1 /*-
2 * Copyright (c) 1980, 1986, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)if.c 8.5 (Berkeley) 1/9/95
30 * $FreeBSD: stable/10/sys/net/if.c 333106 2018-04-30 08:39:23Z royger $
31 */
32
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 #include "opt_inet.h"
36
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/conf.h>
40 #include <sys/malloc.h>
41 #include <sys/sbuf.h>
42 #include <sys/bus.h>
43 #include <sys/mbuf.h>
44 #include <sys/systm.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/kernel.h>
51 #include <sys/lock.h>
52 #include <sys/refcount.h>
53 #include <sys/module.h>
54 #include <sys/rwlock.h>
55 #include <sys/sockio.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/taskqueue.h>
59 #include <sys/domain.h>
60 #include <sys/jail.h>
61 #include <sys/priv.h>
62
63 #include <machine/stdarg.h>
64 #include <vm/uma.h>
65
66 #include <net/if.h>
67 #include <net/if_arp.h>
68 #include <net/if_clone.h>
69 #include <net/if_dl.h>
70 #include <net/if_types.h>
71 #include <net/if_var.h>
72 #include <net/radix.h>
73 #include <net/route.h>
74 #include <net/vnet.h>
75
76 #if defined(INET) || defined(INET6)
77 #include <net/ethernet.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 #include <netinet/ip.h>
81 #include <netinet/ip_carp.h>
82 #ifdef INET
83 #include <netinet/if_ether.h>
84 #endif /* INET */
85 #ifdef INET6
86 #include <netinet6/in6_var.h>
87 #include <netinet6/in6_ifattach.h>
88 #endif /* INET6 */
89 #endif /* INET || INET6 */
90
91 #include <security/mac/mac_framework.h>
92
93 #ifdef COMPAT_FREEBSD32
94 #include <sys/mount.h>
95 #include <compat/freebsd32/freebsd32.h>
96 #endif
97
98 struct ifindex_entry {
99 struct ifnet *ife_ifnet;
100 };
101
102 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
103 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
104
105 TUNABLE_INT("net.link.ifqmaxlen", &ifqmaxlen);
106 SYSCTL_INT(_net_link, OID_AUTO, ifqmaxlen, CTLFLAG_RDTUN,
107 &ifqmaxlen, 0, "max send queue size");
108
109 /* Log link state change events */
110 static int log_link_state_change = 1;
111
112 SYSCTL_INT(_net_link, OID_AUTO, log_link_state_change, CTLFLAG_RW,
113 &log_link_state_change, 0,
114 "log interface link state change events");
115
116 /* Log promiscuous mode change events */
117 static int log_promisc_mode_change = 1;
118
119 TUNABLE_INT("net.link.log_promisc_mode_change", &log_promisc_mode_change);
120 SYSCTL_INT(_net_link, OID_AUTO, log_promisc_mode_change, CTLFLAG_RDTUN,
121 &log_promisc_mode_change, 1,
122 "log promiscuous mode change events");
123
124 /* Interface description */
125 static unsigned int ifdescr_maxlen = 1024;
126 SYSCTL_UINT(_net, OID_AUTO, ifdescr_maxlen, CTLFLAG_RW,
127 &ifdescr_maxlen, 0,
128 "administrative maximum length for interface description");
129
130 static MALLOC_DEFINE(M_IFDESCR, "ifdescr", "ifnet descriptions");
131
132 /* global sx for non-critical path ifdescr */
133 static struct sx ifdescr_sx;
134 SX_SYSINIT(ifdescr_sx, &ifdescr_sx, "ifnet descr");
135
136 void (*bridge_linkstate_p)(struct ifnet *ifp);
137 void (*ng_ether_link_state_p)(struct ifnet *ifp, int state);
138 void (*lagg_linkstate_p)(struct ifnet *ifp, int state);
139 /* These are external hooks for CARP. */
140 void (*carp_linkstate_p)(struct ifnet *ifp);
141 void (*carp_demote_adj_p)(int, char *);
142 int (*carp_master_p)(struct ifaddr *);
143 #if defined(INET) || defined(INET6)
144 int (*carp_forus_p)(struct ifnet *ifp, u_char *dhost);
145 int (*carp_output_p)(struct ifnet *ifp, struct mbuf *m,
146 const struct sockaddr *sa);
147 int (*carp_ioctl_p)(struct ifreq *, u_long, struct thread *);
148 int (*carp_attach_p)(struct ifaddr *, int);
149 void (*carp_detach_p)(struct ifaddr *);
150 #endif
151 #ifdef INET
152 int (*carp_iamatch_p)(struct ifaddr *, uint8_t **);
153 #endif
154 #ifdef INET6
155 struct ifaddr *(*carp_iamatch6_p)(struct ifnet *ifp, struct in6_addr *taddr6);
156 caddr_t (*carp_macmatch6_p)(struct ifnet *ifp, struct mbuf *m,
157 const struct in6_addr *taddr);
158 #endif
159
160 struct mbuf *(*tbr_dequeue_ptr)(struct ifaltq *, int) = NULL;
161
162 /*
163 * XXX: Style; these should be sorted alphabetically, and unprototyped
164 * static functions should be prototyped. Currently they are sorted by
165 * declaration order.
166 */
167 static void if_attachdomain(void *);
168 static void if_attachdomain1(struct ifnet *);
169 static int ifconf(u_long, caddr_t);
170 static void if_freemulti(struct ifmultiaddr *);
171 static void if_init(void *);
172 static void if_grow(void);
173 static void if_input_default(struct ifnet *, struct mbuf *);
174 static void if_route(struct ifnet *, int flag, int fam);
175 static int if_setflag(struct ifnet *, int, int, int *, int);
176 static int if_transmit(struct ifnet *ifp, struct mbuf *m);
177 static void if_unroute(struct ifnet *, int flag, int fam);
178 static void link_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
179 static int if_rtdel(struct radix_node *, void *);
180 static int ifhwioctl(u_long, struct ifnet *, caddr_t, struct thread *);
181 static int if_delmulti_locked(struct ifnet *, struct ifmultiaddr *, int);
182 static void do_link_state_change(void *, int);
183 static int if_getgroup(struct ifgroupreq *, struct ifnet *);
184 static int if_getgroupmembers(struct ifgroupreq *);
185 static void if_delgroups(struct ifnet *);
186 static void if_attach_internal(struct ifnet *, int, struct if_clone *);
187 static int if_detach_internal(struct ifnet *, int, struct if_clone **);
188
189 #ifdef INET6
190 /*
191 * XXX: declare here to avoid to include many inet6 related files..
192 * should be more generalized?
193 */
194 extern void nd6_setmtu(struct ifnet *);
195 #endif
196
197 VNET_DEFINE(int, if_index);
198 int ifqmaxlen = IFQ_MAXLEN;
199 VNET_DEFINE(struct ifnethead, ifnet); /* depend on static init XXX */
200 VNET_DEFINE(struct ifgrouphead, ifg_head);
201
202 static VNET_DEFINE(int, if_indexlim) = 8;
203
204 /* Table of ifnet by index. */
205 VNET_DEFINE(struct ifindex_entry *, ifindex_table);
206
207 #define V_if_indexlim VNET(if_indexlim)
208 #define V_ifindex_table VNET(ifindex_table)
209
210 /*
211 * The global network interface list (V_ifnet) and related state (such as
212 * if_index, if_indexlim, and ifindex_table) are protected by an sxlock and
213 * an rwlock. Either may be acquired shared to stablize the list, but both
214 * must be acquired writable to modify the list. This model allows us to
215 * both stablize the interface list during interrupt thread processing, but
216 * also to stablize it over long-running ioctls, without introducing priority
217 * inversions and deadlocks.
218 */
219 struct rwlock ifnet_rwlock;
220 struct sx ifnet_sxlock;
221
222 /*
223 * The allocation of network interfaces is a rather non-atomic affair; we
224 * need to select an index before we are ready to expose the interface for
225 * use, so will use this pointer value to indicate reservation.
226 */
227 #define IFNET_HOLD (void *)(uintptr_t)(-1)
228
229 static if_com_alloc_t *if_com_alloc[256];
230 static if_com_free_t *if_com_free[256];
231
232 static MALLOC_DEFINE(M_IFNET, "ifnet", "interface internals");
233 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
234 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
235
236 struct ifnet *
ifnet_byindex_locked(u_short idx)237 ifnet_byindex_locked(u_short idx)
238 {
239
240 if (idx > V_if_index)
241 return (NULL);
242 if (V_ifindex_table[idx].ife_ifnet == IFNET_HOLD)
243 return (NULL);
244 return (V_ifindex_table[idx].ife_ifnet);
245 }
246
247 struct ifnet *
ifnet_byindex(u_short idx)248 ifnet_byindex(u_short idx)
249 {
250 struct ifnet *ifp;
251
252 IFNET_RLOCK_NOSLEEP();
253 ifp = ifnet_byindex_locked(idx);
254 IFNET_RUNLOCK_NOSLEEP();
255 return (ifp);
256 }
257
258 struct ifnet *
ifnet_byindex_ref(u_short idx)259 ifnet_byindex_ref(u_short idx)
260 {
261 struct ifnet *ifp;
262
263 IFNET_RLOCK_NOSLEEP();
264 ifp = ifnet_byindex_locked(idx);
265 if (ifp == NULL || (ifp->if_flags & IFF_DYING)) {
266 IFNET_RUNLOCK_NOSLEEP();
267 return (NULL);
268 }
269 if_ref(ifp);
270 IFNET_RUNLOCK_NOSLEEP();
271 return (ifp);
272 }
273
274 /*
275 * Allocate an ifindex array entry; return 0 on success or an error on
276 * failure.
277 */
278 static int
ifindex_alloc_locked(u_short * idxp)279 ifindex_alloc_locked(u_short *idxp)
280 {
281 u_short idx;
282
283 IFNET_WLOCK_ASSERT();
284
285 retry:
286 /*
287 * Try to find an empty slot below V_if_index. If we fail, take the
288 * next slot.
289 */
290 for (idx = 1; idx <= V_if_index; idx++) {
291 if (V_ifindex_table[idx].ife_ifnet == NULL)
292 break;
293 }
294
295 /* Catch if_index overflow. */
296 if (idx < 1)
297 return (ENOSPC);
298 if (idx >= V_if_indexlim) {
299 if_grow();
300 goto retry;
301 }
302 if (idx > V_if_index)
303 V_if_index = idx;
304 *idxp = idx;
305 return (0);
306 }
307
308 static void
ifindex_free_locked(u_short idx)309 ifindex_free_locked(u_short idx)
310 {
311
312 IFNET_WLOCK_ASSERT();
313
314 V_ifindex_table[idx].ife_ifnet = NULL;
315 while (V_if_index > 0 &&
316 V_ifindex_table[V_if_index].ife_ifnet == NULL)
317 V_if_index--;
318 }
319
320 static void
ifindex_free(u_short idx)321 ifindex_free(u_short idx)
322 {
323
324 IFNET_WLOCK();
325 ifindex_free_locked(idx);
326 IFNET_WUNLOCK();
327 }
328
329 static void
ifnet_setbyindex_locked(u_short idx,struct ifnet * ifp)330 ifnet_setbyindex_locked(u_short idx, struct ifnet *ifp)
331 {
332
333 IFNET_WLOCK_ASSERT();
334
335 V_ifindex_table[idx].ife_ifnet = ifp;
336 }
337
338 static void
ifnet_setbyindex(u_short idx,struct ifnet * ifp)339 ifnet_setbyindex(u_short idx, struct ifnet *ifp)
340 {
341
342 IFNET_WLOCK();
343 ifnet_setbyindex_locked(idx, ifp);
344 IFNET_WUNLOCK();
345 }
346
347 struct ifaddr *
ifaddr_byindex(u_short idx)348 ifaddr_byindex(u_short idx)
349 {
350 struct ifnet *ifp;
351 struct ifaddr *ifa = NULL;
352
353 IFNET_RLOCK_NOSLEEP();
354 ifp = ifnet_byindex_locked(idx);
355 if (ifp != NULL && (ifa = ifp->if_addr) != NULL)
356 ifa_ref(ifa);
357 IFNET_RUNLOCK_NOSLEEP();
358 return (ifa);
359 }
360
361 /*
362 * Network interface utility routines.
363 *
364 * Routines with ifa_ifwith* names take sockaddr *'s as
365 * parameters.
366 */
367
368 static void
vnet_if_init(const void * unused __unused)369 vnet_if_init(const void *unused __unused)
370 {
371
372 TAILQ_INIT(&V_ifnet);
373 TAILQ_INIT(&V_ifg_head);
374 IFNET_WLOCK();
375 if_grow(); /* create initial table */
376 IFNET_WUNLOCK();
377 vnet_if_clone_init();
378 }
379 VNET_SYSINIT(vnet_if_init, SI_SUB_INIT_IF, SI_ORDER_SECOND, vnet_if_init,
380 NULL);
381
382 /* ARGSUSED*/
383 static void
if_init(void * dummy __unused)384 if_init(void *dummy __unused)
385 {
386
387 IFNET_LOCK_INIT();
388 if_clone_init();
389 }
390 SYSINIT(interfaces, SI_SUB_INIT_IF, SI_ORDER_FIRST, if_init, NULL);
391
392
393 #ifdef VIMAGE
394 static void
vnet_if_uninit(const void * unused __unused)395 vnet_if_uninit(const void *unused __unused)
396 {
397
398 VNET_ASSERT(TAILQ_EMPTY(&V_ifnet), ("%s:%d tailq &V_ifnet=%p "
399 "not empty", __func__, __LINE__, &V_ifnet));
400 VNET_ASSERT(TAILQ_EMPTY(&V_ifg_head), ("%s:%d tailq &V_ifg_head=%p "
401 "not empty", __func__, __LINE__, &V_ifg_head));
402
403 free((caddr_t)V_ifindex_table, M_IFNET);
404 }
405 VNET_SYSUNINIT(vnet_if_uninit, SI_SUB_INIT_IF, SI_ORDER_FIRST,
406 vnet_if_uninit, NULL);
407 #endif
408
409 static void
if_grow(void)410 if_grow(void)
411 {
412 int oldlim;
413 u_int n;
414 struct ifindex_entry *e;
415
416 IFNET_WLOCK_ASSERT();
417 oldlim = V_if_indexlim;
418 IFNET_WUNLOCK();
419 n = (oldlim << 1) * sizeof(*e);
420 e = malloc(n, M_IFNET, M_WAITOK | M_ZERO);
421 IFNET_WLOCK();
422 if (V_if_indexlim != oldlim) {
423 free(e, M_IFNET);
424 return;
425 }
426 if (V_ifindex_table != NULL) {
427 memcpy((caddr_t)e, (caddr_t)V_ifindex_table, n/2);
428 free((caddr_t)V_ifindex_table, M_IFNET);
429 }
430 V_if_indexlim <<= 1;
431 V_ifindex_table = e;
432 }
433
434 /*
435 * Allocate a struct ifnet and an index for an interface. A layer 2
436 * common structure will also be allocated if an allocation routine is
437 * registered for the passed type.
438 */
439 struct ifnet *
if_alloc(u_char type)440 if_alloc(u_char type)
441 {
442 struct ifnet *ifp;
443 u_short idx;
444
445 ifp = malloc(sizeof(struct ifnet), M_IFNET, M_WAITOK|M_ZERO);
446 IFNET_WLOCK();
447 if (ifindex_alloc_locked(&idx) != 0) {
448 IFNET_WUNLOCK();
449 free(ifp, M_IFNET);
450 return (NULL);
451 }
452 ifnet_setbyindex_locked(idx, IFNET_HOLD);
453 IFNET_WUNLOCK();
454 ifp->if_index = idx;
455 ifp->if_type = type;
456 ifp->if_alloctype = type;
457 if (if_com_alloc[type] != NULL) {
458 ifp->if_l2com = if_com_alloc[type](type, ifp);
459 if (ifp->if_l2com == NULL) {
460 free(ifp, M_IFNET);
461 ifindex_free(idx);
462 return (NULL);
463 }
464 }
465
466 IF_ADDR_LOCK_INIT(ifp);
467 TASK_INIT(&ifp->if_linktask, 0, do_link_state_change, ifp);
468 ifp->if_afdata_initialized = 0;
469 IF_AFDATA_LOCK_INIT(ifp);
470 TAILQ_INIT(&ifp->if_addrhead);
471 TAILQ_INIT(&ifp->if_multiaddrs);
472 TAILQ_INIT(&ifp->if_groups);
473 #ifdef MAC
474 mac_ifnet_init(ifp);
475 #endif
476 ifq_init(&ifp->if_snd, ifp);
477
478 refcount_init(&ifp->if_refcount, 1); /* Index reference. */
479 ifnet_setbyindex(ifp->if_index, ifp);
480 return (ifp);
481 }
482
483 /*
484 * Do the actual work of freeing a struct ifnet, and layer 2 common
485 * structure. This call is made when the last reference to an
486 * interface is released.
