1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2008 The FreeBSD Foundation 5 * Copyright (c) 2009-2021 Bjoern A. Zeeb <bz@FreeBSD.org> 6 * 7 * This software was developed by CK Software GmbH under sponsorship 8 * from the FreeBSD Foundation. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * A pair of virtual back-to-back connected ethernet like interfaces 34 * (``two interfaces with a virtual cross-over cable''). 35 * 36 * This is mostly intended to be used to provide connectivity between 37 * different virtual network stack instances. 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_rss.h" 44 45 #include <sys/param.h> 46 #include <sys/hash.h> 47 #include <sys/jail.h> 48 #include <sys/kernel.h> 49 #include <sys/libkern.h> 50 #include <sys/malloc.h> 51 #include <sys/mbuf.h> 52 #include <sys/module.h> 53 #include <sys/proc.h> 54 #include <sys/queue.h> 55 #include <sys/sched.h> 56 #include <sys/smp.h> 57 #include <sys/socket.h> 58 #include <sys/sockio.h> 59 #include <sys/taskqueue.h> 60 #include <sys/types.h> 61 #include <sys/buf_ring.h> 62 #include <sys/bus.h> 63 #include <sys/interrupt.h> 64 65 #include <net/bpf.h> 66 #include <net/ethernet.h> 67 #include <net/if.h> 68 #include <net/if_var.h> 69 #include <net/if_clone.h> 70 #include <net/if_media.h> 71 #include <net/if_var.h> 72 #include <net/if_types.h> 73 #include <net/netisr.h> 74 #ifdef RSS 75 #include <net/rss_config.h> 76 #include <netinet/in_rss.h> 77 #include <netinet6/in6_rss.h> 78 #endif 79 #include <net/vnet.h> 80 81 static int epair_clone_match(struct if_clone *, const char *); 82 static int epair_clone_create(struct if_clone *, char *, size_t, caddr_t); 83 static int epair_clone_destroy(struct if_clone *, struct ifnet *); 84 85 static const char epairname[] = "epair"; 86 #define RXRSIZE 4096 /* Probably overkill by 4-8x. */ 87 88 static MALLOC_DEFINE(M_EPAIR, epairname, 89 "Pair of virtual cross-over connected Ethernet-like interfaces"); 90 91 VNET_DEFINE_STATIC(struct if_clone *, epair_cloner); 92 #define V_epair_cloner VNET(epair_cloner) 93 94 static unsigned int next_index = 0; 95 #define EPAIR_LOCK_INIT() mtx_init(&epair_n_index_mtx, "epairidx", \ 96 NULL, MTX_DEF) 97 #define EPAIR_LOCK_DESTROY() mtx_destroy(&epair_n_index_mtx) 98 #define EPAIR_LOCK() mtx_lock(&epair_n_index_mtx) 99 #define EPAIR_UNLOCK() mtx_unlock(&epair_n_index_mtx) 100 101 struct epair_softc; 102 struct epair_queue { 103 int id; 104 struct buf_ring *rxring[2]; 105 volatile int ridx; /* 0 || 1 */ 106 struct task tx_task; 107 struct epair_softc *sc; 108 }; 109 110 static struct mtx epair_n_index_mtx; 111 struct epair_softc { 112 struct ifnet *ifp; /* This ifp. */ 113 struct ifnet *oifp; /* other ifp of pair. */ 114 int num_queues; 115 struct epair_queue *queues; 116 struct ifmedia media; /* Media config (fake). */ 117 STAILQ_ENTRY(epair_softc) entry; 118 }; 119 120 struct epair_tasks_t { 121 int tasks; 122 struct taskqueue *tq[MAXCPU]; 123 }; 124 125 static struct epair_tasks_t epair_tasks; 126 127 static void 128 epair_clear_mbuf(struct mbuf *m) 129 { 130 /* Remove any CSUM_SND_TAG as ether_input will barf. */ 131 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) { 132 m_snd_tag_rele(m->m_pkthdr.snd_tag); 133 m->m_pkthdr.snd_tag = NULL; 134 