Home | History | Annotate | Line # | Download | only in npf
npf_inet.c revision 1.33
      1  1.33  mlelstv /*	$NetBSD: npf_inet.c,v 1.33 2015/12/17 12:17:13 mlelstv 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  1.33  mlelstv __KERNEL_RCSID(0, "$NetBSD: npf_inet.c,v 1.33 2015/12/17 12:17:13 mlelstv 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  1.33  mlelstv #include <netinet6/in6_var.h>
     55   1.1    rmind #include <netinet/ip.h>
     56   1.4    rmind #include <netinet/ip6.h>
     57   1.1    rmind #include <netinet/tcp.h>
     58   1.1    rmind #include <netinet/udp.h>
     59   1.1    rmind #include <netinet/ip_icmp.h>
     60   1.1    rmind 
     61   1.1    rmind #include "npf_impl.h"
     62   1.1    rmind 
     63   1.1    rmind /*
     64  1.27    rmind  * npf_fixup{16,32}_cksum: incremental update of the Internet checksum.
     65   1.1    rmind  */
     66   1.1    rmind 
     67   1.1    rmind uint16_t
     68   1.1    rmind npf_fixup16_cksum(uint16_t cksum, uint16_t odatum, uint16_t ndatum)
     69   1.1    rmind {
     70   1.1    rmind 	uint32_t sum;
     71   1.1    rmind 
     72   1.1    rmind 	/*
     73   1.1    rmind 	 * RFC 1624:
     74   1.1    rmind 	 *	HC' = ~(~HC + ~m + m')
     75  1.27    rmind 	 *
     76  1.27    rmind 	 * Note: 1's complement sum is endian-independent (RFC 1071, page 2).
     77   1.1    rmind 	 */
     78  1.27    rmind 	sum = ~cksum & 0xffff;
     79  1.27    rmind 	sum += (~odatum & 0xffff) + ndatum;
     80   1.1    rmind 	sum = (sum >> 16) + (sum & 0xffff);
     81   1.1    rmind 	sum += (sum >> 16);
     82   1.1    rmind 
     83  1.27    rmind 	return ~sum & 0xffff;
     84   1.1    rmind }
     85   1.1    rmind 
     86   1.1    rmind uint16_t
     87   1.1    rmind npf_fixup32_cksum(uint16_t cksum, uint32_t odatum, uint32_t ndatum)
     88   1.1    rmind {
     89  1.27    rmind 	uint32_t sum;
     90  1.27    rmind 
     91  1.27    rmind 	/*
     92  1.27    rmind 	 * Checksum 32-bit datum as as two 16-bit.  Note, the first
     93  1.27    rmind 	 * 32->16 bit reduction is not necessary.
     94  1.27    rmind 	 */
     95  1.27    rmind 	sum = ~cksum & 0xffff;
     96  1.27    rmind 	sum += (~odatum & 0xffff) + (ndatum & 0xffff);
     97   1.1    rmind 
     98  1.27    rmind 	sum += (~odatum >> 16) + (ndatum >> 16);
     99  1.27    rmind 	sum = (sum >> 16) + (sum & 0xffff);
    100  1.27    rmind 	sum += (sum >> 16);
    101  1.27    rmind 	return ~sum & 0xffff;
    102   1.1    rmind }
    103   1.1    rmind 
    104   1.1    rmind /*
    105   1.4    rmind  * npf_addr_cksum: calculate checksum of the address, either IPv4 or IPv6.
    106   1.4    rmind  */
    107   1.4    rmind uint16_t
    108  1.19    rmind npf_addr_cksum(uint16_t cksum, int sz, const npf_addr_t *oaddr,
    109  1.19    rmind     const npf_addr_t *naddr)
    110   1.4    rmind {
    111  1.19    rmind 	const uint32_t *oip32 = (const uint32_t *)oaddr;
    112  1.19    rmind 	const uint32_t *nip32 = (const uint32_t *)naddr;
    113   1.4    rmind 
    114   1.4    rmind 	KASSERT(sz % sizeof(uint32_t) == 0);
    115   1.4    rmind 	do {
    116   1.4    rmind 		cksum = npf_fixup32_cksum(cksum, *oip32++, *nip32++);
    117   1.4    rmind 		sz -= sizeof(uint32_t);
    118   1.4    rmind 	} while (sz);
    119   1.4    rmind 
    120   1.4    rmind 	return cksum;
    121   1.4    rmind }
    122   1.4    rmind 
    123   1.4    rmind /*
    124  1.26    rmind  * npf_addr_sum: provide IP addresses as a XORed 32-bit integer.
    125   1.4    rmind  * Note: used for hash function.
    126   1.1    rmind  */
    127   1.4    rmind uint32_t
    128  1.26    rmind npf_addr_mix(const int sz, const npf_addr_t *a1, const npf_addr_t *a2)
    129   1.1    rmind {
    130   1.4    rmind 	uint32_t mix = 0;
    131   1.1    rmind 
    132   1.5    rmind 	KASSERT(sz > 0 && a1 != NULL && a2 != NULL);
    133   1.5    rmind 
    134  1.26    rmind 	for (int i = 0; i < (sz >> 2); i++) {
    135  1.26    rmind 		mix ^= a1->s6_addr32[i];
    136  1.26    rmind 		mix ^= a2->s6_addr32[i];
    137   1.4    rmind 	}
    138   1.4    rmind 	return mix;
    139   1.4    rmind }
    140   1.1    rmind 
    141  1.13    rmind /*
    142  1.13    rmind  * npf_addr_mask: apply the mask to a given address and store the result.
