FreeBSD/Linux Kernel Cross Reference
sys/net/if_epair.c
1 /*-
2 * Copyright (c) 2008 The FreeBSD Foundation
3 * Copyright (c) 2009-2010 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/8.4/sys/net/if_epair.c 239981 2012-09-01 10:28:20Z trociny $");
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_media.h>
70 #include <net/if_var.h>
71 #include <net/if_types.h>
72 #include <net/netisr.h>
73 #include <net/vnet.h>
74
75 #define EPAIRNAME "epair"
76
77 SYSCTL_DECL(_net_link);
78 SYSCTL_NODE(_net_link, OID_AUTO, epair, CTLFLAG_RW, 0, "epair sysctl");
79
80 #ifdef EPAIR_DEBUG
81 static int epair_debug = 0;
82 SYSCTL_INT(_net_link_epair, OID_AUTO, epair_debug, CTLFLAG_RW,
83 &epair_debug, 0, "if_epair(4) debugging.");
84 #define DPRINTF(fmt, arg...) \
85 if (epair_debug) \
86 printf("[%s:%d] " fmt, __func__, __LINE__, ##arg)
87 #else
88 #define DPRINTF(fmt, arg...)
89 #endif
90
91 static void epair_nh_sintr(struct mbuf *);
92 static struct mbuf *epair_nh_m2cpuid(struct mbuf *, uintptr_t, u_int *);
93 static void epair_nh_drainedcpu(u_int);
94
95 static void epair_start_locked(struct ifnet *);
96 static int epair_media_change(struct ifnet *);
97 static void epair_media_status(struct ifnet *, struct ifmediareq *);
98
99 static int epair_clone_match(struct if_clone *, const char *);
100 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t);
101 static int epair_clone_destroy(struct if_clone *, struct ifnet *);
102
103 /* Netisr realted definitions and sysctl. */
104 static struct netisr_handler epair_nh = {
105 .nh_name = EPAIRNAME,
106 .nh_proto = NETISR_EPAIR,
107 .nh_policy = NETISR_POLICY_CPU,
108 .nh_handler = epair_nh_sintr,
109 .nh_m2cpuid = epair_nh_m2cpuid,
110 .nh_drainedcpu = epair_nh_drainedcpu,
111 };
112
113 static int
114 sysctl_epair_netisr_maxqlen(SYSCTL_HANDLER_ARGS)
115 {
116 int error, qlimit;
117
118 netisr_getqlimit(&epair_nh, &qlimit);
119 error = sysctl_handle_int(oidp, &qlimit, 0, req);
120 if (error || !req->newptr)
121 return (error);
122 if (qlimit < 1)
123 return (EINVAL);
124 return (netisr_setqlimit(&epair_nh, qlimit));
125 }
126 SYSCTL_PROC(_net_link_epair, OID_AUTO, netisr_maxqlen, CTLTYPE_INT|CTLFLAG_RW,
127 0, 0, sysctl_epair_netisr_maxqlen, "I",
128 "Maximum if_epair(4) netisr \"hw\" queue length");
129
130 struct epair_softc {
131 struct ifnet *ifp; /* This ifp. */
132 struct ifnet *oifp; /* other ifp of pair. */
133 struct ifmedia media; /* Media config (fake). */
134 u_int refcount; /* # of mbufs in flight. */
135 u_int cpuid; /* CPU ID assigned upon creation. */
136 void (*if_qflush)(struct ifnet *);
137 /* Original if_qflush routine. */
138 };
139
140 /*
141 * Per-CPU list of ifps with data in the ifq that needs to be flushed
142 * to the netisr ``hw'' queue before we allow any further direct queuing
143 * to the ``hw'' queue.
