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 #include "opt_inet.h"
37 #include "opt_inet6.h"
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/jail.h>
42 #include <sys/socket.h>
43 #include <sys/sockio.h>
44 #include <sys/mbuf.h>
45 #include <sys/errno.h>
46 #include <sys/kernel.h>
47 #include <sys/sysctl.h>
48 #include <sys/malloc.h>
49 #include <sys/proc.h>
50
51 #include <net/ethernet.h>
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/route.h>
55 #include <net/vnet.h>
56
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/ip.h>
60 #include <netinet/ip_var.h>
61 #include <netinet/in_var.h>
62 #include <netinet/ip_encap.h>
63 #include <netinet/ip_ecn.h>
64 #include <netinet/in_fib.h>
65
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #endif
69
70 #include <net/if_gif.h>
71
72 #define GIF_TTL 30
73 VNET_DEFINE_STATIC(int, ip_gif_ttl) = GIF_TTL;
74 #define V_ip_gif_ttl VNET(ip_gif_ttl)
75 SYSCTL_INT(_net_inet_ip, IPCTL_GIF_TTL, gifttl, CTLFLAG_VNET | CTLFLAG_RW,
76 &VNET_NAME(ip_gif_ttl), 0, "Default TTL value for encapsulated packets");
77
78 /*
79 * We keep interfaces in a hash table using src+dst as key.
80 * Interfaces with GIF_IGNORE_SOURCE flag are linked into plain list.
81 */
82 VNET_DEFINE_STATIC(struct gif_list *, ipv4_hashtbl) = NULL;
83 VNET_DEFINE_STATIC(struct gif_list *, ipv4_srchashtbl) = NULL;
84 VNET_DEFINE_STATIC(struct gif_list, ipv4_list) = CK_LIST_HEAD_INITIALIZER();
85 #define V_ipv4_hashtbl VNET(ipv4_hashtbl)
86 #define V_ipv4_srchashtbl VNET(ipv4_srchashtbl)
87 #define V_ipv4_list VNET(ipv4_list)
88
89 #define GIF_HASH(src, dst) (V_ipv4_hashtbl[\
90 in_gif_hashval((src), (dst)) & (GIF_HASH_SIZE - 1)])
91 #define GIF_SRCHASH(src) (V_ipv4_srchashtbl[\
92 fnv_32_buf(&(src), sizeof(src), FNV1_32_INIT) & (GIF_HASH_SIZE - 1)])
93 #define GIF_HASH_SC(sc) GIF_HASH((sc)->gif_iphdr->ip_src.s_addr,\
94 (sc)->gif_iphdr->ip_dst.s_addr)
95 static uint32_t
in_gif_hashval(in_addr_t src,in_addr_t dst)96 in_gif_hashval(in_addr_t src, in_addr_t dst)
97 {
98 uint32_t ret;
99
100 ret = fnv_32_buf(&src, sizeof(src), FNV1_32_INIT);
101 return (fnv_32_buf(&dst, sizeof(dst), ret));
102 }
103
104 static int
in_gif_checkdup(const struct gif_softc * sc,in_addr_t src,in_addr_t dst)105 in_gif_checkdup(const struct gif_softc *sc, in_addr_t src, in_addr_t dst)
106 {
107 struct gif_softc *tmp;
108
109 if (sc->gif_family == AF_INET &&
110 sc->gif_iphdr->ip_src.s_addr == src &&
111 sc->gif_iphdr->ip_dst.s_addr == dst)
112 return (EEXIST);
113
114 CK_LIST_FOREACH(tmp, &GIF_HASH(src, dst), chain) {
115 if (tmp == sc)
116 continue;
117 if (tmp->gif_iphdr->ip_src.s_addr == src &&
118 tmp->gif_iphdr->ip_dst.s_addr == dst)
119 return (EADDRNOTAVAIL);
120 }
121 return (0);
122 }
123
124 /*
125 * Check that ingress address belongs to local host.
