Home | History | Annotate | Line # | Download | only in npf
npf_inet.c revision 1.40 
      1 /*	$NetBSD: npf_inet.c,v 1.40 2018/03/13 09:04:02 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 1.40 2018/03/13 09:04:02 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) - thlen;
    219 	}
    220 	return 0;
    221 }
    222 
    223 /*
    224  * npf_fetch_tcpopts: parse and return TCP options.
    225  */
    226 bool
    227 npf_fetch_tcpopts(npf_cache_t *npc, uint16_t *mss, int *wscale)
    228 {
    229 	nbuf_t *nbuf = npc->npc_nbuf;
    230 	const struct tcphdr *th = npc->npc_l4.tcp;
    231 	int topts_len, step;
    232 	uint8_t *nptr;
    233 	uint8_t val;
    234 	bool ok;
    235 
    236 	KASSERT(npf_iscached(npc, NPC_IP46));
    237 	KASSERT(npf_iscached(npc, NPC_TCP));
    238 
    239 	/* Determine if there are any TCP options, get their length. */
    240 	topts_len = (th->th_off << 2) - sizeof(struct tcphdr);
    241 	if (topts_len <= 0) {
    242 		/* No options. */
    243 		return false;
    244 	}
    245 	KASSERT(topts_len <= MAX_TCPOPTLEN);
    246 
    247 	/* First step: IP and TCP header up to options. */
    248 	step = npc->npc_hlen + sizeof(struct tcphdr);
    249 	nbuf_reset(nbuf);
    250 next:
    251 	if ((nptr = nbuf_advance(nbuf, step, 1)) == NULL) {
    252 		ok = false;
    253 		goto done;
    254 	}
    255 	val = *nptr;
    256 
    257 	switch (val) {
    258 	case TCPOPT_EOL:
    259 		/* Done. */
    260 		ok = true;
    261 		goto done;
    262 	case TCPOPT_NOP:
    263 		topts_len--;
    264 		step = 1;
    265 		break;
    266 	case TCPOPT_MAXSEG:
    267 		if ((nptr = nbuf_ensure_contig(nbuf, TCPOLEN_MAXSEG)) == NULL) {
    268 			ok = false;
    269 			goto done;
    270 		}
    271 		if (mss) {
    272 			if (*mss) {
    273 				memcpy(nptr + 2, mss, sizeof(uint16_t));
    274 			} else {
    275 				memcpy(mss, nptr + 2, sizeof(uint16_t));
    276 			}
    277 		}
    278 		topts_len -= TCPOLEN_MAXSEG;
    279 		step = TCPOLEN_MAXSEG;
    280 		break;
    281 	case TCPOPT_WINDOW:
    282 		/* TCP Window Scaling (RFC 1323). */
    283 		if ((nptr = nbuf_ensure_contig(nbuf, TCPOLEN_WINDOW)) == NULL) {
    284 			ok = false;
    285 			goto done;
    286 		}
    287 		val = *(nptr + 2);
    288 		*wscale = (val > TCP_MAX_WINSHIFT) ? TCP_MAX_WINSHIFT : val;
    289 		topts_len -= TCPOLEN_WINDOW;
    290 		step = TCPOLEN_WINDOW;
    291 		break;
    292 	default:
    293 		if ((nptr = nbuf_ensure_contig(nbuf, 2)) == NULL) {
    294 			ok = false;
    295 			goto done;
    296 		}
    297 		val = *(nptr + 1);
    298 		if (val < 2 || val > topts_len) {
    299 			ok = false;
    300 			goto done;
    301 		}
    302 		topts_len -= val;
    303 		step = val;
    304 	}
    305 
    306 	/* Any options left? */
    307 	if (__predict_true(topts_len > 0)) {
    308 		goto next;
    309 	}
    310 	ok = true;
    311 done:
    312 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    313 		npf_recache(npc);
    314 	}
    315 	return ok;
    316 }
    317 
    318 static int
    319 npf_cache_ip(npf_cache_t *npc, nbuf_t *nbuf)
    320 {
    321 	const void *nptr = nbuf_dataptr(nbuf);
    322 	const uint8_t ver = *(const uint8_t *)nptr;
    323 	int flags = 0;
    324 
    325 	switch (ver >> 4) {
    326 	case IPVERSION: {
    327 		struct ip *ip;
    328 
