Home | History | Annotate | Line # | Download | only in npf
npf_inet.c revision 1.50
      1  1.50      maxv /*	$NetBSD: npf_inet.c,v 1.50 2018/04/08 05:51:45 maxv 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  1.36  christos #ifdef _KERNEL
     42   1.1     rmind #include <sys/cdefs.h>
     43  1.50      maxv __KERNEL_RCSID(0, "$NetBSD: npf_inet.c,v 1.50 2018/04/08 05:51:45 maxv Exp $");
     44   1.1     rmind 
     45   1.1     rmind #include <sys/param.h>
     46  1.11     rmind #include <sys/types.h>
     47   1.1     rmind 
     48   1.4     rmind #include <net/pfil.h>
     49   1.4     rmind #include <net/if.h>
     50   1.4     rmind #include <net/ethertypes.h>
     51   1.4     rmind #include <net/if_ether.h>
     52   1.4     rmind 
     53   1.1     rmind #include <netinet/in_systm.h>
     54   1.1     rmind #include <netinet/in.h>
     55  1.33   mlelstv #include <netinet6/in6_var.h>
     56   1.1     rmind #include <netinet/ip.h>
     57   1.4     rmind #include <netinet/ip6.h>
     58   1.1     rmind #include <netinet/tcp.h>
     59   1.1     rmind #include <netinet/udp.h>
     60   1.1     rmind #include <netinet/ip_icmp.h>
     61  1.36  christos #endif
     62   1.1     rmind 
     63   1.1     rmind #include "npf_impl.h"
     64   1.1     rmind 
     65   1.1     rmind /*
     66  1.27     rmind  * npf_fixup{16,32}_cksum: incremental update of the Internet checksum.
     67   1.1     rmind  */
     68   1.1     rmind 
     69   1.1     rmind uint16_t
     70   1.1     rmind npf_fixup16_cksum(uint16_t cksum, uint16_t odatum, uint16_t ndatum)
     71   1.1     rmind {
     72   1.1     rmind 	uint32_t sum;
     73   1.1     rmind 
     74   1.1     rmind 	/*
     75   1.1     rmind 	 * RFC 1624:
     76   1.1     rmind 	 *	HC' = ~(~HC + ~m + m')
     77  1.27     rmind 	 *
     78  1.27     rmind 	 * Note: 1's complement sum is endian-independent (RFC 1071, page 2).
     79   1.1     rmind 	 */
     80  1.27     rmind 	sum = ~cksum & 0xffff;
     81  1.27     rmind 	sum += (~odatum & 0xffff) + ndatum;
     82   1.1     rmind 	sum = (sum >> 16) + (sum & 0xffff);
     83   1.1     rmind 	sum += (sum >> 16);
     84   1.1     rmind 
     85  1.27     rmind 	return ~sum & 0xffff;
     86   1.1     rmind }
     87   1.1     rmind 
     88   1.1     rmind uint16_t
     89   1.1     rmind npf_fixup32_cksum(uint16_t cksum, uint32_t odatum, uint32_t ndatum)
     90   1.1     rmind {
     91  1.27     rmind 	uint32_t sum;
     92  1.27     rmind 
     93  1.27     rmind 	/*
     94  1.27     rmind 	 * Checksum 32-bit datum as as two 16-bit.  Note, the first
     95  1.27     rmind 	 * 32->16 bit reduction is not necessary.
     96  1.27     rmind 	 */
     97  1.27     rmind 	sum = ~cksum & 0xffff;
     98  1.27     rmind 	sum += (~odatum & 0xffff) + (ndatum & 0xffff);
     99   1.1     rmind 
    100  1.27     rmind 	sum += (~odatum >> 16) + (ndatum >> 16);
    101  1.27     rmind 	sum = (sum >> 16) + (sum & 0xffff);
    102  1.27     rmind 	sum += (sum >> 16);
    103  1.27     rmind 	return ~sum & 0xffff;
    104   1.1     rmind }
    105   1.1     rmind 
    106   1.1     rmind /*
    107   1.4     rmind  * npf_addr_cksum: calculate checksum of the address, either IPv4 or IPv6.
    108   1.4     rmind  */
    109   1.4     rmind uint16_t
    110  1.19     rmind npf_addr_cksum(uint16_t cksum, int sz, const npf_addr_t *oaddr,
    111  1.19     rmind     const npf_addr_t *naddr)
    112   1.4     rmind {
    113  1.19     rmind 	const uint32_t *oip32 = (const uint32_t *)oaddr;
    114  1.19     rmind 	const uint32_t *nip32 = (const uint32_t *)naddr;
    115   1.4     rmind 
    116   1.4     rmind 	KASSERT(sz % sizeof(uint32_t) == 0);
    117   1.4     rmind 	do {
    118   1.4     rmind 		cksum = npf_fixup32_cksum(cksum, *oip32++, *nip32++);
    119   1.4     rmind 		sz -= sizeof(uint32_t);
    120   1.4     rmind 	} while (sz);
    121   1.4     rmind 
    122   1.4     rmind 	return cksum;
    123   1.4     rmind }
    124   1.4     rmind 
    125   1.4     rmind /*
    126  1.26     rmind  * npf_addr_sum: provide IP addresses as a XORed 32-bit integer.
    127   1.4     rmind  * Note: used for hash function.
