xref: /freebsd/sys/netpfil/ipfw/nat64/nat64_translate.c (revision 9f23cbd6cae82fd77edfad7173432fa8dccd0a95)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2015-2019 Yandex LLC
5  * Copyright (c) 2015-2019 Andrey V. Elsukov <ae@FreeBSD.org>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_ipstealth.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/counter.h>
37 #include <sys/errno.h>
38 #include <sys/kernel.h>
39 #include <sys/lock.h>
40 #include <sys/mbuf.h>
41 #include <sys/module.h>
42 #include <sys/rmlock.h>
43 #include <sys/rwlock.h>
44 #include <sys/socket.h>
45 #include <sys/queue.h>
46 
47 #include <net/if.h>
48 #include <net/if_var.h>
49 #include <net/if_private.h>
50 #include <net/if_pflog.h>
51 #include <net/pfil.h>
52 #include <net/netisr.h>
53 #include <net/route.h>
54 #include <net/route/nhop.h>
55 
56 #include <netinet/in.h>
57 #include <netinet/in_fib.h>
58 #include <netinet/in_var.h>
59 #include <netinet/ip.h>
60 #include <netinet/ip_var.h>
61 #include <netinet/ip_fw.h>
62 #include <netinet/ip6.h>
63 #include <netinet/icmp6.h>
64 #include <netinet/ip_icmp.h>
65 #include <netinet/tcp.h>
66 #include <netinet/udp.h>
67 #include <netinet6/in6_var.h>
68 #include <netinet6/in6_fib.h>
69 #include <netinet6/ip6_var.h>
70 #include <netinet6/ip_fw_nat64.h>
71 
72 #include <netpfil/pf/pf.h>
73 #include <netpfil/ipfw/ip_fw_private.h>
74 #include <machine/in_cksum.h>
75 
76 #include "ip_fw_nat64.h"
77 #include "nat64_translate.h"
78 
79 typedef int (*nat64_output_t)(struct ifnet *, struct mbuf *,
80     struct sockaddr *, struct nat64_counters *, void *);
81 typedef int (*nat64_output_one_t)(struct mbuf *, struct nat64_counters *,
82     void *);
83 
84 static struct nhop_object *nat64_find_route4(struct sockaddr_in *,
85     struct mbuf *);
86 static struct nhop_object *nat64_find_route6(struct sockaddr_in6 *,
87     struct mbuf *);
88 static int nat64_output_one(struct mbuf *, struct nat64_counters *, void *);
89 static int nat64_output(struct ifnet *, struct mbuf *, struct sockaddr *,
90     struct nat64_counters *, void *);
91 static int nat64_direct_output_one(struct mbuf *, struct nat64_counters *,
92     void *);
93 static int nat64_direct_output(struct ifnet *, struct mbuf *,
94     struct sockaddr *, struct nat64_counters *, void *);
95 
96 struct nat64_methods {
97 	nat64_output_t		output;
98 	nat64_output_one_t	output_one;
99 };
100 static const struct nat64_methods nat64_netisr = {
101 	.output = nat64_output,
102 	.output_one = nat64_output_one
103 };
104 static const struct nat64_methods nat64_direct = {
105 	.output = nat64_direct_output,
106 	.output_one = nat64_direct_output_one
107 };
108 
109 /* These variables should be initialized explicitly on module loading */
110 VNET_DEFINE_STATIC(const struct nat64_methods *, nat64out);
111 VNET_DEFINE_STATIC(const int *, nat64ipstealth);
112 VNET_DEFINE_STATIC(const int *, nat64ip6stealth);
113 #define	V_nat64out		VNET(nat64out)
114 #define	V_nat64ipstealth	VNET(nat64ipstealth)
115 #define	V_nat64ip6stealth	VNET(nat64ip6stealth)
116 
117 static const int stealth_on = 1;
118 #ifndef IPSTEALTH
119 static const int stealth_off = 0;
120 #endif
121 
122 void
123 nat64_set_output_method(int direct)
124 {
125 
126 	if (direct != 0) {
127 		V_nat64out = &nat64_direct;
128 #ifdef IPSTEALTH
129 		/* Honor corresponding variables, if IPSTEALTH is defined */
130 		V_nat64ipstealth = &V_ipstealth;
131 		V_nat64ip6stealth = &V_ip6stealth;
132 #else
133 		/* otherwise we need to decrement HLIM/TTL for direct case */
134 		V_nat64ipstealth = V_nat64ip6stealth = &stealth_off;
135 #endif
136 	} else {
137 		V_nat64out = &nat64_netisr;
138 		/* Leave TTL/HLIM decrementing to forwarding code */
139 		V_nat64ipstealth = V_nat64ip6stealth = &stealth_on;
140 	}
141 }
142 
143 int
144 nat64_get_output_method(void)
145 {
146 
147 	return (V_nat64out == &nat64_direct ? 1: 0);
148 }
149 
150 static void
151 nat64_log(struct pfloghdr *logdata, struct mbuf *m, sa_family_t family)
152 {
153 
154 	logdata->dir = PF_OUT;
155 	logdata->af = family;
156 	ipfw_bpf_mtap2(logdata, PFLOG_HDRLEN, m);
157 }
158 
159 static int
160 nat64_direct_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
161     struct nat64_counters *stats, void *logdata)
162 {
163 	int error;
164 
165 	if (logdata != NULL)
166 		nat64_log(logdata, m, dst->sa_family);
167 	error = (*ifp->if_output)(ifp, m, dst, NULL);
168 	if (error != 0)
169 		NAT64STAT_INC(stats, oerrors);
170 	return (error);
171 }
172 
173 static int
174 nat64_direct_output_one(struct mbuf *m, struct nat64_counters *stats,
175     void *logdata)
176 {
177 	struct nhop_object *nh4 = NULL;
178 	struct nhop_object *nh6 = NULL;
179 	struct sockaddr_in6 dst6;
180 	struct sockaddr_in dst4;
181 	struct sockaddr *dst;
182 	struct ip6_hdr *ip6;
183 	struct ip *ip4;
184 	struct ifnet *ifp;
185 	int error;
186 
187 	ip4 = mtod(m, struct ip *);
188 	error = 0;
189 	switch (ip4->ip_v) {
190 	case IPVERSION:
191 		dst4.sin_addr = ip4->ip_dst;
192 		nh4 = nat64_find_route4(&dst4, m);
193 		if (nh4 == NULL) {
194 			NAT64STAT_INC(stats, noroute4);
195 			error = EHOSTUNREACH;
196 		} else {
197 			ifp = nh4->nh_ifp;
198 			dst = (struct sockaddr *)&dst4;
199 		}
200 		break;
201 	case (IPV6_VERSION >> 4):
202 		ip6 = mtod(m, struct ip6_hdr *);
203 		dst6.sin6_addr = ip6->ip6_dst;
204 		nh6 = nat64_find_route6(&dst6, m);
205 		if (nh6 == NULL) {
206 			NAT64STAT_INC(stats, noroute6);
207 			error = EHOSTUNREACH;
208 		} else {
209 			ifp = nh6->nh_ifp;
210 			dst = (struct sockaddr *)&dst6;
211 		}
212 		break;
213 	default:
214 		m_freem(m);
215 		NAT64STAT_INC(stats, dropped);
216 		DPRINTF(DP_DROPS, "dropped due to unknown IP version");
217 		return (EAFNOSUPPORT);
218 	}
219 	if (error != 0) {
220 		m_freem(m);
221 		return (EHOSTUNREACH);
222 	}
223 	if (logdata != NULL)
224 		nat64_log(logdata, m, dst->sa_family);
225 	error = (*ifp->if_output)(ifp, m, dst, NULL);
226 	if (error != 0)
227 		NAT64STAT_INC(stats, oerrors);
228 	return (error);
229 }
230 
231 static int
232 nat64_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
233     struct nat64_counters *stats, void *logdata)
234 {
235 	struct ip *ip4;
236 	int ret, af;
237 
238 	ip4 = mtod(m, struct ip *);
239 	switch (ip4->ip_v) {
240 	case IPVERSION:
241 		af = AF_INET;
242 		ret = NETISR_IP;
243 		break;
244 	case (IPV6_VERSION >> 4):
245 		af = AF_INET6;
246 		ret = NETISR_IPV6;
247 		break;
248 	default:
249 		m_freem(m);
250 		NAT64STAT_INC(stats, dropped);
251 		DPRINTF(DP_DROPS, "unknown IP version");
252 		return (EAFNOSUPPORT);
253 	}
254 	if (logdata != NULL)
255 		nat64_log(logdata, m, af);
256 	if (m->m_pkthdr.rcvif == NULL)
257 		m->m_pkthdr.rcvif = V_loif;
258 	ret = netisr_queue(ret, m);
259 	if (ret != 0)
260 		NAT64STAT_INC(stats, oerrors);
261 	return (ret);
262 }
263 
264 static int
265 nat64_output_one(struct mbuf *m, struct nat64_counters *stats, void *logdata)
266 {
267 
268 	return (nat64_output(NULL, m, NULL, stats, logdata));
269 }
270 
271 /*
272  * Check the given IPv6 prefix and length according to RFC6052:
273  *   The prefixes can only have one of the following lengths:
274  *   32, 40, 48, 56, 64, or 96 (The Well-Known Prefix is 96 bits long).
275  * Returns zero on success, otherwise EINVAL.
