vlan_tci = skb->vlan_tci;
792                         data->vlan_proto = skb->vlan_proto;
793                 } else {
794                         data->vlan_proto = 0;
795                 }
796
797                 data->encap_size = nf_bridge_encap_header_len(skb);
798                 data->size = ETH_HLEN + data->encap_size;
799
800                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
801                                                  data->size);
802
803                 return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
804         }
805         if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
806             skb->protocol == htons(ETH_P_IPV6)) {
807                 const struct nf_ipv6_ops *v6ops = nf_get_ipv6_ops();
808                 struct brnf_frag_data *data;
809
810                 if (br_validate_ipv6(net, skb))
811                         goto drop;
812
813                 IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
814
815                 data = this_cpu_ptr(&brnf_frag_data_storage);
816                 data->encap_size = nf_bridge_encap_header_len(skb);
817                 data->size = ETH_HLEN + data->encap_size;
818
819                 skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
820                                                  data->size);
821
822                 if (v6ops)
823                         return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
824
825                 kfree_skb(skb);
826                 return -EMSGSIZE;
827         }
828         nf_bridge_info_free(skb);
829         return br_dev_queue_push_xmit(net, sk, skb);
830  drop:
831         kfree_skb(skb);
832         return 0;
833 }
834
835 /* PF_BRIDGE/POST_ROUTING ********************************************/
836 static unsigned int br_nf_post_routing(void *priv,
837                                        struct sk_buff *skb,
838                                        const struct nf_hook_state *state)
839 {
840         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
841         struct net_device *realoutdev = bridge_parent(skb->dev);
842         u_int8_t pf;
843
844         /* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
845          * on a bridge, but was delivered locally and is now being routed:
846          *
847          * POST_ROUTING was already invoked from the ip stack.
848          */
849         if (!nf_bridge || !nf_bridge->physoutdev)
850                 return NF_ACCEPT;
851
852         if (!realoutdev)
853                 return NF_DROP;
854
855         if (IS_IP(skb) || is_vlan_ip(skb, state->net) ||
856             is_pppoe_ip(skb, state->net))
857                 pf = NFPROTO_IPV4;
858         else if (IS_IPV6(skb) || is_vlan_ipv6(skb, state->net) ||
859                  is_pppoe_ipv6(skb, state->net))
860                 pf = NFPROTO_IPV6;
861         else
862                 return NF_ACCEPT;
863
864         if (skb->pkt_type == PACKET_OTHERHOST) {
865                 skb->pkt_type = PACKET_HOST;
866                 nf_bridge->pkt_otherhost = true;
867         }
868
869         nf_bridge_pull_encap_header(skb);
870         if (pf == NFPROTO_IPV4)
871                 skb->protocol = htons(ETH_P_IP);
872         else
873                 skb->protocol = htons(ETH_P_IPV6);
874
875         NF_HOOK(pf, NF_INET_POST_ROUTING, state->net, state->sk, skb,
876                 NULL, realoutdev,
877                 br_nf_dev_queue_xmit);
878
879         return NF_STOLEN;
880 }
881
882 /* IP/SABOTAGE *****************************************************/
883 /* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
884  * for the second time. */
885 static unsigned int ip_sabotage_in(void *priv,
886                                    struct sk_buff *skb,
887                                    const struct nf_hook_state *state)
888 {
889         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
890
891         if (nf_bridge) {
892                 if (nf_bridge->sabotage_in_done)
893                         return NF_ACCEPT;
894
895                 if (!nf_bridge->in_prerouting &&
896                     !netif_is_l3_master(skb->dev) &&
897                     !netif_is_l3_slave(skb->dev)) {
898                         nf_bridge->sabotage_in_done = 1;
899                         state->okfn(state->net, state->sk, skb);
900                         return NF_STOLEN;
901                 }
902         }
903
904         return NF_ACCEPT;
905 }
906
907 /* This is called when br_netfilter has called into iptables/netfilter,
908  * and DNAT has taken place on a bridge-forwarded packet.
