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 uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features) 439 { 440 VirtIONet *n = VIRTIO_NET(vdev); 441 NetClientState *nc = qemu_get_queue(n->nic); 442 443 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 444 445 if (!peer_has_vnet_hdr(n)) { 446 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM); 447 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4); 448 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6); 449 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN); 450 451 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM); 452 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4); 453 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6); 454 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN); 455 } 456 457 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 458 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO); 459 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO); 460 } 461 462 if (!get_vhost_net(nc->peer)) { 463 return features; 464 } 465 return vhost_net_get_features(get_vhost_net(nc->peer), features); 466 } 467 468 static uint32_t virtio_net_bad_features(VirtIODevice *vdev) 469 { 470 uint32_t features = 0; 471 472 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 473 * but also these: */ 474 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 475 virtio_add_feature(&features, VIRTIO_NET_F_CSUM); 476 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4); 477 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6); 478 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN); 479 480 return features; 481 } 482 483 static void virtio_net_apply_guest_offloads(VirtIONet *n) 484 { 485 qemu_set_offload(qemu_get_queue(n->nic)->peer, 486 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 487 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 488 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 489 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 490 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO))); 491 } 492 493 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features) 494 { 495 static const uint64_t guest_offloads_mask = 496 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 497 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 498 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 499 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 500 (1ULL << VIRTIO_NET_F_GUEST_UFO); 501 502 return guest_offloads_mask & features; 503 } 504 505 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n) 506 { 507 VirtIODevice *vdev = VIRTIO_DEVICE(n); 508 return virtio_net_guest_offloads_by_features(vdev->guest_features); 509 } 510 511 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features) 512 { 513 VirtIONet *n = VIRTIO_NET(vdev); 514 int i; 515 516 virtio_net_set_multiqueue(n, 517 __virtio_has_feature(features, VIRTIO_NET_F_MQ)); 518 519 virtio_net_set_mrg_rx_bufs(n, 520 __virtio_has_feature(features, 521 VIRTIO_NET_F_MRG_RXBUF)); 522 523 if (n->has_vnet_hdr) { 524 n->curr_guest_offloads = 525 virtio_net_guest_offloads_by_features(features); 526 virtio_net_apply_guest_offloads(n); 527 } 528 529 for (i = 0; i < n->max_queues; i++) { 530 NetClientState *nc = qemu_get_subqueue(n->nic, i); 531 532 if (!get_vhost_net(nc->peer)) { 533 continue; 534 } 535 vhost_net_ack_features(get_vhost_net(nc->peer), features); 536 } 537 538 if (__virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) { 539 memset(n->vlans, 0, MAX_VLAN >> 3); 540 } else { 541 memset(n->vlans, 0xff, MAX_VLAN >> 3); 542 } 543 } 544 545 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 546 struct iovec *iov, unsigned int iov_cnt) 547 { 548 uint8_t on; 549 size_t s; 550 NetClientState *nc = qemu_get_queue(n->nic); 551 552 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 553 if (s != sizeof(on)) { 554 return VIRTIO_NET_ERR; 555 } 556 557 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 558 n->promisc = on; 559 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 560 n->allmulti = on; 561 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 562 n->alluni = on; 563 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 564 n->nomulti = on; 565 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 566 n->nouni = on; 567 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 568 n->nobcast = on; 569 } else { 570 return VIRTIO_NET_ERR; 571 } 572 573 rxfilter_notify(nc); 574 575 return VIRTIO_NET_OK; 576 } 577 578 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 579 struct iovec *iov, unsigned int iov_cnt) 580 { 581 VirtIODevice *vdev = VIRTIO_DEVICE(n); 582 uint64_t offloads; 583 size_t s; 584 585 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 586 return VIRTIO_NET_ERR; 587 } 588 589 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 590 if (s != sizeof(offloads)) { 591 return VIRTIO_NET_ERR; 592 } 593 594 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 