xref: /freebsd/sys/netinet/in_gif.c (revision 5bb3134a8c21cb87b30e135ef168483f0333dabb) 
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * Copyright (c) 2018 Andrey V. Elsukov <ae@FreeBSD.org>
6  * All rights reserved.
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  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	$KAME: in_gif.c,v 1.54 2001/05/14 14:02:16 itojun Exp $
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include "opt_inet.h"
39 #include "opt_inet6.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/jail.h>
44 #include <sys/socket.h>
45 #include <sys/sockio.h>
46 #include <sys/mbuf.h>
47 #include <sys/errno.h>
48 #include <sys/kernel.h>
49 #include <sys/sysctl.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 
53 #include <net/ethernet.h>
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/route.h>
57 #include <net/vnet.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/in_systm.h>
61 #include <netinet/ip.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip_encap.h>
65 #include <netinet/ip_ecn.h>
66 #include <netinet/in_fib.h>
67 
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #endif
71 
72 #include <net/if_gif.h>
73 
74 #define GIF_TTL		30
75 VNET_DEFINE_STATIC(int, ip_gif_ttl) = GIF_TTL;
76 #define	V_ip_gif_ttl		VNET(ip_gif_ttl)
77 SYSCTL_INT(_net_inet_ip, IPCTL_GIF_TTL, gifttl, CTLFLAG_VNET | CTLFLAG_RW,
78     &VNET_NAME(ip_gif_ttl), 0, "Default TTL value for encapsulated packets");
79 
80 /*
81  * We keep interfaces in a hash table using src+dst as key.
82  * Interfaces with GIF_IGNORE_SOURCE flag are linked into plain list.
83  */
84 VNET_DEFINE_STATIC(struct gif_list *, ipv4_hashtbl) = NULL;
85 VNET_DEFINE_STATIC(struct gif_list *, ipv4_srchashtbl) = NULL;
86 VNET_DEFINE_STATIC(struct gif_list, ipv4_list) = CK_LIST_HEAD_INITIALIZER();
87 #define	V_ipv4_hashtbl		VNET(ipv4_hashtbl)
88 #define	V_ipv4_srchashtbl	VNET(ipv4_srchashtbl)
89 #define	V_ipv4_list		VNET(ipv4_list)
90 
91 #define	GIF_HASH(src, dst)	(V_ipv4_hashtbl[\
92     in_gif_hashval((src), (dst)) & (GIF_HASH_SIZE - 1)])
93 #define	GIF_SRCHASH(src)	(V_ipv4_srchashtbl[\
94     fnv_32_buf(&(src), sizeof(src), FNV1_32_INIT) & (GIF_HASH_SIZE - 1)])
95 #define	GIF_HASH_SC(sc)		GIF_HASH((sc)->gif_iphdr->ip_src.s_addr,\
96     (sc)->gif_iphdr->ip_dst.s_addr)
97 static uint32_t
98 in_gif_hashval(in_addr_t src, in_addr_t dst)
99 {
100 	uint32_t ret;
101 
102 	ret = fnv_32_buf(&src, sizeof(src), FNV1_32_INIT);
103 	return (fnv_32_buf(&dst, sizeof(dst), ret));
104 }
105 
106 static int
107 in_gif_checkdup(const struct gif_softc *sc, in_addr_t src, in_addr_t dst)
108 {
109 	struct gif_softc *tmp;
110 
111 	if (sc->gif_family == AF_INET &&
112 	    sc->gif_iphdr->ip_src.s_addr == src &&
113 	    sc->gif_iphdr->ip_dst.s_addr == dst)
114 		return (EEXIST);
115 
116 	CK_LIST_FOREACH(tmp, &GIF_HASH(src, dst), chain) {
117 		if (tmp == sc)
118 			continue;
119 		if (tmp->gif_iphdr->ip_src.s_addr == src &&
120 		    tmp->gif_iphdr->ip_dst.s_addr == dst)
121 			return (EADDRNOTAVAIL);
122 	}
123 	return (0);
124 }
125 
126 /*
127  * Check that ingress address belongs to local host.
128  */
129 static void
130 in_gif_set_running(struct gif_softc *sc)
131 {
132 
133 	if (in_localip(sc->gif_iphdr->ip_src))
134 		GIF2IFP(sc)->if_drv_flags |= IFF_DRV_RUNNING;
135 	else
136 		GIF2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING;
137 }
138 
139 /*
140  * ifaddr_event handler.
