sys/net: Add dom_if_up and dom_if_down
[dragonfly.git] / sys / net / if.c
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  * 3. 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        8.3 (Berkeley) 1/4/94
30  * $FreeBSD: src/sys/net/if.c,v 1.185 2004/03/13 02:35:03 brooks Exp $
31  */
32
33 #include "opt_inet6.h"
34 #include "opt_inet.h"
35 #include "opt_ifpoll.h"
36
37 #include <sys/param.h>
38 #include <sys/malloc.h>
39 #include <sys/mbuf.h>
40 #include <sys/systm.h>
41 #include <sys/proc.h>
42 #include <sys/priv.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/socketops.h>
47 #include <sys/kernel.h>
48 #include <sys/ktr.h>
49 #include <sys/mutex.h>
50 #include <sys/lock.h>
51 #include <sys/sockio.h>
52 #include <sys/syslog.h>
53 #include <sys/sysctl.h>
54 #include <sys/domain.h>
55 #include <sys/thread.h>
56 #include <sys/serialize.h>
57 #include <sys/bus.h>
58 #include <sys/jail.h>
59
60 #include <sys/thread2.h>
61 #include <sys/msgport2.h>
62 #include <sys/mutex2.h>
63
64 #include <net/if.h>
65 #include <net/if_arp.h>
66 #include <net/if_dl.h>
67 #include <net/if_types.h>
68 #include <net/if_var.h>
69 #include <net/if_ringmap.h>
70 #include <net/ifq_var.h>
71 #include <net/radix.h>
72 #include <net/route.h>
73 #include <net/if_clone.h>
74 #include <net/netisr2.h>
75 #include <net/netmsg2.h>
76
77 #include <machine/atomic.h>
78 #include <machine/stdarg.h>
79 #include <machine/smp.h>
80
81 #if defined(INET) || defined(INET6)
82 #include <netinet/in.h>
83 #include <netinet/in_var.h>
84 #include <netinet/if_ether.h>
85 #ifdef INET6
86 #include <netinet6/in6_var.h>
87 #include <netinet6/in6_ifattach.h>
88 #endif /* INET6 */
89 #endif /* INET || INET6 */
90
91 struct netmsg_ifaddr {
92         struct netmsg_base base;
93         struct ifaddr   *ifa;
94         struct ifnet    *ifp;
95         int             tail;
96 };
97
98 struct ifsubq_stage_head {
99         TAILQ_HEAD(, ifsubq_stage)      stg_head;
100 } __cachealign;
101
102 struct if_ringmap {
103         int             rm_cnt;
104         int             rm_grid;
105         int             rm_cpumap[];
106 };
107
108 #define RINGMAP_FLAG_NONE               0x0
109 #define RINGMAP_FLAG_POWEROF2           0x1
110
111 /*
112  * System initialization
113  */
114 static void     if_attachdomain(void *);
115 static void     if_attachdomain1(struct ifnet *);
116 static int      ifconf(u_long, caddr_t, struct ucred *);
117 static void     ifinit(void *);
118 static void     ifnetinit(void *);
119 static void     if_slowtimo(void *);
120 static void     link_rtrequest(int, struct rtentry *);
121 static int      if_rtdel(struct radix_node *, void *);
122 static void     if_slowtimo_dispatch(netmsg_t);
123
124 /* Helper functions */
125 static void     ifsq_watchdog_reset(struct ifsubq_watchdog *);
126 static int      if_delmulti_serialized(struct ifnet *, struct sockaddr *);
127 static struct ifnet_array *ifnet_array_alloc(int);
128 static void     ifnet_array_free(struct ifnet_array *);
129 static struct ifnet_array *ifnet_array_add(struct ifnet *,
130                     const struct ifnet_array *);
131 static struct ifnet_array *ifnet_array_del(struct ifnet *,
132                     const struct ifnet_array *);
133 static struct ifg_group *if_creategroup(const char *);
134 static int      if_destroygroup(struct ifg_group *);
135 static int      if_delgroup_locked(struct ifnet *, const char *);
136 static int      if_getgroups(struct ifgroupreq *, struct ifnet *);
137 static int      if_getgroupmembers(struct ifgroupreq *);
138
139 #ifdef INET6
140 /*
141  * XXX: declare here to avoid to include many inet6 related files..
142  * should be more generalized?
143  */
144 extern void     nd6_setmtu(struct ifnet *);
145 #endif
146
147 SYSCTL_NODE(_net, PF_LINK, link, CTLFLAG_RW, 0, "Link layers");
148 SYSCTL_NODE(_net_link, 0, generic, CTLFLAG_RW, 0, "Generic link-management");
149 SYSCTL_NODE(_net_link, OID_AUTO, ringmap, CTLFLAG_RW, 0, "link ringmap");
150
151 static int ifsq_stage_cntmax = 16;
152 TUNABLE_INT("net.link.stage_cntmax", &ifsq_stage_cntmax);
153 SYSCTL_INT(_net_link, OID_AUTO, stage_cntmax, CTLFLAG_RW,
154     &ifsq_stage_cntmax, 0, "ifq staging packet count max");
155
156 static int if_stats_compat = 0;
157 SYSCTL_INT(_net_link, OID_AUTO, stats_compat, CTLFLAG_RW,
158     &if_stats_compat, 0, "Compat the old ifnet stats");
159
160 static int if_ringmap_dumprdr = 0;
161 SYSCTL_INT(_net_link_ringmap, OID_AUTO, dump_rdr, CTLFLAG_RW,
162     &if_ringmap_dumprdr, 0, "dump redirect table");
163
164 SYSINIT(interfaces, SI_SUB_PROTO_IF, SI_ORDER_FIRST, ifinit, NULL);
165 SYSINIT(ifnet, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, ifnetinit, NULL);
166
167 static if_com_alloc_t *if_com_alloc[256];
168 static if_com_free_t *if_com_free[256];
169
170 MALLOC_DEFINE(M_IFADDR, "ifaddr", "interface address");
171 MALLOC_DEFINE(M_IFMADDR, "ether_multi", "link-level multicast address");
172 MALLOC_DEFINE(M_IFNET, "ifnet", "interface structure");
173
174 int                     ifqmaxlen = IFQ_MAXLEN;
175 struct ifnethead        ifnet = TAILQ_HEAD_INITIALIZER(ifnet);
176 struct ifgrouphead      ifg_head = TAILQ_HEAD_INITIALIZER(ifg_head);
177 static struct lock      ifgroup_lock;
178
179 static struct ifnet_array       ifnet_array0;
180 static struct ifnet_array       *ifnet_array = &ifnet_array0;
181
182 static struct callout           if_slowtimo_timer;
183 static struct netmsg_base       if_slowtimo_netmsg;
184
185 int                     if_index = 0;
186 struct ifnet            **ifindex2ifnet = NULL;
187 static struct mtx       ifnet_mtx = MTX_INITIALIZER("ifnet");
188
189 static struct ifsubq_stage_head ifsubq_stage_heads[MAXCPU];
190
191 #ifdef notyet
192 #define IFQ_KTR_STRING          "ifq=%p"
193 #define IFQ_KTR_ARGS            struct ifaltq *ifq
194 #ifndef KTR_IFQ
195 #define KTR_IFQ                 KTR_ALL
196 #endif
197 KTR_INFO_MASTER(ifq);
198 KTR_INFO(KTR_IFQ, ifq, enqueue, 0, IFQ_KTR_STRING, IFQ_KTR_ARGS);
199 KTR_INFO(KTR_IFQ, ifq, dequeue, 1, IFQ_KTR_STRING, IFQ_KTR_ARGS);
200 #define logifq(name, arg)       KTR_LOG(ifq_ ## name, arg)
201
202 #define IF_START_KTR_STRING     "ifp=%p"
203 #define IF_START_KTR_ARGS       struct ifnet *ifp
204 #ifndef KTR_IF_START
205 #define KTR_IF_START            KTR_ALL
206 #endif
207 KTR_INFO_MASTER(if_start);
208 KTR_INFO(KTR_IF_START, if_start, run, 0,
209          IF_START_KTR_STRING, IF_START_KTR_ARGS);
210 KTR_INFO(KTR_IF_START, if_start, sched, 1,
211          IF_START_KTR_STRING, IF_START_KTR_ARGS);
212 KTR_INFO(KTR_IF_START, if_start, avoid, 2,
213          IF_START_KTR_STRING, IF_START_KTR_ARGS);
214 KTR_INFO(KTR_IF_START, if_start, contend_sched, 3,
215          IF_START_KTR_STRING, IF_START_KTR_ARGS);
216 KTR_INFO(KTR_IF_START, if_start, chase_sched, 4,
217          IF_START_KTR_STRING, IF_START_KTR_ARGS);
218 #define logifstart(name, arg)   KTR_LOG(if_start_ ## name, arg)
219 #endif /* notyet */
220
221 /*
222  * Network interface utility routines.
223  *
224  * Routines with ifa_ifwith* names take sockaddr *'s as
225  * parameters.
226  */
227 /* ARGSUSED */
228 static void
229 ifinit(void *dummy)
230 {
231         lockinit(&ifgroup_lock, "ifgroup", 0, 0);
232
233         callout_init_mp(&if_slowtimo_timer);
234         netmsg_init(&if_slowtimo_netmsg, NULL, &netisr_adone_rport,
235             MSGF_PRIORITY, if_slowtimo_dispatch);
236
237         /* Start if_slowtimo */
238         lwkt_sendmsg(netisr_cpuport(0), &if_slowtimo_netmsg.lmsg);
239 }
240
241 static void
242 ifsq_ifstart_ipifunc(void *arg)
243 {
244         struct ifaltq_subque *ifsq = arg;
245         struct lwkt_msg *lmsg = ifsq_get_ifstart_lmsg(ifsq, mycpuid);
246
247         crit_enter();
248         if (lmsg->ms_flags & MSGF_DONE)
249                 lwkt_sendmsg_oncpu(netisr_cpuport(mycpuid), lmsg);
250         crit_exit();
251 }
252
253 static __inline void
254 ifsq_stage_remove(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
255 {
256         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
257         TAILQ_REMOVE(&head->stg_head, stage, stg_link);
258         stage->stg_flags &= ~(IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED);
259         stage->stg_cnt = 0;
260         stage->stg_len = 0;
261 }
262
263 static __inline void
264 ifsq_stage_insert(struct ifsubq_stage_head *head, struct ifsubq_stage *stage)
265 {
266         KKASSERT((stage->stg_flags &
267             (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
268         stage->stg_flags |= IFSQ_STAGE_FLAG_QUED;
269         TAILQ_INSERT_TAIL(&head->stg_head, stage, stg_link);
270 }
271
272 /*
273  * Schedule ifnet.if_start on the subqueue owner CPU
274  */
275 static void
276 ifsq_ifstart_schedule(struct ifaltq_subque *ifsq, int force)
277 {
278         int cpu;
279
280         if (!force && curthread->td_type == TD_TYPE_NETISR &&
281             ifsq_stage_cntmax > 0) {
282                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
283
284                 stage->stg_cnt = 0;
285                 stage->stg_len = 0;
286                 if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
287                         ifsq_stage_insert(&ifsubq_stage_heads[mycpuid], stage);
288                 stage->stg_flags |= IFSQ_STAGE_FLAG_SCHED;
289                 return;
290         }
291
292         cpu = ifsq_get_cpuid(ifsq);
293         if (cpu != mycpuid)
294                 lwkt_send_ipiq(globaldata_find(cpu), ifsq_ifstart_ipifunc, ifsq);
295         else
296                 ifsq_ifstart_ipifunc(ifsq);
297 }
298
299 /*
300  * NOTE:
301  * This function will release ifnet.if_start subqueue interlock,
302  * if ifnet.if_start for the subqueue does not need to be scheduled
303  */
304 static __inline int
305 ifsq_ifstart_need_schedule(struct ifaltq_subque *ifsq, int running)
306 {
307         if (!running || ifsq_is_empty(ifsq)
308 #ifdef ALTQ
309             || ifsq->ifsq_altq->altq_tbr != NULL
310 #endif
311         ) {
312                 ALTQ_SQ_LOCK(ifsq);
313                 /*
314                  * ifnet.if_start subqueue interlock is released, if:
315                  * 1) Hardware can not take any packets, due to
316                  *    o  interface is marked down
317                  *    o  hardware queue is full (ifsq_is_oactive)
318                  *    Under the second situation, hardware interrupt
319                  *    or polling(4) will call/schedule ifnet.if_start
320                  *    on the subqueue when hardware queue is ready
321                  * 2) There is no packet in the subqueue.
322                  *    Further ifq_dispatch or ifq_handoff will call/
323                  *    schedule ifnet.if_start on the subqueue.
324                  * 3) TBR is used and it does not allow further
325                  *    dequeueing.
326                  *    TBR callout will call ifnet.if_start on the
327                  *    subqueue.
328                  */
329                 if (!running || !ifsq_data_ready(ifsq)) {
330                         ifsq_clr_started(ifsq);
331                         ALTQ_SQ_UNLOCK(ifsq);
332                         return 0;
333                 }
334                 ALTQ_SQ_UNLOCK(ifsq);
335         }
336         return 1;
337 }
338
339 static void
340 ifsq_ifstart_dispatch(netmsg_t msg)
341 {
342         struct lwkt_msg *lmsg = &msg->base.lmsg;
343         struct ifaltq_subque *ifsq = lmsg->u.ms_resultp;
344         struct ifnet *ifp = ifsq_get_ifp(ifsq);
345         struct globaldata *gd = mycpu;
346         int running = 0, need_sched;
347
348         crit_enter_gd(gd);
349
350         lwkt_replymsg(lmsg, 0); /* reply ASAP */
351
352         if (gd->gd_cpuid != ifsq_get_cpuid(ifsq)) {
353                 /*
354                  * We need to chase the subqueue owner CPU change.
355                  */
356                 ifsq_ifstart_schedule(ifsq, 1);
357                 crit_exit_gd(gd);
358                 return;
359         }
360
361         ifsq_serialize_hw(ifsq);
362         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
363                 ifp->if_start(ifp, ifsq);
364                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
365                         running = 1;
366         }
367         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
368         ifsq_deserialize_hw(ifsq);
369
370         if (need_sched) {
371                 /*
372                  * More data need to be transmitted, ifnet.if_start is
373                  * scheduled on the subqueue owner CPU, and we keep going.
374                  * NOTE: ifnet.if_start subqueue interlock is not released.
375                  */
376                 ifsq_ifstart_schedule(ifsq, 0);
377         }
378
379         crit_exit_gd(gd);
380 }
381
382 /* Device driver ifnet.if_start helper function */
383 void
384 ifsq_devstart(struct ifaltq_subque *ifsq)
385 {
386         struct ifnet *ifp = ifsq_get_ifp(ifsq);
387         int running = 0;
388
389         ASSERT_ALTQ_SQ_SERIALIZED_HW(ifsq);
390
391         ALTQ_SQ_LOCK(ifsq);
392         if (ifsq_is_started(ifsq) || !ifsq_data_ready(ifsq)) {
393                 ALTQ_SQ_UNLOCK(ifsq);
394                 return;
395         }
396         ifsq_set_started(ifsq);
397         ALTQ_SQ_UNLOCK(ifsq);
398
399         ifp->if_start(ifp, ifsq);
400
401         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
402                 running = 1;
403
404         if (ifsq_ifstart_need_schedule(ifsq, running)) {
405                 /*
406                  * More data need to be transmitted, ifnet.if_start is
407                  * scheduled on ifnet's CPU, and we keep going.
408                  * NOTE: ifnet.if_start interlock is not released.
