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: stable/11/sys/net/if_epair.c 357326 2020-01-31 10:34:38Z kp $"); 52 53#include <sys/param.h> 54#include <sys/kernel.h> 55#include <sys/malloc.h> 56#include <sys/mbuf.h> 57#include <sys/module.h> 58#include <sys/refcount.h> 59#include <sys/queue.h> 60#include <sys/smp.h> 61#include <sys/socket.h> 62#include <sys/sockio.h> 63#include <sys/sysctl.h> 64#include <sys/types.h> 65 66#include <net/bpf.h> 67#include <net/ethernet.h> 68#include <net/if.h> 69#include <net/if_var.h> 70#include <net/if_clone.h> 71#include <net/if_media.h> 72#include <net/if_var.h> 73#include <net/if_types.h> 74#include <net/netisr.h> 75#include <net/vnet.h> 76 77SYSCTL_DECL(_net_link); 78static SYSCTL_NODE(_net_link, OID_AUTO, epair, CTLFLAG_RW, 0, "epair sysctl"); 79 80#ifdef EPAIR_DEBUG 81static int epair_debug = 0; 82SYSCTL_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 91static void epair_nh_sintr(struct mbuf *); 92static struct mbuf *epair_nh_m2cpuid(struct mbuf *, uintptr_t, u_int *); 93static void epair_nh_drainedcpu(u_int); 94 95static void epair_start_locked(struct ifnet *); 96static int epair_media_change(struct ifnet *); 97static void epair_media_status(struct ifnet *, struct ifmediareq *); 98 99static int epair_clone_match(struct if_clone *, const char *); 100static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t); 101static int epair_clone_destroy(struct if_clone *, struct ifnet *); 102 103static const char epairname[] = "epair"; 104 105/* Netisr related definitions and sysctl. */ 106static struct netisr_handler epair_nh = { 107 .nh_name = epairname, 108 .nh_proto = NETISR_EPAIR, 109 .nh_policy = NETISR_POLICY_CPU, 110 .nh_handler = epair_nh_sintr, 111 .nh_m2cpuid = epair_nh_m2cpuid, 112 .nh_drainedcpu = epair_nh_drainedcpu, 113}; 114 115static int 116sysctl_epair_netisr_maxqlen(SYSCTL_HANDLER_ARGS) 117{ 118 int error, qlimit; 119 120 netisr_getqlimit(&epair_nh, &qlimit); 121 error = sysctl_handle_int(oidp, &qlimit, 0, req); 122 if (error || !req->newptr) 123 return (error); 124 if (qlimit < 1) 125 return (EINVAL); 126 return (netisr_setqlimit(&epair_nh, qlimit)); 127} 128SYSCTL_PROC(_net_link_epair, OID_AUTO, netisr_maxqlen, CTLTYPE_INT|CTLFLAG_RW, 129 0, 0, sysctl_epair_netisr_maxqlen, "I", 130 "Maximum if_epair(4) netisr \"hw\" queue length"); 131 132struct epair_softc { 133 struct ifnet *ifp; /* This ifp. */ 134 struct ifnet *oifp; /* other ifp of pair. */ 135 struct ifmedia media; /* Media config (fake). */ 136 u_int refcount; /* # of mbufs in flight. */ 137 u_int cpuid; /* CPU ID assigned upon creation. */ 138 void (*if_qflush)(struct ifnet *); 139 /* Original if_qflush routine. */ 140}; 141 142/* 143 * Per-CPU list of ifps with data in the ifq that needs to be flushed 144 * to the netisr ``hw'' queue before we allow any further direct queuing 145 * to the ``hw'' queue. 146 */ 147struct epair_ifp_drain { 148 STAILQ_ENTRY(epair_ifp_drain) ifp_next; 149 struct ifnet *ifp; 150}; 151STAILQ_HEAD(eid_list, epair_ifp_drain); 152 153#define EPAIR_LOCK_INIT(dpcpu) mtx_init(&(dpcpu)->if_epair_mtx, \ 154 "if_epair", NULL, MTX_DEF) 155#define EPAIR_LOCK_DESTROY(dpcpu) mtx_destroy(&(dpcpu)->if_epair_mtx) 156#define EPAIR_LOCK_ASSERT(dpcpu) mtx_assert(&(dpcpu)->if_epair_mtx, \ 157 MA_OWNED) 158#define EPAIR_LOCK(dpcpu) mtx_lock(&(dpcpu)->if_epair_mtx) 159#define EPAIR_UNLOCK(dpcpu) mtx_unlock(&(dpcpu)->if_epair_mtx) 160 