276  */
277 int
278 nat64_check_prefixlen(int length)
279 {
280 
281 	switch (length) {
282 	case 32:
283 	case 40:
284 	case 48:
285 	case 56:
286 	case 64:
287 	case 96:
288 		return (0);
289 	}
290 	return (EINVAL);
291 }
292 
293 int
294 nat64_check_prefix6(const struct in6_addr *prefix, int length)
295 {
296 
297 	if (nat64_check_prefixlen(length) != 0)
298 		return (EINVAL);
299 
300 	/* Well-known prefix has 96 prefix length */
301 	if (IN6_IS_ADDR_WKPFX(prefix) && length != 96)
302 		return (EINVAL);
303 
304 	/* Bits 64 to 71 must be set to zero */
305 	if (prefix->__u6_addr.__u6_addr8[8] != 0)
306 		return (EINVAL);
307 
308 	/* Some extra checks */
309 	if (IN6_IS_ADDR_MULTICAST(prefix) ||
310 	    IN6_IS_ADDR_UNSPECIFIED(prefix) ||
311 	    IN6_IS_ADDR_LOOPBACK(prefix))
312 		return (EINVAL);
313 	return (0);
314 }
315 
316 int
317 nat64_check_private_ip4(const struct nat64_config *cfg, in_addr_t ia)
318 {
319 
320 	if (cfg->flags & NAT64_ALLOW_PRIVATE)
321 		return (0);
322 
323 	/* WKPFX must not be used to represent non-global IPv4 addresses */
324 	if (cfg->flags & NAT64_WKPFX) {
325 		/* IN_PRIVATE */
326 		if ((ia & htonl(0xff000000)) == htonl(0x0a000000) ||
327 		    (ia & htonl(0xfff00000)) == htonl(0xac100000) ||
328 		    (ia & htonl(0xffff0000)) == htonl(0xc0a80000))
329 			return (1);
330 		/*
331 		 * RFC 5735:
332 		 *  192.0.0.0/24 - reserved for IETF protocol assignments
333 		 *  192.88.99.0/24 - for use as 6to4 relay anycast addresses
334 		 *  198.18.0.0/15 - for use in benchmark tests
335 		 *  192.0.2.0/24, 198.51.100.0/24, 203.0.113.0/24 - for use
336 		 *   in documentation and example code
337 		 */
338 		if ((ia & htonl(0xffffff00)) == htonl(0xc0000000) ||
339 		    (ia & htonl(0xffffff00)) == htonl(0xc0586300) ||
340 		    (ia & htonl(0xfffffe00)) == htonl(0xc6120000) ||
341 		    (ia & htonl(0xffffff00)) == htonl(0xc0000200) ||
342 		    (ia & htonl(0xfffffe00)) == htonl(0xc6336400) ||
343 		    (ia & htonl(0xffffff00)) == htonl(0xcb007100))
344 			return (1);
345 	}
346 	return (0);
347 }
348 
349 /*
350  * Embed @ia IPv4 address into @ip6 IPv6 address.
351  * Place to embedding determined from prefix length @plen.
352  */
353 void
354 nat64_embed_ip4(struct in6_addr *ip6, int plen, in_addr_t ia)
355 {
356 
357 	switch (plen) {
358 	case 32:
359 	case 96:
360 		ip6->s6_addr32[plen / 32] = ia;
361 		break;
362 	case 40:
363 	case 48:
364 	case 56:
365 		/*
366 		 * Preserve prefix bits.
367 		 * Since suffix bits should be zero and reserved for future
368 		 * use, we just overwrite the whole word, where they are.
369 		 */
370 		ip6->s6_addr32[1] &= 0xffffffff << (32 - plen % 32);
371 #if BYTE_ORDER == BIG_ENDIAN
372 		ip6->s6_addr32[1] |= ia >> (plen % 32);
373 		ip6->s6_addr32[2] = ia << (24 - plen % 32);
374 #elif BYTE_ORDER == LITTLE_ENDIAN
375 		ip6->s6_addr32[1] |= ia << (plen % 32);
376 		ip6->s6_addr32[2] = ia >> (24 - plen % 32);
377 #endif
378 		break;
379 	case 64:
380 #if BYTE_ORDER == BIG_ENDIAN
381 		ip6->s6_addr32[2] = ia >> 8;
382 		ip6->s6_addr32[3] = ia << 24;
383 #elif BYTE_ORDER == LITTLE_ENDIAN
384 		ip6->s6_addr32[2] = ia << 8;
385 		ip6->s6_addr32[3] = ia >> 24;
386 #endif
387 		break;
388 	default:
389 		panic("Wrong plen: %d", plen);
390 	};
391 	/*
392 	 * Bits 64 to 71 of the address are reserved for compatibility
393 	 * with the host identifier format defined in the IPv6 addressing
394 	 * architecture [RFC4291]. These bits MUST be set to zero.
395 	 */
396 	ip6->s6_addr8[8] = 0;
397 }
398 
399 in_addr_t
400 nat64_extract_ip4(const struct in6_addr *ip6, int plen)
401 {
402 	in_addr_t ia;
403 
404 	/*
405 	 * According to RFC 6052 p2.2:
406 	 * IPv4-embedded IPv6 addresses are composed of a variable-length
407 	 * prefix, the embedded IPv4 address, and a variable length suffix.
408 	 * The suffix bits are reserved for future extensions and SHOULD
409 	 * be set to zero.
410 	 */
411 	switch (plen) {
412 	case 32:
413 		if (ip6->s6_addr32[3] != 0 || ip6->s6_addr32[2] != 0)
414 			goto badip6;
415 		break;
416 	case 40:
417 		if (ip6->s6_addr32[3] != 0 ||
418 		    (ip6->s6_addr32[2] & htonl(0xff00ffff)) != 0)
419 			goto badip6;
420 		break;
421 	case 48:
422 		if (ip6->s6_addr32[3] != 0 ||
423 		    (ip6->s6_addr32[2] & htonl(0xff0000ff)) != 0)
424 			goto badip6;
425 		break;
426 	case 56:
427 		if (ip6->s6_addr32[3] != 0 || ip6->s6_addr8[8] != 0)
428 			goto badip6;
429 		break;
430 	case 64:
431 		if (ip6->s6_addr8[8] != 0 ||
432 		    (ip6->s6_addr32[3] & htonl(0x00ffffff)) != 0)
433 			goto badip6;
434 	};
435 	switch (plen) {
436 	case 32:
437 	case 96:
438 		ia = ip6->s6_addr32[plen / 32];
439 		break;
440 	case 40:
441 	case 48:
442 	case 56:
443 #if BYTE_ORDER == BIG_ENDIAN
444 		ia = (ip6->s6_addr32[1] << (plen % 32)) |
445 		    (ip6->s6_addr32[2] >> (24 - plen % 32));
446 #elif BYTE_ORDER == LITTLE_ENDIAN
447 		ia = (ip6->s6_addr32[1] >> (plen % 32)) |
448 		    (ip6->s6_addr32[2] << (24 - plen % 32));
449 #endif
450 		break;
451 	case 64:
452 #if BYTE_ORDER == BIG_ENDIAN
453 		ia = (ip6->s6_addr32[2] << 8) | (ip6->s6_addr32[3] >> 24);
454 #elif BYTE_ORDER == LITTLE_ENDIAN
455 		ia = (ip6->s6_addr32[2] >> 8) | (ip6->s6_addr32[3] << 24);
456 #endif
457 		break;
458 	default:
459 		return (0);
460 	};
461 	if (nat64_check_ip4(ia) == 0)
462 		return (ia);
463 
464 	DPRINTF(DP_GENERIC | DP_DROPS,
465 	    "invalid destination address: %08x", ia);
466 	return (0);
467 badip6:
468 	DPRINTF(DP_GENERIC | DP_DROPS, "invalid IPv4-embedded IPv6 address");
469 	return (0);
470 }
471 
472 /*
473  * According to RFC 1624 the equation for incremental checksum update is:
474  *	HC' = ~(~HC + ~m + m')	--	[Eqn. 3]
475  *	HC' = HC - ~m - m'	--	[Eqn. 4]
476  * So, when we are replacing IPv4 addresses to IPv6, we
477  * can assume, that new bytes previously were zeros, and vise versa -
478  * when we replacing IPv6 addresses to IPv4, now unused bytes become
479  * zeros. The payload length in pseudo header has bigger size, but one
480  * half of it should be zero. Using the equation 4 we get:
481  *	HC' = HC - (~m0 + m0')	-- m0 is first changed word
482  *	HC' = (HC - (~m0 + m0')) - (~m1 + m1')	-- m1 is second changed word
483  *	HC' = HC - ~m0 - m0' - ~m1 - m1' - ... =
484  *	  = HC - sum(~m[i] + m'[i])
485  *
486  * The function result should be used as follows:
487  *	IPv6 to IPv4:	HC' = cksum_add(HC, result)
488  *	IPv4 to IPv6:	HC' = cksum_add(HC, ~result)
489  */
490 static uint16_t
491 nat64_cksum_convert(struct ip6_hdr *ip6, struct ip *ip)
492 {
493 	uint32_t sum;
494 	uint16_t *p;
495 
496 	sum = ~ip->ip_src.s_addr >> 16;
497 	sum += ~ip->ip_src.s_addr & 0xffff;
498 	sum += ~ip->ip_dst.s_addr >> 16;
499 	sum += ~ip->ip_dst.s_addr & 0xffff;
500 
501 	for (p = (uint16_t *)&ip6->ip6_src;
502 	    p < (uint16_t *)(&ip6->ip6_src + 2); p++)
503 		sum += *p;
504 
505 	while (sum >> 16)
506 		sum = (sum & 0xffff) + (sum >> 16);
507 	return (sum);
508 }
509 
510 static void
511 nat64_init_ip4hdr(const struct ip6_hdr *ip6, const struct ip6_frag *frag,
512     uint16_t plen, uint8_t proto, struct ip *ip)
513 {
514 
515 	/* assume addresses are already initialized */
516 	ip->ip_v = IPVERSION;
517 	ip->ip_hl = sizeof(*ip) >> 2;
518 	ip->ip_tos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
519 	ip->ip_len = htons(sizeof(*ip) + plen);
520 	ip->ip_ttl = ip6->ip6_hlim;
521 	if (*V_nat64ip6stealth == 0)
522 		ip->ip_ttl -= IPV6_HLIMDEC;
523 	ip->ip_sum = 0;
524 	ip->ip_p = (proto == IPPROTO_ICMPV6) ? IPPROTO_ICMP: proto;
525 	ip_fillid(ip);
526 	if (frag != NULL) {
527 		ip->ip_off = htons(ntohs(frag->ip6f_offlg) >> 3);
528 		if (frag->ip6f_offlg & IP6F_MORE_FRAG)
529 			ip->ip_off |= htons(IP_MF);
530 	} else {
531 		ip->ip_off = htons(IP_DF);
532 	}
533 	ip->ip_sum = in_cksum_hdr(ip);
534 }
535 
536 #define	FRAGSZ(mtu) ((mtu) - sizeof(struct ip6_hdr) - sizeof(struct ip6_frag))
537 static NAT64NOINLINE int
538 nat64_fragment6(struct nat64_counters *stats, struct ip6_hdr *ip6,
539     struct mbufq *mq, struct mbuf *m, uint32_t mtu, uint16_t ip_id,
540     uint16_t ip_off)
541 {
542 	struct ip6_frag ip6f;
543 	struct mbuf *n;
544 	uint16_t hlen, len, offset;
545 	int plen;
546 
547 	plen = ntohs(ip6->ip6_plen);
548 	hlen = sizeof(struct ip6_hdr);
549 
550 	/* Fragmentation isn't needed */
551 	if (ip_off == 0 && plen <= mtu - hlen) {
552 		M_PREPEND(m, hlen, M_NOWAIT);
553 		if (m == NULL) {
554 			NAT64STAT_INC(stats, nomem);
555 			return (ENOMEM);
556 		}
557 		bcopy(ip6, mtod(m, void *), hlen);
558 		if (mbufq_enqueue(mq, m) != 0) {
559 			m_freem(m);
560 			NAT64STAT_INC(stats, dropped);
561 			DPRINTF(DP_DROPS, "dropped due to mbufq overflow");
562 			return (ENOBUFS);
563 		}
564 		return (0);
565 	}
566 
567 	hlen += sizeof(struct ip6_frag);
568 	ip6f.ip6f_reserved = 0;
569 	ip6f.ip6f_nxt = ip6->ip6_nxt;
570 	ip6->ip6_nxt = IPPROTO_FRAGMENT;
571 	if (ip_off != 0) {
572 		/*
573 		 * We have got an IPv4 fragment.