409                  */
410                 ifsq_ifstart_schedule(ifsq, 0);
411         }
412 }
413
414 void
415 if_devstart(struct ifnet *ifp)
416 {
417         ifsq_devstart(ifq_get_subq_default(&ifp->if_snd));
418 }
419
420 /* Device driver ifnet.if_start schedule helper function */
421 void
422 ifsq_devstart_sched(struct ifaltq_subque *ifsq)
423 {
424         ifsq_ifstart_schedule(ifsq, 1);
425 }
426
427 void
428 if_devstart_sched(struct ifnet *ifp)
429 {
430         ifsq_devstart_sched(ifq_get_subq_default(&ifp->if_snd));
431 }
432
433 static void
434 if_default_serialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
435 {
436         lwkt_serialize_enter(ifp->if_serializer);
437 }
438
439 static void
440 if_default_deserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
441 {
442         lwkt_serialize_exit(ifp->if_serializer);
443 }
444
445 static int
446 if_default_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz __unused)
447 {
448         return lwkt_serialize_try(ifp->if_serializer);
449 }
450
451 #ifdef INVARIANTS
452 static void
453 if_default_serialize_assert(struct ifnet *ifp,
454                             enum ifnet_serialize slz __unused,
455                             boolean_t serialized)
456 {
457         if (serialized)
458                 ASSERT_SERIALIZED(ifp->if_serializer);
459         else
460                 ASSERT_NOT_SERIALIZED(ifp->if_serializer);
461 }
462 #endif
463
464 /*
465  * Attach an interface to the list of "active" interfaces.
466  *
467  * The serializer is optional.
468  */
469 void
470 if_attach(struct ifnet *ifp, lwkt_serialize_t serializer)
471 {
472         unsigned socksize;
473         int namelen, masklen;
474         struct sockaddr_dl *sdl, *sdl_addr;
475         struct ifaddr *ifa;
476         struct ifaltq *ifq;
477         struct ifnet **old_ifindex2ifnet = NULL;
478         struct ifnet_array *old_ifnet_array;
479         int i, q, qlen;
480         char qlenname[64];
481
482         static int if_indexlim = 8;
483
484         if (ifp->if_serialize != NULL) {
485                 KASSERT(ifp->if_deserialize != NULL &&
486                         ifp->if_tryserialize != NULL &&
487                         ifp->if_serialize_assert != NULL,
488                         ("serialize functions are partially setup"));
489
490                 /*
491                  * If the device supplies serialize functions,
492                  * then clear if_serializer to catch any invalid
493                  * usage of this field.
494                  */
495                 KASSERT(serializer == NULL,
496                         ("both serialize functions and default serializer "
497                          "are supplied"));
498                 ifp->if_serializer = NULL;
499         } else {
500                 KASSERT(ifp->if_deserialize == NULL &&
501                         ifp->if_tryserialize == NULL &&
502                         ifp->if_serialize_assert == NULL,
503                         ("serialize functions are partially setup"));
504                 ifp->if_serialize = if_default_serialize;
505                 ifp->if_deserialize = if_default_deserialize;
506                 ifp->if_tryserialize = if_default_tryserialize;
507 #ifdef INVARIANTS
508                 ifp->if_serialize_assert = if_default_serialize_assert;
509 #endif
510
511                 /*
512                  * The serializer can be passed in from the device,
513                  * allowing the same serializer to be used for both
514                  * the interrupt interlock and the device queue.
515                  * If not specified, the netif structure will use an
516                  * embedded serializer.
517                  */
518                 if (serializer == NULL) {
519                         serializer = &ifp->if_default_serializer;
520                         lwkt_serialize_init(serializer);
521                 }
522                 ifp->if_serializer = serializer;
523         }
524
525         /*
526          * Make if_addrhead available on all CPUs, since they
527          * could be accessed by any threads.
528          */
529         ifp->if_addrheads = kmalloc(ncpus * sizeof(struct ifaddrhead),
530                                     M_IFADDR, M_WAITOK | M_ZERO);
531         for (i = 0; i < ncpus; ++i)
532                 TAILQ_INIT(&ifp->if_addrheads[i]);
533
534         TAILQ_INIT(&ifp->if_multiaddrs);
535         TAILQ_INIT(&ifp->if_groups);
536         getmicrotime(&ifp->if_lastchange);
537         if_addgroup(ifp, IFG_ALL);
538
539         /*
540          * create a Link Level name for this device
541          */
542         namelen = strlen(ifp->if_xname);
543         masklen = offsetof(struct sockaddr_dl, sdl_data[0]) + namelen;
544         socksize = masklen + ifp->if_addrlen;
545         if (socksize < sizeof(*sdl))
546                 socksize = sizeof(*sdl);
547         socksize = RT_ROUNDUP(socksize);
548         ifa = ifa_create(sizeof(struct ifaddr) + 2 * socksize);
549         sdl = sdl_addr = (struct sockaddr_dl *)(ifa + 1);
550         sdl->sdl_len = socksize;
551         sdl->sdl_family = AF_LINK;
552         bcopy(ifp->if_xname, sdl->sdl_data, namelen);
553         sdl->sdl_nlen = namelen;
554         sdl->sdl_type = ifp->if_type;
555         ifp->if_lladdr = ifa;
556         ifa->ifa_ifp = ifp;
557         ifa->ifa_rtrequest = link_rtrequest;
558         ifa->ifa_addr = (struct sockaddr *)sdl;
559         sdl = (struct sockaddr_dl *)(socksize + (caddr_t)sdl);
560         ifa->ifa_netmask = (struct sockaddr *)sdl;
561         sdl->sdl_len = masklen;
562         while (namelen != 0)
563                 sdl->sdl_data[--namelen] = 0xff;
564         ifa_iflink(ifa, ifp, 0 /* Insert head */);
565
566         /*
567          * Make if_data available on all CPUs, since they could
568          * be updated by hardware interrupt routing, which could
569          * be bound to any CPU.
570          */
571         ifp->if_data_pcpu = kmalloc(ncpus * sizeof(struct ifdata_pcpu),
572                                     M_DEVBUF,
573                                     M_WAITOK | M_ZERO | M_CACHEALIGN);
574
575         if (ifp->if_mapsubq == NULL)
576                 ifp->if_mapsubq = ifq_mapsubq_default;
577
578         ifq = &ifp->if_snd;
579         ifq->altq_type = 0;
580         ifq->altq_disc = NULL;
581         ifq->altq_flags &= ALTQF_CANTCHANGE;
582         ifq->altq_tbr = NULL;
583         ifq->altq_ifp = ifp;
584
585         if (ifq->altq_subq_cnt <= 0)
586                 ifq->altq_subq_cnt = 1;
587         ifq->altq_subq =
588                 kmalloc(ifq->altq_subq_cnt * sizeof(struct ifaltq_subque),
589                         M_DEVBUF,
590                         M_WAITOK | M_ZERO | M_CACHEALIGN);
591
592         if (ifq->altq_maxlen == 0) {
593                 if_printf(ifp, "driver didn't set altq_maxlen\n");
594                 ifq_set_maxlen(ifq, ifqmaxlen);
595         }
596
597         /* Allow user to override driver's setting. */
598         ksnprintf(qlenname, sizeof(qlenname), "net.%s.qlenmax", ifp->if_xname);
599         qlen = -1;
600         TUNABLE_INT_FETCH(qlenname, &qlen);
601         if (qlen > 0) {
602                 if_printf(ifp, "qlenmax -> %d\n", qlen);
603                 ifq_set_maxlen(ifq, qlen);
604         }
605
606         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
607                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
608
609                 ALTQ_SQ_LOCK_INIT(ifsq);
610                 ifsq->ifsq_index = q;
611
612                 ifsq->ifsq_altq = ifq;
613                 ifsq->ifsq_ifp = ifp;
614
615                 ifsq->ifsq_maxlen = ifq->altq_maxlen;
616                 ifsq->ifsq_maxbcnt = ifsq->ifsq_maxlen * MCLBYTES;
617                 ifsq->ifsq_prepended = NULL;
618                 ifsq->ifsq_started = 0;
619                 ifsq->ifsq_hw_oactive = 0;
620                 ifsq_set_cpuid(ifsq, 0);
621                 if (ifp->if_serializer != NULL)
622                         ifsq_set_hw_serialize(ifsq, ifp->if_serializer);
623
624                 /* XXX: netisr_ncpus */
625                 ifsq->ifsq_stage =
626                         kmalloc(ncpus * sizeof(struct ifsubq_stage),
627                                 M_DEVBUF,
628                                 M_WAITOK | M_ZERO | M_CACHEALIGN);
629                 for (i = 0; i < ncpus; ++i)
630                         ifsq->ifsq_stage[i].stg_subq = ifsq;
631
632                 /*
633                  * Allocate one if_start message for each CPU, since
634                  * the hardware TX ring could be assigned to any CPU.
635                  *
636                  * NOTE:
637                  * If the hardware TX ring polling CPU and the hardware
638                  * TX ring interrupt CPU are same, one if_start message
639                  * should be enough.
640                  */
641                 ifsq->ifsq_ifstart_nmsg =
642                     kmalloc(ncpus * sizeof(struct netmsg_base),
643                     M_LWKTMSG, M_WAITOK);
644                 for (i = 0; i < ncpus; ++i) {
645                         netmsg_init(&ifsq->ifsq_ifstart_nmsg[i], NULL,
646                             &netisr_adone_rport, 0, ifsq_ifstart_dispatch);
647                         ifsq->ifsq_ifstart_nmsg[i].lmsg.u.ms_resultp = ifsq;
648                 }
649         }
650         ifq_set_classic(ifq);
651
652         /*
653          * Increase mbuf cluster/jcluster limits for the mbufs that
654          * could sit on the device queues for quite some time.
655          */
656         if (ifp->if_nmbclusters > 0)
657                 mcl_inclimit(ifp->if_nmbclusters);
658         if (ifp->if_nmbjclusters > 0)
659                 mjcl_inclimit(ifp->if_nmbjclusters);
660
661         /*
662          * Install this ifp into ifindex2inet, ifnet queue and ifnet
663          * array after it is setup.
664          *
665          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
666          * by ifnet lock, so that non-netisr threads could get a
667          * consistent view.
668          */
669         ifnet_lock();
670
671         /* Don't update if_index until ifindex2ifnet is setup */
672         ifp->if_index = if_index + 1;
673         sdl_addr->sdl_index = ifp->if_index;
674
675         /*
676          * Install this ifp into ifindex2ifnet
677          */
678         if (ifindex2ifnet == NULL || ifp->if_index >= if_indexlim) {
679                 unsigned int n;
680                 struct ifnet **q;
681
682                 /*
683                  * Grow ifindex2ifnet
684                  */
685                 if_indexlim <<= 1;
686                 n = if_indexlim * sizeof(*q);
687                 q = kmalloc(n, M_IFADDR, M_WAITOK | M_ZERO);
688                 if (ifindex2ifnet != NULL) {
689                         bcopy(ifindex2ifnet, q, n/2);
690                         /* Free old ifindex2ifnet after sync all netisrs */
691                         old_ifindex2ifnet = ifindex2ifnet;
692                 }
693                 ifindex2ifnet = q;
694         }
695         ifindex2ifnet[ifp->if_index] = ifp;
696         /*
697          * Update if_index after this ifp is installed into ifindex2ifnet,
698          * so that netisrs could get a consistent view of ifindex2ifnet.
699          */
700         cpu_sfence();
701         if_index = ifp->if_index;
702
703         /*
704          * Install this ifp into ifnet array.
705          */
706         /* Free old ifnet array after sync all netisrs */
707         old_ifnet_array = ifnet_array;
708         ifnet_array = ifnet_array_add(ifp, old_ifnet_array);
709
710         /*
711          * Install this ifp into ifnet queue.
712          */
713         TAILQ_INSERT_TAIL(&ifnetlist, ifp, if_link);
714
715         ifnet_unlock();
716
717         /*
718          * Sync all netisrs so that the old ifindex2ifnet and ifnet array
719          * are no longer accessed and we can free them safely later on.
720          */
721         netmsg_service_sync();
722         if (old_ifindex2ifnet != NULL)
723                 kfree(old_ifindex2ifnet, M_IFADDR);
724         ifnet_array_free(old_ifnet_array);
725
726         if (!SLIST_EMPTY(&domains))
727                 if_attachdomain1(ifp);
728
729         /* Announce the interface. */
730         EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
731         devctl_notify("IFNET", ifp->if_xname, "ATTACH", NULL);
732         rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
733 }
734
735 static void
736 if_attachdomain(void *dummy)
737 {
738         struct ifnet *ifp;
739
740         ifnet_lock();
741         TAILQ_FOREACH(ifp, &ifnetlist, if_list)
742                 if_attachdomain1(ifp);
743         ifnet_unlock();
744 }
745 SYSINIT(domainifattach, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_FIRST,
746         if_attachdomain, NULL);
747
748 static void
749 if_attachdomain1(struct ifnet *ifp)
750 {
751         struct domain *dp;
752
753         crit_enter();
754
755         /* address family dependent data region */
756         bzero(ifp->if_afdata, sizeof(ifp->if_afdata));
757         SLIST_FOREACH(dp, &domains, dom_next)
758                 if (dp->dom_ifattach)
759                         ifp->if_afdata[dp->dom_family] =
760                                 (*dp->dom_ifattach)(ifp);
761         crit_exit();
762 }
763
764 /*
765  * Purge all addresses whose type is _not_ AF_LINK
766  */
767 static void
768 if_purgeaddrs_nolink_dispatch(netmsg_t nmsg)
769 {
770         struct ifnet *ifp = nmsg->lmsg.u.ms_resultp;
771         struct ifaddr_container *ifac, *next;
772
773         ASSERT_NETISR0;
774
775         /*
776          * The ifaddr processing in the following loop will block,
777          * however, this function is called in netisr0, in which
778          * ifaddr list changes happen, so we don't care about the
779          * blockness of the ifaddr processing here.
780          */
781         TAILQ_FOREACH_MUTABLE(ifac, &ifp->if_addrheads[mycpuid],
782                               ifa_link, next) {
783                 struct ifaddr *ifa = ifac->ifa;
784
785                 /* Ignore marker */
786                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
787                         continue;
788
789                 /* Leave link ifaddr as it is */
790                 if (ifa->ifa_addr->sa_family == AF_LINK)
791                         continue;
792 #ifdef INET
793                 /* XXX: Ugly!! ad hoc just for INET */
794                 if (ifa->ifa_addr->sa_family == AF_INET) {
795                         struct ifaliasreq ifr;
796                         struct sockaddr_in saved_addr, saved_dst;
797 #ifdef IFADDR_DEBUG_VERBOSE
798                         int i;
799
800                         kprintf("purge in4 addr %p: ", ifa);
801                         for (i = 0; i < ncpus; ++i) {
802                                 kprintf("%d ",
803                                     ifa->ifa_containers[i].ifa_refcnt);
804                         }
805                         kprintf("\n");
806 #endif
807
808                         /* Save information for panic. */
809                         memcpy(&saved_addr, ifa->ifa_addr, sizeof(saved_addr));
810                         if (ifa->ifa_dstaddr != NULL) {
811                                 memcpy(&saved_dst, ifa->ifa_dstaddr,
812                                     sizeof(saved_dst));
813                         } else {
814                                 memset(&saved_dst, 0, sizeof(saved_dst));
815                         }
816
817                         bzero(&ifr, sizeof ifr);
818                         ifr.ifra_addr = *ifa->ifa_addr;
819                         if (ifa->ifa_dstaddr)
820                                 ifr.ifra_broadaddr = *ifa->ifa_dstaddr;
821                         if (in_control(SIOCDIFADDR, (caddr_t)&ifr, ifp,
822                                        NULL) == 0)
823                                 continue;
824
825                         /* MUST NOT HAPPEN */
826                         panic("%s: in_control failed %x, dst %x", ifp->if_xname,
827                             ntohl(saved_addr.sin_addr.s_addr),
828                             ntohl(saved_dst.sin_addr.s_addr));
829                 }
830 #endif /* INET */
831 #ifdef INET6
832                 if (ifa->ifa_addr->sa_family == AF_INET6) {
833 #ifdef IFADDR_DEBUG_VERBOSE
834                         int i;
835
836                         kprintf("purge in6 addr %p: ", ifa);
837                         for (i = 0; i < ncpus; ++i) {
838                                 kprintf("%d ",
839                                     ifa->ifa_containers[i].ifa_refcnt);
840                         }
841                         kprintf("\n");
842 #endif
843
844                         in6_purgeaddr(ifa);
845                         /* ifp_addrhead is already updated */
846                         continue;
847                 }
848 #endif /* INET6 */
849                 if_printf(ifp, "destroy ifaddr family %d\n",
850                     ifa->ifa_addr->sa_family);
851                 ifa_ifunlink(ifa, ifp);
852                 ifa_destroy(ifa);
853         }
854
855         netisr_replymsg(&nmsg->base, 0);
856 }
857
858 void
859 if_purgeaddrs_nolink(struct ifnet *ifp)
860 {
861         struct netmsg_base nmsg;
862
863         netmsg_init(&nmsg, NULL, &curthread->td_msgport, 0,
864             if_purgeaddrs_nolink_dispatch);
865         nmsg.lmsg.u.ms_resultp = ifp;
866         netisr_domsg(&nmsg, 0);
867 }
868
869 static void
870 ifq_stage_detach_handler(netmsg_t nmsg)
871 {
872         struct ifaltq *ifq = nmsg->lmsg.u.ms_resultp;
873         int q;
874
875         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
876                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
877                 struct ifsubq_stage *stage = ifsq_get_stage(ifsq, mycpuid);
878
879                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED)
880                         ifsq_stage_remove(&ifsubq_stage_heads[mycpuid], stage);
881         }
882         lwkt_replymsg(&nmsg->lmsg, 0);
883 }
884
885 static void
886 ifq_stage_detach(struct ifaltq *ifq)
887 {
888         struct netmsg_base base;
889         int cpu;
890
891         netmsg_init(&base, NULL, &curthread->td_msgport, 0,
892             ifq_stage_detach_handler);
893         base.lmsg.u.ms_resultp = ifq;
894
895         /* XXX netisr_ncpus */
896         for (cpu = 0; cpu < ncpus; ++cpu)
897                 lwkt_domsg(netisr_cpuport(cpu), &base.lmsg, 0);
898 }
899
900 struct netmsg_if_rtdel {
901         struct netmsg_base      base;
902         struct ifnet            *ifp;
903 };
904
905 static void
906 if_rtdel_dispatch(netmsg_t msg)
907 {
908         struct netmsg_if_rtdel *rmsg = (void *)msg;
909         int i, cpu;
910
911         cpu = mycpuid;
912         ASSERT_NETISR_NCPUS(cpu);
913
914         for (i = 1; i <= AF_MAX; i++) {
915                 struct radix_node_head  *rnh;
916
917                 if ((rnh = rt_tables[cpu][i]) == NULL)
918                         continue;
919                 rnh->rnh_walktree(rnh, if_rtdel, rmsg->ifp);
920         }
921         netisr_forwardmsg(&msg->base, cpu + 1);
922 }
923
924 /*
925  * Detach an interface, removing it from the
926  * list of "active" interfaces.
