xref: /netbsd-src/sys/net/npf/npf_alg_icmp.c (revision 6a493d6bc668897c91594964a732d38505b70cbb)
1 /*	$NetBSD: npf_alg_icmp.c,v 1.18 2013/12/06 01:33:37 rmind Exp $	*/
2 
3 /*-
4  * Copyright (c) 2010 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This material is based upon work partially supported by The
8  * NetBSD Foundation under a contract with Mindaugas Rasiukevicius.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * NPF ALG for ICMP and traceroute translations.
34  */
35 
36 #include <sys/cdefs.h>
37 __KERNEL_RCSID(0, "$NetBSD: npf_alg_icmp.c,v 1.18 2013/12/06 01:33:37 rmind Exp $");
38 
39 #include <sys/param.h>
40 #include <sys/module.h>
41 
42 #include <netinet/in_systm.h>
43 #include <netinet/in.h>
44 #include <netinet/ip.h>
45 #include <netinet/tcp.h>
46 #include <netinet/udp.h>
47 #include <netinet/ip_icmp.h>
48 #include <netinet/icmp6.h>
49 #include <net/pfil.h>
50 
51 #include "npf_impl.h"
52 
53 MODULE(MODULE_CLASS_MISC, npf_alg_icmp, "npf");
54 
55 /*
56  * Traceroute criteria.
57  *
58  * IANA assigned base port: 33434.  However, common practice is to increase
59  * the port, thus monitor [33434-33484] range.  Additional filter is low TTL.
60  */
61 
62 #define	TR_BASE_PORT	33434
63 #define	TR_PORT_RANGE	33484
64 #define	TR_MAX_TTL	48
65 
66 static npf_alg_t *	alg_icmp	__read_mostly;
67 
68 static bool	npfa_icmp_match(npf_cache_t *, nbuf_t *, npf_nat_t *, int);
69 static bool	npfa_icmp_nat(npf_cache_t *, nbuf_t *, npf_nat_t *, int);
70 static npf_session_t *npfa_icmp_session(npf_cache_t *, nbuf_t *, int);
71 
72 /*
73  * npf_alg_icmp_{init,fini,modcmd}: ICMP ALG initialization, destruction
74  * and module interface.
75  */
76 
77 static int
78 npf_alg_icmp_init(void)
79 {
80 	alg_icmp = npf_alg_register("icmp", npfa_icmp_match,
81 	    npfa_icmp_nat, npfa_icmp_session);
82 	return alg_icmp ? 0 : ENOMEM;
83 }
84 
85 static int
86 npf_alg_icmp_fini(void)
87 {
88 	KASSERT(alg_icmp != NULL);
89 	return npf_alg_unregister(alg_icmp);
90 }
91 
92 static int
93 npf_alg_icmp_modcmd(modcmd_t cmd, void *arg)
94 {
95 	switch (cmd) {
96 	case MODULE_CMD_INIT:
97 		return npf_alg_icmp_init();
98 	case MODULE_CMD_FINI:
99 		return npf_alg_icmp_fini();
100 	case MODULE_CMD_AUTOUNLOAD:
101 		return EBUSY;
102 	default:
103 		return ENOTTY;
104 	}
105 	return 0;
106 }
107 
108 /*
109  * npfa_icmp_match: match inspector - determines ALG case and associates
110  * our ALG with the NAT entry.
