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 * (C) 2006-2010 Patrick McHardy <kaber@trash.net>
5 */
6
7#include <linux/types.h>
8#include <linux/timer.h>
9#include <linux/netfilter.h>
10#include <linux/in.h>
11#include <linux/icmp.h>
12#include <linux/seq_file.h>
13#include <net/ip.h>
14#include <net/checksum.h>
15#include <linux/netfilter_ipv4.h>
16#include <net/netfilter/nf_conntrack_tuple.h>
17#include <net/netfilter/nf_conntrack_l4proto.h>
18#include <net/netfilter/nf_conntrack_core.h>
19#include <net/netfilter/nf_conntrack_timeout.h>
20#include <net/netfilter/nf_conntrack_zones.h>
21#include <net/netfilter/nf_log.h>
22
23#include "nf_internals.h"
24
25static const unsigned int nf_ct_icmp_timeout = 30*HZ;
26
27bool icmp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
28		       struct net *net, struct nf_conntrack_tuple *tuple)
29{
30	const struct icmphdr *hp;
31	struct icmphdr _hdr;
32
33	hp = skb_header_pointer(skb, dataoff, sizeof(_hdr), &_hdr);
34	if (hp == NULL)
35		return false;
36
37	tuple->dst.u.icmp.type = hp->type;
38	tuple->src.u.icmp.id = hp->un.echo.id;
39	tuple->dst.u.icmp.code = hp->code;
40
41	return true;
42}
43
44/* Add 1; spaces filled with 0. */
45static const u_int8_t invmap[] = {
46	[ICMP_ECHO] = ICMP_ECHOREPLY + 1,
47	[ICMP_ECHOREPLY] = ICMP_ECHO + 1,
48	[ICMP_TIMESTAMP] = ICMP_TIMESTAMPREPLY + 1,
49	[ICMP_TIMESTAMPREPLY] = ICMP_TIMESTAMP + 1,
50	[ICMP_INFO_REQUEST] = ICMP_INFO_REPLY + 1,
51	[ICMP_INFO_REPLY] = ICMP_INFO_REQUEST + 1,
52	[ICMP_ADDRESS] = ICMP_ADDRESSREPLY + 1,
53	[ICMP_ADDRESSREPLY] = ICMP_ADDRESS + 1
54};
55
56bool nf_conntrack_invert_icmp_tuple(struct nf_conntrack_tuple *tuple,
57				    const struct nf_conntrack_tuple *orig)
58{
59	if (orig->dst.u.icmp.type >= sizeof(invmap) ||
60	    !invmap[orig->dst.u.icmp.type])
61		return false;
62
63	tuple->src.u.icmp.id = orig->src.u.icmp.id;
64	tuple->dst.u.icmp.type = invmap[orig->dst.u.icmp.type] - 1;
65	tuple->dst.u.icmp.code = orig->dst.u.icmp.code;
66	return true;
67}
68
69/* Returns verdict for packet, or -1 for invalid. */
70int nf_conntrack_icmp_packet(struct nf_conn *ct,
71			     struct sk_buff *skb,
72			     enum ip_conntrack_info ctinfo,
73			     const struct nf_hook_state *state)
74{
75	/* Do not immediately delete the connection after the first
76	   successful reply to avoid excessive conntrackd traffic
77	   and also to handle correctly ICMP echo reply duplicates. */
78	unsigned int *timeout = nf_ct_timeout_lookup(ct);
79	static const u_int8_t valid_new[] = {
80		[ICMP_ECHO] = 1,
81		[ICMP_TIMESTAMP] = 1,
82		[ICMP_INFO_REQUEST] = 1,
83		[ICMP_ADDRESS] = 1
84	};
85
86	if (state->pf != NFPROTO_IPV4)
87		return -NF_ACCEPT;
88
89	if (ct->tuplehash[0].tuple.dst.u.icmp.type >= sizeof(valid_new) ||
90	    !valid_new[ct->tuplehash[0].tuple.dst.u.icmp.type]) {
