1 /*
2  * Copyright (c) 2014 Pablo Neira Ayuso <pablo@netfilter.org>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  */
8 
9 #include <linux/kernel.h>
10 #include <linux/init.h>
11 #include <linux/module.h>
12 #include <linux/netlink.h>
13 #include <linux/netfilter.h>
14 #include <linux/netfilter/nf_tables.h>
15 #include <net/netfilter/nf_tables.h>
16 #include <net/netfilter/nft_reject.h>
17 #include <net/netfilter/ipv4/nf_reject.h>
18 #include <net/netfilter/ipv6/nf_reject.h>
19 #include <linux/ip.h>
20 #include <net/ip.h>
21 #include <net/ip6_checksum.h>
22 #include <linux/netfilter_bridge.h>
23 #include <linux/netfilter_ipv6.h>
24 #include "../br_private.h"
25 
26 static void nft_reject_br_push_etherhdr(struct sk_buff *oldskb,
27 					struct sk_buff *nskb)
28 {
29 	struct ethhdr *eth;
30 
31 	eth = skb_push(nskb, ETH_HLEN);
32 	skb_reset_mac_header(nskb);
33 	ether_addr_copy(eth->h_source, eth_hdr(oldskb)->h_dest);
34 	ether_addr_copy(eth->h_dest, eth_hdr(oldskb)->h_source);
35 	eth->h_proto = eth_hdr(oldskb)->h_proto;
36 	skb_pull(nskb, ETH_HLEN);
37 }
38 
39 static int nft_bridge_iphdr_validate(struct sk_buff *skb)
40 {
41 	struct iphdr *iph;
42 	u32 len;
43 
44 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
45 		return 0;
46 
47 	iph = ip_hdr(skb);
48 	if (iph->ihl < 5 || iph->version != 4)
49 		return 0;
50 
51 	len = ntohs(iph->tot_len);
52 	if (skb->len < len)
53 		return 0;
54 	else if (len < (iph->ihl*4))
55 		return 0;
56 
57 	if (!pskb_may_pull(skb, iph->ihl*4))
58 		return 0;
59 
60 	return 1;
61 }
62 
63 /* We cannot use oldskb->dev, it can be either bridge device (NF_BRIDGE INPUT)
64  * or the bridge port (NF_BRIDGE PREROUTING).
65  */
66 static void nft_reject_br_send_v4_tcp_reset(struct net *net,
67 					    struct sk_buff *oldskb,
68 					    const struct net_device *dev,
69 					    int hook)
70 {
71 	struct sk_buff *nskb;
72 	struct iphdr *niph;
73 	const struct tcphdr *oth;
74 	struct tcphdr _oth;
75 
76 	if (!nft_bridge_iphdr_validate(oldskb))
77 		return;
78 
79 	oth = nf_reject_ip_tcphdr_get(oldskb, &_oth, hook);
80 	if (!oth)
81 		return;
82 
83 	nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct tcphdr) +
84 			 LL_MAX_HEADER, GFP_ATOMIC);
85 	if (!nskb)
86 		return;
87 
88 	skb_reserve(nskb, LL_MAX_HEADER);
89 	niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_TCP,
90 				   net->ipv4.sysctl_ip_default_ttl);
91 	nf_reject_ip_tcphdr_put(nskb, oldskb, oth);
92 	niph->tot_len = htons(nskb->len);
93 	ip_send_check(niph);
94 
95 	nft_reject_br_push_etherhdr(oldskb, nskb);
96 
97 	br_forward(br_port_get_rcu(dev), nskb, false, true);
98 }
99 
100 static void nft_reject_br_send_v4_unreach(struct net *net,
101 					  struct sk_buff *oldskb,
102 					  const struct net_device *dev,
103 					  int hook, u8 code)
104 {
105 	struct sk_buff *nskb;
106 	struct iphdr *niph;
107 	struct icmphdr *icmph;
108 	unsigned int len;
109 	__wsum csum;
110 	u8 proto;
111 
112 	if (!nft_bridge_iphdr_validate(oldskb))
113 		return;
114 
115 	/* IP header checks: fragment. */
116 	if (ip_hdr(oldskb)->frag_off & htons(IP_OFFSET))
117 		return;
118 
119 	/* RFC says return as much as we can without exceeding 576 bytes. */
120 	len = min_t(unsigned int, 536, oldskb->len);
121 
122 	if (!pskb_may_pull(oldskb, len))
123 		return;
124 
125 	if (pskb_trim_rcsum(oldskb, ntohs(ip_hdr(oldskb)->tot_len)))
126 		return;
127 
128 	if (ip_hdr(oldskb)->protocol == IPPROTO_TCP ||
129 	    ip_hdr(oldskb)->protocol == IPPROTO_UDP)
130 		proto = ip_hdr(oldskb)->protocol;
131 	else