111  */
112 static bool
113 npfa_icmp_match(npf_cache_t *npc, nbuf_t *nbuf, npf_nat_t *nt, int di)
114 {
115 	const int proto = npc->npc_proto;
116 	const struct ip *ip = npc->npc_ip.v4;
117 	in_port_t dport;
118 
119 	KASSERT(npf_iscached(npc, NPC_IP46));
120 	KASSERT(npf_iscached(npc, NPC_LAYER4));
121 
122 	/* Check for low TTL.  Also, we support outbound NAT only. */
123 	if (ip->ip_ttl > TR_MAX_TTL || di != PFIL_OUT) {
124 		return false;
125 	}
126 
127 	switch (proto) {
128 	case IPPROTO_TCP: {
129 		const struct tcphdr *th = npc->npc_l4.tcp;
130 		dport = ntohs(th->th_dport);
131 		break;
132 	}
133 	case IPPROTO_UDP: {
134 		const struct udphdr *uh = npc->npc_l4.udp;
135 		dport = ntohs(uh->uh_dport);
136 		break;
137 	}
138 	case IPPROTO_ICMP:
139 	case IPPROTO_ICMPV6:
140 		/* Just to pass the test below. */
141 		dport = TR_BASE_PORT;
142 		break;
143 	default:
144 		return false;
145 	}
146 
147 	/* Handle TCP/UDP traceroute - check for port range. */
148 	if (dport < TR_BASE_PORT || dport > TR_PORT_RANGE) {
149 		return false;
150 	}
151 
152 	/* Associate ALG with translation entry. */
153 	npf_nat_setalg(nt, alg_icmp, 0);
154 	return true;
155 }
156 
157 /*
158  * npfa_icmp{4,6}_inspect: retrieve unique identifiers - either ICMP query
159  * ID or TCP/UDP ports of the original packet, which is embedded.
160  */
161 
162 static bool
163 npfa_icmp4_inspect(const int type, npf_cache_t *npc, nbuf_t *nbuf)
164 {
165 	u_int offby;
166 
167 	/* Per RFC 792. */
168 	switch (type) {
169 	case ICMP_UNREACH:
170 	case ICMP_SOURCEQUENCH:
171 	case ICMP_REDIRECT:
172 	case ICMP_TIMXCEED:
173 	case ICMP_PARAMPROB:
174 		if (npc == NULL) {
175 			return false;
176 		}
177 		/* Should contain original IP header. */
178 		if (!nbuf_advance(nbuf, offsetof(struct icmp, icmp_ip), 0)) {
179 			return false;
180 		}
181 		return (npf_cache_all(npc, nbuf) & NPC_LAYER4) != 0;
182 
183 	case ICMP_ECHOREPLY:
184 	case ICMP_ECHO:
185 	case ICMP_TSTAMP:
186 	case ICMP_TSTAMPREPLY:
187 	case ICMP_IREQ:
188 	case ICMP_IREQREPLY:
189 		/* Should contain ICMP query ID - ensure. */
190 		offby = offsetof(struct icmp, icmp_id);
191 		if (!nbuf_advance(nbuf, offby, sizeof(uint16_t))) {
192 			return false;
193 		}
194 		npc->npc_info |= NPC_ICMP_ID;
195 		return true;
196 	default:
197 		break;
198 	}
199 	return false;
200 }
201 
202 static bool
203 npfa_icmp6_inspect(const int type, npf_cache_t *npc, nbuf_t *nbuf)
204 {
205 	u_int offby;
206 
207 	/* Per RFC 4443. */
208 	switch (type) {
209 	case ICMP6_DST_UNREACH:
210 	case ICMP6_PACKET_TOO_BIG:
211 	case ICMP6_TIME_EXCEEDED:
212 	case ICMP6_PARAM_PROB:
213 		if (npc == NULL) {
214 			return false;
215 		}
216 		/* Should contain original IP header. */
217 		if (!nbuf_advance(nbuf, sizeof(struct icmp6_hdr), 0)) {
218 			return false;
219 		}
220 		return (npf_cache_all(npc, nbuf) & NPC_LAYER4) != 0;
221 
222 	case ICMP6_ECHO_REQUEST:
223 	case ICMP6_ECHO_REPLY:
224 		/* Should contain ICMP query ID - ensure. */
225 		offby = offsetof(struct icmp6_hdr, icmp6_id);
226 		if (!nbuf_advance(nbuf, offby, sizeof(uint16_t))) {
227 			return false;
228 		}
229 		npc->npc_info |= NPC_ICMP_ID;
230 		return true;
231 	default:
232 		break;
233 	}
234 	return false;
235 }
236 
237 /*
238  * npfa_icmp_session: ALG ICMP inspector.
239  *
240  * => Returns true if "enpc" is filled.
