1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (C) 2001 WIDE Project. 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 University 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 REGENTS 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 REGENTS 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
33 #include "opt_inet.h"
34
35 #define IN_HISTORICAL_NETS /* include class masks */
36
37 #include <sys/param.h>
38 #include <sys/eventhandler.h>
39 #include <sys/systm.h>
40 #include <sys/sockio.h>
41 #include <sys/malloc.h>
42 #include <sys/priv.h>
43 #include <sys/socket.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/proc.h>
48 #include <sys/sysctl.h>
49 #include <sys/syslog.h>
50 #include <sys/sx.h>
51
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/if_arp.h>
55 #include <net/if_dl.h>
56 #include <net/if_llatbl.h>
57 #include <net/if_private.h>
58 #include <net/if_types.h>
59 #include <net/if_bridgevar.h>
60 #include <net/route.h>
61 #include <net/route/nhop.h>
62 #include <net/route/route_ctl.h>
63 #include <net/vnet.h>
64
65 #include <netinet/if_ether.h>
66 #include <netinet/in.h>
67 #include <netinet/in_fib.h>
68 #include <netinet/in_var.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip_carp.h>
72 #include <netinet/igmp_var.h>
73 #include <netinet/udp.h>
74 #include <netinet/udp_var.h>
75
76 #ifdef MAC
77 #include <security/mac/mac_framework.h>
78 #endif
79
80 static int in_aifaddr_ioctl(u_long, caddr_t, struct ifnet *, struct ucred *);
81 static int in_difaddr_ioctl(u_long, caddr_t, struct ifnet *, struct ucred *);
82 static int in_gifaddr_ioctl(u_long, caddr_t, struct ifnet *, struct ucred *);
83
84 static void in_socktrim(struct sockaddr_in *);
85 static void in_purgemaddrs(struct ifnet *);
86
87 static bool ia_need_loopback_route(const struct in_ifaddr *);
88
89 VNET_DEFINE_STATIC(int, nosameprefix);
90 #define V_nosameprefix VNET(nosameprefix)
91 SYSCTL_INT(_net_inet_ip, OID_AUTO, no_same_prefix, CTLFLAG_VNET | CTLFLAG_RW,
92 &VNET_NAME(nosameprefix), 0,
93 "Refuse to create same prefixes on different interfaces");
94
95 VNET_DEFINE_STATIC(bool, broadcast_lowest);
96 #define V_broadcast_lowest VNET(broadcast_lowest)
97 SYSCTL_BOOL(_net_inet_ip, OID_AUTO, broadcast_lowest, CTLFLAG_VNET | CTLFLAG_RW,
98 &VNET_NAME(broadcast_lowest), 0,
99 "Treat lowest address on a subnet (host 0) as broadcast");
100
101 VNET_DEFINE(bool, ip_allow_net240) = false;
102 #define V_ip_allow_net240 VNET(ip_allow_net240)
103 SYSCTL_BOOL(_net_inet_ip, OID_AUTO, allow_net240,
104 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_allow_net240), 0,
105 "Allow forwarding of and ICMP response to Experimental addresses, aka Class E (240/4)");
106 /* see https://datatracker.ietf.org/doc/draft-schoen-intarea-unicast-240 */
107
108 VNET_DEFINE(bool, ip_allow_net0) = false;
109 SYSCTL_BOOL(_net_inet_ip, OID_AUTO, allow_net0,
110 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip_allow_net0), 0,
111 "Allow forwarding of and ICMP response to addresses in network 0/8");
112 /* see https://datatracker.ietf.org/doc/draft-schoen-intarea-unicast-0 */
113
114 VNET_DEFINE(uint32_t, in_loopback_mask) = IN_LOOPBACK_MASK_DFLT;
115 #define V_in_loopback_mask VNET(in_loopback_mask)
116 static int sysctl_loopback_prefixlen(SYSCTL_HANDLER_ARGS);
117 SYSCTL_PROC(_net_inet_ip, OID_AUTO, loopback_prefixlen,
118 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
119 NULL, 0, sysctl_loopback_prefixlen, "I",
120 "Prefix length of address space reserved for loopback");
121 /* see https://datatracker.ietf.org/doc/draft-schoen-intarea-unicast-127 */
122
123 VNET_DECLARE(struct inpcbinfo, ripcbinfo);
124 #define V_ripcbinfo VNET(ripcbinfo)
125
126 static struct sx in_control_sx;
127 SX_SYSINIT(in_control_sx, &in_control_sx, "in_control");
128
129 /*
130 * Return true if an internet address is for a ``local'' host
131 * (one to which we have a connection).
132 */
133 bool
in_localaddr(struct in_addr in)134 in_localaddr(struct in_addr in)
135 {
136 u_long i = ntohl(in.s_addr);
137 struct in_ifaddr *ia;
138
139 NET_EPOCH_ASSERT();
140
141 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
142 if ((i & ia->ia_subnetmask) == ia->ia_subnet)
143 return (true);
144 }
145
146 return (false);
147 }
148
149 /*
150 * Return true if an internet address is for the local host and configured
151 * on one of its interfaces.
152 */
153 bool
in_localip(struct in_addr in)154 in_localip(struct in_addr in)
155 {
156 struct in_ifaddr *ia;
157
158 NET_EPOCH_ASSERT();
159
160 CK_LIST_FOREACH(ia, INADDR_HASH(in.s_addr), ia_hash)
161 if (IA_SIN(ia)->sin_addr.s_addr == in.s_addr)
162 return (true);
163
164 return (false);
165 }
166
167 /*
168 * Like in_localip(), but FIB-aware and carp(4)-aware.
169 */
170 bool
in_localip_fib(struct in_addr in,uint16_t fib)171 in_localip_fib(struct in_addr in, uint16_t fib)
172 {
173 struct in_ifaddr *ia;
174
175 NET_EPOCH_ASSERT();
176
177 CK_LIST_FOREACH(ia, INADDR_HASH(in.s_addr), ia_hash)
178 if (IA_SIN(ia)->sin_addr.s_addr == in.s_addr &&
179 (ia->ia_ifa.ifa_carp == NULL ||
180 carp_master_p(&ia->ia_ifa)) &&
181 ia->ia_ifa.ifa_ifp->if_fib == fib)
182 return (true);
183
184 return (false);
185 }
186
187 /*
188 * Return true if an internet address is configured on an interface.
189 */
190 bool
in_ifhasaddr(struct ifnet * ifp,struct in_addr in)191 in_ifhasaddr(struct ifnet *ifp, struct in_addr in)
192 {
193 struct ifaddr *ifa;
194 struct in_ifaddr *ia;
195
196 NET_EPOCH_ASSERT();
197
198 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
199 if (ifa->ifa_addr->sa_family != AF_INET)
200 continue;
201 ia = (struct in_ifaddr *)ifa;
202 if (ia->ia_addr.sin_addr.s_addr == in.s_addr)
203 return (true);
204 }
205
206 return (false);
207 }
208
209 /*
210 * Return the first configured IPv4 address on the given ifnet.
211 * Replaces deprecated IFP_TO_IA() from 4.3BSD.
212 * Used only by IPv4 multicast, where source-address selection
213 * has not yet been introduced.
214 *
215 * Returns ifa or NULL.
216 */
217 struct in_ifaddr *
in_ifprimaryaddr(struct ifnet * ifp)218 in_ifprimaryaddr(struct ifnet *ifp)
219 {
220 struct ifaddr *ifa;
221
222 NET_EPOCH_ASSERT();
223
224 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
225 if (ifa->ifa_addr->sa_family == AF_INET)
226 break;
227
228 return ((struct in_ifaddr *)ifa);
229 }
230
231 /*
232 * Return a reference to the interface address which is different to
233 * the supplied one but with same IP address value.
234 */
235 static struct in_ifaddr *
in_localip_more(struct in_ifaddr * original_ia)236 in_localip_more(struct in_ifaddr *original_ia)
237 {
238 struct epoch_tracker et;
239 in_addr_t original_addr = IA_SIN(original_ia)->sin_addr.s_addr;
240 uint32_t original_fib = original_ia->ia_ifa.ifa_ifp->if_fib;
241 struct in_ifaddr *ia;
242
243 NET_EPOCH_ENTER(et);
244 CK_LIST_FOREACH(ia, INADDR_HASH(original_addr), ia_hash) {
245 in_addr_t addr = IA_SIN(ia)->sin_addr.s_addr;
246 uint32_t fib = ia->ia_ifa.ifa_ifp->if_fib;
247 if (!V_rt_add_addr_allfibs && (original_fib != fib))
248 continue;
249 if ((original_ia != ia) && (original_addr == addr)) {
250 ifa_ref(&ia->ia_ifa);
251 NET_EPOCH_EXIT(et);
252 return (ia);
253 }
254 }
255 NET_EPOCH_EXIT(et);
256
257 return (NULL);
258 }
259
260 /*
261 * Tries to find first IPv4 address in the provided fib.
262 * Prefers non-loopback addresses and return loopback IFF
263 * @loopback_ok is set.
264 *
265 * Returns ifa or NULL.
