Home | History | Annotate | Line # | Download | only in npf
      1 /*-
      2  * Copyright (c) 2010 The NetBSD Foundation, Inc.
      3  * All rights reserved.
      4  *
      5  * This material is based upon work partially supported by The
      6  * NetBSD Foundation under a contract with Mindaugas Rasiukevicius.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  *
     17  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     18  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     19  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     20  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     21  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     22  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     23  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     24  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     25  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     26  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     27  * POSSIBILITY OF SUCH DAMAGE.
     28  */
     29 
     30 /*
     31  * NPF ALG for ICMP and traceroute translations.
     32  */
     33 
     34 #ifdef _KERNEL
     35 #include <sys/cdefs.h>
     36 __KERNEL_RCSID(0, "$NetBSD: npf_alg_icmp.c,v 1.33 2020/05/30 14:16:56 rmind Exp $");
     37 
     38 #include <sys/param.h>
     39 #include <sys/module.h>
     40 
     41 #include <netinet/in_systm.h>
     42 #include <netinet/in.h>
     43 #include <netinet/ip.h>
     44 #include <netinet/tcp.h>
     45 #include <netinet/udp.h>
     46 #include <netinet/ip_icmp.h>
     47 #include <netinet/icmp6.h>
     48 #include <net/pfil.h>
     49 #endif
     50 
     51 #include "npf_impl.h"
     52 #include "npf_conn.h"
     53 
     54 MODULE(MODULE_CLASS_MISC, npf_alg_icmp, "npf");
     55 
     56 /*
     57  * Traceroute criteria.
     58  *
     59  * IANA assigned base port: 33434.  However, common practice is to increase
     60  * the port, thus monitor [33434-33484] range.  Additional filter is low TTL.
     61  */
     62 
     63 #define	TR_BASE_PORT	33434
     64 #define	TR_PORT_RANGE	33484
     65 #define	TR_MAX_TTL	48
     66 
     67 static npf_alg_t *	alg_icmp	__read_mostly;
     68 
     69 /*
     70  * npfa_icmp_match: matching inspector determines ALG case and associates
     71  * our ALG with the NAT entry.
     72  */
     73 static bool
     74 npfa_icmp_match(npf_cache_t *npc, npf_nat_t *nt, int di)
     75 {
     76 	const int proto = npc->npc_proto;
     77 	const struct ip *ip = npc->npc_ip.v4;
     78 	in_port_t dport;
     79 
     80 	KASSERT(npf_iscached(npc, NPC_IP46));
     81 	KASSERT(npf_iscached(npc, NPC_LAYER4));
     82 
     83 	/* Check for low TTL.  Also, we support outbound NAT only. */
     84 	if (ip->ip_ttl > TR_MAX_TTL || di != PFIL_OUT) {
     85 		return false;
     86 	}
     87 
     88 	switch (proto) {
     89 	case IPPROTO_TCP: {
     90 		const struct tcphdr *th = npc->npc_l4.tcp;
     91 		dport = ntohs(th->th_dport);
     92 		break;
     93 	}
     94 	case IPPROTO_UDP: {
     95 		const struct udphdr *uh = npc->npc_l4.udp;
     96 		dport = ntohs(uh->uh_dport);
     97 		break;
     98 	}
     99 	case IPPROTO_ICMP:
    100 	case IPPROTO_ICMPV6:
    101 		/* Just to pass the test below. */
    102 		dport = TR_BASE_PORT;
    103 		break;
    104 	default:
    105 		return false;
    106 	}
    107 
    108 	/* Handle TCP/UDP traceroute - check for port range. */
    109 	if (dport < TR_BASE_PORT || dport > TR_PORT_RANGE) {
    110 		return false;
    111 	}
    112 
    113 	/* Associate ALG with translation entry. */
    114 	npf_nat_setalg(nt, alg_icmp, 0);
    115 	return true;
    116 }
    117 
    118 /*
    119  * npfa_icmp{4,6}_inspect: retrieve unique identifiers - either ICMP query
    120  * ID or TCP/UDP ports of the original packet, which is embedded.
