Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/net/if_epair.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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 #include "opt_inet.h" 45 #include "opt_inet6.h" 46 47 #include <sys/param.h> 48 #include <sys/hash.h> 49 #include <sys/jail.h> 50 #include <sys/kernel.h> 51 #include <sys/libkern.h> 52 #include <sys/malloc.h> 53 #include <sys/mbuf.h> 54 #include <sys/module.h> 55 #include <sys/proc.h> 56 #include <sys/queue.h> 57 #include <sys/sched.h> 58 #include <sys/smp.h> 59 #include <sys/socket.h> 60 #include <sys/sockio.h> 61 #include <sys/taskqueue.h> 62 #include <sys/types.h> 63 #include <sys/buf_ring.h> 64 #include <sys/bus.h> 65 #include <sys/interrupt.h> 66 67 #include <net/bpf.h> 68 #include <net/ethernet.h> 69 #include <net/if.h> 70 #include <net/if_var.h> 71 #include <net/if_clone.h> 72 #include <net/if_media.h> 73 #include <net/if_var.h> 74 #include <net/if_private.h> 75 #include <net/if_types.h> 76 #include <net/netisr.h> 77 #ifdef RSS 78 #include <net/rss_config.h> 79 #ifdef INET 80 #include <netinet/in_rss.h> 81 #endif 82 #ifdef INET6 83 #include <netinet6/in6_rss.h> 84 #endif 85 #endif 86 #include <net/vnet.h> 87 88 static const char epairname[] = "epair"; 89 #define RXRSIZE 4096 /* Probably overkill by 4-8x. */ 90 91 static MALLOC_DEFINE(M_EPAIR, epairname, 92 "Pair of virtual cross-over connected Ethernet-like interfaces"); 93 94 VNET_DEFINE_STATIC(struct if_clone *, epair_cloner); 95 #define V_epair_cloner VNET(epair_cloner) 96 97 static unsigned int next_index = 0; 98 #define EPAIR_LOCK_INIT() mtx_init(&epair_n_index_mtx, "epairidx", \ 99 NULL, MTX_DEF) 100 #define EPAIR_LOCK_DESTROY() mtx_destroy(&epair_n_index_mtx) 101 #define EPAIR_LOCK() mtx_lock(&epair_n_index_mtx) 102 #define EPAIR_UNLOCK() mtx_unlock(&epair_n_index_mtx) 103 104 #define BIT_QUEUE_TASK 0 105 #define BIT_MBUF_QUEUED 1 106 107 struct epair_softc; 108 struct epair_queue { 109 int id; 110 struct buf_ring *rxring[2]; 111 volatile int ridx; /* 0 || 1 */ 112 volatile long state; /* taskqueue coordination */ 113 struct task tx_task; 114 struct epair_softc *sc; 115 }; 116 117 static struct mtx epair_n_index_mtx; 118 struct epair_softc { 119 struct ifnet *ifp; /* This ifp. */ 120 struct ifnet *oifp; /* other ifp of pair. */ 121 int num_queues; 122 struct epair_queue *queues; 123 struct ifmedia media; /* Media config (fake). */ 124 STAILQ_ENTRY(epair_softc) entry; 125 }; 126 127 struct epair_tasks_t { 128 int tasks; 129 struct taskqueue *tq[MAXCPU]; 130 }; 131 132 static struct epair_tasks_t epair_tasks; 133 134 static void 135 epair_clear_mbuf(struct mbuf *m) 136 { 137 /* Remove any CSUM_SND_TAG as ether_input will barf. */ 138 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) { 139 m_snd_tag_rele(m->m_pkthdr.snd_tag); 140 m->m_pkthdr.snd_tag = NULL; 141 m->m_pkthdr.csum_flags &= ~CSUM_SND_TAG; 142 } 143 144 m_tag_delete_nonpersistent(m); 145 } 146 147 static void 148 epair_if_input(struct epair_softc *sc, struct epair_queue *q, int ridx) 149 { 150 struct ifnet *ifp; 151 struct mbuf *m; 152 153 ifp = sc->ifp; 154 CURVNET_SET(ifp->if_vnet); 155 while (! buf_ring_empty(q->rxring[ridx])) { 156 m = buf_ring_dequeue_mc(q->rxring[ridx]); 157 if (m == NULL) 158 continue; 159 160 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); 161 (*ifp->if_input)(ifp, m); 162 163 } 164 CURVNET_RESTORE(); 165 } 166 167 static void 168 epair_tx_start_deferred(void *arg, int pending) 169 { 170 struct epair_queue *q = (struct epair_queue *)arg; 171 struct epair_softc *sc = q->sc; 172 int ridx, nidx; 173 174 if_ref(sc->ifp); 175 ridx = atomic_load_int(&q->ridx); 176 do { 177 nidx = (ridx == 0) ? 