1 /*- 2 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice unmodified, this list of conditions, and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27 /* Driver for VirtIO network devices. */ 28 29 #include "opt_ifpoll.h" 30 31 #include <sys/cdefs.h> 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/kernel.h> 36 #include <sys/sockio.h> 37 #include <sys/mbuf.h> 38 #include <sys/malloc.h> 39 #include <sys/module.h> 40 #include <sys/socket.h> 41 #include <sys/sysctl.h> 42 #include <sys/taskqueue.h> 43 #include <sys/random.h> 44 #include <sys/sglist.h> 45 #include <sys/serialize.h> 46 #include <sys/bus.h> 47 #include <sys/rman.h> 48 49 #include <machine/limits.h> 50 51 #include <net/ethernet.h> 52 #include <net/if.h> 53 #include <net/if_arp.h> 54 #include <net/if_dl.h> 55 #include <net/if_types.h> 56 #include <net/if_media.h> 57 #include <net/vlan/if_vlan_var.h> 58 #include <net/vlan/if_vlan_ether.h> 59 #include <net/if_poll.h> 60 #include <net/ifq_var.h> 61 62 #include <net/bpf.h> 63 64 #include <netinet/in_systm.h> 65 #include <netinet/in.h> 66 #include <netinet/ip.h> 67 #include <netinet/ip6.h> 68 #include <netinet/udp.h> 69 #include <netinet/tcp.h> 70 71 #include <dev/virtual/virtio/virtio/virtio.h> 72 #include <dev/virtual/virtio/virtio/virtqueue.h> 73 #include <dev/virtual/virtio/net/virtio_net.h> 74 #include <dev/virtual/virtio/net/if_vtnetvar.h> 75 76 MALLOC_DEFINE(M_VTNET, "VTNET_TX", "Outgoing VTNET TX frame header"); 77 78 static int vtnet_probe(device_t); 79 static int vtnet_attach(device_t); 80 static int vtnet_detach(device_t); 81 static int vtnet_suspend(device_t); 82 static int vtnet_resume(device_t); 83 static int vtnet_shutdown(device_t); 84 85 static void vtnet_negotiate_features(struct vtnet_softc *); 86 #ifdef IFPOLL_ENABLE 87 static void vtnet_npoll(struct ifnet *, struct ifpoll_info *); 88 static void vtnet_npoll_status(struct ifnet *); 89 static void vtnet_npoll_rx(struct ifnet *, void *, int); 90 static void vtnet_npoll_tx(struct ifnet *, void *, int); 91 #endif 92 static void vtnet_serialize(struct ifnet *, enum ifnet_serialize); 93 static void vtnet_deserialize(struct ifnet *, enum ifnet_serialize); 94 static int vtnet_tryserialize(struct ifnet *, enum ifnet_serialize); 95 #ifdef INVARIANTS 96 static void vtnet_serialize_assert(struct ifnet *, enum ifnet_serialize, 97 boolean_t); 98 #endif /* INVARIANTS */ 99 static int vtnet_alloc_intrs(struct vtnet_softc *); 100 static int vtnet_alloc_virtqueues(struct vtnet_softc *); 101 static int vtnet_bind_intrs(struct vtnet_softc *); 102 static void vtnet_get_hwaddr(struct vtnet_softc *); 103 static void vtnet_set_hwaddr(struct vtnet_softc *); 104 static int vtnet_is_link_up(struct vtnet_softc *); 105 static void vtnet_update_link_status(struct vtnet_softc *); 106 static void vtnet_watchdog(struct ifaltq_subque *); 107 static int vtnet_setup_interface(struct vtnet_softc *); 108 static int vtnet_change_mtu(struct vtnet_softc *, int); 109 static int vtnet_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *); 110 111 static int vtnet_init_rx_vq(struct vtnet_softc *); 112 static void vtnet_free_rx_mbufs(struct vtnet_softc *); 113 static void vtnet_free_tx_mbufs(struct vtnet_softc *); 114 static void vtnet_free_ctrl_vq(struct vtnet_softc *); 115 116 static struct mbuf * vtnet_alloc_rxbuf(struct vtnet_softc *, int, 117 struct mbuf **); 118 static int vtnet_replace_rxbuf(struct vtnet_softc *, 119 struct mbuf *, int); 120 static int vtnet_newbuf(struct vtnet_softc *); 121 static void vtnet_discard_merged_rxbuf(struct vtnet_softc *, int); 122 static void vtnet_discard_rxbuf(struct vtnet_softc *, struct mbuf *); 123 static int vtnet_enqueue_rxbuf(struct vtnet_softc *, struct mbuf *); 124 static void vtnet_vlan_tag_remove(struct mbuf *); 125 static int vtnet_rx_csum(struct vtnet_softc *, struct mbuf *, 126 struct virtio_net_hdr *); 127 static int vtnet_rxeof_merged(struct vtnet_softc *, struct mbuf *, int); 128 static int vtnet_rxeof(struct vtnet_softc *, int, int *); 129 static void vtnet_rx_msix_intr(void *); 130 static void vtnet_rx_vq_intr(void *); 131 132 static void vtnet_enqueue_txhdr(struct vtnet_softc *, 133 struct vtnet_tx_header *); 134 static void vtnet_txeof(struct vtnet_softc *); 135 static struct mbuf * vtnet_tx_offload(struct vtnet_softc *, struct mbuf *, 136 struct virtio_net_hdr *); 137 static int vtnet_enqueue_txbuf(struct vtnet_softc *, struct mbuf **, 138 struct vtnet_tx_header *); 139 static int vtnet_encap(struct vtnet_softc *, struct mbuf **); 140 static void vtnet_start(struct ifnet *, struct ifaltq_subque *); 141 142 static void vtnet_config_intr(void *); 143 static void vtnet_tx_msix_intr(void *); 144 static void vtnet_tx_vq_intr(void *); 145 146 static void vtnet_stop(struct vtnet_softc *); 147 static int vtnet_virtio_reinit(struct vtnet_softc *); 148 static void vtnet_init(void *); 149 150 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *, 151 struct sglist *, int, int); 152 153 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *); 154 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int); 155 static int vtnet_set_promisc(struct vtnet_softc *, int); 156 static int vtnet_set_allmulti(struct vtnet_softc *, int); 157 static void vtnet_rx_filter(struct vtnet_softc *sc); 158 static void vtnet_rx_filter_mac(struct vtnet_softc *); 159 160 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t); 161 static void vtnet_rx_filter_vlan(struct vtnet_softc *); 162 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t); 163 static void vtnet_register_vlan(void *, struct ifnet *, uint16_t); 164 static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t); 165 166 static int vtnet_ifmedia_upd(struct ifnet *); 167 static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *); 168 169 static void vtnet_add_statistics(struct vtnet_softc *); 170 171 static int vtnet_enable_rx_intr(struct vtnet_softc *); 172 static int vtnet_enable_tx_intr(struct vtnet_softc *); 173 static void vtnet_disable_rx_intr(struct vtnet_softc *); 174 static void vtnet_disable_tx_intr(struct vtnet_softc *); 175 176 /* Tunables. */ 177 static int vtnet_csum_disable = 0; 178 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable); 179 static int vtnet_tso_disable = 1; 180 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable); 181 static int vtnet_lro_disable = 0; 182 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable); 183 184 /* 185 * Reducing the number of transmit completed interrupts can 186 * improve performance. To do so, the define below keeps the 187 * Tx vq interrupt disabled and adds calls to vtnet_txeof() 188 * in the start path. The price to pay for this is the m_free'ing 189 * of transmitted mbufs may be delayed. 190 */ 191 #define VTNET_TX_INTR_MODERATION 192 193 static struct virtio_feature_desc vtnet_feature_desc[] = { 194 { VIRTIO_NET_F_CSUM, "TxChecksum" }, 195 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" }, 196 { VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, "DynOffload" }, 197 { VIRTIO_NET_F_MAC, "MacAddress" }, 198 { VIRTIO_NET_F_GSO, "TxAllGSO" }, 199 { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" }, 200 { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" }, 201 { VIRTIO_NET_F_GUEST_ECN, "RxECN" }, 202 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" }, 203 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" }, 204 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" }, 205 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" }, 206 { VIRTIO_NET_F_HOST_UFO, "TxUFO" }, 207 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" }, 208 { VIRTIO_NET_F_STATUS, "Status" }, 209 { VIRTIO_NET_F_CTRL_VQ, "ControlVq" }, 210 { VIRTIO_NET_F_CTRL_RX, "RxMode" }, 211 { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" }, 212 { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" }, 213 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" }, 214 { VIRTIO_NET_F_MQ, "Multiqueue" }, 215 { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" }, 216 { 0, NULL } 217 }; 218 219 static device_method_t vtnet_methods[] = { 220 /* Device methods. */ 221 DEVMETHOD(device_probe, vtnet_probe), 222 DEVMETHOD(device_attach, vtnet_attach), 223 DEVMETHOD(device_detach, vtnet_detach), 224 DEVMETHOD(device_suspend, vtnet_suspend), 225 DEVMETHOD(device_resume, vtnet_resume), 226 DEVMETHOD(device_shutdown, vtnet_shutdown), 227 228 DEVMETHOD_END 229 }; 230 231 static driver_t vtnet_driver = { 232 "vtnet", 233 vtnet_methods, 234 sizeof(struct vtnet_softc) 235 }; 236 237 static devclass_t vtnet_devclass; 238 239 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass, NULL, NULL); 240 MODULE_VERSION(vtnet, 1); 241 MODULE_DEPEND(vtnet, virtio, 1, 1, 1); 242 243 static int 244 vtnet_probe(device_t dev) 245 { 246 if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK) 247 return (ENXIO); 248 249 device_set_desc(dev, "VirtIO Networking Adapter"); 250 251 return (BUS_PROBE_DEFAULT); 252 } 253 254 static int 255 vtnet_attach(device_t dev) 256 { 257 struct vtnet_softc *sc; 258 int i, error; 259 260 sc = device_get_softc(dev); 261 sc->vtnet_dev = dev; 262 263 lwkt_serialize_init(&sc->vtnet_slz); 264 lwkt_serialize_init(&sc->vtnet_rx_slz); 265 lwkt_serialize_init(&sc->vtnet_tx_slz); 266 sc->serializes[0] = &sc->vtnet_slz; 267 sc->serializes[1] = &sc->vtnet_rx_slz; 268 sc->serializes[2] = &sc->vtnet_tx_slz; 269 270 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd, 271 vtnet_ifmedia_sts); 272 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL); 273 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE); 274 275 vtnet_add_statistics(sc); 276 SLIST_INIT(&sc->vtnet_txhdr_free); 277 278 /* Register our feature descriptions. */ 279 virtio_set_feature_desc(dev, vtnet_feature_desc); 280 vtnet_negotiate_features(sc); 281 282 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) 283 sc->vtnet_flags |= VTNET_FLAG_INDIRECT; 284 285 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { 286 /* This feature should always be negotiated. */ 287 sc->vtnet_flags |= VTNET_FLAG_MAC; 