    121  *
    122  * => Sets hasqid=true if the packet has a Query Id. In this case neither
    123  *    the nbuf nor npc is touched.
    124  */
    125 
    126 static bool
    127 npfa_icmp4_inspect(const int type, npf_cache_t *npc, bool *hasqid)
    128 {
    129 	nbuf_t *nbuf = npc->npc_nbuf;
    130 
    131 	/* Per RFC 792. */
    132 	switch (type) {
    133 	case ICMP_UNREACH:
    134 	case ICMP_SOURCEQUENCH:
    135 	case ICMP_REDIRECT:
    136 	case ICMP_TIMXCEED:
    137 	case ICMP_PARAMPROB:
    138 		/* Should contain original IP header. */
    139 		if (!nbuf_advance(nbuf, offsetof(struct icmp, icmp_ip), 0)) {
    140 			return false;
    141 		}
    142 		return (npf_cache_all(npc) & NPC_LAYER4) != 0;
    143 
    144 	case ICMP_ECHOREPLY:
    145 	case ICMP_ECHO:
    146 	case ICMP_TSTAMP:
    147 	case ICMP_TSTAMPREPLY:
    148 	case ICMP_IREQ:
    149 	case ICMP_IREQREPLY:
    150 		/* Contains ICMP query ID. */
    151 		*hasqid = true;
    152 		return true;
    153 	default:
    154 		break;
    155 	}
    156 	return false;
    157 }
    158 
    159 static bool
    160 npfa_icmp6_inspect(const int type, npf_cache_t *npc, bool *hasqid)
    161 {
    162 	nbuf_t *nbuf = npc->npc_nbuf;
    163 
    164 	/* Per RFC 4443. */
    165 	switch (type) {
    166 	case ICMP6_DST_UNREACH:
    167 	case ICMP6_PACKET_TOO_BIG:
    168 	case ICMP6_TIME_EXCEEDED:
    169 	case ICMP6_PARAM_PROB:
    170 		/* Should contain original IP header. */
    171 		if (!nbuf_advance(nbuf, sizeof(struct icmp6_hdr), 0)) {
    172 			return false;
    173 		}
    174 		return (npf_cache_all(npc) & NPC_LAYER4) != 0;
    175 
    176 	case ICMP6_ECHO_REQUEST:
    177 	case ICMP6_ECHO_REPLY:
    178 		/* Contains ICMP query ID. */
    179 		*hasqid = true;
    180 		return true;
    181 	default:
    182 		break;
    183 	}
    184 	return false;
    185 }
    186 
    187 /*
    188  * npfa_icmp_inspect: ALG ICMP inspector.
    189  *
    190  * => Returns false if there is a problem with the format.
    191  */
    192 static bool
    193 npfa_icmp_inspect(npf_cache_t *npc, npf_cache_t *enpc)
    194 {
    195 	nbuf_t *nbuf = npc->npc_nbuf;
    196 	bool ret, hasqid = false;
    197 
    198 	KASSERT(npf_iscached(npc, NPC_IP46));
    199 	KASSERT(npf_iscached(npc, NPC_ICMP));
    200 
    201 	/* Advance to ICMP header. */
    202 	nbuf_reset(nbuf);
    203 	if (!nbuf_advance(nbuf, npc->npc_hlen, 0)) {
    204 		return false;
    205 	}
    206 	memset(enpc, 0, sizeof(npf_cache_t));
    207 	enpc->npc_ctx = npc->npc_ctx;
    208 	enpc->npc_nbuf = nbuf;
    209 
    210 	/*
    211 	 * Inspect the ICMP packet.  The relevant data might be in the
    212 	 * embedded packet.  Fill the "enpc" cache, if so.