144 */
145 struct epair_ifp_drain {
146 STAILQ_ENTRY(epair_ifp_drain) ifp_next;
147 struct ifnet *ifp;
148 };
149 STAILQ_HEAD(eid_list, epair_ifp_drain);
150
151 #define EPAIR_LOCK_INIT(dpcpu) mtx_init(&(dpcpu)->if_epair_mtx, \
152 "if_epair", NULL, MTX_DEF)
153 #define EPAIR_LOCK_DESTROY(dpcpu) mtx_destroy(&(dpcpu)->if_epair_mtx)
154 #define EPAIR_LOCK_ASSERT(dpcpu) mtx_assert(&(dpcpu)->if_epair_mtx, \
155 MA_OWNED)
156 #define EPAIR_LOCK(dpcpu) mtx_lock(&(dpcpu)->if_epair_mtx)
157 #define EPAIR_UNLOCK(dpcpu) mtx_unlock(&(dpcpu)->if_epair_mtx)
158
159 #ifdef INVARIANTS
160 #define EPAIR_REFCOUNT_INIT(r, v) refcount_init((r), (v))
161 #define EPAIR_REFCOUNT_AQUIRE(r) refcount_acquire((r))
162 #define EPAIR_REFCOUNT_RELEASE(r) refcount_release((r))
163 #define EPAIR_REFCOUNT_ASSERT(a, p) KASSERT(a, p)
164 #else
165 #define EPAIR_REFCOUNT_INIT(r, v)
166 #define EPAIR_REFCOUNT_AQUIRE(r)
167 #define EPAIR_REFCOUNT_RELEASE(r)
168 #define EPAIR_REFCOUNT_ASSERT(a, p)
169 #endif
170
171 static MALLOC_DEFINE(M_EPAIR, EPAIRNAME,
172 "Pair of virtual cross-over connected Ethernet-like interfaces");
173
174 static struct if_clone epair_cloner = IFC_CLONE_INITIALIZER(
175 EPAIRNAME, NULL, IF_MAXUNIT,
176 NULL, epair_clone_match, epair_clone_create, epair_clone_destroy);
177
178 /*
179 * DPCPU area and functions.
180 */
181 struct epair_dpcpu {
182 struct mtx if_epair_mtx; /* Per-CPU locking. */
183 int epair_drv_flags; /* Per-CPU ``hw'' drv flags. */
184 struct eid_list epair_ifp_drain_list; /* Per-CPU list of ifps with
185 * data in the ifq. */
186 };
187 DPCPU_DEFINE(struct epair_dpcpu, epair_dpcpu);
188
189 static void
190 epair_dpcpu_init(void)
191 {
192 struct epair_dpcpu *epair_dpcpu;
193 struct eid_list *s;
194 u_int cpuid;
195
196 CPU_FOREACH(cpuid) {
197 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
198
199 /* Initialize per-cpu lock. */
200 EPAIR_LOCK_INIT(epair_dpcpu);
201
202 /* Driver flags are per-cpu as are our netisr "hw" queues. */
203 epair_dpcpu->epair_drv_flags = 0;
204
205 /*
206 * Initialize per-cpu drain list.
207 * Manually do what STAILQ_HEAD_INITIALIZER would do.
208 */
209 s = &epair_dpcpu->epair_ifp_drain_list;
210 s->stqh_first = NULL;
211 s->stqh_last = &s->stqh_first;
212 }
213 }
214
215 static void
216 epair_dpcpu_detach(void)
217 {
218 struct epair_dpcpu *epair_dpcpu;
219 u_int cpuid;
220
221 CPU_FOREACH(cpuid) {
222 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
223
224 /* Destroy per-cpu lock. */
225 EPAIR_LOCK_DESTROY(epair_dpcpu);
226 }
227 }
228
229 /*
230 * Helper functions.
231 */
232 static u_int
233 cpuid_from_ifp(struct ifnet *ifp)
234 {
235 struct epair_softc *sc;
236
237 if (ifp == NULL)
238 return (0);
239 sc = ifp->if_softc;
240
241 return (sc->cpuid);
242 }
243
244 /*
245 * Netisr handler functions.