288 } 289 290 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { 291 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; 292 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 293 } else { 294 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 295 } 296 297 sc->vtnet_rx_mbuf_size = MCLBYTES; 298 sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc); 299 300 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { 301 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; 302 303 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 304 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; 305 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) 306 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; 307 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR) && 308 virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 309 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; 310 } 311 312 error = vtnet_alloc_intrs(sc); 313 if (error) { 314 device_printf(dev, "cannot allocate interrupts\n"); 315 goto fail; 316 } 317 318 error = vtnet_alloc_virtqueues(sc); 319 if (error) { 320 device_printf(dev, "cannot allocate virtqueues\n"); 321 goto fail; 322 } 323 324 error = vtnet_bind_intrs(sc); 325 if (error) { 326 device_printf(dev, "cannot bind virtqueues to interrupts\n"); 327 goto fail; 328 } 329 330 /* Read (or generate) the MAC address for the adapter. */ 331 vtnet_get_hwaddr(sc); 332 333 error = vtnet_setup_interface(sc); 334 if (error) { 335 device_printf(dev, "cannot setup interface\n"); 336 goto fail; 337 } 338 339 for (i = 0; i < sc->vtnet_nintr; i++) { 340 error = virtio_setup_intr(dev, i, sc->vtnet_intr_slz[i]); 341 if (error) { 342 device_printf(dev, "cannot setup virtqueue " 343 "interrupts\n"); 344 ether_ifdetach(sc->vtnet_ifp); 345 goto fail; 346 } 347 } 348 349 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) { 350 ifnet_serialize_all(sc->vtnet_ifp); 351 vtnet_set_hwaddr(sc); 352 ifnet_deserialize_all(sc->vtnet_ifp); 353 } 354 355 /* 356 * Device defaults to promiscuous mode for backwards 357 * compatibility. Turn it off if possible. 358 */ 359 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 360 ifnet_serialize_all(sc->vtnet_ifp); 361 if (vtnet_set_promisc(sc, 0) != 0) { 362 sc->vtnet_ifp->if_flags |= IFF_PROMISC; 363 device_printf(dev, 364 "cannot disable promiscuous mode\n"); 365 } 366 ifnet_deserialize_all(sc->vtnet_ifp); 367 } else 368 sc->vtnet_ifp->if_flags |= IFF_PROMISC; 369 370 fail: 371 if (error) 372 vtnet_detach(dev); 373 374 return (error); 375 } 376 377 static int 378 vtnet_detach(device_t dev) 379 { 380 struct vtnet_softc *sc; 381 struct ifnet *ifp; 382 int i; 383 384 sc = device_get_softc(dev); 385 ifp = sc->vtnet_ifp; 386 387 for (i = 0; i < sc->vtnet_nintr; i++) 388 virtio_teardown_intr(dev, i); 389 390 if (device_is_attached(dev)) { 391 ifnet_serialize_all(ifp); 392 vtnet_stop(sc); 393 lwkt_serialize_handler_disable(&sc->vtnet_slz); 394 lwkt_serialize_handler_disable(&sc->vtnet_rx_slz); 395 lwkt_serialize_handler_disable(&sc->vtnet_tx_slz); 396 ifnet_deserialize_all(ifp); 397 398 ether_ifdetach(ifp); 399 } 400 401 if (sc->vtnet_vlan_attach != NULL) { 402 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach); 403 sc->vtnet_vlan_attach = NULL; 404 } 405 if (sc->vtnet_vlan_detach != NULL) { 406 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach); 407 sc->vtnet_vlan_detach = NULL; 408 } 409 410 if (ifp) { 411 if_free(ifp); 412 sc->vtnet_ifp = NULL; 413 } 414 415 if (sc->vtnet_rx_vq != NULL) 416 vtnet_free_rx_mbufs(sc); 417 if (sc->vtnet_tx_vq != NULL) 418 vtnet_free_tx_mbufs(sc); 419 if (sc->vtnet_ctrl_vq != NULL) 420 vtnet_free_ctrl_vq(sc); 421 422 if (sc->vtnet_txhdrarea != NULL) { 423 contigfree(sc->vtnet_txhdrarea, 424 sc->vtnet_txhdrcount * sizeof(struct vtnet_tx_header), 425 M_VTNET); 426 sc->vtnet_txhdrarea = NULL; 427 } 428 SLIST_INIT(&sc->vtnet_txhdr_free); 429 if (sc->vtnet_macfilter != NULL) { 430 contigfree(sc->vtnet_macfilter, 431 sizeof(struct vtnet_mac_filter), M_DEVBUF); 432 sc->vtnet_macfilter = NULL; 433 } 434 435 ifmedia_removeall(&sc->vtnet_media); 436 437 return (0); 438 } 439 440 static int 441 vtnet_suspend(device_t dev) 442 { 443 struct vtnet_softc *sc; 444 445 sc = device_get_softc(dev); 446 447 ifnet_serialize_all(sc->vtnet_ifp); 448 vtnet_stop(sc); 449 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED; 450 ifnet_deserialize_all(sc->vtnet_ifp); 451 452 return (0); 453 } 454 455 static int 456 vtnet_resume(device_t dev) 457 { 458 struct vtnet_softc *sc; 459 struct ifnet *ifp; 460 461 sc = device_get_softc(dev); 462 ifp = sc->vtnet_ifp; 463 464 ifnet_serialize_all(ifp); 465 if (ifp->if_flags & IFF_UP) 466 vtnet_init(sc); 467 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED; 468 ifnet_deserialize_all(ifp); 469 470 return (0); 471 } 472 473 static int 474 vtnet_shutdown(device_t dev) 475 { 476 477 /* 478 * Suspend already does all of what we need to 479 * do here; we just never expect to be resumed. 480 */ 481 return (vtnet_suspend(dev)); 482 } 483 484 static void 485 vtnet_negotiate_features(struct vtnet_softc *sc) 486 { 487 device_t dev; 488 uint64_t mask, features; 489 490 dev = sc->vtnet_dev; 491 mask = 0; 492 493 if (vtnet_csum_disable) 494 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM; 495 496 /* 497 * XXX DragonFly doesn't support receive checksum offload for ipv6 yet, 498 * hence always disable the virtio feature for now. 499 * XXX We need to support the DynOffload feature, in order to 500 * dynamically enable/disable this feature. 501 */ 502 mask |= VIRTIO_NET_F_GUEST_CSUM; 503 504 /* 505 * TSO is only available when the tx checksum offload feature is also 506 * negotiated. 507 */ 508 if (vtnet_csum_disable || vtnet_tso_disable) 509 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | 510 VIRTIO_NET_F_HOST_ECN; 511 512 if (vtnet_lro_disable) 513 mask |= VTNET_LRO_FEATURES; 514 515 features = VTNET_FEATURES & ~mask; 516 features |= VIRTIO_F_NOTIFY_ON_EMPTY; 517 features |= VIRTIO_F_ANY_LAYOUT; 518 sc->vtnet_features = virtio_negotiate_features(dev, features); 519 520 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) && 521 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) { 522 /* 523 * LRO without mergeable buffers requires special care. This 524 * is not ideal because every receive buffer must be large 525 * enough to hold the maximum TCP packet, the Ethernet header, 526 * and the header. This requires up to 34 descriptors with 527 * MCLBYTES clusters. If we do not have indirect descriptors, 528 * LRO is disabled since the virtqueue will not contain very 529 * many receive buffers. 530 */ 531 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) { 532 device_printf(dev, 533 "LRO disabled due to both mergeable buffers and " 534 "indirect descriptors not negotiated\n"); 535 536 features &= ~VTNET_LRO_FEATURES; 537 sc->vtnet_features = 538 virtio_negotiate_features(dev, features); 539 } else 540 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; 541 } 542 } 543 544 static void 545 vtnet_serialize(struct ifnet *ifp, enum ifnet_serialize slz) 546 { 547 struct vtnet_softc *sc = ifp->if_softc; 548 549 ifnet_serialize_array_enter(sc->serializes, 3, slz); 550 } 551 552 static void 553 vtnet_deserialize(struct ifnet *ifp, enum ifnet_serialize slz) 554 { 555 struct vtnet_softc *sc = ifp->if_softc; 556 557 ifnet_serialize_array_exit(sc->serializes, 3, slz); 558 } 559 560 static int 561 vtnet_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz) 562 { 563 struct vtnet_softc *sc = ifp->if_softc; 564 565 return ifnet_serialize_array_try(sc->serializes, 3, slz); 566 } 567 568 #ifdef INVARIANTS 569 570 static void 571 vtnet_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz, 572 boolean_t serialized) 573 { 574 struct vtnet_softc *sc = ifp->if_softc; 575 576 ifnet_serialize_array_assert(sc->serializes, 3, slz, serialized); 577 } 578 579 #endif /* INVARIANTS */ 580 581 static int 582 vtnet_alloc_intrs(struct vtnet_softc *sc) 583 { 584 int cnt, error; 585 int intrcount = virtio_intr_count(sc->vtnet_dev); 586 int i; 587 int use_config; 588 589 if (virtio_with_feature(sc->vtnet_dev, VIRTIO_NET_F_STATUS)) { 590 use_config = 1; 591 /* We can use a maximum of 3 interrupt vectors. */ 592 intrcount = imin(intrcount, 3); 593 } else { 594 /* We can use a maximum of 2 interrupt vectors. */ 595 intrcount = imin(intrcount, 2); 596 } 597 598 if (intrcount < 1) 599 return (ENXIO); 600 601 for (i = 0; i < intrcount; i++) 602 sc->vtnet_cpus[i] = -1; 603 604 cnt = intrcount; 605 error = virtio_intr_alloc(sc->vtnet_dev, &cnt, use_config, 606 sc->vtnet_cpus); 607 if (error != 0) { 608 virtio_intr_release(sc->vtnet_dev); 609 return (error); 610 } 611 sc->vtnet_nintr = cnt; 612 613 return (0); 614 } 615 616 static int 617 vtnet_alloc_virtqueues(struct vtnet_softc *sc) 618 { 619 device_t dev; 620 struct vq_alloc_info vq_info[3]; 621 int nvqs; 622 623 dev = sc->vtnet_dev; 624 nvqs = 2; 625 626 /* 627 * Indirect descriptors are not needed for the Rx 628 * virtqueue when mergeable buffers are negotiated. 629 * The header is placed inline with the data, not 630 * in a separate descriptor, and mbuf clusters are 631 * always physically contiguous. 632 */ 633 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 634 sc->vtnet_rx_nsegs = (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) ? 