m->m_pkthdr.csum_flags &= ~CSUM_SND_TAG; 135 } 136 137 m_tag_delete_nonpersistent(m); 138 } 139 140 static void 141 epair_if_input(struct epair_softc *sc, struct epair_queue *q, int ridx) 142 { 143 struct ifnet *ifp; 144 struct mbuf *m; 145 146 ifp = sc->ifp; 147 CURVNET_SET(ifp->if_vnet); 148 while (! buf_ring_empty(q->rxring[ridx])) { 149 m = buf_ring_dequeue_mc(q->rxring[ridx]); 150 if (m == NULL) 151 continue; 152 153 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); 154 (*ifp->if_input)(ifp, m); 155 156 } 157 CURVNET_RESTORE(); 158 } 159 160 static void 161 epair_tx_start_deferred(void *arg, int pending) 162 { 163 struct epair_queue *q = (struct epair_queue *)arg; 164 struct epair_softc *sc = q->sc; 165 int ridx, nidx; 166 167 if_ref(sc->ifp); 168 ridx = atomic_load_int(&q->ridx); 169 do { 170 nidx = (ridx == 0) ? 1 : 0; 171 } while (!atomic_fcmpset_int(&q->ridx, &ridx, nidx)); 172 epair_if_input(sc, q, ridx); 173 174 if (! buf_ring_empty(q->rxring[nidx])) 175 taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task); 176 177 if_rele(sc->ifp); 178 } 179 180 static int 181 epair_menq(struct mbuf *m, struct epair_softc *osc) 182 { 183 struct ifnet *ifp, *oifp; 184 int len, ret; 185 int ridx; 186 short mflags; 187 struct epair_queue *q = NULL; 188 uint32_t bucket; 189 bool was_empty; 190 #ifdef RSS 191 struct ether_header *eh; 192 #endif 193 194 /* 195 * I know this looks weird. We pass the "other sc" as we need that one 196 * and can get both ifps from it as well. 197 */ 198 oifp = osc->ifp; 199 ifp = osc->oifp; 200 201 M_ASSERTPKTHDR(m); 202 epair_clear_mbuf(m); 203 if_setrcvif(m, oifp); 204 M_SETFIB(m, oifp->if_fib); 205 206 /* Save values as once the mbuf is queued, it's not ours anymore. */ 207 len = m->m_pkthdr.len; 208 mflags = m->m_flags; 209 210 MPASS(m->m_nextpkt == NULL); 211 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); 212 213 #ifdef RSS 214 ret = rss_m2bucket(m, &bucket); 215 if (ret) { 216 /* Actually hash the packet. */ 217 eh = mtod(m, struct ether_header *); 218 219 switch (ntohs(eh->ether_type)) { 220 case ETHERTYPE_IP: 221 rss_soft_m2cpuid_v4(m, 0, &bucket); 222 break; 223 case ETHERTYPE_IPV6: 224 rss_soft_m2cpuid_v6(m, 0, &bucket); 225 break; 226 default: 227 bucket = 0; 228 break; 229 } 230 } 231 bucket %= osc->num_queues; 232 #else 233 bucket = 0; 234 #endif 235 q = &osc->queues[bucket]; 236 237 ridx = atomic_load_int(&q->ridx); 238 was_empty = buf_ring_empty(q->rxring[ridx]); 239 ret = buf_ring_enqueue(q->rxring[ridx], m); 240 if (ret != 0) { 241 /* Ring is full. */ 242 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); 243 m_freem(m); 244 goto done; 245 } 246 247 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 248 /* 249 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics, 250 * but as we bypass all this we have to duplicate 251 * the logic another time. 252 */ 253 if_inc_counter(ifp, IFCOUNTER_OBYTES, len); 254 if (mflags & (M_BCAST|M_MCAST)) 255 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); 256 /* Someone else received the packet. */ 257 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1); 258 259 done: 260 if (was_empty) 261 taskqueue_enqueue(epair_tasks.tq[bucket], &q->tx_task); 262 263 return (0); 264 } 265 266 static void 267 