    329 		ip = nbuf_ensure_contig(nbuf, sizeof(struct ip));
    330 		if (ip == NULL) {
    331 			return NPC_FMTERR;
    332 		}
    333 
    334 		/* Check header length and fragment offset. */
    335 		if ((u_int)(ip->ip_hl << 2) < sizeof(struct ip)) {
    336 			return NPC_FMTERR;
    337 		}
    338 		if (ip->ip_off & ~htons(IP_DF | IP_RF)) {
    339 			/* Note fragmentation. */
    340 			flags |= NPC_IPFRAG;
    341 		}
    342 
    343 		/* Cache: layer 3 - IPv4. */
    344 		npc->npc_alen = sizeof(struct in_addr);
    345 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip->ip_src;
    346 		npc->npc_ips[NPF_DST] = (npf_addr_t *)&ip->ip_dst;
    347 		npc->npc_hlen = ip->ip_hl << 2;
    348 		npc->npc_proto = ip->ip_p;
    349 
    350 		npc->npc_ip.v4 = ip;
    351 		flags |= NPC_IP4;
    352 		break;
    353 	}
    354 
    355 	case (IPV6_VERSION >> 4): {
    356 		struct ip6_hdr *ip6;
    357 		struct ip6_ext *ip6e;
    358 		struct ip6_frag *ip6f;
    359 		size_t off, hlen;
    360 		int frag_present;
    361 
    362 		ip6 = nbuf_ensure_contig(nbuf, sizeof(struct ip6_hdr));
    363 		if (ip6 == NULL) {
    364 			return NPC_FMTERR;
    365 		}
    366 
    367 		/* Set initial next-protocol value. */
    368 		hlen = sizeof(struct ip6_hdr);
    369 		npc->npc_proto = ip6->ip6_nxt;
    370 		npc->npc_hlen = hlen;
    371 
    372 		frag_present = 0;
    373 
    374 		/*
    375 		 * Advance by the length of the current header.
    376 		 */
    377 		off = nbuf_offset(nbuf);
    378 		while ((ip6e = nbuf_advance(nbuf, hlen, sizeof(*ip6e))) != NULL) {
    379 			/*
    380 			 * Determine whether we are going to continue.
    381 			 */
    382 			switch (npc->npc_proto) {
    383 			case IPPROTO_HOPOPTS:
    384 			case IPPROTO_DSTOPTS:
    385 			case IPPROTO_ROUTING:
    386 				hlen = (ip6e->ip6e_len + 1) << 3;
    387 				break;
    388 			case IPPROTO_FRAGMENT:
    389 				if (frag_present++)
    390 					return NPC_FMTERR;
    391 				ip6f = nbuf_ensure_contig(nbuf, sizeof(*ip6f));
    392 				if (ip6f == NULL)
    393 					return NPC_FMTERR;
    394 
    395 				hlen = 0;
    396 				flags |= NPC_IPFRAG;
    397 
    398 				break;
    399 			case IPPROTO_AH:
    400 				hlen = (ip6e->ip6e_len + 2) << 2;
    401 				break;
    402 			default:
    403 				hlen = 0;
    404 				break;
    405 			}
    406 
    407 			if (!hlen) {
    408 				break;
    409 			}
    410 			npc->npc_proto = ip6e->ip6e_nxt;
    411 			npc->npc_hlen += hlen;
    412 		}
    413 
    414 		/*
    415 		 * Re-fetch the header pointers (nbufs might have been
    416 		 * reallocated).  Restore the original offset (if any).
    417 		 */
    418 		nbuf_reset(nbuf);
    419 		ip6 = nbuf_dataptr(nbuf);
    420 		if (off) {
    421 			nbuf_advance(nbuf, off, 0);
    422 		}
    423 
    424 		/* Cache: layer 3 - IPv6. */
    425 		npc->npc_alen = sizeof(struct in6_addr);
    426 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip6->ip6_src;
    427 		npc->npc_ips[NPF_DST]= (npf_addr_t *)&ip6->ip6_dst;
    428 
    429 		npc->npc_ip.v6 = ip6;
    430 		flags |= NPC_IP6;
    431 		break;
    432 	}
    433 	default:
    434 		break;
    435 	}
    436 	return flags;
    437 }
    438 
    439 /*
    440  * npf_cache_all: general routine to cache all relevant IP (v4 or v6)
    441  * and TCP, UDP or ICMP headers.
