Home | History | Annotate | Line # | Download | only in npf
npf_inet.c revision 1.32
      1  1.32   rmind /*	$NetBSD: npf_inet.c,v 1.32 2014/07/20 00:37:41 rmind Exp $	*/
      2   1.1   rmind 
      3   1.1   rmind /*-
      4  1.29   rmind  * Copyright (c) 2009-2014 The NetBSD Foundation, Inc.
      5   1.1   rmind  * All rights reserved.
      6   1.1   rmind  *
      7   1.1   rmind  * This material is based upon work partially supported by The
      8   1.1   rmind  * NetBSD Foundation under a contract with Mindaugas Rasiukevicius.
      9   1.1   rmind  *
     10   1.1   rmind  * Redistribution and use in source and binary forms, with or without
     11   1.1   rmind  * modification, are permitted provided that the following conditions
     12   1.1   rmind  * are met:
     13   1.1   rmind  * 1. Redistributions of source code must retain the above copyright
     14   1.1   rmind  *    notice, this list of conditions and the following disclaimer.
     15   1.1   rmind  * 2. Redistributions in binary form must reproduce the above copyright
     16   1.1   rmind  *    notice, this list of conditions and the following disclaimer in the
     17   1.1   rmind  *    documentation and/or other materials provided with the distribution.
     18   1.1   rmind  *
     19   1.1   rmind  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20   1.1   rmind  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21   1.1   rmind  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22   1.1   rmind  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23   1.1   rmind  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24   1.1   rmind  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25   1.1   rmind  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26   1.1   rmind  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27   1.1   rmind  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28   1.1   rmind  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29   1.1   rmind  * POSSIBILITY OF SUCH DAMAGE.
     30   1.1   rmind  */
     31   1.1   rmind 
     32   1.1   rmind /*
     33  1.22   rmind  * Various protocol related helper routines.
     34  1.12   rmind  *
     35  1.12   rmind  * This layer manipulates npf_cache_t structure i.e. caches requested headers
     36  1.12   rmind  * and stores which information was cached in the information bit field.
     37  1.12   rmind  * It is also responsibility of this layer to update or invalidate the cache
     38  1.12   rmind  * on rewrites (e.g. by translation routines).
     39   1.1   rmind  */
     40   1.1   rmind 
     41   1.1   rmind #include <sys/cdefs.h>
     42  1.32   rmind __KERNEL_RCSID(0, "$NetBSD: npf_inet.c,v 1.32 2014/07/20 00:37:41 rmind Exp $");
     43   1.1   rmind 
     44   1.1   rmind #include <sys/param.h>
     45  1.11   rmind #include <sys/types.h>
     46   1.1   rmind 
     47   1.4   rmind #include <net/pfil.h>
     48   1.4   rmind #include <net/if.h>
     49   1.4   rmind #include <net/ethertypes.h>
     50   1.4   rmind #include <net/if_ether.h>
     51   1.4   rmind 
     52   1.1   rmind #include <netinet/in_systm.h>
     53   1.1   rmind #include <netinet/in.h>
     54   1.1   rmind #include <netinet/ip.h>
     55   1.4   rmind #include <netinet/ip6.h>
     56   1.1   rmind #include <netinet/tcp.h>
     57   1.1   rmind #include <netinet/udp.h>
     58   1.1   rmind #include <netinet/ip_icmp.h>
     59   1.1   rmind 
     60   1.1   rmind #include "npf_impl.h"
     61   1.1   rmind 
     62   1.1   rmind /*
     63  1.27   rmind  * npf_fixup{16,32}_cksum: incremental update of the Internet checksum.
     64   1.1   rmind  */
     65   1.1   rmind 
     66   1.1   rmind uint16_t
     67   1.1   rmind npf_fixup16_cksum(uint16_t cksum, uint16_t odatum, uint16_t ndatum)
     68   1.1   rmind {
     69   1.1   rmind 	uint32_t sum;
     70   1.1   rmind 
     71   1.1   rmind 	/*
     72   1.1   rmind 	 * RFC 1624:
     73   1.1   rmind 	 *	HC' = ~(~HC + ~m + m')
     74  1.27   rmind 	 *
     75  1.27   rmind 	 * Note: 1's complement sum is endian-independent (RFC 1071, page 2).
     76   1.1   rmind 	 */
     77  1.27   rmind 	sum = ~cksum & 0xffff;
     78  1.27   rmind 	sum += (~odatum & 0xffff) + ndatum;
     79   1.1   rmind 	sum = (sum >> 16) + (sum & 0xffff);
     80   1.1   rmind 	sum += (sum >> 16);
     81   1.1   rmind 
     82  1.27   rmind 	return ~sum & 0xffff;
     83   1.1   rmind }
     84   1.1   rmind 
     85   1.1   rmind uint16_t
     86   1.1   rmind npf_fixup32_cksum(uint16_t cksum, uint32_t odatum, uint32_t ndatum)
     87   1.1   rmind {
     88  1.27   rmind 	uint32_t sum;
     89  1.27   rmind 
     90  1.27   rmind 	/*
     91  1.27   rmind 	 * Checksum 32-bit datum as as two 16-bit.  Note, the first
     92  1.27   rmind 	 * 32->16 bit reduction is not necessary.