487 */
488 static void
if_free_internal(struct ifnet * ifp)489 if_free_internal(struct ifnet *ifp)
490 {
491
492 KASSERT((ifp->if_flags & IFF_DYING),
493 ("if_free_internal: interface not dying"));
494
495 if (if_com_free[ifp->if_alloctype] != NULL)
496 if_com_free[ifp->if_alloctype](ifp->if_l2com,
497 ifp->if_alloctype);
498
499 #ifdef MAC
500 mac_ifnet_destroy(ifp);
501 #endif /* MAC */
502 if (ifp->if_description != NULL)
503 free(ifp->if_description, M_IFDESCR);
504 IF_AFDATA_DESTROY(ifp);
505 IF_ADDR_LOCK_DESTROY(ifp);
506 ifq_delete(&ifp->if_snd);
507 free(ifp, M_IFNET);
508 }
509
510 /*
511 * Deregister an interface and free the associated storage.
512 */
513 void
if_free(struct ifnet * ifp)514 if_free(struct ifnet *ifp)
515 {
516
517 ifp->if_flags |= IFF_DYING; /* XXX: Locking */
518
519 CURVNET_SET_QUIET(ifp->if_vnet);
520 IFNET_WLOCK();
521 KASSERT(ifp == ifnet_byindex_locked(ifp->if_index),
522 ("%s: freeing unallocated ifnet", ifp->if_xname));
523
524 ifindex_free_locked(ifp->if_index);
525 IFNET_WUNLOCK();
526
527 if (refcount_release(&ifp->if_refcount))
528 if_free_internal(ifp);
529 CURVNET_RESTORE();
530 }
531
532 /*
533 * Interfaces to keep an ifnet type-stable despite the possibility of the
534 * driver calling if_free(). If there are additional references, we defer
535 * freeing the underlying data structure.
536 */
537 void
if_ref(struct ifnet * ifp)538 if_ref(struct ifnet *ifp)
539 {
540
541 /* We don't assert the ifnet list lock here, but arguably should. */
542 refcount_acquire(&ifp->if_refcount);
543 }
544
545 void
if_rele(struct ifnet * ifp)546 if_rele(struct ifnet *ifp)
547 {
548
549 if (!refcount_release(&ifp->if_refcount))
550 return;
551 if_free_internal(ifp);
552 }
553
554 void
ifq_init(struct ifaltq * ifq,struct ifnet * ifp)555 ifq_init(struct ifaltq *ifq, struct ifnet *ifp)
556 {
557
558 mtx_init(&ifq->ifq_mtx, ifp->if_xname, "if send queue", MTX_DEF);
559
560 if (ifq->ifq_maxlen == 0)
561 ifq->ifq_maxlen = ifqmaxlen;
562
563 ifq->altq_type = 0;
564 ifq->altq_disc = NULL;
565 ifq->altq_flags &= ALTQF_CANTCHANGE;
566 ifq->altq_tbr = NULL;
567 ifq->altq_ifp = ifp;
568 }
569
570 void
ifq_delete(struct ifaltq * ifq)571 ifq_delete(struct ifaltq *ifq)
572 {
573 mtx_destroy(&ifq->ifq_mtx);
574 }
575
576 /*
577 * Perform generic interface initalization tasks and attach the interface
578 * to the list of "active" interfaces. If vmove flag is set on entry
579 * to if_attach_internal(), perform only a limited subset of initialization
580 * tasks, given that we are moving from one vnet to another an ifnet which
581 * has already been fully initialized.
582 *
583 * Note that if_detach_internal() removes group membership unconditionally
584 * even when vmove flag is set, and if_attach_internal() adds only IFG_ALL.
585 * Thus, when if_vmove() is applied to a cloned interface, group membership
586 * is lost while a cloned one always joins a group whose name is
587 * ifc->ifc_name. To recover this after if_detach_internal() and
588 * if_attach_internal(), the cloner should be specified to
589 * if_attach_internal() via ifc. If it is non-NULL, if_attach_internal()
590 * attempts to join a group whose name is ifc->ifc_name.
591 *
592 * XXX:
593 * - The decision to return void and thus require this function to
594 * succeed is questionable.
595 * - We should probably do more sanity checking. For instance we don't
596 * do anything to insure if_xname is unique or non-empty.
597 */
598 void
if_attach(struct ifnet * ifp)599 if_attach(struct ifnet *ifp)
600 {
601
602 if_attach_internal(ifp, 0, NULL);
603 }
604
605 /*
606 * Compute the least common TSO limit.
607 */
608 void
if_hw_tsomax_common(struct ifnet * ifp,struct ifnet_hw_tsomax * pmax)609 if_hw_tsomax_common(struct ifnet *ifp, struct ifnet_hw_tsomax *pmax)
610 {
611 /*
612 * 1) If there is no limit currently, take the limit from
613 * the network adapter.
614 *
615 * 2) If the network adapter has a limit below the current
616 * limit, apply it.
617 */
618 if (pmax->tsomaxbytes == 0 || (ifp->if_hw_tsomax != 0 &&
619 ifp->if_hw_tsomax < pmax->tsomaxbytes)) {
620 pmax->tsomaxbytes = ifp->if_hw_tsomax;
621 }
622 if (pmax->tsomaxsegcount == 0 || (ifp->if_hw_tsomaxsegcount != 0 &&
623 ifp->if_hw_tsomaxsegcount < pmax->tsomaxsegcount)) {
624 pmax->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
625 }
626 if (pmax->tsomaxsegsize == 0 || (ifp->if_hw_tsomaxsegsize != 0 &&
627 ifp->if_hw_tsomaxsegsize < pmax->tsomaxsegsize)) {
628 pmax->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
629 }
630 }
631
632 /*
633 * Update TSO limit of a network adapter.
634 *
635 * Returns zero if no change. Else non-zero.
636 */
637 int
if_hw_tsomax_update(struct ifnet * ifp,struct ifnet_hw_tsomax * pmax)638 if_hw_tsomax_update(struct ifnet *ifp, struct ifnet_hw_tsomax *pmax)
639 {
640 int retval = 0;
641 if (ifp->if_hw_tsomax != pmax->tsomaxbytes) {
642 ifp->if_hw_tsomax = pmax->tsomaxbytes;
643 retval++;
644 }
645 if (ifp->if_hw_tsomaxsegsize != pmax->tsomaxsegsize) {
646 ifp->if_hw_tsomaxsegsize = pmax->tsomaxsegsize;
647 retval++;
648 }
649 if (ifp->if_hw_tsomaxsegcount != pmax->tsomaxsegcount) {
650 ifp->if_hw_tsomaxsegcount = pmax->tsomaxsegcount;
651 retval++;
652 }
653 return (retval);
654 }
655
656 static void
if_attach_internal(struct ifnet * ifp,int vmove,struct if_clone * ifc)657 if_attach_internal(struct ifnet *ifp, int vmove, struct if_clone *ifc)
658 {
659 unsigned socksize, ifasize;
660 int namelen, masklen;
661 struct sockaddr_dl *sdl;
662 struct ifaddr *ifa;
663
664 if (ifp->if_index == 0 || ifp != ifnet_byindex(ifp->if_index))
665 panic ("%s: BUG: if_attach called without if_alloc'd input()\n",
666 ifp->if_xname);
667
668 #ifdef VIMAGE
669 ifp->if_vnet = curvnet;
670 if (ifp->if_home_vnet == NULL)
671 ifp->if_home_vnet = curvnet;
672 #endif
673
674 if_addgroup(ifp, IFG_ALL);
675
676 /* Restore group membership for cloned interfaces. */
677 if (vmove && ifc != NULL)
678 if_clone_addgroup(ifp, ifc);
679
680 getmicrotime(&ifp->if_lastchange);
681 ifp->if_data.ifi_epoch = time_uptime;
682 ifp->if_data.ifi_datalen = sizeof(struct if_data);
683
684 KASSERT((ifp->if_transmit == NULL && ifp->if_qflush == NULL) ||
685 (ifp->if_transmit != NULL && ifp->if_qflush != NULL),
686 ("transmit and qflush must both either be set or both be NULL"));
687 if (ifp->if_transmit == NULL) {
688 ifp->if_transmit = if_transmit;
689 ifp->if_qflush = if_qflush;
690 }
691 if (ifp->if_input == NULL)
692 ifp->if_input = if_input_default;
693
694 if (!vmove) {
695 #ifdef MAC
696 mac_ifnet_create(ifp);
697 #endif
698
699 /*
700 * Create a Link Level name for this device.
701 */
702 namelen = strlen(ifp->if_xname);
703 /*
704 * Always save enough space for any possiable name so we
705 * can do a rename in place later.
706 */
707 masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + IFNAMSIZ;
708 socksize = masklen + ifp->if_addrlen;
709 if (socksize < sizeof(*sdl))
710 socksize = sizeof(*sdl);
711 socksize = roundup2(socksize, sizeof(long));
712 ifasize = sizeof(*ifa) + 2 * socksize;
713 ifa = malloc(ifasize, M_IFADDR, M_WAITOK | M_ZERO);
714 ifa_init(ifa);
715 sdl = (struct sockaddr_dl *)(ifa + 1);
716 sdl->sdl_len = socksize;
717 sdl->sdl_family = AF_LINK;
718 bcopy(ifp->if_xname, sdl->sdl_data, namelen);
719 sdl->sdl_nlen = namelen;
720 sdl->sdl_index = ifp->if_index;
721 sdl->sdl_type = ifp->if_type;
722 ifp->if_addr = ifa;
723 ifa->ifa_ifp = ifp;
724 ifa->ifa_rtrequest = link_rtrequest;
725 ifa->ifa_addr = (struct sockaddr *)sdl;
726 sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
727 ifa->ifa_netmask = (struct sockaddr *)sdl;
728 sdl->sdl_len = masklen;
729 while (namelen != 0)
730 sdl->sdl_data[--namelen] = 0xff;
731 TAILQ_INSERT_HEAD(&ifp->if_addrhead, ifa, ifa_link);
732 /* Reliably crash if used uninitialized. */
733 ifp->if_broadcastaddr = NULL;
734
735 if (ifp->if_type == IFT_ETHER) {
736 ifp->if_hw_addr = malloc(ifp->if_addrlen, M_IFADDR,
737 M_WAITOK | M_ZERO);
738 }
739
740 #if defined(INET) || defined(INET6)
741 /* Use defaults for TSO, if nothing is set */
742 if (ifp->if_hw_tsomax == 0 &&
743 ifp->if_hw_tsomaxsegcount == 0 &&
744 ifp->if_hw_tsomaxsegsize == 0) {
745 /*
746 * The TSO defaults needs to be such that an
747 * NFS mbuf list of 35 mbufs totalling just
748 * below 64K works and that a chain of mbufs
749 * can be defragged into at most 32 segments:
750 */
751 ifp->if_hw_tsomax = min(IP_MAXPACKET, (32 * MCLBYTES) -
752 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN));
753 ifp->if_hw_tsomaxsegcount = 35;
754 ifp->if_hw_tsomaxsegsize = 2048; /* 2K */
755
756 /* XXX some drivers set IFCAP_TSO after ethernet attach */
757 if (ifp->if_capabilities & IFCAP_TSO) {
758 if_printf(ifp, "Using defaults for TSO: %u/%u/%u\n",
759 ifp->if_hw_tsomax,
760 ifp->if_hw_tsomaxsegcount,
761 ifp->if_hw_tsomaxsegsize);
762 }
763 }
764 #endif
765 }
766 #ifdef VIMAGE
767 else {
768 /*
769 * Update the interface index in the link layer address
770 * of the interface.
771 */
772 for (ifa = ifp->if_addr; ifa != NULL;
773 ifa = TAILQ_NEXT(ifa, ifa_link)) {
774 if (ifa->ifa_addr->sa_family == AF_LINK) {
775 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
776 sdl->sdl_index = ifp->if_index;
777 }
778 }
779 }
780 #endif
781
782 IFNET_WLOCK();
783 TAILQ_INSERT_TAIL(&V_ifnet, ifp, if_link);
784 #ifdef VIMAGE
785 curvnet->vnet_ifcnt++;
786 #endif
787 IFNET_WUNLOCK();
788
789 if (domain_init_status >= 2)
790 if_attachdomain1(ifp);
791
792 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
793 if (IS_DEFAULT_VNET(curvnet))
794 devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
795
796 /* Announce the interface. */
797 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
798 }
799
800 static void
if_attachdomain(void * dummy)801 if_attachdomain(void *dummy)
802 {
803 struct ifnet *ifp;
804
805 TAILQ_FOREACH(ifp, &V_ifnet, if_link)
806 if_attachdomain1(ifp);
807 }
808 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_SECOND,
809 if_attachdomain, NULL);
810
811 static void
if_attachdomain1(struct ifnet * ifp)812 if_attachdomain1(struct ifnet *ifp)
813 {
814 struct domain *dp;
815
816 /*
817 * Since dp->dom_ifattach calls malloc() with M_WAITOK, we
818 * cannot lock ifp->if_afdata initialization, entirely.
819 */
820 if (IF_AFDATA_TRYLOCK(ifp) == 0)
821 return;
822 if (ifp->if_afdata_initialized >= domain_init_status) {
823 IF_AFDATA_UNLOCK(ifp);
824 log(LOG_WARNING, "%s called more than once on %s\n",
825 __func__, ifp->if_xname);
826 return;
827 }
828 ifp->if_afdata_initialized = domain_init_status;
829 IF_AFDATA_UNLOCK(ifp);
830
831 /* address family dependent data region */
832 bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
833 for (dp = domains; dp; dp = dp->dom_next) {
834 if (dp->dom_ifattach)
835 ifp->if_afdata[dp->dom_family] =
836 (*dp->dom_ifattach)(ifp);
837 }
838 }
839
840 /*
841 * Remove any unicast or broadcast network addresses from an interface.
842 */
843 void
if_purgeaddrs(struct ifnet * ifp)844 if_purgeaddrs(struct ifnet *ifp)
845 {
846 struct ifaddr *ifa, *next;
847
848 TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
849 if (ifa->ifa_addr->sa_family == AF_LINK)
850 continue;
851 #ifdef INET
852 /* XXX: Ugly!! ad hoc just for INET */
853 if (ifa->ifa_addr->sa_family == AF_INET) {
854 struct ifaliasreq ifr;
855
856 bzero(&ifr, sizeof(ifr));
857 ifr.ifra_addr = *ifa->ifa_addr;
858 if (ifa->ifa_dstaddr)
859 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
860 if (in_control(NULL, SIOCDIFADDR, (caddr_t)&ifr, ifp,
861 NULL) == 0)
862 continue;
863 }
864 #endif /* INET */
865 #ifdef INET6
866 if (ifa->ifa_addr->sa_family == AF_INET6) {
867 in6_purgeaddr(ifa);
868 /* ifp_addrhead is already updated */
869 continue;
870 }
871 #endif /* INET6 */
872 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
873 ifa_free(ifa);
874 }
875 }
876
877 /*
878 * Remove any multicast network addresses from an interface when an ifnet
879 * is going away.
880 */
881 static void
if_purgemaddrs(struct ifnet * ifp)882 if_purgemaddrs(struct ifnet *ifp)
883 {
884 struct ifmultiaddr *ifma;
885 struct ifmultiaddr *next;
886
887 IF_ADDR_WLOCK(ifp);
888 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
889 if_delmulti_locked(ifp, ifma, 1);
890 IF_ADDR_WUNLOCK(ifp);
891 }
892
893 /*
894 * Detach an interface, removing it from the list of "active" interfaces.
895 * If vmove flag is set on entry to if_detach_internal(), perform only a
896 * limited subset of cleanup tasks, given that we are moving an ifnet from
897 * one vnet to another, where it must be fully operational.
898 *
899 * XXXRW: There are some significant questions about event ordering, and
900 * how to prevent things from starting to use the interface during detach.
901 */
902 void
if_detach(struct ifnet * ifp)903 if_detach(struct ifnet *ifp)
904 {
905
906 CURVNET_SET_QUIET(ifp->if_vnet);
907 if_detach_internal(ifp, 0, NULL);
908 CURVNET_RESTORE();
909 }
910
911 static int
if_detach_internal(struct ifnet * ifp,int vmove,struct if_clone ** ifcp)912 if_detach_internal(struct ifnet *ifp, int vmove, struct if_clone **ifcp)
913 {
914 struct ifaddr *ifa;
915 struct radix_node_head *rnh;
916 int i, j;
917 struct domain *dp;
918 struct ifnet *iter;
919 int found = 0;
920
921 IFNET_WLOCK();
922 TAILQ_FOREACH(iter, &V_ifnet, if_link)
923 if (iter == ifp) {
924 TAILQ_REMOVE(&V_ifnet, ifp, if_link);
925 found = 1;
926 break;
927 }
928 #ifdef VIMAGE
929 if (found)
930 curvnet->vnet_ifcnt--;
931 #endif
932 IFNET_WUNLOCK();
933 if (!found) {
934 /*
935 * While we would want to panic here, we cannot
936 * guarantee that the interface is indeed still on
937 * the list given we don't hold locks all the way.