266 */
267 struct in_ifaddr *
in_findlocal(uint32_t fibnum,bool loopback_ok)268 in_findlocal(uint32_t fibnum, bool loopback_ok)
269 {
270 struct in_ifaddr *ia = NULL, *ia_lo = NULL;
271
272 NET_EPOCH_ASSERT();
273
274 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
275 uint32_t ia_fib = ia->ia_ifa.ifa_ifp->if_fib;
276 if (!V_rt_add_addr_allfibs && (fibnum != ia_fib))
277 continue;
278
279 if (!IN_LOOPBACK(ntohl(IA_SIN(ia)->sin_addr.s_addr)))
280 break;
281 if (loopback_ok)
282 ia_lo = ia;
283 }
284
285 if (ia == NULL)
286 ia = ia_lo;
287
288 return (ia);
289 }
290
291 /*
292 * Determine whether an IP address is in a reserved set of addresses
293 * that may not be forwarded, or whether datagrams to that destination
294 * may be forwarded.
295 */
296 bool
in_canforward(struct in_addr in)297 in_canforward(struct in_addr in)
298 {
299 u_long i = ntohl(in.s_addr);
300
301 if (IN_MULTICAST(i) || IN_LINKLOCAL(i) || IN_LOOPBACK(i) ||
302 in_nullhost(in))
303 return (false);
304 if (IN_EXPERIMENTAL(i) && !V_ip_allow_net240)
305 return (false);
306 if (IN_ZERONET(i) && !V_ip_allow_net0)
307 return (false);
308 return (true);
309 }
310
311 /*
312 * Sysctl to manage prefix of reserved loopback network; translate
313 * to/from mask. The mask is always contiguous high-order 1 bits
314 * followed by all 0 bits.
315 */
316 static int
sysctl_loopback_prefixlen(SYSCTL_HANDLER_ARGS)317 sysctl_loopback_prefixlen(SYSCTL_HANDLER_ARGS)
318 {
319 int error, preflen;
320
321 /* ffs is 1-based; compensate. */
322 preflen = 33 - ffs(V_in_loopback_mask);
323 error = sysctl_handle_int(oidp, &preflen, 0, req);
324 if (error || !req->newptr)
325 return (error);
326 if (preflen < 8 || preflen > 31)
327 return (EINVAL);
328 V_in_loopback_mask = 0xffffffff << (32 - preflen);
329 return (0);
330 }
331
332 /*
333 * Trim a mask in a sockaddr
334 */
335 static void
in_socktrim(struct sockaddr_in * ap)336 in_socktrim(struct sockaddr_in *ap)
337 {
338 char *cplim = (char *) &ap->sin_addr;
339 char *cp = (char *) (&ap->sin_addr + 1);
340
341 ap->sin_len = 0;
342 while (--cp >= cplim)
343 if (*cp) {
344 (ap)->sin_len = cp - (char *) (ap) + 1;
345 break;
346 }
347 }
348
349 /*
350 * Generic internet control operations (ioctl's).
351 */
352 int
in_control_ioctl(u_long cmd,void * data,struct ifnet * ifp,struct ucred * cred)353 in_control_ioctl(u_long cmd, void *data, struct ifnet *ifp,
354 struct ucred *cred)
355 {
356 struct ifreq *ifr = (struct ifreq *)data;
357 struct sockaddr_in *addr = (struct sockaddr_in *)&ifr->ifr_addr;
358 struct epoch_tracker et;
359 struct ifaddr *ifa;
360 struct in_ifaddr *ia;
361 int error;
362
363 if (ifp == NULL)
364 return (EADDRNOTAVAIL);
365
366 /*
367 * Filter out 4 ioctls we implement directly. Forward the rest
368 * to specific functions and ifp->if_ioctl().
369 */
370 switch (cmd) {
371 case SIOCGIFADDR:
372 case SIOCGIFBRDADDR:
373 case SIOCGIFDSTADDR:
374 case SIOCGIFNETMASK:
375 break;
376 case SIOCGIFALIAS:
377 sx_xlock(&in_control_sx);
378 error = in_gifaddr_ioctl(cmd, data, ifp, cred);
379 sx_xunlock(&in_control_sx);
380 return (error);
381 case SIOCDIFADDR:
382 sx_xlock(&in_control_sx);
383 error = in_difaddr_ioctl(cmd, data, ifp, cred);
384 sx_xunlock(&in_control_sx);
385 return (error);
386 case OSIOCAIFADDR: /* 9.x compat */
387 case SIOCAIFADDR:
388 sx_xlock(&in_control_sx);
389 error = in_aifaddr_ioctl(cmd, data, ifp, cred);
390 sx_xunlock(&in_control_sx);
391 return (error);
392 case SIOCSIFADDR:
393 case SIOCSIFBRDADDR:
394 case SIOCSIFDSTADDR:
395 case SIOCSIFNETMASK:
396 /* We no longer support that old commands. */
397 return (EINVAL);
398 default:
399 if (ifp->if_ioctl == NULL)
400 return (EOPNOTSUPP);
401 return ((*ifp->if_ioctl)(ifp, cmd, data));
402 }
403
404 if (addr->sin_addr.s_addr != INADDR_ANY &&
405 prison_check_ip4(cred, &addr->sin_addr) != 0)
406 return (EADDRNOTAVAIL);
407
408 /*
409 * Find address for this interface, if it exists. If an
410 * address was specified, find that one instead of the
411 * first one on the interface, if possible.
412 */
413 NET_EPOCH_ENTER(et);
414 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
415 if (ifa->ifa_addr->sa_family != AF_INET)
416 continue;
417 ia = (struct in_ifaddr *)ifa;
418 if (ia->ia_addr.sin_addr.s_addr == addr->sin_addr.s_addr)
419 break;
420 }
421 if (ifa == NULL)
422 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
423 if (ifa->ifa_addr->sa_family == AF_INET) {
424 ia = (struct in_ifaddr *)ifa;
425 if (prison_check_ip4(cred,
426 &ia->ia_addr.sin_addr) == 0)
427 break;
428 }
429
430 if (ifa == NULL) {
431 NET_EPOCH_EXIT(et);
432 return (EADDRNOTAVAIL);
433 }
434
435 error = 0;
436 switch (cmd) {
437 case SIOCGIFADDR:
438 *addr = ia->ia_addr;
439 break;
440
441 case SIOCGIFBRDADDR:
442 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
443 error = EINVAL;
444 break;
445 }
446 *addr = ia->ia_broadaddr;
447 break;
448
449 case SIOCGIFDSTADDR:
450 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
451 error = EINVAL;
452 break;
453 }
454 *addr = ia->ia_dstaddr;
455 break;
456
457 case SIOCGIFNETMASK:
458 *addr = ia->ia_sockmask;
459 break;
460 }
461
462 NET_EPOCH_EXIT(et);
463
464 return (error);
465 }
466
467 int
in_mask2len(struct in_addr * mask)468 in_mask2len(struct in_addr *mask)
469 {
470 int x, y;
471 u_char *p;
472
473 p = (u_char *)mask;
474 for (x = 0; x < sizeof(*mask); x++) {
475 if (p[x] != 0xff)
476 break;
477 }
478 y = 0;
479 if (x < sizeof(*mask)) {
480 for (y = 0; y < 8; y++) {
481 if ((p[x] & (0x80 >> y)) == 0)
482 break;
483 }
484 }
485 return (x * 8 + y);
486 }
487
488 int
in_control(struct socket * so,u_long cmd,void * data,struct ifnet * ifp,struct thread * td)489 in_control(struct socket *so, u_long cmd, void *data, struct ifnet *ifp,
490 struct thread *td)
491 {
492 return (in_control_ioctl(cmd, data, ifp, td ? td->td_ucred : NULL));
493 }
494
495 static int
in_aifaddr_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp,struct ucred * cred)496 in_aifaddr_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp, struct ucred *cred)
497 {
498 const struct in_aliasreq *ifra = (struct in_aliasreq *)data;
499 const struct sockaddr_in *addr = &ifra->ifra_addr;
500 const struct sockaddr_in *broadaddr = &ifra->ifra_broadaddr;
501 const struct sockaddr_in *mask = &ifra->ifra_mask;
502 const struct sockaddr_in *dstaddr = &ifra->ifra_dstaddr;
503 const int vhid = (cmd == SIOCAIFADDR) ? ifra->ifra_vhid : 0;
504 struct epoch_tracker et;
505 struct ifaddr *ifa;
506 struct in_ifaddr *ia;
507 bool iaIsFirst;
508 int error = 0;
509
510 error = priv_check_cred(cred, PRIV_NET_ADDIFADDR);
511 if (error)
512 return (error);
513
514 /*
515 * ifra_addr must be present and be of INET family.
516 * ifra_broadaddr/ifra_dstaddr and ifra_mask are optional.
517 */
518 if (addr->sin_len != sizeof(struct sockaddr_in) ||
519 addr->sin_family != AF_INET)
520 return (EINVAL);
521 if (broadaddr->sin_len != 0 &&
522 (broadaddr->sin_len != sizeof(struct sockaddr_in) ||
523 broadaddr->sin_family != AF_INET))
524 return (EINVAL);
525 if (mask->sin_len != 0 &&
526 (mask->sin_len != sizeof(struct sockaddr_in) ||
527 mask->sin_family != AF_INET))
528 return (EINVAL);
529 if ((ifp->if_flags & IFF_POINTOPOINT) &&
530 (dstaddr->sin_len != sizeof(struct sockaddr_in) ||
531 dstaddr->sin_addr.s_addr == INADDR_ANY))
532 return (EDESTADDRREQ);
533 if (vhid != 0 && carp_attach_p == NULL)
534 return (EPROTONOSUPPORT);
535
536 #ifdef MAC
537 /* Check if a MAC policy disallows setting the IPv4 address. */
538 error = mac_inet_check_add_addr(cred, &addr->sin_addr, ifp);
539 if (error != 0)
540 return (error);
541 #endif
542
543 /*
544 * Check if bridge wants to allow adding addrs to member interfaces.