    128   1.1     rmind  */
    129   1.4     rmind uint32_t
    130  1.26     rmind npf_addr_mix(const int sz, const npf_addr_t *a1, const npf_addr_t *a2)
    131   1.1     rmind {
    132   1.4     rmind 	uint32_t mix = 0;
    133   1.1     rmind 
    134   1.5     rmind 	KASSERT(sz > 0 && a1 != NULL && a2 != NULL);
    135   1.5     rmind 
    136  1.26     rmind 	for (int i = 0; i < (sz >> 2); i++) {
    137  1.36  christos 		mix ^= a1->word32[i];
    138  1.36  christos 		mix ^= a2->word32[i];
    139   1.4     rmind 	}
    140   1.4     rmind 	return mix;
    141   1.4     rmind }
    142   1.1     rmind 
    143  1.13     rmind /*
    144  1.13     rmind  * npf_addr_mask: apply the mask to a given address and store the result.
    145  1.13     rmind  */
    146  1.13     rmind void
    147  1.13     rmind npf_addr_mask(const npf_addr_t *addr, const npf_netmask_t mask,
    148  1.13     rmind     const int alen, npf_addr_t *out)
    149  1.12     rmind {
    150  1.13     rmind 	const int nwords = alen >> 2;
    151  1.12     rmind 	uint_fast8_t length = mask;
    152  1.12     rmind 
    153  1.12     rmind 	/* Note: maximum length is 32 for IPv4 and 128 for IPv6. */
    154  1.12     rmind 	KASSERT(length <= NPF_MAX_NETMASK);
    155  1.12     rmind 
    156  1.13     rmind 	for (int i = 0; i < nwords; i++) {
    157  1.13     rmind 		uint32_t wordmask;
    158  1.13     rmind 
    159  1.12     rmind 		if (length >= 32) {
    160  1.13     rmind 			wordmask = htonl(0xffffffff);
    161  1.12     rmind 			length -= 32;
    162  1.13     rmind 		} else if (length) {
    163  1.13     rmind 			wordmask = htonl(0xffffffff << (32 - length));
    164  1.13     rmind 			length = 0;
    165  1.12     rmind 		} else {
    166  1.13     rmind 			wordmask = 0;
    167  1.12     rmind 		}
    168  1.36  christos 		out->word32[i] = addr->word32[i] & wordmask;
    169  1.12     rmind 	}
    170  1.12     rmind }
    171  1.12     rmind 
    172  1.12     rmind /*
    173  1.12     rmind  * npf_addr_cmp: compare two addresses, either IPv4 or IPv6.
    174  1.12     rmind  *
    175  1.13     rmind  * => Return 0 if equal and negative/positive if less/greater accordingly.
    176  1.12     rmind  * => Ignore the mask, if NPF_NO_NETMASK is specified.
    177  1.12     rmind  */
    178  1.12     rmind int
    179  1.12     rmind npf_addr_cmp(const npf_addr_t *addr1, const npf_netmask_t mask1,
    180  1.13     rmind     const npf_addr_t *addr2, const npf_netmask_t mask2, const int alen)
    181  1.12     rmind {
    182  1.13     rmind 	npf_addr_t realaddr1, realaddr2;
    183  1.12     rmind 
    184  1.12     rmind 	if (mask1 != NPF_NO_NETMASK) {
    185  1.13     rmind 		npf_addr_mask(addr1, mask1, alen, &realaddr1);
    186  1.13     rmind 		addr1 = &realaddr1;
    187  1.12     rmind 	}
    188  1.12     rmind 	if (mask2 != NPF_NO_NETMASK) {
    189  1.13     rmind 		npf_addr_mask(addr2, mask2, alen, &realaddr2);
    190  1.13     rmind 		addr2 = &realaddr2;
    191  1.12     rmind 	}
    192  1.13     rmind 	return memcmp(addr1, addr2, alen);
    193  1.12     rmind }
    194  1.12     rmind 
    195   1.4     rmind /*
    196   1.4     rmind  * npf_tcpsaw: helper to fetch SEQ, ACK, WIN and return TCP data length.
    197  1.12     rmind  *
    198  1.12     rmind  * => Returns all values in host byte-order.
    199   1.4     rmind  */
    200   1.4     rmind int
    201  1.12     rmind npf_tcpsaw(const npf_cache_t *npc, tcp_seq *seq, tcp_seq *ack, uint32_t *win)
    202   1.4     rmind {
    203  1.19     rmind 	const struct tcphdr *th = npc->npc_l4.tcp;
    204   1.8     rmind 	u_int thlen;
    205   1.1     rmind 
    206   1.7    zoltan 	KASSERT(npf_iscached(npc, NPC_TCP));
    207   1.1     rmind 
    208   1.4     rmind 	*seq = ntohl(th->th_seq);
    209   1.4     rmind 	*ack = ntohl(th->th_ack);
    210   1.4     rmind 	*win = (uint32_t)ntohs(th->th_win);
    211   1.8     rmind 	thlen = th->th_off << 2;
    212   1.1     rmind 
    213   1.7    zoltan 	if (npf_iscached(npc, NPC_IP4)) {
    214  1.19     rmind 		const struct ip *ip = npc->npc_ip.v4;
    215  1.21     rmind 		return ntohs(ip->ip_len) - npc->npc_hlen - thlen;
    216  1.12     rmind 	} else if (npf_iscached(npc, NPC_IP6)) {
    217  1.19     rmind 		const struct ip6_hdr *ip6 = npc->npc_ip.v6;
    218  1.42      maxv 		return ntohs(ip6->ip6_plen) -
    219  1.42      maxv 		    (npc->npc_hlen - sizeof(*ip6)) - thlen;
    220   1.7    zoltan 	}
    221   1.7    zoltan 	return 0;
    222   1.1     rmind }
    223   1.1     rmind 
    224   1.1     rmind /*
    225   1.4     rmind  * npf_fetch_tcpopts: parse and return TCP options.
