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#include <net/netfilter/br_netfilter.h>
39
40#include <asm/uaccess.h>
41#include "br_private.h"
42#ifdef CONFIG_SYSCTL
43#include <linux/sysctl.h>
44#endif
45
46#ifdef CONFIG_SYSCTL
47static struct ctl_table_header *brnf_sysctl_header;
48static int brnf_call_iptables __read_mostly = 1;
49static int brnf_call_ip6tables __read_mostly = 1;
50static int brnf_call_arptables __read_mostly = 1;
51static int brnf_filter_vlan_tagged __read_mostly = 0;
52static int brnf_filter_pppoe_tagged __read_mostly = 0;
53static int brnf_pass_vlan_indev __read_mostly = 0;
54#else
55#define brnf_call_iptables 1
56#define brnf_call_ip6tables 1
57#define brnf_call_arptables 1
58#define brnf_filter_vlan_tagged 0
59#define brnf_filter_pppoe_tagged 0
60#define brnf_pass_vlan_indev 0
61#endif
62
63#define IS_IP(skb) \
64	(!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IP))
65
66#define IS_IPV6(skb) \
67	(!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_IPV6))
68
69#define IS_ARP(skb) \
70	(!skb_vlan_tag_present(skb) && skb->protocol == htons(ETH_P_ARP))
71
72static inline __be16 vlan_proto(const struct sk_buff *skb)
73{
74	if (skb_vlan_tag_present(skb))
75		return skb->protocol;
76	else if (skb->protocol == htons(ETH_P_8021Q))
77		return vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
78	else
79		return 0;
80}
81
82#define IS_VLAN_IP(skb) \
83	(vlan_proto(skb) == htons(ETH_P_IP) && \
84	 brnf_filter_vlan_tagged)
85
86#define IS_VLAN_IPV6(skb) \
87	(vlan_proto(skb) == htons(ETH_P_IPV6) && \
88	 brnf_filter_vlan_tagged)
89
90#define IS_VLAN_ARP(skb) \
91	(vlan_proto(skb) == htons(ETH_P_ARP) &&	\
92	 brnf_filter_vlan_tagged)
93
94static inline __be16 pppoe_proto(const struct sk_buff *skb)
95{
96	return *((__be16 *)(skb_mac_header(skb) + ETH_HLEN +
97			    sizeof(struct pppoe_hdr)));
98}
99
100#define IS_PPPOE_IP(skb) \
101	(skb->protocol == htons(ETH_P_PPP_SES) && \
102	 pppoe_proto(skb) == htons(PPP_IP) && \
103	 brnf_filter_pppoe_tagged)
104
105#define IS_PPPOE_IPV6(skb) \
106	(skb->protocol == htons(ETH_P_PPP_SES) && \
107	 pppoe_proto(skb) == htons(PPP_IPV6) && \
108	 brnf_filter_pppoe_tagged)
109
110/* largest possible L2 header, see br_nf_dev_queue_xmit() */
111#define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
112
113#if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
114struct brnf_frag_data {
115	char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
116	u8 encap_size;
117	u8 size;
118};
119
120static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
121#endif
122
123static struct nf_bridge_info *nf_bridge_info_get(const struct sk_buff *skb)
124{
125	return skb->nf_bridge;
126}
127
128static inline struct rtable *bridge_parent_rtable(const struct net_device *dev)
129{
130	struct net_bridge_port *port;
131
132	port = br_port_get_rcu(dev);
133	return port ? &port->br->fake_rtable : NULL;
134}
135
136static inline struct net_device *bridge_parent(const struct net_device *dev)
137{
138	struct net_bridge_port *port;
139
140	port = br_port_get_rcu(dev);
141	return port ? port->br->dev : NULL;
142}
143
144static inline struct nf_bridge_info *nf_bridge_alloc(struct sk_buff *skb)
145{
146	skb->nf_bridge = kzalloc(sizeof(struct nf_bridge_info), GFP_ATOMIC);
147	if (likely(skb->nf_bridge))
148		atomic_set(&(skb->nf_bridge->use), 1);
149
150	return skb->nf_bridge;
151}
152
153static inline struct nf_bridge_info *nf_bridge_unshare(struct sk_buff *skb)
154{
155	struct nf_bridge_info *nf_bridge = skb->nf_bridge;
156
157	if (atomic_read(&nf_bridge->use) > 1) {
158		struct nf_bridge_info *tmp = nf_bridge_alloc(skb);
159
160		if (tmp) {
161			memcpy(tmp, nf_bridge, sizeof(struct nf_bridge_info));
162			atomic_set(&tmp->use, 1);
163		}
164		nf_bridge_put(nf_bridge);
165		nf_bridge = tmp;
166	}
167	return nf_bridge;
168}
169
170static unsigned int nf_bridge_encap_header_len(const struct sk_buff *skb)
171{
172	switch (skb->protocol) {
173	case __cpu_to_be16(ETH_P_8021Q):
174		return VLAN_HLEN;
175	case __cpu_to_be16(ETH_P_PPP_SES):
176		return PPPOE_SES_HLEN;
177	default:
178		return 0;
179	}
180}
181
182static inline void nf_bridge_push_encap_header(struct sk_buff *skb)
183{