    143  1.13    rmind  */
    144  1.13    rmind void
    145  1.13    rmind npf_addr_mask(const npf_addr_t *addr, const npf_netmask_t mask,
    146  1.13    rmind     const int alen, npf_addr_t *out)
    147  1.12    rmind {
    148  1.13    rmind 	const int nwords = alen >> 2;
    149  1.12    rmind 	uint_fast8_t length = mask;
    150  1.12    rmind 
    151  1.12    rmind 	/* Note: maximum length is 32 for IPv4 and 128 for IPv6. */
    152  1.12    rmind 	KASSERT(length <= NPF_MAX_NETMASK);
    153  1.12    rmind 
    154  1.13    rmind 	for (int i = 0; i < nwords; i++) {
    155  1.13    rmind 		uint32_t wordmask;
    156  1.13    rmind 
    157  1.12    rmind 		if (length >= 32) {
    158  1.13    rmind 			wordmask = htonl(0xffffffff);
    159  1.12    rmind 			length -= 32;
    160  1.13    rmind 		} else if (length) {
    161  1.13    rmind 			wordmask = htonl(0xffffffff << (32 - length));
    162  1.13    rmind 			length = 0;
    163  1.12    rmind 		} else {
    164  1.13    rmind 			wordmask = 0;
    165  1.12    rmind 		}
    166  1.13    rmind 		out->s6_addr32[i] = addr->s6_addr32[i] & wordmask;
    167  1.12    rmind 	}
    168  1.12    rmind }
    169  1.12    rmind 
    170  1.12    rmind /*
    171  1.12    rmind  * npf_addr_cmp: compare two addresses, either IPv4 or IPv6.
    172  1.12    rmind  *
    173  1.13    rmind  * => Return 0 if equal and negative/positive if less/greater accordingly.
    174  1.12    rmind  * => Ignore the mask, if NPF_NO_NETMASK is specified.
    175  1.12    rmind  */
    176  1.12    rmind int
    177  1.12    rmind npf_addr_cmp(const npf_addr_t *addr1, const npf_netmask_t mask1,
    178  1.13    rmind     const npf_addr_t *addr2, const npf_netmask_t mask2, const int alen)
    179  1.12    rmind {
    180  1.13    rmind 	npf_addr_t realaddr1, realaddr2;
    181  1.12    rmind 
    182  1.12    rmind 	if (mask1 != NPF_NO_NETMASK) {
    183  1.13    rmind 		npf_addr_mask(addr1, mask1, alen, &realaddr1);
    184  1.13    rmind 		addr1 = &realaddr1;
    185  1.12    rmind 	}
    186  1.12    rmind 	if (mask2 != NPF_NO_NETMASK) {
    187  1.13    rmind 		npf_addr_mask(addr2, mask2, alen, &realaddr2);
    188  1.13    rmind 		addr2 = &realaddr2;
    189  1.12    rmind 	}
    190  1.13    rmind 	return memcmp(addr1, addr2, alen);
    191  1.12    rmind }
    192  1.12    rmind 
    193   1.4    rmind /*
    194   1.4    rmind  * npf_tcpsaw: helper to fetch SEQ, ACK, WIN and return TCP data length.
    195  1.12    rmind  *
    196  1.12    rmind  * => Returns all values in host byte-order.
    197   1.4    rmind  */
    198   1.4    rmind int
    199  1.12    rmind npf_tcpsaw(const npf_cache_t *npc, tcp_seq *seq, tcp_seq *ack, uint32_t *win)
    200   1.4    rmind {
    201  1.19    rmind 	const struct tcphdr *th = npc->npc_l4.tcp;
    202   1.8    rmind 	u_int thlen;
    203   1.1    rmind 
    204   1.7   zoltan 	KASSERT(npf_iscached(npc, NPC_TCP));
    205   1.1    rmind 
    206   1.4    rmind 	*seq = ntohl(th->th_seq);
    207   1.4    rmind 	*ack = ntohl(th->th_ack);
    208   1.4    rmind 	*win = (uint32_t)ntohs(th->th_win);
    209   1.8    rmind 	thlen = th->th_off << 2;
    210   1.1    rmind 
    211   1.7   zoltan 	if (npf_iscached(npc, NPC_IP4)) {
    212  1.19    rmind 		const struct ip *ip = npc->npc_ip.v4;
    213  1.21    rmind 		return ntohs(ip->ip_len) - npc->npc_hlen - thlen;
    214  1.12    rmind 	} else if (npf_iscached(npc, NPC_IP6)) {
    215  1.19    rmind 		const struct ip6_hdr *ip6 = npc->npc_ip.v6;
    216   1.8    rmind 		return ntohs(ip6->ip6_plen) - thlen;
    217   1.7   zoltan 	}
    218   1.7   zoltan 	return 0;
    219   1.1    rmind }
    220   1.1    rmind 
    221   1.1    rmind /*
    222   1.4    rmind  * npf_fetch_tcpopts: parse and return TCP options.