161#ifdef INVARIANTS 162#define EPAIR_REFCOUNT_INIT(r, v) refcount_init((r), (v)) 163#define EPAIR_REFCOUNT_AQUIRE(r) refcount_acquire((r)) 164#define EPAIR_REFCOUNT_RELEASE(r) refcount_release((r)) 165#define EPAIR_REFCOUNT_ASSERT(a, p) KASSERT(a, p) 166#else 167#define EPAIR_REFCOUNT_INIT(r, v) 168#define EPAIR_REFCOUNT_AQUIRE(r) 169#define EPAIR_REFCOUNT_RELEASE(r) 170#define EPAIR_REFCOUNT_ASSERT(a, p) 171#endif 172 173static MALLOC_DEFINE(M_EPAIR, epairname, 174 "Pair of virtual cross-over connected Ethernet-like interfaces"); 175 176static VNET_DEFINE(struct if_clone *, epair_cloner); 177#define V_epair_cloner VNET(epair_cloner) 178 179/* 180 * DPCPU area and functions. 181 */ 182struct epair_dpcpu { 183 struct mtx if_epair_mtx; /* Per-CPU locking. */ 184 int epair_drv_flags; /* Per-CPU ``hw'' drv flags. */ 185 struct eid_list epair_ifp_drain_list; /* Per-CPU list of ifps with 186 * data in the ifq. */ 187}; 188DPCPU_DEFINE(struct epair_dpcpu, epair_dpcpu); 189 190static void 191epair_dpcpu_init(void) 192{ 193 struct epair_dpcpu *epair_dpcpu; 194 struct eid_list *s; 195 u_int cpuid; 196 197 CPU_FOREACH(cpuid) { 198 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu); 199 200 /* Initialize per-cpu lock. */ 201 EPAIR_LOCK_INIT(epair_dpcpu); 202 203 /* Driver flags are per-cpu as are our netisr "hw" queues. */ 204 epair_dpcpu->epair_drv_flags = 0; 205 206 /* 207 * Initialize per-cpu drain list. 208 * Manually do what STAILQ_HEAD_INITIALIZER would do. 209 */ 210 s = &epair_dpcpu->epair_ifp_drain_list; 211 s->stqh_first = NULL; 212 s->stqh_last = &s->stqh_first; 213 } 214} 215 216static void 217epair_dpcpu_detach(void) 218{ 219 struct epair_dpcpu *epair_dpcpu; 220 u_int cpuid; 221 222 CPU_FOREACH(cpuid) { 223 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu); 224 225 /* Destroy per-cpu lock. */ 226 EPAIR_LOCK_DESTROY(epair_dpcpu); 227 } 228} 229 230/* 231 * Helper functions. 232 */ 233static u_int 234cpuid_from_ifp(struct ifnet *ifp) 235{ 236 struct epair_softc *sc; 237 238 if (ifp == NULL) 239 return (0); 240 sc = ifp->if_softc; 241 242 return (sc->cpuid); 243} 244 245/* 246 * Netisr handler functions. 247 */ 248static void 249epair_nh_sintr(struct mbuf *m) 250{ 251 struct ifnet *ifp; 252 struct epair_softc *sc; 253 254 ifp = m->m_pkthdr.rcvif; 255 (*ifp->if_input)(ifp, m); 256 sc = ifp->if_softc; 257 EPAIR_REFCOUNT_RELEASE(&sc->refcount); 258 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, 259 ("%s: ifp=%p sc->refcount not >= 1: %d", 260 __func__, ifp, sc->refcount)); 261 DPRINTF("ifp=%p refcount=%u\n", ifp, sc->refcount); 262} 263 264static struct mbuf * 265epair_nh_m2cpuid(struct mbuf *m, uintptr_t source, u_int *cpuid) 266{ 267 268 *cpuid = cpuid_from_ifp(m->m_pkthdr.rcvif); 269 270 return (m); 271} 272 273static void 274epair_nh_drainedcpu(u_int cpuid) 275{ 276 struct epair_dpcpu *epair_dpcpu; 277 struct epair_ifp_drain *elm, *tvar; 278 struct ifnet *ifp; 279 280 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu); 281 EPAIR_LOCK(epair_dpcpu); 282 /* 283 * Assume our "hw" queue and possibly ifq will be emptied 284 * again. In case we will overflow the "hw" queue while 285 * draining, epair_start_locked will set IFF_DRV_OACTIVE 286 * again and we will stop and return. 