184	unsigned int len = nf_bridge_encap_header_len(skb);
185
186	skb_push(skb, len);
187	skb->network_header -= len;
188}
189
190static inline void nf_bridge_pull_encap_header(struct sk_buff *skb)
191{
192	unsigned int len = nf_bridge_encap_header_len(skb);
193
194	skb_pull(skb, len);
195	skb->network_header += len;
196}
197
198static inline void nf_bridge_pull_encap_header_rcsum(struct sk_buff *skb)
199{
200	unsigned int len = nf_bridge_encap_header_len(skb);
201
202	skb_pull_rcsum(skb, len);
203	skb->network_header += len;
204}
205
206/* When handing a packet over to the IP layer
207 * check whether we have a skb that is in the
208 * expected format
209 */
210
211static int br_parse_ip_options(struct sk_buff *skb)
212{
213	const struct iphdr *iph;
214	struct net_device *dev = skb->dev;
215	u32 len;
216
217	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
218		goto inhdr_error;
219
220	iph = ip_hdr(skb);
221
222	/* Basic sanity checks */
223	if (iph->ihl < 5 || iph->version != 4)
224		goto inhdr_error;
225
226	if (!pskb_may_pull(skb, iph->ihl*4))
227		goto inhdr_error;
228
229	iph = ip_hdr(skb);
230	if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
231		goto inhdr_error;
232
233	len = ntohs(iph->tot_len);
234	if (skb->len < len) {
235		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INTRUNCATEDPKTS);
236		goto drop;
237	} else if (len < (iph->ihl*4))
238		goto inhdr_error;
239
240	if (pskb_trim_rcsum(skb, len)) {
241		IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INDISCARDS);
242		goto drop;
243	}
244
245	memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
246	/* We should really parse IP options here but until
247	 * somebody who actually uses IP options complains to
248	 * us we'll just silently ignore the options because
249	 * we're lazy!
250	 */
251	return 0;
252
253inhdr_error:
254	IP_INC_STATS_BH(dev_net(dev), IPSTATS_MIB_INHDRERRORS);
255drop:
256	return -1;
257}
258
259static void nf_bridge_update_protocol(struct sk_buff *skb)
260{
261	switch (skb->nf_bridge->orig_proto) {
262	case BRNF_PROTO_8021Q:
263		skb->protocol = htons(ETH_P_8021Q);
264		break;
265	case BRNF_PROTO_PPPOE:
266		skb->protocol = htons(ETH_P_PPP_SES);
267		break;
268	case BRNF_PROTO_UNCHANGED:
269		break;
270	}
271}
272
273/* PF_BRIDGE/PRE_ROUTING *********************************************/
274/* Undo the changes made for ip6tables PREROUTING and continue the
275 * bridge PRE_ROUTING hook. */
276static int br_nf_pre_routing_finish_ipv6(struct sock *sk, struct sk_buff *skb)
277{
278	struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
279	struct rtable *rt;
280
281	if (nf_bridge->pkt_otherhost) {
282		skb->pkt_type = PACKET_OTHERHOST;
283		nf_bridge->pkt_otherhost = false;
284	}
285	nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
286
287	rt = bridge_parent_rtable(nf_bridge->physindev);
288	if (!rt) {
289		kfree_skb(skb);
290		return 0;
291	}
292	skb_dst_set_noref(skb, &rt->dst);
293
294	skb->dev = nf_bridge->physindev;
295	nf_bridge_update_protocol(skb);
296	nf_bridge_push_encap_header(skb);
297	NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, sk, skb,
298		       skb->dev, NULL,
299		       br_handle_frame_finish, 1);
300
301	return 0;
302}
303
304/* Obtain the correct destination MAC address, while preserving the original
305 * source MAC address. If we already know this address, we just copy it. If we
306 * don't, we use the neighbour framework to find out. In both cases, we make
307 * sure that br_handle_frame_finish() is called afterwards.
308 */
309static int br_nf_pre_routing_finish_bridge(struct sock *sk, struct sk_buff *skb)
310{
311	struct neighbour *neigh;
312	struct dst_entry *dst;
313
314	skb->dev = bridge_parent(skb->dev);
315	if (!skb->dev)
316		goto free_skb;
317	dst = skb_dst(skb);
318	neigh = dst_neigh_lookup_skb(dst, skb);
319	if (neigh) {
320		struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
321		int ret;
322
323		if (neigh->hh.hh_len) {
324			neigh_hh_bridge(&neigh->hh, skb);
325			skb->dev = nf_bridge->physindev;
326			ret = br_handle_frame_finish(sk, skb);
327		} else {
328			/* the neighbour function below overwrites the complete
329			 * MAC header, so we save the Ethernet source address and
330			 * protocol number.
