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 "monitor/monitor.h" 26 27 #define VIRTIO_NET_VM_VERSION 11 28 29 #define MAC_TABLE_ENTRIES 64 30 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */ 31 32 /* 33 * Calculate the number of bytes up to and including the given 'field' of 34 * 'container'. 35 */ 36 #define endof(container, field) \ 37 (offsetof(container, field) + sizeof(((container *)0)->field)) 38 39 typedef struct VirtIOFeature { 40 uint32_t flags; 41 size_t end; 42 } VirtIOFeature; 43 44 static VirtIOFeature feature_sizes[] = { 45 {.flags = 1 << VIRTIO_NET_F_MAC, 46 .end = endof(struct virtio_net_config, mac)}, 47 {.flags = 1 << VIRTIO_NET_F_STATUS, 48 .end = endof(struct virtio_net_config, status)}, 49 {.flags = 1 << VIRTIO_NET_F_MQ, 50 .end = endof(struct virtio_net_config, max_virtqueue_pairs)}, 51 {} 52 }; 53 54 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc) 55 { 56 VirtIONet *n = qemu_get_nic_opaque(nc); 57 58 return &n->vqs[nc->queue_index]; 59 } 60 61 static int vq2q(int queue_index) 62 { 63 return queue_index / 2; 64 } 65 66 /* TODO 67 * - we could suppress RX interrupt if we were so inclined. 68 */ 69 70 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config) 71 { 72 VirtIONet *n = VIRTIO_NET(vdev); 73 struct virtio_net_config netcfg; 74 75 stw_p(&netcfg.status, n->status); 76 stw_p(&netcfg.max_virtqueue_pairs, n->max_queues); 77 memcpy(netcfg.mac, n->mac, ETH_ALEN); 78 memcpy(config, &netcfg, n->config_size); 79 } 80 81 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config) 82 { 83 VirtIONet *n = VIRTIO_NET(vdev); 84 struct virtio_net_config netcfg = {}; 85 86 memcpy(&netcfg, config, n->config_size); 87 88 if (!(vdev->guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) && 89 memcmp(netcfg.mac, n->mac, ETH_ALEN)) { 90 memcpy(n->mac, netcfg.mac, ETH_ALEN); 91 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 92 } 93 } 94 95 static bool virtio_net_started(VirtIONet *n, uint8_t status) 96 { 97 VirtIODevice *vdev = VIRTIO_DEVICE(n); 98 return (status & VIRTIO_CONFIG_S_DRIVER_OK) && 99 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running; 100 } 101 102 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status) 103 { 104 VirtIODevice *vdev = VIRTIO_DEVICE(n); 105 NetClientState *nc = qemu_get_queue(n->nic); 106 int queues = n->multiqueue ? n->max_queues : 1; 107 108 if (!nc->peer) { 109 return; 110 } 111 if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 112 return; 113 } 114 115 if (!tap_get_vhost_net(nc->peer)) { 116 return; 117 } 118 119 if (!!n->vhost_started == 120 (virtio_net_started(n, status) && !nc->peer->link_down)) { 121 return; 122 } 123 if (!n->vhost_started) { 124 int r; 125 if (!vhost_net_query(tap_get_vhost_net(nc->peer), vdev)) { 126 return; 127 } 128 n->vhost_started = 1; 129 r = vhost_net_start(vdev, n->nic->ncs, queues); 130 if (r < 0) { 131 error_report("unable to start vhost net: %d: " 132 "falling back on userspace virtio", -r); 133 n->vhost_started = 0; 134 } 135 } else { 136 vhost_net_stop(vdev, n->nic->ncs, queues); 137 n->vhost_started = 0; 138 } 139 } 140 141 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status) 142 { 143 VirtIONet *n = VIRTIO_NET(vdev); 144 VirtIONetQueue *q; 145 int i; 146 uint8_t queue_status; 147 148 virtio_net_vhost_status(n, status); 149 150 for (i = 0; i < n->max_queues; i++) { 151 q = &n->vqs[i]; 152 153 if ((!n->multiqueue && i != 0) || i >= n->curr_queues) { 154 queue_status = 0; 155 } else { 156 queue_status = status; 157 } 158 159 if (!q->tx_waiting) { 160 continue; 161 } 162 163 if (virtio_net_started(n, queue_status) && !n->vhost_started) { 164 if (q->tx_timer) { 165 timer_mod(q->tx_timer, 166 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 167 } else { 168 qemu_bh_schedule(q->tx_bh); 169 } 170 } else { 171 if (q->tx_timer) { 172 timer_del(q->tx_timer); 173 } else { 174 qemu_bh_cancel(q->tx_bh); 175 } 176 } 177 } 178 } 179 180 static void virtio_net_set_link_status(NetClientState *nc) 181 { 182 VirtIONet *n = qemu_get_nic_opaque(nc); 183 VirtIODevice *vdev = VIRTIO_DEVICE(n); 184 uint16_t old_status = n->status; 185 186 if (nc->link_down) 187 n->status &= ~VIRTIO_NET_S_LINK_UP; 188 else 189 n->status |= VIRTIO_NET_S_LINK_UP; 190 191 if (n->status != old_status) 192 virtio_notify_config(vdev); 193 194 virtio_net_set_status(vdev, vdev->status); 195 } 196 197 static void rxfilter_notify(NetClientState *nc) 198 { 199 QObject *event_data; 200 VirtIONet *n = qemu_get_nic_opaque(nc); 201 202 if (nc->rxfilter_notify_enabled) { 203 if (n->netclient_name) { 204 event_data = qobject_from_jsonf("{ 'name': %s, 'path': %s }", 205 n->netclient_name, 206 object_get_canonical_path(OBJECT(n->qdev))); 207 } else { 208 event_data = qobject_from_jsonf("{ 'path': %s }", 209 object_get_canonical_path(OBJECT(n->qdev))); 210 } 211 monitor_protocol_event(QEVENT_NIC_RX_FILTER_CHANGED, event_data); 212 qobject_decref(event_data); 213 214 /* disable event notification to avoid events flooding */ 215 nc->rxfilter_notify_enabled = 0; 216 } 217 } 218 219 static char *mac_strdup_printf(const uint8_t *mac) 220 { 221 return g_strdup_printf("%.2x:%.2x:%.2x:%.2x:%.2x:%.2x", mac[0], 222 mac[1], mac[2], mac[3], mac[4], mac[5]); 223 } 224 225 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc) 226 { 227 VirtIONet *n = qemu_get_nic_opaque(nc); 228 RxFilterInfo *info; 229 strList *str_list, *entry; 230 intList *int_list, *int_entry; 231 int i, j; 232 233 info = g_malloc0(sizeof(*info)); 234 info->name = g_strdup(nc->name); 235 info->promiscuous = n->promisc; 236 237 if (n->nouni) { 238 info->unicast = RX_STATE_NONE; 239 } else if (n->alluni) { 240 info->unicast = RX_STATE_ALL; 241 } else { 242 info->unicast = RX_STATE_NORMAL; 243 } 244 245 if (n->nomulti) { 246 info->multicast = RX_STATE_NONE; 247 } else if (n->allmulti) { 248 