    223   1.1    rmind  */
    224   1.1    rmind bool
    225  1.32    rmind npf_fetch_tcpopts(npf_cache_t *npc, uint16_t *mss, int *wscale)
    226   1.1    rmind {
    227  1.32    rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    228  1.19    rmind 	const struct tcphdr *th = npc->npc_l4.tcp;
    229   1.4    rmind 	int topts_len, step;
    230  1.19    rmind 	void *nptr;
    231   1.4    rmind 	uint8_t val;
    232  1.19    rmind 	bool ok;
    233   1.4    rmind 
    234   1.7   zoltan 	KASSERT(npf_iscached(npc, NPC_IP46));
    235   1.7   zoltan 	KASSERT(npf_iscached(npc, NPC_TCP));
    236  1.10    rmind 
    237   1.4    rmind 	/* Determine if there are any TCP options, get their length. */
    238   1.4    rmind 	topts_len = (th->th_off << 2) - sizeof(struct tcphdr);
    239   1.4    rmind 	if (topts_len <= 0) {
    240   1.4    rmind 		/* No options. */
    241   1.1    rmind 		return false;
    242   1.4    rmind 	}
    243   1.4    rmind 	KASSERT(topts_len <= MAX_TCPOPTLEN);
    244   1.1    rmind 
    245   1.4    rmind 	/* First step: IP and TCP header up to options. */
    246  1.21    rmind 	step = npc->npc_hlen + sizeof(struct tcphdr);
    247  1.19    rmind 	nbuf_reset(nbuf);
    248   1.4    rmind next:
    249  1.19    rmind 	if ((nptr = nbuf_advance(nbuf, step, 1)) == NULL) {
    250  1.19    rmind 		ok = false;
    251  1.19    rmind 		goto done;
    252   1.4    rmind 	}
    253  1.19    rmind 	val = *(uint8_t *)nptr;
    254  1.12    rmind 
    255   1.4    rmind 	switch (val) {
    256   1.4    rmind 	case TCPOPT_EOL:
    257   1.4    rmind 		/* Done. */
    258  1.19    rmind 		ok = true;
    259  1.19    rmind 		goto done;
    260   1.4    rmind 	case TCPOPT_NOP:
    261   1.4    rmind 		topts_len--;
    262   1.4    rmind 		step = 1;
    263   1.4    rmind 		break;
    264   1.4    rmind 	case TCPOPT_MAXSEG:
    265  1.19    rmind 		if ((nptr = nbuf_advance(nbuf, 2, 2)) == NULL) {
    266  1.19    rmind 			ok = false;
    267  1.19    rmind 			goto done;
    268   1.4    rmind 		}
    269   1.4    rmind 		if (mss) {
    270  1.19    rmind 			if (*mss) {
    271  1.19    rmind 				memcpy(nptr, mss, sizeof(uint16_t));
    272  1.19    rmind 			} else {
    273  1.19    rmind 				memcpy(mss, nptr, sizeof(uint16_t));
    274  1.19    rmind 			}
    275   1.4    rmind 		}
    276   1.4    rmind 		topts_len -= TCPOLEN_MAXSEG;
    277  1.19    rmind 		step = 2;
    278   1.4    rmind 		break;
    279   1.4    rmind 	case TCPOPT_WINDOW:
    280  1.10    rmind 		/* TCP Window Scaling (RFC 1323). */
    281  1.19    rmind 		if ((nptr = nbuf_advance(nbuf, 2, 1)) == NULL) {
    282  1.19    rmind 			ok = false;
    283  1.19    rmind 			goto done;
    284   1.4    rmind 		}
    285  1.19    rmind 		val = *(uint8_t *)nptr;
    286   1.4    rmind 		*wscale = (val > TCP_MAX_WINSHIFT) ? TCP_MAX_WINSHIFT : val;
    287   1.4    rmind 		topts_len -= TCPOLEN_WINDOW;
    288  1.19    rmind 		step = 1;
    289   1.4    rmind 		break;
    290   1.4    rmind 	default:
    291  1.19    rmind 		if ((nptr = nbuf_advance(nbuf, 1, 1)) == NULL) {
    292  1.19    rmind 			ok = false;
    293  1.19    rmind 			goto done;
    294   1.4    rmind 		}
    295  1.19    rmind 		val = *(uint8_t *)nptr;
    296  1.16    rmind 		if (val < 2 || val > topts_len) {
    297  1.19    rmind 			ok = false;
    298  1.19    rmind 			goto done;
    299   1.4    rmind 		}
    300   1.4    rmind 		topts_len -= val;
    301   1.4    rmind 		step = val - 1;
    302   1.4    rmind 	}
    303  1.12    rmind 
    304   1.6    rmind 	/* Any options left? */
    305   1.4    rmind 	if (__predict_true(topts_len > 0)) {
    306   1.4    rmind 		goto next;
    307   1.4    rmind 	}
    308  1.19    rmind 	ok = true;
    309  1.19    rmind done:
    310  1.19    rmind 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    311  1.32    rmind 		npf_recache(npc);
    312  1.19    rmind 	}
    313  1.19    rmind 	return ok;
    314   1.1    rmind }
    315   1.1    rmind 
    316  1.19    rmind static int
    317  1.19    rmind npf_cache_ip(npf_cache_t *npc, nbuf_t *nbuf)