241  */
242 static bool
243 npfa_icmp_inspect(npf_cache_t *npc, nbuf_t *nbuf, npf_cache_t *enpc)
244 {
245 	bool ret;
246 
247 	KASSERT(npf_iscached(npc, NPC_IP46));
248 	KASSERT(npf_iscached(npc, NPC_ICMP));
249 
250 	/* Advance to ICMP header. */
251 	nbuf_reset(nbuf);
252 	if (!nbuf_advance(nbuf, npc->npc_hlen, 0)) {
253 		return false;
254 	}
255 	enpc->npc_info = 0;
256 
257 	/*
258 	 * Inspect the ICMP packet.  The relevant data might be in the
259 	 * embedded packet.  Fill the "enpc" cache, if so.
260 	 */
261 	if (npf_iscached(npc, NPC_IP4)) {
262 		const struct icmp *ic = npc->npc_l4.icmp;
263 		ret = npfa_icmp4_inspect(ic->icmp_type, enpc, nbuf);
264 	} else if (npf_iscached(npc, NPC_IP6)) {
265 		const struct icmp6_hdr *ic6 = npc->npc_l4.icmp6;
266 		ret = npfa_icmp6_inspect(ic6->icmp6_type, enpc, nbuf);
267 	} else {
268 		ret = false;
269 	}
270 	if (!ret) {
271 		return false;
272 	}
273 
274 	/* ICMP ID is the original packet, just indicate it. */
275 	if (npf_iscached(enpc, NPC_ICMP_ID)) {
276 		npc->npc_info |= NPC_ICMP_ID;
277 		return false;
278 	}
279 
280 	/* Indicate that embedded packet is in the cache. */
281 	return true;
282 }
283 
284 static npf_session_t *
285 npfa_icmp_session(npf_cache_t *npc, nbuf_t *nbuf, int di)
286 {
287 	npf_cache_t enpc;
288 
289 	/* Inspect ICMP packet for an embedded packet. */
290 	if (!npf_iscached(npc, NPC_ICMP))
291 		return NULL;
292 	if (!npfa_icmp_inspect(npc, nbuf, &enpc))
293 		return NULL;
294 
295 	/*
296 	 * Invert the identifiers of the embedded packet.
297 	 * If it is ICMP, then ensure ICMP ID.
298 	 */
299 	union l4 {
300 		struct tcphdr th;
301 		struct udphdr uh;
302 	} l4;
303 	bool ret, forw;
304 
305 	#define	SWAP(type, x, y) { type tmp = x; x = y; y = tmp; }
306 	SWAP(npf_addr_t *, enpc.npc_ips[NPF_SRC], enpc.npc_ips[NPF_DST]);
307 
308 	switch (enpc.npc_proto) {
309 	case IPPROTO_TCP:
310 		l4.th.th_sport = enpc.npc_l4.tcp->th_dport;
311 		l4.th.th_dport = enpc.npc_l4.tcp->th_sport;
312 		enpc.npc_l4.tcp = &l4.th;
313 		break;
314 	case IPPROTO_UDP:
315 		l4.uh.uh_sport = enpc.npc_l4.udp->uh_dport;
316 		l4.uh.uh_dport = enpc.npc_l4.udp->uh_sport;
317 		enpc.npc_l4.udp = &l4.uh;
318 		break;
319 	case IPPROTO_ICMP: {
320 		const struct icmp *ic = enpc.npc_l4.icmp;
321 		ret = npfa_icmp4_inspect(ic->icmp_type, &enpc, nbuf);
322 		if (!ret || !npf_iscached(&enpc, NPC_ICMP_ID))
323 			return false;
324 		break;
325 	}
326 	case IPPROTO_ICMPV6: {
327 		const struct icmp6_hdr *ic6 = enpc.npc_l4.icmp6;
328 		ret = npfa_icmp6_inspect(ic6->icmp6_type, &enpc, nbuf);
329 		if (!ret || !npf_iscached(&enpc, NPC_ICMP_ID))
330 			return false;
331 		break;
332 	}
333 	default:
334 		return false;
335 	}
336 
337 	/* Lookup for a session using embedded packet. */
338 	return npf_session_lookup(&enpc, nbuf, di, &forw);
339 }
340 
341 /*
342  * npfa_icmp_nat: ALG translator - rewrites IP address in the IP header
343  * which is embedded in ICMP packet.  Note: backwards stream only.
