xref: /freebsd/sys/netinet/in_gif.c (revision f540a43052c12c76d3453ead881248d5467a1ab0) 
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/if_private.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/ip.h>
63 #include <netinet/ip_var.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip_encap.h>
66 #include <netinet/ip_ecn.h>
67 #include <netinet/in_fib.h>
68 
69 #ifdef INET6
70 #include <netinet/ip6.h>
71 #endif
72 
73 #include <net/if_gif.h>
74 
75 #define GIF_TTL		30
76 VNET_DEFINE_STATIC(int, ip_gif_ttl) = GIF_TTL;
77 #define	V_ip_gif_ttl		VNET(ip_gif_ttl)
78 SYSCTL_INT(_net_inet_ip, IPCTL_GIF_TTL, gifttl, CTLFLAG_VNET | CTLFLAG_RW,
79     &VNET_NAME(ip_gif_ttl), 0, "Default TTL value for encapsulated packets");
80 
81 /*
82  * We keep interfaces in a hash table using src+dst as key.
83  * Interfaces with GIF_IGNORE_SOURCE flag are linked into plain list.
84  */
85 VNET_DEFINE_STATIC(struct gif_list *, ipv4_hashtbl) = NULL;
86 VNET_DEFINE_STATIC(struct gif_list *, ipv4_srchashtbl) = NULL;
87 VNET_DEFINE_STATIC(struct gif_list, ipv4_list) = CK_LIST_HEAD_INITIALIZER();
88 #define	V_ipv4_hashtbl		VNET(ipv4_hashtbl)
89 #define	V_ipv4_srchashtbl	VNET(ipv4_srchashtbl)
90 #define	V_ipv4_list		VNET(ipv4_list)
91 
92 #define	GIF_HASH(src, dst)	(V_ipv4_hashtbl[\
93     in_gif_hashval((src), (dst)) & (GIF_HASH_SIZE - 1)])
94 #define	GIF_SRCHASH(src)	(V_ipv4_srchashtbl[\
95     fnv_32_buf(&(src), sizeof(src), FNV1_32_INIT) & (GIF_HASH_SIZE - 1)])
96 #define	GIF_HASH_SC(sc)		GIF_HASH((sc)->gif_iphdr->ip_src.s_addr,\
97     (sc)->gif_iphdr->ip_dst.s_addr)
98 static uint32_t
99 in_gif_hashval(in_addr_t src, in_addr_t dst)
100 {
101 	uint32_t ret;
102 
103 	ret = fnv_32_buf(&src, sizeof(src), FNV1_32_INIT);
104 	return (fnv_32_buf(&dst, sizeof(dst), ret));
105 }
106 
107 static int
108 in_gif_checkdup(const struct gif_softc *sc, in_addr_t src, in_addr_t dst)
109 {
110 	struct gif_softc *tmp;
111 
112 	if (sc->gif_family == AF_INET &&
113 	    sc->gif_iphdr->ip_src.s_addr == src &&
114 	    sc->gif_iphdr->ip_dst.s_addr == dst)
115 		return (EEXIST);
116 
117 	CK_LIST_FOREACH(tmp, &GIF_HASH(src, dst), chain) {
118 		if (tmp == sc)
119 			continue;
120 		if (tmp->gif_iphdr->ip_src.s_addr == src &&
121 		    tmp->gif_iphdr->ip_dst.s_addr == dst)
122 			return (EADDRNOTAVAIL);
123 	}
124 	return (0);
125 }
126 
127 /*
128  * Check that ingress address belongs to local host.
129  */
130 static void
131 in_gif_set_running(struct gif_softc *sc)
132 {
133 
134 	if (in_localip(sc->gif_iphdr->ip_src))
135 		GIF2IFP(sc)->if_drv_flags |= IFF_DRV_RUNNING;
136 	else
137 		GIF2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING;
138 }
139 
140 /*
141  * ifaddr_event handler.
142  * Clear IFF_DRV_RUNNING flag when ingress address disappears to prevent
143  * source address spoofing.
