1 /* 2 * Virtio Network Device 3 * 4 * Copyright IBM, Corp. 2007 5 * 6 * Authors: 7 * Anthony Liguori <aliguori@us.ibm.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2. See 10 * the COPYING file in the top-level directory. 11 * 12 */ 13 14 #include "qemu/osdep.h" 15 #include "qemu/iov.h" 16 #include "hw/virtio/virtio.h" 17 #include "net/net.h" 18 #include "net/checksum.h" 19 #include "net/tap.h" 20 #include "qemu/error-report.h" 21 #include "qemu/timer.h" 22 #include "hw/virtio/virtio-net.h" 23 #include "net/vhost_net.h" 24 #include "hw/virtio/virtio-bus.h" 25 #include "qapi/qmp/qjson.h" 26 #include "qapi-event.h" 27 #include "hw/virtio/virtio-access.h" 28 #include "migration/misc.h" 29 30 #define VIRTIO_NET_VM_VERSION 11 31 32 #define MAC_TABLE_ENTRIES 64 33 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */ 34 35 /* previously fixed value */ 36 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256 37 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256 38 39 /* for now, only allow larger queues; with virtio-1, guest can downsize */ 40 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 41 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 42 43 /* 44 * Calculate the number of bytes up to and including the given 'field' of 45 * 'container'. 46 */ 47 #define endof(container, field) \ 48 (offsetof(container, field) + sizeof(((container *)0)->field)) 49 50 typedef struct VirtIOFeature { 51 uint32_t flags; 52 size_t end; 53 } VirtIOFeature; 54 55 static VirtIOFeature feature_sizes[] = { 56 {.flags = 1 << VIRTIO_NET_F_MAC, 57 .end = endof(struct virtio_net_config, mac)}, 58 {.flags = 1 << VIRTIO_NET_F_STATUS, 59 .end = endof(struct virtio_net_config, status)}, 60 {.flags = 1 << VIRTIO_NET_F_MQ, 61 .end = endof(struct virtio_net_config, max_virtqueue_pairs)}, 62 {.flags = 1 << VIRTIO_NET_F_MTU, 63 .end = endof(struct virtio_net_config, mtu)}, 64 {} 65 }; 66 67 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc) 68 { 69 VirtIONet *n = qemu_get_nic_opaque(nc); 70 71 return &n->vqs[nc->queue_index]; 72 } 73 74 static int vq2q(int queue_index) 75 { 76 return queue_index / 2; 77 } 78 79 /* TODO 80 * - we could suppress RX interrupt if we were so inclined. 81 */ 82 83 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config) 84 { 85 VirtIONet *n = VIRTIO_NET(vdev); 86 struct virtio_net_config netcfg; 87 88 virtio_stw_p(vdev, &netcfg.status, n->status); 89 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues); 90 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu); 91 memcpy(netcfg.mac, n->mac, ETH_ALEN); 92 memcpy(config, &netcfg, n->config_size); 93 } 94 95 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config) 96 { 97 VirtIONet *n = VIRTIO_NET(vdev); 98 struct virtio_net_config netcfg = {}; 99 100 memcpy(&netcfg, config, n->config_size); 101 102 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) && 103 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) && 104 memcmp(netcfg.mac, n->mac, ETH_ALEN)) { 105 memcpy(n->mac, netcfg.mac, ETH_ALEN); 106 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 107 } 108 } 109 110 static bool virtio_net_started(VirtIONet *n, uint8_t status) 111 { 112 VirtIODevice *vdev = VIRTIO_DEVICE(n); 113 return (status & VIRTIO_CONFIG_S_DRIVER_OK) && 114 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running; 115 } 116 117 static void virtio_net_announce_timer(void *opaque) 118 { 119 VirtIONet *n = opaque; 120 VirtIODevice *vdev = VIRTIO_DEVICE(n); 121 122 n->announce_counter--; 123 n->status |= VIRTIO_NET_S_ANNOUNCE; 124 virtio_notify_config(vdev); 125 } 126 127 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status) 128 { 129 VirtIODevice *vdev = VIRTIO_DEVICE(n); 130 NetClientState *nc = qemu_get_queue(n->nic); 131 int queues = n->multiqueue ? n->max_queues : 1; 132 133 if (!get_vhost_net(nc->peer)) { 134 return; 135 } 136 137 if ((virtio_net_started(n, status) && !nc->peer->link_down) == 138 !!n->vhost_started) { 139 return; 140 } 141 if (!n->vhost_started) { 142 int r, i; 143 144 if (n->needs_vnet_hdr_swap) { 145 error_report("backend does not support %s vnet headers; " 146 "falling back on userspace virtio", 147 virtio_is_big_endian(vdev) ? "BE" : "LE"); 148 return; 149 } 150 151 /* Any packets outstanding? Purge them to avoid touching rings 152 * when vhost is running. 153 */ 154 for (i = 0; i < queues; i++) { 155 NetClientState *qnc = qemu_get_subqueue(n->nic, i); 156 157 /* Purge both directions: TX and RX. */ 158 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc); 159 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer); 160 } 161 162 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) { 163 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu); 164 if (r < 0) { 165 error_report("%uBytes MTU not supported by the backend", 166 n->net_conf.mtu); 167 168 return; 169 } 170 } 171 172 n->vhost_started = 1; 173 r = vhost_net_start(vdev, n->nic->ncs, queues); 174 if (r < 0) { 175 error_report("unable to start vhost net: %d: " 176 "falling back on userspace virtio", -r); 177 n->vhost_started = 0; 178 } 179 } else { 180 vhost_net_stop(vdev, n->nic->ncs, queues); 181 n->vhost_started = 0; 182 } 183 } 184 185 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev, 186 NetClientState *peer, 187 bool enable) 188 { 189 if (virtio_is_big_endian(vdev)) { 190 return qemu_set_vnet_be(peer, enable); 191 } else { 192 return qemu_set_vnet_le(peer, enable); 193 } 194 } 195 196 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs, 197 int queues, bool enable) 198 { 199 int i; 200 201 for (i = 0; i < queues; i++) { 202 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 && 203 enable) { 204 while (--i >= 0) { 205 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false); 206 } 207 208 return true; 209 } 210 } 211 212 return false; 213 } 214 215 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status) 216 { 217 VirtIODevice *vdev = VIRTIO_DEVICE(n); 218 int queues = n->multiqueue ? n->max_queues : 1; 219 220 if (virtio_net_started(n, status)) { 221 /* Before using the device, we tell the network backend about the 222 * endianness to use when parsing vnet headers. If the backend 223 * can't do it, we fallback onto fixing the headers in the core 224 * virtio-net code. 225 */ 226 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs, 227 queues, true); 228 } else if (virtio_net_started(n, vdev->status)) { 229 /* After using the device, we need to reset the network backend to 230 * the default (guest native endianness), otherwise the guest may 231 * lose network connectivity if it is rebooted into a different 232 * endianness. 233 */ 234 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false); 235 } 236 } 237 238 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq) 239 { 240 unsigned int dropped = virtqueue_drop_all(vq); 241 if (dropped) { 242 virtio_notify(vdev, vq); 243 } 244 } 245 246 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status) 247 { 248 VirtIONet *n = VIRTIO_NET(vdev); 249 VirtIONetQueue *q; 250 int i; 251 uint8_t queue_status; 252 253 virtio_net_vnet_endian_status(n, status); 254 virtio_net_vhost_status(n, status); 255 256 for (i = 0; i < n->max_queues; i++) { 257 NetClientState *ncs = qemu_get_subqueue(n->nic, i); 258 bool queue_started; 259 q = &n->vqs[i]; 260 261 if ((!n->multiqueue && i != 0) || i >= n->curr_queues) { 262 queue_status = 0; 263 } else { 264 queue_status = status; 265 } 266 queue_started = 267 virtio_net_started(n, queue_status) && !n->vhost_started; 268 269 if (queue_started) { 270 qemu_flush_queued_packets(ncs); 271 } 272 273 if (!q->tx_waiting) { 274 continue; 275 } 276 277 if (queue_started) { 278 if (q->tx_timer) { 279 timer_mod(q->tx_timer, 280 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 281 } else { 282 qemu_bh_schedule(q->tx_bh); 283 } 284 } else { 285 if (q->tx_timer) { 286 timer_del(q->tx_timer); 287 } else { 288 qemu_bh_cancel(q->tx_bh); 289 } 290 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 && 291 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK)) { 292 /* if tx is waiting we are likely have some packets