595 uint64_t supported_offloads; 596 597 if (!n->has_vnet_hdr) { 598 return VIRTIO_NET_ERR; 599 } 600 601 supported_offloads = virtio_net_supported_guest_offloads(n); 602 if (offloads & ~supported_offloads) { 603 return VIRTIO_NET_ERR; 604 } 605 606 n->curr_guest_offloads = offloads; 607 virtio_net_apply_guest_offloads(n); 608 609 return VIRTIO_NET_OK; 610 } else { 611 return VIRTIO_NET_ERR; 612 } 613 } 614 615 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 616 struct iovec *iov, unsigned int iov_cnt) 617 { 618 VirtIODevice *vdev = VIRTIO_DEVICE(n); 619 struct virtio_net_ctrl_mac mac_data; 620 size_t s; 621 NetClientState *nc = qemu_get_queue(n->nic); 622 623 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 624 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 625 return VIRTIO_NET_ERR; 626 } 627 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 628 assert(s == sizeof(n->mac)); 629 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 630 rxfilter_notify(nc); 631 632 return VIRTIO_NET_OK; 633 } 634 635 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 636 return VIRTIO_NET_ERR; 637 } 638 639 int in_use = 0; 640 int first_multi = 0; 641 uint8_t uni_overflow = 0; 642 uint8_t multi_overflow = 0; 643 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 644 645 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 646 sizeof(mac_data.entries)); 647 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 648 if (s != sizeof(mac_data.entries)) { 649 goto error; 650 } 651 iov_discard_front(&iov, &iov_cnt, s); 652 653 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 654 goto error; 655 } 656 657 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 658 s = iov_to_buf(iov, iov_cnt, 0, macs, 659 mac_data.entries * ETH_ALEN); 660 if (s != mac_data.entries * ETH_ALEN) { 661 goto error; 662 } 663 in_use += mac_data.entries; 664 } else { 665 uni_overflow = 1; 666 } 667 668 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 669 670 first_multi = in_use; 671 672 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 673 sizeof(mac_data.entries)); 674 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 675 if (s != sizeof(mac_data.entries)) { 676 goto error; 677 } 678 679 iov_discard_front(&iov, &iov_cnt, s); 680 681 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 682 goto error; 683 } 684 685 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) { 686 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN], 687 mac_data.entries * ETH_ALEN); 688 if (s != mac_data.entries * ETH_ALEN) { 689 goto error; 690 } 691 in_use += mac_data.entries; 692 } else { 693 multi_overflow = 1; 694 } 695 696 n->mac_table.in_use = in_use; 697 n->mac_table.first_multi = first_multi; 698 n->mac_table.uni_overflow = uni_overflow; 699 n->mac_table.multi_overflow = multi_overflow; 700 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN); 701 g_free(macs); 702 rxfilter_notify(nc); 703 704 return VIRTIO_NET_OK; 705 706 error: 707 g_free(macs); 708 return VIRTIO_NET_ERR; 709 } 710 711 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 712 struct iovec *iov, unsigned int iov_cnt) 713 { 714 VirtIODevice *vdev = VIRTIO_DEVICE(n); 715 uint16_t vid; 716 size_t s; 717 NetClientState *nc = qemu_get_queue(n->nic); 718 719 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 720 vid = virtio_lduw_p(vdev, &vid); 721 if (s != sizeof(vid)) { 722 return VIRTIO_NET_ERR; 723 } 724 725 if (vid >= MAX_VLAN) 726 return VIRTIO_NET_ERR; 727 728 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 729 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 730 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 731 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 732 else 733 return VIRTIO_NET_ERR; 734 735 rxfilter_notify(nc); 736 737 return VIRTIO_NET_OK; 738 } 739 740 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd, 741 struct iovec *iov, unsigned int iov_cnt) 742 { 743 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK && 744 n->status & VIRTIO_NET_S_ANNOUNCE) { 745 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 746 if (n->announce_counter) { 747 timer_mod(n->announce_timer, 748 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 749 self_announce_delay(n->announce_counter)); 750 } 751 return VIRTIO_NET_OK; 752 } else { 753 return VIRTIO_NET_ERR; 754 } 755 } 756 757 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 758 struct iovec *iov, unsigned int iov_cnt) 759 { 760 VirtIODevice *vdev = VIRTIO_DEVICE(n); 761 struct virtio_net_ctrl_mq mq; 762 size_t s; 763 uint16_t queues; 764 765 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 766 if (s != sizeof(mq)) { 767 return VIRTIO_NET_ERR; 768 } 769 770 if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 771 return VIRTIO_NET_ERR; 772 } 773 774 queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs); 775 776 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 777 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 778 