545 */
546 if (ifp->if_bridge != NULL && ifp->if_type != IFT_GIF &&
547 bridge_member_ifaddrs_p != NULL) {
548 if (bridge_member_ifaddrs_p())
549 if_printf(ifp, "WARNING: Assigning an IP address to "
550 "an interface which is also a bridge member is "
551 "deprecated and will be unsupported in a future "
552 "release.\n");
553 else
554 return (EINVAL);
555 }
556
557 /*
558 * See whether address already exist.
559 */
560 iaIsFirst = true;
561 ia = NULL;
562 NET_EPOCH_ENTER(et);
563 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
564 struct in_ifaddr *it;
565
566 if (ifa->ifa_addr->sa_family != AF_INET)
567 continue;
568
569 it = (struct in_ifaddr *)ifa;
570 if (it->ia_addr.sin_addr.s_addr == addr->sin_addr.s_addr &&
571 prison_check_ip4(cred, &addr->sin_addr) == 0)
572 ia = it;
573 else
574 iaIsFirst = false;
575 }
576 NET_EPOCH_EXIT(et);
577
578 if (ia != NULL)
579 (void )in_difaddr_ioctl(cmd, data, ifp, cred);
580
581 ifa = ifa_alloc(sizeof(struct in_ifaddr), M_WAITOK);
582 ia = (struct in_ifaddr *)ifa;
583 ifa->ifa_addr = (struct sockaddr *)&ia->ia_addr;
584 ifa->ifa_dstaddr = (struct sockaddr *)&ia->ia_dstaddr;
585 ifa->ifa_netmask = (struct sockaddr *)&ia->ia_sockmask;
586 callout_init_rw(&ia->ia_garp_timer, &ifp->if_addr_lock,
587 CALLOUT_RETURNUNLOCKED);
588
589 ia->ia_ifp = ifp;
590 ia->ia_addr = *addr;
591 if (mask->sin_len != 0) {
592 ia->ia_sockmask = *mask;
593 ia->ia_subnetmask = ntohl(ia->ia_sockmask.sin_addr.s_addr);
594 } else {
595 in_addr_t i = ntohl(addr->sin_addr.s_addr);
596
597 /*
598 * If netmask isn't supplied, use historical default.
599 * This is deprecated for interfaces other than loopback
600 * or point-to-point; warn in other cases. In the future
601 * we should return an error rather than warning.
602 */
603 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0)
604 printf("%s: set address: WARNING: network mask "
605 "should be specified; using historical default\n",
606 ifp->if_xname);
607 if (IN_CLASSA(i))
608 ia->ia_subnetmask = IN_CLASSA_NET;
609 else if (IN_CLASSB(i))
610 ia->ia_subnetmask = IN_CLASSB_NET;
611 else
612 ia->ia_subnetmask = IN_CLASSC_NET;
613 ia->ia_sockmask.sin_addr.s_addr = htonl(ia->ia_subnetmask);
614 }
615 ia->ia_subnet = ntohl(addr->sin_addr.s_addr) & ia->ia_subnetmask;
616 in_socktrim(&ia->ia_sockmask);
617
618 if (ifp->if_flags & IFF_BROADCAST) {
619 if (broadaddr->sin_len != 0) {
620 ia->ia_broadaddr = *broadaddr;
621 } else if (ia->ia_subnetmask == IN_RFC3021_MASK) {
622 ia->ia_broadaddr.sin_addr.s_addr = INADDR_BROADCAST;
623 ia->ia_broadaddr.sin_len = sizeof(struct sockaddr_in);
624 ia->ia_broadaddr.sin_family = AF_INET;
625 } else {
626 ia->ia_broadaddr.sin_addr.s_addr =
627 htonl(ia->ia_subnet | ~ia->ia_subnetmask);
628 ia->ia_broadaddr.sin_len = sizeof(struct sockaddr_in);
629 ia->ia_broadaddr.sin_family = AF_INET;
630 }
631 }
632
633 if (ifp->if_flags & IFF_POINTOPOINT)
634 ia->ia_dstaddr = *dstaddr;
635
636 if (vhid != 0) {
637 error = (*carp_attach_p)(&ia->ia_ifa, vhid);
638 if (error)
639 return (error);
640 }
641
642 /* if_addrhead is already referenced by ifa_alloc() */
643 IF_ADDR_WLOCK(ifp);
644 CK_STAILQ_INSERT_TAIL(&ifp->if_addrhead, ifa, ifa_link);
645 IF_ADDR_WUNLOCK(ifp);
646
647 ifa_ref(ifa); /* in_ifaddrhead */
648 sx_assert(&in_control_sx, SA_XLOCKED);
649 CK_STAILQ_INSERT_TAIL(&V_in_ifaddrhead, ia, ia_link);
650 CK_LIST_INSERT_HEAD(INADDR_HASH(ia->ia_addr.sin_addr.s_addr), ia,
651 ia_hash);
652
653 /*
654 * Give the interface a chance to initialize
655 * if this is its first address,
656 * and to validate the address if necessary.
657 */
658 if (ifp->if_ioctl != NULL) {
659 error = (*ifp->if_ioctl)(ifp, SIOCSIFADDR, (caddr_t)ia);
660 if (error)
661 goto fail1;
662 }
663
664 /*
665 * Add route for the network.
666 */
667 if (vhid == 0) {
668 error = in_addprefix(ia);
669 if (error)
670 goto fail1;
671 }
672
673 /*
674 * Add a loopback route to self.
675 */
676 if (vhid == 0 && ia_need_loopback_route(ia)) {
677 struct in_ifaddr *eia;
678
679 eia = in_localip_more(ia);
680
681 if (eia == NULL) {
682 error = ifa_add_loopback_route((struct ifaddr *)ia,
683 (struct sockaddr *)&ia->ia_addr);
684 if (error)
685 goto fail2;
686 } else
687 ifa_free(&eia->ia_ifa);
688 }
689
690 if (iaIsFirst && (ifp->if_flags & IFF_MULTICAST)) {
691 struct in_addr allhosts_addr;
692 struct in_ifinfo *ii;
693
694 ii = (struct in_ifinfo *)ifp->if_inet;
695 allhosts_addr.s_addr = htonl(INADDR_ALLHOSTS_GROUP);
696
697 error = in_joingroup(ifp, &allhosts_addr, NULL,
698 &ii->ii_allhosts);
699 }
700
701 /*
702 * Note: we don't need extra reference for ifa, since we called
703 * with sx lock held, and ifaddr can not be deleted in concurrent
704 * thread.
705 */
706 EVENTHANDLER_INVOKE(ifaddr_event_ext, ifp, ifa, IFADDR_EVENT_ADD);
707
708 return (error);
709
710 fail2:
711 if (vhid == 0)
712 (void )in_scrubprefix(ia, LLE_STATIC);
713
714 fail1:
715 if (ia->ia_ifa.ifa_carp)
716 (*carp_detach_p)(&ia->ia_ifa, false);
717
718 IF_ADDR_WLOCK(ifp);
719 CK_STAILQ_REMOVE(&ifp->if_addrhead, &ia->ia_ifa, ifaddr, ifa_link);
720 IF_ADDR_WUNLOCK(ifp);
721 ifa_free(&ia->ia_ifa); /* if_addrhead */
722
723 sx_assert(&in_control_sx, SA_XLOCKED);
724 CK_STAILQ_REMOVE(&V_in_ifaddrhead, ia, in_ifaddr, ia_link);
725 CK_LIST_REMOVE(ia, ia_hash);
726 ifa_free(&ia->ia_ifa); /* in_ifaddrhead */
727
728 return (error);
729 }
730
731 static int
in_difaddr_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp,struct ucred * cred)732 in_difaddr_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp, struct ucred *cred)
733 {
734 const struct ifreq *ifr = (struct ifreq *)data;
735 const struct sockaddr_in *addr = (const struct sockaddr_in *)
736 &ifr->ifr_addr;
737 struct ifaddr *ifa;
738 struct in_ifaddr *ia;
739 bool deleteAny, iaIsLast;
740 int error;
741
742 if (cred != NULL) {
743 error = priv_check_cred(cred, PRIV_NET_DELIFADDR);
744 if (error)
745 return (error);
746 }
747
748 if (addr->sin_len != sizeof(struct sockaddr_in) ||
749 addr->sin_family != AF_INET)
750 deleteAny = true;
751 else
752 deleteAny = false;
753
754 iaIsLast = true;
755 ia = NULL;
756 IF_ADDR_WLOCK(ifp);
757 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
758 struct in_ifaddr *it;
759
760 if (ifa->ifa_addr->sa_family != AF_INET)
761 continue;
762
763 it = (struct in_ifaddr *)ifa;
764 if (deleteAny && ia == NULL && (cred == NULL ||
765 prison_check_ip4(cred, &it->ia_addr.sin_addr) == 0))
766 ia = it;
767
768 if (it->ia_addr.sin_addr.s_addr == addr->sin_addr.s_addr &&
769 (cred == NULL || prison_check_ip4(cred,
770 &addr->sin_addr) == 0))
771 ia = it;
772
773 if (it != ia)
774 iaIsLast = false;
775 }
776
777 if (ia == NULL) {
778 IF_ADDR_WUNLOCK(ifp);
779 return (EADDRNOTAVAIL);
780 }
781
782 CK_STAILQ_REMOVE(&ifp->if_addrhead, &ia->ia_ifa, ifaddr, ifa_link);
783 IF_ADDR_WUNLOCK(ifp);
784 ifa_free(&ia->ia_ifa); /* if_addrhead */
785
786 sx_assert(&in_control_sx, SA_XLOCKED);
787 CK_STAILQ_REMOVE(&V_in_ifaddrhead, ia, in_ifaddr, ia_link);
788 CK_LIST_REMOVE(ia, ia_hash);
789
790 /*
791 * in_scrubprefix() kills the interface route.