     93  1.27   rmind 	 */
     94  1.27   rmind 	sum = ~cksum & 0xffff;
     95  1.27   rmind 	sum += (~odatum & 0xffff) + (ndatum & 0xffff);
     96   1.1   rmind 
     97  1.27   rmind 	sum += (~odatum >> 16) + (ndatum >> 16);
     98  1.27   rmind 	sum = (sum >> 16) + (sum & 0xffff);
     99  1.27   rmind 	sum += (sum >> 16);
    100  1.27   rmind 	return ~sum & 0xffff;
    101   1.1   rmind }
    102   1.1   rmind 
    103   1.1   rmind /*
    104   1.4   rmind  * npf_addr_cksum: calculate checksum of the address, either IPv4 or IPv6.
    105   1.4   rmind  */
    106   1.4   rmind uint16_t
    107  1.19   rmind npf_addr_cksum(uint16_t cksum, int sz, const npf_addr_t *oaddr,
    108  1.19   rmind     const npf_addr_t *naddr)
    109   1.4   rmind {
    110  1.19   rmind 	const uint32_t *oip32 = (const uint32_t *)oaddr;
    111  1.19   rmind 	const uint32_t *nip32 = (const uint32_t *)naddr;
    112   1.4   rmind 
    113   1.4   rmind 	KASSERT(sz % sizeof(uint32_t) == 0);
    114   1.4   rmind 	do {
    115   1.4   rmind 		cksum = npf_fixup32_cksum(cksum, *oip32++, *nip32++);
    116   1.4   rmind 		sz -= sizeof(uint32_t);
    117   1.4   rmind 	} while (sz);
    118   1.4   rmind 
    119   1.4   rmind 	return cksum;
    120   1.4   rmind }
    121   1.4   rmind 
    122   1.4   rmind /*
    123  1.26   rmind  * npf_addr_sum: provide IP addresses as a XORed 32-bit integer.
    124   1.4   rmind  * Note: used for hash function.
    125   1.1   rmind  */
    126   1.4   rmind uint32_t
    127  1.26   rmind npf_addr_mix(const int sz, const npf_addr_t *a1, const npf_addr_t *a2)
    128   1.1   rmind {
    129   1.4   rmind 	uint32_t mix = 0;
    130   1.1   rmind 
    131   1.5   rmind 	KASSERT(sz > 0 && a1 != NULL && a2 != NULL);
    132   1.5   rmind 
    133  1.26   rmind 	for (int i = 0; i < (sz >> 2); i++) {
    134  1.26   rmind 		mix ^= a1->s6_addr32[i];
    135  1.26   rmind 		mix ^= a2->s6_addr32[i];
    136   1.4   rmind 	}
    137   1.4   rmind 	return mix;
    138   1.4   rmind }
    139   1.1   rmind 
    140  1.13   rmind /*
    141  1.13   rmind  * npf_addr_mask: apply the mask to a given address and store the result.
    142  1.13   rmind  */
    143  1.13   rmind void
    144  1.13   rmind npf_addr_mask(const npf_addr_t *addr, const npf_netmask_t mask,
    145  1.13   rmind     const int alen, npf_addr_t *out)
    146  1.12   rmind {
    147  1.13   rmind 	const int nwords = alen >> 2;
    148  1.12   rmind 	uint_fast8_t length = mask;
    149  1.12   rmind 
    150  1.12   rmind 	/* Note: maximum length is 32 for IPv4 and 128 for IPv6. */
    151  1.12   rmind 	KASSERT(length <= NPF_MAX_NETMASK);
    152  1.12   rmind 
    153  1.13   rmind 	for (int i = 0; i < nwords; i++) {
    154  1.13   rmind 		uint32_t wordmask;
    155  1.13   rmind 
    156  1.12   rmind 		if (length >= 32) {
    157  1.13   rmind 			wordmask = htonl(0xffffffff);
    158  1.12   rmind 			length -= 32;
    159  1.13   rmind 		} else if (length) {
    160  1.13   rmind 			wordmask = htonl(0xffffffff << (32 - length));
    161  1.13   rmind 			length = 0;
    162  1.12   rmind 		} else {
    163  1.13   rmind 			wordmask = 0;
    164  1.12   rmind 		}
    165  1.13   rmind 		out->s6_addr32[i] = addr->s6_addr32[i] & wordmask;
    166  1.12   rmind 	}
    167  1.12   rmind }
    168  1.12   rmind 
    169  1.12   rmind /*
    170  1.12   rmind  * npf_addr_cmp: compare two addresses, either IPv4 or IPv6.
    171  1.12   rmind  *
    172  1.13   rmind  * => Return 0 if equal and negative/positive if less/greater accordingly.
    173  1.12   rmind  * => Ignore the mask, if NPF_NO_NETMASK is specified.
    174  1.12   rmind  */
    175  1.12   rmind int
    176  1.12   rmind npf_addr_cmp(const npf_addr_t *addr1, const npf_netmask_t mask1,
    177  1.13   rmind     const npf_addr_t *addr2, const npf_netmask_t mask2, const int alen)
    178  1.12   rmind {
    179  1.13   rmind 	npf_addr_t realaddr1, realaddr2;
    180  1.12   rmind 
    181  1.12   rmind 	if (mask1 != NPF_NO_NETMASK) {
    182  1.13   rmind 		npf_addr_mask(addr1, mask1, alen, &realaddr1);
    183  1.13   rmind 		addr1 = &realaddr1;
    184  1.12   rmind 	}
    185  1.12   rmind 	if (mask2 != NPF_NO_NETMASK) {
    186  1.13   rmind 		npf_addr_mask(addr2, mask2, alen, &realaddr2);
    187  1.13   rmind 		addr2 = &realaddr2;
    188  1.12   rmind 	}
    189  1.13   rmind 	return memcmp(addr1, addr2, alen);
    190  1.12   rmind }
    191  1.12   rmind 
    192   1.4   rmind /*
    193   1.4   rmind  * npf_tcpsaw: helper to fetch SEQ, ACK, WIN and return TCP data length.