909  *
910  * neigh->output has created a new MAC header, with local br0 MAC
911  * as saddr.
912  *
913  * This restores the original MAC saddr of the bridged packet
914  * before invoking bridge forward logic to transmit the packet.
915  */
916 static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
917 {
918         struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
919         struct net_device *br_indev;
920
921         br_indev = nf_bridge_get_physindev(skb, dev_net(skb->dev));
922         if (!br_indev) {
923                 kfree_skb(skb);
924                 return;
925         }
926
927         skb_pull(skb, ETH_HLEN);
928         nf_bridge->bridged_dnat = 0;
929
930         BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
931
932         skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
933                                        nf_bridge->neigh_header,
934                                        ETH_HLEN - ETH_ALEN);
935         skb->dev = br_indev;
936
937         nf_bridge->physoutdev = NULL;
938         br_handle_frame_finish(dev_net(skb->dev), NULL, skb);
939 }
940
941 static int br_nf_dev_xmit(struct sk_buff *skb)
942 {
943         const struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
944
945         if (nf_bridge && nf_bridge->bridged_dnat) {
946                 br_nf_pre_routing_finish_bridge_slow(skb);
947                 return 1;
948         }
949         return 0;
950 }
951
952 static const struct nf_br_ops br_ops = {
953         .br_dev_xmit_hook =     br_nf_dev_xmit,
954 };
955
956 /* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
957  * br_dev_queue_push_xmit is called afterwards */
958 static const struct nf_hook_ops br_nf_ops[] = {
959         {
960                 .hook = br_nf_pre_routing,
961                 .pf = NFPROTO_BRIDGE,
962                 .hooknum = NF_BR_PRE_ROUTING,
963                 .priority = NF_BR_PRI_BRNF,
964         },
965         {
966                 .hook = br_nf_forward_ip,
967                 .pf = NFPROTO_BRIDGE,
968                 .hooknum = NF_BR_FORWARD,
969                 .priority = NF_BR_PRI_BRNF - 1,
970         },
971         {
972                 .hook = br_nf_forward_arp,
973                 .pf = NFPROTO_BRIDGE,
974                 .hooknum = NF_BR_FORWARD,
975                 .priority = NF_BR_PRI_BRNF,
976         },
977         {
978                 .hook = br_nf_post_routing,
979                 .pf = NFPROTO_BRIDGE,
980                 .hooknum = NF_BR_POST_ROUTING,
981                 .priority = NF_BR_PRI_LAST,
982         },
983         {
984                 .hook = ip_sabotage_in,
985                 .pf = NFPROTO_IPV4,
986                 .hooknum = NF_INET_PRE_ROUTING,
987                 .priority = NF_IP_PRI_FIRST,
988         },
989         {
990                 .hook = ip_sabotage_in,
991                 .pf = NFPROTO_IPV6,
992                 .hooknum = NF_INET_PRE_ROUTING,
993                 .priority = NF_IP6_PRI_FIRST,
994         },
995 };
996
997 static int brnf_device_event(struct notifier_block *unused, unsigned long event,
998                              void *ptr)
999 {
1000         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1001         struct brnf_net *brnet;
1002         struct net *net;
1003         int ret;
1004
1005         if (event != NETDEV_REGISTER || !netif_is_bridge_master(dev))
1006                 return NOTIFY_DONE;
1007
1008         ASSERT_RTNL();
1009
1010         net = dev_net(dev);
1011         brnet = net_generic(net, brnf_net_id);
1012         if (brnet->enabled)
1013                 return NOTIFY_OK;
1014
1015         ret = nf_register_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
1016         if (ret)
1017                 return NOTIFY_BAD;
1018
1019         brnet->enabled = true;
1020         return NOTIFY_OK;
1021 }
1022
1023 static struct notifier_block brnf_notifier __read_mostly = {
1024         .notifier_call = brnf_device_event,
1025 };
1026
1027 /* recursively invokes nf_hook_slow (again), skipping already-called
1028  * hooks (< NF_BR_PRI_BRNF).