in tx queue 293 * and disabled notification */ 294 q->tx_waiting = 0; 295 virtio_queue_set_notification(q->tx_vq, 1); 296 virtio_net_drop_tx_queue_data(vdev, q->tx_vq); 297 } 298 } 299 } 300 } 301 302 static void virtio_net_set_link_status(NetClientState *nc) 303 { 304 VirtIONet *n = qemu_get_nic_opaque(nc); 305 VirtIODevice *vdev = VIRTIO_DEVICE(n); 306 uint16_t old_status = n->status; 307 308 if (nc->link_down) 309 n->status &= ~VIRTIO_NET_S_LINK_UP; 310 else 311 n->status |= VIRTIO_NET_S_LINK_UP; 312 313 if (n->status != old_status) 314 virtio_notify_config(vdev); 315 316 virtio_net_set_status(vdev, vdev->status); 317 } 318 319 static void rxfilter_notify(NetClientState *nc) 320 { 321 VirtIONet *n = qemu_get_nic_opaque(nc); 322 323 if (nc->rxfilter_notify_enabled) { 324 gchar *path = object_get_canonical_path(OBJECT(n->qdev)); 325 qapi_event_send_nic_rx_filter_changed(!!n->netclient_name, 326 n->netclient_name, path, &error_abort); 327 g_free(path); 328 329 /* disable event notification to avoid events flooding */ 330 nc->rxfilter_notify_enabled = 0; 331 } 332 } 333 334 static intList *get_vlan_table(VirtIONet *n) 335 { 336 intList *list, *entry; 337 int i, j; 338 339 list = NULL; 340 for (i = 0; i < MAX_VLAN >> 5; i++) { 341 for (j = 0; n->vlans[i] && j <= 0x1f; j++) { 342 if (n->vlans[i] & (1U << j)) { 343 entry = g_malloc0(sizeof(*entry)); 344 entry->value = (i << 5) + j; 345 entry->next = list; 346 list = entry; 347 } 348 } 349 } 350 351 return list; 352 } 353 354 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc) 355 { 356 VirtIONet *n = qemu_get_nic_opaque(nc); 357 VirtIODevice *vdev = VIRTIO_DEVICE(n); 358 RxFilterInfo *info; 359 strList *str_list, *entry; 360 int i; 361 362 info = g_malloc0(sizeof(*info)); 363 info->name = g_strdup(nc->name); 364 info->promiscuous = n->promisc; 365 366 if (n->nouni) { 367 info->unicast = RX_STATE_NONE; 368 } else if (n->alluni) { 369 info->unicast = RX_STATE_ALL; 370 } else { 371 info->unicast = RX_STATE_NORMAL; 372 } 373 374 if (n->nomulti) { 375 info->multicast = RX_STATE_NONE; 376 } else if (n->allmulti) { 377 info->multicast = RX_STATE_ALL; 378 } else { 379 info->multicast = RX_STATE_NORMAL; 380 } 381 382 info->broadcast_allowed = n->nobcast; 383 info->multicast_overflow = n->mac_table.multi_overflow; 384 info->unicast_overflow = n->mac_table.uni_overflow; 385 386 info->main_mac = qemu_mac_strdup_printf(n->mac); 387 388 str_list = NULL; 389 for (i = 0; i < n->mac_table.first_multi; i++) { 390 entry = g_malloc0(sizeof(*entry)); 391 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN); 392 entry->next = str_list; 393 str_list = entry; 394 } 395 info->unicast_table = str_list; 396 397 str_list = NULL; 398 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 399 entry = g_malloc0(sizeof(*entry)); 400 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN); 401 entry->next = str_list; 402 str_list = entry; 403 } 404 info->multicast_table = str_list; 405 info->vlan_table = get_vlan_table(n); 406 407 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) { 408 info->vlan = RX_STATE_ALL; 409 } else if (!info->vlan_table) { 410 info->vlan = RX_STATE_NONE; 411 } else { 412 info->vlan = RX_STATE_NORMAL; 413 } 414 415 /* enable event notification after query */ 416 nc->rxfilter_notify_enabled = 1; 417 418 return info; 419 } 420 421 static void virtio_net_reset(VirtIODevice *vdev) 422 { 423 VirtIONet *n = VIRTIO_NET(vdev); 424 425 /* Reset back to compatibility mode */ 426 n->promisc = 1; 427 n->allmulti = 0; 428 n->alluni = 0; 429 n->nomulti = 0; 430 n->nouni = 0; 431 n->nobcast = 0; 432 /* multiqueue is disabled by default */ 433 n->curr_queues = 1; 434 timer_del(n->announce_timer); 435 n->announce_counter = 0; 436 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 437 438 /* Flush any MAC and VLAN filter table state */ 439 n->mac_table.in_use = 0; 440 n->mac_table.first_multi = 0; 441 n->mac_table.multi_overflow = 0; 442 n->mac_table.uni_overflow = 0; 443 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN); 444 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac)); 445 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 446 memset(n->vlans, 0, MAX_VLAN >> 3); 447 } 448 449 static void peer_test_vnet_hdr(VirtIONet *n) 450 { 451 NetClientState *nc = qemu_get_queue(n->nic); 452 if (!nc->peer) { 453 return; 454 } 455 456 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer); 457 } 458 459 static int peer_has_vnet_hdr(VirtIONet *n) 460 { 461 return n->has_vnet_hdr; 462 } 463 464 static int peer_has_ufo(VirtIONet *n) 465 { 466 if (!peer_has_vnet_hdr(n)) 467 return 0; 468 469 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer); 470 471 return n->has_ufo; 472 } 473 474 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs, 475 int version_1) 476 { 477 int i; 478 NetClientState *nc; 479 480 n->mergeable_rx_bufs = mergeable_rx_bufs; 481 482 if (version_1) { 483 n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); 484 } else { 485 n->guest_hdr_len = n->mergeable_rx_bufs ? 486 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 487 sizeof(struct virtio_net_hdr); 488 } 489 490 for (i = 0; i < n->max_queues; i++) { 491 nc = qemu_get_subqueue(n->nic, i); 492 493 if (peer_has_vnet_hdr(n) && 494 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) { 495 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len); 496 n->host_hdr_len = n->guest_hdr_len; 497 } 498 } 499 } 500 501 static int virtio_net_max_tx_queue_size(VirtIONet *n) 502 { 503 NetClientState *peer = n->nic_conf.peers.ncs[0]; 504 505 /* 506 * Backends other than vhost-user don't support max queue size. 507 */ 508 if (!peer) { 509 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 510 } 511 512 if (peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) { 513 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 514 } 515 516 return VIRTQUEUE_MAX_SIZE; 517 } 518 519 static int peer_attach(VirtIONet *n, int index) 520 { 521 NetClientState *nc = qemu_get_subqueue(n->nic, index); 522 523 if (!nc->peer) { 524 return 0; 525 } 526 527 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 528 vhost_set_vring_enable(nc->peer, 1); 529 } 530 531 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 532 return 0; 533 } 534 535 if (n->max_queues == 1) { 536 return 0; 537 } 538 539 return tap_enable(nc->peer); 540 } 541 542 static int peer_detach(VirtIONet *n, int index) 543 { 544 NetClientState *nc = qemu_get_subqueue(n->nic, index); 545 546 if (!nc->peer) { 547 return 0; 548 } 549 550 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 551 vhost_set_vring_enable(nc->peer, 0); 552 } 553 554 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 555 return 0; 556 } 557 558 return tap_disable(nc->peer); 559 } 560 561 static void virtio_net_set_queues(VirtIONet *n) 562 { 563 int i; 564 int r; 565 566 if (n->nic->peer_deleted) { 567 return; 568 } 569 570 for (i = 0; i < n->max_queues; i++) { 571 if (i < n->curr_queues) { 572 r = peer_attach(n, i); 573 assert(!r); 574 } else { 575 r = peer_detach(n, i); 576 assert(!r); 577 } 578 } 579 } 580 581 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); 582 583 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features, 584 Error **errp) 585 { 586 VirtIONet *n = VIRTIO_NET(vdev); 587 NetClientState *nc = qemu_get_queue(n->nic); 588 589 /* Firstly sync all virtio-net possible supported features */ 590 features |= n->host_features; 591 592 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 593 594 if (!peer_has_vnet_hdr(n)) { 595 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM); 596 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4); 597 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6); 598 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN); 599 600 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM); 601 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4); 602 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6); 603 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN); 604 } 605 606 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 607 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO); 