635 VTNET_MAX_RX_SEGS : VTNET_MIN_RX_SEGS; 636 } else 637 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS; 638 639 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || 640 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 641 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS; 642 else 643 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS; 644 645 VQ_ALLOC_INFO_INIT(&vq_info[0], sc->vtnet_rx_nsegs, &sc->vtnet_rx_vq, 646 "%s receive", device_get_nameunit(dev)); 647 648 VQ_ALLOC_INFO_INIT(&vq_info[1], sc->vtnet_tx_nsegs, &sc->vtnet_tx_vq, 649 "%s transmit", device_get_nameunit(dev)); 650 651 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 652 nvqs++; 653 654 VQ_ALLOC_INFO_INIT(&vq_info[2], 0, &sc->vtnet_ctrl_vq, 655 "%s control", device_get_nameunit(dev)); 656 } 657 658 return (virtio_alloc_virtqueues(dev, nvqs, vq_info)); 659 } 660 661 static int 662 vtnet_bind_intrs(struct vtnet_softc *sc) 663 { 664 int error = 0; 665 int i; 666 667 for (i = 0; i < 3; i++) 668 sc->vtnet_intr_slz[i] = &sc->vtnet_slz; 669 670 /* Possible "Virtqueue <-> IRQ" configurations */ 671 switch (sc->vtnet_nintr) { 672 case 1: 673 sc->vtnet_irqmap[0] = (struct irqmap){0, vtnet_rx_vq_intr}; 674 sc->vtnet_irqmap[1] = (struct irqmap){0, vtnet_tx_vq_intr}; 675 break; 676 case 2: 677 if (virtio_with_feature(sc->vtnet_dev, VIRTIO_NET_F_STATUS)) { 678 sc->vtnet_irqmap[0] = 679 (struct irqmap){1, vtnet_rx_vq_intr}; 680 sc->vtnet_irqmap[1] = 681 (struct irqmap){1, vtnet_tx_vq_intr}; 682 } else { 683 sc->vtnet_irqmap[0] = 684 (struct irqmap){0, vtnet_rx_msix_intr}; 685 sc->vtnet_irqmap[1] = 686 (struct irqmap){1, vtnet_tx_msix_intr}; 687 sc->vtnet_intr_slz[0] = &sc->vtnet_rx_slz; 688 sc->vtnet_intr_slz[1] = &sc->vtnet_tx_slz; 689 } 690 break; 691 case 3: 692 sc->vtnet_irqmap[0] = (struct irqmap){1, vtnet_rx_msix_intr}; 693 sc->vtnet_irqmap[1] = (struct irqmap){2, vtnet_tx_msix_intr}; 694 sc->vtnet_intr_slz[1] = &sc->vtnet_rx_slz; 695 sc->vtnet_intr_slz[2] = &sc->vtnet_tx_slz; 696 break; 697 default: 698 device_printf(sc->vtnet_dev, 699 "Invalid interrupt vector count: %d\n", sc->vtnet_nintr); 700 error = EINVAL; 701 goto fail; 702 } 703 704 for (i = 0; i < 2; i++) { 705 error = virtio_bind_intr(sc->vtnet_dev, 706 sc->vtnet_irqmap[i].irq, i, sc->vtnet_irqmap[i].handler, 707 sc); 708 if (error) { 709 device_printf(sc->vtnet_dev, 710 "cannot bind virtqueue IRQs\n"); 711 goto fail; 712 } 713 } 714 if (virtio_with_feature(sc->vtnet_dev, VIRTIO_NET_F_STATUS)) { 715 error = virtio_bind_intr(sc->vtnet_dev, 0, -1, 716 vtnet_config_intr, sc); 717 if (error) { 718 device_printf(sc->vtnet_dev, 719 "cannot bind config_change IRQ\n"); 720 goto fail; 721 } 722 } 723 724 fail: 725 return (error); 726 } 727 728 static int 729 vtnet_setup_interface(struct vtnet_softc *sc) 730 { 731 device_t dev; 732 struct ifnet *ifp; 733 int i; 734 735 dev = sc->vtnet_dev; 736 737 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER); 738 if (ifp == NULL) { 739 device_printf(dev, "cannot allocate ifnet structure\n"); 740 return (ENOSPC); 741 } 742 743 ifp->if_softc = sc; 744 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 745 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 746 ifp->if_init = vtnet_init; 747 ifp->if_start = vtnet_start; 748 #ifdef IFPOLL_ENABLE 749 ifp->if_npoll = vtnet_npoll; 750 #endif 751 ifp->if_serialize = vtnet_serialize; 752 ifp->if_deserialize = vtnet_deserialize; 753 ifp->if_tryserialize = vtnet_tryserialize; 754 #ifdef INVARIANTS 755 ifp->if_serialize_assert = vtnet_serialize_assert; 756 #endif 757 ifp->if_ioctl = vtnet_ioctl; 758 759 sc->vtnet_rx_process_limit = virtqueue_size(sc->vtnet_rx_vq); 760 sc->vtnet_tx_size = virtqueue_size(sc->vtnet_tx_vq); 761 if (sc->vtnet_flags & VTNET_FLAG_INDIRECT) 762 sc->vtnet_txhdrcount = sc->vtnet_tx_size; 763 else 764 sc->vtnet_txhdrcount = (sc->vtnet_tx_size / 2) + 1; 765 sc->vtnet_txhdrarea = contigmalloc( 766 sc->vtnet_txhdrcount * sizeof(struct vtnet_tx_header), 767 M_VTNET, M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0); 768 if (sc->vtnet_txhdrarea == NULL) { 769 device_printf(dev, "cannot contigmalloc the tx headers\n"); 770 return (ENOMEM); 771 } 772 for (i = 0; i < sc->vtnet_txhdrcount; i++) 773 vtnet_enqueue_txhdr(sc, &sc->vtnet_txhdrarea[i]); 774 sc->vtnet_macfilter = contigmalloc( 775 sizeof(struct vtnet_mac_filter), 776 M_DEVBUF, M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0); 777 if (sc->vtnet_macfilter == NULL) { 778 device_printf(dev, 779 "cannot contigmalloc the mac filter table\n"); 780 return (ENOMEM); 781 } 782 ifq_set_maxlen(&ifp->if_snd, sc->vtnet_tx_size - 1); 783 ifq_set_ready(&ifp->if_snd); 784 785 ether_ifattach(ifp, sc->vtnet_hwaddr, NULL); 786 787 /* The Tx IRQ is currently always the last allocated interrupt. */ 788 ifq_set_cpuid(&ifp->if_snd, sc->vtnet_cpus[sc->vtnet_nintr - 1]); 789 ifsq_watchdog_init(&sc->vtnet_tx_watchdog, 790 ifq_get_subq_default(&ifp->if_snd), vtnet_watchdog); 791 ifq_set_hw_serialize(&ifp->if_snd, &sc->vtnet_tx_slz); 792 793 /* Tell the upper layer(s) we support long frames. */ 794 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); 795 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU; 796 797 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { 798 ifp->if_capabilities |= IFCAP_TXCSUM; 799 800 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) 801 ifp->if_capabilities |= IFCAP_TSO4; 802 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 803 ifp->if_capabilities |= IFCAP_TSO6; 804 if (ifp->if_capabilities & IFCAP_TSO) 805 ifp->if_capabilities |= IFCAP_VLAN_HWTSO; 806 807 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) 808 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 809 } 810 811 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) 812 ifp->if_capabilities |= IFCAP_RXCSUM; 813 814 #if 0 /* IFCAP_LRO doesn't exist in DragonFly. */ 815 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) || 816 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6)) 817 ifp->if_capabilities |= IFCAP_LRO; 818 #endif 819 820 if ((ifp->if_capabilities & IFCAP_HWCSUM) == IFCAP_HWCSUM) { 821 /* 822 * VirtIO does not support VLAN tagging, but we can fake 823 * it by inserting and removing the 802.1Q header during 824 * transmit and receive. We are then able to do checksum 825 * offloading of VLAN frames. 826 */ 827 ifp->if_capabilities |= 828 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM; 829 } 830 831 ifp->if_capenable = ifp->if_capabilities; 832 833 /* 834 * Capabilities after here are not enabled by default. 835 */ 836 837 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 838 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; 839 840 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 841 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 842 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 843 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 844 } 845 846 return (0); 847 } 848 849 static void 850 vtnet_set_hwaddr(struct vtnet_softc *sc) 851 { 852 device_t dev; 853 854 dev = sc->vtnet_dev; 855 856 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) && 857 (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)) { 858 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0) 859 device_printf(dev, "unable to set MAC address\n"); 860 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) { 861 virtio_write_device_config(dev, 862 offsetof(struct virtio_net_config, mac), 863 sc->vtnet_hwaddr, ETHER_ADDR_LEN); 864 } 865 } 866 867 static void 868 vtnet_get_hwaddr(struct vtnet_softc *sc) 869 { 870 device_t dev; 871 872 dev = sc->vtnet_dev; 873 874 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) { 875 /* 876 * Generate a random locally administered unicast address. 877 * 878 * It would be nice to generate the same MAC address across 879 * reboots, but it seems all the hosts currently available 880 * support the MAC feature, so this isn't too important. 881 */ 882 sc->vtnet_hwaddr[0] = 0xB2; 883 karc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1); 884 return; 885 } 886 887 virtio_read_device_config(dev, 888 offsetof(struct virtio_net_config, mac), 889 sc->vtnet_hwaddr, ETHER_ADDR_LEN); 890 } 891 892 static int 893 vtnet_is_link_up(struct vtnet_softc *sc) 894 { 895 device_t dev; 896 struct ifnet *ifp; 897 uint16_t status; 898 899 dev = sc->vtnet_dev; 900 ifp = sc->vtnet_ifp; 901 902 ASSERT_SERIALIZED(&sc->vtnet_slz); 903 904 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) { 905 status = virtio_read_dev_config_2(dev, 906 offsetof(struct virtio_net_config, status)); 907 } else { 908 status = VIRTIO_NET_S_LINK_UP; 909 } 910 911 return ((status & VIRTIO_NET_S_LINK_UP) != 0); 912 } 913 914 static void 915 vtnet_update_link_status(struct vtnet_softc *sc) 916 { 917 device_t dev; 918 struct ifnet *ifp; 919 struct ifaltq_subque *ifsq; 920 int link; 921 922 dev = sc->vtnet_dev; 923 ifp = sc->vtnet_ifp; 924 ifsq = ifq_get_subq_default(&ifp->if_snd); 925 926 link = vtnet_is_link_up(sc); 927 928 if (link && ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0)) { 929 sc->vtnet_flags |= VTNET_FLAG_LINK; 930 if (bootverbose) 931 device_printf(dev, "Link is up\n"); 932 ifp->if_link_state = LINK_STATE_UP; 933 if_link_state_change(ifp); 934 if (!ifsq_is_empty(ifsq)) 935 ifsq_devstart_sched(ifsq); 936 } else if (!link && (sc->vtnet_flags & VTNET_FLAG_LINK)) { 937 sc->vtnet_flags &= ~VTNET_FLAG_LINK; 938 if (bootverbose) 939 device_printf(dev, "Link is down\n"); 940 941 ifp->if_link_state = LINK_STATE_DOWN; 942 if_link_state_change(ifp); 943 } 944 } 945 946 static void 947 vtnet_watchdog(struct ifaltq_subque *ifsq) 948 { 949 struct ifnet *ifp; 950 struct vtnet_softc *sc; 951 952 ifp = ifsq_get_ifp(ifsq); 953 sc = ifp->if_softc; 954 ASSERT_IFNET_SERIALIZED_ALL(ifp); 955 956 if (virtqueue_empty(sc->vtnet_tx_vq)) { 957 if_printf(ifp, "Spurious TX watchdog timeout -- ignoring\n"); 958 sc->vtnet_tx_watchdog.wd_timer = 0; 959 return; 960 } 961 962 if_printf(ifp, "TX watchdog timeout -- resetting\n"); 963 #ifdef VTNET_DEBUG 964 virtqueue_dump(sc->vtnet_tx_vq); 965 #endif 966 ifp->if_oerrors++; 967 ifp->if_flags &= ~IFF_RUNNING; 968 vtnet_init(sc); 969 ifsq_devstart_sched(ifsq); 970 } 971 972 static int 973 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data,struct ucred *cr) 974 { 975 struct vtnet_softc *sc; 976 struct ifreq *ifr; 977 int reinit, mask, error; 978 979 sc = ifp->if_softc; 980 ifr = (struct ifreq *) data; 981 reinit = 0; 982 error = 0; 983 984 switch (cmd) { 985 case SIOCSIFMTU: 986 if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > VTNET_MAX_MTU) 987 error = EINVAL; 988 else if (ifp->if_mtu != ifr->ifr_mtu) 989 error = vtnet_change_mtu(sc, ifr->ifr_mtu); 990 break; 991 992 