574 		 * Use offset value and ip_id from original fragment.
575 		 */
576 		ip6f.ip6f_ident = htonl(ntohs(ip_id));
577 		offset = (ntohs(ip_off) & IP_OFFMASK) << 3;
578 		NAT64STAT_INC(stats, ifrags);
579 	} else {
580 		/* The packet size exceeds interface MTU */
581 		ip6f.ip6f_ident = htonl(ip6_randomid());
582 		offset = 0; /* First fragment*/
583 	}
584 	while (plen > 0 && m != NULL) {
585 		n = NULL;
586 		len = FRAGSZ(mtu) & ~7;
587 		if (len > plen)
588 			len = plen;
589 		ip6->ip6_plen = htons(len + sizeof(ip6f));
590 		ip6f.ip6f_offlg = ntohs(offset);
591 		if (len < plen || (ip_off & htons(IP_MF)) != 0)
592 			ip6f.ip6f_offlg |= IP6F_MORE_FRAG;
593 		offset += len;
594 		plen -= len;
595 		if (plen > 0) {
596 			n = m_split(m, len, M_NOWAIT);
597 			if (n == NULL)
598 				goto fail;
599 		}
600 		M_PREPEND(m, hlen, M_NOWAIT);
601 		if (m == NULL)
602 			goto fail;
603 		bcopy(ip6, mtod(m, void *), sizeof(struct ip6_hdr));
604 		bcopy(&ip6f, mtodo(m, sizeof(struct ip6_hdr)),
605 		    sizeof(struct ip6_frag));
606 		if (mbufq_enqueue(mq, m) != 0)
607 			goto fail;
608 		m = n;
609 	}
610 	NAT64STAT_ADD(stats, ofrags, mbufq_len(mq));
611 	return (0);
612 fail:
613 	if (m != NULL)
614 		m_freem(m);
615 	if (n != NULL)
616 		m_freem(n);
617 	mbufq_drain(mq);
618 	NAT64STAT_INC(stats, nomem);
619 	return (ENOMEM);
620 }
621 
622 static struct nhop_object *
623 nat64_find_route6(struct sockaddr_in6 *dst, struct mbuf *m)
624 {
625 	struct nhop_object *nh;
626 
627 	NET_EPOCH_ASSERT();
628 	nh = fib6_lookup(M_GETFIB(m), &dst->sin6_addr, 0, NHR_NONE, 0);
629 	if (nh == NULL)
630 		return (NULL);
631 	if (nh->nh_flags & (NHF_BLACKHOLE | NHF_REJECT))
632 		return (NULL);
633 
634 	dst->sin6_family = AF_INET6;
635 	dst->sin6_len = sizeof(*dst);
636 	if (nh->nh_flags & NHF_GATEWAY)
637 		dst->sin6_addr = nh->gw6_sa.sin6_addr;
638 	dst->sin6_port = 0;
639 	dst->sin6_scope_id = 0;
640 	dst->sin6_flowinfo = 0;
641 	return (nh);
642 }
643 
644 #define	NAT64_ICMP6_PLEN	64
645 static NAT64NOINLINE void
646 nat64_icmp6_reflect(struct mbuf *m, uint8_t type, uint8_t code, uint32_t mtu,
647     struct nat64_counters *stats, void *logdata)
648 {
649 	struct icmp6_hdr *icmp6;
650 	struct ip6_hdr *ip6, *oip6;
651 	struct mbuf *n;
652 	int len, plen, proto;
653 
654 	len = 0;
655 	proto = nat64_getlasthdr(m, &len);
656 	if (proto < 0) {
657 		DPRINTF(DP_DROPS, "mbuf isn't contigious");
658 		goto freeit;
659 	}
660 	/*
661 	 * Do not send ICMPv6 in reply to ICMPv6 errors.
662 	 */
663 	if (proto == IPPROTO_ICMPV6) {
664 		if (m->m_len < len + sizeof(*icmp6)) {
665 			DPRINTF(DP_DROPS, "mbuf isn't contigious");
666 			goto freeit;
667 		}
668 		icmp6 = mtodo(m, len);
669 		if (icmp6->icmp6_type < ICMP6_ECHO_REQUEST ||
670 		    icmp6->icmp6_type == ND_REDIRECT) {
671 			DPRINTF(DP_DROPS, "do not send ICMPv6 in reply to "
672 			    "ICMPv6 errors");
673 			goto freeit;
674 		}
675 		/*
676 		 * If there are extra headers between IPv6 and ICMPv6,
677 		 * strip off them.
678 		 */
679 		if (len > sizeof(struct ip6_hdr)) {
680 			/*
681 			 * NOTE: ipfw_chk already did m_pullup() and it is
682 			 * expected that data is contigious from the start
683 			 * of IPv6 header up to the end of ICMPv6 header.
684 			 */
685 			bcopy(mtod(m, caddr_t),
686 			    mtodo(m, len - sizeof(struct ip6_hdr)),
687 			    sizeof(struct ip6_hdr));
688 			m_adj(m, len - sizeof(struct ip6_hdr));
689 		}
690 	}
691 	/*
692 	if (icmp6_ratelimit(&ip6->ip6_src, type, code))
693 		goto freeit;
694 		*/
695 	ip6 = mtod(m, struct ip6_hdr *);
696 	switch (type) {
697 	case ICMP6_DST_UNREACH:
698 	case ICMP6_PACKET_TOO_BIG:
699 	case ICMP6_TIME_EXCEEDED:
700 	case ICMP6_PARAM_PROB:
701 		break;
702 	default:
703 		goto freeit;
704 	}
705 	/* Calculate length of ICMPv6 payload */
706 	len = (m->m_pkthdr.len > NAT64_ICMP6_PLEN) ? NAT64_ICMP6_PLEN:
707 	    m->m_pkthdr.len;
708 
709 	/* Create new ICMPv6 datagram */
710 	plen = len + sizeof(struct icmp6_hdr);
711 	n = m_get2(sizeof(struct ip6_hdr) + plen + max_hdr, M_NOWAIT,
712 	    MT_HEADER, M_PKTHDR);
713 	if (n == NULL) {
714 		NAT64STAT_INC(stats, nomem);
715 		m_freem(m);
716 		return;
717 	}
718 	/*
719 	 * Move pkthdr from original mbuf. We should have initialized some
720 	 * fields, because we can reinject this mbuf to netisr and it will
721 	 * go through input path (it requires at least rcvif should be set).
722 	 * Also do M_ALIGN() to reduce chances of need to allocate new mbuf
723 	 * in the chain, when we will do M_PREPEND() or make some type of
724 	 * tunneling.
725 	 */
726 	m_move_pkthdr(n, m);
727 	M_ALIGN(n, sizeof(struct ip6_hdr) + plen + max_hdr);
728 
729 	n->m_len = n->m_pkthdr.len = sizeof(struct ip6_hdr) + plen;
730 	oip6 = mtod(n, struct ip6_hdr *);
731 	/*
732 	 * Make IPv6 source address selection for reflected datagram.
733 	 * nat64_check_ip6() doesn't allow scoped addresses, therefore
734 	 * we use zero scopeid.