287 */ 288 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list, 289 ifp_next, tvar) { 290 ifp = elm->ifp; 291 epair_dpcpu->epair_drv_flags &= ~IFF_DRV_OACTIVE; 292 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 293 epair_start_locked(ifp); 294 295 IFQ_LOCK(&ifp->if_snd); 296 if (IFQ_IS_EMPTY(&ifp->if_snd)) { 297 struct epair_softc *sc; 298 299 STAILQ_REMOVE(&epair_dpcpu->epair_ifp_drain_list, 300 elm, epair_ifp_drain, ifp_next); 301 /* The cached ifp goes off the list. */ 302 sc = ifp->if_softc; 303 EPAIR_REFCOUNT_RELEASE(&sc->refcount); 304 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, 305 ("%s: ifp=%p sc->refcount not >= 1: %d", 306 __func__, ifp, sc->refcount)); 307 free(elm, M_EPAIR); 308 } 309 IFQ_UNLOCK(&ifp->if_snd); 310 311 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) != 0) { 312 /* Our "hw"q overflew again. */ 313 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE; 314 DPRINTF("hw queue length overflow at %u\n", 315 epair_nh.nh_qlimit); 316 break; 317 } 318 } 319 EPAIR_UNLOCK(epair_dpcpu); 320} 321 322/* 323 * Network interface (`if') related functions. 324 */ 325static void 326epair_remove_ifp_from_draining(struct ifnet *ifp) 327{ 328 struct epair_dpcpu *epair_dpcpu; 329 struct epair_ifp_drain *elm, *tvar; 330 u_int cpuid; 331 332 CPU_FOREACH(cpuid) { 333 epair_dpcpu = DPCPU_ID_PTR(cpuid, epair_dpcpu); 334 EPAIR_LOCK(epair_dpcpu); 335 STAILQ_FOREACH_SAFE(elm, &epair_dpcpu->epair_ifp_drain_list, 336 ifp_next, tvar) { 337 if (ifp == elm->ifp) { 338 struct epair_softc *sc; 339 340 STAILQ_REMOVE( 341 &epair_dpcpu->epair_ifp_drain_list, elm, 342 epair_ifp_drain, ifp_next); 343 /* The cached ifp goes off the list. */ 344 sc = ifp->if_softc; 345 EPAIR_REFCOUNT_RELEASE(&sc->refcount); 346 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, 347 ("%s: ifp=%p sc->refcount not >= 1: %d", 348 __func__, ifp, sc->refcount)); 349 free(elm, M_EPAIR); 350 } 351 } 352 EPAIR_UNLOCK(epair_dpcpu); 353 } 354} 355 356static int 357epair_add_ifp_for_draining(struct ifnet *ifp) 358{ 359 struct epair_dpcpu *epair_dpcpu; 360 struct epair_softc *sc; 361 struct epair_ifp_drain *elm = NULL; 362 363 sc = ifp->if_softc; 364 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu); 365 EPAIR_LOCK_ASSERT(epair_dpcpu); 366 STAILQ_FOREACH(elm, &epair_dpcpu->epair_ifp_drain_list, ifp_next) 367 if (elm->ifp == ifp) 368 break; 369 /* If the ifp is there already, return success. */ 370 if (elm != NULL) 371 return (0); 372 373 elm = malloc(sizeof(struct epair_ifp_drain), M_EPAIR, M_NOWAIT|M_ZERO); 374 if (elm == NULL) 375 return (ENOMEM); 376 377 elm->ifp = ifp; 378 /* Add a reference for the ifp pointer on the list. */ 379 EPAIR_REFCOUNT_AQUIRE(&sc->refcount); 380 STAILQ_INSERT_TAIL(&epair_dpcpu->epair_ifp_drain_list, elm, ifp_next); 381 382 return (0); 383} 384 385static void 386epair_start_locked(struct ifnet *ifp) 387{ 388 struct epair_dpcpu *epair_dpcpu; 389 struct mbuf *m; 390 struct epair_softc *sc; 391 struct ifnet *oifp; 392 int error; 393 394 DPRINTF("ifp=%p\n", ifp); 395 sc = ifp->if_softc; 396 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu); 397 EPAIR_LOCK_ASSERT(epair_dpcpu); 398 399 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 400 return; 401 if ((ifp->if_flags & IFF_UP) == 0) 402 return; 403 404 /* 405 * We get packets here from ether_output via if_handoff() 406 * and need to put them into the input queue of the oifp 407 * and call oifp->if_input() via netisr/epair_sintr(). 408 */ 409 oifp = sc->oifp; 410 sc = oifp->if_softc; 411 for (;;) { 412 IFQ_DEQUEUE(&ifp->if_snd, m); 413 if (m == NULL) 414 break; 415 BPF_MTAP(ifp, m); 416 417 /* 418 * In case the outgoing interface is not usable, 419 * drop the packet. 