141  * Clear IFF_DRV_RUNNING flag when ingress address disappears to prevent
142  * source address spoofing.
143  */
144 static void
145 in_gif_srcaddr(void *arg __unused, const struct sockaddr *sa,
146     int event __unused)
147 {
148 	const struct sockaddr_in *sin;
149 	struct gif_softc *sc;
150 
151 	/* Check that VNET is ready */
152 	if (V_ipv4_hashtbl == NULL)
153 		return;
154 
155 	NET_EPOCH_ASSERT();
156 	sin = (const struct sockaddr_in *)sa;
157 	CK_LIST_FOREACH(sc, &GIF_SRCHASH(sin->sin_addr.s_addr), srchash) {
158 		if (sc->gif_iphdr->ip_src.s_addr != sin->sin_addr.s_addr)
159 			continue;
160 		in_gif_set_running(sc);
161 	}
162 }
163 
164 static void
165 in_gif_attach(struct gif_softc *sc)
166 {
167 
168 	if (sc->gif_options & GIF_IGNORE_SOURCE)
169 		CK_LIST_INSERT_HEAD(&V_ipv4_list, sc, chain);
170 	else
171 		CK_LIST_INSERT_HEAD(&GIF_HASH_SC(sc), sc, chain);
172 
173 	CK_LIST_INSERT_HEAD(&GIF_SRCHASH(sc->gif_iphdr->ip_src.s_addr),
174 	    sc, srchash);
175 }
176 
177 int
178 in_gif_setopts(struct gif_softc *sc, u_int options)
179 {
180 
181 	/* NOTE: we are protected with gif_ioctl_sx lock */
182 	MPASS(sc->gif_family == AF_INET);
183 	MPASS(sc->gif_options != options);
184 
185 	if ((options & GIF_IGNORE_SOURCE) !=
186 	    (sc->gif_options & GIF_IGNORE_SOURCE)) {
187 		CK_LIST_REMOVE(sc, srchash);
188 		CK_LIST_REMOVE(sc, chain);
189 		sc->gif_options = options;
190 		in_gif_attach(sc);
191 	}
192 	return (0);
193 }
194 
195 int
196 in_gif_ioctl(struct gif_softc *sc, u_long cmd, caddr_t data)
197 {
198 	struct ifreq *ifr = (struct ifreq *)data;
199 	struct epoch_tracker et;
200 	struct sockaddr_in *dst, *src;
201 	struct ip *ip;
202 	int error;
203 
204 	/* NOTE: we are protected with gif_ioctl_sx lock */
205 	error = EINVAL;
206 	switch (cmd) {
207 	case SIOCSIFPHYADDR:
208 		src = &((struct in_aliasreq *)data)->ifra_addr;
209 		dst = &((struct in_aliasreq *)data)->ifra_dstaddr;
210 
211 		/* sanity checks */
212 		if (src->sin_family != dst->sin_family ||
213 		    src->sin_family != AF_INET ||
214 		    src->sin_len != dst->sin_len ||
215 		    src->sin_len != sizeof(*src))
216 			break;
217 		if (src->sin_addr.s_addr == INADDR_ANY ||
218 		    dst->sin_addr.s_addr == INADDR_ANY) {
219 			error = EADDRNOTAVAIL;
220 			break;
221 		}
222 		if (V_ipv4_hashtbl == NULL) {
223 			V_ipv4_hashtbl = gif_hashinit();
224 			V_ipv4_srchashtbl = gif_hashinit();
225 		}
226 		error = in_gif_checkdup(sc, src->sin_addr.s_addr,
227 		    dst->sin_addr.s_addr);
228 		if (error == EADDRNOTAVAIL)
229 			break;
230 		if (error == EEXIST) {
231 			/* Addresses are the same. Just return. */
232 			error = 0;
233 			break;
234 		}
235 		ip = malloc(sizeof(*ip), M_GIF, M_WAITOK | M_ZERO);
236 		ip->ip_src.s_addr = src->sin_addr.s_addr;
237 		ip->ip_dst.s_addr = dst->sin_addr.s_addr;
238 		if (sc->gif_family != 0) {
239 			/* Detach existing tunnel first */
240 			CK_LIST_REMOVE(sc, srchash);
241 			CK_LIST_REMOVE(sc, chain);
242 			GIF_WAIT();
243 			free(sc->gif_hdr, M_GIF);
244 			/* XXX: should we notify about link state change? */
245 		}
246 		sc->gif_family = AF_INET;
247 		sc->gif_iphdr = ip;
248 		in_gif_attach(sc);
249 		NET_EPOCH_ENTER(et);
250 		in_gif_set_running(sc);
251 		NET_EPOCH_EXIT(et);
252 		break;
253 	case SIOCGIFPSRCADDR:
254 	case SIOCGIFPDSTADDR:
255 		if (sc->gif_family != AF_INET) {
256 			error = EADDRNOTAVAIL;
257 			break;
258 		}
259 		src = (struct sockaddr_in *)&ifr->ifr_addr;
260 		memset(src, 0, sizeof(*src));
261 		src->sin_family = AF_INET;
262 		src->sin_len = sizeof(*src);
263 		src->sin_addr = (cmd == SIOCGIFPSRCADDR) ?