queues > n->max_queues || 779 !n->multiqueue) { 780 return VIRTIO_NET_ERR; 781 } 782 783 n->curr_queues = queues; 784 /* stop the backend before changing the number of queues to avoid handling a 785 * disabled queue */ 786 virtio_net_set_status(vdev, vdev->status); 787 virtio_net_set_queues(n); 788 789 return VIRTIO_NET_OK; 790 } 791 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 792 { 793 VirtIONet *n = VIRTIO_NET(vdev); 794 struct virtio_net_ctrl_hdr ctrl; 795 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 796 VirtQueueElement elem; 797 size_t s; 798 struct iovec *iov, *iov2; 799 unsigned int iov_cnt; 800 801 while (virtqueue_pop(vq, &elem)) { 802 if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) || 803 iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) { 804 error_report("virtio-net ctrl missing headers"); 805 exit(1); 806 } 807 808 iov_cnt = elem.out_num; 809 iov2 = iov = g_memdup(elem.out_sg, sizeof(struct iovec) * elem.out_num); 810 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl)); 811 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl)); 812 if (s != sizeof(ctrl)) { 813 status = VIRTIO_NET_ERR; 814 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 815 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt); 816 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 817 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt); 818 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 819 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt); 820 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) { 821 status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt); 822 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 823 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt); 824 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 825 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt); 826 } 827 828 s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status)); 829 assert(s == sizeof(status)); 830 831 virtqueue_push(vq, &elem, sizeof(status)); 832 virtio_notify(vdev, vq); 833 g_free(iov2); 834 } 835 } 836 837 /* RX */ 838 839 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 840 { 841 VirtIONet *n = VIRTIO_NET(vdev); 842 int queue_index = vq2q(virtio_get_queue_index(vq)); 843 844 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 845 } 846 847 static int virtio_net_can_receive(NetClientState *nc) 848 { 849 VirtIONet *n = qemu_get_nic_opaque(nc); 850 VirtIODevice *vdev = VIRTIO_DEVICE(n); 851 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 852 853 if (!vdev->vm_running) { 854 return 0; 855 } 856 857 if (nc->queue_index >= n->curr_queues) { 858 return 0; 859 } 860 861 if (!virtio_queue_ready(q->rx_vq) || 862 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 863 return 0; 864 } 865 866 return 1; 867 } 868 869 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 870 { 871 VirtIONet *n = q->n; 872 if (virtio_queue_empty(q->rx_vq) || 873 (n->mergeable_rx_bufs && 874 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 875 virtio_queue_set_notification(q->rx_vq, 1); 876 877 /* To avoid a race condition where the guest has made some buffers 878 * available after the above check but before notification was 879 * enabled, check for available buffers again. 880 */ 881 if (virtio_queue_empty(q->rx_vq) || 882 (n->mergeable_rx_bufs && 883 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 884 return 0; 885 } 886 } 887 888 virtio_queue_set_notification(q->rx_vq, 0); 889 return 1; 890 } 891 892 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr) 893 { 894 virtio_tswap16s(vdev, &hdr->hdr_len); 895 virtio_tswap16s(vdev, &hdr->gso_size); 896 virtio_tswap16s(vdev, &hdr->csum_start); 897 virtio_tswap16s(vdev, &hdr->csum_offset); 898 } 899 900 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 901 * it never finds out that the packets don't have valid checksums. This 902 * causes dhclient to get upset. Fedora's carried a patch for ages to 903 * fix this with Xen but it hasn't appeared in an upstream release of 904 * dhclient yet. 905 * 906 * To avoid breaking existing guests, we catch udp packets and add 907 * checksums. This is terrible but it's better than hacking the guest 908 * kernels. 909 * 910 * N.B. if we introduce a zero-copy API, this operation is no longer free so 911 * we should provide a mechanism to disable it to avoid polluting the host 912 * cache. 