927  */
928 void
929 if_detach(struct ifnet *ifp)
930 {
931         struct ifnet_array *old_ifnet_array;
932         struct ifg_list *ifgl;
933         struct netmsg_if_rtdel msg;
934         struct domain *dp;
935         int q;
936
937         /* Announce that the interface is gone. */
938         EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
939         rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
940         devctl_notify("IFNET", ifp->if_xname, "DETACH", NULL);
941
942         /*
943          * Remove this ifp from ifindex2inet, ifnet queue and ifnet
944          * array before it is whacked.
945          *
946          * Protect ifindex2ifnet, ifnet queue and ifnet array changes
947          * by ifnet lock, so that non-netisr threads could get a
948          * consistent view.
949          */
950         ifnet_lock();
951
952         /*
953          * Remove this ifp from ifindex2ifnet and maybe decrement if_index.
954          */
955         ifindex2ifnet[ifp->if_index] = NULL;
956         while (if_index > 0 && ifindex2ifnet[if_index] == NULL)
957                 if_index--;
958
959         /*
960          * Remove this ifp from ifnet queue.
961          */
962         TAILQ_REMOVE(&ifnetlist, ifp, if_link);
963
964         /*
965          * Remove this ifp from ifnet array.
966          */
967         /* Free old ifnet array after sync all netisrs */
968         old_ifnet_array = ifnet_array;
969         ifnet_array = ifnet_array_del(ifp, old_ifnet_array);
970
971         ifnet_unlock();
972
973         ifgroup_lockmgr(LK_EXCLUSIVE);
974         while ((ifgl = TAILQ_FIRST(&ifp->if_groups)) != NULL)
975                 if_delgroup_locked(ifp, ifgl->ifgl_group->ifg_group);
976         ifgroup_lockmgr(LK_RELEASE);
977
978         /*
979          * Sync all netisrs so that the old ifnet array is no longer
980          * accessed and we can free it safely later on.
981          */
982         netmsg_service_sync();
983         ifnet_array_free(old_ifnet_array);
984
985         /*
986          * Remove routes and flush queues.
987          */
988         crit_enter();
989 #ifdef IFPOLL_ENABLE
990         if (ifp->if_flags & IFF_NPOLLING)
991                 ifpoll_deregister(ifp);
992 #endif
993         if_down(ifp);
994
995         /* Decrease the mbuf clusters/jclusters limits increased by us */
996         if (ifp->if_nmbclusters > 0)
997                 mcl_inclimit(-ifp->if_nmbclusters);
998         if (ifp->if_nmbjclusters > 0)
999                 mjcl_inclimit(-ifp->if_nmbjclusters);
1000
1001 #ifdef ALTQ
1002         if (ifq_is_enabled(&ifp->if_snd))
1003                 altq_disable(&ifp->if_snd);
1004         if (ifq_is_attached(&ifp->if_snd))
1005                 altq_detach(&ifp->if_snd);
1006 #endif
1007
1008         /*
1009          * Clean up all addresses.
1010          */
1011         ifp->if_lladdr = NULL;
1012
1013         if_purgeaddrs_nolink(ifp);
1014         if (!TAILQ_EMPTY(&ifp->if_addrheads[mycpuid])) {
1015                 struct ifaddr *ifa;
1016
1017                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
1018                 KASSERT(ifa->ifa_addr->sa_family == AF_LINK,
1019                         ("non-link ifaddr is left on if_addrheads"));
1020
1021                 ifa_ifunlink(ifa, ifp);
1022                 ifa_destroy(ifa);
1023                 KASSERT(TAILQ_EMPTY(&ifp->if_addrheads[mycpuid]),
1024                         ("there are still ifaddrs left on if_addrheads"));
1025         }
1026
1027 #ifdef INET
1028         /*
1029          * Remove all IPv4 kernel structures related to ifp.
1030          */
1031         in_ifdetach(ifp);
1032 #endif
1033
1034 #ifdef INET6
1035         /*
1036          * Remove all IPv6 kernel structs related to ifp.  This should be done
1037          * before removing routing entries below, since IPv6 interface direct
1038          * routes are expected to be removed by the IPv6-specific kernel API.
1039          * Otherwise, the kernel will detect some inconsistency and bark it.
1040          */
1041         in6_ifdetach(ifp);
1042 #endif
1043
1044         /*
1045          * Delete all remaining routes using this interface
1046          */
1047         netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY,
1048             if_rtdel_dispatch);
1049         msg.ifp = ifp;
1050         netisr_domsg_global(&msg.base);
1051
1052         SLIST_FOREACH(dp, &domains, dom_next) {
1053                 if (dp->dom_ifdetach && ifp->if_afdata[dp->dom_family])
1054                         (*dp->dom_ifdetach)(ifp,
1055                                 ifp->if_afdata[dp->dom_family]);
1056         }
1057
1058         kfree(ifp->if_addrheads, M_IFADDR);
1059
1060         lwkt_synchronize_ipiqs("if_detach");
1061         ifq_stage_detach(&ifp->if_snd);
1062
1063         for (q = 0; q < ifp->if_snd.altq_subq_cnt; ++q) {
1064                 struct ifaltq_subque *ifsq = &ifp->if_snd.altq_subq[q];
1065
1066                 kfree(ifsq->ifsq_ifstart_nmsg, M_LWKTMSG);
1067                 kfree(ifsq->ifsq_stage, M_DEVBUF);
1068         }
1069         kfree(ifp->if_snd.altq_subq, M_DEVBUF);
1070
1071         kfree(ifp->if_data_pcpu, M_DEVBUF);
1072
1073         crit_exit();
1074 }
1075
1076 int
1077 ifgroup_lockmgr(u_int flags)
1078 {
1079         return lockmgr(&ifgroup_lock, flags);
1080 }
1081
1082 /*
1083  * Create an empty interface group.
1084  */
1085 static struct ifg_group *
1086 if_creategroup(const char *groupname)
1087 {
1088         struct ifg_group *ifg;
1089
1090         ifg = kmalloc(sizeof(*ifg), M_IFNET, M_WAITOK);
1091         strlcpy(ifg->ifg_group, groupname, sizeof(ifg->ifg_group));
1092         ifg->ifg_refcnt = 0;
1093         ifg->ifg_carp_demoted = 0;
1094         TAILQ_INIT(&ifg->ifg_members);
1095
1096         ifgroup_lockmgr(LK_EXCLUSIVE);
1097         TAILQ_INSERT_TAIL(&ifg_head, ifg, ifg_next);
1098         ifgroup_lockmgr(LK_RELEASE);
1099
1100         EVENTHANDLER_INVOKE(group_attach_event, ifg);
1101
1102         return (ifg);
1103 }
1104
1105 /*
1106  * Destroy an empty interface group.
1107  */
1108 static int
1109 if_destroygroup(struct ifg_group *ifg)
1110 {
1111         KASSERT(ifg->ifg_refcnt == 0,
1112                 ("trying to delete a non-empty interface group"));
1113
1114         ifgroup_lockmgr(LK_EXCLUSIVE);
1115         TAILQ_REMOVE(&ifg_head, ifg, ifg_next);
1116         ifgroup_lockmgr(LK_RELEASE);
1117
1118         EVENTHANDLER_INVOKE(group_detach_event, ifg);
1119         kfree(ifg, M_IFNET);
1120
1121         return (0);
1122 }
1123
1124 /*
1125  * Add the interface to a group.
1126  * The target group will be created if it doesn't exist.
1127  */
1128 int
1129 if_addgroup(struct ifnet *ifp, const char *groupname)
1130 {
1131         struct ifg_list *ifgl;
1132         struct ifg_group *ifg;
1133         struct ifg_member *ifgm;
1134
1135         if (groupname[0] &&
1136             groupname[strlen(groupname) - 1] >= '0' &&
1137             groupname[strlen(groupname) - 1] <= '9')
1138                 return (EINVAL);
1139
1140         ifgroup_lockmgr(LK_SHARED);
1141
1142         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1143                 if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0) {
1144                         ifgroup_lockmgr(LK_RELEASE);
1145                         return (EEXIST);
1146                 }
1147         }
1148
1149         TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1150                 if (strcmp(ifg->ifg_group, groupname) == 0)
1151                         break;
1152         }
1153
1154         ifgroup_lockmgr(LK_RELEASE);
1155
1156         if (ifg == NULL)
1157                 ifg = if_creategroup(groupname);
1158
1159         ifgl = kmalloc(sizeof(*ifgl), M_IFNET, M_WAITOK);
1160         ifgm = kmalloc(sizeof(*ifgm), M_IFNET, M_WAITOK);
1161         ifgl->ifgl_group = ifg;
1162         ifgm->ifgm_ifp = ifp;
1163         ifg->ifg_refcnt++;
1164
1165         ifgroup_lockmgr(LK_EXCLUSIVE);
1166         TAILQ_INSERT_TAIL(&ifg->ifg_members, ifgm, ifgm_next);
1167         TAILQ_INSERT_TAIL(&ifp->if_groups, ifgl, ifgl_next);
1168         ifgroup_lockmgr(LK_RELEASE);
1169
1170         EVENTHANDLER_INVOKE(group_change_event, groupname);
1171
1172         return (0);
1173 }
1174
1175 /*
1176  * Remove the interface from a group.
1177  * The group will be destroyed if it becomes empty.
1178  *
1179  * The 'ifgroup_lock' must be hold exclusively when calling this.
1180  */
1181 static int
1182 if_delgroup_locked(struct ifnet *ifp, const char *groupname)
1183 {
1184         struct ifg_list *ifgl;
1185         struct ifg_member *ifgm;
1186
1187         KKASSERT(lockstatus(&ifgroup_lock, curthread) == LK_EXCLUSIVE);
1188
1189         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1190                 if (strcmp(ifgl->ifgl_group->ifg_group, groupname) == 0)
1191                         break;
1192         }
1193         if (ifgl == NULL)
1194                 return (ENOENT);
1195
1196         TAILQ_REMOVE(&ifp->if_groups, ifgl, ifgl_next);
1197
1198         TAILQ_FOREACH(ifgm, &ifgl->ifgl_group->ifg_members, ifgm_next) {
1199                 if (ifgm->ifgm_ifp == ifp)
1200                         break;
1201         }
1202
1203         if (ifgm != NULL) {
1204                 TAILQ_REMOVE(&ifgl->ifgl_group->ifg_members, ifgm, ifgm_next);
1205
1206                 ifgroup_lockmgr(LK_RELEASE);
1207                 EVENTHANDLER_INVOKE(group_change_event, groupname);
1208                 ifgroup_lockmgr(LK_EXCLUSIVE);
1209
1210                 kfree(ifgm, M_IFNET);
1211                 ifgl->ifgl_group->ifg_refcnt--;
1212         }
1213
1214         if (ifgl->ifgl_group->ifg_refcnt == 0) {
1215                 ifgroup_lockmgr(LK_RELEASE);
1216                 if_destroygroup(ifgl->ifgl_group);
1217                 ifgroup_lockmgr(LK_EXCLUSIVE);
1218         }
1219
1220         kfree(ifgl, M_IFNET);
1221
1222         return (0);
1223 }
1224
1225 int
1226 if_delgroup(struct ifnet *ifp, const char *groupname)
1227 {
1228         int error;
1229
1230         ifgroup_lockmgr(LK_EXCLUSIVE);
1231         error = if_delgroup_locked(ifp, groupname);
1232         ifgroup_lockmgr(LK_RELEASE);
1233
1234         return (error);
1235 }
1236
1237 /*
1238  * Store all the groups that the interface belongs to in memory
1239  * pointed to by data.
1240  */
1241 static int
1242 if_getgroups(struct ifgroupreq *ifgr, struct ifnet *ifp)
1243 {
1244         struct ifg_list *ifgl;
1245         struct ifg_req *ifgrq, *p;
1246         int len, error;
1247
1248         len = 0;
1249         ifgroup_lockmgr(LK_SHARED);
1250         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next)
1251                 len += sizeof(struct ifg_req);
1252         ifgroup_lockmgr(LK_RELEASE);
1253
1254         if (ifgr->ifgr_len == 0) {
1255                 /*
1256                  * Caller is asking how much memory should be allocated in
1257                  * the next request in order to hold all the groups.
1258                  */
1259                 ifgr->ifgr_len = len;
1260                 return (0);
1261         } else if (ifgr->ifgr_len != len) {
1262                 return (EINVAL);
1263         }
1264
1265         ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1266         if (ifgrq == NULL)
1267                 return (ENOMEM);
1268
1269         ifgroup_lockmgr(LK_SHARED);
1270         p = ifgrq;
1271         TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
1272                 if (len < sizeof(struct ifg_req)) {
1273                         ifgroup_lockmgr(LK_RELEASE);
1274                         return (EINVAL);
1275                 }
1276
1277                 strlcpy(p->ifgrq_group, ifgl->ifgl_group->ifg_group,
1278                         sizeof(ifgrq->ifgrq_group));
1279                 len -= sizeof(struct ifg_req);
1280                 p++;
1281         }
1282         ifgroup_lockmgr(LK_RELEASE);
1283
1284         error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1285         kfree(ifgrq, M_TEMP);
1286         if (error)
1287                 return (error);
1288
1289         return (0);
1290 }
1291
1292 /*
1293  * Store all the members of a group in memory pointed to by data.
