xref: /linux/net/ipv4/netfilter/nf_reject_ipv4.c (revision 2da68a77) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* (C) 1999-2001 Paul `Rusty' Russell
3  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
4  */
5 
6 #include <linux/module.h>
7 #include <net/ip.h>
8 #include <net/tcp.h>
9 #include <net/route.h>
10 #include <net/dst.h>
11 #include <net/netfilter/ipv4/nf_reject.h>
12 #include <linux/netfilter_ipv4.h>
13 #include <linux/netfilter_bridge.h>
14 
15 static int nf_reject_iphdr_validate(struct sk_buff *skb)
16 {
17 	struct iphdr *iph;
18 	u32 len;
19 
20 	if (!pskb_may_pull(skb, sizeof(struct iphdr)))
21 		return 0;
22 
23 	iph = ip_hdr(skb);
24 	if (iph->ihl < 5 || iph->version != 4)
25 		return 0;
26 
27 	len = ntohs(iph->tot_len);
28 	if (skb->len < len)
29 		return 0;
30 	else if (len < (iph->ihl*4))
31 		return 0;
32 
33 	if (!pskb_may_pull(skb, iph->ihl*4))
34 		return 0;
35 
36 	return 1;
37 }
38 
39 struct sk_buff *nf_reject_skb_v4_tcp_reset(struct net *net,
40 					   struct sk_buff *oldskb,
41 					   const struct net_device *dev,
42 					   int hook)
43 {
44 	const struct tcphdr *oth;
45 	struct sk_buff *nskb;
46 	struct iphdr *niph;
47 	struct tcphdr _oth;
48 
49 	if (!nf_reject_iphdr_validate(oldskb))
50 		return NULL;
51 
52 	oth = nf_reject_ip_tcphdr_get(oldskb, &_oth, hook);
53 	if (!oth)
54 		return NULL;
55 
56 	nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct tcphdr) +
57 			 LL_MAX_HEADER, GFP_ATOMIC);
58 	if (!nskb)
59 		return NULL;
60 
61 	nskb->dev = (struct net_device *)dev;
62 
63 	skb_reserve(nskb, LL_MAX_HEADER);
64 	niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_TCP,
65 				   READ_ONCE(net->ipv4.sysctl_ip_default_ttl));
66 	nf_reject_ip_tcphdr_put(nskb, oldskb, oth);
67 	niph->tot_len = htons(nskb->len);
68 	ip_send_check(niph);
69 
70 	return nskb;
71 }
72 EXPORT_SYMBOL_GPL(nf_reject_skb_v4_tcp_reset);
73 
74 struct sk_buff *nf_reject_skb_v4_unreach(struct net *net,
75 					 struct sk_buff *oldskb,
76 					 const struct net_device *dev,
77 					 int hook, u8 code)
78 {
79 	struct sk_buff *nskb;
80 	struct iphdr *niph;
81 	struct icmphdr *icmph;
82 	unsigned int len;
83 	int dataoff;
84 	__wsum csum;
85 	u8 proto;
86 
87 	if (!nf_reject_iphdr_validate(oldskb))
88 		return NULL;
89 
90 	/* IP header checks: fragment. */
91 	if (ip_hdr(oldskb)->frag_off & htons(IP_OFFSET))
92 		return NULL;
93 
94 	/* RFC says return as much as we can without exceeding 576 bytes. */
95 	len = min_t(unsigned int, 536, oldskb->len);
96 
97 	if (!pskb_may_pull(oldskb, len))
98 		return NULL;
99 
100 	if (pskb_trim_rcsum(oldskb, ntohs(ip_hdr(oldskb)->tot_len)))
101 		return NULL;
102 
103 	dataoff = ip_hdrlen(oldskb);
104 	proto = ip_hdr(oldskb)->protocol;
105 
106 	if (!skb_csum_unnecessary(oldskb) &&
107 	    nf_reject_verify_csum(oldskb, dataoff, proto) &&
108 	    nf_ip_checksum(oldskb, hook, ip_hdrlen(oldskb), proto))
109 		return NULL;
110 
111 	nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct icmphdr) +
112 			 LL_MAX_HEADER + len, GFP_ATOMIC);
113 	if (!nskb)
114 		return NULL;
115 
116 	nskb->dev = (struct net_device *)dev;
117 
118 	skb_reserve(nskb, LL_MAX_HEADER);
119 	niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_ICMP,
