Home | History | Annotate | Line # | Download | only in npf
npf_inet.c revision 1.46
      1 /*	$NetBSD: npf_inet.c,v 1.46 2018/03/22 09:04:25 maxv Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2009-2014 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  * Various protocol related helper routines.
     34  *
     35  * This layer manipulates npf_cache_t structure i.e. caches requested headers
     36  * and stores which information was cached in the information bit field.
     37  * It is also responsibility of this layer to update or invalidate the cache
     38  * on rewrites (e.g. by translation routines).
     39  */
     40 
     41 #ifdef _KERNEL
     42 #include <sys/cdefs.h>
     43 __KERNEL_RCSID(0, "$NetBSD: npf_inet.c,v 1.46 2018/03/22 09:04:25 maxv Exp $");
     44 
     45 #include <sys/param.h>
     46 #include <sys/types.h>
     47 
     48 #include <net/pfil.h>
     49 #include <net/if.h>
     50 #include <net/ethertypes.h>
     51 #include <net/if_ether.h>
     52 
     53 #include <netinet/in_systm.h>
     54 #include <netinet/in.h>
     55 #include <netinet6/in6_var.h>
     56 #include <netinet/ip.h>
     57 #include <netinet/ip6.h>
     58 #include <netinet/tcp.h>
     59 #include <netinet/udp.h>
     60 #include <netinet/ip_icmp.h>
     61 #endif
     62 
     63 #include "npf_impl.h"
     64 
     65 /*
     66  * npf_fixup{16,32}_cksum: incremental update of the Internet checksum.
     67  */
     68 
     69 uint16_t
     70 npf_fixup16_cksum(uint16_t cksum, uint16_t odatum, uint16_t ndatum)
     71 {
     72 	uint32_t sum;
     73 
     74 	/*
     75 	 * RFC 1624:
     76 	 *	HC' = ~(~HC + ~m + m')
     77 	 *
     78 	 * Note: 1's complement sum is endian-independent (RFC 1071, page 2).
     79 	 */
     80 	sum = ~cksum & 0xffff;
     81 	sum += (~odatum & 0xffff) + ndatum;
     82 	sum = (sum >> 16) + (sum & 0xffff);
     83 	sum += (sum >> 16);
     84 
     85 	return ~sum & 0xffff;
     86 }
     87 
     88 uint16_t
     89 npf_fixup32_cksum(uint16_t cksum, uint32_t odatum, uint32_t ndatum)
     90 {
     91 	uint32_t sum;
     92 
     93 	/*
     94 	 * Checksum 32-bit datum as as two 16-bit.  Note, the first
     95 	 * 32->16 bit reduction is not necessary.
     96 	 */
     97 	sum = ~cksum & 0xffff;
     98 	sum += (~odatum & 0xffff) + (ndatum & 0xffff);
     99 
    100 	sum += (~odatum >> 16) + (ndatum >> 16);
    101 	sum = (sum >> 16) + (sum & 0xffff);
    102 	sum += (sum >> 16);
    103 	return ~sum & 0xffff;
    104 }
    105 
    106 /*
    107  * npf_addr_cksum: calculate checksum of the address, either IPv4 or IPv6.
    108  */
    109 uint16_t
    110 npf_addr_cksum(uint16_t cksum, int sz, const npf_addr_t *oaddr,
    111     const npf_addr_t *naddr)
    112 {
    113 	const uint32_t *oip32 = (const uint32_t *)oaddr;
    114 	const uint32_t *nip32 = (const uint32_t *)naddr;
    115 
    116 	KASSERT(sz % sizeof(uint32_t) == 0);
    117 	do {
    118 		cksum = npf_fixup32_cksum(cksum, *oip32++, *nip32++);
    119 		sz -= sizeof(uint32_t);
    120 	} while (sz);
    121 
    122 	return cksum;
    123 }
    124 
    125 /*
    126  * npf_addr_sum: provide IP addresses as a XORed 32-bit integer.
    127  * Note: used for hash function.
    128  */
    129 uint32_t
    130 npf_addr_mix(const int sz, const npf_addr_t *a1, const npf_addr_t *a2)
    131 {
    132 	uint32_t mix = 0;
    133 
    134 	KASSERT(sz > 0 && a1 != NULL && a2 != NULL);
    135 
    136 	for (int i = 0; i < (sz >> 2); i++) {
    137 		mix ^= a1->word32[i];
    138 		mix ^= a2->word32[i];
    139 	}
    140 	return mix;
    141 }
    142 
    143 /*
    144  * npf_addr_mask: apply the mask to a given address and store the result.