144  */
145 static void
146 in_gif_srcaddr(void *arg __unused, const struct sockaddr *sa,
147     int event __unused)
148 {
149 	const struct sockaddr_in *sin;
150 	struct gif_softc *sc;
151 
152 	/* Check that VNET is ready */
153 	if (V_ipv4_hashtbl == NULL)
154 		return;
155 
156 	NET_EPOCH_ASSERT();
157 	sin = (const struct sockaddr_in *)sa;
158 	CK_LIST_FOREACH(sc, &GIF_SRCHASH(sin->sin_addr.s_addr), srchash) {
159 		if (sc->gif_iphdr->ip_src.s_addr != sin->sin_addr.s_addr)
160 			continue;
161 		in_gif_set_running(sc);
162 	}
163 }
164 
165 static void
166 in_gif_attach(struct gif_softc *sc)
167 {
168 
169 	if (sc->gif_options & GIF_IGNORE_SOURCE)
170 		CK_LIST_INSERT_HEAD(&V_ipv4_list, sc, chain);
171 	else
172 		CK_LIST_INSERT_HEAD(&GIF_HASH_SC(sc), sc, chain);
173 
174 	CK_LIST_INSERT_HEAD(&GIF_SRCHASH(sc->gif_iphdr->ip_src.s_addr),
175 	    sc, srchash);
176 }
177 
178 int
179 in_gif_setopts(struct gif_softc *sc, u_int options)
180 {
181 
182 	/* NOTE: we are protected with gif_ioctl_sx lock */
183 	MPASS(sc->gif_family == AF_INET);
184 	MPASS(sc->gif_options != options);
185 
186 	if ((options & GIF_IGNORE_SOURCE) !=
187 	    (sc->gif_options & GIF_IGNORE_SOURCE)) {
188 		CK_LIST_REMOVE(sc, srchash);
189 		CK_LIST_REMOVE(sc, chain);
190 		sc->gif_options = options;
191 		in_gif_attach(sc);
192 	}
193 	return (0);
194 }
195 
196 int
197 in_gif_ioctl(struct gif_softc *sc, u_long cmd, caddr_t data)
198 {
199 	struct ifreq *ifr = (struct ifreq *)data;
200 	struct epoch_tracker et;
201 	struct sockaddr_in *dst, *src;
202 	struct ip *ip;
203 	int error;
204 
205 	/* NOTE: we are protected with gif_ioctl_sx lock */
206 	error = EINVAL;
207 	switch (cmd) {
208 	case SIOCSIFPHYADDR:
209 		src = &((struct in_aliasreq *)data)->ifra_addr;
210 		dst = &((struct in_aliasreq *)data)->ifra_dstaddr;
211 
212 		/* sanity checks */
213 		if (src->sin_family != dst->sin_family ||
214 		    src->sin_family != AF_INET ||
215 		    src->sin_len != dst->sin_len ||
216 		    src->sin_len != sizeof(*src))
217 			break;
218 		if (src->sin_addr.s_addr == INADDR_ANY ||
219 		    dst->sin_addr.s_addr == INADDR_ANY) {
220 			error = EADDRNOTAVAIL;
221 			break;
222 		}
223 		if (V_ipv4_hashtbl == NULL) {
224 			V_ipv4_hashtbl = gif_hashinit();
225 			V_ipv4_srchashtbl = gif_hashinit();
226 		}
227 		error = in_gif_checkdup(sc, src->sin_addr.s_addr,
228 		    dst->sin_addr.s_addr);
229 		if (error == EADDRNOTAVAIL)
230 			break;
231 		if (error == EEXIST) {
232 			/* Addresses are the same. Just return. */
233 			error = 0;
234 			break;
235 		}
236 		ip = malloc(sizeof(*ip), M_GIF, M_WAITOK | M_ZERO);
237 		ip->ip_src.s_addr = src->sin_addr.s_addr;
238 		ip->ip_dst.s_addr = dst->sin_addr.s_addr;
239 		if (sc->gif_family != 0) {
240 			/* Detach existing tunnel first */
241 			CK_LIST_REMOVE(sc, srchash);
242 			CK_LIST_REMOVE(sc, chain);
243 			GIF_WAIT();
244 			free(sc->gif_hdr, M_GIF);
245 			/* XXX: should we notify about link state change? */
246 		}
247 		sc->gif_family = AF_INET;
248 		sc->gif_iphdr = ip;
249 		in_gif_attach(sc);
250 		NET_EPOCH_ENTER(et);
251 		in_gif_set_running(sc);
252 		NET_EPOCH_EXIT(et);
253 		break;
254 	case SIOCGIFPSRCADDR:
255 	case SIOCGIFPDSTADDR:
256 		if (sc->gif_family != AF_INET) {
257 			error = EADDRNOTAVAIL;
258 			break;
259 		}
260 		src = (struct sockaddr_in *)&ifr->ifr_addr;
261 		memset(src, 0, sizeof(*src));
262 		src->sin_family = AF_INET;
263 		src->sin_len = sizeof(*src);
264 		src->sin_addr = (cmd == SIOCGIFPSRCADDR) ?