792 */
793 in_scrubprefix(ia, LLE_STATIC);
794
795 /*
796 * in_ifadown gets rid of all the rest of
797 * the routes. This is not quite the right
798 * thing to do, but at least if we are running
799 * a routing process they will come back.
800 */
801 in_ifadown(&ia->ia_ifa, 1);
802
803 if (ia->ia_ifa.ifa_carp)
804 (*carp_detach_p)(&ia->ia_ifa, cmd == SIOCAIFADDR);
805
806 /*
807 * If this is the last IPv4 address configured on this
808 * interface, leave the all-hosts group.
809 * No state-change report need be transmitted.
810 */
811 if (iaIsLast && (ifp->if_flags & IFF_MULTICAST)) {
812 struct in_ifinfo *ii;
813
814 ii = (struct in_ifinfo *)ifp->if_inet;
815 if (ii->ii_allhosts) {
816 (void)in_leavegroup(ii->ii_allhosts, NULL);
817 ii->ii_allhosts = NULL;
818 }
819 }
820
821 IF_ADDR_WLOCK(ifp);
822 if (callout_stop(&ia->ia_garp_timer) == 1) {
823 ifa_free(&ia->ia_ifa);
824 }
825 IF_ADDR_WUNLOCK(ifp);
826
827 EVENTHANDLER_INVOKE(ifaddr_event_ext, ifp, &ia->ia_ifa,
828 IFADDR_EVENT_DEL);
829 ifa_free(&ia->ia_ifa); /* in_ifaddrhead */
830
831 return (0);
832 }
833
834 static int
in_gifaddr_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp,struct ucred * cred)835 in_gifaddr_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp, struct ucred *cred)
836 {
837 struct in_aliasreq *ifra = (struct in_aliasreq *)data;
838 const struct sockaddr_in *addr = &ifra->ifra_addr;
839 struct epoch_tracker et;
840 struct ifaddr *ifa;
841 struct in_ifaddr *ia;
842
843 /*
844 * ifra_addr must be present and be of INET family.
845 */
846 if (addr->sin_len != sizeof(struct sockaddr_in) ||
847 addr->sin_family != AF_INET)
848 return (EINVAL);
849
850 /*
851 * See whether address exist.
852 */
853 ia = NULL;
854 NET_EPOCH_ENTER(et);
855 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
856 struct in_ifaddr *it;
857
858 if (ifa->ifa_addr->sa_family != AF_INET)
859 continue;
860
861 it = (struct in_ifaddr *)ifa;
862 if (it->ia_addr.sin_addr.s_addr == addr->sin_addr.s_addr &&
863 prison_check_ip4(cred, &addr->sin_addr) == 0) {
864 ia = it;
865 break;
866 }
867 }
868 if (ia == NULL) {
869 NET_EPOCH_EXIT(et);
870 return (EADDRNOTAVAIL);
871 }
872
873 ifra->ifra_mask = ia->ia_sockmask;
874 if ((ifp->if_flags & IFF_POINTOPOINT) &&
875 ia->ia_dstaddr.sin_family == AF_INET)
876 ifra->ifra_dstaddr = ia->ia_dstaddr;
877 else if ((ifp->if_flags & IFF_BROADCAST) &&
878 ia->ia_broadaddr.sin_family == AF_INET)
879 ifra->ifra_broadaddr = ia->ia_broadaddr;
880 else
881 memset(&ifra->ifra_broadaddr, 0,
882 sizeof(ifra->ifra_broadaddr));
883
884 NET_EPOCH_EXIT(et);
885 return (0);
886 }
887
888 static int
in_match_ifaddr(const struct rtentry * rt,const struct nhop_object * nh,void * arg)889 in_match_ifaddr(const struct rtentry *rt, const struct nhop_object *nh, void *arg)
890 {
891
892 if (nh->nh_ifa == (struct ifaddr *)arg)
893 return (1);
894
895 return (0);
896 }
897
898 static int
in_handle_prefix_route(uint32_t fibnum,int cmd,struct sockaddr_in * dst,struct sockaddr_in * netmask,struct ifaddr * ifa,struct ifnet * ifp)899 in_handle_prefix_route(uint32_t fibnum, int cmd,
900 struct sockaddr_in *dst, struct sockaddr_in *netmask, struct ifaddr *ifa,
901 struct ifnet *ifp)
902 {
903
904 NET_EPOCH_ASSERT();
905
906 /* Prepare gateway */
907 struct sockaddr_dl_short sdl = {
908 .sdl_family = AF_LINK,
909 .sdl_len = sizeof(struct sockaddr_dl_short),
910 .sdl_type = ifa->ifa_ifp->if_type,
911 .sdl_index = ifa->ifa_ifp->if_index,
912 };
913
914 struct rt_addrinfo info = {
915 .rti_ifa = ifa,
916 .rti_ifp = ifp,
917 .rti_flags = RTF_PINNED | ((netmask != NULL) ? 0 : RTF_HOST),
918 .rti_info = {
919 [RTAX_DST] = (struct sockaddr *)dst,
920 [RTAX_NETMASK] = (struct sockaddr *)netmask,
921 [RTAX_GATEWAY] = (struct sockaddr *)&sdl,
922 },
923 /* Ensure we delete the prefix IFF prefix ifa matches */
924 .rti_filter = in_match_ifaddr,
925 .rti_filterdata = ifa,
926 };
927
928 return (rib_handle_ifaddr_info(fibnum, cmd, &info));
929 }
930
931 /*
932 * Routing table interaction with interface addresses.
933 *
934 * In general, two types of routes needs to be installed:
935 * a) "interface" or "prefix" route, telling user that the addresses
936 * behind the ifa prefix are reached directly.
937 * b) "loopback" route installed for the ifa address, telling user that
938 * the address belongs to local system.
939 *
940 * Handling for (a) and (b) differs in multi-fib aspects, hence they
941 * are implemented in different functions below.
942 *
943 * The cases above may intersect - /32 interface aliases results in
944 * the same prefix produced by (a) and (b). This blurs the definition
945 * of the "loopback" route and complicate interactions. The interaction
946 * table is defined below. The case numbers are used in the multiple
947 * functions below to refer to the particular test case.
948 *
949 * There can be multiple options:
950 * 1) Adding address with prefix on non-p2p/non-loopback interface.
951 * Example: 192.0.2.1/24. Action:
952 * * add "prefix" route towards 192.0.2.0/24 via @ia interface,
953 * using @ia as an address source.
954 * * add "loopback" route towards 192.0.2.1 via V_loif, saving
955 * @ia ifp in the gateway and using @ia as an address source.
956 *
957 * 2) Adding address with /32 mask to non-p2p/non-loopback interface.
958 * Example: 192.0.2.2/32. Action:
959 * * add "prefix" host route via V_loif, using @ia as an address source.
960 *
961 * 3) Adding address with or without prefix to p2p interface.
962 * Example: 10.0.0.1/24->10.0.0.2. Action:
963 * * add "prefix" host route towards 10.0.0.2 via this interface, using @ia
964 * as an address source. Note: no sense in installing full /24 as the interface
965 * is point-to-point.
966 * * add "loopback" route towards 10.0.9.1 via V_loif, saving
967 * @ia ifp in the gateway and using @ia as an address source.
968 *
969 * 4) Adding address with or without prefix to loopback interface.
970 * Example: 192.0.2.1/24. Action:
971 * * add "prefix" host route via @ia interface, using @ia as an address source.
972 * Note: Skip installing /24 prefix as it would introduce TTL loop
973 * for the traffic destined to these addresses.
974 */
975
976 /*
977 * Checks if @ia needs to install loopback route to @ia address via
978 * ifa_maintain_loopback_route().
979 *
980 * Return true on success.
981 */
982 static bool
ia_need_loopback_route(const struct in_ifaddr * ia)983 ia_need_loopback_route(const struct in_ifaddr *ia)
984 {
985 struct ifnet *ifp = ia->ia_ifp;
986
987 /* Case 4: Skip loopback interfaces */
988 if ((ifp->if_flags & IFF_LOOPBACK) ||
989 (ia->ia_addr.sin_addr.s_addr == INADDR_ANY))
990 return (false);
991
992 /* Clash avoidance: Skip p2p interfaces with both addresses are equal */
993 if ((ifp->if_flags & IFF_POINTOPOINT) &&
994 ia->ia_dstaddr.sin_addr.s_addr == ia->ia_addr.sin_addr.s_addr)
995 return (false);
996
997 /* Case 2: skip /32 prefixes */
998 if (!(ifp->if_flags & IFF_POINTOPOINT) &&
999 (ia->ia_sockmask.sin_addr.s_addr == INADDR_BROADCAST))
1000 return (false);
1001
1002 return (true);
1003 }
1004
1005 /*
1006 * Calculate "prefix" route corresponding to @ia.