    145  */
    146 void
    147 npf_addr_mask(const npf_addr_t *addr, const npf_netmask_t mask,
    148     const int alen, npf_addr_t *out)
    149 {
    150 	const int nwords = alen >> 2;
    151 	uint_fast8_t length = mask;
    152 
    153 	/* Note: maximum length is 32 for IPv4 and 128 for IPv6. */
    154 	KASSERT(length <= NPF_MAX_NETMASK);
    155 
    156 	for (int i = 0; i < nwords; i++) {
    157 		uint32_t wordmask;
    158 
    159 		if (length >= 32) {
    160 			wordmask = htonl(0xffffffff);
    161 			length -= 32;
    162 		} else if (length) {
    163 			wordmask = htonl(0xffffffff << (32 - length));
    164 			length = 0;
    165 		} else {
    166 			wordmask = 0;
    167 		}
    168 		out->word32[i] = addr->word32[i] & wordmask;
    169 	}
    170 }
    171 
    172 /*
    173  * npf_addr_cmp: compare two addresses, either IPv4 or IPv6.
    174  *
    175  * => Return 0 if equal and negative/positive if less/greater accordingly.
    176  * => Ignore the mask, if NPF_NO_NETMASK is specified.
    177  */
    178 int
    179 npf_addr_cmp(const npf_addr_t *addr1, const npf_netmask_t mask1,
    180     const npf_addr_t *addr2, const npf_netmask_t mask2, const int alen)
    181 {
    182 	npf_addr_t realaddr1, realaddr2;
    183 
    184 	if (mask1 != NPF_NO_NETMASK) {
    185 		npf_addr_mask(addr1, mask1, alen, &realaddr1);
    186 		addr1 = &realaddr1;
    187 	}
    188 	if (mask2 != NPF_NO_NETMASK) {
    189 		npf_addr_mask(addr2, mask2, alen, &realaddr2);
    190 		addr2 = &realaddr2;
    191 	}
    192 	return memcmp(addr1, addr2, alen);
    193 }
    194 
    195 /*
    196  * npf_tcpsaw: helper to fetch SEQ, ACK, WIN and return TCP data length.
    197  *
    198  * => Returns all values in host byte-order.
    199  */
    200 int
    201 npf_tcpsaw(const npf_cache_t *npc, tcp_seq *seq, tcp_seq *ack, uint32_t *win)
    202 {
    203 	const struct tcphdr *th = npc->npc_l4.tcp;
    204 	u_int thlen;
    205 
    206 	KASSERT(npf_iscached(npc, NPC_TCP));
    207 
    208 	*seq = ntohl(th->th_seq);
    209 	*ack = ntohl(th->th_ack);
    210 	*win = (uint32_t)ntohs(th->th_win);
    211 	thlen = th->th_off << 2;
    212 
    213 	if (npf_iscached(npc, NPC_IP4)) {
    214 		const struct ip *ip = npc->npc_ip.v4;
    215 		return ntohs(ip->ip_len) - npc->npc_hlen - thlen;
    216 	} else if (npf_iscached(npc, NPC_IP6)) {
    217 		const struct ip6_hdr *ip6 = npc->npc_ip.v6;
    218 		return ntohs(ip6->ip6_plen) -
    219 		    (npc->npc_hlen - sizeof(*ip6)) - thlen;
    220 	}
    221 	return 0;
    222 }
    223 
    224 /*
    225  * npf_fetch_tcpopts: parse and return TCP options.