1294  */
1295 static int
1296 if_getgroupmembers(struct ifgroupreq *ifgr)
1297 {
1298         struct ifg_group *ifg;
1299         struct ifg_member *ifgm;
1300         struct ifg_req *ifgrq, *p;
1301         int len, error;
1302
1303         ifgroup_lockmgr(LK_SHARED);
1304
1305         TAILQ_FOREACH(ifg, &ifg_head, ifg_next) {
1306                 if (strcmp(ifg->ifg_group, ifgr->ifgr_name) == 0)
1307                         break;
1308         }
1309         if (ifg == NULL) {
1310                 ifgroup_lockmgr(LK_RELEASE);
1311                 return (ENOENT);
1312         }
1313
1314         len = 0;
1315         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next)
1316                 len += sizeof(struct ifg_req);
1317
1318         ifgroup_lockmgr(LK_RELEASE);
1319
1320         if (ifgr->ifgr_len == 0) {
1321                 ifgr->ifgr_len = len;
1322                 return (0);
1323         } else if (ifgr->ifgr_len != len) {
1324                 return (EINVAL);
1325         }
1326
1327         ifgrq = kmalloc(len, M_TEMP, M_INTWAIT | M_NULLOK | M_ZERO);
1328         if (ifgrq == NULL)
1329                 return (ENOMEM);
1330
1331         ifgroup_lockmgr(LK_SHARED);
1332         p = ifgrq;
1333         TAILQ_FOREACH(ifgm, &ifg->ifg_members, ifgm_next) {
1334                 if (len < sizeof(struct ifg_req)) {
1335                         ifgroup_lockmgr(LK_RELEASE);
1336                         return (EINVAL);
1337                 }
1338
1339                 strlcpy(p->ifgrq_member, ifgm->ifgm_ifp->if_xname,
1340                         sizeof(p->ifgrq_member));
1341                 len -= sizeof(struct ifg_req);
1342                 p++;
1343         }
1344         ifgroup_lockmgr(LK_RELEASE);
1345
1346         error = copyout(ifgrq, ifgr->ifgr_groups, ifgr->ifgr_len);
1347         kfree(ifgrq, M_TEMP);
1348         if (error)
1349                 return (error);
1350
1351         return (0);
1352 }
1353
1354 /*
1355  * Delete Routes for a Network Interface
1356  *
1357  * Called for each routing entry via the rnh->rnh_walktree() call above
1358  * to delete all route entries referencing a detaching network interface.
1359  *
1360  * Arguments:
1361  *      rn      pointer to node in the routing table
1362  *      arg     argument passed to rnh->rnh_walktree() - detaching interface
1363  *
1364  * Returns:
1365  *      0       successful
1366  *      errno   failed - reason indicated
1367  *
1368  */
1369 static int
1370 if_rtdel(struct radix_node *rn, void *arg)
1371 {
1372         struct rtentry  *rt = (struct rtentry *)rn;
1373         struct ifnet    *ifp = arg;
1374         int             err;
1375
1376         if (rt->rt_ifp == ifp) {
1377
1378                 /*
1379                  * Protect (sorta) against walktree recursion problems
1380                  * with cloned routes
1381                  */
1382                 if (!(rt->rt_flags & RTF_UP))
1383                         return (0);
1384
1385                 err = rtrequest(RTM_DELETE, rt_key(rt), rt->rt_gateway,
1386                                 rt_mask(rt), rt->rt_flags,
1387                                 NULL);
1388                 if (err) {
1389                         log(LOG_WARNING, "if_rtdel: error %d\n", err);
1390                 }
1391         }
1392
1393         return (0);
1394 }
1395
1396 static __inline boolean_t
1397 ifa_prefer(const struct ifaddr *cur_ifa, const struct ifaddr *old_ifa)
1398 {
1399         if (old_ifa == NULL)
1400                 return TRUE;
1401
1402         if ((old_ifa->ifa_ifp->if_flags & IFF_UP) == 0 &&
1403             (cur_ifa->ifa_ifp->if_flags & IFF_UP))
1404                 return TRUE;
1405         if ((old_ifa->ifa_flags & IFA_ROUTE) == 0 &&
1406             (cur_ifa->ifa_flags & IFA_ROUTE))
1407                 return TRUE;
1408         return FALSE;
1409 }
1410
1411 /*
1412  * Locate an interface based on a complete address.
1413  */
1414 struct ifaddr *
1415 ifa_ifwithaddr(struct sockaddr *addr)
1416 {
1417         const struct ifnet_array *arr;
1418         int i;
1419
1420         arr = ifnet_array_get();
1421         for (i = 0; i < arr->ifnet_count; ++i) {
1422                 struct ifnet *ifp = arr->ifnet_arr[i];
1423                 struct ifaddr_container *ifac;
1424
1425                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1426                         struct ifaddr *ifa = ifac->ifa;
1427
1428                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1429                                 continue;
1430                         if (sa_equal(addr, ifa->ifa_addr))
1431                                 return (ifa);
1432                         if ((ifp->if_flags & IFF_BROADCAST) &&
1433                             ifa->ifa_broadaddr &&
1434                             /* IPv6 doesn't have broadcast */
1435                             ifa->ifa_broadaddr->sa_len != 0 &&
1436                             sa_equal(ifa->ifa_broadaddr, addr))
1437                                 return (ifa);
1438                 }
1439         }
1440         return (NULL);
1441 }
1442
1443 /*
1444  * Locate the point to point interface with a given destination address.
1445  */
1446 struct ifaddr *
1447 ifa_ifwithdstaddr(struct sockaddr *addr)
1448 {
1449         const struct ifnet_array *arr;
1450         int i;
1451
1452         arr = ifnet_array_get();
1453         for (i = 0; i < arr->ifnet_count; ++i) {
1454                 struct ifnet *ifp = arr->ifnet_arr[i];
1455                 struct ifaddr_container *ifac;
1456
1457                 if (!(ifp->if_flags & IFF_POINTOPOINT))
1458                         continue;
1459
1460                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1461                         struct ifaddr *ifa = ifac->ifa;
1462
1463                         if (ifa->ifa_addr->sa_family != addr->sa_family)
1464                                 continue;
1465                         if (ifa->ifa_dstaddr &&
1466                             sa_equal(addr, ifa->ifa_dstaddr))
1467                                 return (ifa);
1468                 }
1469         }
1470         return (NULL);
1471 }
1472
1473 /*
1474  * Find an interface on a specific network.  If many, choice
1475  * is most specific found.
1476  */
1477 struct ifaddr *
1478 ifa_ifwithnet(struct sockaddr *addr)
1479 {
1480         struct ifaddr *ifa_maybe = NULL;
1481         u_int af = addr->sa_family;
1482         char *addr_data = addr->sa_data, *cplim;
1483         const struct ifnet_array *arr;
1484         int i;
1485
1486         /*
1487          * AF_LINK addresses can be looked up directly by their index number,
1488          * so do that if we can.
1489          */
1490         if (af == AF_LINK) {
1491                 struct sockaddr_dl *sdl = (struct sockaddr_dl *)addr;
1492
1493                 if (sdl->sdl_index && sdl->sdl_index <= if_index)
1494                         return (ifindex2ifnet[sdl->sdl_index]->if_lladdr);
1495         }
1496
1497         /*
1498          * Scan though each interface, looking for ones that have
1499          * addresses in this address family.
1500          */
1501         arr = ifnet_array_get();
1502         for (i = 0; i < arr->ifnet_count; ++i) {
1503                 struct ifnet *ifp = arr->ifnet_arr[i];
1504                 struct ifaddr_container *ifac;
1505
1506                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1507                         struct ifaddr *ifa = ifac->ifa;
1508                         char *cp, *cp2, *cp3;
1509
1510                         if (ifa->ifa_addr->sa_family != af)
1511 next:                           continue;
1512                         if (af == AF_INET && ifp->if_flags & IFF_POINTOPOINT) {
1513                                 /*
1514                                  * This is a bit broken as it doesn't
1515                                  * take into account that the remote end may
1516                                  * be a single node in the network we are
1517                                  * looking for.
1518                                  * The trouble is that we don't know the
1519                                  * netmask for the remote end.
1520                                  */
1521                                 if (ifa->ifa_dstaddr != NULL &&
1522                                     sa_equal(addr, ifa->ifa_dstaddr))
1523                                         return (ifa);
1524                         } else {
1525                                 /*
1526                                  * if we have a special address handler,
1527                                  * then use it instead of the generic one.
1528                                  */
1529                                 if (ifa->ifa_claim_addr) {
1530                                         if ((*ifa->ifa_claim_addr)(ifa, addr)) {
1531                                                 return (ifa);
1532                                         } else {
1533                                                 continue;
1534                                         }
1535                                 }
1536
1537                                 /*
1538                                  * Scan all the bits in the ifa's address.
1539                                  * If a bit dissagrees with what we are
1540                                  * looking for, mask it with the netmask
1541                                  * to see if it really matters.
1542                                  * (A byte at a time)
1543                                  */
1544                                 if (ifa->ifa_netmask == 0)
1545                                         continue;
1546                                 cp = addr_data;
1547                                 cp2 = ifa->ifa_addr->sa_data;
1548                                 cp3 = ifa->ifa_netmask->sa_data;
1549                                 cplim = ifa->ifa_netmask->sa_len +
1550                                         (char *)ifa->ifa_netmask;
1551                                 while (cp3 < cplim)
1552                                         if ((*cp++ ^ *cp2++) & *cp3++)
1553                                                 goto next; /* next address! */
1554                                 /*
1555                                  * If the netmask of what we just found
1556                                  * is more specific than what we had before
1557                                  * (if we had one) then remember the new one
1558                                  * before continuing to search for an even
1559                                  * better one.  If the netmasks are equal,
1560                                  * we prefer the this ifa based on the result
1561                                  * of ifa_prefer().
1562                                  */
1563                                 if (ifa_maybe == NULL ||
1564                                     rn_refines((char *)ifa->ifa_netmask,
1565                                         (char *)ifa_maybe->ifa_netmask) ||
1566                                     (sa_equal(ifa_maybe->ifa_netmask,
1567                                         ifa->ifa_netmask) &&
1568                                      ifa_prefer(ifa, ifa_maybe)))
1569                                         ifa_maybe = ifa;
1570                         }
1571                 }
1572         }
1573         return (ifa_maybe);
1574 }
1575
1576 /*
1577  * Find an interface address specific to an interface best matching
1578  * a given address.
1579  */
1580 struct ifaddr *
1581 ifaof_ifpforaddr(struct sockaddr *addr, struct ifnet *ifp)
1582 {
1583         struct ifaddr_container *ifac;
1584         char *cp, *cp2, *cp3;
1585         char *cplim;
1586         struct ifaddr *ifa_maybe = NULL;
1587         u_int af = addr->sa_family;
1588
1589         if (af >= AF_MAX)
1590                 return (0);
1591         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1592                 struct ifaddr *ifa = ifac->ifa;
1593
1594                 if (ifa->ifa_addr->sa_family != af)
1595                         continue;
1596                 if (ifa_maybe == NULL)
1597                         ifa_maybe = ifa;
1598                 if (ifa->ifa_netmask == NULL) {
1599                         if (sa_equal(addr, ifa->ifa_addr) ||
1600                             (ifa->ifa_dstaddr != NULL &&
1601                              sa_equal(addr, ifa->ifa_dstaddr)))
1602                                 return (ifa);
1603                         continue;
1604                 }
1605                 if (ifp->if_flags & IFF_POINTOPOINT) {
1606                         if (sa_equal(addr, ifa->ifa_dstaddr))
1607                                 return (ifa);
1608                 } else {
1609                         cp = addr->sa_data;
1610                         cp2 = ifa->ifa_addr->sa_data;
1611                         cp3 = ifa->ifa_netmask->sa_data;
1612                         cplim = ifa->ifa_netmask->sa_len + (char *)ifa->ifa_netmask;
1613                         for (; cp3 < cplim; cp3++)
1614                                 if ((*cp++ ^ *cp2++) & *cp3)
1615                                         break;
1616                         if (cp3 == cplim)
1617                                 return (ifa);
1618                 }
1619         }
1620         return (ifa_maybe);
1621 }
1622
1623 /*
1624  * Default action when installing a route with a Link Level gateway.
1625  * Lookup an appropriate real ifa to point to.
1626  * This should be moved to /sys/net/link.c eventually.
1627  */
1628 static void
1629 link_rtrequest(int cmd, struct rtentry *rt)
1630 {
1631         struct ifaddr *ifa;
1632         struct sockaddr *dst;
1633         struct ifnet *ifp;
1634
1635         if (cmd != RTM_ADD || (ifa = rt->rt_ifa) == NULL ||
1636             (ifp = ifa->ifa_ifp) == NULL || (dst = rt_key(rt)) == NULL)
1637                 return;
1638         ifa = ifaof_ifpforaddr(dst, ifp);
1639         if (ifa != NULL) {
1640                 IFAFREE(rt->rt_ifa);
1641                 IFAREF(ifa);
1642                 rt->rt_ifa = ifa;
1643                 if (ifa->ifa_rtrequest && ifa->ifa_rtrequest != link_rtrequest)
1644                         ifa->ifa_rtrequest(cmd, rt);
1645         }
1646 }
1647
1648 struct netmsg_ifroute {
1649         struct netmsg_base      base;
1650         struct ifnet            *ifp;
1651         int                     flag;
1652         int                     fam;
1653 };
1654
1655 /*
1656  * Mark an interface down and notify protocols of the transition.
1657  */
1658 static void
1659 if_unroute_dispatch(netmsg_t nmsg)
1660 {
1661         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1662         struct ifnet *ifp = msg->ifp;
1663         int flag = msg->flag, fam = msg->fam;
1664         struct ifaddr_container *ifac;
1665         struct domain *dp;
1666
1667         ASSERT_NETISR0;
1668
1669         ifp->if_flags &= ~flag;
1670         getmicrotime(&ifp->if_lastchange);
1671         rt_ifmsg(ifp);
1672
1673         /*
1674          * The ifaddr processing in the following loop will block,
1675          * however, this function is called in netisr0, in which
1676          * ifaddr list changes happen, so we don't care about the
1677          * blockness of the ifaddr processing here.
1678          */
1679         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1680                 struct ifaddr *ifa = ifac->ifa;
1681
1682                 /* Ignore marker */
1683                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1684                         continue;
1685
1686                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1687                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1688         }
1689
1690         SLIST_FOREACH(dp, &domains, dom_next)
1691                 if (dp->dom_if_down != NULL)
1692                         dp->dom_if_down(ifp);
1693
1694         ifq_purge_all(&ifp->if_snd);
1695         netisr_replymsg(&nmsg->base, 0);
1696 }
1697
1698 static void
1699 if_unroute(struct ifnet *ifp, int flag, int fam)
1700 {
1701         struct netmsg_ifroute msg;
1702
1703         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1704             if_unroute_dispatch);
1705         msg.ifp = ifp;
1706         msg.flag = flag;
1707         msg.fam = fam;
1708         netisr_domsg(&msg.base, 0);
1709 }
1710
1711 /*
1712  * Mark an interface up and notify protocols of the transition.
1713  */
1714 static void
1715 if_route_dispatch(netmsg_t nmsg)
1716 {
1717         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1718         struct ifnet *ifp = msg->ifp;
1719         int flag = msg->flag, fam = msg->fam;
1720         struct ifaddr_container *ifac;
1721         struct domain *dp;
1722
1723         ASSERT_NETISR0;
1724
1725         ifq_purge_all(&ifp->if_snd);
1726         ifp->if_flags |= flag;
1727         getmicrotime(&ifp->if_lastchange);
1728         rt_ifmsg(ifp);
1729
1730         /*
1731          * The ifaddr processing in the following loop will block,
1732          * however, this function is called in netisr0, in which
1733          * ifaddr list changes happen, so we don't care about the
1734          * blockness of the ifaddr processing here.
1735          */
1736         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1737                 struct ifaddr *ifa = ifac->ifa;
1738
1739                 /* Ignore marker */
1740                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1741                         continue;
1742
1743                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1744                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1745         }
1746
1747         SLIST_FOREACH(dp, &domains, dom_next)
1748                 if (dp->dom_if_up != NULL)
1749                         dp->dom_if_up(ifp);
1750 #ifdef INET6
1751         in6_if_up(ifp);
1752 #endif
1753
1754         netisr_replymsg(&nmsg->base, 0);
1755 }
1756
1757 static void
1758 if_route(struct ifnet *ifp, int flag, int fam)
1759 {
1760         struct netmsg_ifroute msg;
1761
1762         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1763             if_route_dispatch);
1764         msg.ifp = ifp;
1765         msg.flag = flag;
1766         msg.fam = fam;
1767         netisr_domsg(&msg.base, 0);
1768 }
1769
1770 /*
1771  * Mark an interface down and notify protocols of the transition.  An
1772  * interface going down is also considered to be a synchronizing event.