938 */
939 return (ENOENT);
940 #if 0
941 if (vmove)
942 panic("%s: ifp=%p not on the ifnet tailq %p",
943 __func__, ifp, &V_ifnet);
944 else
945 return; /* XXX this should panic as well? */
946 #endif
947 }
948
949 /* Check if this is a cloned interface or not. */
950 if (vmove && ifcp != NULL)
951 *ifcp = if_clone_findifc(ifp);
952
953 /*
954 * Remove/wait for pending events.
955 */
956 taskqueue_drain(taskqueue_swi, &ifp->if_linktask);
957
958 /*
959 * Remove routes and flush queues.
960 */
961 if_down(ifp);
962 #ifdef ALTQ
963 if (ALTQ_IS_ENABLED(&ifp->if_snd))
964 altq_disable(&ifp->if_snd);
965 if (ALTQ_IS_ATTACHED(&ifp->if_snd))
966 altq_detach(&ifp->if_snd);
967 #endif
968
969 if_purgeaddrs(ifp);
970
971 #ifdef INET
972 in_ifdetach(ifp);
973 #endif
974
975 #ifdef INET6
976 /*
977 * Remove all IPv6 kernel structs related to ifp. This should be done
978 * before removing routing entries below, since IPv6 interface direct
979 * routes are expected to be removed by the IPv6-specific kernel API.
980 * Otherwise, the kernel will detect some inconsistency and bark it.
981 */
982 in6_ifdetach(ifp);
983 #endif
984 if_purgemaddrs(ifp);
985
986 /* Announce that the interface is gone. */
987 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
988 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
989 if (IS_DEFAULT_VNET(curvnet))
990 devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
991
992 if (!vmove) {
993 /*
994 * Prevent further calls into the device driver via ifnet.
995 */
996 if_dead(ifp);
997
998 /*
999 * Remove link ifaddr pointer and maybe decrement if_index.
1000 * Clean up all addresses.
1001 */
1002 free(ifp->if_hw_addr, M_IFADDR);
1003 ifp->if_hw_addr = NULL;
1004 ifp->if_addr = NULL;
1005
1006 /* We can now free link ifaddr. */
1007 if (!TAILQ_EMPTY(&ifp->if_addrhead)) {
1008 ifa = TAILQ_FIRST(&ifp->if_addrhead);
1009 TAILQ_REMOVE(&ifp->if_addrhead, ifa, ifa_link);
1010 ifa_free(ifa);
1011 }
1012 }
1013
1014 /*
1015 * Delete all remaining routes using this interface
1016 * Unfortuneatly the only way to do this is to slog through
1017 * the entire routing table looking for routes which point
1018 * to this interface...oh well...
1019 */
1020 for (i = 1; i <= AF_MAX; i++) {
1021 for (j = 0; j < rt_numfibs; j++) {
1022 rnh = rt_tables_get_rnh(j, i);
1023 if (rnh == NULL)
1024 continue;
1025 RADIX_NODE_HEAD_LOCK(rnh);
1026 (void) rnh->rnh_walktree(rnh, if_rtdel, ifp);
1027 RADIX_NODE_HEAD_UNLOCK(rnh);
1028 }
1029 }
1030
1031 if_delgroups(ifp);
1032
1033 /*
1034 * We cannot hold the lock over dom_ifdetach calls as they might
1035 * sleep, for example trying to drain a callout, thus open up the
1036 * theoretical race with re-attaching.
1037 */
1038 IF_AFDATA_LOCK(ifp);
1039 i = ifp->if_afdata_initialized;
1040 ifp->if_afdata_initialized = 0;
1041 IF_AFDATA_UNLOCK(ifp);
1042 for (dp = domains; i > 0 && dp; dp = dp->dom_next) {
1043 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1044 (*dp->dom_ifdetach)(ifp,
1045 ifp->if_afdata[dp->dom_family]);
1046 }
1047
1048 return (0);
1049 }
1050
1051 #ifdef VIMAGE
1052 /*
1053 * if_vmove() performs a limited version of if_detach() in current
1054 * vnet and if_attach()es the ifnet to the vnet specified as 2nd arg.
1055 * An attempt is made to shrink if_index in current vnet, find an
1056 * unused if_index in target vnet and calls if_grow() if necessary,
1057 * and finally find an unused if_xname for the target vnet.
1058 */
1059 void
if_vmove(struct ifnet * ifp,struct vnet * new_vnet)1060 if_vmove(struct ifnet *ifp, struct vnet *new_vnet)
1061 {
1062 u_short idx;
1063 struct if_clone *ifc;
1064 int rc;
1065
1066 /*
1067 * Detach from current vnet, but preserve LLADDR info, do not
1068 * mark as dead etc. so that the ifnet can be reattached later.
1069 * If we cannot find it, we lost the race to someone else.
1070 */
1071 rc = if_detach_internal(ifp, 1, &ifc);
1072 if (rc != 0)
1073 return;
1074
1075 /*
1076 * Unlink the ifnet from ifindex_table[] in current vnet, and shrink
1077 * the if_index for that vnet if possible.
1078 *
1079 * NOTE: IFNET_WLOCK/IFNET_WUNLOCK() are assumed to be unvirtualized,
1080 * or we'd lock on one vnet and unlock on another.
1081 */
1082 IFNET_WLOCK();
1083 ifindex_free_locked(ifp->if_index);
1084 IFNET_WUNLOCK();
1085
1086 /*
1087 * Perform interface-specific reassignment tasks, if provided by
1088 * the driver.
1089 */
1090 if (ifp->if_reassign != NULL)
1091 ifp->if_reassign(ifp, new_vnet, NULL);
1092
1093 /*
1094 * Switch to the context of the target vnet.
1095 */
1096 CURVNET_SET_QUIET(new_vnet);
1097
1098 IFNET_WLOCK();
1099 if (ifindex_alloc_locked(&idx) != 0) {
1100 IFNET_WUNLOCK();
1101 panic("if_index overflow");
1102 }
1103 ifp->if_index = idx;
1104 ifnet_setbyindex_locked(ifp->if_index, ifp);
1105 IFNET_WUNLOCK();
1106
1107 if_attach_internal(ifp, 1, ifc);
1108
1109 CURVNET_RESTORE();
1110 }
1111
1112 /*
1113 * Move an ifnet to or from another child prison/vnet, specified by the jail id.
1114 */
1115 static int
if_vmove_loan(struct thread * td,struct ifnet * ifp,char * ifname,int jid)1116 if_vmove_loan(struct thread *td, struct ifnet *ifp, char *ifname, int jid)
1117 {
1118 struct prison *pr;
1119 struct ifnet *difp;
1120
1121 /* Try to find the prison within our visibility. */
1122 sx_slock(&allprison_lock);
1123 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1124 sx_sunlock(&allprison_lock);
1125 if (pr == NULL)
1126 return (ENXIO);
1127 prison_hold_locked(pr);
1128 mtx_unlock(&pr->pr_mtx);
1129
1130 /* Do not try to move the iface from and to the same prison. */
1131 if (pr->pr_vnet == ifp->if_vnet) {
1132 prison_free(pr);
1133 return (EEXIST);
1134 }
1135
1136 /* Make sure the named iface does not exists in the dst. prison/vnet. */
1137 /* XXX Lock interfaces to avoid races. */
1138 CURVNET_SET_QUIET(pr->pr_vnet);
1139 difp = ifunit(ifname);
1140 CURVNET_RESTORE();
1141 if (difp != NULL) {
1142 prison_free(pr);
1143 return (EEXIST);
1144 }
1145
1146 /* Move the interface into the child jail/vnet. */
1147 if_vmove(ifp, pr->pr_vnet);
1148
1149 /* Report the new if_xname back to the userland. */
1150 sprintf(ifname, "%s", ifp->if_xname);
1151
1152 prison_free(pr);
1153 return (0);
1154 }
1155
1156 static int
if_vmove_reclaim(struct thread * td,char * ifname,int jid)1157 if_vmove_reclaim(struct thread *td, char *ifname, int jid)
1158 {
1159 struct prison *pr;
1160 struct vnet *vnet_dst;
1161 struct ifnet *ifp;
1162
1163 /* Try to find the prison within our visibility. */
1164 sx_slock(&allprison_lock);
1165 pr = prison_find_child(td->td_ucred->cr_prison, jid);
1166 sx_sunlock(&allprison_lock);
1167 if (pr == NULL)
1168 return (ENXIO);
1169 prison_hold_locked(pr);
1170 mtx_unlock(&pr->pr_mtx);
1171
1172 /* Make sure the named iface exists in the source prison/vnet. */
1173 CURVNET_SET(pr->pr_vnet);
1174 ifp = ifunit(ifname); /* XXX Lock to avoid races. */
1175 if (ifp == NULL) {
1176 CURVNET_RESTORE();
1177 prison_free(pr);
1178 return (ENXIO);
1179 }
1180
1181 /* Do not try to move the iface from and to the same prison. */
1182 vnet_dst = TD_TO_VNET(td);
1183 if (vnet_dst == ifp->if_vnet) {
1184 CURVNET_RESTORE();
1185 prison_free(pr);
1186 return (EEXIST);
1187 }
1188
1189 /* Get interface back from child jail/vnet. */
1190 if_vmove(ifp, vnet_dst);
1191 CURVNET_RESTORE();
1192
1193 /* Report the new if_xname back to the userland. */
1194 sprintf(ifname, "%s", ifp->if_xname);
1195
1196 prison_free(pr);
1197 return (0);
1198 }
1199 #endif /* VIMAGE */
1200
1201 /*
1202 * Add a group to an interface
1203 */
1204 int
if_addgroup(struct ifnet * ifp,const char * groupname)1205 if_addgroup(struct ifnet *ifp, const char *groupname)
1206 {
1207 struct ifg_list *ifgl;
1208 struct ifg_group *ifg = NULL;
1209 struct ifg_member *ifgm;
1210 int new = 0;
1211
1212 if (groupname[0] && groupname[strlen(groupname) - 1] >= '0' &&
1213 groupname[strlen(groupname) - 1] <= '9')
1214 return (EINVAL);
1215
1216 IFNET_WLOCK();
1217 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1218 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname)) {
1219 IFNET_WUNLOCK();
1220 return (EEXIST);
1221 }
1222
1223 if ((ifgl = (struct ifg_list *)malloc(sizeof(struct ifg_list), M_TEMP,
1224 M_NOWAIT)) == NULL) {
1225 IFNET_WUNLOCK();
1226 return (ENOMEM);
1227 }
1228
1229 if ((ifgm = (struct ifg_member *)malloc(sizeof(struct ifg_member),
1230 M_TEMP, M_NOWAIT)) == NULL) {
1231 free(ifgl, M_TEMP);
1232 IFNET_WUNLOCK();
1233 return (ENOMEM);
1234 }
1235
1236 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1237 if (!strcmp(ifg->ifg_group, groupname))
1238 break;
1239
1240 if (ifg == NULL) {
1241 if ((ifg = (struct ifg_group *)malloc(sizeof(struct ifg_group),
1242 M_TEMP, M_NOWAIT)) == NULL) {
1243 free(ifgl, M_TEMP);
1244 free(ifgm, M_TEMP);
1245 IFNET_WUNLOCK();
1246 return (ENOMEM);
1247 }
1248 strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1249 ifg->ifg_refcnt = 0;
1250 TAILQ_INIT(&ifg->ifg_members);
1251 TAILQ_INSERT_TAIL(&V_ifg_head, ifg, ifg_next);
1252 new = 1;
1253 }
1254
1255 ifg->ifg_refcnt++;
1256 ifgl->ifgl_group = ifg;
1257 ifgm->ifgm_ifp = ifp;
1258
1259 IF_ADDR_WLOCK(ifp);
1260 TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1261 TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1262 IF_ADDR_WUNLOCK(ifp);
1263
1264 IFNET_WUNLOCK();
1265
1266 if (new)
1267 EVENTHANDLER_INVOKE(group_attach_event, ifg);
1268 EVENTHANDLER_INVOKE(group_change_event, groupname);
1269
1270 return (0);
1271 }
1272
1273 /*
1274 * Remove a group from an interface
1275 */
1276 int
if_delgroup(struct ifnet * ifp,const char * groupname)1277 if_delgroup(struct ifnet *ifp, const char *groupname)
1278 {
1279 struct ifg_list *ifgl;
1280 struct ifg_member *ifgm;
1281
1282 IFNET_WLOCK();
1283 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1284 if (!strcmp(ifgl->ifgl_group->ifg_group, groupname))
1285 break;
1286 if (ifgl == NULL) {
1287 IFNET_WUNLOCK();
1288 return (ENOENT);
1289 }
1290
1291 IF_ADDR_WLOCK(ifp);
1292 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1293 IF_ADDR_WUNLOCK(ifp);
1294
1295 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1296 if (ifgm->ifgm_ifp == ifp)
1297 break;
1298
1299 if (ifgm != NULL) {
1300 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1301 free(ifgm, M_TEMP);
1302 }
1303
1304 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1305 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1306 IFNET_WUNLOCK();
1307 EVENTHANDLER_INVOKE(group_detach_event, ifgl->ifgl_group);
1308 free(ifgl->ifgl_group, M_TEMP);
1309 } else
1310 IFNET_WUNLOCK();
1311
1312 free(ifgl, M_TEMP);
1313
1314 EVENTHANDLER_INVOKE(group_change_event, groupname);
1315
1316 return (0);
1317 }
1318
1319 /*
1320 * Remove an interface from all groups
1321 */
1322 static void
if_delgroups(struct ifnet * ifp)1323 if_delgroups(struct ifnet *ifp)
1324 {
1325 struct ifg_list *ifgl;
1326 struct ifg_member *ifgm;
1327 char groupname[IFNAMSIZ];
1328
1329 IFNET_WLOCK();
1330 while (!TAILQ_EMPTY(&ifp->if_groups)) {
1331 ifgl = TAILQ_FIRST(&ifp->if_groups);
1332
1333 strlcpy(groupname, ifgl->ifgl_group->ifg_group, IFNAMSIZ);
1334
1335 IF_ADDR_WLOCK(ifp);
1336 TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1337 IF_ADDR_WUNLOCK(ifp);
1338
1339 TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next)
1340 if (ifgm->ifgm_ifp == ifp)
1341 break;
1342
1343 if (ifgm != NULL) {
1344 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm,
1345 ifgm_next);
1346 free(ifgm, M_TEMP);
1347 }
1348
1349 if (--ifgl->ifgl_group->ifg_refcnt == 0) {
1350 TAILQ_REMOVE(&V_ifg_head, ifgl->ifgl_group, ifg_next);
1351 IFNET_WUNLOCK();
1352 EVENTHANDLER_INVOKE(group_detach_event,
1353 ifgl->ifgl_group);
1354 free(ifgl->ifgl_group, M_TEMP);
1355 } else
1356 IFNET_WUNLOCK();
1357
1358 free(ifgl, M_TEMP);
1359
1360 EVENTHANDLER_INVOKE(group_change_event, groupname);
1361
1362 IFNET_WLOCK();
1363 }
1364 IFNET_WUNLOCK();
1365 }
1366
1367 /*
1368 * Stores all groups from an interface in memory pointed
1369 * to by data
1370 */
1371 static int
if_getgroup(struct ifgroupreq * data,struct ifnet * ifp)1372 if_getgroup(struct ifgroupreq *data, struct ifnet *ifp)
1373 {
1374 int len, error;
1375 struct ifg_list *ifgl;
1376 struct ifg_req ifgrq, *ifgp;
1377 struct ifgroupreq *ifgr = data;
1378
1379 if (ifgr->ifgr_len == 0) {
1380 IF_ADDR_RLOCK(ifp);
1381 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1382 ifgr->ifgr_len += sizeof(struct ifg_req);
1383 IF_ADDR_RUNLOCK(ifp);
1384 return (0);
1385 }
1386
1387 len = ifgr->ifgr_len;
1388 ifgp = ifgr->ifgr_groups;
1389 /* XXX: wire */
1390 IF_ADDR_RLOCK(ifp);
1391 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1392 if (len < sizeof(ifgrq)) {
1393 IF_ADDR_RUNLOCK(ifp);
1394 return (EINVAL);
1395 }
1396 bzero(&ifgrq, sizeof ifgrq);
1397 strlcpy(ifgrq.ifgrq_group, ifgl->ifgl_group->ifg_group,
1398 sizeof(ifgrq.ifgrq_group));
1399 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1400 IF_ADDR_RUNLOCK(ifp);
1401 return (error);
1402 }
1403 len -= sizeof(ifgrq);
1404 ifgp++;
1405 }
1406 IF_ADDR_RUNLOCK(ifp);
1407
1408 return (0);
1409 }
1410
1411 /*
1412 * Stores all members of a group in memory pointed to by data
1413 */
1414 static int
if_getgroupmembers(struct ifgroupreq * data)1415 if_getgroupmembers(struct ifgroupreq *data)
1416 {
1417 struct ifgroupreq *ifgr = data;
1418 struct ifg_group *ifg;
1419 struct ifg_member *ifgm;
1420 struct ifg_req ifgrq, *ifgp;
1421 int len, error;
1422
1423 IFNET_RLOCK();
1424 TAILQ_FOREACH(ifg, &V_ifg_head, ifg_next)
1425 if (!strcmp(ifg->ifg_group, ifgr->ifgr_name))
1426 break;
1427 if (ifg == NULL) {
1428 IFNET_RUNLOCK();
1429 return (ENOENT);
1430 }
1431
1432 if (ifgr->ifgr_len == 0) {
1433 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1434 ifgr->ifgr_len += sizeof(ifgrq);
1435 IFNET_RUNLOCK();
1436 return (0);
1437 }
1438
1439 len = ifgr->ifgr_len;
1440 ifgp = ifgr->ifgr_groups;
1441 TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1442 if (len < sizeof(ifgrq)) {
1443 IFNET_RUNLOCK();
1444 return (EINVAL);
1445 }
1446 bzero(&ifgrq, sizeof ifgrq);
1447 strlcpy(ifgrq.ifgrq_member, ifgm->ifgm_ifp->if_xname,
1448 sizeof(ifgrq.ifgrq_member));
1449 if ((error = copyout(&ifgrq, ifgp, sizeof(struct ifg_req)))) {
1450 IFNET_RUNLOCK();
1451 return (error);
1452 }
1453 len -= sizeof(ifgrq);
1454 ifgp++;
1455 }
1456 IFNET_RUNLOCK();
1457
1458 return (0);
1459 }
1460
1461 /*
1462 * Delete Routes for a Network Interface
1463 *
1464 * Called for each routing entry via the rnh->rnh_walktree() call above
1465 * to delete all route entries referencing a detaching network interface.