    442  *
    443  * => nbuf offset shall be set accordingly.
    444  */
    445 int
    446 npf_cache_all(npf_cache_t *npc)
    447 {
    448 	nbuf_t *nbuf = npc->npc_nbuf;
    449 	int flags, l4flags;
    450 	u_int hlen;
    451 
    452 	/*
    453 	 * This routine is a main point where the references are cached,
    454 	 * therefore clear the flag as we reset.
    455 	 */
    456 again:
    457 	nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    458 
    459 	/*
    460 	 * First, cache the L3 header (IPv4 or IPv6).  If IP packet is
    461 	 * fragmented, then we cannot look into L4.
    462 	 */
    463 	flags = npf_cache_ip(npc, nbuf);
    464 	if ((flags & NPC_IP46) == 0 || (flags & NPC_IPFRAG) != 0 ||
    465 	    (flags & NPC_FMTERR) != 0) {
    466 		nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    467 		npc->npc_info |= flags;
    468 		return flags;
    469 	}
    470 	hlen = npc->npc_hlen;
    471 
    472 	switch (npc->npc_proto) {
    473 	case IPPROTO_TCP:
    474 		/* Cache: layer 4 - TCP. */
    475 		npc->npc_l4.tcp = nbuf_advance(nbuf, hlen,
    476 		    sizeof(struct tcphdr));
    477 		l4flags = NPC_LAYER4 | NPC_TCP;
    478 		break;
    479 	case IPPROTO_UDP:
    480 		/* Cache: layer 4 - UDP. */
    481 		npc->npc_l4.udp = nbuf_advance(nbuf, hlen,
    482 		    sizeof(struct udphdr));
    483 		l4flags = NPC_LAYER4 | NPC_UDP;
    484 		break;
    485 	case IPPROTO_ICMP:
    486 		/* Cache: layer 4 - ICMPv4. */
    487 		npc->npc_l4.icmp = nbuf_advance(nbuf, hlen,
    488 		    offsetof(struct icmp, icmp_void));
    489 		l4flags = NPC_LAYER4 | NPC_ICMP;
    490 		break;
    491 	case IPPROTO_ICMPV6:
    492 		/* Cache: layer 4 - ICMPv6. */
    493 		npc->npc_l4.icmp6 = nbuf_advance(nbuf, hlen,
    494 		    offsetof(struct icmp6_hdr, icmp6_data32));
    495 		l4flags = NPC_LAYER4 | NPC_ICMP;
    496 		break;
    497 	default:
    498 		l4flags = 0;
    499 		break;
    500 	}
    501 
    502 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    503 		goto again;
    504 	}
    505 
    506 	/* Add the L4 flags if nbuf_advance() succeeded. */
    507 	if (l4flags && npc->npc_l4.hdr) {
    508 		flags |= l4flags;
    509 	}
    510 	npc->npc_info |= flags;
    511 	return flags;
    512 }
    513 
    514 void
    515 npf_recache(npf_cache_t *npc)
    516 {
    517 	nbuf_t *nbuf = npc->npc_nbuf;
    518 	const int mflags __diagused = npc->npc_info & (NPC_IP46 | NPC_LAYER4);
    519 	int flags __diagused;
    520 
    521 	nbuf_reset(nbuf);
    522 	npc->npc_info = 0;
    523 	flags = npf_cache_all(npc);
    524 
    525 	KASSERT((flags & mflags) == mflags);
    526 	KASSERT(nbuf_flag_p(nbuf, NBUF_DATAREF_RESET) == 0);
    527 }
    528 
    529 /*
    530  * npf_rwrip: rewrite required IP address.
    531  */
    532 bool
    533 npf_rwrip(const npf_cache_t *npc, u_int which, const npf_addr_t *addr)
    534 {
    535 	KASSERT(npf_iscached(npc, NPC_IP46));
    536 	KASSERT(which == NPF_SRC || which == NPF_DST);
    537 
    538 	memcpy(npc->npc_ips[which], addr, npc->npc_alen);
    539 	return true;
    540 }
    541 
    542 /*
    543  * npf_rwrport: rewrite required TCP/UDP port.