331			 */
332			skb_copy_from_linear_data_offset(skb,
333							 -(ETH_HLEN-ETH_ALEN),
334							 nf_bridge->neigh_header,
335							 ETH_HLEN-ETH_ALEN);
336			/* tell br_dev_xmit to continue with forwarding */
337			nf_bridge->mask |= BRNF_BRIDGED_DNAT;
338			/* FIXME Need to refragment */
339			ret = neigh->output(neigh, skb);
340		}
341		neigh_release(neigh);
342		return ret;
343	}
344free_skb:
345	kfree_skb(skb);
346	return 0;
347}
348
349static bool daddr_was_changed(const struct sk_buff *skb,
350			      const struct nf_bridge_info *nf_bridge)
351{
352	return ip_hdr(skb)->daddr != nf_bridge->ipv4_daddr;
353}
354
355/* This requires some explaining. If DNAT has taken place,
356 * we will need to fix up the destination Ethernet address.
357 * This is also true when SNAT takes place (for the reply direction).
358 *
359 * There are two cases to consider:
360 * 1. The packet was DNAT'ed to a device in the same bridge
361 *    port group as it was received on. We can still bridge
362 *    the packet.
363 * 2. The packet was DNAT'ed to a different device, either
364 *    a non-bridged device or another bridge port group.
365 *    The packet will need to be routed.
366 *
367 * The correct way of distinguishing between these two cases is to
368 * call ip_route_input() and to look at skb->dst->dev, which is
369 * changed to the destination device if ip_route_input() succeeds.
370 *
371 * Let's first consider the case that ip_route_input() succeeds:
372 *
373 * If the output device equals the logical bridge device the packet
374 * came in on, we can consider this bridging. The corresponding MAC
375 * address will be obtained in br_nf_pre_routing_finish_bridge.
376 * Otherwise, the packet is considered to be routed and we just
377 * change the destination MAC address so that the packet will
378 * later be passed up to the IP stack to be routed. For a redirected
379 * packet, ip_route_input() will give back the localhost as output device,
380 * which differs from the bridge device.
381 *
382 * Let's now consider the case that ip_route_input() fails:
383 *
384 * This can be because the destination address is martian, in which case
385 * the packet will be dropped.
386 * If IP forwarding is disabled, ip_route_input() will fail, while
387 * ip_route_output_key() can return success. The source
388 * address for ip_route_output_key() is set to zero, so ip_route_output_key()
389 * thinks we're handling a locally generated packet and won't care
390 * if IP forwarding is enabled. If the output device equals the logical bridge
391 * device, we proceed as if ip_route_input() succeeded. If it differs from the
392 * logical bridge port or if ip_route_output_key() fails we drop the packet.
393 */
394static int br_nf_pre_routing_finish(struct sock *sk, struct sk_buff *skb)
395{
396	struct net_device *dev = skb->dev;
397	struct iphdr *iph = ip_hdr(skb);
398	struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
399	struct rtable *rt;
400	int err;
401	int frag_max_size;
402
403	frag_max_size = IPCB(skb)->frag_max_size;
404	BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
405
406	if (nf_bridge->pkt_otherhost) {
407		skb->pkt_type = PACKET_OTHERHOST;
408		nf_bridge->pkt_otherhost = false;
409	}
410	nf_bridge->mask ^= BRNF_NF_BRIDGE_PREROUTING;
411	if (daddr_was_changed(skb, nf_bridge)) {
412		if ((err = ip_route_input(skb, iph->daddr, iph->saddr, iph->tos, dev))) {
413			struct in_device *in_dev = __in_dev_get_rcu(dev);
414
415			/* If err equals -EHOSTUNREACH the error is due to a
416			 * martian destination or due to the fact that
417			 * forwarding is disabled. For most martian packets,
418			 * ip_route_output_key() will fail. It won't fail for 2 types of
419			 * martian destinations: loopback destinations and destination
420			 * 0.0.0.0. In both cases the packet will be dropped because the
421			 * destination is the loopback device and not the bridge. */
422			if (err != -EHOSTUNREACH || !in_dev || IN_DEV_FORWARD(in_dev))
423				goto free_skb;
424
425			rt = ip_route_output(dev_net(dev), iph->daddr, 0,
426					     RT_TOS(iph->tos), 0);
427			if (!IS_ERR(rt)) {
428				/* - Bridged-and-DNAT'ed traffic doesn't
429				 *   require ip_forwarding. */
430				if (rt->dst.dev == dev) {
431					skb_dst_set(skb, &rt->dst);
432					goto bridged_dnat;
433				}
434				ip_rt_put(rt);
435			}
436free_skb:
437			kfree_skb(skb);
438			return 0;
439		} else {
440			if (skb_dst(skb)->dev == dev) {
441bridged_dnat:
442				skb->dev = nf_bridge->physindev;
443				nf_bridge_update_protocol(skb);