120 				   READ_ONCE(net->ipv4.sysctl_ip_default_ttl));
121 
122 	skb_reset_transport_header(nskb);
123 	icmph = skb_put_zero(nskb, sizeof(struct icmphdr));
124 	icmph->type     = ICMP_DEST_UNREACH;
125 	icmph->code	= code;
126 
127 	skb_put_data(nskb, skb_network_header(oldskb), len);
128 
129 	csum = csum_partial((void *)icmph, len + sizeof(struct icmphdr), 0);
130 	icmph->checksum = csum_fold(csum);
131 
132 	niph->tot_len	= htons(nskb->len);
133 	ip_send_check(niph);
134 
135 	return nskb;
136 }
137 EXPORT_SYMBOL_GPL(nf_reject_skb_v4_unreach);
138 
139 const struct tcphdr *nf_reject_ip_tcphdr_get(struct sk_buff *oldskb,
140 					     struct tcphdr *_oth, int hook)
141 {
142 	const struct tcphdr *oth;
143 
144 	/* IP header checks: fragment. */
145 	if (ip_hdr(oldskb)->frag_off & htons(IP_OFFSET))
146 		return NULL;
147 
148 	if (ip_hdr(oldskb)->protocol != IPPROTO_TCP)
149 		return NULL;
150 
151 	oth = skb_header_pointer(oldskb, ip_hdrlen(oldskb),
152 				 sizeof(struct tcphdr), _oth);
153 	if (oth == NULL)
154 		return NULL;
155 
156 	/* No RST for RST. */
157 	if (oth->rst)
158 		return NULL;
159 
160 	/* Check checksum */
161 	if (nf_ip_checksum(oldskb, hook, ip_hdrlen(oldskb), IPPROTO_TCP))
162 		return NULL;
163 
164 	return oth;
165 }
166 EXPORT_SYMBOL_GPL(nf_reject_ip_tcphdr_get);
167 
168 struct iphdr *nf_reject_iphdr_put(struct sk_buff *nskb,
169 				  const struct sk_buff *oldskb,
170 				  __u8 protocol, int ttl)
171 {
172 	struct iphdr *niph, *oiph = ip_hdr(oldskb);
173 
174 	skb_reset_network_header(nskb);
175 	niph = skb_put(nskb, sizeof(struct iphdr));
176 	niph->version	= 4;
177 	niph->ihl	= sizeof(struct iphdr) / 4;
178 	niph->tos	= 0;
179 	niph->id	= 0;
180 	niph->frag_off	= htons(IP_DF);
181 	niph->protocol	= protocol;
182 	niph->check	= 0;
183 	niph->saddr	= oiph->daddr;
184 	niph->daddr	= oiph->saddr;
185 	niph->ttl	= ttl;
186 
187 	nskb->protocol = htons(ETH_P_IP);
188 
189 	return niph;
190 }
191 EXPORT_SYMBOL_GPL(nf_reject_iphdr_put);
192 
193 void nf_reject_ip_tcphdr_put(struct sk_buff *nskb, const struct sk_buff *oldskb,
194 			  const struct tcphdr *oth)
195 {
196 	struct iphdr *niph = ip_hdr(nskb);
197 	struct tcphdr *tcph;
198 
199 	skb_reset_transport_header(nskb);
200 	tcph = skb_put_zero(nskb, sizeof(struct tcphdr));
201 	tcph->source	= oth->dest;
202 	tcph->dest	= oth->source;
203 	tcph->doff	= sizeof(struct tcphdr) / 4;
204 
205 	if (oth->ack) {
206 		tcph->seq = oth->ack_seq;
207 	} else {
208 		tcph->ack_seq = htonl(ntohl(oth->seq) + oth->syn + oth->fin +
209 				      oldskb->len - ip_hdrlen(oldskb) -
210 				      (oth->doff << 2));
211 		tcph->ack = 1;
212 	}
213 
214 	tcph->rst	= 1;
215 	tcph->check = ~tcp_v4_check(sizeof(struct tcphdr), niph->saddr,
216 				    niph->daddr, 0);
217 	nskb->ip_summed = CHECKSUM_PARTIAL;
218 	nskb->csum_start = (unsigned char *)tcph - nskb->head;
219 	nskb->csum_offset = offsetof(struct tcphdr, check);
220 }
221 EXPORT_SYMBOL_GPL(nf_reject_ip_tcphdr_put);
222 
223 static int nf_reject_fill_skb_dst(struct sk_buff *skb_in)
224 {
225 	struct dst_entry *dst = NULL;
226 	struct flowi fl;
227 
228 	memset(&fl, 0, sizeof(struct flowi));
229 	fl.u.ip4.daddr = ip_hdr(skb_in)->saddr;