epair_start(struct ifnet *ifp) 268 { 269 struct mbuf *m; 270 struct epair_softc *sc; 271 struct ifnet *oifp; 272 273 /* 274 * We get packets here from ether_output via if_handoff() 275 * and need to put them into the input queue of the oifp 276 * and will put the packet into the receive-queue (rxq) of the 277 * other interface (oifp) of our pair. 278 */ 279 sc = ifp->if_softc; 280 oifp = sc->oifp; 281 sc = oifp->if_softc; 282 for (;;) { 283 IFQ_DEQUEUE(&ifp->if_snd, m); 284 if (m == NULL) 285 break; 286 M_ASSERTPKTHDR(m); 287 BPF_MTAP(ifp, m); 288 289 /* In case either interface is not usable drop the packet. */ 290 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 291 (ifp->if_flags & IFF_UP) == 0 || 292 (oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 293 (oifp->if_flags & IFF_UP) == 0) { 294 m_freem(m); 295 continue; 296 } 297 298 (void) epair_menq(m, sc); 299 } 300 } 301 302 static int 303 epair_transmit(struct ifnet *ifp, struct mbuf *m) 304 { 305 struct epair_softc *sc; 306 struct ifnet *oifp; 307 int error; 308 #ifdef ALTQ 309 int len; 310 short mflags; 311 #endif 312 313 if (m == NULL) 314 return (0); 315 M_ASSERTPKTHDR(m); 316 317 /* 318 * We are not going to use the interface en/dequeue mechanism 319 * on the TX side. We are called from ether_output_frame() 320 * and will put the packet into the receive-queue (rxq) of the 321 * other interface (oifp) of our pair. 322 */ 323 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 324 m_freem(m); 325 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 326 return (ENXIO); 327 } 328 if ((ifp->if_flags & IFF_UP) == 0) { 329 m_freem(m); 330 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 331 return (ENETDOWN); 332 } 333 334 BPF_MTAP(ifp, m); 335 336 /* 337 * In case the outgoing interface is not usable, 338 * drop the packet. 339 */ 340 sc = ifp->if_softc; 341 oifp = sc->oifp; 342 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 343 (oifp->if_flags & IFF_UP) == 0) { 344 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 345 m_freem(m); 346 return (0); 347 } 348 349 #ifdef ALTQ 350 len = m->m_pkthdr.len; 351 mflags = m->m_flags; 352 353 /* Support ALTQ via the classic if_start() path. */ 354 IF_LOCK(&ifp->if_snd); 355 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 356 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error); 357 if (error) 358 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); 359 IF_UNLOCK(&ifp->if_snd); 360 if (!error) { 361 if_inc_counter(ifp, IFCOUNTER_OBYTES, len); 362 if (mflags & (M_BCAST|M_MCAST)) 363 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); 364 epair_start(ifp); 365 } 366 return (error); 367 } 368 IF_UNLOCK(&ifp->if_snd); 369 #endif 370 371 error = epair_menq(m, oifp->if_softc); 372 return (error); 373 } 374 375 static int 376 epair_media_change(struct ifnet *ifp __unused) 377 { 378 379 /* Do nothing. */ 380 return (0); 381 } 382 383 static void 384 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr) 385 { 386 387 imr->ifm_status = IFM_AVALID | IFM_ACTIVE; 388 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX; 389 } 390 391 static int 392 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 393 { 394 struct epair_softc *sc; 395 struct ifreq *ifr; 396 int error; 397 398 ifr = (struct ifreq *)data; 399 switch (cmd) { 400 case SIOCSIFFLAGS: 401 case SIOCADDMULTI: 402 case SIOCDELMULTI: 403 error = 0; 404 break; 405 406 case SIOCSIFMEDIA: 407 case SIOCGIFMEDIA: 408 sc = ifp->if_softc; 409 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd); 410 break; 411 412 case SIOCSIFMTU: 413 /* We basically allow all kinds of MTUs. */ 414 ifp->if_mtu = ifr->ifr_mtu; 415 error = 0; 416 break; 417 418 default: 419 /* Let the common ethernet handler process this. */ 420 error = ether_ioctl(ifp, cmd, data); 421 break; 422 } 423 424 return (error); 425 } 426 427 static void 428 epair_init(void *dummy __unused) 429 { 430 } 431 432 /* 433 * Interface cloning functions. 434 * We use our private ones so that we can create/destroy our secondary 435 * device along with the primary one. 436 */ 437 static int 438 epair_clone_match(struct if_clone *ifc, const char *name) 439 { 440 const char *cp; 441 442 /* 443 * Our base name is epair. 444 * Our interfaces will be named epair<n>[ab]. 445 * So accept anything of the following list: 446 * - epair 447 * - epair<n> 448 * but not the epair<n>[ab] versions. 449 */ 450 if (strncmp(epairname, name, sizeof(epairname)-1) != 0) 451 return (0); 452 453 for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) { 454 if (*cp < '0' || *cp > '9') 455 return (0); 456 } 457 458 return (1); 459 } 460 461 static void 462 epair_clone_add(struct if_clone *ifc, struct epair_softc *scb) 463 { 464 struct ifnet *ifp; 465 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ 466 467 ifp = scb->ifp; 468 /* Copy epairNa etheraddr and change the last byte. */ 469 memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN); 470 eaddr[5] = 0x0b; 471 ether_ifattach(ifp, eaddr); 472 473 if_clone_addif(ifc, ifp); 474 } 475 476 static int 477 epair_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params) 478 { 479 struct epair_softc *sca, *scb; 480 struct ifnet *ifp; 481 char *dp; 482 int error, unit, wildcard; 483 uint64_t hostid; 484 uint32_t key[3]; 485 uint32_t hash; 486 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ 487 488 /* Try to see if a special unit was requested. */ 489 error = ifc_name2unit(name, &unit); 490 if (error != 0) 491 return (error); 492 wildcard = (unit < 0); 493 494 error = ifc_alloc_unit(ifc, &unit); 495 if (error != 0) 496 return (error); 497 498 /* 499 * If no unit had been given, we need to adjust the ifName. 500 * Also make sure there is space for our extra [ab] suffix. 501 */ 502 for (dp = name; *dp != '\0'; dp++); 503 if (wildcard) { 504 error = snprintf(dp, len - (dp - name), "%d", unit); 505 if (error > len - (dp - name) - 1) { 506 /* ifName too long. */ 507 ifc_free_unit(ifc, unit); 508 return (ENOSPC); 509 } 510 dp += error; 511 } 512 if (len - (dp - name) - 1 < 1) { 513 /* No space left for our [ab] suffix. */ 514 ifc_free_unit(ifc, unit); 515 return (ENOSPC); 516 } 517 *dp = 'b'; 518 /* Must not change dp so we can replace 'a' by 'b' later. */ 519 *(dp+1) = '\0'; 520 521 /* Check if 'a' and 'b' interfaces already exist. */ 522 if (ifunit(name) != NULL) 523 return (EEXIST); 524 *dp = 'a'; 525 if (ifunit(name) != NULL) 526 return (EEXIST); 527 528 /* Allocate memory for both [ab] interfaces */ 529 sca = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); 