    544  */
    545 bool
    546 npf_rwrport(const npf_cache_t *npc, u_int which, const in_port_t port)
    547 {
    548 	const int proto = npc->npc_proto;
    549 	in_port_t *oport;
    550 
    551 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    552 	KASSERT(proto == IPPROTO_TCP || proto == IPPROTO_UDP);
    553 	KASSERT(which == NPF_SRC || which == NPF_DST);
    554 
    555 	/* Get the offset and store the port in it. */
    556 	if (proto == IPPROTO_TCP) {
    557 		struct tcphdr *th = npc->npc_l4.tcp;
    558 		oport = (which == NPF_SRC) ? &th->th_sport : &th->th_dport;
    559 	} else {
    560 		struct udphdr *uh = npc->npc_l4.udp;
    561 		oport = (which == NPF_SRC) ? &uh->uh_sport : &uh->uh_dport;
    562 	}
    563 	memcpy(oport, &port, sizeof(in_port_t));
    564 	return true;
    565 }
    566 
    567 /*
    568  * npf_rwrcksum: rewrite IPv4 and/or TCP/UDP checksum.
    569  */
    570 bool
    571 npf_rwrcksum(const npf_cache_t *npc, u_int which,
    572     const npf_addr_t *addr, const in_port_t port)
    573 {
    574 	const npf_addr_t *oaddr = npc->npc_ips[which];
    575 	const int proto = npc->npc_proto;
    576 	const int alen = npc->npc_alen;
    577 	uint16_t *ocksum;
    578 	in_port_t oport;
    579 
    580 	KASSERT(npf_iscached(npc, NPC_LAYER4));
    581 	KASSERT(which == NPF_SRC || which == NPF_DST);
    582 
    583 	if (npf_iscached(npc, NPC_IP4)) {
    584 		struct ip *ip = npc->npc_ip.v4;
    585 		uint16_t ipsum = ip->ip_sum;
    586 
    587 		/* Recalculate IPv4 checksum and rewrite. */
    588 		ip->ip_sum = npf_addr_cksum(ipsum, alen, oaddr, addr);
    589 	} else {
    590 		/* No checksum for IPv6. */
    591 		KASSERT(npf_iscached(npc, NPC_IP6));
    592 	}
    593 
    594 	/* Nothing else to do for ICMP. */
    595 	if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
    596 		return true;
    597 	}
    598 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    599 
    600 	/*
    601 	 * Calculate TCP/UDP checksum:
    602 	 * - Skip if UDP and the current checksum is zero.
    603 	 * - Fixup the IP address change.
    604 	 * - Fixup the port change, if required (non-zero).
    605 	 */
    606 	if (proto == IPPROTO_TCP) {
    607 		struct tcphdr *th = npc->npc_l4.tcp;
    608 
    609 		ocksum = &th->th_sum;
    610 		oport = (which == NPF_SRC) ? th->th_sport : th->th_dport;
    611 	} else {
    612 		struct udphdr *uh = npc->npc_l4.udp;
    613 
    614 		KASSERT(proto == IPPROTO_UDP);
    615 		ocksum = &uh->uh_sum;
    616 		if (*ocksum == 0) {
    617 			/* No need to update. */
    618 			return true;
    619 		}
    620 		oport = (which == NPF_SRC) ? uh->uh_sport : uh->uh_dport;
    621 	}
    622 
    623 	uint16_t cksum = npf_addr_cksum(*ocksum, alen, oaddr, addr);
    624 	if (port) {
    625 		cksum = npf_fixup16_cksum(cksum, oport, port);
    626 	}
    627 
    628 	/* Rewrite TCP/UDP checksum. */
    629 	memcpy(ocksum, &cksum, sizeof(uint16_t));
    630 	return true;
    631 }
    632 
    633 /*
    634  * npf_napt_rwr: perform address and/or port translation.
    635  */
    636 int
    637 npf_napt_rwr(const npf_cache_t *npc, u_int which,
    638     const npf_addr_t *addr, const in_addr_t port)
    639 {
    640 	const unsigned proto = npc->npc_proto;
    641 
    642 	/*
    643 	 * Rewrite IP and/or TCP/UDP checksums first, since we need the
    644 	 * current (old) address/port for the calculations.  Then perform
    645 	 * the address translation i.e. rewrite source or destination.