    194  1.12   rmind  *
    195  1.12   rmind  * => Returns all values in host byte-order.
    196   1.4   rmind  */
    197   1.4   rmind int
    198  1.12   rmind npf_tcpsaw(const npf_cache_t *npc, tcp_seq *seq, tcp_seq *ack, uint32_t *win)
    199   1.4   rmind {
    200  1.19   rmind 	const struct tcphdr *th = npc->npc_l4.tcp;
    201   1.8   rmind 	u_int thlen;
    202   1.1   rmind 
    203   1.7  zoltan 	KASSERT(npf_iscached(npc, NPC_TCP));
    204   1.1   rmind 
    205   1.4   rmind 	*seq = ntohl(th->th_seq);
    206   1.4   rmind 	*ack = ntohl(th->th_ack);
    207   1.4   rmind 	*win = (uint32_t)ntohs(th->th_win);
    208   1.8   rmind 	thlen = th->th_off << 2;
    209   1.1   rmind 
    210   1.7  zoltan 	if (npf_iscached(npc, NPC_IP4)) {
    211  1.19   rmind 		const struct ip *ip = npc->npc_ip.v4;
    212  1.21   rmind 		return ntohs(ip->ip_len) - npc->npc_hlen - thlen;
    213  1.12   rmind 	} else if (npf_iscached(npc, NPC_IP6)) {
    214  1.19   rmind 		const struct ip6_hdr *ip6 = npc->npc_ip.v6;
    215   1.8   rmind 		return ntohs(ip6->ip6_plen) - thlen;
    216   1.7  zoltan 	}
    217   1.7  zoltan 	return 0;
    218   1.1   rmind }
    219   1.1   rmind 
    220   1.1   rmind /*
    221   1.4   rmind  * npf_fetch_tcpopts: parse and return TCP options.
    222   1.1   rmind  */
    223   1.1   rmind bool
    224  1.32   rmind npf_fetch_tcpopts(npf_cache_t *npc, uint16_t *mss, int *wscale)
    225   1.1   rmind {
    226  1.32   rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    227  1.19   rmind 	const struct tcphdr *th = npc->npc_l4.tcp;
    228   1.4   rmind 	int topts_len, step;
    229  1.19   rmind 	void *nptr;
    230   1.4   rmind 	uint8_t val;
    231  1.19   rmind 	bool ok;
    232   1.4   rmind 
    233   1.7  zoltan 	KASSERT(npf_iscached(npc, NPC_IP46));
    234   1.7  zoltan 	KASSERT(npf_iscached(npc, NPC_TCP));
    235  1.10   rmind 
    236   1.4   rmind 	/* Determine if there are any TCP options, get their length. */
    237   1.4   rmind 	topts_len = (th->th_off << 2) - sizeof(struct tcphdr);
    238   1.4   rmind 	if (topts_len <= 0) {
    239   1.4   rmind 		/* No options. */
    240   1.1   rmind 		return false;
    241   1.4   rmind 	}
    242   1.4   rmind 	KASSERT(topts_len <= MAX_TCPOPTLEN);
    243   1.1   rmind 
    244   1.4   rmind 	/* First step: IP and TCP header up to options. */
    245  1.21   rmind 	step = npc->npc_hlen + sizeof(struct tcphdr);
    246  1.19   rmind 	nbuf_reset(nbuf);
    247   1.4   rmind next:
    248  1.19   rmind 	if ((nptr = nbuf_advance(nbuf, step, 1)) == NULL) {
    249  1.19   rmind 		ok = false;
    250  1.19   rmind 		goto done;
    251   1.4   rmind 	}
    252  1.19   rmind 	val = *(uint8_t *)nptr;
    253  1.12   rmind 
    254   1.4   rmind 	switch (val) {
    255   1.4   rmind 	case TCPOPT_EOL:
    256   1.4   rmind 		/* Done. */
    257  1.19   rmind 		ok = true;
    258  1.19   rmind 		goto done;
    259   1.4   rmind 	case TCPOPT_NOP:
    260   1.4   rmind 		topts_len--;
    261   1.4   rmind 		step = 1;
    262   1.4   rmind 		break;
    263   1.4   rmind 	case TCPOPT_MAXSEG:
    264  1.19   rmind 		if ((nptr = nbuf_advance(nbuf, 2, 2)) == NULL) {
    265  1.19   rmind 			ok = false;
    266  1.19   rmind 			goto done;
    267   1.4   rmind 		}
    268   1.4   rmind 		if (mss) {
    269  1.19   rmind 			if (*mss) {
    270  1.19   rmind 				memcpy(nptr, mss, sizeof(uint16_t));
    271  1.19   rmind 			} else {
    272  1.19   rmind 				memcpy(mss, nptr, sizeof(uint16_t));
    273  1.19   rmind 			}
    274   1.4   rmind 		}
    275   1.4   rmind 		topts_len -= TCPOLEN_MAXSEG;
    276  1.19   rmind 		step = 2;
    277   1.4   rmind 		break;
    278   1.4   rmind 	case TCPOPT_WINDOW:
    279  1.10   rmind 		/* TCP Window Scaling (RFC 1323). */
    280  1.19   rmind 		if ((nptr = nbuf_advance(nbuf, 2, 1)) == NULL) {
    281  1.19   rmind 			ok = false;
    282  1.19   rmind 			goto done;
    283   1.4   rmind 		}
    284  1.19   rmind 		val = *(uint8_t *)nptr;