1466 *
1467 * Arguments:
1468 * rn pointer to node in the routing table
1469 * arg argument passed to rnh->rnh_walktree() - detaching interface
1470 *
1471 * Returns:
1472 * 0 successful
1473 * errno failed - reason indicated
1474 *
1475 */
1476 static int
if_rtdel(struct radix_node * rn,void * arg)1477 if_rtdel(struct radix_node *rn, void *arg)
1478 {
1479 struct rtentry *rt = (struct rtentry *)rn;
1480 struct ifnet *ifp = arg;
1481 int err;
1482
1483 if (rt->rt_ifp == ifp) {
1484
1485 /*
1486 * Protect (sorta) against walktree recursion problems
1487 * with cloned routes
1488 */
1489 if ((rt->rt_flags & RTF_UP) == 0)
1490 return (0);
1491
1492 err = rtrequest_fib(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1493 rt_mask(rt),
1494 rt->rt_flags|RTF_RNH_LOCKED|RTF_PINNED,
1495 (struct rtentry **) NULL, rt->rt_fibnum);
1496 if (err) {
1497 log(LOG_WARNING, "if_rtdel: error %d\n", err);
1498 }
1499 }
1500
1501 return (0);
1502 }
1503
1504 /*
1505 * A compatibility function returns ifnet counter values.
1506 */
1507 uint64_t
if_get_counter_default(struct ifnet * ifp,ift_counter cnt)1508 if_get_counter_default(struct ifnet *ifp, ift_counter cnt)
1509 {
1510
1511 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1512 switch (cnt) {
1513 case IFCOUNTER_IPACKETS:
1514 return (ifp->if_ipackets);
1515 case IFCOUNTER_IERRORS:
1516 return (ifp->if_ierrors);
1517 case IFCOUNTER_OPACKETS:
1518 return (ifp->if_opackets);
1519 case IFCOUNTER_OERRORS:
1520 return (ifp->if_oerrors);
1521 case IFCOUNTER_COLLISIONS:
1522 return (ifp->if_collisions);
1523 case IFCOUNTER_IBYTES:
1524 return (ifp->if_ibytes);
1525 case IFCOUNTER_OBYTES:
1526 return (ifp->if_obytes);
1527 case IFCOUNTER_IMCASTS:
1528 return (ifp->if_imcasts);
1529 case IFCOUNTER_OMCASTS:
1530 return (ifp->if_omcasts);
1531 case IFCOUNTER_IQDROPS:
1532 return (ifp->if_iqdrops);
1533 #ifdef _IFI_OQDROPS
1534 case IFCOUNTER_OQDROPS:
1535 return (ifp->if_oqdrops);
1536 #endif
1537 case IFCOUNTER_NOPROTO:
1538 return (ifp->if_noproto);
1539 default:
1540 break;
1541 };
1542 return (0);
1543 }
1544
1545 /*
1546 * Increase an ifnet counter. Usually used for counters shared
1547 * between the stack and a driver, but function supports them all.
1548 */
1549 void
if_inc_counter(struct ifnet * ifp,ift_counter cnt,int64_t inc)1550 if_inc_counter(struct ifnet *ifp, ift_counter cnt, int64_t inc)
1551 {
1552
1553 KASSERT(cnt < IFCOUNTERS, ("%s: invalid cnt %d", __func__, cnt));
1554 switch (cnt) {
1555 case IFCOUNTER_IPACKETS:
1556 ifp->if_ipackets += inc;
1557 break;
1558 case IFCOUNTER_IERRORS:
1559 ifp->if_ierrors += inc;
1560 break;
1561 case IFCOUNTER_OPACKETS:
1562 ifp->if_opackets += inc;
1563 break;
1564 case IFCOUNTER_OERRORS:
1565 ifp->if_oerrors += inc;
1566 break;
1567 case IFCOUNTER_COLLISIONS:
1568 ifp->if_collisions += inc;
1569 break;
1570 case IFCOUNTER_IBYTES:
1571 ifp->if_ibytes += inc;
1572 break;
1573 case IFCOUNTER_OBYTES:
1574 ifp->if_obytes += inc;
1575 break;
1576 case IFCOUNTER_IMCASTS:
1577 ifp->if_imcasts += inc;
1578 break;
1579 case IFCOUNTER_OMCASTS:
1580 ifp->if_omcasts += inc;
1581 break;
1582 case IFCOUNTER_IQDROPS:
1583 ifp->if_iqdrops += inc;
1584 break;
1585 #ifdef _IFI_OQDROPS
1586 case IFCOUNTER_OQDROPS:
1587 ifp->if_oqdrops += inc;
1588 break;
1589 #endif
1590 case IFCOUNTER_NOPROTO:
1591 ifp->if_noproto += inc;
1592 break;
1593 default:
1594 break;
1595 };
1596 }
1597
1598 /*
1599 * Wrapper functions for struct ifnet address list locking macros. These are
1600 * used by kernel modules to avoid encoding programming interface or binary
1601 * interface assumptions that may be violated when kernel-internal locking
1602 * approaches change.
1603 */
1604 void
if_addr_rlock(struct ifnet * ifp)1605 if_addr_rlock(struct ifnet *ifp)
1606 {
1607
1608 IF_ADDR_RLOCK(ifp);
1609 }
1610
1611 void
if_addr_runlock(struct ifnet * ifp)1612 if_addr_runlock(struct ifnet *ifp)
1613 {
1614
1615 IF_ADDR_RUNLOCK(ifp);
1616 }
1617
1618 void
if_maddr_rlock(struct ifnet * ifp)1619 if_maddr_rlock(struct ifnet *ifp)
1620 {
1621
1622 IF_ADDR_RLOCK(ifp);
1623 }
1624
1625 void
if_maddr_runlock(struct ifnet * ifp)1626 if_maddr_runlock(struct ifnet *ifp)
1627 {
1628
1629 IF_ADDR_RUNLOCK(ifp);
1630 }
1631
1632 /*
1633 * Initialization, destruction and refcounting functions for ifaddrs.
1634 */
1635 void
ifa_init(struct ifaddr * ifa)1636 ifa_init(struct ifaddr *ifa)
1637 {
1638
1639 mtx_init(&ifa->ifa_mtx, "ifaddr", NULL, MTX_DEF);
1640 refcount_init(&ifa->ifa_refcnt, 1);
1641 ifa->if_data.ifi_datalen = sizeof(ifa->if_data);
1642 }
1643
1644 void
ifa_ref(struct ifaddr * ifa)1645 ifa_ref(struct ifaddr *ifa)
1646 {
1647
1648 refcount_acquire(&ifa->ifa_refcnt);
1649 }
1650
1651 void
ifa_free(struct ifaddr * ifa)1652 ifa_free(struct ifaddr *ifa)
1653 {
1654
1655 if (refcount_release(&ifa->ifa_refcnt)) {
1656 mtx_destroy(&ifa->ifa_mtx);
1657 free(ifa, M_IFADDR);
1658 }
1659 }
1660
1661 int
ifa_add_loopback_route(struct ifaddr * ifa,struct sockaddr * ia)1662 ifa_add_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1663 {
1664 int error = 0;
1665 struct rtentry *rt = NULL;
1666 struct rt_addrinfo info;
1667 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK};
1668
1669 bzero(&info, sizeof(info));
1670 info.rti_ifp = V_loif;
1671 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC | RTF_PINNED;
1672 info.rti_info[RTAX_DST] = ia;
1673 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1674 error = rtrequest1_fib(RTM_ADD, &info, &rt, ifa->ifa_ifp->if_fib);
1675
1676 if (error == 0 && rt != NULL) {
1677 RT_LOCK(rt);
1678 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type =
1679 ifa->ifa_ifp->if_type;
1680 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index =
1681 ifa->ifa_ifp->if_index;
1682 RT_REMREF(rt);
1683 RT_UNLOCK(rt);
1684 } else if (error != 0)
1685 log(LOG_DEBUG, "%s: insertion failed: %u\n", __func__, error);
1686
1687 return (error);
1688 }
1689
1690 int
ifa_del_loopback_route(struct ifaddr * ifa,struct sockaddr * ia)1691 ifa_del_loopback_route(struct ifaddr *ifa, struct sockaddr *ia)
1692 {
1693 int error = 0;
1694 struct rt_addrinfo info;
1695 struct sockaddr_dl null_sdl;
1696
1697 bzero(&null_sdl, sizeof(null_sdl));
1698 null_sdl.sdl_len = sizeof(null_sdl);
1699 null_sdl.sdl_family = AF_LINK;
1700 null_sdl.sdl_type = ifa->ifa_ifp->if_type;
1701 null_sdl.sdl_index = ifa->ifa_ifp->if_index;
1702 bzero(&info, sizeof(info));
1703 info.rti_flags = ifa->ifa_flags | RTF_HOST | RTF_STATIC | RTF_PINNED;
1704 info.rti_info[RTAX_DST] = ia;
1705 info.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&null_sdl;
1706 error = rtrequest1_fib(RTM_DELETE, &info, NULL, ifa->ifa_ifp->if_fib);
1707
1708 if (error != 0)
1709 log(LOG_DEBUG, "%s: deletion failed: %u\n", __func__, error);
1710
1711 return (error);
1712 }
1713
1714 /*
1715 * XXX: Because sockaddr_dl has deeper structure than the sockaddr
1716 * structs used to represent other address families, it is necessary
1717 * to perform a different comparison.
1718 */
1719
1720 #define sa_equal(a1, a2) \
1721 (bcmp((a1), (a2), ((a1))->sa_len) == 0)
1722
1723 #define sa_dl_equal(a1, a2) \
1724 ((((struct sockaddr_dl *)(a1))->sdl_len == \
1725 ((struct sockaddr_dl *)(a2))->sdl_len) && \
1726 (bcmp(LLADDR((struct sockaddr_dl *)(a1)), \
1727 LLADDR((struct sockaddr_dl *)(a2)), \
1728 ((struct sockaddr_dl *)(a1))->sdl_alen) == 0))
1729
1730 /*
1731 * Locate an interface based on a complete address.
1732 */
1733 /*ARGSUSED*/
1734 static struct ifaddr *
ifa_ifwithaddr_internal(struct sockaddr * addr,int getref)1735 ifa_ifwithaddr_internal(struct sockaddr *addr, int getref)
1736 {
1737 struct ifnet *ifp;
1738 struct ifaddr *ifa;
1739
1740 IFNET_RLOCK_NOSLEEP();
1741 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1742 IF_ADDR_RLOCK(ifp);
1743 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1744 if (ifa->ifa_addr->sa_family != addr->sa_family)
1745 continue;
1746 if (sa_equal(addr, ifa->ifa_addr)) {
1747 if (getref)
1748 ifa_ref(ifa);
1749 IF_ADDR_RUNLOCK(ifp);
1750 goto done;
1751 }
1752 /* IP6 doesn't have broadcast */
1753 if ((ifp->if_flags & IFF_BROADCAST) &&
1754 ifa->ifa_broadaddr &&
1755 ifa->ifa_broadaddr->sa_len != 0 &&
1756 sa_equal(ifa->ifa_broadaddr, addr)) {
1757 if (getref)
1758 ifa_ref(ifa);
1759 IF_ADDR_RUNLOCK(ifp);
1760 goto done;
1761 }
1762 }
1763 IF_ADDR_RUNLOCK(ifp);
1764 }
1765 ifa = NULL;
1766 done:
1767 IFNET_RUNLOCK_NOSLEEP();
1768 return (ifa);
1769 }
1770
1771 struct ifaddr *
ifa_ifwithaddr(struct sockaddr * addr)1772 ifa_ifwithaddr(struct sockaddr *addr)
1773 {
1774
1775 return (ifa_ifwithaddr_internal(addr, 1));
1776 }
1777
1778 int
ifa_ifwithaddr_check(struct sockaddr * addr)1779 ifa_ifwithaddr_check(struct sockaddr *addr)
1780 {
1781
1782 return (ifa_ifwithaddr_internal(addr, 0) != NULL);
1783 }
1784
1785 /*
1786 * Locate an interface based on the broadcast address.
1787 */
1788 /* ARGSUSED */
1789 struct ifaddr *
ifa_ifwithbroadaddr(struct sockaddr * addr)1790 ifa_ifwithbroadaddr(struct sockaddr *addr)
1791 {
1792 struct ifnet *ifp;
1793 struct ifaddr *ifa;
1794
1795 IFNET_RLOCK_NOSLEEP();
1796 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1797 IF_ADDR_RLOCK(ifp);
1798 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1799 if (ifa->ifa_addr->sa_family != addr->sa_family)
1800 continue;
1801 if ((ifp->if_flags & IFF_BROADCAST) &&
1802 ifa->ifa_broadaddr &&
1803 ifa->ifa_broadaddr->sa_len != 0 &&
1804 sa_equal(ifa->ifa_broadaddr, addr)) {
1805 ifa_ref(ifa);
1806 IF_ADDR_RUNLOCK(ifp);
1807 goto done;
1808 }
1809 }
1810 IF_ADDR_RUNLOCK(ifp);
1811 }
1812 ifa = NULL;
1813 done:
1814 IFNET_RUNLOCK_NOSLEEP();
1815 return (ifa);
1816 }
1817
1818 /*
1819 * Locate the point to point interface with a given destination address.
1820 */
1821 /*ARGSUSED*/
1822 struct ifaddr *
ifa_ifwithdstaddr_fib(struct sockaddr * addr,int fibnum)1823 ifa_ifwithdstaddr_fib(struct sockaddr *addr, int fibnum)
1824 {
1825 struct ifnet *ifp;
1826 struct ifaddr *ifa;
1827
1828 IFNET_RLOCK_NOSLEEP();
1829 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1830 if ((ifp->if_flags & IFF_POINTOPOINT) == 0)
1831 continue;
1832 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1833 continue;
1834 IF_ADDR_RLOCK(ifp);
1835 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1836 if (ifa->ifa_addr->sa_family != addr->sa_family)
1837 continue;
1838 if (ifa->ifa_dstaddr != NULL &&
1839 sa_equal(addr, ifa->ifa_dstaddr)) {
1840 ifa_ref(ifa);
1841 IF_ADDR_RUNLOCK(ifp);
1842 goto done;
1843 }
1844 }
1845 IF_ADDR_RUNLOCK(ifp);
1846 }
1847 ifa = NULL;
1848 done:
1849 IFNET_RUNLOCK_NOSLEEP();
1850 return (ifa);
1851 }
1852
1853 struct ifaddr *
ifa_ifwithdstaddr(struct sockaddr * addr)1854 ifa_ifwithdstaddr(struct sockaddr *addr)
1855 {
1856
1857 return (ifa_ifwithdstaddr_fib(addr, RT_ALL_FIBS));
1858 }
1859
1860 /*
1861 * Find an interface on a specific network. If many, choice
1862 * is most specific found.