344  */
345 static bool
346 npfa_icmp_nat(npf_cache_t *npc, nbuf_t *nbuf, npf_nat_t *nt, int forw)
347 {
348 	npf_cache_t enpc;
349 
350 	if (forw || !npf_iscached(npc, NPC_ICMP))
351 		return false;
352 	if (!npfa_icmp_inspect(npc, nbuf, &enpc))
353 		return false;
354 
355 	KASSERT(npf_iscached(&enpc, NPC_IP46));
356 	KASSERT(npf_iscached(&enpc, NPC_LAYER4));
357 
358 	struct icmp *ic = npc->npc_l4.icmp;
359 	uint16_t cksum = ic->icmp_cksum;
360 
361 	CTASSERT(offsetof(struct icmp, icmp_cksum) ==
362 	    offsetof(struct icmp6_hdr, icmp6_cksum));
363 
364 	/*
365 	 * Retrieve the original address and port, then calculate ICMP
366 	 * checksum for these changes in the embedded packet.  While data
367 	 * is not rewritten in the cache, save IP and TCP/UDP checksums.
368 	 *
369 	 * XXX: Assumes NPF_NATOUT (source address/port).  Currently,
370 	 * npfa_icmp_match() matches only for the PFIL_OUT traffic.
371 	 */
372 	const int proto = enpc.npc_proto;
373 	uint16_t ipcksum = 0, l4cksum = 0;
374 	npf_addr_t *addr;
375 	in_port_t port;
376 
377 	npf_nat_getorig(nt, &addr, &port);
378 
379 	if (npf_iscached(&enpc, NPC_IP4)) {
380 		const struct ip *eip = enpc.npc_ip.v4;
381 		ipcksum = eip->ip_sum;
382 	}
383 	cksum = npf_addr_cksum(cksum, enpc.npc_alen, enpc.npc_ips[NPF_SRC], addr);
384 
385 	switch (proto) {
386 	case IPPROTO_TCP: {
387 		const struct tcphdr *th = enpc.npc_l4.tcp;
388 		cksum = npf_fixup16_cksum(cksum, th->th_sport, port);
389 		l4cksum = th->th_sum;
390 		break;
391 	}
392 	case IPPROTO_UDP: {
393 		const struct udphdr *uh = enpc.npc_l4.udp;
394 		cksum = npf_fixup16_cksum(cksum, uh->uh_sport, port);
395 		l4cksum = uh->uh_sum;
396 		break;
397 	}
398 	case IPPROTO_ICMP:
399 	case IPPROTO_ICMPV6:
400 		break;
401 	default:
402 		return false;
403 	}
404 
405 	/*
406 	 * Rewrite the source IP address and port of the embedded IP header,
407 	 * which represents the original packet.  This updates the checksums
408 	 * in the embedded packet.
409 	 */
410 	if (npf_nat_translate(&enpc, nbuf, nt, forw)) {
411 		return false;
412 	}
413 
414 	/*
415 	 * Finish calculation of the ICMP checksum: include the checksum
416 	 * change in the embedded packet.
417 	 */
418 	if (npf_iscached(&enpc, NPC_IP4)) {
419 		const struct ip *eip = enpc.npc_ip.v4;
420 		cksum = npf_fixup16_cksum(cksum, ipcksum, eip->ip_sum);
421 	}
422 	switch (proto) {
423 	case IPPROTO_TCP: {
424 		const struct tcphdr *th = enpc.npc_l4.tcp;
425 		cksum = npf_fixup16_cksum(cksum, l4cksum, th->th_sum);
426 		break;
427 	}
428 	case IPPROTO_UDP:
429 		if (l4cksum) {
430 			const struct udphdr *uh = enpc.npc_l4.udp;
431 			cksum = npf_fixup16_cksum(cksum, l4cksum, uh->uh_sum);
432 		}
433 		break;
434 	}
435 	ic->icmp_cksum = cksum;
436 	return true;
437 }
438