info->multicast = RX_STATE_ALL; 249 } else { 250 info->multicast = RX_STATE_NORMAL; 251 } 252 253 info->broadcast_allowed = n->nobcast; 254 info->multicast_overflow = n->mac_table.multi_overflow; 255 info->unicast_overflow = n->mac_table.uni_overflow; 256 257 info->main_mac = mac_strdup_printf(n->mac); 258 259 str_list = NULL; 260 for (i = 0; i < n->mac_table.first_multi; i++) { 261 entry = g_malloc0(sizeof(*entry)); 262 entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN); 263 entry->next = str_list; 264 str_list = entry; 265 } 266 info->unicast_table = str_list; 267 268 str_list = NULL; 269 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 270 entry = g_malloc0(sizeof(*entry)); 271 entry->value = mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN); 272 entry->next = str_list; 273 str_list = entry; 274 } 275 info->multicast_table = str_list; 276 277 int_list = NULL; 278 for (i = 0; i < MAX_VLAN >> 5; i++) { 279 for (j = 0; n->vlans[i] && j < 0x1f; j++) { 280 if (n->vlans[i] & (1U << j)) { 281 int_entry = g_malloc0(sizeof(*int_entry)); 282 int_entry->value = (i << 5) + j; 283 int_entry->next = int_list; 284 int_list = int_entry; 285 } 286 } 287 } 288 info->vlan_table = int_list; 289 290 /* enable event notification after query */ 291 nc->rxfilter_notify_enabled = 1; 292 293 return info; 294 } 295 296 static void virtio_net_reset(VirtIODevice *vdev) 297 { 298 VirtIONet *n = VIRTIO_NET(vdev); 299 300 /* Reset back to compatibility mode */ 301 n->promisc = 1; 302 n->allmulti = 0; 303 n->alluni = 0; 304 n->nomulti = 0; 305 n->nouni = 0; 306 n->nobcast = 0; 307 /* multiqueue is disabled by default */ 308 n->curr_queues = 1; 309 310 /* Flush any MAC and VLAN filter table state */ 311 n->mac_table.in_use = 0; 312 n->mac_table.first_multi = 0; 313 n->mac_table.multi_overflow = 0; 314 n->mac_table.uni_overflow = 0; 315 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN); 316 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac)); 317 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 318 memset(n->vlans, 0, MAX_VLAN >> 3); 319 } 320 321 static void peer_test_vnet_hdr(VirtIONet *n) 322 { 323 NetClientState *nc = qemu_get_queue(n->nic); 324 if (!nc->peer) { 325 return; 326 } 327 328 if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 329 return; 330 } 331 332 n->has_vnet_hdr = tap_has_vnet_hdr(nc->peer); 333 } 334 335 static int peer_has_vnet_hdr(VirtIONet *n) 336 { 337 return n->has_vnet_hdr; 338 } 339 340 static int peer_has_ufo(VirtIONet *n) 341 { 342 if (!peer_has_vnet_hdr(n)) 343 return 0; 344 345 n->has_ufo = tap_has_ufo(qemu_get_queue(n->nic)->peer); 346 347 return n->has_ufo; 348 } 349 350 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs) 351 { 352 int i; 353 NetClientState *nc; 354 355 n->mergeable_rx_bufs = mergeable_rx_bufs; 356 357 n->guest_hdr_len = n->mergeable_rx_bufs ? 358 sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr); 359 360 for (i = 0; i < n->max_queues; i++) { 361 nc = qemu_get_subqueue(n->nic, i); 362 363 if (peer_has_vnet_hdr(n) && 364 tap_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) { 365 tap_set_vnet_hdr_len(nc->peer, n->guest_hdr_len); 366 n->host_hdr_len = n->guest_hdr_len; 367 } 368 } 369 } 370 371 static int peer_attach(VirtIONet *n, int index) 372 { 373 NetClientState *nc = qemu_get_subqueue(n->nic, index); 374 375 if (!nc->peer) { 376 return 0; 377 } 378 379 if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 380 return 0; 381 } 382 383 return tap_enable(nc->peer); 384 } 385 386 static int peer_detach(VirtIONet *n, int index) 387 { 388 NetClientState *nc = qemu_get_subqueue(n->nic, index); 389 390 if (!nc->peer) { 391 return 0; 392 } 393 394 if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 395 return 0; 396 } 397 398 return tap_disable(nc->peer); 399 } 400 401 static void virtio_net_set_queues(VirtIONet *n) 402 { 403 int i; 404 405 for (i = 0; i < n->max_queues; i++) { 406 if (i < n->curr_queues) { 407 assert(!peer_attach(n, i)); 408 } else { 409 assert(!peer_detach(n, i)); 410 } 411 } 412 } 413 414 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); 415 416 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features) 417 { 418 VirtIONet *n = VIRTIO_NET(vdev); 419 NetClientState *nc = qemu_get_queue(n->nic); 420 421 features |= (1 << VIRTIO_NET_F_MAC); 422 423 if (!peer_has_vnet_hdr(n)) { 424 features &= ~(0x1 << VIRTIO_NET_F_CSUM); 425 features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4); 426 features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6); 427 features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN); 428 429 features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM); 430 features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4); 431 features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6); 432 features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN); 433 } 434 435 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 436 features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO); 437 features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO); 438 } 439 440 if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 441 return features; 442 } 443 if (!tap_get_vhost_net(nc->peer)) { 444 return features; 445 } 446 return vhost_net_get_features(tap_get_vhost_net(nc->peer), features); 447 } 448 449 static uint32_t virtio_net_bad_features(VirtIODevice *vdev) 450 { 451 uint32_t features = 0; 452 453 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 454 * but also these: */ 455 features |= (1 << VIRTIO_NET_F_MAC); 456 features |= (1 << VIRTIO_NET_F_CSUM); 457 features |= (1 << VIRTIO_NET_F_HOST_TSO4); 458 features |= (1 << VIRTIO_NET_F_HOST_TSO6); 459 features |= (1 << VIRTIO_NET_F_HOST_ECN); 460 461 return features; 462 } 463 464 static void virtio_net_apply_guest_offloads(VirtIONet *n) 465 { 466 tap_set_offload(qemu_get_subqueue(n->nic, 0)->peer, 467 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 468 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 469 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 470 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 471 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO))); 472 } 473 474 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features) 475 { 476 static const uint64_t guest_offloads_mask = 477 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 478 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 479 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 480 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 481 (1ULL << VIRTIO_NET_F_GUEST_UFO); 482 483 return guest_offloads_mask & features; 484 } 485 486 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n) 487 { 488 VirtIODevice *vdev = VIRTIO_DEVICE(n); 489 return virtio_net_guest_offloads_by_features(vdev->guest_features); 490 } 491 492 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features) 493 { 494 VirtIONet *n = VIRTIO_NET(vdev); 495 int i; 496 497 virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ))); 498 499 virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF))); 500 501 if (n->has_vnet_hdr) { 502 n->curr_guest_offloads = 503 virtio_net_guest_offloads_by_features(features); 504 virtio_net_apply_guest_offloads(n); 505 } 506 507 for (i = 0; i < n->max_queues; i++) { 508 NetClientState *nc = qemu_get_subqueue(n->nic, i); 509 510 if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 511 continue; 512 } 513 if (!tap_get_vhost_net(nc->peer)) { 514 continue; 515 } 516 vhost_net_ack_features(tap_get_vhost_net(nc->peer), features); 517 } 518 } 519 520 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 521 struct iovec *iov, unsigned int iov_cnt) 522 { 523 uint8_t on; 524 size_t s; 525 NetClientState *nc = qemu_get_queue(n->nic); 526 527 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 528 if (s != sizeof(on)) { 529 return VIRTIO_NET_ERR; 530 } 531 532 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 533 n->promisc = on; 534 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 535 n->allmulti = on; 536 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 537 n->alluni = on; 538 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 539 n->nomulti = on; 540 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 541 n->nouni = on; 542 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 543 n->nobcast = on; 544 } else { 545 return VIRTIO_NET_ERR; 546 } 547 548 rxfilter_notify(nc); 549 550 return VIRTIO_NET_OK; 551 } 552 553 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 554 struct iovec *iov, unsigned int iov_cnt) 555 { 556 VirtIODevice *vdev = VIRTIO_DEVICE(n); 557 uint64_t offloads; 558 size_t s; 559 560 if (!((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features)) { 561 return VIRTIO_NET_ERR; 562 } 563 564 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 565 if (s != sizeof(offloads)) { 566 return VIRTIO_NET_ERR; 567 } 568 569 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 570 uint64_t supported_offloads; 571 572 if (!n->has_vnet_hdr) { 573 return VIRTIO_NET_ERR; 574 } 575 576 supported_offloads = virtio_net_supported_guest_offloads(n); 577 if (offloads & ~supported_offloads) { 578 return VIRTIO_NET_ERR; 579 } 580 581 n->curr_guest_offloads = offloads; 582 virtio_net_apply_guest_offloads(n); 583 584 return VIRTIO_NET_OK; 585 } else { 586 return VIRTIO_NET_ERR; 587 } 588 } 589 590 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 591 struct iovec *iov, unsigned int iov_cnt) 592 { 593 struct virtio_net_ctrl_mac mac_data; 594 size_t s; 595 NetClientState *nc = qemu_get_queue(n->nic); 596 597 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 598 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 599 return VIRTIO_NET_ERR; 600 } 601 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 602 assert(s == sizeof(n->mac)); 603 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 604 rxfilter_notify(nc); 605 606 return VIRTIO_NET_OK; 607 } 608 609 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 610 return VIRTIO_NET_ERR; 611 } 612 613 n->mac_table.in_use = 0; 614 n->mac_table.first_multi = 0; 615 n->mac_table.uni_overflow = 0; 616 n->mac_table.multi_overflow = 0; 617 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN); 618 619 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 620 sizeof(mac_data.entries)); 621 mac_data.entries = ldl_p(&mac_data.entries); 622 if (s != sizeof(mac_data.entries)) { 623 goto error; 624 } 625 iov_discard_front(&iov, &iov_cnt, s); 626 627 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 628 goto error; 629 } 630 631 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 632 s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs, 633 mac_data.entries * ETH_ALEN); 634 if (s != mac_data.entries * ETH_ALEN) { 635 goto error; 636 } 637 n->mac_table.in_use += mac_data.entries; 638 } else { 639 n->mac_table.uni_overflow = 1; 640 } 641 642 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 643 644 n->mac_table.first_multi = n->mac_table.in_use; 645 646 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 647 sizeof(mac_data.entries)); 648 mac_data.entries = ldl_p(&mac_data.entries); 649 if (s != sizeof(mac_data.entries)) { 650 goto error; 651 } 652 653 iov_discard_front(&iov, &iov_cnt, s); 654 655 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 656 goto error; 657 } 658 659 if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) { 660 s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs, 661 mac_data.entries * ETH_ALEN); 662 if (s != mac_data.entries * ETH_ALEN) { 663 goto error; 664 } 665 n->mac_table.in_use += mac_data.entries; 666 } else { 667 