1029  *
1030  * Called with rcu read lock held.
1031  */
1032 int br_nf_hook_thresh(unsigned int hook, struct net *net,
1033                       struct sock *sk, struct sk_buff *skb,
1034                       struct net_device *indev,
1035                       struct net_device *outdev,
1036                       int (*okfn)(struct net *, struct sock *,
1037                                   struct sk_buff *))
1038 {
1039         const struct nf_hook_entries *e;
1040         struct nf_hook_state state;
1041         struct nf_hook_ops **ops;
1042         unsigned int i;
1043         int ret;
1044
1045         e = rcu_dereference(net->nf.hooks_bridge[hook]);
1046         if (!e)
1047                 return okfn(net, sk, skb);
1048
1049         ops = nf_hook_entries_get_hook_ops(e);
1050         for (i = 0; i < e->num_hook_entries; i++) {
1051                 /* These hooks have already been called */
1052                 if (ops[i]->priority < NF_BR_PRI_BRNF)
1053                         continue;
1054
1055                 /* These hooks have not been called yet, run them. */
1056                 if (ops[i]->priority > NF_BR_PRI_BRNF)
1057                         break;
1058
1059                 /* take a closer look at NF_BR_PRI_BRNF. */
1060                 if (ops[i]->hook == br_nf_pre_routing) {
1061                         /* This hook diverted the skb to this function,
1062                          * hooks after this have not been run yet.
1063                          */
1064                         i++;
1065                         break;
1066                 }
1067         }
1068
1069         nf_hook_state_init(&state, hook, NFPROTO_BRIDGE, indev, outdev,
1070                            sk, net, okfn);
1071
1072         ret = nf_hook_slow(skb, &state, e, i);
1073         if (ret == 1)
1074                 ret = okfn(net, sk, skb);
1075
1076         return ret;
1077 }
1078
1079 #ifdef CONFIG_SYSCTL
1080 static
1081 int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1082                             void *buffer, size_t *lenp, loff_t *ppos)
1083 {
1084         int ret;
1085
1086         ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1087
1088         if (write && *(int *)(ctl->data))
1089                 *(int *)(ctl->data) = 1;
1090         return ret;
1091 }
1092
1093 static struct ctl_table brnf_table[] = {
1094         {
1095                 .procname       = "bridge-nf-call-arptables",
1096                 .maxlen         = sizeof(int),
1097                 .mode           = 0644,
1098                 .proc_handler   = brnf_sysctl_call_tables,
1099         },
1100         {
1101                 .procname       = "bridge-nf-call-iptables",
1102                 .maxlen         = sizeof(int),
1103                 .mode           = 0644,
1104                 .proc_handler   = brnf_sysctl_call_tables,
1105         },
1106         {
1107                 .procname       = "bridge-nf-call-ip6tables",
1108                 .maxlen         = sizeof(int),
1109                 .mode           = 0644,
1110                 .proc_handler   = brnf_sysctl_call_tables,
1111         },
1112         {
1113                 .procname       = "bridge-nf-filter-vlan-tagged",
1114                 .maxlen         = sizeof(int),
1115                 .mode           = 0644,
1116                 .proc_handler   = brnf_sysctl_call_tables,
1117         },
1118         {
1119                 .procname       = "bridge-nf-filter-pppoe-tagged",
1120                 .maxlen         = sizeof(int),
1121                 .mode           = 0644,
1122                 .proc_handler   = brnf_sysctl_call_tables,
1123         },
1124         {
1125                 .procname       = "bridge-nf-pass-vlan-input-dev",
1126                 .maxlen         = sizeof(int),
1127                 .mode           = 0644,
1128                 .proc_handler   = brnf_sysctl_call_tables,