913 */ 914 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 915 uint8_t *buf, size_t size) 916 { 917 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 918 (size > 27 && size < 1500) && /* normal sized MTU */ 919 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 920 (buf[23] == 17) && /* ip.protocol == UDP */ 921 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 922 net_checksum_calculate(buf, size); 923 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 924 } 925 } 926 927 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt, 928 const void *buf, size_t size) 929 { 930 if (n->has_vnet_hdr) { 931 /* FIXME this cast is evil */ 932 void *wbuf = (void *)buf; 933 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, 934 size - n->host_hdr_len); 935 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf); 936 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr)); 937 } else { 938 struct virtio_net_hdr hdr = { 939 .flags = 0, 940 .gso_type = VIRTIO_NET_HDR_GSO_NONE 941 }; 942 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr); 943 } 944 } 945 946 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 947 { 948 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 949 static const uint8_t vlan[] = {0x81, 0x00}; 950 uint8_t *ptr = (uint8_t *)buf; 951 int i; 952 953 if (n->promisc) 954 return 1; 955 956 ptr += n->host_hdr_len; 957 958 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 959 int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff; 960 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 961 return 0; 962 } 963 964 if (ptr[0] & 1) { // multicast 965 if (!memcmp(ptr, bcast, sizeof(bcast))) { 966 return !n->nobcast; 967 } else if (n->nomulti) { 968 return 0; 969 } else if (n->allmulti || n->mac_table.multi_overflow) { 970 return 1; 971 } 972 973 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 974 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 975 return 1; 976 } 977 } 978 } else { // unicast 979 if (n->nouni) { 980 return 0; 981 } else if (n->alluni || n->mac_table.uni_overflow) { 982 return 1; 983 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 984 return 1; 985 } 986 987 for (i = 0; i < n->mac_table.first_multi; i++) { 988 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 989 return 1; 990 } 991 } 992 } 993 994 return 0; 995 } 996 997 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size) 998 { 999 VirtIONet *n = qemu_get_nic_opaque(nc); 1000 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1001 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1002 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 1003 struct virtio_net_hdr_mrg_rxbuf mhdr; 1004 unsigned mhdr_cnt = 0; 1005 size_t offset, i, guest_offset; 1006 1007 if (!virtio_net_can_receive(nc)) { 1008 return -1; 1009 } 1010 1011 /* hdr_len refers to the header we supply to the guest */ 1012 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 1013 return 0; 1014 } 1015 1016 if (!receive_filter(n, buf, size)) 1017 return size; 1018 1019 offset = i = 0; 1020 1021 while (offset < size) { 1022 VirtQueueElement elem; 1023 int len, total; 1024 const struct iovec *sg = elem.in_sg; 1025 1026 total = 0; 1027 1028 if (virtqueue_pop(q->rx_vq, &elem) == 0) { 1029 if (i == 0) 1030 return -1; 1031 error_report("virtio-net unexpected empty queue: " 1032 "i %zd mergeable %d offset %zd, size %zd, " 1033 "guest hdr len %zd, host hdr len %zd guest features 0x%x", 1034 i, n->mergeable_rx_bufs, offset, size, 1035 n->guest_hdr_len, n->host_hdr_len, vdev->guest_features); 1036 exit(1); 1037 } 1038 1039 if (elem.in_num < 1) { 1040 error_report("virtio-net receive queue contains no in buffers"); 1041 exit(1); 1042 } 1043 1044 if (i == 0) { 1045 assert(offset == 0); 1046 if (n->mergeable_rx_bufs) { 1047 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 1048 sg, elem.in_num, 1049 offsetof(typeof(mhdr), num_buffers), 1050 sizeof(mhdr.num_buffers)); 1051 } 1052 1053 receive_header(n, sg, elem.in_num, buf, size); 1054 offset = n->host_hdr_len; 1055 total += n->guest_hdr_len; 1056 guest_offset = n->guest_hdr_len; 1057 } else { 1058 guest_offset = 0; 1059 } 1060 1061 /* copy in packet. ugh */ 1062 len = iov_from_buf(sg, elem.in_num, guest_offset, 1063 buf + offset, size - offset); 1064 total += len; 1065 offset += len; 1066 /* If buffers can't be merged, at this point we 1067 * must have consumed the complete packet. 1068 * Otherwise, drop it. */ 1069 if (!n->mergeable_rx_bufs && offset < size) { 1070 #if 0 1071 error_report("virtio-net truncated non-mergeable packet: " 1072 "i %zd mergeable %d offset %zd, size %zd, " 1073 "guest hdr len %zd, host hdr len %zd", 1074 i, n->mergeable_rx_bufs, 1075 offset, size, n->guest_hdr_len, n->host_hdr_len); 1076 #endif 1077 return size; 1078 } 1079 1080 /* signal other side */ 1081 virtqueue_fill(q->rx_vq, &elem, total, i++); 1082 } 1083 1084 if (mhdr_cnt) { 1085 virtio_stw_p(vdev, &mhdr.num_buffers, i); 1086 iov_from_buf(mhdr_sg, mhdr_cnt, 1087 0, 1088 &mhdr.num_buffers, sizeof mhdr.num_buffers); 1089 } 1090 1091 virtqueue_flush(q->rx_vq, i); 1092 virtio_notify(vdev, q->rx_vq); 1093 1094 return size; 1095 } 1096 1097 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 1098 1099 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 1100 { 1101 VirtIONet *n = qemu_get_nic_opaque(nc); 1102 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1103 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1104 1105 virtqueue_push(q->tx_vq, &q->async_tx.elem, 0); 1106 virtio_notify(vdev, q->tx_vq); 1107 1108 