1773  * We must ensure that all packet processing related to the interface
1774  * has completed before we return so e.g. the caller can free the ifnet
1775  * structure that the mbufs may be referencing.
1776  *
1777  * NOTE: must be called at splnet or eqivalent.
1778  */
1779 void
1780 if_down(struct ifnet *ifp)
1781 {
1782         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
1783         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1784         netmsg_service_sync();
1785 }
1786
1787 /*
1788  * Mark an interface up and notify protocols of
1789  * the transition.
1790  * NOTE: must be called at splnet or eqivalent.
1791  */
1792 void
1793 if_up(struct ifnet *ifp)
1794 {
1795         if_route(ifp, IFF_UP, AF_UNSPEC);
1796         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
1797 }
1798
1799 /*
1800  * Process a link state change.
1801  * NOTE: must be called at splsoftnet or equivalent.
1802  */
1803 void
1804 if_link_state_change(struct ifnet *ifp)
1805 {
1806         int link_state = ifp->if_link_state;
1807
1808         rt_ifmsg(ifp);
1809         devctl_notify("IFNET", ifp->if_xname,
1810             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1811
1812         EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
1813 }
1814
1815 /*
1816  * Handle interface watchdog timer routines.  Called
1817  * from softclock, we decrement timers (if set) and
1818  * call the appropriate interface routine on expiration.
1819  */
1820 static void
1821 if_slowtimo_dispatch(netmsg_t nmsg)
1822 {
1823         struct globaldata *gd = mycpu;
1824         const struct ifnet_array *arr;
1825         int i;
1826
1827         ASSERT_NETISR0;
1828
1829         crit_enter_gd(gd);
1830         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1831         crit_exit_gd(gd);
1832
1833         arr = ifnet_array_get();
1834         for (i = 0; i < arr->ifnet_count; ++i) {
1835                 struct ifnet *ifp = arr->ifnet_arr[i];
1836
1837                 crit_enter_gd(gd);
1838
1839                 if (if_stats_compat) {
1840                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1841                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1842                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1843                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1844                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1845                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1846                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1847                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1848                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1849                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1850                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1851                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1852                 }
1853
1854                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1855                         crit_exit_gd(gd);
1856                         continue;
1857                 }
1858                 if (ifp->if_watchdog) {
1859                         if (ifnet_tryserialize_all(ifp)) {
1860                                 (*ifp->if_watchdog)(ifp);
1861                                 ifnet_deserialize_all(ifp);
1862                         } else {
1863                                 /* try again next timeout */
1864                                 ++ifp->if_timer;
1865                         }
1866                 }
1867
1868                 crit_exit_gd(gd);
1869         }
1870
1871         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1872 }
1873
1874 static void
1875 if_slowtimo(void *arg __unused)
1876 {
1877         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1878
1879         KASSERT(mycpuid == 0, ("not on cpu0"));
1880         crit_enter();
1881         if (lmsg->ms_flags & MSGF_DONE)
1882                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1883         crit_exit();
1884 }
1885
1886 /*
1887  * Map interface name to
1888  * interface structure pointer.
1889  */
1890 struct ifnet *
1891 ifunit(const char *name)
1892 {
1893         struct ifnet *ifp;
1894
1895         /*
1896          * Search all the interfaces for this name/number
1897          */
1898         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1899
1900         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1901                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1902                         break;
1903         }
1904         return (ifp);
1905 }
1906
1907 struct ifnet *
1908 ifunit_netisr(const char *name)
1909 {
1910         const struct ifnet_array *arr;
1911         int i;
1912
1913         /*
1914          * Search all the interfaces for this name/number
1915          */
1916
1917         arr = ifnet_array_get();
1918         for (i = 0; i < arr->ifnet_count; ++i) {
1919                 struct ifnet *ifp = arr->ifnet_arr[i];
1920
1921                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1922                         return ifp;
1923         }
1924         return NULL;
1925 }
1926
1927 /*
1928  * Interface ioctls.
1929  */
1930 int
1931 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1932 {
1933         struct ifnet *ifp;
1934         struct ifgroupreq *ifgr;
1935         struct ifreq *ifr;
1936         struct ifstat *ifs;
1937         int error, do_ifup = 0;
1938         short oif_flags;
1939         int new_flags;
1940         size_t namelen, onamelen;
1941         char new_name[IFNAMSIZ];
1942         struct ifaddr *ifa;
1943         struct sockaddr_dl *sdl;
1944
1945         switch (cmd) {
1946         case SIOCGIFCONF:
1947                 return (ifconf(cmd, data, cred));
1948         default:
1949                 break;
1950         }
1951
1952         ifr = (struct ifreq *)data;
1953
1954         switch (cmd) {
1955         case SIOCIFCREATE:
1956         case SIOCIFCREATE2:
1957                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1958                         return (error);
1959                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1960                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1961         case SIOCIFDESTROY:
1962                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1963                         return (error);
1964                 return (if_clone_destroy(ifr->ifr_name));
1965         case SIOCIFGCLONERS:
1966                 return (if_clone_list((struct if_clonereq *)data));
1967         case SIOCGIFGMEMB:
1968                 return (if_getgroupmembers((struct ifgroupreq *)data));
1969         default:
1970                 break;
1971         }
1972
1973         /*
1974          * Nominal ioctl through interface, lookup the ifp and obtain a
1975          * lock to serialize the ifconfig ioctl operation.
1976          */
1977         ifnet_lock();
1978
1979         ifp = ifunit(ifr->ifr_name);
1980         if (ifp == NULL) {
1981                 ifnet_unlock();
1982                 return (ENXIO);
1983         }
1984         error = 0;
1985
1986         switch (cmd) {
1987         case SIOCGIFINDEX:
1988                 ifr->ifr_index = ifp->if_index;
1989                 break;
1990
1991         case SIOCGIFFLAGS:
1992                 ifr->ifr_flags = ifp->if_flags;
1993                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1994                 break;
1995
1996         case SIOCGIFCAP:
1997                 ifr->ifr_reqcap = ifp->if_capabilities;
1998                 ifr->ifr_curcap = ifp->if_capenable;
1999                 break;
2000
2001         case SIOCGIFMETRIC:
2002                 ifr->ifr_metric = ifp->if_metric;
2003                 break;
2004
2005         case SIOCGIFMTU:
2006                 ifr->ifr_mtu = ifp->if_mtu;
2007                 break;
2008
2009         case SIOCGIFTSOLEN:
2010                 ifr->ifr_tsolen = ifp->if_tsolen;
2011                 break;
2012
2013         case SIOCGIFDATA:
2014                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
2015                                 sizeof(ifp->if_data));
2016                 break;
2017
2018         case SIOCGIFPHYS:
2019                 ifr->ifr_phys = ifp->if_physical;
2020                 break;
2021
2022         case SIOCGIFPOLLCPU:
2023                 ifr->ifr_pollcpu = -1;
2024                 break;
2025
2026         case SIOCSIFPOLLCPU:
2027                 break;
2028
2029         case SIOCSIFFLAGS:
2030                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2031                 if (error)
2032                         break;
2033                 new_flags = (ifr->ifr_flags & 0xffff) |
2034                     (ifr->ifr_flagshigh << 16);
2035                 if (ifp->if_flags & IFF_SMART) {
2036                         /* Smart drivers twiddle their own routes */
2037                 } else if (ifp->if_flags & IFF_UP &&
2038                     (new_flags & IFF_UP) == 0) {
2039                         if_down(ifp);
2040                 } else if (new_flags & IFF_UP &&
2041                     (ifp->if_flags & IFF_UP) == 0) {
2042                         do_ifup = 1;
2043                 }
2044
2045 #ifdef IFPOLL_ENABLE
2046                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
2047                         if (new_flags & IFF_NPOLLING)
2048                                 ifpoll_register(ifp);
2049                         else
2050                                 ifpoll_deregister(ifp);
2051                 }
2052 #endif
2053
2054                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2055                         (new_flags &~ IFF_CANTCHANGE);
2056                 if (new_flags & IFF_PPROMISC) {
2057                         /* Permanently promiscuous mode requested */
2058                         ifp->if_flags |= IFF_PROMISC;
2059                 } else if (ifp->if_pcount == 0) {
2060                         ifp->if_flags &= ~IFF_PROMISC;
2061                 }
2062                 if (ifp->if_ioctl) {
2063                         ifnet_serialize_all(ifp);
2064                         ifp->if_ioctl(ifp, cmd, data, cred);
2065                         ifnet_deserialize_all(ifp);
2066                 }
2067                 if (do_ifup)
2068                         if_up(ifp);
2069                 getmicrotime(&ifp->if_lastchange);
2070                 break;
2071
2072         case SIOCSIFCAP:
2073                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2074                 if (error)
2075                         break;
2076                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
2077                         error = EINVAL;
2078                         break;
2079                 }
2080                 ifnet_serialize_all(ifp);
2081                 ifp->if_ioctl(ifp, cmd, data, cred);
2082                 ifnet_deserialize_all(ifp);
2083                 break;
2084
2085         case SIOCSIFNAME:
2086                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2087                 if (error)
2088                         break;
2089                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2090                 if (error)
2091                         break;
2092                 if (new_name[0] == '\0') {
2093                         error = EINVAL;
2094                         break;
2095                 }
2096                 if (ifunit(new_name) != NULL) {
2097                         error = EEXIST;
2098                         break;
2099                 }
2100
2101                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
2102
2103                 /* Announce the departure of the interface. */
2104                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2105
2106                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2107                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2108                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2109                 namelen = strlen(new_name);
2110                 onamelen = sdl->sdl_nlen;
2111                 /*
2112                  * Move the address if needed.  This is safe because we
2113                  * allocate space for a name of length IFNAMSIZ when we
2114                  * create this in if_attach().
2115                  */
2116                 if (namelen != onamelen) {
2117                         bcopy(sdl->sdl_data + onamelen,
2118                             sdl->sdl_data + namelen, sdl->sdl_alen);
2119                 }
2120                 bcopy(new_name, sdl->sdl_data, namelen);
2121                 sdl->sdl_nlen = namelen;
2122                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2123                 bzero(sdl->sdl_data, onamelen);
2124                 while (namelen != 0)
2125                         sdl->sdl_data[--namelen] = 0xff;
2126
2127                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2128
2129                 /* Announce the return of the interface. */
2130                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2131                 break;
2132
2133         case SIOCSIFMETRIC:
2134                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2135                 if (error)
2136                         break;
2137                 ifp->if_metric = ifr->ifr_metric;
2138                 getmicrotime(&ifp->if_lastchange);
2139                 break;
2140
2141         case SIOCSIFPHYS:
2142                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2143                 if (error)
2144                         break;
2145                 if (ifp->if_ioctl == NULL) {
2146                         error = EOPNOTSUPP;
2147                         break;
2148                 }
2149                 ifnet_serialize_all(ifp);
2150                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2151                 ifnet_deserialize_all(ifp);
2152                 if (error == 0)
2153                         getmicrotime(&ifp->if_lastchange);
2154                 break;
2155
2156         case SIOCSIFMTU:
2157         {
2158                 u_long oldmtu = ifp->if_mtu;
2159
2160                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2161                 if (error)
2162                         break;
2163                 if (ifp->if_ioctl == NULL) {
2164                         error = EOPNOTSUPP;
2165                         break;
2166                 }
2167                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2168                         error = EINVAL;
2169                         break;
2170                 }
2171                 ifnet_serialize_all(ifp);
2172                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2173                 ifnet_deserialize_all(ifp);
2174                 if (error == 0) {
2175                         getmicrotime(&ifp->if_lastchange);
2176                         rt_ifmsg(ifp);
2177                 }
2178                 /*
2179                  * If the link MTU changed, do network layer specific procedure.
2180                  */
2181                 if (ifp->if_mtu != oldmtu) {
2182 #ifdef INET6
2183                         nd6_setmtu(ifp);
2184 #endif
2185                 }
2186                 break;
2187         }
2188
2189         case SIOCSIFTSOLEN:
2190                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2191                 if (error)
2192                         break;
2193
2194                 /* XXX need driver supplied upper limit */
2195                 if (ifr->ifr_tsolen <= 0) {
2196                         error = EINVAL;
2197                         break;
2198                 }
2199                 ifp->if_tsolen = ifr->ifr_tsolen;
2200                 break;
2201
2202         case SIOCADDMULTI:
2203         case SIOCDELMULTI:
2204                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2205                 if (error)
2206                         break;
2207
2208                 /* Don't allow group membership on non-multicast interfaces. */
2209                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2210                         error = EOPNOTSUPP;
2211                         break;
2212                 }
2213
2214                 /* Don't let users screw up protocols' entries. */
2215                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2216                         error = EINVAL;
2217                         break;
2218                 }
2219
2220                 if (cmd == SIOCADDMULTI) {
2221                         struct ifmultiaddr *ifma;
2222                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2223                 } else {
2224                         error = if_delmulti(ifp, &ifr->ifr_addr);
2225                 }
2226                 if (error == 0)
2227                         getmicrotime(&ifp->if_lastchange);
2228                 break;
2229
2230         case SIOCSIFPHYADDR:
2231         case SIOCDIFPHYADDR:
2232 #ifdef INET6
2233         case SIOCSIFPHYADDR_IN6:
2234 #endif
2235         case SIOCSLIFPHYADDR:
2236         case SIOCSIFMEDIA:
2237         case SIOCSIFGENERIC:
2238                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2239                 if (error)
2240                         break;
2241                 if (ifp->if_ioctl == NULL) {
2242                         error = EOPNOTSUPP;
2243                         break;
2244                 }
2245                 ifnet_serialize_all(ifp);
2246                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2247                 ifnet_deserialize_all(ifp);
2248                 if (error == 0)
2249                         getmicrotime(&ifp->if_lastchange);
2250                 break;
2251
2252         case SIOCGIFSTATUS:
2253                 ifs = (struct ifstat *)data;
2254                 ifs->ascii[0] = '\0';
2255                 /* fall through */
2256         case SIOCGIFPSRCADDR:
2257         case SIOCGIFPDSTADDR:
2258         case SIOCGLIFPHYADDR:
2259         case SIOCGIFMEDIA:
2260         case SIOCGIFGENERIC:
2261                 if (ifp->if_ioctl == NULL) {
2262                         error = EOPNOTSUPP;
2263                         break;
2264                 }
2265                 ifnet_serialize_all(ifp);
2266                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2267                 ifnet_deserialize_all(ifp);
2268                 break;
2269
2270         case SIOCSIFLLADDR:
2271                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2272                 if (error)
2273                         break;
2274                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2275                                      ifr->ifr_addr.sa_len);
2276                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2277                 break;
2278
2279         case SIOCAIFGROUP:
2280                 ifgr = (struct ifgroupreq *)ifr;
2281                 if ((error = priv_check_cred(cred, PRIV_NET_ADDIFGROUP, 0)))
2282                         return (error);
2283                 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2284                         return (error);
2285                 break;
2286
2287         case SIOCDIFGROUP:
2288                 ifgr = (struct ifgroupreq *)ifr;
2289                 if ((error = priv_check_cred(cred, PRIV_NET_DELIFGROUP, 0)))
2290                         return (error);
2291                 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2292                         return (error);
2293                 break;
2294
2295         case SIOCGIFGROUP:
2296                 ifgr = (struct ifgroupreq *)ifr;
2297                 if ((error = if_getgroups(ifgr, ifp)))
2298                         return (error);
2299                 break;
2300
2301         default:
2302                 oif_flags = ifp->if_flags;
2303                 if (so->so_proto == 0) {
2304                         error = EOPNOTSUPP;
2305                         break;
2306                 }
2307                 error = so_pru_control_direct(so, cmd, data, ifp);
2308
2309                 /*
2310                  * If the socket control method returns EOPNOTSUPP, pass the
2311                  * request directly to the interface.
2312                  *
2313                  * Exclude the SIOCSIF{ADDR,BRDADDR,DSTADDR,NETMASK} ioctls,
2314                  * because drivers may trust these ioctls to come from an
2315                  * already privileged layer and thus do not perform credentials
2316                  * checks or input validation.