1863 */
1864 struct ifaddr *
ifa_ifwithnet_fib(struct sockaddr * addr,int ignore_ptp,int fibnum)1865 ifa_ifwithnet_fib(struct sockaddr *addr, int ignore_ptp, int fibnum)
1866 {
1867 struct ifnet *ifp;
1868 struct ifaddr *ifa;
1869 struct ifaddr *ifa_maybe = NULL;
1870 u_int af = addr->sa_family;
1871 char *addr_data = addr->sa_data, *cplim;
1872
1873 /*
1874 * AF_LINK addresses can be looked up directly by their index number,
1875 * so do that if we can.
1876 */
1877 if (af == AF_LINK) {
1878 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1879 if (sdl->sdl_index && sdl->sdl_index <= V_if_index)
1880 return (ifaddr_byindex(sdl->sdl_index));
1881 }
1882
1883 /*
1884 * Scan though each interface, looking for ones that have addresses
1885 * in this address family and the requested fib. Maintain a reference
1886 * on ifa_maybe once we find one, as we release the IF_ADDR_RLOCK() that
1887 * kept it stable when we move onto the next interface.
1888 */
1889 IFNET_RLOCK_NOSLEEP();
1890 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
1891 if ((fibnum != RT_ALL_FIBS) && (ifp->if_fib != fibnum))
1892 continue;
1893 IF_ADDR_RLOCK(ifp);
1894 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1895 char *cp, *cp2, *cp3;
1896
1897 if (ifa->ifa_addr->sa_family != af)
1898 next: continue;
1899 if (af == AF_INET &&
1900 ifp->if_flags & IFF_POINTOPOINT && !ignore_ptp) {
1901 /*
1902 * This is a bit broken as it doesn't
1903 * take into account that the remote end may
1904 * be a single node in the network we are
1905 * looking for.
1906 * The trouble is that we don't know the
1907 * netmask for the remote end.
1908 */
1909 if (ifa->ifa_dstaddr != NULL &&
1910 sa_equal(addr, ifa->ifa_dstaddr)) {
1911 ifa_ref(ifa);
1912 IF_ADDR_RUNLOCK(ifp);
1913 goto done;
1914 }
1915 } else {
1916 /*
1917 * if we have a special address handler,
1918 * then use it instead of the generic one.
1919 */
1920 if (ifa->ifa_claim_addr) {
1921 if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1922 ifa_ref(ifa);
1923 IF_ADDR_RUNLOCK(ifp);
1924 goto done;
1925 }
1926 continue;
1927 }
1928
1929 /*
1930 * Scan all the bits in the ifa's address.
1931 * If a bit dissagrees with what we are
1932 * looking for, mask it with the netmask
1933 * to see if it really matters.
1934 * (A byte at a time)
1935 */
1936 if (ifa->ifa_netmask == 0)
1937 continue;
1938 cp = addr_data;
1939 cp2 = ifa->ifa_addr->sa_data;
1940 cp3 = ifa->ifa_netmask->sa_data;
1941 cplim = ifa->ifa_netmask->sa_len
1942 + (char *)ifa->ifa_netmask;
1943 while (cp3 < cplim)
1944 if ((*cp++ ^ *cp2++) & *cp3++)
1945 goto next; /* next address! */
1946 /*
1947 * If the netmask of what we just found
1948 * is more specific than what we had before
1949 * (if we had one), or if the virtual status
1950 * of new prefix is better than of the old one,
1951 * then remember the new one before continuing
1952 * to search for an even better one.
1953 */
1954 if (ifa_maybe == NULL ||
1955 ifa_preferred(ifa_maybe, ifa) ||
1956 rn_refines((caddr_t)ifa->ifa_netmask,
1957 (caddr_t)ifa_maybe->ifa_netmask)) {
1958 if (ifa_maybe != NULL)
1959 ifa_free(ifa_maybe);
1960 ifa_maybe = ifa;
1961 ifa_ref(ifa_maybe);
1962 }
1963 }
1964 }
1965 IF_ADDR_RUNLOCK(ifp);
1966 }
1967 ifa = ifa_maybe;
1968 ifa_maybe = NULL;
1969 done:
1970 IFNET_RUNLOCK_NOSLEEP();
1971 if (ifa_maybe != NULL)
1972 ifa_free(ifa_maybe);
1973 return (ifa);
1974 }
1975
1976 struct ifaddr *
ifa_ifwithnet(struct sockaddr * addr,int ignore_ptp)1977 ifa_ifwithnet(struct sockaddr *addr, int ignore_ptp)
1978 {
1979
1980 return (ifa_ifwithnet_fib(addr, ignore_ptp, RT_ALL_FIBS));
1981 }
1982
1983 /*
1984 * Find an interface address specific to an interface best matching
1985 * a given address.
1986 */
1987 struct ifaddr *
ifaof_ifpforaddr(struct sockaddr * addr,struct ifnet * ifp)1988 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1989 {
1990 struct ifaddr *ifa;
1991 char *cp, *cp2, *cp3;
1992 char *cplim;
1993 struct ifaddr *ifa_maybe = NULL;
1994 u_int af = addr->sa_family;
1995
1996 if (af >= AF_MAX)
1997 return (NULL);
1998 IF_ADDR_RLOCK(ifp);
1999 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
2000 if (ifa->ifa_addr->sa_family != af)
2001 continue;
2002 if (ifa_maybe == NULL)
2003 ifa_maybe = ifa;
2004 if (ifa->ifa_netmask == 0) {
2005 if (sa_equal(addr, ifa->ifa_addr) ||
2006 (ifa->ifa_dstaddr &&
2007 sa_equal(addr, ifa->ifa_dstaddr)))
2008 goto done;
2009 continue;
2010 }
2011 if (ifp->if_flags & IFF_POINTOPOINT) {
2012 if (sa_equal(addr, ifa->ifa_dstaddr))
2013 goto done;
2014 } else {
2015 cp = addr->sa_data;
2016 cp2 = ifa->ifa_addr->sa_data;
2017 cp3 = ifa->ifa_netmask->sa_data;
2018 cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
2019 for (; cp3 < cplim; cp3++)
2020 if ((*cp++ ^ *cp2++) & *cp3)
2021 break;
2022 if (cp3 == cplim)
2023 goto done;
2024 }
2025 }
2026 ifa = ifa_maybe;
2027 done:
2028 if (ifa != NULL)
2029 ifa_ref(ifa);
2030 IF_ADDR_RUNLOCK(ifp);
2031 return (ifa);
2032 }
2033
2034 /*
2035 * See whether new ifa is better than current one:
2036 * 1) A non-virtual one is preferred over virtual.
2037 * 2) A virtual in master state preferred over any other state.
2038 *
2039 * Used in several address selecting functions.
2040 */
2041 int
ifa_preferred(struct ifaddr * cur,struct ifaddr * next)2042 ifa_preferred(struct ifaddr *cur, struct ifaddr *next)
2043 {
2044
2045 return (cur->ifa_carp && (!next->ifa_carp ||
2046 ((*carp_master_p)(next) && !(*carp_master_p)(cur))));
2047 }
2048
2049 #include <net/if_llatbl.h>
2050
2051 /*
2052 * Default action when installing a route with a Link Level gateway.
2053 * Lookup an appropriate real ifa to point to.
2054 * This should be moved to /sys/net/link.c eventually.
2055 */
2056 static void
link_rtrequest(int cmd,struct rtentry * rt,struct rt_addrinfo * info)2057 link_rtrequest(int cmd, struct rtentry *rt, struct rt_addrinfo *info)
2058 {
2059 struct ifaddr *ifa, *oifa;
2060 struct sockaddr *dst;
2061 struct ifnet *ifp;
2062
2063 RT_LOCK_ASSERT(rt);
2064
2065 if (cmd != RTM_ADD || ((ifa = rt->rt_ifa) == 0) ||
2066 ((ifp = ifa->ifa_ifp) == 0) || ((dst = rt_key(rt)) == 0))
2067 return;
2068 ifa = ifaof_ifpforaddr(dst, ifp);
2069 if (ifa) {
2070 oifa = rt->rt_ifa;
2071 rt->rt_ifa = ifa;
2072 ifa_free(oifa);
2073 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
2074 ifa->ifa_rtrequest(cmd, rt, info);
2075 }
2076 }
2077
2078 /*
2079 * Mark an interface down and notify protocols of
2080 * the transition.
2081 */
2082 static void
if_unroute(struct ifnet * ifp,int flag,int fam)2083 if_unroute(struct ifnet *ifp, int flag, int fam)
2084 {
2085 struct ifaddr *ifa;
2086
2087 KASSERT(flag == IFF_UP, ("if_unroute: flag != IFF_UP"));
2088
2089 ifp->if_flags &= ~flag;
2090 getmicrotime(&ifp->if_lastchange);
2091 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
2092 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
2093 pfctlinput(PRC_IFDOWN, ifa->ifa_addr);
2094 ifp->if_qflush(ifp);
2095
2096 if (ifp->if_carp)
2097 (*carp_linkstate_p)(ifp);
2098 rt_ifmsg(ifp);
2099 }
2100
2101 /*
2102 * Mark an interface up and notify protocols of
2103 * the transition.
2104 */
2105 static void
if_route(struct ifnet * ifp,int flag,int fam)2106 if_route(struct ifnet *ifp, int flag, int fam)
2107 {
2108 struct ifaddr *ifa;
2109
2110 KASSERT(flag == IFF_UP, ("if_route: flag != IFF_UP"));
2111
2112 ifp->if_flags |= flag;
2113 getmicrotime(&ifp->if_lastchange);
2114 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
2115 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
2116 pfctlinput(PRC_IFUP, ifa->ifa_addr);
2117 if (ifp->if_carp)
2118 (*carp_linkstate_p)(ifp);
2119 rt_ifmsg(ifp);
2120 #ifdef INET6
2121 in6_if_up(ifp);
2122 #endif
2123 }
2124
2125 void (*vlan_link_state_p)(struct ifnet *); /* XXX: private from if_vlan */
2126 void (*vlan_trunk_cap_p)(struct ifnet *); /* XXX: private from if_vlan */
2127 struct ifnet *(*vlan_trunkdev_p)(struct ifnet *);
2128 struct ifnet *(*vlan_devat_p)(struct ifnet *, uint16_t);
2129 int (*vlan_tag_p)(struct ifnet *, uint16_t *);
2130 int (*vlan_setcookie_p)(struct ifnet *, void *);
2131 void *(*vlan_cookie_p)(struct ifnet *);
2132
2133 /*
2134 * Handle a change in the interface link state. To avoid LORs
2135 * between driver lock and upper layer locks, as well as possible
2136 * recursions, we post event to taskqueue, and all job
2137 * is done in static do_link_state_change().
2138 */
2139 void
if_link_state_change(struct ifnet * ifp,int link_state)2140 if_link_state_change(struct ifnet *ifp, int link_state)
2141 {
2142 /* Return if state hasn't changed. */
2143 if (ifp->if_link_state == link_state)
2144 return;
2145
2146 ifp->if_link_state = link_state;
2147
2148 taskqueue_enqueue(taskqueue_swi, &ifp->if_linktask);
2149 }
2150
2151 static void
do_link_state_change(void * arg,int pending)2152 do_link_state_change(void *arg, int pending)
2153 {
2154 struct ifnet *ifp = (struct ifnet *)arg;
2155 int link_state = ifp->if_link_state;
2156 CURVNET_SET(ifp->if_vnet);
2157
2158 /* Notify that the link state has changed. */
2159 rt_ifmsg(ifp);
2160 if (ifp->if_vlantrunk != NULL)
2161 (*vlan_link_state_p)(ifp);
2162
2163 if ((ifp->if_type == IFT_ETHER || ifp->if_type == IFT_L2VLAN) &&
2164 IFP2AC(ifp)->ac_netgraph != NULL)
2165 (*ng_ether_link_state_p)(ifp, link_state);
2166 if (ifp->if_carp)
2167 (*carp_linkstate_p)(ifp);
2168 if (ifp->if_bridge)
2169 (*bridge_linkstate_p)(ifp);
2170 if (ifp->if_lagg)
2171 (*lagg_linkstate_p)(ifp, link_state);
2172
2173 if (IS_DEFAULT_VNET(curvnet))
2174 devctl_notify("IFNET", ifp->if_xname,
2175 (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN",
2176 NULL);
2177 if (pending > 1)
2178 if_printf(ifp, "%d link states coalesced\n", pending);
2179 if (log_link_state_change)
2180 log(LOG_NOTICE, "%s: link state changed to %s\n", ifp->if_xname,
2181 (link_state == LINK_STATE_UP) ? "UP" : "DOWN" );
2182 EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
2183 CURVNET_RESTORE();
2184 }
2185
2186 /*
2187 * Mark an interface down and notify protocols of
2188 * the transition.
2189 */
2190 void
if_down(struct ifnet * ifp)2191 if_down(struct ifnet *ifp)
2192 {
2193
2194 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
2195 if_unroute(ifp, IFF_UP, AF_UNSPEC);
2196 }
2197
2198 /*
2199 * Mark an interface up and notify protocols of
2200 * the transition.
2201 */
2202 void
if_up(struct ifnet * ifp)2203 if_up(struct ifnet *ifp)
2204 {
2205
2206 if_route(ifp, IFF_UP, AF_UNSPEC);
2207 EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
2208 }
2209
2210 /*
2211 * Flush an interface queue.
2212 */
2213 void
if_qflush(struct ifnet * ifp)2214 if_qflush(struct ifnet *ifp)
2215 {
2216 struct mbuf *m, *n;
2217 struct ifaltq *ifq;
2218
2219 ifq = &ifp->if_snd;
2220 IFQ_LOCK(ifq);
2221 #ifdef ALTQ
2222 if (ALTQ_IS_ENABLED(ifq))
2223 ALTQ_PURGE(ifq);
2224 #endif
2225 n = ifq->ifq_head;
2226 while ((m = n) != 0) {
2227 n = m->m_nextpkt;
2228 m_freem(m);
2229 }
2230 ifq->ifq_head = 0;
2231 ifq->ifq_tail = 0;
2232 ifq->ifq_len = 0;
2233 IFQ_UNLOCK(ifq);
2234 }
2235
2236 /*
2237 * Map interface name to interface structure pointer, with or without
2238 * returning a reference.
2239 */
2240 struct ifnet *
ifunit_ref(const char * name)2241 ifunit_ref(const char *name)
2242 {
2243 struct ifnet *ifp;
2244
2245 IFNET_RLOCK_NOSLEEP();
2246 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2247 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0 &&
2248 !(ifp->if_flags & IFF_DYING))
2249 break;
2250 }
2251 if (ifp != NULL)
2252 if_ref(ifp);
2253 IFNET_RUNLOCK_NOSLEEP();
2254 return (ifp);
2255 }
2256
2257 struct ifnet *
ifunit(const char * name)2258 ifunit(const char *name)
2259 {
2260 struct ifnet *ifp;
2261
2262 IFNET_RLOCK_NOSLEEP();
2263 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2264 if (strncmp(name, ifp->if_xname, IFNAMSIZ) == 0)
2265 break;
2266 }
2267 IFNET_RUNLOCK_NOSLEEP();
2268 return (ifp);
2269 }
2270
2271 /*
2272 * Hardware specific interface ioctls.
2273 */
2274 static int
ifhwioctl(u_long cmd,struct ifnet * ifp,caddr_t data,struct thread * td)2275 ifhwioctl(u_long cmd, struct ifnet *ifp, caddr_t data, struct thread *td)
2276 {
2277 struct ifreq *ifr;
2278 struct ifstat *ifs;
2279 int error = 0, do_ifup = 0;
2280 int new_flags, temp_flags;
2281 size_t namelen, onamelen;
2282 size_t descrlen;
2283 char *descrbuf, *odescrbuf;
2284 char new_name[IFNAMSIZ];
2285 struct ifaddr *ifa;
2286 struct sockaddr_dl *sdl;
2287
2288 ifr = (struct ifreq *)data;
2289 switch (cmd) {
2290 case SIOCGIFINDEX:
2291 ifr->ifr_index = ifp->if_index;
2292 break;
2293
2294 case SIOCGIFFLAGS:
2295 temp_flags = ifp->if_flags | ifp->if_drv_flags;
2296 ifr->ifr_flags = temp_flags & 0xffff;
2297 ifr->ifr_flagshigh = temp_flags >> 16;
2298 break;
2299
2300 case SIOCGIFCAP:
2301 ifr->ifr_reqcap = ifp->if_capabilities;
2302 ifr->ifr_curcap = ifp->if_capenable;
2303 break;
2304
2305 #ifdef MAC
2306 case SIOCGIFMAC:
2307 error = mac_ifnet_ioctl_get(td->td_ucred, ifr, ifp);
2308 break;
2309 #endif
2310
2311 case SIOCGIFMETRIC:
2312 ifr->ifr_metric = ifp->if_metric;
2313 break;
2314
2315 case SIOCGIFMTU:
2316 ifr->ifr_mtu = ifp->if_mtu;
2317 break;
2318
2319 case SIOCGIFPHYS:
2320 ifr->ifr_phys = ifp->if_physical;
2321 break;
2322
2323 case SIOCGIFDESCR:
2324 error = 0;
2325 sx_slock(&ifdescr_sx);
2326 if (ifp->if_description == NULL)
2327 error = ENOMSG;
2328 else {
2329 /* space for terminating nul */
2330 descrlen = strlen(ifp->if_description) + 1;
2331 if (ifr->ifr_buffer.length < descrlen)
2332 ifr->ifr_buffer.buffer = NULL;
2333 else
2334 error = copyout(ifp->if_description,
2335 ifr->ifr_buffer.buffer, descrlen);
2336 ifr->ifr_buffer.length = descrlen;
2337 }
2338 sx_sunlock(&ifdescr_sx);
2339 break;
2340
2341 case SIOCSIFDESCR:
2342 error = priv_check(td, PRIV_NET_SETIFDESCR);
2343 if (error)
2344 return (error);
2345
2346 /*
2347 * Copy only (length-1) bytes to make sure that
2348 * if_description is always nul terminated. The
2349 * length parameter is supposed to count the
2350 * terminating nul in.