    226   1.1     rmind  */
    227   1.1     rmind bool
    228  1.32     rmind npf_fetch_tcpopts(npf_cache_t *npc, uint16_t *mss, int *wscale)
    229   1.1     rmind {
    230  1.32     rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    231  1.19     rmind 	const struct tcphdr *th = npc->npc_l4.tcp;
    232  1.49      maxv 	int cnt, optlen = 0;
    233  1.48      maxv 	bool setmss = false;
    234  1.49      maxv 	uint8_t *cp, opt;
    235   1.4     rmind 	uint8_t val;
    236  1.19     rmind 	bool ok;
    237   1.4     rmind 
    238   1.7    zoltan 	KASSERT(npf_iscached(npc, NPC_IP46));
    239   1.7    zoltan 	KASSERT(npf_iscached(npc, NPC_TCP));
    240  1.10     rmind 
    241   1.4     rmind 	/* Determine if there are any TCP options, get their length. */
    242  1.49      maxv 	cnt = (th->th_off << 2) - sizeof(struct tcphdr);
    243  1.49      maxv 	if (cnt <= 0) {
    244   1.4     rmind 		/* No options. */
    245   1.1     rmind 		return false;
    246   1.4     rmind 	}
    247  1.49      maxv 	KASSERT(cnt <= MAX_TCPOPTLEN);
    248   1.1     rmind 
    249  1.48      maxv 	/* Determine if we want to set or get the mss. */
    250  1.48      maxv 	if (mss) {
    251  1.48      maxv 		setmss = (*mss != 0);
    252  1.48      maxv 	}
    253  1.48      maxv 
    254  1.49      maxv 	/* Fetch all the options at once. */
    255  1.19     rmind 	nbuf_reset(nbuf);
    256  1.49      maxv 	const int step = npc->npc_hlen + sizeof(struct tcphdr);
    257  1.49      maxv 	if ((cp = nbuf_advance(nbuf, step, cnt)) == NULL) {
    258  1.19     rmind 		ok = false;
    259  1.19     rmind 		goto done;
    260   1.4     rmind 	}
    261  1.12     rmind 
    262  1.49      maxv 	/* Scan the options. */
    263  1.49      maxv 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
    264  1.49      maxv 		opt = cp[0];
    265  1.49      maxv 		if (opt == TCPOPT_EOL)
    266  1.49      maxv 			break;
    267  1.49      maxv 		if (opt == TCPOPT_NOP)
    268  1.49      maxv 			optlen = 1;
    269  1.49      maxv 		else {
    270  1.49      maxv 			if (cnt < 2)
    271  1.49      maxv 				break;
    272  1.49      maxv 			optlen = cp[1];
    273  1.49      maxv 			if (optlen < 2 || optlen > cnt)
    274  1.49      maxv 				break;
    275  1.49      maxv 		}
    276  1.49      maxv 
    277  1.49      maxv 		switch (opt) {
    278  1.49      maxv 		case TCPOPT_MAXSEG:
    279  1.49      maxv 			if (optlen != TCPOLEN_MAXSEG)
    280  1.49      maxv 				continue;
    281  1.49      maxv 			if (mss) {
    282  1.49      maxv 				if (setmss) {
    283  1.49      maxv 					memcpy(cp + 2, mss, sizeof(uint16_t));
    284  1.49      maxv 				} else {
    285  1.49      maxv 					memcpy(mss, cp + 2, sizeof(uint16_t));
    286  1.49      maxv 				}
    287  1.19     rmind 			}
    288  1.49      maxv 			break;
    289  1.49      maxv 		case TCPOPT_WINDOW:
    290  1.50      maxv 			if (optlen != TCPOLEN_WINDOW)
    291  1.49      maxv 				continue;
    292  1.49      maxv 			val = *(cp + 2);
    293  1.49      maxv 			*wscale = (val > TCP_MAX_WINSHIFT) ? TCP_MAX_WINSHIFT : val;
    294  1.49      maxv 			break;
    295  1.49      maxv 		default:
    296  1.49      maxv 			break;
    297   1.4     rmind 		}
    298   1.4     rmind 	}
    299  1.49      maxv 
    300  1.19     rmind 	ok = true;
    301  1.19     rmind done:
    302  1.19     rmind 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    303  1.32     rmind 		npf_recache(npc);
    304  1.19     rmind 	}
    305  1.19     rmind 	return ok;
    306   1.1     rmind }
    307   1.1     rmind 
    308  1.19     rmind static int
    309  1.19     rmind npf_cache_ip(npf_cache_t *npc, nbuf_t *nbuf)
    310   1.1     rmind {
    311  1.19     rmind 	const void *nptr = nbuf_dataptr(nbuf);
    312  1.19     rmind 	const uint8_t ver = *(const uint8_t *)nptr;
    313  1.19     rmind 	int flags = 0;
    314  1.12     rmind 
    315  1.43      maxv 	/*
    316  1.43      maxv 	 * We intentionally don't read the L4 payload after IPPROTO_AH.