    285   1.4   rmind 		*wscale = (val > TCP_MAX_WINSHIFT) ? TCP_MAX_WINSHIFT : val;
    286   1.4   rmind 		topts_len -= TCPOLEN_WINDOW;
    287  1.19   rmind 		step = 1;
    288   1.4   rmind 		break;
    289   1.4   rmind 	default:
    290  1.19   rmind 		if ((nptr = nbuf_advance(nbuf, 1, 1)) == NULL) {
    291  1.19   rmind 			ok = false;
    292  1.19   rmind 			goto done;
    293   1.4   rmind 		}
    294  1.19   rmind 		val = *(uint8_t *)nptr;
    295  1.16   rmind 		if (val < 2 || val > topts_len) {
    296  1.19   rmind 			ok = false;
    297  1.19   rmind 			goto done;
    298   1.4   rmind 		}
    299   1.4   rmind 		topts_len -= val;
    300   1.4   rmind 		step = val - 1;
    301   1.4   rmind 	}
    302  1.12   rmind 
    303   1.6   rmind 	/* Any options left? */
    304   1.4   rmind 	if (__predict_true(topts_len > 0)) {
    305   1.4   rmind 		goto next;
    306   1.4   rmind 	}
    307  1.19   rmind 	ok = true;
    308  1.19   rmind done:
    309  1.19   rmind 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    310  1.32   rmind 		npf_recache(npc);
    311  1.19   rmind 	}
    312  1.19   rmind 	return ok;
    313   1.1   rmind }
    314   1.1   rmind 
    315  1.19   rmind static int
    316  1.19   rmind npf_cache_ip(npf_cache_t *npc, nbuf_t *nbuf)
    317   1.1   rmind {
    318  1.19   rmind 	const void *nptr = nbuf_dataptr(nbuf);
    319  1.19   rmind 	const uint8_t ver = *(const uint8_t *)nptr;
    320  1.19   rmind 	int flags = 0;
    321  1.12   rmind 
    322   1.4   rmind 	switch (ver >> 4) {
    323  1.12   rmind 	case IPVERSION: {
    324  1.19   rmind 		struct ip *ip;
    325  1.12   rmind 
    326  1.19   rmind 		ip = nbuf_ensure_contig(nbuf, sizeof(struct ip));
    327  1.19   rmind 		if (ip == NULL) {
    328  1.19   rmind 			return 0;
    329   1.4   rmind 		}
    330  1.12   rmind 
    331   1.4   rmind 		/* Check header length and fragment offset. */
    332  1.10   rmind 		if ((u_int)(ip->ip_hl << 2) < sizeof(struct ip)) {
    333  1.19   rmind 			return 0;
    334   1.4   rmind 		}
    335   1.4   rmind 		if (ip->ip_off & ~htons(IP_DF | IP_RF)) {
    336   1.4   rmind 			/* Note fragmentation. */
    337  1.19   rmind 			flags |= NPC_IPFRAG;
    338   1.4   rmind 		}
    339  1.12   rmind 
    340   1.4   rmind 		/* Cache: layer 3 - IPv4. */
    341  1.14   rmind 		npc->npc_alen = sizeof(struct in_addr);
    342  1.28   rmind 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip->ip_src;
    343  1.28   rmind 		npc->npc_ips[NPF_DST] = (npf_addr_t *)&ip->ip_dst;
    344   1.7  zoltan 		npc->npc_hlen = ip->ip_hl << 2;
    345  1.19   rmind 		npc->npc_proto = ip->ip_p;
    346  1.19   rmind 
    347  1.19   rmind 		npc->npc_ip.v4 = ip;
    348  1.19   rmind 		flags |= NPC_IP4;
    349   1.4   rmind 		break;
    350  1.12   rmind 	}
    351   1.4   rmind 
    352  1.12   rmind 	case (IPV6_VERSION >> 4): {
    353  1.19   rmind 		struct ip6_hdr *ip6;
    354  1.19   rmind 		struct ip6_ext *ip6e;
    355  1.19   rmind 		size_t off, hlen;
    356  1.19   rmind 
    357  1.19   rmind 		ip6 = nbuf_ensure_contig(nbuf, sizeof(struct ip6_hdr));
    358  1.19   rmind 		if (ip6 == NULL) {
    359  1.19   rmind 			return 0;
    360   1.7  zoltan 		}
    361  1.19   rmind 
    362  1.19   rmind 		/* Set initial next-protocol value. */
    363  1.19   rmind 		hlen = sizeof(struct ip6_hdr);
    364  1.19   rmind 		npc->npc_proto = ip6->ip6_nxt;
    365  1.13   rmind 		npc->npc_hlen = hlen;
    366   1.7  zoltan 
    367  1.12   rmind 		/*
    368  1.19   rmind 		 * Advance by the length of the current header.
    369  1.12   rmind 		 */
    370  1.19   rmind 		off = nbuf_offset(nbuf);
    371  1.19   rmind 		while (nbuf_advance(nbuf, hlen, 0) != NULL) {
    372  1.19   rmind 			ip6e = nbuf_ensure_contig(nbuf, sizeof(*ip6e));
    373  1.19   rmind 			if (ip6e == NULL) {
    374  1.19   rmind 				return 0;
    375  1.19   rmind 			}
    376  1.19   rmind 
    377  1.13   rmind 			/*
    378  1.13   rmind 			 * Determine whether we are going to continue.