444				nf_bridge_push_encap_header(skb);
445				NF_HOOK_THRESH(NFPROTO_BRIDGE,
446					       NF_BR_PRE_ROUTING,
447					       sk, skb, skb->dev, NULL,
448					       br_nf_pre_routing_finish_bridge,
449					       1);
450				return 0;
451			}
452			ether_addr_copy(eth_hdr(skb)->h_dest, dev->dev_addr);
453			skb->pkt_type = PACKET_HOST;
454		}
455	} else {
456		rt = bridge_parent_rtable(nf_bridge->physindev);
457		if (!rt) {
458			kfree_skb(skb);
459			return 0;
460		}
461		skb_dst_set_noref(skb, &rt->dst);
462	}
463
464	skb->dev = nf_bridge->physindev;
465	nf_bridge_update_protocol(skb);
466	nf_bridge_push_encap_header(skb);
467	NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_PRE_ROUTING, sk, skb,
468		       skb->dev, NULL,
469		       br_handle_frame_finish, 1);
470
471	return 0;
472}
473
474static struct net_device *brnf_get_logical_dev(struct sk_buff *skb, const struct net_device *dev)
475{
476	struct net_device *vlan, *br;
477
478	br = bridge_parent(dev);
479	if (brnf_pass_vlan_indev == 0 || !skb_vlan_tag_present(skb))
480		return br;
481
482	vlan = __vlan_find_dev_deep_rcu(br, skb->vlan_proto,
483				    skb_vlan_tag_get(skb) & VLAN_VID_MASK);
484
485	return vlan ? vlan : br;
486}
487
488/* Some common code for IPv4/IPv6 */
489static struct net_device *setup_pre_routing(struct sk_buff *skb)
490{
491	struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
492
493	if (skb->pkt_type == PACKET_OTHERHOST) {
494		skb->pkt_type = PACKET_HOST;
495		nf_bridge->pkt_otherhost = true;
496	}
497
498	nf_bridge->mask |= BRNF_NF_BRIDGE_PREROUTING;
499	nf_bridge->physindev = skb->dev;
500	skb->dev = brnf_get_logical_dev(skb, skb->dev);
501
502	if (skb->protocol == htons(ETH_P_8021Q))
503		nf_bridge->orig_proto = BRNF_PROTO_8021Q;
504	else if (skb->protocol == htons(ETH_P_PPP_SES))
505		nf_bridge->orig_proto = BRNF_PROTO_PPPOE;
506
507	/* Must drop socket now because of tproxy. */
508	skb_orphan(skb);
509	return skb->dev;
510}
511
512/* We only check the length. A bridge shouldn't do any hop-by-hop stuff anyway */
513static int check_hbh_len(struct sk_buff *skb)
514{
515	unsigned char *raw = (u8 *)(ipv6_hdr(skb) + 1);
516	u32 pkt_len;
517	const unsigned char *nh = skb_network_header(skb);
518	int off = raw - nh;
519	int len = (raw[1] + 1) << 3;
520
521	if ((raw + len) - skb->data > skb_headlen(skb))
522		goto bad;
523
524	off += 2;
525	len -= 2;
526
527	while (len > 0) {
528		int optlen = nh[off + 1] + 2;
529
530		switch (nh[off]) {
531		case IPV6_TLV_PAD1:
532			optlen = 1;
533			break;
534
535		case IPV6_TLV_PADN:
536			break;
537
538		case IPV6_TLV_JUMBO:
539			if (nh[off + 1] != 4 || (off & 3) != 2)
540				goto bad;
541			pkt_len = ntohl(*(__be32 *) (nh + off + 2));
542			if (pkt_len <= IPV6_MAXPLEN ||
543			    ipv6_hdr(skb)->payload_len)
544				goto bad;
545			if (pkt_len > skb->len - sizeof(struct ipv6hdr))
546				goto bad;
547			if (pskb_trim_rcsum(skb,
548					    pkt_len + sizeof(struct ipv6hdr)))
549				goto bad;
550			nh = skb_network_header(skb);
551			break;
552		default:
553			if (optlen > len)
554				goto bad;
555			break;
556		}
557		off += optlen;
558		len -= optlen;
559	}
560	if (len == 0)
561		return 0;
562bad:
563	return -1;
564
565}
566
567/* Replicate the checks that IPv6 does on packet reception and pass the packet
568 * to ip6tables, which doesn't support NAT, so things are fairly simple. */
569static unsigned int br_nf_pre_routing_ipv6(const struct nf_hook_ops *ops,
570					   struct sk_buff *skb,
571					   const struct nf_hook_state *state)
572{
573	const struct ipv6hdr *hdr;
574	u32 pkt_len;
575
576	if (skb->len < sizeof(struct ipv6hdr))
577		return NF_DROP;
578
579	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
580		return NF_DROP;
581
582	hdr = ipv6_hdr(skb);
583
584	if (hdr->version != 6)
585		return NF_DROP;
586
587	pkt_len = ntohs(hdr->payload_len);
588
589	if (pkt_len || hdr->nexthdr != NEXTHDR_HOP) {
590		if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
591			return NF_DROP;
592		if (pskb_trim_rcsum(skb, pkt_len + sizeof(struct ipv6hdr)))
593			return NF_DROP;
594	}
595	if (hdr->nexthdr == NEXTHDR_HOP && check_hbh_len(skb))
596		return NF_DROP;
597
598	nf_bridge_put(skb->nf_bridge);
599	if (!nf_bridge_alloc(skb))
600		return NF_DROP;
601	if (!setup_pre_routing(skb))
602		return NF_DROP;
603
604	skb->protocol = htons(ETH_P_IPV6);
605	NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING, state->sk, skb,
606		skb->dev, NULL,
607		br_nf_pre_routing_finish_ipv6);
608
609	return NF_STOLEN;
610}
611
612/* Direct IPv6 traffic to br_nf_pre_routing_ipv6.