1 : 0; 178 } while (!atomic_fcmpset_int(&q->ridx, &ridx, nidx)); 179 epair_if_input(sc, q, ridx); 180 181 atomic_clear_long(&q->state, (1 << BIT_QUEUE_TASK)); 182 if (atomic_testandclear_long(&q->state, BIT_MBUF_QUEUED)) 183 taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task); 184 185 if_rele(sc->ifp); 186 } 187 188 static struct epair_queue * 189 epair_select_queue(struct epair_softc *sc, struct mbuf *m) 190 { 191 uint32_t bucket; 192 #ifdef RSS 193 struct ether_header *eh; 194 int ret; 195 196 ret = rss_m2bucket(m, &bucket); 197 if (ret) { 198 /* Actually hash the packet. */ 199 eh = mtod(m, struct ether_header *); 200 201 switch (ntohs(eh->ether_type)) { 202 #ifdef INET 203 case ETHERTYPE_IP: 204 rss_soft_m2cpuid_v4(m, 0, &bucket); 205 break; 206 #endif 207 #ifdef INET6 208 case ETHERTYPE_IPV6: 209 rss_soft_m2cpuid_v6(m, 0, &bucket); 210 break; 211 #endif 212 default: 213 bucket = 0; 214 break; 215 } 216 } 217 bucket %= sc->num_queues; 218 #else 219 bucket = 0; 220 #endif 221 return (&sc->queues[bucket]); 222 } 223 224 static void 225 epair_prepare_mbuf(struct mbuf *m, struct ifnet *src_ifp) 226 { 227 M_ASSERTPKTHDR(m); 228 epair_clear_mbuf(m); 229 if_setrcvif(m, src_ifp); 230 M_SETFIB(m, src_ifp->if_fib); 231 232 MPASS(m->m_nextpkt == NULL); 233 MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0); 234 } 235 236 static void 237 epair_menq(struct mbuf *m, struct epair_softc *osc) 238 { 239 struct ifnet *ifp, *oifp; 240 int len, ret; 241 int ridx; 242 short mflags; 243 244 /* 245 * I know this looks weird. We pass the "other sc" as we need that one 246 * and can get both ifps from it as well. 247 */ 248 oifp = osc->ifp; 249 ifp = osc->oifp; 250 251 epair_prepare_mbuf(m, oifp); 252 253 /* Save values as once the mbuf is queued, it's not ours anymore. */ 254 len = m->m_pkthdr.len; 255 mflags = m->m_flags; 256 257 struct epair_queue *q = epair_select_queue(osc, m); 258 259 atomic_set_long(&q->state, (1 << BIT_MBUF_QUEUED)); 260 ridx = atomic_load_int(&q->ridx); 261 ret = buf_ring_enqueue(q->rxring[ridx], m); 262 if (ret != 0) { 263 /* Ring is full. */ 264 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); 265 m_freem(m); 266 return; 267 } 268 269 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 270 /* 271 * IFQ_HANDOFF_ADJ/ip_handoff() update statistics, 272 * but as we bypass all this we have to duplicate 273 * the logic another time. 274 */ 275 if_inc_counter(ifp, IFCOUNTER_OBYTES, len); 276 if (mflags & (M_BCAST|M_MCAST)) 277 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); 278 /* Someone else received the packet. */ 279 if_inc_counter(oifp, IFCOUNTER_IPACKETS, 1); 280 281 if (!atomic_testandset_long(&q->state, BIT_QUEUE_TASK)) 282 taskqueue_enqueue(epair_tasks.tq[q->id], &q->tx_task); 283 } 284 285 static void 286 epair_start(struct ifnet *ifp) 287 { 