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/iov.h" 15 #include "hw/virtio/virtio.h" 16 #include "net/net.h" 17 #include "net/checksum.h" 18 #include "net/tap.h" 19 #include "qemu/error-report.h" 20 #include "qemu/timer.h" 21 #include "hw/virtio/virtio-net.h" 22 #include "net/vhost_net.h" 23 #include "hw/virtio/virtio-bus.h" 24 #include "qapi/qmp/qjson.h" 25 #include "qapi-event.h" 26 #include "hw/virtio/virtio-access.h" 27 28 #define VIRTIO_NET_VM_VERSION 11 29 30 #define MAC_TABLE_ENTRIES 64 31 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */ 32 33 /* 34 * Calculate the number of bytes up to and including the given 'field' of 35 * 'container'. 36 */ 37 #define endof(container, field) \ 38 (offsetof(container, field) + sizeof(((container *)0)->field)) 39 40 typedef struct VirtIOFeature { 41 uint32_t flags; 42 size_t end; 43 } VirtIOFeature; 44 45 static VirtIOFeature feature_sizes[] = { 46 {.flags = 1 << VIRTIO_NET_F_MAC, 47 .end = endof(struct virtio_net_config, mac)}, 48 {.flags = 1 << VIRTIO_NET_F_STATUS, 49 .end = endof(struct virtio_net_config, status)}, 50 {.flags = 1 << VIRTIO_NET_F_MQ, 51 .end = endof(struct virtio_net_config, max_virtqueue_pairs)}, 52 {} 53 }; 54 55 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc) 56 { 57 VirtIONet *n = qemu_get_nic_opaque(nc); 58 59 return &n->vqs[nc->queue_index]; 60 } 61 62 static int vq2q(int queue_index) 63 { 64 return queue_index / 2; 65 } 66 67 /* TODO 68 * - we could suppress RX interrupt if we were so inclined. 69 */ 70 71 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config) 72 { 73 VirtIONet *n = VIRTIO_NET(vdev); 74 struct virtio_net_config netcfg; 75 76 virtio_stw_p(vdev, &netcfg.status, n->status); 77 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues); 78 memcpy(netcfg.mac, n->mac, ETH_ALEN); 79 memcpy(config, &netcfg, n->config_size); 80 } 81 82 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config) 83 { 84 VirtIONet *n = VIRTIO_NET(vdev); 85 struct virtio_net_config netcfg = {}; 86 87 memcpy(&netcfg, config, n->config_size); 88 89 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) && 90 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1) && 91 memcmp(netcfg.mac, n->mac, ETH_ALEN)) { 92 memcpy(n->mac, netcfg.mac, ETH_ALEN); 93 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 94 } 95 } 96 97 static bool virtio_net_started(VirtIONet *n, uint8_t status) 98 { 99 VirtIODevice *vdev = VIRTIO_DEVICE(n); 100 return (status & VIRTIO_CONFIG_S_DRIVER_OK) && 101 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running; 102 } 103 104 static void virtio_net_announce_timer(void *opaque) 105 { 106 VirtIONet *n = opaque; 107 VirtIODevice *vdev = VIRTIO_DEVICE(n); 108 109 n->announce_counter--; 110 n->status |= VIRTIO_NET_S_ANNOUNCE; 111 virtio_notify_config(vdev); 112 } 113 114 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status) 115 { 116 VirtIODevice *vdev = VIRTIO_DEVICE(n); 117 NetClientState *nc = qemu_get_queue(n->nic); 118 int queues = n->multiqueue ? n->max_queues : 1; 119 120 if (!get_vhost_net(nc->peer)) { 121 return; 122 } 123 124 if ((virtio_net_started(n, status) && !nc->peer->link_down) == 125 !!n->vhost_started) { 126 return; 127 } 128 if (!n->vhost_started) { 129 int r, i; 130 131 if (!vhost_net_query(get_vhost_net(nc->peer), vdev)) { 132 return; 133 } 134 135 /* Any packets outstanding? Purge them to avoid touching rings 136 * when vhost is running. 137 */ 138 for (i = 0; i < queues; i++) { 139 NetClientState *qnc = qemu_get_subqueue(n->nic, i); 140 141 /* Purge both directions: TX and RX. */ 142 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc); 143 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer); 144 } 145 146 n->vhost_started = 1; 147 r = vhost_net_start(vdev, n->nic->ncs, queues); 148 if (r < 0) { 149 error_report("unable to start vhost net: %d: " 150 "falling back on userspace virtio", -r); 151 n->vhost_started = 0; 152 } 153 } else { 154 vhost_net_stop(vdev, n->nic->ncs, queues); 155 n->vhost_started = 0; 156 } 157 } 158 159 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status) 160 { 161 VirtIONet *n = VIRTIO_NET(vdev); 162 VirtIONetQueue *q; 163 int i; 164 uint8_t queue_status; 165 166 virtio_net_vhost_status(n, status); 167 168 for (i = 0; i < n->max_queues; i++) { 169 q = &n->vqs[i]; 170 171 if ((!n->multiqueue && i != 0) || i >= n->curr_queues) { 172 queue_status = 0; 173 } else { 174 queue_status = status; 175 } 176 177 if (!q->tx_waiting) { 178 continue; 179 } 180 181 if (virtio_net_started(n, queue_status) && !n->vhost_started) { 182 if (q->tx_timer) { 183 timer_mod(q->tx_timer, 184 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 185 } else { 186 qemu_bh_schedule(q->tx_bh); 187 } 188 } else { 189 if (q->tx_timer) { 190 timer_del(q->tx_timer); 191 } else { 192 qemu_bh_cancel(q->tx_bh); 193 } 194 } 195 } 196 } 197 198 static void virtio_net_set_link_status(NetClientState *nc) 199 { 200 VirtIONet *n = qemu_get_nic_opaque(nc); 201 VirtIODevice *vdev = VIRTIO_DEVICE(n); 202 uint16_t old_status = n->status; 203 204 if (nc->link_down) 205 n->status &= ~VIRTIO_NET_S_LINK_UP; 206 else 207 n->status |= VIRTIO_NET_S_LINK_UP; 208 209 if (n->status != old_status) 210 virtio_notify_config(vdev); 211 212 virtio_net_set_status(vdev, vdev->status); 213 } 214 215 static void rxfilter_notify(NetClientState *nc) 216 { 217 VirtIONet *n = qemu_get_nic_opaque(nc); 218 219 if (nc->rxfilter_notify_enabled) { 220 gchar *path = object_get_canonical_path(OBJECT(n->qdev)); 221 qapi_event_send_nic_rx_filter_changed(!!n->netclient_name, 222 n->netclient_name, path, &error_abort); 223 g_free(path); 224 225 /* disable event notification to avoid events flooding */ 226 nc->rxfilter_notify_enabled = 0; 227 } 228 } 229 230 static intList *get_vlan_table(VirtIONet *n) 231 { 232 intList *list, *entry; 233 int i, j; 234 235 list = NULL; 236 for (i = 0; i < MAX_VLAN >> 5; i++) { 237 for (j = 0; n->vlans[i] && j <= 0x1f; j++) { 238 if (n->vlans[i] & (1U << j)) { 239 entry = g_malloc0(sizeof(*entry)); 240 entry->value = (i << 5) + j; 241 entry->next = list; 242 list = entry; 243 } 244 } 245 } 246 247 return list; 248 } 249 250 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc) 251 { 252 VirtIONet *n = qemu_get_nic_opaque(nc); 253 VirtIODevice *vdev = VIRTIO_DEVICE(n); 254 RxFilterInfo *info; 255 strList *str_list, *entry; 256 int i; 257 258 info = g_malloc0(sizeof(*info)); 259 info->name = g_strdup(nc->name); 260 info->promiscuous = n->promisc; 261 262 if (n->nouni) { 263 info->unicast = RX_STATE_NONE; 