91		/* Can't create a new ICMP `conn' with this. */
92		pr_debug("icmp: can't create new conn with type %u\n",
93			 ct->tuplehash[0].tuple.dst.u.icmp.type);
94		nf_ct_dump_tuple_ip(&ct->tuplehash[0].tuple);
95		return -NF_ACCEPT;
96	}
97
98	if (!timeout)
99		timeout = &nf_icmp_pernet(nf_ct_net(ct))->timeout;
100
101	nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
102	return NF_ACCEPT;
103}
104
105/* Check inner header is related to any of the existing connections */
106int nf_conntrack_inet_error(struct nf_conn *tmpl, struct sk_buff *skb,
107			    unsigned int dataoff,
108			    const struct nf_hook_state *state,
109			    u8 l4proto, union nf_inet_addr *outer_daddr)
110{
111	struct nf_conntrack_tuple innertuple, origtuple;
112	const struct nf_conntrack_tuple_hash *h;
113	const struct nf_conntrack_zone *zone;
114	enum ip_conntrack_info ctinfo;
115	struct nf_conntrack_zone tmp;
116	union nf_inet_addr *ct_daddr;
117	enum ip_conntrack_dir dir;
118	struct nf_conn *ct;
119
120	WARN_ON(skb_nfct(skb));
121	zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
122
123	/* Are they talking about one of our connections? */
124	if (!nf_ct_get_tuplepr(skb, dataoff,
125			       state->pf, state->net, &origtuple))
126		return -NF_ACCEPT;
127
128	/* Ordinarily, we'd expect the inverted tupleproto, but it's
129	   been preserved inside the ICMP. */
130	if (!nf_ct_invert_tuple(&innertuple, &origtuple))
131		return -NF_ACCEPT;
132
133	h = nf_conntrack_find_get(state->net, zone, &innertuple);
134	if (!h)
135		return -NF_ACCEPT;
136
137	/* Consider: A -> T (=This machine) -> B
138	 *   Conntrack entry will look like this:
139	 *      Original:  A->B
140	 *      Reply:     B->T (SNAT case) OR A
141	 *
142	 * When this function runs, we got packet that looks like this:
143	 * iphdr|icmphdr|inner_iphdr|l4header (tcp, udp, ..).
144	 *
145	 * Above nf_conntrack_find_get() makes lookup based on inner_hdr,
146	 * so we should expect that destination of the found connection
147	 * matches outer header destination address.
148	 *
149	 * In above example, we can consider these two cases:
150	 *  1. Error coming in reply direction from B or M (middle box) to
151	 *     T (SNAT case) or A.
152	 *     Inner saddr will be B, dst will be T or A.
153	 *     The found conntrack will be reply tuple (B->T/A).
154	 *  2. Error coming in original direction from A or M to B.
155	 *     Inner saddr will be A, inner daddr will be B.
156	 *     The found conntrack will be original tuple (A->B).
157	 *
158	 * In both cases, conntrack[dir].dst == inner.dst.
159	 *
160	 * A bogus packet could look like this:
161	 *   Inner: B->T
162	 *   Outer: B->X (other machine reachable by T).
163	 *
164	 * In this case, lookup yields connection A->B and will
165	 * set packet from B->X as *RELATED*, even though no connection
166	 * from X was ever seen.
