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 int r; 401 402 for (i = 0; i < n->max_queues; i++) { 403 if (i < n->curr_queues) { 404 r = peer_attach(n, i); 405 assert(!r); 406 } else { 407 r = peer_detach(n, i); 408 assert(!r); 409 } 410 } 411 } 412 413 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); 414 415 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features) 416 { 417 VirtIONet *n = VIRTIO_NET(vdev); 418 NetClientState *nc = qemu_get_queue(n->nic); 419 420 features |= (1 << VIRTIO_NET_F_MAC); 421 422 if (!peer_has_vnet_hdr(n)) { 423 features &= ~(0x1 << VIRTIO_NET_F_CSUM); 424 features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4); 425 features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6); 426 features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN); 427 428 features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM); 429 features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4); 430 features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6); 431 features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN); 432 } 433 434 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 435 features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO); 436 features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO); 437 } 438 439 if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 440 return features; 441 } 442 if (!tap_get_vhost_net(nc->peer)) { 443 return features; 444 } 445 return vhost_net_get_features(tap_get_vhost_net(nc->peer), features); 446 } 447 448 static uint32_t virtio_net_bad_features(VirtIODevice *vdev) 449 { 450 uint32_t features = 0; 451 452 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 453 * but also these: */ 454 features |= (1 << VIRTIO_NET_F_MAC); 455 features |= (1 << VIRTIO_NET_F_CSUM); 456 features |= (1 << VIRTIO_NET_F_HOST_TSO4); 457 features |= (1 << VIRTIO_NET_F_HOST_TSO6); 458 features |= (1 << VIRTIO_NET_F_HOST_ECN); 459 460 return features; 461 } 462 463 static void virtio_net_apply_guest_offloads(VirtIONet *n) 464 { 465 qemu_set_offload(qemu_get_queue(n->nic)->peer, 466 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 467 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 468 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 469 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 470 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO))); 471 } 472 473 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features) 474 { 475 static const uint64_t guest_offloads_mask = 476 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 477 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 478 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 479 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 480 (1ULL << VIRTIO_NET_F_GUEST_UFO); 481 482 return guest_offloads_mask & features; 483 } 484 485 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n) 486 { 487 VirtIODevice *vdev = VIRTIO_DEVICE(n); 488 return virtio_net_guest_offloads_by_features(vdev->guest_features); 489 } 490 491 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features) 492 { 493 VirtIONet *n = VIRTIO_NET(vdev); 494 int i; 495 496 virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ))); 497 498 virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF))); 499 500 if (n->has_vnet_hdr) { 501 n->curr_guest_offloads = 502 virtio_net_guest_offloads_by_features(features); 503 virtio_net_apply_guest_offloads(n); 504 } 505 506 for (i = 0; i < n->max_queues; i++) { 507 NetClientState *nc = qemu_get_subqueue(n->nic, i); 508 509 if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) { 510 continue; 511 } 512 if (!tap_get_vhost_net(nc->peer)) { 513 continue; 514 } 515 vhost_net_ack_features(tap_get_vhost_net(nc->peer), features); 516 } 517 518 if ((1 << VIRTIO_NET_F_CTRL_VLAN) & features) { 519 memset(n->vlans, 0, MAX_VLAN >> 3); 520 } else { 521 memset(n->vlans, 0xff, MAX_VLAN >> 3); 522 } 523 } 524 525 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 526 struct iovec *iov, unsigned int iov_cnt) 527 { 528 uint8_t on; 529 size_t s; 530 NetClientState *nc = qemu_get_queue(n->nic); 531 532 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 533 if (s != sizeof(on)) { 534 return VIRTIO_NET_ERR; 535 } 536 537 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 538 n->promisc = on; 539 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 540 n->allmulti = on; 541 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 542 n->alluni = on; 543 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 544 n->nomulti = on; 545 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 546 n->nouni = on; 547 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 548 n->nobcast = on; 549 } else { 550 return VIRTIO_NET_ERR; 551 } 552 553 rxfilter_notify(nc); 554 555 return VIRTIO_NET_OK; 556 } 557 558 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 559 struct iovec *iov, unsigned int iov_cnt) 560 { 561 VirtIODevice *vdev = VIRTIO_DEVICE(n); 562 uint64_t offloads; 563 size_t s; 564 565 if (!