735 	 */
736 	if (in6_selectsrc_addr(M_GETFIB(n), &ip6->ip6_src, 0,
737 	    n->m_pkthdr.rcvif, &oip6->ip6_src, NULL) != 0) {
738 		/*
739 		 * Failed to find proper source address, drop the packet.
740 		 */
741 		m_freem(n);
742 		goto freeit;
743 	}
744 	oip6->ip6_dst = ip6->ip6_src;
745 	oip6->ip6_nxt = IPPROTO_ICMPV6;
746 	oip6->ip6_flow = 0;
747 	oip6->ip6_vfc |= IPV6_VERSION;
748 	oip6->ip6_hlim = V_ip6_defhlim;
749 	oip6->ip6_plen = htons(plen);
750 
751 	icmp6 = mtodo(n, sizeof(struct ip6_hdr));
752 	icmp6->icmp6_cksum = 0;
753 	icmp6->icmp6_type = type;
754 	icmp6->icmp6_code = code;
755 	icmp6->icmp6_mtu = htonl(mtu);
756 
757 	m_copydata(m, 0, len, mtodo(n, sizeof(struct ip6_hdr) +
758 	    sizeof(struct icmp6_hdr)));
759 	icmp6->icmp6_cksum = in6_cksum(n, IPPROTO_ICMPV6,
760 	    sizeof(struct ip6_hdr), plen);
761 	m_freem(m);
762 	V_nat64out->output_one(n, stats, logdata);
763 	return;
764 freeit:
765 	NAT64STAT_INC(stats, dropped);
766 	m_freem(m);
767 }
768 
769 static struct nhop_object *
770 nat64_find_route4(struct sockaddr_in *dst, struct mbuf *m)
771 {
772 	struct nhop_object *nh;
773 
774 	NET_EPOCH_ASSERT();
775 	nh = fib4_lookup(M_GETFIB(m), dst->sin_addr, 0, NHR_NONE, 0);
776 	if (nh == NULL)
777 		return (NULL);
778 	if (nh->nh_flags & (NHF_BLACKHOLE | NHF_BROADCAST | NHF_REJECT))
779 		return (NULL);
780 
781 	dst->sin_family = AF_INET;
782 	dst->sin_len = sizeof(*dst);
783 	if (nh->nh_flags & NHF_GATEWAY)
784 		dst->sin_addr = nh->gw4_sa.sin_addr;
785 	dst->sin_port = 0;
786 	return (nh);
787 }
788 
789 #define	NAT64_ICMP_PLEN	64
790 static NAT64NOINLINE void
791 nat64_icmp_reflect(struct mbuf *m, uint8_t type,
792     uint8_t code, uint16_t mtu, struct nat64_counters *stats, void *logdata)
793 {
794 	struct icmp *icmp;
795 	struct ip *ip, *oip;
796 	struct mbuf *n;
797 	int len, plen;
798 
799 	ip = mtod(m, struct ip *);
800 	/* Do not send ICMP error if packet is not the first fragment */
801 	if (ip->ip_off & ~ntohs(IP_MF|IP_DF)) {
802 		DPRINTF(DP_DROPS, "not first fragment");
803 		goto freeit;
804 	}
805 	/* Do not send ICMP in reply to ICMP errors */
806 	if (ip->ip_p == IPPROTO_ICMP) {
807 		if (m->m_len < (ip->ip_hl << 2)) {
808 			DPRINTF(DP_DROPS, "mbuf isn't contigious");
809 			goto freeit;
810 		}
811 		icmp = mtodo(m, ip->ip_hl << 2);
812 		if (!ICMP_INFOTYPE(icmp->icmp_type)) {
813 			DPRINTF(DP_DROPS, "do not send ICMP in reply to "
814 			    "ICMP errors");
815 			goto freeit;
816 		}
817 	}
818 	switch (type) {
819 	case ICMP_UNREACH:
820 	case ICMP_TIMXCEED:
821 	case ICMP_PARAMPROB:
822 		break;
823 	default:
824 		goto freeit;
825 	}
826 	/* Calculate length of ICMP payload */
827 	len = (m->m_pkthdr.len > NAT64_ICMP_PLEN) ? (ip->ip_hl << 2) + 8:
828 	    m->m_pkthdr.len;
829 
830 	/* Create new ICMPv4 datagram */
831 	plen = len + sizeof(struct icmphdr) + sizeof(uint32_t);
832 	n = m_get2(sizeof(struct ip) + plen + max_hdr, M_NOWAIT,
833 	    MT_HEADER, M_PKTHDR);
834 	if (n == NULL) {
835 		NAT64STAT_INC(stats, nomem);
836 		m_freem(m);
837 		return;
838 	}
839 	m_move_pkthdr(n, m);
840 	M_ALIGN(n, sizeof(struct ip) + plen + max_hdr);
841 
842 	n->m_len = n->m_pkthdr.len = sizeof(struct ip) + plen;
843 	oip = mtod(n, struct ip *);
844 	oip->ip_v = IPVERSION;
845 	oip->ip_hl = sizeof(struct ip) >> 2;
846 	oip->ip_tos = 0;
847 	oip->ip_len = htons(n->m_pkthdr.len);
848 	oip->ip_ttl = V_ip_defttl;
849 	oip->ip_p = IPPROTO_ICMP;
850 	ip_fillid(oip);
851 	oip->ip_off = htons(IP_DF);
852 	oip->ip_src = ip->ip_dst;
853 	oip->ip_dst = ip->ip_src;
854 	oip->ip_sum = 0;
855 	oip->ip_sum = in_cksum_hdr(oip);
856 
857 	icmp = mtodo(n, sizeof(struct ip));
858 	icmp->icmp_type = type;
859 	icmp->icmp_code = code;
860 	icmp->icmp_cksum = 0;
861 	icmp->icmp_pmvoid = 0;
862 	icmp->icmp_nextmtu = htons(mtu);
863 	m_copydata(m, 0, len, mtodo(n, sizeof(struct ip) +
864 	    sizeof(struct icmphdr) + sizeof(uint32_t)));
865 	icmp->icmp_cksum = in_cksum_skip(n, sizeof(struct ip) + plen,
866 	    sizeof(struct ip));
867 	m_freem(m);
868 	V_nat64out->output_one(n, stats, logdata);
869 	return;
870 freeit:
871 	NAT64STAT_INC(stats, dropped);
872 	m_freem(m);
873 }
874 
875 /* Translate ICMP echo request/reply into ICMPv6 */
876 static void
877 nat64_icmp_handle_echo(struct ip6_hdr *ip6, struct icmp6_hdr *icmp6,
878     uint16_t id, uint8_t type)
879 {
880 	uint16_t old;
881 
882 	old = *(uint16_t *)icmp6;	/* save type+code in one word */
883 	icmp6->icmp6_type = type;
884 	/* Reflect ICMPv6 -> ICMPv4 type translation in the cksum */
885 	icmp6->icmp6_cksum = cksum_adjust(icmp6->icmp6_cksum,
886 	    old, *(uint16_t *)icmp6);
887 	if (id != 0) {
888 		old = icmp6->icmp6_id;
889 		icmp6->icmp6_id = id;
890 		/* Reflect ICMP id translation in the cksum */
891 		icmp6->icmp6_cksum = cksum_adjust(icmp6->icmp6_cksum,
892 		    old, id);
893 	}
894 	/* Reflect IPv6 pseudo header in the cksum */
895 	icmp6->icmp6_cksum = ~in6_cksum_pseudo(ip6, ntohs(ip6->ip6_plen),
896 	    IPPROTO_ICMPV6, ~icmp6->icmp6_cksum);
897 }
898 
899 static NAT64NOINLINE struct mbuf *
900 nat64_icmp_translate(struct mbuf *m, struct ip6_hdr *ip6, uint16_t icmpid,
901     int offset, struct nat64_config *cfg)
902 {
903 	struct ip ip;
904 	struct icmp *icmp;
905 	struct tcphdr *tcp;
906 	struct udphdr *udp;
907 	struct ip6_hdr *eip6;
908 	struct mbuf *n;
909 	uint32_t mtu;
910 	int len, hlen, plen;
911 	uint8_t type, code;
912 
913 	if (m->m_len < offset + ICMP_MINLEN)
914 		m = m_pullup(m, offset + ICMP_MINLEN);
915 	if (m == NULL) {
916 		NAT64STAT_INC(&cfg->stats, nomem);
917 		return (m);
918 	}
919 	mtu = 0;
920 	icmp = mtodo(m, offset);
921 	/* RFC 7915 p4.2 */
922 	switch (icmp->icmp_type) {
923 	case ICMP_ECHOREPLY:
924 		type = ICMP6_ECHO_REPLY;
925 		code = 0;
926 		break;
927 	case ICMP_UNREACH:
928 		type = ICMP6_DST_UNREACH;
929 		switch (icmp->icmp_code) {
930 		case ICMP_UNREACH_NET:
931 		case ICMP_UNREACH_HOST:
932 		case ICMP_UNREACH_SRCFAIL:
933 		case ICMP_UNREACH_NET_UNKNOWN:
934 		case ICMP_UNREACH_HOST_UNKNOWN:
935 		case ICMP_UNREACH_TOSNET:
936 		case ICMP_UNREACH_TOSHOST:
937 			code = ICMP6_DST_UNREACH_NOROUTE;
938 			break;
939 		case ICMP_UNREACH_PROTOCOL:
940 			type = ICMP6_PARAM_PROB;
941 			code = ICMP6_PARAMPROB_NEXTHEADER;
942 			break;
943 		case ICMP_UNREACH_PORT:
944 			code = ICMP6_DST_UNREACH_NOPORT;
945 			break;
946 		case ICMP_UNREACH_NEEDFRAG:
947 			type = ICMP6_PACKET_TOO_BIG;
948 			code = 0;
949 			/* XXX: needs an additional look */
950 			mtu = max(IPV6_MMTU, ntohs(icmp->icmp_nextmtu) + 20);
951 			break;
952 		case ICMP_UNREACH_NET_PROHIB:
953 		case ICMP_UNREACH_HOST_PROHIB:
954 		case ICMP_UNREACH_FILTER_PROHIB:
955 		case ICMP_UNREACH_PRECEDENCE_CUTOFF:
956 			code = ICMP6_DST_UNREACH_ADMIN;
957 			break;
958 		default:
959 			DPRINTF(DP_DROPS, "Unsupported ICMP type %d, code %d",
960 			    icmp->icmp_type, icmp->icmp_code);
961 			goto freeit;
962 		}
963 		break;
964 	case ICMP_TIMXCEED:
965 		type = ICMP6_TIME_EXCEEDED;
966 		code = icmp->icmp_code;
967 		break;
968 	case ICMP_ECHO:
969 		type = ICMP6_ECHO_REQUEST;
970 		code = 0;
971 		break;
972 	case ICMP_PARAMPROB:
973 		type = ICMP6_PARAM_PROB;
974 		switch (icmp->icmp_code) {
975 		case ICMP_PARAMPROB_ERRATPTR:
976 		case ICMP_PARAMPROB_LENGTH:
977 			code = ICMP6_PARAMPROB_HEADER;
978 			switch (icmp->icmp_pptr) {
979 			case 0: /* Version/IHL */
980 			case 1: /* Type Of Service */
981 				mtu = icmp->icmp_pptr;
982 				break;
983 			case 2: /* Total Length */
984 			case 3: mtu = 4; /* Payload Length */
985 				break;
986 			case 8: /* Time to Live */
987 				mtu = 7; /* Hop Limit */
988 				break;
989 			case 9: /* Protocol */
990 				mtu = 6; /* Next Header */
991 				break;
992 			case 12: /* Source address */
993 			case 13:
994 			case 14:
995 			case 15:
996 				mtu = 8;
997 				break;
998 			case 16: /* Destination address */
999 			case 17:
1000 			case 18:
1001 			case 19:
1002 				mtu = 24;
1003 				break;
1004 			default: /* Silently drop */
1005 				DPRINTF(DP_DROPS, "Unsupported ICMP type %d,"
1006 				    " code %d, pptr %d", icmp->icmp_type,
1007 				    icmp->icmp_code, icmp->icmp_pptr);
1008 				goto freeit;
1009 			}
1010 			break;
1011 		default:
1012 			DPRINTF(DP_DROPS, "Unsupported ICMP type %d,"
1013 			    " code %d, pptr %d", icmp->icmp_type,
1014 			    icmp->icmp_code, icmp->icmp_pptr);
1015 			goto freeit;
1016 		}
1017 		break;
1018 	default:
1019 		DPRINTF(DP_DROPS, "Unsupported ICMP type %d, code %d",
1020 		    icmp->icmp_type, icmp->icmp_code);
1021 		goto freeit;
1022 	}
1023 	/*
1024 	 * For echo request/reply we can use original payload,
1025 	 * but we need adjust icmp_cksum, because ICMPv6 cksum covers
1026 	 * IPv6 pseudo header and ICMPv6 types differs from ICMPv4.
1027 	 */
1028 	if (type == ICMP6_ECHO_REQUEST || type == ICMP6_ECHO_REPLY) {
1029 		nat64_icmp_handle_echo(ip6, ICMP6(icmp), icmpid, type);
1030 		return (m);
1031 	}
1032 	/*
1033 	 * For other types of ICMP messages we need to translate inner
1034 	 * IPv4 header to IPv6 header.
1035 	 * Assume ICMP src is the same as payload dst
1036 	 * E.g. we have ( GWsrc1 , NATIP1 ) in outer header
1037 	 * and          ( NATIP1, Hostdst1 ) in ICMP copy header.
1038 	 * In that case, we already have map for NATIP1 and GWsrc1.
1039 	 * The only thing we need is to copy IPv6 map prefix to
1040 	 * Hostdst1.
1041 	 */
1042 	hlen = offset + ICMP_MINLEN;
1043 	if (m->m_pkthdr.len < hlen + sizeof(struct ip) + ICMP_MINLEN) {
1044 		DPRINTF(DP_DROPS, "Message is too short %d",
1045 		    m->m_pkthdr.len);
1046 		goto freeit;
1047 	}
1048 	m_copydata(m, hlen, sizeof(struct ip), (char *)&ip);
1049 	if (ip.ip_v != IPVERSION) {
1050 		DPRINTF(DP_DROPS, "Wrong IP version %d", ip.ip_v);
1051 		goto freeit;
1052 	}
1053 	hlen += ip.ip_hl << 2; /* Skip inner IP header */
1054 	if (nat64_check_ip4(ip.ip_src.s_addr) != 0 ||
1055 	    nat64_check_ip4(ip.ip_dst.s_addr) != 0 ||
1056 	    nat64_check_private_ip4(cfg, ip.ip_src.s_addr) != 0 ||
1057 	    nat64_check_private_ip4(cfg, ip.ip_dst.s_addr) != 0) {
1058 		DPRINTF(DP_DROPS, "IP addresses checks failed %04x -> %04x",
1059 		    ntohl(ip.ip_src.s_addr), ntohl(ip.ip_dst.s_addr));
1060 		goto freeit;
1061 	}
1062 	if (m->m_pkthdr.len < hlen + ICMP_MINLEN) {
1063 		DPRINTF(DP_DROPS, "Message is too short %d",
1064 		    m->m_pkthdr.len);
1065 		goto freeit;
1066 	}
1067 #if 0
1068 	/*
1069 	 * Check that inner source matches the outer destination.
1070 	 * XXX: We need some method to convert IPv4 into IPv6 address here,
1071 	 *	and compare IPv6 addresses.
1072 	 */
1073 	if (ip.ip_src.s_addr != nat64_get_ip4(&ip6->ip6_dst)) {
1074 		DPRINTF(DP_GENERIC, "Inner source doesn't match destination ",
1075 		    "%04x vs %04x", ip.ip_src.s_addr,
1076 		    nat64_get_ip4(&ip6->ip6_dst));
1077 		goto freeit;
1078 	}
1079 #endif
1080 	/*
1081 	 * Create new mbuf for ICMPv6 datagram.
1082 	 * NOTE: len is data length just after inner IP header.
1083 	 */
1084 	len = m->m_pkthdr.len - hlen;
1085 	if (sizeof(struct ip6_hdr) +
1086 	    sizeof(struct icmp6_hdr) + len > NAT64_ICMP6_PLEN)
1087 		len = NAT64_ICMP6_PLEN - sizeof(struct icmp6_hdr) -
1088 		    sizeof(struct ip6_hdr);
1089 	plen = sizeof(struct icmp6_hdr) + sizeof(struct ip6_hdr) + len;
1090 	n = m_get2(offset + plen + max_hdr, M_NOWAIT, MT_HEADER, M_PKTHDR);
1091 	if (n == NULL) {
1092 		NAT64STAT_INC(&cfg->stats, nomem);
1093 		m_freem(m);
1094 		return (NULL);
1095 	}
1096 	m_move_pkthdr(n, m);
1097 	M_ALIGN(n, offset + plen + max_hdr);
1098 	n->m_len = n->m_pkthdr.len = offset + plen;
1099 	/* Adjust ip6_plen in outer header */
1100 	ip6->ip6_plen = htons(plen);
1101 	/* Construct new inner IPv6 header */
1102 	eip6 = mtodo(n, offset + sizeof(struct icmp6_hdr));
1103 	eip6->ip6_src = ip6->ip6_dst;
1104 
1105 	/* Use the same prefix that we have in outer header */
1106 	eip6->ip6_dst = ip6->ip6_src;
1107 	MPASS(cfg->flags & NAT64_PLATPFX);
1108 	nat64_embed_ip4(&eip6->ip6_dst, cfg->plat_plen, ip.ip_dst.s_addr);
1109 
1110 	eip6->ip6_flow = htonl(ip.ip_tos << 20);
1111 	eip6->ip6_vfc |= IPV6_VERSION;
1112 	eip6->ip6_hlim = ip.ip_ttl;
1113 	eip6->ip6_plen = htons(ntohs(ip.ip_len) - (ip.ip_hl << 2));
1114 	eip6->ip6_nxt = (ip.ip_p == IPPROTO_ICMP) ? IPPROTO_ICMPV6: ip.ip_p;
1115 	m_copydata(m, hlen, len, (char *)(eip6 + 1));
1116 	/*
1117 	 * We need to translate source port in the inner ULP header,
1118 	 * and adjust ULP checksum.
1119 	 */
1120 	switch (ip.ip_p) {
1121 	case IPPROTO_TCP:
1122 		if (len < offsetof(struct tcphdr, th_sum))
1123 			break;
1124 		tcp = TCP(eip6 + 1);
1125 		if (icmpid != 0) {
1126 			tcp->th_sum = cksum_adjust(tcp->th_sum,
1127 			    tcp->th_sport, icmpid);
1128 			tcp->th_sport = icmpid;
1129 		}
1130 		tcp->th_sum = cksum_add(tcp->th_sum,
1131 		    ~nat64_cksum_convert(eip6, &ip));
1132 		break;
1133 	case IPPROTO_UDP:
1134 		if (len < offsetof(struct udphdr, uh_sum))
1135 			break;
1136 		udp = UDP(eip6 + 1);
1137 		if (icmpid != 0) {
1138 			udp->uh_sum = cksum_adjust(udp->uh_sum,
1139 			    udp->uh_sport, icmpid);
1140 			udp->uh_sport = icmpid;
1141 		}
1142 		udp->uh_sum = cksum_add(udp->uh_sum,
1143 		    ~nat64_cksum_convert(eip6, &ip));
1144 		break;
1145 	case IPPROTO_ICMP:
1146 		/*
1147 		 * Check if this is an ICMP error message for echo request
1148 		 * that we sent. I.e. ULP in the data containing invoking
1149 		 * packet is IPPROTO_ICMP and its type is ICMP_ECHO.