126 */
127 static void
in_gif_set_running(struct gif_softc * sc)128 in_gif_set_running(struct gif_softc *sc)
129 {
130
131 if (in_localip(sc->gif_iphdr->ip_src))
132 GIF2IFP(sc)->if_drv_flags |= IFF_DRV_RUNNING;
133 else
134 GIF2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING;
135 }
136
137 /*
138 * ifaddr_event handler.
139 * Clear IFF_DRV_RUNNING flag when ingress address disappears to prevent
140 * source address spoofing.
141 */
142 static void
in_gif_srcaddr(void * arg __unused,const struct sockaddr * sa,int event __unused)143 in_gif_srcaddr(void *arg __unused, const struct sockaddr *sa,
144 int event __unused)
145 {
146 const struct sockaddr_in *sin;
147 struct gif_softc *sc;
148
149 /* Check that VNET is ready */
150 if (V_ipv4_hashtbl == NULL)
151 return;
152
153 NET_EPOCH_ASSERT();
154 sin = (const struct sockaddr_in *)sa;
155 CK_LIST_FOREACH(sc, &GIF_SRCHASH(sin->sin_addr.s_addr), srchash) {
156 if (sc->gif_iphdr->ip_src.s_addr != sin->sin_addr.s_addr)
157 continue;
158 in_gif_set_running(sc);
159 }
160 }
161
162 static void
in_gif_attach(struct gif_softc * sc)163 in_gif_attach(struct gif_softc *sc)
164 {
165
166 if (sc->gif_options & GIF_IGNORE_SOURCE)
167 CK_LIST_INSERT_HEAD(&V_ipv4_list, sc, chain);
168 else
169 CK_LIST_INSERT_HEAD(&GIF_HASH_SC(sc), sc, chain);
170
171 CK_LIST_INSERT_HEAD(&GIF_SRCHASH(sc->gif_iphdr->ip_src.s_addr),
172 sc, srchash);
173 }
174
175 int
in_gif_setopts(struct gif_softc * sc,u_int options)176 in_gif_setopts(struct gif_softc *sc, u_int options)
177 {
178
179 /* NOTE: we are protected with gif_ioctl_sx lock */
180 MPASS(sc->gif_family == AF_INET);
181 MPASS(sc->gif_options != options);
182
183 if ((options & GIF_IGNORE_SOURCE) !=
184 (sc->gif_options & GIF_IGNORE_SOURCE)) {
185 CK_LIST_REMOVE(sc, srchash);
186 CK_LIST_REMOVE(sc, chain);
187 sc->gif_options = options;
188 in_gif_attach(sc);
189 }
190 return (0);
191 }
192
193 int
in_gif_ioctl(struct gif_softc * sc,u_long cmd,caddr_t data)194 in_gif_ioctl(struct gif_softc *sc, u_long cmd, caddr_t data)
195 {
196 struct ifreq *ifr = (struct ifreq *)data;
197 struct sockaddr_in *dst, *src;
198 struct ip *ip;
199 int error;
200
201 /* NOTE: we are protected with gif_ioctl_sx lock */
202 error = EINVAL;
203 switch (cmd) {
204 case SIOCSIFPHYADDR:
205 src = &((struct in_aliasreq *)data)->ifra_addr;
206 dst = &((struct in_aliasreq *)data)->ifra_dstaddr;
207
208 /* sanity checks */
209 if (src->sin_family != dst->sin_family ||
210 src->sin_family != AF_INET ||
211 src->sin_len != dst->sin_len ||
212 src->sin_len != sizeof(*src))
213 break;
214 if (src->sin_addr.s_addr == INADDR_ANY ||
215 dst->sin_addr.s_addr == INADDR_ANY) {
216 error = EADDRNOTAVAIL;
217 break;
218 }
219 if (V_ipv4_hashtbl == NULL) {
220 V_ipv4_hashtbl = gif_hashinit();
221 V_ipv4_srchashtbl = gif_hashinit();
222 }
223 error = in_gif_checkdup(sc, src->sin_addr.s_addr,
224 dst->sin_addr.s_addr);
225 if (error == EADDRNOTAVAIL)
226 break;
227 if (error == EEXIST) {
228 /* Addresses are the same. Just return. */
229 error = 0;
230 break;
231 }
232 ip = malloc(sizeof(*ip), M_GIF, M_WAITOK | M_ZERO);
233 ip->ip_src.s_addr = src->sin_addr.s_addr;
234 ip->ip_dst.s_addr = dst->sin_addr.s_addr;
235 if (sc->gif_family != 0) {
236 /* Detach existing tunnel first */
237 CK_LIST_REMOVE(sc, srchash);
238 CK_LIST_REMOVE(sc, chain);
239 GIF_WAIT();
240 free(sc->gif_hdr, M_GIF);
241 /* XXX: should we notify about link state change? */
242 }
243 sc->gif_family = AF_INET;
244 sc->gif_iphdr = ip;
245 in_gif_attach(sc);
246 in_gif_set_running(sc);
247 break;
248 case SIOCGIFPSRCADDR:
249 case SIOCGIFPDSTADDR:
250 if (sc->gif_family != AF_INET) {
251 error = EADDRNOTAVAIL;
252 break;
253 }
254 src = (struct sockaddr_in *)&ifr->ifr_addr;
255 memset(src, 0, sizeof(*src));
256 src->sin_family = AF_INET;
257 src->sin_len = sizeof(*src);
258 src->sin_addr = (cmd == SIOCGIFPSRCADDR) ?