case SIOCSIFFLAGS: 993 if ((ifp->if_flags & IFF_UP) == 0) { 994 if (ifp->if_flags & IFF_RUNNING) 995 vtnet_stop(sc); 996 } else if (ifp->if_flags & IFF_RUNNING) { 997 if ((ifp->if_flags ^ sc->vtnet_if_flags) & 998 (IFF_PROMISC | IFF_ALLMULTI)) { 999 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) 1000 vtnet_rx_filter(sc); 1001 else 1002 error = ENOTSUP; 1003 } 1004 } else { 1005 vtnet_init(sc); 1006 } 1007 1008 if (error == 0) 1009 sc->vtnet_if_flags = ifp->if_flags; 1010 break; 1011 1012 case SIOCADDMULTI: 1013 case SIOCDELMULTI: 1014 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) && 1015 (ifp->if_flags & IFF_RUNNING)) 1016 vtnet_rx_filter_mac(sc); 1017 break; 1018 1019 case SIOCSIFMEDIA: 1020 case SIOCGIFMEDIA: 1021 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); 1022 break; 1023 1024 case SIOCSIFCAP: 1025 mask = ifr->ifr_reqcap ^ ifp->if_capenable; 1026 1027 1028 if (mask & IFCAP_TXCSUM) { 1029 ifp->if_capenable ^= IFCAP_TXCSUM; 1030 if (ifp->if_capenable & IFCAP_TXCSUM) 1031 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD; 1032 else 1033 ifp->if_hwassist &= ~VTNET_CSUM_OFFLOAD; 1034 } 1035 1036 if (mask & IFCAP_TSO4) { 1037 ifp->if_capenable ^= IFCAP_TSO4; 1038 if (ifp->if_capenable & IFCAP_TSO4) 1039 ifp->if_hwassist |= CSUM_TSO; 1040 else 1041 ifp->if_hwassist &= ~CSUM_TSO; 1042 } 1043 1044 if (mask & IFCAP_RXCSUM) { 1045 ifp->if_capenable ^= IFCAP_RXCSUM; 1046 reinit = 1; 1047 } 1048 1049 #if 0 /* IFCAP_LRO doesn't exist in DragonFly. */ 1050 if (mask & IFCAP_LRO) { 1051 ifp->if_capenable ^= IFCAP_LRO; 1052 reinit = 1; 1053 } 1054 #endif 1055 1056 if (mask & IFCAP_VLAN_HWFILTER) { 1057 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; 1058 reinit = 1; 1059 } 1060 1061 if (mask & IFCAP_VLAN_HWTSO) 1062 ifp->if_capenable ^= IFCAP_VLAN_HWTSO; 1063 1064 if (mask & IFCAP_VLAN_HWTAGGING) 1065 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; 1066 1067 if (reinit && (ifp->if_flags & IFF_RUNNING)) { 1068 ifp->if_flags &= ~IFF_RUNNING; 1069 vtnet_init(sc); 1070 } 1071 //VLAN_CAPABILITIES(ifp); 1072 1073 break; 1074 1075 default: 1076 error = ether_ioctl(ifp, cmd, data); 1077 break; 1078 } 1079 1080 return (error); 1081 } 1082 1083 static int 1084 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu) 1085 { 1086 struct ifnet *ifp; 1087 int new_frame_size, clsize; 1088 1089 ifp = sc->vtnet_ifp; 1090 1091 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1092 new_frame_size = sizeof(struct vtnet_rx_header) + 1093 sizeof(struct ether_vlan_header) + new_mtu; 1094 1095 if (new_frame_size > MJUM9BYTES) 1096 return (EINVAL); 1097 1098 if (new_frame_size <= MCLBYTES) 1099 clsize = MCLBYTES; 1100 else 1101 clsize = MJUM9BYTES; 1102 } else { 1103 new_frame_size = sizeof(struct virtio_net_hdr_mrg_rxbuf) + 1104 sizeof(struct ether_vlan_header) + new_mtu; 1105 1106 if (new_frame_size <= MCLBYTES) 1107 clsize = MCLBYTES; 1108 else 1109 clsize = MJUMPAGESIZE; 1110 } 1111 1112 sc->vtnet_rx_mbuf_size = clsize; 1113 sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc); 1114 KASSERT(sc->vtnet_rx_mbuf_count < VTNET_MAX_RX_SEGS, 1115 ("too many rx mbufs: %d", sc->vtnet_rx_mbuf_count)); 1116 1117 ifp->if_mtu = new_mtu; 1118 1119 if (ifp->if_flags & IFF_RUNNING) { 1120 ifp->if_flags &= ~IFF_RUNNING; 1121 vtnet_init(sc); 1122 } 1123 1124 return (0); 1125 } 1126 1127 static int 1128 vtnet_init_rx_vq(struct vtnet_softc *sc) 1129 { 1130 struct virtqueue *vq; 1131 int nbufs, error; 1132 1133 vq = sc->vtnet_rx_vq; 1134 nbufs = 0; 1135 error = ENOSPC; 1136 1137 while (!virtqueue_full(vq)) { 1138 if ((error = vtnet_newbuf(sc)) != 0) 1139 break; 1140 nbufs++; 1141 } 1142 1143 if (nbufs > 0) { 1144 virtqueue_notify(vq, NULL); 1145 1146 /* 1147 * EMSGSIZE signifies the virtqueue did not have enough 1148 * entries available to hold the last mbuf. This is not 1149 * an error. We should not get ENOSPC since we check if 1150 * the virtqueue is full before attempting to add a 1151 * buffer. 1152 */ 1153 if (error == EMSGSIZE) 1154 error = 0; 1155 } 1156 1157 return (error); 1158 } 1159 1160 static void 1161 vtnet_free_rx_mbufs(struct vtnet_softc *sc) 1162 { 1163 struct virtqueue *vq; 1164 struct mbuf *m; 1165 int last; 1166 1167 vq = sc->vtnet_rx_vq; 1168 last = 0; 1169 1170 while ((m = virtqueue_drain(vq, &last)) != NULL) 1171 m_freem(m); 1172 1173 KASSERT(virtqueue_empty(vq), ("mbufs remaining in Rx Vq")); 1174 } 1175 1176 static void 1177 vtnet_free_tx_mbufs(struct vtnet_softc *sc) 1178 { 1179 struct virtqueue *vq; 1180 struct vtnet_tx_header *txhdr; 1181 int last; 1182 1183 vq = sc->vtnet_tx_vq; 1184 last = 0; 1185 1186 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { 1187 m_freem(txhdr->vth_mbuf); 1188 vtnet_enqueue_txhdr(sc, txhdr); 1189 } 1190 1191 KASSERT(virtqueue_empty(vq), ("mbufs remaining in Tx Vq")); 1192 } 1193 1194 static void 1195 vtnet_free_ctrl_vq(struct vtnet_softc *sc) 1196 { 1197 /* 1198 * The control virtqueue is only polled, therefore 1199 * it should already be empty. 1200 */ 1201 KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq), 1202 ("Ctrl Vq not empty")); 1203 } 1204 1205 static struct mbuf * 1206 vtnet_alloc_rxbuf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) 1207 { 1208 struct mbuf *m_head, *m_tail, *m; 1209 int i, clsize; 1210 1211 clsize = sc->vtnet_rx_mbuf_size; 1212 1213 /*use getcl instead of getjcl. see if_mxge.c comment line 2398*/ 1214 //m_head = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, clsize); 1215 m_head = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR ); 1216 if (m_head == NULL) 1217 goto fail; 1218 1219 m_head->m_len = clsize; 1220 m_tail = m_head; 1221 1222 if (nbufs > 1) { 1223 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1224 ("chained Rx mbuf requested without LRO_NOMRG")); 1225 1226 for (i = 0; i < nbufs - 1; i++) { 1227 //m = m_getjcl(M_DONTWAIT, MT_DATA, 0, clsize); 1228 m = m_getcl(M_NOWAIT, MT_DATA, 0); 1229 if (m == NULL) 1230 goto fail; 1231 1232 m->m_len = clsize; 1233 m_tail->m_next = m; 1234 m_tail = m; 1235 } 1236 } 1237 1238 if (m_tailp != NULL) 1239 *m_tailp = m_tail; 1240 1241 return (m_head); 1242 1243 fail: 1244 sc->vtnet_stats.mbuf_alloc_failed++; 1245 m_freem(m_head); 1246 1247 return (NULL); 1248 } 1249 1250 static int 1251 vtnet_replace_rxbuf(struct vtnet_softc *sc, struct mbuf *m0, int len0) 1252 { 1253 struct mbuf *m, *m_prev; 1254 struct mbuf *m_new, *m_tail; 1255 int len, clsize, nreplace, error; 1256 1257 m = m0; 1258 m_prev = NULL; 1259 len = len0; 1260 1261 m_tail = NULL; 1262 clsize = sc->vtnet_rx_mbuf_size; 1263 nreplace = 0; 1264 1265 if (m->m_next != NULL) 1266 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1267 ("chained Rx mbuf without LRO_NOMRG")); 1268 1269 /* 1270 * Since LRO_NOMRG mbuf chains are so large, we want to avoid 1271 * allocating an entire chain for each received frame. When 1272 * the received frame's length is less than that of the chain, 1273 * the unused mbufs are reassigned to the new chain. 1274 */ 1275 while (len > 0) { 1276 /* 1277 * Something is seriously wrong if we received 1278 * a frame larger than the mbuf chain. Drop it. 1279 */ 1280 if (m == NULL) { 1281 sc->vtnet_stats.rx_frame_too_large++; 1282 return (EMSGSIZE); 1283 } 1284 1285 KASSERT(m->m_len == clsize, 1286 ("mbuf length not expected cluster size: %d", 1287 m->m_len)); 1288 1289 m->m_len = MIN(m->m_len, len); 1290 len -= m->m_len; 1291 1292 m_prev = m; 1293 m = m->m_next; 1294 nreplace++; 1295 } 1296 1297 KASSERT(m_prev != NULL, ("m_prev == NULL")); 1298 KASSERT(nreplace <= sc->vtnet_rx_mbuf_count, 1299 ("too many replacement mbufs: %d/%d", nreplace, 1300 sc->vtnet_rx_mbuf_count)); 1301 1302 m_new = vtnet_alloc_rxbuf(sc, nreplace, &m_tail); 1303 if (m_new == NULL) { 1304 m_prev->m_len = clsize; 1305 return (ENOBUFS); 1306 } 1307 1308 /* 1309 * Move unused mbufs, if any, from the original chain 1310 * onto the end of the new chain. 1311 */ 1312 if (m_prev->m_next != NULL) { 1313 m_tail->m_next = m_prev->m_next; 1314 m_prev->m_next = NULL; 1315 } 1316 1317 error = vtnet_enqueue_rxbuf(sc, m_new); 1318 if (error) { 1319 /* 1320 * BAD! We could not enqueue the replacement mbuf chain. We 1321 * must restore the m0 chain to the original state if it was 1322 * modified so we can subsequently discard it. 1323 * 1324 * NOTE: The replacement is suppose to be an identical copy 1325 * to the one just dequeued so this is an unexpected error. 1326 */ 1327 sc->vtnet_stats.rx_enq_replacement_failed++; 1328 1329 if (m_tail->m_next != NULL) { 1330 m_prev->m_next = m_tail->m_next; 1331 m_tail->m_next = NULL; 1332 } 1333 1334 m_prev->m_len = clsize; 1335 m_freem(m_new); 1336 } 1337 1338 return (error); 1339 } 1340 1341 static int 1342 vtnet_newbuf(struct vtnet_softc *sc) 1343 { 1344 struct mbuf *m; 1345 int error; 1346 1347 m = vtnet_alloc_rxbuf(sc, sc->vtnet_rx_mbuf_count, NULL); 1348 if (m == NULL) 1349 return (ENOBUFS); 1350 1351 error = vtnet_enqueue_rxbuf(sc, m); 1352 if (error) 1353 m_freem(m); 1354 1355 return (error); 1356 } 1357 1358 static void 1359 vtnet_discard_merged_rxbuf(struct vtnet_softc *sc, int nbufs) 1360 { 1361 struct virtqueue *vq; 1362 struct mbuf *m; 1363 1364 vq = sc->vtnet_rx_vq; 1365 1366 while (--nbufs > 0) { 1367 if ((m = virtqueue_dequeue(vq, NULL)) == NULL) 1368 break; 1369 vtnet_discard_rxbuf(sc, m); 1370 } 1371 } 1372 1373 static void 1374 vtnet_discard_rxbuf(struct vtnet_softc *sc, struct mbuf *m) 1375 { 1376 int error; 1377 1378 /* 1379 * Requeue the discarded mbuf. This should always be 1380 * successful since it was just dequeued. 