246 */
247 static void
248 epair_nh_sintr(struct mbuf *m)
249 {
250 struct ifnet *ifp;
251 struct epair_softc *sc;
252
253 ifp = m->m_pkthdr.rcvif;
254 (*ifp->if_input)(ifp, m);
255 sc = ifp->if_softc;
256 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
257 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
258 ("%s: ifp=%p sc->refcount not >= 1: %d",
259 __func__, ifp, sc->refcount));
260 DPRINTF("ifp=%p refcount=%u\n", ifp, sc->refcount);
261 }
262
263 static struct mbuf *
264 epair_nh_m2cpuid(struct mbuf *m, uintptr_t source, u_int *cpuid)
265 {
266
267 *cpuid = cpuid_from_ifp(m->m_pkthdr.rcvif);
268
269 return (m);
270 }
271
272 static void
273 epair_nh_drainedcpu(u_int cpuid)
274 {
275 struct epair_dpcpu *epair_dpcpu;
276 struct epair_ifp_drain *elm, *tvar;
277 struct ifnet *ifp;
278
279 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
280 EPAIR_LOCK(epair_dpcpu);
281 /*
282 * Assume our "hw" queue and possibly ifq will be emptied
283 * again. In case we will overflow the "hw" queue while
284 * draining, epair_start_locked will set IFF_DRV_OACTIVE
285 * again and we will stop and return.
286 */
287 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list,
288 ifp_next, tvar) {
289 ifp = elm->ifp;
290 epair_dpcpu->epair_drv_flags &= ~IFF_DRV_OACTIVE;
291 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
292 epair_start_locked(ifp);
293
294 IFQ_LOCK(&ifp->if_snd);
295 if (IFQ_IS_EMPTY(&ifp->if_snd)) {
296 struct epair_softc *sc;
297
298 STAILQ_REMOVE(&epair_dpcpu->epair_ifp_drain_list,
299 elm, epair_ifp_drain, ifp_next);
300 /* The cached ifp goes off the list. */
301 sc = ifp->if_softc;
302 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
303 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
304 ("%s: ifp=%p sc->refcount not >= 1: %d",
305 __func__, ifp, sc->refcount));
306 free(elm, M_EPAIR);
307 }
308 IFQ_UNLOCK(&ifp->if_snd);
309
310 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) != 0) {
311 /* Our "hw"q overflew again. */
312 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
313 DPRINTF("hw queue length overflow at %u\n",
314 epair_nh.nh_qlimit);
315 break;
316 }
317 }
318 EPAIR_UNLOCK(epair_dpcpu);
319 }
320
321 /*
322 * Network interface (`if') related functions.
323 */
324 static void
325 epair_remove_ifp_from_draining(struct ifnet *ifp)
326 {
327 struct epair_dpcpu *epair_dpcpu;
328 struct epair_ifp_drain *elm, *tvar;
329 u_int cpuid;
330
331 CPU_FOREACH(cpuid) {
332 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu);
333 EPAIR_LOCK(epair_dpcpu);
334 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list,
335 ifp_next, tvar) {
336 if (ifp == elm->ifp) {
337 struct epair_softc *sc;
338
339 STAILQ_REMOVE(
340 &epair_dpcpu->epair_ifp_drain_list, elm,
341 epair_ifp_drain, ifp_next);
342 /* The cached ifp goes off the list. */
343 sc = ifp->if_softc;
344 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
345 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
346 ("%s: ifp=%p sc->refcount not >= 1: %d",
347 __func__, ifp, sc->refcount));
348 free(elm, M_EPAIR);
349 }
350 }
351 EPAIR_UNLOCK(epair_dpcpu);
352 }
353 }
354
355 static int
356 epair_add_ifp_for_draining(struct ifnet *ifp)
357 {
358 struct epair_dpcpu *epair_dpcpu;
359 struct epair_softc *sc;
360 struct epair_ifp_drain *elm = NULL;
361
362 sc = ifp->if_softc;
363 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
364 EPAIR_LOCK_ASSERT(epair_dpcpu);
365 STAILQ_FOREACH(elm, &epair_dpcpu->epair_ifp_drain_list, ifp_next)
366 if (elm->ifp == ifp)
367 break;
368 /* If the ifp is there already, return success. */
369 if (elm != NULL)
370 return (0);
371
372 elm = malloc(sizeof(struct epair_ifp_drain), M_EPAIR, M_NOWAIT|M_ZERO);
373 if (elm == NULL)
374 return (ENOMEM);
375
376 elm->ifp = ifp;
377 /* Add a reference for the ifp pointer on the list. */
378 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
379 STAILQ_INSERT_TAIL(&epair_dpcpu->epair_ifp_drain_list, elm, ifp_next);
380
381 return (0);
382 }
383
384 static void
385 epair_start_locked(struct ifnet *ifp)
386 {
387 struct epair_dpcpu *epair_dpcpu;
388 struct mbuf *m;
389 struct epair_softc *sc;
390 struct ifnet *oifp;
391 int error;
392
393 DPRINTF("ifp=%p\n", ifp);
394 sc = ifp->if_softc;
395 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
396 EPAIR_LOCK_ASSERT(epair_dpcpu);
397
398 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
399 return;
400 if ((ifp->if_flags & IFF_UP) == 0)
401 return;
402
403 /*
404 * We get patckets here from ether_output via if_handoff()
405 * and ned to put them into the input queue of the oifp
406 * and call oifp->if_input() via netisr/epair_sintr().