    379  1.13   rmind 			 */
    380  1.19   rmind 			switch (npc->npc_proto) {
    381  1.13   rmind 			case IPPROTO_HOPOPTS:
    382   1.7  zoltan 			case IPPROTO_DSTOPTS:
    383   1.7  zoltan 			case IPPROTO_ROUTING:
    384  1.19   rmind 				hlen = (ip6e->ip6e_len + 1) << 3;
    385   1.7  zoltan 				break;
    386   1.7  zoltan 			case IPPROTO_FRAGMENT:
    387  1.13   rmind 				hlen = sizeof(struct ip6_frag);
    388  1.19   rmind 				flags |= NPC_IPFRAG;
    389   1.7  zoltan 				break;
    390   1.7  zoltan 			case IPPROTO_AH:
    391  1.19   rmind 				hlen = (ip6e->ip6e_len + 2) << 2;
    392   1.7  zoltan 				break;
    393   1.7  zoltan 			default:
    394  1.13   rmind 				hlen = 0;
    395  1.13   rmind 				break;
    396  1.13   rmind 			}
    397  1.13   rmind 
    398  1.13   rmind 			if (!hlen) {
    399   1.7  zoltan 				break;
    400   1.7  zoltan 			}
    401  1.19   rmind 			npc->npc_proto = ip6e->ip6e_nxt;
    402  1.13   rmind 			npc->npc_hlen += hlen;
    403  1.13   rmind 		}
    404   1.7  zoltan 
    405  1.23   rmind 		/*
    406  1.23   rmind 		 * Re-fetch the header pointers (nbufs might have been
    407  1.23   rmind 		 * reallocated).  Restore the original offset (if any).
    408  1.23   rmind 		 */
    409  1.19   rmind 		nbuf_reset(nbuf);
    410  1.23   rmind 		ip6 = nbuf_dataptr(nbuf);
    411  1.19   rmind 		if (off) {
    412  1.19   rmind 			nbuf_advance(nbuf, off, 0);
    413  1.19   rmind 		}
    414  1.19   rmind 
    415  1.12   rmind 		/* Cache: layer 3 - IPv6. */
    416  1.14   rmind 		npc->npc_alen = sizeof(struct in6_addr);
    417  1.28   rmind 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip6->ip6_src;
    418  1.28   rmind 		npc->npc_ips[NPF_DST]= (npf_addr_t *)&ip6->ip6_dst;
    419  1.19   rmind 
    420  1.19   rmind 		npc->npc_ip.v6 = ip6;
    421  1.19   rmind 		flags |= NPC_IP6;
    422   1.7  zoltan 		break;
    423  1.12   rmind 	}
    424   1.4   rmind 	default:
    425  1.19   rmind 		break;
    426   1.4   rmind 	}
    427  1.19   rmind 	return flags;
    428   1.1   rmind }
    429   1.1   rmind 
    430   1.1   rmind /*
    431   1.4   rmind  * npf_cache_all: general routine to cache all relevant IP (v4 or v6)
    432  1.12   rmind  * and TCP, UDP or ICMP headers.
    433  1.19   rmind  *
    434  1.19   rmind  * => nbuf offset shall be set accordingly.
    435   1.1   rmind  */
    436  1.10   rmind int
    437  1.32   rmind npf_cache_all(npf_cache_t *npc)
    438   1.1   rmind {
    439  1.32   rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    440  1.19   rmind 	int flags, l4flags;
    441  1.19   rmind 	u_int hlen;
    442  1.19   rmind 
    443  1.19   rmind 	/*
    444  1.19   rmind 	 * This routine is a main point where the references are cached,
    445  1.19   rmind 	 * therefore clear the flag as we reset.
    446  1.19   rmind 	 */
    447  1.19   rmind again:
    448  1.19   rmind 	nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    449   1.1   rmind 
    450  1.19   rmind 	/*
    451  1.19   rmind 	 * First, cache the L3 header (IPv4 or IPv6).  If IP packet is
    452  1.19   rmind 	 * fragmented, then we cannot look into L4.