    226  */
    227 bool
    228 npf_fetch_tcpopts(npf_cache_t *npc, uint16_t *mss, int *wscale)
    229 {
    230 	nbuf_t *nbuf = npc->npc_nbuf;
    231 	const struct tcphdr *th = npc->npc_l4.tcp;
    232 	int topts_len, step;
    233 	uint8_t *nptr;
    234 	uint8_t val;
    235 	bool ok;
    236 
    237 	KASSERT(npf_iscached(npc, NPC_IP46));
    238 	KASSERT(npf_iscached(npc, NPC_TCP));
    239 
    240 	/* Determine if there are any TCP options, get their length. */
    241 	topts_len = (th->th_off << 2) - sizeof(struct tcphdr);
    242 	if (topts_len <= 0) {
    243 		/* No options. */
    244 		return false;
    245 	}
    246 	KASSERT(topts_len <= MAX_TCPOPTLEN);
    247 
    248 	/* First step: IP and TCP header up to options. */
    249 	step = npc->npc_hlen + sizeof(struct tcphdr);
    250 	nbuf_reset(nbuf);
    251 next:
    252 	if ((nptr = nbuf_advance(nbuf, step, 1)) == NULL) {
    253 		ok = false;
    254 		goto done;
    255 	}
    256 	val = *nptr;
    257 
    258 	switch (val) {
    259 	case TCPOPT_EOL:
    260 		/* Done. */
    261 		ok = true;
    262 		goto done;
    263 	case TCPOPT_NOP:
    264 		topts_len--;
    265 		step = 1;
    266 		break;
    267 	case TCPOPT_MAXSEG:
    268 		if ((nptr = nbuf_ensure_contig(nbuf, TCPOLEN_MAXSEG)) == NULL) {
    269 			ok = false;
    270 			goto done;
    271 		}
    272 		if (mss) {
    273 			if (*mss) {
    274 				memcpy(nptr + 2, mss, sizeof(uint16_t));
    275 			} else {
    276 				memcpy(mss, nptr + 2, sizeof(uint16_t));
    277 			}
    278 		}
    279 		topts_len -= TCPOLEN_MAXSEG;
    280 		step = TCPOLEN_MAXSEG;
    281 		break;
    282 	case TCPOPT_WINDOW:
    283 		/* TCP Window Scaling (RFC 1323). */
    284 		if ((nptr = nbuf_ensure_contig(nbuf, TCPOLEN_WINDOW)) == NULL) {
    285 			ok = false;
    286 			goto done;
    287 		}
    288 		val = *(nptr + 2);
    289 		*wscale = (val > TCP_MAX_WINSHIFT) ? TCP_MAX_WINSHIFT : val;
    290 		topts_len -= TCPOLEN_WINDOW;
    291 		step = TCPOLEN_WINDOW;
    292 		break;
    293 	default:
    294 		if ((nptr = nbuf_ensure_contig(nbuf, 2)) == NULL) {
    295 			ok = false;
    296 			goto done;
    297 		}
    298 		val = *(nptr + 1);
    299 		if (val < 2 || val > topts_len) {
    300 			ok = false;
    301 			goto done;
    302 		}
    303 		topts_len -= val;
    304 		step = val;
    305 	}
    306 
    307 	/* Any options left? */
    308 	if (__predict_true(topts_len > 0)) {
    309 		goto next;
    310 	}
    311 	ok = true;
    312 done:
    313 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    314 		npf_recache(npc);
    315 	}
    316 	return ok;
    317 }
    318 
    319 static int
    320 npf_cache_ip(npf_cache_t *npc, nbuf_t *nbuf)
    321 {
    322 	const void *nptr = nbuf_dataptr(nbuf);
    323 	const uint8_t ver = *(const uint8_t *)nptr;
    324 	int flags = 0;
    325 
    326 	/*
    327 	 * We intentionally don't read the L4 payload after IPPROTO_AH.