n->mac_table.multi_overflow = 1; 668 } 669 670 rxfilter_notify(nc); 671 672 return VIRTIO_NET_OK; 673 674 error: 675 rxfilter_notify(nc); 676 return VIRTIO_NET_ERR; 677 } 678 679 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 680 struct iovec *iov, unsigned int iov_cnt) 681 { 682 uint16_t vid; 683 size_t s; 684 NetClientState *nc = qemu_get_queue(n->nic); 685 686 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 687 vid = lduw_p(&vid); 688 if (s != sizeof(vid)) { 689 return VIRTIO_NET_ERR; 690 } 691 692 if (vid >= MAX_VLAN) 693 return VIRTIO_NET_ERR; 694 695 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 696 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 697 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 698 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 699 else 700 return VIRTIO_NET_ERR; 701 702 rxfilter_notify(nc); 703 704 return VIRTIO_NET_OK; 705 } 706 707 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 708 struct iovec *iov, unsigned int iov_cnt) 709 { 710 VirtIODevice *vdev = VIRTIO_DEVICE(n); 711 struct virtio_net_ctrl_mq mq; 712 size_t s; 713 uint16_t queues; 714 715 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 716 if (s != sizeof(mq)) { 717 return VIRTIO_NET_ERR; 718 } 719 720 if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 721 return VIRTIO_NET_ERR; 722 } 723 724 queues = lduw_p(&mq.virtqueue_pairs); 725 726 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 727 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 728 queues > n->max_queues || 729 !n->multiqueue) { 730 return VIRTIO_NET_ERR; 731 } 732 733 n->curr_queues = queues; 734 /* stop the backend before changing the number of queues to avoid handling a 735 * disabled queue */ 736 virtio_net_set_status(vdev, vdev->status); 737 virtio_net_set_queues(n); 738 739 return VIRTIO_NET_OK; 740 } 741 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 742 { 743 VirtIONet *n = VIRTIO_NET(vdev); 744 struct virtio_net_ctrl_hdr ctrl; 745 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 746 VirtQueueElement elem; 747 size_t s; 748 struct iovec *iov; 749 unsigned int iov_cnt; 750 751 while (virtqueue_pop(vq, &elem)) { 752 if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) || 753 iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) { 754 error_report("virtio-net ctrl missing headers"); 755 exit(1); 756 } 757 758 iov = elem.out_sg; 759 iov_cnt = elem.out_num; 760 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl)); 761 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl)); 762 if (s != sizeof(ctrl)) { 763 status = VIRTIO_NET_ERR; 764 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 765 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt); 766 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 767 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt); 768 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 769 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt); 770 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 771 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt); 772 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 773 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt); 774 } 775 776 s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status)); 777 assert(s == sizeof(status)); 778 779 virtqueue_push(vq, &elem, sizeof(status)); 780 virtio_notify(vdev, vq); 781 } 782 } 783 784 /* RX */ 785 786 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 787 { 788 VirtIONet *n = VIRTIO_NET(vdev); 789 int queue_index = vq2q(virtio_get_queue_index(vq)); 790 791 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 792 } 793 794 static int virtio_net_can_receive(NetClientState *nc) 795 { 796 VirtIONet *n = qemu_get_nic_opaque(nc); 797 VirtIODevice *vdev = VIRTIO_DEVICE(n); 798 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 799 800 if (!vdev->vm_running) { 801 return 0; 802 } 803 804 if (nc->queue_index >= n->curr_queues) { 805 return 0; 806 } 807 808 if (!virtio_queue_ready(q->rx_vq) || 809 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 810 return 0; 811 } 812 813 return 1; 814 } 815 816 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 817 { 818 VirtIONet *n = q->n; 819 if (virtio_queue_empty(q->rx_vq) || 820 (n->mergeable_rx_bufs && 821 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 822 virtio_queue_set_notification(q->rx_vq, 1); 823 824 /* To avoid a race condition where the guest has made some buffers 825 * available after the above check but before notification was 826 * enabled, check for available buffers again. 827 */ 828 if (virtio_queue_empty(q->rx_vq) || 829 (n->mergeable_rx_bufs && 830 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 831 return 0; 832 } 833 } 834 835 virtio_queue_set_notification(q->rx_vq, 0); 836 return 1; 837 } 838 839 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 840 * it never finds out that the packets don't have valid checksums. This 841 * causes dhclient to get upset. Fedora's carried a patch for ages to 842 * fix this with Xen but it hasn't appeared in an upstream release of 843 * dhclient yet. 844 * 845 * To avoid breaking existing guests, we catch udp packets and add 846 * checksums. This is terrible but it's better than hacking the guest 847 * kernels. 848 * 849 * N.B. if we introduce a zero-copy API, this operation is no longer free so 850 * we should provide a mechanism to disable it to avoid polluting the host 851 * cache. 