    213 	 */
    214 	if (npf_iscached(npc, NPC_IP4) &&
    215 	    npc->npc_proto == IPPROTO_ICMP) {
    216 		const struct icmp *ic = npc->npc_l4.icmp;
    217 		ret = npfa_icmp4_inspect(ic->icmp_type, enpc, &hasqid);
    218 	} else if (npf_iscached(npc, NPC_IP6) &&
    219 	    npc->npc_proto == IPPROTO_ICMPV6) {
    220 		const struct icmp6_hdr *ic6 = npc->npc_l4.icmp6;
    221 		ret = npfa_icmp6_inspect(ic6->icmp6_type, enpc, &hasqid);
    222 	} else {
    223 		ret = false;
    224 	}
    225 	if (!ret) {
    226 		return false;
    227 	}
    228 
    229 	/* ICMP ID is the original packet, just indicate it. */
    230 	if (hasqid) {
    231 		npc->npc_info |= NPC_ICMP_ID;
    232 	}
    233 
    234 	return true;
    235 }
    236 
    237 static npf_conn_t *
    238 npfa_icmp_conn(npf_cache_t *npc, int di)
    239 {
    240 	npf_conn_t *conn = NULL;
    241 	npf_cache_t enpc;
    242 	bool hasqid = false;
    243 
    244 	/* Inspect ICMP packet for an embedded packet. */
    245 	if (!npf_iscached(npc, NPC_ICMP))
    246 		return NULL;
    247 	if (!npfa_icmp_inspect(npc, &enpc))
    248 		goto out;
    249 
    250 	/*
    251 	 * If the ICMP packet had a Query Id, leave now. The packet didn't get
    252 	 * modified, so no need to recache npc.
    253 	 */
    254 	if (npf_iscached(npc, NPC_ICMP_ID)) {
    255 		KASSERT(!nbuf_flag_p(npc->npc_nbuf, NBUF_DATAREF_RESET));
    256 		return NULL;
    257 	}
    258 
    259 	/*
    260 	 * Invert the identifiers of the embedded packet.
    261 	 * If it is ICMP, then ensure ICMP ID.
    262 	 */
    263 	union l4 {
    264 		struct tcphdr th;
    265 		struct udphdr uh;
    266 	} l4;
    267 	npf_flow_t flow;
    268 	bool ret;
    269 
    270 	#define	SWAP(type, x, y) { type tmp = x; x = y; y = tmp; }
    271 	SWAP(npf_addr_t *, enpc.npc_ips[NPF_SRC], enpc.npc_ips[NPF_DST]);
    272 
    273 	switch (enpc.npc_proto) {
    274 	case IPPROTO_TCP:
    275 		l4.th.th_sport = enpc.npc_l4.tcp->th_dport;
    276 		l4.th.th_dport = enpc.npc_l4.tcp->th_sport;
    277 		enpc.npc_l4.tcp = &l4.th;
    278 		break;
    279 	case IPPROTO_UDP:
    280 		l4.uh.uh_sport = enpc.npc_l4.udp->uh_dport;
    281 		l4.uh.uh_dport = enpc.npc_l4.udp->uh_sport;
    282 		enpc.npc_l4.udp = &l4.uh;
    283 		break;
    284 	case IPPROTO_ICMP: {
    285 		const struct icmp *ic = enpc.npc_l4.icmp;
    286 		ret = npfa_icmp4_inspect(ic->icmp_type, &enpc, &hasqid);
    287 		if (!ret || !hasqid)
    288 			goto out;
    289 		enpc.npc_info |= NPC_ICMP_ID;
    290 		break;
    291 	}
    292 	case IPPROTO_ICMPV6: {
    293 		const struct icmp6_hdr *ic6 = enpc.npc_l4.icmp6;
    294 		ret = npfa_icmp6_inspect(ic6->icmp6_type, &enpc, &hasqid);
    295 		if (!ret || !hasqid)
    296 			goto out;
    297 		enpc.npc_info |= NPC_ICMP_ID;
    298 		break;
    299 	}
    300 	default:
    301 		goto out;
    302 	}
    303 
    304 	/* Lookup a connection using the embedded packet. */
    305 	conn = npf_conn_lookup(&enpc, di, &flow);
    306 out:
    307 	/*
    308 	 * Recache npc. The nbuf may have been updated as a result of
    309 	 * caching enpc.