1150 		 */
1151 		icmp = (struct icmp *)(eip6 + 1);
1152 		if (icmp->icmp_type != ICMP_ECHO) {
1153 			m_freem(n);
1154 			goto freeit;
1155 		}
1156 		/*
1157 		 * For our client this original datagram should looks
1158 		 * like it was ICMPv6 datagram with type ICMP6_ECHO_REQUEST.
1159 		 * Thus we need adjust icmp_cksum and convert type from
1160 		 * ICMP_ECHO to ICMP6_ECHO_REQUEST.
1161 		 */
1162 		nat64_icmp_handle_echo(eip6, ICMP6(icmp), icmpid,
1163 		    ICMP6_ECHO_REQUEST);
1164 	}
1165 	m_freem(m);
1166 	/* Convert ICMPv4 into ICMPv6 header */
1167 	icmp = mtodo(n, offset);
1168 	ICMP6(icmp)->icmp6_type = type;
1169 	ICMP6(icmp)->icmp6_code = code;
1170 	ICMP6(icmp)->icmp6_mtu = htonl(mtu);
1171 	ICMP6(icmp)->icmp6_cksum = 0;
1172 	ICMP6(icmp)->icmp6_cksum = cksum_add(
1173 	    ~in6_cksum_pseudo(ip6, plen, IPPROTO_ICMPV6, 0),
1174 	    in_cksum_skip(n, n->m_pkthdr.len, offset));
1175 	return (n);
1176 freeit:
1177 	m_freem(m);
1178 	NAT64STAT_INC(&cfg->stats, dropped);
1179 	return (NULL);
1180 }
1181 
1182 int
1183 nat64_getlasthdr(struct mbuf *m, int *offset)
1184 {
1185 	struct ip6_hdr *ip6;
1186 	struct ip6_hbh *hbh;
1187 	int proto, hlen;
1188 
1189 	if (offset != NULL)
1190 		hlen = *offset;
1191 	else
1192 		hlen = 0;
1193 
1194 	if (m->m_len < hlen + sizeof(*ip6))
1195 		return (-1);
1196 
1197 	ip6 = mtodo(m, hlen);
1198 	hlen += sizeof(*ip6);
1199 	proto = ip6->ip6_nxt;
1200 	/* Skip extension headers */
1201 	while (proto == IPPROTO_HOPOPTS || proto == IPPROTO_ROUTING ||
1202 	    proto == IPPROTO_DSTOPTS) {
1203 		hbh = mtodo(m, hlen);
1204 		/*
1205 		 * We expect mbuf has contigious data up to
1206 		 * upper level header.
1207 		 */
1208 		if (m->m_len < hlen)
1209 			return (-1);
1210 		/*
1211 		 * We doesn't support Jumbo payload option,
1212 		 * so return error.
1213 		 */
1214 		if (proto == IPPROTO_HOPOPTS && ip6->ip6_plen == 0)
1215 			return (-1);
1216 		proto = hbh->ip6h_nxt;
1217 		hlen += (hbh->ip6h_len + 1) << 3;
1218 	}
1219 	if (offset != NULL)
1220 		*offset = hlen;
1221 	return (proto);
1222 }
1223 
1224 int
1225 nat64_do_handle_ip4(struct mbuf *m, struct in6_addr *saddr,
1226     struct in6_addr *daddr, uint16_t lport, struct nat64_config *cfg,
1227     void *logdata)
1228 {
1229 	struct nhop_object *nh;
1230 	struct ip6_hdr ip6;
1231 	struct sockaddr_in6 dst;
1232 	struct ip *ip;
1233 	struct mbufq mq;
1234 	uint16_t ip_id, ip_off;
1235 	uint16_t *csum;
1236 	int plen, hlen;
1237 	uint8_t proto;
1238 
1239 	ip = mtod(m, struct ip*);
1240 
1241 	if (*V_nat64ipstealth == 0 && ip->ip_ttl <= IPTTLDEC) {
1242 		nat64_icmp_reflect(m, ICMP_TIMXCEED,
1243 		    ICMP_TIMXCEED_INTRANS, 0, &cfg->stats, logdata);
1244 		return (NAT64RETURN);
1245 	}
1246 
1247 	ip6.ip6_dst = *daddr;
1248 	ip6.ip6_src = *saddr;
1249 
1250 	hlen = ip->ip_hl << 2;
1251 	plen = ntohs(ip->ip_len) - hlen;
1252 	proto = ip->ip_p;
1253 
1254 	/* Save ip_id and ip_off, both are in network byte order */
1255 	ip_id = ip->ip_id;
1256 	ip_off = ip->ip_off & htons(IP_OFFMASK | IP_MF);
1257 
1258 	/* Fragment length must be multiple of 8 octets */
1259 	if ((ip->ip_off & htons(IP_MF)) != 0 && (plen & 0x7) != 0) {
1260 		nat64_icmp_reflect(m, ICMP_PARAMPROB,
1261 		    ICMP_PARAMPROB_LENGTH, 0, &cfg->stats, logdata);
1262 		return (NAT64RETURN);
1263 	}
1264 	/* Fragmented ICMP is unsupported */
1265 	if (proto == IPPROTO_ICMP && ip_off != 0) {
1266 		DPRINTF(DP_DROPS, "dropped due to fragmented ICMP");
1267 		NAT64STAT_INC(&cfg->stats, dropped);
1268 		return (NAT64MFREE);
1269 	}
1270 
1271 	dst.sin6_addr = ip6.ip6_dst;
1272 	nh = nat64_find_route6(&dst, m);
1273 	if (nh == NULL) {
1274 		NAT64STAT_INC(&cfg->stats, noroute6);
1275 		nat64_icmp_reflect(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0,
1276 		    &cfg->stats, logdata);
1277 		return (NAT64RETURN);
1278 	}
1279 	if (nh->nh_mtu < plen + sizeof(ip6) &&
1280 	    (ip->ip_off & htons(IP_DF)) != 0) {
1281 		nat64_icmp_reflect(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG,
1282 		    FRAGSZ(nh->nh_mtu) + sizeof(struct ip), &cfg->stats, logdata);
1283 		return (NAT64RETURN);
1284 	}
1285 
1286 	ip6.ip6_flow = htonl(ip->ip_tos << 20);
1287 	ip6.ip6_vfc |= IPV6_VERSION;
1288 	ip6.ip6_hlim = ip->ip_ttl;
1289 	if (*V_nat64ipstealth == 0)
1290 		ip6.ip6_hlim -= IPTTLDEC;
1291 	ip6.ip6_plen = htons(plen);
1292 	ip6.ip6_nxt = (proto == IPPROTO_ICMP) ? IPPROTO_ICMPV6: proto;
1293 
1294 	/* Handle delayed checksums if needed. */
1295 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1296 		in_delayed_cksum(m);
1297 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1298 	}
1299 	/* Convert checksums. */
1300 	switch (proto) {
1301 	case IPPROTO_TCP:
1302 		csum = &TCP(mtodo(m, hlen))->th_sum;
1303 		if (lport != 0) {
1304 			struct tcphdr *tcp = TCP(mtodo(m, hlen));
1305 			*csum = cksum_adjust(*csum, tcp->th_dport, lport);
1306 			tcp->th_dport = lport;
1307 		}
1308 		*csum = cksum_add(*csum, ~nat64_cksum_convert(&ip6, ip));
1309 		break;
1310 	case IPPROTO_UDP:
1311 		csum = &UDP(mtodo(m, hlen))->uh_sum;
1312 		if (lport != 0) {
1313 			struct udphdr *udp = UDP(mtodo(m, hlen));
1314 			*csum = cksum_adjust(*csum, udp->uh_dport, lport);
1315 			udp->uh_dport = lport;
1316 		}
1317 		*csum = cksum_add(*csum, ~nat64_cksum_convert(&ip6, ip));
1318 		break;
1319 	case IPPROTO_ICMP:
1320 		m = nat64_icmp_translate(m, &ip6, lport, hlen, cfg);
1321 		if (m == NULL)	/* stats already accounted */
1322 			return (NAT64RETURN);
1323 	}
1324 
1325 	m_adj(m, hlen);
1326 	mbufq_init(&mq, 255);
1327 	nat64_fragment6(&cfg->stats, &ip6, &mq, m, nh->nh_mtu, ip_id, ip_off);
1328 	while ((m = mbufq_dequeue(&mq)) != NULL) {
1329 		if (V_nat64out->output(nh->nh_ifp, m, (struct sockaddr *)&dst,
1330 		    &cfg->stats, logdata) != 0)
1331 			break;
1332 		NAT64STAT_INC(&cfg->stats, opcnt46);
1333 	}
1334 	mbufq_drain(&mq);
1335 	return (NAT64RETURN);
1336 }
1337 
1338 int
1339 nat64_handle_icmp6(struct mbuf *m, int hlen, uint32_t aaddr, uint16_t aport,
1340     struct nat64_config *cfg, void *logdata)
1341 {
1342 	struct ip ip;
1343 	struct icmp6_hdr *icmp6;
1344 	struct ip6_frag *ip6f;
1345 	struct ip6_hdr *ip6, *ip6i;
1346 	uint32_t mtu;
1347 	int plen, proto;
1348 	uint8_t type, code;
1349 
1350 	if (hlen == 0) {
1351 		ip6 = mtod(m, struct ip6_hdr *);
1352 		if (nat64_check_ip6(&ip6->ip6_src) != 0 ||
1353 		    nat64_check_ip6(&ip6->ip6_dst) != 0)
1354 			return (NAT64SKIP);
1355 
1356 		proto = nat64_getlasthdr(m, &hlen);
1357 		if (proto != IPPROTO_ICMPV6) {
1358 			DPRINTF(DP_DROPS,
1359 			    "dropped due to mbuf isn't contigious");
1360 			NAT64STAT_INC(&cfg->stats, dropped);
1361 			return (NAT64MFREE);
1362 		}
1363 	}
1364 
1365 	/*
1366 	 * Translate ICMPv6 type and code to ICMPv4 (RFC7915).