132 		proto = 0;
133 
134 	if (!skb_csum_unnecessary(oldskb) &&
135 	    nf_ip_checksum(oldskb, hook, ip_hdrlen(oldskb), proto))
136 		return;
137 
138 	nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct icmphdr) +
139 			 LL_MAX_HEADER + len, GFP_ATOMIC);
140 	if (!nskb)
141 		return;
142 
143 	skb_reserve(nskb, LL_MAX_HEADER);
144 	niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_ICMP,
145 				   net->ipv4.sysctl_ip_default_ttl);
146 
147 	skb_reset_transport_header(nskb);
148 	icmph = skb_put_zero(nskb, sizeof(struct icmphdr));
149 	icmph->type     = ICMP_DEST_UNREACH;
150 	icmph->code	= code;
151 
152 	skb_put_data(nskb, skb_network_header(oldskb), len);
153 
154 	csum = csum_partial((void *)icmph, len + sizeof(struct icmphdr), 0);
155 	icmph->checksum = csum_fold(csum);
156 
157 	niph->tot_len	= htons(nskb->len);
158 	ip_send_check(niph);
159 
160 	nft_reject_br_push_etherhdr(oldskb, nskb);
161 
162 	br_forward(br_port_get_rcu(dev), nskb, false, true);
163 }
164 
165 static int nft_bridge_ip6hdr_validate(struct sk_buff *skb)
166 {
167 	struct ipv6hdr *hdr;
168 	u32 pkt_len;
169 
170 	if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
171 		return 0;
172 
173 	hdr = ipv6_hdr(skb);
174 	if (hdr->version != 6)
175 		return 0;
176 
177 	pkt_len = ntohs(hdr->payload_len);
178 	if (pkt_len + sizeof(struct ipv6hdr) > skb->len)
179 		return 0;
180 
181 	return 1;
182 }
183 
184 static void nft_reject_br_send_v6_tcp_reset(struct net *net,
185 					    struct sk_buff *oldskb,
186 					    const struct net_device *dev,
187 					    int hook)
188 {
189 	struct sk_buff *nskb;
190 	const struct tcphdr *oth;
191 	struct tcphdr _oth;
192 	unsigned int otcplen;
193 	struct ipv6hdr *nip6h;
194 
195 	if (!nft_bridge_ip6hdr_validate(oldskb))
196 		return;
197 
198 	oth = nf_reject_ip6_tcphdr_get(oldskb, &_oth, &otcplen, hook);
199 	if (!oth)
200 		return;
201 
202 	nskb = alloc_skb(sizeof(struct ipv6hdr) + sizeof(struct tcphdr) +
203 			 LL_MAX_HEADER, GFP_ATOMIC);
204 	if (!nskb)
205 		return;
206 
207 	skb_reserve(nskb, LL_MAX_HEADER);
208 	nip6h = nf_reject_ip6hdr_put(nskb, oldskb, IPPROTO_TCP,
209 				     net->ipv6.devconf_all->hop_limit);
210 	nf_reject_ip6_tcphdr_put(nskb, oldskb, oth, otcplen);
211 	nip6h->payload_len = htons(nskb->len - sizeof(struct ipv6hdr));
212 
213 	nft_reject_br_push_etherhdr(oldskb, nskb);
214 
215 	br_forward(br_port_get_rcu(dev), nskb, false, true);
216 }
217 
218 static bool reject6_br_csum_ok(struct sk_buff *skb, int hook)
219 {
220 	const struct ipv6hdr *ip6h = ipv6_hdr(skb);
221 	int thoff;
222 	__be16 fo;
223 	u8 proto = ip6h->nexthdr;
224 
225 	if (skb_csum_unnecessary(skb))
226 		return true;
227 
228 	if (ip6h->payload_len &&
229 	    pskb_trim_rcsum(skb, ntohs(ip6h->payload_len) + sizeof(*ip6h)))
230 		return false;
231 
232 	thoff = ipv6_skip_exthdr(skb, ((u8*)(ip6h+1) - skb->data), &proto, &fo);
233 	if (thoff < 0 || thoff >= skb->len || (fo & htons(~0x7)) != 0)
234 		return false;
235 
236 	return nf_ip6_checksum(skb, hook, thoff, proto) == 0;
237 }
238 
239 static void nft_reject_br_send_v6_unreach(struct net *net,
240 					  struct sk_buff *oldskb,
241 					  const struct net_device *dev,
242 					  int hook, u8 code)
243 {
244 	struct sk_buff *nskb;
245 	struct ipv6hdr *nip6h;
246 	struct icmp6hdr *icmp6h;
247 	unsigned int len;
248 
249 	if (!nft_bridge_ip6hdr_validate(oldskb))
250 		return;
251 
252 	/* Include "As much of invoking packet as possible without the ICMPv6
253 	 * packet exceeding the minimum IPv6 MTU" in the ICMP payload.