407 */
408 oifp = sc->oifp;
409 sc = oifp->if_softc;
410 for (;;) {
411 IFQ_DEQUEUE(&ifp->if_snd, m);
412 if (m == NULL)
413 break;
414 BPF_MTAP(ifp, m);
415
416 /*
417 * In case the outgoing interface is not usable,
418 * drop the packet.
419 */
420 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
421 (oifp->if_flags & IFF_UP) ==0) {
422 ifp->if_oerrors++;
423 m_freem(m);
424 continue;
425 }
426 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname);
427
428 /*
429 * Add a reference so the interface cannot go while the
430 * packet is in transit as we rely on rcvif to stay valid.
431 */
432 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
433 m->m_pkthdr.rcvif = oifp;
434 CURVNET_SET_QUIET(oifp->if_vnet);
435 error = netisr_queue(NETISR_EPAIR, m);
436 CURVNET_RESTORE();
437 if (!error) {
438 ifp->if_opackets++;
439 /* Someone else received the packet. */
440 oifp->if_ipackets++;
441 } else {
442 /* The packet was freed already. */
443 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
444 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
445 (void) epair_add_ifp_for_draining(ifp);
446 ifp->if_oerrors++;
447 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
448 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
449 ("%s: ifp=%p sc->refcount not >= 1: %d",
450 __func__, oifp, sc->refcount));
451 }
452 }
453 }
454
455 static void
456 epair_start(struct ifnet *ifp)
457 {
458 struct epair_dpcpu *epair_dpcpu;
459
460 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu);
461 EPAIR_LOCK(epair_dpcpu);
462 epair_start_locked(ifp);
463 EPAIR_UNLOCK(epair_dpcpu);
464 }
465
466 static int
467 epair_transmit_locked(struct ifnet *ifp, struct mbuf *m)
468 {
469 struct epair_dpcpu *epair_dpcpu;
470 struct epair_softc *sc;
471 struct ifnet *oifp;
472 int error, len;
473 short mflags;
474
475 DPRINTF("ifp=%p m=%p\n", ifp, m);
476 sc = ifp->if_softc;
477 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu);
478 EPAIR_LOCK_ASSERT(epair_dpcpu);
479
480 if (m == NULL)
481 return (0);
482
483 /*
484 * We are not going to use the interface en/dequeue mechanism
485 * on the TX side. We are called from ether_output_frame()
486 * and will put the packet into the incoming queue of the
487 * other interface of our pair via the netsir.
488 */
489 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
490 m_freem(m);
491 return (ENXIO);
492 }
493 if ((ifp->if_flags & IFF_UP) == 0) {
494 m_freem(m);
495 return (ENETDOWN);
496 }
497
498 BPF_MTAP(ifp, m);
499
500 /*
501 * In case the outgoing interface is not usable,
502 * drop the packet.