    318   1.1    rmind {
    319  1.19    rmind 	const void *nptr = nbuf_dataptr(nbuf);
    320  1.19    rmind 	const uint8_t ver = *(const uint8_t *)nptr;
    321  1.19    rmind 	int flags = 0;
    322  1.12    rmind 
    323   1.4    rmind 	switch (ver >> 4) {
    324  1.12    rmind 	case IPVERSION: {
    325  1.19    rmind 		struct ip *ip;
    326  1.12    rmind 
    327  1.19    rmind 		ip = nbuf_ensure_contig(nbuf, sizeof(struct ip));
    328  1.19    rmind 		if (ip == NULL) {
    329  1.19    rmind 			return 0;
    330   1.4    rmind 		}
    331  1.12    rmind 
    332   1.4    rmind 		/* Check header length and fragment offset. */
    333  1.10    rmind 		if ((u_int)(ip->ip_hl << 2) < sizeof(struct ip)) {
    334  1.19    rmind 			return 0;
    335   1.4    rmind 		}
    336   1.4    rmind 		if (ip->ip_off & ~htons(IP_DF | IP_RF)) {
    337   1.4    rmind 			/* Note fragmentation. */
    338  1.19    rmind 			flags |= NPC_IPFRAG;
    339   1.4    rmind 		}
    340  1.12    rmind 
    341   1.4    rmind 		/* Cache: layer 3 - IPv4. */
    342  1.14    rmind 		npc->npc_alen = sizeof(struct in_addr);
    343  1.28    rmind 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip->ip_src;
    344  1.28    rmind 		npc->npc_ips[NPF_DST] = (npf_addr_t *)&ip->ip_dst;
    345   1.7   zoltan 		npc->npc_hlen = ip->ip_hl << 2;
    346  1.19    rmind 		npc->npc_proto = ip->ip_p;
    347  1.19    rmind 
    348  1.19    rmind 		npc->npc_ip.v4 = ip;
    349  1.19    rmind 		flags |= NPC_IP4;
    350   1.4    rmind 		break;
    351  1.12    rmind 	}
    352   1.4    rmind 
    353  1.12    rmind 	case (IPV6_VERSION >> 4): {
    354  1.19    rmind 		struct ip6_hdr *ip6;
    355  1.19    rmind 		struct ip6_ext *ip6e;
    356  1.19    rmind 		size_t off, hlen;
    357  1.19    rmind 
    358  1.19    rmind 		ip6 = nbuf_ensure_contig(nbuf, sizeof(struct ip6_hdr));
    359  1.19    rmind 		if (ip6 == NULL) {
    360  1.19    rmind 			return 0;
    361   1.7   zoltan 		}
    362  1.19    rmind 
    363  1.19    rmind 		/* Set initial next-protocol value. */
    364  1.19    rmind 		hlen = sizeof(struct ip6_hdr);
    365  1.19    rmind 		npc->npc_proto = ip6->ip6_nxt;
    366  1.13    rmind 		npc->npc_hlen = hlen;
    367   1.7   zoltan 
    368  1.12    rmind 		/*
    369  1.19    rmind 		 * Advance by the length of the current header.
    370  1.12    rmind 		 */
    371  1.19    rmind 		off = nbuf_offset(nbuf);
    372  1.19    rmind 		while (nbuf_advance(nbuf, hlen, 0) != NULL) {
    373  1.19    rmind 			ip6e = nbuf_ensure_contig(nbuf, sizeof(*ip6e));
    374  1.19    rmind 			if (ip6e == NULL) {
    375  1.19    rmind 				return 0;
    376  1.19    rmind 			}
    377  1.19    rmind 
    378  1.13    rmind 			/*
    379  1.13    rmind 			 * Determine whether we are going to continue.
    380  1.13    rmind 			 */
    381  1.19    rmind 			switch (npc->npc_proto) {
    382  1.13    rmind 			case IPPROTO_HOPOPTS:
    383   1.7   zoltan 			case IPPROTO_DSTOPTS:
    384   1.7   zoltan 			case IPPROTO_ROUTING:
    385  1.19    rmind 				hlen = (ip6e->ip6e_len + 1) << 3;
    386   1.7   zoltan 				break;
    387   1.7   zoltan 			case IPPROTO_FRAGMENT:
    388  1.13    rmind 				hlen = sizeof(struct ip6_frag);
    389  1.19    rmind 				flags |= NPC_IPFRAG;
    390   1.7   zoltan 				break;
    391   1.7   zoltan 			case IPPROTO_AH:
    392  1.19    rmind 				hlen = (ip6e->ip6e_len + 2) << 2;
    393   1.7   zoltan 				break;
    394   1.7   zoltan 			default:
    395  1.13    rmind 				hlen = 0;
    396  1.13    rmind 				break;
    397  1.13    rmind 			}
    398  1.13    rmind 
    399  1.13    rmind 			if (!hlen) {
    400   1.7   zoltan 				break;
    401   1.7   zoltan 			}
    402  1.19    rmind 			npc->npc_proto = ip6e->ip6e_nxt;
    403  1.13    rmind 			npc->npc_hlen += hlen;
    404  1.13    rmind 		}
    405   1.7   zoltan 
    406  1.23    rmind 		/*
    407  1.23    rmind 		 * Re-fetch the header pointers (nbufs might have been
    408  1.23    rmind 		 * reallocated).  Restore the original offset (if any).