    328 	 */
    329 
    330 	switch (ver >> 4) {
    331 	case IPVERSION: {
    332 		struct ip *ip;
    333 
    334 		ip = nbuf_ensure_contig(nbuf, sizeof(struct ip));
    335 		if (ip == NULL) {
    336 			return NPC_FMTERR;
    337 		}
    338 
    339 		/* Retrieve the complete header. */
    340 		if ((u_int)(ip->ip_hl << 2) < sizeof(struct ip)) {
    341 			return NPC_FMTERR;
    342 		}
    343 		ip = nbuf_ensure_contig(nbuf, (u_int)(ip->ip_hl << 2));
    344 		if (ip == NULL) {
    345 			return NPC_FMTERR;
    346 		}
    347 
    348 		if (ip->ip_off & ~htons(IP_DF | IP_RF)) {
    349 			/* Note fragmentation. */
    350 			flags |= NPC_IPFRAG;
    351 		}
    352 
    353 		/* Cache: layer 3 - IPv4. */
    354 		npc->npc_alen = sizeof(struct in_addr);
    355 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip->ip_src;
    356 		npc->npc_ips[NPF_DST] = (npf_addr_t *)&ip->ip_dst;
    357 		npc->npc_hlen = ip->ip_hl << 2;
    358 		npc->npc_proto = ip->ip_p;
    359 
    360 		npc->npc_ip.v4 = ip;
    361 		flags |= NPC_IP4;
    362 		break;
    363 	}
    364 
    365 	case (IPV6_VERSION >> 4): {
    366 		struct ip6_hdr *ip6;
    367 		struct ip6_ext *ip6e;
    368 		struct ip6_frag *ip6f;
    369 		size_t off, hlen;
    370 		int frag_present;
    371 
    372 		ip6 = nbuf_ensure_contig(nbuf, sizeof(struct ip6_hdr));
    373 		if (ip6 == NULL) {
    374 			return NPC_FMTERR;
    375 		}
    376 
    377 		/*
    378 		 * XXX: We don't handle IPv6 Jumbograms.
    379 		 */
    380 
    381 		/* Set initial next-protocol value. */
    382 		hlen = sizeof(struct ip6_hdr);
    383 		npc->npc_proto = ip6->ip6_nxt;
    384 		npc->npc_hlen = hlen;
    385 
    386 		frag_present = 0;
    387 
    388 		/*
    389 		 * Advance by the length of the current header.
    390 		 */
    391 		off = nbuf_offset(nbuf);
    392 		while ((ip6e = nbuf_advance(nbuf, hlen, sizeof(*ip6e))) != NULL) {
    393 			/*
    394 			 * Determine whether we are going to continue.
    395 			 */
    396 			switch (npc->npc_proto) {
    397 			case IPPROTO_HOPOPTS:
    398 			case IPPROTO_DSTOPTS:
    399 			case IPPROTO_ROUTING:
    400 				hlen = (ip6e->ip6e_len + 1) << 3;
    401 				break;
    402 			case IPPROTO_FRAGMENT:
    403 				if (frag_present++)
    404 					return NPC_FMTERR;
    405 				ip6f = nbuf_ensure_contig(nbuf, sizeof(*ip6f));
    406 				if (ip6f == NULL)
    407 					return NPC_FMTERR;
    408 
    409 				/* RFC6946: Skip dummy fragments. */
    410 				if (!ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK) &&
    411 				    !(ip6f->ip6f_offlg & IP6F_MORE_FRAG)) {
    412 					hlen = sizeof(struct ip6_frag);
    413 					break;
    414 				}
    415 
    416 				hlen = 0;
    417 				flags |= NPC_IPFRAG;
    418 
    419 				break;
    420 			default:
    421 				hlen = 0;
    422 				break;
    423 			}
    424 
    425 			if (!hlen) {
    426 				break;
    427 			}
    428 			npc->npc_proto = ip6e->ip6e_nxt;
    429 			npc->npc_hlen += hlen;
    430 		}
    431 
    432 		if (ip6e == NULL) {
    433 			return NPC_FMTERR;
    434 		}
    435 
    436 		/*
    437 		 * Re-fetch the header pointers (nbufs might have been
    438 		 * reallocated).  Restore the original offset (if any).
    439 		 */
    440 		nbuf_reset(nbuf);
    441 		ip6 = nbuf_dataptr(nbuf);
    442 		if (off) {
    443 			nbuf_advance(nbuf, off, 0);
    444 		}
    445 
    446 		/* Cache: layer 3 - IPv6. */
    447 		npc->npc_alen = sizeof(struct in6_addr);
    448 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip6->ip6_src;
    449 		npc->npc_ips[NPF_DST] = (npf_addr_t *)&ip6->ip6_dst;
    450 
    451 		npc->npc_ip.v6 = ip6;
    452 		flags |= NPC_IP6;
    453 		break;
    454 	}
    455 	default:
    456 		break;
    457 	}
    458 	return flags;
    459 }
    460 
    461 /*
    462  * npf_cache_all: general routine to cache all relevant IP (v4 or v6)
    463  * and TCP, UDP or ICMP headers.
    464  *
    465  * => nbuf offset shall be set accordingly.