530 sca->ifp = if_alloc(IFT_ETHER); 531 sca->num_queues = epair_tasks.tasks; 532 if (sca->ifp == NULL) { 533 free(sca, M_EPAIR); 534 ifc_free_unit(ifc, unit); 535 return (ENOSPC); 536 } 537 sca->queues = mallocarray(sca->num_queues, sizeof(struct epair_queue), 538 M_EPAIR, M_WAITOK); 539 for (int i = 0; i < sca->num_queues; i++) { 540 struct epair_queue *q = &sca->queues[i]; 541 q->id = i; 542 q->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); 543 q->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); 544 q->ridx = 0; 545 q->sc = sca; 546 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q); 547 } 548 549 scb = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); 550 scb->ifp = if_alloc(IFT_ETHER); 551 scb->num_queues = epair_tasks.tasks; 552 if (scb->ifp == NULL) { 553 free(scb, M_EPAIR); 554 if_free(sca->ifp); 555 free(sca, M_EPAIR); 556 ifc_free_unit(ifc, unit); 557 return (ENOSPC); 558 } 559 scb->queues = mallocarray(scb->num_queues, sizeof(struct epair_queue), 560 M_EPAIR, M_WAITOK); 561 for (int i = 0; i < scb->num_queues; i++) { 562 struct epair_queue *q = &scb->queues[i]; 563 q->id = i; 564 q->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); 565 q->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); 566 q->ridx = 0; 567 q->sc = scb; 568 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q); 569 } 570 571 /* 572 * Cross-reference the interfaces so we will be able to free both. 573 */ 574 sca->oifp = scb->ifp; 575 scb->oifp = sca->ifp; 576 577 EPAIR_LOCK(); 578 #ifdef SMP 579 /* Get an approximate distribution. */ 580 hash = next_index % mp_ncpus; 581 #else 582 hash = 0; 583 #endif 584 EPAIR_UNLOCK(); 585 586 /* Initialise pseudo media types. */ 587 ifmedia_init(&sca->media, 0, epair_media_change, epair_media_status); 588 ifmedia_add(&sca->media, IFM_ETHER | IFM_10G_T, 0, NULL); 589 ifmedia_set(&sca->media, IFM_ETHER | IFM_10G_T); 590 ifmedia_init(&scb->media, 0, epair_media_change, epair_media_status); 591 ifmedia_add(&scb->media, IFM_ETHER | IFM_10G_T, 0, NULL); 592 ifmedia_set(&scb->media, IFM_ETHER | IFM_10G_T); 593 594 /* Finish initialization of interface <n>a. */ 595 ifp = sca->ifp; 596 ifp->if_softc = sca; 597 strlcpy(ifp->if_xname, name, IFNAMSIZ); 598 ifp->if_dname = epairname; 599 ifp->if_dunit = unit; 600 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 601 ifp->if_flags |= IFF_KNOWSEPOCH; 602 ifp->if_capabilities = IFCAP_VLAN_MTU; 603 ifp->if_capenable = IFCAP_VLAN_MTU; 604 ifp->if_start = epair_start; 605 ifp->if_ioctl = epair_ioctl; 606 ifp->if_init = epair_init; 607 if_setsendqlen(ifp, ifqmaxlen); 608 if_setsendqready(ifp); 609 610 /* 611 * Calculate the etheraddr hashing the hostid and the 612 * interface index. The result would be hopefully unique. 613 * Note that the "a" component of an epair instance may get moved 614 * to a different VNET after creation. In that case its index 615 * will be freed and the index can get reused by new epair instance. 616 * Make sure we do not create same etheraddr again. 