    646 	 */
    647 	if (!npf_rwrcksum(npc, which, addr, port)) {
    648 		return EINVAL;
    649 	}
    650 	if (!npf_rwrip(npc, which, addr)) {
    651 		return EINVAL;
    652 	}
    653 	if (port == 0) {
    654 		/* Done. */
    655 		return 0;
    656 	}
    657 
    658 	switch (proto) {
    659 	case IPPROTO_TCP:
    660 	case IPPROTO_UDP:
    661 		/* Rewrite source/destination port. */
    662 		if (!npf_rwrport(npc, which, port)) {
    663 			return EINVAL;
    664 		}
    665 		break;
    666 	case IPPROTO_ICMP:
    667 	case IPPROTO_ICMPV6:
    668 		KASSERT(npf_iscached(npc, NPC_ICMP));
    669 		/* Nothing. */
    670 		break;
    671 	default:
    672 		return ENOTSUP;
    673 	}
    674 	return 0;
    675 }
    676 
    677 /*
    678  * IPv6-to-IPv6 Network Prefix Translation (NPTv6), as per RFC 6296.
    679  */
    680 
    681 int
    682 npf_npt66_rwr(const npf_cache_t *npc, u_int which, const npf_addr_t *pref,
    683     npf_netmask_t len, uint16_t adj)
    684 {
    685 	npf_addr_t *addr = npc->npc_ips[which];
    686 	unsigned remnant, word, preflen = len >> 4;
    687 	uint32_t sum;
    688 
    689 	KASSERT(which == NPF_SRC || which == NPF_DST);
    690 
    691 	if (!npf_iscached(npc, NPC_IP6)) {
    692 		return EINVAL;
    693 	}
    694 	if (len <= 48) {
    695 		/*
    696 		 * The word to adjust.  Cannot translate the 0xffff
    697 		 * subnet if /48 or shorter.
    698 		 */
    699 		word = 3;
    700 		if (addr->word16[word] == 0xffff) {
    701 			return EINVAL;
    702 		}
    703 	} else {
    704 		/*
    705 		 * Also, all 0s or 1s in the host part are disallowed for
    706 		 * longer than /48 prefixes.
    707 		 */
    708 		if ((addr->word32[2] == 0 && addr->word32[3] == 0) ||
    709 		    (addr->word32[2] == ~0U && addr->word32[3] == ~0U))
    710 			return EINVAL;
    711 
    712 		/* Determine the 16-bit word to adjust. */
    713 		for (word = 4; word < 8; word++)
    714 			if (addr->word16[word] != 0xffff)
    715 				break;
    716 	}
    717 
    718 	/* Rewrite the prefix. */
    719 	for (unsigned i = 0; i < preflen; i++) {
    720 		addr->word16[i] = pref->word16[i];
    721 	}
    722 
    723 	/*
    724 	 * If prefix length is within a 16-bit word (not dividable by 16),
    725 	 * then prepare a mask, determine the word and adjust it.
    726 	 */
    727 	if ((remnant = len - (preflen << 4)) != 0) {
    728 		const uint16_t wordmask = (1U << remnant) - 1;
    729 		const unsigned i = preflen;
    730 
    731 		addr->word16[i] = (pref->word16[i] & wordmask) |
    732 		    (addr->word16[i] & ~wordmask);
    733 	}
    734 
    735 	/*
    736 	 * Performing 1's complement sum/difference.
    737 	 */
    738 	sum = addr->word16[word] + adj;
    739 	while (sum >> 16) {
    740 		sum = (sum >> 16) + (sum & 0xffff);
    741 	}
    742 	if (sum == 0xffff) {
    743 		/* RFC 1071. */
    744 		sum = 0x0000;
    745 	}
    746 	addr->word16[word] = sum;
    747 	return 0;
    748 }
    749 
    750 #if defined(DDB) || defined(_NPF_TESTING)
    751 
    752 const char *
    753 npf_addr_dump(const npf_addr_t *addr, int alen)
    754 {
    755 	if (alen == sizeof(struct in_addr)) {
    756 		struct in_addr ip;
    757 		memcpy(&ip, addr, alen);
    758 		return inet_ntoa(ip);
    759 	}
    760 	return "[IPv6]";
    761 }
    762 
    763 #endif
    764