    466  */
    467 int
    468 npf_cache_all(npf_cache_t *npc)
    469 {
    470 	nbuf_t *nbuf = npc->npc_nbuf;
    471 	int flags, l4flags;
    472 	u_int hlen;
    473 
    474 	/*
    475 	 * This routine is a main point where the references are cached,
    476 	 * therefore clear the flag as we reset.
    477 	 */
    478 again:
    479 	nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    480 
    481 	/*
    482 	 * First, cache the L3 header (IPv4 or IPv6).  If IP packet is
    483 	 * fragmented, then we cannot look into L4.
    484 	 */
    485 	flags = npf_cache_ip(npc, nbuf);
    486 	if ((flags & NPC_IP46) == 0 || (flags & NPC_IPFRAG) != 0 ||
    487 	    (flags & NPC_FMTERR) != 0) {
    488 		nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    489 		npc->npc_info |= flags;
    490 		return flags;
    491 	}
    492 	hlen = npc->npc_hlen;
    493 
    494 	/*
    495 	 * Note: we guarantee that the potential "Query Id" field of the
    496 	 * ICMPv4/ICMPv6 packets is in the nbuf. This field is used in the
    497 	 * ICMP ALG.
    498 	 */
    499 	switch (npc->npc_proto) {
    500 	case IPPROTO_TCP:
    501 		/* Cache: layer 4 - TCP. */
    502 		npc->npc_l4.tcp = nbuf_advance(nbuf, hlen,
    503 		    sizeof(struct tcphdr));
    504 		l4flags = NPC_LAYER4 | NPC_TCP;
    505 		break;
    506 	case IPPROTO_UDP:
    507 		/* Cache: layer 4 - UDP. */
    508 		npc->npc_l4.udp = nbuf_advance(nbuf, hlen,
    509 		    sizeof(struct udphdr));
    510 		l4flags = NPC_LAYER4 | NPC_UDP;
    511 		break;
    512 	case IPPROTO_ICMP:
    513 		/* Cache: layer 4 - ICMPv4. */
    514 		npc->npc_l4.icmp = nbuf_advance(nbuf, hlen,
    515 		    ICMP_MINLEN);
    516 		l4flags = NPC_LAYER4 | NPC_ICMP;
    517 		break;
    518 	case IPPROTO_ICMPV6:
    519 		/* Cache: layer 4 - ICMPv6. */
    520 		npc->npc_l4.icmp6 = nbuf_advance(nbuf, hlen,
    521 		    sizeof(struct icmp6_hdr));
    522 		l4flags = NPC_LAYER4 | NPC_ICMP;
    523 		break;
    524 	default:
    525 		l4flags = 0;
    526 		break;
    527 	}
    528 
    529 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    530 		goto again;
    531 	}
    532 
    533 	/* Add the L4 flags if nbuf_advance() succeeded. */
    534 	if (l4flags && npc->npc_l4.hdr) {
    535 		flags |= l4flags;
    536 	}
    537 	npc->npc_info |= flags;
    538 	return flags;
    539 }
    540 
    541 void
    542 npf_recache(npf_cache_t *npc)
    543 {
    544 	nbuf_t *nbuf = npc->npc_nbuf;
    545 	const int mflags __diagused = npc->npc_info & (NPC_IP46 | NPC_LAYER4);
    546 	int flags __diagused;
    547 
    548 	nbuf_reset(nbuf);
    549 	npc->npc_info = 0;
    550 	flags = npf_cache_all(npc);
    551 
    552 	KASSERT((flags & mflags) == mflags);
    553 	KASSERT(nbuf_flag_p(nbuf, NBUF_DATAREF_RESET) == 0);
    554 }
    555 
    556 /*
    557  * npf_rwrip: rewrite required IP address.
    558  */
    559 bool
    560 npf_rwrip(const npf_cache_t *npc, u_int which, const npf_addr_t *addr)
    561 {
    562 	KASSERT(npf_iscached(npc, NPC_IP46));
    563 	KASSERT(which == NPF_SRC || which == NPF_DST);
    564 
    565 	memcpy(npc->npc_ips[which], addr, npc->npc_alen);
    566 	return true;
    567 }
    568 
    569 /*
    570  * npf_rwrport: rewrite required TCP/UDP port.
