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>
7 #include <linux/types.h>
8 #include <linux/timer.h>
9 #include <linux/netfilter.h>
11 #include <linux/icmp.h>
12 #include <linux/seq_file.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>
23 static const unsigned int nf_ct_icmp_timeout = 30*HZ;
25 bool icmp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
26 struct net *net, struct nf_conntrack_tuple *tuple)
28 const struct icmphdr *hp;
31 hp = skb_header_pointer(skb, dataoff, sizeof(_hdr), &_hdr);
35 tuple->dst.u.icmp.type = hp->type;
36 tuple->src.u.icmp.id = hp->un.echo.id;
37 tuple->dst.u.icmp.code = hp->code;
42 /* Add 1; spaces filled with 0. */
43 static const u_int8_t invmap[] = {
44 [ICMP_ECHO] = ICMP_ECHOREPLY + 1,
45 [ICMP_ECHOREPLY] = ICMP_ECHO + 1,
46 [ICMP_TIMESTAMP] = ICMP_TIMESTAMPREPLY + 1,
47 [ICMP_TIMESTAMPREPLY] = ICMP_TIMESTAMP + 1,
48 [ICMP_INFO_REQUEST] = ICMP_INFO_REPLY + 1,
49 [ICMP_INFO_REPLY] = ICMP_INFO_REQUEST + 1,
50 [ICMP_ADDRESS] = ICMP_ADDRESSREPLY + 1,
51 [ICMP_ADDRESSREPLY] = ICMP_ADDRESS + 1
54 bool nf_conntrack_invert_icmp_tuple(struct nf_conntrack_tuple *tuple,
55 const struct nf_conntrack_tuple *orig)
57 if (orig->dst.u.icmp.type >= sizeof(invmap) ||
58 !invmap[orig->dst.u.icmp.type])
61 tuple->src.u.icmp.id = orig->src.u.icmp.id;
62 tuple->dst.u.icmp.type = invmap[orig->dst.u.icmp.type] - 1;
63 tuple->dst.u.icmp.code = orig->dst.u.icmp.code;
67 /* Returns verdict for packet, or -1 for invalid. */
68 int nf_conntrack_icmp_packet(struct nf_conn *ct,
70 enum ip_conntrack_info ctinfo,
71 const struct nf_hook_state *state)
73 /* Do not immediately delete the connection after the first
74 successful reply to avoid excessive conntrackd traffic
75 and also to handle correctly ICMP echo reply duplicates. */
76 unsigned int *timeout = nf_ct_timeout_lookup(ct);
77 static const u_int8_t valid_new[] = {
80 [ICMP_INFO_REQUEST] = 1,
84 if (state->pf != NFPROTO_IPV4)
87 if (ct->tuplehash[0].tuple.dst.u.icmp.type >= sizeof(valid_new) ||
88 !valid_new[ct->tuplehash[0].tuple.dst.u.icmp.type]) {
89 /* Can't create a new ICMP `conn' with this. */
90 pr_debug("icmp: can't create new conn with type %u\n",
91 ct->tuplehash[0].tuple.dst.u.icmp.type);
92 nf_ct_dump_tuple_ip(&ct->tuplehash[0].tuple);
97 timeout = &nf_icmp_pernet(nf_ct_net(ct))->timeout;
99 nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
103 /* Check inner header is related to any of the existing connections */
104 int nf_conntrack_inet_error(struct nf_conn *tmpl, struct sk_buff *skb,
105 unsigned int dataoff,
106 const struct nf_hook_state *state,
107 u8 l4proto, union nf_inet_addr *outer_daddr)
109 struct nf_conntrack_tuple innertuple, origtuple;
110 const struct nf_conntrack_tuple_hash *h;
111 const struct nf_conntrack_zone *zone;
112 enum ip_conntrack_info ctinfo;
113 struct nf_conntrack_zone tmp;
114 union nf_inet_addr *ct_daddr;
115 enum ip_conntrack_dir dir;
118 WARN_ON(skb_nfct(skb));
119 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
121 /* Are they talking about one of our connections? */
122 if (!nf_ct_get_tuplepr(skb, dataoff,
123 state->pf, state->net, &origtuple))
126 /* Ordinarily, we'd expect the inverted tupleproto, but it's
127 been preserved inside the ICMP. */
128 if (!nf_ct_invert_tuple(&innertuple, &origtuple))
131 h = nf_conntrack_find_get(state->net, zone, &innertuple);
135 /* Consider: A -> T (=This machine) -> B
136 * Conntrack entry will look like this:
138 * Reply: B->T (SNAT case) OR A
140 * When this function runs, we got packet that looks like this:
141 * iphdr|icmphdr|inner_iphdr|l4header (tcp, udp, ..).