264 } else if (n->alluni) { 265 info->unicast = RX_STATE_ALL; 266 } else { 267 info->unicast = RX_STATE_NORMAL; 268 } 269 270 if (n->nomulti) { 271 info->multicast = RX_STATE_NONE; 272 } else if (n->allmulti) { 273 info->multicast = RX_STATE_ALL; 274 } else { 275 info->multicast = RX_STATE_NORMAL; 276 } 277 278 info->broadcast_allowed = n->nobcast; 279 info->multicast_overflow = n->mac_table.multi_overflow; 280 info->unicast_overflow = n->mac_table.uni_overflow; 281 282 info->main_mac = qemu_mac_strdup_printf(n->mac); 283 284 str_list = NULL; 285 for (i = 0; i < n->mac_table.first_multi; i++) { 286 entry = g_malloc0(sizeof(*entry)); 287 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN); 288 entry->next = str_list; 289 str_list = entry; 290 } 291 info->unicast_table = str_list; 292 293 str_list = NULL; 294 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 295 entry = g_malloc0(sizeof(*entry)); 296 entry->value = qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN); 297 entry->next = str_list; 298 str_list = entry; 299 } 300 info->multicast_table = str_list; 301 info->vlan_table = get_vlan_table(n); 302 303 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) { 304 info->vlan = RX_STATE_ALL; 305 } else if (!info->vlan_table) { 306 info->vlan = RX_STATE_NONE; 307 } else { 308 info->vlan = RX_STATE_NORMAL; 309 } 310 311 /* enable event notification after query */ 312 nc->rxfilter_notify_enabled = 1; 313 314 return info; 315 } 316 317 static void virtio_net_reset(VirtIODevice *vdev) 318 { 319 VirtIONet *n = VIRTIO_NET(vdev); 320 321 /* Reset back to compatibility mode */ 322 n->promisc = 1; 323 n->allmulti = 0; 324 n->alluni = 0; 325 n->nomulti = 0; 326 n->nouni = 0; 327 n->nobcast = 0; 328 /* multiqueue is disabled by default */ 329 n->curr_queues = 1; 330 timer_del(n->announce_timer); 331 n->announce_counter = 0; 332 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 333 334 /* Flush any MAC and VLAN filter table state */ 335 n->mac_table.in_use = 0; 336 n->mac_table.first_multi = 0; 337 n->mac_table.multi_overflow = 0; 338 n->mac_table.uni_overflow = 0; 339 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN); 340 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac)); 341 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 342 memset(n->vlans, 0, MAX_VLAN >> 3); 343 } 344 345 static void peer_test_vnet_hdr(VirtIONet *n) 346 { 347 NetClientState *nc = qemu_get_queue(n->nic); 348 if (!nc->peer) { 349 return; 350 } 351 352 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer); 353 } 354 355 static int peer_has_vnet_hdr(VirtIONet *n) 356 { 357 return n->has_vnet_hdr; 358 } 359 360 static int peer_has_ufo(VirtIONet *n) 361 { 362 if (!peer_has_vnet_hdr(n)) 363 return 0; 364 365 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer); 366 367 return n->has_ufo; 368 } 369 370 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs, 371 int version_1) 372 { 373 int i; 374 NetClientState *nc; 375 376 n->mergeable_rx_bufs = mergeable_rx_bufs; 377 378 if (version_1) { 379 n->guest_hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); 380 } else { 381 n->guest_hdr_len = n->mergeable_rx_bufs ? 382 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 383 sizeof(struct virtio_net_hdr); 384 } 385 386 for (i = 0; i < n->max_queues; i++) { 387 nc = qemu_get_subqueue(n->nic, i); 388 389 if (peer_has_vnet_hdr(n) && 390 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) { 391 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len); 392 n->host_hdr_len = n->guest_hdr_len; 393 } 394 } 395 } 396 397 static int peer_attach(VirtIONet *n, int index) 398 { 399 NetClientState *nc = qemu_get_subqueue(n->nic, index); 400 401 if (!nc->peer) { 402 return 0; 403 } 404 405 if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 406 return 0; 407 } 408 409 return tap_enable(nc->peer); 410 } 411 412 static int peer_detach(VirtIONet *n, int index) 413 { 414 NetClientState *nc = qemu_get_subqueue(n->nic, index); 415 416 if (!nc->peer) { 417 return 0; 418 } 419 420 if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 421 return 0; 422 } 423 424 return tap_disable(nc->peer); 425 } 426 427 static void virtio_net_set_queues(VirtIONet *n) 428 { 429 int i; 430 int r; 431 432 for (i = 0; i < n->max_queues; i++) { 433 if (i < n->curr_queues) { 434 r = peer_attach(n, i); 435 assert(!r); 436 } else { 437 r = peer_detach(n, i); 438 assert(!r); 439 } 440 } 441 } 442 443 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); 444 445 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features) 446 { 447 VirtIONet *n = VIRTIO_NET(vdev); 448 NetClientState *nc = qemu_get_queue(n->nic); 449 450 /* Firstly sync all virtio-net possible supported features */ 451 features |= n->host_features; 452 453 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 454 455 if (!peer_has_vnet_hdr(n)) { 456 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM); 457 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4); 458 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6); 459 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN); 460 461 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM); 462 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4); 463 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6); 464 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN); 465 } 466 467 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 468 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO); 469 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO); 470 } 471 472 if (!get_vhost_net(nc->peer)) { 473 virtio_add_feature(&features, VIRTIO_F_VERSION_1); 474 return features; 475 } 476 return vhost_net_get_features(get_vhost_net(nc->peer), features); 477 } 478 479 static uint64_t virtio_net_bad_features(VirtIODevice *vdev) 480 { 481 uint64_t features = 0; 482 483 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 484 * but also these: */ 485 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 486 virtio_add_feature(&features, VIRTIO_NET_F_CSUM); 487 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4); 488 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6); 489 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN); 490 491 return features; 492 } 493 494 static void virtio_net_apply_guest_offloads(VirtIONet *n) 495 { 496 qemu_set_offload(qemu_get_queue(n->nic)->peer, 497 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 498 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 499 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 