((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features)) { 566 return VIRTIO_NET_ERR; 567 } 568 569 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 570 if (s != sizeof(offloads)) { 571 return VIRTIO_NET_ERR; 572 } 573 574 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 575 uint64_t supported_offloads; 576 577 if (!n->has_vnet_hdr) { 578 return VIRTIO_NET_ERR; 579 } 580 581 supported_offloads = virtio_net_supported_guest_offloads(n); 582 if (offloads & ~supported_offloads) { 583 return VIRTIO_NET_ERR; 584 } 585 586 n->curr_guest_offloads = offloads; 587 virtio_net_apply_guest_offloads(n); 588 589 return VIRTIO_NET_OK; 590 } else { 591 return VIRTIO_NET_ERR; 592 } 593 } 594 595 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 596 struct iovec *iov, unsigned int iov_cnt) 597 { 598 struct virtio_net_ctrl_mac mac_data; 599 size_t s; 600 NetClientState *nc = qemu_get_queue(n->nic); 601 602 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 603 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 604 return VIRTIO_NET_ERR; 605 } 606 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 607 assert(s == sizeof(n->mac)); 608 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 609 rxfilter_notify(nc); 610 611 return VIRTIO_NET_OK; 612 } 613 614 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 615 return VIRTIO_NET_ERR; 616 } 617 618 int in_use = 0; 619 int first_multi = 0; 620 uint8_t uni_overflow = 0; 621 uint8_t multi_overflow = 0; 622 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 623 624 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 625 sizeof(mac_data.entries)); 626 mac_data.entries = ldl_p(&mac_data.entries); 627 if (s != sizeof(mac_data.entries)) { 628 goto error; 629 } 630 iov_discard_front(&iov, &iov_cnt, s); 631 632 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 633 goto error; 634 } 635 636 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 637 s = iov_to_buf(iov, iov_cnt, 0, macs, 638 mac_data.entries * ETH_ALEN); 639 if (s != mac_data.entries * ETH_ALEN) { 640 goto error; 641 } 642 in_use += mac_data.entries; 643 } else { 644 uni_overflow = 1; 645 } 646 647 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 648 649 first_multi = in_use; 650 651 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 652 sizeof(mac_data.entries)); 653 mac_data.entries = ldl_p(&mac_data.entries); 654 if (s != sizeof(mac_data.entries)) { 655 goto error; 656 } 657 658 iov_discard_front(&iov, &iov_cnt, s); 659 660 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 661 goto error; 662 } 663 664 if (in_use + mac_data.entries <= MAC_TABLE_ENTRIES) { 665 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN], 666 mac_data.entries * ETH_ALEN); 667 if (s != mac_data.entries * ETH_ALEN) { 668 goto error; 669 } 670 in_use += mac_data.entries; 671 } else { 672 multi_overflow = 1; 673 } 674 675 n->mac_table.in_use = in_use; 676 n->mac_table.first_multi = first_multi; 677 n->mac_table.uni_overflow = uni_overflow; 678 n->mac_table.multi_overflow = multi_overflow; 679 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN); 680 g_free(macs); 681 rxfilter_notify(nc); 682 683 return VIRTIO_NET_OK; 684 685 error: 686 g_free(macs); 687 return VIRTIO_NET_ERR; 688 } 689 690 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 691 struct iovec *iov, unsigned int iov_cnt) 692 { 693 uint16_t vid; 694 size_t s; 695 NetClientState *nc = qemu_get_queue(n->nic); 696 697 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 698 vid = lduw_p(&vid); 699 if (s != sizeof(vid)) { 700 return VIRTIO_NET_ERR; 701 } 702 703 if (vid >= MAX_VLAN) 704 return VIRTIO_NET_ERR; 705 706 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 707 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 708 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 709 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 710 else 711 return VIRTIO_NET_ERR; 712 713 rxfilter_notify(nc); 714 715 return VIRTIO_NET_OK; 716 } 717 718 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 719 struct iovec *iov, unsigned int iov_cnt) 720 { 721 VirtIODevice *vdev = VIRTIO_DEVICE(n); 722 struct virtio_net_ctrl_mq mq; 723 size_t s; 724 uint16_t queues; 725 726 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 727 if (s != sizeof(mq)) { 728 return VIRTIO_NET_ERR; 729 } 730 731 if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 732 return VIRTIO_NET_ERR; 733 } 734 735 queues = lduw_p(&mq.virtqueue_pairs); 736 737 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 738 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 739 queues > n->max_queues || 740 !n->multiqueue) { 741 return VIRTIO_NET_ERR; 742 } 743 744 n->curr_queues = queues; 745 /* stop the backend before changing the number of queues to avoid handling a 746 * disabled queue */ 747 virtio_net_set_status(vdev, vdev->status); 748 virtio_net_set_queues(n); 749 750 return VIRTIO_NET_OK; 751 } 752 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 753 { 754 VirtIONet *n = VIRTIO_NET(vdev); 755 struct virtio_net_ctrl_hdr ctrl; 756 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 757 VirtQueueElement elem; 758 size_t s; 759 struct iovec *iov; 760 unsigned int iov_cnt; 761 762 while (virtqueue_pop(vq, &elem)) { 763 if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) || 764 iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) { 765 error_report("virtio-net ctrl missing headers"); 766 exit(1); 767 } 768 769 iov = elem.out_sg; 770 iov_cnt = elem.out_num; 771 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl)); 772 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl)); 773 if (s != sizeof(ctrl)) { 774 status = VIRTIO_NET_ERR; 775 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 776 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt); 777 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 778 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt); 779 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 780 