be6670c63a7fc0979af9f380842b449f8f516d16
[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
1666         ASSERT_NETISR0;
1667
1668         ifp->if_flags &= ~flag;
1669         getmicrotime(&ifp->if_lastchange);
1670         /*
1671          * The ifaddr processing in the following loop will block,
1672          * however, this function is called in netisr0, in which
1673          * ifaddr list changes happen, so we don't care about the
1674          * blockness of the ifaddr processing here.
1675          */
1676         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1677                 struct ifaddr *ifa = ifac->ifa;
1678
1679                 /* Ignore marker */
1680                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1681                         continue;
1682
1683                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1684                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1685         }
1686         ifq_purge_all(&ifp->if_snd);
1687         rt_ifmsg(ifp);
1688
1689         netisr_replymsg(&nmsg->base, 0);
1690 }
1691
1692 void
1693 if_unroute(struct ifnet *ifp, int flag, int fam)
1694 {
1695         struct netmsg_ifroute msg;
1696
1697         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1698             if_unroute_dispatch);
1699         msg.ifp = ifp;
1700         msg.flag = flag;
1701         msg.fam = fam;
1702         netisr_domsg(&msg.base, 0);
1703 }
1704
1705 /*
1706  * Mark an interface up and notify protocols of the transition.
1707  */
1708 static void
1709 if_route_dispatch(netmsg_t nmsg)
1710 {
1711         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1712         struct ifnet *ifp = msg->ifp;
1713         int flag = msg->flag, fam = msg->fam;
1714         struct ifaddr_container *ifac;
1715
1716         ASSERT_NETISR0;
1717
1718         ifq_purge_all(&ifp->if_snd);
1719         ifp->if_flags |= flag;
1720         getmicrotime(&ifp->if_lastchange);
1721         /*
1722          * The ifaddr processing in the following loop will block,
1723          * however, this function is called in netisr0, in which
1724          * ifaddr list changes happen, so we don't care about the
1725          * blockness of the ifaddr processing here.
1726          */
1727         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1728                 struct ifaddr *ifa = ifac->ifa;
1729
1730                 /* Ignore marker */
1731                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1732                         continue;
1733
1734                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1735                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1736         }
1737         rt_ifmsg(ifp);
1738 #ifdef INET6
1739         in6_if_up(ifp);
1740 #endif
1741
1742         netisr_replymsg(&nmsg->base, 0);
1743 }
1744
1745 void
1746 if_route(struct ifnet *ifp, int flag, int fam)
1747 {
1748         struct netmsg_ifroute msg;
1749
1750         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1751             if_route_dispatch);
1752         msg.ifp = ifp;
1753         msg.flag = flag;
1754         msg.fam = fam;
1755         netisr_domsg(&msg.base, 0);
1756 }
1757
1758 /*
1759  * Mark an interface down and notify protocols of the transition.  An
1760  * interface going down is also considered to be a synchronizing event.
1761  * We must ensure that all packet processing related to the interface
1762  * has completed before we return so e.g. the caller can free the ifnet
1763  * structure that the mbufs may be referencing.
1764  *
1765  * NOTE: must be called at splnet or eqivalent.
1766  */
1767 void
1768 if_down(struct ifnet *ifp)
1769 {
1770         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
1771         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1772         netmsg_service_sync();
1773 }
1774
1775 /*
1776  * Mark an interface up and notify protocols of
1777  * the transition.
1778  * NOTE: must be called at splnet or eqivalent.
1779  */
1780 void
1781 if_up(struct ifnet *ifp)
1782 {
1783         if_route(ifp, IFF_UP, AF_UNSPEC);
1784         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
1785 }
1786
1787 /*
1788  * Process a link state change.
1789  * NOTE: must be called at splsoftnet or equivalent.
1790  */
1791 void
1792 if_link_state_change(struct ifnet *ifp)
1793 {
1794         int link_state = ifp->if_link_state;
1795
1796         rt_ifmsg(ifp);
1797         devctl_notify("IFNET", ifp->if_xname,
1798             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1799
1800         EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
1801 }
1802
1803 /*
1804  * Handle interface watchdog timer routines.  Called
1805  * from softclock, we decrement timers (if set) and
1806  * call the appropriate interface routine on expiration.
1807  */
1808 static void
1809 if_slowtimo_dispatch(netmsg_t nmsg)
1810 {
1811         struct globaldata *gd = mycpu;
1812         const struct ifnet_array *arr;
1813         int i;
1814
1815         ASSERT_NETISR0;
1816
1817         crit_enter_gd(gd);
1818         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1819         crit_exit_gd(gd);
1820
1821         arr = ifnet_array_get();
1822         for (i = 0; i < arr->ifnet_count; ++i) {
1823                 struct ifnet *ifp = arr->ifnet_arr[i];
1824
1825                 crit_enter_gd(gd);
1826
1827                 if (if_stats_compat) {
1828                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1829                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1830                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1831                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1832                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1833                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1834                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1835                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1836                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1837                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1838                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1839                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1840                 }
1841
1842                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1843                         crit_exit_gd(gd);
1844                         continue;
1845                 }
1846                 if (ifp->if_watchdog) {
1847                         if (ifnet_tryserialize_all(ifp)) {
1848                                 (*ifp->if_watchdog)(ifp);
1849                                 ifnet_deserialize_all(ifp);
1850                         } else {
1851                                 /* try again next timeout */
1852                                 ++ifp->if_timer;
1853                         }
1854                 }
1855
1856                 crit_exit_gd(gd);
1857         }
1858
1859         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1860 }
1861
1862 static void
1863 if_slowtimo(void *arg __unused)
1864 {
1865         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1866
1867         KASSERT(mycpuid == 0, ("not on cpu0"));
1868         crit_enter();
1869         if (lmsg->ms_flags & MSGF_DONE)
1870                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1871         crit_exit();
1872 }
1873
1874 /*
1875  * Map interface name to
1876  * interface structure pointer.
1877  */
1878 struct ifnet *
1879 ifunit(const char *name)
1880 {
1881         struct ifnet *ifp;
1882
1883         /*
1884          * Search all the interfaces for this name/number
1885          */
1886         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1887
1888         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1889                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1890                         break;
1891         }
1892         return (ifp);
1893 }
1894
1895 struct ifnet *
1896 ifunit_netisr(const char *name)
1897 {
1898         const struct ifnet_array *arr;
1899         int i;
1900
1901         /*
1902          * Search all the interfaces for this name/number
1903          */
1904
1905         arr = ifnet_array_get();
1906         for (i = 0; i < arr->ifnet_count; ++i) {
1907                 struct ifnet *ifp = arr->ifnet_arr[i];
1908
1909                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1910                         return ifp;
1911         }
1912         return NULL;
1913 }
1914
1915 /*
1916  * Interface ioctls.
1917  */
1918 int
1919 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1920 {
1921         struct ifnet *ifp;
1922         struct ifgroupreq *ifgr;
1923         struct ifreq *ifr;
1924         struct ifstat *ifs;
1925         int error, do_ifup = 0;
1926         short oif_flags;
1927         int new_flags;
1928         size_t namelen, onamelen;
1929         char new_name[IFNAMSIZ];
1930         struct ifaddr *ifa;
1931         struct sockaddr_dl *sdl;
1932
1933         switch (cmd) {
1934         case SIOCGIFCONF:
1935                 return (ifconf(cmd, data, cred));
1936         default:
1937                 break;
1938         }
1939
1940         ifr = (struct ifreq *)data;
1941
1942         switch (cmd) {
1943         case SIOCIFCREATE:
1944         case SIOCIFCREATE2:
1945                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1946                         return (error);
1947                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1948                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1949         case SIOCIFDESTROY:
1950                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1951                         return (error);
1952                 return (if_clone_destroy(ifr->ifr_name));
1953         case SIOCIFGCLONERS:
1954                 return (if_clone_list((struct if_clonereq *)data));
1955         case SIOCGIFGMEMB:
1956                 return (if_getgroupmembers((struct ifgroupreq *)data));
1957         default:
1958                 break;
1959         }
1960
1961         /*
1962          * Nominal ioctl through interface, lookup the ifp and obtain a
1963          * lock to serialize the ifconfig ioctl operation.
1964          */
1965         ifnet_lock();
1966
1967         ifp = ifunit(ifr->ifr_name);
1968         if (ifp == NULL) {
1969                 ifnet_unlock();
1970                 return (ENXIO);
1971         }
1972         error = 0;
1973
1974         switch (cmd) {
1975         case SIOCGIFINDEX:
1976                 ifr->ifr_index = ifp->if_index;
1977                 break;
1978
1979         case SIOCGIFFLAGS:
1980                 ifr->ifr_flags = ifp->if_flags;
1981                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1982                 break;
1983
1984         case SIOCGIFCAP:
1985                 ifr->ifr_reqcap = ifp->if_capabilities;
1986                 ifr->ifr_curcap = ifp->if_capenable;
1987                 break;
1988
1989         case SIOCGIFMETRIC:
1990                 ifr->ifr_metric = ifp->if_metric;
1991                 break;
1992
1993         case SIOCGIFMTU:
1994                 ifr->ifr_mtu = ifp->if_mtu;
1995                 break;
1996
1997         case SIOCGIFTSOLEN:
1998                 ifr->ifr_tsolen = ifp->if_tsolen;
1999                 break;
2000
2001         case SIOCGIFDATA:
2002                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
2003                                 sizeof(ifp->if_data));
2004                 break;
2005
2006         case SIOCGIFPHYS:
2007                 ifr->ifr_phys = ifp->if_physical;
2008                 break;
2009
2010         case SIOCGIFPOLLCPU:
2011                 ifr->ifr_pollcpu = -1;
2012                 break;
2013
2014         case SIOCSIFPOLLCPU:
2015                 break;
2016
2017         case SIOCSIFFLAGS:
2018                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2019                 if (error)
2020                         break;
2021                 new_flags = (ifr->ifr_flags & 0xffff) |
2022                     (ifr->ifr_flagshigh << 16);
2023                 if (ifp->if_flags & IFF_SMART) {
2024                         /* Smart drivers twiddle their own routes */
2025                 } else if (ifp->if_flags & IFF_UP &&
2026                     (new_flags & IFF_UP) == 0) {
2027                         if_down(ifp);
2028                 } else if (new_flags & IFF_UP &&
2029                     (ifp->if_flags & IFF_UP) == 0) {
2030                         do_ifup = 1;
2031                 }
2032
2033 #ifdef IFPOLL_ENABLE
2034                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
2035                         if (new_flags & IFF_NPOLLING)
2036                                 ifpoll_register(ifp);
2037                         else
2038                                 ifpoll_deregister(ifp);
2039                 }
2040 #endif
2041
2042                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2043                         (new_flags &~ IFF_CANTCHANGE);
2044                 if (new_flags & IFF_PPROMISC) {
2045                         /* Permanently promiscuous mode requested */
2046                         ifp->if_flags |= IFF_PROMISC;
2047                 } else if (ifp->if_pcount == 0) {
2048                         ifp->if_flags &= ~IFF_PROMISC;
2049                 }
2050                 if (ifp->if_ioctl) {
2051                         ifnet_serialize_all(ifp);
2052                         ifp->if_ioctl(ifp, cmd, data, cred);
2053                         ifnet_deserialize_all(ifp);
2054                 }
2055                 if (do_ifup)
2056                         if_up(ifp);
2057                 getmicrotime(&ifp->if_lastchange);
2058                 break;
2059
2060         case SIOCSIFCAP:
2061                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2062                 if (error)
2063                         break;
2064                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
2065                         error = EINVAL;
2066                         break;
2067                 }
2068                 ifnet_serialize_all(ifp);
2069                 ifp->if_ioctl(ifp, cmd, data, cred);
2070                 ifnet_deserialize_all(ifp);
2071                 break;
2072
2073         case SIOCSIFNAME:
2074                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2075                 if (error)
2076                         break;
2077                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2078                 if (error)
2079                         break;
2080                 if (new_name[0] == '\0') {
2081                         error = EINVAL;
2082                         break;
2083                 }
2084                 if (ifunit(new_name) != NULL) {
2085                         error = EEXIST;
2086                         break;
2087                 }
2088
2089                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
2090
2091                 /* Announce the departure of the interface. */
2092                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2093
2094                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2095                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2096                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2097                 namelen = strlen(new_name);
2098                 onamelen = sdl->sdl_nlen;
2099                 /*
2100                  * Move the address if needed.  This is safe because we
2101                  * allocate space for a name of length IFNAMSIZ when we
2102                  * create this in if_attach().
2103                  */
2104                 if (namelen != onamelen) {
2105                         bcopy(sdl->sdl_data + onamelen,
2106                             sdl->sdl_data + namelen, sdl->sdl_alen);
2107                 }
2108                 bcopy(new_name, sdl->sdl_data, namelen);
2109                 sdl->sdl_nlen = namelen;
2110                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2111                 bzero(sdl->sdl_data, onamelen);
2112                 while (namelen != 0)
2113                         sdl->sdl_data[--namelen] = 0xff;
2114
2115                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2116
2117                 /* Announce the return of the interface. */
2118                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2119                 break;
2120
2121         case SIOCSIFMETRIC:
2122                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2123                 if (error)
2124                         break;
2125                 ifp->if_metric = ifr->ifr_metric;
2126                 getmicrotime(&ifp->if_lastchange);
2127                 break;
2128
2129         case SIOCSIFPHYS:
2130                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2131                 if (error)
2132                         break;
2133                 if (ifp->if_ioctl == NULL) {
2134                         error = EOPNOTSUPP;
2135                         break;
2136                 }
2137                 ifnet_serialize_all(ifp);
2138                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2139                 ifnet_deserialize_all(ifp);
2140                 if (error == 0)
2141                         getmicrotime(&ifp->if_lastchange);
2142                 break;
2143
2144         case SIOCSIFMTU:
2145         {
2146                 u_long oldmtu = ifp->if_mtu;
2147
2148                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2149                 if (error)
2150                         break;
2151                 if (ifp->if_ioctl == NULL) {
2152                         error = EOPNOTSUPP;
2153                         break;
2154                 }
2155                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2156                         error = EINVAL;
2157                         break;
2158                 }
2159                 ifnet_serialize_all(ifp);
2160                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2161                 ifnet_deserialize_all(ifp);
2162                 if (error == 0) {
2163                         getmicrotime(&ifp->if_lastchange);
2164                         rt_ifmsg(ifp);
2165                 }
2166                 /*
2167                  * If the link MTU changed, do network layer specific procedure.
2168                  */
2169                 if (ifp->if_mtu != oldmtu) {
2170 #ifdef INET6
2171                         nd6_setmtu(ifp);
2172 #endif
2173                 }
2174                 break;
2175         }
2176
2177         case SIOCSIFTSOLEN:
2178                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2179                 if (error)
2180                         break;
2181
2182                 /* XXX need driver supplied upper limit */
2183                 if (ifr->ifr_tsolen <= 0) {
2184                         error = EINVAL;
2185                         break;
2186                 }
2187                 ifp->if_tsolen = ifr->ifr_tsolen;
2188                 break;
2189
2190         case SIOCADDMULTI:
2191         case SIOCDELMULTI:
2192                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2193                 if (error)
2194                         break;
2195
2196                 /* Don't allow group membership on non-multicast interfaces. */
2197                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2198                         error = EOPNOTSUPP;
2199                         break;
2200                 }
2201
2202                 /* Don't let users screw up protocols' entries. */
2203                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2204                         error = EINVAL;
2205                         break;
2206                 }
2207
2208                 if (cmd == SIOCADDMULTI) {
2209                         struct ifmultiaddr *ifma;
2210                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2211                 } else {
2212                         error = if_delmulti(ifp, &ifr->ifr_addr);
2213                 }
2214                 if (error == 0)
2215                         getmicrotime(&ifp->if_lastchange);
2216                 break;
2217
2218         case SIOCSIFPHYADDR:
2219         case SIOCDIFPHYADDR:
2220 #ifdef INET6
2221         case SIOCSIFPHYADDR_IN6:
2222 #endif
2223         case SIOCSLIFPHYADDR:
2224         case SIOCSIFMEDIA:
2225         case SIOCSIFGENERIC:
2226                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2227                 if (error)
2228                         break;
2229                 if (ifp->if_ioctl == NULL) {
2230                         error = EOPNOTSUPP;
2231                         break;
2232                 }
2233                 ifnet_serialize_all(ifp);
2234                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2235                 ifnet_deserialize_all(ifp);
2236                 if (error == 0)
2237                         getmicrotime(&ifp->if_lastchange);
2238                 break;
2239
2240         case SIOCGIFSTATUS:
2241                 ifs = (struct ifstat *)data;
2242                 ifs->ascii[0] = '\0';
2243                 /* fall through */
2244         case SIOCGIFPSRCADDR:
2245         case SIOCGIFPDSTADDR:
2246         case SIOCGLIFPHYADDR:
2247         case SIOCGIFMEDIA:
2248         case SIOCGIFGENERIC:
2249                 if (ifp->if_ioctl == NULL) {
2250                         error = EOPNOTSUPP;
2251                         break;
2252                 }
2253                 ifnet_serialize_all(ifp);
2254                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2255                 ifnet_deserialize_all(ifp);
2256                 break;
2257
2258         case SIOCSIFLLADDR:
2259                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2260                 if (error)
2261                         break;
2262                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2263                                      ifr->ifr_addr.sa_len);
2264                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2265                 break;
2266
2267         case SIOCAIFGROUP:
2268                 ifgr = (struct ifgroupreq *)ifr;
2269                 if ((error = priv_check_cred(cred, PRIV_NET_ADDIFGROUP, 0)))
2270                         return (error);
2271                 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2272                         return (error);
2273                 break;
2274
2275         case SIOCDIFGROUP:
2276                 ifgr = (struct ifgroupreq *)ifr;
2277                 if ((error = priv_check_cred(cred, PRIV_NET_DELIFGROUP, 0)))
2278                         return (error);
2279                 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2280                         return (error);
2281                 break;
2282
2283         case SIOCGIFGROUP:
2284                 ifgr = (struct ifgroupreq *)ifr;
2285                 if ((error = if_getgroups(ifgr, ifp)))
2286                         return (error);
2287                 break;
2288
2289         default:
2290                 oif_flags = ifp->if_flags;
2291                 if (so->so_proto == 0) {
2292                         error = EOPNOTSUPP;
2293                         break;
2294                 }
2295                 error = so_pru_control_direct(so, cmd, data, ifp);
2296
2297                 /*
2298                  * If the socket control method returns EOPNOTSUPP, pass the
2299                  * request directly to the interface.
2300                  *
2301                  * Exclude the SIOCSIF{ADDR,BRDADDR,DSTADDR,NETMASK} ioctls,
2302                  * because drivers may trust these ioctls to come from an
2303                  * already privileged layer and thus do not perform credentials
2304                  * checks or input validation.
2305                  */
2306                 if (error == EOPNOTSUPP &&
2307                     ifp->if_ioctl != NULL &&
2308                     cmd != SIOCSIFADDR &&
2309                     cmd != SIOCSIFBRDADDR &&
2310                     cmd != SIOCSIFDSTADDR &&
2311                     cmd != SIOCSIFNETMASK) {
2312                         ifnet_serialize_all(ifp);
2313                         error = ifp->if_ioctl(ifp, cmd, data, cred);
2314                         ifnet_deserialize_all(ifp);
2315                 }
2316
2317                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2318 #ifdef INET6
2319                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2320                         if (ifp->if_flags & IFF_UP) {
2321                                 crit_enter();
2322                                 in6_if_up(ifp);
2323                                 crit_exit();
2324                         }
2325 #endif
2326                 }
2327                 break;
2328         }
2329
2330         ifnet_unlock();
2331         return (error);
2332 }
2333
2334 /*
2335  * Set/clear promiscuous mode on interface ifp based on the truth value
2336  * of pswitch.  The calls are reference counted so that only the first
2337  * "on" request actually has an effect, as does the final "off" request.
2338  * Results are undefined if the "off" and "on" requests are not matched.
2339  */
2340 int
2341 ifpromisc(struct ifnet *ifp, int pswitch)
2342 {
2343         struct ifreq ifr;
2344         int error;
2345         int oldflags;
2346
2347         oldflags = ifp->if_flags;
2348         if (ifp->if_flags & IFF_PPROMISC) {
2349                 /* Do nothing if device is in permanently promiscuous mode */
2350                 ifp->if_pcount += pswitch ? 1 : -1;
2351                 return (0);
2352         }
2353         if (pswitch) {
2354                 /*
2355                  * If the device is not configured up, we cannot put it in
2356                  * promiscuous mode.
2357                  */
2358                 if ((ifp->if_flags & IFF_UP) == 0)
2359                         return (ENETDOWN);
2360                 if (ifp->if_pcount++ != 0)
2361                         return (0);
2362                 ifp->if_flags |= IFF_PROMISC;
2363                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2364                     ifp->if_xname);
2365         } else {
2366                 if (--ifp->if_pcount > 0)
2367                         return (0);
2368                 ifp->if_flags &= ~IFF_PROMISC;
2369                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2370                     ifp->if_xname);
2371         }
2372         ifr.ifr_flags = ifp->if_flags;
2373         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2374         ifnet_serialize_all(ifp);
2375         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2376         ifnet_deserialize_all(ifp);
2377         if (error == 0)
2378                 rt_ifmsg(ifp);
2379         else
2380                 ifp->if_flags = oldflags;
2381         return error;
2382 }
2383
2384 /*
2385  * Return interface configuration
2386  * of system.  List may be used
2387  * in later ioctl's (above) to get
2388  * other information.
2389  */
2390 static int
2391 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2392 {
2393         struct ifconf *ifc = (struct ifconf *)data;
2394         struct ifnet *ifp;
2395         struct sockaddr *sa;
2396         struct ifreq ifr, *ifrp;
2397         int space = ifc->ifc_len, error = 0;
2398
2399         ifrp = ifc->ifc_req;
2400
2401         ifnet_lock();
2402         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2403                 struct ifaddr_container *ifac, *ifac_mark;
2404                 struct ifaddr_marker mark;
2405                 struct ifaddrhead *head;
2406                 int addrs;
2407
2408                 if (space <= sizeof ifr)
2409                         break;
2410
2411                 /*
2412                  * Zero the stack declared structure first to prevent
2413                  * memory disclosure.
2414                  */
2415                 bzero(&ifr, sizeof(ifr));
2416                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2417                     >= sizeof(ifr.ifr_name)) {
2418                         error = ENAMETOOLONG;
2419                         break;
2420                 }
2421
2422                 /*
2423                  * Add a marker, since copyout() could block and during that
2424                  * period the list could be changed.  Inserting the marker to
2425                  * the header of the list will not cause trouble for the code
2426                  * assuming that the first element of the list is AF_LINK; the
2427                  * marker will be moved to the next position w/o blocking.
2428                  */
2429                 ifa_marker_init(&mark, ifp);
2430                 ifac_mark = &mark.ifac;
2431                 head = &ifp->if_addrheads[mycpuid];
2432
2433                 addrs = 0;
2434                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2435                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2436                         struct ifaddr *ifa = ifac->ifa;
2437
2438                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2439                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2440
2441                         /* Ignore marker */
2442                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2443                                 continue;
2444
2445                         if (space <= sizeof ifr)
2446                                 break;
2447                         sa = ifa->ifa_addr;
2448                         if (cred->cr_prison &&
2449                             prison_if(cred, sa))
2450                                 continue;
2451                         addrs++;
2452                         /*
2453                          * Keep a reference on this ifaddr, so that it will
2454                          * not be destroyed when its address is copied to
2455                          * the userland, which could block.
2456                          */
2457                         IFAREF(ifa);
2458                         if (sa->sa_len <= sizeof(*sa)) {
2459                                 ifr.ifr_addr = *sa;
2460                                 error = copyout(&ifr, ifrp, sizeof ifr);
2461                                 ifrp++;
2462                         } else {
2463                                 if (space < (sizeof ifr) + sa->sa_len -
2464                                             sizeof(*sa)) {
2465                                         IFAFREE(ifa);
2466                                         break;
2467                                 }
2468                                 space -= sa->sa_len - sizeof(*sa);
2469                                 error = copyout(&ifr, ifrp,
2470                                                 sizeof ifr.ifr_name);
2471                                 if (error == 0)
2472                                         error = copyout(sa, &ifrp->ifr_addr,
2473                                                         sa->sa_len);
2474                                 ifrp = (struct ifreq *)
2475                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2476                         }
2477                         IFAFREE(ifa);
2478                         if (error)
2479                                 break;
2480                         space -= sizeof ifr;
2481                 }
2482                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2483                 if (error)
2484                         break;
2485                 if (!addrs) {
2486                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2487                         error = copyout(&ifr, ifrp, sizeof ifr);
2488                         if (error)
2489                                 break;
2490                         space -= sizeof ifr;
2491                         ifrp++;
2492                 }
2493         }
2494         ifnet_unlock();
2495
2496         ifc->ifc_len -= space;
2497         return (error);
2498 }
2499
2500 /*
2501  * Just like if_promisc(), but for all-multicast-reception mode.
2502  */
2503 int
2504 if_allmulti(struct ifnet *ifp, int onswitch)
2505 {
2506         int error = 0;
2507         struct ifreq ifr;
2508
2509         crit_enter();
2510
2511         if (onswitch) {
2512                 if (ifp->if_amcount++ == 0) {
2513                         ifp->if_flags |= IFF_ALLMULTI;
2514                         ifr.ifr_flags = ifp->if_flags;
2515                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2516                         ifnet_serialize_all(ifp);
2517                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2518                                               NULL);
2519                         ifnet_deserialize_all(ifp);
2520                 }
2521         } else {
2522                 if (ifp->if_amcount > 1) {
2523                         ifp->if_amcount--;
2524                 } else {
2525                         ifp->if_amcount = 0;
2526                         ifp->if_flags &= ~IFF_ALLMULTI;
2527                         ifr.ifr_flags = ifp->if_flags;
2528                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2529                         ifnet_serialize_all(ifp);
2530                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2531                                               NULL);
2532                         ifnet_deserialize_all(ifp);
2533                 }
2534         }
2535
2536         crit_exit();
2537
2538         if (error == 0)
2539                 rt_ifmsg(ifp);
2540         return error;
2541 }
2542
2543 /*
2544  * Add a multicast listenership to the interface in question.
2545  * The link layer provides a routine which converts
2546  */
2547 int
2548 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2549     struct ifmultiaddr **retifma)
2550 {
2551         struct sockaddr *llsa, *dupsa;
2552         int error;
2553         struct ifmultiaddr *ifma;
2554
2555         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2556
2557         /*
2558          * If the matching multicast address already exists
2559          * then don't add a new one, just add a reference
2560          */
2561         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2562                 if (sa_equal(sa, ifma->ifma_addr)) {
2563                         ifma->ifma_refcount++;
2564                         if (retifma)
2565                                 *retifma = ifma;
2566                         return 0;
2567                 }
2568         }
2569
2570         /*
2571          * Give the link layer a chance to accept/reject it, and also
2572          * find out which AF_LINK address this maps to, if it isn't one
2573          * already.
2574          */
2575         if (ifp->if_resolvemulti) {
2576                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2577                 if (error)
2578                         return error;
2579         } else {
2580                 llsa = NULL;
2581         }
2582
2583         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2584         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2585         bcopy(sa, dupsa, sa->sa_len);
2586
2587         ifma->ifma_addr = dupsa;
2588         ifma->ifma_lladdr = llsa;
2589         ifma->ifma_ifp = ifp;
2590         ifma->ifma_refcount = 1;
2591         ifma->ifma_protospec = NULL;
2592         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2593
2594         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2595         if (retifma)
2596                 *retifma = ifma;
2597
2598         if (llsa != NULL) {
2599                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2600                         if (sa_equal(ifma->ifma_addr, llsa))
2601                                 break;
2602                 }
2603                 if (ifma) {
2604                         ifma->ifma_refcount++;
2605                 } else {
2606                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2607                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2608                         bcopy(llsa, dupsa, llsa->sa_len);
2609                         ifma->ifma_addr = dupsa;
2610                         ifma->ifma_ifp = ifp;
2611                         ifma->ifma_refcount = 1;
2612                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2613                 }
2614         }
2615         /*
2616          * We are certain we have added something, so call down to the
2617          * interface to let them know about it.
2618          */
2619         if (ifp->if_ioctl)
2620                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2621
2622         return 0;
2623 }
2624
2625 int
2626 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2627     struct ifmultiaddr **retifma)
2628 {
2629         int error;
2630
2631         ifnet_serialize_all(ifp);
2632         error = if_addmulti_serialized(ifp, sa, retifma);
2633         ifnet_deserialize_all(ifp);
2634
2635         return error;
2636 }
2637
2638 /*
2639  * Remove a reference to a multicast address on this interface.  Yell
2640  * if the request does not match an existing membership.
2641  */
2642 static int
2643 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2644 {
2645         struct ifmultiaddr *ifma;
2646
2647         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2648
2649         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2650                 if (sa_equal(sa, ifma->ifma_addr))
2651                         break;
2652         if (ifma == NULL)
2653                 return ENOENT;
2654
2655         if (ifma->ifma_refcount > 1) {
2656                 ifma->ifma_refcount--;
2657                 return 0;
2658         }
2659
2660         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2661         sa = ifma->ifma_lladdr;
2662         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2663         /*
2664          * Make sure the interface driver is notified
2665          * in the case of a link layer mcast group being left.
2666          */
2667         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2668                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2669         kfree(ifma->ifma_addr, M_IFMADDR);
2670         kfree(ifma, M_IFMADDR);
2671         if (sa == NULL)
2672                 return 0;
2673
2674         /*
2675          * Now look for the link-layer address which corresponds to
2676          * this network address.  It had been squirreled away in
2677          * ifma->ifma_lladdr for this purpose (so we don't have
2678          * to call ifp->if_resolvemulti() again), and we saved that
2679          * value in sa above.  If some nasty deleted the
2680          * link-layer address out from underneath us, we can deal because
2681          * the address we stored was is not the same as the one which was
2682          * in the record for the link-layer address.  (So we don't complain
2683          * in that case.)
2684          */
2685         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2686                 if (sa_equal(sa, ifma->ifma_addr))
2687                         break;
2688         if (ifma == NULL)
2689                 return 0;
2690
2691         if (ifma->ifma_refcount > 1) {
2692                 ifma->ifma_refcount--;
2693                 return 0;
2694         }
2695
2696         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2697         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2698         kfree(ifma->ifma_addr, M_IFMADDR);
2699         kfree(sa, M_IFMADDR);
2700         kfree(ifma, M_IFMADDR);
2701
2702         return 0;
2703 }
2704
2705 int
2706 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2707 {
2708         int error;
2709
2710         ifnet_serialize_all(ifp);
2711         error = if_delmulti_serialized(ifp, sa);
2712         ifnet_deserialize_all(ifp);
2713
2714         return error;
2715 }
2716
2717 /*
2718  * Delete all multicast group membership for an interface.
2719  * Should be used to quickly flush all multicast filters.
2720  */
2721 void
2722 if_delallmulti_serialized(struct ifnet *ifp)
2723 {
2724         struct ifmultiaddr *ifma, mark;
2725         struct sockaddr sa;
2726
2727         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2728
2729         bzero(&sa, sizeof(sa));
2730         sa.sa_family = AF_UNSPEC;
2731         sa.sa_len = sizeof(sa);
2732
2733         bzero(&mark, sizeof(mark));
2734         mark.ifma_addr = &sa;
2735
2736         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2737         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2738                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2739                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2740                     ifma_link);
2741
2742                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2743                         continue;
2744
2745                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2746         }
2747         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2748 }
2749
2750
2751 /*
2752  * Set the link layer address on an interface.
2753  *
2754  * At this time we only support certain types of interfaces,
2755  * and we don't allow the length of the address to change.
2756  */
2757 int
2758 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2759 {
2760         struct sockaddr_dl *sdl;
2761         struct ifreq ifr;
2762
2763         sdl = IF_LLSOCKADDR(ifp);
2764         if (sdl == NULL)
2765                 return (EINVAL);
2766         if (len != sdl->sdl_alen)       /* don't allow length to change */
2767                 return (EINVAL);
2768         switch (ifp->if_type) {
2769         case IFT_ETHER:                 /* these types use struct arpcom */
2770         case IFT_XETHER:
2771         case IFT_L2VLAN:
2772         case IFT_IEEE8023ADLAG:
2773                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2774                 bcopy(lladdr, LLADDR(sdl), len);
2775                 break;
2776         default:
2777                 return (ENODEV);
2778         }
2779         /*
2780          * If the interface is already up, we need
2781          * to re-init it in order to reprogram its
2782          * address filter.
2783          */
2784         ifnet_serialize_all(ifp);
2785         if ((ifp->if_flags & IFF_UP) != 0) {
2786 #ifdef INET
2787                 struct ifaddr_container *ifac;
2788 #endif
2789
2790                 ifp->if_flags &= ~IFF_UP;
2791                 ifr.ifr_flags = ifp->if_flags;
2792                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2793                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2794                               NULL);
2795                 ifp->if_flags |= IFF_UP;
2796                 ifr.ifr_flags = ifp->if_flags;
2797                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2798                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2799                                  NULL);
2800 #ifdef INET
2801                 /*
2802                  * Also send gratuitous ARPs to notify other nodes about
2803                  * the address change.
2804                  */
2805                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2806                         struct ifaddr *ifa = ifac->ifa;
2807
2808                         if (ifa->ifa_addr != NULL &&
2809                             ifa->ifa_addr->sa_family == AF_INET)
2810                                 arp_gratuitous(ifp, ifa);
2811                 }
2812 #endif
2813         }
2814         ifnet_deserialize_all(ifp);
2815         return (0);
2816 }
2817
2818 struct ifmultiaddr *
2819 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2820 {
2821         struct ifmultiaddr *ifma;
2822
2823         /* TODO: need ifnet_serialize_main */
2824         ifnet_serialize_all(ifp);
2825         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2826                 if (sa_equal(ifma->ifma_addr, sa))
2827                         break;
2828         ifnet_deserialize_all(ifp);
2829
2830         return ifma;
2831 }
2832
2833 /*
2834  * This function locates the first real ethernet MAC from a network
2835  * card and loads it into node, returning 0 on success or ENOENT if
2836  * no suitable interfaces were found.  It is used by the uuid code to
2837  * generate a unique 6-byte number.
2838  */
2839 int
2840 if_getanyethermac(uint16_t *node, int minlen)
2841 {
2842         struct ifnet *ifp;
2843         struct sockaddr_dl *sdl;
2844
2845         ifnet_lock();
2846         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2847                 if (ifp->if_type != IFT_ETHER)
2848                         continue;
2849                 sdl = IF_LLSOCKADDR(ifp);
2850                 if (sdl->sdl_alen < minlen)
2851                         continue;
2852                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2853                       minlen);
2854                 ifnet_unlock();
2855                 return(0);
2856         }
2857         ifnet_unlock();
2858         return (ENOENT);
2859 }
2860
2861 /*
2862  * The name argument must be a pointer to storage which will last as
2863  * long as the interface does.  For physical devices, the result of
2864  * device_get_name(dev) is a good choice and for pseudo-devices a
2865  * static string works well.
2866  */
2867 void
2868 if_initname(struct ifnet *ifp, const char *name, int unit)
2869 {
2870         ifp->if_dname = name;
2871         ifp->if_dunit = unit;
2872         if (unit != IF_DUNIT_NONE)
2873                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2874         else
2875                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2876 }
2877
2878 int
2879 if_printf(struct ifnet *ifp, const char *fmt, ...)
2880 {
2881         __va_list ap;
2882         int retval;
2883
2884         retval = kprintf("%s: ", ifp->if_xname);
2885         __va_start(ap, fmt);
2886         retval += kvprintf(fmt, ap);
2887         __va_end(ap);
2888         return (retval);
2889 }
2890
2891 struct ifnet *
2892 if_alloc(uint8_t type)
2893 {
2894         struct ifnet *ifp;
2895         size_t size;
2896
2897         /*
2898          * XXX temporary hack until arpcom is setup in if_l2com
2899          */
2900         if (type == IFT_ETHER)
2901                 size = sizeof(struct arpcom);
2902         else
2903                 size = sizeof(struct ifnet);
2904
2905         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2906
2907         ifp->if_type = type;
2908
2909         if (if_com_alloc[type] != NULL) {
2910                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2911                 if (ifp->if_l2com == NULL) {
2912                         kfree(ifp, M_IFNET);
2913                         return (NULL);
2914                 }
2915         }
2916         return (ifp);
2917 }
2918
2919 void
2920 if_free(struct ifnet *ifp)
2921 {
2922         kfree(ifp, M_IFNET);
2923 }
2924
2925 void
2926 ifq_set_classic(struct ifaltq *ifq)
2927 {
2928         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2929             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2930 }
2931
2932 void
2933 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2934     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2935 {
2936         int q;
2937
2938         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2939         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2940         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2941         KASSERT(request != NULL, ("request is not specified"));
2942
2943         ifq->altq_mapsubq = mapsubq;
2944         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2945                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2946
2947                 ifsq->ifsq_enqueue = enqueue;
2948                 ifsq->ifsq_dequeue = dequeue;
2949                 ifsq->ifsq_request = request;
2950         }
2951 }
2952
2953 static void
2954 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2955 {
2956
2957         classq_add(&ifsq->ifsq_norm, m);
2958         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2959 }
2960
2961 static void
2962 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2963 {
2964
2965         classq_add(&ifsq->ifsq_prio, m);
2966         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2967         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2968 }
2969
2970 static struct mbuf *
2971 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2972 {
2973         struct mbuf *m;
2974
2975         m = classq_get(&ifsq->ifsq_norm);
2976         if (m != NULL)
2977                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2978         return (m);
2979 }
2980
2981 static struct mbuf *
2982 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
2983 {
2984         struct mbuf *m;
2985
2986         m = classq_get(&ifsq->ifsq_prio);
2987         if (m != NULL) {
2988                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
2989                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
2990         }
2991         return (m);
2992 }
2993
2994 int
2995 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
2996     struct altq_pktattr *pa __unused)
2997 {
2998
2999         M_ASSERTPKTHDR(m);
3000 again:
3001         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
3002             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
3003                 struct mbuf *m_drop;
3004
3005                 if (m->m_flags & M_PRIO) {
3006                         m_drop = NULL;
3007                         if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
3008                             ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
3009                                 /* Try dropping some from normal queue. */
3010                                 m_drop = ifsq_norm_dequeue(ifsq);
3011                         }
3012                         if (m_drop == NULL)
3013                                 m_drop = ifsq_prio_dequeue(ifsq);
3014                 } else {
3015                         m_drop = ifsq_norm_dequeue(ifsq);
3016                 }
3017                 if (m_drop != NULL) {
3018                         IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
3019                         m_freem(m_drop);
3020                         goto again;
3021                 }
3022                 /*
3023                  * No old packets could be dropped!
3024                  * NOTE: Caller increases oqdrops.
3025                  */
3026                 m_freem(m);
3027                 return (ENOBUFS);
3028         } else {
3029                 if (m->m_flags & M_PRIO)
3030                         ifsq_prio_enqueue(ifsq, m);
3031                 else
3032                         ifsq_norm_enqueue(ifsq, m);
3033                 return (0);
3034         }
3035 }
3036
3037 struct mbuf *
3038 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
3039 {
3040         struct mbuf *m;
3041
3042         switch (op) {
3043         case ALTDQ_POLL:
3044                 m = classq_head(&ifsq->ifsq_prio);
3045                 if (m == NULL)
3046                         m = classq_head(&ifsq->ifsq_norm);
3047                 break;
3048
3049         case ALTDQ_REMOVE:
3050                 m = ifsq_prio_dequeue(ifsq);
3051                 if (m == NULL)
3052                         m = ifsq_norm_dequeue(ifsq);
3053                 break;
3054
3055         default:
3056                 panic("unsupported ALTQ dequeue op: %d", op);
3057         }
3058         return m;
3059 }
3060
3061 int
3062 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
3063 {
3064         switch (req) {
3065         case ALTRQ_PURGE:
3066                 for (;;) {
3067                         struct mbuf *m;
3068
3069                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
3070                         if (m == NULL)
3071                                 break;
3072                         m_freem(m);
3073                 }
3074                 break;
3075
3076         default:
3077                 panic("unsupported ALTQ request: %d", req);
3078         }
3079         return 0;
3080 }
3081
3082 static void
3083 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
3084 {
3085         struct ifnet *ifp = ifsq_get_ifp(ifsq);
3086         int running = 0, need_sched;
3087
3088         /*
3089          * Try to do direct ifnet.if_start on the subqueue first, if there is
3090          * contention on the subqueue hardware serializer, ifnet.if_start on
3091          * the subqueue will be scheduled on the subqueue owner CPU.
3092          */
3093         if (!ifsq_tryserialize_hw(ifsq)) {
3094                 /*
3095                  * Subqueue hardware serializer contention happened,
3096                  * ifnet.if_start on the subqueue is scheduled on
3097                  * the subqueue owner CPU, and we keep going.
3098                  */
3099                 ifsq_ifstart_schedule(ifsq, 1);
3100                 return;
3101         }
3102
3103         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
3104                 ifp->if_start(ifp, ifsq);
3105                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
3106                         running = 1;
3107         }
3108         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
3109
3110         ifsq_deserialize_hw(ifsq);
3111
3112         if (need_sched) {
3113                 /*
3114                  * More data need to be transmitted, ifnet.if_start on the
3115                  * subqueue is scheduled on the subqueue owner CPU, and we
3116                  * keep going.
3117                  * NOTE: ifnet.if_start subqueue interlock is not released.
3118                  */
3119                 ifsq_ifstart_schedule(ifsq, force_sched);
3120         }
3121 }
3122
3123 /*
3124  * Subqeue packets staging mechanism:
3125  *
3126  * The packets enqueued into the subqueue are staged to a certain amount
3127  * before the ifnet.if_start on the subqueue is called.  In this way, the
3128  * driver could avoid writing to hardware registers upon every packet,
3129  * instead, hardware registers could be written when certain amount of
3130  * packets are put onto hardware TX ring.  The measurement on several modern
3131  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3132  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3133  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3134  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3135  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3136  *
3137  * Subqueue packets staging is performed for two entry points into drivers'
3138  * transmission function:
3139  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3140  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3141  *
3142  * Subqueue packets staging will be stopped upon any of the following
3143  * conditions:
3144  * - If the count of packets enqueued on the current CPU is great than or
3145  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3146  * - If the total length of packets enqueued on the current CPU is great
3147  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3148  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3149  *   is usually less than hardware's MTU.
3150  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3151  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3152  *   released.
3153  * - The if_start_rollup(), which is registered as low priority netisr
3154  *   rollup function, is called; probably because no more work is pending
3155  *   for netisr.
3156  *
3157  * NOTE:
3158  * Currently subqueue packet staging is only performed in netisr threads.
3159  */
3160 int
3161 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3162 {
3163         struct ifaltq *ifq = &ifp->if_snd;
3164         struct ifaltq_subque *ifsq;
3165         int error, start = 0, len, mcast = 0, avoid_start = 0;
3166         struct ifsubq_stage_head *head = NULL;
3167         struct ifsubq_stage *stage = NULL;
3168         struct globaldata *gd = mycpu;
3169         struct thread *td = gd->gd_curthread;
3170
3171         crit_enter_quick(td);
3172
3173         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3174         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3175
3176         len = m->m_pkthdr.len;
3177         if (m->m_flags & M_MCAST)
3178                 mcast = 1;
3179
3180         if (td->td_type == TD_TYPE_NETISR) {
3181                 head = &ifsubq_stage_heads[mycpuid];
3182                 stage = ifsq_get_stage(ifsq, mycpuid);
3183
3184                 stage->stg_cnt++;
3185                 stage->stg_len += len;
3186                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3187                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3188                         avoid_start = 1;
3189         }
3190
3191         ALTQ_SQ_LOCK(ifsq);
3192         error = ifsq_enqueue_locked(ifsq, m, pa);
3193         if (error) {
3194                 IFNET_STAT_INC(ifp, oqdrops, 1);
3195                 if (!ifsq_data_ready(ifsq)) {
3196                         ALTQ_SQ_UNLOCK(ifsq);
3197                         crit_exit_quick(td);
3198                         return error;
3199                 }
3200                 avoid_start = 0;
3201         }
3202         if (!ifsq_is_started(ifsq)) {
3203                 if (avoid_start) {
3204                         ALTQ_SQ_UNLOCK(ifsq);
3205
3206                         KKASSERT(!error);
3207                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3208                                 ifsq_stage_insert(head, stage);
3209
3210                         IFNET_STAT_INC(ifp, obytes, len);
3211                         if (mcast)
3212                                 IFNET_STAT_INC(ifp, omcasts, 1);
3213                         crit_exit_quick(td);
3214                         return error;
3215                 }
3216
3217                 /*
3218                  * Hold the subqueue interlock of ifnet.if_start
3219                  */
3220                 ifsq_set_started(ifsq);
3221                 start = 1;
3222         }
3223         ALTQ_SQ_UNLOCK(ifsq);
3224
3225         if (!error) {
3226                 IFNET_STAT_INC(ifp, obytes, len);
3227                 if (mcast)
3228                         IFNET_STAT_INC(ifp, omcasts, 1);
3229         }
3230
3231         if (stage != NULL) {
3232                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3233                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3234                         if (!avoid_start) {
3235                                 ifsq_stage_remove(head, stage);
3236                                 ifsq_ifstart_schedule(ifsq, 1);
3237                         }
3238                         crit_exit_quick(td);
3239                         return error;
3240                 }
3241
3242                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3243                         ifsq_stage_remove(head, stage);
3244                 } else {
3245                         stage->stg_cnt = 0;
3246                         stage->stg_len = 0;
3247                 }
3248         }
3249
3250         if (!start) {
3251                 crit_exit_quick(td);
3252                 return error;
3253         }
3254
3255         ifsq_ifstart_try(ifsq, 0);
3256
3257         crit_exit_quick(td);
3258         return error;
3259 }
3260
3261 void *
3262 ifa_create(int size)
3263 {
3264         struct ifaddr *ifa;
3265         int i;
3266
3267         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3268
3269         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3270
3271         /*
3272          * Make ifa_container availabel on all CPUs, since they
3273          * could be accessed by any threads.
3274          */
3275         ifa->ifa_containers =
3276                 kmalloc(ncpus * sizeof(struct ifaddr_container),
3277                         M_IFADDR,
3278                         M_INTWAIT | M_ZERO | M_CACHEALIGN);
3279
3280         ifa->ifa_ncnt = ncpus;
3281         for (i = 0; i < ncpus; ++i) {
3282                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3283
3284                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3285                 ifac->ifa = ifa;
3286                 ifac->ifa_refcnt = 1;
3287         }
3288 #ifdef IFADDR_DEBUG
3289         kprintf("alloc ifa %p %d\n", ifa, size);
3290 #endif
3291         return ifa;
3292 }
3293
3294 void
3295 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3296 {
3297         struct ifaddr *ifa = ifac->ifa;
3298
3299         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3300         KKASSERT(ifac->ifa_refcnt == 0);
3301         KASSERT(ifac->ifa_listmask == 0,
3302                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3303
3304         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3305
3306 #ifdef IFADDR_DEBUG_VERBOSE
3307         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3308 #endif
3309
3310         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3311                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3312         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3313 #ifdef IFADDR_DEBUG
3314                 kprintf("free ifa %p\n", ifa);
3315 #endif
3316                 kfree(ifa->ifa_containers, M_IFADDR);
3317                 kfree(ifa, M_IFADDR);
3318         }
3319 }
3320
3321 static void
3322 ifa_iflink_dispatch(netmsg_t nmsg)
3323 {
3324         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3325         struct ifaddr *ifa = msg->ifa;
3326         struct ifnet *ifp = msg->ifp;
3327         int cpu = mycpuid;
3328         struct ifaddr_container *ifac;
3329
3330         crit_enter();
3331
3332         ifac = &ifa->ifa_containers[cpu];
3333         ASSERT_IFAC_VALID(ifac);
3334         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3335                 ("ifaddr is on if_addrheads"));
3336
3337         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3338         if (msg->tail)
3339                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3340         else
3341                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3342
3343         crit_exit();
3344
3345         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3346 }
3347
3348 void
3349 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3350 {
3351         struct netmsg_ifaddr msg;
3352
3353         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3354                     0, ifa_iflink_dispatch);
3355         msg.ifa = ifa;
3356         msg.ifp = ifp;
3357         msg.tail = tail;
3358
3359         netisr_domsg(&msg.base, 0);
3360 }
3361
3362 static void
3363 ifa_ifunlink_dispatch(netmsg_t nmsg)
3364 {
3365         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3366         struct ifaddr *ifa = msg->ifa;
3367         struct ifnet *ifp = msg->ifp;
3368         int cpu = mycpuid;
3369         struct ifaddr_container *ifac;
3370
3371         crit_enter();
3372
3373         ifac = &ifa->ifa_containers[cpu];
3374         ASSERT_IFAC_VALID(ifac);
3375         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3376                 ("ifaddr is not on if_addrhead"));
3377
3378         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3379         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3380
3381         crit_exit();
3382
3383         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3384 }
3385
3386 void
3387 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3388 {
3389         struct netmsg_ifaddr msg;
3390
3391         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3392                     0, ifa_ifunlink_dispatch);
3393         msg.ifa = ifa;
3394         msg.ifp = ifp;
3395
3396         netisr_domsg(&msg.base, 0);
3397 }
3398
3399 static void
3400 ifa_destroy_dispatch(netmsg_t nmsg)
3401 {
3402         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3403
3404         IFAFREE(msg->ifa);
3405         netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3406 }
3407
3408 void
3409 ifa_destroy(struct ifaddr *ifa)
3410 {
3411         struct netmsg_ifaddr msg;
3412
3413         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3414                     0, ifa_destroy_dispatch);
3415         msg.ifa = ifa;
3416
3417         netisr_domsg(&msg.base, 0);
3418 }
3419
3420 static void
3421 if_start_rollup(void)
3422 {
3423         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3424         struct ifsubq_stage *stage;
3425
3426         crit_enter();
3427
3428         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3429                 struct ifaltq_subque *ifsq = stage->stg_subq;
3430                 int is_sched = 0;
3431
3432                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3433                         is_sched = 1;
3434                 ifsq_stage_remove(head, stage);
3435
3436                 if (is_sched) {
3437                         ifsq_ifstart_schedule(ifsq, 1);
3438                 } else {
3439                         int start = 0;
3440
3441                         ALTQ_SQ_LOCK(ifsq);
3442                         if (!ifsq_is_started(ifsq)) {
3443                                 /*
3444                                  * Hold the subqueue interlock of
3445                                  * ifnet.if_start
3446                                  */
3447                                 ifsq_set_started(ifsq);
3448                                 start = 1;
3449                         }
3450                         ALTQ_SQ_UNLOCK(ifsq);
3451
3452                         if (start)
3453                                 ifsq_ifstart_try(ifsq, 1);
3454                 }
3455                 KKASSERT((stage->stg_flags &
3456                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3457         }
3458
3459         crit_exit();
3460 }
3461
3462 static void
3463 ifnetinit(void *dummy __unused)
3464 {
3465         int i;
3466
3467         /* XXX netisr_ncpus */
3468         for (i = 0; i < ncpus; ++i)
3469                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3470         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3471 }
3472
3473 void
3474 if_register_com_alloc(u_char type,
3475     if_com_alloc_t *a, if_com_free_t *f)
3476 {
3477
3478         KASSERT(if_com_alloc[type] == NULL,
3479             ("if_register_com_alloc: %d already registered", type));
3480         KASSERT(if_com_free[type] == NULL,
3481             ("if_register_com_alloc: %d free already registered", type));
3482
3483         if_com_alloc[type] = a;
3484         if_com_free[type] = f;
3485 }
3486
3487 void
3488 if_deregister_com_alloc(u_char type)
3489 {
3490
3491         KASSERT(if_com_alloc[type] != NULL,
3492             ("if_deregister_com_alloc: %d not registered", type));
3493         KASSERT(if_com_free[type] != NULL,
3494             ("if_deregister_com_alloc: %d free not registered", type));
3495         if_com_alloc[type] = NULL;
3496         if_com_free[type] = NULL;
3497 }
3498
3499 void
3500 ifq_set_maxlen(struct ifaltq *ifq, int len)
3501 {
3502         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3503 }
3504
3505 int
3506 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3507 {
3508         return ALTQ_SUBQ_INDEX_DEFAULT;
3509 }
3510
3511 int
3512 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3513 {
3514
3515         return (cpuid % ifq->altq_subq_mappriv);
3516 }
3517
3518 static void
3519 ifsq_watchdog(void *arg)
3520 {
3521         struct ifsubq_watchdog *wd = arg;
3522         struct ifnet *ifp;
3523
3524         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3525                 goto done;
3526
3527         ifp = ifsq_get_ifp(wd->wd_subq);
3528         if (ifnet_tryserialize_all(ifp)) {
3529                 wd->wd_watchdog(wd->wd_subq);
3530                 ifnet_deserialize_all(ifp);
3531         } else {
3532                 /* try again next timeout */
3533                 wd->wd_timer = 1;
3534         }
3535 done:
3536         ifsq_watchdog_reset(wd);
3537 }
3538
3539 static void
3540 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3541 {
3542         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3543             ifsq_get_cpuid(wd->wd_subq));
3544 }
3545
3546 void
3547 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3548     ifsq_watchdog_t watchdog)
3549 {
3550         callout_init_mp(&wd->wd_callout);
3551         wd->wd_timer = 0;
3552         wd->wd_subq = ifsq;
3553         wd->wd_watchdog = watchdog;
3554 }
3555
3556 void
3557 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3558 {
3559         wd->wd_timer = 0;
3560         ifsq_watchdog_reset(wd);
3561 }
3562
3563 void
3564 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3565 {
3566         wd->wd_timer = 0;
3567         callout_stop(&wd->wd_callout);
3568 }
3569
3570 void
3571 ifnet_lock(void)
3572 {
3573         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3574             ("try holding ifnet lock in netisr"));
3575         mtx_lock(&ifnet_mtx);
3576 }
3577
3578 void
3579 ifnet_unlock(void)
3580 {
3581         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3582             ("try holding ifnet lock in netisr"));
3583         mtx_unlock(&ifnet_mtx);
3584 }
3585
3586 static struct ifnet_array *
3587 ifnet_array_alloc(int count)
3588 {
3589         struct ifnet_array *arr;
3590
3591         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3592             M_IFNET, M_WAITOK);
3593         arr->ifnet_count = count;
3594
3595         return arr;
3596 }
3597
3598 static void
3599 ifnet_array_free(struct ifnet_array *arr)
3600 {
3601         if (arr == &ifnet_array0)
3602                 return;
3603         kfree(arr, M_IFNET);
3604 }
3605
3606 static struct ifnet_array *
3607 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3608 {
3609         struct ifnet_array *arr;
3610         int count, i;
3611
3612         KASSERT(old_arr->ifnet_count >= 0,
3613             ("invalid ifnet array count %d", old_arr->ifnet_count));
3614         count = old_arr->ifnet_count + 1;
3615         arr = ifnet_array_alloc(count);
3616
3617         /*
3618          * Save the old ifnet array and append this ifp to the end of
3619          * the new ifnet array.
3620          */
3621         for (i = 0; i < old_arr->ifnet_count; ++i) {
3622                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3623                     ("%s is already in ifnet array", ifp->if_xname));
3624                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3625         }
3626         KASSERT(i == count - 1,
3627             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3628              ifp->if_xname, count - 1, i));
3629         arr->ifnet_arr[i] = ifp;
3630
3631         return arr;
3632 }
3633
3634 static struct ifnet_array *
3635 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3636 {
3637         struct ifnet_array *arr;
3638         int count, i, idx, found = 0;
3639
3640         KASSERT(old_arr->ifnet_count > 0,
3641             ("invalid ifnet array count %d", old_arr->ifnet_count));
3642         count = old_arr->ifnet_count - 1;
3643         arr = ifnet_array_alloc(count);
3644
3645         /*
3646          * Save the old ifnet array, but skip this ifp.