613 * Replicate the checks that IPv4 does on packet reception.
614 * Set skb->dev to the bridge device (i.e. parent of the
615 * receiving device) to make netfilter happy, the REDIRECT
616 * target in particular.  Save the original destination IP
617 * address to be able to detect DNAT afterwards. */
618static unsigned int br_nf_pre_routing(const struct nf_hook_ops *ops,
619				      struct sk_buff *skb,
620				      const struct nf_hook_state *state)
621{
622	struct nf_bridge_info *nf_bridge;
623	struct net_bridge_port *p;
624	struct net_bridge *br;
625	__u32 len = nf_bridge_encap_header_len(skb);
626
627	if (unlikely(!pskb_may_pull(skb, len)))
628		return NF_DROP;
629
630	p = br_port_get_rcu(state->in);
631	if (p == NULL)
632		return NF_DROP;
633	br = p->br;
634
635	if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb)) {
636		if (!brnf_call_ip6tables && !br->nf_call_ip6tables)
637			return NF_ACCEPT;
638
639		nf_bridge_pull_encap_header_rcsum(skb);
640		return br_nf_pre_routing_ipv6(ops, skb, state);
641	}
642
643	if (!brnf_call_iptables && !br->nf_call_iptables)
644		return NF_ACCEPT;
645
646	if (!IS_IP(skb) && !IS_VLAN_IP(skb) && !IS_PPPOE_IP(skb))
647		return NF_ACCEPT;
648
649	nf_bridge_pull_encap_header_rcsum(skb);
650
651	if (br_parse_ip_options(skb))
652		return NF_DROP;
653
654	nf_bridge_put(skb->nf_bridge);
655	if (!nf_bridge_alloc(skb))
656		return NF_DROP;
657	if (!setup_pre_routing(skb))
658		return NF_DROP;
659
660	nf_bridge = nf_bridge_info_get(skb);
661	nf_bridge->ipv4_daddr = ip_hdr(skb)->daddr;
662
663	skb->protocol = htons(ETH_P_IP);
664
665	NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING, state->sk, skb,
666		skb->dev, NULL,
667		br_nf_pre_routing_finish);
668
669	return NF_STOLEN;
670}
671
672
673/* PF_BRIDGE/LOCAL_IN ************************************************/
674/* The packet is locally destined, which requires a real
675 * dst_entry, so detach the fake one.  On the way up, the
676 * packet would pass through PRE_ROUTING again (which already
677 * took place when the packet entered the bridge), but we
678 * register an IPv4 PRE_ROUTING 'sabotage' hook that will
679 * prevent this from happening. */
680static unsigned int br_nf_local_in(const struct nf_hook_ops *ops,
681				   struct sk_buff *skb,
682				   const struct nf_hook_state *state)
683{
684	br_drop_fake_rtable(skb);
685	return NF_ACCEPT;
686}
687
688/* PF_BRIDGE/FORWARD *************************************************/
689static int br_nf_forward_finish(struct sock *sk, struct sk_buff *skb)
690{
691	struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
692	struct net_device *in;
693
694	if (!IS_ARP(skb) && !IS_VLAN_ARP(skb)) {
695		int frag_max_size;
696
697		if (skb->protocol == htons(ETH_P_IP)) {
698			frag_max_size = IPCB(skb)->frag_max_size;
699			BR_INPUT_SKB_CB(skb)->frag_max_size = frag_max_size;
700		}
701
702		in = nf_bridge->physindev;
703		if (nf_bridge->pkt_otherhost) {
704			skb->pkt_type = PACKET_OTHERHOST;
705			nf_bridge->pkt_otherhost = false;
706		}
707		nf_bridge_update_protocol(skb);
708	} else {
709		in = *((struct net_device **)(skb->cb));
710	}
711	nf_bridge_push_encap_header(skb);
712
713	NF_HOOK_THRESH(NFPROTO_BRIDGE, NF_BR_FORWARD, sk, skb,
714		       in, skb->dev, br_forward_finish, 1);
715	return 0;
716}
717
718
719/* This is the 'purely bridged' case.  For IP, we pass the packet to
720 * netfilter with indev and outdev set to the bridge device,
721 * but we are still able to filter on the 'real' indev/outdev
722 * because of the physdev module. For ARP, indev and outdev are the
723 * bridge ports. */
724static unsigned int br_nf_forward_ip(const struct nf_hook_ops *ops,
725				     struct sk_buff *skb,
726				     const struct nf_hook_state *state)
727{
728	struct nf_bridge_info *nf_bridge;
729	struct net_device *parent;
730	u_int8_t pf;
731
732	if (!skb->nf_bridge)
733		return NF_ACCEPT;
734
735	/* Need exclusive nf_bridge_info since we might have multiple
736	 * different physoutdevs. */
737	if (!nf_bridge_unshare(skb))
738		return NF_DROP;
739
740	nf_bridge = nf_bridge_info_get(skb);
741	if (!nf_bridge)
742		return NF_DROP;
743
744	parent = bridge_parent(state->out);
745	if (!parent)
746		return NF_DROP;