500 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 501 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO))); 502 } 503 504 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features) 505 { 506 static const uint64_t guest_offloads_mask = 507 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 508 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 509 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 510 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 511 (1ULL << VIRTIO_NET_F_GUEST_UFO); 512 513 return guest_offloads_mask & features; 514 } 515 516 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n) 517 { 518 VirtIODevice *vdev = VIRTIO_DEVICE(n); 519 return virtio_net_guest_offloads_by_features(vdev->guest_features); 520 } 521 522 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features) 523 { 524 VirtIONet *n = VIRTIO_NET(vdev); 525 int i; 526 527 virtio_net_set_multiqueue(n, 528 __virtio_has_feature(features, VIRTIO_NET_F_MQ)); 529 530 virtio_net_set_mrg_rx_bufs(n, 531 __virtio_has_feature(features, 532 VIRTIO_NET_F_MRG_RXBUF), 533 __virtio_has_feature(features, 534 VIRTIO_F_VERSION_1)); 535 536 if (n->has_vnet_hdr) { 537 n->curr_guest_offloads = 538 virtio_net_guest_offloads_by_features(features); 539 virtio_net_apply_guest_offloads(n); 540 } 541 542 for (i = 0; i < n->max_queues; i++) { 543 NetClientState *nc = qemu_get_subqueue(n->nic, i); 544 545 if (!get_vhost_net(nc->peer)) { 546 continue; 547 } 548 vhost_net_ack_features(get_vhost_net(nc->peer), features); 549 } 550 551 if (__virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) { 552 memset(n->vlans, 0, MAX_VLAN >> 3); 553 } else { 554 memset(n->vlans, 0xff, MAX_VLAN >> 3); 555 } 556 } 557 558 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 559 struct iovec *iov, unsigned int iov_cnt) 560 { 561 uint8_t on; 562 size_t s; 563 NetClientState *nc = qemu_get_queue(n->nic); 564 565 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 566 if (s != sizeof(on)) { 567 return VIRTIO_NET_ERR; 568 } 569 570 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 571 n->promisc = on; 572 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 573 n->allmulti = on; 574 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 575 n->alluni = on; 576 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 577 n->nomulti = on; 578 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 579 n->nouni = on; 580 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 581 n->nobcast = on; 582 } else { 583 return VIRTIO_NET_ERR; 584 } 585 586 rxfilter_notify(nc); 587 588 return VIRTIO_NET_OK; 589 } 590 591 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 592 struct iovec *iov, unsigned int iov_cnt) 593 { 594 VirtIODevice *vdev = VIRTIO_DEVICE(n); 595 uint64_t offloads; 596 size_t s; 597 598 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 599 return VIRTIO_NET_ERR; 600 } 601 602 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 603 if (s != sizeof(offloads)) { 604 return VIRTIO_NET_ERR; 605 } 606 607 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 608 uint64_t supported_offloads; 609 610 if (!n->has_vnet_hdr) { 611 return VIRTIO_NET_ERR; 612 } 613 614 supported_offloads = virtio_net_supported_guest_offloads(n); 615 if (offloads & ~supported_offloads) { 616 return VIRTIO_NET_ERR; 617 } 618 619 n->curr_guest_offloads = offloads; 620 virtio_net_apply_guest_offloads(n); 621 622 return VIRTIO_NET_OK; 623 } else { 624 return VIRTIO_NET_ERR; 625 } 626 } 627 628 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 629 struct iovec *iov, unsigned int iov_cnt) 630 { 631 VirtIODevice *vdev = VIRTIO_DEVICE(n); 632 struct virtio_net_ctrl_mac mac_data; 633 size_t s; 634 NetClientState *nc = qemu_get_queue(n->nic); 635 636 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 637 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 638 return VIRTIO_NET_ERR; 639 } 640 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 641 assert(s == sizeof(n->mac)); 642 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 643 rxfilter_notify(nc); 644 645 return VIRTIO_NET_OK; 646 } 647 648 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 649 return VIRTIO_NET_ERR; 650 } 651 652 int in_use = 0; 653 int first_multi = 0; 654 uint8_t uni_overflow = 0; 655 uint8_t multi_overflow = 0; 656 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 657 658 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 659 sizeof(mac_data.entries)); 660 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 661 if (s != sizeof(mac_data.entries)) { 662 goto error; 663 } 664 iov_discard_front(&iov, &iov_cnt, s); 665 666 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 667 goto error; 668 } 669 670 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 671 s = iov_to_buf(iov, iov_cnt, 0, macs, 672 mac_data.entries * ETH_ALEN); 673 if (s != mac_data.entries * ETH_ALEN) { 674 goto error; 675 } 676 in_use += mac_data.entries; 677 } else { 678 uni_overflow = 1; 679 } 680 681 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 682 683 first_multi = in_use; 684 685 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 686 sizeof(mac_data.entries)); 687 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 688 if (s != sizeof(mac_data.entries)) { 689 goto error; 690 } 691 692 iov_discard_front(&iov, &iov_cnt, s); 693 694 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 695 goto error; 696 } 697 698 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) { 699 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN], 700 mac_data.entries * ETH_ALEN); 701 if (s != mac_data.entries * ETH_ALEN) { 702 goto error; 703 } 704 in_use += mac_data.entries; 705 } else { 706 multi_overflow = 1; 707 } 708 709 n->mac_table.in_use = in_use; 710 n->mac_table.first_multi = first_multi; 711 n->mac_table.uni_overflow = uni_overflow; 712 n->mac_table.multi_overflow = multi_overflow; 713 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN); 714 g_free(macs); 715 rxfilter_notify(nc); 716 717 return VIRTIO_NET_OK; 718 719 error: 720 g_free(macs); 721 return VIRTIO_NET_ERR; 722 } 723 724 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 725 struct iovec *iov, unsigned int iov_cnt) 726 { 727 VirtIODevice *vdev = VIRTIO_DEVICE(n); 728 uint16_t vid; 729 size_t s; 730 NetClientState *nc = qemu_get_queue(n->nic); 731 732 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 733 vid = virtio_lduw_p(vdev, &vid); 734 if (s != sizeof(vid)) { 735 return VIRTIO_NET_ERR; 736 } 737 738 if (vid >= MAX_VLAN) 739 