1367 	 * NOTE: ICMPv6 echo handled by nat64_do_handle_ip6().
1368 	 */
1369 	icmp6 = mtodo(m, hlen);
1370 	mtu = 0;
1371 	switch (icmp6->icmp6_type) {
1372 	case ICMP6_DST_UNREACH:
1373 		type = ICMP_UNREACH;
1374 		switch (icmp6->icmp6_code) {
1375 		case ICMP6_DST_UNREACH_NOROUTE:
1376 		case ICMP6_DST_UNREACH_BEYONDSCOPE:
1377 		case ICMP6_DST_UNREACH_ADDR:
1378 			code = ICMP_UNREACH_HOST;
1379 			break;
1380 		case ICMP6_DST_UNREACH_ADMIN:
1381 			code = ICMP_UNREACH_HOST_PROHIB;
1382 			break;
1383 		case ICMP6_DST_UNREACH_NOPORT:
1384 			code = ICMP_UNREACH_PORT;
1385 			break;
1386 		default:
1387 			DPRINTF(DP_DROPS, "Unsupported ICMPv6 type %d,"
1388 			    " code %d", icmp6->icmp6_type,
1389 			    icmp6->icmp6_code);
1390 			NAT64STAT_INC(&cfg->stats, dropped);
1391 			return (NAT64MFREE);
1392 		}
1393 		break;
1394 	case ICMP6_PACKET_TOO_BIG:
1395 		type = ICMP_UNREACH;
1396 		code = ICMP_UNREACH_NEEDFRAG;
1397 		mtu = ntohl(icmp6->icmp6_mtu);
1398 		if (mtu < IPV6_MMTU) {
1399 			DPRINTF(DP_DROPS, "Wrong MTU %d in ICMPv6 type %d,"
1400 			    " code %d", mtu, icmp6->icmp6_type,
1401 			    icmp6->icmp6_code);
1402 			NAT64STAT_INC(&cfg->stats, dropped);
1403 			return (NAT64MFREE);
1404 		}
1405 		/*
1406 		 * Adjust MTU to reflect difference between
1407 		 * IPv6 an IPv4 headers.
1408 		 */
1409 		mtu -= sizeof(struct ip6_hdr) - sizeof(struct ip);
1410 		break;
1411 	case ICMP6_TIME_EXCEEDED:
1412 		type = ICMP_TIMXCEED;
1413 		code = icmp6->icmp6_code;
1414 		break;
1415 	case ICMP6_PARAM_PROB:
1416 		switch (icmp6->icmp6_code) {
1417 		case ICMP6_PARAMPROB_HEADER:
1418 			type = ICMP_PARAMPROB;
1419 			code = ICMP_PARAMPROB_ERRATPTR;
1420 			mtu = ntohl(icmp6->icmp6_pptr);
1421 			switch (mtu) {
1422 			case 0: /* Version/Traffic Class */
1423 			case 1: /* Traffic Class/Flow Label */
1424 				break;
1425 			case 4: /* Payload Length */
1426 			case 5:
1427 				mtu = 2;
1428 				break;
1429 			case 6: /* Next Header */
1430 				mtu = 9;
1431 				break;
1432 			case 7: /* Hop Limit */
1433 				mtu = 8;
1434 				break;
1435 			default:
1436 				if (mtu >= 8 && mtu <= 23) {
1437 					mtu = 12; /* Source address */
1438 					break;
1439 				}
1440 				if (mtu >= 24 && mtu <= 39) {
1441 					mtu = 16; /* Destination address */
1442 					break;
1443 				}
1444 				DPRINTF(DP_DROPS, "Unsupported ICMPv6 type %d,"
1445 				    " code %d, pptr %d", icmp6->icmp6_type,
1446 				    icmp6->icmp6_code, mtu);
1447 				NAT64STAT_INC(&cfg->stats, dropped);
1448 				return (NAT64MFREE);
1449 			}
1450 		case ICMP6_PARAMPROB_NEXTHEADER:
1451 			type = ICMP_UNREACH;
1452 			code = ICMP_UNREACH_PROTOCOL;
1453 			break;
1454 		default:
1455 			DPRINTF(DP_DROPS, "Unsupported ICMPv6 type %d,"
1456 			    " code %d, pptr %d", icmp6->icmp6_type,
1457 			    icmp6->icmp6_code, ntohl(icmp6->icmp6_pptr));
1458 			NAT64STAT_INC(&cfg->stats, dropped);
1459 			return (NAT64MFREE);
1460 		}
1461 		break;
1462 	default:
1463 		DPRINTF(DP_DROPS, "Unsupported ICMPv6 type %d, code %d",
1464 		    icmp6->icmp6_type, icmp6->icmp6_code);
1465 		NAT64STAT_INC(&cfg->stats, dropped);
1466 		return (NAT64MFREE);
1467 	}
1468 
1469 	hlen += sizeof(struct icmp6_hdr);
1470 	if (m->m_pkthdr.len < hlen + sizeof(struct ip6_hdr) + ICMP_MINLEN) {
1471 		NAT64STAT_INC(&cfg->stats, dropped);
1472 		DPRINTF(DP_DROPS, "Message is too short %d",
1473 		    m->m_pkthdr.len);
1474 		return (NAT64MFREE);
1475 	}
1476 	/*
1477 	 * We need at least ICMP_MINLEN bytes of original datagram payload
1478 	 * to generate ICMP message. It is nice that ICMP_MINLEN is equal
1479 	 * to sizeof(struct ip6_frag). So, if embedded datagram had a fragment
1480 	 * header we will not have to do m_pullup() again.
1481 	 *
1482 	 * What we have here:
1483 	 * Outer header: (IPv6iGW, v4mapPRefix+v4exthost)
1484 	 * Inner header: (v4mapPRefix+v4host, IPv6iHost) [sport, dport]
1485 	 * We need to translate it to:
1486 	 *
1487 	 * Outer header: (alias_host, v4exthost)
1488 	 * Inner header: (v4exthost, alias_host) [sport, alias_port]
1489 	 *
1490 	 * Assume caller function has checked if v4mapPRefix+v4host
1491 	 * matches configured prefix.
1492 	 * The only two things we should be provided with are mapping between
1493 	 * IPv6iHost <> alias_host and between dport and alias_port.
1494 	 */
1495 	if (m->m_len < hlen + sizeof(struct ip6_hdr) + ICMP_MINLEN)
1496 		m = m_pullup(m, hlen + sizeof(struct ip6_hdr) + ICMP_MINLEN);
1497 	if (m == NULL) {
1498 		NAT64STAT_INC(&cfg->stats, nomem);
1499 		return (NAT64RETURN);
1500 	}
1501 	ip6 = mtod(m, struct ip6_hdr *);
1502 	ip6i = mtodo(m, hlen);
1503 	ip6f = NULL;
1504 	proto = ip6i->ip6_nxt;
1505 	plen = ntohs(ip6i->ip6_plen);
1506 	hlen += sizeof(struct ip6_hdr);
1507 	if (proto == IPPROTO_FRAGMENT) {
1508 		if (m->m_pkthdr.len < hlen + sizeof(struct ip6_frag) +
1509 		    ICMP_MINLEN)
1510 			goto fail;
1511 		ip6f = mtodo(m, hlen);
1512 		proto = ip6f->ip6f_nxt;
1513 		plen -= sizeof(struct ip6_frag);
1514 		hlen += sizeof(struct ip6_frag);
1515 		/* Ajust MTU to reflect frag header size */
1516 		if (type == ICMP_UNREACH && code == ICMP_UNREACH_NEEDFRAG)
1517 			mtu -= sizeof(struct ip6_frag);
1518 	}
1519 	if (proto != IPPROTO_TCP && proto != IPPROTO_UDP) {
1520 		DPRINTF(DP_DROPS, "Unsupported proto %d in the inner header",
1521 		    proto);
1522 		goto fail;
1523 	}
1524 	if (nat64_check_ip6(&ip6i->ip6_src) != 0 ||
1525 	    nat64_check_ip6(&ip6i->ip6_dst) != 0) {
1526 		DPRINTF(DP_DROPS, "Inner addresses do not passes the check");
1527 		goto fail;
1528 	}
1529 	/* Check if outer dst is the same as inner src */
1530 	if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6i->ip6_src)) {
1531 		DPRINTF(DP_DROPS, "Inner src doesn't match outer dst");
1532 		goto fail;
1533 	}
1534 
1535 	/* Now we need to make a fake IPv4 packet to generate ICMP message */
1536 	ip.ip_dst.s_addr = aaddr;
1537 	ip.ip_src.s_addr = nat64_extract_ip4(&ip6i->ip6_src, cfg->plat_plen);
1538 	if (ip.ip_src.s_addr == 0)
1539 		goto fail;
1540 	/* XXX: Make fake ulp header */
1541 	if (V_nat64out == &nat64_direct) /* init_ip4hdr will decrement it */
1542 		ip6i->ip6_hlim += IPV6_HLIMDEC;
1543 	nat64_init_ip4hdr(ip6i, ip6f, plen, proto, &ip);
1544 	m_adj(m, hlen - sizeof(struct ip));
1545 	bcopy(&ip, mtod(m, void *), sizeof(ip));
1546 	nat64_icmp_reflect(m, type, code, (uint16_t)mtu, &cfg->stats,
1547 	    logdata);
1548 	return (NAT64RETURN);
1549 fail:
1550 	/*
1551 	 * We must call m_freem() because mbuf pointer could be
1552 	 * changed with m_pullup().