420 */ 421 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 422 (oifp->if_flags & IFF_UP) ==0) { 423 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 424 m_freem(m); 425 continue; 426 } 427 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname); 428 429 /* 430 * Add a reference so the interface cannot go while the 431 * packet is in transit as we rely on rcvif to stay valid. 432 */ 433 EPAIR_REFCOUNT_AQUIRE(&sc->refcount); 434 m->m_pkthdr.rcvif = oifp; 435 CURVNET_SET_QUIET(oifp->if_vnet); 436 error = netisr_queue(NETISR_EPAIR, m); 437 CURVNET_RESTORE(); 438 if (!error) { 439 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 440 /* Someone else received the packet. */ 441 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1); 442 } else { 443 /* The packet was freed already. */ 444 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE; 445 ifp->if_drv_flags |= IFF_DRV_OACTIVE; 446 (void) epair_add_ifp_for_draining(ifp); 447 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 448 EPAIR_REFCOUNT_RELEASE(&sc->refcount); 449 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, 450 ("%s: ifp=%p sc->refcount not >= 1: %d", 451 __func__, oifp, sc->refcount)); 452 } 453 } 454} 455 456static void 457epair_start(struct ifnet *ifp) 458{ 459 struct epair_dpcpu *epair_dpcpu; 460 461 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu); 462 EPAIR_LOCK(epair_dpcpu); 463 epair_start_locked(ifp); 464 EPAIR_UNLOCK(epair_dpcpu); 465} 466 467static int 468epair_transmit_locked(struct ifnet *ifp, struct mbuf *m) 469{ 470 struct epair_dpcpu *epair_dpcpu; 471 struct epair_softc *sc; 472 struct ifnet *oifp; 473 int error, len; 474 short mflags; 475 476 DPRINTF("ifp=%p m=%p\n", ifp, m); 477 sc = ifp->if_softc; 478 epair_dpcpu = DPCPU_ID_PTR(sc->cpuid, epair_dpcpu); 479 EPAIR_LOCK_ASSERT(epair_dpcpu); 480 481 if (m == NULL) 482 return (0); 483 484 /* 485 * We are not going to use the interface en/dequeue mechanism 486 * on the TX side. We are called from ether_output_frame() 487 * and will put the packet into the incoming queue of the 488 * other interface of our pair via the netsir. 489 */ 490 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 491 m_freem(m); 492 return (ENXIO); 493 } 494 if ((ifp->if_flags & IFF_UP) == 0) { 495 m_freem(m); 496 return (ENETDOWN); 497 } 498 499 BPF_MTAP(ifp, m); 500 501 /* 502 * In case the outgoing interface is not usable, 503 * drop the packet. 504 */ 505 oifp = sc->oifp; 506 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 507 (oifp->if_flags & IFF_UP) ==0) { 508 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 509 m_freem(m); 510 return (0); 511 } 512 len = m->m_pkthdr.len; 513 mflags = m->m_flags; 514 DPRINTF("packet %s -> %s\n", ifp->if_xname, oifp->if_xname); 515 516#ifdef ALTQ 517 /* Support ALTQ via the classic if_start() path. */ 518 IF_LOCK(&ifp->if_snd); 519 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 520 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error); 521 if (error) 522 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); 523 IF_UNLOCK(&ifp->if_snd); 524 if (!error) { 525 if_inc_counter(ifp, IFCOUNTER_OBYTES, len); 526 if (mflags & (M_BCAST|M_MCAST)) 527 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); 528 529 if ((ifp->if_drv_flags & IFF_DRV_OACTIVE) == 0) 530 epair_start_locked(ifp); 531 else 532 (void)epair_add_ifp_for_draining(ifp); 533 } 534 return (error); 535 } 536 IF_UNLOCK(&ifp->if_snd); 537#endif 538 539 if ((epair_dpcpu->epair_drv_flags & IFF_DRV_OACTIVE) != 0) { 540 /* 541 * Our hardware queue is full, try to fall back 542 * queuing to the ifq but do not call ifp->if_start. 