747
748	if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
749		pf = NFPROTO_IPV4;
750	else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
751		pf = NFPROTO_IPV6;
752	else
753		return NF_ACCEPT;
754
755	nf_bridge_pull_encap_header(skb);
756
757	if (skb->pkt_type == PACKET_OTHERHOST) {
758		skb->pkt_type = PACKET_HOST;
759		nf_bridge->pkt_otherhost = true;
760	}
761
762	if (pf == NFPROTO_IPV4) {
763		int frag_max = BR_INPUT_SKB_CB(skb)->frag_max_size;
764
765		if (br_parse_ip_options(skb))
766			return NF_DROP;
767
768		IPCB(skb)->frag_max_size = frag_max;
769	}
770
771	nf_bridge->physoutdev = skb->dev;
772	if (pf == NFPROTO_IPV4)
773		skb->protocol = htons(ETH_P_IP);
774	else
775		skb->protocol = htons(ETH_P_IPV6);
776
777	NF_HOOK(pf, NF_INET_FORWARD, NULL, skb,
778		brnf_get_logical_dev(skb, state->in),
779		parent,	br_nf_forward_finish);
780
781	return NF_STOLEN;
782}
783
784static unsigned int br_nf_forward_arp(const struct nf_hook_ops *ops,
785				      struct sk_buff *skb,
786				      const struct nf_hook_state *state)
787{
788	struct net_bridge_port *p;
789	struct net_bridge *br;
790	struct net_device **d = (struct net_device **)(skb->cb);
791
792	p = br_port_get_rcu(state->out);
793	if (p == NULL)
794		return NF_ACCEPT;
795	br = p->br;
796
797	if (!brnf_call_arptables && !br->nf_call_arptables)
798		return NF_ACCEPT;
799
800	if (!IS_ARP(skb)) {
801		if (!IS_VLAN_ARP(skb))
802			return NF_ACCEPT;
803		nf_bridge_pull_encap_header(skb);
804	}
805
806	if (arp_hdr(skb)->ar_pln != 4) {
807		if (IS_VLAN_ARP(skb))
808			nf_bridge_push_encap_header(skb);
809		return NF_ACCEPT;
810	}
811	*d = state->in;
812	NF_HOOK(NFPROTO_ARP, NF_ARP_FORWARD, state->sk, skb,
813		state->in, state->out, br_nf_forward_finish);
814
815	return NF_STOLEN;
816}
817
818#if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4)
819static int br_nf_push_frag_xmit(struct sock *sk, struct sk_buff *skb)
820{
821	struct brnf_frag_data *data;
822	int err;
823
824	data = this_cpu_ptr(&brnf_frag_data_storage);
825	err = skb_cow_head(skb, data->size);
826
827	if (err) {
828		kfree_skb(skb);
829		return 0;
830	}
831
832	skb_copy_to_linear_data_offset(skb, -data->size, data->mac, data->size);
833	__skb_push(skb, data->encap_size);
834
835	return br_dev_queue_push_xmit(sk, skb);
836}
837
838static int br_nf_dev_queue_xmit(struct sock *sk, struct sk_buff *skb)
839{
840	int ret;
841	int frag_max_size;
842	unsigned int mtu_reserved;
843
844	if (skb_is_gso(skb) || skb->protocol != htons(ETH_P_IP))
845		return br_dev_queue_push_xmit(sk, skb);
846
847	mtu_reserved = nf_bridge_mtu_reduction(skb);
848	/* This is wrong! We should preserve the original fragment
849	 * boundaries by preserving frag_list rather than refragmenting.
850	 */
851	if (skb->len + mtu_reserved > skb->dev->mtu) {
852		struct brnf_frag_data *data;
853
854		frag_max_size = BR_INPUT_SKB_CB(skb)->frag_max_size;
855		if (br_parse_ip_options(skb))
856			/* Drop invalid packet */
857			return NF_DROP;
858		IPCB(skb)->frag_max_size = frag_max_size;
859
860		nf_bridge_update_protocol(skb);
861
862		data = this_cpu_ptr(&brnf_frag_data_storage);
863		data->encap_size = nf_bridge_encap_header_len(skb);
864		data->size = ETH_HLEN + data->encap_size;
865
866		skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
867						 data->size);
868
869		ret = ip_fragment(sk, skb, br_nf_push_frag_xmit);
870	} else {
871		ret = br_dev_queue_push_xmit(sk, skb);
872	}
873
874	return ret;
875}
876#else
877static int br_nf_dev_queue_xmit(struct sock *sk, struct sk_buff *skb)
878{
879        return br_dev_queue_push_xmit(sk, skb);
880}
881#endif
882
883/* PF_BRIDGE/POST_ROUTING ********************************************/
884static unsigned int br_nf_post_routing(const struct nf_hook_ops *ops,
885				       struct sk_buff *skb,
886				       const struct nf_hook_state *state)
887{
888	struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
889	struct net_device *realoutdev = bridge_parent(skb->dev);
890	u_int8_t pf;
891
892	/* if nf_bridge is set, but ->physoutdev is NULL, this packet came in
893	 * on a bridge, but was delivered locally and is now being routed:
894	 *
895	 * POST_ROUTING was already invoked from the ip stack.