    310 	 */
    311 	npf_recache(npc);
    312 	return conn;
    313 }
    314 
    315 /*
    316  * npfa_icmp_nat: ALG translator - rewrites IP address in the IP header
    317  * which is embedded in ICMP packet.  Note: backwards stream only.
    318  */
    319 static bool
    320 npfa_icmp_nat(npf_cache_t *npc, npf_nat_t *nt, npf_flow_t flow)
    321 {
    322 	const unsigned which = NPF_SRC;
    323 	npf_cache_t enpc;
    324 	struct icmp *ic;
    325 	uint16_t cksum;
    326 
    327 	if (flow == NPF_FLOW_FORW || !npf_iscached(npc, NPC_ICMP))
    328 		return false;
    329 
    330 	/*
    331 	 * ICMP: fetch the current checksum we are going to fixup.
    332 	 */
    333 	ic = npc->npc_l4.icmp;
    334 	cksum = ic->icmp_cksum;
    335 
    336 	if (!npfa_icmp_inspect(npc, &enpc))
    337 		goto err;
    338 
    339 	/*
    340 	 * If the ICMP packet had a Query Id, leave now. The packet didn't get
    341 	 * modified, so no need to recache npc.
    342 	 */
    343 	if (npf_iscached(npc, NPC_ICMP_ID)) {
    344 		KASSERT(!nbuf_flag_p(npc->npc_nbuf, NBUF_DATAREF_RESET));
    345 		return false;
    346 	}
    347 
    348 	KASSERT(npf_iscached(&enpc, NPC_IP46));
    349 	KASSERT(npf_iscached(&enpc, NPC_LAYER4));
    350 
    351 	CTASSERT(offsetof(struct icmp, icmp_cksum) ==
    352 	    offsetof(struct icmp6_hdr, icmp6_cksum));
    353 
    354 	/*
    355 	 * Fetch the IP and port in the _embedded_ packet.  Also, fetch
    356 	 * the IPv4 and TCP/UDP checksums before they are rewritten.
    357 	 */
    358 	const int proto = enpc.npc_proto;
    359 	uint16_t ipcksum = 0, l4cksum = 0;
    360 	in_port_t old_port = 0;
    361 
    362 	if (npf_iscached(&enpc, NPC_IP4)) {
    363 		const struct ip *eip = enpc.npc_ip.v4;
    364 		ipcksum = eip->ip_sum;
    365 	}
    366 	switch (proto) {
    367 	case IPPROTO_TCP: {
    368 		const struct tcphdr *th = enpc.npc_l4.tcp;
    369 		old_port = th->th_sport;
    370 		l4cksum = th->th_sum;
    371 		break;
    372 	}
    373 	case IPPROTO_UDP: {
    374 		const struct udphdr *uh = enpc.npc_l4.udp;
    375 		old_port = uh->uh_sport;
    376 		l4cksum = uh->uh_sum;
    377 		break;
    378 	}
    379 	case IPPROTO_ICMP:
    380 	case IPPROTO_ICMPV6:
    381 		break;
    382 	default:
    383 		goto err;
    384 	}
    385 
    386 	/*
    387 	 * Get the original IP address and port.
    388 	 * Calculate the part of the ICMP checksum fixup.