608 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO); 609 } 610 611 if (!get_vhost_net(nc->peer)) { 612 return features; 613 } 614 features = vhost_net_get_features(get_vhost_net(nc->peer), features); 615 vdev->backend_features = features; 616 617 if (n->mtu_bypass_backend && 618 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) { 619 features |= (1ULL << VIRTIO_NET_F_MTU); 620 } 621 622 return features; 623 } 624 625 static uint64_t virtio_net_bad_features(VirtIODevice *vdev) 626 { 627 uint64_t features = 0; 628 629 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 630 * but also these: */ 631 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 632 virtio_add_feature(&features, VIRTIO_NET_F_CSUM); 633 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4); 634 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6); 635 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN); 636 637 return features; 638 } 639 640 static void virtio_net_apply_guest_offloads(VirtIONet *n) 641 { 642 qemu_set_offload(qemu_get_queue(n->nic)->peer, 643 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 644 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 645 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 646 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 647 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO))); 648 } 649 650 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features) 651 { 652 static const uint64_t guest_offloads_mask = 653 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 654 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 655 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 656 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 657 (1ULL << VIRTIO_NET_F_GUEST_UFO); 658 659 return guest_offloads_mask & features; 660 } 661 662 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n) 663 { 664 VirtIODevice *vdev = VIRTIO_DEVICE(n); 665 return virtio_net_guest_offloads_by_features(vdev->guest_features); 666 } 667 668 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features) 669 { 670 VirtIONet *n = VIRTIO_NET(vdev); 671 int i; 672 673 if (n->mtu_bypass_backend && 674 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) { 675 features &= ~(1ULL << VIRTIO_NET_F_MTU); 676 } 677 678 virtio_net_set_multiqueue(n, 679 virtio_has_feature(features, VIRTIO_NET_F_MQ)); 680 681 virtio_net_set_mrg_rx_bufs(n, 682 virtio_has_feature(features, 683 VIRTIO_NET_F_MRG_RXBUF), 684 virtio_has_feature(features, 685 VIRTIO_F_VERSION_1)); 686 687 if (n->has_vnet_hdr) { 688 n->curr_guest_offloads = 689 virtio_net_guest_offloads_by_features(features); 690 virtio_net_apply_guest_offloads(n); 691 } 692 693 for (i = 0; i < n->max_queues; i++) { 694 NetClientState *nc = qemu_get_subqueue(n->nic, i); 695 696 if (!get_vhost_net(nc->peer)) { 697 continue; 698 } 699 vhost_net_ack_features(get_vhost_net(nc->peer), features); 700 } 701 702 if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) { 703 memset(n->vlans, 0, MAX_VLAN >> 3); 704 } else { 705 memset(n->vlans, 0xff, MAX_VLAN >> 3); 706 } 707 } 708 709 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 710 struct iovec *iov, unsigned int iov_cnt) 711 { 712 uint8_t on; 713 size_t s; 714 NetClientState *nc = qemu_get_queue(n->nic); 715 716 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 717 if (s != sizeof(on)) { 718 return VIRTIO_NET_ERR; 719 } 720 721 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 722 n->promisc = on; 723 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 724 n->allmulti = on; 725 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 726 n->alluni = on; 727 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 728 n->nomulti = on; 729 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 730 n->nouni = on; 731 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 732 n->nobcast = on; 733 } else { 734 return VIRTIO_NET_ERR; 735 } 736 737 rxfilter_notify(nc); 738 739 return VIRTIO_NET_OK; 740 } 741 742 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 743 struct iovec *iov, unsigned int iov_cnt) 744 { 745 VirtIODevice *vdev = VIRTIO_DEVICE(n); 746 uint64_t offloads; 747 size_t s; 748 749 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 750 return VIRTIO_NET_ERR; 751 } 752 753 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 754 if (s != sizeof(offloads)) { 755 return VIRTIO_NET_ERR; 756 } 757 758 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 759 uint64_t supported_offloads; 760 761 if (!n->has_vnet_hdr) { 762 return VIRTIO_NET_ERR; 763 } 764 765 supported_offloads = virtio_net_supported_guest_offloads(n); 766 if (offloads & ~supported_offloads) { 767 return VIRTIO_NET_ERR; 768 } 769 770 n->curr_guest_offloads = offloads; 771 virtio_net_apply_guest_offloads(n); 772 773 return VIRTIO_NET_OK; 774 } else { 775 return VIRTIO_NET_ERR; 776 } 777 } 778 779 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 780 struct iovec *iov, unsigned int iov_cnt) 781 { 782 VirtIODevice *vdev = VIRTIO_DEVICE(n); 783 struct virtio_net_ctrl_mac mac_data; 784 size_t s; 785 NetClientState *nc = qemu_get_queue(n->nic); 786 787 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 788 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 789 return VIRTIO_NET_ERR; 790 } 791 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 792 assert(s == sizeof(n->mac)); 793 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 794 rxfilter_notify(nc); 795 796 return VIRTIO_NET_OK; 797 } 798 799 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 800 return VIRTIO_NET_ERR; 801 } 802 803 int in_use = 0; 804 int first_multi = 0; 805 uint8_t uni_overflow = 0; 806 uint8_t multi_overflow = 0; 807 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 808 809 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 810 sizeof(mac_data.entries)); 811 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 812 if (s != sizeof(mac_data.entries)) { 813 goto error; 814 } 815 iov_discard_front(&iov, &iov_cnt, s); 816 817 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 818 goto error; 819 } 820 821 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 822 s = iov_to_buf(iov, iov_cnt, 0, macs, 823 mac_data.entries * ETH_ALEN); 824 if (s != mac_data.entries * ETH_ALEN) { 825 goto error; 826 } 827 in_use += mac_data.entries; 828 } else { 829 uni_overflow = 1; 830 } 831 832 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 833 834 first_multi = in_use; 835 836 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 837 sizeof(mac_data.entries)); 838 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 839 if (s != sizeof(mac_data.entries)) { 840 goto error; 841 } 842 843 iov_discard_front(&iov, &iov_cnt, s); 844 845 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 846 goto error; 847 } 848 849 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) { 850 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN], 851 mac_data.entries * ETH_ALEN); 852 if (s != mac_data.entries * ETH_ALEN) { 853 goto error; 854 } 855 in_use += mac_data.entries; 856 } else { 857 multi_overflow = 1; 858 } 859 860 n->mac_table.in_use = in_use; 861 n->mac_table.first_multi = first_multi; 862 n->mac_table.uni_overflow = uni_overflow; 863 n->mac_table.multi_overflow = multi_overflow; 864 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN); 865 g_free(macs); 866 rxfilter_notify(nc); 867 868 return VIRTIO_NET_OK; 869 870 error: 871 g_free(macs); 872 return VIRTIO_NET_ERR; 873 } 874 875 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 876 struct iovec *iov, unsigned int iov_cnt) 877 { 878 VirtIODevice *vdev = VIRTIO_DEVICE(n); 879 uint16_t vid; 880 size_t s; 881 NetClientState *nc = qemu_get_queue(n->nic); 882 883 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 884 vid = virtio_lduw_p(vdev, &vid); 885 if (s != sizeof(vid)) { 886 return VIRTIO_NET_ERR; 887 } 888 889 if (vid >= MAX_VLAN) 890 return VIRTIO_NET_ERR; 891 892 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 893 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 894 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 