    453  1.19   rmind 	 */
    454  1.19   rmind 	flags = npf_cache_ip(npc, nbuf);
    455  1.19   rmind 	if ((flags & NPC_IP46) == 0 || (flags & NPC_IPFRAG) != 0) {
    456  1.23   rmind 		nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    457  1.19   rmind 		npc->npc_info |= flags;
    458  1.19   rmind 		return flags;
    459   1.1   rmind 	}
    460  1.19   rmind 	hlen = npc->npc_hlen;
    461  1.19   rmind 
    462  1.19   rmind 	switch (npc->npc_proto) {
    463   1.1   rmind 	case IPPROTO_TCP:
    464  1.19   rmind 		/* Cache: layer 4 - TCP. */
    465  1.19   rmind 		npc->npc_l4.tcp = nbuf_advance(nbuf, hlen,
    466  1.19   rmind 		    sizeof(struct tcphdr));
    467  1.19   rmind 		l4flags = NPC_LAYER4 | NPC_TCP;
    468  1.10   rmind 		break;
    469   1.1   rmind 	case IPPROTO_UDP:
    470  1.19   rmind 		/* Cache: layer 4 - UDP. */
    471  1.19   rmind 		npc->npc_l4.udp = nbuf_advance(nbuf, hlen,
    472  1.19   rmind 		    sizeof(struct udphdr));
    473  1.19   rmind 		l4flags = NPC_LAYER4 | NPC_UDP;
    474  1.10   rmind 		break;
    475   1.1   rmind 	case IPPROTO_ICMP:
    476  1.19   rmind 		/* Cache: layer 4 - ICMPv4. */
    477  1.19   rmind 		npc->npc_l4.icmp = nbuf_advance(nbuf, hlen,
    478  1.19   rmind 		    offsetof(struct icmp, icmp_void));
    479  1.19   rmind 		l4flags = NPC_LAYER4 | NPC_ICMP;
    480  1.19   rmind 		break;
    481  1.15     spz 	case IPPROTO_ICMPV6:
    482  1.19   rmind 		/* Cache: layer 4 - ICMPv6. */
    483  1.19   rmind 		npc->npc_l4.icmp6 = nbuf_advance(nbuf, hlen,
    484  1.19   rmind 		    offsetof(struct icmp6_hdr, icmp6_data32));
    485  1.19   rmind 		l4flags = NPC_LAYER4 | NPC_ICMP;
    486  1.19   rmind 		break;
    487  1.19   rmind 	default:
    488  1.19   rmind 		l4flags = 0;
    489  1.10   rmind 		break;
    490   1.1   rmind 	}
    491  1.19   rmind 
    492  1.19   rmind 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    493  1.19   rmind 		goto again;
    494  1.19   rmind 	}
    495  1.19   rmind 
    496  1.19   rmind 	/* Add the L4 flags if nbuf_advance() succeeded. */
    497  1.19   rmind 	if (l4flags && npc->npc_l4.hdr) {
    498  1.19   rmind 		flags |= l4flags;
    499  1.19   rmind 	}
    500  1.19   rmind 	npc->npc_info |= flags;
    501  1.19   rmind 	return flags;
    502  1.19   rmind }
    503  1.19   rmind 
    504  1.19   rmind void
    505  1.32   rmind npf_recache(npf_cache_t *npc)
    506  1.19   rmind {
    507  1.32   rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    508  1.24  martin 	const int mflags __diagused = npc->npc_info & (NPC_IP46 | NPC_LAYER4);
    509  1.25     mrg 	int flags __diagused;
    510  1.19   rmind 
    511  1.19   rmind 	nbuf_reset(nbuf);
    512  1.19   rmind 	npc->npc_info = 0;
    513  1.32   rmind 	flags = npf_cache_all(npc);
    514  1.32   rmind 
    515  1.19   rmind 	KASSERT((flags & mflags) == mflags);
    516  1.19   rmind 	KASSERT(nbuf_flag_p(nbuf, NBUF_DATAREF_RESET) == 0);
    517   1.1   rmind }
    518   1.1   rmind 
    519   1.1   rmind /*
    520  1.19   rmind  * npf_rwrip: rewrite required IP address.
    521   1.4   rmind  */
    522   1.4   rmind bool
    523  1.28   rmind npf_rwrip(const npf_cache_t *npc, u_int which, const npf_addr_t *addr)
    524   1.4   rmind {
    525   1.4   rmind 	KASSERT(npf_iscached(npc, NPC_IP46));
    526  1.28   rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    527   1.4   rmind 
    528  1.28   rmind 	memcpy(npc->npc_ips[which], addr, npc->npc_alen);
    529   1.4   rmind 	return true;
    530   1.4   rmind }
    531   1.4   rmind 
    532   1.4   rmind /*
    533  1.19   rmind  * npf_rwrport: rewrite required TCP/UDP port.
    534   1.1   rmind  */
    535   1.1   rmind bool
    536  1.28   rmind npf_rwrport(const npf_cache_t *npc, u_int which, const in_port_t port)
    537   1.1   rmind {
    538  1.21   rmind 	const int proto = npc->npc_proto;
    539   1.4   rmind 	in_port_t *oport;
    540   1.1   rmind 
    541   1.4   rmind 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    542   1.1   rmind 	KASSERT(proto == IPPROTO_TCP || proto == IPPROTO_UDP);
    543  1.28   rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    544   1.1   rmind 
    545  1.19   rmind 	/* Get the offset and store the port in it. */
    546   1.4   rmind 	if (proto == IPPROTO_TCP) {
    547  1.19   rmind 		struct tcphdr *th = npc->npc_l4.tcp;
    548  1.28   rmind 		oport = (which == NPF_SRC) ? &th->th_sport : &th->th_dport;
    549   1.1   rmind 	} else {
    550  1.19   rmind 		struct udphdr *uh = npc->npc_l4.udp;
    551  1.28   rmind 		oport = (which == NPF_SRC) ? &uh->uh_sport : &uh->uh_dport;
    552   1.1   rmind 	}
    553  1.19   rmind 	memcpy(oport, &port, sizeof(in_port_t));
    554   1.1   rmind 	return true;
    555   1.1   rmind }
    556   1.1   rmind 
    557   1.1   rmind /*
    558  1.19   rmind  * npf_rwrcksum: rewrite IPv4 and/or TCP/UDP checksum.