230 	nf_ip_route(dev_net(skb_in->dev), &dst, &fl, false);
231 	if (!dst)
232 		return -1;
233 
234 	skb_dst_set(skb_in, dst);
235 	return 0;
236 }
237 
238 /* Send RST reply */
239 void nf_send_reset(struct net *net, struct sock *sk, struct sk_buff *oldskb,
240 		   int hook)
241 {
242 	struct net_device *br_indev __maybe_unused;
243 	struct sk_buff *nskb;
244 	struct iphdr *niph;
245 	const struct tcphdr *oth;
246 	struct tcphdr _oth;
247 
248 	oth = nf_reject_ip_tcphdr_get(oldskb, &_oth, hook);
249 	if (!oth)
250 		return;
251 
252 	if ((hook == NF_INET_PRE_ROUTING || hook == NF_INET_INGRESS) &&
253 	    nf_reject_fill_skb_dst(oldskb) < 0)
254 		return;
255 
256 	if (skb_rtable(oldskb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
257 		return;
258 
259 	nskb = alloc_skb(sizeof(struct iphdr) + sizeof(struct tcphdr) +
260 			 LL_MAX_HEADER, GFP_ATOMIC);
261 	if (!nskb)
262 		return;
263 
264 	/* ip_route_me_harder expects skb->dst to be set */
265 	skb_dst_set_noref(nskb, skb_dst(oldskb));
266 
267 	nskb->mark = IP4_REPLY_MARK(net, oldskb->mark);
268 
269 	skb_reserve(nskb, LL_MAX_HEADER);
270 	niph = nf_reject_iphdr_put(nskb, oldskb, IPPROTO_TCP,
271 				   ip4_dst_hoplimit(skb_dst(nskb)));
272 	nf_reject_ip_tcphdr_put(nskb, oldskb, oth);
273 	if (ip_route_me_harder(net, sk, nskb, RTN_UNSPEC))
274 		goto free_nskb;
275 
276 	niph = ip_hdr(nskb);
277 
278 	/* "Never happens" */
279 	if (nskb->len > dst_mtu(skb_dst(nskb)))
280 		goto free_nskb;
281 
282 	nf_ct_attach(nskb, oldskb);
283 
284 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
285 	/* If we use ip_local_out for bridged traffic, the MAC source on
286 	 * the RST will be ours, instead of the destination's.  This confuses
287 	 * some routers/firewalls, and they drop the packet.  So we need to
288 	 * build the eth header using the original destination's MAC as the
289 	 * source, and send the RST packet directly.
290 	 */
291 	br_indev = nf_bridge_get_physindev(oldskb);
292 	if (br_indev) {
293 		struct ethhdr *oeth = eth_hdr(oldskb);
294 
295 		nskb->dev = br_indev;
296 		niph->tot_len = htons(nskb->len);
297 		ip_send_check(niph);
298 		if (dev_hard_header(nskb, nskb->dev, ntohs(nskb->protocol),
299 				    oeth->h_source, oeth->h_dest, nskb->len) < 0)
300 			goto free_nskb;
301 		dev_queue_xmit(nskb);
302 	} else
303 #endif
304 		ip_local_out(net, nskb->sk, nskb);
305 
306 	return;
307 
308  free_nskb:
309 	kfree_skb(nskb);
310 }
311 EXPORT_SYMBOL_GPL(nf_send_reset);
312 
313 void nf_send_unreach(struct sk_buff *skb_in, int code, int hook)
314 {
315 	struct iphdr *iph = ip_hdr(skb_in);
316 	int dataoff = ip_hdrlen(skb_in);
317 	u8 proto = iph->protocol;
318 
319 	if (iph->frag_off & htons(IP_OFFSET))
320 		return;
321 
322 	if ((hook == NF_INET_PRE_ROUTING || hook == NF_INET_INGRESS) &&
323 	    nf_reject_fill_skb_dst(skb_in) < 0)
324 		return;
325 
326 	if (skb_csum_unnecessary(skb_in) ||
327 	    !nf_reject_verify_csum(skb_in, dataoff, proto)) {
328 		icmp_send(skb_in, ICMP_DEST_UNREACH, code, 0);
329 		return;
330 	}
331 
332 	if (nf_ip_checksum(skb_in, hook, dataoff, proto) == 0)
333 		icmp_send(skb_in, ICMP_DEST_UNREACH, code, 0);
334 }
335 EXPORT_SYMBOL_GPL(nf_send_unreach);
336 
337 MODULE_LICENSE("GPL");
338