503 */
504 oifp = sc->oifp;
505 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
506 (oifp->if_flags & IFF_UP) ==0) {
507 ifp->if_oerrors++;
508 m_freem(m);
509 return (0);
510 }
511 len = m->m_pkthdr.len;
512 mflags = m->m_flags;
513 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname);
514
515 #ifdef ALTQ
516 /* Support ALTQ via the clasic if_start() path. */
517 IF_LOCK(&ifp->if_snd);
518 if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
519 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error);
520 if (error)
521 ifp->if_snd.ifq_drops++;
522 IF_UNLOCK(&ifp->if_snd);
523 if (!error) {
524 ifp->if_obytes += len;
525 if (mflags & (M_BCAST|M_MCAST))
526 ifp->if_omcasts++;
527
528 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) == 0)
529 epair_start_locked(ifp);
530 else
531 (void)epair_add_ifp_for_draining(ifp);
532 }
533 return (error);
534 }
535 IF_UNLOCK(&ifp->if_snd);
536 #endif
537
538 if ((epair_dpcpu->epair_drv_flags & IFF_DRV_OACTIVE) != 0) {
539 /*
540 * Our hardware queue is full, try to fall back
541 * queuing to the ifq but do not call ifp->if_start.
542 * Either we are lucky or the packet is gone.
543 */
544 IFQ_ENQUEUE(&ifp->if_snd, m, error);
545 if (!error)
546 (void)epair_add_ifp_for_draining(ifp);
547 return (error);
548 }
549 sc = oifp->if_softc;
550 /*
551 * Add a reference so the interface cannot go while the
552 * packet is in transit as we rely on rcvif to stay valid.
553 */
554 EPAIR_REFCOUNT_AQUIRE(&sc->refcount);
555 m->m_pkthdr.rcvif = oifp;
556 CURVNET_SET_QUIET(oifp->if_vnet);
557 error = netisr_queue(NETISR_EPAIR, m);
558 CURVNET_RESTORE();
559 if (!error) {
560 ifp->if_opackets++;
561 /*
562 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics,
563 * but as we bypass all this we have to duplicate
564 * the logic another time.
565 */
566 ifp->if_obytes += len;
567 if (mflags & (M_BCAST|M_MCAST))
568 ifp->if_omcasts++;
569 /* Someone else received the packet. */
570 oifp->if_ipackets++;
571 } else {
572 /* The packet was freed already. */
573 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE;
574 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
575 ifp->if_oerrors++;
576 EPAIR_REFCOUNT_RELEASE(&sc->refcount);
577 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1,
578 ("%s: ifp=%p sc->refcount not >= 1: %d",
579 __func__, oifp, sc->refcount));
580 }
581
582 return (error);
583 }
584
585 static int
586 epair_transmit(struct ifnet *ifp, struct mbuf *m)
587 {
588 struct epair_dpcpu *epair_dpcpu;
589 int error;
590
591 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu);
592 EPAIR_LOCK(epair_dpcpu);
593 error = epair_transmit_locked(ifp, m);
594 EPAIR_UNLOCK(epair_dpcpu);
595 return (error);
596 }
597
598 static void
599 epair_qflush(struct ifnet *ifp)
600 {
601 struct epair_softc *sc;
602
603 sc = ifp->if_softc;
604 KASSERT(sc != NULL, ("%s: ifp=%p, epair_softc gone? sc=%p\n",
605 __func__, ifp, sc));
606 /*
607 * Remove this ifp from all backpointer lists. The interface will not
608 * usable for flushing anyway nor should it have anything to flush
609 * after if_qflush().