    389 	 */
    390 	npf_addr_t *addr;
    391 	in_port_t port;
    392 
    393 	npf_nat_getorig(nt, &addr, &port);
    394 
    395 	cksum = npf_addr_cksum(cksum, enpc.npc_alen, enpc.npc_ips[which], addr);
    396 	if (port) {
    397 		cksum = npf_fixup16_cksum(cksum, old_port, port);
    398 	}
    399 
    400 	/*
    401 	 * Translate the embedded packet.  The following changes will
    402 	 * be performed by npf_napt_rwr():
    403 	 *
    404 	 *	1) Rewrite the IP address and, if not ICMP, port.
    405 	 *	2) Rewrite the TCP/UDP checksum (if not ICMP).
    406 	 *	3) Rewrite the IPv4 checksum for (1) and (2).
    407 	 *
    408 	 * XXX: Assumes NPF_NATOUT (source address/port).  Currently,
    409 	 * npfa_icmp_match() matches only for the PFIL_OUT traffic.
    410 	 */
    411 	if (npf_napt_rwr(&enpc, which, addr, port)) {
    412 		goto err;
    413 	}
    414 
    415 	/*
    416 	 * Finally, finish the ICMP checksum fixup: include the checksum
    417 	 * changes in the embedded packet.
    418 	 */
    419 	if (npf_iscached(&enpc, NPC_IP4)) {
    420 		const struct ip *eip = enpc.npc_ip.v4;
    421 		cksum = npf_fixup16_cksum(cksum, ipcksum, eip->ip_sum);
    422 	}
    423 	switch (proto) {
    424 	case IPPROTO_TCP: {
    425 		const struct tcphdr *th = enpc.npc_l4.tcp;
    426 		cksum = npf_fixup16_cksum(cksum, l4cksum, th->th_sum);
    427 		break;
    428 	}
    429 	case IPPROTO_UDP:
    430 		if (l4cksum) {
    431 			const struct udphdr *uh = enpc.npc_l4.udp;
    432 			cksum = npf_fixup16_cksum(cksum, l4cksum, uh->uh_sum);
    433 		}
    434 		break;
    435 	}
    436 	npf_recache(npc);
    437 	KASSERT(npf_iscached(npc, NPC_ICMP));
    438 	ic = npc->npc_l4.icmp;
    439 	ic->icmp_cksum = cksum;
    440 	return true;
    441 
    442 err:
    443 	/*
    444 	 * Recache npc. The nbuf may have been updated as a result of
    445 	 * caching enpc.
    446 	 */
    447 	npf_recache(npc);
    448 	return false;
    449 }
    450 
    451 /*
    452  * npf_alg_icmp_{init,fini,modcmd}: ICMP ALG initialization, destruction
    453  * and module interface.
    454  */
    455 
    456 __dso_public int
    457 npf_alg_icmp_init(npf_t *npf)
    458 {
    459 	static const npfa_funcs_t icmp = {
    460 		.match		= npfa_icmp_match,
    461 		.translate	= npfa_icmp_nat,
    462 		.inspect	= npfa_icmp_conn,
    463 		.destroy	= NULL,
    464 	};
    465 	alg_icmp = npf_alg_register(npf, "icmp", &icmp);
    466 	return alg_icmp ? 0 : ENOMEM;
    467 }
    468 
    469 __dso_public int
    470 npf_alg_icmp_fini(npf_t *npf)
    471 {
    472 	KASSERT(alg_icmp != NULL);
    473 	return npf_alg_unregister(npf, alg_icmp);
    474 }
    475 
    476 #ifdef _KERNEL
    477 static int
    478 npf_alg_icmp_modcmd(modcmd_t cmd, void *arg)
    479 {
    480 	npf_t *npf = npf_getkernctx();
    481 
    482 	switch (cmd) {
    483 	case MODULE_CMD_INIT:
    484 		return npf_alg_icmp_init(npf);
    485 	case MODULE_CMD_FINI:
    486 		return npf_alg_icmp_fini(npf);
    487 	case MODULE_CMD_AUTOUNLOAD:
    488 		return EBUSY;
    489 	default:
    490 		return ENOTTY;
    491 	}
    492 	return 0;
    493 }
    494 #endif
    495