895 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 896 else 897 return VIRTIO_NET_ERR; 898 899 rxfilter_notify(nc); 900 901 return VIRTIO_NET_OK; 902 } 903 904 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd, 905 struct iovec *iov, unsigned int iov_cnt) 906 { 907 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK && 908 n->status & VIRTIO_NET_S_ANNOUNCE) { 909 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 910 if (n->announce_counter) { 911 timer_mod(n->announce_timer, 912 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 913 self_announce_delay(n->announce_counter)); 914 } 915 return VIRTIO_NET_OK; 916 } else { 917 return VIRTIO_NET_ERR; 918 } 919 } 920 921 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 922 struct iovec *iov, unsigned int iov_cnt) 923 { 924 VirtIODevice *vdev = VIRTIO_DEVICE(n); 925 struct virtio_net_ctrl_mq mq; 926 size_t s; 927 uint16_t queues; 928 929 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 930 if (s != sizeof(mq)) { 931 return VIRTIO_NET_ERR; 932 } 933 934 if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 935 return VIRTIO_NET_ERR; 936 } 937 938 queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs); 939 940 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 941 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 942 queues > n->max_queues || 943 !n->multiqueue) { 944 return VIRTIO_NET_ERR; 945 } 946 947 n->curr_queues = queues; 948 /* stop the backend before changing the number of queues to avoid handling a 949 * disabled queue */ 950 virtio_net_set_status(vdev, vdev->status); 951 virtio_net_set_queues(n); 952 953 return VIRTIO_NET_OK; 954 } 955 956 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 957 { 958 VirtIONet *n = VIRTIO_NET(vdev); 959 struct virtio_net_ctrl_hdr ctrl; 960 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 961 VirtQueueElement *elem; 962 size_t s; 963 struct iovec *iov, *iov2; 964 unsigned int iov_cnt; 965 966 for (;;) { 967 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 968 if (!elem) { 969 break; 970 } 971 if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) || 972 iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) { 973 virtio_error(vdev, "virtio-net ctrl missing headers"); 974 virtqueue_detach_element(vq, elem, 0); 975 g_free(elem); 976 break; 977 } 978 979 iov_cnt = elem->out_num; 980 iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num); 981 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl)); 982 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl)); 983 if (s != sizeof(ctrl)) { 984 status = VIRTIO_NET_ERR; 985 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 986 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt); 987 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 988 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt); 989 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 990 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt); 991 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) { 992 status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt); 993 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 994 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt); 995 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 996 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt); 997 } 998 999 s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status)); 1000 assert(s == sizeof(status)); 1001 1002 virtqueue_push(vq, elem, sizeof(status)); 1003 virtio_notify(vdev, vq); 1004 g_free(iov2); 1005 g_free(elem); 1006 } 1007 } 1008 1009 /* RX */ 1010 1011 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 1012 { 1013 VirtIONet *n = VIRTIO_NET(vdev); 1014 int queue_index = vq2q(virtio_get_queue_index(vq)); 1015 1016 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 1017 } 1018 1019 static int virtio_net_can_receive(NetClientState *nc) 1020 { 1021 VirtIONet *n = qemu_get_nic_opaque(nc); 1022 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1023 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1024 1025 if (!vdev->vm_running) { 1026 return 0; 1027 } 1028 1029 if (nc->queue_index >= n->curr_queues) { 1030 return 0; 1031 } 1032 1033 if (!virtio_queue_ready(q->rx_vq) || 1034 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1035 return 0; 1036 } 1037 1038 return 1; 1039 } 1040 1041 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 1042 { 1043 VirtIONet *n = q->n; 1044 if (virtio_queue_empty(q->rx_vq) || 1045 (n->mergeable_rx_bufs && 1046 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 1047 virtio_queue_set_notification(q->rx_vq, 1); 1048 1049 /* To avoid a race condition where the guest has made some buffers 1050 * available after the above check but before notification was 1051 * enabled, check for available buffers again. 1052 */ 1053 if (virtio_queue_empty(q->rx_vq) || 1054 (n->mergeable_rx_bufs && 1055 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 1056 return 0; 1057 } 1058 } 1059 1060 virtio_queue_set_notification(q->rx_vq, 0); 1061 return 1; 1062 } 1063 1064 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr) 1065 { 1066 virtio_tswap16s(vdev, &hdr->hdr_len); 1067 virtio_tswap16s(vdev, &hdr->gso_size); 1068 virtio_tswap16s(vdev, &hdr->csum_start); 1069 virtio_tswap16s(vdev, &hdr->csum_offset); 1070 } 1071 1072 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 1073 * it never finds out that the packets don't have valid checksums. This 1074 * causes dhclient to get upset. Fedora's carried a patch for ages to 1075 * fix this with Xen but it hasn't appeared in an upstream release of 1076 * dhclient yet. 1077 * 1078 * To avoid breaking existing guests, we catch udp packets and add 1079 * checksums. This is terrible but it's better than hacking the guest 1080 * kernels. 1081 * 1082 * N.B. if we introduce a zero-copy API, this operation is no longer free so 1083 * we should provide a mechanism to disable it to avoid polluting the host 1084 * cache. 1085 */ 1086 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 1087 uint8_t *buf, size_t size) 1088 { 1089 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 1090 (size > 27 && size < 1500) && /* normal sized MTU */ 1091 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 1092 (buf[23] == 17) && /* ip.protocol == UDP */ 1093 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 1094 net_checksum_calculate(buf, size); 1095 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 1096 } 1097 } 1098 1099 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt, 1100 const void *buf, size_t size) 1101 { 1102 if (n->has_vnet_hdr) { 1103 /* FIXME this cast is evil */ 1104 void *wbuf = (void *)buf; 1105 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, 1106 size - n->host_hdr_len); 1107 1108 if (n->needs_vnet_hdr_swap) { 1109 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf); 1110 } 1111 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr)); 1112 } else { 1113 struct virtio_net_hdr hdr = { 1114 .flags = 0, 1115 .gso_type = VIRTIO_NET_HDR_GSO_NONE 1116 }; 1117 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr); 1118 } 1119 } 1120 1121 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 1122 { 1123 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 1124 static const uint8_t vlan[] = {0x81, 0x00}; 1125 uint8_t *ptr = (uint8_t *)buf; 1126 int i; 1127 1128 if (n->promisc) 1129 return 1; 1130 1131 ptr += n->host_hdr_len; 1132 1133 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 1134 int vid = lduw_be_p(ptr + 14) & 0xfff; 1135 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 1136 return 0; 1137 } 1138 1139 if (ptr[0] & 1) { // multicast 1140 if (!memcmp(ptr, bcast, sizeof(bcast))) { 1141 return !n->nobcast; 1142 } else if (n->nomulti) { 1143 return 0; 1144 } else if (n->allmulti || n->mac_table.multi_overflow) { 1145 return 1; 1146 } 1147 1148 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 1149 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1150 return 1; 1151 } 1152 } 1153 } else { // unicast 1154 if (n->nouni) { 1155 return 0; 1156 } else if (n->alluni || n->mac_table.uni_overflow) { 1157 return 1; 1158 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 1159 return 1; 1160 } 1161 1162 for (i = 0; i < n->mac_table.first_multi; i++) { 1163 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1164 return 1; 1165 } 1166 } 1167 } 1168 1169 return 0; 1170 } 1171 1172 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, 1173 size_t size) 1174 { 1175 VirtIONet *n = qemu_get_nic_opaque(nc); 1176 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1177 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1178 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 1179 struct virtio_net_hdr_mrg_rxbuf mhdr; 1180 unsigned mhdr_cnt = 0; 1181 size_t offset, i, guest_offset; 1182 1183 if (!virtio_net_can_receive(nc)) { 1184 return -1; 1185 } 1186 1187 /* hdr_len refers to the header we supply to the guest */ 1188 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 1189 return 0; 1190 } 1191 1192 if (!receive_filter(n, buf, size)) 1193 return size; 1194 1195 offset = i = 0; 1196 1197 while (offset < size) { 1198 VirtQueueElement *elem; 1199 int len, total; 1200 const struct iovec *sg; 1201 1202 total = 0; 1203 1204 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement)); 1205 if (!elem) { 1206 if (i) { 1207 virtio_error(vdev, "virtio-net unexpected empty queue: " 1208 "i %zd mergeable %d offset %zd, size %zd, " 1209 "guest hdr len %zd, host hdr len %zd " 1210 "guest features 0x%" PRIx64, 1211 i, n->mergeable_rx_bufs, offset, size, 1212 n->guest_hdr_len, n->host_hdr_len, 1213 vdev->guest_features); 1214 } 1215 return -1; 1216 } 1217 1218 if (elem->in_num < 1) { 1219 virtio_error(vdev, 1220 "virtio-net receive queue contains no in buffers"); 1221 virtqueue_detach_element(q->rx_vq, elem, 0); 1222 g_free(elem); 1223 return -1; 1224 } 1225 1226 sg = elem->in_sg; 1227 if (i == 0) { 1228 assert(offset == 0); 1229 if (n->mergeable_rx_bufs) { 1230 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 1231 sg, elem->in_num, 1232 offsetof(typeof(mhdr), num_buffers), 1233 sizeof(mhdr.num_buffers)); 1234 } 1235 1236 receive_header(n, sg, elem->in_num, buf, size); 1237 offset = n->host_hdr_len; 1238 total += n->guest_hdr_len; 1239 guest_offset = n->guest_hdr_len; 1240 } else { 1241 guest_offset = 0; 1242 } 1243 1244 /* copy in packet. ugh */ 1245 len = iov_from_buf(sg, elem->in_num, guest_offset, 1246 buf + offset, size - offset); 1247 total += len; 1248 offset += len; 1249 /* If buffers can't be merged, at this point we 1250 * must have consumed the complete packet. 1251 * Otherwise, drop it. */ 1252 if (!n->mergeable_rx_bufs && offset < size) { 1253 virtqueue_unpop(q->rx_vq, elem, total); 1254 g_free(elem); 1255 return size; 1256 } 1257 1258 /* signal other side */ 1259 virtqueue_fill(q->rx_vq, elem, total, i++); 1260 g_free(elem); 1261 } 1262 1263 if (mhdr_cnt) { 1264 virtio_stw_p(vdev, &mhdr.num_buffers, i); 1265 iov_from_buf(mhdr_sg, mhdr_cnt, 1266 0, 1267 &mhdr.num_buffers, sizeof mhdr.num_buffers); 1268 } 1269 1270 virtqueue_flush(q->rx_vq, i); 1271 virtio_notify(vdev, q->rx_vq); 1272 1273 return size; 1274 } 1275 1276 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, 1277 size_t size) 1278 { 1279 ssize_t r; 1280 1281 rcu_read_lock(); 1282 r = virtio_net_receive_rcu(nc, buf, size); 1283 rcu_read_unlock(); 1284 return r; 1285 } 1286 1287 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 1288 1289 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 1290 { 1291 VirtIONet *n = qemu_get_nic_opaque(nc); 1292 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1293 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1294 1295 virtqueue_push(q->tx_vq, q->async_tx.elem, 0); 1296 virtio_notify(vdev, q->tx_vq); 1297 1298 g_free(q->async_tx.elem); 1299 q->async_tx.elem = NULL; 1300 1301 virtio_queue_set_notification(q->tx_vq, 1); 1302 virtio_net_flush_tx(q); 1303 } 1304 1305 /* TX */ 1306 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 1307 { 1308 VirtIONet *n = q->n; 1309 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1310 VirtQueueElement *elem; 1311 int32_t num_packets = 0; 1312 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 1313 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1314 return num_packets; 1315 } 1316 1317 if (q->async_tx.elem) { 1318 virtio_queue_set_notification(q->tx_vq, 0); 1319 return num_packets; 1320 } 1321 1322 for (;;) { 1323 ssize_t ret; 1324 unsigned int out_num; 1325 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg; 1326 struct virtio_net_hdr_mrg_rxbuf mhdr; 1327 1328 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement)); 1329 if (!elem) { 1330 break; 1331 } 1332 1333 out_num = elem->out_num; 1334 out_sg = elem->out_sg; 1335 if (out_num < 1) { 1336 virtio_error(vdev, "virtio-net header not in first element"); 1337 virtqueue_detach_element(q->tx_vq, elem, 0); 1338 g_free(elem); 1339 return -EINVAL; 1340 } 1341 1342 if (n->has_vnet_hdr) { 1343 if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) < 1344 n->guest_hdr_len) { 1345 virtio_error(vdev, "virtio-net header incorrect"); 1346 virtqueue_detach_element(q->tx_vq, elem, 0); 1347 g_free(elem); 1348 return -EINVAL; 1349 } 1350 if (n->needs_vnet_hdr_swap) { 1351 virtio_net_hdr_swap(vdev, (void *) &mhdr); 1352 sg2[0].iov_base = &mhdr; 1353 sg2[0].iov_len = n->guest_hdr_len; 1354 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1, 1355 out_sg, out_num, 1356 n->guest_hdr_len, -1); 1357 if (out_num == VIRTQUEUE_MAX_SIZE) { 1358 goto drop; 1359 } 1360 out_num += 1; 1361 out_sg = sg2; 1362 } 1363 } 1364 /* 1365 * If host wants to see the guest header as is, we can 1366 * pass it on unchanged. Otherwise, copy just the parts 1367 * that host is interested in. 1368 */ 1369 assert(n->host_hdr_len <= n->guest_hdr_len); 1370 if (n->host_hdr_len != n->guest_hdr_len) { 1371 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 1372 out_sg, out_num, 1373 0, n->host_hdr_len); 1374 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 1375 out_sg, out_num, 1376 n->guest_hdr_len, -1); 1377 out_num = sg_num; 1378 out_sg = sg; 1379 } 1380 1381 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 1382 out_sg, out_num, virtio_net_tx_complete); 1383 if (ret == 0) { 1384 virtio_queue_set_notification(q->tx_vq, 0); 1385 q->async_tx.elem = elem; 1386 return -EBUSY; 1387 } 1388 1389 drop: 1390 virtqueue_push(q->tx_vq, elem, 0); 1391 virtio_notify(vdev, q->tx_vq); 1392 g_free(elem); 1393 1394 if (++num_packets >= n->tx_burst) { 1395 break; 1396 } 1397 } 1398 return num_packets; 1399 } 1400 1401 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 1402 { 1403 VirtIONet *n = VIRTIO_NET(vdev); 1404 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1405 1406 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 1407 virtio_net_drop_tx_queue_data(vdev, vq); 1408 return; 1409 } 1410 1411 /* This happens when device was stopped but VCPU wasn't. */ 1412 if (!vdev->vm_running) { 1413 q->tx_waiting = 1; 1414 return; 1415 } 1416 1417 if (q->tx_waiting) { 1418 virtio_queue_set_notification(vq, 1); 1419 timer_del(q->tx_timer); 1420 q->tx_waiting = 0; 1421 if (virtio_net_flush_tx(q) == -EINVAL) { 1422 return; 1423 } 1424 } else { 1425 timer_mod(q->tx_timer, 1426 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 1427 q->tx_waiting = 1; 1428 virtio_queue_set_notification(vq, 0); 1429 } 1430 } 1431 1432 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 1433 { 1434 VirtIONet *n = VIRTIO_NET(vdev); 1435 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1436 1437 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 1438 virtio_net_drop_tx_queue_data(vdev, vq); 1439 return; 1440 } 1441 1442 if (unlikely(q->tx_waiting)) { 1443 return; 1444 } 1445 q->tx_waiting = 1; 1446 /* This happens when device was stopped but VCPU wasn't. */ 1447 if (!vdev->vm_running) { 1448 return; 1449 } 1450 virtio_queue_set_notification(vq, 0); 1451 qemu_bh_schedule(q->tx_bh); 1452 } 1453 1454 static void virtio_net_tx_timer(void *opaque) 1455 { 1456 VirtIONetQueue *q = opaque; 1457 VirtIONet *n = q->n; 1458 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1459 /* This happens when device was stopped but BH wasn't. */ 1460 if (!vdev->vm_running) { 1461 /* Make sure tx waiting is set, so we'll run when restarted. */ 1462 assert(q->tx_waiting); 1463 return; 1464 } 1465 1466 q->tx_waiting = 0; 1467 1468 /* Just in case the driver is not ready on more */ 1469 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1470 return; 1471 } 1472 1473 virtio_queue_set_notification(q->tx_vq, 1); 1474 virtio_net_flush_tx(q); 1475 } 1476 1477 static void virtio_net_tx_bh(void *opaque) 1478 { 1479 VirtIONetQueue *q = opaque; 1480 VirtIONet *n = q->n; 1481 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1482 int32_t ret; 1483 1484 /* This happens when device was stopped but BH wasn't. */ 1485 if (!vdev->vm_running) { 1486 /* Make sure tx waiting is set, so we'll run when restarted. */ 1487 assert(q->tx_waiting); 1488 return; 1489 } 1490 1491 q->tx_waiting = 0; 1492 1493 /* Just in case the driver is not ready on more */ 1494 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 1495 return; 1496 } 1497 1498 ret = virtio_net_flush_tx(q); 1499 if (ret == -EBUSY || ret == -EINVAL) { 1500 return; /* Notification re-enable handled by tx_complete or device 1501 * broken */ 1502 } 1503 1504 /* If we flush a full burst of packets, assume there are 1505 * more coming and immediately reschedule */ 1506 if (ret >= n->tx_burst) { 1507 qemu_bh_schedule(q->tx_bh); 1508 q->tx_waiting = 1; 1509 return; 1510 } 1511 1512 /* If less than a full burst, re-enable notification and flush 1513 * anything that may have come in while we weren't looking. If 1514 * we find something, assume the guest is still active and reschedule */ 1515 virtio_queue_set_notification(q->tx_vq, 1); 1516 ret = virtio_net_flush_tx(q); 1517 if (ret == -EINVAL) { 1518 return; 1519 } else if (ret > 0) { 1520 virtio_queue_set_notification(q->tx_vq, 0); 1521 qemu_bh_schedule(q->tx_bh); 1522 q->tx_waiting = 1; 1523 } 1524 } 1525 1526 static void virtio_net_add_queue(VirtIONet *n, int index) 1527 { 1528 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1529 1530 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size, 1531 virtio_net_handle_rx); 1532 1533 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 1534 n->vqs[index].tx_vq = 1535 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 1536 virtio_net_handle_tx_timer); 1537 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 1538 virtio_net_tx_timer, 1539 &n->vqs[index]); 1540 } else { 1541 n->vqs[index].tx_vq = 1542 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 1543 virtio_net_handle_tx_bh); 1544 n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]); 1545 } 1546 1547 n->vqs[index].tx_waiting = 0; 1548 n->vqs[index].n = n; 1549 } 1550 1551 static void virtio_net_del_queue(VirtIONet *n, int index) 1552 { 1553 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1554 VirtIONetQueue *q = &n->vqs[index]; 1555 NetClientState *nc = qemu_get_subqueue(n->nic, index); 1556 1557 qemu_purge_queued_packets(nc); 1558 1559 virtio_del_queue(vdev, index * 2); 1560 if (q->tx_timer) { 1561 timer_del(q->tx_timer); 1562 timer_free(q->tx_timer); 1563 q->tx_timer = NULL; 1564 } else { 1565 qemu_bh_delete(q->tx_bh); 1566 q->tx_bh = NULL; 1567 } 1568 q->tx_waiting = 0; 1569 virtio_del_queue(vdev, index * 2 + 1); 1570 } 1571 1572 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues) 1573 { 1574 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1575 int old_num_queues = virtio_get_num_queues(vdev); 1576 int new_num_queues = new_max_queues * 2 + 1; 1577 int i; 1578 1579 assert(old_num_queues >= 3); 1580 assert(old_num_queues % 2 == 1); 1581 1582 if (old_num_queues == new_num_queues) { 1583 return; 1584 } 1585 1586 /* 1587 * We always need to remove and add ctrl vq if 1588 * old_num_queues != new_num_queues. Remove ctrl_vq first, 1589 * and then we only enter one of the following too loops. 1590 */ 1591 virtio_del_queue(vdev, old_num_queues - 1); 1592 1593 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) { 1594 /* new_num_queues < old_num_queues */ 1595 virtio_net_del_queue(n, i / 2); 1596 } 1597 1598 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) { 1599 /* new_num_queues > old_num_queues */ 1600 virtio_net_add_queue(n, i / 2); 1601 } 1602 1603 /* add ctrl_vq last */ 1604 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1605 } 1606 1607 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 1608 { 1609 int max = multiqueue ? n->max_queues : 1; 1610 1611 n->multiqueue = multiqueue; 1612 virtio_net_change_num_queues(n, max); 1613 1614 virtio_net_set_queues(n); 1615 } 1616 1617 static int virtio_net_post_load_device(void *opaque, int version_id) 1618 { 1619 VirtIONet *n = opaque; 1620 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1621 int i, link_down; 1622 1623 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs, 1624 virtio_vdev_has_feature(vdev, 1625 VIRTIO_F_VERSION_1)); 1626 1627 /* MAC_TABLE_ENTRIES may be different from the saved image */ 1628 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) { 1629 n->mac_table.in_use = 0; 1630 } 1631 1632 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 1633 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 1634 } 1635 1636 if (peer_has_vnet_hdr(n)) { 1637 virtio_net_apply_guest_offloads(n); 1638 } 1639 1640 virtio_net_set_queues(n); 1641 1642 /* Find the first multicast entry in the saved MAC filter */ 1643 for (i = 0; i < n->mac_table.in_use; i++) { 1644 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 1645 break; 1646 } 1647 } 1648 n->mac_table.first_multi = i; 1649 1650 /* nc.link_down can't be migrated, so infer link_down according 1651 * to link status bit in n->status */ 1652 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 1653 for (i = 0; i < n->max_queues; i++) { 1654 qemu_get_subqueue(n->nic, i)->link_down = link_down; 1655 } 1656 1657 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 1658 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 1659 n->announce_counter = SELF_ANNOUNCE_ROUNDS; 1660 timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL)); 1661 } 1662 1663 return 0; 1664 } 1665 1666 /* tx_waiting field of a VirtIONetQueue */ 1667 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = { 1668 .name = "virtio-net-queue-tx_waiting", 1669 .fields = (VMStateField[]) { 1670 VMSTATE_UINT32(tx_waiting, VirtIONetQueue), 1671 VMSTATE_END_OF_LIST() 1672 }, 1673 }; 1674 1675 static bool max_queues_gt_1(void *opaque, int version_id) 1676 { 1677 return VIRTIO_NET(opaque)->max_queues > 1; 1678 } 1679 1680 static bool has_ctrl_guest_offloads(void *opaque, int version_id) 1681 { 1682 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque), 1683 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS); 1684 } 1685 1686 static bool mac_table_fits(void *opaque, int version_id) 1687 { 1688 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES; 1689 } 1690 1691 static bool mac_table_doesnt_fit(void *opaque, int version_id) 1692 { 1693 return !mac_table_fits(opaque, version_id); 1694 } 1695 1696 /* This temporary type is shared by all the WITH_TMP methods 1697 * although only some fields are used by each. 1698 */ 1699 struct VirtIONetMigTmp { 1700 VirtIONet *parent; 1701 VirtIONetQueue *vqs_1; 1702 uint16_t curr_queues_1; 1703 uint8_t has_ufo; 1704 uint32_t has_vnet_hdr; 1705 }; 1706 1707 /* The 2nd and subsequent tx_waiting flags are loaded later than 1708 * the 1st entry in the queues and only if there's more than one 1709 * entry. We use the tmp mechanism to calculate a temporary 1710 * pointer and count and also validate the count. 