    409  1.23    rmind 		 */
    410  1.19    rmind 		nbuf_reset(nbuf);
    411  1.23    rmind 		ip6 = nbuf_dataptr(nbuf);
    412  1.19    rmind 		if (off) {
    413  1.19    rmind 			nbuf_advance(nbuf, off, 0);
    414  1.19    rmind 		}
    415  1.19    rmind 
    416  1.12    rmind 		/* Cache: layer 3 - IPv6. */
    417  1.14    rmind 		npc->npc_alen = sizeof(struct in6_addr);
    418  1.28    rmind 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip6->ip6_src;
    419  1.28    rmind 		npc->npc_ips[NPF_DST]= (npf_addr_t *)&ip6->ip6_dst;
    420  1.19    rmind 
    421  1.19    rmind 		npc->npc_ip.v6 = ip6;
    422  1.19    rmind 		flags |= NPC_IP6;
    423   1.7   zoltan 		break;
    424  1.12    rmind 	}
    425   1.4    rmind 	default:
    426  1.19    rmind 		break;
    427   1.4    rmind 	}
    428  1.19    rmind 	return flags;
    429   1.1    rmind }
    430   1.1    rmind 
    431   1.1    rmind /*
    432   1.4    rmind  * npf_cache_all: general routine to cache all relevant IP (v4 or v6)
    433  1.12    rmind  * and TCP, UDP or ICMP headers.
    434  1.19    rmind  *
    435  1.19    rmind  * => nbuf offset shall be set accordingly.
    436   1.1    rmind  */
    437  1.10    rmind int
    438  1.32    rmind npf_cache_all(npf_cache_t *npc)
    439   1.1    rmind {
    440  1.32    rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    441  1.19    rmind 	int flags, l4flags;
    442  1.19    rmind 	u_int hlen;
    443  1.19    rmind 
    444  1.19    rmind 	/*
    445  1.19    rmind 	 * This routine is a main point where the references are cached,
    446  1.19    rmind 	 * therefore clear the flag as we reset.
    447  1.19    rmind 	 */
    448  1.19    rmind again:
    449  1.19    rmind 	nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    450   1.1    rmind 
    451  1.19    rmind 	/*
    452  1.19    rmind 	 * First, cache the L3 header (IPv4 or IPv6).  If IP packet is
    453  1.19    rmind 	 * fragmented, then we cannot look into L4.
    454  1.19    rmind 	 */
    455  1.19    rmind 	flags = npf_cache_ip(npc, nbuf);
    456  1.19    rmind 	if ((flags & NPC_IP46) == 0 || (flags & NPC_IPFRAG) != 0) {
    457  1.23    rmind 		nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    458  1.19    rmind 		npc->npc_info |= flags;
    459  1.19    rmind 		return flags;
    460   1.1    rmind 	}
    461  1.19    rmind 	hlen = npc->npc_hlen;
    462  1.19    rmind 
    463  1.19    rmind 	switch (npc->npc_proto) {
    464   1.1    rmind 	case IPPROTO_TCP:
    465  1.19    rmind 		/* Cache: layer 4 - TCP. */
    466  1.19    rmind 		npc->npc_l4.tcp = nbuf_advance(nbuf, hlen,
    467  1.19    rmind 		    sizeof(struct tcphdr));
    468  1.19    rmind 		l4flags = NPC_LAYER4 | NPC_TCP;
    469  1.10    rmind 		break;
    470   1.1    rmind 	case IPPROTO_UDP:
    471  1.19    rmind 		/* Cache: layer 4 - UDP. */
    472  1.19    rmind 		npc->npc_l4.udp = nbuf_advance(nbuf, hlen,
    473  1.19    rmind 		    sizeof(struct udphdr));
    474  1.19    rmind 		l4flags = NPC_LAYER4 | NPC_UDP;
    475  1.10    rmind 		break;
    476   1.1    rmind 	case IPPROTO_ICMP:
    477  1.19    rmind 		/* Cache: layer 4 - ICMPv4. */
    478  1.19    rmind 		npc->npc_l4.icmp = nbuf_advance(nbuf, hlen,
    479  1.19    rmind 		    offsetof(struct icmp, icmp_void));
    480  1.19    rmind 		l4flags = NPC_LAYER4 | NPC_ICMP;
    481  1.19    rmind 		break;
    482  1.15      spz 	case IPPROTO_ICMPV6:
    483  1.19    rmind 		/* Cache: layer 4 - ICMPv6. */
