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