288 struct mbuf *m; 289 struct epair_softc *sc; 290 struct ifnet *oifp; 291 292 /* 293 * We get packets here from ether_output via if_handoff() 294 * and need to put them into the input queue of the oifp 295 * and will put the packet into the receive-queue (rxq) of the 296 * other interface (oifp) of our pair. 297 */ 298 sc = ifp->if_softc; 299 oifp = sc->oifp; 300 sc = oifp->if_softc; 301 for (;;) { 302 IFQ_DEQUEUE(&ifp->if_snd, m); 303 if (m == NULL) 304 break; 305 M_ASSERTPKTHDR(m); 306 BPF_MTAP(ifp, m); 307 308 /* In case either interface is not usable drop the packet. */ 309 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 310 (ifp->if_flags & IFF_UP) == 0 || 311 (oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 312 (oifp->if_flags & IFF_UP) == 0) { 313 m_freem(m); 314 continue; 315 } 316 317 epair_menq(m, sc); 318 } 319 } 320 321 static int 322 epair_transmit(struct ifnet *ifp, struct mbuf *m) 323 { 324 struct epair_softc *sc; 325 struct ifnet *oifp; 326 #ifdef ALTQ 327 int len; 328 short mflags; 329 #endif 330 331 if (m == NULL) 332 return (0); 333 M_ASSERTPKTHDR(m); 334 335 /* 336 * We are not going to use the interface en/dequeue mechanism 337 * on the TX side. We are called from ether_output_frame() 338 * and will put the packet into the receive-queue (rxq) of the 339 * other interface (oifp) of our pair. 340 */ 341 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 342 m_freem(m); 343 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 344 return (ENXIO); 345 } 346 if ((ifp->if_flags & IFF_UP) == 0) { 347 m_freem(m); 348 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 349 return (ENETDOWN); 350 } 351 352 BPF_MTAP(ifp, m); 353 354 /* 355 * In case the outgoing interface is not usable, 356 * drop the packet. 357 */ 358 sc = ifp->if_softc; 359 oifp = sc->oifp; 360 if ((oifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 361 (oifp->if_flags & IFF_UP) == 0) { 362 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 363 m_freem(m); 364 return (0); 365 } 366 367 #ifdef ALTQ 368 len = m->m_pkthdr.len; 369 mflags = m->m_flags; 370 int error = 0; 371 372 /* Support ALTQ via the classic if_start() path. */ 373 IF_LOCK(&ifp->if_snd); 374 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 375 ALTQ_ENQUEUE(&ifp->if_snd, m, NULL, error); 376 if (error) 377 if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1); 378 IF_UNLOCK(&ifp->if_snd); 379 if (!error) { 380 if_inc_counter(ifp, IFCOUNTER_OBYTES, len); 381 if (mflags & (M_BCAST|M_MCAST)) 382 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1); 383 epair_start(ifp); 384 } 385 return (error); 386 } 387 IF_UNLOCK(&ifp->if_snd); 388 #endif 389 390 epair_menq(m, oifp->if_softc); 391 return (0); 392 } 393 394 static void 395 epair_qflush(struct ifnet *ifp __unused) 396 { 397 } 398 399 static int 400 epair_media_change(struct ifnet *ifp __unused) 401 { 402 403 /* Do nothing. */ 404 return (0); 405 } 406 407 static void 408 epair_media_status(struct ifnet *ifp __unused, struct ifmediareq *imr) 409 { 410 411 imr->ifm_status = IFM_AVALID | IFM_ACTIVE; 412 imr->ifm_active = IFM_ETHER | IFM_10G_T | IFM_FDX; 413 } 414 415 static int 416 epair_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 417 { 418 struct epair_softc *sc; 