1711 */ 1712 1713 static void virtio_net_tx_waiting_pre_save(void *opaque) 1714 { 1715 struct VirtIONetMigTmp *tmp = opaque; 1716 1717 tmp->vqs_1 = tmp->parent->vqs + 1; 1718 tmp->curr_queues_1 = tmp->parent->curr_queues - 1; 1719 if (tmp->parent->curr_queues == 0) { 1720 tmp->curr_queues_1 = 0; 1721 } 1722 } 1723 1724 static int virtio_net_tx_waiting_pre_load(void *opaque) 1725 { 1726 struct VirtIONetMigTmp *tmp = opaque; 1727 1728 /* Reuse the pointer setup from save */ 1729 virtio_net_tx_waiting_pre_save(opaque); 1730 1731 if (tmp->parent->curr_queues > tmp->parent->max_queues) { 1732 error_report("virtio-net: curr_queues %x > max_queues %x", 1733 tmp->parent->curr_queues, tmp->parent->max_queues); 1734 1735 return -EINVAL; 1736 } 1737 1738 return 0; /* all good */ 1739 } 1740 1741 static const VMStateDescription vmstate_virtio_net_tx_waiting = { 1742 .name = "virtio-net-tx_waiting", 1743 .pre_load = virtio_net_tx_waiting_pre_load, 1744 .pre_save = virtio_net_tx_waiting_pre_save, 1745 .fields = (VMStateField[]) { 1746 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp, 1747 curr_queues_1, 1748 vmstate_virtio_net_queue_tx_waiting, 1749 struct VirtIONetQueue), 1750 VMSTATE_END_OF_LIST() 1751 }, 1752 }; 1753 1754 /* the 'has_ufo' flag is just tested; if the incoming stream has the 1755 * flag set we need to check that we have it 1756 */ 1757 static int virtio_net_ufo_post_load(void *opaque, int version_id) 1758 { 1759 struct VirtIONetMigTmp *tmp = opaque; 1760 1761 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) { 1762 error_report("virtio-net: saved image requires TUN_F_UFO support"); 1763 return -EINVAL; 1764 } 1765 1766 return 0; 1767 } 1768 1769 static void virtio_net_ufo_pre_save(void *opaque) 1770 { 1771 struct VirtIONetMigTmp *tmp = opaque; 1772 1773 tmp->has_ufo = tmp->parent->has_ufo; 1774 } 1775 1776 static const VMStateDescription vmstate_virtio_net_has_ufo = { 1777 .name = "virtio-net-ufo", 1778 .post_load = virtio_net_ufo_post_load, 1779 .pre_save = virtio_net_ufo_pre_save, 1780 .fields = (VMStateField[]) { 1781 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp), 1782 VMSTATE_END_OF_LIST() 1783 }, 1784 }; 1785 1786 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the 1787 * flag set we need to check that we have it 1788 */ 1789 static int virtio_net_vnet_post_load(void *opaque, int version_id) 1790 { 1791 struct VirtIONetMigTmp *tmp = opaque; 1792 1793 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) { 1794 error_report("virtio-net: saved image requires vnet_hdr=on"); 1795 return -EINVAL; 1796 } 1797 1798 return 0; 1799 } 1800 1801 static void virtio_net_vnet_pre_save(void *opaque) 1802 { 1803 struct VirtIONetMigTmp *tmp = opaque; 1804 1805 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr; 1806 } 1807 1808 static const VMStateDescription vmstate_virtio_net_has_vnet = { 1809 .name = "virtio-net-vnet", 1810 .post_load = virtio_net_vnet_post_load, 1811 .pre_save = virtio_net_vnet_pre_save, 1812 .fields = (VMStateField[]) { 1813 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp), 1814 VMSTATE_END_OF_LIST() 1815 }, 1816 }; 1817 1818 static const VMStateDescription vmstate_virtio_net_device = { 1819 .name = "virtio-net-device", 1820 .version_id = VIRTIO_NET_VM_VERSION, 1821 .minimum_version_id = VIRTIO_NET_VM_VERSION, 1822 .post_load = virtio_net_post_load_device, 1823 .fields = (VMStateField[]) { 1824 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN), 1825 VMSTATE_STRUCT_POINTER(vqs, VirtIONet, 1826 vmstate_virtio_net_queue_tx_waiting, 1827 VirtIONetQueue), 1828 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet), 1829 VMSTATE_UINT16(status, VirtIONet), 1830 VMSTATE_UINT8(promisc, VirtIONet), 1831 VMSTATE_UINT8(allmulti, VirtIONet), 1832 VMSTATE_UINT32(mac_table.in_use, VirtIONet), 1833 1834 /* Guarded pair: If it fits we load it, else we throw it away 1835 * - can happen if source has a larger MAC table.; post-load 1836 * sets flags in this case. 1837 */ 1838 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet, 1839 0, mac_table_fits, mac_table.in_use, 1840 ETH_ALEN), 1841 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0, 1842 mac_table.in_use, ETH_ALEN), 1843 1844 /* Note: This is an array of uint32's that's always been saved as a 1845 * buffer; hold onto your endiannesses; it's actually used as a bitmap 1846 * but based on the uint. 1847 */ 1848 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3), 1849 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 1850 vmstate_virtio_net_has_vnet), 1851 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet), 1852 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet), 1853 VMSTATE_UINT8(alluni, VirtIONet), 1854 VMSTATE_UINT8(nomulti, VirtIONet), 1855 VMSTATE_UINT8(nouni, VirtIONet), 1856 VMSTATE_UINT8(nobcast, VirtIONet), 1857 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 1858 vmstate_virtio_net_has_ufo), 1859 VMSTATE_SINGLE_TEST(max_queues, VirtIONet, max_queues_gt_1, 0, 1860 vmstate_info_uint16_equal, uint16_t), 1861 VMSTATE_UINT16_TEST(curr_queues, VirtIONet, max_queues_gt_1), 1862 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 1863 vmstate_virtio_net_tx_waiting), 1864 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet, 1865 has_ctrl_guest_offloads), 1866 VMSTATE_END_OF_LIST() 1867 }, 1868 }; 1869 1870 static NetClientInfo net_virtio_info = { 1871 .type = NET_CLIENT_DRIVER_NIC, 1872 .size = sizeof(NICState), 1873 .can_receive = virtio_net_can_receive, 1874 .receive = virtio_net_receive, 1875 .link_status_changed = virtio_net_set_link_status, 1876 .query_rx_filter = virtio_net_query_rxfilter, 1877 }; 1878 1879 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 1880 { 1881 VirtIONet *n = VIRTIO_NET(vdev); 1882 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1883 assert(n->vhost_started); 1884 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx); 1885 } 1886 1887 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 1888 bool mask) 1889 { 1890 VirtIONet *n = VIRTIO_NET(vdev); 1891 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1892 assert(n->vhost_started); 1893 vhost_net_virtqueue_mask(get_vhost_net(nc->peer), 1894 vdev, idx, mask); 1895 } 1896 1897 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features) 1898 { 1899 int i, config_size = 0; 1900 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC); 1901 1902 for (i = 0; feature_sizes[i].flags != 0; i++) { 1903 if (host_features & feature_sizes[i].flags) { 1904 config_size = MAX(feature_sizes[i].end, config_size); 1905 } 1906 } 1907 n->config_size = config_size; 1908 } 1909 1910 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 1911 const char *type) 1912 { 1913 /* 1914 * The name can be NULL, the netclient name will be type.x. 1915 */ 1916 assert(type != NULL); 1917 1918 g_free(n->netclient_name); 1919 g_free(n->netclient_type); 1920 n->netclient_name = g_strdup(name); 1921 n->netclient_type = g_strdup(type); 1922 } 1923 1924 static void virtio_net_device_realize(DeviceState *dev, Error **errp) 1925 { 1926 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1927 VirtIONet *n = VIRTIO_NET(dev); 1928 NetClientState *nc; 1929 int i; 1930 1931 if (n->net_conf.mtu) { 1932 n->host_features |= (0x1 << VIRTIO_NET_F_MTU); 1933 } 1934 1935 virtio_net_set_config_size(n, n->host_features); 1936 virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size); 1937 1938 /* 1939 * We set a lower limit on RX queue size to what it always was. 1940 * Guests that want a smaller ring can always resize it without 1941 * help from us (using virtio 1 and up). 