return VIRTIO_NET_ERR; 740 741 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 742 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 743 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 744 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 745 else 746 return VIRTIO_NET_ERR; 747 748 rxfilter_notify(nc); 749 750 return VIRTIO_NET_OK; 751 } 752 753 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd, 754 struct iovec *iov, unsigned int iov_cnt) 755 { 756 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK && 757 n->status & VIRTIO_NET_S_ANNOUNCE) { 758 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 759 if (n->announce_counter) { 760 timer_mod(n->announce_timer, 761 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 762 self_announce_delay(n->announce_counter)); 763 } 764 return VIRTIO_NET_OK; 765 } else { 766 return VIRTIO_NET_ERR; 767 } 768 } 769 770 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 771 struct iovec *iov, unsigned int iov_cnt) 772 { 773 VirtIODevice *vdev = VIRTIO_DEVICE(n); 774 struct virtio_net_ctrl_mq mq; 775 size_t s; 776 uint16_t queues; 777 778 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 779 if (s != sizeof(mq)) { 780 return VIRTIO_NET_ERR; 781 } 782 783 if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 784 return VIRTIO_NET_ERR; 785 } 786 787 queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs); 788 789 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 790 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 791 queues > n->max_queues || 792 !n->multiqueue) { 793 return VIRTIO_NET_ERR; 794 } 795 796 n->curr_queues = queues; 797 /* stop the backend before changing the number of queues to avoid handling a 798 * disabled queue */ 799 virtio_net_set_status(vdev, vdev->status); 800 virtio_net_set_queues(n); 801 802 return VIRTIO_NET_OK; 803 } 804 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 805 { 806 VirtIONet *n = VIRTIO_NET(vdev); 807 struct virtio_net_ctrl_hdr ctrl; 808 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 809 VirtQueueElement elem; 810 size_t s; 811 struct iovec *iov, *iov2; 812 unsigned int iov_cnt; 813 814 while (virtqueue_pop(vq, &elem)) { 815 if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) || 816 iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) { 817 error_report("virtio-net ctrl missing headers"); 818 exit(1); 819 } 820 821 iov_cnt = elem.out_num; 822 iov2 = iov = g_memdup(elem.out_sg, sizeof(struct iovec) * elem.out_num); 823 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl)); 824 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl)); 825 if (s != sizeof(ctrl)) { 826 status = VIRTIO_NET_ERR; 827 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 828 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt); 829 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 830 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt); 831 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 832 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt); 833 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) { 834 status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt); 835 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 836 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt); 837 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 838 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt); 839 } 840 841 s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status)); 842 assert(s == sizeof(status)); 843 844 virtqueue_push(vq, &elem, sizeof(status)); 845 virtio_notify(vdev, vq); 846 g_free(iov2); 847 } 848 } 849 850 /* RX */ 851 852 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 853 { 854 VirtIONet *n = VIRTIO_NET(vdev); 855 int queue_index = vq2q(virtio_get_queue_index(vq)); 856 857 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 858 } 859 860 static int virtio_net_can_receive(NetClientState *nc) 861 { 862 VirtIONet *n = qemu_get_nic_opaque(nc); 863 VirtIODevice *vdev = VIRTIO_DEVICE(n); 864 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 865 866 if (!vdev->vm_running) { 867 return 0; 868 } 869 870 if (nc->queue_index >= n->curr_queues) { 871 return 0; 872 } 873 874 if (!virtio_queue_ready(q->rx_vq) || 875 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 876 return 0; 877 } 878 879 return 1; 880 } 881 882 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 883 { 884 VirtIONet *n = q->n; 885 if (virtio_queue_empty(q->rx_vq) || 886 (n->mergeable_rx_bufs && 887 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 888 virtio_queue_set_notification(q->rx_vq, 1); 889 890 /* To avoid a race condition where the guest has made some buffers 891 * available after the above check but before notification was 892 * enabled, check for available buffers again. 893 */ 894 if (virtio_queue_empty(q->rx_vq) || 895 (n->mergeable_rx_bufs && 896 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 897 return 0; 898 } 899 } 900 901 virtio_queue_set_notification(q->rx_vq, 0); 902 return 1; 903 } 904 905 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr) 906 { 907 virtio_tswap16s(vdev, &hdr->hdr_len); 908 virtio_tswap16s(vdev, &hdr->gso_size); 909 virtio_tswap16s(vdev, &hdr->csum_start); 910 virtio_tswap16s(vdev, &hdr->csum_offset); 911 } 912 913 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 914 * it never finds out that the packets don't have valid checksums. This 915 * causes dhclient to get upset. Fedora's carried a patch for ages to 916 * fix this with Xen but it hasn't appeared in an upstream release of 917 * dhclient yet. 918 * 919 * To avoid breaking existing guests, we catch udp packets and add 920 * checksums. This is terrible but it's better than hacking the guest 921 * kernels. 922 * 923 * N.B. if we introduce a zero-copy API, this operation is no longer free so 924 * we should provide a mechanism to disable it to avoid polluting the host 925 * cache. 