1381 */ 1382 error = vtnet_enqueue_rxbuf(sc, m); 1383 KASSERT(error == 0, ("cannot requeue discarded mbuf")); 1384 } 1385 1386 static int 1387 vtnet_enqueue_rxbuf(struct vtnet_softc *sc, struct mbuf *m) 1388 { 1389 struct sglist sg; 1390 struct sglist_seg segs[VTNET_MAX_RX_SEGS]; 1391 struct vtnet_rx_header *rxhdr; 1392 struct virtio_net_hdr *hdr; 1393 uint8_t *mdata; 1394 int offset, error; 1395 1396 ASSERT_SERIALIZED(&sc->vtnet_rx_slz); 1397 if ((sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) == 0) 1398 KASSERT(m->m_next == NULL, ("chained Rx mbuf")); 1399 1400 sglist_init(&sg, sc->vtnet_rx_nsegs, segs); 1401 1402 mdata = mtod(m, uint8_t *); 1403 offset = 0; 1404 1405 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1406 rxhdr = (struct vtnet_rx_header *) mdata; 1407 hdr = &rxhdr->vrh_hdr; 1408 offset += sizeof(struct vtnet_rx_header); 1409 1410 error = sglist_append(&sg, hdr, sc->vtnet_hdr_size); 1411 KASSERT(error == 0, ("cannot add header to sglist")); 1412 } 1413 1414 error = sglist_append(&sg, mdata + offset, m->m_len - offset); 1415 if (error) 1416 return (error); 1417 1418 if (m->m_next != NULL) { 1419 error = sglist_append_mbuf(&sg, m->m_next); 1420 if (error) 1421 return (error); 1422 } 1423 1424 return (virtqueue_enqueue(sc->vtnet_rx_vq, m, &sg, 0, sg.sg_nseg)); 1425 } 1426 1427 #ifdef IFPOLL_ENABLE 1428 1429 static void 1430 vtnet_npoll_status(struct ifnet *ifp) 1431 { 1432 struct vtnet_softc *sc = ifp->if_softc; 1433 1434 ASSERT_SERIALIZED(&sc->vtnet_slz); 1435 1436 vtnet_update_link_status(sc); 1437 } 1438 1439 static void 1440 vtnet_npoll_rx(struct ifnet *ifp, void *arg __unused, int cycle) 1441 { 1442 struct vtnet_softc *sc = ifp->if_softc; 1443 1444 vtnet_rxeof(sc, cycle, NULL); 1445 } 1446 1447 static void 1448 vtnet_npoll_tx(struct ifnet *ifp, void *arg __unused, int cycle __unused) 1449 { 1450 struct vtnet_softc *sc = ifp->if_softc; 1451 1452 ASSERT_SERIALIZED(&sc->vtnet_tx_slz); 1453 1454 vtnet_txeof(sc); 1455 if (!ifq_is_empty(&ifp->if_snd)) 1456 if_devstart(ifp); 1457 } 1458 1459 static void 1460 vtnet_npoll(struct ifnet *ifp, struct ifpoll_info *info) 1461 { 1462 struct vtnet_softc *sc = ifp->if_softc; 1463 int i; 1464 1465 ASSERT_IFNET_SERIALIZED_ALL(ifp); 1466 1467 if (info) { 1468 int cpu; 1469 1470 info->ifpi_status.status_func = vtnet_npoll_status; 1471 info->ifpi_status.serializer = &sc->vtnet_slz; 1472 1473 /* Use the same cpu for rx and tx. */ 1474 cpu = device_get_unit(device_get_parent(sc->vtnet_dev)); 1475 /* Shuffle a bit. */ 1476 cpu = (cpu * 61) % netisr_ncpus; 1477 KKASSERT(cpu < netisr_ncpus); 1478 info->ifpi_tx[cpu].poll_func = vtnet_npoll_tx; 1479 info->ifpi_tx[cpu].arg = NULL; 1480 info->ifpi_tx[cpu].serializer = &sc->vtnet_tx_slz; 1481 ifq_set_cpuid(&ifp->if_snd, cpu); 1482 1483 info->ifpi_rx[cpu].poll_func = vtnet_npoll_rx; 1484 info->ifpi_rx[cpu].arg = NULL; 1485 info->ifpi_rx[cpu].serializer = &sc->vtnet_rx_slz; 1486 1487 for (i = 0; i < 3; i++) 1488 lwkt_serialize_handler_disable(sc->serializes[i]); 1489 vtnet_disable_rx_intr(sc); 1490 vtnet_disable_tx_intr(sc); 1491 for (i = 0; i < sc->vtnet_nintr; i++) 1492 virtio_teardown_intr(sc->vtnet_dev, i); 1493 if (virtio_with_feature(sc->vtnet_dev, VIRTIO_NET_F_STATUS)) 1494 virtio_unbind_intr(sc->vtnet_dev, -1); 1495 for (i = 0; i < 2; i++) 1496 virtio_unbind_intr(sc->vtnet_dev, i); 1497 } else { 1498 int error; 1499 1500 ifq_set_cpuid(&ifp->if_snd, 1501 sc->vtnet_cpus[sc->vtnet_nintr - 1]); 1502 for (i = 0; i < 3; i++) 1503 lwkt_serialize_handler_enable(sc->serializes[i]); 1504 for (i = 0; i < 2; i++) { 1505 error = virtio_bind_intr(sc->vtnet_dev, 1506 sc->vtnet_irqmap[i].irq, i, 1507 sc->vtnet_irqmap[i].handler, sc); 1508 if (error) { 1509 device_printf(sc->vtnet_dev, 1510 "cannot re-bind virtqueue IRQs\n"); 1511 } 1512 } 1513 if (virtio_with_feature(sc->vtnet_dev, VIRTIO_NET_F_STATUS)) { 1514 error = virtio_bind_intr(sc->vtnet_dev, 0, -1, 1515 vtnet_config_intr, sc); 1516 if (error) { 1517 device_printf(sc->vtnet_dev, 1518 "cannot re-bind config_change IRQ\n"); 1519 } 1520 } 1521 for (i = 0; i < sc->vtnet_nintr; i++) { 1522 error = virtio_setup_intr(sc->vtnet_dev, i, 1523 sc->vtnet_intr_slz[i]); 1524 if (error) { 1525 device_printf(sc->vtnet_dev, 1526 "cannot setup virtqueue interrupts\n"); 1527 } 1528 } 1529 vtnet_enable_rx_intr(sc); 1530 vtnet_enable_tx_intr(sc); 1531 } 1532 } 1533 1534 #endif /* IFPOLL_ENABLE */ 1535 1536 static void 1537 vtnet_vlan_tag_remove(struct mbuf *m) 1538 { 1539 struct ether_vlan_header *evl; 1540 1541 evl = mtod(m, struct ether_vlan_header *); 1542 1543 m->m_pkthdr.ether_vlantag = ntohs(evl->evl_tag); 1544 m->m_flags |= M_VLANTAG; 1545 1546 /* Strip the 802.1Q header. */ 1547 bcopy((char *) evl, (char *) evl + ETHER_VLAN_ENCAP_LEN, 1548 ETHER_HDR_LEN - ETHER_TYPE_LEN); 1549 m_adj(m, ETHER_VLAN_ENCAP_LEN); 1550 } 1551 1552 /* 1553 * Alternative method of doing receive checksum offloading. Rather 1554 * than parsing the received frame down to the IP header, use the 1555 * csum_offset to determine which CSUM_* flags are appropriate. We 1556 * can get by with doing this only because the checksum offsets are 1557 * unique for the things we care about. 1558 */ 1559 static int 1560 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m, 1561 struct virtio_net_hdr *hdr) 1562 { 1563 struct ether_header *eh; 1564 struct ether_vlan_header *evh; 1565 struct udphdr *udp; 1566 int csum_len; 1567 uint16_t eth_type; 1568 1569 csum_len = hdr->csum_start + hdr->csum_offset; 1570 1571 if (csum_len < sizeof(struct ether_header) + sizeof(struct ip)) 1572 return (1); 1573 if (m->m_len < csum_len) 1574 return (1); 1575 1576 eh = mtod(m, struct ether_header *); 1577 eth_type = ntohs(eh->ether_type); 1578 if (eth_type == ETHERTYPE_VLAN) { 1579 evh = mtod(m, struct ether_vlan_header *); 1580 eth_type = ntohs(evh->evl_proto); 1581 } 1582 1583 if (eth_type != ETHERTYPE_IP && eth_type != ETHERTYPE_IPV6) { 1584 sc->vtnet_stats.rx_csum_bad_ethtype++; 1585 return (1); 1586 } 1587 1588 /* Use the offset to determine the appropriate CSUM_* flags. */ 1589 switch (hdr->csum_offset) { 1590 case offsetof(struct udphdr, uh_sum): 1591 if (m->m_len < hdr->csum_start + sizeof(struct udphdr)) 1592 return (1); 1593 udp = (struct udphdr *)(mtod(m, uint8_t *) + hdr->csum_start); 1594 if (udp->uh_sum == 0) 1595 return (0); 1596 1597 /* FALLTHROUGH */ 1598 1599 case offsetof(struct tcphdr, th_sum): 1600 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1601 m->m_pkthdr.csum_data = 0xFFFF; 1602 break; 1603 1604 default: 1605 sc->vtnet_stats.rx_csum_bad_offset++; 1606 return (1); 1607 } 1608 1609 sc->vtnet_stats.rx_csum_offloaded++; 1610 1611 return (0); 1612 } 1613 1614 static int 1615 vtnet_rxeof_merged(struct vtnet_softc *sc, struct mbuf *m_head, int nbufs) 1616 { 1617 struct ifnet *ifp; 1618 struct virtqueue *vq; 1619 struct mbuf *m, *m_tail; 1620 int len; 1621 1622 ifp = sc->vtnet_ifp; 1623 vq = sc->vtnet_rx_vq; 1624 m_tail = m_head; 1625 1626 while (--nbufs > 0) { 1627 m = virtqueue_dequeue(vq, &len); 1628 if (m == NULL) { 1629 ifp->if_ierrors++; 1630 goto fail; 1631 } 1632 1633 if (vtnet_newbuf(sc) != 0) { 1634 ifp->if_iqdrops++; 1635 vtnet_discard_rxbuf(sc, m); 1636 if (nbufs > 1) 1637 vtnet_discard_merged_rxbuf(sc, nbufs); 1638 goto fail; 1639 } 1640 1641 if (m->m_len < len) 1642 len = m->m_len; 1643 1644 m->m_len = len; 1645 m->m_flags &= ~M_PKTHDR; 1646 1647 m_head->m_pkthdr.len += len; 1648 m_tail->m_next = m; 1649 m_tail = m; 1650 } 1651 1652 return (0); 1653 1654 fail: 1655 sc->vtnet_stats.rx_mergeable_failed++; 1656 m_freem(m_head); 1657 1658 return (1); 1659 } 1660 1661 static int 1662 vtnet_rxeof(struct vtnet_softc *sc, int count, int *rx_npktsp) 1663 { 1664 struct virtio_net_hdr lhdr; 1665 struct ifnet *ifp; 1666 struct virtqueue *vq; 1667 struct mbuf *m; 1668 struct ether_header *eh; 1669 struct virtio_net_hdr *hdr; 1670 struct virtio_net_hdr_mrg_rxbuf *mhdr; 1671 int len, deq, nbufs, adjsz, rx_npkts; 1672 1673 ifp = sc->vtnet_ifp; 1674 vq = sc->vtnet_rx_vq; 1675 hdr = &lhdr; 1676 deq = 0; 1677 rx_npkts = 0; 1678 1679 while (--count >= 0) { 1680 m = virtqueue_dequeue(vq, &len); 1681 if (m == NULL) 1682 break; 1683 deq++; 1684 1685 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { 1686 ifp->if_ierrors++; 1687 vtnet_discard_rxbuf(sc, m); 1688 continue; 1689 } 1690 1691 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1692 nbufs = 1; 1693 adjsz = sizeof(struct vtnet_rx_header); 1694 /* 1695 * Account for our pad between the header and 1696 * the actual start of the frame. 1697 */ 1698 len += VTNET_RX_HEADER_PAD; 1699 } else { 1700 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *); 1701 nbufs = mhdr->num_buffers; 1702 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); 1703 } 1704 1705 if (vtnet_replace_rxbuf(sc, m, len) != 0) { 1706 ifp->if_iqdrops++; 1707 vtnet_discard_rxbuf(sc, m); 1708 if (nbufs > 1) 1709 vtnet_discard_merged_rxbuf(sc, nbufs); 1710 continue; 1711 } 1712 1713 m->m_pkthdr.len = len; 1714 m->m_pkthdr.rcvif = ifp; 1715 m->m_pkthdr.csum_flags = 0; 1716 1717 if (nbufs > 1) { 1718 if (vtnet_rxeof_merged(sc, m, nbufs) != 0) 1719 continue; 1720 } 1721 1722 ifp->if_ipackets++; 1723 1724 /* 1725 * Save copy of header before we strip it. For both mergeable 1726 * and non-mergeable, the VirtIO header is placed first in the 1727 * mbuf's data. We no longer need num_buffers, so always use a 1728 * virtio_net_hdr. 1729 */ 1730 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr)); 1731 m_adj(m, adjsz); 1732 1733 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { 1734 eh = mtod(m, struct ether_header *); 1735 if (eh->ether_type == htons(ETHERTYPE_VLAN)) { 1736 vtnet_vlan_tag_remove(m); 1737 1738 /* 1739 * With the 802.1Q header removed, update the 1740 * checksum starting location accordingly. 1741 */ 1742 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1743 hdr->csum_start -= 1744 ETHER_VLAN_ENCAP_LEN; 1745 } 1746 } 1747 1748 if (ifp->if_capenable & IFCAP_RXCSUM && 1749 hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 1750 if (vtnet_rx_csum(sc, m, hdr) != 0) 1751 sc->vtnet_stats.rx_csum_failed++; 1752 } 1753 1754 rx_npkts++; 1755 ifp->if_input(ifp, m, NULL, mycpuid); 1756 1757 /* 1758 * The interface may have been stopped while we were 1759 * passing the packet up the network stack. 1760 */ 1761 if ((ifp->if_flags & IFF_RUNNING) == 0) 1762 break; 1763 } 1764 1765 if (deq > 0) 1766 virtqueue_notify(vq, NULL); 1767 1768 if (rx_npktsp != NULL) 1769 *rx_npktsp = rx_npkts; 1770 1771 return (count > 0 ? 