543 * Either we are lucky or the packet is gone. 544 */ 545 IFQ_ENQUEUE(&ifp->if_snd, m, error); 546 if (!error) 547 (void)epair_add_ifp_for_draining(ifp); 548 return (error); 549 } 550 sc = oifp->if_softc; 551 /* 552 * Add a reference so the interface cannot go while the 553 * packet is in transit as we rely on rcvif to stay valid. 554 */ 555 EPAIR_REFCOUNT_AQUIRE(&sc->refcount); 556 m->m_pkthdr.rcvif = oifp; 557 CURVNET_SET_QUIET(oifp->if_vnet); 558 error = netisr_queue(NETISR_EPAIR, m); 559 CURVNET_RESTORE(); 560 if (!error) { 561 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 562 /* 563 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics, 564 * but as we bypass all this we have to duplicate 565 * the logic another time. 566 */ 567 if_inc_counter(ifp, IFCOUNTER_OBYTES, len); 568 if (mflags & (M_BCAST|M_MCAST)) 569 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); 570 /* Someone else received the packet. */ 571 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1); 572 } else { 573 /* The packet was freed already. */ 574 epair_dpcpu->epair_drv_flags |= IFF_DRV_OACTIVE; 575 ifp->if_drv_flags |= IFF_DRV_OACTIVE; 576 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 577 EPAIR_REFCOUNT_RELEASE(&sc->refcount); 578 EPAIR_REFCOUNT_ASSERT((int)sc->refcount >= 1, 579 ("%s: ifp=%p sc->refcount not >= 1: %d", 580 __func__, oifp, sc->refcount)); 581 } 582 583 return (error); 584} 585 586static int 587epair_transmit(struct ifnet *ifp, struct mbuf *m) 588{ 589 struct epair_dpcpu *epair_dpcpu; 590 int error; 591 592 epair_dpcpu = DPCPU_ID_PTR(cpuid_from_ifp(ifp), epair_dpcpu); 593 EPAIR_LOCK(epair_dpcpu); 594 error = epair_transmit_locked(ifp, m); 595 EPAIR_UNLOCK(epair_dpcpu); 596 return (error); 597} 598 599static void 600epair_qflush(struct ifnet *ifp) 601{ 602 struct epair_softc *sc; 603 604 sc = ifp->if_softc; 605 KASSERT(sc != NULL, ("%s: ifp=%p, epair_softc gone? sc=%p\n", 606 __func__, ifp, sc)); 607 /* 608 * Remove this ifp from all backpointer lists. The interface will not 609 * usable for flushing anyway nor should it have anything to flush 610 * after if_qflush(). 611 */ 612 epair_remove_ifp_from_draining(ifp); 613 614 if (sc->if_qflush) 615 sc->if_qflush(ifp); 616} 617 618static int 619epair_media_change(struct ifnet *ifp __unused) 620{ 621 622 /* Do nothing. */ 623 return (0); 624} 625 626static void 627epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr) 628{ 629 630 imr->ifm_status = IFM_AVALID | IFM_ACTIVE; 631 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX; 632} 633 634static int 635epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 636{ 637 struct epair_softc *sc; 638 struct ifreq *ifr; 639 int error; 640 641 ifr = (struct ifreq *)data; 642 switch (cmd) { 643 case SIOCSIFFLAGS: 644 case SIOCADDMULTI: 645 case SIOCDELMULTI: 646 error = 0; 647 break; 648 649 case SIOCSIFMEDIA: 650 case SIOCGIFMEDIA: 651 sc = ifp->if_softc; 652 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd); 653 break; 654 655 case SIOCSIFMTU: 656 /* We basically allow all kinds of MTUs. */ 657 ifp->if_mtu = ifr->ifr_mtu; 658 error = 0; 659 break; 660 661 default: 662 /* Let the common ethernet handler process this. */ 663 error = ether_ioctl(ifp, cmd, data); 664 break; 665 } 666 667 return (error); 668} 669 670static void 671epair_init(void *dummy __unused) 672{ 673} 674 675 676/* 677 * Interface cloning functions. 