265 		    sc->gif_iphdr->ip_src: sc->gif_iphdr->ip_dst;
266 		error = prison_if(curthread->td_ucred, (struct sockaddr *)src);
267 		if (error != 0)
268 			memset(src, 0, sizeof(*src));
269 		break;
270 	}
271 	return (error);
272 }
273 
274 int
275 in_gif_output(struct ifnet *ifp, struct mbuf *m, int proto, uint8_t ecn)
276 {
277 	struct gif_softc *sc = ifp->if_softc;
278 	struct ip *ip;
279 	int len;
280 
281 	/* prepend new IP header */
282 	NET_EPOCH_ASSERT();
283 	len = sizeof(struct ip);
284 #ifndef __NO_STRICT_ALIGNMENT
285 	if (proto == IPPROTO_ETHERIP)
286 		len += ETHERIP_ALIGN;
287 #endif
288 	M_PREPEND(m, len, M_NOWAIT);
289 	if (m == NULL)
290 		return (ENOBUFS);
291 #ifndef __NO_STRICT_ALIGNMENT
292 	if (proto == IPPROTO_ETHERIP) {
293 		len = mtod(m, vm_offset_t) & 3;
294 		KASSERT(len == 0 || len == ETHERIP_ALIGN,
295 		    ("in_gif_output: unexpected misalignment"));
296 		m->m_data += len;
297 		m->m_len -= ETHERIP_ALIGN;
298 	}
299 #endif
300 	ip = mtod(m, struct ip *);
301 
302 	MPASS(sc->gif_family == AF_INET);
303 	bcopy(sc->gif_iphdr, ip, sizeof(struct ip));
304 	ip->ip_p = proto;
305 	/* version will be set in ip_output() */
306 	ip->ip_ttl = V_ip_gif_ttl;
307 	ip->ip_len = htons(m->m_pkthdr.len);
308 	ip->ip_tos = ecn;
309 
310 	return (ip_output(m, NULL, NULL, 0, NULL, NULL));
311 }
312 
313 static int
314 in_gif_input(struct mbuf *m, int off, int proto, void *arg)
315 {
316 	struct gif_softc *sc = arg;
317 	struct ifnet *gifp;
318 	struct ip *ip;
319 	uint8_t ecn;
320 
321 	NET_EPOCH_ASSERT();
322 	if (sc == NULL) {
323 		m_freem(m);
324 		KMOD_IPSTAT_INC(ips_nogif);
325 		return (IPPROTO_DONE);
326 	}
327 	gifp = GIF2IFP(sc);
328 	if ((gifp->if_flags & IFF_UP) != 0) {
329 		ip = mtod(m, struct ip *);
330 		ecn = ip->ip_tos;
331 		m_adj(m, off);
332 		gif_input(m, gifp, proto, ecn);
333 	} else {
334 		m_freem(m);
335 		KMOD_IPSTAT_INC(ips_nogif);
336 	}
337 	return (IPPROTO_DONE);
338 }
339 
340 static int
341 in_gif_lookup(const struct mbuf *m, int off, int proto, void **arg)
342 {
343 	const struct ip *ip;
344 	struct gif_softc *sc;
345 	int ret;
346 
347 	if (V_ipv4_hashtbl == NULL)
348 		return (0);
349 
350 	NET_EPOCH_ASSERT();
351 	ip = mtod(m, const struct ip *);
352 	/*
353 	 * NOTE: it is safe to iterate without any locking here, because softc
354 	 * can be reclaimed only when we are not within net_epoch_preempt
355 	 * section, but ip_encap lookup+input are executed in epoch section.