0 : EAGAIN); 1772 } 1773 1774 static void 1775 vtnet_rx_msix_intr(void *xsc) 1776 { 1777 struct vtnet_softc *sc; 1778 struct ifnet *ifp; 1779 int more; 1780 1781 sc = xsc; 1782 ifp = sc->vtnet_ifp; 1783 1784 if (!virtqueue_pending(sc->vtnet_rx_vq)) 1785 return; 1786 1787 vtnet_disable_rx_intr(sc); 1788 next: 1789 if ((ifp->if_flags & IFF_RUNNING) == 0) { 1790 vtnet_enable_rx_intr(sc); 1791 return; 1792 } 1793 1794 more = vtnet_rxeof(sc, sc->vtnet_rx_process_limit, NULL); 1795 if (!more && vtnet_enable_rx_intr(sc) != 0) { 1796 vtnet_disable_rx_intr(sc); 1797 more = 1; 1798 } 1799 1800 if (more) { 1801 sc->vtnet_stats.rx_task_rescheduled++; 1802 goto next; 1803 } 1804 } 1805 1806 static void 1807 vtnet_rx_vq_intr(void *xsc) 1808 { 1809 struct vtnet_softc *sc = xsc; 1810 1811 lwkt_serialize_enter(&sc->vtnet_rx_slz); 1812 vtnet_rx_msix_intr(xsc); 1813 lwkt_serialize_exit(&sc->vtnet_rx_slz); 1814 } 1815 1816 static void 1817 vtnet_enqueue_txhdr(struct vtnet_softc *sc, struct vtnet_tx_header *txhdr) 1818 { 1819 bzero(txhdr, sizeof(*txhdr)); 1820 SLIST_INSERT_HEAD(&sc->vtnet_txhdr_free, txhdr, link); 1821 } 1822 1823 static void 1824 vtnet_txeof(struct vtnet_softc *sc) 1825 { 1826 struct virtqueue *vq; 1827 struct ifnet *ifp; 1828 struct vtnet_tx_header *txhdr; 1829 int deq; 1830 1831 vq = sc->vtnet_tx_vq; 1832 ifp = sc->vtnet_ifp; 1833 deq = 0; 1834 1835 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { 1836 deq++; 1837 ifp->if_opackets++; 1838 m_freem(txhdr->vth_mbuf); 1839 vtnet_enqueue_txhdr(sc, txhdr); 1840 } 1841 1842 if (deq > 0) { 1843 ifq_clr_oactive(&ifp->if_snd); 1844 if (virtqueue_empty(vq)) 1845 sc->vtnet_tx_watchdog.wd_timer = 0; 1846 else 1847 sc->vtnet_tx_watchdog.wd_timer = VTNET_WATCHDOG_TIMEOUT; 1848 } 1849 } 1850 1851 static struct mbuf * 1852 vtnet_tx_offload(struct vtnet_softc *sc, struct mbuf *m, 1853 struct virtio_net_hdr *hdr) 1854 { 1855 struct ifnet *ifp; 1856 struct ether_header *eh; 1857 struct ether_vlan_header *evh; 1858 struct ip *ip; 1859 struct ip6_hdr *ip6; 1860 struct tcphdr *tcp; 1861 int ip_offset; 1862 uint16_t eth_type, csum_start; 1863 uint8_t ip_proto, gso_type; 1864 1865 ifp = sc->vtnet_ifp; 1866 M_ASSERTPKTHDR(m); 1867 1868 ip_offset = sizeof(struct ether_header); 1869 if (m->m_len < ip_offset) { 1870 if ((m = m_pullup(m, ip_offset)) == NULL) 1871 return (NULL); 1872 } 1873 1874 eh = mtod(m, struct ether_header *); 1875 eth_type = ntohs(eh->ether_type); 1876 if (eth_type == ETHERTYPE_VLAN) { 1877 ip_offset = sizeof(struct ether_vlan_header); 1878 if (m->m_len < ip_offset) { 1879 if ((m = m_pullup(m, ip_offset)) == NULL) 1880 return (NULL); 1881 } 1882 evh = mtod(m, struct ether_vlan_header *); 1883 eth_type = ntohs(evh->evl_proto); 1884 } 1885 1886 switch (eth_type) { 1887 case ETHERTYPE_IP: 1888 if (m->m_len < ip_offset + sizeof(struct ip)) { 1889 m = m_pullup(m, ip_offset + sizeof(struct ip)); 1890 if (m == NULL) 1891 return (NULL); 1892 } 1893 1894 ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset); 1895 ip_proto = ip->ip_p; 1896 csum_start = ip_offset + (ip->ip_hl << 2); 1897 gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 1898 break; 1899 1900 case ETHERTYPE_IPV6: 1901 if (m->m_len < ip_offset + sizeof(struct ip6_hdr)) { 1902 m = m_pullup(m, ip_offset + sizeof(struct ip6_hdr)); 1903 if (m == NULL) 1904 return (NULL); 1905 } 1906 1907 ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset); 1908 /* 1909 * XXX Assume no extension headers are present. Presently, 1910 * this will always be true in the case of TSO, and FreeBSD 1911 * does not perform checksum offloading of IPv6 yet. 1912 */ 1913 ip_proto = ip6->ip6_nxt; 1914 csum_start = ip_offset + sizeof(struct ip6_hdr); 1915 gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 1916 break; 1917 1918 default: 1919 return (m); 1920 } 1921 1922 if (m->m_pkthdr.csum_flags & VTNET_CSUM_OFFLOAD) { 1923 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; 1924 hdr->csum_start = csum_start; 1925 hdr->csum_offset = m->m_pkthdr.csum_data; 1926 1927 sc->vtnet_stats.tx_csum_offloaded++; 1928 } 1929 1930 if (m->m_pkthdr.csum_flags & CSUM_TSO) { 1931 if (ip_proto != IPPROTO_TCP) 1932 return (m); 1933 1934 if (m->m_len < csum_start + sizeof(struct tcphdr)) { 1935 m = m_pullup(m, csum_start + sizeof(struct tcphdr)); 1936 if (m == NULL) 1937 return (NULL); 1938 } 1939 1940 tcp = (struct tcphdr *)(mtod(m, uint8_t *) + csum_start); 1941 hdr->gso_type = gso_type; 1942 hdr->hdr_len = csum_start + (tcp->th_off << 2); 1943 hdr->gso_size = m->m_pkthdr.tso_segsz; 1944 1945 if (tcp->th_flags & TH_CWR) { 1946 /* 1947 * Drop if we did not negotiate VIRTIO_NET_F_HOST_ECN. 1948 * ECN support is only configurable globally with the 1949 * net.inet.tcp.ecn.enable sysctl knob. 1950 */ 1951 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { 1952 if_printf(ifp, "TSO with ECN not supported " 1953 "by host\n"); 1954 m_freem(m); 1955 return (NULL); 1956 } 1957 1958 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 1959 } 1960 1961 sc->vtnet_stats.tx_tso_offloaded++; 1962 } 1963 1964 return (m); 1965 } 1966 1967 static int 1968 vtnet_enqueue_txbuf(struct vtnet_softc *sc, struct mbuf **m_head, 1969 struct vtnet_tx_header *txhdr) 1970 { 1971 struct sglist sg; 1972 struct sglist_seg segs[VTNET_MAX_TX_SEGS]; 1973 struct virtqueue *vq; 1974 struct mbuf *m; 1975 int error; 1976 1977 vq = sc->vtnet_tx_vq; 1978 m = *m_head; 1979 1980 sglist_init(&sg, sc->vtnet_tx_nsegs, segs); 1981 error = sglist_append(&sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); 1982 KASSERT(error == 0 && sg.sg_nseg == 1, 1983 ("%s: error %d adding header to sglist", __func__, error)); 1984 1985 error = sglist_append_mbuf(&sg, m); 1986 if (error) { 1987 m = m_defrag(m, M_NOWAIT); 1988 if (m == NULL) 1989 goto fail; 1990 1991 *m_head = m; 1992 sc->vtnet_stats.tx_defragged++; 1993 1994 error = sglist_append_mbuf(&sg, m); 1995 if (error) 1996 goto fail; 1997 } 1998 1999 txhdr->vth_mbuf = m; 2000 error = virtqueue_enqueue(vq, txhdr, &sg, sg.sg_nseg, 0); 2001 2002 return (error); 2003 2004 fail: 2005 sc->vtnet_stats.tx_defrag_failed++; 2006 m_freem(*m_head); 2007 *m_head = NULL; 2008 2009 return (ENOBUFS); 2010 } 2011 2012 static struct mbuf * 2013 vtnet_vlan_tag_insert(struct mbuf *m) 2014 { 2015 struct mbuf *n; 2016 struct ether_vlan_header *evl; 2017 2018 if (M_WRITABLE(m) == 0) { 2019 n = m_dup(m, M_NOWAIT); 2020 m_freem(m); 2021 if ((m = n) == NULL) 2022 return (NULL); 2023 } 2024 2025 M_PREPEND(m, ETHER_VLAN_ENCAP_LEN, M_NOWAIT); 2026 if (m == NULL) 2027 return (NULL); 2028 if (m->m_len < sizeof(struct ether_vlan_header)) { 2029 m = m_pullup(m, sizeof(struct ether_vlan_header)); 2030 if (m == NULL) 2031 return (NULL); 2032 } 2033 2034 /* Insert 802.1Q header into the existing Ethernet header. */ 2035 evl = mtod(m, struct ether_vlan_header *); 2036 bcopy((char *) evl + ETHER_VLAN_ENCAP_LEN, 2037 (char *) evl, ETHER_HDR_LEN - ETHER_TYPE_LEN); 2038 evl->evl_encap_proto = htons(ETHERTYPE_VLAN); 2039 evl->evl_tag = htons(m->m_pkthdr.ether_vlantag); 2040 m->m_flags &= ~M_VLANTAG; 2041 2042 return (m); 2043 } 2044 2045 static int 2046 vtnet_encap(struct vtnet_softc *sc, struct mbuf **m_head) 2047 { 2048 struct vtnet_tx_header *txhdr; 2049 struct virtio_net_hdr *hdr; 2050 struct mbuf *m; 2051 int error; 2052 2053 txhdr = SLIST_FIRST(&sc->vtnet_txhdr_free); 2054 if (txhdr == NULL) 2055 return (ENOBUFS); 2056 SLIST_REMOVE_HEAD(&sc->vtnet_txhdr_free, link); 2057 2058 /* 2059 * Always use the non-mergeable header to simplify things. When 2060 * the mergeable feature is negotiated, the num_buffers field 2061 * must be set to zero. We use vtnet_hdr_size later to enqueue 2062 * the correct header size to the host. 2063 */ 2064 hdr = &txhdr->vth_uhdr.hdr; 2065 m = *m_head; 2066 2067 error = ENOBUFS; 2068 2069 if (m->m_flags & M_VLANTAG) { 2070 //m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); 2071 m = vtnet_vlan_tag_insert(m); 2072 if ((*m_head = m) == NULL) 2073 goto fail; 2074 m->m_flags &= ~M_VLANTAG; 2075 } 2076 2077 if (m->m_pkthdr.csum_flags != 0) { 2078 m = vtnet_tx_offload(sc, m, hdr); 2079 if ((*m_head = m) == NULL) 2080 goto fail; 2081 } 2082 2083 error = vtnet_enqueue_txbuf(sc, m_head, txhdr); 2084 fail: 2085 if (error != 0) 2086 vtnet_enqueue_txhdr(sc, txhdr); 2087 return (error); 2088 } 2089 2090 static void 2091 vtnet_start(struct ifnet *ifp, struct ifaltq_subque *ifsq) 2092 { 2093 struct vtnet_softc *sc; 2094 struct virtqueue *vq; 2095 struct mbuf *m0; 2096 int enq; 2097 2098 sc = ifp->if_softc; 2099 vq = sc->vtnet_tx_vq; 2100 enq = 0; 2101 2102 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq); 2103 ASSERT_SERIALIZED(&sc->vtnet_tx_slz); 2104 2105 if ((ifp->if_flags & (IFF_RUNNING)) != 2106 IFF_RUNNING || ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0)) 2107 return; 2108 2109 #ifdef VTNET_TX_INTR_MODERATION 2110 if (virtqueue_nused(vq) >= sc->vtnet_tx_size / 2) 2111 vtnet_txeof(sc); 2112 #endif 2113 2114 while (!ifsq_is_empty(ifsq)) { 2115 if (virtqueue_full(vq)) { 2116 ifsq_set_oactive(ifsq); 2117 break; 2118 } 2119 2120 m0 = ifsq_dequeue(ifsq); 2121 if (m0 == NULL) 2122 break; 2123 2124 if (vtnet_encap(sc, &m0) != 0) { 2125 if (m0 == NULL) 2126 break; 2127 ifsq_prepend(ifsq, m0); 2128 ifsq_set_oactive(ifsq); 2129 break; 2130 } 2131 2132 enq++; 2133 ETHER_BPF_MTAP(ifp, m0); 2134 } 2135 2136 if (enq > 0) { 2137 virtqueue_notify(vq, NULL); 2138 sc->vtnet_tx_watchdog.wd_timer = VTNET_WATCHDOG_TIMEOUT; 2139 } 2140 } 2141 2142 static void 2143 vtnet_tx_msix_intr(void *xsc) 2144 { 2145 struct vtnet_softc *sc; 2146 struct ifnet *ifp; 2147 struct ifaltq_subque *ifsq; 2148 2149 sc = xsc; 2150 ifp = sc->vtnet_ifp; 2151 ifsq = ifq_get_subq_default(&ifp->if_snd); 2152 2153 if (!virtqueue_pending(sc->vtnet_tx_vq)) 2154 return; 2155 2156 vtnet_disable_tx_intr(sc); 2157 next: 2158 if ((ifp->if_flags & IFF_RUNNING) == 0) { 2159 vtnet_enable_tx_intr(sc); 2160 return; 2161 } 2162 2163 vtnet_txeof(sc); 2164 2165 if (!ifsq_is_empty(ifsq)) 2166 ifsq_devstart(ifsq); 2167 2168 if (vtnet_enable_tx_intr(sc) != 0) { 2169 vtnet_disable_tx_intr(sc); 2170 sc->vtnet_stats.tx_task_rescheduled++; 2171 goto next; 2172 } 2173 } 2174 2175 static void 2176 vtnet_tx_vq_intr(void *xsc) 2177 { 2178 struct vtnet_softc *sc = xsc; 2179 2180 lwkt_serialize_enter(&sc->vtnet_tx_slz); 2181 vtnet_tx_msix_intr(xsc); 2182 lwkt_serialize_exit(&sc->vtnet_tx_slz); 2183 } 2184 2185 static void 2186 vtnet_config_intr(void *arg) 2187 { 2188 struct vtnet_softc *sc; 2189 2190 sc = arg; 2191 2192 vtnet_update_link_status(sc); 2193 } 2194 2195 static void 2196 vtnet_stop(struct vtnet_softc *sc) 2197 { 2198 device_t dev; 2199 struct ifnet *ifp; 2200 2201 dev = sc->vtnet_dev; 2202 ifp = sc->vtnet_ifp; 2203 2204 ASSERT_IFNET_SERIALIZED_ALL(ifp); 2205 2206 ifq_clr_oactive(&ifp->if_snd); 2207 ifsq_watchdog_stop(&sc->vtnet_tx_watchdog); 2208 ifp->if_flags &= ~(IFF_RUNNING); 2209 2210 vtnet_disable_rx_intr(sc); 2211 vtnet_disable_tx_intr(sc); 2212 2213 /* 2214 * Stop the host VirtIO adapter. Note this will reset the host 2215 * adapter's state back to the pre-initialized state, so in 2216 * order to make the device usable again, we must drive it 2217 * through virtio_reinit() and virtio_reinit_complete(). 