    317  1.43      maxv 	 */
    318  1.43      maxv 
    319   1.4     rmind 	switch (ver >> 4) {
    320  1.12     rmind 	case IPVERSION: {
    321  1.19     rmind 		struct ip *ip;
    322  1.12     rmind 
    323  1.19     rmind 		ip = nbuf_ensure_contig(nbuf, sizeof(struct ip));
    324  1.19     rmind 		if (ip == NULL) {
    325  1.38      maxv 			return NPC_FMTERR;
    326   1.4     rmind 		}
    327  1.12     rmind 
    328  1.46      maxv 		/* Retrieve the complete header. */
    329  1.10     rmind 		if ((u_int)(ip->ip_hl << 2) < sizeof(struct ip)) {
    330  1.38      maxv 			return NPC_FMTERR;
    331   1.4     rmind 		}
    332  1.46      maxv 		ip = nbuf_ensure_contig(nbuf, (u_int)(ip->ip_hl << 2));
    333  1.46      maxv 		if (ip == NULL) {
    334  1.46      maxv 			return NPC_FMTERR;
    335  1.46      maxv 		}
    336  1.46      maxv 
    337   1.4     rmind 		if (ip->ip_off & ~htons(IP_DF | IP_RF)) {
    338   1.4     rmind 			/* Note fragmentation. */
    339  1.19     rmind 			flags |= NPC_IPFRAG;
    340   1.4     rmind 		}
    341  1.12     rmind 
    342   1.4     rmind 		/* Cache: layer 3 - IPv4. */
    343  1.14     rmind 		npc->npc_alen = sizeof(struct in_addr);
    344  1.28     rmind 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip->ip_src;
    345  1.28     rmind 		npc->npc_ips[NPF_DST] = (npf_addr_t *)&ip->ip_dst;
    346   1.7    zoltan 		npc->npc_hlen = ip->ip_hl << 2;
    347  1.19     rmind 		npc->npc_proto = ip->ip_p;
    348  1.19     rmind 
    349  1.19     rmind 		npc->npc_ip.v4 = ip;
    350  1.19     rmind 		flags |= NPC_IP4;
    351   1.4     rmind 		break;
    352  1.12     rmind 	}
    353   1.4     rmind 
    354  1.12     rmind 	case (IPV6_VERSION >> 4): {
    355  1.19     rmind 		struct ip6_hdr *ip6;
    356  1.19     rmind 		struct ip6_ext *ip6e;
    357  1.37  christos 		struct ip6_frag *ip6f;
    358  1.19     rmind 		size_t off, hlen;
    359  1.38      maxv 		int frag_present;
    360  1.19     rmind 
    361  1.19     rmind 		ip6 = nbuf_ensure_contig(nbuf, sizeof(struct ip6_hdr));
    362  1.19     rmind 		if (ip6 == NULL) {
    363  1.38      maxv 			return NPC_FMTERR;
    364   1.7    zoltan 		}
    365  1.19     rmind 
    366  1.44      maxv 		/*
    367  1.44      maxv 		 * XXX: We don't handle IPv6 Jumbograms.
    368  1.44      maxv 		 */
    369  1.44      maxv 
    370  1.19     rmind 		/* Set initial next-protocol value. */
    371  1.19     rmind 		hlen = sizeof(struct ip6_hdr);
    372  1.19     rmind 		npc->npc_proto = ip6->ip6_nxt;
    373  1.13     rmind 		npc->npc_hlen = hlen;
    374   1.7    zoltan 
    375  1.38      maxv 		frag_present = 0;
    376  1.38      maxv 
    377  1.12     rmind 		/*
    378  1.19     rmind 		 * Advance by the length of the current header.
    379  1.12     rmind 		 */
    380  1.19     rmind 		off = nbuf_offset(nbuf);
    381  1.38      maxv 		while ((ip6e = nbuf_advance(nbuf, hlen, sizeof(*ip6e))) != NULL) {
    382  1.13     rmind 			/*
    383  1.13     rmind 			 * Determine whether we are going to continue.
    384  1.13     rmind 			 */
    385  1.19     rmind 			switch (npc->npc_proto) {
    386  1.13     rmind 			case IPPROTO_HOPOPTS:
    387   1.7    zoltan 			case IPPROTO_DSTOPTS:
    388   1.7    zoltan 			case IPPROTO_ROUTING:
    389  1.19     rmind 				hlen = (ip6e->ip6e_len + 1) << 3;
    390   1.7    zoltan 				break;
    391   1.7    zoltan 			case IPPROTO_FRAGMENT:
    392  1.38      maxv 				if (frag_present++)
    393  1.38      maxv 					return NPC_FMTERR;
    394  1.37  christos 				ip6f = nbuf_ensure_contig(nbuf, sizeof(*ip6f));
    395  1.37  christos 				if (ip6f == NULL)
    396  1.38      maxv 					return NPC_FMTERR;
    397  1.38      maxv 
    398  1.41      maxv 				/* RFC6946: Skip dummy fragments. */
    399  1.41      maxv 				if (!ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK) &&
    400  1.41      maxv 				    !(ip6f->ip6f_offlg & IP6F_MORE_FRAG)) {
    401  1.41      maxv 					hlen = sizeof(struct ip6_frag);
    402  1.41      maxv 					break;
    403  1.41      maxv 				}
    404  1.41      maxv 
    405  1.40      maxv 				hlen = 0;
    406  1.40      maxv 				flags |= NPC_IPFRAG;
    407  1.37  christos 
    408   1.7    zoltan 				break;
    409   1.7    zoltan 			default:
    410  1.13     rmind 				hlen = 0;
    411  1.13     rmind 				break;
    412  1.13     rmind 			}
    413  1.13     rmind 
    414  1.13     rmind 			if (!hlen) {
    415   1.7    zoltan 				break;
    416   1.7    zoltan 			}
    417  1.19     rmind 			npc->npc_proto = ip6e->ip6e_nxt;
    418  1.13     rmind 			npc->npc_hlen += hlen;
    419  1.13     rmind 		}
    420   1.7    zoltan 
    421  1.46      maxv 		if (ip6e == NULL) {
    422  1.46      maxv 			return NPC_FMTERR;
    423  1.46      maxv 		}
    424  1.46      maxv 
    425  1.23     rmind 		/*
    426  1.23     rmind 		 * Re-fetch the header pointers (nbufs might have been
    427  1.23     rmind 		 * reallocated).  Restore the original offset (if any).