610 */
611 epair_remove_ifp_from_draining(ifp);
612
613 if (sc->if_qflush)
614 sc->if_qflush(ifp);
615 }
616
617 static int
618 epair_media_change(struct ifnet *ifp __unused)
619 {
620
621 /* Do nothing. */
622 return (0);
623 }
624
625 static void
626 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr)
627 {
628
629 imr->ifm_status = IFM_AVALID | IFM_ACTIVE;
630 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX;
631 }
632
633 static int
634 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
635 {
636 struct epair_softc *sc;
637 struct ifreq *ifr;
638 int error;
639
640 ifr = (struct ifreq *)data;
641 switch (cmd) {
642 case SIOCSIFFLAGS:
643 case SIOCADDMULTI:
644 case SIOCDELMULTI:
645 error = 0;
646 break;
647
648 case SIOCSIFMEDIA:
649 case SIOCGIFMEDIA:
650 sc = ifp->if_softc;
651 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
652 break;
653
654 case SIOCSIFMTU:
655 /* We basically allow all kinds of MTUs. */
656 ifp->if_mtu = ifr->ifr_mtu;
657 error = 0;
658 break;
659
660 default:
661 /* Let the common ethernet handler process this. */
662 error = ether_ioctl(ifp, cmd, data);
663 break;
664 }
665
666 return (error);
667 }
668
669 static void
670 epair_init(void *dummy __unused)
671 {
672 }
673
674
675 /*
676 * Interface cloning functions.
677 * We use our private ones so that we can create/destroy our secondary
678 * device along with the primary one.
679 */
680 static int
681 epair_clone_match(struct if_clone *ifc, const char *name)
682 {
683 const char *cp;
684
685 DPRINTF("name='%s'\n", name);
686
687 /*
688 * Our base name is epair.
689 * Our interfaces will be named epair<n>[ab].
690 * So accept anything of the following list:
691 * - epair
692 * - epair<n>
693 * but not the epair<n>[ab] versions.
694 */
695 if (strncmp(EPAIRNAME, name, sizeof(EPAIRNAME)-1) != 0)
696 return (0);
697
698 for (cp = name + sizeof(EPAIRNAME) - 1; *cp != '\0'; cp++) {
699 if (*cp < '' || *cp > '9')
700 return (0);
701 }
702
703 return (1);
704 }
705
706 static int
707 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
708 {
709 struct epair_softc *sca, *scb;
710 struct ifnet *ifp;
711 char *dp;
712 int error, unit, wildcard;
713 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */
714
715 /*
716 * We are abusing params to create our second interface.
717 * Actually we already created it and called if_clone_createif()
718 * for it to do the official insertion procedure the moment we knew
719 * it cannot fail anymore. So just do attach it here.
720 */
721 if (params) {
722 scb = (struct epair_softc *)params;
723 ifp = scb->ifp;
724 /* Assign a hopefully unique, locally administered etheraddr. */
725 eaddr[0] = 0x02;
726 eaddr[3] = (ifp->if_index >> 8) & 0xff;
727 eaddr[4] = ifp->if_index & 0xff;
728 eaddr[5] = 0x0b;
729 ether_ifattach(ifp, eaddr);
730 /* Correctly set the name for the cloner list. */
731 strlcpy(name, scb->ifp->if_xname, len);
732 return (0);
733 }
734
735 /* Try to see if a special unit was requested. */
736 error = ifc_name2unit(name, &unit);
737 if (error != 0)
738 return (error);
739 wildcard = (unit < 0);
740
741 error = ifc_alloc_unit(ifc, &unit);
742 if (error != 0)
743 return (error);
744
745 /*
746 * If no unit had been given, we need to adjust the ifName.
747 * Also make sure there is space for our extra [ab] suffix.
748 */
749 for (dp = name; *dp != '\0'; dp++);
750 if (wildcard) {
751 error = snprintf(dp, len - (dp - name), "%d", unit);
752 if (error > len - (dp - name) - 1) {
753 /* ifName too long. */
754 ifc_free_unit(ifc, unit);
755 return (ENOSPC);
756 }
757 dp += error;
758 }
759 if (len - (dp - name) - 1 < 1) {
760 /* No space left for our [ab] suffix. */
761 ifc_free_unit(ifc, unit);
762 return (ENOSPC);
763 }
764 *dp = 'a';
765 /* Must not change dp so we can replace 'a' by 'b' later. */
766 *(dp+1) = '\0';
767
768 /* Allocate memory for both [ab] interfaces */
769 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
770 EPAIR_REFCOUNT_INIT(&sca->refcount, 1);
771 sca->ifp = if_alloc(IFT_ETHER);
772 if (sca->ifp == NULL) {
773 free(sca, M_EPAIR);
774 ifc_free_unit(ifc, unit);
775 return (ENOSPC);
776 }
777
778 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO);
779 EPAIR_REFCOUNT_INIT(&scb->refcount, 1);
780 scb->ifp = if_alloc(IFT_ETHER);
781 if (scb->ifp == NULL) {
782 free(scb, M_EPAIR);
783 if_free(sca->ifp);
784 free(sca, M_EPAIR);
785 ifc_free_unit(ifc, unit);
786 return (ENOSPC);
787 }
788
789 /*
790 * Cross-reference the interfaces so we will be able to free both.