926 */ 927 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 928 uint8_t *buf, size_t size) 929 { 930 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 931 (size > 27 && size < 1500) && /* normal sized MTU */ 932 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 933 (buf[23] == 17) && /* ip.protocol == UDP */ 934 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 935 net_checksum_calculate(buf, size); 936 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 937 } 938 } 939 940 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt, 941 const void *buf, size_t size) 942 { 943 if (n->has_vnet_hdr) { 944 /* FIXME this cast is evil */ 945 void *wbuf = (void *)buf; 946 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, 947 size - n->host_hdr_len); 948 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf); 949 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr)); 950 } else { 951 struct virtio_net_hdr hdr = { 952 .flags = 0, 953 .gso_type = VIRTIO_NET_HDR_GSO_NONE 954 }; 955 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr); 956 } 957 } 958 959 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 960 { 961 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 962 static const uint8_t vlan[] = {0x81, 0x00}; 963 uint8_t *ptr = (uint8_t *)buf; 964 int i; 965 966 if (n->promisc) 967 return 1; 968 969 ptr += n->host_hdr_len; 970 971 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 972 int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff; 973 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 974 return 0; 975 } 976 977 if (ptr[0] & 1) { // multicast 978 if (!memcmp(ptr, bcast, sizeof(bcast))) { 979 return !n->nobcast; 980 } else if (n->nomulti) { 981 return 0; 982 } else if (n->allmulti || n->mac_table.multi_overflow) { 983 return 1; 984 } 985 986 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 987 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 988 return 1; 989 } 990 } 991 } else { // unicast 992 if (n->nouni) { 993 return 0; 994 } else if (n->alluni || n->mac_table.uni_overflow) { 995 return 1; 996 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 997 return 1; 998 } 999 1000 for (i = 0; i < n->mac_table.first_multi; i++) { 1001 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1002 return 1; 1003 } 1004 } 1005 } 1006 1007 return 0; 1008 } 1009 1010 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size) 1011 { 1012 VirtIONet *n = qemu_get_nic_opaque(nc); 1013 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1014 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1015 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 1016 struct virtio_net_hdr_mrg_rxbuf mhdr; 1017 unsigned mhdr_cnt = 0; 1018 size_t offset, i, guest_offset; 1019 1020 if (!virtio_net_can_receive(nc)) { 1021 return -1; 1022 } 1023 1024 /* hdr_len refers to the header we supply to the guest */ 1025 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 1026 return 0; 1027 } 1028 1029 if (!receive_filter(n, buf, size)) 1030 return size; 1031 1032 offset = i = 0; 1033 1034 while (offset < size) { 1035 VirtQueueElement elem; 1036 int len, total; 1037 const struct iovec *sg = elem.in_sg; 1038 1039 total = 0; 1040 1041 if (virtqueue_pop(q->rx_vq, &elem) == 0) { 1042 if (i == 0) 1043 return -1; 1044 error_report("virtio-net unexpected empty queue: " 1045 "i %zd mergeable %d offset %zd, size %zd, " 1046 "guest hdr len %zd, host hdr len %zd " 1047 "guest features 0x%" PRIx64, 1048 i, n->mergeable_rx_bufs, offset, size, 1049 n->guest_hdr_len, n->host_hdr_len, 1050 vdev->guest_features); 1051 exit(1); 1052 } 1053 1054 if (elem.in_num < 1) { 1055 error_report("virtio-net receive queue contains no in buffers"); 1056 exit(1); 1057 } 1058 1059 if (i == 0) { 1060 assert(offset == 0); 1061 if (n->mergeable_rx_bufs) { 1062 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 1063 sg, elem.in_num, 1064 offsetof(typeof(mhdr), num_buffers), 1065 sizeof(mhdr.num_buffers)); 1066 } 1067 1068 receive_header(n, sg, elem.in_num, buf, size); 1069 offset = n->host_hdr_len; 1070 total += n->guest_hdr_len; 1071 guest_offset = n->guest_hdr_len; 1072 } else { 1073 guest_offset = 0; 1074 } 1075 1076 /* copy in packet. ugh */ 1077 len = iov_from_buf(sg, elem.in_num, guest_offset, 1078 buf + offset, size - offset); 1079 total += len; 1080 offset += len; 1081 /* If buffers can't be merged, at this point we 1082 * must have consumed the complete packet. 1083 * Otherwise, drop it. */ 1084 if (!n->mergeable_rx_bufs && offset < size) { 1085 #if 0 1086 error_report("virtio-net truncated non-mergeable packet: " 1087 "i %zd mergeable %d offset %zd, size %zd, " 1088 "guest hdr len %zd, host hdr len %zd", 1089 i, n->mergeable_rx_bufs, 1090 offset, size, n->guest_hdr_len, n->host_hdr_len); 1091 #endif 1092 return size; 1093 } 1094 1095 /* signal other side */ 1096 virtqueue_fill(q->rx_vq, &elem, total, i++); 1097 } 1098 1099 if (mhdr_cnt) { 1100 virtio_stw_p(vdev, &mhdr.num_buffers, i); 1101 iov_from_buf(mhdr_sg, mhdr_cnt, 1102 0, 1103 &mhdr.num_buffers, sizeof mhdr.num_buffers); 1104 } 1105 1106 virtqueue_flush(q->rx_vq, i); 1107 virtio_notify(vdev, q->rx_vq); 1108 1109 return size; 1110 } 1111 1112 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 1113 1114 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 1115 { 1116 VirtIONet *n = qemu_get_nic_opaque(nc); 1117 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1118 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1119 1120 virtqueue_push(q->tx_vq, &q->async_tx.elem, 0); 1121 virtio_notify(vdev, q->tx_vq); 1122 1123 q->async_tx.elem.out_num = q->async_tx.len = 0; 1124 1125 virtio_queue_set_notification(q->tx_vq, 1); 1126 virtio_net_flush_tx(q); 1127 } 1128 1129 /* TX */ 1130 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 1131 { 1132 VirtIONet *n = q->n; 1133 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1134 VirtQueueElement elem; 1135 int32_t num_packets = 0; 1136 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 1137 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1138 return num_packets; 1139 } 1140 1141 if (q->async_tx.elem.out_num) { 1142 virtio_queue_set_notification(q->tx_vq, 0); 1143 return num_packets; 1144 } 1145 1146 while (virtqueue_pop(q->tx_vq, &elem)) { 1147 ssize_t ret, len; 1148 unsigned int out_num = elem.out_num; 1149 struct iovec *out_sg = &elem.out_sg[0]; 1150 struct iovec sg[VIRTQUEUE_MAX_SIZE]; 1151 1152 if (out_num < 1) { 1153 error_report("virtio-net header not in first element"); 1154 exit(1); 1155 } 1156 1157 if (n->has_vnet_hdr) { 1158 if (out_sg[0].iov_len < n->guest_hdr_len) { 1159 error_report("virtio-net header incorrect"); 1160 exit(1); 1161 } 1162 virtio_net_hdr_swap(vdev, (void *) out_sg[0].iov_base); 1163 } 1164 1165 /* 1166 * If host wants to see the guest header as is, we can 1167 * pass it on unchanged. Otherwise, copy just the parts 1168 * that host is interested in. 1169 */ 1170 assert(n->host_hdr_len <= n->guest_hdr_len); 1171 if (n->host_hdr_len != n->guest_hdr_len) { 1172 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 1173 out_sg, out_num, 1174 0, n->host_hdr_len); 1175 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 