896	 */
897	if (!nf_bridge || !nf_bridge->physoutdev)
898		return NF_ACCEPT;
899
900	if (!realoutdev)
901		return NF_DROP;
902
903	if (IS_IP(skb) || IS_VLAN_IP(skb) || IS_PPPOE_IP(skb))
904		pf = NFPROTO_IPV4;
905	else if (IS_IPV6(skb) || IS_VLAN_IPV6(skb) || IS_PPPOE_IPV6(skb))
906		pf = NFPROTO_IPV6;
907	else
908		return NF_ACCEPT;
909
910	/* We assume any code from br_dev_queue_push_xmit onwards doesn't care
911	 * about the value of skb->pkt_type. */
912	if (skb->pkt_type == PACKET_OTHERHOST) {
913		skb->pkt_type = PACKET_HOST;
914		nf_bridge->pkt_otherhost = true;
915	}
916
917	nf_bridge_pull_encap_header(skb);
918	if (pf == NFPROTO_IPV4)
919		skb->protocol = htons(ETH_P_IP);
920	else
921		skb->protocol = htons(ETH_P_IPV6);
922
923	NF_HOOK(pf, NF_INET_POST_ROUTING, state->sk, skb,
924		NULL, realoutdev,
925		br_nf_dev_queue_xmit);
926
927	return NF_STOLEN;
928}
929
930/* IP/SABOTAGE *****************************************************/
931/* Don't hand locally destined packets to PF_INET(6)/PRE_ROUTING
932 * for the second time. */
933static unsigned int ip_sabotage_in(const struct nf_hook_ops *ops,
934				   struct sk_buff *skb,
935				   const struct nf_hook_state *state)
936{
937	if (skb->nf_bridge &&
938	    !(skb->nf_bridge->mask & BRNF_NF_BRIDGE_PREROUTING)) {
939		return NF_STOP;
940	}
941
942	return NF_ACCEPT;
943}
944
945/* This is called when br_netfilter has called into iptables/netfilter,
946 * and DNAT has taken place on a bridge-forwarded packet.
947 *
948 * neigh->output has created a new MAC header, with local br0 MAC
949 * as saddr.
950 *
951 * This restores the original MAC saddr of the bridged packet
952 * before invoking bridge forward logic to transmit the packet.
953 */
954static void br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
955{
956	struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
957
958	skb_pull(skb, ETH_HLEN);
959	nf_bridge->mask &= ~BRNF_BRIDGED_DNAT;
960
961	BUILD_BUG_ON(sizeof(nf_bridge->neigh_header) != (ETH_HLEN - ETH_ALEN));
962
963	skb_copy_to_linear_data_offset(skb, -(ETH_HLEN - ETH_ALEN),
964				       nf_bridge->neigh_header,
965				       ETH_HLEN - ETH_ALEN);
966	skb->dev = nf_bridge->physindev;
967	br_handle_frame_finish(NULL, skb);
968}
969
970static int br_nf_dev_xmit(struct sk_buff *skb)
971{
972	if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
973		br_nf_pre_routing_finish_bridge_slow(skb);
974		return 1;
975	}
976	return 0;
977}
978
979static const struct nf_br_ops br_ops = {
980	.br_dev_xmit_hook =	br_nf_dev_xmit,
981};
982
983void br_netfilter_enable(void)
984{
985}
986EXPORT_SYMBOL_GPL(br_netfilter_enable);
987
988/* For br_nf_post_routing, we need (prio = NF_BR_PRI_LAST), because
989 * br_dev_queue_push_xmit is called afterwards */
990static struct nf_hook_ops br_nf_ops[] __read_mostly = {
991	{
992		.hook = br_nf_pre_routing,
993		.owner = THIS_MODULE,
994		.pf = NFPROTO_BRIDGE,
995		.hooknum = NF_BR_PRE_ROUTING,
996		.priority = NF_BR_PRI_BRNF,
997	},
998	{
999		.hook = br_nf_local_in,
1000		.owner = THIS_MODULE,
1001		.pf = NFPROTO_BRIDGE,
1002		.hooknum = NF_BR_LOCAL_IN,
1003		.priority = NF_BR_PRI_BRNF,
1004	},
1005	{
1006		.hook = br_nf_forward_ip,
1007		.owner = THIS_MODULE,
1008		.pf = NFPROTO_BRIDGE,
1009		.hooknum = NF_BR_FORWARD,
1010		.priority = NF_BR_PRI_BRNF - 1,
1011	},
1012	{
1013		.hook = br_nf_forward_arp,
1014		.owner = THIS_MODULE,
1015		.pf = NFPROTO_BRIDGE,