852 */ 853 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 854 uint8_t *buf, size_t size) 855 { 856 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 857 (size > 27 && size < 1500) && /* normal sized MTU */ 858 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 859 (buf[23] == 17) && /* ip.protocol == UDP */ 860 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 861 net_checksum_calculate(buf, size); 862 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 863 } 864 } 865 866 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt, 867 const void *buf, size_t size) 868 { 869 if (n->has_vnet_hdr) { 870 /* FIXME this cast is evil */ 871 void *wbuf = (void *)buf; 872 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, 873 size - n->host_hdr_len); 874 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr)); 875 } else { 876 struct virtio_net_hdr hdr = { 877 .flags = 0, 878 .gso_type = VIRTIO_NET_HDR_GSO_NONE 879 }; 880 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr); 881 } 882 } 883 884 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 885 { 886 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 887 static const uint8_t vlan[] = {0x81, 0x00}; 888 uint8_t *ptr = (uint8_t *)buf; 889 int i; 890 891 if (n->promisc) 892 return 1; 893 894 ptr += n->host_hdr_len; 895 896 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 897 int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff; 898 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 899 return 0; 900 } 901 902 if (ptr[0] & 1) { // multicast 903 if (!memcmp(ptr, bcast, sizeof(bcast))) { 904 return !n->nobcast; 905 } else if (n->nomulti) { 906 return 0; 907 } else if (n->allmulti || n->mac_table.multi_overflow) { 908 return 1; 909 } 910 911 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 912 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 913 return 1; 914 } 915 } 916 } else { // unicast 917 if (n->nouni) { 918 return 0; 919 } else if (n->alluni || n->mac_table.uni_overflow) { 920 return 1; 921 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 922 return 1; 923 } 924 925 for (i = 0; i < n->mac_table.first_multi; i++) { 926 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 927 return 1; 928 } 929 } 930 } 931 932 return 0; 933 } 934 935 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size) 936 { 937 VirtIONet *n = qemu_get_nic_opaque(nc); 938 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 939 VirtIODevice *vdev = VIRTIO_DEVICE(n); 940 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 941 struct virtio_net_hdr_mrg_rxbuf mhdr; 942 unsigned mhdr_cnt = 0; 943 size_t offset, i, guest_offset; 944 945 if (!virtio_net_can_receive(nc)) { 946 return -1; 947 } 948 949 /* hdr_len refers to the header we supply to the guest */ 950 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 951 return 0; 952 } 953 954 if (!receive_filter(n, buf, size)) 955 return size; 956 957 offset = i = 0; 958 959 while (offset < size) { 960 VirtQueueElement elem; 961 int len, total; 962 const struct iovec *sg = elem.in_sg; 963 964 total = 0; 965 966 if (virtqueue_pop(q->rx_vq, &elem) == 0) { 967 if (i == 0) 968 return -1; 969 error_report("virtio-net unexpected empty queue: " 970 "i %zd mergeable %d offset %zd, size %zd, " 971 "guest hdr len %zd, host hdr len %zd guest features 0x%x", 972 i, n->mergeable_rx_bufs, offset, size, 973 n->guest_hdr_len, n->host_hdr_len, vdev->guest_features); 974 exit(1); 975 } 976 977 if (elem.in_num < 1) { 978 error_report("virtio-net receive queue contains no in buffers"); 979 exit(1); 980 } 981 982 if (i == 0) { 983 assert(offset == 0); 984 if (n->mergeable_rx_bufs) { 985 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 986 sg, elem.in_num, 987 offsetof(typeof(mhdr), num_buffers), 988 sizeof(mhdr.num_buffers)); 989 } 990 991 receive_header(n, sg, elem.in_num, buf, size); 992 offset = n->host_hdr_len; 993 total += n->guest_hdr_len; 994 guest_offset = n->guest_hdr_len; 995 } else { 996 guest_offset = 0; 997 } 998 999 /* copy in packet. ugh */ 1000 len = iov_from_buf(sg, elem.in_num, guest_offset, 1001 buf + offset, size - offset); 1002 total += len; 1003 offset += len; 1004 /* If buffers can't be merged, at this point we 1005 * must have consumed the complete packet. 1006 * Otherwise, drop it. */ 1007 if (!n->mergeable_rx_bufs && offset < size) { 1008 #if 0 1009 error_report("virtio-net truncated non-mergeable packet: " 1010 "i %zd mergeable %d offset %zd, size %zd, " 1011 "guest hdr len %zd, host hdr len %zd", 1012 i, n->mergeable_rx_bufs, 1013 offset, size, n->guest_hdr_len, n->host_hdr_len); 1014 #endif 1015 return size; 1016 } 1017 1018 /* signal other side */ 1019 virtqueue_fill(q->rx_vq, &elem, total, i++); 1020 } 1021 1022 if (mhdr_cnt) { 1023 stw_p(&mhdr.num_buffers, i); 1024 iov_from_buf(mhdr_sg, mhdr_cnt, 1025 0, 1026 &mhdr.num_buffers, sizeof mhdr.num_buffers); 1027 } 1028 1029 virtqueue_flush(q->rx_vq, i); 1030 virtio_notify(vdev, q->rx_vq); 1031 1032 return size; 1033 } 1034 1035 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 1036 1037 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 1038 { 1039 VirtIONet *n = qemu_get_nic_opaque(nc); 1040 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1041 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1042 1043 virtqueue_push(q->tx_vq, &q->async_tx.elem, 0); 1044 virtio_notify(vdev, q->tx_vq); 1045 1046 q->async_tx.elem.out_num = q->async_tx.len = 0; 1047 1048 virtio_queue_set_notification(q->tx_vq, 1); 1049 virtio_net_flush_tx(q); 1050 } 1051 1052 /* TX */ 1053 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 1054 { 1055 VirtIONet *n = q->n; 1056 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1057 VirtQueueElement elem; 1058 int32_t num_packets = 0; 1059 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 1060 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1061 return num_packets; 1062 } 1063 1064 assert(vdev->vm_running); 1065 1066 if (q->async_tx.elem.out_num) { 1067 virtio_queue_set_notification(q->tx_vq, 0); 1068 return num_packets; 1069 } 1070 1071 while (virtqueue_pop(q->tx_vq, &elem)) { 1072 ssize_t ret, len; 1073 unsigned int out_num = elem.out_num; 1074 struct iovec *out_sg = &elem.out_sg[0]; 1075 struct iovec sg[VIRTQUEUE_MAX_SIZE]; 1076 1077 if (out_num < 1) { 1078 error_report("virtio-net header not in first element"); 1079 exit(1); 1080 } 1081 1082 /* 1083 * If host wants to see the guest header as is, we can 1084 * pass it on unchanged. Otherwise, copy just the parts 1085 * that host is interested in. 1086 */ 1087 assert(n->host_hdr_len <= n->guest_hdr_len); 1088 if (n->host_hdr_len != n->guest_hdr_len) { 1089 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 1090 out_sg, out_num, 1091 0, n->host_hdr_len); 1092 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 1093 out_sg, out_num, 1094 n->guest_hdr_len, -1); 1095 out_num = sg_num; 1096 out_sg = sg; 1097 } 1098 1099 len = n->guest_hdr_len; 1100 1101 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 1102 out_sg, out_num, virtio_net_tx_complete); 1103 if (ret == 0) { 1104 virtio_queue_set_notification(q->tx_vq, 0); 1105 q->async_tx.elem = elem; 1106 q->async_tx.len = len; 1107 return -EBUSY; 1108 } 1109 1110 len += ret; 1111 1112 virtqueue_push(q->tx_vq, &elem, 0); 1113 virtio_notify(vdev, q->tx_vq); 1114 1115 if (++num_packets >= n->tx_burst) { 1116 break; 1117 } 1118 } 1119 return num_packets; 1120 } 1121 1122 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 1123 { 1124 VirtIONet *n = VIRTIO_NET(vdev); 1125 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1126 1127 /* This happens when device was stopped but VCPU wasn't. */ 1128 if (!vdev->vm_running) { 1129 q->tx_waiting = 1; 1130 return; 1131 } 1132 1133 if (q->tx_waiting) { 1134 virtio_queue_set_notification(vq, 1); 1135 timer_del(q->tx_timer); 1136 q->tx_waiting = 0; 1137 virtio_net_flush_tx(q); 1138 } else { 1139 timer_mod(q->tx_timer, 1140 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 1141 q->tx_waiting = 1; 1142 virtio_queue_set_notification(vq, 0); 1143 } 1144 } 1145 1146 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 1147 { 1148 VirtIONet *n = VIRTIO_NET(vdev); 1149 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1150 1151 if (unlikely(q->tx_waiting)) { 1152 return; 1153 } 1154 q->tx_waiting = 1; 1155 /* This happens when device was stopped but VCPU wasn't. */ 1156 if (!vdev->vm_running) { 1157 return; 1158 } 1159 virtio_queue_set_notification(vq, 0); 1160 qemu_bh_schedule(q->tx_bh); 1161 } 1162 1163 static void virtio_net_tx_timer(void *opaque) 1164 { 1165 VirtIONetQueue *q = opaque; 1166 VirtIONet *n = q->n; 1167 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1168 assert(vdev->vm_running); 1169 1170 q->tx_waiting = 0; 1171 1172 /* Just in case the driver is not ready on more */ 1173 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1174 return; 1175 } 1176 1177 virtio_queue_set_notification(q->tx_vq, 1); 1178 virtio_net_flush_tx(q); 1179 } 1180 1181 static void virtio_net_tx_bh(void *opaque) 1182 { 1183 VirtIONetQueue *q = opaque; 1184 VirtIONet *n = q->n; 1185 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1186 int32_t ret; 1187 1188 assert(vdev->vm_running); 1189 1190 q->tx_waiting = 0; 1191 1192 /* Just in case the driver is not ready on more */ 1193 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 1194 return; 1195 } 1196 1197 ret = virtio_net_flush_tx(q); 1198 if (ret == -EBUSY) { 1199 return; /* Notification re-enable handled by tx_complete */ 1200 } 1201 1202 /* If we flush a full burst of packets, assume there are 1203 * more coming and immediately reschedule */ 1204 if (ret >= n->tx_burst) { 1205 qemu_bh_schedule(q->tx_bh); 1206 q->tx_waiting = 1; 1207 return; 1208 } 1209 1210 /* If less than a full burst, re-enable notification and flush 1211 * anything that may have come in while we weren't looking. If 1212 * we find something, assume the guest is still active and reschedule */ 1213 virtio_queue_set_notification(q->tx_vq, 1); 1214 if (virtio_net_flush_tx(q) > 0) { 1215 virtio_queue_set_notification(q->tx_vq, 0); 1216 qemu_bh_schedule(q->tx_bh); 1217 q->tx_waiting = 1; 1218 } 1219 } 1220 1221 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 1222 { 1223 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1224 int i, max = multiqueue ? n->max_queues : 1; 1225 1226 n->multiqueue = multiqueue; 1227 1228 for (i = 2; i <= n->max_queues * 2 + 1; i++) { 1229 virtio_del_queue(vdev, i); 1230 } 1231 1232 for (i = 1; i < max; i++) { 1233 n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx); 1234 if (n->vqs[i].tx_timer) { 1235 n->vqs[i].tx_vq = 1236 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer); 1237 n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 1238 virtio_net_tx_timer, 1239 &n->vqs[i]); 1240 } else { 1241 n->vqs[i].tx_vq = 1242 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh); 1243 n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]); 1244 } 1245 1246 n->vqs[i].tx_waiting = 0; 1247 n->vqs[i].n = n; 1248 } 1249 1250 /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack 1251 * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid 1252 * breaking them. 1253 */ 1254 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1255 1256 virtio_net_set_queues(n); 1257 } 1258 1259 static void virtio_net_save(QEMUFile *f, void *opaque) 1260 { 1261 int i; 1262 VirtIONet *n = opaque; 1263 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1264 1265 /* At this point, backend must be stopped, otherwise 1266 * it might keep writing to memory. */ 1267 assert(!n->vhost_started); 1268 virtio_save(vdev, f); 1269 1270 qemu_put_buffer(f, n->mac, ETH_ALEN); 1271 qemu_put_be32(f, n->vqs[0].tx_waiting); 1272 qemu_put_be32(f, n->mergeable_rx_bufs); 1273 qemu_put_be16(f, n->status); 1274 qemu_put_byte(f, n->promisc); 1275 qemu_put_byte(f, n->allmulti); 1276 qemu_put_be32(f, n->mac_table.in_use); 1277 qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN); 1278 qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3); 