143 * Above nf_conntrack_find_get() makes lookup based on inner_hdr,
144 * so we should expect that destination of the found connection
145 * matches outer header destination address.
147 * In above example, we can consider these two cases:
148 * 1. Error coming in reply direction from B or M (middle box) to
149 * T (SNAT case) or A.
150 * Inner saddr will be B, dst will be T or A.
151 * The found conntrack will be reply tuple (B->T/A).
152 * 2. Error coming in original direction from A or M to B.
153 * Inner saddr will be A, inner daddr will be B.
154 * The found conntrack will be original tuple (A->B).
156 * In both cases, conntrack[dir].dst == inner.dst.
158 * A bogus packet could look like this:
160 * Outer: B->X (other machine reachable by T).
162 * In this case, lookup yields connection A->B and will
163 * set packet from B->X as *RELATED*, even though no connection
164 * from X was ever seen.
166 ct = nf_ct_tuplehash_to_ctrack(h);
167 dir = NF_CT_DIRECTION(h);
168 ct_daddr = &ct->tuplehash[dir].tuple.dst.u3;
169 if (!nf_inet_addr_cmp(outer_daddr, ct_daddr)) {
170 if (state->pf == AF_INET) {
171 nf_l4proto_log_invalid(skb, state->net, state->pf,
173 "outer daddr %pI4 != inner %pI4",
174 &outer_daddr->ip, &ct_daddr->ip);
175 } else if (state->pf == AF_INET6) {
176 nf_l4proto_log_invalid(skb, state->net, state->pf,
178 "outer daddr %pI6 != inner %pI6",
179 &outer_daddr->ip6, &ct_daddr->ip6);
185 ctinfo = IP_CT_RELATED;
186 if (dir == IP_CT_DIR_REPLY)
187 ctinfo += IP_CT_IS_REPLY;
189 /* Update skb to refer to this connection */
190 nf_ct_set(skb, ct, ctinfo);
194 static void icmp_error_log(const struct sk_buff *skb,
195 const struct nf_hook_state *state,
198 nf_l4proto_log_invalid(skb, state->net, state->pf,
199 IPPROTO_ICMP, "%s", msg);
202 /* Small and modified version of icmp_rcv */
203 int nf_conntrack_icmpv4_error(struct nf_conn *tmpl,
204 struct sk_buff *skb, unsigned int dataoff,
205 const struct nf_hook_state *state)
207 union nf_inet_addr outer_daddr;
208 const struct icmphdr *icmph;
211 /* Not enough header? */
212 icmph = skb_header_pointer(skb, dataoff, sizeof(_ih), &_ih);
214 icmp_error_log(skb, state, "short packet");
218 /* See nf_conntrack_proto_tcp.c */
219 if (state->net->ct.sysctl_checksum &&
220 state->hook == NF_INET_PRE_ROUTING &&
221 nf_ip_checksum(skb, state->hook, dataoff, IPPROTO_ICMP)) {
222 icmp_error_log(skb, state, "bad hw icmp checksum");
227 * 18 is the highest 'known' ICMP type. Anything else is a mystery
229 * RFC 1122: 3.2.2 Unknown ICMP messages types MUST be silently
232 if (icmph->type > NR_ICMP_TYPES) {
233 icmp_error_log(skb, state, "invalid icmp type");
237 /* Need to track icmp error message? */
238 if (!icmp_is_err(icmph->type))
241 memset(&outer_daddr, 0, sizeof(outer_daddr));
242 outer_daddr.ip = ip_hdr(skb)->daddr;
244 dataoff += sizeof(*icmph);
245 return nf_conntrack_inet_error(tmpl, skb, dataoff, state,
246 IPPROTO_ICMP, &outer_daddr);
249 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