    484  1.19    rmind 		npc->npc_l4.icmp6 = nbuf_advance(nbuf, hlen,
    485  1.19    rmind 		    offsetof(struct icmp6_hdr, icmp6_data32));
    486  1.19    rmind 		l4flags = NPC_LAYER4 | NPC_ICMP;
    487  1.19    rmind 		break;
    488  1.19    rmind 	default:
    489  1.19    rmind 		l4flags = 0;
    490  1.10    rmind 		break;
    491   1.1    rmind 	}
    492  1.19    rmind 
    493  1.19    rmind 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    494  1.19    rmind 		goto again;
    495  1.19    rmind 	}
    496  1.19    rmind 
    497  1.19    rmind 	/* Add the L4 flags if nbuf_advance() succeeded. */
    498  1.19    rmind 	if (l4flags && npc->npc_l4.hdr) {
    499  1.19    rmind 		flags |= l4flags;
    500  1.19    rmind 	}
    501  1.19    rmind 	npc->npc_info |= flags;
    502  1.19    rmind 	return flags;
    503  1.19    rmind }
    504  1.19    rmind 
    505  1.19    rmind void
    506  1.32    rmind npf_recache(npf_cache_t *npc)
    507  1.19    rmind {
    508  1.32    rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    509  1.24   martin 	const int mflags __diagused = npc->npc_info & (NPC_IP46 | NPC_LAYER4);
    510  1.25      mrg 	int flags __diagused;
    511  1.19    rmind 
    512  1.19    rmind 	nbuf_reset(nbuf);
    513  1.19    rmind 	npc->npc_info = 0;
    514  1.32    rmind 	flags = npf_cache_all(npc);
    515  1.32    rmind 
    516  1.19    rmind 	KASSERT((flags & mflags) == mflags);
    517  1.19    rmind 	KASSERT(nbuf_flag_p(nbuf, NBUF_DATAREF_RESET) == 0);
    518   1.1    rmind }
    519   1.1    rmind 
    520   1.1    rmind /*
    521  1.19    rmind  * npf_rwrip: rewrite required IP address.
    522   1.4    rmind  */
    523   1.4    rmind bool
    524  1.28    rmind npf_rwrip(const npf_cache_t *npc, u_int which, const npf_addr_t *addr)
    525   1.4    rmind {
    526   1.4    rmind 	KASSERT(npf_iscached(npc, NPC_IP46));
    527  1.28    rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    528   1.4    rmind 
    529  1.28    rmind 	memcpy(npc->npc_ips[which], addr, npc->npc_alen);
    530   1.4    rmind 	return true;
    531   1.4    rmind }
    532   1.4    rmind 
    533   1.4    rmind /*
    534  1.19    rmind  * npf_rwrport: rewrite required TCP/UDP port.
    535   1.1    rmind  */
    536   1.1    rmind bool
    537  1.28    rmind npf_rwrport(const npf_cache_t *npc, u_int which, const in_port_t port)
    538   1.1    rmind {
    539  1.21    rmind 	const int proto = npc->npc_proto;
    540   1.4    rmind 	in_port_t *oport;
    541   1.1    rmind 
    542   1.4    rmind 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    543   1.1    rmind 	KASSERT(proto == IPPROTO_TCP || proto == IPPROTO_UDP);
    544  1.28    rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    545   1.1    rmind 
    546  1.19    rmind 	/* Get the offset and store the port in it. */
    547   1.4    rmind 	if (proto == IPPROTO_TCP) {
    548  1.19    rmind 		struct tcphdr *th = npc->npc_l4.tcp;
    549  1.28    rmind 		oport = (which == NPF_SRC) ? &th->th_sport : &th->th_dport;
    550   1.1    rmind 	} else {
    551  1.19    rmind 		struct udphdr *uh = npc->npc_l4.udp;
    552  1.28    rmind 		oport = (which == NPF_SRC) ? &uh->uh_sport : &uh->uh_dport;
    553   1.1    rmind 	}
    554  1.19    rmind 	memcpy(oport, &port, sizeof(in_port_t));
    555   1.1    rmind 	return true;
    556   1.1    rmind }
    557   1.1    rmind 
    558   1.1    rmind /*
    559  1.19    rmind  * npf_rwrcksum: rewrite IPv4 and/or TCP/UDP checksum.