1007 */
1008 static void
ia_getrtprefix(const struct in_ifaddr * ia,struct in_addr * prefix,struct in_addr * mask)1009 ia_getrtprefix(const struct in_ifaddr *ia, struct in_addr *prefix, struct in_addr *mask)
1010 {
1011
1012 if (ia->ia_ifp->if_flags & IFF_POINTOPOINT) {
1013 /* Case 3: return host route for dstaddr */
1014 *prefix = ia->ia_dstaddr.sin_addr;
1015 mask->s_addr = INADDR_BROADCAST;
1016 } else if (ia->ia_ifp->if_flags & IFF_LOOPBACK) {
1017 /* Case 4: return host route for ifaddr */
1018 *prefix = ia->ia_addr.sin_addr;
1019 mask->s_addr = INADDR_BROADCAST;
1020 } else {
1021 /* Cases 1,2: return actual ia prefix */
1022 *prefix = ia->ia_addr.sin_addr;
1023 *mask = ia->ia_sockmask.sin_addr;
1024 prefix->s_addr &= mask->s_addr;
1025 }
1026 }
1027
1028 /*
1029 * Adds or delete interface "prefix" route corresponding to @ifa.
1030 * Returns 0 on success or errno.
1031 */
1032 static int
in_handle_ifaddr_route(int cmd,struct in_ifaddr * ia)1033 in_handle_ifaddr_route(int cmd, struct in_ifaddr *ia)
1034 {
1035 struct ifaddr *ifa = &ia->ia_ifa;
1036 struct in_addr daddr, maddr;
1037 struct sockaddr_in *pmask;
1038 struct epoch_tracker et;
1039 int error;
1040
1041 ia_getrtprefix(ia, &daddr, &maddr);
1042
1043 struct sockaddr_in mask = {
1044 .sin_family = AF_INET,
1045 .sin_len = sizeof(struct sockaddr_in),
1046 .sin_addr = maddr,
1047 };
1048
1049 pmask = (maddr.s_addr != INADDR_BROADCAST) ? &mask : NULL;
1050
1051 struct sockaddr_in dst = {
1052 .sin_family = AF_INET,
1053 .sin_len = sizeof(struct sockaddr_in),
1054 .sin_addr.s_addr = daddr.s_addr & maddr.s_addr,
1055 };
1056
1057 struct ifnet *ifp = ia->ia_ifp;
1058
1059 if ((maddr.s_addr == INADDR_BROADCAST) &&
1060 (!(ia->ia_ifp->if_flags & (IFF_POINTOPOINT|IFF_LOOPBACK)))) {
1061 /* Case 2: host route on broadcast interface */
1062 ifp = V_loif;
1063 }
1064
1065 uint32_t fibnum = ifa->ifa_ifp->if_fib;
1066 NET_EPOCH_ENTER(et);
1067 error = in_handle_prefix_route(fibnum, cmd, &dst, pmask, ifa, ifp);
1068 NET_EPOCH_EXIT(et);
1069
1070 return (error);
1071 }
1072
1073 /*
1074 * Check if we have a route for the given prefix already.
1075 */
1076 static bool
in_hasrtprefix(struct in_ifaddr * target)1077 in_hasrtprefix(struct in_ifaddr *target)
1078 {
1079 struct epoch_tracker et;
1080 struct in_ifaddr *ia;
1081 struct in_addr prefix, mask, p, m;
1082 bool result = false;
1083
1084 ia_getrtprefix(target, &prefix, &mask);
1085
1086 /* Look for an existing address with the same prefix, mask, and fib */
1087 NET_EPOCH_ENTER(et);
1088 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1089 ia_getrtprefix(ia, &p, &m);
1090
1091 if (prefix.s_addr != p.s_addr ||
1092 mask.s_addr != m.s_addr)
1093 continue;
1094
1095 if (target->ia_ifp->if_fib != ia->ia_ifp->if_fib)
1096 continue;
1097
1098 /*
1099 * If we got a matching prefix route inserted by other
1100 * interface address, we are done here.
1101 */
1102 if (ia->ia_flags & IFA_ROUTE) {
1103 result = true;
1104 break;
1105 }
1106 }
1107 NET_EPOCH_EXIT(et);
1108
1109 return (result);
1110 }
1111
1112 int
in_addprefix(struct in_ifaddr * target)1113 in_addprefix(struct in_ifaddr *target)
1114 {
1115 int error;
1116
1117 if (in_hasrtprefix(target)) {
1118 if (V_nosameprefix)
1119 return (EEXIST);
1120 else {
1121 rt_addrmsg(RTM_ADD, &target->ia_ifa,
1122 target->ia_ifp->if_fib);
1123 return (0);
1124 }
1125 }
1126
1127 /*
1128 * No-one seem to have this prefix route, so we try to insert it.
1129 */
1130 rt_addrmsg(RTM_ADD, &target->ia_ifa, target->ia_ifp->if_fib);
1131 error = in_handle_ifaddr_route(RTM_ADD, target);
1132 if (!error)
1133 target->ia_flags |= IFA_ROUTE;
1134 return (error);
1135 }
1136
1137 /*
1138 * Removes either all lle entries for given @ia, or lle
1139 * corresponding to @ia address.
1140 */
1141 static void
in_scrubprefixlle(struct in_ifaddr * ia,int all,u_int flags)1142 in_scrubprefixlle(struct in_ifaddr *ia, int all, u_int flags)
1143 {
1144 struct sockaddr_in addr, mask;
1145 struct sockaddr *saddr, *smask;
1146 struct ifnet *ifp;
1147
1148 saddr = (struct sockaddr *)&addr;
1149 bzero(&addr, sizeof(addr));
1150 addr.sin_len = sizeof(addr);
1151 addr.sin_family = AF_INET;
1152 smask = (struct sockaddr *)&mask;
1153 bzero(&mask, sizeof(mask));
1154 mask.sin_len = sizeof(mask);
1155 mask.sin_family = AF_INET;
1156 mask.sin_addr.s_addr = ia->ia_subnetmask;
1157 ifp = ia->ia_ifp;
1158
1159 if (all) {
1160 /*
1161 * Remove all L2 entries matching given prefix.
1162 * Convert address to host representation to avoid
1163 * doing this on every callback. ia_subnetmask is already
1164 * stored in host representation.
1165 */
1166 addr.sin_addr.s_addr = ntohl(ia->ia_addr.sin_addr.s_addr);
1167 lltable_prefix_free(AF_INET, saddr, smask, flags);
1168 } else {
1169 /* Remove interface address only */
1170 addr.sin_addr.s_addr = ia->ia_addr.sin_addr.s_addr;
1171 lltable_delete_addr(LLTABLE(ifp), LLE_IFADDR, saddr);
1172 }
1173 }
1174
1175 /*
1176 * If there is no other address in the system that can serve a route to the
1177 * same prefix, remove the route. Hand over the route to the new address
1178 * otherwise.
1179 */
1180 int
in_scrubprefix(struct in_ifaddr * target,u_int flags)1181 in_scrubprefix(struct in_ifaddr *target, u_int flags)
1182 {
1183 struct epoch_tracker et;
1184 struct in_ifaddr *ia;
1185 struct in_addr prefix, mask, p, m;
1186 int error = 0;
1187
1188 /*
1189 * Remove the loopback route to the interface address.
1190 */
1191 if (ia_need_loopback_route(target) && (flags & LLE_STATIC)) {
1192 struct in_ifaddr *eia;
1193
1194 eia = in_localip_more(target);
1195
1196 if (eia != NULL) {
1197 error = ifa_switch_loopback_route((struct ifaddr *)eia,
1198 (struct sockaddr *)&target->ia_addr);
1199 ifa_free(&eia->ia_ifa);
1200 } else {
1201 error = ifa_del_loopback_route((struct ifaddr *)target,
1202 (struct sockaddr *)&target->ia_addr);
1203 }
1204 }
1205
1206 ia_getrtprefix(target, &prefix, &mask);
1207
1208 if ((target->ia_flags & IFA_ROUTE) == 0) {
1209 rt_addrmsg(RTM_DELETE, &target->ia_ifa, target->ia_ifp->if_fib);
1210
1211 /*
1212 * Removing address from !IFF_UP interface or
1213 * prefix which exists on other interface (along with route).
1214 * No entries should exist here except target addr.
1215 * Given that, delete this entry only.
1216 */
1217 in_scrubprefixlle(target, 0, flags);
1218 return (0);
1219 }
1220
1221 NET_EPOCH_ENTER(et);
1222 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1223 ia_getrtprefix(ia, &p, &m);
1224
1225 if (prefix.s_addr != p.s_addr ||
1226 mask.s_addr != m.s_addr)
1227 continue;
1228
1229 if ((ia->ia_ifp->if_flags & IFF_UP) == 0)
1230 continue;
1231
1232 /*
1233 * If we got a matching prefix address, move IFA_ROUTE and
1234 * the route itself to it. Make sure that routing daemons
1235 * get a heads-up.
1236 */
1237 if ((ia->ia_flags & IFA_ROUTE) == 0) {
1238 ifa_ref(&ia->ia_ifa);
1239 NET_EPOCH_EXIT(et);
1240 error = in_handle_ifaddr_route(RTM_DELETE, target);
1241 if (error == 0)
1242 target->ia_flags &= ~IFA_ROUTE;
1243 else
1244 log(LOG_INFO, "in_scrubprefix: err=%d, old prefix delete failed\n",
1245 error);
1246 /* Scrub all entries IFF interface is different */
1247 in_scrubprefixlle(target, target->ia_ifp != ia->ia_ifp,
1248 flags);
1249 error = in_handle_ifaddr_route(RTM_ADD, ia);
1250 if (error == 0)
1251 ia->ia_flags |= IFA_ROUTE;
1252 else
1253 log(LOG_INFO, "in_scrubprefix: err=%d, new prefix add failed\n",
1254 error);
1255 ifa_free(&ia->ia_ifa);
1256 return (error);
1257 }
1258 }
1259 NET_EPOCH_EXIT(et);
1260
1261 /*
1262 * remove all L2 entries on the given prefix
1263 */
1264 in_scrubprefixlle(target, 1, flags);
1265
1266 /*
1267 * As no-one seem to have this prefix, we can remove the route.