259 sc->gif_iphdr->ip_src: sc->gif_iphdr->ip_dst;
260 error = prison_if(curthread->td_ucred, (struct sockaddr *)src);
261 if (error != 0)
262 memset(src, 0, sizeof(*src));
263 break;
264 }
265 return (error);
266 }
267
268 int
in_gif_output(struct ifnet * ifp,struct mbuf * m,int proto,uint8_t ecn)269 in_gif_output(struct ifnet *ifp, struct mbuf *m, int proto, uint8_t ecn)
270 {
271 struct gif_softc *sc = ifp->if_softc;
272 struct ip *ip;
273
274 /* prepend new IP header */
275 NET_EPOCH_ASSERT();
276 M_PREPEND(m, sizeof(struct ip), M_NOWAIT);
277 if (m == NULL)
278 return (ENOBUFS);
279 ip = mtod(m, struct ip *);
280
281 MPASS(sc->gif_family == AF_INET);
282 memcpy(ip, sc->gif_iphdr, sizeof(struct ip));
283 ip->ip_p = proto;
284 /* version will be set in ip_output() */
285 ip->ip_ttl = V_ip_gif_ttl;
286 ip->ip_len = htons(m->m_pkthdr.len);
287 ip->ip_tos = ecn;
288
289 return (ip_output(m, NULL, NULL, 0, NULL, NULL));
290 }
291
292 static int
in_gif_input(struct mbuf * m,int off,int proto,void * arg)293 in_gif_input(struct mbuf *m, int off, int proto, void *arg)
294 {
295 struct gif_softc *sc = arg;
296 struct ifnet *gifp;
297 struct ip *ip;
298 uint8_t ecn;
299
300 NET_EPOCH_ASSERT();
301 if (sc == NULL) {
302 m_freem(m);
303 KMOD_IPSTAT_INC(ips_nogif);
304 return (IPPROTO_DONE);
305 }
306 gifp = GIF2IFP(sc);
307 if ((gifp->if_flags & IFF_UP) != 0) {
308 ip = mtod(m, struct ip *);
309 ecn = ip->ip_tos;
310 m_adj(m, off);
311 gif_input(m, gifp, proto, ecn);
312 } else {
313 m_freem(m);
314 KMOD_IPSTAT_INC(ips_nogif);
315 }
316 return (IPPROTO_DONE);
317 }
318
319 static int
in_gif_lookup(const struct mbuf * m,int off,int proto,void ** arg)320 in_gif_lookup(const struct mbuf *m, int off, int proto, void **arg)
321 {
322 const struct ip *ip;
323 struct gif_softc *sc;
324 int ret;
325
326 if (V_ipv4_hashtbl == NULL)
327 return (0);
328
329 NET_EPOCH_ASSERT();
330 ip = mtod(m, const struct ip *);
331 /*
332 * NOTE: it is safe to iterate without any locking here, because softc
333 * can be reclaimed only when we are not within net_epoch_preempt
334 * section, but ip_encap lookup+input are executed in epoch section.