2317                  */
2318                 if (error == EOPNOTSUPP &&
2319                     ifp->if_ioctl != NULL &&
2320                     cmd != SIOCSIFADDR &&
2321                     cmd != SIOCSIFBRDADDR &&
2322                     cmd != SIOCSIFDSTADDR &&
2323                     cmd != SIOCSIFNETMASK) {
2324                         ifnet_serialize_all(ifp);
2325                         error = ifp->if_ioctl(ifp, cmd, data, cred);
2326                         ifnet_deserialize_all(ifp);
2327                 }
2328
2329                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2330 #ifdef INET6
2331                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2332                         if (ifp->if_flags & IFF_UP) {
2333                                 crit_enter();
2334                                 in6_if_up(ifp);
2335                                 crit_exit();
2336                         }
2337 #endif
2338                 }
2339                 break;
2340         }
2341
2342         ifnet_unlock();
2343         return (error);
2344 }
2345
2346 /*
2347  * Set/clear promiscuous mode on interface ifp based on the truth value
2348  * of pswitch.  The calls are reference counted so that only the first
2349  * "on" request actually has an effect, as does the final "off" request.
2350  * Results are undefined if the "off" and "on" requests are not matched.
2351  */
2352 int
2353 ifpromisc(struct ifnet *ifp, int pswitch)
2354 {
2355         struct ifreq ifr;
2356         int error;
2357         int oldflags;
2358
2359         oldflags = ifp->if_flags;
2360         if (ifp->if_flags & IFF_PPROMISC) {
2361                 /* Do nothing if device is in permanently promiscuous mode */
2362                 ifp->if_pcount += pswitch ? 1 : -1;
2363                 return (0);
2364         }
2365         if (pswitch) {
2366                 /*
2367                  * If the device is not configured up, we cannot put it in
2368                  * promiscuous mode.
2369                  */
2370                 if ((ifp->if_flags & IFF_UP) == 0)
2371                         return (ENETDOWN);
2372                 if (ifp->if_pcount++ != 0)
2373                         return (0);
2374                 ifp->if_flags |= IFF_PROMISC;
2375                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2376                     ifp->if_xname);
2377         } else {
2378                 if (--ifp->if_pcount > 0)
2379                         return (0);
2380                 ifp->if_flags &= ~IFF_PROMISC;
2381                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2382                     ifp->if_xname);
2383         }
2384         ifr.ifr_flags = ifp->if_flags;
2385         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2386         ifnet_serialize_all(ifp);
2387         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2388         ifnet_deserialize_all(ifp);
2389         if (error == 0)
2390                 rt_ifmsg(ifp);
2391         else
2392                 ifp->if_flags = oldflags;
2393         return error;
2394 }
2395
2396 /*
2397  * Return interface configuration
2398  * of system.  List may be used
2399  * in later ioctl's (above) to get
2400  * other information.
2401  */
2402 static int
2403 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2404 {
2405         struct ifconf *ifc = (struct ifconf *)data;
2406         struct ifnet *ifp;
2407         struct sockaddr *sa;
2408         struct ifreq ifr, *ifrp;
2409         int space = ifc->ifc_len, error = 0;
2410
2411         ifrp = ifc->ifc_req;
2412
2413         ifnet_lock();
2414         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2415                 struct ifaddr_container *ifac, *ifac_mark;
2416                 struct ifaddr_marker mark;
2417                 struct ifaddrhead *head;
2418                 int addrs;
2419
2420                 if (space <= sizeof ifr)
2421                         break;
2422
2423                 /*
2424                  * Zero the stack declared structure first to prevent
2425                  * memory disclosure.
2426                  */
2427                 bzero(&ifr, sizeof(ifr));
2428                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2429                     >= sizeof(ifr.ifr_name)) {
2430                         error = ENAMETOOLONG;
2431                         break;
2432                 }
2433
2434                 /*
2435                  * Add a marker, since copyout() could block and during that
2436                  * period the list could be changed.  Inserting the marker to
2437                  * the header of the list will not cause trouble for the code
2438                  * assuming that the first element of the list is AF_LINK; the
2439                  * marker will be moved to the next position w/o blocking.
2440                  */
2441                 ifa_marker_init(&mark, ifp);
2442                 ifac_mark = &mark.ifac;
2443                 head = &ifp->if_addrheads[mycpuid];
2444
2445                 addrs = 0;
2446                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2447                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2448                         struct ifaddr *ifa = ifac->ifa;
2449
2450                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2451                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2452
2453                         /* Ignore marker */
2454                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2455                                 continue;
2456
2457                         if (space <= sizeof ifr)
2458                                 break;
2459                         sa = ifa->ifa_addr;
2460                         if (cred->cr_prison &&
2461                             prison_if(cred, sa))
2462                                 continue;
2463                         addrs++;
2464                         /*
2465                          * Keep a reference on this ifaddr, so that it will
2466                          * not be destroyed when its address is copied to
2467                          * the userland, which could block.
2468                          */
2469                         IFAREF(ifa);
2470                         if (sa->sa_len <= sizeof(*sa)) {
2471                                 ifr.ifr_addr = *sa;
2472                                 error = copyout(&ifr, ifrp, sizeof ifr);
2473                                 ifrp++;
2474                         } else {
2475                                 if (space < (sizeof ifr) + sa->sa_len -
2476                                             sizeof(*sa)) {
2477                                         IFAFREE(ifa);
2478                                         break;
2479                                 }
2480                                 space -= sa->sa_len - sizeof(*sa);
2481                                 error = copyout(&ifr, ifrp,
2482                                                 sizeof ifr.ifr_name);
2483                                 if (error == 0)
2484                                         error = copyout(sa, &ifrp->ifr_addr,
2485                                                         sa->sa_len);
2486                                 ifrp = (struct ifreq *)
2487                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2488                         }
2489                         IFAFREE(ifa);
2490                         if (error)
2491                                 break;
2492                         space -= sizeof ifr;
2493                 }
2494                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2495                 if (error)
2496                         break;
2497                 if (!addrs) {
2498                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2499                         error = copyout(&ifr, ifrp, sizeof ifr);
2500                         if (error)
2501                                 break;
2502                         space -= sizeof ifr;
2503                         ifrp++;
2504                 }
2505         }
2506         ifnet_unlock();
2507
2508         ifc->ifc_len -= space;
2509         return (error);
2510 }
2511
2512 /*
2513  * Just like if_promisc(), but for all-multicast-reception mode.
2514  */
2515 int
2516 if_allmulti(struct ifnet *ifp, int onswitch)
2517 {
2518         int error = 0;
2519         struct ifreq ifr;
2520
2521         crit_enter();
2522
2523         if (onswitch) {
2524                 if (ifp->if_amcount++ == 0) {
2525                         ifp->if_flags |= IFF_ALLMULTI;
2526                         ifr.ifr_flags = ifp->if_flags;
2527                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2528                         ifnet_serialize_all(ifp);
2529                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2530                                               NULL);
2531                         ifnet_deserialize_all(ifp);
2532                 }
2533         } else {
2534                 if (ifp->if_amcount > 1) {
2535                         ifp->if_amcount--;
2536                 } else {
2537                         ifp->if_amcount = 0;
2538                         ifp->if_flags &= ~IFF_ALLMULTI;
2539                         ifr.ifr_flags = ifp->if_flags;
2540                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2541                         ifnet_serialize_all(ifp);
2542                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2543                                               NULL);
2544                         ifnet_deserialize_all(ifp);
2545                 }
2546         }
2547
2548         crit_exit();
2549
2550         if (error == 0)
2551                 rt_ifmsg(ifp);
2552         return error;
2553 }
2554
2555 /*
2556  * Add a multicast listenership to the interface in question.
2557  * The link layer provides a routine which converts
2558  */
2559 int
2560 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2561     struct ifmultiaddr **retifma)
2562 {
2563         struct sockaddr *llsa, *dupsa;
2564         int error;
2565         struct ifmultiaddr *ifma;
2566
2567         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2568
2569         /*
2570          * If the matching multicast address already exists
2571          * then don't add a new one, just add a reference
2572          */
2573         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2574                 if (sa_equal(sa, ifma->ifma_addr)) {
2575                         ifma->ifma_refcount++;
2576                         if (retifma)
2577                                 *retifma = ifma;
2578                         return 0;
2579                 }
2580         }
2581
2582         /*
2583          * Give the link layer a chance to accept/reject it, and also
2584          * find out which AF_LINK address this maps to, if it isn't one
2585          * already.
2586          */
2587         if (ifp->if_resolvemulti) {
2588                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2589                 if (error)
2590                         return error;
2591         } else {
2592                 llsa = NULL;
2593         }
2594
2595         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2596         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2597         bcopy(sa, dupsa, sa->sa_len);
2598
2599         ifma->ifma_addr = dupsa;
2600         ifma->ifma_lladdr = llsa;
2601         ifma->ifma_ifp = ifp;
2602         ifma->ifma_refcount = 1;
2603         ifma->ifma_protospec = NULL;
2604         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2605
2606         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2607         if (retifma)
2608                 *retifma = ifma;
2609
2610         if (llsa != NULL) {
2611                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2612                         if (sa_equal(ifma->ifma_addr, llsa))
2613                                 break;
2614                 }
2615                 if (ifma) {
2616                         ifma->ifma_refcount++;
2617                 } else {
2618                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2619                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2620                         bcopy(llsa, dupsa, llsa->sa_len);
2621                         ifma->ifma_addr = dupsa;
2622                         ifma->ifma_ifp = ifp;
2623                         ifma->ifma_refcount = 1;
2624                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2625                 }
2626         }
2627         /*
2628          * We are certain we have added something, so call down to the
2629          * interface to let them know about it.
2630          */
2631         if (ifp->if_ioctl)
2632                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2633
2634         return 0;
2635 }
2636
2637 int
2638 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2639     struct ifmultiaddr **retifma)
2640 {
2641         int error;
2642
2643         ifnet_serialize_all(ifp);
2644         error = if_addmulti_serialized(ifp, sa, retifma);
2645         ifnet_deserialize_all(ifp);
2646
2647         return error;
2648 }
2649
2650 /*
2651  * Remove a reference to a multicast address on this interface.  Yell
2652  * if the request does not match an existing membership.
2653  */
2654 static int
2655 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2656 {
2657         struct ifmultiaddr *ifma;
2658
2659         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2660
2661         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2662                 if (sa_equal(sa, ifma->ifma_addr))
2663                         break;
2664         if (ifma == NULL)
2665                 return ENOENT;
2666
2667         if (ifma->ifma_refcount > 1) {
2668                 ifma->ifma_refcount--;
2669                 return 0;
2670         }
2671
2672         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2673         sa = ifma->ifma_lladdr;
2674         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2675         /*
2676          * Make sure the interface driver is notified
2677          * in the case of a link layer mcast group being left.
2678          */
2679         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2680                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2681         kfree(ifma->ifma_addr, M_IFMADDR);
2682         kfree(ifma, M_IFMADDR);
2683         if (sa == NULL)
2684                 return 0;
2685
2686         /*
2687          * Now look for the link-layer address which corresponds to
2688          * this network address.  It had been squirreled away in
2689          * ifma->ifma_lladdr for this purpose (so we don't have
2690          * to call ifp->if_resolvemulti() again), and we saved that
2691          * value in sa above.  If some nasty deleted the
2692          * link-layer address out from underneath us, we can deal because
2693          * the address we stored was is not the same as the one which was
2694          * in the record for the link-layer address.  (So we don't complain
2695          * in that case.)
2696          */
2697         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2698                 if (sa_equal(sa, ifma->ifma_addr))
2699                         break;
2700         if (ifma == NULL)
2701                 return 0;
2702
2703         if (ifma->ifma_refcount > 1) {
2704                 ifma->ifma_refcount--;
2705                 return 0;
2706         }
2707
2708         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2709         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2710         kfree(ifma->ifma_addr, M_IFMADDR);
2711         kfree(sa, M_IFMADDR);
2712         kfree(ifma, M_IFMADDR);
2713
2714         return 0;
2715 }
2716
2717 int
2718 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2719 {
2720         int error;
2721
2722         ifnet_serialize_all(ifp);
2723         error = if_delmulti_serialized(ifp, sa);
2724         ifnet_deserialize_all(ifp);
2725
2726         return error;
2727 }
2728
2729 /*
2730  * Delete all multicast group membership for an interface.
2731  * Should be used to quickly flush all multicast filters.
2732  */
2733 void
2734 if_delallmulti_serialized(struct ifnet *ifp)
2735 {
2736         struct ifmultiaddr *ifma, mark;
2737         struct sockaddr sa;
2738
2739         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2740
2741         bzero(&sa, sizeof(sa));
2742         sa.sa_family = AF_UNSPEC;
2743         sa.sa_len = sizeof(sa);
2744
2745         bzero(&mark, sizeof(mark));
2746         mark.ifma_addr = &sa;
2747
2748         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2749         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2750                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2751                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2752                     ifma_link);
2753
2754                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2755                         continue;
2756
2757                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2758         }
2759         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2760 }
2761
2762
2763 /*
2764  * Set the link layer address on an interface.
2765  *
2766  * At this time we only support certain types of interfaces,
2767  * and we don't allow the length of the address to change.
2768  */
2769 int
2770 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2771 {
2772         struct sockaddr_dl *sdl;
2773         struct ifreq ifr;
2774
2775         sdl = IF_LLSOCKADDR(ifp);
2776         if (sdl == NULL)
2777                 return (EINVAL);
2778         if (len != sdl->sdl_alen)       /* don't allow length to change */
2779                 return (EINVAL);
2780         switch (ifp->if_type) {
2781         case IFT_ETHER:                 /* these types use struct arpcom */
2782         case IFT_XETHER:
2783         case IFT_L2VLAN:
2784         case IFT_IEEE8023ADLAG:
2785                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2786                 bcopy(lladdr, LLADDR(sdl), len);
2787                 break;
2788         default:
2789                 return (ENODEV);
2790         }
2791         /*
2792          * If the interface is already up, we need
2793          * to re-init it in order to reprogram its
2794          * address filter.
2795          */
2796         ifnet_serialize_all(ifp);
2797         if ((ifp->if_flags & IFF_UP) != 0) {
2798 #ifdef INET
2799                 struct ifaddr_container *ifac;
2800 #endif
2801
2802                 ifp->if_flags &= ~IFF_UP;
2803                 ifr.ifr_flags = ifp->if_flags;
2804                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2805                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2806                               NULL);
2807                 ifp->if_flags |= IFF_UP;
2808                 ifr.ifr_flags = ifp->if_flags;
2809                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2810                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2811                                  NULL);
2812 #ifdef INET
2813                 /*
2814                  * Also send gratuitous ARPs to notify other nodes about
2815                  * the address change.
2816                  */
2817                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2818                         struct ifaddr *ifa = ifac->ifa;
2819
2820                         if (ifa->ifa_addr != NULL &&
2821                             ifa->ifa_addr->sa_family == AF_INET)
2822                                 arp_gratuitous(ifp, ifa);
2823                 }
2824 #endif
2825         }
2826         ifnet_deserialize_all(ifp);
2827         return (0);
2828 }
2829
2830
2831 /*
2832  * Locate an interface based on a complete address.
2833  */
2834 struct ifnet *
2835 if_bylla(const void *lla, unsigned char lla_len)
2836 {
2837         const struct ifnet_array *arr;
2838         struct ifnet *ifp;
2839         struct sockaddr_dl *sdl;
2840         int i;
2841
2842         arr = ifnet_array_get();
2843         for (i = 0; i < arr->ifnet_count; ++i) {
2844                 ifp = arr->ifnet_arr[i];
2845                 if (ifp->if_addrlen != lla_len)
2846                         continue;
2847
2848                 sdl = IF_LLSOCKADDR(ifp);
2849                 if (memcmp(lla, LLADDR(sdl), lla_len) == 0)
2850                         return (ifp);
2851         }
2852         return (NULL);
2853 }
2854
2855 struct ifmultiaddr *
2856 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2857 {
2858         struct ifmultiaddr *ifma;
2859
2860         /* TODO: need ifnet_serialize_main */
2861         ifnet_serialize_all(ifp);
2862         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2863                 if (sa_equal(ifma->ifma_addr, sa))
2864                         break;
2865         ifnet_deserialize_all(ifp);
2866
2867         return ifma;
2868 }
2869
2870 /*
2871  * This function locates the first real ethernet MAC from a network
2872  * card and loads it into node, returning 0 on success or ENOENT if
2873  * no suitable interfaces were found.  It is used by the uuid code to
2874  * generate a unique 6-byte number.