    559   1.1   rmind  */
    560   1.1   rmind bool
    561  1.28   rmind npf_rwrcksum(const npf_cache_t *npc, u_int which,
    562  1.19   rmind     const npf_addr_t *addr, const in_port_t port)
    563   1.1   rmind {
    564  1.28   rmind 	const npf_addr_t *oaddr = npc->npc_ips[which];
    565  1.21   rmind 	const int proto = npc->npc_proto;
    566  1.19   rmind 	const int alen = npc->npc_alen;
    567  1.18   rmind 	uint16_t *ocksum;
    568  1.18   rmind 	in_port_t oport;
    569  1.18   rmind 
    570  1.19   rmind 	KASSERT(npf_iscached(npc, NPC_LAYER4));
    571  1.28   rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    572  1.18   rmind 
    573   1.4   rmind 	if (npf_iscached(npc, NPC_IP4)) {
    574  1.19   rmind 		struct ip *ip = npc->npc_ip.v4;
    575  1.19   rmind 		uint16_t ipsum = ip->ip_sum;
    576   1.4   rmind 
    577  1.19   rmind 		/* Recalculate IPv4 checksum and rewrite. */
    578  1.19   rmind 		ip->ip_sum = npf_addr_cksum(ipsum, alen, oaddr, addr);
    579   1.4   rmind 	} else {
    580   1.4   rmind 		/* No checksum for IPv6. */
    581   1.4   rmind 		KASSERT(npf_iscached(npc, NPC_IP6));
    582   1.4   rmind 	}
    583   1.4   rmind 
    584  1.18   rmind 	/* Nothing else to do for ICMP. */
    585  1.30   rmind 	if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
    586   1.4   rmind 		return true;
    587   1.4   rmind 	}
    588   1.7  zoltan 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    589   1.4   rmind 
    590  1.18   rmind 	/*
    591  1.18   rmind 	 * Calculate TCP/UDP checksum:
    592  1.18   rmind 	 * - Skip if UDP and the current checksum is zero.
    593  1.18   rmind 	 * - Fixup the IP address change.
    594  1.18   rmind 	 * - Fixup the port change, if required (non-zero).
    595  1.18   rmind 	 */
    596   1.4   rmind 	if (proto == IPPROTO_TCP) {
    597  1.19   rmind 		struct tcphdr *th = npc->npc_l4.tcp;
    598   1.4   rmind 
    599  1.18   rmind 		ocksum = &th->th_sum;
    600  1.28   rmind 		oport = (which == NPF_SRC) ? th->th_sport : th->th_dport;
    601   1.4   rmind 	} else {
    602  1.19   rmind 		struct udphdr *uh = npc->npc_l4.udp;
    603   1.4   rmind 
    604   1.4   rmind 		KASSERT(proto == IPPROTO_UDP);
    605  1.18   rmind 		ocksum = &uh->uh_sum;
    606  1.18   rmind 		if (*ocksum == 0) {
    607   1.4   rmind 			/* No need to update. */
    608   1.4   rmind 			return true;
    609   1.4   rmind 		}
    610  1.28   rmind 		oport = (which == NPF_SRC) ? uh->uh_sport : uh->uh_dport;
    611  1.18   rmind 	}
    612  1.18   rmind 
    613  1.19   rmind 	uint16_t cksum = npf_addr_cksum(*ocksum, alen, oaddr, addr);
    614  1.18   rmind 	if (port) {
    615  1.18   rmind 		cksum = npf_fixup16_cksum(cksum, oport, port);
    616   1.4   rmind 	}
    617   1.1   rmind 
    618  1.19   rmind 	/* Rewrite TCP/UDP checksum. */
    619  1.19   rmind 	memcpy(ocksum, &cksum, sizeof(uint16_t));
    620   1.4   rmind 	return true;
    621   1.4   rmind }
    622   1.4   rmind 
    623  1.29   rmind /*
    624  1.30   rmind  * npf_napt_rwr: perform address and/or port translation.
    625  1.30   rmind  */
    626  1.30   rmind int
    627  1.30   rmind npf_napt_rwr(const npf_cache_t *npc, u_int which,
    628  1.30   rmind     const npf_addr_t *addr, const in_addr_t port)
    629  1.30   rmind {
    630  1.30   rmind 	const unsigned proto = npc->npc_proto;
    631  1.30   rmind 
    632  1.30   rmind 	/*
    633  1.30   rmind 	 * Rewrite IP and/or TCP/UDP checksums first, since we need the
    634  1.30   rmind 	 * current (old) address/port for the calculations.  Then perform
    635  1.30   rmind 	 * the address translation i.e. rewrite source or destination.
    636  1.30   rmind 	 */
    637  1.30   rmind 	if (!npf_rwrcksum(npc, which, addr, port)) {
    638  1.30   rmind 		return EINVAL;
    639  1.30   rmind 	}
    640  1.30   rmind 	if (!npf_rwrip(npc, which, addr)) {
    641  1.30   rmind 		return EINVAL;
    642  1.30   rmind 	}
    643  1.30   rmind 	if (port == 0) {
    644  1.30   rmind 		/* Done. */
    645  1.30   rmind 		return 0;
    646  1.30   rmind 	}
    647  1.30   rmind 
    648  1.30   rmind 	switch (proto) {
    649  1.30   rmind 	case IPPROTO_TCP:
    650  1.30   rmind 	case IPPROTO_UDP:
    651  1.30   rmind 		/* Rewrite source/destination port. */
    652  1.30   rmind 		if (!npf_rwrport(npc, which, port)) {
    653  1.30   rmind 			return EINVAL;
    654  1.30   rmind 		}
    655  1.30   rmind 		break;
    656  1.30   rmind 	case IPPROTO_ICMP:
    657  1.30   rmind 	case IPPROTO_ICMPV6:
    658  1.30   rmind 		KASSERT(npf_iscached(npc, NPC_ICMP));
    659  1.30   rmind 		/* Nothing. */
    660  1.30   rmind 		break;
    661  1.30   rmind 	default:
    662  1.30   rmind 		return ENOTSUP;
    663  1.30   rmind 	}
    664  1.30   rmind 	return 0;
    665  1.30   rmind }
    666  1.30   rmind 
    667  1.30   rmind /*
    668  1.29   rmind  * IPv6-to-IPv6 Network Prefix Translation (NPTv6), as per RFC 6296.