335 */
336 ret = 0;
337 CK_LIST_FOREACH(sc, &GIF_HASH(ip->ip_dst.s_addr,
338 ip->ip_src.s_addr), chain) {
339 /*
340 * This is an inbound packet, its ip_dst is source address
341 * in softc.
342 */
343 if (sc->gif_iphdr->ip_src.s_addr == ip->ip_dst.s_addr &&
344 sc->gif_iphdr->ip_dst.s_addr == ip->ip_src.s_addr) {
345 ret = ENCAP_DRV_LOOKUP;
346 goto done;
347 }
348 }
349 /*
350 * No exact match.
351 * Check the list of interfaces with GIF_IGNORE_SOURCE flag.
352 */
353 CK_LIST_FOREACH(sc, &V_ipv4_list, chain) {
354 if (sc->gif_iphdr->ip_src.s_addr == ip->ip_dst.s_addr) {
355 ret = 32 + 8; /* src + proto */
356 goto done;
357 }
358 }
359 return (0);
360 done:
361 if ((GIF2IFP(sc)->if_flags & IFF_UP) == 0)
362 return (0);
363 /* ingress filters on outer source */
364 if ((GIF2IFP(sc)->if_flags & IFF_LINK2) == 0) {
365 if (fib4_check_urpf(sc->gif_fibnum, ip->ip_src, 0, NHR_NONE,
366 m->m_pkthdr.rcvif) == 0)
367 return (0);
368 }
369 *arg = sc;
370 return (ret);
371 }
372
373 static const struct srcaddrtab *ipv4_srcaddrtab;
374 static struct {
375 const struct encap_config encap;
376 const struct encaptab *cookie;
377 } ipv4_encap_cfg[] = {
378 {
379 .encap = {
380 .proto = IPPROTO_IPV4,
381 .min_length = 2 * sizeof(struct ip),
382 .exact_match = ENCAP_DRV_LOOKUP,
383 .lookup = in_gif_lookup,
384 .input = in_gif_input
385 },
386 },
387 #ifdef INET6
388 {
389 .encap = {
390 .proto = IPPROTO_IPV6,
391 .min_length = sizeof(struct ip) +
392 sizeof(struct ip6_hdr),
393 .exact_match = ENCAP_DRV_LOOKUP,
394 .lookup = in_gif_lookup,
395 .input = in_gif_input
396 },
397 },
398 #endif
399 {
400 .encap = {
401 .proto = IPPROTO_ETHERIP,
402 .min_length = sizeof(struct ip) +
403 sizeof(struct etherip_header) +
404 sizeof(struct ether_header),
405 .exact_match = ENCAP_DRV_LOOKUP,
406 .lookup = in_gif_lookup,
407 .input = in_gif_input
408 },
409 }
410 };
411
412 void
in_gif_init(void)413 in_gif_init(void)
414 {
415 int i;
416
417 if (!IS_DEFAULT_VNET(curvnet))
418 return;
419
420 ipv4_srcaddrtab = ip_encap_register_srcaddr(in_gif_srcaddr,
421 NULL, M_WAITOK);
422 for (i = 0; i < nitems(ipv4_encap_cfg); i++)
423 ipv4_encap_cfg[i].cookie = ip_encap_attach(
424 &ipv4_encap_cfg[i].encap, NULL, M_WAITOK);
425 }
426
427 void
in_gif_uninit(void)428 in_gif_uninit(void)
429 {
430 int i;
431
432 if (IS_DEFAULT_VNET(curvnet)) {
433 for (i = 0; i < nitems(ipv4_encap_cfg); i++)
434 ip_encap_detach(ipv4_encap_cfg[i].cookie);
435 ip_encap_unregister_srcaddr(ipv4_srcaddrtab);
436 }
437 if (V_ipv4_hashtbl != NULL) {
438 gif_hashdestroy(V_ipv4_hashtbl);
439 V_ipv4_hashtbl = NULL;
440 GIF_WAIT();
441 gif_hashdestroy(V_ipv4_srchashtbl);
442 }
443 }
444