254 	 */
255 	len = min_t(unsigned int, 1220, oldskb->len);
256 
257 	if (!pskb_may_pull(oldskb, len))
258 		return;
259 
260 	if (!reject6_br_csum_ok(oldskb, hook))
261 		return;
262 
263 	nskb = alloc_skb(sizeof(struct ipv6hdr) + sizeof(struct icmp6hdr) +
264 			 LL_MAX_HEADER + len, GFP_ATOMIC);
265 	if (!nskb)
266 		return;
267 
268 	skb_reserve(nskb, LL_MAX_HEADER);
269 	nip6h = nf_reject_ip6hdr_put(nskb, oldskb, IPPROTO_ICMPV6,
270 				     net->ipv6.devconf_all->hop_limit);
271 
272 	skb_reset_transport_header(nskb);
273 	icmp6h = skb_put_zero(nskb, sizeof(struct icmp6hdr));
274 	icmp6h->icmp6_type = ICMPV6_DEST_UNREACH;
275 	icmp6h->icmp6_code = code;
276 
277 	skb_put_data(nskb, skb_network_header(oldskb), len);
278 	nip6h->payload_len = htons(nskb->len - sizeof(struct ipv6hdr));
279 
280 	icmp6h->icmp6_cksum =
281 		csum_ipv6_magic(&nip6h->saddr, &nip6h->daddr,
282 				nskb->len - sizeof(struct ipv6hdr),
283 				IPPROTO_ICMPV6,
284 				csum_partial(icmp6h,
285 					     nskb->len - sizeof(struct ipv6hdr),
286 					     0));
287 
288 	nft_reject_br_push_etherhdr(oldskb, nskb);
289 
290 	br_forward(br_port_get_rcu(dev), nskb, false, true);
291 }
292 
293 static void nft_reject_bridge_eval(const struct nft_expr *expr,
294 				   struct nft_regs *regs,
295 				   const struct nft_pktinfo *pkt)
296 {
297 	struct nft_reject *priv = nft_expr_priv(expr);
298 	const unsigned char *dest = eth_hdr(pkt->skb)->h_dest;
299 
300 	if (is_broadcast_ether_addr(dest) ||
301 	    is_multicast_ether_addr(dest))
302 		goto out;
303 
304 	switch (eth_hdr(pkt->skb)->h_proto) {
305 	case htons(ETH_P_IP):
306 		switch (priv->type) {
307 		case NFT_REJECT_ICMP_UNREACH:
308 			nft_reject_br_send_v4_unreach(nft_net(pkt), pkt->skb,
309 						      nft_in(pkt),
310 						      nft_hook(pkt),
311 						      priv->icmp_code);
312 			break;
313 		case NFT_REJECT_TCP_RST:
314 			nft_reject_br_send_v4_tcp_reset(nft_net(pkt), pkt->skb,
315 							nft_in(pkt),
316 							nft_hook(pkt));
317 			break;
318 		case NFT_REJECT_ICMPX_UNREACH:
319 			nft_reject_br_send_v4_unreach(nft_net(pkt), pkt->skb,
320 						      nft_in(pkt),
321 						      nft_hook(pkt),
322 						      nft_reject_icmp_code(priv->icmp_code));
323 			break;
324 		}
325 		break;
326 	case htons(ETH_P_IPV6):
327 		switch (priv->type) {
328 		case NFT_REJECT_ICMP_UNREACH:
329 			nft_reject_br_send_v6_unreach(nft_net(pkt), pkt->skb,
330 						      nft_in(pkt),
331 						      nft_hook(pkt),
332 						      priv->icmp_code);
333 			break;
334 		case NFT_REJECT_TCP_RST:
335 			nft_reject_br_send_v6_tcp_reset(nft_net(pkt), pkt->skb,
336 							nft_in(pkt),
337 							nft_hook(pkt));
338 			break;
339 		case NFT_REJECT_ICMPX_UNREACH:
340 			nft_reject_br_send_v6_unreach(nft_net(pkt), pkt->skb,
341 						      nft_in(pkt),
342 						      nft_hook(pkt),
343 						      nft_reject_icmpv6_code(priv->icmp_code));
344 			break;
345 		}
346 		break;
347 	default:
348 		/* No explicit way to reject this protocol, drop it. */
