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