    571  */
    572 bool
    573 npf_rwrport(const npf_cache_t *npc, u_int which, const in_port_t port)
    574 {
    575 	const int proto = npc->npc_proto;
    576 	in_port_t *oport;
    577 
    578 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    579 	KASSERT(proto == IPPROTO_TCP || proto == IPPROTO_UDP);
    580 	KASSERT(which == NPF_SRC || which == NPF_DST);
    581 
    582 	/* Get the offset and store the port in it. */
    583 	if (proto == IPPROTO_TCP) {
    584 		struct tcphdr *th = npc->npc_l4.tcp;
    585 		oport = (which == NPF_SRC) ? &th->th_sport : &th->th_dport;
    586 	} else {
    587 		struct udphdr *uh = npc->npc_l4.udp;
    588 		oport = (which == NPF_SRC) ? &uh->uh_sport : &uh->uh_dport;
    589 	}
    590 	memcpy(oport, &port, sizeof(in_port_t));
    591 	return true;
    592 }
    593 
    594 /*
    595  * npf_rwrcksum: rewrite IPv4 and/or TCP/UDP checksum.
    596  */
    597 bool
    598 npf_rwrcksum(const npf_cache_t *npc, u_int which,
    599     const npf_addr_t *addr, const in_port_t port)
    600 {
    601 	const npf_addr_t *oaddr = npc->npc_ips[which];
    602 	const int proto = npc->npc_proto;
    603 	const int alen = npc->npc_alen;
    604 	uint16_t *ocksum;
    605 	in_port_t oport;
    606 
    607 	KASSERT(npf_iscached(npc, NPC_LAYER4));
    608 	KASSERT(which == NPF_SRC || which == NPF_DST);
    609 
    610 	if (npf_iscached(npc, NPC_IP4)) {
    611 		struct ip *ip = npc->npc_ip.v4;
    612 		uint16_t ipsum = ip->ip_sum;
    613 
    614 		/* Recalculate IPv4 checksum and rewrite. */
    615 		ip->ip_sum = npf_addr_cksum(ipsum, alen, oaddr, addr);
    616 	} else {
    617 		/* No checksum for IPv6. */
    618 		KASSERT(npf_iscached(npc, NPC_IP6));
    619 	}
    620 
    621 	/* Nothing else to do for ICMP. */
    622 	if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
    623 		return true;
    624 	}
    625 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    626 
    627 	/*
    628 	 * Calculate TCP/UDP checksum:
    629 	 * - Skip if UDP and the current checksum is zero.
    630 	 * - Fixup the IP address change.
    631 	 * - Fixup the port change, if required (non-zero).
    632 	 */
    633 	if (proto == IPPROTO_TCP) {
    634 		struct tcphdr *th = npc->npc_l4.tcp;
    635 
    636 		ocksum = &th->th_sum;
    637 		oport = (which == NPF_SRC) ? th->th_sport : th->th_dport;
    638 	} else {
    639 		struct udphdr *uh = npc->npc_l4.udp;
    640 
    641 		KASSERT(proto == IPPROTO_UDP);
    642 		ocksum = &uh->uh_sum;
    643 		if (*ocksum == 0) {
    644 			/* No need to update. */
    645 			return true;
    646 		}
    647 		oport = (which == NPF_SRC) ? uh->uh_sport : uh->uh_dport;
    648 	}
    649 
    650 	uint16_t cksum = npf_addr_cksum(*ocksum, alen, oaddr, addr);
    651 	if (port) {
    652 		cksum = npf_fixup16_cksum(cksum, oport, port);
    653 	}
    654 
    655 	/* Rewrite TCP/UDP checksum. */
    656 	memcpy(ocksum, &cksum, sizeof(uint16_t));
    657 	return true;
    658 }
    659 
    660 /*
    661  * npf_napt_rwr: perform address and/or port translation.
    662  */
    663 int
    664 npf_napt_rwr(const npf_cache_t *npc, u_int which,
    665     const npf_addr_t *addr, const in_addr_t port)
    666 {
    667 	const unsigned proto = npc->npc_proto;
    668 
    669 	/*
    670 	 * Rewrite IP and/or TCP/UDP checksums first, since we need the
    671 	 * current (old) address/port for the calculations.  Then perform
    672 	 * the address translation i.e. rewrite source or destination.