1268 */
1269 rt_addrmsg(RTM_DELETE, &target->ia_ifa, target->ia_ifp->if_fib);
1270 error = in_handle_ifaddr_route(RTM_DELETE, target);
1271 if (error == 0)
1272 target->ia_flags &= ~IFA_ROUTE;
1273 else
1274 log(LOG_INFO, "in_scrubprefix: err=%d, prefix delete failed\n", error);
1275 return (error);
1276 }
1277
1278 void
in_ifscrub_all(void)1279 in_ifscrub_all(void)
1280 {
1281 struct ifnet *ifp;
1282 struct ifaddr *ifa, *nifa;
1283 struct ifreq ifr;
1284
1285 IFNET_RLOCK();
1286 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1287 /* Cannot lock here - lock recursion. */
1288 /* NET_EPOCH_ENTER(et); */
1289 CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, nifa) {
1290 if (ifa->ifa_addr->sa_family != AF_INET)
1291 continue;
1292
1293 /*
1294 * This is ugly but the only way for legacy IP to
1295 * cleanly remove addresses and everything attached.
1296 */
1297 bzero(&ifr, sizeof(ifr));
1298 ifr.ifr_addr = *ifa->ifa_addr;
1299 (void)in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr,
1300 ifp, NULL);
1301 }
1302 /* NET_EPOCH_EXIT(et); */
1303 in_purgemaddrs(ifp);
1304 igmp_domifdetach(ifp);
1305 }
1306 IFNET_RUNLOCK();
1307 }
1308
1309 bool
in_ifaddr_broadcast(struct in_addr in,struct in_ifaddr * ia)1310 in_ifaddr_broadcast(struct in_addr in, struct in_ifaddr *ia)
1311 {
1312
1313 return ((in.s_addr == ia->ia_broadaddr.sin_addr.s_addr ||
1314 /*
1315 * Optionally check for old-style (host 0) broadcast, but
1316 * taking into account that RFC 3021 obsoletes it.
1317 */
1318 __predict_false(V_broadcast_lowest &&
1319 ia->ia_subnetmask != IN_RFC3021_MASK &&
1320 ntohl(in.s_addr) == ia->ia_subnet)) &&
1321 /*
1322 * Check for an all one subnetmask. These
1323 * only exist when an interface gets a secondary
1324 * address.
1325 */
1326 ia->ia_subnetmask != (u_long)0xffffffff);
1327 }
1328
1329 /*
1330 * Return true if the address might be a local broadcast address.
1331 */
1332 bool
in_ifnet_broadcast(struct in_addr in,struct ifnet * ifp)1333 in_ifnet_broadcast(struct in_addr in, struct ifnet *ifp)
1334 {
1335 struct ifaddr *ifa;
1336
1337 NET_EPOCH_ASSERT();
1338
1339 if (in_broadcast(in))
1340 return (true);
1341 if ((ifp->if_flags & IFF_BROADCAST) == 0)
1342 return (false);
1343 /*
1344 * Look through the list of addresses for a match
1345 * with a broadcast address.
1346 */
1347 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1348 if (ifa->ifa_addr->sa_family == AF_INET &&
1349 in_ifaddr_broadcast(in, (struct in_ifaddr *)ifa))
1350 return (true);
1351 return (false);
1352 }
1353
1354
1355 static struct lltable *in_lltattach(struct ifnet *);
1356 void
in_ifattach(void * arg __unused,struct ifnet * ifp)1357 in_ifattach(void *arg __unused, struct ifnet *ifp)
1358 {
1359 struct in_ifinfo *ii;
1360
1361 ii = malloc(sizeof(struct in_ifinfo), M_IFADDR, M_WAITOK|M_ZERO);
1362 ii->ii_llt = in_lltattach(ifp);
1363 ii->ii_igmp = igmp_domifattach(ifp);
1364 ifp->if_inet = ii;
1365 }
1366 EVENTHANDLER_DEFINE(ifnet_arrival_event, in_ifattach, NULL,
1367 EVENTHANDLER_PRI_ANY);
1368
1369 /*
1370 * On interface removal, clean up IPv4 data structures hung off of the ifnet.
1371 */
1372 static void
in_ifdetach(void * arg __unused,struct ifnet * ifp)1373 in_ifdetach(void *arg __unused, struct ifnet *ifp)
1374 {
1375 struct in_ifinfo *ii = ifp->if_inet;
1376
1377 #ifdef VIMAGE
1378 /*
1379 * On VNET shutdown abort here as the stack teardown will do all
1380 * the work top-down for us. XXXGL: this logic is copied from
1381 * if_detach() before dom_ifattach removal. Ideally we'd like to have
1382 * same logic for VNET shutdown and normal detach. This means that
1383 * interfaces should be detach before protocols destroyed during VNET
1384 * shutdown.
1385 */
1386 if (!VNET_IS_SHUTTING_DOWN(ifp->if_vnet))
1387 #endif
1388 {
1389 IN_MULTI_LOCK();
1390 in_pcbpurgeif0(&V_ripcbinfo, ifp);
1391 in_pcbpurgeif0(&V_udbinfo, ifp);
1392 in_pcbpurgeif0(&V_ulitecbinfo, ifp);
1393 in_purgemaddrs(ifp);
1394 IN_MULTI_UNLOCK();
1395
1396 /*
1397 * Make sure all multicast deletions invoking if_ioctl() are
1398 * completed before returning. Else we risk accessing a freed
1399 * ifnet structure pointer.
1400 */
1401 inm_release_wait(NULL);
1402 }
1403
1404 igmp_domifdetach(ifp);
1405 lltable_free(ii->ii_llt);
1406 free(ii, M_IFADDR);
1407 }
1408 EVENTHANDLER_DEFINE(ifnet_departure_event, in_ifdetach, NULL,
1409 EVENTHANDLER_PRI_ANY);
1410
1411 static void
in_ifnet_event(void * arg __unused,struct ifnet * ifp,int event)1412 in_ifnet_event(void *arg __unused, struct ifnet *ifp, int event)
1413 {
1414 struct epoch_tracker et;
1415 struct ifaddr *ifa;
1416 struct in_ifaddr *ia;
1417 int error;
1418
1419 NET_EPOCH_ENTER(et);
1420 switch (event) {
1421 case IFNET_EVENT_DOWN:
1422 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1423 if (ifa->ifa_addr->sa_family != AF_INET)
1424 continue;
1425 ia = (struct in_ifaddr *)ifa;
1426 if ((ia->ia_flags & IFA_ROUTE) == 0)
1427 continue;
1428 ifa_ref(ifa);
1429 /*
1430 * in_scrubprefix() kills the interface route.
1431 */
1432 in_scrubprefix(ia, 0);
1433 /*
1434 * in_ifadown gets rid of all the rest of the
1435 * routes. This is not quite the right thing
1436 * to do, but at least if we are running a
1437 * routing process they will come back.
1438 */
1439 in_ifadown(ifa, 0);
1440 ifa_free(ifa);
1441 }
1442 break;
1443
1444 case IFNET_EVENT_UP:
1445 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1446 if (ifa->ifa_addr->sa_family != AF_INET)
1447 continue;
1448 ia = (struct in_ifaddr *)ifa;
1449 if (ia->ia_flags & IFA_ROUTE)
1450 continue;
1451 ifa_ref(ifa);
1452 error = ifa_del_loopback_route(ifa, ifa->ifa_addr);
1453 rt_addrmsg(RTM_ADD, ifa, ifa->ifa_ifp->if_fib);
1454 error = in_handle_ifaddr_route(RTM_ADD, ia);
1455 if (error == 0)
1456 ia->ia_flags |= IFA_ROUTE;
1457 error = ifa_add_loopback_route(ifa, ifa->ifa_addr);
1458 ifa_free(ifa);
1459 }
1460 break;
1461 }
1462 NET_EPOCH_EXIT(et);
1463 }
1464 EVENTHANDLER_DEFINE(ifnet_event, in_ifnet_event, NULL, EVENTHANDLER_PRI_ANY);
1465
1466 /*
1467 * Delete all IPv4 multicast address records, and associated link-layer
1468 * multicast address records, associated with ifp.
1469 * XXX It looks like domifdetach runs AFTER the link layer cleanup.
1470 * XXX This should not race with ifma_protospec being set during
1471 * a new allocation, if it does, we have bigger problems.
1472 */
1473 static void
in_purgemaddrs(struct ifnet * ifp)1474 in_purgemaddrs(struct ifnet *ifp)
1475 {
1476 struct epoch_tracker et;
1477 struct in_multi_head purgeinms;
1478 struct in_multi *inm;
1479 struct ifmultiaddr *ifma;
1480
1481 SLIST_INIT(&purgeinms);
1482 IN_MULTI_LIST_LOCK();
1483
1484 /*
1485 * Extract list of in_multi associated with the detaching ifp
1486 * which the PF_INET layer is about to release.
1487 * We need to do this as IF_ADDR_LOCK() may be re-acquired
1488 * by code further down.