167	 */
168	ct = nf_ct_tuplehash_to_ctrack(h);
169	dir = NF_CT_DIRECTION(h);
170	ct_daddr = &ct->tuplehash[dir].tuple.dst.u3;
171	if (!nf_inet_addr_cmp(outer_daddr, ct_daddr)) {
172		if (state->pf == AF_INET) {
173			nf_l4proto_log_invalid(skb, state,
174					       l4proto,
175					       "outer daddr %pI4 != inner %pI4",
176					       &outer_daddr->ip, &ct_daddr->ip);
177		} else if (state->pf == AF_INET6) {
178			nf_l4proto_log_invalid(skb, state,
179					       l4proto,
180					       "outer daddr %pI6 != inner %pI6",
181					       &outer_daddr->ip6, &ct_daddr->ip6);
182		}
183		nf_ct_put(ct);
184		return -NF_ACCEPT;
185	}
186
187	ctinfo = IP_CT_RELATED;
188	if (dir == IP_CT_DIR_REPLY)
189		ctinfo += IP_CT_IS_REPLY;
190
191	/* Update skb to refer to this connection */
192	nf_ct_set(skb, ct, ctinfo);
193	return NF_ACCEPT;
194}
195
196static void icmp_error_log(const struct sk_buff *skb,
197			   const struct nf_hook_state *state,
198			   const char *msg)
199{
200	nf_l4proto_log_invalid(skb, state, IPPROTO_ICMP, "%s", msg);
201}
202
203/* Small and modified version of icmp_rcv */
204int nf_conntrack_icmpv4_error(struct nf_conn *tmpl,
205			      struct sk_buff *skb, unsigned int dataoff,
206			      const struct nf_hook_state *state)
207{
208	union nf_inet_addr outer_daddr;
209	const struct icmphdr *icmph;
210	struct icmphdr _ih;
211
212	/* Not enough header? */
213	icmph = skb_header_pointer(skb, dataoff, sizeof(_ih), &_ih);
214	if (icmph == NULL) {
215		icmp_error_log(skb, state, "short packet");
216		return -NF_ACCEPT;
217	}
218
219	/* See nf_conntrack_proto_tcp.c */
220	if (state->net->ct.sysctl_checksum &&
221	    state->hook == NF_INET_PRE_ROUTING &&
222	    nf_ip_checksum(skb, state->hook, dataoff, IPPROTO_ICMP)) {
223		icmp_error_log(skb, state, "bad hw icmp checksum");
224		return -NF_ACCEPT;
225	}
226
227	/*
228	 *	18 is the highest 'known' ICMP type. Anything else is a mystery
229	 *
230	 *	RFC 1122: 3.2.2  Unknown ICMP messages types MUST be silently
231	 *		  discarded.
232	 */
233	if (icmph->type > NR_ICMP_TYPES) {
234		icmp_error_log(skb, state, "invalid icmp type");
235		return -NF_ACCEPT;
236	}
237
238	/* Need to track icmp error message? */
239	if (!icmp_is_err(icmph->type))
240		return NF_ACCEPT;
241
242	memset(&outer_daddr, 0, sizeof(outer_daddr));
243	outer_daddr.ip = ip_hdr(skb)->daddr;
244
245	dataoff += sizeof(*icmph);
246	return nf_conntrack_inet_error(tmpl, skb, dataoff, state,
247				       IPPROTO_ICMP, &outer_daddr);
248}
249
250#if IS_ENABLED(CONFIG_NF_CT_NETLINK)
251
252#include <linux/netfilter/nfnetlink.h>
253#include <linux/netfilter/nfnetlink_conntrack.h>
254
255static int icmp_tuple_to_nlattr(struct sk_buff *skb,
256				const struct nf_conntrack_tuple *t)
257{
258	if (nla_put_be16(skb, CTA_PROTO_ICMP_ID, t->src.u.icmp.id) ||
259	    nla_put_u8(skb, CTA_PROTO_ICMP_TYPE, t->dst.u.icmp.type) ||
260	    nla_put_u8(skb, CTA_PROTO_ICMP_CODE, t->dst.u.icmp.code))
261		goto nla_put_failure;
262	return 0;
263
264nla_put_failure:
265	return -1;
266}
267
268static const struct nla_policy icmp_nla_policy[CTA_PROTO_MAX+1] = {
269	[CTA_PROTO_ICMP_TYPE]	= { .type = NLA_U8 },