1129         },
1130         { }
1131 };
1132
1133 static inline void br_netfilter_sysctl_default(struct brnf_net *brnf)
1134 {
1135         brnf->call_iptables = 1;
1136         brnf->call_ip6tables = 1;
1137         brnf->call_arptables = 1;
1138         brnf->filter_vlan_tagged = 0;
1139         brnf->filter_pppoe_tagged = 0;
1140         brnf->pass_vlan_indev = 0;
1141 }
1142
1143 static int br_netfilter_sysctl_init_net(struct net *net)
1144 {
1145         struct ctl_table *table = brnf_table;
1146         struct brnf_net *brnet;
1147
1148         if (!net_eq(net, &init_net)) {
1149                 table = kmemdup(table, sizeof(brnf_table), GFP_KERNEL);
1150                 if (!table)
1151                         return -ENOMEM;
1152         }
1153
1154         brnet = net_generic(net, brnf_net_id);
1155         table[0].data = &brnet->call_arptables;
1156         table[1].data = &brnet->call_iptables;
1157         table[2].data = &brnet->call_ip6tables;
1158         table[3].data = &brnet->filter_vlan_tagged;
1159         table[4].data = &brnet->filter_pppoe_tagged;
1160         table[5].data = &brnet->pass_vlan_indev;
1161
1162         br_netfilter_sysctl_default(brnet);
1163
1164         brnet->ctl_hdr = register_net_sysctl_sz(net, "net/bridge", table,
1165                                                 ARRAY_SIZE(brnf_table));
1166         if (!brnet->ctl_hdr) {
1167                 if (!net_eq(net, &init_net))
1168                         kfree(table);
1169
1170                 return -ENOMEM;
1171         }
1172
1173         return 0;
1174 }
1175
1176 static void br_netfilter_sysctl_exit_net(struct net *net,
1177                                          struct brnf_net *brnet)
1178 {
1179         struct ctl_table *table = brnet->ctl_hdr->ctl_table_arg;
1180
1181         unregister_net_sysctl_table(brnet->ctl_hdr);
1182         if (!net_eq(net, &init_net))
1183                 kfree(table);
1184 }
1185
1186 static int __net_init brnf_init_net(struct net *net)
1187 {
1188         return br_netfilter_sysctl_init_net(net);
1189 }
1190 #endif
1191
1192 static void __net_exit brnf_exit_net(struct net *net)
1193 {
1194         struct brnf_net *brnet;
1195
1196         brnet = net_generic(net, brnf_net_id);
1197         if (brnet->enabled) {
1198                 nf_unregister_net_hooks(net, br_nf_ops, ARRAY_SIZE(br_nf_ops));
1199                 brnet->enabled = false;
1200         }
1201
1202 #ifdef CONFIG_SYSCTL
1203         br_netfilter_sysctl_exit_net(net, brnet);
1204 #endif
1205 }
1206
1207 static struct pernet_operations brnf_net_ops __read_mostly = {
1208 #ifdef CONFIG_SYSCTL
1209         .init = brnf_init_net,
1210 #endif
1211         .exit = brnf_exit_net,
1212         .id   = &brnf_net_id,
1213         .size = sizeof(struct brnf_net),
1214 };
1215
1216 static int __init br_netfilter_init(void)
1217 {
1218         int ret;
1219
1220         ret = register_pernet_subsys(&brnf_net_ops);
1221         if (ret < 0)
1222                 return ret;
1223
1224         ret = register_netdevice_notifier(&brnf_notifier);
1225         if (ret < 0) {
1226                 unregister_pernet_subsys(&brnf_net_ops);
1227                 return ret;
1228         }
1229
1230         RCU_INIT_POINTER(nf_br_ops, &br_ops);
1231         printk(KERN_NOTICE "Bridge firewalling registered\n");
1232         return 0;
1233 }
1234
1235 static void __exit br_netfilter_fini(void)
1236 {
1237         RCU_INIT_POINTER(nf_br_ops, NULL);
1238         unregister_netdevice_notifier(&brnf_notifier);
1239         unregister_pernet_subsys(&brnf_net_ops);
1240 }
1241
1242 module_init(br_netfilter_init);
1243 module_exit(br_netfilter_fini);
1244
1245 MODULE_LICENSE("GPL");
1246 MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1247 MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1248 MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");