1279 qemu_put_be32(f, n->has_vnet_hdr); 1280 qemu_put_byte(f, n->mac_table.multi_overflow); 1281 qemu_put_byte(f, n->mac_table.uni_overflow); 1282 qemu_put_byte(f, n->alluni); 1283 qemu_put_byte(f, n->nomulti); 1284 qemu_put_byte(f, n->nouni); 1285 qemu_put_byte(f, n->nobcast); 1286 qemu_put_byte(f, n->has_ufo); 1287 if (n->max_queues > 1) { 1288 qemu_put_be16(f, n->max_queues); 1289 qemu_put_be16(f, n->curr_queues); 1290 for (i = 1; i < n->curr_queues; i++) { 1291 qemu_put_be32(f, n->vqs[i].tx_waiting); 1292 } 1293 } 1294 1295 if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) { 1296 qemu_put_be64(f, n->curr_guest_offloads); 1297 } 1298 } 1299 1300 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id) 1301 { 1302 VirtIONet *n = opaque; 1303 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1304 int ret, i, link_down; 1305 1306 if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION) 1307 return -EINVAL; 1308 1309 ret = virtio_load(vdev, f); 1310 if (ret) { 1311 return ret; 1312 } 1313 1314 qemu_get_buffer(f, n->mac, ETH_ALEN); 1315 n->vqs[0].tx_waiting = qemu_get_be32(f); 1316 1317 virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f)); 1318 1319 if (version_id >= 3) 1320 n->status = qemu_get_be16(f); 1321 1322 if (version_id >= 4) { 1323 if (version_id < 8) { 1324 n->promisc = qemu_get_be32(f); 1325 n->allmulti = qemu_get_be32(f); 1326 } else { 1327 n->promisc = qemu_get_byte(f); 1328 n->allmulti = qemu_get_byte(f); 1329 } 1330 } 1331 1332 if (version_id >= 5) { 1333 n->mac_table.in_use = qemu_get_be32(f); 1334 /* MAC_TABLE_ENTRIES may be different from the saved image */ 1335 if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) { 1336 qemu_get_buffer(f, n->mac_table.macs, 1337 n->mac_table.in_use * ETH_ALEN); 1338 } else if (n->mac_table.in_use) { 1339 uint8_t *buf = g_malloc0(n->mac_table.in_use); 1340 qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN); 1341 g_free(buf); 1342 n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1; 1343 n->mac_table.in_use = 0; 1344 } 1345 } 1346 1347 if (version_id >= 6) 1348 qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3); 1349 1350 if (version_id >= 7) { 1351 if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) { 1352 error_report("virtio-net: saved image requires vnet_hdr=on"); 1353 return -1; 1354 } 1355 } 1356 1357 if (version_id >= 9) { 1358 n->mac_table.multi_overflow = qemu_get_byte(f); 1359 n->mac_table.uni_overflow = qemu_get_byte(f); 1360 } 1361 1362 if (version_id >= 10) { 1363 n->alluni = qemu_get_byte(f); 1364 n->nomulti = qemu_get_byte(f); 1365 n->nouni = qemu_get_byte(f); 1366 n->nobcast = qemu_get_byte(f); 1367 } 1368 1369 if (version_id >= 11) { 1370 if (qemu_get_byte(f) && !peer_has_ufo(n)) { 1371 error_report("virtio-net: saved image requires TUN_F_UFO support"); 1372 return -1; 1373 } 1374 } 1375 1376 if (n->max_queues > 1) { 1377 if (n->max_queues != qemu_get_be16(f)) { 1378 error_report("virtio-net: different max_queues "); 1379 return -1; 1380 } 1381 1382 n->curr_queues = qemu_get_be16(f); 1383 for (i = 1; i < n->curr_queues; i++) { 1384 n->vqs[i].tx_waiting = qemu_get_be32(f); 1385 } 1386 } 1387 1388 if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) { 1389 n->curr_guest_offloads = qemu_get_be64(f); 1390 } else { 1391 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 1392 } 1393 1394 if (peer_has_vnet_hdr(n)) { 1395 virtio_net_apply_guest_offloads(n); 1396 } 1397 1398 virtio_net_set_queues(n); 1399 1400 /* Find the first multicast entry in the saved MAC filter */ 1401 for (i = 0; i < n->mac_table.in_use; i++) { 1402 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 1403 break; 1404 } 1405 } 1406 n->mac_table.first_multi = i; 1407 1408 /* nc.link_down can't be migrated, so infer link_down according 1409 * to link status bit in n->status */ 1410 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 1411 for (i = 0; i < n->max_queues; i++) { 1412 qemu_get_subqueue(n->nic, i)->link_down = link_down; 1413 } 1414 1415 return 0; 1416 } 1417 1418 static void virtio_net_cleanup(NetClientState *nc) 1419 { 1420 VirtIONet *n = qemu_get_nic_opaque(nc); 1421 1422 n->nic = NULL; 1423 } 1424 1425 static NetClientInfo net_virtio_info = { 1426 .type = NET_CLIENT_OPTIONS_KIND_NIC, 1427 .size = sizeof(NICState), 1428 .can_receive = virtio_net_can_receive, 1429 .receive = virtio_net_receive, 1430 .cleanup = virtio_net_cleanup, 1431 .link_status_changed = virtio_net_set_link_status, 1432 .query_rx_filter = virtio_net_query_rxfilter, 1433 }; 1434 1435 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 1436 { 1437 VirtIONet *n = VIRTIO_NET(vdev); 1438 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1439 assert(n->vhost_started); 1440 return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx); 1441 } 1442 1443 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 1444 bool mask) 1445 { 1446 VirtIONet *n = VIRTIO_NET(vdev); 1447 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1448 assert(n->vhost_started); 1449 vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer), 1450 vdev, idx, mask); 1451 } 1452 1453 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features) 1454 { 1455 int i, config_size = 0; 1456 host_features |= (1 << VIRTIO_NET_F_MAC); 1457 for (i = 0; feature_sizes[i].flags != 0; i++) { 1458 if (host_features & feature_sizes[i].flags) { 1459 config_size = MAX(feature_sizes[i].end, config_size); 1460 } 1461 } 1462 n->config_size = config_size; 1463 } 1464 1465 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 1466 const char *type) 1467 { 1468 /* 1469 * The name can be NULL, the netclient name will be type.x. 1470 */ 1471 assert(type != NULL); 1472 1473 if (n->netclient_name) { 1474 g_free(n->netclient_name); 1475 n->netclient_name = NULL; 1476 } 1477 if (n->netclient_type) { 1478 g_free(n->netclient_type); 