1553 	 */
1554 	m_freem(m);
1555 	NAT64STAT_INC(&cfg->stats, dropped);
1556 	return (NAT64RETURN);
1557 }
1558 
1559 int
1560 nat64_do_handle_ip6(struct mbuf *m, uint32_t aaddr, uint16_t aport,
1561     struct nat64_config *cfg, void *logdata)
1562 {
1563 	struct ip ip;
1564 	struct nhop_object *nh;
1565 	struct sockaddr_in dst;
1566 	struct ip6_frag *frag;
1567 	struct ip6_hdr *ip6;
1568 	struct icmp6_hdr *icmp6;
1569 	uint16_t *csum;
1570 	int plen, hlen, proto;
1571 
1572 	/*
1573 	 * XXX: we expect ipfw_chk() did m_pullup() up to upper level
1574 	 * protocol's headers. Also we skip some checks, that ip6_input(),
1575 	 * ip6_forward(), ip6_fastfwd() and ipfw_chk() already did.
1576 	 */
1577 	ip6 = mtod(m, struct ip6_hdr *);
1578 	if (nat64_check_ip6(&ip6->ip6_src) != 0 ||
1579 	    nat64_check_ip6(&ip6->ip6_dst) != 0) {
1580 		return (NAT64SKIP);
1581 	}
1582 
1583 	/* Starting from this point we must not return zero */
1584 	ip.ip_src.s_addr = aaddr;
1585 	if (nat64_check_ip4(ip.ip_src.s_addr) != 0) {
1586 		DPRINTF(DP_GENERIC | DP_DROPS, "invalid source address: %08x",
1587 		    ip.ip_src.s_addr);
1588 		NAT64STAT_INC(&cfg->stats, dropped);
1589 		return (NAT64MFREE);
1590 	}
1591 
1592 	ip.ip_dst.s_addr = nat64_extract_ip4(&ip6->ip6_dst, cfg->plat_plen);
1593 	if (ip.ip_dst.s_addr == 0) {
1594 		NAT64STAT_INC(&cfg->stats, dropped);
1595 		return (NAT64MFREE);
1596 	}
1597 
1598 	if (*V_nat64ip6stealth == 0 && ip6->ip6_hlim <= IPV6_HLIMDEC) {
1599 		nat64_icmp6_reflect(m, ICMP6_TIME_EXCEEDED,
1600 		    ICMP6_TIME_EXCEED_TRANSIT, 0, &cfg->stats, logdata);
1601 		return (NAT64RETURN);
1602 	}
1603 
1604 	hlen = 0;
1605 	plen = ntohs(ip6->ip6_plen);
1606 	proto = nat64_getlasthdr(m, &hlen);
1607 	if (proto < 0) {
1608 		DPRINTF(DP_DROPS, "dropped due to mbuf isn't contigious");
1609 		NAT64STAT_INC(&cfg->stats, dropped);
1610 		return (NAT64MFREE);
1611 	}
1612 	frag = NULL;
1613 	if (proto == IPPROTO_FRAGMENT) {
1614 		/* ipfw_chk should m_pullup up to frag header */
1615 		if (m->m_len < hlen + sizeof(*frag)) {
1616 			DPRINTF(DP_DROPS,
1617 			    "dropped due to mbuf isn't contigious");
1618 			NAT64STAT_INC(&cfg->stats, dropped);
1619 			return (NAT64MFREE);
1620 		}
1621 		frag = mtodo(m, hlen);
1622 		proto = frag->ip6f_nxt;
1623 		hlen += sizeof(*frag);
1624 		/* Fragmented ICMPv6 is unsupported */
1625 		if (proto == IPPROTO_ICMPV6) {
1626 			DPRINTF(DP_DROPS, "dropped due to fragmented ICMPv6");
1627 			NAT64STAT_INC(&cfg->stats, dropped);
1628 			return (NAT64MFREE);
1629 		}
1630 		/* Fragment length must be multiple of 8 octets */
1631 		if ((frag->ip6f_offlg & IP6F_MORE_FRAG) != 0 &&
1632 		    ((plen + sizeof(struct ip6_hdr) - hlen) & 0x7) != 0) {
1633 			nat64_icmp6_reflect(m, ICMP6_PARAM_PROB,
1634 			    ICMP6_PARAMPROB_HEADER,
1635 			    offsetof(struct ip6_hdr, ip6_plen), &cfg->stats,
1636 			    logdata);
1637 			return (NAT64RETURN);
1638 		}
1639 	}
1640 	plen -= hlen - sizeof(struct ip6_hdr);
1641 	if (plen < 0 || m->m_pkthdr.len < plen + hlen) {
1642 		DPRINTF(DP_DROPS, "plen %d, pkthdr.len %d, hlen %d",
1643 		    plen, m->m_pkthdr.len, hlen);
1644 		NAT64STAT_INC(&cfg->stats, dropped);
1645 		return (NAT64MFREE);
1646 	}
1647 
1648 	icmp6 = NULL;	/* Make gcc happy */
1649 	if (proto == IPPROTO_ICMPV6) {
1650 		icmp6 = mtodo(m, hlen);
1651 		if (icmp6->icmp6_type != ICMP6_ECHO_REQUEST &&
1652 		    icmp6->icmp6_type != ICMP6_ECHO_REPLY)
1653 			return (nat64_handle_icmp6(m, hlen, aaddr, aport,
1654 			    cfg, logdata));
1655 	}
1656 	dst.sin_addr.s_addr = ip.ip_dst.s_addr;
1657 	nh = nat64_find_route4(&dst, m);
1658 	if (nh == NULL) {
1659 		NAT64STAT_INC(&cfg->stats, noroute4);
1660 		nat64_icmp6_reflect(m, ICMP6_DST_UNREACH,
1661 		    ICMP6_DST_UNREACH_NOROUTE, 0, &cfg->stats, logdata);
1662 		return (NAT64RETURN);
1663 	}
1664 	if (nh->nh_mtu < plen + sizeof(ip)) {
1665 		nat64_icmp6_reflect(m, ICMP6_PACKET_TOO_BIG, 0, nh->nh_mtu,
1666 		    &cfg->stats, logdata);
1667 		return (NAT64RETURN);
1668 	}
1669 	nat64_init_ip4hdr(ip6, frag, plen, proto, &ip);
1670 
1671 	/* Handle delayed checksums if needed. */
1672 	if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) {
1673 		in6_delayed_cksum(m, plen, hlen);
1674 		m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6;
1675 	}
1676 	/* Convert checksums. */
1677 	switch (proto) {
1678 	case IPPROTO_TCP:
1679 		csum = &TCP(mtodo(m, hlen))->th_sum;
1680 		if (aport != 0) {
1681 			struct tcphdr *tcp = TCP(mtodo(m, hlen));
1682 			*csum = cksum_adjust(*csum, tcp->th_sport, aport);
1683 			tcp->th_sport = aport;
1684 		}
1685 		*csum = cksum_add(*csum, nat64_cksum_convert(ip6, &ip));
1686 		break;
1687 	case IPPROTO_UDP:
1688 		csum = &UDP(mtodo(m, hlen))->uh_sum;
1689 		if (aport != 0) {
1690 			struct udphdr *udp = UDP(mtodo(m, hlen));
1691 			*csum = cksum_adjust(*csum, udp->uh_sport, aport);
1692 			udp->uh_sport = aport;
1693 		}
1694 		*csum = cksum_add(*csum, nat64_cksum_convert(ip6, &ip));
1695 		break;
1696 	case IPPROTO_ICMPV6:
1697 		/* Checksum in ICMPv6 covers pseudo header */
1698 		csum = &icmp6->icmp6_cksum;
1699 		*csum = cksum_add(*csum, in6_cksum_pseudo(ip6, plen,
1700 		    IPPROTO_ICMPV6, 0));
1701 		/* Convert ICMPv6 types to ICMP */
1702 		proto = *(uint16_t *)icmp6; /* save old word for cksum_adjust */
1703 		if (icmp6->icmp6_type == ICMP6_ECHO_REQUEST)
1704 			icmp6->icmp6_type = ICMP_ECHO;
1705 		else /* ICMP6_ECHO_REPLY */
1706 			icmp6->icmp6_type = ICMP_ECHOREPLY;
1707 		*csum = cksum_adjust(*csum, (uint16_t)proto,
1708 		    *(uint16_t *)icmp6);
1709 		if (aport != 0) {
1710 			uint16_t old_id = icmp6->icmp6_id;
1711 			icmp6->icmp6_id = aport;
1712 			*csum = cksum_adjust(*csum, old_id, aport);
1713 		}
1714 		break;
1715 	};
1716 
1717 	m_adj(m, hlen - sizeof(ip));
1718 	bcopy(&ip, mtod(m, void *), sizeof(ip));
1719 	if (V_nat64out->output(nh->nh_ifp, m, (struct sockaddr *)&dst,
1720 	    &cfg->stats, logdata) == 0)
1721 		NAT64STAT_INC(&cfg->stats, opcnt64);
1722 	return (NAT64RETURN);
1723 }
1724