270	[CTA_PROTO_ICMP_CODE]	= { .type = NLA_U8 },
271	[CTA_PROTO_ICMP_ID]	= { .type = NLA_U16 },
272};
273
274static int icmp_nlattr_to_tuple(struct nlattr *tb[],
275				struct nf_conntrack_tuple *tuple,
276				u_int32_t flags)
277{
278	if (flags & CTA_FILTER_FLAG(CTA_PROTO_ICMP_TYPE)) {
279		if (!tb[CTA_PROTO_ICMP_TYPE])
280			return -EINVAL;
281
282		tuple->dst.u.icmp.type = nla_get_u8(tb[CTA_PROTO_ICMP_TYPE]);
283		if (tuple->dst.u.icmp.type >= sizeof(invmap) ||
284		    !invmap[tuple->dst.u.icmp.type])
285			return -EINVAL;
286	}
287
288	if (flags & CTA_FILTER_FLAG(CTA_PROTO_ICMP_CODE)) {
289		if (!tb[CTA_PROTO_ICMP_CODE])
290			return -EINVAL;
291
292		tuple->dst.u.icmp.code = nla_get_u8(tb[CTA_PROTO_ICMP_CODE]);
293	}
294
295	if (flags & CTA_FILTER_FLAG(CTA_PROTO_ICMP_ID)) {
296		if (!tb[CTA_PROTO_ICMP_ID])
297			return -EINVAL;
298
299		tuple->src.u.icmp.id = nla_get_be16(tb[CTA_PROTO_ICMP_ID]);
300	}
301
302	return 0;
303}
304
305static unsigned int icmp_nlattr_tuple_size(void)
306{
307	static unsigned int size __read_mostly;
308
309	if (!size)
310		size = nla_policy_len(icmp_nla_policy, CTA_PROTO_MAX + 1);
311
312	return size;
313}
314#endif
315
316#ifdef CONFIG_NF_CONNTRACK_TIMEOUT
317
318#include <linux/netfilter/nfnetlink.h>
319#include <linux/netfilter/nfnetlink_cttimeout.h>
320
321static int icmp_timeout_nlattr_to_obj(struct nlattr *tb[],
322				      struct net *net, void *data)
323{
324	unsigned int *timeout = data;
325	struct nf_icmp_net *in = nf_icmp_pernet(net);
326
327	if (tb[CTA_TIMEOUT_ICMP_TIMEOUT]) {
328		if (!timeout)
329			timeout = &in->timeout;
330		*timeout =
331			ntohl(nla_get_be32(tb[CTA_TIMEOUT_ICMP_TIMEOUT])) * HZ;
332	} else if (timeout) {
333		/* Set default ICMP timeout. */
334		*timeout = in->timeout;
335	}
336	return 0;
337}
338
339static int
340icmp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
341{
342	const unsigned int *timeout = data;
343
344	if (nla_put_be32(skb, CTA_TIMEOUT_ICMP_TIMEOUT, htonl(*timeout / HZ)))
345		goto nla_put_failure;
346	return 0;
347
348nla_put_failure:
349	return -ENOSPC;
350}
351
352static const struct nla_policy
353icmp_timeout_nla_policy[CTA_TIMEOUT_ICMP_MAX+1] = {
354	[CTA_TIMEOUT_ICMP_TIMEOUT]	= { .type = NLA_U32 },
355};
356#endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
357
358void nf_conntrack_icmp_init_net(struct net *net)
359{
360	struct nf_icmp_net *in = nf_icmp_pernet(net);
361
362	in->timeout = nf_ct_icmp_timeout;
363}
364
365const struct nf_conntrack_l4proto nf_conntrack_l4proto_icmp =
366{
367	.l4proto		= IPPROTO_ICMP,
368#if IS_ENABLED(CONFIG_NF_CT_NETLINK)
369	.tuple_to_nlattr	= icmp_tuple_to_nlattr,
370	.nlattr_tuple_size	= icmp_nlattr_tuple_size,
371	.nlattr_to_tuple	= icmp_nlattr_to_tuple,
372	.nla_policy		= icmp_nla_policy,
373#endif
374#ifdef CONFIG_NF_CONNTRACK_TIMEOUT
375	.ctnl_timeout		= {
376		.nlattr_to_obj	= icmp_timeout_nlattr_to_obj,
377		.obj_to_nlattr	= icmp_timeout_obj_to_nlattr,
378		.nlattr_max	= CTA_TIMEOUT_ICMP_MAX,
379		.obj_size	= sizeof(unsigned int),
380		.nla_policy	= icmp_timeout_nla_policy,
381	},
382#endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
383};
384