q->async_tx.elem.out_num = q->async_tx.len = 0; 1109 1110 virtio_queue_set_notification(q->tx_vq, 1); 1111 virtio_net_flush_tx(q); 1112 } 1113 1114 /* TX */ 1115 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 1116 { 1117 VirtIONet *n = q->n; 1118 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1119 VirtQueueElement elem; 1120 int32_t num_packets = 0; 1121 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 1122 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1123 return num_packets; 1124 } 1125 1126 if (q->async_tx.elem.out_num) { 1127 virtio_queue_set_notification(q->tx_vq, 0); 1128 return num_packets; 1129 } 1130 1131 while (virtqueue_pop(q->tx_vq, &elem)) { 1132 ssize_t ret, len; 1133 unsigned int out_num = elem.out_num; 1134 struct iovec *out_sg = &elem.out_sg[0]; 1135 struct iovec sg[VIRTQUEUE_MAX_SIZE]; 1136 1137 if (out_num < 1) { 1138 error_report("virtio-net header not in first element"); 1139 exit(1); 1140 } 1141 1142 if (n->has_vnet_hdr) { 1143 if (out_sg[0].iov_len < n->guest_hdr_len) { 1144 error_report("virtio-net header incorrect"); 1145 exit(1); 1146 } 1147 virtio_net_hdr_swap(vdev, (void *) out_sg[0].iov_base); 1148 } 1149 1150 /* 1151 * If host wants to see the guest header as is, we can 1152 * pass it on unchanged. Otherwise, copy just the parts 1153 * that host is interested in. 1154 */ 1155 assert(n->host_hdr_len <= n->guest_hdr_len); 1156 if (n->host_hdr_len != n->guest_hdr_len) { 1157 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 1158 out_sg, out_num, 1159 0, n->host_hdr_len); 1160 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 1161 out_sg, out_num, 1162 n->guest_hdr_len, -1); 1163 out_num = sg_num; 1164 out_sg = sg; 1165 } 1166 1167 len = n->guest_hdr_len; 1168 1169 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 1170 out_sg, out_num, virtio_net_tx_complete); 1171 if (ret == 0) { 1172 virtio_queue_set_notification(q->tx_vq, 0); 1173 q->async_tx.elem = elem; 1174 q->async_tx.len = len; 1175 return -EBUSY; 1176 } 1177 1178 len += ret; 1179 1180 virtqueue_push(q->tx_vq, &elem, 0); 1181 virtio_notify(vdev, q->tx_vq); 1182 1183 if (++num_packets >= n->tx_burst) { 1184 break; 1185 } 1186 } 1187 return num_packets; 1188 } 1189 1190 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 1191 { 1192 VirtIONet *n = VIRTIO_NET(vdev); 1193 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1194 1195 /* This happens when device was stopped but VCPU wasn't. */ 1196 if (!vdev->vm_running) { 1197 q->tx_waiting = 1; 1198 return; 1199 } 1200 1201 if (q->tx_waiting) { 1202 virtio_queue_set_notification(vq, 1); 1203 timer_del(q->tx_timer); 1204 q->tx_waiting = 0; 1205 virtio_net_flush_tx(q); 1206 } else { 1207 timer_mod(q->tx_timer, 1208 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 1209 q->tx_waiting = 1; 1210 virtio_queue_set_notification(vq, 0); 1211 } 1212 } 1213 1214 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 1215 { 1216 VirtIONet *n = VIRTIO_NET(vdev); 1217 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1218 1219 if (unlikely(q->tx_waiting)) { 1220 return; 1221 } 1222 q->tx_waiting = 1; 1223 /* This happens when device was stopped but VCPU wasn't. */ 1224 if (!vdev->vm_running) { 1225 return; 1226 } 1227 virtio_queue_set_notification(vq, 0); 1228 qemu_bh_schedule(q->tx_bh); 1229 } 1230 1231 static void virtio_net_tx_timer(void *opaque) 1232 { 1233 VirtIONetQueue *q = opaque; 1234 VirtIONet *n = q->n; 1235 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1236 /* This happens when device was stopped but BH wasn't. */ 1237 if (!vdev->vm_running) { 1238 /* Make sure tx waiting is set, so we'll run when restarted. */ 1239 assert(q->tx_waiting); 1240 return; 1241 } 1242 1243 q->tx_waiting = 0; 1244 1245 /* Just in case the driver is not ready on more */ 1246 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1247 return; 1248 } 1249 1250 virtio_queue_set_notification(q->tx_vq, 1); 1251 virtio_net_flush_tx(q); 1252 } 1253 1254 static void virtio_net_tx_bh(void *opaque) 1255 { 1256 VirtIONetQueue *q = opaque; 1257 VirtIONet *n = q->n; 1258 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1259 int32_t ret; 1260 1261 /* This happens when device was stopped but BH wasn't. */ 1262 if (!vdev->vm_running) { 1263 /* Make sure tx waiting is set, so we'll run when restarted. */ 1264 assert(q->tx_waiting); 1265 return; 1266 } 1267 1268 q->tx_waiting = 0; 1269 1270 /* Just in case the driver is not ready on more */ 1271 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 1272 return; 1273 } 1274 1275 ret = virtio_net_flush_tx(q); 1276 if (ret == -EBUSY) { 1277 return; /* Notification re-enable handled by tx_complete */ 1278 } 1279 1280 /* If we flush a full burst of packets, assume there are 1281 * more coming and immediately reschedule */ 1282 if (ret >= n->tx_burst) { 1283 qemu_bh_schedule(q->tx_bh); 1284 q->tx_waiting = 1; 1285 return; 1286 } 1287 1288 /* If less than a full burst, re-enable notification and flush 1289 * anything