1176 out_sg, out_num, 1177 n->guest_hdr_len, -1); 1178 out_num = sg_num; 1179 out_sg = sg; 1180 } 1181 1182 len = n->guest_hdr_len; 1183 1184 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 1185 out_sg, out_num, virtio_net_tx_complete); 1186 if (ret == 0) { 1187 virtio_queue_set_notification(q->tx_vq, 0); 1188 q->async_tx.elem = elem; 1189 q->async_tx.len = len; 1190 return -EBUSY; 1191 } 1192 1193 len += ret; 1194 1195 virtqueue_push(q->tx_vq, &elem, 0); 1196 virtio_notify(vdev, q->tx_vq); 1197 1198 if (++num_packets >= n->tx_burst) { 1199 break; 1200 } 1201 } 1202 return num_packets; 1203 } 1204 1205 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 1206 { 1207 VirtIONet *n = VIRTIO_NET(vdev); 1208 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1209 1210 /* This happens when device was stopped but VCPU wasn't. */ 1211 if (!vdev->vm_running) { 1212 q->tx_waiting = 1; 1213 return; 1214 } 1215 1216 if (q->tx_waiting) { 1217 virtio_queue_set_notification(vq, 1); 1218 timer_del(q->tx_timer); 1219 q->tx_waiting = 0; 1220 virtio_net_flush_tx(q); 1221 } else { 1222 timer_mod(q->tx_timer, 1223 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 1224 q->tx_waiting = 1; 1225 virtio_queue_set_notification(vq, 0); 1226 } 1227 } 1228 1229 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 1230 { 1231 VirtIONet *n = VIRTIO_NET(vdev); 1232 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1233 1234 if (unlikely(q->tx_waiting)) { 1235 return; 1236 } 1237 q->tx_waiting = 1; 1238 /* This happens when device was stopped but VCPU wasn't. */ 1239 if (!vdev->vm_running) { 1240 return; 1241 } 1242 virtio_queue_set_notification(vq, 0); 1243 qemu_bh_schedule(q->tx_bh); 1244 } 1245 1246 static void virtio_net_tx_timer(void *opaque) 1247 { 1248 VirtIONetQueue *q = opaque; 1249 VirtIONet *n = q->n; 1250 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1251 /* This happens when device was stopped but BH wasn't. */ 1252 if (!vdev->vm_running) { 1253 /* Make sure tx waiting is set, so we'll run when restarted. */ 1254 assert(q->tx_waiting); 1255 return; 1256 } 1257 1258 q->tx_waiting = 0; 1259 1260 /* Just in case the driver is not ready on more */ 1261 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1262 return; 1263 } 1264 1265 virtio_queue_set_notification(q->tx_vq, 1); 1266 virtio_net_flush_tx(q); 1267 } 1268 1269 static void virtio_net_tx_bh(void *opaque) 1270 { 1271 VirtIONetQueue *q = opaque; 1272 VirtIONet *n = q->n; 1273 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1274 int32_t ret; 1275 1276 /* This happens when device was stopped but BH wasn't. */ 1277 if (!vdev->vm_running) { 1278 /* Make sure tx waiting is set, so we'll run when restarted. */ 1279 assert(q->tx_waiting); 1280 return; 1281 } 1282 1283 q->tx_waiting = 0; 1284 1285 /* Just in case the driver is not ready on more */ 1286 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 1287 return; 1288 } 1289 1290 ret = virtio_net_flush_tx(q); 1291 if (ret == -EBUSY) { 1292 return; /* Notification re-enable handled by tx_complete */ 1293 } 1294 1295 /* If we flush a full burst of packets, assume there are 1296 * more coming and immediately reschedule */ 1297 if (ret >= n->tx_burst) { 1298 qemu_bh_schedule(q->tx_bh); 1299 q->tx_waiting = 1; 1300 return; 1301 } 1302 1303 /* If less than a full burst, re-enable notification and flush 1304 * anything that may have come in while we weren't looking. If 1305 * we find something, assume the guest is still active and reschedule */ 1306 virtio_queue_set_notification(q->tx_vq, 1); 1307 if (virtio_net_flush_tx(q) > 0) { 1308 virtio_queue_set_notification(q->tx_vq, 0); 1309 qemu_bh_schedule(q->tx_bh); 1310 q->tx_waiting = 1; 1311 } 1312 } 1313 1314 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 1315 { 1316 n->multiqueue = multiqueue; 1317 1318 virtio_net_set_queues(n); 1319 } 1320 1321 static void virtio_net_save(QEMUFile *f, void *opaque) 1322 { 1323 VirtIONet *n = opaque; 1324 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1325 1326 /* At this point, backend must be stopped, otherwise 1327 * it might keep writing to memory. */ 1328 assert(!n->vhost_started); 1329 virtio_save(vdev, f); 1330 } 1331 1332 static void virtio_net_save_device(VirtIODevice *vdev, QEMUFile *f) 1333 { 1334 VirtIONet *n = VIRTIO_NET(vdev); 1335 int i; 1336 1337 qemu_put_buffer(f, n->mac, ETH_ALEN); 1338 qemu_put_be32(f, n->vqs[0].tx_waiting); 1339 qemu_put_be32(f, n->mergeable_rx_bufs); 1340 qemu_put_be16(f, n->status); 1341 qemu_put_byte(f, n->promisc); 1342 qemu_put_byte(f, n->allmulti); 1343 qemu_put_be32(f, n->mac_table.in_use); 1344 qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN); 1345 qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3); 1346 qemu_put_be32(f, n->has_vnet_hdr); 1347 qemu_put_byte(f, n->mac_table.multi_overflow); 1348 qemu_put_byte(f, n->mac_table.uni_overflow); 1349 qemu_put_byte(f, n->alluni); 1350 qemu_put_byte(f, n->nomulti); 1351 qemu_put_byte(f, n->nouni); 1352 qemu_put_byte(f, n->nobcast); 1353 qemu_put_byte(f, n->has_ufo); 1354 if (n->max_queues > 1) { 1355 qemu_put_be16(f, n->max_queues); 1356 qemu_put_be16(f, n->curr_queues); 1357 for (i = 1; i < n->curr_queues; i++) { 1358 qemu_put_be32(f, n->vqs[i].tx_waiting); 1359 } 1360 } 1361 1362 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 1363 qemu_put_be64(f, n->curr_guest_offloads); 1364 } 1365 } 1366 1367 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id) 1368 { 1369 VirtIONet *n = opaque; 1370 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1371 1372 if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION) 1373 return -EINVAL; 1374 1375 return virtio_load(vdev, f, version_id); 1376 } 1377 1378 static int virtio_net_load_device(VirtIODevice *vdev, QEMUFile *f, 1379 int version_id) 1380 { 1381 VirtIONet *n = VIRTIO_NET(vdev); 1382 int i, link_down; 1383 1384 qemu_get_buffer(f, n->mac, ETH_ALEN); 1385 n->vqs[0].tx_waiting = qemu_get_be32(f); 1386 1387 virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f), 1388 virtio_has_feature(vdev, VIRTIO_F_VERSION_1)); 1389 1390 if (version_id >= 3) 1391 n->status = qemu_get_be16(f); 1392 1393 if (version_id >= 4) { 1394 if (version_id < 8) { 1395 n->promisc = qemu_get_be32(f); 1396 n->allmulti = qemu_get_be32(f); 1397 } else { 1398 n->promisc = qemu_get_byte(f); 1399 n->allmulti = qemu_get_byte(f); 1400 } 1401 } 1402 1403 if (version_id >= 5) { 1404 n->mac_table.in_use = qemu_get_be32(f); 1405 /* MAC_TABLE_ENTRIES may be different from the saved image */ 1406 if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) { 1407 qemu_get_buffer(f, n->mac_table.macs, 1408 n->mac_table.in_use * ETH_ALEN); 1409 } else { 1410 int64_t i; 1411 1412 /* Overflow detected - can happen if source has a larger MAC table. 