    673 	 */
    674 	if (!npf_rwrcksum(npc, which, addr, port)) {
    675 		return EINVAL;
    676 	}
    677 	if (!npf_rwrip(npc, which, addr)) {
    678 		return EINVAL;
    679 	}
    680 	if (port == 0) {
    681 		/* Done. */
    682 		return 0;
    683 	}
    684 
    685 	switch (proto) {
    686 	case IPPROTO_TCP:
    687 	case IPPROTO_UDP:
    688 		/* Rewrite source/destination port. */
    689 		if (!npf_rwrport(npc, which, port)) {
    690 			return EINVAL;
    691 		}
    692 		break;
    693 	case IPPROTO_ICMP:
    694 	case IPPROTO_ICMPV6:
    695 		KASSERT(npf_iscached(npc, NPC_ICMP));
    696 		/* Nothing. */
    697 		break;
    698 	default:
    699 		return ENOTSUP;
    700 	}
    701 	return 0;
    702 }
    703 
    704 /*
    705  * IPv6-to-IPv6 Network Prefix Translation (NPTv6), as per RFC 6296.
    706  */
    707 
    708 int
    709 npf_npt66_rwr(const npf_cache_t *npc, u_int which, const npf_addr_t *pref,
    710     npf_netmask_t len, uint16_t adj)
    711 {
    712 	npf_addr_t *addr = npc->npc_ips[which];
    713 	unsigned remnant, word, preflen = len >> 4;
    714 	uint32_t sum;
    715 
    716 	KASSERT(which == NPF_SRC || which == NPF_DST);
    717 
    718 	if (!npf_iscached(npc, NPC_IP6)) {
    719 		return EINVAL;
    720 	}
    721 	if (len <= 48) {
    722 		/*
    723 		 * The word to adjust.  Cannot translate the 0xffff
    724 		 * subnet if /48 or shorter.
    725 		 */
    726 		word = 3;
    727 		if (addr->word16[word] == 0xffff) {
    728 			return EINVAL;
    729 		}
    730 	} else {
    731 		/*
    732 		 * Also, all 0s or 1s in the host part are disallowed for
    733 		 * longer than /48 prefixes.
    734 		 */
    735 		if ((addr->word32[2] == 0 && addr->word32[3] == 0) ||
    736 		    (addr->word32[2] == ~0U && addr->word32[3] == ~0U))
    737 			return EINVAL;
    738 
    739 		/* Determine the 16-bit word to adjust. */
    740 		for (word = 4; word < 8; word++)
    741 			if (addr->word16[word] != 0xffff)
    742 				break;
    743 	}
    744 
    745 	/* Rewrite the prefix. */
    746 	for (unsigned i = 0; i < preflen; i++) {
    747 		addr->word16[i] = pref->word16[i];
    748 	}
    749 
    750 	/*
    751 	 * If prefix length is within a 16-bit word (not dividable by 16),
    752 	 * then prepare a mask, determine the word and adjust it.
    753 	 */
    754 	if ((remnant = len - (preflen << 4)) != 0) {
    755 		const uint16_t wordmask = (1U << remnant) - 1;
    756 		const unsigned i = preflen;
    757 
    758 		addr->word16[i] = (pref->word16[i] & wordmask) |
    759 		    (addr->word16[i] & ~wordmask);
    760 	}
    761 
    762 	/*
    763 	 * Performing 1's complement sum/difference.
    764 	 */
    765 	sum = addr->word16[word] + adj;
    766 	while (sum >> 16) {
    767 		sum = (sum >> 16) + (sum & 0xffff);
    768 	}
    769 	if (sum == 0xffff) {
    770 		/* RFC 1071. */
    771 		sum = 0x0000;
    772 	}
    773 	addr->word16[word] = sum;
    774 	return 0;
    775 }
    776 
    777 #if defined(DDB) || defined(_NPF_TESTING)
    778 
    779 const char *
    780 npf_addr_dump(const npf_addr_t *addr, int alen)
    781 {
    782 	if (alen == sizeof(struct in_addr)) {
    783 		struct in_addr ip;
    784 		memcpy(&ip, addr, alen);
    785 		return inet_ntoa(ip);
    786 	}
    787 	return "[IPv6]";
    788 }
    789 
    790 #endif
    791