356 	 */
357 	ret = 0;
358 	CK_LIST_FOREACH(sc, &GIF_HASH(ip->ip_dst.s_addr,
359 	    ip->ip_src.s_addr), chain) {
360 		/*
361 		 * This is an inbound packet, its ip_dst is source address
362 		 * in softc.
363 		 */
364 		if (sc->gif_iphdr->ip_src.s_addr == ip->ip_dst.s_addr &&
365 		    sc->gif_iphdr->ip_dst.s_addr == ip->ip_src.s_addr) {
366 			ret = ENCAP_DRV_LOOKUP;
367 			goto done;
368 		}
369 	}
370 	/*
371 	 * No exact match.
372 	 * Check the list of interfaces with GIF_IGNORE_SOURCE flag.
373 	 */
374 	CK_LIST_FOREACH(sc, &V_ipv4_list, chain) {
375 		if (sc->gif_iphdr->ip_src.s_addr == ip->ip_dst.s_addr) {
376 			ret = 32 + 8; /* src + proto */
377 			goto done;
378 		}
379 	}
380 	return (0);
381 done:
382 	if ((GIF2IFP(sc)->if_flags & IFF_UP) == 0)
383 		return (0);
384 	/* ingress filters on outer source */
385 	if ((GIF2IFP(sc)->if_flags & IFF_LINK2) == 0) {
386 		if (fib4_check_urpf(sc->gif_fibnum, ip->ip_src, 0, NHR_NONE,
387 					m->m_pkthdr.rcvif) == 0)
388 			return (0);
389 	}
390 	*arg = sc;
391 	return (ret);
392 }
393 
394 static const struct srcaddrtab *ipv4_srcaddrtab;
395 static struct {
396 	const struct encap_config encap;
397 	const struct encaptab *cookie;
398 } ipv4_encap_cfg[] = {
399 	{
400 		.encap = {
401 			.proto = IPPROTO_IPV4,
402 			.min_length = 2 * sizeof(struct ip),
403 			.exact_match = ENCAP_DRV_LOOKUP,
404 			.lookup = in_gif_lookup,
405 			.input = in_gif_input
406 		},
407 	},
408 #ifdef INET6
409 	{
410 		.encap = {
411 			.proto = IPPROTO_IPV6,
412 			.min_length = sizeof(struct ip) +
413 			    sizeof(struct ip6_hdr),
414 			.exact_match = ENCAP_DRV_LOOKUP,
415 			.lookup = in_gif_lookup,
416 			.input = in_gif_input
417 		},
418 	},
419 #endif
420 	{
421 		.encap = {
422 			.proto = IPPROTO_ETHERIP,
423 			.min_length = sizeof(struct ip) +
424 			    sizeof(struct etherip_header) +
425 			    sizeof(struct ether_header),
426 			.exact_match = ENCAP_DRV_LOOKUP,
427 			.lookup = in_gif_lookup,
428 			.input = in_gif_input
429 		},
430 	}
431 };
432 
433 void
434 in_gif_init(void)
435 {
436 	int i;
437 
438 	if (!IS_DEFAULT_VNET(curvnet))
439 		return;
440 
441 	ipv4_srcaddrtab = ip_encap_register_srcaddr(in_gif_srcaddr,
442 	    NULL, M_WAITOK);
443 	for (i = 0; i < nitems(ipv4_encap_cfg); i++)
444 		ipv4_encap_cfg[i].cookie = ip_encap_attach(
445 		    &ipv4_encap_cfg[i].encap, NULL, M_WAITOK);
446 }
447 
448 void
449 in_gif_uninit(void)
450 {
451 	int i;
452 
453 	if (IS_DEFAULT_VNET(curvnet)) {
454 		for (i = 0; i < nitems(ipv4_encap_cfg); i++)
455 			ip_encap_detach(ipv4_encap_cfg[i].cookie);
456 		ip_encap_unregister_srcaddr(ipv4_srcaddrtab);
457 	}
458 	if (V_ipv4_hashtbl != NULL) {
459 		gif_hashdestroy(V_ipv4_hashtbl);
460 		V_ipv4_hashtbl = NULL;
461 		GIF_WAIT();
462 		gif_hashdestroy(V_ipv4_srchashtbl);
463 	}
464 }
465