2875  */
2876 int
2877 if_getanyethermac(uint16_t *node, int minlen)
2878 {
2879         struct ifnet *ifp;
2880         struct sockaddr_dl *sdl;
2881
2882         ifnet_lock();
2883         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2884                 if (ifp->if_type != IFT_ETHER)
2885                         continue;
2886                 sdl = IF_LLSOCKADDR(ifp);
2887                 if (sdl->sdl_alen < minlen)
2888                         continue;
2889                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2890                       minlen);
2891                 ifnet_unlock();
2892                 return(0);
2893         }
2894         ifnet_unlock();
2895         return (ENOENT);
2896 }
2897
2898 /*
2899  * The name argument must be a pointer to storage which will last as
2900  * long as the interface does.  For physical devices, the result of
2901  * device_get_name(dev) is a good choice and for pseudo-devices a
2902  * static string works well.
2903  */
2904 void
2905 if_initname(struct ifnet *ifp, const char *name, int unit)
2906 {
2907         ifp->if_dname = name;
2908         ifp->if_dunit = unit;
2909         if (unit != IF_DUNIT_NONE)
2910                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2911         else
2912                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2913 }
2914
2915 int
2916 if_printf(struct ifnet *ifp, const char *fmt, ...)
2917 {
2918         __va_list ap;
2919         int retval;
2920
2921         retval = kprintf("%s: ", ifp->if_xname);
2922         __va_start(ap, fmt);
2923         retval += kvprintf(fmt, ap);
2924         __va_end(ap);
2925         return (retval);
2926 }
2927
2928 struct ifnet *
2929 if_alloc(uint8_t type)
2930 {
2931         struct ifnet *ifp;
2932         size_t size;
2933
2934         /*
2935          * XXX temporary hack until arpcom is setup in if_l2com
2936          */
2937         if (type == IFT_ETHER)
2938                 size = sizeof(struct arpcom);
2939         else
2940                 size = sizeof(struct ifnet);
2941
2942         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2943
2944         ifp->if_type = type;
2945
2946         if (if_com_alloc[type] != NULL) {
2947                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2948                 if (ifp->if_l2com == NULL) {
2949                         kfree(ifp, M_IFNET);
2950                         return (NULL);
2951                 }
2952         }
2953         return (ifp);
2954 }
2955
2956 void
2957 if_free(struct ifnet *ifp)
2958 {
2959         kfree(ifp, M_IFNET);
2960 }
2961
2962 void
2963 ifq_set_classic(struct ifaltq *ifq)
2964 {
2965         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2966             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2967 }
2968
2969 void
2970 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2971     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2972 {
2973         int q;
2974
2975         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2976         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2977         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2978         KASSERT(request != NULL, ("request is not specified"));
2979
2980         ifq->altq_mapsubq = mapsubq;
2981         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2982                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2983
2984                 ifsq->ifsq_enqueue = enqueue;
2985                 ifsq->ifsq_dequeue = dequeue;
2986                 ifsq->ifsq_request = request;
2987         }
2988 }
2989
2990 static void
2991 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2992 {
2993
2994         classq_add(&ifsq->ifsq_norm, m);
2995         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2996 }
2997
2998 static void
2999 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
3000 {
3001
3002         classq_add(&ifsq->ifsq_prio, m);
3003         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
3004         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
3005 }
3006
3007 static struct mbuf *
3008 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
3009 {
3010         struct mbuf *m;
3011
3012         m = classq_get(&ifsq->ifsq_norm);
3013         if (m != NULL)
3014                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
3015         return (m);
3016 }
3017
3018 static struct mbuf *
3019 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
3020 {
3021         struct mbuf *m;
3022
3023         m = classq_get(&ifsq->ifsq_prio);
3024         if (m != NULL) {
3025                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
3026                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
3027         }
3028         return (m);
3029 }
3030
3031 int
3032 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
3033     struct altq_pktattr *pa __unused)
3034 {
3035
3036         M_ASSERTPKTHDR(m);
3037 again:
3038         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
3039             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
3040                 struct mbuf *m_drop;
3041
3042                 if (m->m_flags & M_PRIO) {
3043                         m_drop = NULL;
3044                         if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
3045                             ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
3046                                 /* Try dropping some from normal queue. */
3047                                 m_drop = ifsq_norm_dequeue(ifsq);
3048                         }
3049                         if (m_drop == NULL)
3050                                 m_drop = ifsq_prio_dequeue(ifsq);
3051                 } else {
3052                         m_drop = ifsq_norm_dequeue(ifsq);
3053                 }
3054                 if (m_drop != NULL) {
3055                         IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
3056                         m_freem(m_drop);
3057                         goto again;
3058                 }
3059                 /*
3060                  * No old packets could be dropped!
3061                  * NOTE: Caller increases oqdrops.
3062                  */
3063                 m_freem(m);
3064                 return (ENOBUFS);
3065         } else {
3066                 if (m->m_flags & M_PRIO)
3067                         ifsq_prio_enqueue(ifsq, m);
3068                 else
3069                         ifsq_norm_enqueue(ifsq, m);
3070                 return (0);
3071         }
3072 }
3073
3074 struct mbuf *
3075 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
3076 {
3077         struct mbuf *m;
3078
3079         switch (op) {
3080         case ALTDQ_POLL:
3081                 m = classq_head(&ifsq->ifsq_prio);
3082                 if (m == NULL)
3083                         m = classq_head(&ifsq->ifsq_norm);
3084                 break;
3085
3086         case ALTDQ_REMOVE:
3087                 m = ifsq_prio_dequeue(ifsq);
3088                 if (m == NULL)
3089                         m = ifsq_norm_dequeue(ifsq);
3090                 break;
3091
3092         default:
3093                 panic("unsupported ALTQ dequeue op: %d", op);
3094         }
3095         return m;
3096 }
3097
3098 int
3099 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
3100 {
3101         switch (req) {
3102         case ALTRQ_PURGE:
3103                 for (;;) {
3104                         struct mbuf *m;
3105
3106                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
3107                         if (m == NULL)
3108                                 break;
3109                         m_freem(m);
3110                 }
3111                 break;
3112
3113         default:
3114                 panic("unsupported ALTQ request: %d", req);
3115         }
3116         return 0;
3117 }
3118
3119 static void
3120 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
3121 {
3122         struct ifnet *ifp = ifsq_get_ifp(ifsq);
3123         int running = 0, need_sched;
3124
3125         /*
3126          * Try to do direct ifnet.if_start on the subqueue first, if there is
3127          * contention on the subqueue hardware serializer, ifnet.if_start on
3128          * the subqueue will be scheduled on the subqueue owner CPU.
3129          */
3130         if (!ifsq_tryserialize_hw(ifsq)) {
3131                 /*
3132                  * Subqueue hardware serializer contention happened,
3133                  * ifnet.if_start on the subqueue is scheduled on
3134                  * the subqueue owner CPU, and we keep going.
3135                  */
3136                 ifsq_ifstart_schedule(ifsq, 1);
3137                 return;
3138         }
3139
3140         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
3141                 ifp->if_start(ifp, ifsq);
3142                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
3143                         running = 1;
3144         }
3145         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
3146
3147         ifsq_deserialize_hw(ifsq);
3148
3149         if (need_sched) {
3150                 /*
3151                  * More data need to be transmitted, ifnet.if_start on the
3152                  * subqueue is scheduled on the subqueue owner CPU, and we
3153                  * keep going.
3154                  * NOTE: ifnet.if_start subqueue interlock is not released.
3155                  */
3156                 ifsq_ifstart_schedule(ifsq, force_sched);
3157         }
3158 }
3159
3160 /*
3161  * Subqeue packets staging mechanism:
3162  *
3163  * The packets enqueued into the subqueue are staged to a certain amount
3164  * before the ifnet.if_start on the subqueue is called.  In this way, the
3165  * driver could avoid writing to hardware registers upon every packet,
3166  * instead, hardware registers could be written when certain amount of
3167  * packets are put onto hardware TX ring.  The measurement on several modern
3168  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3169  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3170  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3171  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3172  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3173  *
3174  * Subqueue packets staging is performed for two entry points into drivers'
3175  * transmission function:
3176  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3177  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3178  *
3179  * Subqueue packets staging will be stopped upon any of the following
3180  * conditions:
3181  * - If the count of packets enqueued on the current CPU is great than or
3182  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3183  * - If the total length of packets enqueued on the current CPU is great
3184  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3185  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3186  *   is usually less than hardware's MTU.
3187  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3188  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3189  *   released.
3190  * - The if_start_rollup(), which is registered as low priority netisr
3191  *   rollup function, is called; probably because no more work is pending
3192  *   for netisr.
3193  *
3194  * NOTE:
3195  * Currently subqueue packet staging is only performed in netisr threads.
3196  */
3197 int
3198 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3199 {
3200         struct ifaltq *ifq = &ifp->if_snd;
3201         struct ifaltq_subque *ifsq;
3202         int error, start = 0, len, mcast = 0, avoid_start = 0;
3203         struct ifsubq_stage_head *head = NULL;
3204         struct ifsubq_stage *stage = NULL;
3205         struct globaldata *gd = mycpu;
3206         struct thread *td = gd->gd_curthread;
3207
3208         crit_enter_quick(td);
3209
3210         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3211         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3212
3213         len = m->m_pkthdr.len;
3214         if (m->m_flags & M_MCAST)
3215                 mcast = 1;
3216
3217         if (td->td_type == TD_TYPE_NETISR) {
3218                 head = &ifsubq_stage_heads[mycpuid];
3219                 stage = ifsq_get_stage(ifsq, mycpuid);
3220
3221                 stage->stg_cnt++;
3222                 stage->stg_len += len;
3223                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3224                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3225                         avoid_start = 1;
3226         }
3227
3228         ALTQ_SQ_LOCK(ifsq);
3229         error = ifsq_enqueue_locked(ifsq, m, pa);
3230         if (error) {
3231                 IFNET_STAT_INC(ifp, oqdrops, 1);
3232                 if (!ifsq_data_ready(ifsq)) {
3233                         ALTQ_SQ_UNLOCK(ifsq);
3234                         crit_exit_quick(td);
3235                         return error;
3236                 }
3237                 avoid_start = 0;
3238         }
3239         if (!ifsq_is_started(ifsq)) {
3240                 if (avoid_start) {
3241                         ALTQ_SQ_UNLOCK(ifsq);
3242
3243                         KKASSERT(!error);
3244                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3245                                 ifsq_stage_insert(head, stage);
3246
3247                         IFNET_STAT_INC(ifp, obytes, len);
3248                         if (mcast)
3249                                 IFNET_STAT_INC(ifp, omcasts, 1);
3250                         crit_exit_quick(td);
3251                         return error;
3252                 }
3253
3254                 /*
3255                  * Hold the subqueue interlock of ifnet.if_start
3256                  */
3257                 ifsq_set_started(ifsq);
3258                 start = 1;
3259         }
3260         ALTQ_SQ_UNLOCK(ifsq);
3261
3262         if (!error) {
3263                 IFNET_STAT_INC(ifp, obytes, len);
3264                 if (mcast)
3265                         IFNET_STAT_INC(ifp, omcasts, 1);
3266         }
3267
3268         if (stage != NULL) {
3269                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3270                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3271                         if (!avoid_start) {
3272                                 ifsq_stage_remove(head, stage);
3273                                 ifsq_ifstart_schedule(ifsq, 1);
3274                         }
3275                         crit_exit_quick(td);
3276                         return error;
3277                 }
3278
3279                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3280                         ifsq_stage_remove(head, stage);
3281                 } else {
3282                         stage->stg_cnt = 0;
3283                         stage->stg_len = 0;
3284                 }
3285         }
3286
3287         if (!start) {
3288                 crit_exit_quick(td);
3289                 return error;
3290         }
3291
3292         ifsq_ifstart_try(ifsq, 0);
3293
3294         crit_exit_quick(td);
3295         return error;
3296 }
3297
3298 void *
3299 ifa_create(int size)
3300 {
3301         struct ifaddr *ifa;
3302         int i;
3303
3304         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3305
3306         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3307
3308         /*
3309          * Make ifa_container availabel on all CPUs, since they
3310          * could be accessed by any threads.