1413 * We simply set overflow flag so there's no need to maintain the 1414 * table of addresses, discard them all. 1415 * Note: 64 bit math to avoid integer overflow. 1416 */ 1417 for (i = 0; i < (int64_t)n->mac_table.in_use * ETH_ALEN; ++i) { 1418 qemu_get_byte(f); 1419 } 1420 n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1; 1421 n->mac_table.in_use = 0; 1422 } 1423 } 1424 1425 if (version_id >= 6) 1426 qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3); 1427 1428 if (version_id >= 7) { 1429 if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) { 1430 error_report("virtio-net: saved image requires vnet_hdr=on"); 1431 return -1; 1432 } 1433 } 1434 1435 if (version_id >= 9) { 1436 n->mac_table.multi_overflow = qemu_get_byte(f); 1437 n->mac_table.uni_overflow = qemu_get_byte(f); 1438 } 1439 1440 if (version_id >= 10) { 1441 n->alluni = qemu_get_byte(f); 1442 n->nomulti = qemu_get_byte(f); 1443 n->nouni = qemu_get_byte(f); 1444 n->nobcast = qemu_get_byte(f); 1445 } 1446 1447 if (version_id >= 11) { 1448 if (qemu_get_byte(f) && !peer_has_ufo(n)) { 1449 error_report("virtio-net: saved image requires TUN_F_UFO support"); 1450 return -1; 1451 } 1452 } 1453 1454 if (n->max_queues > 1) { 1455 if (n->max_queues != qemu_get_be16(f)) { 1456 error_report("virtio-net: different max_queues "); 1457 return -1; 1458 } 1459 1460 n->curr_queues = qemu_get_be16(f); 1461 if (n->curr_queues > n->max_queues) { 1462 error_report("virtio-net: curr_queues %x > max_queues %x", 1463 n->curr_queues, n->max_queues); 1464 return -1; 1465 } 1466 for (i = 1; i < n->curr_queues; i++) { 1467 n->vqs[i].tx_waiting = qemu_get_be32(f); 1468 } 1469 } 1470 1471 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 1472 n->curr_guest_offloads = qemu_get_be64(f); 1473 } else { 1474 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 1475 } 1476 1477 if (peer_has_vnet_hdr(n)) { 1478 virtio_net_apply_guest_offloads(n); 1479 } 1480 1481 virtio_net_set_queues(n); 1482 1483 /* Find the first multicast entry in the saved MAC filter */ 1484 for (i = 0; i < n->mac_table.in_use; i++) { 1485 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 1486 break; 1487 } 1488 } 1489 n->mac_table.first_multi = i; 1490 1491 /* nc.link_down can't be migrated, so infer link_down according 1492 * to link status bit in n->status */ 1493 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 1494 for (i = 0; i < n->max_queues; i++) { 1495 qemu_get_subqueue(n->nic, i)->link_down = link_down; 1496 } 1497 1498 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 1499 virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 1500 n->announce_counter = SELF_ANNOUNCE_ROUNDS; 1501 timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL)); 1502 } 1503 1504 return 0; 1505 } 1506 1507 static NetClientInfo net_virtio_info = { 1508 .type = NET_CLIENT_OPTIONS_KIND_NIC, 1509 .size = sizeof(NICState), 1510 .can_receive = virtio_net_can_receive, 1511 .receive = virtio_net_receive, 1512 .link_status_changed = virtio_net_set_link_status, 1513 .query_rx_filter = virtio_net_query_rxfilter, 1514 }; 1515 1516 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 1517 { 1518 VirtIONet *n = VIRTIO_NET(vdev); 1519 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1520 assert(n->vhost_started); 1521 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx); 1522 } 1523 1524 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 1525 bool mask) 1526 { 1527 VirtIONet *n = VIRTIO_NET(vdev); 1528 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1529 assert(n->vhost_started); 1530 vhost_net_virtqueue_mask(get_vhost_net(nc->peer), 1531 vdev, idx, mask); 1532 } 1533 1534 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features) 1535 { 1536 int i, config_size = 0; 1537 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC); 1538 for (i = 0; feature_sizes[i].flags != 0; i++) { 1539 if (host_features & feature_sizes[i].flags) { 1540 config_size = MAX(feature_sizes[i].end, config_size); 1541 } 1542 } 1543 n->config_size = config_size; 1544 } 1545 1546 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 1547 const char *type) 1548 { 1549 /* 1550 * The name can be NULL, the netclient name will be type.x. 1551 */ 1552 assert(type != NULL); 1553 1554 g_free(n->netclient_name); 1555 g_free(n->netclient_type); 1556 n->netclient_name = g_strdup(name); 1557 n->netclient_type = g_strdup(type); 1558 } 1559 1560 static void virtio_net_device_realize(DeviceState *dev, Error **errp) 1561 { 1562 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1563 VirtIONet *n = VIRTIO_NET(dev); 1564 NetClientState *nc; 1565 int i; 1566 1567 virtio_net_set_config_size(n, n->host_features); 1568 virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size); 1569 1570 n->max_queues = MAX(n->nic_conf.peers.queues, 1); 1571 if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) { 1572 error_setg(errp, "Invalid number of queues (= %" PRIu32 "), " 1573 "must be a positive integer less than %d.", 1574 n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2); 1575 virtio_cleanup(vdev); 1576 return; 1577 } 1578 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues); 1579 n->curr_queues = 1; 1580 n->tx_timeout = n->net_conf.txtimer; 1581 1582 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 1583 && strcmp(n->net_conf.tx, "bh")) { 1584 error_report("virtio-net: " 1585 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 1586 n->net_conf.tx); 1587 error_report("Defaulting to \"bh\""); 1588 } 1589 1590 for (i = 0; i < n->max_queues; i++) { 1591 n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx); 1592 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 1593 n->vqs[i].tx_vq = 1594 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer); 1595 n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 1596 virtio_net_tx_timer, 1597 &n->vqs[i]); 1598 } else { 1599 n->vqs[i].tx_vq = 1600 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh); 1601 n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]); 1602 } 1603 1604 n->vqs[i].tx_waiting = 0; 1605 n->vqs[i].n = n; 1606 } 1607 1608 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1609 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 1610 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 1611 n->status = VIRTIO_NET_S_LINK_UP; 1612 n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, 1613 virtio_net_announce_timer, n); 1614 1615 if (n->netclient_type) { 1616 /* 1617 * Happen when virtio_net_set_netclient_name has been called. 