2351 */
2352 if (ifr->ifr_buffer.length > ifdescr_maxlen)
2353 return (ENAMETOOLONG);
2354 else if (ifr->ifr_buffer.length == 0)
2355 descrbuf = NULL;
2356 else {
2357 descrbuf = malloc(ifr->ifr_buffer.length, M_IFDESCR,
2358 M_WAITOK | M_ZERO);
2359 error = copyin(ifr->ifr_buffer.buffer, descrbuf,
2360 ifr->ifr_buffer.length - 1);
2361 if (error) {
2362 free(descrbuf, M_IFDESCR);
2363 break;
2364 }
2365 }
2366
2367 sx_xlock(&ifdescr_sx);
2368 odescrbuf = ifp->if_description;
2369 ifp->if_description = descrbuf;
2370 sx_xunlock(&ifdescr_sx);
2371
2372 getmicrotime(&ifp->if_lastchange);
2373 free(odescrbuf, M_IFDESCR);
2374 break;
2375
2376 case SIOCGIFFIB:
2377 ifr->ifr_fib = ifp->if_fib;
2378 break;
2379
2380 case SIOCSIFFIB:
2381 error = priv_check(td, PRIV_NET_SETIFFIB);
2382 if (error)
2383 return (error);
2384 if (ifr->ifr_fib >= rt_numfibs)
2385 return (EINVAL);
2386
2387 ifp->if_fib = ifr->ifr_fib;
2388 break;
2389
2390 case SIOCSIFFLAGS:
2391 error = priv_check(td, PRIV_NET_SETIFFLAGS);
2392 if (error)
2393 return (error);
2394 /*
2395 * Currently, no driver owned flags pass the IFF_CANTCHANGE
2396 * check, so we don't need special handling here yet.
2397 */
2398 new_flags = (ifr->ifr_flags & 0xffff) |
2399 (ifr->ifr_flagshigh << 16);
2400 if (ifp->if_flags & IFF_SMART) {
2401 /* Smart drivers twiddle their own routes */
2402 } else if (ifp->if_flags & IFF_UP &&
2403 (new_flags & IFF_UP) == 0) {
2404 if_down(ifp);
2405 } else if (new_flags & IFF_UP &&
2406 (ifp->if_flags & IFF_UP) == 0) {
2407 do_ifup = 1;
2408 }
2409 /* See if permanently promiscuous mode bit is about to flip */
2410 if ((ifp->if_flags ^ new_flags) & IFF_PPROMISC) {
2411 if (new_flags & IFF_PPROMISC)
2412 ifp->if_flags |= IFF_PROMISC;
2413 else if (ifp->if_pcount == 0)
2414 ifp->if_flags &= ~IFF_PROMISC;
2415 if (log_promisc_mode_change)
2416 log(LOG_INFO, "%s: permanently promiscuous mode %s\n",
2417 ifp->if_xname,
2418 ((new_flags & IFF_PPROMISC) ?
2419 "enabled" : "disabled"));
2420 }
2421 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2422 (new_flags &~ IFF_CANTCHANGE);
2423 if (ifp->if_ioctl) {
2424 (void) (*ifp->if_ioctl)(ifp, cmd, data);
2425 }
2426 if (do_ifup)
2427 if_up(ifp);
2428 getmicrotime(&ifp->if_lastchange);
2429 break;
2430
2431 case SIOCSIFCAP:
2432 error = priv_check(td, PRIV_NET_SETIFCAP);
2433 if (error)
2434 return (error);
2435 if (ifp->if_ioctl == NULL)
2436 return (EOPNOTSUPP);
2437 if (ifr->ifr_reqcap & ~ifp->if_capabilities)
2438 return (EINVAL);
2439 error = (*ifp->if_ioctl)(ifp, cmd, data);
2440 if (error == 0)
2441 getmicrotime(&ifp->if_lastchange);
2442 break;
2443
2444 #ifdef MAC
2445 case SIOCSIFMAC:
2446 error = mac_ifnet_ioctl_set(td->td_ucred, ifr, ifp);
2447 break;
2448 #endif
2449
2450 case SIOCSIFNAME:
2451 error = priv_check(td, PRIV_NET_SETIFNAME);
2452 if (error)
2453 return (error);
2454 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2455 if (error != 0)
2456 return (error);
2457 if (new_name[0] == '\0')
2458 return (EINVAL);
2459 if (new_name[IFNAMSIZ-1] != '\0') {
2460 new_name[IFNAMSIZ-1] = '\0';
2461 if (strlen(new_name) == IFNAMSIZ-1)
2462 return (EINVAL);
2463 }
2464 if (ifunit(new_name) != NULL)
2465 return (EEXIST);
2466
2467 /*
2468 * XXX: Locking. Nothing else seems to lock if_flags,
2469 * and there are numerous other races with the
2470 * ifunit() checks not being atomic with namespace
2471 * changes (renames, vmoves, if_attach, etc).
2472 */
2473 ifp->if_flags |= IFF_RENAMING;
2474
2475 /* Announce the departure of the interface. */
2476 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2477 EVENTHANDLER_INVOKE(ifnet_departure_event, ifp);
2478
2479 log(LOG_INFO, "%s: changing name to '%s'\n",
2480 ifp->if_xname, new_name);
2481
2482 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2483 ifa = ifp->if_addr;
2484 IFA_LOCK(ifa);
2485 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2486 namelen = strlen(new_name);
2487 onamelen = sdl->sdl_nlen;
2488 /*
2489 * Move the address if needed. This is safe because we
2490 * allocate space for a name of length IFNAMSIZ when we
2491 * create this in if_attach().
2492 */
2493 if (namelen != onamelen) {
2494 bcopy(sdl->sdl_data + onamelen,
2495 sdl->sdl_data + namelen, sdl->sdl_alen);
2496 }
2497 bcopy(new_name, sdl->sdl_data, namelen);
2498 sdl->sdl_nlen = namelen;
2499 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2500 bzero(sdl->sdl_data, onamelen);
2501 while (namelen != 0)
2502 sdl->sdl_data[--namelen] = 0xff;
2503 IFA_UNLOCK(ifa);
2504
2505 EVENTHANDLER_INVOKE(ifnet_arrival_event, ifp);
2506 /* Announce the return of the interface. */
2507 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2508
2509 ifp->if_flags &= ~IFF_RENAMING;
2510 break;
2511
2512 #ifdef VIMAGE
2513 case SIOCSIFVNET:
2514 error = priv_check(td, PRIV_NET_SETIFVNET);
2515 if (error)
2516 return (error);
2517 error = if_vmove_loan(td, ifp, ifr->ifr_name, ifr->ifr_jid);
2518 break;
2519 #endif
2520
2521 case SIOCSIFMETRIC:
2522 error = priv_check(td, PRIV_NET_SETIFMETRIC);
2523 if (error)
2524 return (error);
2525 ifp->if_metric = ifr->ifr_metric;
2526 getmicrotime(&ifp->if_lastchange);
2527 break;
2528
2529 case SIOCSIFPHYS:
2530 error = priv_check(td, PRIV_NET_SETIFPHYS);
2531 if (error)
2532 return (error);
2533 if (ifp->if_ioctl == NULL)
2534 return (EOPNOTSUPP);
2535 error = (*ifp->if_ioctl)(ifp, cmd, data);
2536 if (error == 0)
2537 getmicrotime(&ifp->if_lastchange);
2538 break;
2539
2540 case SIOCSIFMTU:
2541 {
2542 u_long oldmtu = ifp->if_mtu;
2543
2544 error = priv_check(td, PRIV_NET_SETIFMTU);
2545 if (error)
2546 return (error);
2547 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU)
2548 return (EINVAL);
2549 if (ifp->if_ioctl == NULL)
2550 return (EOPNOTSUPP);
2551 error = (*ifp->if_ioctl)(ifp, cmd, data);
2552 if (error == 0) {
2553 getmicrotime(&ifp->if_lastchange);
2554 rt_ifmsg(ifp);
2555 }
2556 /*
2557 * If the link MTU changed, do network layer specific procedure.
2558 */
2559 if (ifp->if_mtu != oldmtu) {
2560 #ifdef INET6
2561 nd6_setmtu(ifp);
2562 #endif
2563 }
2564 break;
2565 }
2566
2567 case SIOCADDMULTI:
2568 case SIOCDELMULTI:
2569 if (cmd == SIOCADDMULTI)
2570 error = priv_check(td, PRIV_NET_ADDMULTI);
2571 else
2572 error = priv_check(td, PRIV_NET_DELMULTI);
2573 if (error)
2574 return (error);
2575
2576 /* Don't allow group membership on non-multicast interfaces. */
2577 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2578 return (EOPNOTSUPP);
2579
2580 /* Don't let users screw up protocols' entries. */
2581 if (ifr->ifr_addr.sa_family != AF_LINK)
2582 return (EINVAL);
2583
2584 if (cmd == SIOCADDMULTI) {
2585 struct ifmultiaddr *ifma;
2586
2587 /*
2588 * Userland is only permitted to join groups once
2589 * via the if_addmulti() KPI, because it cannot hold
2590 * struct ifmultiaddr * between calls. It may also
2591 * lose a race while we check if the membership
2592 * already exists.
2593 */
2594 IF_ADDR_RLOCK(ifp);
2595 ifma = if_findmulti(ifp, &ifr->ifr_addr);
2596 IF_ADDR_RUNLOCK(ifp);
2597 if (ifma != NULL)
2598 error = EADDRINUSE;
2599 else
2600 error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2601 } else {
2602 error = if_delmulti(ifp, &ifr->ifr_addr);
2603 }
2604 if (error == 0)
2605 getmicrotime(&ifp->if_lastchange);
2606 break;
2607
2608 case SIOCSIFPHYADDR:
2609 case SIOCDIFPHYADDR:
2610 #ifdef INET6
2611 case SIOCSIFPHYADDR_IN6:
2612 #endif
2613 case SIOCSLIFPHYADDR:
2614 case SIOCSIFMEDIA:
2615 case SIOCSIFGENERIC:
2616 error = priv_check(td, PRIV_NET_HWIOCTL);
2617 if (error)
2618 return (error);
2619 if (ifp->if_ioctl == NULL)
2620 return (EOPNOTSUPP);
2621 error = (*ifp->if_ioctl)(ifp, cmd, data);
2622 if (error == 0)
2623 getmicrotime(&ifp->if_lastchange);
2624 break;
2625
2626 case SIOCGIFSTATUS:
2627 ifs = (struct ifstat *)data;
2628 ifs->ascii[0] = '\0';
2629
2630 case SIOCGIFPSRCADDR:
2631 case SIOCGIFPDSTADDR:
2632 case SIOCGLIFPHYADDR:
2633 case SIOCGIFMEDIA:
2634 case SIOCGIFXMEDIA:
2635 case SIOCGIFGENERIC:
2636 case SIOCGIFRSSKEY:
2637 case SIOCGIFRSSHASH:
2638 if (ifp->if_ioctl == NULL)
2639 return (EOPNOTSUPP);
2640 error = (*ifp->if_ioctl)(ifp, cmd, data);
2641 break;
2642
2643 case SIOCSIFLLADDR:
2644 error = priv_check(td, PRIV_NET_SETLLADDR);
2645 if (error)
2646 return (error);
2647 error = if_setlladdr(ifp,
2648 ifr->ifr_addr.sa_data, ifr->ifr_addr.sa_len);
2649 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2650 break;
2651
2652 case SIOCGHWADDR:
2653 error = if_gethwaddr(ifp, ifr);
2654 break;
2655
2656 case SIOCAIFGROUP:
2657 {
2658 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2659
2660 error = priv_check(td, PRIV_NET_ADDIFGROUP);
2661 if (error)
2662 return (error);
2663 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2664 return (error);
2665 break;
2666 }
2667
2668 case SIOCGIFGROUP:
2669 if ((error = if_getgroup((struct ifgroupreq *)ifr, ifp)))
2670 return (error);
2671 break;
2672
2673 case SIOCDIFGROUP:
2674 {
2675 struct ifgroupreq *ifgr = (struct ifgroupreq *)ifr;
2676
2677 error = priv_check(td, PRIV_NET_DELIFGROUP);
2678 if (error)
2679 return (error);
2680 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2681 return (error);
2682 break;
2683 }
2684
2685 default:
2686 error = ENOIOCTL;
2687 break;
2688 }
2689 return (error);
2690 }
2691
2692 #ifdef COMPAT_FREEBSD32
2693 struct ifconf32 {
2694 int32_t ifc_len;
2695 union {
2696 uint32_t ifcu_buf;
2697 uint32_t ifcu_req;
2698 } ifc_ifcu;
2699 };
2700 #define SIOCGIFCONF32 _IOWR('i', 36, struct ifconf32)
2701 #endif
2702
2703 /*
2704 * Interface ioctls.
2705 */
2706 int
ifioctl(struct socket * so,u_long cmd,caddr_t data,struct thread * td)2707 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct thread *td)
2708 {
2709 struct ifnet *ifp;
2710 struct ifreq *ifr;
2711 int error;
2712 int oif_flags;
2713
2714 CURVNET_SET(so->so_vnet);
2715 switch (cmd) {
2716 case SIOCGIFCONF:
2717 case OSIOCGIFCONF:
2718 error = ifconf(cmd, data);
2719 CURVNET_RESTORE();
2720 return (error);
2721
2722 #ifdef COMPAT_FREEBSD32
2723 case SIOCGIFCONF32:
2724 {
2725 struct ifconf32 *ifc32;
2726 struct ifconf ifc;
2727
2728 ifc32 = (struct ifconf32 *)data;
2729 ifc.ifc_len = ifc32->ifc_len;
2730 ifc.ifc_buf = PTRIN(ifc32->ifc_buf);
2731
2732 error = ifconf(SIOCGIFCONF, (void *)&ifc);
2733 CURVNET_RESTORE();
2734 if (error == 0)
2735 ifc32->ifc_len = ifc.ifc_len;
2736 return (error);
2737 }
2738 #endif
2739 }
2740 ifr = (struct ifreq *)data;
2741
2742 switch (cmd) {
2743 #ifdef VIMAGE
2744 case SIOCSIFRVNET:
2745 error = priv_check(td, PRIV_NET_SETIFVNET);
2746 if (error == 0)
2747 error = if_vmove_reclaim(td, ifr->ifr_name,
2748 ifr->ifr_jid);
2749 CURVNET_RESTORE();
2750 return (error);
2751 #endif
2752 case SIOCIFCREATE:
2753 case SIOCIFCREATE2:
2754 error = priv_check(td, PRIV_NET_IFCREATE);
2755 if (error == 0)
2756 error = if_clone_create(ifr->ifr_name,
2757 sizeof(ifr->ifr_name),
2758 cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL);
2759 CURVNET_RESTORE();
2760 return (error);
2761 case SIOCIFDESTROY:
2762 error = priv_check(td, PRIV_NET_IFDESTROY);
2763 if (error == 0)
2764 error = if_clone_destroy(ifr->ifr_name);
2765 CURVNET_RESTORE();
2766 return (error);
2767
2768 case SIOCIFGCLONERS:
2769 error = if_clone_list((struct if_clonereq *)data);
2770 CURVNET_RESTORE();
2771 return (error);
2772 case SIOCGIFGMEMB:
2773 error = if_getgroupmembers((struct ifgroupreq *)data);
2774 CURVNET_RESTORE();
2775 return (error);
2776 #if defined(INET) || defined(INET6)
2777 case SIOCSVH:
2778 case SIOCGVH:
2779 if (carp_ioctl_p == NULL)
2780 error = EPROTONOSUPPORT;
2781 else
2782 error = (*carp_ioctl_p)(ifr, cmd, td);
2783 CURVNET_RESTORE();
2784 return (error);
2785 #endif
2786 }
2787
2788 ifp = ifunit_ref(ifr->ifr_name);
2789 if (ifp == NULL) {
2790 CURVNET_RESTORE();
2791 return (ENXIO);
2792 }
2793
2794 error = ifhwioctl(cmd, ifp, data, td);
2795 if (error != ENOIOCTL) {
2796 if_rele(ifp);
2797 CURVNET_RESTORE();
2798 return (error);
2799 }
2800
2801 oif_flags = ifp->if_flags;
2802 if (so->so_proto == NULL) {
2803 if_rele(ifp);
2804 CURVNET_RESTORE();
2805 return (EOPNOTSUPP);
2806 }
2807
2808 /*
2809 * Pass the request on to the socket control method, and if the
2810 * latter returns EOPNOTSUPP, directly to the interface.