264 		    sc->gif_iphdr->ip_src: sc->gif_iphdr->ip_dst;
265 		error = prison_if(curthread->td_ucred, (struct sockaddr *)src);
266 		if (error != 0)
267 			memset(src, 0, sizeof(*src));
268 		break;
269 	}
270 	return (error);
271 }
272 
273 int
274 in_gif_output(struct ifnet *ifp, struct mbuf *m, int proto, uint8_t ecn)
275 {
276 	struct gif_softc *sc = ifp->if_softc;
277 	struct ip *ip;
278 	int len;
279 
280 	/* prepend new IP header */
281 	NET_EPOCH_ASSERT();
282 	len = sizeof(struct ip);
283 #ifndef __NO_STRICT_ALIGNMENT
284 	if (proto == IPPROTO_ETHERIP)
285 		len += ETHERIP_ALIGN;
286 #endif
287 	M_PREPEND(m, len, M_NOWAIT);
288 	if (m == NULL)
289 		return (ENOBUFS);
290 #ifndef __NO_STRICT_ALIGNMENT
291 	if (proto == IPPROTO_ETHERIP) {
292 		len = mtod(m, vm_offset_t) & 3;
293 		KASSERT(len == 0 || len == ETHERIP_ALIGN,
294 		    ("in_gif_output: unexpected misalignment"));
295 		m->m_data += len;
296 		m->m_len -= ETHERIP_ALIGN;
297 	}
298 #endif
299 	ip = mtod(m, struct ip *);
300 
301 	MPASS(sc->gif_family == AF_INET);
302 	bcopy(sc->gif_iphdr, ip, sizeof(struct ip));
303 	ip->ip_p = proto;
304 	/* version will be set in ip_output() */
305 	ip->ip_ttl = V_ip_gif_ttl;
306 	ip->ip_len = htons(m->m_pkthdr.len);
307 	ip->ip_tos = ecn;
308 
309 	return (ip_output(m, NULL, NULL, 0, NULL, NULL));
310 }
311 
312 static int
313 in_gif_input(struct mbuf *m, int off, int proto, void *arg)
314 {
315 	struct gif_softc *sc = arg;
316 	struct ifnet *gifp;
317 	struct ip *ip;
318 	uint8_t ecn;
319 
320 	NET_EPOCH_ASSERT();
321 	if (sc == NULL) {
322 		m_freem(m);
323 		KMOD_IPSTAT_INC(ips_nogif);
324 		return (IPPROTO_DONE);
325 	}
326 	gifp = GIF2IFP(sc);
327 	if ((gifp->if_flags & IFF_UP) != 0) {
328 		ip = mtod(m, struct ip *);
329 		ecn = ip->ip_tos;
330 		m_adj(m, off);
331 		gif_input(m, gifp, proto, ecn);
332 	} else {
333 		m_freem(m);
334 		KMOD_IPSTAT_INC(ips_nogif);
335 	}
336 	return (IPPROTO_DONE);
337 }
338 
339 static int
340 in_gif_lookup(const struct mbuf *m, int off, int proto, void **arg)
341 {
342 	const struct ip *ip;
343 	struct gif_softc *sc;
344 	int ret;
345 
346 	if (V_ipv4_hashtbl == NULL)
347 		return (0);
348 
349 	NET_EPOCH_ASSERT();
350 	ip = mtod(m, const struct ip *);
351 	/*
352 	 * NOTE: it is safe to iterate without any locking here, because softc
353 	 * can be reclaimed only when we are not within net_epoch_preempt
354 	 * section, but ip_encap lookup+input are executed in epoch section.