    560   1.1    rmind  */
    561   1.1    rmind bool
    562  1.28    rmind npf_rwrcksum(const npf_cache_t *npc, u_int which,
    563  1.19    rmind     const npf_addr_t *addr, const in_port_t port)
    564   1.1    rmind {
    565  1.28    rmind 	const npf_addr_t *oaddr = npc->npc_ips[which];
    566  1.21    rmind 	const int proto = npc->npc_proto;
    567  1.19    rmind 	const int alen = npc->npc_alen;
    568  1.18    rmind 	uint16_t *ocksum;
    569  1.18    rmind 	in_port_t oport;
    570  1.18    rmind 
    571  1.19    rmind 	KASSERT(npf_iscached(npc, NPC_LAYER4));
    572  1.28    rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    573  1.18    rmind 
    574   1.4    rmind 	if (npf_iscached(npc, NPC_IP4)) {
    575  1.19    rmind 		struct ip *ip = npc->npc_ip.v4;
    576  1.19    rmind 		uint16_t ipsum = ip->ip_sum;
    577   1.4    rmind 
    578  1.19    rmind 		/* Recalculate IPv4 checksum and rewrite. */
    579  1.19    rmind 		ip->ip_sum = npf_addr_cksum(ipsum, alen, oaddr, addr);
    580   1.4    rmind 	} else {
    581   1.4    rmind 		/* No checksum for IPv6. */
    582   1.4    rmind 		KASSERT(npf_iscached(npc, NPC_IP6));
    583   1.4    rmind 	}
    584   1.4    rmind 
    585  1.18    rmind 	/* Nothing else to do for ICMP. */
    586  1.30    rmind 	if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
    587   1.4    rmind 		return true;
    588   1.4    rmind 	}
    589   1.7   zoltan 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    590   1.4    rmind 
    591  1.18    rmind 	/*
    592  1.18    rmind 	 * Calculate TCP/UDP checksum:
    593  1.18    rmind 	 * - Skip if UDP and the current checksum is zero.
    594  1.18    rmind 	 * - Fixup the IP address change.
    595  1.18    rmind 	 * - Fixup the port change, if required (non-zero).
    596  1.18    rmind 	 */
    597   1.4    rmind 	if (proto == IPPROTO_TCP) {
    598  1.19    rmind 		struct tcphdr *th = npc->npc_l4.tcp;
    599   1.4    rmind 
    600  1.18    rmind 		ocksum = &th->th_sum;
    601  1.28    rmind 		oport = (which == NPF_SRC) ? th->th_sport : th->th_dport;
    602   1.4    rmind 	} else {
    603  1.19    rmind 		struct udphdr *uh = npc->npc_l4.udp;
    604   1.4    rmind 
    605   1.4    rmind 		KASSERT(proto == IPPROTO_UDP);
    606  1.18    rmind 		ocksum = &uh->uh_sum;
    607  1.18    rmind 		if (*ocksum == 0) {
    608   1.4    rmind 			/* No need to update. */
    609   1.4    rmind 			return true;
    610   1.4    rmind 		}
    611  1.28    rmind 		oport = (which == NPF_SRC) ? uh->uh_sport : uh->uh_dport;
    612  1.18    rmind 	}
    613  1.18    rmind 
    614  1.19    rmind 	uint16_t cksum = npf_addr_cksum(*ocksum, alen, oaddr, addr);
    615  1.18    rmind 	if (port) {
    616  1.18    rmind 		cksum = npf_fixup16_cksum(cksum, oport, port);
    617   1.4    rmind 	}
    618   1.1    rmind 
    619  1.19    rmind 	/* Rewrite TCP/UDP checksum. */
    620  1.19    rmind 	memcpy(ocksum, &cksum, sizeof(uint16_t));
    621   1.4    rmind 	return true;
    622   1.4    rmind }
    623   1.4    rmind 
    624  1.29    rmind /*
    625  1.30    rmind  * npf_napt_rwr: perform address and/or port translation.
    626  1.30    rmind  */
    627  1.30    rmind int
    628  1.30    rmind npf_napt_rwr(const npf_cache_t *npc, u_int which,
    629  1.30    rmind     const npf_addr_t *addr, const in_addr_t port)
    630  1.30    rmind {
    631  1.30    rmind 	const unsigned proto = npc->npc_proto;
    632  1.30    rmind 
    633  1.30    rmind 	/*
    634  1.30    rmind 	 * Rewrite IP and/or TCP/UDP checksums first, since we need the
    635  1.30    rmind 	 * current (old) address/port for the calculations.  Then perform
    636  1.30    rmind 	 * the address translation i.e. rewrite source or destination.
    637  1.30    rmind 	 */
    638  1.30    rmind 	if (!npf_rwrcksum(npc, which, addr, port)) {
    639  1.30    rmind 		return EINVAL;
    640  1.30    rmind 	}
    641  1.30    rmind 	if (!npf_rwrip(npc, which, addr)) {
    642  1.30    rmind 		return EINVAL;
    643  1.30    rmind 	}
    644  1.30    rmind 	if (port == 0) {
    645  1.30    rmind 		/* Done. */
    646  1.30    rmind 		return 0;
    647  1.30    rmind 	}
    648  1.30    rmind 
    649  1.30    rmind 	switch (proto) {
    650  1.30    rmind 	case IPPROTO_TCP:
    651  1.30    rmind 	case IPPROTO_UDP:
    652  1.30    rmind 		/* Rewrite source/destination port. */
    653  1.30    rmind 		if (!npf_rwrport(npc, which, port)) {
    654  1.30    rmind 			return EINVAL;
    655  1.30    rmind 		}
    656  1.30    rmind 		break;
    657  1.30    rmind 	case IPPROTO_ICMP:
    658  1.30    rmind 	case IPPROTO_ICMPV6:
    659  1.30    rmind 		KASSERT(npf_iscached(npc, NPC_ICMP));
    660  1.30    rmind 		/* Nothing. */
    661  1.30    rmind 		break;
    662  1.30    rmind 	default:
    663  1.30    rmind 		return ENOTSUP;
    664  1.30    rmind 	}
    665  1.30    rmind 	return 0;
    666  1.30    rmind }
    667  1.30    rmind 
    668  1.30    rmind /*
    669  1.29    rmind  * IPv6-to-IPv6 Network Prefix Translation (NPTv6), as per RFC 6296.