that may have come in while we weren't looking. If 1290 * we find something, assume the guest is still active and reschedule */ 1291 virtio_queue_set_notification(q->tx_vq, 1); 1292 if (virtio_net_flush_tx(q) > 0) { 1293 virtio_queue_set_notification(q->tx_vq, 0); 1294 qemu_bh_schedule(q->tx_bh); 1295 q->tx_waiting = 1; 1296 } 1297 } 1298 1299 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 1300 { 1301 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1302 int i, max = multiqueue ? n->max_queues : 1; 1303 1304 n->multiqueue = multiqueue; 1305 1306 for (i = 2; i <= n->max_queues * 2 + 1; i++) { 1307 virtio_del_queue(vdev, i); 1308 } 1309 1310 for (i = 1; i < max; i++) { 1311 n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx); 1312 if (n->vqs[i].tx_timer) { 1313 n->vqs[i].tx_vq = 1314 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer); 1315 n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 1316 virtio_net_tx_timer, 1317 &n->vqs[i]); 1318 } else { 1319 n->vqs[i].tx_vq = 1320 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh); 1321 n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]); 1322 } 1323 1324 n->vqs[i].tx_waiting = 0; 1325 n->vqs[i].n = n; 1326 } 1327 1328 /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack 1329 * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid 1330 * breaking them. 1331 */ 1332 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1333 1334 virtio_net_set_queues(n); 1335 } 1336 1337 static void virtio_net_save(QEMUFile *f, void *opaque) 1338 { 1339 VirtIONet *n = opaque; 1340 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1341 1342 /* At this point, backend must be stopped, otherwise 1343 * it might keep writing to memory. */ 1344 assert(!n->vhost_started); 1345 virtio_save(vdev, f); 1346 } 1347 1348 static void virtio_net_save_device(VirtIODevice *vdev, QEMUFile *f) 1349 { 1350 VirtIONet *n = VIRTIO_NET(vdev); 1351 int i; 1352 1353 qemu_put_buffer(f, n->mac, ETH_ALEN); 1354 qemu_put_be32(f, n->vqs[0].tx_waiting); 1355 qemu_put_be32(f, n->mergeable_rx_bufs); 1356 qemu_put_be16(f, n->status); 1357 qemu_put_byte(f, n->promisc); 1358 qemu_put_byte(f, n->allmulti); 1359 qemu_put_be32(f, n->mac_table.in_use); 1360 qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN); 1361 qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3); 1362 qemu_put_be32(f, n->has_vnet_hdr); 1363 qemu_put_byte(f, n->mac_table.multi_overflow); 1364 qemu_put_byte(f, n->mac_table.uni_overflow); 1365 qemu_put_byte(f, n->alluni); 1366 qemu_put_byte(f, n->nomulti); 1367 qemu_put_byte(f, n->nouni); 1368 qemu_put_byte(f, n->nobcast); 1369 qemu_put_byte(f, n->has_ufo); 1370 if (n->max_queues > 1) { 1371 qemu_put_be16(f, n->max_queues); 1372 qemu_put_be16(f, n->curr_queues); 1373 for (i = 1; i < n->curr_queues; i++) { 1374 qemu_put_be32(f, n->vqs[i].tx_waiting); 1375 } 1376 } 1377 1378 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 1379 qemu_put_be64(f, n->curr_guest_offloads); 1380 } 1381 } 1382 1383 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id) 1384 { 1385 VirtIONet *n = opaque; 1386 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1387 1388 if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION) 1389 return -EINVAL; 1390 1391 return virtio_load(vdev, f, version_id); 1392 } 1393 1394 static int virtio_net_load_device(VirtIODevice *vdev, QEMUFile *f, 1395 int version_id) 1396 { 1397 VirtIONet *n = VIRTIO_NET(vdev); 1398 int i, link_down; 1399 1400 qemu_get_buffer(f, n->mac, ETH_ALEN); 1401 n->vqs[0].tx_waiting = qemu_get_be32(f); 1402 1403 virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f)); 1404 1405 if (version_id >= 3) 1406 n->status = qemu_get_be16(f); 1407 1408 if (version_id >= 4) { 1409 if (version_id < 8) { 1410 n->promisc = qemu_get_be32(f); 1411 n->allmulti = qemu_get_be32(f); 1412 } else { 1413 n->promisc = qemu_get_byte(f); 1414 n->allmulti = qemu_get_byte(f); 1415 } 1416 } 1417 1418 if (version_id >= 5) { 1419 n->mac_table.in_use = qemu_get_be32(f); 1420 /* MAC_TABLE_ENTRIES may be different from the saved image */ 1421 if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) { 1422 qemu_get_buffer(f, n->mac_table.macs, 1423 n->mac_table.in_use * ETH_ALEN); 1424 } else { 1425 int64_t i; 1426 1427 /* Overflow detected - can happen if source has a larger MAC table. 1428 * We simply set overflow flag so there's no need to maintain the 1429 * table of addresses, discard them all. 1430 * Note: 64 bit math to avoid integer overflow. 