3647          */
3648         idx = 0;
3649         for (i = 0; i < old_arr->ifnet_count; ++i) {
3650                 if (old_arr->ifnet_arr[i] == ifp) {
3651                         KASSERT(!found,
3652                             ("dup %s is in ifnet array", ifp->if_xname));
3653                         found = 1;
3654                         continue;
3655                 }
3656                 KASSERT(idx < count,
3657                     ("invalid ifnet array index %d, count %d", idx, count));
3658                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3659                 ++idx;
3660         }
3661         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3662         KASSERT(idx == count,
3663             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3664              ifp->if_xname, count, idx));
3665
3666         return arr;
3667 }
3668
3669 const struct ifnet_array *
3670 ifnet_array_get(void)
3671 {
3672         const struct ifnet_array *ret;
3673
3674         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3675         ret = ifnet_array;
3676         /* Make sure 'ret' is really used. */
3677         cpu_ccfence();
3678         return (ret);
3679 }
3680
3681 int
3682 ifnet_array_isempty(void)
3683 {
3684         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3685         if (ifnet_array->ifnet_count == 0)
3686                 return 1;
3687         else
3688                 return 0;
3689 }
3690
3691 void
3692 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3693 {
3694         struct ifaddr *ifa;
3695
3696         memset(mark, 0, sizeof(*mark));
3697         ifa = &mark->ifa;
3698
3699         mark->ifac.ifa = ifa;
3700
3701         ifa->ifa_addr = &mark->addr;
3702         ifa->ifa_dstaddr = &mark->dstaddr;
3703         ifa->ifa_netmask = &mark->netmask;
3704         ifa->ifa_ifp = ifp;
3705 }
3706
3707 static int
3708 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3709 {
3710
3711         KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3712
3713         if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3714                 ring_cnt = ring_cntmax;
3715         if (ring_cnt > netisr_ncpus)
3716                 ring_cnt = netisr_ncpus;
3717         return (ring_cnt);
3718 }
3719
3720 static void
3721 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3722 {
3723         int i, offset;
3724
3725         KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3726         KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3727             grid, rm->rm_cnt));
3728         rm->rm_grid = grid;
3729
3730         offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3731         for (i = 0; i < rm->rm_cnt; ++i) {
3732                 rm->rm_cpumap[i] = offset + i;
3733                 KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3734                     ("invalid cpumap[%d] = %d, offset %d", i,
3735                      rm->rm_cpumap[i], offset));
3736         }
3737 }
3738
3739 static struct if_ringmap *
3740 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3741     uint32_t flags)
3742 {
3743         struct if_ringmap *rm;
3744         int i, grid = 0, prev_grid;
3745
3746         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3747         rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3748             M_DEVBUF, M_WAITOK | M_ZERO);
3749
3750         rm->rm_cnt = ring_cnt;
3751         if (flags & RINGMAP_FLAG_POWEROF2)
3752                 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3753
3754         prev_grid = netisr_ncpus;
3755         for (i = 0; i < netisr_ncpus; ++i) {
3756                 if (netisr_ncpus % (i + 1) != 0)
3757                         continue;
3758
3759                 grid = netisr_ncpus / (i + 1);
3760                 if (rm->rm_cnt > grid) {
3761                         grid = prev_grid;
3762                         break;
3763                 }
3764
3765                 if (rm->rm_cnt > netisr_ncpus / (i + 2))
3766                         break;
3767                 prev_grid = grid;
3768         }
3769         if_ringmap_set_grid(dev, rm, grid);
3770
3771         return (rm);
3772 }
3773
3774 struct if_ringmap *
3775 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3776 {
3777
3778         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3779             RINGMAP_FLAG_NONE));
3780 }
3781
3782 struct if_ringmap *
3783 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3784 {
3785
3786         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3787             RINGMAP_FLAG_POWEROF2));
3788 }
3789
3790 void
3791 if_ringmap_free(struct if_ringmap *rm)
3792 {
3793
3794         kfree(rm, M_DEVBUF);
3795 }
3796
3797 /*
3798  * Align the two ringmaps.
3799  *
3800  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3801  *
3802  * Before:
3803  *
3804  * CPU      0  1  2  3   4  5  6  7
3805  * NIC_RX               n0 n1 n2 n3
3806  * NIC_TX        N0 N1
3807  *
3808  * After:
3809  *
3810  * CPU      0  1  2  3   4  5  6  7
3811  * NIC_RX               n0 n1 n2 n3
3812  * NIC_TX               N0 N1
3813  */
3814 void
3815 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3816 {
3817
3818         if (rm0->rm_grid > rm1->rm_grid)
3819                 if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3820         else if (rm0->rm_grid < rm1->rm_grid)
3821                 if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3822 }
3823
3824 void
3825 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3826 {
3827         int subset_grid, cnt, divisor, mod, offset, i;
3828         struct if_ringmap *subset_rm, *rm;
3829         int old_rm0_grid, old_rm1_grid;
3830
3831         if (rm0->rm_grid == rm1->rm_grid)
3832                 return;
3833
3834         /* Save grid for later use */
3835         old_rm0_grid = rm0->rm_grid;
3836         old_rm1_grid = rm1->rm_grid;
3837
3838         if_ringmap_align(dev, rm0, rm1);
3839
3840         /*
3841          * Re-shuffle rings to get more even distribution.
3842          *
3843          * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3844          *
3845          * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3846          *
3847          * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3848          * NIC_TX   A0 A1        B0 B1        C0 C1
3849          *
3850          * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3851          * NIC_TX         D0 D1        E0 E1        F0 F1
3852          */
3853
3854         if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3855                 cnt = rm0->rm_cnt;
3856                 subset_grid = old_rm1_grid;
3857                 subset_rm = rm1;
3858                 rm = rm0;
3859         } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3860                 cnt = rm1->rm_cnt;
3861                 subset_grid = old_rm0_grid;
3862                 subset_rm = rm0;
3863                 rm = rm1;
3864         } else {
3865                 /* No space to shuffle. */
3866                 return;
3867         }
3868
3869         mod = cnt / subset_grid;
3870         KKASSERT(mod >= 2);
3871         divisor = netisr_ncpus / rm->rm_grid;
3872         offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3873
3874         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3875                 subset_rm->rm_cpumap[i] += offset;
3876                 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3877                     ("match: invalid cpumap[%d] = %d, offset %d",
3878                      i, subset_rm->rm_cpumap[i], offset));
3879         }
3880 #ifdef INVARIANTS
3881         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3882                 int j;
3883
3884                 for (j = 0; j < rm->rm_cnt; ++j) {
3885                         if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3886                                 break;
3887                 }
3888                 KASSERT(j < rm->rm_cnt,
3889                     ("subset cpumap[%d] = %d not found in superset",
3890                      i, subset_rm->rm_cpumap[i]));
3891         }
3892 #endif
3893 }
3894
3895 int
3896 if_ringmap_count(const struct if_ringmap *rm)
3897 {
3898
3899         return (rm->rm_cnt);
3900 }
3901
3902 int
3903 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3904 {
3905
3906         KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3907         return (rm->rm_cpumap[ring]);
3908 }
3909
3910 void
3911 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3912 {
3913         int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3914
3915         KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3916             ("invalid redirect table entries %d", table_nent));
3917
3918         grid_idx = 0;
3919         for (i = 0; i < NETISR_CPUMAX; ++i) {
3920                 table[i] = grid_idx++ % rm->rm_cnt;
3921
3922                 if (grid_idx == rm->rm_grid)
3923                         grid_idx = 0;
3924         }
3925
3926         /*
3927          * Make the ring distributed more evenly for the remainder
3928          * of each grid.
3929          *
3930          * e.g. 12 netisrs, rm contains 8 rings.
3931          *
3932          * Redirect table before:
3933          *
3934          *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3935          *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3936          *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3937          *  ....
3938          *
3939          * Redirect table after being patched (pX, patched entries):
3940          *
3941          *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3942          *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3943          * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3944          *  ....
3945          */
3946         patch_cnt = rm->rm_grid % rm->rm_cnt;
3947         if (patch_cnt == 0)
3948                 goto done;
3949         patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3950
3951         grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3952         grid_idx = 0;
3953         for (i = 0; i < grid_cnt; ++i) {
3954                 int j;
3955
3956                 for (j = 0; j < patch_cnt; ++j) {
3957                         int fix_idx;
3958
3959                         fix_idx = (i * rm->rm_grid) + patch_off + j;
3960                         if (fix_idx >= NETISR_CPUMAX)
3961                                 goto done;
3962                         table[fix_idx] = grid_idx++ % rm->rm_cnt;
3963                 }
3964         }
3965 done:
3966         /*
3967          * If the device supports larger redirect table, duplicate
3968          * the first NETISR_CPUMAX entries to the rest of the table,
3969          * so that it matches upper layer's expectation:
3970          * (hash & NETISR_CPUMASK) % netisr_ncpus
3971          */
3972         ncopy = table_nent / NETISR_CPUMAX;
3973         for (i = 1; i < ncopy; ++i) {
3974                 memcpy(&table[i * NETISR_CPUMAX], table,
3975                     NETISR_CPUMAX * sizeof(table[0]));
3976         }
3977         if (if_ringmap_dumprdr) {
3978                 for (i = 0; i < table_nent; ++i) {
3979                         if (i != 0 && i % 16 == 0)
3980                                 kprintf("\n");
3981                         kprintf("%03d ", table[i]);
3982                 }
3983                 kprintf("\n");
3984         }
3985 }
3986
3987 int
3988 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
3989 {
3990         struct if_ringmap *rm = arg1;
3991         int i, error = 0;
3992
3993         for (i = 0; i < rm->rm_cnt; ++i) {
3994                 int cpu = rm->rm_cpumap[i];
3995
3996                 error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
3997                 if (error)
3998                         break;
3999         }
4000         return (error);
4001 }