1479 n->netclient_type = NULL; 1480 } 1481 1482 if (name != NULL) { 1483 n->netclient_name = g_strdup(name); 1484 } 1485 n->netclient_type = g_strdup(type); 1486 } 1487 1488 static int virtio_net_device_init(VirtIODevice *vdev) 1489 { 1490 int i; 1491 1492 DeviceState *qdev = DEVICE(vdev); 1493 VirtIONet *n = VIRTIO_NET(vdev); 1494 NetClientState *nc; 1495 1496 virtio_init(VIRTIO_DEVICE(n), "virtio-net", VIRTIO_ID_NET, 1497 n->config_size); 1498 1499 n->max_queues = MAX(n->nic_conf.queues, 1); 1500 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues); 1501 n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx); 1502 n->curr_queues = 1; 1503 n->vqs[0].n = n; 1504 n->tx_timeout = n->net_conf.txtimer; 1505 1506 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 1507 && strcmp(n->net_conf.tx, "bh")) { 1508 error_report("virtio-net: " 1509 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 1510 n->net_conf.tx); 1511 error_report("Defaulting to \"bh\""); 1512 } 1513 1514 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 1515 n->vqs[0].tx_vq = virtio_add_queue(vdev, 256, 1516 virtio_net_handle_tx_timer); 1517 n->vqs[0].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, virtio_net_tx_timer, 1518 &n->vqs[0]); 1519 } else { 1520 n->vqs[0].tx_vq = virtio_add_queue(vdev, 256, 1521 virtio_net_handle_tx_bh); 1522 n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]); 1523 } 1524 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1525 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 1526 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 1527 n->status = VIRTIO_NET_S_LINK_UP; 1528 1529 if (n->netclient_type) { 1530 /* 1531 * Happen when virtio_net_set_netclient_name has been called. 1532 */ 1533 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 1534 n->netclient_type, n->netclient_name, n); 1535 } else { 1536 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 1537 object_get_typename(OBJECT(qdev)), qdev->id, n); 1538 } 1539 1540 peer_test_vnet_hdr(n); 1541 if (peer_has_vnet_hdr(n)) { 1542 for (i = 0; i < n->max_queues; i++) { 1543 tap_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true); 1544 } 1545 n->host_hdr_len = sizeof(struct virtio_net_hdr); 1546 } else { 1547 n->host_hdr_len = 0; 1548 } 1549 1550 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 1551 1552 n->vqs[0].tx_waiting = 0; 1553 n->tx_burst = n->net_conf.txburst; 1554 virtio_net_set_mrg_rx_bufs(n, 0); 1555 n->promisc = 1; /* for compatibility */ 1556 1557 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 1558 1559 n->vlans = g_malloc0(MAX_VLAN >> 3); 1560 1561 nc = qemu_get_queue(n->nic); 1562 nc->rxfilter_notify_enabled = 1; 1563 1564 n->qdev = qdev; 1565 register_savevm(qdev, "virtio-net", -1, VIRTIO_NET_VM_VERSION, 1566 virtio_net_save, virtio_net_load, n); 1567 1568 add_boot_device_path(n->nic_conf.bootindex, qdev, "/ethernet-phy@0"); 1569 return 0; 1570 } 1571 1572 static int virtio_net_device_exit(DeviceState *qdev) 1573 { 1574 VirtIONet *n = VIRTIO_NET(qdev); 1575 VirtIODevice *vdev = VIRTIO_DEVICE(qdev); 1576 int i; 1577 1578 /* This will stop vhost backend if appropriate. */ 1579 virtio_net_set_status(vdev, 0); 1580 1581 unregister_savevm(qdev, "virtio-net", n); 1582 1583 if (n->netclient_name) { 1584 g_free(n->netclient_name); 1585 n->netclient_name = NULL; 1586 } 1587 if (n->netclient_type) { 1588 g_free(n->netclient_type); 1589 n->netclient_type = NULL; 1590 } 1591 1592 g_free(n->mac_table.macs); 1593 g_free(n->vlans); 1594 1595 for (i = 0; i < n->max_queues; i++) { 1596 VirtIONetQueue *q = &n->vqs[i]; 1597 NetClientState *nc = qemu_get_subqueue(n->nic, i); 1598 1599 qemu_purge_queued_packets(nc); 1600 1601 if (q->tx_timer) { 1602 timer_del(q->tx_timer); 1603 timer_free(q->tx_timer); 1604 } else { 1605 qemu_bh_delete(q->tx_bh); 1606 } 1607 } 1608 1609 g_free(n->vqs); 1610 qemu_del_nic(n->nic); 1611 virtio_cleanup(vdev); 1612 1613 return 0; 1614 } 1615 1616 static void virtio_net_instance_init(Object *obj) 1617 { 1618 VirtIONet *n = VIRTIO_NET(obj); 1619 1620 /* 1621 * The default config_size is sizeof(struct virtio_net_config). 1622 * Can be overriden with virtio_net_set_config_size. 1623 */ 1624 n->config_size = sizeof(struct virtio_net_config); 1625 } 1626 1627 static Property virtio_net_properties[] = { 1628 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 1629 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 1630 TX_TIMER_INTERVAL), 1631 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 1632 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 1633 DEFINE_PROP_END_OF_LIST(), 1634 }; 1635 1636 static void virtio_net_class_init(ObjectClass *klass, void *data) 1637 { 1638 DeviceClass *dc = DEVICE_CLASS(klass); 1639 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 1640 dc->exit = virtio_net_device_exit; 1641 dc->props = virtio_net_properties; 1642 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 1643 vdc->init = virtio_net_device_init; 1644 vdc->get_config = virtio_net_get_config; 1645 vdc->set_config = virtio_net_set_config; 1646 vdc->get_features = virtio_net_get_features; 1647 vdc->set_features = virtio_net_set_features; 1648 vdc->bad_features = virtio_net_bad_features; 1649 vdc->reset = virtio_net_reset; 1650 vdc->set_status = virtio_net_set_status; 1651 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 1652 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 1653 } 1654 1655 static const TypeInfo virtio_net_info = { 1656 .name = TYPE_VIRTIO_NET, 1657 .parent = TYPE_VIRTIO_DEVICE, 1658 .instance_size = sizeof(VirtIONet), 1659 .instance_init = virtio_net_instance_init, 1660 .class_init = virtio_net_class_init, 1661 }; 1662 1663 static void virtio_register_types(void) 1664 { 1665 type_register_static(&virtio_net_info); 1666 } 1667 1668 type_init(virtio_register_types) 1669