1489 */
1490 IF_ADDR_WLOCK(ifp);
1491 NET_EPOCH_ENTER(et);
1492 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1493 inm = inm_ifmultiaddr_get_inm(ifma);
1494 if (inm == NULL)
1495 continue;
1496 inm_rele_locked(&purgeinms, inm);
1497 }
1498 NET_EPOCH_EXIT(et);
1499 IF_ADDR_WUNLOCK(ifp);
1500
1501 inm_release_list_deferred(&purgeinms);
1502 igmp_ifdetach(ifp);
1503 IN_MULTI_LIST_UNLOCK();
1504 }
1505
1506 struct in_llentry {
1507 struct llentry base;
1508 };
1509
1510 #define IN_LLTBL_DEFAULT_HSIZE 32
1511 #define IN_LLTBL_HASH(k, h) \
1512 (((((((k >> 8) ^ k) >> 8) ^ k) >> 8) ^ k) & ((h) - 1))
1513
1514 /*
1515 * Do actual deallocation of @lle.
1516 */
1517 static void
in_lltable_destroy_lle_unlocked(epoch_context_t ctx)1518 in_lltable_destroy_lle_unlocked(epoch_context_t ctx)
1519 {
1520 struct llentry *lle;
1521
1522 lle = __containerof(ctx, struct llentry, lle_epoch_ctx);
1523 LLE_LOCK_DESTROY(lle);
1524 LLE_REQ_DESTROY(lle);
1525 free(lle, M_LLTABLE);
1526 }
1527
1528 /*
1529 * Called by LLE_FREE_LOCKED when number of references
1530 * drops to zero.
1531 */
1532 static void
in_lltable_destroy_lle(struct llentry * lle)1533 in_lltable_destroy_lle(struct llentry *lle)
1534 {
1535
1536 LLE_WUNLOCK(lle);
1537 NET_EPOCH_CALL(in_lltable_destroy_lle_unlocked, &lle->lle_epoch_ctx);
1538 }
1539
1540 static struct llentry *
in_lltable_new(struct in_addr addr4,u_int flags)1541 in_lltable_new(struct in_addr addr4, u_int flags)
1542 {
1543 struct in_llentry *lle;
1544
1545 lle = malloc(sizeof(struct in_llentry), M_LLTABLE, M_NOWAIT | M_ZERO);
1546 if (lle == NULL) /* NB: caller generates msg */
1547 return NULL;
1548
1549 /*
1550 * For IPv4 this will trigger "arpresolve" to generate
1551 * an ARP request.
1552 */
1553 lle->base.la_expire = time_uptime; /* mark expired */
1554 lle->base.r_l3addr.addr4 = addr4;
1555 lle->base.lle_refcnt = 1;
1556 lle->base.lle_free = in_lltable_destroy_lle;
1557 LLE_LOCK_INIT(&lle->base);
1558 LLE_REQ_INIT(&lle->base);
1559 callout_init(&lle->base.lle_timer, 1);
1560
1561 return (&lle->base);
1562 }
1563
1564 static int
in_lltable_match_prefix(const struct sockaddr * saddr,const struct sockaddr * smask,u_int flags,struct llentry * lle)1565 in_lltable_match_prefix(const struct sockaddr *saddr,
1566 const struct sockaddr *smask, u_int flags, struct llentry *lle)
1567 {
1568 struct in_addr addr, mask, lle_addr;
1569
1570 addr = ((const struct sockaddr_in *)saddr)->sin_addr;
1571 mask = ((const struct sockaddr_in *)smask)->sin_addr;
1572 lle_addr.s_addr = ntohl(lle->r_l3addr.addr4.s_addr);
1573
1574 if (IN_ARE_MASKED_ADDR_EQUAL(lle_addr, addr, mask) == 0)
1575 return (0);
1576
1577 if (lle->la_flags & LLE_IFADDR) {
1578 /*
1579 * Delete LLE_IFADDR records IFF address & flag matches.
1580 * Note that addr is the interface address within prefix
1581 * being matched.
1582 * Note also we should handle 'ifdown' cases without removing
1583 * ifaddr macs.
1584 */
1585 if (addr.s_addr == lle_addr.s_addr && (flags & LLE_STATIC) != 0)
1586 return (1);
1587 return (0);
1588 }
1589
1590 /* flags & LLE_STATIC means deleting both dynamic and static entries */
1591 if ((flags & LLE_STATIC) || !(lle->la_flags & LLE_STATIC))
1592 return (1);
1593
1594 return (0);
1595 }
1596
1597 static void
in_lltable_free_entry(struct lltable * llt,struct llentry * lle)1598 in_lltable_free_entry(struct lltable *llt, struct llentry *lle)
1599 {
1600 size_t pkts_dropped;
1601
1602 LLE_WLOCK_ASSERT(lle);
1603 KASSERT(llt != NULL, ("lltable is NULL"));
1604
1605 /* Unlink entry from table if not already */
1606 if ((lle->la_flags & LLE_LINKED) != 0) {
1607 LLTABLE_LOCK_ASSERT(llt);
1608 lltable_unlink_entry(llt, lle);
1609 }
1610
1611 /* Drop hold queue */
1612 pkts_dropped = llentry_free(lle);
1613 ARPSTAT_ADD(dropped, pkts_dropped);
1614 }
1615
1616 static int
in_lltable_rtcheck(struct ifnet * ifp,u_int flags,const struct sockaddr * l3addr)1617 in_lltable_rtcheck(struct ifnet *ifp, u_int flags, const struct sockaddr *l3addr)
1618 {
1619 struct nhop_object *nh;
1620 struct in_addr addr;
1621
1622 KASSERT(l3addr->sa_family == AF_INET,
1623 ("sin_family %d", l3addr->sa_family));
1624
1625 addr = ((const struct sockaddr_in *)l3addr)->sin_addr;
1626
1627 nh = fib4_lookup(ifp->if_fib, addr, 0, NHR_NONE, 0);
1628 if (nh == NULL)
1629 return (EINVAL);
1630
1631 /*
1632 * If the gateway for an existing host route matches the target L3
1633 * address, which is a special route inserted by some implementation
1634 * such as MANET, and the interface is of the correct type, then
1635 * allow for ARP to proceed.
1636 */
1637 if (nh->nh_flags & NHF_GATEWAY) {
1638 if (!(nh->nh_flags & NHF_HOST) || nh->nh_ifp->if_type != IFT_ETHER ||
1639 (nh->nh_ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) != 0 ||
1640 memcmp(nh->gw_sa.sa_data, l3addr->sa_data,
1641 sizeof(in_addr_t)) != 0) {
1642 return (EINVAL);
1643 }
1644 }
1645
1646 /*
1647 * Make sure that at least the destination address is covered
1648 * by the route. This is for handling the case where 2 or more
1649 * interfaces have the same prefix. An incoming packet arrives
1650 * on one interface and the corresponding outgoing packet leaves
1651 * another interface.
1652 */
1653 if ((nh->nh_ifp != ifp) && (nh->nh_flags & NHF_HOST) == 0) {
1654 struct in_ifaddr *ia = (struct in_ifaddr *)ifaof_ifpforaddr(l3addr, ifp);
1655 struct in_addr dst_addr, mask_addr;
1656
1657 if (ia == NULL)
1658 return (EINVAL);
1659
1660 /*
1661 * ifaof_ifpforaddr() returns _best matching_ IFA.
1662 * It is possible that ifa prefix does not cover our address.
1663 * Explicitly verify and fail if that's the case.
1664 */
1665 dst_addr = IA_SIN(ia)->sin_addr;
1666 mask_addr.s_addr = htonl(ia->ia_subnetmask);
1667
1668 if (!IN_ARE_MASKED_ADDR_EQUAL(dst_addr, addr, mask_addr))
1669 return (EINVAL);
1670 }
1671
1672 return (0);
1673 }
1674
1675 static inline uint32_t
in_lltable_hash_dst(const struct in_addr dst,uint32_t hsize)1676 in_lltable_hash_dst(const struct in_addr dst, uint32_t hsize)
1677 {
1678
1679 return (IN_LLTBL_HASH(dst.s_addr, hsize));
1680 }
1681
1682 static uint32_t
in_lltable_hash(const struct llentry * lle,uint32_t hsize)1683 in_lltable_hash(const struct llentry *lle, uint32_t hsize)
1684 {
1685
1686 return (in_lltable_hash_dst(lle->r_l3addr.addr4, hsize));
1687 }
1688
1689 static void
in_lltable_fill_sa_entry(const struct llentry * lle,struct sockaddr * sa)1690 in_lltable_fill_sa_entry(const struct llentry *lle, struct sockaddr *sa)
1691 {
1692 struct sockaddr_in *sin;
1693
1694 sin = (struct sockaddr_in *)sa;
1695 bzero(sin, sizeof(*sin));
1696 sin->sin_family = AF_INET;
1697 sin->sin_len = sizeof(*sin);
1698 sin->sin_addr = lle->r_l3addr.addr4;
1699 }
1700
1701 static inline struct llentry *
in_lltable_find_dst(struct lltable * llt,struct in_addr dst)1702 in_lltable_find_dst(struct lltable *llt, struct in_addr dst)
1703 {
1704 struct llentry *lle;
1705 struct llentries *lleh;
1706 u_int hashidx;
1707
1708 hashidx = in_lltable_hash_dst(dst, llt->llt_hsize);
1709 lleh = &llt->lle_head[hashidx];
1710 CK_LIST_FOREACH(lle, lleh, lle_next) {
1711 if (lle->la_flags & LLE_DELETED)
1712 continue;
1713 if (lle->r_l3addr.addr4.s_addr == dst.s_addr)
1714 break;
1715 }
1716
1717 return (lle);
1718 }
1719
1720 static void
in_lltable_delete_entry(struct lltable * llt,struct llentry * lle)1721 in_lltable_delete_entry(struct lltable *llt, struct llentry *lle)
1722 {
1723
1724 lle->la_flags |= LLE_DELETED;
1725 EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_DELETED);
1726 #ifdef DIAGNOSTIC
1727 log(LOG_INFO, "ifaddr cache = %p is deleted\n", lle);
1728 #endif
1729 llentry_free(lle);
1730 }
1731
1732 static struct llentry *
in_lltable_alloc(struct lltable * llt,u_int flags,const struct sockaddr * l3addr)1733 in_lltable_alloc(struct lltable *llt, u_int flags, const struct sockaddr *l3addr)
1734 {
1735 const struct sockaddr_in *sin = (const struct sockaddr_in *)l3addr;
1736 struct ifnet *ifp = llt->llt_ifp;
1737 struct llentry *lle;
1738 char linkhdr[LLE_MAX_LINKHDR];
1739 size_t linkhdrsize;
1740 int lladdr_off;
1741
1742 KASSERT(l3addr->sa_family == AF_INET,
1743 ("sin_family %d", l3addr->sa_family));
1744
1745 /*
1746 * A route that covers the given address must have
1747 * been installed 1st because we are doing a resolution,
1748 * verify this.