419 struct ifreq *ifr; 420 int error; 421 422 ifr = (struct ifreq *)data; 423 switch (cmd) { 424 case SIOCSIFFLAGS: 425 case SIOCADDMULTI: 426 case SIOCDELMULTI: 427 error = 0; 428 break; 429 430 case SIOCSIFMEDIA: 431 case SIOCGIFMEDIA: 432 sc = ifp->if_softc; 433 error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd); 434 break; 435 436 case SIOCSIFMTU: 437 /* We basically allow all kinds of MTUs. */ 438 ifp->if_mtu = ifr->ifr_mtu; 439 error = 0; 440 break; 441 442 default: 443 /* Let the common ethernet handler process this. */ 444 error = ether_ioctl(ifp, cmd, data); 445 break; 446 } 447 448 return (error); 449 } 450 451 static void 452 epair_init(void *dummy __unused) 453 { 454 } 455 456 /* 457 * Interface cloning functions. 458 * We use our private ones so that we can create/destroy our secondary 459 * device along with the primary one. 460 */ 461 static int 462 epair_clone_match(struct if_clone *ifc, const char *name) 463 { 464 const char *cp; 465 466 /* 467 * Our base name is epair. 468 * Our interfaces will be named epair<n>[ab]. 469 * So accept anything of the following list: 470 * - epair 471 * - epair<n> 472 * but not the epair<n>[ab] versions. 473 */ 474 if (strncmp(epairname, name, sizeof(epairname)-1) != 0) 475 return (0); 476 477 for (cp = name + sizeof(epairname) - 1; *cp != '\0'; cp++) { 478 if (*cp < '' || *cp > '9') 479 return (0); 480 } 481 482 return (1); 483 } 484 485 static void 486 epair_clone_add(struct if_clone *ifc, struct epair_softc *scb) 487 { 488 struct ifnet *ifp; 489 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ 490 491 ifp = scb->ifp; 492 /* Copy epairNa etheraddr and change the last byte. */ 493 memcpy(eaddr, scb->oifp->if_hw_addr, ETHER_ADDR_LEN); 494 eaddr[5] = 0x0b; 495 ether_ifattach(ifp, eaddr); 496 497 if_clone_addif(ifc, ifp); 498 } 499 500 static struct epair_softc * 501 epair_alloc_sc(struct if_clone *ifc) 502 { 503 struct epair_softc *sc; 504 505 struct ifnet *ifp = if_alloc(IFT_ETHER); 506 if (ifp == NULL) 507 return (NULL); 508 509 sc = malloc(sizeof(struct epair_softc), M_EPAIR, M_WAITOK | M_ZERO); 510 sc->ifp = ifp; 511 sc->num_queues = epair_tasks.tasks; 512 sc->queues = mallocarray(sc->num_queues, sizeof(struct epair_queue), 513 M_EPAIR, M_WAITOK); 514 for (int i = 0; i < sc->num_queues; i++) { 515 struct epair_queue *q = &sc->queues[i]; 516 q->id = i; 517 q->rxring[0] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); 518 q->rxring[1] = buf_ring_alloc(RXRSIZE, M_EPAIR, M_WAITOK, NULL); 519 q->ridx = 0; 520 q->state = 0; 521 q->sc = sc; 522 NET_TASK_INIT(&q->tx_task, 0, epair_tx_start_deferred, q); 523 } 524 525 /* Initialise pseudo media types. */ 526 ifmedia_init(&sc->media, 0, epair_media_change, epair_media_status); 527 ifmedia_add(&sc->media, IFM_ETHER | IFM_10G_T, 0, NULL); 528 ifmedia_set(&sc->media, IFM_ETHER | IFM_10G_T); 529 530 return (sc); 531 } 532 533 static void 534 epair_setup_ifp(struct epair_softc *sc, char *name, int unit) 535 { 536 struct ifnet *ifp = sc->ifp; 537 538 ifp->if_softc = sc; 539 strlcpy(ifp->if_xname, name, IFNAMSIZ); 540 ifp->if_dname = epairname; 541 ifp->if_dunit = unit; 542 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 543 ifp->if_flags |= IFF_KNOWSEPOCH; 544 ifp->if_capabilities = IFCAP_VLAN_MTU; 