617 */ 618 getcredhostid(curthread->td_ucred, (unsigned long *)&hostid); 619 if (hostid == 0) 620 arc4rand(&hostid, sizeof(hostid), 0); 621 622 EPAIR_LOCK(); 623 if (ifp->if_index > next_index) 624 next_index = ifp->if_index; 625 else 626 next_index++; 627 628 key[0] = (uint32_t)next_index; 629 EPAIR_UNLOCK(); 630 key[1] = (uint32_t)(hostid & 0xffffffff); 631 key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff); 632 hash = jenkins_hash32(key, 3, 0); 633 634 eaddr[0] = 0x02; 635 memcpy(&eaddr[1], &hash, 4); 636 eaddr[5] = 0x0a; 637 ether_ifattach(ifp, eaddr); 638 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */ 639 ifp->if_transmit = epair_transmit; 640 641 /* Swap the name and finish initialization of interface <n>b. */ 642 *dp = 'b'; 643 644 ifp = scb->ifp; 645 ifp->if_softc = scb; 646 strlcpy(ifp->if_xname, name, IFNAMSIZ); 647 ifp->if_dname = epairname; 648 ifp->if_dunit = unit; 649 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 650 ifp->if_flags |= IFF_KNOWSEPOCH; 651 ifp->if_capabilities = IFCAP_VLAN_MTU; 652 ifp->if_capenable = IFCAP_VLAN_MTU; 653 ifp->if_start = epair_start; 654 ifp->if_ioctl = epair_ioctl; 655 ifp->if_init = epair_init; 656 if_setsendqlen(ifp, ifqmaxlen); 657 if_setsendqready(ifp); 658 /* We need to play some tricks here for the second interface. */ 659 strlcpy(name, epairname, len); 660 661 /* Correctly set the name for the cloner list. */ 662 strlcpy(name, scb->ifp->if_xname, len); 663 epair_clone_add(ifc, scb); 664 665 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */ 666 ifp->if_transmit = epair_transmit; 667 668 /* 669 * Restore name to <n>a as the ifp for this will go into the 670 * cloner list for the initial call. 671 */ 672 strlcpy(name, sca->ifp->if_xname, len); 673 674 /* Tell the world, that we are ready to rock. */ 675 sca->ifp->if_drv_flags |= IFF_DRV_RUNNING; 676 if_link_state_change(sca->ifp, LINK_STATE_UP); 677 scb->ifp->if_drv_flags |= IFF_DRV_RUNNING; 678 if_link_state_change(scb->ifp, LINK_STATE_UP); 679 680 return (0); 681 } 682 683 static void 684 epair_drain_rings(struct epair_softc *sc) 685 { 686 int ridx; 687 struct mbuf *m; 688 689 for (ridx = 0; ridx < 2; ridx++) { 690 for (int i = 0; i < sc->num_queues; i++) { 691 struct epair_queue *q = &sc->queues[i]; 692 do { 693 m = buf_ring_dequeue_sc(q->rxring[ridx]); 694 if (m == NULL) 695 break; 696 m_freem(m); 697 } while (1); 698 } 699 } 700 } 701 702 static int 703 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp) 704 { 705 struct ifnet *oifp; 706 struct epair_softc *sca, *scb; 707 int unit, error; 708 709 /* 710 * In case we called into if_clone_destroyif() ourselves 711 * again to remove the second interface, the softc will be 712 * NULL. In that case so not do anything but return success. 713 */ 714 if (ifp->if_softc == NULL) 715 return (0); 716 717 unit = ifp->if_dunit; 718 sca = ifp->if_softc; 719 oifp = sca->oifp; 720 scb = oifp->if_softc; 721 722 /* Frist get the interfaces down and detached. */ 723 if_link_state_change(ifp, LINK_STATE_DOWN); 724 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 725 if_link_state_change(oifp, LINK_STATE_DOWN); 726 oifp->if_drv_flags &= ~IFF_DRV_RUNNING; 727 728 ether_ifdetach(ifp); 729 ether_ifdetach(oifp); 730 731 /* Third free any queued packets and all the resources. */ 732 CURVNET_SET_QUIET(oifp->if_vnet); 733 epair_drain_rings(scb); 734 oifp->if_softc = NULL; 735 error = if_clone_destroyif(ifc, oifp); 736 if (error) 737 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d", 