    428  1.23     rmind 		 */
    429  1.19     rmind 		nbuf_reset(nbuf);
    430  1.23     rmind 		ip6 = nbuf_dataptr(nbuf);
    431  1.19     rmind 		if (off) {
    432  1.19     rmind 			nbuf_advance(nbuf, off, 0);
    433  1.19     rmind 		}
    434  1.19     rmind 
    435  1.12     rmind 		/* Cache: layer 3 - IPv6. */
    436  1.14     rmind 		npc->npc_alen = sizeof(struct in6_addr);
    437  1.28     rmind 		npc->npc_ips[NPF_SRC] = (npf_addr_t *)&ip6->ip6_src;
    438  1.44      maxv 		npc->npc_ips[NPF_DST] = (npf_addr_t *)&ip6->ip6_dst;
    439  1.19     rmind 
    440  1.19     rmind 		npc->npc_ip.v6 = ip6;
    441  1.19     rmind 		flags |= NPC_IP6;
    442   1.7    zoltan 		break;
    443  1.12     rmind 	}
    444   1.4     rmind 	default:
    445  1.19     rmind 		break;
    446   1.4     rmind 	}
    447  1.19     rmind 	return flags;
    448   1.1     rmind }
    449   1.1     rmind 
    450   1.1     rmind /*
    451   1.4     rmind  * npf_cache_all: general routine to cache all relevant IP (v4 or v6)
    452  1.12     rmind  * and TCP, UDP or ICMP headers.
    453  1.19     rmind  *
    454  1.19     rmind  * => nbuf offset shall be set accordingly.
    455   1.1     rmind  */
    456  1.10     rmind int
    457  1.32     rmind npf_cache_all(npf_cache_t *npc)
    458   1.1     rmind {
    459  1.32     rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    460  1.19     rmind 	int flags, l4flags;
    461  1.19     rmind 	u_int hlen;
    462  1.19     rmind 
    463  1.19     rmind 	/*
    464  1.19     rmind 	 * This routine is a main point where the references are cached,
    465  1.19     rmind 	 * therefore clear the flag as we reset.
    466  1.19     rmind 	 */
    467  1.19     rmind again:
    468  1.19     rmind 	nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    469   1.1     rmind 
    470  1.19     rmind 	/*
    471  1.19     rmind 	 * First, cache the L3 header (IPv4 or IPv6).  If IP packet is
    472  1.19     rmind 	 * fragmented, then we cannot look into L4.
    473  1.19     rmind 	 */
    474  1.19     rmind 	flags = npf_cache_ip(npc, nbuf);
    475  1.38      maxv 	if ((flags & NPC_IP46) == 0 || (flags & NPC_IPFRAG) != 0 ||
    476  1.38      maxv 	    (flags & NPC_FMTERR) != 0) {
    477  1.47      maxv 		goto out;
    478   1.1     rmind 	}
    479  1.19     rmind 	hlen = npc->npc_hlen;
    480  1.19     rmind 
    481  1.45      maxv 	/*
    482  1.45      maxv 	 * Note: we guarantee that the potential "Query Id" field of the
    483  1.45      maxv 	 * ICMPv4/ICMPv6 packets is in the nbuf. This field is used in the
    484  1.45      maxv 	 * ICMP ALG.