355 	 */
356 	ret = 0;
357 	CK_LIST_FOREACH(sc, &GIF_HASH(ip->ip_dst.s_addr,
358 	    ip->ip_src.s_addr), chain) {
359 		/*
360 		 * This is an inbound packet, its ip_dst is source address
361 		 * in softc.
362 		 */
363 		if (sc->gif_iphdr->ip_src.s_addr == ip->ip_dst.s_addr &&
364 		    sc->gif_iphdr->ip_dst.s_addr == ip->ip_src.s_addr) {
365 			ret = ENCAP_DRV_LOOKUP;
366 			goto done;
367 		}
368 	}
369 	/*
370 	 * No exact match.
371 	 * Check the list of interfaces with GIF_IGNORE_SOURCE flag.
372 	 */
373 	CK_LIST_FOREACH(sc, &V_ipv4_list, chain) {
374 		if (sc->gif_iphdr->ip_src.s_addr == ip->ip_dst.s_addr) {
375 			ret = 32 + 8; /* src + proto */
376 			goto done;
377 		}
378 	}
379 	return (0);
380 done:
381 	if ((GIF2IFP(sc)->if_flags & IFF_UP) == 0)
382 		return (0);
383 	/* ingress filters on outer source */
384 	if ((GIF2IFP(sc)->if_flags & IFF_LINK2) == 0) {
385 		if (fib4_check_urpf(sc->gif_fibnum, ip->ip_src, 0, NHR_NONE,
386 					m->m_pkthdr.rcvif) == 0)
387 			return (0);
388 	}
389 	*arg = sc;
390 	return (ret);
391 }
392 
393 static const struct srcaddrtab *ipv4_srcaddrtab;
394 static struct {
395 	const struct encap_config encap;
396 	const struct encaptab *cookie;
397 } ipv4_encap_cfg[] = {
398 	{
399 		.encap = {
400 			.proto = IPPROTO_IPV4,
401 			.min_length = 2 * sizeof(struct ip),
402 			.exact_match = ENCAP_DRV_LOOKUP,
403 			.lookup = in_gif_lookup,
404 			.input = in_gif_input
405 		},
406 	},
407 #ifdef INET6
408 	{
409 		.encap = {
410 			.proto = IPPROTO_IPV6,
411 			.min_length = sizeof(struct ip) +
412 			    sizeof(struct ip6_hdr),
413 			.exact_match = ENCAP_DRV_LOOKUP,
414 			.lookup = in_gif_lookup,
415 			.input = in_gif_input
416 		},
417 	},
418 #endif
419 	{
420 		.encap = {
421 			.proto = IPPROTO_ETHERIP,
422 			.min_length = sizeof(struct ip) +
423 			    sizeof(struct etherip_header) +
424 			    sizeof(struct ether_header),
425 			.exact_match = ENCAP_DRV_LOOKUP,
426 			.lookup = in_gif_lookup,
427 			.input = in_gif_input
428 		},
429 	}
430 };
431 
432 void
433 in_gif_init(void)
434 {
435 	int i;
436 
437 	if (!IS_DEFAULT_VNET(curvnet))
438 		return;
439 
440 	ipv4_srcaddrtab = ip_encap_register_srcaddr(in_gif_srcaddr,
441 	    NULL, M_WAITOK);
442 	for (i = 0; i < nitems(ipv4_encap_cfg); i++)
443 		ipv4_encap_cfg[i].cookie = ip_encap_attach(
444 		    &ipv4_encap_cfg[i].encap, NULL, M_WAITOK);
445 }
446 
447 void
448 in_gif_uninit(void)
449 {
450 	int i;
451 
452 	if (IS_DEFAULT_VNET(curvnet)) {
453 		for (i = 0; i < nitems(ipv4_encap_cfg); i++)
454 			ip_encap_detach(ipv4_encap_cfg[i].cookie);
455 		ip_encap_unregister_srcaddr(ipv4_srcaddrtab);
456 	}
457 	if (V_ipv4_hashtbl != NULL) {
458 		gif_hashdestroy(V_ipv4_hashtbl);
459 		V_ipv4_hashtbl = NULL;
460 		GIF_WAIT();
461 		gif_hashdestroy(V_ipv4_srchashtbl);
462 	}
463 }
464