2811 *
2812 * Make an exception for the legacy SIOCSIF* requests. Drivers
2813 * trust SIOCSIFADDR et al to come from an already privileged
2814 * layer, and do not perform any credentials checks or input
2815 * validation.
2816 */
2817 #ifndef COMPAT_43
2818 error = ((*so->so_proto->pr_usrreqs->pru_control)(so, cmd,
2819 data,
2820 ifp, td));
2821 if (error == EOPNOTSUPP && ifp != NULL && ifp->if_ioctl != NULL &&
2822 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2823 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2824 error = (*ifp->if_ioctl)(ifp, cmd, data);
2825 #else
2826 {
2827 u_long ocmd = cmd;
2828
2829 switch (cmd) {
2830
2831 case SIOCSIFDSTADDR:
2832 case SIOCSIFADDR:
2833 case SIOCSIFBRDADDR:
2834 case SIOCSIFNETMASK:
2835 #if BYTE_ORDER != BIG_ENDIAN
2836 if (ifr->ifr_addr.sa_family == 0 &&
2837 ifr->ifr_addr.sa_len < 16) {
2838 ifr->ifr_addr.sa_family = ifr->ifr_addr.sa_len;
2839 ifr->ifr_addr.sa_len = 16;
2840 }
2841 #else
2842 if (ifr->ifr_addr.sa_len == 0)
2843 ifr->ifr_addr.sa_len = 16;
2844 #endif
2845 break;
2846
2847 case OSIOCGIFADDR:
2848 cmd = SIOCGIFADDR;
2849 break;
2850
2851 case OSIOCGIFDSTADDR:
2852 cmd = SIOCGIFDSTADDR;
2853 break;
2854
2855 case OSIOCGIFBRDADDR:
2856 cmd = SIOCGIFBRDADDR;
2857 break;
2858
2859 case OSIOCGIFNETMASK:
2860 cmd = SIOCGIFNETMASK;
2861 }
2862 error = ((*so->so_proto->pr_usrreqs->pru_control)(so,
2863 cmd,
2864 data,
2865 ifp, td));
2866 if (error == EOPNOTSUPP && ifp != NULL &&
2867 ifp->if_ioctl != NULL &&
2868 cmd != SIOCSIFADDR && cmd != SIOCSIFBRDADDR &&
2869 cmd != SIOCSIFDSTADDR && cmd != SIOCSIFNETMASK)
2870 error = (*ifp->if_ioctl)(ifp, cmd, data);
2871 switch (ocmd) {
2872
2873 case OSIOCGIFADDR:
2874 case OSIOCGIFDSTADDR:
2875 case OSIOCGIFBRDADDR:
2876 case OSIOCGIFNETMASK:
2877 *(u_short *)&ifr->ifr_addr = ifr->ifr_addr.sa_family;
2878
2879 }
2880 }
2881 #endif /* COMPAT_43 */
2882
2883 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2884 #ifdef INET6
2885 if (ifp->if_flags & IFF_UP)
2886 in6_if_up(ifp);
2887 #endif
2888 }
2889 if_rele(ifp);
2890 CURVNET_RESTORE();
2891 return (error);
2892 }
2893
2894 /*
2895 * The code common to handling reference counted flags,
2896 * e.g., in ifpromisc() and if_allmulti().
2897 * The "pflag" argument can specify a permanent mode flag to check,
2898 * such as IFF_PPROMISC for promiscuous mode; should be 0 if none.
2899 *
2900 * Only to be used on stack-owned flags, not driver-owned flags.
2901 */
2902 static int
if_setflag(struct ifnet * ifp,int flag,int pflag,int * refcount,int onswitch)2903 if_setflag(struct ifnet *ifp, int flag, int pflag, int *refcount, int onswitch)
2904 {
2905 struct ifreq ifr;
2906 int error;
2907 int oldflags, oldcount;
2908
2909 /* Sanity checks to catch programming errors */
2910 KASSERT((flag & (IFF_DRV_OACTIVE|IFF_DRV_RUNNING)) == 0,
2911 ("%s: setting driver-owned flag %d", __func__, flag));
2912
2913 if (onswitch)
2914 KASSERT(*refcount >= 0,
2915 ("%s: increment negative refcount %d for flag %d",
2916 __func__, *refcount, flag));
2917 else
2918 KASSERT(*refcount > 0,
2919 ("%s: decrement non-positive refcount %d for flag %d",
2920 __func__, *refcount, flag));
2921
2922 /* In case this mode is permanent, just touch refcount */
2923 if (ifp->if_flags & pflag) {
2924 *refcount += onswitch ? 1 : -1;
2925 return (0);
2926 }
2927
2928 /* Save ifnet parameters for if_ioctl() may fail */
2929 oldcount = *refcount;
2930 oldflags = ifp->if_flags;
2931
2932 /*
2933 * See if we aren't the only and touching refcount is enough.
2934 * Actually toggle interface flag if we are the first or last.
2935 */
2936 if (onswitch) {
2937 if ((*refcount)++)
2938 return (0);
2939 ifp->if_flags |= flag;
2940 } else {
2941 if (--(*refcount))
2942 return (0);
2943 ifp->if_flags &= ~flag;
2944 }
2945
2946 /* Call down the driver since we've changed interface flags */
2947 if (ifp->if_ioctl == NULL) {
2948 error = EOPNOTSUPP;
2949 goto recover;
2950 }
2951 ifr.ifr_flags = ifp->if_flags & 0xffff;
2952 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2953 error = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
2954 if (error)
2955 goto recover;
2956 /* Notify userland that interface flags have changed */
2957 rt_ifmsg(ifp);
2958 return (0);
2959
2960 recover:
2961 /* Recover after driver error */
2962 *refcount = oldcount;
2963 ifp->if_flags = oldflags;
2964 return (error);
2965 }
2966
2967 /*
2968 * Set/clear promiscuous mode on interface ifp based on the truth value
2969 * of pswitch. The calls are reference counted so that only the first
2970 * "on" request actually has an effect, as does the final "off" request.
2971 * Results are undefined if the "off" and "on" requests are not matched.
2972 */
2973 int
ifpromisc(struct ifnet * ifp,int pswitch)2974 ifpromisc(struct ifnet *ifp, int pswitch)
2975 {
2976 int error;
2977 int oldflags = ifp->if_flags;
2978
2979 error = if_setflag(ifp, IFF_PROMISC, IFF_PPROMISC,
2980 &ifp->if_pcount, pswitch);
2981 /* If promiscuous mode status has changed, log a message */
2982 if (error == 0 && ((ifp->if_flags ^ oldflags) & IFF_PROMISC) &&
2983 log_promisc_mode_change)
2984 log(LOG_INFO, "%s: promiscuous mode %s\n",
2985 ifp->if_xname,
2986 (ifp->if_flags & IFF_PROMISC) ? "enabled" : "disabled");
2987 return (error);
2988 }
2989
2990 /*
2991 * Return interface configuration
2992 * of system. List may be used
2993 * in later ioctl's (above) to get
2994 * other information.
2995 */
2996 /*ARGSUSED*/
2997 static int
ifconf(u_long cmd,caddr_t data)2998 ifconf(u_long cmd, caddr_t data)
2999 {
3000 struct ifconf *ifc = (struct ifconf *)data;
3001 struct ifnet *ifp;
3002 struct ifaddr *ifa;
3003 struct ifreq ifr;
3004 struct sbuf *sb;
3005 int error, full = 0, valid_len, max_len;
3006
3007 /* Limit initial buffer size to MAXPHYS to avoid DoS from userspace. */
3008 max_len = MAXPHYS - 1;
3009
3010 /* Prevent hostile input from being able to crash the system */
3011 if (ifc->ifc_len <= 0)
3012 return (EINVAL);
3013
3014 again:
3015 if (ifc->ifc_len <= max_len) {
3016 max_len = ifc->ifc_len;
3017 full = 1;
3018 }
3019 sb = sbuf_new(NULL, NULL, max_len + 1, SBUF_FIXEDLEN);
3020 max_len = 0;
3021 valid_len = 0;
3022
3023 IFNET_RLOCK();
3024 TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
3025 int addrs;
3026
3027 /*
3028 * Zero the ifr to make sure we don't disclose the contents
3029 * of the stack.
3030 */
3031 memset(&ifr, 0, sizeof(ifr));
3032
3033 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
3034 >= sizeof(ifr.ifr_name)) {
3035 sbuf_delete(sb);
3036 IFNET_RUNLOCK();
3037 return (ENAMETOOLONG);
3038 }
3039
3040 addrs = 0;
3041 IF_ADDR_RLOCK(ifp);
3042 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3043 struct sockaddr *sa = ifa->ifa_addr;
3044
3045 if (prison_if(curthread->td_ucred, sa) != 0)
3046 continue;
3047 addrs++;
3048 #ifdef COMPAT_43
3049 if (cmd == OSIOCGIFCONF) {
3050 struct osockaddr *osa =
3051 (struct osockaddr *)&ifr.ifr_addr;
3052 ifr.ifr_addr = *sa;
3053 osa->sa_family = sa->sa_family;
3054 sbuf_bcat(sb, &ifr, sizeof(ifr));
3055 max_len += sizeof(ifr);
3056 } else
3057 #endif
3058 if (sa->sa_len <= sizeof(*sa)) {
3059 if (sa->sa_len < sizeof(*sa)) {
3060 memset(&ifr.ifr_ifru.ifru_addr, 0,
3061 sizeof(ifr.ifr_ifru.ifru_addr));
3062 memcpy(&ifr.ifr_ifru.ifru_addr, sa,
3063 sa->sa_len);
3064 } else
3065 ifr.ifr_ifru.ifru_addr = *sa;
3066 sbuf_bcat(sb, &ifr, sizeof(ifr));
3067 max_len += sizeof(ifr);
3068 } else {
3069 sbuf_bcat(sb, &ifr,
3070 offsetof(struct ifreq, ifr_addr));
3071 max_len += offsetof(struct ifreq, ifr_addr);
3072 sbuf_bcat(sb, sa, sa->sa_len);
3073 max_len += sa->sa_len;
3074 }
3075
3076 if (sbuf_error(sb) == 0)
3077 valid_len = sbuf_len(sb);
3078 }
3079 IF_ADDR_RUNLOCK(ifp);
3080 if (addrs == 0) {
3081 sbuf_bcat(sb, &ifr, sizeof(ifr));
3082 max_len += sizeof(ifr);
3083
3084 if (sbuf_error(sb) == 0)
3085 valid_len = sbuf_len(sb);
3086 }
3087 }
3088 IFNET_RUNLOCK();
3089
3090 /*
3091 * If we didn't allocate enough space (uncommon), try again. If
3092 * we have already allocated as much space as we are allowed,
3093 * return what we've got.
3094 */
3095 if (valid_len != max_len && !full) {
3096 sbuf_delete(sb);
3097 goto again;
3098 }
3099
3100 ifc->ifc_len = valid_len;
3101 sbuf_finish(sb);
3102 error = copyout(sbuf_data(sb), ifc->ifc_req, ifc->ifc_len);
3103 sbuf_delete(sb);
3104 return (error);
3105 }
3106
3107 /*
3108 * Just like ifpromisc(), but for all-multicast-reception mode.
3109 */
3110 int
if_allmulti(struct ifnet * ifp,int onswitch)3111 if_allmulti(struct ifnet *ifp, int onswitch)
3112 {
3113
3114 return (if_setflag(ifp, IFF_ALLMULTI, 0, &ifp->if_amcount, onswitch));
3115 }
3116
3117 struct ifmultiaddr *
if_findmulti(struct ifnet * ifp,struct sockaddr * sa)3118 if_findmulti(struct ifnet *ifp, struct sockaddr *sa)
3119 {
3120 struct ifmultiaddr *ifma;
3121
3122 IF_ADDR_LOCK_ASSERT(ifp);
3123
3124 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3125 if (sa->sa_family == AF_LINK) {
3126 if (sa_dl_equal(ifma->ifma_addr, sa))
3127 break;
3128 } else {
3129 if (sa_equal(ifma->ifma_addr, sa))
3130 break;
3131 }
3132 }
3133
3134 return ifma;
3135 }
3136
3137 /*
3138 * Allocate a new ifmultiaddr and initialize based on passed arguments. We
3139 * make copies of passed sockaddrs. The ifmultiaddr will not be added to
3140 * the ifnet multicast address list here, so the caller must do that and
3141 * other setup work (such as notifying the device driver). The reference
3142 * count is initialized to 1.
3143 */
3144 static struct ifmultiaddr *
if_allocmulti(struct ifnet * ifp,struct sockaddr * sa,struct sockaddr * llsa,int mflags)3145 if_allocmulti(struct ifnet *ifp, struct sockaddr *sa, struct sockaddr *llsa,
3146 int mflags)
3147 {
3148 struct ifmultiaddr *ifma;
3149 struct sockaddr *dupsa;
3150
3151 ifma = malloc(sizeof *ifma, M_IFMADDR, mflags |
3152 M_ZERO);
3153 if (ifma == NULL)
3154 return (NULL);
3155
3156 dupsa = malloc(sa->sa_len, M_IFMADDR, mflags);
3157 if (dupsa == NULL) {
3158 free(ifma, M_IFMADDR);
3159 return (NULL);
3160 }
3161 bcopy(sa, dupsa, sa->sa_len);
3162 ifma->ifma_addr = dupsa;
3163
3164 ifma->ifma_ifp = ifp;
3165 ifma->ifma_refcount = 1;
3166 ifma->ifma_protospec = NULL;
3167
3168 if (llsa == NULL) {
3169 ifma->ifma_lladdr = NULL;
3170 return (ifma);
3171 }
3172
3173 dupsa = malloc(llsa->sa_len, M_IFMADDR, mflags);
3174 if (dupsa == NULL) {
3175 free(ifma->ifma_addr, M_IFMADDR);
3176 free(ifma, M_IFMADDR);
3177 return (NULL);
3178 }
3179 bcopy(llsa, dupsa, llsa->sa_len);
3180 ifma->ifma_lladdr = dupsa;
3181
3182 return (ifma);
3183 }
3184
3185 /*
3186 * if_freemulti: free ifmultiaddr structure and possibly attached related
3187 * addresses. The caller is responsible for implementing reference
3188 * counting, notifying the driver, handling routing messages, and releasing
3189 * any dependent link layer state.
3190 */
3191 static void
if_freemulti(struct ifmultiaddr * ifma)3192 if_freemulti(struct ifmultiaddr *ifma)
3193 {
3194
3195 KASSERT(ifma->ifma_refcount == 0, ("if_freemulti: refcount %d",
3196 ifma->ifma_refcount));
3197 KASSERT(ifma->ifma_protospec == NULL,
3198 ("if_freemulti: protospec not NULL"));
3199
3200 if (ifma->ifma_lladdr != NULL)
3201 free(ifma->ifma_lladdr, M_IFMADDR);
3202 free(ifma->ifma_addr, M_IFMADDR);
3203 free(ifma, M_IFMADDR);
3204 }
3205
3206 /*
3207 * Register an additional multicast address with a network interface.
3208 *
3209 * - If the address is already present, bump the reference count on the
3210 * address and return.
3211 * - If the address is not link-layer, look up a link layer address.
3212 * - Allocate address structures for one or both addresses, and attach to the
3213 * multicast address list on the interface. If automatically adding a link
3214 * layer address, the protocol address will own a reference to the link
3215 * layer address, to be freed when it is freed.
3216 * - Notify the network device driver of an addition to the multicast address
3217 * list.
3218 *
3219 * 'sa' points to caller-owned memory with the desired multicast address.
3220 *
3221 * 'retifma' will be used to return a pointer to the resulting multicast
3222 * address reference, if desired.
3223 */
3224 int
if_addmulti(struct ifnet * ifp,struct sockaddr * sa,struct ifmultiaddr ** retifma)3225 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
3226 struct ifmultiaddr **retifma)
3227 {
3228 struct ifmultiaddr *ifma, *ll_ifma;
3229 struct sockaddr *llsa;
3230 int error;
3231
3232 /*
3233 * If the address is already present, return a new reference to it;
3234 * otherwise, allocate storage and set up a new address.
3235 */
3236 IF_ADDR_WLOCK(ifp);
3237 ifma = if_findmulti(ifp, sa);
3238 if (ifma != NULL) {
3239 ifma->ifma_refcount++;
3240 if (retifma != NULL)
3241 *retifma = ifma;
3242 IF_ADDR_WUNLOCK(ifp);
3243 return (0);
3244 }
3245
3246 /*
3247 * The address isn't already present; resolve the protocol address
3248 * into a link layer address, and then look that up, bump its
3249 * refcount or allocate an ifma for that also. If 'llsa' was
3250 * returned, we will need to free it later.