678 * We use our private ones so that we can create/destroy our secondary 679 * device along with the primary one. 680 */ 681static int 682epair_clone_match(struct if_clone *ifc, const char *name) 683{ 684 const char *cp; 685 686 DPRINTF("name='%s'\n", name); 687 688 /* 689 * Our base name is epair. 690 * Our interfaces will be named epair<n>[ab]. 691 * So accept anything of the following list: 692 * - epair 693 * - epair<n> 694 * but not the epair<n>[ab] versions. 695 */ 696 if (strncmp(epairname, name, sizeof(epairname)-1) != 0) 697 return (0); 698 699 for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) { 700 if (*cp < '0' || *cp > '9') 701 return (0); 702 } 703 704 return (1); 705} 706 707static void 708epair_clone_add(struct if_clone *ifc, struct epair_softc *scb) 709{ 710 struct ifnet *ifp; 711 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ 712 713 ifp = scb->ifp; 714 /* Assign a hopefully unique, locally administered etheraddr. */ 715 eaddr[0] = 0x02; 716 eaddr[3] = (ifp->if_index >> 8) & 0xff; 717 eaddr[4] = ifp->if_index & 0xff; 718 eaddr[5] = 0x0b; 719 ether_ifattach(ifp, eaddr); 720 721 if_clone_addif(ifc, ifp); 722} 723 724static int 725epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) 726{ 727 struct epair_softc *sca, *scb; 728 struct ifnet *ifp; 729 char *dp; 730 int error, unit, wildcard; 731 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ 732 733 /* Try to see if a special unit was requested. */ 734 error = ifc_name2unit(name, &unit); 735 if (error != 0) 736 return (error); 737 wildcard = (unit < 0); 738 739 error = ifc_alloc_unit(ifc, &unit); 740 if (error != 0) 741 return (error); 742 743 /* 744 * If no unit had been given, we need to adjust the ifName. 745 * Also make sure there is space for our extra [ab] suffix. 746 */ 747 for (dp = name; *dp != '\0'; dp++); 748 if (wildcard) { 749 error = snprintf(dp, len - (dp - name), "%d", unit); 750 if (error > len - (dp - name) - 1) { 751 /* ifName too long. */ 752 ifc_free_unit(ifc, unit); 753 return (ENOSPC); 754 } 755 dp += error; 756 } 757 if (len - (dp - name) - 1 < 1) { 758 /* No space left for our [ab] suffix. */ 759 ifc_free_unit(ifc, unit); 760 return (ENOSPC); 761 } 762 *dp = 'b'; 763 /* Must not change dp so we can replace 'a' by 'b' later. */ 764 *(dp+1) = '\0'; 765 766 /* Check if 'a' and 'b' interfaces already exist. */ 767 if (ifunit(name) != NULL) 768 return (EEXIST); 769 *dp = 'a'; 770 if (ifunit(name) != NULL) 771 return (EEXIST); 772 773 /* Allocate memory for both [ab] interfaces */ 774 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); 775 EPAIR_REFCOUNT_INIT(&sca->refcount, 1); 776 sca->ifp = if_alloc(IFT_ETHER); 777 if (sca->ifp == NULL) { 778 free(sca, M_EPAIR); 779 ifc_free_unit(ifc, unit); 780 return (ENOSPC); 781 } 782 783 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); 784 EPAIR_REFCOUNT_INIT(&scb->refcount, 1); 785 scb->ifp = if_alloc(IFT_ETHER); 786 if (scb->ifp == NULL) { 787 free(scb, M_EPAIR); 788 if_free(sca->ifp); 789 free(sca, M_EPAIR); 790 ifc_free_unit(ifc, unit); 791 return (ENOSPC); 792 } 793 794 /* 795 * Cross-reference the interfaces so we will be able to free both. 796 */ 797 sca->oifp = scb->ifp; 798 scb->oifp = sca->ifp; 799 800 /* 801 * Calculate the cpuid for netisr queueing based on the 802 * ifIndex of the interfaces. As long as we cannot configure 803 * this or use cpuset information easily we cannot guarantee 804 * cache locality but we can at least allow parallelism. 