791 */
792 sca->oifp = scb->ifp;
793 scb->oifp = sca->ifp;
794
795 /*
796 * Calculate the cpuid for netisr queueing based on the
797 * ifIndex of the interfaces. As long as we cannot configure
798 * this or use cpuset information easily we cannot guarantee
799 * cache locality but we can at least allow parallelism.
800 */
801 sca->cpuid =
802 netisr_get_cpuid(sca->ifp->if_index % netisr_get_cpucount());
803 scb->cpuid =
804 netisr_get_cpuid(scb->ifp->if_index % netisr_get_cpucount());
805
806 /* Finish initialization of interface <n>a. */
807 ifp = sca->ifp;
808 ifp->if_softc = sca;
809 strlcpy(ifp->if_xname, name, IFNAMSIZ);
810 ifp->if_dname = ifc->ifc_name;
811 ifp->if_dunit = unit;
812 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
813 ifp->if_capabilities = IFCAP_VLAN_MTU;
814 ifp->if_capenable = IFCAP_VLAN_MTU;
815 ifp->if_start = epair_start;
816 ifp->if_ioctl = epair_ioctl;
817 ifp->if_init = epair_init;
818 ifp->if_snd.ifq_maxlen = ifqmaxlen;
819 /* Assign a hopefully unique, locally administered etheraddr. */
820 eaddr[0] = 0x02;
821 eaddr[3] = (ifp->if_index >> 8) & 0xff;
822 eaddr[4] = ifp->if_index & 0xff;
823 eaddr[5] = 0x0a;
824 ether_ifattach(ifp, eaddr);
825 sca->if_qflush = ifp->if_qflush;
826 ifp->if_qflush = epair_qflush;
827 ifp->if_transmit = epair_transmit;
828 ifp->if_baudrate = IF_Gbps(10UL); /* arbitrary maximum */
829
830 /* Swap the name and finish initialization of interface <n>b. */
831 *dp = 'b';
832
833 ifp = scb->ifp;
834 ifp->if_softc = scb;
835 strlcpy(ifp->if_xname, name, IFNAMSIZ);
836 ifp->if_dname = ifc->ifc_name;
837 ifp->if_dunit = unit;
838 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
839 ifp->if_capabilities = IFCAP_VLAN_MTU;
840 ifp->if_capenable = IFCAP_VLAN_MTU;
841 ifp->if_start = epair_start;
842 ifp->if_ioctl = epair_ioctl;
843 ifp->if_init = epair_init;
844 ifp->if_snd.ifq_maxlen = ifqmaxlen;
845 /* We need to play some tricks here for the second interface. */
846 strlcpy(name, EPAIRNAME, len);
847 error = if_clone_create(name, len, (caddr_t)scb);
848 if (error)
849 panic("%s: if_clone_createif() for our 2nd iface failed: %d",
850 __func__, error);
851 scb->if_qflush = ifp->if_qflush;
852 ifp->if_qflush = epair_qflush;
853 ifp->if_transmit = epair_transmit;
854 ifp->if_baudrate = IF_Gbps(10UL); /* arbitrary maximum */
855
856 /*
857 * Restore name to <n>a as the ifp for this will go into the
858 * cloner list for the initial call.