3251 */
3252 llsa = NULL;
3253 ll_ifma = NULL;
3254 if (ifp->if_resolvemulti != NULL) {
3255 error = ifp->if_resolvemulti(ifp, &llsa, sa);
3256 if (error)
3257 goto unlock_out;
3258 }
3259
3260 /*
3261 * Allocate the new address. Don't hook it up yet, as we may also
3262 * need to allocate a link layer multicast address.
3263 */
3264 ifma = if_allocmulti(ifp, sa, llsa, M_NOWAIT);
3265 if (ifma == NULL) {
3266 error = ENOMEM;
3267 goto free_llsa_out;
3268 }
3269
3270 /*
3271 * If a link layer address is found, we'll need to see if it's
3272 * already present in the address list, or allocate is as well.
3273 * When this block finishes, the link layer address will be on the
3274 * list.
3275 */
3276 if (llsa != NULL) {
3277 ll_ifma = if_findmulti(ifp, llsa);
3278 if (ll_ifma == NULL) {
3279 ll_ifma = if_allocmulti(ifp, llsa, NULL, M_NOWAIT);
3280 if (ll_ifma == NULL) {
3281 --ifma->ifma_refcount;
3282 if_freemulti(ifma);
3283 error = ENOMEM;
3284 goto free_llsa_out;
3285 }
3286 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ll_ifma,
3287 ifma_link);
3288 } else
3289 ll_ifma->ifma_refcount++;
3290 ifma->ifma_llifma = ll_ifma;
3291 }
3292
3293 /*
3294 * We now have a new multicast address, ifma, and possibly a new or
3295 * referenced link layer address. Add the primary address to the
3296 * ifnet address list.
3297 */
3298 TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
3299
3300 if (retifma != NULL)
3301 *retifma = ifma;
3302
3303 /*
3304 * Must generate the message while holding the lock so that 'ifma'
3305 * pointer is still valid.
3306 */
3307 rt_newmaddrmsg(RTM_NEWMADDR, ifma);
3308 IF_ADDR_WUNLOCK(ifp);
3309
3310 /*
3311 * We are certain we have added something, so call down to the
3312 * interface to let them know about it.
3313 */
3314 if (ifp->if_ioctl != NULL) {
3315 (void) (*ifp->if_ioctl)(ifp, SIOCADDMULTI, 0);
3316 }
3317
3318 if (llsa != NULL)
3319 free(llsa, M_IFMADDR);
3320
3321 return (0);
3322
3323 free_llsa_out:
3324 if (llsa != NULL)
3325 free(llsa, M_IFMADDR);
3326
3327 unlock_out:
3328 IF_ADDR_WUNLOCK(ifp);
3329 return (error);
3330 }
3331
3332 /*
3333 * Delete a multicast group membership by network-layer group address.
3334 *
3335 * Returns ENOENT if the entry could not be found. If ifp no longer
3336 * exists, results are undefined. This entry point should only be used
3337 * from subsystems which do appropriate locking to hold ifp for the
3338 * duration of the call.
3339 * Network-layer protocol domains must use if_delmulti_ifma().
3340 */
3341 int
if_delmulti(struct ifnet * ifp,struct sockaddr * sa)3342 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
3343 {
3344 struct ifmultiaddr *ifma;
3345 int lastref;
3346 #ifdef INVARIANTS
3347 struct ifnet *oifp;
3348
3349 IFNET_RLOCK_NOSLEEP();
3350 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3351 if (ifp == oifp)
3352 break;
3353 if (ifp != oifp)
3354 ifp = NULL;
3355 IFNET_RUNLOCK_NOSLEEP();
3356
3357 KASSERT(ifp != NULL, ("%s: ifnet went away", __func__));
3358 #endif
3359 if (ifp == NULL)
3360 return (ENOENT);
3361
3362 IF_ADDR_WLOCK(ifp);
3363 lastref = 0;
3364 ifma = if_findmulti(ifp, sa);
3365 if (ifma != NULL)
3366 lastref = if_delmulti_locked(ifp, ifma, 0);
3367 IF_ADDR_WUNLOCK(ifp);
3368
3369 if (ifma == NULL)
3370 return (ENOENT);
3371
3372 if (lastref && ifp->if_ioctl != NULL) {
3373 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3374 }
3375
3376 return (0);
3377 }
3378
3379 /*
3380 * Delete all multicast group membership for an interface.
3381 * Should be used to quickly flush all multicast filters.
3382 */
3383 void
if_delallmulti(struct ifnet * ifp)3384 if_delallmulti(struct ifnet *ifp)
3385 {
3386 struct ifmultiaddr *ifma;
3387 struct ifmultiaddr *next;
3388
3389 IF_ADDR_WLOCK(ifp);
3390 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next)
3391 if_delmulti_locked(ifp, ifma, 0);
3392 IF_ADDR_WUNLOCK(ifp);
3393 }
3394
3395 /*
3396 * Delete a multicast group membership by group membership pointer.
3397 * Network-layer protocol domains must use this routine.
3398 *
3399 * It is safe to call this routine if the ifp disappeared.
3400 */
3401 void
if_delmulti_ifma(struct ifmultiaddr * ifma)3402 if_delmulti_ifma(struct ifmultiaddr *ifma)
3403 {
3404 struct ifnet *ifp;
3405 int lastref;
3406
3407 ifp = ifma->ifma_ifp;
3408 #ifdef DIAGNOSTIC
3409 if (ifp == NULL) {
3410 printf("%s: ifma_ifp seems to be detached\n", __func__);
3411 } else {
3412 struct ifnet *oifp;
3413
3414 IFNET_RLOCK_NOSLEEP();
3415 TAILQ_FOREACH(oifp, &V_ifnet, if_link)
3416 if (ifp == oifp)
3417 break;
3418 if (ifp != oifp) {
3419 printf("%s: ifnet %p disappeared\n", __func__, ifp);
3420 ifp = NULL;
3421 }
3422 IFNET_RUNLOCK_NOSLEEP();
3423 }
3424 #endif
3425 /*
3426 * If and only if the ifnet instance exists: Acquire the address lock.
3427 */
3428 if (ifp != NULL)
3429 IF_ADDR_WLOCK(ifp);
3430
3431 lastref = if_delmulti_locked(ifp, ifma, 0);
3432
3433 if (ifp != NULL) {
3434 /*
3435 * If and only if the ifnet instance exists:
3436 * Release the address lock.
3437 * If the group was left: update the hardware hash filter.
3438 */
3439 IF_ADDR_WUNLOCK(ifp);
3440 if (lastref && ifp->if_ioctl != NULL) {
3441 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, 0);
3442 }
3443 }
3444 }
3445
3446 /*
3447 * Perform deletion of network-layer and/or link-layer multicast address.
3448 *
3449 * Return 0 if the reference count was decremented.
3450 * Return 1 if the final reference was released, indicating that the
3451 * hardware hash filter should be reprogrammed.
3452 */
3453 static int
if_delmulti_locked(struct ifnet * ifp,struct ifmultiaddr * ifma,int detaching)3454 if_delmulti_locked(struct ifnet *ifp, struct ifmultiaddr *ifma, int detaching)
3455 {
3456 struct ifmultiaddr *ll_ifma;
3457
3458 if (ifp != NULL && ifma->ifma_ifp != NULL) {
3459 KASSERT(ifma->ifma_ifp == ifp,
3460 ("%s: inconsistent ifp %p", __func__, ifp));
3461 IF_ADDR_WLOCK_ASSERT(ifp);
3462 }
3463
3464 ifp = ifma->ifma_ifp;
3465
3466 /*
3467 * If the ifnet is detaching, null out references to ifnet,
3468 * so that upper protocol layers will notice, and not attempt
3469 * to obtain locks for an ifnet which no longer exists. The
3470 * routing socket announcement must happen before the ifnet
3471 * instance is detached from the system.
3472 */
3473 if (detaching) {
3474 #ifdef DIAGNOSTIC
3475 printf("%s: detaching ifnet instance %p\n", __func__, ifp);
3476 #endif
3477 /*
3478 * ifp may already be nulled out if we are being reentered
3479 * to delete the ll_ifma.
3480 */
3481 if (ifp != NULL) {
3482 rt_newmaddrmsg(RTM_DELMADDR, ifma);
3483 ifma->ifma_ifp = NULL;
3484 }
3485 }
3486
3487 if (--ifma->ifma_refcount > 0)
3488 return 0;
3489
3490 /*
3491 * If this ifma is a network-layer ifma, a link-layer ifma may
3492 * have been associated with it. Release it first if so.
3493 */
3494 ll_ifma = ifma->ifma_llifma;
3495 if (ll_ifma != NULL) {
3496 KASSERT(ifma->ifma_lladdr != NULL,
3497 ("%s: llifma w/o lladdr", __func__));
3498 if (detaching)
3499 ll_ifma->ifma_ifp = NULL; /* XXX */
3500 if (--ll_ifma->ifma_refcount == 0) {
3501 if (ifp != NULL) {
3502 TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma,
3503 ifma_link);
3504 }
3505 if_freemulti(ll_ifma);
3506 }
3507 }
3508
3509 if (ifp != NULL)
3510 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
3511
3512 if_freemulti(ifma);
3513
3514 /*
3515 * The last reference to this instance of struct ifmultiaddr
3516 * was released; the hardware should be notified of this change.
3517 */
3518 return 1;
3519 }
3520
3521 /*
3522 * Set the link layer address on an interface.
3523 *
3524 * At this time we only support certain types of interfaces,
3525 * and we don't allow the length of the address to change.
3526 */
3527 int
if_setlladdr(struct ifnet * ifp,const u_char * lladdr,int len)3528 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
3529 {
3530 struct sockaddr_dl *sdl;
3531 struct ifaddr *ifa;
3532 struct ifreq ifr;
3533
3534 IF_ADDR_RLOCK(ifp);
3535 ifa = ifp->if_addr;
3536 if (ifa == NULL) {
3537 IF_ADDR_RUNLOCK(ifp);
3538 return (EINVAL);
3539 }
3540 ifa_ref(ifa);
3541 IF_ADDR_RUNLOCK(ifp);
3542 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
3543 if (sdl == NULL) {
3544 ifa_free(ifa);
3545 return (EINVAL);
3546 }
3547 if (len != sdl->sdl_alen) { /* don't allow length to change */
3548 ifa_free(ifa);
3549 return (EINVAL);
3550 }
3551 switch (ifp->if_type) {
3552 case IFT_ETHER:
3553 case IFT_FDDI:
3554 case IFT_XETHER:
3555 case IFT_ISO88025:
3556 case IFT_L2VLAN:
3557 case IFT_BRIDGE:
3558 case IFT_ARCNET:
3559 case IFT_IEEE8023ADLAG:
3560 case IFT_IEEE80211:
3561 bcopy(lladdr, LLADDR(sdl), len);
3562 ifa_free(ifa);
3563 break;
3564 default:
3565 ifa_free(ifa);
3566 return (ENODEV);
3567 }
3568
3569 /*
3570 * If the interface is already up, we need
3571 * to re-init it in order to reprogram its
3572 * address filter.
3573 */
3574 if ((ifp->if_flags & IFF_UP) != 0) {
3575 if (ifp->if_ioctl) {
3576 ifp->if_flags &= ~IFF_UP;
3577 ifr.ifr_flags = ifp->if_flags & 0xffff;
3578 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3579 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3580 ifp->if_flags |= IFF_UP;
3581 ifr.ifr_flags = ifp->if_flags & 0xffff;
3582 ifr.ifr_flagshigh = ifp->if_flags >> 16;
3583 (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
3584 }
3585 #ifdef INET
3586 /*
3587 * Also send gratuitous ARPs to notify other nodes about
3588 * the address change.
3589 */
3590 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
3591 if (ifa->ifa_addr->sa_family == AF_INET)
3592 arp_ifinit(ifp, ifa);
3593 }
3594 #endif
3595 }
3596 return (0);
3597 }
3598
3599 /*
3600 * Get the link layer address that was read from the hardware at attach.
3601 *
3602 * This is only set by Ethernet NICs (IFT_ETHER), but laggX interfaces re-type
3603 * their component interfaces as IFT_IEEE8023ADLAG.
3604 */
3605 int
if_gethwaddr(struct ifnet * ifp,struct ifreq * ifr)3606 if_gethwaddr(struct ifnet *ifp, struct ifreq *ifr)
3607 {
3608
3609 if (ifp->if_hw_addr == NULL)
3610 return (ENODEV);
3611
3612 switch (ifp->if_type) {
3613 case IFT_ETHER:
3614 case IFT_IEEE8023ADLAG:
3615 bcopy(ifp->if_hw_addr, ifr->ifr_addr.sa_data, ifp->if_addrlen);
3616 return (0);
3617 default:
3618 return (ENODEV);
3619 }
3620 }
3621
3622 /*
3623 * The name argument must be a pointer to storage which will last as
3624 * long as the interface does. For physical devices, the result of
3625 * device_get_name(dev) is a good choice and for pseudo-devices a
3626 * static string works well.
3627 */
3628 void
if_initname(struct ifnet * ifp,const char * name,int unit)3629 if_initname(struct ifnet *ifp, const char *name, int unit)
3630 {
3631 ifp->if_dname = name;
3632 ifp->if_dunit = unit;
3633 if (unit != IF_DUNIT_NONE)
3634 snprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
3635 else
3636 strlcpy(ifp->if_xname, name, IFNAMSIZ);
3637 }
3638
3639 int
if_printf(struct ifnet * ifp,const char * fmt,...)3640 if_printf(struct ifnet *ifp, const char * fmt, ...)
3641 {
3642 va_list ap;
3643 int retval;
3644
3645 retval = printf("%s: ", ifp->if_xname);
3646 va_start(ap, fmt);
3647 retval += vprintf(fmt, ap);
3648 va_end(ap);
3649 return (retval);
3650 }
3651
3652 void
if_start(struct ifnet * ifp)3653 if_start(struct ifnet *ifp)
3654 {
3655
3656 (*(ifp)->if_start)(ifp);
3657 }
3658
3659 /*
3660 * Backwards compatibility interface for drivers
3661 * that have not implemented it
3662 */
3663 static int
if_transmit(struct ifnet * ifp,struct mbuf * m)3664 if_transmit(struct ifnet *ifp, struct mbuf *m)
3665 {
3666 int error;
3667
3668 IFQ_HANDOFF(ifp, m, error);
3669 return (error);
3670 }
3671
3672 static void
if_input_default(struct ifnet * ifp __unused,struct mbuf * m)3673 if_input_default(struct ifnet *ifp __unused, struct mbuf *m)
3674 {
3675
3676 m_freem(m);
3677 }
3678
3679 int
if_handoff(struct ifqueue * ifq,struct mbuf * m,struct ifnet * ifp,int adjust)3680 if_handoff(struct ifqueue *ifq, struct mbuf *m, struct ifnet *ifp, int adjust)
3681 {
3682 int active = 0;
3683
3684 IF_LOCK(ifq);
3685 if (_IF_QFULL(ifq)) {
3686 _IF_DROP(ifq);
3687 IF_UNLOCK(ifq);
3688 m_freem(m);
3689 return (0);
3690 }
3691 if (ifp != NULL) {
3692 ifp->if_obytes += m->m_pkthdr.len + adjust;
3693 if (m->m_flags & (M_BCAST|M_MCAST))
3694 ifp->if_omcasts++;
3695 active = ifp->if_drv_flags & IFF_DRV_OACTIVE;
3696 }
3697 _IF_ENQUEUE(ifq, m);
3698 IF_UNLOCK(ifq);
3699 if (ifp != NULL && !active)
3700 (*(ifp)->if_start)(ifp);
3701 return (1);
3702 }
3703
3704 void
if_register_com_alloc(u_char type,if_com_alloc_t * a,if_com_free_t * f)3705 if_register_com_alloc(u_char type,
3706 if_com_alloc_t *a, if_com_free_t *f)
3707 {
3708
3709 KASSERT(if_com_alloc[type] == NULL,
3710 ("if_register_com_alloc: %d already registered", type));
3711 KASSERT(if_com_free[type] == NULL,
3712 ("if_register_com_alloc: %d free already registered", type));
3713
3714 if_com_alloc[type] = a;
3715 if_com_free[type] = f;
3716 }
3717
3718 void
if_deregister_com_alloc(u_char type)3719 if_deregister_com_alloc(u_char type)
3720 {
3721
3722 KASSERT(if_com_alloc[type] != NULL,
3723 ("if_deregister_com_alloc: %d not registered", type));
3724 KASSERT(if_com_free[type] != NULL,
3725 ("if_deregister_com_alloc: %d free not registered", type));
3726 if_com_alloc[type] = NULL;
3727 if_com_free[type] = NULL;
3728 }
3729