    670  1.29    rmind  */
    671  1.29    rmind 
    672  1.29    rmind int
    673  1.29    rmind npf_npt66_rwr(const npf_cache_t *npc, u_int which, const npf_addr_t *pref,
    674  1.29    rmind     npf_netmask_t len, uint16_t adj)
    675  1.29    rmind {
    676  1.29    rmind 	npf_addr_t *addr = npc->npc_ips[which];
    677  1.29    rmind 	unsigned remnant, word, preflen = len >> 4;
    678  1.29    rmind 	uint32_t sum;
    679  1.29    rmind 
    680  1.29    rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    681  1.29    rmind 
    682  1.29    rmind 	if (!npf_iscached(npc, NPC_IP6)) {
    683  1.29    rmind 		return EINVAL;
    684  1.29    rmind 	}
    685  1.29    rmind 	if (len <= 48) {
    686  1.29    rmind 		/*
    687  1.29    rmind 		 * The word to adjust.  Cannot translate the 0xffff
    688  1.29    rmind 		 * subnet if /48 or shorter.
    689  1.29    rmind 		 */
    690  1.29    rmind 		word = 3;
    691  1.29    rmind 		if (addr->s6_addr16[word] == 0xffff) {
    692  1.29    rmind 			return EINVAL;
    693  1.29    rmind 		}
    694  1.29    rmind 	} else {
    695  1.29    rmind 		/*
    696  1.29    rmind 		 * Also, all 0s or 1s in the host part are disallowed for
    697  1.29    rmind 		 * longer than /48 prefixes.
    698  1.29    rmind 		 */
    699  1.29    rmind 		if ((addr->s6_addr32[2] == 0 && addr->s6_addr32[3] == 0) ||
    700  1.29    rmind 		    (addr->s6_addr32[2] == ~0U && addr->s6_addr32[3] == ~0U))
    701  1.29    rmind 			return EINVAL;
    702  1.29    rmind 
    703  1.29    rmind 		/* Determine the 16-bit word to adjust. */
    704  1.29    rmind 		for (word = 4; word < 8; word++)
    705  1.29    rmind 			if (addr->s6_addr16[word] != 0xffff)
    706  1.29    rmind 				break;
    707  1.29    rmind 	}
    708  1.29    rmind 
    709  1.29    rmind 	/* Rewrite the prefix. */
    710  1.29    rmind 	for (unsigned i = 0; i < preflen; i++) {
    711  1.29    rmind 		addr->s6_addr16[i] = pref->s6_addr16[i];
    712  1.29    rmind 	}
    713  1.29    rmind 
    714  1.29    rmind 	/*
    715  1.29    rmind 	 * If prefix length is within a 16-bit word (not dividable by 16),
    716  1.29    rmind 	 * then prepare a mask, determine the word and adjust it.
    717  1.29    rmind 	 */
    718  1.29    rmind 	if ((remnant = len - (preflen << 4)) != 0) {
    719  1.29    rmind 		const uint16_t wordmask = (1U << remnant) - 1;
    720  1.29    rmind 		const unsigned i = preflen;
    721  1.29    rmind 
    722  1.29    rmind 		addr->s6_addr16[i] = (pref->s6_addr16[i] & wordmask) |
    723  1.29    rmind 		    (addr->s6_addr16[i] & ~wordmask);
    724  1.29    rmind 	}
    725  1.29    rmind 
    726  1.29    rmind 	/*
    727  1.29    rmind 	 * Performing 1's complement sum/difference.
    728  1.29    rmind 	 */
    729  1.29    rmind 	sum = addr->s6_addr16[word] + adj;
    730  1.29    rmind 	while (sum >> 16) {
    731  1.29    rmind 		sum = (sum >> 16) + (sum & 0xffff);
    732  1.29    rmind 	}
    733  1.29    rmind 	if (sum == 0xffff) {
    734  1.29    rmind 		/* RFC 1071. */
    735  1.29    rmind 		sum = 0x0000;
    736  1.29    rmind 	}
    737  1.29    rmind 	addr->s6_addr16[word] = sum;
    738  1.29    rmind 	return 0;
    739  1.29    rmind }
    740  1.29    rmind 
    741  1.13    rmind #if defined(DDB) || defined(_NPF_TESTING)
    742  1.13    rmind 
    743  1.31    rmind const char *
    744  1.31    rmind npf_addr_dump(const npf_addr_t *addr, int alen)
    745  1.13    rmind {
    746  1.31    rmind 	if (alen == sizeof(struct in_addr)) {
    747  1.31    rmind 		struct in_addr ip;
    748  1.31    rmind 		memcpy(&ip, addr, alen);
    749  1.31    rmind 		return inet_ntoa(ip);
    750  1.31    rmind 	}
    751  1.33  mlelstv 	return ip6_sprintf(addr);
    752  1.13    rmind }
    753  1.13    rmind 
    754  1.13    rmind #endif
    755