545 ifp->if_capenable = IFCAP_VLAN_MTU; 546 ifp->if_transmit = epair_transmit; 547 ifp->if_qflush = epair_qflush; 548 ifp->if_start = epair_start; 549 ifp->if_ioctl = epair_ioctl; 550 ifp->if_init = epair_init; 551 if_setsendqlen(ifp, ifqmaxlen); 552 if_setsendqready(ifp); 553 554 ifp->if_baudrate = IF_Gbps(10); /* arbitrary maximum */ 555 } 556 557 static void 558 epair_generate_mac(struct epair_softc *sc, uint8_t *eaddr) 559 { 560 uint32_t key[3]; 561 uint32_t hash; 562 uint64_t hostid; 563 564 EPAIR_LOCK(); 565 #ifdef SMP 566 /* Get an approximate distribution. */ 567 hash = next_index % mp_ncpus; 568 #else 569 hash = 0; 570 #endif 571 EPAIR_UNLOCK(); 572 573 /* 574 * Calculate the etheraddr hashing the hostid and the 575 * interface index. The result would be hopefully unique. 576 * Note that the "a" component of an epair instance may get moved 577 * to a different VNET after creation. In that case its index 578 * will be freed and the index can get reused by new epair instance. 579 * Make sure we do not create same etheraddr again. 580 */ 581 getcredhostid(curthread->td_ucred, (unsigned long *)&hostid); 582 if (hostid == 0) 583 arc4rand(&hostid, sizeof(hostid), 0); 584 585 struct ifnet *ifp = sc->ifp; 586 EPAIR_LOCK(); 587 if (ifp->if_index > next_index) 588 next_index = ifp->if_index; 589 else 590 next_index++; 591 592 key[0] = (uint32_t)next_index; 593 EPAIR_UNLOCK(); 594 key[1] = (uint32_t)(hostid & 0xffffffff); 595 key[2] = (uint32_t)((hostid >> 32) & 0xfffffffff); 596 hash = jenkins_hash32(key, 3, 0); 597 598 eaddr[0] = 0x02; 599 memcpy(&eaddr[1], &hash, 4); 600 eaddr[5] = 0x0a; 601 } 602 603 static void 604 epair_free_sc(struct epair_softc *sc) 605 { 606 if (sc == NULL) 607 return; 608 609 if_free(sc->ifp); 610 ifmedia_removeall(&sc->media); 611 for (int i = 0; i < sc->num_queues; i++) { 612 struct epair_queue *q = &sc->queues[i]; 613 buf_ring_free(q->rxring[0], M_EPAIR); 614 buf_ring_free(q->rxring[1], M_EPAIR); 615 } 616 free(sc->queues, M_EPAIR); 617 free(sc, M_EPAIR); 618 } 619 620 static void 621 epair_set_state(struct ifnet *ifp, bool running) 622 { 623 if (running) { 624 ifp->if_drv_flags |= IFF_DRV_RUNNING; 625 if_link_state_change(ifp, LINK_STATE_UP); 626 } else { 627 if_link_state_change(ifp, LINK_STATE_DOWN); 628 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 629 } 630 } 631 632 static int 633 epair_handle_unit(struct if_clone *ifc, char *name, size_t len, int *punit) 634 { 635 int error = 0, unit, wildcard; 636 char *dp; 637 638 /* Try to see if a special unit was requested. */ 639 error = ifc_name2unit(name, &unit); 640 if (error != 0) 641 return (error); 642 wildcard = (unit < 0); 643 644 error = ifc_alloc_unit(ifc, &unit); 645 if (error != 0) 646 return (error); 647 648 /* 649 * If no unit had been given, we need to adjust the ifName. 650 * Also make sure there is space for our extra [ab] suffix. 