3311          */
3312         ifa->ifa_containers =
3313                 kmalloc(ncpus * sizeof(struct ifaddr_container),
3314                         M_IFADDR,
3315                         M_INTWAIT | M_ZERO | M_CACHEALIGN);
3316
3317         ifa->ifa_ncnt = ncpus;
3318         for (i = 0; i < ncpus; ++i) {
3319                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3320
3321                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3322                 ifac->ifa = ifa;
3323                 ifac->ifa_refcnt = 1;
3324         }
3325 #ifdef IFADDR_DEBUG
3326         kprintf("alloc ifa %p %d\n", ifa, size);
3327 #endif
3328         return ifa;
3329 }
3330
3331 void
3332 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3333 {
3334         struct ifaddr *ifa = ifac->ifa;
3335
3336         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3337         KKASSERT(ifac->ifa_refcnt == 0);
3338         KASSERT(ifac->ifa_listmask == 0,
3339                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3340
3341         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3342
3343 #ifdef IFADDR_DEBUG_VERBOSE
3344         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3345 #endif
3346
3347         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3348                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3349         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3350 #ifdef IFADDR_DEBUG
3351                 kprintf("free ifa %p\n", ifa);
3352 #endif
3353                 kfree(ifa->ifa_containers, M_IFADDR);
3354                 kfree(ifa, M_IFADDR);
3355         }
3356 }
3357
3358 static void
3359 ifa_iflink_dispatch(netmsg_t nmsg)
3360 {
3361         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3362         struct ifaddr *ifa = msg->ifa;
3363         struct ifnet *ifp = msg->ifp;
3364         int cpu = mycpuid;
3365         struct ifaddr_container *ifac;
3366
3367         crit_enter();
3368
3369         ifac = &ifa->ifa_containers[cpu];
3370         ASSERT_IFAC_VALID(ifac);
3371         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3372                 ("ifaddr is on if_addrheads"));
3373
3374         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3375         if (msg->tail)
3376                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3377         else
3378                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3379
3380         crit_exit();
3381
3382         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3383 }
3384
3385 void
3386 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3387 {
3388         struct netmsg_ifaddr msg;
3389
3390         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3391                     0, ifa_iflink_dispatch);
3392         msg.ifa = ifa;
3393         msg.ifp = ifp;
3394         msg.tail = tail;
3395
3396         netisr_domsg(&msg.base, 0);
3397 }
3398
3399 static void
3400 ifa_ifunlink_dispatch(netmsg_t nmsg)
3401 {
3402         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3403         struct ifaddr *ifa = msg->ifa;
3404         struct ifnet *ifp = msg->ifp;
3405         int cpu = mycpuid;
3406         struct ifaddr_container *ifac;
3407
3408         crit_enter();
3409
3410         ifac = &ifa->ifa_containers[cpu];
3411         ASSERT_IFAC_VALID(ifac);
3412         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3413                 ("ifaddr is not on if_addrhead"));
3414
3415         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3416         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3417
3418         crit_exit();
3419
3420         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3421 }
3422
3423 void
3424 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3425 {
3426         struct netmsg_ifaddr msg;
3427
3428         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3429                     0, ifa_ifunlink_dispatch);
3430         msg.ifa = ifa;
3431         msg.ifp = ifp;
3432
3433         netisr_domsg(&msg.base, 0);
3434 }
3435
3436 static void
3437 ifa_destroy_dispatch(netmsg_t nmsg)
3438 {
3439         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3440
3441         IFAFREE(msg->ifa);
3442         netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3443 }
3444
3445 void
3446 ifa_destroy(struct ifaddr *ifa)
3447 {
3448         struct netmsg_ifaddr msg;
3449
3450         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3451                     0, ifa_destroy_dispatch);
3452         msg.ifa = ifa;
3453
3454         netisr_domsg(&msg.base, 0);
3455 }
3456
3457 static void
3458 if_start_rollup(void)
3459 {
3460         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3461         struct ifsubq_stage *stage;
3462
3463         crit_enter();
3464
3465         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3466                 struct ifaltq_subque *ifsq = stage->stg_subq;
3467                 int is_sched = 0;
3468
3469                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3470                         is_sched = 1;
3471                 ifsq_stage_remove(head, stage);
3472
3473                 if (is_sched) {
3474                         ifsq_ifstart_schedule(ifsq, 1);
3475                 } else {
3476                         int start = 0;
3477
3478                         ALTQ_SQ_LOCK(ifsq);
3479                         if (!ifsq_is_started(ifsq)) {
3480                                 /*
3481                                  * Hold the subqueue interlock of
3482                                  * ifnet.if_start
3483                                  */
3484                                 ifsq_set_started(ifsq);
3485                                 start = 1;
3486                         }
3487                         ALTQ_SQ_UNLOCK(ifsq);
3488
3489                         if (start)
3490                                 ifsq_ifstart_try(ifsq, 1);
3491                 }
3492                 KKASSERT((stage->stg_flags &
3493                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3494         }
3495
3496         crit_exit();
3497 }
3498
3499 static void
3500 ifnetinit(void *dummy __unused)
3501 {
3502         int i;
3503
3504         /* XXX netisr_ncpus */
3505         for (i = 0; i < ncpus; ++i)
3506                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3507         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3508 }
3509
3510 void
3511 if_register_com_alloc(u_char type,
3512     if_com_alloc_t *a, if_com_free_t *f)
3513 {
3514
3515         KASSERT(if_com_alloc[type] == NULL,
3516             ("if_register_com_alloc: %d already registered", type));
3517         KASSERT(if_com_free[type] == NULL,
3518             ("if_register_com_alloc: %d free already registered", type));
3519
3520         if_com_alloc[type] = a;
3521         if_com_free[type] = f;
3522 }
3523
3524 void
3525 if_deregister_com_alloc(u_char type)
3526 {
3527
3528         KASSERT(if_com_alloc[type] != NULL,
3529             ("if_deregister_com_alloc: %d not registered", type));
3530         KASSERT(if_com_free[type] != NULL,
3531             ("if_deregister_com_alloc: %d free not registered", type));
3532         if_com_alloc[type] = NULL;
3533         if_com_free[type] = NULL;
3534 }
3535
3536 void
3537 ifq_set_maxlen(struct ifaltq *ifq, int len)
3538 {
3539         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3540 }
3541
3542 int
3543 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3544 {
3545         return ALTQ_SUBQ_INDEX_DEFAULT;
3546 }
3547
3548 int
3549 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3550 {
3551
3552         return (cpuid % ifq->altq_subq_mappriv);
3553 }
3554
3555 static void
3556 ifsq_watchdog(void *arg)
3557 {
3558         struct ifsubq_watchdog *wd = arg;
3559         struct ifnet *ifp;
3560
3561         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3562                 goto done;
3563
3564         ifp = ifsq_get_ifp(wd->wd_subq);
3565         if (ifnet_tryserialize_all(ifp)) {
3566                 wd->wd_watchdog(wd->wd_subq);
3567                 ifnet_deserialize_all(ifp);
3568         } else {
3569                 /* try again next timeout */
3570                 wd->wd_timer = 1;
3571         }
3572 done:
3573         ifsq_watchdog_reset(wd);
3574 }
3575
3576 static void
3577 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3578 {
3579         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3580             ifsq_get_cpuid(wd->wd_subq));
3581 }
3582
3583 void
3584 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3585     ifsq_watchdog_t watchdog)
3586 {
3587         callout_init_mp(&wd->wd_callout);
3588         wd->wd_timer = 0;
3589         wd->wd_subq = ifsq;
3590         wd->wd_watchdog = watchdog;
3591 }
3592
3593 void
3594 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3595 {
3596         wd->wd_timer = 0;
3597         ifsq_watchdog_reset(wd);
3598 }
3599
3600 void
3601 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3602 {
3603         wd->wd_timer = 0;
3604         callout_stop(&wd->wd_callout);
3605 }
3606
3607 void
3608 ifnet_lock(void)
3609 {
3610         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3611             ("try holding ifnet lock in netisr"));
3612         mtx_lock(&ifnet_mtx);
3613 }
3614
3615 void
3616 ifnet_unlock(void)
3617 {
3618         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3619             ("try holding ifnet lock in netisr"));
3620         mtx_unlock(&ifnet_mtx);
3621 }
3622
3623 static struct ifnet_array *
3624 ifnet_array_alloc(int count)
3625 {
3626         struct ifnet_array *arr;
3627
3628         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3629             M_IFNET, M_WAITOK);
3630         arr->ifnet_count = count;
3631
3632         return arr;
3633 }
3634
3635 static void
3636 ifnet_array_free(struct ifnet_array *arr)
3637 {
3638         if (arr == &ifnet_array0)
3639                 return;
3640         kfree(arr, M_IFNET);
3641 }
3642
3643 static struct ifnet_array *
3644 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3645 {
3646         struct ifnet_array *arr;
3647         int count, i;
3648
3649         KASSERT(old_arr->ifnet_count >= 0,
3650             ("invalid ifnet array count %d", old_arr->ifnet_count));
3651         count = old_arr->ifnet_count + 1;
3652         arr = ifnet_array_alloc(count);
3653
3654         /*
3655          * Save the old ifnet array and append this ifp to the end of
3656          * the new ifnet array.
3657          */
3658         for (i = 0; i < old_arr->ifnet_count; ++i) {
3659                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3660                     ("%s is already in ifnet array", ifp->if_xname));
3661                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3662         }
3663         KASSERT(i == count - 1,
3664             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3665              ifp->if_xname, count - 1, i));
3666         arr->ifnet_arr[i] = ifp;
3667
3668         return arr;
3669 }
3670
3671 static struct ifnet_array *
3672 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3673 {
3674         struct ifnet_array *arr;
3675         int count, i, idx, found = 0;
3676
3677         KASSERT(old_arr->ifnet_count > 0,
3678             ("invalid ifnet array count %d", old_arr->ifnet_count));
3679         count = old_arr->ifnet_count - 1;
3680         arr = ifnet_array_alloc(count);
3681
3682         /*
3683          * Save the old ifnet array, but skip this ifp.
3684          */
3685         idx = 0;
3686         for (i = 0; i < old_arr->ifnet_count; ++i) {
3687                 if (old_arr->ifnet_arr[i] == ifp) {
3688                         KASSERT(!found,
3689                             ("dup %s is in ifnet array", ifp->if_xname));
3690                         found = 1;
3691                         continue;
3692                 }
3693                 KASSERT(idx < count,
3694                     ("invalid ifnet array index %d, count %d", idx, count));
3695                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3696                 ++idx;
3697         }
3698         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3699         KASSERT(idx == count,
3700             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3701              ifp->if_xname, count, idx));
3702
3703         return arr;
3704 }
3705
3706 const struct ifnet_array *
3707 ifnet_array_get(void)
3708 {
3709         const struct ifnet_array *ret;
3710
3711         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3712         ret = ifnet_array;
3713         /* Make sure 'ret' is really used. */
3714         cpu_ccfence();
3715         return (ret);
3716 }
3717
3718 int
3719 ifnet_array_isempty(void)
3720 {
3721         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3722         if (ifnet_array->ifnet_count == 0)
3723                 return 1;
3724         else
3725                 return 0;
3726 }
3727
3728 void
3729 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3730 {
3731         struct ifaddr *ifa;
3732
3733         memset(mark, 0, sizeof(*mark));
3734         ifa = &mark->ifa;
3735
3736         mark->ifac.ifa = ifa;
3737
3738         ifa->ifa_addr = &mark->addr;
3739         ifa->ifa_dstaddr = &mark->dstaddr;
3740         ifa->ifa_netmask = &mark->netmask;
3741         ifa->ifa_ifp = ifp;
3742 }
3743
3744 static int
3745 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3746 {
3747
3748         KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3749
3750         if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3751                 ring_cnt = ring_cntmax;
3752         if (ring_cnt > netisr_ncpus)
3753                 ring_cnt = netisr_ncpus;
3754         return (ring_cnt);
3755 }
3756
3757 static void
3758 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3759 {
3760         int i, offset;
3761
3762         KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3763         KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3764             grid, rm->rm_cnt));
3765         rm->rm_grid = grid;
3766
3767         offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3768         for (i = 0; i < rm->rm_cnt; ++i) {
3769                 rm->rm_cpumap[i] = offset + i;
3770                 KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3771                     ("invalid cpumap[%d] = %d, offset %d", i,
3772                      rm->rm_cpumap[i], offset));
3773         }
3774 }
3775
3776 static struct if_ringmap *
3777 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3778     uint32_t flags)
3779 {
3780         struct if_ringmap *rm;
3781         int i, grid = 0, prev_grid;
3782
3783         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3784         rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3785             M_DEVBUF, M_WAITOK | M_ZERO);
3786
3787         rm->rm_cnt = ring_cnt;
3788         if (flags & RINGMAP_FLAG_POWEROF2)
3789                 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3790
3791         prev_grid = netisr_ncpus;
3792         for (i = 0; i < netisr_ncpus; ++i) {
3793                 if (netisr_ncpus % (i + 1) != 0)
3794                         continue;
3795
3796                 grid = netisr_ncpus / (i + 1);
3797                 if (rm->rm_cnt > grid) {
3798                         grid = prev_grid;
3799                         break;
3800                 }
3801
3802                 if (rm->rm_cnt > netisr_ncpus / (i + 2))
3803                         break;
3804                 prev_grid = grid;
3805         }
3806         if_ringmap_set_grid(dev, rm, grid);
3807
3808         return (rm);
3809 }
3810
3811 struct if_ringmap *
3812 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3813 {
3814
3815         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3816             RINGMAP_FLAG_NONE));
3817 }
3818
3819 struct if_ringmap *
3820 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3821 {
3822
3823         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3824             RINGMAP_FLAG_POWEROF2));
3825 }
3826
3827 void
3828 if_ringmap_free(struct if_ringmap *rm)
3829 {
3830
3831         kfree(rm, M_DEVBUF);
3832 }
3833
3834 /*
3835  * Align the two ringmaps.
3836  *
3837  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3838  *
3839  * Before:
3840  *
3841  * CPU      0  1  2  3   4  5  6  7
3842  * NIC_RX               n0 n1 n2 n3
3843  * NIC_TX        N0 N1
3844  *
3845  * After:
3846  *
3847  * CPU      0  1  2  3   4  5  6  7
3848  * NIC_RX               n0 n1 n2 n3
3849  * NIC_TX               N0 N1
3850  */
3851 void
3852 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3853 {
3854
3855         if (rm0->rm_grid > rm1->rm_grid)
3856                 if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3857         else if (rm0->rm_grid < rm1->rm_grid)
3858                 if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3859 }
3860
3861 void
3862 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3863 {
3864         int subset_grid, cnt, divisor, mod, offset, i;
3865         struct if_ringmap *subset_rm, *rm;
3866         int old_rm0_grid, old_rm1_grid;
3867
3868         if (rm0->rm_grid == rm1->rm_grid)
3869                 return;
3870
3871         /* Save grid for later use */
3872         old_rm0_grid = rm0->rm_grid;
3873         old_rm1_grid = rm1->rm_grid;
3874
3875         if_ringmap_align(dev, rm0, rm1);
3876
3877         /*
3878          * Re-shuffle rings to get more even distribution.
3879          *
3880          * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3881          *
3882          * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3883          *
3884          * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3885          * NIC_TX   A0 A1        B0 B1        C0 C1
3886          *
3887          * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3888          * NIC_TX         D0 D1        E0 E1        F0 F1
3889          */
3890
3891         if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3892                 cnt = rm0->rm_cnt;
3893                 subset_grid = old_rm1_grid;
3894                 subset_rm = rm1;
3895                 rm = rm0;
3896         } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3897                 cnt = rm1->rm_cnt;
3898                 subset_grid = old_rm0_grid;
3899                 subset_rm = rm0;
3900                 rm = rm1;
3901         } else {
3902                 /* No space to shuffle. */
3903                 return;
3904         }
3905
3906         mod = cnt / subset_grid;
3907         KKASSERT(mod >= 2);
3908         divisor = netisr_ncpus / rm->rm_grid;
3909         offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3910
3911         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3912                 subset_rm->rm_cpumap[i] += offset;
3913                 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3914                     ("match: invalid cpumap[%d] = %d, offset %d",
3915                      i, subset_rm->rm_cpumap[i], offset));
3916         }
3917 #ifdef INVARIANTS
3918         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3919                 int j;
3920
3921                 for (j = 0; j < rm->rm_cnt; ++j) {
3922                         if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3923                                 break;
3924                 }
3925                 KASSERT(j < rm->rm_cnt,
3926                     ("subset cpumap[%d] = %d not found in superset",
3927                      i, subset_rm->rm_cpumap[i]));
3928         }
3929 #endif
3930 }
3931
3932 int
3933 if_ringmap_count(const struct if_ringmap *rm)
3934 {
3935
3936         return (rm->rm_cnt);
3937 }
3938
3939 int
3940 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3941 {
3942
3943         KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3944         return (rm->rm_cpumap[ring]);
3945 }
3946
3947 void
3948 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3949 {
3950         int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3951
3952         KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3953             ("invalid redirect table entries %d", table_nent));
3954
3955         grid_idx = 0;
3956         for (i = 0; i < NETISR_CPUMAX; ++i) {
3957                 table[i] = grid_idx++ % rm->rm_cnt;
3958
3959                 if (grid_idx == rm->rm_grid)
3960                         grid_idx = 0;
3961         }
3962
3963         /*
3964          * Make the ring distributed more evenly for the remainder
3965          * of each grid.
3966          *
3967          * e.g. 12 netisrs, rm contains 8 rings.
3968          *
3969          * Redirect table before:
3970          *
3971          *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3972          *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3973          *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3974          *  ....
3975          *
3976          * Redirect table after being patched (pX, patched entries):
3977          *
3978          *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3979          *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3980          * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3981          *  ....
3982          */
3983         patch_cnt = rm->rm_grid % rm->rm_cnt;
3984         if (patch_cnt == 0)
3985                 goto done;
3986         patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3987
3988         grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3989         grid_idx = 0;
3990         for (i = 0; i < grid_cnt; ++i) {
3991                 int j;
3992
3993                 for (j = 0; j < patch_cnt; ++j) {
3994                         int fix_idx;
3995
3996                         fix_idx = (i * rm->rm_grid) + patch_off + j;
3997                         if (fix_idx >= NETISR_CPUMAX)
3998                                 goto done;
3999                         table[fix_idx] = grid_idx++ % rm->rm_cnt;
4000                 }
4001         }
4002 done:
4003         /*
4004          * If the device supports larger redirect table, duplicate
4005          * the first NETISR_CPUMAX entries to the rest of the table,
4006          * so that it matches upper layer's expectation:
4007          * (hash & NETISR_CPUMASK) % netisr_ncpus
4008          */
4009         ncopy = table_nent / NETISR_CPUMAX;
4010         for (i = 1; i < ncopy; ++i) {
4011                 memcpy(&table[i * NETISR_CPUMAX], table,
4012                     NETISR_CPUMAX * sizeof(table[0]));
4013         }
4014         if (if_ringmap_dumprdr) {
4015                 for (i = 0; i < table_nent; ++i) {
4016                         if (i != 0 && i % 16 == 0)
4017                                 kprintf("\n");
4018                         kprintf("%03d ", table[i]);
4019                 }
4020                 kprintf("\n");
4021         }
4022 }
4023
4024 int
4025 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
4026 {
4027         struct if_ringmap *rm = arg1;
4028         int i, error = 0;
4029
4030         for (i = 0; i < rm->rm_cnt; ++i) {
4031                 int cpu = rm->rm_cpumap[i];
4032
4033                 error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
4034                 if (error)
4035                         break;
4036         }
4037         return (error);
4038 }