859 */
860 strlcpy(name, sca->ifp->if_xname, len);
861 DPRINTF("name='%s/%db' created sca=%p scb=%p\n", name, unit, sca, scb);
862
863 /* Initialise pseudo media types. */
864 ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status);
865 ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL);
866 ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T);
867 ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status);
868 ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL);
869 ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T);
870
871 /* Tell the world, that we are ready to rock. */
872 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING;
873 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING;
874 if_link_state_change(sca->ifp, LINK_STATE_UP);
875 if_link_state_change(scb->ifp, LINK_STATE_UP);
876
877 return (0);
878 }
879
880 static int
881 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp)
882 {
883 struct ifnet *oifp;
884 struct epair_softc *sca, *scb;
885 int unit, error;
886
887 DPRINTF("ifp=%p\n", ifp);
888
889 /*
890 * In case we called into if_clone_destroyif() ourselves
891 * again to remove the second interface, the softc will be
892 * NULL. In that case so not do anything but return success.
893 */
894 if (ifp->if_softc == NULL)
895 return (0);
896
897 unit = ifp->if_dunit;
898 sca = ifp->if_softc;
899 oifp = sca->oifp;
900 scb = oifp->if_softc;
901
902 DPRINTF("ifp=%p oifp=%p\n", ifp, oifp);
903 if_link_state_change(ifp, LINK_STATE_DOWN);
904 if_link_state_change(oifp, LINK_STATE_DOWN);
905 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
906 oifp->if_drv_flags &= ~IFF_DRV_RUNNING;
907
908 /*
909 * Get rid of our second half. As the other of the two
910 * interfaces may reside in a different vnet, we need to
911 * switch before freeing them.
912 */
913 CURVNET_SET_QUIET(oifp->if_vnet);
914 ether_ifdetach(oifp);
915 /*
916 * Wait for all packets to be dispatched to if_input.
917 * The numbers can only go down as the interface is
918 * detached so there is no need to use atomics.
919 */
920 DPRINTF("scb refcnt=%u\n", scb->refcount);
921 EPAIR_REFCOUNT_ASSERT(scb->refcount == 1,
922 ("%s: ifp=%p scb->refcount!=1: %d", __func__, oifp, scb->refcount));
923 oifp->if_softc = NULL;
924 error = if_clone_destroyif(ifc, oifp);
925 if (error)
926 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d",
927 __func__, error);
928 if_free(oifp);
929 ifmedia_removeall(&scb->media);
930 free(scb, M_EPAIR);
931 CURVNET_RESTORE();
932
933 ether_ifdetach(ifp);
934 /*
935 * Wait for all packets to be dispatched to if_input.
936 */
937 DPRINTF("sca refcnt=%u\n", sca->refcount);
938 EPAIR_REFCOUNT_ASSERT(sca->refcount == 1,
939 ("%s: ifp=%p sca->refcount!=1: %d", __func__, ifp, sca->refcount));
940 if_free(ifp);
941 ifmedia_removeall(&sca->media);
942 free(sca, M_EPAIR);
943 ifc_free_unit(ifc, unit);
944
945 return (0);
946 }
947
948 static int
949 epair_modevent(module_t mod, int type, void *data)
950 {
951 int qlimit;
952
953 switch (type) {
954 case MOD_LOAD:
955 /* For now limit us to one global mutex and one inq. */
956 epair_dpcpu_init();
957 epair_nh.nh_qlimit = 42 * ifqmaxlen; /* 42 shall be the number. */
958 if (TUNABLE_INT_FETCH("net.link.epair.netisr_maxqlen", &qlimit))
959 epair_nh.nh_qlimit = qlimit;
960 netisr_register(&epair_nh);
961 if_clone_attach(&epair_cloner);
962 if (bootverbose)
963 printf("%s initialized.\n", EPAIRNAME);
964 break;
965 case MOD_UNLOAD:
966 if_clone_detach(&epair_cloner);
967 netisr_unregister(&epair_nh);
968 epair_dpcpu_detach();
969 if (bootverbose)
970 printf("%s unloaded.\n", EPAIRNAME);
971 break;
972 default:
973 return (EOPNOTSUPP);
974 }
975 return (0);
976 }
977
978 static moduledata_t epair_mod = {
979 "if_epair",
980 epair_modevent,
981 0
982 };
983
984 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
985 MODULE_VERSION(if_epair, 1);
Cache object: 06b8cb2d94f833f68a9bb231cfcaad5d
|