805 */ 806 sca->cpuid = 807 netisr_get_cpuid(sca->ifp->if_index); 808 scb->cpuid = 809 netisr_get_cpuid(scb->ifp->if_index); 810 811 /* Initialise pseudo media types. */ 812 ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status); 813 ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL); 814 ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T); 815 ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status); 816 ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL); 817 ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T); 818 819 /* Finish initialization of interface <n>a. */ 820 ifp = sca->ifp; 821 ifp->if_softc = sca; 822 strlcpy(ifp->if_xname, name, IFNAMSIZ); 823 ifp->if_dname = epairname; 824 ifp->if_dunit = unit; 825 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 826 ifp->if_capabilities = IFCAP_VLAN_MTU; 827 ifp->if_capenable = IFCAP_VLAN_MTU; 828 ifp->if_start = epair_start; 829 ifp->if_ioctl = epair_ioctl; 830 ifp->if_init = epair_init; 831 ifp->if_snd.ifq_maxlen = ifqmaxlen; 832 /* Assign a hopefully unique, locally administered etheraddr. */ 833 eaddr[0] = 0x02; 834 eaddr[3] = (ifp->if_index >> 8) & 0xff; 835 eaddr[4] = ifp->if_index & 0xff; 836 eaddr[5] = 0x0a; 837 ether_ifattach(ifp, eaddr); 838 sca->if_qflush = ifp->if_qflush; 839 ifp->if_qflush = epair_qflush; 840 ifp->if_transmit = epair_transmit; 841 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */ 842 843 /* Swap the name and finish initialization of interface <n>b. */ 844 *dp = 'b'; 845 846 ifp = scb->ifp; 847 ifp->if_softc = scb; 848 strlcpy(ifp->if_xname, name, IFNAMSIZ); 849 ifp->if_dname = epairname; 850 ifp->if_dunit = unit; 851 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 852 ifp->if_capabilities = IFCAP_VLAN_MTU; 853 ifp->if_capenable = IFCAP_VLAN_MTU; 854 ifp->if_start = epair_start; 855 ifp->if_ioctl = epair_ioctl; 856 ifp->if_init = epair_init; 857 ifp->if_snd.ifq_maxlen = ifqmaxlen; 858 /* We need to play some tricks here for the second interface. */ 859 strlcpy(name, epairname, len); 860 861 /* Correctly set the name for the cloner list. */ 862 strlcpy(name, scb->ifp->if_xname, len); 863 epair_clone_add(ifc, scb); 864 865 scb->if_qflush = ifp->if_qflush; 866 ifp->if_qflush = epair_qflush; 867 ifp->if_transmit = epair_transmit; 868 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */ 869 870 /* 871 * Restore name to <n>a as the ifp for this will go into the 872 * cloner list for the initial call. 873 */ 874 strlcpy(name, sca->ifp->if_xname, len); 875 DPRINTF("name='%s/%db' created sca=%p scb=%p\n", name, unit, sca, scb); 876 877 /* Tell the world, that we are ready to rock. */ 878 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING; 879 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING; 880 if_link_state_change(sca->ifp, LINK_STATE_UP); 881 if_link_state_change(scb->ifp, LINK_STATE_UP); 882 883 return (0); 884} 885 886static int 887epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp) 888{ 889 struct ifnet *oifp; 890 struct epair_softc *sca, *scb; 891 int unit, error; 892 893 DPRINTF("ifp=%p\n", ifp); 894 895 /* 896 * In case we called into if_clone_destroyif() ourselves 897 * again to remove the second interface, the softc will be 898 * NULL. In that case so not do anything but return success. 