251 #include <linux/netfilter/nfnetlink.h>
252 #include <linux/netfilter/nfnetlink_conntrack.h>
254 static int icmp_tuple_to_nlattr(struct sk_buff *skb,
255 const struct nf_conntrack_tuple *t)
257 if (nla_put_be16(skb, CTA_PROTO_ICMP_ID, t->src.u.icmp.id) ||
258 nla_put_u8(skb, CTA_PROTO_ICMP_TYPE, t->dst.u.icmp.type) ||
259 nla_put_u8(skb, CTA_PROTO_ICMP_CODE, t->dst.u.icmp.code))
260 goto nla_put_failure;
267 static const struct nla_policy icmp_nla_policy[CTA_PROTO_MAX+1] = {
268 [CTA_PROTO_ICMP_TYPE] = { .type = NLA_U8 },
269 [CTA_PROTO_ICMP_CODE] = { .type = NLA_U8 },
270 [CTA_PROTO_ICMP_ID] = { .type = NLA_U16 },
273 static int icmp_nlattr_to_tuple(struct nlattr *tb[],
274 struct nf_conntrack_tuple *tuple)
276 if (!tb[CTA_PROTO_ICMP_TYPE] ||
277 !tb[CTA_PROTO_ICMP_CODE] ||
278 !tb[CTA_PROTO_ICMP_ID])
281 tuple->dst.u.icmp.type = nla_get_u8(tb[CTA_PROTO_ICMP_TYPE]);
282 tuple->dst.u.icmp.code = nla_get_u8(tb[CTA_PROTO_ICMP_CODE]);
283 tuple->src.u.icmp.id = nla_get_be16(tb[CTA_PROTO_ICMP_ID]);
285 if (tuple->dst.u.icmp.type >= sizeof(invmap) ||
286 !invmap[tuple->dst.u.icmp.type])
292 static unsigned int icmp_nlattr_tuple_size(void)
294 static unsigned int size __read_mostly;
297 size = nla_policy_len(icmp_nla_policy, CTA_PROTO_MAX + 1);
303 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
305 #include <linux/netfilter/nfnetlink.h>
306 #include <linux/netfilter/nfnetlink_cttimeout.h>
308 static int icmp_timeout_nlattr_to_obj(struct nlattr *tb[],
309 struct net *net, void *data)
311 unsigned int *timeout = data;
312 struct nf_icmp_net *in = nf_icmp_pernet(net);
314 if (tb[CTA_TIMEOUT_ICMP_TIMEOUT]) {
316 timeout = &in->timeout;
318 ntohl(nla_get_be32(tb[CTA_TIMEOUT_ICMP_TIMEOUT])) * HZ;
319 } else if (timeout) {
320 /* Set default ICMP timeout. */
321 *timeout = in->timeout;
327 icmp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
329 const unsigned int *timeout = data;
331 if (nla_put_be32(skb, CTA_TIMEOUT_ICMP_TIMEOUT, htonl(*timeout / HZ)))
332 goto nla_put_failure;
339 static const struct nla_policy
340 icmp_timeout_nla_policy[CTA_TIMEOUT_ICMP_MAX+1] = {
341 [CTA_TIMEOUT_ICMP_TIMEOUT] = { .type = NLA_U32 },
343 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
345 void nf_conntrack_icmp_init_net(struct net *net)
347 struct nf_icmp_net *in = nf_icmp_pernet(net);
349 in->timeout = nf_ct_icmp_timeout;
352 const struct nf_conntrack_l4proto nf_conntrack_l4proto_icmp =
354 .l4proto = IPPROTO_ICMP,
355 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
356 .tuple_to_nlattr = icmp_tuple_to_nlattr,
357 .nlattr_tuple_size = icmp_nlattr_tuple_size,
358 .nlattr_to_tuple = icmp_nlattr_to_tuple,
359 .nla_policy = icmp_nla_policy,
361 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
363 .nlattr_to_obj = icmp_timeout_nlattr_to_obj,
364 .obj_to_nlattr = icmp_timeout_obj_to_nlattr,
365 .nlattr_max = CTA_TIMEOUT_ICMP_MAX,
366 .obj_size = sizeof(unsigned int),
367 .nla_policy = icmp_timeout_nla_policy,
369 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */