FreeBSD/Linux Kernel Cross Reference
sys/net/if_epair.c
1 /*-
2 * Copyright (c) 2008 The FreeBSD Foundation
3 * Copyright (c) 2009 Bjoern A. Zeeb <bz@FreeBSD.org>
4 * All rights reserved.
5 *
6 * This software was developed by CK Software GmbH under sponsorship
7 * from the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31 /*
32 * A pair of virtual back-to-back connected ethernet like interfaces
33 * (``two interfaces with a virtual cross-over cable'').
34 *
35 * This is mostly intended to be used to provide connectivity between
36 * different virtual network stack instances.
37 */
38 /*
39 * Things to re-think once we have more experience:
40 * - ifp->if_reassign function once we can test with vimage. Depending on
41 * how if_vmove() is going to be improved.
42 * - Real random etheraddrs that are checked to be uniquish; we would need
43 * to re-do them in case we move the interface between network stacks
44 * in a private if_reassign function.
45 * In case we bridge to a real interface/network or between indepedent
46 * epairs on multiple stacks/machines, we may need this.
47 * For now let the user handle that case.
48 */
49
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD: releng/8.0/sys/net/if_epair.c 196019 2009-08-01 19:26:27Z rwatson $");
52
53 #include <sys/param.h>
54 #include <sys/kernel.h>
55 #include <sys/mbuf.h>
56 #include <sys/module.h>
57 #include <sys/refcount.h>
58 #include <sys/queue.h>
59 #include <sys/smp.h>
60 #include <sys/socket.h>
61 #include <sys/sockio.h>
62 #include <sys/sysctl.h>
63 #include <sys/types.h>
64
65 #include <net/bpf.h>
66 #include <net/ethernet.h>
67 #include <net/if.h>
68 #include <net/if_clone.h>
69 #include <net/if_var.h>
70 #include <net/if_types.h>
71 #include <net/netisr.h>
72 #include <net/vnet.h>
73
74 #define EPAIRNAME "epair"
75
76 SYSCTL_DECL(_net_link);
77 SYSCTL_NODE(_net_link, OID_AUTO, epair, CTLFLAG_RW, 0, "epair sysctl");
78
79 #ifdef EPAIR_DEBUG
80 static int epair_debug = 0;
81 SYSCTL_XINT(_net_link_epair, OID_AUTO, epair_debug, CTLFLAG_RW,
82 &epair_debug, 0, "if_epair(4) debugging.");
83 #define DPRINTF(fmt, arg...) \
84 if (epair_debug) \
85 printf("[%s:%d] " fmt, __func__, __LINE__, ##arg)
86 #else
87 #define DPRINTF(fmt, arg...)
88 #endif
89
90 static void epair_nh_sintr(struct mbuf *);
91 static struct mbuf *epair_nh_m2cpuid(struct mbuf *, uintptr_t, u_int *);
92 static void epair_nh_drainedcpu(u_int);
93
94 static void epair_start_locked(struct ifnet *);
95
96 static int epair_clone_match(struct if_clone *, const char *);
97 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t);
98 static int epair_clone_destroy(struct if_clone *, struct ifnet *);
99
100 /* Netisr realted definitions and sysctl. */
101 static struct netisr_handler epair_nh = {
102 .nh_name = EPAIRNAME,
103 .nh_proto = NETISR_EPAIR,
104 .nh_policy = NETISR_POLICY_CPU,
105 .nh_handler = epair_nh_sintr,
106 .nh_m2cpuid = epair_nh_m2cpuid,
107 .nh_drainedcpu = epair_nh_drainedcpu,
108 };
109
110 static int
111 sysctl_epair_netisr_maxqlen(SYSCTL_HANDLER_ARGS)
112 {
113 int error, qlimit;
114
115 netisr_getqlimit(&epair_nh, &qlimit);
116 error = sysctl_handle_int(oidp, &qlimit, 0, req);
117 if (error || !req->newptr)
118 return (error);
119 if (qlimit < 1)
120 return (EINVAL);
121 return (netisr_setqlimit(&epair_nh, qlimit));
122 }
123 SYSCTL_PROC(_net_link_epair, OID_AUTO, netisr_maxqlen, CTLTYPE_INT|CTLFLAG_RW,
124 0, 0, sysctl_epair_netisr_maxqlen, "I",
125 "Maximum if_epair(4) netisr \"hw\" queue length");
126
127 struct epair_softc {
128 struct ifnet *ifp; /* This ifp. */
129 struct ifnet *oifp; /* other ifp of pair. */
130 u_int refcount; /* # of mbufs in flight. */
131 u_int cpuid; /* CPU ID assigned upon creation. */
132 void (*if_qflush)(struct ifnet *);
133 /* Original if_qflush routine. */
134 };
135
136 /*
137 * Per-CPU list of ifps with data in the ifq that needs to be flushed
138 * to the netisr ``hw'' queue before we allow any further direct queuing
139 * to the ``hw'' queue.
140 */
141 struct epair_ifp_drain {
142 STAILQ_ENTRY(epair_ifp_drain) ifp_next;
143 struct ifnet *ifp;
144 };
145 STAILQ_HEAD(eid_list, epair_ifp_drain);
146
147 #define EPAIR_LOCK_INIT(dpcpu) mtx_init(&(dpcpu)->if_epair_mtx, \
148 "if_epair", NULL, MTX_DEF)
149 #define EPAIR_LOCK_DESTROY(dpcpu) mtx_destroy(&(dpcpu)->if_epair_mtx)
150 #define EPAIR_LOCK_ASSERT(dpcpu) mtx_assert(&(dpcpu)->if_epair_mtx, \
151 MA_OWNED)
152 #define EPAIR_LOCK(dpcpu) mtx_lock(&(dpcpu)->if_epair_mtx)
153 #define EPAIR_UNLOCK(dpcpu) mtx_unlock(&(dpcpu)->if_epair_mtx)
154
155 #ifdef INVARIANTS
156 #define EPAIR_REFCOUNT_INIT(r, v) refcount_init((r), (v))
157 #define EPAIR_REFCOUNT_AQUIRE(r) refcount_acquire((r))
158 #define EPAIR_REFCOUNT_RELEASE(r) refcount_release((r))
159 #define EPAIR_REFCOUNT_ASSERT(a, p) KASSERT(a, p)
160 #else
161 #define EPAIR_REFCOUNT_INIT(r, v)
162 #define EPAIR_REFCOUNT_AQUIRE(r)
163 #define EPAIR_REFCOUNT_RELEASE(r)
164 #define EPAIR_REFCOUNT_ASSERT(a, p)
165 #endif
166
167 static MALLOC_DEFINE(M_EPAIR, EPAIRNAME,
168 "Pair of virtual cross-over connected Ethernet-like interfaces");
169
170 static struct if_clone epair_cloner = IFC_CLONE_INITIALIZER(
171 EPAIRNAME, NULL, IF_MAXUNIT,
172 NULL, epair_clone_match, epair_clone_create, epair_clone_destroy);
173
174 /*
175 * DPCPU area and functions.
176 */
177 struct epair_dpcpu {
178 struct mtx if_epair_mtx; /* Per-CPU locking. */
179 int epair_drv_flags; /* Per-CPU ``hw'' drv flags. */
180 struct eid_list epair_ifp_drain_list; /* Per-CPU list of ifps with
181 * data in the ifq. */
182 };
183 DPCPU_DEFINE(struct epair_dpcpu, epair_dpcpu);
184
185 static void
186 epair_dpcpu_init(void)
187 {
188 struct epair_dpcpu *epair_dpcpu;
189 struct eid_list *s;
190 u_int cpuid;
191
192 for (cpuid = 0; cpuid <= mp_maxid; cpuid++) {
193 if (CPU_ABSENT(cpuid))
194 continue;
195
196 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
197
198 /* Initialize per-cpu lock. */
199 EPAIR_LOCK_INIT(epair_dpcpu);
200
201 /* Driver flags are per-cpu as are our netisr "hw" queues. */
202 epair_dpcpu->epair_drv_flags = 0;
203
204 /*
205 * Initialize per-cpu drain list.
206 * Manually do what STAILQ_HEAD_INITIALIZER would do.
207 */
208 s = &epair_dpcpu->epair_ifp_drain_list;
209 s->stqh_first = NULL;
210 s->stqh_last = &s->stqh_first;
211 }
212 }
213
214 static void
215 epair_dpcpu_detach(void)
216 {
217 struct epair_dpcpu *epair_dpcpu;
218 u_int cpuid;
219
220 for (cpuid = 0; cpuid <= mp_maxid; cpuid++) {
221 if (CPU_ABSENT(cpuid))
222 continue;
223
224 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
225
226 /* Destroy per-cpu lock. */
227 EPAIR_LOCK_DESTROY(epair_dpcpu);
228 }
229 }
230
231 /*
232 * Helper functions.
233 */
234 static u_int
235 cpuid_from_ifp(struct ifnet *ifp)
236 {
237 struct epair_softc *sc;
238
239 if (ifp == NULL)
240 return (0);
241 sc = ifp->if_softc;
242
243 return (sc->cpuid);
244 }
245
246 /*
247 * Netisr handler functions.
248 */
249 static void
250 epair_nh_sintr(struct mbuf *m)
251 {
252 struct ifnet *ifp;
253 struct epair_softc *sc;
254
255 ifp = m->m_pkthdr.rcvif;
256 (*ifp->if_input)(ifp, m);
257 sc = ifp->if_softc;
258 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
259 DPRINTF("ifp=%p refcount=%u\n", ifp, sc->refcount);
260 }
261
262 static struct mbuf *
263 epair_nh_m2cpuid(struct mbuf *m, uintptr_t source, u_int *cpuid)
264 {
265
266 *cpuid = cpuid_from_ifp(m->m_pkthdr.rcvif);
267
268 return (m);
269 }
270
271 static void
272 epair_nh_drainedcpu(u_int cpuid)
273 {
274 struct epair_dpcpu *epair_dpcpu;
275 struct epair_ifp_drain *elm, *tvar;
276 struct ifnet *ifp;
277
278 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
279 EPAIR_LOCK(epair_dpcpu);
280 /*
281 * Assume our "hw" queue and possibly ifq will be emptied
282 * again. In case we will overflow the "hw" queue while
283 * draining, epair_start_locked will set IFF_DRV_OACTIVE
284 * again and we will stop and return.
285 */
286 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list,
287 ifp_next, tvar) {
288 ifp = elm->ifp;
289 epair_dpcpu->epair_drv_flags &= ~IFF_DRV_OACTIVE;
290 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
291 epair_start_locked(ifp);
292
293 IFQ_LOCK(&ifp->if_snd);
294 if (IFQ_IS_EMPTY(&ifp->if_snd)) {
295 STAILQ_REMOVE(&epair_dpcpu->epair_ifp_drain_list,
296 elm, epair_ifp_drain, ifp_next);
297 free(elm, M_EPAIR);
298 }
299 IFQ_UNLOCK(&ifp->if_snd);
300
301 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) != 0) {
302 /* Our "hw"q overflew again. */
303 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE
304 DPRINTF("hw queue length overflow at %u\n",
305 epair_nh.nh_qlimit);
306 break;
307 }
308 }
309 EPAIR_UNLOCK(epair_dpcpu);
310 }
311
312 /*
313 * Network interface (`if') related functions.
314 */
315 static int
316 epair_add_ifp_for_draining(struct ifnet *ifp)
317 {
318 struct epair_dpcpu *epair_dpcpu;
319 struct epair_softc *sc = sc = ifp->if_softc;
320 struct epair_ifp_drain *elm = NULL;
321
322 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
323 STAILQ_FOREACH(elm, &epair_dpcpu->epair_ifp_drain_list, ifp_next)
324 if (elm->ifp == ifp)
325 break;
326 /* If the ipf is there already, return success. */
327 if (elm != NULL)
328 return (0);
329
330 elm = malloc(sizeof(struct epair_ifp_drain), M_EPAIR, M_NOWAIT|M_ZERO);
331 if (elm == NULL)
332 return (ENOMEM);
333
334 elm->ifp = ifp;
335 STAILQ_INSERT_TAIL(&epair_dpcpu->epair_ifp_drain_list, elm, ifp_next);
336
337 return (0);
338 }
339
340 static void
341 epair_start_locked(struct ifnet *ifp)
342 {
343 struct epair_dpcpu *epair_dpcpu;
344 struct mbuf *m;
345 struct epair_softc *sc;
346 struct ifnet *oifp;
347 int error;
348
349 DPRINTF("ifp=%p\n", ifp);
350 sc = ifp->if_softc;
351 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
352 EPAIR_LOCK_ASSERT(epair_dpcpu);
353
354 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
355 return;
356 if ((ifp->if_flags & IFF_UP) == 0)
357 return;
358
359 /*
360 * We get patckets here from ether_output via if_handoff()
361 * and ned to put them into the input queue of the oifp
362 * and call oifp->if_input() via netisr/epair_sintr().
363 */
364 oifp = sc->oifp;
365 sc = oifp->if_softc;
366 for (;;) {
367 IFQ_DEQUEUE(&ifp->if_snd, m);
368 if (m == NULL)
369 break;
370 BPF_MTAP(ifp, m);
371
372 /*
373 * In case the outgoing interface is not usable,
374 * drop the packet.
375 */
376 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
377 (oifp->if_flags & IFF_UP) ==0) {
378 ifp->if_oerrors++;
379 m_freem(m);
380 continue;
381 }
382 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname);
383
384 /*
385 * Add a reference so the interface cannot go while the
386 * packet is in transit as we rely on rcvif to stay valid.
387 */
388 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
389 m->m_pkthdr.rcvif = oifp;
390 CURVNET_SET_QUIET(oifp->if_vnet);
391 error = netisr_queue(NETISR_EPAIR, m);
392 CURVNET_RESTORE();
393 if (!error) {
394 ifp->if_opackets++;
395 /* Someone else received the packet. */
396 oifp->if_ipackets++;
397 } else {
398 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
399 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
400 if (epair_add_ifp_for_draining(ifp)) {
401 ifp->if_oerrors++;
402 m_freem(m);
403 }
404 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
405 }
406 }
407 }
408
409 static void
410 epair_start(struct ifnet *ifp)
411 {
412 struct epair_dpcpu *epair_dpcpu;
413
414 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu);
415 EPAIR_LOCK(epair_dpcpu);
416 epair_start_locked(ifp);
417 EPAIR_UNLOCK(epair_dpcpu);
418 }
419
420 static int
421 epair_transmit_locked(struct ifnet *ifp, struct mbuf *m)
422 {
423 struct epair_dpcpu *epair_dpcpu;
424 struct epair_softc *sc;
425 struct ifnet *oifp;
426 int error, len;
427 short mflags;
428
429 DPRINTF("ifp=%p m=%p\n", ifp, m);
430 sc = ifp->if_softc;
431 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
432 EPAIR_LOCK_ASSERT(epair_dpcpu);
433
434 if (m == NULL)
435 return (0);
436
437 /*
438 * We are not going to use the interface en/dequeue mechanism
439 * on the TX side. We are called from ether_output_frame()
440 * and will put the packet into the incoming queue of the
441 * other interface of our pair via the netsir.
442 */
443 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
444 m_freem(m);
445 return (ENXIO);
446 }
447 if ((ifp->if_flags & IFF_UP) == 0) {
448 m_freem(m);
449 return (ENETDOWN);
450 }
451
452 BPF_MTAP(ifp, m);
453
454 /*
455 * In case the outgoing interface is not usable,
456 * drop the packet.
457 */
458 oifp = sc->oifp;
459 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
460 (oifp->if_flags & IFF_UP) ==0) {
461 ifp->if_oerrors++;
462 m_freem(m);
463 return (0);
464 }
465 len = m->m_pkthdr.len;
466 mflags = m->m_flags;
467 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname);
468
469 #ifdef ALTQ
470 /* Support ALTQ via the clasic if_start() path. */
471 IF_LOCK(&ifp->if_snd);
472 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
473 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
474 if (error)
475 ifp->if_snd.ifq_drops++;
476 IF_UNLOCK(&ifp->if_snd);
477 if (!error) {
478 ifp->if_obytes += len;
479 if (mflags & (M_BCAST|M_MCAST))
480 ifp->if_omcasts++;
481
482 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) == 0)
483 epair_start_locked(ifp);
484 else
485 (void)epair_add_ifp_for_draining(ifp);
486 }
487 return (error);
488 }
489 IF_UNLOCK(&ifp->if_snd);
490 #endif
491
492 if ((epair_dpcpu->epair_drv_flags & IFF_DRV_OACTIVE) != 0) {
493 /*
494 * Our hardware queue is full, try to fall back
495 * queuing to the ifq but do not call ifp->if_start.
496 * Either we are lucky or the packet is gone.
497 */
498 IFQ_ENQUEUE(&ifp->if_snd, m, error);
499 if (!error)
500 (void)epair_add_ifp_for_draining(ifp);
501 return (error);
502 }
503 sc = oifp->if_softc;
504 /*
505 * Add a reference so the interface cannot go while the
506 * packet is in transit as we rely on rcvif to stay valid.
507 */
508 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
509 m->m_pkthdr.rcvif = oifp;
510 CURVNET_SET_QUIET(oifp->if_vnet);
511 error = netisr_queue(NETISR_EPAIR, m);
512 CURVNET_RESTORE();
513 if (!error) {
514 ifp->if_opackets++;
515 /*
516 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics,
517 * but as we bypass all this we have to duplicate
518 * the logic another time.
519 */
520 ifp->if_obytes += len;
521 if (mflags & (M_BCAST|M_MCAST))
522 ifp->if_omcasts++;
523 /* Someone else received the packet. */
524 oifp->if_ipackets++;
525 } else {
526 /* The packet was freed already. */
527 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
528 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
529 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
530 }
531
532 return (error);
533 }
534
535 static int
536 epair_transmit(struct ifnet *ifp, struct mbuf *m)
537 {
538 struct epair_dpcpu *epair_dpcpu;
539 int error;
540
541 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu);
542 EPAIR_LOCK(epair_dpcpu);
543 error = epair_transmit_locked(ifp, m);
544 EPAIR_UNLOCK(epair_dpcpu);
545 return (error);
546 }
547
548 static void
549 epair_qflush(struct ifnet *ifp)
550 {
551 struct epair_dpcpu *epair_dpcpu;
552 struct epair_softc *sc;
553 struct ifaltq *ifq;
554
555 sc = ifp->if_softc;
556 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
557 EPAIR_LOCK(epair_dpcpu);
558 ifq = &ifp->if_snd;
559 DPRINTF("ifp=%p sc refcnt=%u ifq_len=%u\n",
560 ifp, sc->refcount, ifq->ifq_len);
561 /*
562 * Instead of calling EPAIR_REFCOUNT_RELEASE(&sc->refcount);
563 * n times, just subtract for the cleanup.
564 */
565 sc->refcount -= ifq->ifq_len;
566 EPAIR_UNLOCK(epair_dpcpu);
567 if (sc->if_qflush)
568 sc->if_qflush(ifp);
569 }
570
571 static int
572 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
573 {
574 struct ifreq *ifr;
575 int error;
576
577 ifr = (struct ifreq *)data;
578 switch (cmd) {
579 case SIOCSIFFLAGS:
580 case SIOCADDMULTI:
581 case SIOCDELMULTI:
582 error = 0;
583 break;
584
585 case SIOCSIFMTU:
586 /* We basically allow all kinds of MTUs. */
587 ifp->if_mtu = ifr->ifr_mtu;
588 error = 0;
589 break;
590
591 default:
592 /* Let the common ethernet handler process this. */
593 error = ether_ioctl(ifp, cmd, data);
594 break;
595 }
596
597 return (error);
598 }
599
600 static void
601 epair_init(void *dummy __unused)
602 {
603 }
604
605
606 /*
607 * Interface cloning functions.
608 * We use our private ones so that we can create/destroy our secondary
609 * device along with the primary one.
610 */
611 static int
612 epair_clone_match(struct if_clone *ifc, const char *name)
613 {
614 const char *cp;
615
616 DPRINTF("name='%s'\n", name);
617
618 /*
619 * Our base name is epair.
620 * Our interfaces will be named epair<n>[ab].
621 * So accept anything of the following list:
622 * - epair
623 * - epair<n>
624 * but not the epair<n>[ab] versions.
625 */
626 if (strncmp(EPAIRNAME, name, sizeof(EPAIRNAME)-1) != 0)
627 return (0);
628
629 for (cp = name + sizeof(EPAIRNAME) - 1; *cp != '\0'; cp++) {
630 if (*cp < '' || *cp > '9')
631 return (0);
632 }
633
634 return (1);
635 }
636
637 static int
638 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
639 {
640 struct epair_softc *sca, *scb;
641 struct ifnet *ifp;
642 char *dp;
643 int error, unit, wildcard;
644 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
645
646 /*
647 * We are abusing params to create our second interface.
648 * Actually we already created it and called if_clone_createif()
649 * for it to do the official insertion procedure the moment we knew
650 * it cannot fail anymore. So just do attach it here.
651 */
652 if (params) {
653 scb = (struct epair_softc *)params;
654 ifp = scb->ifp;
655 /* Assign a hopefully unique, locally administered etheraddr. */
656 eaddr[0] = 0x02;
657 eaddr[3] = (ifp->if_index >> 8) & 0xff;
658 eaddr[4] = ifp->if_index & 0xff;
659 eaddr[5] = 0x0b;
660 ether_ifattach(ifp, eaddr);
661 /* Correctly set the name for the cloner list. */
662 strlcpy(name, scb->ifp->if_xname, len);
663 return (0);
664 }
665
666 /* Try to see if a special unit was requested. */
667 error = ifc_name2unit(name, &unit);
668 if (error != 0)
669 return (error);
670 wildcard = (unit < 0);
671
672 error = ifc_alloc_unit(ifc, &unit);
673 if (error != 0)
674 return (error);
675
676 /*
677 * If no unit had been given, we need to adjust the ifName.
678 * Also make sure there is space for our extra [ab] suffix.
679 */
680 for (dp = name; *dp != '\0'; dp++);
681 if (wildcard) {
682 error = snprintf(dp, len - (dp - name), "%d", unit);
683 if (error > len - (dp - name) - 1) {
684 /* ifName too long. */
685 ifc_free_unit(ifc, unit);
686 return (ENOSPC);
687 }
688 dp += error;
689 }
690 if (len - (dp - name) - 1 < 1) {
691 /* No space left for our [ab] suffix. */
692 ifc_free_unit(ifc, unit);
693 return (ENOSPC);
694 }
695 *dp = 'a';
696 /* Must not change dp so we can replace 'a' by 'b' later. */
697 *(dp+1) = '\0';
698
699 /* Allocate memory for both [ab] interfaces */
700 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
701 EPAIR_REFCOUNT_INIT(&sca->refcount, 1);
702 sca->ifp = if_alloc(IFT_ETHER);
703 if (sca->ifp == NULL) {
704 free(sca, M_EPAIR);
705 ifc_free_unit(ifc, unit);
706 return (ENOSPC);
707 }
708
709 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
710 EPAIR_REFCOUNT_INIT(&scb->refcount, 1);
711 scb->ifp = if_alloc(IFT_ETHER);
712 if (scb->ifp == NULL) {
713 free(scb, M_EPAIR);
714 if_free(sca->ifp);
715 free(sca, M_EPAIR);
716 ifc_free_unit(ifc, unit);
717 return (ENOSPC);
718 }
719
720 /*
721 * Cross-reference the interfaces so we will be able to free both.
722 */
723 sca->oifp = scb->ifp;
724 scb->oifp = sca->ifp;
725
726 /*
727 * Calculate the cpuid for netisr queueing based on the
728 * ifIndex of the interfaces. As long as we cannot configure
729 * this or use cpuset information easily we cannot guarantee
730 * cache locality but we can at least allow parallelism.
731 */
732 sca->cpuid =
733 netisr_get_cpuid(sca->ifp->if_index % netisr_get_cpucount());
734 scb->cpuid =
735 netisr_get_cpuid(scb->ifp->if_index % netisr_get_cpucount());
736
737 /* Finish initialization of interface <n>a. */
738 ifp = sca->ifp;
739 ifp->if_softc = sca;
740 strlcpy(ifp->if_xname, name, IFNAMSIZ);
741 ifp->if_dname = ifc->ifc_name;
742 ifp->if_dunit = unit;
743 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
744 ifp->if_start = epair_start;
745 ifp->if_ioctl = epair_ioctl;
746 ifp->if_init = epair_init;
747 ifp->if_snd.ifq_maxlen = ifqmaxlen;
748 /* Assign a hopefully unique, locally administered etheraddr. */
749 eaddr[0] = 0x02;
750 eaddr[3] = (ifp->if_index >> 8) & 0xff;
751 eaddr[4] = ifp->if_index & 0xff;
752 eaddr[5] = 0x0a;
753 ether_ifattach(ifp, eaddr);
754 sca->if_qflush = ifp->if_qflush;
755 ifp->if_qflush = epair_qflush;
756 ifp->if_transmit = epair_transmit;
757 ifp->if_baudrate = IF_Gbps(10UL); /* arbitrary maximum */
758
759 /* Swap the name and finish initialization of interface <n>b. */
760 *dp = 'b';
761
762 ifp = scb->ifp;
763 ifp->if_softc = scb;
764 strlcpy(ifp->if_xname, name, IFNAMSIZ);
765 ifp->if_dname = ifc->ifc_name;
766 ifp->if_dunit = unit;
767 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
768 ifp->if_start = epair_start;
769 ifp->if_ioctl = epair_ioctl;
770 ifp->if_init = epair_init;
771 ifp->if_snd.ifq_maxlen = ifqmaxlen;
772 /* We need to play some tricks here for the second interface. */
773 strlcpy(name, EPAIRNAME, len);
774 error = if_clone_create(name, len, (caddr_t)scb);
775 if (error)
776 panic("%s: if_clone_createif() for our 2nd iface failed: %d",
777 __func__, error);
778 scb->if_qflush = ifp->if_qflush;
779 ifp->if_qflush = epair_qflush;
780 ifp->if_transmit = epair_transmit;
781 ifp->if_baudrate = IF_Gbps(10UL); /* arbitrary maximum */
782
783 /*
784 * Restore name to <n>a as the ifp for this will go into the
785 * cloner list for the initial call.
786 */
787 strlcpy(name, sca->ifp->if_xname, len);
788 DPRINTF("name='%s/%db' created sca=%p scb=%p\n", name, unit, sca, scb);
789
790 /* Tell the world, that we are ready to rock. */
791 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING;
792 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING;
793
794 return (0);
795 }
796
797 static int
798 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
799 {
800 struct ifnet *oifp;
801 struct epair_softc *sca, *scb;
802 int unit, error;
803
804 DPRINTF("ifp=%p\n", ifp);
805
806 /*
807 * In case we called into if_clone_destroyif() ourselves
808 * again to remove the second interface, the softc will be
809 * NULL. In that case so not do anything but return success.
810 */
811 if (ifp->if_softc == NULL)
812 return (0);
813
814 unit = ifp->if_dunit;
815 sca = ifp->if_softc;
816 oifp = sca->oifp;
817 scb = oifp->if_softc;
818
819 DPRINTF("ifp=%p oifp=%p\n", ifp, oifp);
820 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
821 oifp->if_drv_flags &= ~IFF_DRV_RUNNING;
822 ether_ifdetach(oifp);
823 ether_ifdetach(ifp);
824 /*
825 * Wait for all packets to be dispatched to if_input.
826 * The numbers can only go down as the interfaces are
827 * detached so there is no need to use atomics.
828 */
829 DPRINTF("sca refcnt=%u scb refcnt=%u\n", sca->refcount, scb->refcount);
830 EPAIR_REFCOUNT_ASSERT(sca->refcount == 1 && scb->refcount == 1,
831 ("%s: sca->refcount!=1: %d || scb->refcount!=1: %d",
832 __func__, sca->refcount, scb->refcount));
833
834 /*
835 * Get rid of our second half.
836 */
837 oifp->if_softc = NULL;
838 error = if_clone_destroyif(ifc, oifp);
839 if (error)
840 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
841 __func__, error);
842
843 /*
844 * Finish cleaning up. Free them and release the unit.
845 * As the other of the two interfaces my reside in a different vnet,
846 * we need to switch before freeing them.
847 */
848 CURVNET_SET_QUIET(oifp->if_vnet);
849 if_free_type(oifp, IFT_ETHER);
850 CURVNET_RESTORE();
851 if_free_type(ifp, IFT_ETHER);
852 free(scb, M_EPAIR);
853 free(sca, M_EPAIR);
854 ifc_free_unit(ifc, unit);
855
856 return (0);
857 }
858
859 static int
860 epair_modevent(module_t mod, int type, void *data)
861 {
862 int qlimit;
863
864 switch (type) {
865 case MOD_LOAD:
866 /* For now limit us to one global mutex and one inq. */
867 epair_dpcpu_init();
868 epair_nh.nh_qlimit = 42 * ifqmaxlen; /* 42 shall be the number. */
869 if (TUNABLE_INT_FETCH("net.link.epair.netisr_maxqlen", &qlimit))
870 epair_nh.nh_qlimit = qlimit;
871 netisr_register(&epair_nh);
872 if_clone_attach(&epair_cloner);
873 if (bootverbose)
874 printf("%s initialized.\n", EPAIRNAME);
875 break;
876 case MOD_UNLOAD:
877 if_clone_detach(&epair_cloner);
878 netisr_unregister(&epair_nh);
879 epair_dpcpu_detach();
880 if (bootverbose)
881 printf("%s unloaded.\n", EPAIRNAME);
882 break;
883 default:
884 return (EOPNOTSUPP);
885 }
886 return (0);
887 }
888
889 static moduledata_t epair_mod = {
890 "if_epair",
891 epair_modevent,
892 0
893 };
894
895 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
896 MODULE_VERSION(if_epair, 1);
Cache object: e31f281dec4acc891bca41ae331705f5
|