738 __func__, error); 739 if_free(oifp); 740 ifmedia_removeall(&scb->media); 741 for (int i = 0; i < scb->num_queues; i++) { 742 struct epair_queue *q = &scb->queues[i]; 743 buf_ring_free(q->rxring[0], M_EPAIR); 744 buf_ring_free(q->rxring[1], M_EPAIR); 745 } 746 free(scb->queues, M_EPAIR); 747 free(scb, M_EPAIR); 748 CURVNET_RESTORE(); 749 750 epair_drain_rings(sca); 751 if_free(ifp); 752 ifmedia_removeall(&sca->media); 753 for (int i = 0; i < sca->num_queues; i++) { 754 struct epair_queue *q = &sca->queues[i]; 755 buf_ring_free(q->rxring[0], M_EPAIR); 756 buf_ring_free(q->rxring[1], M_EPAIR); 757 } 758 free(sca->queues, M_EPAIR); 759 free(sca, M_EPAIR); 760 761 /* Last free the cloner unit. */ 762 ifc_free_unit(ifc, unit); 763 764 return (0); 765 } 766 767 static void 768 vnet_epair_init(const void *unused __unused) 769 { 770 771 V_epair_cloner = if_clone_advanced(epairname, 0, 772 epair_clone_match, epair_clone_create, epair_clone_destroy); 773 } 774 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY, 775 vnet_epair_init, NULL); 776 777 static void 778 vnet_epair_uninit(const void *unused __unused) 779 { 780 781 if_clone_detach(V_epair_cloner); 782 } 783 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, 784 vnet_epair_uninit, NULL); 785 786 static int 787 epair_mod_init(void) 788 { 789 char name[32]; 790 epair_tasks.tasks = 0; 791 792 #ifdef RSS 793 struct pcpu *pcpu; 794 int cpu; 795 796 CPU_FOREACH(cpu) { 797 cpuset_t cpu_mask; 798 799 /* Pin to this CPU so we get appropriate NUMA allocations. */ 800 pcpu = pcpu_find(cpu); 801 thread_lock(curthread); 802 sched_bind(curthread, cpu); 803 thread_unlock(curthread); 804 805 snprintf(name, sizeof(name), "epair_task_%d", cpu); 806 807 epair_tasks.tq[cpu] = taskqueue_create(name, M_WAITOK, 808 taskqueue_thread_enqueue, 809 &epair_tasks.tq[cpu]); 810 CPU_SETOF(cpu, &cpu_mask); 811 taskqueue_start_threads_cpuset(&epair_tasks.tq[cpu], 1, PI_NET, 812 &cpu_mask, "%s", name); 813 814 epair_tasks.tasks++; 815 } 816 #else 817 snprintf(name, sizeof(name), "epair_task"); 818 819 epair_tasks.tq[0] = taskqueue_create(name, M_WAITOK, 820 taskqueue_thread_enqueue, 821 &epair_tasks.tq[0]); 822 taskqueue_start_threads(&epair_tasks.tq[0], 1, PI_NET, "%s", name); 823 824 epair_tasks.tasks = 1; 825 #endif 826 827 return (0); 828 } 829 830 static void 831 epair_mod_cleanup(void) 832 { 833 834 for (int i = 0; i < epair_tasks.tasks; i++) { 835 taskqueue_drain_all(epair_tasks.tq[i]); 836 taskqueue_free(epair_tasks.tq[i]); 837 } 838 } 839 840 static int 841 epair_modevent(module_t mod, int type, void *data) 842 { 843 int ret; 844 845 switch (type) { 846 case MOD_LOAD: 847 EPAIR_LOCK_INIT(); 848 ret = epair_mod_init(); 849 if (ret != 0) 850 return (ret); 851 if (bootverbose) 852 printf("%s: %s initialized.\n", __func__, epairname); 853 break; 854 case MOD_UNLOAD: 855 epair_mod_cleanup(); 856 EPAIR_LOCK_DESTROY(); 857 if (bootverbose) 858 printf("%s: %s unloaded.\n", __func__, epairname); 859 break; 860 default: 861 return (EOPNOTSUPP); 862 } 863 return (0); 864 } 865 866 static moduledata_t epair_mod = { 867 "if_epair", 868 epair_modevent, 869 0 870 }; 871 872 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE); 873 MODULE_VERSION(if_epair, 3); 874