1431 */ 1432 for (i = 0; i < (int64_t)n->mac_table.in_use * ETH_ALEN; ++i) { 1433 qemu_get_byte(f); 1434 } 1435 n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1; 1436 n->mac_table.in_use = 0; 1437 } 1438 } 1439 1440 if (version_id >= 6) 1441 qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3); 1442 1443 if (version_id >= 7) { 1444 if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) { 1445 error_report("virtio-net: saved image requires vnet_hdr=on"); 1446 return -1; 1447 } 1448 } 1449 1450 if (version_id >= 9) { 1451 n->mac_table.multi_overflow = qemu_get_byte(f); 1452 n->mac_table.uni_overflow = qemu_get_byte(f); 1453 } 1454 1455 if (version_id >= 10) { 1456 n->alluni = qemu_get_byte(f); 1457 n->nomulti = qemu_get_byte(f); 1458 n->nouni = qemu_get_byte(f); 1459 n->nobcast = qemu_get_byte(f); 1460 } 1461 1462 if (version_id >= 11) { 1463 if (qemu_get_byte(f) && !peer_has_ufo(n)) { 1464 error_report("virtio-net: saved image requires TUN_F_UFO support"); 1465 return -1; 1466 } 1467 } 1468 1469 if (n->max_queues > 1) { 1470 if (n->max_queues != qemu_get_be16(f)) { 1471 error_report("virtio-net: different max_queues "); 1472 return -1; 1473 } 1474 1475 n->curr_queues = qemu_get_be16(f); 1476 if (n->curr_queues > n->max_queues) { 1477 error_report("virtio-net: curr_queues %x > max_queues %x", 1478 n->curr_queues, n->max_queues); 1479 return -1; 1480 } 1481 for (i = 1; i < n->curr_queues; i++) { 1482 n->vqs[i].tx_waiting = qemu_get_be32(f); 1483 } 1484 } 1485 1486 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 1487 n->curr_guest_offloads = qemu_get_be64(f); 1488 } else { 1489 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 1490 } 1491 1492 if (peer_has_vnet_hdr(n)) { 1493 virtio_net_apply_guest_offloads(n); 1494 } 1495 1496 virtio_net_set_queues(n); 1497 1498 /* Find the first multicast entry in the saved MAC filter */ 1499 for (i = 0; i < n->mac_table.in_use; i++) { 1500 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 1501 break; 1502 } 1503 } 1504 n->mac_table.first_multi = i; 1505 1506 /* nc.link_down can't be migrated, so infer link_down according 1507 * to link status bit in n->status */ 1508 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 1509 for (i = 0; i < n->max_queues; i++) { 1510 qemu_get_subqueue(n->nic, i)->link_down = link_down; 1511 } 1512 1513 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 1514 virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 1515 n->announce_counter = SELF_ANNOUNCE_ROUNDS; 1516 timer_mod(n->announce_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL)); 1517 } 1518 1519 return 0; 1520 } 1521 1522 static NetClientInfo net_virtio_info = { 1523 .type = NET_CLIENT_OPTIONS_KIND_NIC, 1524 .size = sizeof(NICState), 1525 .can_receive = virtio_net_can_receive, 1526 .receive = virtio_net_receive, 1527 .link_status_changed = virtio_net_set_link_status, 1528 .query_rx_filter = virtio_net_query_rxfilter, 1529 }; 1530 1531 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 1532 { 1533 VirtIONet *n = VIRTIO_NET(vdev); 1534 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1535 assert(n->vhost_started); 1536 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx); 1537 } 1538 1539 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 1540 bool mask) 1541 { 1542 VirtIONet *n = VIRTIO_NET(vdev); 1543 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1544 assert(n->vhost_started); 1545 vhost_net_virtqueue_mask(get_vhost_net(nc->peer), 1546 vdev, idx, mask); 1547 } 1548 1549 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features) 1550 { 1551 int i, config_size = 0; 1552 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC); 1553 for (i = 0; feature_sizes[i].flags != 0; i++) { 1554 if (host_features & feature_sizes[i].flags) { 1555 config_size = MAX(feature_sizes[i].end, config_size); 1556 } 1557 } 1558 n->config_size = config_size; 1559 } 1560 1561 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 1562 const char *type) 1563 { 1564 /* 1565 * The name can be NULL, the netclient name will be type.x. 1566 */ 1567 assert(type != NULL); 1568 1569 g_free(n->netclient_name); 1570 g_free(n->netclient_type); 1571 n->netclient_name = g_strdup(name); 1572 n->netclient_type = g_strdup(type); 1573 } 1574 1575 static void virtio_net_device_realize(DeviceState *dev, Error **errp) 1576 { 1577 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1578 VirtIONet *n = VIRTIO_NET(dev); 1579 NetClientState *nc; 1580 int i; 1581 1582 virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size); 1583 1584 n->max_queues = MAX(n->nic_conf.peers.queues, 1); 1585 if (n->max_queues * 2 + 1 > VIRTIO_PCI_QUEUE_MAX) { 1586 error_setg(errp, "Invalid number of queues (= %" PRIu32 "), " 1587 "must be a positive integer less than %d.", 1588 n->max_queues, (VIRTIO_PCI_QUEUE_MAX - 1) / 2); 1589 virtio_cleanup(vdev); 1590 return; 1591 } 1592 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues); 1593 n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx); 1594 n->curr_queues = 1; 1595 n->vqs[0].n = n; 1596 n->tx_timeout = n->net_conf.txtimer; 1597 1598 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 1599 && strcmp(n->net_conf.tx, "bh")) { 1600 error_report("virtio-net: " 1601 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 1602 n->net_conf.tx); 1603 error_report("Defaulting to \"bh\""); 1604 } 1605 1606 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 1607 n->vqs[0].tx_vq = virtio_add_queue(vdev, 256, 1608 virtio_net_handle_tx_timer); 1609 n->vqs[0].