1016		.hooknum = NF_BR_FORWARD,
1017		.priority = NF_BR_PRI_BRNF,
1018	},
1019	{
1020		.hook = br_nf_post_routing,
1021		.owner = THIS_MODULE,
1022		.pf = NFPROTO_BRIDGE,
1023		.hooknum = NF_BR_POST_ROUTING,
1024		.priority = NF_BR_PRI_LAST,
1025	},
1026	{
1027		.hook = ip_sabotage_in,
1028		.owner = THIS_MODULE,
1029		.pf = NFPROTO_IPV4,
1030		.hooknum = NF_INET_PRE_ROUTING,
1031		.priority = NF_IP_PRI_FIRST,
1032	},
1033	{
1034		.hook = ip_sabotage_in,
1035		.owner = THIS_MODULE,
1036		.pf = NFPROTO_IPV6,
1037		.hooknum = NF_INET_PRE_ROUTING,
1038		.priority = NF_IP6_PRI_FIRST,
1039	},
1040};
1041
1042#ifdef CONFIG_SYSCTL
1043static
1044int brnf_sysctl_call_tables(struct ctl_table *ctl, int write,
1045			    void __user *buffer, size_t *lenp, loff_t *ppos)
1046{
1047	int ret;
1048
1049	ret = proc_dointvec(ctl, write, buffer, lenp, ppos);
1050
1051	if (write && *(int *)(ctl->data))
1052		*(int *)(ctl->data) = 1;
1053	return ret;
1054}
1055
1056static struct ctl_table brnf_table[] = {
1057	{
1058		.procname	= "bridge-nf-call-arptables",
1059		.data		= &brnf_call_arptables,
1060		.maxlen		= sizeof(int),
1061		.mode		= 0644,
1062		.proc_handler	= brnf_sysctl_call_tables,
1063	},
1064	{
1065		.procname	= "bridge-nf-call-iptables",
1066		.data		= &brnf_call_iptables,
1067		.maxlen		= sizeof(int),
1068		.mode		= 0644,
1069		.proc_handler	= brnf_sysctl_call_tables,
1070	},
1071	{
1072		.procname	= "bridge-nf-call-ip6tables",
1073		.data		= &brnf_call_ip6tables,
1074		.maxlen		= sizeof(int),
1075		.mode		= 0644,
1076		.proc_handler	= brnf_sysctl_call_tables,
1077	},
1078	{
1079		.procname	= "bridge-nf-filter-vlan-tagged",
1080		.data		= &brnf_filter_vlan_tagged,
1081		.maxlen		= sizeof(int),
1082		.mode		= 0644,
1083		.proc_handler	= brnf_sysctl_call_tables,
1084	},
1085	{
1086		.procname	= "bridge-nf-filter-pppoe-tagged",
1087		.data		= &brnf_filter_pppoe_tagged,
1088		.maxlen		= sizeof(int),
1089		.mode		= 0644,
1090		.proc_handler	= brnf_sysctl_call_tables,
1091	},
1092	{
1093		.procname	= "bridge-nf-pass-vlan-input-dev",
1094		.data		= &brnf_pass_vlan_indev,
1095		.maxlen		= sizeof(int),
1096		.mode		= 0644,
1097		.proc_handler	= brnf_sysctl_call_tables,
1098	},
1099	{ }
1100};
1101#endif
1102
1103static int __init br_netfilter_init(void)
1104{
1105	int ret;
1106
1107	ret = nf_register_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1108	if (ret < 0)
1109		return ret;
1110
1111#ifdef CONFIG_SYSCTL
1112	brnf_sysctl_header = register_net_sysctl(&init_net, "net/bridge", brnf_table);
1113	if (brnf_sysctl_header == NULL) {
1114		printk(KERN_WARNING
1115		       "br_netfilter: can't register to sysctl.\n");
1116		nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1117		return -ENOMEM;
1118	}
1119#endif
1120	RCU_INIT_POINTER(nf_br_ops, &br_ops);
1121	printk(KERN_NOTICE "Bridge firewalling registered\n");
1122	return 0;
1123}
1124
1125static void __exit br_netfilter_fini(void)
1126{
1127	RCU_INIT_POINTER(nf_br_ops, NULL);
1128	nf_unregister_hooks(br_nf_ops, ARRAY_SIZE(br_nf_ops));
1129#ifdef CONFIG_SYSCTL
1130	unregister_net_sysctl_table(brnf_sysctl_header);
1131#endif
1132}
1133
1134module_init(br_netfilter_init);
1135module_exit(br_netfilter_fini);
1136
1137MODULE_LICENSE("GPL");
1138MODULE_AUTHOR("Lennert Buytenhek <buytenh@gnu.org>");
1139MODULE_AUTHOR("Bart De Schuymer <bdschuym@pandora.be>");
1140MODULE_DESCRIPTION("Linux ethernet netfilter firewall bridge");
1141