651 */ 652 for (dp = name; *dp != '\0'; dp++); 653 if (wildcard) { 654 int slen = snprintf(dp, len - (dp - name), "%d", unit); 655 if (slen > len - (dp - name) - 1) { 656 /* ifName too long. */ 657 error = ENOSPC; 658 goto done; 659 } 660 dp += slen; 661 } 662 if (len - (dp - name) - 1 < 1) { 663 /* No space left for our [ab] suffix. */ 664 error = ENOSPC; 665 goto done; 666 } 667 *dp = 'b'; 668 /* Must not change dp so we can replace 'a' by 'b' later. */ 669 *(dp+1) = '\0'; 670 671 /* Check if 'a' and 'b' interfaces already exist. */ 672 if (ifunit(name) != NULL) { 673 error = EEXIST; 674 goto done; 675 } 676 677 *dp = 'a'; 678 if (ifunit(name) != NULL) { 679 error = EEXIST; 680 goto done; 681 } 682 *punit = unit; 683 done: 684 if (error != 0) 685 ifc_free_unit(ifc, unit); 686 687 return (error); 688 } 689 690 static int 691 epair_clone_create(struct if_clone *ifc, char *name, size_t len, 692 struct ifc_data *ifd, struct ifnet **ifpp) 693 { 694 struct epair_softc *sca, *scb; 695 struct ifnet *ifp; 696 char *dp; 697 int error, unit; 698 uint8_t eaddr[ETHER_ADDR_LEN]; /* 00:00:00:00:00:00 */ 699 700 error = epair_handle_unit(ifc, name, len, &unit); 701 if (error != 0) 702 return (error); 703 704 /* Allocate memory for both [ab] interfaces */ 705 sca = epair_alloc_sc(ifc); 706 scb = epair_alloc_sc(ifc); 707 if (sca == NULL || scb == NULL) { 708 epair_free_sc(sca); 709 epair_free_sc(scb); 710 ifc_free_unit(ifc, unit); 711 return (ENOSPC); 712 } 713 714 /* 715 * Cross-reference the interfaces so we will be able to free both. 716 */ 717 sca->oifp = scb->ifp; 718 scb->oifp = sca->ifp; 719 720 /* Finish initialization of interface <n>a. */ 721 ifp = sca->ifp; 722 epair_setup_ifp(sca, name, unit); 723 epair_generate_mac(sca, eaddr); 724 725 ether_ifattach(ifp, eaddr); 726 727 /* Swap the name and finish initialization of interface <n>b. */ 728 dp = name + strlen(name) - 1; 729 *dp = 'b'; 730 731 epair_setup_ifp(scb, name, unit); 732 733 ifp = scb->ifp; 734 /* We need to play some tricks here for the second interface. */ 735 strlcpy(name, epairname, len); 736 /* Correctly set the name for the cloner list. */ 737 strlcpy(name, scb->ifp->if_xname, len); 738 739 epair_clone_add(ifc, scb); 740 741 /* 742 * Restore name to <n>a as the ifp for this will go into the 743 * cloner list for the initial call. 744 */ 745 strlcpy(name, sca->ifp->if_xname, len); 746 747 /* Tell the world, that we are ready to rock. */ 748 epair_set_state(sca->ifp, true); 749 epair_set_state(scb->ifp, true); 750 751 *ifpp = sca->ifp; 752 753 return (0); 754 } 755 756 static void 757 epair_drain_rings(struct epair_softc *sc) 758 { 759 int ridx; 760 struct mbuf *m; 761 762 for (ridx = 0; ridx < 2; ridx++) { 763 for (int i = 0; i < sc->num_queues; i++) { 764 struct epair_queue *q = &sc->queues[i]; 765 do { 766 m = buf_ring_dequeue_sc(q->rxring[ridx]); 767 if (m == NULL) 768 break; 769 m_freem(m); 770 } while (1); 771 } 772 } 773 } 774 775 static int 776 epair_clone_destroy(struct if_clone *ifc, struct ifnet *ifp, uint32_t flags) 777 { 778 struct ifnet *oifp; 779 struct epair_softc *sca, *scb; 780 int unit, error; 781 782 /* 783 * In case we called into if_clone_destroyif() ourselves 784 * again to remove the second interface, the softc will be 785 * NULL. In that case so not do anything but return success. 