349 		break;
350 	}
351 out:
352 	regs->verdict.code = NF_DROP;
353 }
354 
355 static int nft_reject_bridge_validate(const struct nft_ctx *ctx,
356 				      const struct nft_expr *expr,
357 				      const struct nft_data **data)
358 {
359 	return nft_chain_validate_hooks(ctx->chain, (1 << NF_BR_PRE_ROUTING) |
360 						    (1 << NF_BR_LOCAL_IN));
361 }
362 
363 static int nft_reject_bridge_init(const struct nft_ctx *ctx,
364 				  const struct nft_expr *expr,
365 				  const struct nlattr * const tb[])
366 {
367 	struct nft_reject *priv = nft_expr_priv(expr);
368 	int icmp_code;
369 
370 	if (tb[NFTA_REJECT_TYPE] == NULL)
371 		return -EINVAL;
372 
373 	priv->type = ntohl(nla_get_be32(tb[NFTA_REJECT_TYPE]));
374 	switch (priv->type) {
375 	case NFT_REJECT_ICMP_UNREACH:
376 	case NFT_REJECT_ICMPX_UNREACH:
377 		if (tb[NFTA_REJECT_ICMP_CODE] == NULL)
378 			return -EINVAL;
379 
380 		icmp_code = nla_get_u8(tb[NFTA_REJECT_ICMP_CODE]);
381 		if (priv->type == NFT_REJECT_ICMPX_UNREACH &&
382 		    icmp_code > NFT_REJECT_ICMPX_MAX)
383 			return -EINVAL;
384 
385 		priv->icmp_code = icmp_code;
386 		break;
387 	case NFT_REJECT_TCP_RST:
388 		break;
389 	default:
390 		return -EINVAL;
391 	}
392 	return 0;
393 }
394 
395 static int nft_reject_bridge_dump(struct sk_buff *skb,
396 				  const struct nft_expr *expr)
397 {
398 	const struct nft_reject *priv = nft_expr_priv(expr);
399 
400 	if (nla_put_be32(skb, NFTA_REJECT_TYPE, htonl(priv->type)))
401 		goto nla_put_failure;
402 
403 	switch (priv->type) {
404 	case NFT_REJECT_ICMP_UNREACH:
405 	case NFT_REJECT_ICMPX_UNREACH:
406 		if (nla_put_u8(skb, NFTA_REJECT_ICMP_CODE, priv->icmp_code))
407 			goto nla_put_failure;
408 		break;
409 	default:
410 		break;
411 	}
412 
413 	return 0;
414 
415 nla_put_failure:
416 	return -1;
417 }
418 
419 static struct nft_expr_type nft_reject_bridge_type;
420 static const struct nft_expr_ops nft_reject_bridge_ops = {
421 	.type		= &nft_reject_bridge_type,
422 	.size		= NFT_EXPR_SIZE(sizeof(struct nft_reject)),
423 	.eval		= nft_reject_bridge_eval,
424 	.init		= nft_reject_bridge_init,
425 	.dump		= nft_reject_bridge_dump,
426 	.validate	= nft_reject_bridge_validate,
427 };
428 
429 static struct nft_expr_type nft_reject_bridge_type __read_mostly = {
430 	.family		= NFPROTO_BRIDGE,
431 	.name		= "reject",
432 	.ops		= &nft_reject_bridge_ops,
433 	.policy		= nft_reject_policy,
434 	.maxattr	= NFTA_REJECT_MAX,
435 	.owner		= THIS_MODULE,
436 };
437 
438 static int __init nft_reject_bridge_module_init(void)
439 {
440 	return nft_register_expr(&nft_reject_bridge_type);
441 }
442 
443 static void __exit nft_reject_bridge_module_exit(void)
444 {
445 	nft_unregister_expr(&nft_reject_bridge_type);
446 }
447 
448 module_init(nft_reject_bridge_module_init);
449 module_exit(nft_reject_bridge_module_exit);
450 
451 MODULE_LICENSE("GPL");
452 MODULE_AUTHOR("Pablo Neira Ayuso <pablo@netfilter.org>");
453 MODULE_ALIAS_NFT_AF_EXPR(AF_BRIDGE, "reject");
454