1749 */
1750 if (!(flags & LLE_IFADDR) &&
1751 in_lltable_rtcheck(ifp, flags, l3addr) != 0)
1752 return (NULL);
1753
1754 lle = in_lltable_new(sin->sin_addr, flags);
1755 if (lle == NULL) {
1756 log(LOG_INFO, "lla_lookup: new lle malloc failed\n");
1757 return (NULL);
1758 }
1759 lle->la_flags = flags;
1760 if (flags & LLE_STATIC)
1761 lle->r_flags |= RLLE_VALID;
1762 if ((flags & LLE_IFADDR) == LLE_IFADDR) {
1763 linkhdrsize = LLE_MAX_LINKHDR;
1764 if (lltable_calc_llheader(ifp, AF_INET, IF_LLADDR(ifp),
1765 linkhdr, &linkhdrsize, &lladdr_off) != 0) {
1766 in_lltable_free_entry(llt, lle);
1767 return (NULL);
1768 }
1769 lltable_set_entry_addr(ifp, lle, linkhdr, linkhdrsize,
1770 lladdr_off);
1771 lle->la_flags |= LLE_STATIC;
1772 lle->r_flags |= (RLLE_VALID | RLLE_IFADDR);
1773 lle->la_expire = 0;
1774 }
1775
1776 return (lle);
1777 }
1778
1779 /*
1780 * Return NULL if not found or marked for deletion.
1781 * If found return lle read locked.
1782 */
1783 static struct llentry *
in_lltable_lookup(struct lltable * llt,u_int flags,const struct sockaddr * l3addr)1784 in_lltable_lookup(struct lltable *llt, u_int flags, const struct sockaddr *l3addr)
1785 {
1786 const struct sockaddr_in *sin = (const struct sockaddr_in *)l3addr;
1787 struct llentry *lle;
1788
1789 LLTABLE_RLOCK_ASSERT(llt);
1790 KASSERT(l3addr->sa_family == AF_INET,
1791 ("sin_family %d", l3addr->sa_family));
1792 KASSERT((flags & (LLE_UNLOCKED | LLE_EXCLUSIVE)) !=
1793 (LLE_UNLOCKED | LLE_EXCLUSIVE),
1794 ("wrong lle request flags: %#x", flags));
1795
1796 lle = in_lltable_find_dst(llt, sin->sin_addr);
1797 if (lle == NULL)
1798 return (NULL);
1799 if (flags & LLE_UNLOCKED)
1800 return (lle);
1801
1802 if (flags & LLE_EXCLUSIVE)
1803 LLE_WLOCK(lle);
1804 else
1805 LLE_RLOCK(lle);
1806
1807 /*
1808 * If the afdata lock is not held, the LLE may have been unlinked while
1809 * we were blocked on the LLE lock. Check for this case.
1810 */
1811 if (__predict_false((lle->la_flags & LLE_LINKED) == 0)) {
1812 if (flags & LLE_EXCLUSIVE)
1813 LLE_WUNLOCK(lle);
1814 else
1815 LLE_RUNLOCK(lle);
1816 return (NULL);
1817 }
1818 return (lle);
1819 }
1820
1821 static int
in_lltable_dump_entry(struct lltable * llt,struct llentry * lle,struct sysctl_req * wr)1822 in_lltable_dump_entry(struct lltable *llt, struct llentry *lle,
1823 struct sysctl_req *wr)
1824 {
1825 struct ifnet *ifp = llt->llt_ifp;
1826 /* XXX stack use */
1827 struct {
1828 struct rt_msghdr rtm;
1829 struct sockaddr_in sin;
1830 struct sockaddr_dl sdl;
1831 } arpc;
1832 struct sockaddr_dl *sdl;
1833 int error;
1834
1835 bzero(&arpc, sizeof(arpc));
1836 /* skip deleted entries */
1837 if ((lle->la_flags & LLE_DELETED) == LLE_DELETED)
1838 return (0);
1839 /* Skip if jailed and not a valid IP of the prison. */
1840 lltable_fill_sa_entry(lle,(struct sockaddr *)&arpc.sin);
1841 if (prison_if(wr->td->td_ucred, (struct sockaddr *)&arpc.sin) != 0)
1842 return (0);
1843 /*
1844 * produce a msg made of:
1845 * struct rt_msghdr;
1846 * struct sockaddr_in; (IPv4)
1847 * struct sockaddr_dl;
1848 */
1849 arpc.rtm.rtm_msglen = sizeof(arpc);
1850 arpc.rtm.rtm_version = RTM_VERSION;
1851 arpc.rtm.rtm_type = RTM_GET;
1852 arpc.rtm.rtm_flags = RTF_UP;
1853 arpc.rtm.rtm_addrs = RTA_DST | RTA_GATEWAY;
1854
1855 /* publish */
1856 if (lle->la_flags & LLE_PUB)
1857 arpc.rtm.rtm_flags |= RTF_ANNOUNCE;
1858
1859 sdl = &arpc.sdl;
1860 sdl->sdl_family = AF_LINK;
1861 sdl->sdl_len = sizeof(*sdl);
1862 sdl->sdl_index = ifp->if_index;
1863 sdl->sdl_type = ifp->if_type;
1864 if ((lle->la_flags & LLE_VALID) == LLE_VALID) {
1865 sdl->sdl_alen = ifp->if_addrlen;
1866 bcopy(lle->ll_addr, LLADDR(sdl), ifp->if_addrlen);
1867 } else {
1868 sdl->sdl_alen = 0;
1869 bzero(LLADDR(sdl), ifp->if_addrlen);
1870 }
1871
1872 arpc.rtm.rtm_rmx.rmx_expire =
1873 lle->la_flags & LLE_STATIC ? 0 : lle->la_expire;
1874 arpc.rtm.rtm_flags |= (RTF_HOST | RTF_LLDATA);
1875 if (lle->la_flags & LLE_STATIC)
1876 arpc.rtm.rtm_flags |= RTF_STATIC;
1877 if (lle->la_flags & LLE_IFADDR)
1878 arpc.rtm.rtm_flags |= RTF_PINNED;
1879 arpc.rtm.rtm_index = ifp->if_index;
1880 error = SYSCTL_OUT(wr, &arpc, sizeof(arpc));
1881
1882 return (error);
1883 }
1884
1885 static void
in_lltable_post_resolved(struct lltable * llt,struct llentry * lle)1886 in_lltable_post_resolved(struct lltable *llt, struct llentry *lle)
1887 {
1888 struct ifnet *ifp = llt->llt_ifp;
1889
1890 /* gratuitous ARP */
1891 if ((lle->la_flags & LLE_PUB) != 0)
1892 arprequest(ifp, &lle->r_l3addr.addr4, &lle->r_l3addr.addr4,
1893 lle->ll_addr);
1894 }
1895
1896 static struct lltable *
in_lltattach(struct ifnet * ifp)1897 in_lltattach(struct ifnet *ifp)
1898 {
1899 struct lltable *llt;
1900
1901 llt = lltable_allocate_htbl(IN_LLTBL_DEFAULT_HSIZE);
1902 llt->llt_af = AF_INET;
1903 llt->llt_ifp = ifp;
1904
1905 llt->llt_lookup = in_lltable_lookup;
1906 llt->llt_alloc_entry = in_lltable_alloc;
1907 llt->llt_delete_entry = in_lltable_delete_entry;
1908 llt->llt_dump_entry = in_lltable_dump_entry;
1909 llt->llt_hash = in_lltable_hash;
1910 llt->llt_fill_sa_entry = in_lltable_fill_sa_entry;
1911 llt->llt_free_entry = in_lltable_free_entry;
1912 llt->llt_match_prefix = in_lltable_match_prefix;
1913 llt->llt_mark_used = llentry_mark_used;
1914 llt->llt_post_resolved = in_lltable_post_resolved;
1915 lltable_link(llt);
1916
1917 return (llt);
1918 }
1919
1920 struct lltable *
in_lltable_get(struct ifnet * ifp)1921 in_lltable_get(struct ifnet *ifp)
1922 {
1923 /* XXXGL: ??? */
1924 if (ifp->if_inet == NULL)
1925 return (NULL);
1926
1927 return (((struct in_ifinfo *)ifp->if_inet)->ii_llt);
1928 }
1929