786 */ 787 if (ifp->if_softc == NULL) 788 return (0); 789 790 unit = ifp->if_dunit; 791 sca = ifp->if_softc; 792 oifp = sca->oifp; 793 scb = oifp->if_softc; 794 795 /* Frist get the interfaces down and detached. */ 796 epair_set_state(ifp, false); 797 epair_set_state(oifp, false); 798 799 ether_ifdetach(ifp); 800 ether_ifdetach(oifp); 801 802 /* Third free any queued packets and all the resources. */ 803 CURVNET_SET_QUIET(oifp->if_vnet); 804 epair_drain_rings(scb); 805 oifp->if_softc = NULL; 806 error = if_clone_destroyif(ifc, oifp); 807 if (error) 808 panic("%s: if_clone_destroyif() for our 2nd iface failed: %d", 809 __func__, error); 810 epair_free_sc(scb); 811 CURVNET_RESTORE(); 812 813 epair_drain_rings(sca); 814 epair_free_sc(sca); 815 816 /* Last free the cloner unit. */ 817 ifc_free_unit(ifc, unit); 818 819 return (0); 820 } 821 822 static void 823 vnet_epair_init(const void *unused __unused) 824 { 825 struct if_clone_addreq req = { 826 .match_f = epair_clone_match, 827 .create_f = epair_clone_create, 828 .destroy_f = epair_clone_destroy, 829 }; 830 V_epair_cloner = ifc_attach_cloner(epairname, &req); 831 } 832 VNET_SYSINIT(vnet_epair_init, SI_SUB_PSEUDO, SI_ORDER_ANY, 833 vnet_epair_init, NULL); 834 835 static void 836 vnet_epair_uninit(const void *unused __unused) 837 { 838 839 ifc_detach_cloner(V_epair_cloner); 840 } 841 VNET_SYSUNINIT(vnet_epair_uninit, SI_SUB_INIT_IF, SI_ORDER_ANY, 842 vnet_epair_uninit, NULL); 843 844 static int 845 epair_mod_init(void) 846 { 847 char name[32]; 848 epair_tasks.tasks = 0; 849 850 #ifdef RSS 851 int cpu; 852 853 CPU_FOREACH(cpu) { 854 cpuset_t cpu_mask; 855 856 /* Pin to this CPU so we get appropriate NUMA allocations. */ 857 thread_lock(curthread); 858 sched_bind(curthread, cpu); 859 thread_unlock(curthread); 860 861 snprintf(name, sizeof(name), "epair_task_%d", cpu); 862 863 epair_tasks.tq[cpu] = taskqueue_create(name, M_WAITOK, 864 taskqueue_thread_enqueue, 865 &epair_tasks.tq[cpu]); 866 CPU_SETOF(cpu, &cpu_mask); 867 taskqueue_start_threads_cpuset(&epair_tasks.tq[cpu], 1, PI_NET, 868 &cpu_mask, "%s", name); 869 870 epair_tasks.tasks++; 871 } 872 thread_lock(curthread); 873 sched_unbind(curthread); 874 thread_unlock(curthread); 875 #else 876 snprintf(name, sizeof(name), "epair_task"); 877 878 epair_tasks.tq[0] = taskqueue_create(name, M_WAITOK, 879 taskqueue_thread_enqueue, 880 &epair_tasks.tq[0]); 881 taskqueue_start_threads(&epair_tasks.tq[0], 1, PI_NET, "%s", name); 882 883 epair_tasks.tasks = 1; 884 #endif 885 886 return (0); 887 } 888 889 static void 890 epair_mod_cleanup(void) 891 { 892 893 for (int i = 0; i < epair_tasks.tasks; i++) { 894 taskqueue_drain_all(epair_tasks.tq[i]); 895 taskqueue_free(epair_tasks.tq[i]); 896 } 897 } 898 899 static int 900 epair_modevent(module_t mod, int type, void *data) 901 { 902 int ret; 903 904 switch (type) { 905 case MOD_LOAD: 906 EPAIR_LOCK_INIT(); 907 ret = epair_mod_init(); 908 if (ret != 0) 909 return (ret); 910 if (bootverbose) 911 printf("%s: %s initialized.\n", __func__, epairname); 912 break; 913 case MOD_UNLOAD: 914 epair_mod_cleanup(); 915 EPAIR_LOCK_DESTROY(); 916 if (bootverbose) 917 printf("%s: %s unloaded.\n", __func__, epairname); 918 break; 919 default: 920 return (EOPNOTSUPP); 921 } 922 return (0); 923 } 924 925 static moduledata_t epair_mod = { 926 "if_epair", 927 epair_modevent, 928 0 929 }; 930 931 DECLARE_MODULE(if_epair, epair_mod, SI_SUB_PSEUDO, SI_ORDER_MIDDLE); 932 MODULE_VERSION(if_epair, 3);
Cache object: 0c174572daf2bd446fecca6dde49beb4
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]