b3d8cfb302a45ddf823d2ab4880e42d57bd807df
[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         rt_ifmsg(ifp);
1671
1672         /*
1673          * The ifaddr processing in the following loop will block,
1674          * however, this function is called in netisr0, in which
1675          * ifaddr list changes happen, so we don't care about the
1676          * blockness of the ifaddr processing here.
1677          */
1678         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1679                 struct ifaddr *ifa = ifac->ifa;
1680
1681                 /* Ignore marker */
1682                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1683                         continue;
1684
1685                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1686                         kpfctlinput(PRC_IFDOWN, ifa->ifa_addr);
1687         }
1688
1689         ifq_purge_all(&ifp->if_snd);
1690         netisr_replymsg(&nmsg->base, 0);
1691 }
1692
1693 static void
1694 if_unroute(struct ifnet *ifp, int flag, int fam)
1695 {
1696         struct netmsg_ifroute msg;
1697
1698         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1699             if_unroute_dispatch);
1700         msg.ifp = ifp;
1701         msg.flag = flag;
1702         msg.fam = fam;
1703         netisr_domsg(&msg.base, 0);
1704 }
1705
1706 /*
1707  * Mark an interface up and notify protocols of the transition.
1708  */
1709 static void
1710 if_route_dispatch(netmsg_t nmsg)
1711 {
1712         struct netmsg_ifroute *msg = (struct netmsg_ifroute *)nmsg;
1713         struct ifnet *ifp = msg->ifp;
1714         int flag = msg->flag, fam = msg->fam;
1715         struct ifaddr_container *ifac;
1716
1717         ASSERT_NETISR0;
1718
1719         ifq_purge_all(&ifp->if_snd);
1720         ifp->if_flags |= flag;
1721         getmicrotime(&ifp->if_lastchange);
1722         rt_ifmsg(ifp);
1723
1724         /*
1725          * The ifaddr processing in the following loop will block,
1726          * however, this function is called in netisr0, in which
1727          * ifaddr list changes happen, so we don't care about the
1728          * blockness of the ifaddr processing here.
1729          */
1730         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
1731                 struct ifaddr *ifa = ifac->ifa;
1732
1733                 /* Ignore marker */
1734                 if (ifa->ifa_addr->sa_family == AF_UNSPEC)
1735                         continue;
1736
1737                 if (fam == PF_UNSPEC || (fam == ifa->ifa_addr->sa_family))
1738                         kpfctlinput(PRC_IFUP, ifa->ifa_addr);
1739         }
1740 #ifdef INET6
1741         in6_if_up(ifp);
1742 #endif
1743
1744         netisr_replymsg(&nmsg->base, 0);
1745 }
1746
1747 static void
1748 if_route(struct ifnet *ifp, int flag, int fam)
1749 {
1750         struct netmsg_ifroute msg;
1751
1752         netmsg_init(&msg.base, NULL, &curthread->td_msgport, 0,
1753             if_route_dispatch);
1754         msg.ifp = ifp;
1755         msg.flag = flag;
1756         msg.fam = fam;
1757         netisr_domsg(&msg.base, 0);
1758 }
1759
1760 /*
1761  * Mark an interface down and notify protocols of the transition.  An
1762  * interface going down is also considered to be a synchronizing event.
1763  * We must ensure that all packet processing related to the interface
1764  * has completed before we return so e.g. the caller can free the ifnet
1765  * structure that the mbufs may be referencing.
1766  *
1767  * NOTE: must be called at splnet or eqivalent.
1768  */
1769 void
1770 if_down(struct ifnet *ifp)
1771 {
1772         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_DOWN);
1773         if_unroute(ifp, IFF_UP, AF_UNSPEC);
1774         netmsg_service_sync();
1775 }
1776
1777 /*
1778  * Mark an interface up and notify protocols of
1779  * the transition.
1780  * NOTE: must be called at splnet or eqivalent.
1781  */
1782 void
1783 if_up(struct ifnet *ifp)
1784 {
1785         if_route(ifp, IFF_UP, AF_UNSPEC);
1786         EVENTHANDLER_INVOKE(ifnet_event, ifp, IFNET_EVENT_UP);
1787 }
1788
1789 /*
1790  * Process a link state change.
1791  * NOTE: must be called at splsoftnet or equivalent.
1792  */
1793 void
1794 if_link_state_change(struct ifnet *ifp)
1795 {
1796         int link_state = ifp->if_link_state;
1797
1798         rt_ifmsg(ifp);
1799         devctl_notify("IFNET", ifp->if_xname,
1800             (link_state == LINK_STATE_UP) ? "LINK_UP" : "LINK_DOWN", NULL);
1801
1802         EVENTHANDLER_INVOKE(ifnet_link_event, ifp, link_state);
1803 }
1804
1805 /*
1806  * Handle interface watchdog timer routines.  Called
1807  * from softclock, we decrement timers (if set) and
1808  * call the appropriate interface routine on expiration.
1809  */
1810 static void
1811 if_slowtimo_dispatch(netmsg_t nmsg)
1812 {
1813         struct globaldata *gd = mycpu;
1814         const struct ifnet_array *arr;
1815         int i;
1816
1817         ASSERT_NETISR0;
1818
1819         crit_enter_gd(gd);
1820         lwkt_replymsg(&nmsg->lmsg, 0);  /* reply ASAP */
1821         crit_exit_gd(gd);
1822
1823         arr = ifnet_array_get();
1824         for (i = 0; i < arr->ifnet_count; ++i) {
1825                 struct ifnet *ifp = arr->ifnet_arr[i];
1826
1827                 crit_enter_gd(gd);
1828
1829                 if (if_stats_compat) {
1830                         IFNET_STAT_GET(ifp, ipackets, ifp->if_ipackets);
1831                         IFNET_STAT_GET(ifp, ierrors, ifp->if_ierrors);
1832                         IFNET_STAT_GET(ifp, opackets, ifp->if_opackets);
1833                         IFNET_STAT_GET(ifp, oerrors, ifp->if_oerrors);
1834                         IFNET_STAT_GET(ifp, collisions, ifp->if_collisions);
1835                         IFNET_STAT_GET(ifp, ibytes, ifp->if_ibytes);
1836                         IFNET_STAT_GET(ifp, obytes, ifp->if_obytes);
1837                         IFNET_STAT_GET(ifp, imcasts, ifp->if_imcasts);
1838                         IFNET_STAT_GET(ifp, omcasts, ifp->if_omcasts);
1839                         IFNET_STAT_GET(ifp, iqdrops, ifp->if_iqdrops);
1840                         IFNET_STAT_GET(ifp, noproto, ifp->if_noproto);
1841                         IFNET_STAT_GET(ifp, oqdrops, ifp->if_oqdrops);
1842                 }
1843
1844                 if (ifp->if_timer == 0 || --ifp->if_timer) {
1845                         crit_exit_gd(gd);
1846                         continue;
1847                 }
1848                 if (ifp->if_watchdog) {
1849                         if (ifnet_tryserialize_all(ifp)) {
1850                                 (*ifp->if_watchdog)(ifp);
1851                                 ifnet_deserialize_all(ifp);
1852                         } else {
1853                                 /* try again next timeout */
1854                                 ++ifp->if_timer;
1855                         }
1856                 }
1857
1858                 crit_exit_gd(gd);
1859         }
1860
1861         callout_reset(&if_slowtimo_timer, hz / IFNET_SLOWHZ, if_slowtimo, NULL);
1862 }
1863
1864 static void
1865 if_slowtimo(void *arg __unused)
1866 {
1867         struct lwkt_msg *lmsg = &if_slowtimo_netmsg.lmsg;
1868
1869         KASSERT(mycpuid == 0, ("not on cpu0"));
1870         crit_enter();
1871         if (lmsg->ms_flags & MSGF_DONE)
1872                 lwkt_sendmsg_oncpu(netisr_cpuport(0), lmsg);
1873         crit_exit();
1874 }
1875
1876 /*
1877  * Map interface name to
1878  * interface structure pointer.
1879  */
1880 struct ifnet *
1881 ifunit(const char *name)
1882 {
1883         struct ifnet *ifp;
1884
1885         /*
1886          * Search all the interfaces for this name/number
1887          */
1888         KASSERT(mtx_owned(&ifnet_mtx), ("ifnet is not locked"));
1889
1890         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
1891                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1892                         break;
1893         }
1894         return (ifp);
1895 }
1896
1897 struct ifnet *
1898 ifunit_netisr(const char *name)
1899 {
1900         const struct ifnet_array *arr;
1901         int i;
1902
1903         /*
1904          * Search all the interfaces for this name/number
1905          */
1906
1907         arr = ifnet_array_get();
1908         for (i = 0; i < arr->ifnet_count; ++i) {
1909                 struct ifnet *ifp = arr->ifnet_arr[i];
1910
1911                 if (strncmp(ifp->if_xname, name, IFNAMSIZ) == 0)
1912                         return ifp;
1913         }
1914         return NULL;
1915 }
1916
1917 /*
1918  * Interface ioctls.
1919  */
1920 int
1921 ifioctl(struct socket *so, u_long cmd, caddr_t data, struct ucred *cred)
1922 {
1923         struct ifnet *ifp;
1924         struct ifgroupreq *ifgr;
1925         struct ifreq *ifr;
1926         struct ifstat *ifs;
1927         int error, do_ifup = 0;
1928         short oif_flags;
1929         int new_flags;
1930         size_t namelen, onamelen;
1931         char new_name[IFNAMSIZ];
1932         struct ifaddr *ifa;
1933         struct sockaddr_dl *sdl;
1934
1935         switch (cmd) {
1936         case SIOCGIFCONF:
1937                 return (ifconf(cmd, data, cred));
1938         default:
1939                 break;
1940         }
1941
1942         ifr = (struct ifreq *)data;
1943
1944         switch (cmd) {
1945         case SIOCIFCREATE:
1946         case SIOCIFCREATE2:
1947                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1948                         return (error);
1949                 return (if_clone_create(ifr->ifr_name, sizeof(ifr->ifr_name),
1950                         cmd == SIOCIFCREATE2 ? ifr->ifr_data : NULL));
1951         case SIOCIFDESTROY:
1952                 if ((error = priv_check_cred(cred, PRIV_ROOT, 0)) != 0)
1953                         return (error);
1954                 return (if_clone_destroy(ifr->ifr_name));
1955         case SIOCIFGCLONERS:
1956                 return (if_clone_list((struct if_clonereq *)data));
1957         case SIOCGIFGMEMB:
1958                 return (if_getgroupmembers((struct ifgroupreq *)data));
1959         default:
1960                 break;
1961         }
1962
1963         /*
1964          * Nominal ioctl through interface, lookup the ifp and obtain a
1965          * lock to serialize the ifconfig ioctl operation.
1966          */
1967         ifnet_lock();
1968
1969         ifp = ifunit(ifr->ifr_name);
1970         if (ifp == NULL) {
1971                 ifnet_unlock();
1972                 return (ENXIO);
1973         }
1974         error = 0;
1975
1976         switch (cmd) {
1977         case SIOCGIFINDEX:
1978                 ifr->ifr_index = ifp->if_index;
1979                 break;
1980
1981         case SIOCGIFFLAGS:
1982                 ifr->ifr_flags = ifp->if_flags;
1983                 ifr->ifr_flagshigh = ifp->if_flags >> 16;
1984                 break;
1985
1986         case SIOCGIFCAP:
1987                 ifr->ifr_reqcap = ifp->if_capabilities;
1988                 ifr->ifr_curcap = ifp->if_capenable;
1989                 break;
1990
1991         case SIOCGIFMETRIC:
1992                 ifr->ifr_metric = ifp->if_metric;
1993                 break;
1994
1995         case SIOCGIFMTU:
1996                 ifr->ifr_mtu = ifp->if_mtu;
1997                 break;
1998
1999         case SIOCGIFTSOLEN:
2000                 ifr->ifr_tsolen = ifp->if_tsolen;
2001                 break;
2002
2003         case SIOCGIFDATA:
2004                 error = copyout((caddr_t)&ifp->if_data, ifr->ifr_data,
2005                                 sizeof(ifp->if_data));
2006                 break;
2007
2008         case SIOCGIFPHYS:
2009                 ifr->ifr_phys = ifp->if_physical;
2010                 break;
2011
2012         case SIOCGIFPOLLCPU:
2013                 ifr->ifr_pollcpu = -1;
2014                 break;
2015
2016         case SIOCSIFPOLLCPU:
2017                 break;
2018
2019         case SIOCSIFFLAGS:
2020                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2021                 if (error)
2022                         break;
2023                 new_flags = (ifr->ifr_flags & 0xffff) |
2024                     (ifr->ifr_flagshigh << 16);
2025                 if (ifp->if_flags & IFF_SMART) {
2026                         /* Smart drivers twiddle their own routes */
2027                 } else if (ifp->if_flags & IFF_UP &&
2028                     (new_flags & IFF_UP) == 0) {
2029                         if_down(ifp);
2030                 } else if (new_flags & IFF_UP &&
2031                     (ifp->if_flags & IFF_UP) == 0) {
2032                         do_ifup = 1;
2033                 }
2034
2035 #ifdef IFPOLL_ENABLE
2036                 if ((new_flags ^ ifp->if_flags) & IFF_NPOLLING) {
2037                         if (new_flags & IFF_NPOLLING)
2038                                 ifpoll_register(ifp);
2039                         else
2040                                 ifpoll_deregister(ifp);
2041                 }
2042 #endif
2043
2044                 ifp->if_flags = (ifp->if_flags & IFF_CANTCHANGE) |
2045                         (new_flags &~ IFF_CANTCHANGE);
2046                 if (new_flags & IFF_PPROMISC) {
2047                         /* Permanently promiscuous mode requested */
2048                         ifp->if_flags |= IFF_PROMISC;
2049                 } else if (ifp->if_pcount == 0) {
2050                         ifp->if_flags &= ~IFF_PROMISC;
2051                 }
2052                 if (ifp->if_ioctl) {
2053                         ifnet_serialize_all(ifp);
2054                         ifp->if_ioctl(ifp, cmd, data, cred);
2055                         ifnet_deserialize_all(ifp);
2056                 }
2057                 if (do_ifup)
2058                         if_up(ifp);
2059                 getmicrotime(&ifp->if_lastchange);
2060                 break;
2061
2062         case SIOCSIFCAP:
2063                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2064                 if (error)
2065                         break;
2066                 if (ifr->ifr_reqcap & ~ifp->if_capabilities) {
2067                         error = EINVAL;
2068                         break;
2069                 }
2070                 ifnet_serialize_all(ifp);
2071                 ifp->if_ioctl(ifp, cmd, data, cred);
2072                 ifnet_deserialize_all(ifp);
2073                 break;
2074
2075         case SIOCSIFNAME:
2076                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2077                 if (error)
2078                         break;
2079                 error = copyinstr(ifr->ifr_data, new_name, IFNAMSIZ, NULL);
2080                 if (error)
2081                         break;
2082                 if (new_name[0] == '\0') {
2083                         error = EINVAL;
2084                         break;
2085                 }
2086                 if (ifunit(new_name) != NULL) {
2087                         error = EEXIST;
2088                         break;
2089                 }
2090
2091                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
2092
2093                 /* Announce the departure of the interface. */
2094                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
2095
2096                 strlcpy(ifp->if_xname, new_name, sizeof(ifp->if_xname));
2097                 ifa = TAILQ_FIRST(&ifp->if_addrheads[mycpuid])->ifa;
2098                 sdl = (struct sockaddr_dl *)ifa->ifa_addr;
2099                 namelen = strlen(new_name);
2100                 onamelen = sdl->sdl_nlen;
2101                 /*
2102                  * Move the address if needed.  This is safe because we
2103                  * allocate space for a name of length IFNAMSIZ when we
2104                  * create this in if_attach().
2105                  */
2106                 if (namelen != onamelen) {
2107                         bcopy(sdl->sdl_data + onamelen,
2108                             sdl->sdl_data + namelen, sdl->sdl_alen);
2109                 }
2110                 bcopy(new_name, sdl->sdl_data, namelen);
2111                 sdl->sdl_nlen = namelen;
2112                 sdl = (struct sockaddr_dl *)ifa->ifa_netmask;
2113                 bzero(sdl->sdl_data, onamelen);
2114                 while (namelen != 0)
2115                         sdl->sdl_data[--namelen] = 0xff;
2116
2117                 EVENTHANDLER_INVOKE(ifnet_attach_event, ifp);
2118
2119                 /* Announce the return of the interface. */
2120                 rt_ifannouncemsg(ifp, IFAN_ARRIVAL);
2121                 break;
2122
2123         case SIOCSIFMETRIC:
2124                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2125                 if (error)
2126                         break;
2127                 ifp->if_metric = ifr->ifr_metric;
2128                 getmicrotime(&ifp->if_lastchange);
2129                 break;
2130
2131         case SIOCSIFPHYS:
2132                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2133                 if (error)
2134                         break;
2135                 if (ifp->if_ioctl == NULL) {
2136                         error = EOPNOTSUPP;
2137                         break;
2138                 }
2139                 ifnet_serialize_all(ifp);
2140                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2141                 ifnet_deserialize_all(ifp);
2142                 if (error == 0)
2143                         getmicrotime(&ifp->if_lastchange);
2144                 break;
2145
2146         case SIOCSIFMTU:
2147         {
2148                 u_long oldmtu = ifp->if_mtu;
2149
2150                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2151                 if (error)
2152                         break;
2153                 if (ifp->if_ioctl == NULL) {
2154                         error = EOPNOTSUPP;
2155                         break;
2156                 }
2157                 if (ifr->ifr_mtu < IF_MINMTU || ifr->ifr_mtu > IF_MAXMTU) {
2158                         error = EINVAL;
2159                         break;
2160                 }
2161                 ifnet_serialize_all(ifp);
2162                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2163                 ifnet_deserialize_all(ifp);
2164                 if (error == 0) {
2165                         getmicrotime(&ifp->if_lastchange);
2166                         rt_ifmsg(ifp);
2167                 }
2168                 /*
2169                  * If the link MTU changed, do network layer specific procedure.
2170                  */
2171                 if (ifp->if_mtu != oldmtu) {
2172 #ifdef INET6
2173                         nd6_setmtu(ifp);
2174 #endif
2175                 }
2176                 break;
2177         }
2178
2179         case SIOCSIFTSOLEN:
2180                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2181                 if (error)
2182                         break;
2183
2184                 /* XXX need driver supplied upper limit */
2185                 if (ifr->ifr_tsolen <= 0) {
2186                         error = EINVAL;
2187                         break;
2188                 }
2189                 ifp->if_tsolen = ifr->ifr_tsolen;
2190                 break;
2191
2192         case SIOCADDMULTI:
2193         case SIOCDELMULTI:
2194                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2195                 if (error)
2196                         break;
2197
2198                 /* Don't allow group membership on non-multicast interfaces. */
2199                 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
2200                         error = EOPNOTSUPP;
2201                         break;
2202                 }
2203
2204                 /* Don't let users screw up protocols' entries. */
2205                 if (ifr->ifr_addr.sa_family != AF_LINK) {
2206                         error = EINVAL;
2207                         break;
2208                 }
2209
2210                 if (cmd == SIOCADDMULTI) {
2211                         struct ifmultiaddr *ifma;
2212                         error = if_addmulti(ifp, &ifr->ifr_addr, &ifma);
2213                 } else {
2214                         error = if_delmulti(ifp, &ifr->ifr_addr);
2215                 }
2216                 if (error == 0)
2217                         getmicrotime(&ifp->if_lastchange);
2218                 break;
2219
2220         case SIOCSIFPHYADDR:
2221         case SIOCDIFPHYADDR:
2222 #ifdef INET6
2223         case SIOCSIFPHYADDR_IN6:
2224 #endif
2225         case SIOCSLIFPHYADDR:
2226         case SIOCSIFMEDIA:
2227         case SIOCSIFGENERIC:
2228                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2229                 if (error)
2230                         break;
2231                 if (ifp->if_ioctl == NULL) {
2232                         error = EOPNOTSUPP;
2233                         break;
2234                 }
2235                 ifnet_serialize_all(ifp);
2236                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2237                 ifnet_deserialize_all(ifp);
2238                 if (error == 0)
2239                         getmicrotime(&ifp->if_lastchange);
2240                 break;
2241
2242         case SIOCGIFSTATUS:
2243                 ifs = (struct ifstat *)data;
2244                 ifs->ascii[0] = '\0';
2245                 /* fall through */
2246         case SIOCGIFPSRCADDR:
2247         case SIOCGIFPDSTADDR:
2248         case SIOCGLIFPHYADDR:
2249         case SIOCGIFMEDIA:
2250         case SIOCGIFGENERIC:
2251                 if (ifp->if_ioctl == NULL) {
2252                         error = EOPNOTSUPP;
2253                         break;
2254                 }
2255                 ifnet_serialize_all(ifp);
2256                 error = ifp->if_ioctl(ifp, cmd, data, cred);
2257                 ifnet_deserialize_all(ifp);
2258                 break;
2259
2260         case SIOCSIFLLADDR:
2261                 error = priv_check_cred(cred, PRIV_ROOT, 0);
2262                 if (error)
2263                         break;
2264                 error = if_setlladdr(ifp, ifr->ifr_addr.sa_data,
2265                                      ifr->ifr_addr.sa_len);
2266                 EVENTHANDLER_INVOKE(iflladdr_event, ifp);
2267                 break;
2268
2269         case SIOCAIFGROUP:
2270                 ifgr = (struct ifgroupreq *)ifr;
2271                 if ((error = priv_check_cred(cred, PRIV_NET_ADDIFGROUP, 0)))
2272                         return (error);
2273                 if ((error = if_addgroup(ifp, ifgr->ifgr_group)))
2274                         return (error);
2275                 break;
2276
2277         case SIOCDIFGROUP:
2278                 ifgr = (struct ifgroupreq *)ifr;
2279                 if ((error = priv_check_cred(cred, PRIV_NET_DELIFGROUP, 0)))
2280                         return (error);
2281                 if ((error = if_delgroup(ifp, ifgr->ifgr_group)))
2282                         return (error);
2283                 break;
2284
2285         case SIOCGIFGROUP:
2286                 ifgr = (struct ifgroupreq *)ifr;
2287                 if ((error = if_getgroups(ifgr, ifp)))
2288                         return (error);
2289                 break;
2290
2291         default:
2292                 oif_flags = ifp->if_flags;
2293                 if (so->so_proto == 0) {
2294                         error = EOPNOTSUPP;
2295                         break;
2296                 }
2297                 error = so_pru_control_direct(so, cmd, data, ifp);
2298
2299                 /*
2300                  * If the socket control method returns EOPNOTSUPP, pass the
2301                  * request directly to the interface.
2302                  *
2303                  * Exclude the SIOCSIF{ADDR,BRDADDR,DSTADDR,NETMASK} ioctls,
2304                  * because drivers may trust these ioctls to come from an
2305                  * already privileged layer and thus do not perform credentials
2306                  * checks or input validation.
2307                  */
2308                 if (error == EOPNOTSUPP &&
2309                     ifp->if_ioctl != NULL &&
2310                     cmd != SIOCSIFADDR &&
2311                     cmd != SIOCSIFBRDADDR &&
2312                     cmd != SIOCSIFDSTADDR &&
2313                     cmd != SIOCSIFNETMASK) {
2314                         ifnet_serialize_all(ifp);
2315                         error = ifp->if_ioctl(ifp, cmd, data, cred);
2316                         ifnet_deserialize_all(ifp);
2317                 }
2318
2319                 if ((oif_flags ^ ifp->if_flags) & IFF_UP) {
2320 #ifdef INET6
2321                         DELAY(100);/* XXX: temporary workaround for fxp issue*/
2322                         if (ifp->if_flags & IFF_UP) {
2323                                 crit_enter();
2324                                 in6_if_up(ifp);
2325                                 crit_exit();
2326                         }
2327 #endif
2328                 }
2329                 break;
2330         }
2331
2332         ifnet_unlock();
2333         return (error);
2334 }
2335
2336 /*
2337  * Set/clear promiscuous mode on interface ifp based on the truth value
2338  * of pswitch.  The calls are reference counted so that only the first
2339  * "on" request actually has an effect, as does the final "off" request.
2340  * Results are undefined if the "off" and "on" requests are not matched.
2341  */
2342 int
2343 ifpromisc(struct ifnet *ifp, int pswitch)
2344 {
2345         struct ifreq ifr;
2346         int error;
2347         int oldflags;
2348
2349         oldflags = ifp->if_flags;
2350         if (ifp->if_flags & IFF_PPROMISC) {
2351                 /* Do nothing if device is in permanently promiscuous mode */
2352                 ifp->if_pcount += pswitch ? 1 : -1;
2353                 return (0);
2354         }
2355         if (pswitch) {
2356                 /*
2357                  * If the device is not configured up, we cannot put it in
2358                  * promiscuous mode.
2359                  */
2360                 if ((ifp->if_flags & IFF_UP) == 0)
2361                         return (ENETDOWN);
2362                 if (ifp->if_pcount++ != 0)
2363                         return (0);
2364                 ifp->if_flags |= IFF_PROMISC;
2365                 log(LOG_INFO, "%s: promiscuous mode enabled\n",
2366                     ifp->if_xname);
2367         } else {
2368                 if (--ifp->if_pcount > 0)
2369                         return (0);
2370                 ifp->if_flags &= ~IFF_PROMISC;
2371                 log(LOG_INFO, "%s: promiscuous mode disabled\n",
2372                     ifp->if_xname);
2373         }
2374         ifr.ifr_flags = ifp->if_flags;
2375         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2376         ifnet_serialize_all(ifp);
2377         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr, NULL);
2378         ifnet_deserialize_all(ifp);
2379         if (error == 0)
2380                 rt_ifmsg(ifp);
2381         else
2382                 ifp->if_flags = oldflags;
2383         return error;
2384 }
2385
2386 /*
2387  * Return interface configuration
2388  * of system.  List may be used
2389  * in later ioctl's (above) to get
2390  * other information.
2391  */
2392 static int
2393 ifconf(u_long cmd, caddr_t data, struct ucred *cred)
2394 {
2395         struct ifconf *ifc = (struct ifconf *)data;
2396         struct ifnet *ifp;
2397         struct sockaddr *sa;
2398         struct ifreq ifr, *ifrp;
2399         int space = ifc->ifc_len, error = 0;
2400
2401         ifrp = ifc->ifc_req;
2402
2403         ifnet_lock();
2404         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2405                 struct ifaddr_container *ifac, *ifac_mark;
2406                 struct ifaddr_marker mark;
2407                 struct ifaddrhead *head;
2408                 int addrs;
2409
2410                 if (space <= sizeof ifr)
2411                         break;
2412
2413                 /*
2414                  * Zero the stack declared structure first to prevent
2415                  * memory disclosure.
2416                  */
2417                 bzero(&ifr, sizeof(ifr));
2418                 if (strlcpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name))
2419                     >= sizeof(ifr.ifr_name)) {
2420                         error = ENAMETOOLONG;
2421                         break;
2422                 }
2423
2424                 /*
2425                  * Add a marker, since copyout() could block and during that
2426                  * period the list could be changed.  Inserting the marker to
2427                  * the header of the list will not cause trouble for the code
2428                  * assuming that the first element of the list is AF_LINK; the
2429                  * marker will be moved to the next position w/o blocking.
2430                  */
2431                 ifa_marker_init(&mark, ifp);
2432                 ifac_mark = &mark.ifac;
2433                 head = &ifp->if_addrheads[mycpuid];
2434
2435                 addrs = 0;
2436                 TAILQ_INSERT_HEAD(head, ifac_mark, ifa_link);
2437                 while ((ifac = TAILQ_NEXT(ifac_mark, ifa_link)) != NULL) {
2438                         struct ifaddr *ifa = ifac->ifa;
2439
2440                         TAILQ_REMOVE(head, ifac_mark, ifa_link);
2441                         TAILQ_INSERT_AFTER(head, ifac, ifac_mark, ifa_link);
2442
2443                         /* Ignore marker */
2444                         if (ifa->ifa_addr->sa_family == AF_UNSPEC)
2445                                 continue;
2446
2447                         if (space <= sizeof ifr)
2448                                 break;
2449                         sa = ifa->ifa_addr;
2450                         if (cred->cr_prison &&
2451                             prison_if(cred, sa))
2452                                 continue;
2453                         addrs++;
2454                         /*
2455                          * Keep a reference on this ifaddr, so that it will
2456                          * not be destroyed when its address is copied to
2457                          * the userland, which could block.
2458                          */
2459                         IFAREF(ifa);
2460                         if (sa->sa_len <= sizeof(*sa)) {
2461                                 ifr.ifr_addr = *sa;
2462                                 error = copyout(&ifr, ifrp, sizeof ifr);
2463                                 ifrp++;
2464                         } else {
2465                                 if (space < (sizeof ifr) + sa->sa_len -
2466                                             sizeof(*sa)) {
2467                                         IFAFREE(ifa);
2468                                         break;
2469                                 }
2470                                 space -= sa->sa_len - sizeof(*sa);
2471                                 error = copyout(&ifr, ifrp,
2472                                                 sizeof ifr.ifr_name);
2473                                 if (error == 0)
2474                                         error = copyout(sa, &ifrp->ifr_addr,
2475                                                         sa->sa_len);
2476                                 ifrp = (struct ifreq *)
2477                                         (sa->sa_len + (caddr_t)&ifrp->ifr_addr);
2478                         }
2479                         IFAFREE(ifa);
2480                         if (error)
2481                                 break;
2482                         space -= sizeof ifr;
2483                 }
2484                 TAILQ_REMOVE(head, ifac_mark, ifa_link);
2485                 if (error)
2486                         break;
2487                 if (!addrs) {
2488                         bzero(&ifr.ifr_addr, sizeof ifr.ifr_addr);
2489                         error = copyout(&ifr, ifrp, sizeof ifr);
2490                         if (error)
2491                                 break;
2492                         space -= sizeof ifr;
2493                         ifrp++;
2494                 }
2495         }
2496         ifnet_unlock();
2497
2498         ifc->ifc_len -= space;
2499         return (error);
2500 }
2501
2502 /*
2503  * Just like if_promisc(), but for all-multicast-reception mode.
2504  */
2505 int
2506 if_allmulti(struct ifnet *ifp, int onswitch)
2507 {
2508         int error = 0;
2509         struct ifreq ifr;
2510
2511         crit_enter();
2512
2513         if (onswitch) {
2514                 if (ifp->if_amcount++ == 0) {
2515                         ifp->if_flags |= IFF_ALLMULTI;
2516                         ifr.ifr_flags = ifp->if_flags;
2517                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2518                         ifnet_serialize_all(ifp);
2519                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2520                                               NULL);
2521                         ifnet_deserialize_all(ifp);
2522                 }
2523         } else {
2524                 if (ifp->if_amcount > 1) {
2525                         ifp->if_amcount--;
2526                 } else {
2527                         ifp->if_amcount = 0;
2528                         ifp->if_flags &= ~IFF_ALLMULTI;
2529                         ifr.ifr_flags = ifp->if_flags;
2530                         ifr.ifr_flagshigh = ifp->if_flags >> 16;
2531                         ifnet_serialize_all(ifp);
2532                         error = ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2533                                               NULL);
2534                         ifnet_deserialize_all(ifp);
2535                 }
2536         }
2537
2538         crit_exit();
2539
2540         if (error == 0)
2541                 rt_ifmsg(ifp);
2542         return error;
2543 }
2544
2545 /*
2546  * Add a multicast listenership to the interface in question.
2547  * The link layer provides a routine which converts
2548  */
2549 int
2550 if_addmulti_serialized(struct ifnet *ifp, struct sockaddr *sa,
2551     struct ifmultiaddr **retifma)
2552 {
2553         struct sockaddr *llsa, *dupsa;
2554         int error;
2555         struct ifmultiaddr *ifma;
2556
2557         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2558
2559         /*
2560          * If the matching multicast address already exists
2561          * then don't add a new one, just add a reference
2562          */
2563         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2564                 if (sa_equal(sa, ifma->ifma_addr)) {
2565                         ifma->ifma_refcount++;
2566                         if (retifma)
2567                                 *retifma = ifma;
2568                         return 0;
2569                 }
2570         }
2571
2572         /*
2573          * Give the link layer a chance to accept/reject it, and also
2574          * find out which AF_LINK address this maps to, if it isn't one
2575          * already.
2576          */
2577         if (ifp->if_resolvemulti) {
2578                 error = ifp->if_resolvemulti(ifp, &llsa, sa);
2579                 if (error)
2580                         return error;
2581         } else {
2582                 llsa = NULL;
2583         }
2584
2585         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2586         dupsa = kmalloc(sa->sa_len, M_IFMADDR, M_INTWAIT);
2587         bcopy(sa, dupsa, sa->sa_len);
2588
2589         ifma->ifma_addr = dupsa;
2590         ifma->ifma_lladdr = llsa;
2591         ifma->ifma_ifp = ifp;
2592         ifma->ifma_refcount = 1;
2593         ifma->ifma_protospec = NULL;
2594         rt_newmaddrmsg(RTM_NEWMADDR, ifma);
2595
2596         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2597         if (retifma)
2598                 *retifma = ifma;
2599
2600         if (llsa != NULL) {
2601                 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2602                         if (sa_equal(ifma->ifma_addr, llsa))
2603                                 break;
2604                 }
2605                 if (ifma) {
2606                         ifma->ifma_refcount++;
2607                 } else {
2608                         ifma = kmalloc(sizeof *ifma, M_IFMADDR, M_INTWAIT);
2609                         dupsa = kmalloc(llsa->sa_len, M_IFMADDR, M_INTWAIT);
2610                         bcopy(llsa, dupsa, llsa->sa_len);
2611                         ifma->ifma_addr = dupsa;
2612                         ifma->ifma_ifp = ifp;
2613                         ifma->ifma_refcount = 1;
2614                         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, ifma, ifma_link);
2615                 }
2616         }
2617         /*
2618          * We are certain we have added something, so call down to the
2619          * interface to let them know about it.
2620          */
2621         if (ifp->if_ioctl)
2622                 ifp->if_ioctl(ifp, SIOCADDMULTI, 0, NULL);
2623
2624         return 0;
2625 }
2626
2627 int
2628 if_addmulti(struct ifnet *ifp, struct sockaddr *sa,
2629     struct ifmultiaddr **retifma)
2630 {
2631         int error;
2632
2633         ifnet_serialize_all(ifp);
2634         error = if_addmulti_serialized(ifp, sa, retifma);
2635         ifnet_deserialize_all(ifp);
2636
2637         return error;
2638 }
2639
2640 /*
2641  * Remove a reference to a multicast address on this interface.  Yell
2642  * if the request does not match an existing membership.
2643  */
2644 static int
2645 if_delmulti_serialized(struct ifnet *ifp, struct sockaddr *sa)
2646 {
2647         struct ifmultiaddr *ifma;
2648
2649         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2650
2651         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2652                 if (sa_equal(sa, ifma->ifma_addr))
2653                         break;
2654         if (ifma == NULL)
2655                 return ENOENT;
2656
2657         if (ifma->ifma_refcount > 1) {
2658                 ifma->ifma_refcount--;
2659                 return 0;
2660         }
2661
2662         rt_newmaddrmsg(RTM_DELMADDR, ifma);
2663         sa = ifma->ifma_lladdr;
2664         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2665         /*
2666          * Make sure the interface driver is notified
2667          * in the case of a link layer mcast group being left.
2668          */
2669         if (ifma->ifma_addr->sa_family == AF_LINK && sa == NULL)
2670                 ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2671         kfree(ifma->ifma_addr, M_IFMADDR);
2672         kfree(ifma, M_IFMADDR);
2673         if (sa == NULL)
2674                 return 0;
2675
2676         /*
2677          * Now look for the link-layer address which corresponds to
2678          * this network address.  It had been squirreled away in
2679          * ifma->ifma_lladdr for this purpose (so we don't have
2680          * to call ifp->if_resolvemulti() again), and we saved that
2681          * value in sa above.  If some nasty deleted the
2682          * link-layer address out from underneath us, we can deal because
2683          * the address we stored was is not the same as the one which was
2684          * in the record for the link-layer address.  (So we don't complain
2685          * in that case.)
2686          */
2687         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2688                 if (sa_equal(sa, ifma->ifma_addr))
2689                         break;
2690         if (ifma == NULL)
2691                 return 0;
2692
2693         if (ifma->ifma_refcount > 1) {
2694                 ifma->ifma_refcount--;
2695                 return 0;
2696         }
2697
2698         TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
2699         ifp->if_ioctl(ifp, SIOCDELMULTI, 0, NULL);
2700         kfree(ifma->ifma_addr, M_IFMADDR);
2701         kfree(sa, M_IFMADDR);
2702         kfree(ifma, M_IFMADDR);
2703
2704         return 0;
2705 }
2706
2707 int
2708 if_delmulti(struct ifnet *ifp, struct sockaddr *sa)
2709 {
2710         int error;
2711
2712         ifnet_serialize_all(ifp);
2713         error = if_delmulti_serialized(ifp, sa);
2714         ifnet_deserialize_all(ifp);
2715
2716         return error;
2717 }
2718
2719 /*
2720  * Delete all multicast group membership for an interface.
2721  * Should be used to quickly flush all multicast filters.
2722  */
2723 void
2724 if_delallmulti_serialized(struct ifnet *ifp)
2725 {
2726         struct ifmultiaddr *ifma, mark;
2727         struct sockaddr sa;
2728
2729         ASSERT_IFNET_SERIALIZED_ALL(ifp);
2730
2731         bzero(&sa, sizeof(sa));
2732         sa.sa_family = AF_UNSPEC;
2733         sa.sa_len = sizeof(sa);
2734
2735         bzero(&mark, sizeof(mark));
2736         mark.ifma_addr = &sa;
2737
2738         TAILQ_INSERT_HEAD(&ifp->if_multiaddrs, &mark, ifma_link);
2739         while ((ifma = TAILQ_NEXT(&mark, ifma_link)) != NULL) {
2740                 TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2741                 TAILQ_INSERT_AFTER(&ifp->if_multiaddrs, ifma, &mark,
2742                     ifma_link);
2743
2744                 if (ifma->ifma_addr->sa_family == AF_UNSPEC)
2745                         continue;
2746
2747                 if_delmulti_serialized(ifp, ifma->ifma_addr);
2748         }
2749         TAILQ_REMOVE(&ifp->if_multiaddrs, &mark, ifma_link);
2750 }
2751
2752
2753 /*
2754  * Set the link layer address on an interface.
2755  *
2756  * At this time we only support certain types of interfaces,
2757  * and we don't allow the length of the address to change.
2758  */
2759 int
2760 if_setlladdr(struct ifnet *ifp, const u_char *lladdr, int len)
2761 {
2762         struct sockaddr_dl *sdl;
2763         struct ifreq ifr;
2764
2765         sdl = IF_LLSOCKADDR(ifp);
2766         if (sdl == NULL)
2767                 return (EINVAL);
2768         if (len != sdl->sdl_alen)       /* don't allow length to change */
2769                 return (EINVAL);
2770         switch (ifp->if_type) {
2771         case IFT_ETHER:                 /* these types use struct arpcom */
2772         case IFT_XETHER:
2773         case IFT_L2VLAN:
2774         case IFT_IEEE8023ADLAG:
2775                 bcopy(lladdr, ((struct arpcom *)ifp->if_softc)->ac_enaddr, len);
2776                 bcopy(lladdr, LLADDR(sdl), len);
2777                 break;
2778         default:
2779                 return (ENODEV);
2780         }
2781         /*
2782          * If the interface is already up, we need
2783          * to re-init it in order to reprogram its
2784          * address filter.
2785          */
2786         ifnet_serialize_all(ifp);
2787         if ((ifp->if_flags & IFF_UP) != 0) {
2788 #ifdef INET
2789                 struct ifaddr_container *ifac;
2790 #endif
2791
2792                 ifp->if_flags &= ~IFF_UP;
2793                 ifr.ifr_flags = ifp->if_flags;
2794                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2795                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2796                               NULL);
2797                 ifp->if_flags |= IFF_UP;
2798                 ifr.ifr_flags = ifp->if_flags;
2799                 ifr.ifr_flagshigh = ifp->if_flags >> 16;
2800                 ifp->if_ioctl(ifp, SIOCSIFFLAGS, (caddr_t)&ifr,
2801                                  NULL);
2802 #ifdef INET
2803                 /*
2804                  * Also send gratuitous ARPs to notify other nodes about
2805                  * the address change.
2806                  */
2807                 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2808                         struct ifaddr *ifa = ifac->ifa;
2809
2810                         if (ifa->ifa_addr != NULL &&
2811                             ifa->ifa_addr->sa_family == AF_INET)
2812                                 arp_gratuitous(ifp, ifa);
2813                 }
2814 #endif
2815         }
2816         ifnet_deserialize_all(ifp);
2817         return (0);
2818 }
2819
2820
2821 /*
2822  * Locate an interface based on a complete address.
2823  */
2824 struct ifnet *
2825 if_bylla(const void *lla, unsigned char lla_len)
2826 {
2827         const struct ifnet_array *arr;
2828         struct ifnet *ifp;
2829         struct sockaddr_dl *sdl;
2830         int i;
2831
2832         arr = ifnet_array_get();
2833         for (i = 0; i < arr->ifnet_count; ++i) {
2834                 ifp = arr->ifnet_arr[i];
2835                 if (ifp->if_addrlen != lla_len)
2836                         continue;
2837
2838                 sdl = IF_LLSOCKADDR(ifp);
2839                 if (memcmp(lla, LLADDR(sdl), lla_len) == 0)
2840                         return (ifp);
2841         }
2842         return (NULL);
2843 }
2844
2845 struct ifmultiaddr *
2846 ifmaof_ifpforaddr(struct sockaddr *sa, struct ifnet *ifp)
2847 {
2848         struct ifmultiaddr *ifma;
2849
2850         /* TODO: need ifnet_serialize_main */
2851         ifnet_serialize_all(ifp);
2852         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
2853                 if (sa_equal(ifma->ifma_addr, sa))
2854                         break;
2855         ifnet_deserialize_all(ifp);
2856
2857         return ifma;
2858 }
2859
2860 /*
2861  * This function locates the first real ethernet MAC from a network
2862  * card and loads it into node, returning 0 on success or ENOENT if
2863  * no suitable interfaces were found.  It is used by the uuid code to
2864  * generate a unique 6-byte number.
2865  */
2866 int
2867 if_getanyethermac(uint16_t *node, int minlen)
2868 {
2869         struct ifnet *ifp;
2870         struct sockaddr_dl *sdl;
2871
2872         ifnet_lock();
2873         TAILQ_FOREACH(ifp, &ifnetlist, if_link) {
2874                 if (ifp->if_type != IFT_ETHER)
2875                         continue;
2876                 sdl = IF_LLSOCKADDR(ifp);
2877                 if (sdl->sdl_alen < minlen)
2878                         continue;
2879                 bcopy(((struct arpcom *)ifp->if_softc)->ac_enaddr, node,
2880                       minlen);
2881                 ifnet_unlock();
2882                 return(0);
2883         }
2884         ifnet_unlock();
2885         return (ENOENT);
2886 }
2887
2888 /*
2889  * The name argument must be a pointer to storage which will last as
2890  * long as the interface does.  For physical devices, the result of
2891  * device_get_name(dev) is a good choice and for pseudo-devices a
2892  * static string works well.
2893  */
2894 void
2895 if_initname(struct ifnet *ifp, const char *name, int unit)
2896 {
2897         ifp->if_dname = name;
2898         ifp->if_dunit = unit;
2899         if (unit != IF_DUNIT_NONE)
2900                 ksnprintf(ifp->if_xname, IFNAMSIZ, "%s%d", name, unit);
2901         else
2902                 strlcpy(ifp->if_xname, name, IFNAMSIZ);
2903 }
2904
2905 int
2906 if_printf(struct ifnet *ifp, const char *fmt, ...)
2907 {
2908         __va_list ap;
2909         int retval;
2910
2911         retval = kprintf("%s: ", ifp->if_xname);
2912         __va_start(ap, fmt);
2913         retval += kvprintf(fmt, ap);
2914         __va_end(ap);
2915         return (retval);
2916 }
2917
2918 struct ifnet *
2919 if_alloc(uint8_t type)
2920 {
2921         struct ifnet *ifp;
2922         size_t size;
2923
2924         /*
2925          * XXX temporary hack until arpcom is setup in if_l2com
2926          */
2927         if (type == IFT_ETHER)
2928                 size = sizeof(struct arpcom);
2929         else
2930                 size = sizeof(struct ifnet);
2931
2932         ifp = kmalloc(size, M_IFNET, M_WAITOK|M_ZERO);
2933
2934         ifp->if_type = type;
2935
2936         if (if_com_alloc[type] != NULL) {
2937                 ifp->if_l2com = if_com_alloc[type](type, ifp);
2938                 if (ifp->if_l2com == NULL) {
2939                         kfree(ifp, M_IFNET);
2940                         return (NULL);
2941                 }
2942         }
2943         return (ifp);
2944 }
2945
2946 void
2947 if_free(struct ifnet *ifp)
2948 {
2949         kfree(ifp, M_IFNET);
2950 }
2951
2952 void
2953 ifq_set_classic(struct ifaltq *ifq)
2954 {
2955         ifq_set_methods(ifq, ifq->altq_ifp->if_mapsubq,
2956             ifsq_classic_enqueue, ifsq_classic_dequeue, ifsq_classic_request);
2957 }
2958
2959 void
2960 ifq_set_methods(struct ifaltq *ifq, altq_mapsubq_t mapsubq,
2961     ifsq_enqueue_t enqueue, ifsq_dequeue_t dequeue, ifsq_request_t request)
2962 {
2963         int q;
2964
2965         KASSERT(mapsubq != NULL, ("mapsubq is not specified"));
2966         KASSERT(enqueue != NULL, ("enqueue is not specified"));
2967         KASSERT(dequeue != NULL, ("dequeue is not specified"));
2968         KASSERT(request != NULL, ("request is not specified"));
2969
2970         ifq->altq_mapsubq = mapsubq;
2971         for (q = 0; q < ifq->altq_subq_cnt; ++q) {
2972                 struct ifaltq_subque *ifsq = &ifq->altq_subq[q];
2973
2974                 ifsq->ifsq_enqueue = enqueue;
2975                 ifsq->ifsq_dequeue = dequeue;
2976                 ifsq->ifsq_request = request;
2977         }
2978 }
2979
2980 static void
2981 ifsq_norm_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2982 {
2983
2984         classq_add(&ifsq->ifsq_norm, m);
2985         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2986 }
2987
2988 static void
2989 ifsq_prio_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m)
2990 {
2991
2992         classq_add(&ifsq->ifsq_prio, m);
2993         ALTQ_SQ_CNTR_INC(ifsq, m->m_pkthdr.len);
2994         ALTQ_SQ_PRIO_CNTR_INC(ifsq, m->m_pkthdr.len);
2995 }
2996
2997 static struct mbuf *
2998 ifsq_norm_dequeue(struct ifaltq_subque *ifsq)
2999 {
3000         struct mbuf *m;
3001
3002         m = classq_get(&ifsq->ifsq_norm);
3003         if (m != NULL)
3004                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
3005         return (m);
3006 }
3007
3008 static struct mbuf *
3009 ifsq_prio_dequeue(struct ifaltq_subque *ifsq)
3010 {
3011         struct mbuf *m;
3012
3013         m = classq_get(&ifsq->ifsq_prio);
3014         if (m != NULL) {
3015                 ALTQ_SQ_CNTR_DEC(ifsq, m->m_pkthdr.len);
3016                 ALTQ_SQ_PRIO_CNTR_DEC(ifsq, m->m_pkthdr.len);
3017         }
3018         return (m);
3019 }
3020
3021 int
3022 ifsq_classic_enqueue(struct ifaltq_subque *ifsq, struct mbuf *m,
3023     struct altq_pktattr *pa __unused)
3024 {
3025
3026         M_ASSERTPKTHDR(m);
3027 again:
3028         if (ifsq->ifsq_len >= ifsq->ifsq_maxlen ||
3029             ifsq->ifsq_bcnt >= ifsq->ifsq_maxbcnt) {
3030                 struct mbuf *m_drop;
3031
3032                 if (m->m_flags & M_PRIO) {
3033                         m_drop = NULL;
3034                         if (ifsq->ifsq_prio_len < (ifsq->ifsq_maxlen >> 1) &&
3035                             ifsq->ifsq_prio_bcnt < (ifsq->ifsq_maxbcnt >> 1)) {
3036                                 /* Try dropping some from normal queue. */
3037                                 m_drop = ifsq_norm_dequeue(ifsq);
3038                         }
3039                         if (m_drop == NULL)
3040                                 m_drop = ifsq_prio_dequeue(ifsq);
3041                 } else {
3042                         m_drop = ifsq_norm_dequeue(ifsq);
3043                 }
3044                 if (m_drop != NULL) {
3045                         IFNET_STAT_INC(ifsq->ifsq_ifp, oqdrops, 1);
3046                         m_freem(m_drop);
3047                         goto again;
3048                 }
3049                 /*
3050                  * No old packets could be dropped!
3051                  * NOTE: Caller increases oqdrops.
3052                  */
3053                 m_freem(m);
3054                 return (ENOBUFS);
3055         } else {
3056                 if (m->m_flags & M_PRIO)
3057                         ifsq_prio_enqueue(ifsq, m);
3058                 else
3059                         ifsq_norm_enqueue(ifsq, m);
3060                 return (0);
3061         }
3062 }
3063
3064 struct mbuf *
3065 ifsq_classic_dequeue(struct ifaltq_subque *ifsq, int op)
3066 {
3067         struct mbuf *m;
3068
3069         switch (op) {
3070         case ALTDQ_POLL:
3071                 m = classq_head(&ifsq->ifsq_prio);
3072                 if (m == NULL)
3073                         m = classq_head(&ifsq->ifsq_norm);
3074                 break;
3075
3076         case ALTDQ_REMOVE:
3077                 m = ifsq_prio_dequeue(ifsq);
3078                 if (m == NULL)
3079                         m = ifsq_norm_dequeue(ifsq);
3080                 break;
3081
3082         default:
3083                 panic("unsupported ALTQ dequeue op: %d", op);
3084         }
3085         return m;
3086 }
3087
3088 int
3089 ifsq_classic_request(struct ifaltq_subque *ifsq, int req, void *arg)
3090 {
3091         switch (req) {
3092         case ALTRQ_PURGE:
3093                 for (;;) {
3094                         struct mbuf *m;
3095
3096                         m = ifsq_classic_dequeue(ifsq, ALTDQ_REMOVE);
3097                         if (m == NULL)
3098                                 break;
3099                         m_freem(m);
3100                 }
3101                 break;
3102
3103         default:
3104                 panic("unsupported ALTQ request: %d", req);
3105         }
3106         return 0;
3107 }
3108
3109 static void
3110 ifsq_ifstart_try(struct ifaltq_subque *ifsq, int force_sched)
3111 {
3112         struct ifnet *ifp = ifsq_get_ifp(ifsq);
3113         int running = 0, need_sched;
3114
3115         /*
3116          * Try to do direct ifnet.if_start on the subqueue first, if there is
3117          * contention on the subqueue hardware serializer, ifnet.if_start on
3118          * the subqueue will be scheduled on the subqueue owner CPU.
3119          */
3120         if (!ifsq_tryserialize_hw(ifsq)) {
3121                 /*
3122                  * Subqueue hardware serializer contention happened,
3123                  * ifnet.if_start on the subqueue is scheduled on
3124                  * the subqueue owner CPU, and we keep going.
3125                  */
3126                 ifsq_ifstart_schedule(ifsq, 1);
3127                 return;
3128         }
3129
3130         if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq)) {
3131                 ifp->if_start(ifp, ifsq);
3132                 if ((ifp->if_flags & IFF_RUNNING) && !ifsq_is_oactive(ifsq))
3133                         running = 1;
3134         }
3135         need_sched = ifsq_ifstart_need_schedule(ifsq, running);
3136
3137         ifsq_deserialize_hw(ifsq);
3138
3139         if (need_sched) {
3140                 /*
3141                  * More data need to be transmitted, ifnet.if_start on the
3142                  * subqueue is scheduled on the subqueue owner CPU, and we
3143                  * keep going.
3144                  * NOTE: ifnet.if_start subqueue interlock is not released.
3145                  */
3146                 ifsq_ifstart_schedule(ifsq, force_sched);
3147         }
3148 }
3149
3150 /*
3151  * Subqeue packets staging mechanism:
3152  *
3153  * The packets enqueued into the subqueue are staged to a certain amount
3154  * before the ifnet.if_start on the subqueue is called.  In this way, the
3155  * driver could avoid writing to hardware registers upon every packet,
3156  * instead, hardware registers could be written when certain amount of
3157  * packets are put onto hardware TX ring.  The measurement on several modern
3158  * NICs (emx(4), igb(4), bnx(4), bge(4), jme(4)) shows that the hardware
3159  * registers writing aggregation could save ~20% CPU time when 18bytes UDP
3160  * datagrams are transmitted at 1.48Mpps.  The performance improvement by
3161  * hardware registers writing aggeregation is also mentioned by Luigi Rizzo's
3162  * netmap paper (http://info.iet.unipi.it/~luigi/netmap/).
3163  *
3164  * Subqueue packets staging is performed for two entry points into drivers'
3165  * transmission function:
3166  * - Direct ifnet.if_start calling on the subqueue, i.e. ifsq_ifstart_try()
3167  * - ifnet.if_start scheduling on the subqueue, i.e. ifsq_ifstart_schedule()
3168  *
3169  * Subqueue packets staging will be stopped upon any of the following
3170  * conditions:
3171  * - If the count of packets enqueued on the current CPU is great than or
3172  *   equal to ifsq_stage_cntmax. (XXX this should be per-interface)
3173  * - If the total length of packets enqueued on the current CPU is great
3174  *   than or equal to the hardware's MTU - max_protohdr.  max_protohdr is
3175  *   cut from the hardware's MTU mainly bacause a full TCP segment's size
3176  *   is usually less than hardware's MTU.
3177  * - ifsq_ifstart_schedule() is not pending on the current CPU and
3178  *   ifnet.if_start subqueue interlock (ifaltq_subq.ifsq_started) is not
3179  *   released.
3180  * - The if_start_rollup(), which is registered as low priority netisr
3181  *   rollup function, is called; probably because no more work is pending
3182  *   for netisr.
3183  *
3184  * NOTE:
3185  * Currently subqueue packet staging is only performed in netisr threads.
3186  */
3187 int
3188 ifq_dispatch(struct ifnet *ifp, struct mbuf *m, struct altq_pktattr *pa)
3189 {
3190         struct ifaltq *ifq = &ifp->if_snd;
3191         struct ifaltq_subque *ifsq;
3192         int error, start = 0, len, mcast = 0, avoid_start = 0;
3193         struct ifsubq_stage_head *head = NULL;
3194         struct ifsubq_stage *stage = NULL;
3195         struct globaldata *gd = mycpu;
3196         struct thread *td = gd->gd_curthread;
3197
3198         crit_enter_quick(td);
3199
3200         ifsq = ifq_map_subq(ifq, gd->gd_cpuid);
3201         ASSERT_ALTQ_SQ_NOT_SERIALIZED_HW(ifsq);
3202
3203         len = m->m_pkthdr.len;
3204         if (m->m_flags & M_MCAST)
3205                 mcast = 1;
3206
3207         if (td->td_type == TD_TYPE_NETISR) {
3208                 head = &ifsubq_stage_heads[mycpuid];
3209                 stage = ifsq_get_stage(ifsq, mycpuid);
3210
3211                 stage->stg_cnt++;
3212                 stage->stg_len += len;
3213                 if (stage->stg_cnt < ifsq_stage_cntmax &&
3214                     stage->stg_len < (ifp->if_mtu - max_protohdr))
3215                         avoid_start = 1;
3216         }
3217
3218         ALTQ_SQ_LOCK(ifsq);
3219         error = ifsq_enqueue_locked(ifsq, m, pa);
3220         if (error) {
3221                 IFNET_STAT_INC(ifp, oqdrops, 1);
3222                 if (!ifsq_data_ready(ifsq)) {
3223                         ALTQ_SQ_UNLOCK(ifsq);
3224                         crit_exit_quick(td);
3225                         return error;
3226                 }
3227                 avoid_start = 0;
3228         }
3229         if (!ifsq_is_started(ifsq)) {
3230                 if (avoid_start) {
3231                         ALTQ_SQ_UNLOCK(ifsq);
3232
3233                         KKASSERT(!error);
3234                         if ((stage->stg_flags & IFSQ_STAGE_FLAG_QUED) == 0)
3235                                 ifsq_stage_insert(head, stage);
3236
3237                         IFNET_STAT_INC(ifp, obytes, len);
3238                         if (mcast)
3239                                 IFNET_STAT_INC(ifp, omcasts, 1);
3240                         crit_exit_quick(td);
3241                         return error;
3242                 }
3243
3244                 /*
3245                  * Hold the subqueue interlock of ifnet.if_start
3246                  */
3247                 ifsq_set_started(ifsq);
3248                 start = 1;
3249         }
3250         ALTQ_SQ_UNLOCK(ifsq);
3251
3252         if (!error) {
3253                 IFNET_STAT_INC(ifp, obytes, len);
3254                 if (mcast)
3255                         IFNET_STAT_INC(ifp, omcasts, 1);
3256         }
3257
3258         if (stage != NULL) {
3259                 if (!start && (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)) {
3260                         KKASSERT(stage->stg_flags & IFSQ_STAGE_FLAG_QUED);
3261                         if (!avoid_start) {
3262                                 ifsq_stage_remove(head, stage);
3263                                 ifsq_ifstart_schedule(ifsq, 1);
3264                         }
3265                         crit_exit_quick(td);
3266                         return error;
3267                 }
3268
3269                 if (stage->stg_flags & IFSQ_STAGE_FLAG_QUED) {
3270                         ifsq_stage_remove(head, stage);
3271                 } else {
3272                         stage->stg_cnt = 0;
3273                         stage->stg_len = 0;
3274                 }
3275         }
3276
3277         if (!start) {
3278                 crit_exit_quick(td);
3279                 return error;
3280         }
3281
3282         ifsq_ifstart_try(ifsq, 0);
3283
3284         crit_exit_quick(td);
3285         return error;
3286 }
3287
3288 void *
3289 ifa_create(int size)
3290 {
3291         struct ifaddr *ifa;
3292         int i;
3293
3294         KASSERT(size >= sizeof(*ifa), ("ifaddr size too small"));
3295
3296         ifa = kmalloc(size, M_IFADDR, M_INTWAIT | M_ZERO);
3297
3298         /*
3299          * Make ifa_container availabel on all CPUs, since they
3300          * could be accessed by any threads.
3301          */
3302         ifa->ifa_containers =
3303                 kmalloc(ncpus * sizeof(struct ifaddr_container),
3304                         M_IFADDR,
3305                         M_INTWAIT | M_ZERO | M_CACHEALIGN);
3306
3307         ifa->ifa_ncnt = ncpus;
3308         for (i = 0; i < ncpus; ++i) {
3309                 struct ifaddr_container *ifac = &ifa->ifa_containers[i];
3310
3311                 ifac->ifa_magic = IFA_CONTAINER_MAGIC;
3312                 ifac->ifa = ifa;
3313                 ifac->ifa_refcnt = 1;
3314         }
3315 #ifdef IFADDR_DEBUG
3316         kprintf("alloc ifa %p %d\n", ifa, size);
3317 #endif
3318         return ifa;
3319 }
3320
3321 void
3322 ifac_free(struct ifaddr_container *ifac, int cpu_id)
3323 {
3324         struct ifaddr *ifa = ifac->ifa;
3325
3326         KKASSERT(ifac->ifa_magic == IFA_CONTAINER_MAGIC);
3327         KKASSERT(ifac->ifa_refcnt == 0);
3328         KASSERT(ifac->ifa_listmask == 0,
3329                 ("ifa is still on %#x lists", ifac->ifa_listmask));
3330
3331         ifac->ifa_magic = IFA_CONTAINER_DEAD;
3332
3333 #ifdef IFADDR_DEBUG_VERBOSE
3334         kprintf("try free ifa %p cpu_id %d\n", ifac->ifa, cpu_id);
3335 #endif
3336
3337         KASSERT(ifa->ifa_ncnt > 0 && ifa->ifa_ncnt <= ncpus,
3338                 ("invalid # of ifac, %d", ifa->ifa_ncnt));
3339         if (atomic_fetchadd_int(&ifa->ifa_ncnt, -1) == 1) {
3340 #ifdef IFADDR_DEBUG
3341                 kprintf("free ifa %p\n", ifa);
3342 #endif
3343                 kfree(ifa->ifa_containers, M_IFADDR);
3344                 kfree(ifa, M_IFADDR);
3345         }
3346 }
3347
3348 static void
3349 ifa_iflink_dispatch(netmsg_t nmsg)
3350 {
3351         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3352         struct ifaddr *ifa = msg->ifa;
3353         struct ifnet *ifp = msg->ifp;
3354         int cpu = mycpuid;
3355         struct ifaddr_container *ifac;
3356
3357         crit_enter();
3358
3359         ifac = &ifa->ifa_containers[cpu];
3360         ASSERT_IFAC_VALID(ifac);
3361         KASSERT((ifac->ifa_listmask & IFA_LIST_IFADDRHEAD) == 0,
3362                 ("ifaddr is on if_addrheads"));
3363
3364         ifac->ifa_listmask |= IFA_LIST_IFADDRHEAD;
3365         if (msg->tail)
3366                 TAILQ_INSERT_TAIL(&ifp->if_addrheads[cpu], ifac, ifa_link);
3367         else
3368                 TAILQ_INSERT_HEAD(&ifp->if_addrheads[cpu], ifac, ifa_link);
3369
3370         crit_exit();
3371
3372         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3373 }
3374
3375 void
3376 ifa_iflink(struct ifaddr *ifa, struct ifnet *ifp, int tail)
3377 {
3378         struct netmsg_ifaddr msg;
3379
3380         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3381                     0, ifa_iflink_dispatch);
3382         msg.ifa = ifa;
3383         msg.ifp = ifp;
3384         msg.tail = tail;
3385
3386         netisr_domsg(&msg.base, 0);
3387 }
3388
3389 static void
3390 ifa_ifunlink_dispatch(netmsg_t nmsg)
3391 {
3392         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3393         struct ifaddr *ifa = msg->ifa;
3394         struct ifnet *ifp = msg->ifp;
3395         int cpu = mycpuid;
3396         struct ifaddr_container *ifac;
3397
3398         crit_enter();
3399
3400         ifac = &ifa->ifa_containers[cpu];
3401         ASSERT_IFAC_VALID(ifac);
3402         KASSERT(ifac->ifa_listmask & IFA_LIST_IFADDRHEAD,
3403                 ("ifaddr is not on if_addrhead"));
3404
3405         TAILQ_REMOVE(&ifp->if_addrheads[cpu], ifac, ifa_link);
3406         ifac->ifa_listmask &= ~IFA_LIST_IFADDRHEAD;
3407
3408         crit_exit();
3409
3410         netisr_forwardmsg_all(&nmsg->base, cpu + 1);
3411 }
3412
3413 void
3414 ifa_ifunlink(struct ifaddr *ifa, struct ifnet *ifp)
3415 {
3416         struct netmsg_ifaddr msg;
3417
3418         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3419                     0, ifa_ifunlink_dispatch);
3420         msg.ifa = ifa;
3421         msg.ifp = ifp;
3422
3423         netisr_domsg(&msg.base, 0);
3424 }
3425
3426 static void
3427 ifa_destroy_dispatch(netmsg_t nmsg)
3428 {
3429         struct netmsg_ifaddr *msg = (struct netmsg_ifaddr *)nmsg;
3430
3431         IFAFREE(msg->ifa);
3432         netisr_forwardmsg_all(&nmsg->base, mycpuid + 1);
3433 }
3434
3435 void
3436 ifa_destroy(struct ifaddr *ifa)
3437 {
3438         struct netmsg_ifaddr msg;
3439
3440         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
3441                     0, ifa_destroy_dispatch);
3442         msg.ifa = ifa;
3443
3444         netisr_domsg(&msg.base, 0);
3445 }
3446
3447 static void
3448 if_start_rollup(void)
3449 {
3450         struct ifsubq_stage_head *head = &ifsubq_stage_heads[mycpuid];
3451         struct ifsubq_stage *stage;
3452
3453         crit_enter();
3454
3455         while ((stage = TAILQ_FIRST(&head->stg_head)) != NULL) {
3456                 struct ifaltq_subque *ifsq = stage->stg_subq;
3457                 int is_sched = 0;
3458
3459                 if (stage->stg_flags & IFSQ_STAGE_FLAG_SCHED)
3460                         is_sched = 1;
3461                 ifsq_stage_remove(head, stage);
3462
3463                 if (is_sched) {
3464                         ifsq_ifstart_schedule(ifsq, 1);
3465                 } else {
3466                         int start = 0;
3467
3468                         ALTQ_SQ_LOCK(ifsq);
3469                         if (!ifsq_is_started(ifsq)) {
3470                                 /*
3471                                  * Hold the subqueue interlock of
3472                                  * ifnet.if_start
3473                                  */
3474                                 ifsq_set_started(ifsq);
3475                                 start = 1;
3476                         }
3477                         ALTQ_SQ_UNLOCK(ifsq);
3478
3479                         if (start)
3480                                 ifsq_ifstart_try(ifsq, 1);
3481                 }
3482                 KKASSERT((stage->stg_flags &
3483                     (IFSQ_STAGE_FLAG_QUED | IFSQ_STAGE_FLAG_SCHED)) == 0);
3484         }
3485
3486         crit_exit();
3487 }
3488
3489 static void
3490 ifnetinit(void *dummy __unused)
3491 {
3492         int i;
3493
3494         /* XXX netisr_ncpus */
3495         for (i = 0; i < ncpus; ++i)
3496                 TAILQ_INIT(&ifsubq_stage_heads[i].stg_head);
3497         netisr_register_rollup(if_start_rollup, NETISR_ROLLUP_PRIO_IFSTART);
3498 }
3499
3500 void
3501 if_register_com_alloc(u_char type,
3502     if_com_alloc_t *a, if_com_free_t *f)
3503 {
3504
3505         KASSERT(if_com_alloc[type] == NULL,
3506             ("if_register_com_alloc: %d already registered", type));
3507         KASSERT(if_com_free[type] == NULL,
3508             ("if_register_com_alloc: %d free already registered", type));
3509
3510         if_com_alloc[type] = a;
3511         if_com_free[type] = f;
3512 }
3513
3514 void
3515 if_deregister_com_alloc(u_char type)
3516 {
3517
3518         KASSERT(if_com_alloc[type] != NULL,
3519             ("if_deregister_com_alloc: %d not registered", type));
3520         KASSERT(if_com_free[type] != NULL,
3521             ("if_deregister_com_alloc: %d free not registered", type));
3522         if_com_alloc[type] = NULL;
3523         if_com_free[type] = NULL;
3524 }
3525
3526 void
3527 ifq_set_maxlen(struct ifaltq *ifq, int len)
3528 {
3529         ifq->altq_maxlen = len + (ncpus * ifsq_stage_cntmax);
3530 }
3531
3532 int
3533 ifq_mapsubq_default(struct ifaltq *ifq __unused, int cpuid __unused)
3534 {
3535         return ALTQ_SUBQ_INDEX_DEFAULT;
3536 }
3537
3538 int
3539 ifq_mapsubq_modulo(struct ifaltq *ifq, int cpuid)
3540 {
3541
3542         return (cpuid % ifq->altq_subq_mappriv);
3543 }
3544
3545 static void
3546 ifsq_watchdog(void *arg)
3547 {
3548         struct ifsubq_watchdog *wd = arg;
3549         struct ifnet *ifp;
3550
3551         if (__predict_true(wd->wd_timer == 0 || --wd->wd_timer))
3552                 goto done;
3553
3554         ifp = ifsq_get_ifp(wd->wd_subq);
3555         if (ifnet_tryserialize_all(ifp)) {
3556                 wd->wd_watchdog(wd->wd_subq);
3557                 ifnet_deserialize_all(ifp);
3558         } else {
3559                 /* try again next timeout */
3560                 wd->wd_timer = 1;
3561         }
3562 done:
3563         ifsq_watchdog_reset(wd);
3564 }
3565
3566 static void
3567 ifsq_watchdog_reset(struct ifsubq_watchdog *wd)
3568 {
3569         callout_reset_bycpu(&wd->wd_callout, hz, ifsq_watchdog, wd,
3570             ifsq_get_cpuid(wd->wd_subq));
3571 }
3572
3573 void
3574 ifsq_watchdog_init(struct ifsubq_watchdog *wd, struct ifaltq_subque *ifsq,
3575     ifsq_watchdog_t watchdog)
3576 {
3577         callout_init_mp(&wd->wd_callout);
3578         wd->wd_timer = 0;
3579         wd->wd_subq = ifsq;
3580         wd->wd_watchdog = watchdog;
3581 }
3582
3583 void
3584 ifsq_watchdog_start(struct ifsubq_watchdog *wd)
3585 {
3586         wd->wd_timer = 0;
3587         ifsq_watchdog_reset(wd);
3588 }
3589
3590 void
3591 ifsq_watchdog_stop(struct ifsubq_watchdog *wd)
3592 {
3593         wd->wd_timer = 0;
3594         callout_stop(&wd->wd_callout);
3595 }
3596
3597 void
3598 ifnet_lock(void)
3599 {
3600         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3601             ("try holding ifnet lock in netisr"));
3602         mtx_lock(&ifnet_mtx);
3603 }
3604
3605 void
3606 ifnet_unlock(void)
3607 {
3608         KASSERT(curthread->td_type != TD_TYPE_NETISR,
3609             ("try holding ifnet lock in netisr"));
3610         mtx_unlock(&ifnet_mtx);
3611 }
3612
3613 static struct ifnet_array *
3614 ifnet_array_alloc(int count)
3615 {
3616         struct ifnet_array *arr;
3617
3618         arr = kmalloc(__offsetof(struct ifnet_array, ifnet_arr[count]),
3619             M_IFNET, M_WAITOK);
3620         arr->ifnet_count = count;
3621
3622         return arr;
3623 }
3624
3625 static void
3626 ifnet_array_free(struct ifnet_array *arr)
3627 {
3628         if (arr == &ifnet_array0)
3629                 return;
3630         kfree(arr, M_IFNET);
3631 }
3632
3633 static struct ifnet_array *
3634 ifnet_array_add(struct ifnet *ifp, const struct ifnet_array *old_arr)
3635 {
3636         struct ifnet_array *arr;
3637         int count, i;
3638
3639         KASSERT(old_arr->ifnet_count >= 0,
3640             ("invalid ifnet array count %d", old_arr->ifnet_count));
3641         count = old_arr->ifnet_count + 1;
3642         arr = ifnet_array_alloc(count);
3643
3644         /*
3645          * Save the old ifnet array and append this ifp to the end of
3646          * the new ifnet array.
3647          */
3648         for (i = 0; i < old_arr->ifnet_count; ++i) {
3649                 KASSERT(old_arr->ifnet_arr[i] != ifp,
3650                     ("%s is already in ifnet array", ifp->if_xname));
3651                 arr->ifnet_arr[i] = old_arr->ifnet_arr[i];
3652         }
3653         KASSERT(i == count - 1,
3654             ("add %s, ifnet array index mismatch, should be %d, but got %d",
3655              ifp->if_xname, count - 1, i));
3656         arr->ifnet_arr[i] = ifp;
3657
3658         return arr;
3659 }
3660
3661 static struct ifnet_array *
3662 ifnet_array_del(struct ifnet *ifp, const struct ifnet_array *old_arr)
3663 {
3664         struct ifnet_array *arr;
3665         int count, i, idx, found = 0;
3666
3667         KASSERT(old_arr->ifnet_count > 0,
3668             ("invalid ifnet array count %d", old_arr->ifnet_count));
3669         count = old_arr->ifnet_count - 1;
3670         arr = ifnet_array_alloc(count);
3671
3672         /*
3673          * Save the old ifnet array, but skip this ifp.
3674          */
3675         idx = 0;
3676         for (i = 0; i < old_arr->ifnet_count; ++i) {
3677                 if (old_arr->ifnet_arr[i] == ifp) {
3678                         KASSERT(!found,
3679                             ("dup %s is in ifnet array", ifp->if_xname));
3680                         found = 1;
3681                         continue;
3682                 }
3683                 KASSERT(idx < count,
3684                     ("invalid ifnet array index %d, count %d", idx, count));
3685                 arr->ifnet_arr[idx] = old_arr->ifnet_arr[i];
3686                 ++idx;
3687         }
3688         KASSERT(found, ("%s is not in ifnet array", ifp->if_xname));
3689         KASSERT(idx == count,
3690             ("del %s, ifnet array count mismatch, should be %d, but got %d ",
3691              ifp->if_xname, count, idx));
3692
3693         return arr;
3694 }
3695
3696 const struct ifnet_array *
3697 ifnet_array_get(void)
3698 {
3699         const struct ifnet_array *ret;
3700
3701         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3702         ret = ifnet_array;
3703         /* Make sure 'ret' is really used. */
3704         cpu_ccfence();
3705         return (ret);
3706 }
3707
3708 int
3709 ifnet_array_isempty(void)
3710 {
3711         KASSERT(curthread->td_type == TD_TYPE_NETISR, ("not in netisr"));
3712         if (ifnet_array->ifnet_count == 0)
3713                 return 1;
3714         else
3715                 return 0;
3716 }
3717
3718 void
3719 ifa_marker_init(struct ifaddr_marker *mark, struct ifnet *ifp)
3720 {
3721         struct ifaddr *ifa;
3722
3723         memset(mark, 0, sizeof(*mark));
3724         ifa = &mark->ifa;
3725
3726         mark->ifac.ifa = ifa;
3727
3728         ifa->ifa_addr = &mark->addr;
3729         ifa->ifa_dstaddr = &mark->dstaddr;
3730         ifa->ifa_netmask = &mark->netmask;
3731         ifa->ifa_ifp = ifp;
3732 }
3733
3734 static int
3735 if_ringcnt_fixup(int ring_cnt, int ring_cntmax)
3736 {
3737
3738         KASSERT(ring_cntmax > 0, ("invalid ring count max %d", ring_cntmax));
3739
3740         if (ring_cnt <= 0 || ring_cnt > ring_cntmax)
3741                 ring_cnt = ring_cntmax;
3742         if (ring_cnt > netisr_ncpus)
3743                 ring_cnt = netisr_ncpus;
3744         return (ring_cnt);
3745 }
3746
3747 static void
3748 if_ringmap_set_grid(device_t dev, struct if_ringmap *rm, int grid)
3749 {
3750         int i, offset;
3751
3752         KASSERT(grid > 0, ("invalid if_ringmap grid %d", grid));
3753         KASSERT(grid >= rm->rm_cnt, ("invalid if_ringmap grid %d, count %d",
3754             grid, rm->rm_cnt));
3755         rm->rm_grid = grid;
3756
3757         offset = (rm->rm_grid * device_get_unit(dev)) % netisr_ncpus;
3758         for (i = 0; i < rm->rm_cnt; ++i) {
3759                 rm->rm_cpumap[i] = offset + i;
3760                 KASSERT(rm->rm_cpumap[i] < netisr_ncpus,
3761                     ("invalid cpumap[%d] = %d, offset %d", i,
3762                      rm->rm_cpumap[i], offset));
3763         }
3764 }
3765
3766 static struct if_ringmap *
3767 if_ringmap_alloc_flags(device_t dev, int ring_cnt, int ring_cntmax,
3768     uint32_t flags)
3769 {
3770         struct if_ringmap *rm;
3771         int i, grid = 0, prev_grid;
3772
3773         ring_cnt = if_ringcnt_fixup(ring_cnt, ring_cntmax);
3774         rm = kmalloc(__offsetof(struct if_ringmap, rm_cpumap[ring_cnt]),
3775             M_DEVBUF, M_WAITOK | M_ZERO);
3776
3777         rm->rm_cnt = ring_cnt;
3778         if (flags & RINGMAP_FLAG_POWEROF2)
3779                 rm->rm_cnt = 1 << (fls(rm->rm_cnt) - 1);
3780
3781         prev_grid = netisr_ncpus;
3782         for (i = 0; i < netisr_ncpus; ++i) {
3783                 if (netisr_ncpus % (i + 1) != 0)
3784                         continue;
3785
3786                 grid = netisr_ncpus / (i + 1);
3787                 if (rm->rm_cnt > grid) {
3788                         grid = prev_grid;
3789                         break;
3790                 }
3791
3792                 if (rm->rm_cnt > netisr_ncpus / (i + 2))
3793                         break;
3794                 prev_grid = grid;
3795         }
3796         if_ringmap_set_grid(dev, rm, grid);
3797
3798         return (rm);
3799 }
3800
3801 struct if_ringmap *
3802 if_ringmap_alloc(device_t dev, int ring_cnt, int ring_cntmax)
3803 {
3804
3805         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3806             RINGMAP_FLAG_NONE));
3807 }
3808
3809 struct if_ringmap *
3810 if_ringmap_alloc2(device_t dev, int ring_cnt, int ring_cntmax)
3811 {
3812
3813         return (if_ringmap_alloc_flags(dev, ring_cnt, ring_cntmax,
3814             RINGMAP_FLAG_POWEROF2));
3815 }
3816
3817 void
3818 if_ringmap_free(struct if_ringmap *rm)
3819 {
3820
3821         kfree(rm, M_DEVBUF);
3822 }
3823
3824 /*
3825  * Align the two ringmaps.
3826  *
3827  * e.g. 8 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3828  *
3829  * Before:
3830  *
3831  * CPU      0  1  2  3   4  5  6  7
3832  * NIC_RX               n0 n1 n2 n3
3833  * NIC_TX        N0 N1
3834  *
3835  * After:
3836  *
3837  * CPU      0  1  2  3   4  5  6  7
3838  * NIC_RX               n0 n1 n2 n3
3839  * NIC_TX               N0 N1
3840  */
3841 void
3842 if_ringmap_align(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3843 {
3844
3845         if (rm0->rm_grid > rm1->rm_grid)
3846                 if_ringmap_set_grid(dev, rm1, rm0->rm_grid);
3847         else if (rm0->rm_grid < rm1->rm_grid)
3848                 if_ringmap_set_grid(dev, rm0, rm1->rm_grid);
3849 }
3850
3851 void
3852 if_ringmap_match(device_t dev, struct if_ringmap *rm0, struct if_ringmap *rm1)
3853 {
3854         int subset_grid, cnt, divisor, mod, offset, i;
3855         struct if_ringmap *subset_rm, *rm;
3856         int old_rm0_grid, old_rm1_grid;
3857
3858         if (rm0->rm_grid == rm1->rm_grid)
3859                 return;
3860
3861         /* Save grid for later use */
3862         old_rm0_grid = rm0->rm_grid;
3863         old_rm1_grid = rm1->rm_grid;
3864
3865         if_ringmap_align(dev, rm0, rm1);
3866
3867         /*
3868          * Re-shuffle rings to get more even distribution.
3869          *
3870          * e.g. 12 netisrs, rm0 contains 4 rings, rm1 contains 2 rings.
3871          *
3872          * CPU       0  1  2  3   4  5  6  7   8  9 10 11
3873          *
3874          * NIC_RX   a0 a1 a2 a3  b0 b1 b2 b3  c0 c1 c2 c3
3875          * NIC_TX   A0 A1        B0 B1        C0 C1
3876          *
3877          * NIC_RX   d0 d1 d2 d3  e0 e1 e2 e3  f0 f1 f2 f3
3878          * NIC_TX         D0 D1        E0 E1        F0 F1
3879          */
3880
3881         if (rm0->rm_cnt >= (2 * old_rm1_grid)) {
3882                 cnt = rm0->rm_cnt;
3883                 subset_grid = old_rm1_grid;
3884                 subset_rm = rm1;
3885                 rm = rm0;
3886         } else if (rm1->rm_cnt > (2 * old_rm0_grid)) {
3887                 cnt = rm1->rm_cnt;
3888                 subset_grid = old_rm0_grid;
3889                 subset_rm = rm0;
3890                 rm = rm1;
3891         } else {
3892                 /* No space to shuffle. */
3893                 return;
3894         }
3895
3896         mod = cnt / subset_grid;
3897         KKASSERT(mod >= 2);
3898         divisor = netisr_ncpus / rm->rm_grid;
3899         offset = ((device_get_unit(dev) / divisor) % mod) * subset_grid;
3900
3901         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3902                 subset_rm->rm_cpumap[i] += offset;
3903                 KASSERT(subset_rm->rm_cpumap[i] < netisr_ncpus,
3904                     ("match: invalid cpumap[%d] = %d, offset %d",
3905                      i, subset_rm->rm_cpumap[i], offset));
3906         }
3907 #ifdef INVARIANTS
3908         for (i = 0; i < subset_rm->rm_cnt; ++i) {
3909                 int j;
3910
3911                 for (j = 0; j < rm->rm_cnt; ++j) {
3912                         if (rm->rm_cpumap[j] == subset_rm->rm_cpumap[i])
3913                                 break;
3914                 }
3915                 KASSERT(j < rm->rm_cnt,
3916                     ("subset cpumap[%d] = %d not found in superset",
3917                      i, subset_rm->rm_cpumap[i]));
3918         }
3919 #endif
3920 }
3921
3922 int
3923 if_ringmap_count(const struct if_ringmap *rm)
3924 {
3925
3926         return (rm->rm_cnt);
3927 }
3928
3929 int
3930 if_ringmap_cpumap(const struct if_ringmap *rm, int ring)
3931 {
3932
3933         KASSERT(ring >= 0 && ring < rm->rm_cnt, ("invalid ring %d", ring));
3934         return (rm->rm_cpumap[ring]);
3935 }
3936
3937 void
3938 if_ringmap_rdrtable(const struct if_ringmap *rm, int table[], int table_nent)
3939 {
3940         int i, grid_idx, grid_cnt, patch_off, patch_cnt, ncopy;
3941
3942         KASSERT(table_nent > 0 && (table_nent & NETISR_CPUMASK) == 0,
3943             ("invalid redirect table entries %d", table_nent));
3944
3945         grid_idx = 0;
3946         for (i = 0; i < NETISR_CPUMAX; ++i) {
3947                 table[i] = grid_idx++ % rm->rm_cnt;
3948
3949                 if (grid_idx == rm->rm_grid)
3950                         grid_idx = 0;
3951         }
3952
3953         /*
3954          * Make the ring distributed more evenly for the remainder
3955          * of each grid.
3956          *
3957          * e.g. 12 netisrs, rm contains 8 rings.
3958          *
3959          * Redirect table before:
3960          *
3961          *  0  1  2  3  4  5  6  7  0  1  2  3  0  1  2  3
3962          *  4  5  6  7  0  1  2  3  0  1  2  3  4  5  6  7
3963          *  0  1  2  3  0  1  2  3  4  5  6  7  0  1  2  3
3964          *  ....
3965          *
3966          * Redirect table after being patched (pX, patched entries):
3967          *
3968          *  0  1  2  3  4  5  6  7 p0 p1 p2 p3  0  1  2  3
3969          *  4  5  6  7 p4 p5 p6 p7  0  1  2  3  4  5  6  7
3970          * p0 p1 p2 p3  0  1  2  3  4  5  6  7 p4 p5 p6 p7
3971          *  ....
3972          */
3973         patch_cnt = rm->rm_grid % rm->rm_cnt;
3974         if (patch_cnt == 0)
3975                 goto done;
3976         patch_off = rm->rm_grid - (rm->rm_grid % rm->rm_cnt);
3977
3978         grid_cnt = roundup(NETISR_CPUMAX, rm->rm_grid) / rm->rm_grid;
3979         grid_idx = 0;
3980         for (i = 0; i < grid_cnt; ++i) {
3981                 int j;
3982
3983                 for (j = 0; j < patch_cnt; ++j) {
3984                         int fix_idx;
3985
3986                         fix_idx = (i * rm->rm_grid) + patch_off + j;
3987                         if (fix_idx >= NETISR_CPUMAX)
3988                                 goto done;
3989                         table[fix_idx] = grid_idx++ % rm->rm_cnt;
3990                 }
3991         }
3992 done:
3993         /*
3994          * If the device supports larger redirect table, duplicate
3995          * the first NETISR_CPUMAX entries to the rest of the table,
3996          * so that it matches upper layer's expectation:
3997          * (hash & NETISR_CPUMASK) % netisr_ncpus
3998          */
3999         ncopy = table_nent / NETISR_CPUMAX;
4000         for (i = 1; i < ncopy; ++i) {
4001                 memcpy(&table[i * NETISR_CPUMAX], table,
4002                     NETISR_CPUMAX * sizeof(table[0]));
4003         }
4004         if (if_ringmap_dumprdr) {
4005                 for (i = 0; i < table_nent; ++i) {
4006                         if (i != 0 && i % 16 == 0)
4007                                 kprintf("\n");
4008                         kprintf("%03d ", table[i]);
4009                 }
4010                 kprintf("\n");
4011         }
4012 }
4013
4014 int
4015 if_ringmap_cpumap_sysctl(SYSCTL_HANDLER_ARGS)
4016 {
4017         struct if_ringmap *rm = arg1;
4018         int i, error = 0;
4019
4020         for (i = 0; i < rm->rm_cnt; ++i) {
4021                 int cpu = rm->rm_cpumap[i];
4022
4023                 error = SYSCTL_OUT(req, &cpu, sizeof(cpu));
4024                 if (error)
4025                         break;
4026         }
4027         return (error);
4028 }