1942 */ 1943 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE || 1944 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE || 1945 (n->net_conf.rx_queue_size & (n->net_conf.rx_queue_size - 1))) { 1946 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), " 1947 "must be a power of 2 between %d and %d.", 1948 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE, 1949 VIRTQUEUE_MAX_SIZE); 1950 virtio_cleanup(vdev); 1951 return; 1952 } 1953 1954 if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE || 1955 n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE || 1956 !is_power_of_2(n->net_conf.tx_queue_size)) { 1957 error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), " 1958 "must be a power of 2 between %d and %d", 1959 n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE, 1960 VIRTQUEUE_MAX_SIZE); 1961 virtio_cleanup(vdev); 1962 return; 1963 } 1964 1965 n->max_queues = MAX(n->nic_conf.peers.queues, 1); 1966 if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) { 1967 error_setg(errp, "Invalid number of queues (= %" PRIu32 "), " 1968 "must be a positive integer less than %d.", 1969 n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2); 1970 virtio_cleanup(vdev); 1971 return; 1972 } 1973 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues); 1974 n->curr_queues = 1; 1975 n->tx_timeout = n->net_conf.txtimer; 1976 1977 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 1978 && strcmp(n->net_conf.tx, "bh")) { 1979 error_report("virtio-net: " 1980 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 1981 n->net_conf.tx); 1982 error_report("Defaulting to \"bh\""); 1983 } 1984 1985 n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n), 1986 n->net_conf.tx_queue_size); 1987 1988 for (i = 0; i < n->max_queues; i++) { 1989 virtio_net_add_queue(n, i); 1990 } 1991 1992 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1993 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 1994 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 1995 n->status = VIRTIO_NET_S_LINK_UP; 1996 n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, 1997 virtio_net_announce_timer, n); 1998 1999 if (n->netclient_type) { 2000 /* 2001 * Happen when virtio_net_set_netclient_name has been called. 2002 */ 2003 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 2004 n->netclient_type, n->netclient_name, n); 2005 } else { 2006 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 2007 object_get_typename(OBJECT(dev)), dev->id, n); 2008 } 2009 2010 peer_test_vnet_hdr(n); 2011 if (peer_has_vnet_hdr(n)) { 2012 for (i = 0; i < n->max_queues; i++) { 2013 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true); 2014 } 2015 n->host_hdr_len = sizeof(struct virtio_net_hdr); 2016 } else { 2017 n->host_hdr_len = 0; 2018 } 2019 2020 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 2021 2022 n->vqs[0].tx_waiting = 0; 2023 n->tx_burst = n->net_conf.txburst; 2024 virtio_net_set_mrg_rx_bufs(n, 0, 0); 2025 n->promisc = 1; /* for compatibility */ 2026 2027 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 2028 2029 n->vlans = g_malloc0(MAX_VLAN >> 3); 2030 2031 nc = qemu_get_queue(n->nic); 2032 nc->rxfilter_notify_enabled = 1; 2033 2034 n->qdev = dev; 2035 } 2036 2037 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp) 2038 { 2039 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 2040 VirtIONet *n = VIRTIO_NET(dev); 2041 int i, max_queues; 2042 2043 /* This will stop vhost backend if appropriate. */ 2044 virtio_net_set_status(vdev, 0); 2045 2046 g_free(n->netclient_name); 2047 n->netclient_name = NULL; 2048 g_free(n->netclient_type); 2049 n->netclient_type = NULL; 2050 2051 g_free(n->mac_table.macs); 2052 g_free(n->vlans); 2053 2054 max_queues = n->multiqueue ? n->max_queues : 1; 2055 for (i = 0; i < max_queues; i++) { 2056 virtio_net_del_queue(n, i); 2057 } 2058 2059 timer_del(n->announce_timer); 2060 timer_free(n->announce_timer); 2061 g_free(n->vqs); 2062 qemu_del_nic(n->nic); 2063 virtio_cleanup(vdev); 2064 } 2065 2066 static void virtio_net_instance_init(Object *obj) 2067 { 2068 VirtIONet *n = VIRTIO_NET(obj); 2069 2070 /* 2071 * The default config_size is sizeof(struct virtio_net_config). 2072 * Can be overriden with virtio_net_set_config_size. 2073 */ 2074 n->config_size = sizeof(struct virtio_net_config); 2075 device_add_bootindex_property(obj, &n->nic_conf.bootindex, 2076 "bootindex", "/ethernet-phy@0", 2077 DEVICE(n), NULL); 2078 } 2079 2080 static void virtio_net_pre_save(void *opaque) 2081 { 2082 VirtIONet *n = opaque; 2083 2084 /* At this point, backend must be stopped, otherwise 2085 * it might keep writing to memory. */ 2086 assert(!n->vhost_started); 2087 } 2088 2089 static const VMStateDescription vmstate_virtio_net = { 2090 .name = "virtio-net", 2091 .minimum_version_id = VIRTIO_NET_VM_VERSION, 2092 .version_id = VIRTIO_NET_VM_VERSION, 2093 .fields = (VMStateField[]) { 2094 VMSTATE_VIRTIO_DEVICE, 2095 VMSTATE_END_OF_LIST() 2096 }, 2097 .pre_save = virtio_net_pre_save, 2098 }; 2099 2100 static Property virtio_net_properties[] = { 2101 DEFINE_PROP_BIT("csum", VirtIONet, host_features, VIRTIO_NET_F_CSUM, true), 2102 DEFINE_PROP_BIT("guest_csum", VirtIONet, host_features, 2103 VIRTIO_NET_F_GUEST_CSUM, true), 2104 DEFINE_PROP_BIT("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true), 2105 DEFINE_PROP_BIT("guest_tso4", VirtIONet, host_features, 2106 VIRTIO_NET_F_GUEST_TSO4, true), 2107 DEFINE_PROP_BIT("guest_tso6", VirtIONet, host_features, 2108 VIRTIO_NET_F_GUEST_TSO6, true), 2109 DEFINE_PROP_BIT("guest_ecn", VirtIONet, host_features, 2110 VIRTIO_NET_F_GUEST_ECN, true), 2111 DEFINE_PROP_BIT("guest_ufo", VirtIONet, host_features, 2112 VIRTIO_NET_F_GUEST_UFO, true), 2113 DEFINE_PROP_BIT("guest_announce", VirtIONet, host_features, 2114 VIRTIO_NET_F_GUEST_ANNOUNCE, true), 2115 DEFINE_PROP_BIT("host_tso4", VirtIONet, host_features, 2116 VIRTIO_NET_F_HOST_TSO4, true), 2117 DEFINE_PROP_BIT("host_tso6", VirtIONet, host_features, 2118 VIRTIO_NET_F_HOST_TSO6, true), 2119 DEFINE_PROP_BIT("host_ecn", VirtIONet, host_features, 2120 VIRTIO_NET_F_HOST_ECN, true), 2121 DEFINE_PROP_BIT("host_ufo", VirtIONet, host_features, 2122 VIRTIO_NET_F_HOST_UFO, true), 2123 DEFINE_PROP_BIT("mrg_rxbuf", VirtIONet, host_features, 2124 VIRTIO_NET_F_MRG_RXBUF, true), 2125 DEFINE_PROP_BIT("status", VirtIONet, host_features, 2126 VIRTIO_NET_F_STATUS, true), 2127 DEFINE_PROP_BIT("ctrl_vq", VirtIONet, host_features, 2128 VIRTIO_NET_F_CTRL_VQ, true), 2129 DEFINE_PROP_BIT("ctrl_rx", VirtIONet, host_features, 2130 VIRTIO_NET_F_CTRL_RX, true), 2131 DEFINE_PROP_BIT("ctrl_vlan", VirtIONet, host_features, 2132 VIRTIO_NET_F_CTRL_VLAN, true), 2133 DEFINE_PROP_BIT("ctrl_rx_extra", VirtIONet, host_features, 2134 VIRTIO_NET_F_CTRL_RX_EXTRA, true), 2135 DEFINE_PROP_BIT("ctrl_mac_addr", VirtIONet, host_features, 2136 VIRTIO_NET_F_CTRL_MAC_ADDR, true), 2137 DEFINE_PROP_BIT("ctrl_guest_offloads", VirtIONet, host_features, 2138 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true), 2139 DEFINE_PROP_BIT("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false), 2140 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 2141 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 2142 TX_TIMER_INTERVAL), 2143 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 2144 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 2145 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size, 2146 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE), 2147 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size, 2148 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE), 2149 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0), 2150 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend, 2151 true), 2152 DEFINE_PROP_END_OF_LIST(), 2153 }; 2154 2155 static void virtio_net_class_init(ObjectClass *klass, void *data) 2156 { 2157 DeviceClass *dc = DEVICE_CLASS(klass); 2158 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 2159 2160 dc->props = virtio_net_properties; 2161 dc->vmsd = &vmstate_virtio_net; 2162 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 2163 vdc->realize = virtio_net_device_realize; 2164 vdc->unrealize = virtio_net_device_unrealize; 2165 vdc->get_config = virtio_net_get_config; 2166 vdc->set_config = virtio_net_set_config; 2167 vdc->get_features = virtio_net_get_features; 2168 vdc->set_features = virtio_net_set_features; 2169 vdc->bad_features = virtio_net_bad_features; 2170 vdc->reset = virtio_net_reset; 2171 vdc->set_status = virtio_net_set_status; 2172 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 2173 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 2174 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO); 2175 vdc->vmsd = &vmstate_virtio_net_device; 2176 } 2177 2178 static const TypeInfo virtio_net_info = { 2179 .name = TYPE_VIRTIO_NET, 2180 .parent = TYPE_VIRTIO_DEVICE, 2181 .instance_size = sizeof(VirtIONet), 2182 .instance_init = virtio_net_instance_init, 2183 .class_init = virtio_net_class_init, 2184 }; 2185 2186 static void virtio_register_types(void) 2187 { 2188 type_register_static(&virtio_net_info); 2189 } 2190 2191 type_init(virtio_register_types) 2192