899 */ 900 if (ifp->if_softc == NULL) 901 return (0); 902 903 unit = ifp->if_dunit; 904 sca = ifp->if_softc; 905 oifp = sca->oifp; 906 scb = oifp->if_softc; 907 908 DPRINTF("ifp=%p oifp=%p\n", ifp, oifp); 909 if_link_state_change(ifp, LINK_STATE_DOWN); 910 if_link_state_change(oifp, LINK_STATE_DOWN); 911 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 912 oifp->if_drv_flags &= ~IFF_DRV_RUNNING; 913 914 /* 915 * Get rid of our second half. As the other of the two 916 * interfaces may reside in a different vnet, we need to 917 * switch before freeing them. 918 */ 919 CURVNET_SET_QUIET(oifp->if_vnet); 920 ether_ifdetach(oifp); 921 /* 922 * Wait for all packets to be dispatched to if_input. 923 * The numbers can only go down as the interface is 924 * detached so there is no need to use atomics. 925 */ 926 DPRINTF("scb refcnt=%u\n", scb->refcount); 927 EPAIR_REFCOUNT_ASSERT(scb->refcount == 1, 928 ("%s: ifp=%p scb->refcount!=1: %d", __func__, oifp, scb->refcount)); 929 oifp->if_softc = NULL; 930 error = if_clone_destroyif(ifc, oifp); 931 if (error) 932 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d", 933 __func__, error); 934 if_free(oifp); 935 ifmedia_removeall(&scb->media); 936 free(scb, M_EPAIR); 937 CURVNET_RESTORE(); 938 939 ether_ifdetach(ifp); 940 /* 941 * Wait for all packets to be dispatched to if_input. 942 */ 943 DPRINTF("sca refcnt=%u\n", sca->refcount); 944 EPAIR_REFCOUNT_ASSERT(sca->refcount == 1, 945 ("%s: ifp=%p sca->refcount!=1: %d", __func__, ifp, sca->refcount)); 946 if_free(ifp); 947 ifmedia_removeall(&sca->media); 948 free(sca, M_EPAIR); 949 ifc_free_unit(ifc, unit); 950 951 return (0); 952} 953 954static void 955vnet_epair_init(const void *unused __unused) 956{ 957 958 V_epair_cloner = if_clone_advanced(epairname, 0, 959 epair_clone_match, epair_clone_create, epair_clone_destroy); 960#ifdef VIMAGE 961 netisr_register_vnet(&epair_nh); 962#endif 963} 964VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY, 965 vnet_epair_init, NULL); 966 967static void 968vnet_epair_uninit(const void *unused __unused) 969{ 970 971#ifdef VIMAGE 972 netisr_unregister_vnet(&epair_nh); 973#endif 974 if_clone_detach(V_epair_cloner); 975} 976VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, 977 vnet_epair_uninit, NULL); 978 979static int 980epair_modevent(module_t mod, int type, void *data) 981{ 982 int qlimit; 983 984 switch (type) { 985 case MOD_LOAD: 986 /* For now limit us to one global mutex and one inq. */ 987 epair_dpcpu_init(); 988 epair_nh.nh_qlimit = 42 * ifqmaxlen; /* 42 shall be the number. */ 989 if (TUNABLE_INT_FETCH("net.link.epair.netisr_maxqlen", &qlimit)) 990 epair_nh.nh_qlimit = qlimit; 991 netisr_register(&epair_nh); 992 if (bootverbose) 993 printf("%s initialized.\n", epairname); 994 break; 995 case MOD_UNLOAD: 996 netisr_unregister(&epair_nh); 997 epair_dpcpu_detach(); 998 if (bootverbose) 999 printf("%s unloaded.\n", epairname); 1000 break; 1001 default: 1002 return (EOPNOTSUPP); 1003 } 1004 return (0); 1005} 1006 1007static moduledata_t epair_mod = { 1008 "if_epair", 1009 epair_modevent, 1010 0 1011}; 1012 1013DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE); 1014MODULE_VERSION(if_epair, 1); 1015