1618 */ 1619 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 1620 n->netclient_type, n->netclient_name, n); 1621 } else { 1622 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 1623 object_get_typename(OBJECT(dev)), dev->id, n); 1624 } 1625 1626 peer_test_vnet_hdr(n); 1627 if (peer_has_vnet_hdr(n)) { 1628 for (i = 0; i < n->max_queues; i++) { 1629 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true); 1630 } 1631 n->host_hdr_len = sizeof(struct virtio_net_hdr); 1632 } else { 1633 n->host_hdr_len = 0; 1634 } 1635 1636 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 1637 1638 n->vqs[0].tx_waiting = 0; 1639 n->tx_burst = n->net_conf.txburst; 1640 virtio_net_set_mrg_rx_bufs(n, 0, 0); 1641 n->promisc = 1; /* for compatibility */ 1642 1643 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 1644 1645 n->vlans = g_malloc0(MAX_VLAN >> 3); 1646 1647 nc = qemu_get_queue(n->nic); 1648 nc->rxfilter_notify_enabled = 1; 1649 1650 n->qdev = dev; 1651 register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION, 1652 virtio_net_save, virtio_net_load, n); 1653 } 1654 1655 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp) 1656 { 1657 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1658 VirtIONet *n = VIRTIO_NET(dev); 1659 int i; 1660 1661 /* This will stop vhost backend if appropriate. */ 1662 virtio_net_set_status(vdev, 0); 1663 1664 unregister_savevm(dev, "virtio-net", n); 1665 1666 g_free(n->netclient_name); 1667 n->netclient_name = NULL; 1668 g_free(n->netclient_type); 1669 n->netclient_type = NULL; 1670 1671 g_free(n->mac_table.macs); 1672 g_free(n->vlans); 1673 1674 for (i = 0; i < n->max_queues; i++) { 1675 VirtIONetQueue *q = &n->vqs[i]; 1676 NetClientState *nc = qemu_get_subqueue(n->nic, i); 1677 1678 qemu_purge_queued_packets(nc); 1679 1680 if (q->tx_timer) { 1681 timer_del(q->tx_timer); 1682 timer_free(q->tx_timer); 1683 } else if (q->tx_bh) { 1684 qemu_bh_delete(q->tx_bh); 1685 } 1686 } 1687 1688 timer_del(n->announce_timer); 1689 timer_free(n->announce_timer); 1690 g_free(n->vqs); 1691 qemu_del_nic(n->nic); 1692 virtio_cleanup(vdev); 1693 } 1694 1695 static void virtio_net_instance_init(Object *obj) 1696 { 1697 VirtIONet *n = VIRTIO_NET(obj); 1698 1699 /* 1700 * The default config_size is sizeof(struct virtio_net_config). 1701 * Can be overriden with virtio_net_set_config_size. 1702 */ 1703 n->config_size = sizeof(struct virtio_net_config); 1704 device_add_bootindex_property(obj, &n->nic_conf.bootindex, 1705 "bootindex", "/ethernet-phy@0", 1706 DEVICE(n), NULL); 1707 } 1708 1709 static Property virtio_net_properties[] = { 1710 DEFINE_PROP_BIT("any_layout", VirtIONet, host_features, 1711 VIRTIO_F_ANY_LAYOUT, true), 1712 DEFINE_PROP_BIT("csum", VirtIONet, host_features, VIRTIO_NET_F_CSUM, true), 1713 DEFINE_PROP_BIT("guest_csum", VirtIONet, host_features, 1714 VIRTIO_NET_F_GUEST_CSUM, true), 1715 DEFINE_PROP_BIT("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true), 1716 DEFINE_PROP_BIT("guest_tso4", VirtIONet, host_features, 1717 VIRTIO_NET_F_GUEST_TSO4, true), 1718 DEFINE_PROP_BIT("guest_tso6", VirtIONet, host_features, 1719 VIRTIO_NET_F_GUEST_TSO6, true), 1720 DEFINE_PROP_BIT("guest_ecn", VirtIONet, host_features, 1721 VIRTIO_NET_F_GUEST_ECN, true), 1722 DEFINE_PROP_BIT("guest_ufo", VirtIONet, host_features, 1723 VIRTIO_NET_F_GUEST_UFO, true), 1724 DEFINE_PROP_BIT("guest_announce", VirtIONet, host_features, 1725 VIRTIO_NET_F_GUEST_ANNOUNCE, true), 1726 DEFINE_PROP_BIT("host_tso4", VirtIONet, host_features, 1727 VIRTIO_NET_F_HOST_TSO4, true), 1728 DEFINE_PROP_BIT("host_tso6", VirtIONet, host_features, 1729 VIRTIO_NET_F_HOST_TSO6, true), 1730 DEFINE_PROP_BIT("host_ecn", VirtIONet, host_features, 1731 VIRTIO_NET_F_HOST_ECN, true), 1732 DEFINE_PROP_BIT("host_ufo", VirtIONet, host_features, 1733 VIRTIO_NET_F_HOST_UFO, true), 1734 DEFINE_PROP_BIT("mrg_rxbuf", VirtIONet, host_features, 1735 VIRTIO_NET_F_MRG_RXBUF, true), 1736 DEFINE_PROP_BIT("status", VirtIONet, host_features, 1737 VIRTIO_NET_F_STATUS, true), 1738 DEFINE_PROP_BIT("ctrl_vq", VirtIONet, host_features, 1739 VIRTIO_NET_F_CTRL_VQ, true), 1740 DEFINE_PROP_BIT("ctrl_rx", VirtIONet, host_features, 1741 VIRTIO_NET_F_CTRL_RX, true), 1742 DEFINE_PROP_BIT("ctrl_vlan", VirtIONet, host_features, 1743 VIRTIO_NET_F_CTRL_VLAN, true), 1744 DEFINE_PROP_BIT("ctrl_rx_extra", VirtIONet, host_features, 1745 VIRTIO_NET_F_CTRL_RX_EXTRA, true), 1746 DEFINE_PROP_BIT("ctrl_mac_addr", VirtIONet, host_features, 1747 VIRTIO_NET_F_CTRL_MAC_ADDR, true), 1748 DEFINE_PROP_BIT("ctrl_guest_offloads", VirtIONet, host_features, 1749 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true), 1750 DEFINE_PROP_BIT("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false), 1751 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 1752 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 1753 TX_TIMER_INTERVAL), 1754 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 1755 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 1756 DEFINE_PROP_END_OF_LIST(), 1757 }; 1758 1759 static void virtio_net_class_init(ObjectClass *klass, void *data) 1760 { 1761 DeviceClass *dc = DEVICE_CLASS(klass); 1762 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 1763 1764 dc->props = virtio_net_properties; 1765 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 1766 vdc->realize = virtio_net_device_realize; 1767 vdc->unrealize = virtio_net_device_unrealize; 1768 vdc->get_config = virtio_net_get_config; 1769 vdc->set_config = virtio_net_set_config; 1770 vdc->get_features = virtio_net_get_features; 1771 vdc->set_features = virtio_net_set_features; 1772 vdc->bad_features = virtio_net_bad_features; 1773 vdc->reset = virtio_net_reset; 1774 vdc->set_status = virtio_net_set_status; 1775 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 1776 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 1777 vdc->load = virtio_net_load_device; 1778 vdc->save = virtio_net_save_device; 1779 } 1780 1781 static const TypeInfo virtio_net_info = { 1782 .name = TYPE_VIRTIO_NET, 1783 .parent = TYPE_VIRTIO_DEVICE, 1784 .instance_size = sizeof(VirtIONet), 1785 .instance_init = virtio_net_instance_init, 1786 .class_init = virtio_net_class_init, 1787 }; 1788 1789 static void virtio_register_types(void) 1790 { 1791 type_register_static(&virtio_net_info); 1792 } 1793 1794 type_init(virtio_register_types) 1795