    485  1.45      maxv 	 */
    486  1.19     rmind 	switch (npc->npc_proto) {
    487   1.1     rmind 	case IPPROTO_TCP:
    488  1.19     rmind 		/* Cache: layer 4 - TCP. */
    489  1.19     rmind 		npc->npc_l4.tcp = nbuf_advance(nbuf, hlen,
    490  1.19     rmind 		    sizeof(struct tcphdr));
    491  1.19     rmind 		l4flags = NPC_LAYER4 | NPC_TCP;
    492  1.10     rmind 		break;
    493   1.1     rmind 	case IPPROTO_UDP:
    494  1.19     rmind 		/* Cache: layer 4 - UDP. */
    495  1.19     rmind 		npc->npc_l4.udp = nbuf_advance(nbuf, hlen,
    496  1.19     rmind 		    sizeof(struct udphdr));
    497  1.19     rmind 		l4flags = NPC_LAYER4 | NPC_UDP;
    498  1.10     rmind 		break;
    499   1.1     rmind 	case IPPROTO_ICMP:
    500  1.19     rmind 		/* Cache: layer 4 - ICMPv4. */
    501  1.19     rmind 		npc->npc_l4.icmp = nbuf_advance(nbuf, hlen,
    502  1.45      maxv 		    ICMP_MINLEN);
    503  1.19     rmind 		l4flags = NPC_LAYER4 | NPC_ICMP;
    504  1.19     rmind 		break;
    505  1.15       spz 	case IPPROTO_ICMPV6:
    506  1.19     rmind 		/* Cache: layer 4 - ICMPv6. */
    507  1.19     rmind 		npc->npc_l4.icmp6 = nbuf_advance(nbuf, hlen,
    508  1.45      maxv 		    sizeof(struct icmp6_hdr));
    509  1.19     rmind 		l4flags = NPC_LAYER4 | NPC_ICMP;
    510  1.19     rmind 		break;
    511  1.19     rmind 	default:
    512  1.19     rmind 		l4flags = 0;
    513  1.10     rmind 		break;
    514   1.1     rmind 	}
    515  1.19     rmind 
    516  1.47      maxv 	/* Error out if nbuf_advance failed. */
    517  1.47      maxv 	if (l4flags && npc->npc_l4.hdr == NULL) {
    518  1.47      maxv 		goto err;
    519  1.47      maxv 	}
    520  1.47      maxv 
    521  1.19     rmind 	if (nbuf_flag_p(nbuf, NBUF_DATAREF_RESET)) {
    522  1.19     rmind 		goto again;
    523  1.19     rmind 	}
    524  1.19     rmind 
    525  1.47      maxv 	flags |= l4flags;
    526  1.47      maxv 	npc->npc_info |= flags;
    527  1.47      maxv 	return flags;
    528  1.47      maxv 
    529  1.47      maxv err:
    530  1.47      maxv 	flags = NPC_FMTERR;
    531  1.47      maxv out:
    532  1.47      maxv 	nbuf_unset_flag(nbuf, NBUF_DATAREF_RESET);
    533  1.19     rmind 	npc->npc_info |= flags;
    534  1.19     rmind 	return flags;
    535  1.19     rmind }
    536  1.19     rmind 
    537  1.19     rmind void
    538  1.32     rmind npf_recache(npf_cache_t *npc)
    539  1.19     rmind {
    540  1.32     rmind 	nbuf_t *nbuf = npc->npc_nbuf;
    541  1.24    martin 	const int mflags __diagused = npc->npc_info & (NPC_IP46 | NPC_LAYER4);
    542  1.25       mrg 	int flags __diagused;
    543  1.19     rmind 
    544  1.19     rmind 	nbuf_reset(nbuf);
    545  1.19     rmind 	npc->npc_info = 0;
    546  1.32     rmind 	flags = npf_cache_all(npc);
    547  1.32     rmind 
    548  1.19     rmind 	KASSERT((flags & mflags) == mflags);
    549  1.19     rmind 	KASSERT(nbuf_flag_p(nbuf, NBUF_DATAREF_RESET) == 0);
    550   1.1     rmind }
    551   1.1     rmind 
    552   1.1     rmind /*
    553  1.19     rmind  * npf_rwrip: rewrite required IP address.
    554   1.4     rmind  */
    555   1.4     rmind bool
    556  1.28     rmind npf_rwrip(const npf_cache_t *npc, u_int which, const npf_addr_t *addr)
    557   1.4     rmind {
    558   1.4     rmind 	KASSERT(npf_iscached(npc, NPC_IP46));
    559  1.28     rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    560   1.4     rmind 
    561  1.28     rmind 	memcpy(npc->npc_ips[which], addr, npc->npc_alen);
    562   1.4     rmind 	return true;
    563   1.4     rmind }
    564   1.4     rmind 
    565   1.4     rmind /*
    566  1.19     rmind  * npf_rwrport: rewrite required TCP/UDP port.
    567   1.1     rmind  */
    568   1.1     rmind bool
    569  1.28     rmind npf_rwrport(const npf_cache_t *npc, u_int which, const in_port_t port)
    570   1.1     rmind {
    571  1.21     rmind 	const int proto = npc->npc_proto;
    572   1.4     rmind 	in_port_t *oport;
    573   1.1     rmind 
    574   1.4     rmind 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    575   1.1     rmind 	KASSERT(proto == IPPROTO_TCP || proto == IPPROTO_UDP);
    576  1.28     rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    577   1.1     rmind 
    578  1.19     rmind 	/* Get the offset and store the port in it. */
    579   1.4     rmind 	if (proto == IPPROTO_TCP) {
    580  1.19     rmind 		struct tcphdr *th = npc->npc_l4.tcp;
    581  1.28     rmind 		oport = (which == NPF_SRC) ? &th->th_sport : &th->th_dport;
    582   1.1     rmind 	} else {
    583  1.19     rmind 		struct udphdr *uh = npc->npc_l4.udp;
    584  1.28     rmind 		oport = (which == NPF_SRC) ? &uh->uh_sport : &uh->uh_dport;
    585   1.1     rmind 	}
    586  1.19     rmind 	memcpy(oport, &port, sizeof(in_port_t));
    587   1.1     rmind 	return true;
    588   1.1     rmind }
    589   1.1     rmind 
    590   1.1     rmind /*
    591  1.19     rmind  * npf_rwrcksum: rewrite IPv4 and/or TCP/UDP checksum.