    669  1.29   rmind  */
    670  1.29   rmind 
    671  1.29   rmind int
    672  1.29   rmind npf_npt66_rwr(const npf_cache_t *npc, u_int which, const npf_addr_t *pref,
    673  1.29   rmind     npf_netmask_t len, uint16_t adj)
    674  1.29   rmind {
    675  1.29   rmind 	npf_addr_t *addr = npc->npc_ips[which];
    676  1.29   rmind 	unsigned remnant, word, preflen = len >> 4;
    677  1.29   rmind 	uint32_t sum;
    678  1.29   rmind 
    679  1.29   rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    680  1.29   rmind 
    681  1.29   rmind 	if (!npf_iscached(npc, NPC_IP6)) {
    682  1.29   rmind 		return EINVAL;
    683  1.29   rmind 	}
    684  1.29   rmind 	if (len <= 48) {
    685  1.29   rmind 		/*
    686  1.29   rmind 		 * The word to adjust.  Cannot translate the 0xffff
    687  1.29   rmind 		 * subnet if /48 or shorter.
    688  1.29   rmind 		 */
    689  1.29   rmind 		word = 3;
    690  1.29   rmind 		if (addr->s6_addr16[word] == 0xffff) {
    691  1.29   rmind 			return EINVAL;
    692  1.29   rmind 		}
    693  1.29   rmind 	} else {
    694  1.29   rmind 		/*
    695  1.29   rmind 		 * Also, all 0s or 1s in the host part are disallowed for
    696  1.29   rmind 		 * longer than /48 prefixes.
    697  1.29   rmind 		 */
    698  1.29   rmind 		if ((addr->s6_addr32[2] == 0 && addr->s6_addr32[3] == 0) ||
    699  1.29   rmind 		    (addr->s6_addr32[2] == ~0U && addr->s6_addr32[3] == ~0U))
    700  1.29   rmind 			return EINVAL;
    701  1.29   rmind 
    702  1.29   rmind 		/* Determine the 16-bit word to adjust. */
    703  1.29   rmind 		for (word = 4; word < 8; word++)
    704  1.29   rmind 			if (addr->s6_addr16[word] != 0xffff)
    705  1.29   rmind 				break;
    706  1.29   rmind 	}
    707  1.29   rmind 
    708  1.29   rmind 	/* Rewrite the prefix. */
    709  1.29   rmind 	for (unsigned i = 0; i < preflen; i++) {
    710  1.29   rmind 		addr->s6_addr16[i] = pref->s6_addr16[i];
    711  1.29   rmind 	}
    712  1.29   rmind 
    713  1.29   rmind 	/*
    714  1.29   rmind 	 * If prefix length is within a 16-bit word (not dividable by 16),
    715  1.29   rmind 	 * then prepare a mask, determine the word and adjust it.
    716  1.29   rmind 	 */
    717  1.29   rmind 	if ((remnant = len - (preflen << 4)) != 0) {
    718  1.29   rmind 		const uint16_t wordmask = (1U << remnant) - 1;
    719  1.29   rmind 		const unsigned i = preflen;
    720  1.29   rmind 
    721  1.29   rmind 		addr->s6_addr16[i] = (pref->s6_addr16[i] & wordmask) |
    722  1.29   rmind 		    (addr->s6_addr16[i] & ~wordmask);
    723  1.29   rmind 	}
    724  1.29   rmind 
    725  1.29   rmind 	/*
    726  1.29   rmind 	 * Performing 1's complement sum/difference.
    727  1.29   rmind 	 */
    728  1.29   rmind 	sum = addr->s6_addr16[word] + adj;
    729  1.29   rmind 	while (sum >> 16) {
    730  1.29   rmind 		sum = (sum >> 16) + (sum & 0xffff);
    731  1.29   rmind 	}
    732  1.29   rmind 	if (sum == 0xffff) {
    733  1.29   rmind 		/* RFC 1071. */
    734  1.29   rmind 		sum = 0x0000;
    735  1.29   rmind 	}
    736  1.29   rmind 	addr->s6_addr16[word] = sum;
    737  1.29   rmind 	return 0;
    738  1.29   rmind }
    739  1.29   rmind 
    740  1.13   rmind #if defined(DDB) || defined(_NPF_TESTING)
    741  1.13   rmind 
    742  1.31   rmind const char *
    743  1.31   rmind npf_addr_dump(const npf_addr_t *addr, int alen)
    744  1.13   rmind {
    745  1.31   rmind 	if (alen == sizeof(struct in_addr)) {
    746  1.31   rmind 		struct in_addr ip;
    747  1.31   rmind 		memcpy(&ip, addr, alen);
    748  1.31   rmind 		return inet_ntoa(ip);
    749  1.31   rmind 	}
    750  1.31   rmind 	return "[IPv6]"; // XXX
    751  1.13   rmind }
    752  1.13   rmind 
    753  1.13   rmind #endif
    754