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, virtio_net_tx_timer, 1610 &n->vqs[0]); 1611 } else { 1612 n->vqs[0].tx_vq = virtio_add_queue(vdev, 256, 1613 virtio_net_handle_tx_bh); 1614 n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]); 1615 } 1616 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1617 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 1618 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 1619 n->status = VIRTIO_NET_S_LINK_UP; 1620 n->announce_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, 1621 virtio_net_announce_timer, n); 1622 1623 if (n->netclient_type) { 1624 /* 1625 * Happen when virtio_net_set_netclient_name has been called. 1626 */ 1627 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 1628 n->netclient_type, n->netclient_name, n); 1629 } else { 1630 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 1631 object_get_typename(OBJECT(dev)), dev->id, n); 1632 } 1633 1634 peer_test_vnet_hdr(n); 1635 if (peer_has_vnet_hdr(n)) { 1636 for (i = 0; i < n->max_queues; i++) { 1637 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true); 1638 } 1639 n->host_hdr_len = sizeof(struct virtio_net_hdr); 1640 } else { 1641 n->host_hdr_len = 0; 1642 } 1643 1644 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 1645 1646 n->vqs[0].tx_waiting = 0; 1647 n->tx_burst = n->net_conf.txburst; 1648 virtio_net_set_mrg_rx_bufs(n, 0); 1649 n->promisc = 1; /* for compatibility */ 1650 1651 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 1652 1653 n->vlans = g_malloc0(MAX_VLAN >> 3); 1654 1655 nc = qemu_get_queue(n->nic); 1656 nc->rxfilter_notify_enabled = 1; 1657 1658 n->qdev = dev; 1659 register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION, 1660 virtio_net_save, virtio_net_load, n); 1661 } 1662 1663 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp) 1664 { 1665 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1666 VirtIONet *n = VIRTIO_NET(dev); 1667 int i; 1668 1669 /* This will stop vhost backend if appropriate. */ 1670 virtio_net_set_status(vdev, 0); 1671 1672 unregister_savevm(dev, "virtio-net", n); 1673 1674 g_free(n->netclient_name); 1675 n->netclient_name = NULL; 1676 g_free(n->netclient_type); 1677 n->netclient_type = NULL; 1678 1679 g_free(n->mac_table.macs); 1680 g_free(n->vlans); 1681 1682 for (i = 0; i < n->max_queues; i++) { 1683 VirtIONetQueue *q = &n->vqs[i]; 1684 NetClientState *nc = qemu_get_subqueue(n->nic, i); 1685 1686 qemu_purge_queued_packets(nc); 1687 1688 if (q->tx_timer) { 1689 timer_del(q->tx_timer); 1690 timer_free(q->tx_timer); 1691 } else if (q->tx_bh) { 1692 qemu_bh_delete(q->tx_bh); 1693 } 1694 } 1695 1696 timer_del(n->announce_timer); 1697 timer_free(n->announce_timer); 1698 g_free(n->vqs); 1699 qemu_del_nic(n->nic); 1700 virtio_cleanup(vdev); 1701 } 1702 1703 static void virtio_net_instance_init(Object *obj) 1704 { 1705 VirtIONet *n = VIRTIO_NET(obj); 1706 1707 /* 1708 * The default config_size is sizeof(struct virtio_net_config). 1709 * Can be overriden with virtio_net_set_config_size. 1710 */ 1711 n->config_size = sizeof(struct virtio_net_config); 1712 device_add_bootindex_property(obj, &n->nic_conf.bootindex, 1713 "bootindex", "/ethernet-phy@0", 1714 DEVICE(n), NULL); 1715 } 1716 1717 static Property virtio_net_properties[] = { 1718 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 1719 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 1720 TX_TIMER_INTERVAL), 1721 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 1722 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 1723 DEFINE_PROP_END_OF_LIST(), 1724 }; 1725 1726 static void virtio_net_class_init(ObjectClass *klass, void *data) 1727 { 1728 DeviceClass *dc = DEVICE_CLASS(klass); 1729 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 1730 1731 dc->props = virtio_net_properties; 1732 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 1733 vdc->realize = virtio_net_device_realize; 1734 vdc->unrealize = virtio_net_device_unrealize; 1735 vdc->get_config = virtio_net_get_config; 1736 vdc->set_config = virtio_net_set_config; 1737 vdc->get_features = virtio_net_get_features; 1738 vdc->set_features = virtio_net_set_features; 1739 vdc->bad_features = virtio_net_bad_features; 1740 vdc->reset = virtio_net_reset; 1741 vdc->set_status = virtio_net_set_status; 1742 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 1743 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 1744 vdc->load = virtio_net_load_device; 1745 vdc->save = virtio_net_save_device; 1746 } 1747 1748 static const TypeInfo virtio_net_info = { 1749 .name = TYPE_VIRTIO_NET, 1750 .parent = TYPE_VIRTIO_DEVICE, 1751 .instance_size = sizeof(VirtIONet), 1752 .instance_init = virtio_net_instance_init, 1753 .class_init = virtio_net_class_init, 1754 }; 1755 1756 static void virtio_register_types(void) 1757 { 1758 type_register_static(&virtio_net_info); 1759 } 1760 1761 type_init(virtio_register_types) 1762