    592   1.1     rmind  */
    593   1.1     rmind bool
    594  1.28     rmind npf_rwrcksum(const npf_cache_t *npc, u_int which,
    595  1.19     rmind     const npf_addr_t *addr, const in_port_t port)
    596   1.1     rmind {
    597  1.28     rmind 	const npf_addr_t *oaddr = npc->npc_ips[which];
    598  1.21     rmind 	const int proto = npc->npc_proto;
    599  1.19     rmind 	const int alen = npc->npc_alen;
    600  1.18     rmind 	uint16_t *ocksum;
    601  1.18     rmind 	in_port_t oport;
    602  1.18     rmind 
    603  1.19     rmind 	KASSERT(npf_iscached(npc, NPC_LAYER4));
    604  1.28     rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    605  1.18     rmind 
    606   1.4     rmind 	if (npf_iscached(npc, NPC_IP4)) {
    607  1.19     rmind 		struct ip *ip = npc->npc_ip.v4;
    608  1.19     rmind 		uint16_t ipsum = ip->ip_sum;
    609   1.4     rmind 
    610  1.19     rmind 		/* Recalculate IPv4 checksum and rewrite. */
    611  1.19     rmind 		ip->ip_sum = npf_addr_cksum(ipsum, alen, oaddr, addr);
    612   1.4     rmind 	} else {
    613   1.4     rmind 		/* No checksum for IPv6. */
    614   1.4     rmind 		KASSERT(npf_iscached(npc, NPC_IP6));
    615   1.4     rmind 	}
    616   1.4     rmind 
    617  1.18     rmind 	/* Nothing else to do for ICMP. */
    618  1.30     rmind 	if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
    619   1.4     rmind 		return true;
    620   1.4     rmind 	}
    621   1.7    zoltan 	KASSERT(npf_iscached(npc, NPC_TCP) || npf_iscached(npc, NPC_UDP));
    622   1.4     rmind 
    623  1.18     rmind 	/*
    624  1.18     rmind 	 * Calculate TCP/UDP checksum:
    625  1.18     rmind 	 * - Skip if UDP and the current checksum is zero.
    626  1.18     rmind 	 * - Fixup the IP address change.
    627  1.18     rmind 	 * - Fixup the port change, if required (non-zero).
    628  1.18     rmind 	 */
    629   1.4     rmind 	if (proto == IPPROTO_TCP) {
    630  1.19     rmind 		struct tcphdr *th = npc->npc_l4.tcp;
    631   1.4     rmind 
    632  1.18     rmind 		ocksum = &th->th_sum;
    633  1.28     rmind 		oport = (which == NPF_SRC) ? th->th_sport : th->th_dport;
    634   1.4     rmind 	} else {
    635  1.19     rmind 		struct udphdr *uh = npc->npc_l4.udp;
    636   1.4     rmind 
    637   1.4     rmind 		KASSERT(proto == IPPROTO_UDP);
    638  1.18     rmind 		ocksum = &uh->uh_sum;
    639  1.18     rmind 		if (*ocksum == 0) {
    640   1.4     rmind 			/* No need to update. */
    641   1.4     rmind 			return true;
    642   1.4     rmind 		}
    643  1.28     rmind 		oport = (which == NPF_SRC) ? uh->uh_sport : uh->uh_dport;
    644  1.18     rmind 	}
    645  1.18     rmind 
    646  1.19     rmind 	uint16_t cksum = npf_addr_cksum(*ocksum, alen, oaddr, addr);
    647  1.18     rmind 	if (port) {
    648  1.18     rmind 		cksum = npf_fixup16_cksum(cksum, oport, port);
    649   1.4     rmind 	}
    650   1.1     rmind 
    651  1.19     rmind 	/* Rewrite TCP/UDP checksum. */
    652  1.19     rmind 	memcpy(ocksum, &cksum, sizeof(uint16_t));
    653   1.4     rmind 	return true;
    654   1.4     rmind }
    655   1.4     rmind 
    656  1.29     rmind /*
    657  1.30     rmind  * npf_napt_rwr: perform address and/or port translation.
    658  1.30     rmind  */
    659  1.30     rmind int
    660  1.30     rmind npf_napt_rwr(const npf_cache_t *npc, u_int which,
    661  1.30     rmind     const npf_addr_t *addr, const in_addr_t port)
    662  1.30     rmind {
    663  1.30     rmind 	const unsigned proto = npc->npc_proto;
    664  1.30     rmind 
    665  1.30     rmind 	/*
    666  1.30     rmind 	 * Rewrite IP and/or TCP/UDP checksums first, since we need the
    667  1.30     rmind 	 * current (old) address/port for the calculations.  Then perform
    668  1.30     rmind 	 * the address translation i.e. rewrite source or destination.
    669  1.30     rmind 	 */
    670  1.30     rmind 	if (!npf_rwrcksum(npc, which, addr, port)) {
    671  1.30     rmind 		return EINVAL;
    672  1.30     rmind 	}
    673  1.30     rmind 	if (!npf_rwrip(npc, which, addr)) {
    674  1.30     rmind 		return EINVAL;
    675  1.30     rmind 	}
    676  1.30     rmind 	if (port == 0) {
    677  1.30     rmind 		/* Done. */
    678  1.30     rmind 		return 0;
    679  1.30     rmind 	}
    680  1.30     rmind 
    681  1.30     rmind 	switch (proto) {
    682  1.30     rmind 	case IPPROTO_TCP:
    683  1.30     rmind 	case IPPROTO_UDP:
    684  1.30     rmind 		/* Rewrite source/destination port. */
    685  1.30     rmind 		if (!npf_rwrport(npc, which, port)) {
    686  1.30     rmind 			return EINVAL;
    687  1.30     rmind 		}
    688  1.30     rmind 		break;
    689  1.30     rmind 	case IPPROTO_ICMP:
    690  1.30     rmind 	case IPPROTO_ICMPV6:
    691  1.30     rmind 		KASSERT(npf_iscached(npc, NPC_ICMP));
    692  1.30     rmind 		/* Nothing. */
    693  1.30     rmind 		break;
    694  1.30     rmind 	default:
    695  1.30     rmind 		return ENOTSUP;
    696  1.30     rmind 	}
    697  1.30     rmind 	return 0;
    698  1.30     rmind }
    699  1.30     rmind 
    700  1.30     rmind /*
    701  1.29     rmind  * IPv6-to-IPv6 Network Prefix Translation (NPTv6), as per RFC 6296.