2218 */ 2219 virtio_stop(dev); 2220 2221 sc->vtnet_flags &= ~VTNET_FLAG_LINK; 2222 2223 vtnet_free_rx_mbufs(sc); 2224 vtnet_free_tx_mbufs(sc); 2225 } 2226 2227 static int 2228 vtnet_virtio_reinit(struct vtnet_softc *sc) 2229 { 2230 device_t dev; 2231 struct ifnet *ifp; 2232 uint64_t features; 2233 int error; 2234 2235 dev = sc->vtnet_dev; 2236 ifp = sc->vtnet_ifp; 2237 features = sc->vtnet_features; 2238 2239 /* 2240 * Re-negotiate with the host, removing any disabled receive 2241 * features. Transmit features are disabled only on our side 2242 * via if_capenable and if_hwassist. 2243 */ 2244 2245 if (ifp->if_capabilities & IFCAP_RXCSUM) { 2246 if ((ifp->if_capenable & IFCAP_RXCSUM) == 0) 2247 features &= ~VIRTIO_NET_F_GUEST_CSUM; 2248 } 2249 2250 #if 0 /* IFCAP_LRO doesn't exist in DragonFly. */ 2251 if (ifp->if_capabilities & IFCAP_LRO) { 2252 if ((ifp->if_capenable & IFCAP_LRO) == 0) 2253 features &= ~VTNET_LRO_FEATURES; 2254 } 2255 #endif 2256 2257 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) { 2258 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0) 2259 features &= ~VIRTIO_NET_F_CTRL_VLAN; 2260 } 2261 2262 error = virtio_reinit(dev, features); 2263 if (error) 2264 device_printf(dev, "virtio reinit error %d\n", error); 2265 2266 return (error); 2267 } 2268 2269 static void 2270 vtnet_init(void *xsc) 2271 { 2272 struct vtnet_softc *sc; 2273 device_t dev; 2274 struct ifnet *ifp; 2275 int error; 2276 2277 sc = xsc; 2278 dev = sc->vtnet_dev; 2279 ifp = sc->vtnet_ifp; 2280 2281 ASSERT_IFNET_SERIALIZED_ALL(ifp); 2282 2283 if (ifp->if_flags & IFF_RUNNING) 2284 return; 2285 2286 /* Stop host's adapter, cancel any pending I/O. */ 2287 vtnet_stop(sc); 2288 2289 /* Reinitialize the host device. */ 2290 error = vtnet_virtio_reinit(sc); 2291 if (error) { 2292 device_printf(dev, 2293 "reinitialization failed, stopping device...\n"); 2294 vtnet_stop(sc); 2295 return; 2296 } 2297 2298 /* Update host with assigned MAC address. */ 2299 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); 2300 vtnet_set_hwaddr(sc); 2301 2302 ifp->if_hwassist = 0; 2303 if (ifp->if_capenable & IFCAP_TXCSUM) 2304 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD; 2305 if (ifp->if_capenable & IFCAP_TSO4) 2306 ifp->if_hwassist |= CSUM_TSO; 2307 2308 error = vtnet_init_rx_vq(sc); 2309 if (error) { 2310 device_printf(dev, 2311 "cannot allocate mbufs for Rx virtqueue\n"); 2312 vtnet_stop(sc); 2313 return; 2314 } 2315 2316 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 2317 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 2318 /* Restore promiscuous and all-multicast modes. */ 2319 vtnet_rx_filter(sc); 2320 2321 /* Restore filtered MAC addresses. */ 2322 vtnet_rx_filter_mac(sc); 2323 } 2324 2325 /* Restore VLAN filters. */ 2326 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) 2327 vtnet_rx_filter_vlan(sc); 2328 } 2329 2330 #ifdef IFPOLL_ENABLE 2331 if (!(ifp->if_flags & IFF_NPOLLING)) 2332 #endif 2333 { 2334 vtnet_enable_rx_intr(sc); 2335 vtnet_enable_tx_intr(sc); 2336 } 2337 2338 ifp->if_flags |= IFF_RUNNING; 2339 ifq_clr_oactive(&ifp->if_snd); 2340 ifsq_watchdog_start(&sc->vtnet_tx_watchdog); 2341 2342 virtio_reinit_complete(dev); 2343 2344 vtnet_update_link_status(sc); 2345 } 2346 2347 static void 2348 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, 2349 struct sglist *sg, int readable, int writable) 2350 { 2351 struct virtqueue *vq; 2352 void *c; 2353 2354 vq = sc->vtnet_ctrl_vq; 2355 2356 ASSERT_IFNET_SERIALIZED_ALL(sc->vtnet_ifp); 2357 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ, 2358 ("no control virtqueue")); 2359 KASSERT(virtqueue_empty(vq), 2360 ("control command already enqueued")); 2361 2362 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0) 2363 return; 2364 2365 /* 2366 * XXX We can safely drop the serializer between here, and the end of 2367 * the function, when we can correctly sleep for this command to 2368 * be finished. 2369 */ 2370 virtqueue_notify(vq, NULL); 2371 2372 /* 2373 * Poll until the command is complete. Previously, we would 2374 * sleep until the control virtqueue interrupt handler woke 2375 * us up, but dropping the VTNET_MTX leads to serialization 2376 * difficulties. 2377 * 2378 * Furthermore, it appears QEMU/KVM only allocates three MSIX 2379 * vectors. Two of those vectors are needed for the Rx and Tx 2380 * virtqueues. We do not support sharing both a Vq and config 2381 * changed notification on the same MSIX vector. 2382 */ 2383 c = virtqueue_poll(vq, NULL); 2384 KASSERT(c == cookie, ("unexpected control command response")); 2385 } 2386 2387 static int 2388 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) 2389 { 2390 struct { 2391 struct virtio_net_ctrl_hdr hdr __aligned(2); 2392 uint8_t pad1; 2393 char aligned_hwaddr[ETHER_ADDR_LEN] __aligned(8); 2394 uint8_t pad2; 2395 uint8_t ack; 2396 } s; 2397 struct sglist_seg segs[3]; 2398 struct sglist sg; 2399 int error; 2400 2401 s.hdr.class = VIRTIO_NET_CTRL_MAC; 2402 s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 2403 s.ack = VIRTIO_NET_ERR; 2404 2405 /* Copy the mac address into physically contiguous memory */ 2406 memcpy(s.aligned_hwaddr, hwaddr, ETHER_ADDR_LEN); 2407 2408 sglist_init(&sg, 3, segs); 2409 error = 0; 2410 error |= sglist_append(&sg, &s.hdr, 2411 sizeof(struct virtio_net_ctrl_hdr)); 2412 error |= sglist_append(&sg, s.aligned_hwaddr, ETHER_ADDR_LEN); 2413 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 2414 KASSERT(error == 0 && sg.sg_nseg == 3, 2415 ("%s: error %d adding set MAC msg to sglist", __func__, error)); 2416 2417 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 2418 2419 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 2420 } 2421 2422 static void 2423 vtnet_rx_filter(struct vtnet_softc *sc) 2424 { 2425 device_t dev; 2426 struct ifnet *ifp; 2427 2428 dev = sc->vtnet_dev; 2429 ifp = sc->vtnet_ifp; 2430 2431 ASSERT_IFNET_SERIALIZED_ALL(ifp); 2432 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 2433 ("CTRL_RX feature not negotiated")); 2434 2435 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) 2436 device_printf(dev, "cannot %s promiscuous mode\n", 2437 (ifp->if_flags & IFF_PROMISC) ? "enable" : "disable"); 2438 2439 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) 2440 device_printf(dev, "cannot %s all-multicast mode\n", 2441 (ifp->if_flags & IFF_ALLMULTI) ? "enable" : "disable"); 2442 } 2443 2444 static int 2445 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on) 2446 { 2447 struct sglist_seg segs[3]; 2448 struct sglist sg; 2449 struct { 2450 struct virtio_net_ctrl_hdr hdr __aligned(2); 2451 uint8_t pad1; 2452 uint8_t onoff; 2453 uint8_t pad2; 2454 uint8_t ack; 2455 } s; 2456 int error; 2457 2458 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 2459 ("%s: CTRL_RX feature not negotiated", __func__)); 2460 2461 s.hdr.class = VIRTIO_NET_CTRL_RX; 2462 s.hdr.cmd = cmd; 2463 s.onoff = !!on; 2464 s.ack = VIRTIO_NET_ERR; 2465 2466 sglist_init(&sg, 3, segs); 2467 error = 0; 2468 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 2469 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); 2470 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 2471 KASSERT(error == 0 && sg.sg_nseg == 3, 2472 ("%s: error %d adding Rx message to sglist", __func__, error)); 2473 2474 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 2475 2476 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 2477 } 2478 2479 static int 2480 vtnet_set_promisc(struct vtnet_softc *sc, int on) 2481 { 2482 2483 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); 2484 } 2485 2486 static int 2487 vtnet_set_allmulti(struct vtnet_softc *sc, int on) 2488 { 2489 2490 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); 2491 } 2492 2493 static void 2494 vtnet_rx_filter_mac(struct vtnet_softc *sc) 2495 { 2496 struct virtio_net_ctrl_hdr hdr __aligned(2); 2497 struct vtnet_mac_filter *filter; 2498 struct sglist_seg segs[4]; 2499 struct sglist sg; 2500 struct ifnet *ifp; 2501 struct ifaddr *ifa; 2502 struct ifaddr_container *ifac; 2503 struct ifmultiaddr *ifma; 2504 int ucnt, mcnt, promisc, allmulti, error; 2505 uint8_t ack; 2506 2507 ifp = sc->vtnet_ifp; 2508 ucnt = 0; 2509 mcnt = 0; 2510 promisc = 0; 2511 allmulti = 0; 2512 2513 ASSERT_IFNET_SERIALIZED_ALL(ifp); 2514 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 2515 ("%s: CTRL_RX feature not negotiated", __func__)); 2516 2517 /* Use the MAC filtering table allocated in vtnet_attach. */ 2518 filter = sc->vtnet_macfilter; 2519 memset(filter, 0, sizeof(struct vtnet_mac_filter)); 2520 2521 /* Unicast MAC addresses: */ 2522 //if_addr_rlock(ifp); 2523 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) { 2524 ifa = ifac->ifa; 2525 if (ifa->ifa_addr->sa_family != AF_LINK) 2526 continue; 2527 else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), 2528 sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) 2529 continue; 2530 else if (ucnt == VTNET_MAX_MAC_ENTRIES) { 2531 promisc = 1; 2532 break; 2533 } 2534 2535 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), 2536 &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN); 2537 ucnt++; 2538 } 2539 //if_addr_runlock(ifp); 2540 2541 if (promisc != 0) { 2542 filter->vmf_unicast.nentries = 0; 2543 if_printf(ifp, "more than %d MAC addresses assigned, " 2544 "falling back to promiscuous mode\n", 2545 VTNET_MAX_MAC_ENTRIES); 2546 } else 2547 filter->vmf_unicast.nentries = ucnt; 2548 2549 /* Multicast MAC addresses: */ 2550 //if_maddr_rlock(ifp); 2551 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2552 if (ifma->ifma_addr->sa_family != AF_LINK) 2553 continue; 2554 else if (mcnt == VTNET_MAX_MAC_ENTRIES) { 2555 allmulti = 1; 2556 break; 2557 } 2558 2559 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 2560 &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN); 2561 mcnt++; 2562 } 2563 //if_maddr_runlock(ifp); 2564 2565 if (allmulti != 0) { 2566 filter->vmf_multicast.nentries = 0; 2567 if_printf(ifp, "more than %d multicast MAC addresses " 2568 "assigned, falling back to all-multicast mode\n", 2569 VTNET_MAX_MAC_ENTRIES); 2570 } else 2571 filter->vmf_multicast.nentries = mcnt; 2572 2573 if (promisc != 0 && allmulti != 0) 2574 goto out; 2575 2576 hdr.class = VIRTIO_NET_CTRL_MAC; 2577 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 2578 ack = VIRTIO_NET_ERR; 2579 2580 sglist_init(&sg, 4, segs); 2581 error = 0; 2582 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 2583 error |= sglist_append(&sg, &filter->vmf_unicast, 2584 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN); 2585 error |= sglist_append(&sg, &filter->vmf_multicast, 2586 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN); 2587 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 2588 KASSERT(error == 0 && sg.sg_nseg == 4, 2589 ("%s: error %d adding MAC filter msg to sglist", __func__, error)); 2590 2591 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 2592 2593 if (ack != VIRTIO_NET_OK) 2594 if_printf(ifp, "error setting host MAC filter table\n"); 2595 2596 out: 2597 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0) 2598 if_printf(ifp, "cannot enable promiscuous mode\n"); 2599 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0) 2600 if_printf(ifp, "cannot enable all-multicast mode\n"); 2601 } 2602 2603 static int 2604 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 2605 { 2606 struct sglist_seg segs[3]; 2607 struct sglist sg; 2608 struct { 2609 struct virtio_net_ctrl_hdr hdr __aligned(2); 2610 uint8_t pad1; 2611 uint16_t tag; 2612 uint8_t pad2; 2613 uint8_t ack; 2614 } s; 2615 int error; 2616 2617 s.hdr.class = VIRTIO_NET_CTRL_VLAN; 2618 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 2619 s.tag = tag; 2620 s.ack = VIRTIO_NET_ERR; 2621 2622 sglist_init(&sg, 3, segs); 2623 error = 0; 2624 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 2625 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); 2626 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 2627 KASSERT(error == 0 && sg.sg_nseg == 3, 2628 ("%s: error %d adding VLAN message to sglist", __func__, error)); 2629 2630 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 2631 2632 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 2633 } 2634 2635 static void 2636 vtnet_rx_filter_vlan(struct vtnet_softc *sc) 2637 { 2638 uint32_t w; 2639 uint16_t tag; 2640 int i, bit, nvlans; 2641 2642 ASSERT_IFNET_SERIALIZED_ALL(sc->vtnet_ifp); 2643 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER, 2644 ("%s: VLAN_FILTER feature not negotiated", __func__)); 2645 2646 nvlans = sc->vtnet_nvlans; 2647 2648 /* Enable the filter for each configured VLAN. */ 2649 for (i = 0; i < VTNET_VLAN_SHADOW_SIZE && nvlans > 0; i++) { 2650 w = sc->vtnet_vlan_shadow[i]; 2651 while ((bit = ffs(w) - 1) != -1) { 2652 w &= ~(1 << bit); 2653 tag = sizeof(w) * CHAR_BIT * i + bit; 2654 nvlans--; 2655 2656 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { 2657 device_printf(sc->vtnet_dev, 2658 "cannot enable VLAN %d filter\n", tag); 2659 } 2660 } 2661 } 2662 2663 KASSERT(nvlans == 0, ("VLAN count incorrect")); 2664 } 2665 2666 static void 2667 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 2668 { 2669 struct ifnet *ifp; 2670 int idx, bit; 2671 2672 ifp = sc->vtnet_ifp; 2673 idx = (tag >> 5) & 0x7F; 2674 bit = tag & 0x1F; 2675 2676 if (tag == 0 || tag > 4095) 2677 return; 2678 2679 ifnet_serialize_all(ifp); 2680 2681 /* Update shadow VLAN table. */ 2682 if (add) { 2683 sc->vtnet_nvlans++; 2684 sc->vtnet_vlan_shadow[idx] |= (1 << bit); 2685 } else { 2686 sc->vtnet_nvlans--; 2687 sc->vtnet_vlan_shadow[idx] &= ~(1 << bit); 2688 } 2689 2690 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER && 2691 vtnet_exec_vlan_filter(sc, add, tag) != 0) { 2692 device_printf(sc->vtnet_dev, 2693 "cannot %s VLAN %d %s the host filter table\n", 2694 add ? "add" : "remove", tag, add ? "to" : "from"); 2695 } 2696 2697 ifnet_deserialize_all(ifp); 2698 } 2699 2700 static void 2701 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 2702 { 2703 2704 if (ifp->if_softc != arg) 2705 return; 2706 2707 vtnet_update_vlan_filter(arg, 1, tag); 2708 } 2709 2710 static void 2711 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 2712 { 2713 2714 if (ifp->if_softc != arg) 2715 return; 2716 2717 vtnet_update_vlan_filter(arg, 0, tag); 2718 } 2719 2720 static int 2721 vtnet_ifmedia_upd(struct ifnet *ifp) 2722 { 2723 struct vtnet_softc *sc; 2724 struct ifmedia *ifm; 2725 2726 sc = ifp->if_softc; 2727 ifm = &sc->vtnet_media; 2728 2729 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) 2730 return (EINVAL); 2731 2732 return (0); 2733 } 2734 2735 static void 2736 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 2737 { 2738 struct vtnet_softc *sc; 2739 2740 sc = ifp->if_softc; 2741 2742 ifmr->ifm_status = IFM_AVALID; 2743 ifmr->ifm_active = IFM_ETHER; 2744 2745 if (vtnet_is_link_up(sc) != 0) { 2746 ifmr->ifm_status |= IFM_ACTIVE; 2747 ifmr->ifm_active |= VTNET_MEDIATYPE; 2748 } else 2749 ifmr->ifm_active |= IFM_NONE; 2750 } 2751 2752 static void 2753 vtnet_add_statistics(struct vtnet_softc *sc) 2754 { 2755 device_t dev; 2756 struct vtnet_statistics *stats; 2757 struct sysctl_ctx_list *ctx; 2758 struct sysctl_oid *tree; 2759 struct sysctl_oid_list *child; 2760 2761 dev = sc->vtnet_dev; 2762 stats = &sc->vtnet_stats; 2763 ctx = device_get_sysctl_ctx(dev); 2764 tree = device_get_sysctl_tree(dev); 2765 child = SYSCTL_CHILDREN(tree); 2766 2767 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", 2768 CTLFLAG_RD, &stats->mbuf_alloc_failed, 0, 2769 "Mbuf cluster allocation failures"); 2770 2771 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", 2772 CTLFLAG_RD, &stats->rx_frame_too_large, 0, 2773 "Received frame larger than the mbuf chain"); 2774 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", 2775 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 0, 2776 "Enqueuing the replacement receive mbuf failed"); 2777 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", 2778 CTLFLAG_RD, &stats->rx_mergeable_failed, 0, 2779 "Mergeable buffers receive failures"); 2780 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", 2781 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 0, 2782 "Received checksum offloaded buffer with unsupported " 2783 "Ethernet type"); 2784 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", 2785 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 0, 2786 "Received checksum offloaded buffer with incorrect IP protocol"); 2787 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", 2788 CTLFLAG_RD, &stats->rx_csum_bad_offset, 0, 2789 "Received checksum offloaded buffer with incorrect offset"); 2790 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed", 2791 CTLFLAG_RD, &stats->rx_csum_failed, 0, 2792 "Received buffer checksum offload failed"); 2793 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded", 2794 CTLFLAG_RD, &stats->rx_csum_offloaded, 0, 2795 "Received buffer checksum offload succeeded"); 2796 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled", 2797 CTLFLAG_RD, &stats->rx_task_rescheduled, 0, 2798 "Times the receive interrupt task rescheduled itself"); 2799 2800 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype", 2801 CTLFLAG_RD, &stats->tx_csum_bad_ethtype, 0, 2802 "Aborted transmit of checksum offloaded buffer with unknown " 2803 "Ethernet type"); 2804 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype", 2805 CTLFLAG_RD, &stats->tx_tso_bad_ethtype, 0, 2806 "Aborted transmit of TSO buffer with unknown Ethernet type"); 2807 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", 2808 CTLFLAG_RD, &stats->tx_defragged, 0, 2809 "Transmit mbufs defragged"); 2810 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", 2811 CTLFLAG_RD, &stats->tx_defrag_failed, 0, 2812 "Aborted transmit of buffer because defrag failed"); 2813 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded", 2814 CTLFLAG_RD, &stats->tx_csum_offloaded, 0, 2815 "Offloaded checksum of transmitted buffer"); 2816 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded", 2817 CTLFLAG_RD, &stats->tx_tso_offloaded, 0, 2818 "Segmentation offload of transmitted buffer"); 2819 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled", 2820 CTLFLAG_RD, &stats->tx_task_rescheduled, 0, 2821 "Times the transmit interrupt task rescheduled itself"); 2822 } 2823 2824 static int 2825 vtnet_enable_rx_intr(struct vtnet_softc *sc) 2826 { 2827 2828 return (virtqueue_enable_intr(sc->vtnet_rx_vq)); 2829 } 2830 2831 static void 2832 vtnet_disable_rx_intr(struct vtnet_softc *sc) 2833 { 2834 2835 virtqueue_disable_intr(sc->vtnet_rx_vq); 2836 } 2837 2838 static int 2839 vtnet_enable_tx_intr(struct vtnet_softc *sc) 2840 { 2841 2842 #ifdef VTNET_TX_INTR_MODERATION 2843 return (0); 2844 #else 2845 return (virtqueue_enable_intr(sc->vtnet_tx_vq)); 2846 #endif 2847 } 2848 2849 static void 2850 vtnet_disable_tx_intr(struct vtnet_softc *sc) 2851 { 2852 2853 virtqueue_disable_intr(sc->vtnet_tx_vq); 2854 } 2855