    702  1.29     rmind  */
    703  1.29     rmind 
    704  1.29     rmind int
    705  1.29     rmind npf_npt66_rwr(const npf_cache_t *npc, u_int which, const npf_addr_t *pref,
    706  1.29     rmind     npf_netmask_t len, uint16_t adj)
    707  1.29     rmind {
    708  1.29     rmind 	npf_addr_t *addr = npc->npc_ips[which];
    709  1.29     rmind 	unsigned remnant, word, preflen = len >> 4;
    710  1.29     rmind 	uint32_t sum;
    711  1.29     rmind 
    712  1.29     rmind 	KASSERT(which == NPF_SRC || which == NPF_DST);
    713  1.29     rmind 
    714  1.29     rmind 	if (!npf_iscached(npc, NPC_IP6)) {
    715  1.29     rmind 		return EINVAL;
    716  1.29     rmind 	}
    717  1.29     rmind 	if (len <= 48) {
    718  1.29     rmind 		/*
    719  1.29     rmind 		 * The word to adjust.  Cannot translate the 0xffff
    720  1.29     rmind 		 * subnet if /48 or shorter.
    721  1.29     rmind 		 */
    722  1.29     rmind 		word = 3;
    723  1.36  christos 		if (addr->word16[word] == 0xffff) {
    724  1.29     rmind 			return EINVAL;
    725  1.29     rmind 		}
    726  1.29     rmind 	} else {
    727  1.29     rmind 		/*
    728  1.29     rmind 		 * Also, all 0s or 1s in the host part are disallowed for
    729  1.29     rmind 		 * longer than /48 prefixes.
    730  1.29     rmind 		 */
    731  1.36  christos 		if ((addr->word32[2] == 0 && addr->word32[3] == 0) ||
    732  1.36  christos 		    (addr->word32[2] == ~0U && addr->word32[3] == ~0U))
    733  1.29     rmind 			return EINVAL;
    734  1.29     rmind 
    735  1.29     rmind 		/* Determine the 16-bit word to adjust. */
    736  1.29     rmind 		for (word = 4; word < 8; word++)
    737  1.36  christos 			if (addr->word16[word] != 0xffff)
    738  1.29     rmind 				break;
    739  1.29     rmind 	}
    740  1.29     rmind 
    741  1.29     rmind 	/* Rewrite the prefix. */
    742  1.29     rmind 	for (unsigned i = 0; i < preflen; i++) {
    743  1.36  christos 		addr->word16[i] = pref->word16[i];
    744  1.29     rmind 	}
    745  1.29     rmind 
    746  1.29     rmind 	/*
    747  1.29     rmind 	 * If prefix length is within a 16-bit word (not dividable by 16),
    748  1.29     rmind 	 * then prepare a mask, determine the word and adjust it.
    749  1.29     rmind 	 */
    750  1.29     rmind 	if ((remnant = len - (preflen << 4)) != 0) {
    751  1.29     rmind 		const uint16_t wordmask = (1U << remnant) - 1;
    752  1.29     rmind 		const unsigned i = preflen;
    753  1.29     rmind 
    754  1.36  christos 		addr->word16[i] = (pref->word16[i] & wordmask) |
    755  1.36  christos 		    (addr->word16[i] & ~wordmask);
    756  1.29     rmind 	}
    757  1.29     rmind 
    758  1.29     rmind 	/*
    759  1.29     rmind 	 * Performing 1's complement sum/difference.
    760  1.29     rmind 	 */
    761  1.36  christos 	sum = addr->word16[word] + adj;
    762  1.29     rmind 	while (sum >> 16) {
    763  1.29     rmind 		sum = (sum >> 16) + (sum & 0xffff);
    764  1.29     rmind 	}
    765  1.29     rmind 	if (sum == 0xffff) {
    766  1.29     rmind 		/* RFC 1071. */
    767  1.29     rmind 		sum = 0x0000;
    768  1.29     rmind 	}
    769  1.36  christos 	addr->word16[word] = sum;
    770  1.29     rmind 	return 0;
    771  1.29     rmind }
    772  1.29     rmind 
    773  1.13     rmind #if defined(DDB) || defined(_NPF_TESTING)
    774  1.13     rmind 
    775  1.31     rmind const char *
    776  1.31     rmind npf_addr_dump(const npf_addr_t *addr, int alen)
    777  1.13     rmind {
    778  1.31     rmind 	if (alen == sizeof(struct in_addr)) {
    779  1.31     rmind 		struct in_addr ip;
    780  1.31     rmind 		memcpy(&ip, addr, alen);
    781  1.31     rmind 		return inet_ntoa(ip);
    782  1.31     rmind 	}
    783  1.36  christos 	return "[IPv6]";
    784  1.13     rmind }
    785  1.13     rmind 
    786  1.13     rmind #endif
    787