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt); 781 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 782 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt); 783 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 784 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt); 785 } 786 787 s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status)); 788 assert(s == sizeof(status)); 789 790 virtqueue_push(vq, &elem, sizeof(status)); 791 virtio_notify(vdev, vq); 792 } 793 } 794 795 /* RX */ 796 797 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 798 { 799 VirtIONet *n = VIRTIO_NET(vdev); 800 int queue_index = vq2q(virtio_get_queue_index(vq)); 801 802 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 803 } 804 805 static int virtio_net_can_receive(NetClientState *nc) 806 { 807 VirtIONet *n = qemu_get_nic_opaque(nc); 808 VirtIODevice *vdev = VIRTIO_DEVICE(n); 809 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 810 811 if (!vdev->vm_running) { 812 return 0; 813 } 814 815 if (nc->queue_index >= n->curr_queues) { 816 return 0; 817 } 818 819 if (!virtio_queue_ready(q->rx_vq) || 820 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 821 return 0; 822 } 823 824 return 1; 825 } 826 827 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 828 { 829 VirtIONet *n = q->n; 830 if (virtio_queue_empty(q->rx_vq) || 831 (n->mergeable_rx_bufs && 832 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 833 virtio_queue_set_notification(q->rx_vq, 1); 834 835 /* To avoid a race condition where the guest has made some buffers 836 * available after the above check but before notification was 837 * enabled, check for available buffers again. 838 */ 839 if (virtio_queue_empty(q->rx_vq) || 840 (n->mergeable_rx_bufs && 841 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 842 return 0; 843 } 844 } 845 846 virtio_queue_set_notification(q->rx_vq, 0); 847 return 1; 848 } 849 850 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 851 * it never finds out that the packets don't have valid checksums. This 852 * causes dhclient to get upset. Fedora's carried a patch for ages to 853 * fix this with Xen but it hasn't appeared in an upstream release of 854 * dhclient yet. 855 * 856 * To avoid breaking existing guests, we catch udp packets and add 857 * checksums. This is terrible but it's better than hacking the guest 858 * kernels. 859 * 860 * N.B. if we introduce a zero-copy API, this operation is no longer free so 861 * we should provide a mechanism to disable it to avoid polluting the host 862 * cache. 863 */ 864 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 865 uint8_t *buf, size_t size) 866 { 867 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 868 (size > 27 && size < 1500) && /* normal sized MTU */ 869 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 870 (buf[23] == 17) && /* ip.protocol == UDP */ 871 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 872 net_checksum_calculate(buf, size); 873 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 874 } 875 } 876 877 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt, 878 const void *buf, size_t size) 879 { 880 if (n->has_vnet_hdr) { 881 /* FIXME this cast is evil */ 882 void *wbuf = (void *)buf; 883 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, 884 size - n->host_hdr_len); 885 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr)); 886 } else { 887 struct virtio_net_hdr hdr = { 888 .flags = 0, 889 .gso_type = VIRTIO_NET_HDR_GSO_NONE 890 }; 891 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr); 892 } 893 } 894 895 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 896 { 897 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 898 static const uint8_t vlan[] = {0x81, 0x00}; 899 uint8_t *ptr = (uint8_t *)buf; 900 int i; 901 902 if (n->promisc) 903 return 1; 904 905 ptr += n->host_hdr_len; 906 907 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 908 int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff; 909 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 910 return 0; 911 } 912 913 if (ptr[0] & 1) { // multicast 914 if (!memcmp(ptr, bcast, sizeof(bcast))) { 915 return !n->nobcast; 916 } else if (n->nomulti) { 917 return 0; 918 } else if (n->allmulti || n->mac_table.multi_overflow) { 919 return 1; 920 } 921 922 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 923 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 924 return 1; 925 } 926 } 927 } else { // unicast 928 if (n->nouni) { 929 return 0; 930 } else if (n->alluni || n->mac_table.uni_overflow) { 931 return 1; 932 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 933 return 1; 934 } 935 936 for (i = 0; i < n->mac_table.first_multi; i++) { 937 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 938 return 1; 939 } 940 } 941 } 942 943 return 0; 944 } 945 946 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size) 947 { 948 VirtIONet *n = qemu_get_nic_opaque(nc); 949 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 950 VirtIODevice *vdev = VIRTIO_DEVICE(n); 951 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 952 struct virtio_net_hdr_mrg_rxbuf mhdr; 953 unsigned mhdr_cnt = 0; 954 size_t offset, i, guest_offset; 955 956 if (!virtio_net_can_receive(nc)) { 957 return -1; 958 } 959 960 /* hdr_len refers to the header we supply to the guest */ 961 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 962 return 0; 963 } 964 965 if (!receive_filter(n, buf, size)) 966 return size; 967 968 offset = i = 0; 969 970 while (offset < size) { 971 VirtQueueElement elem; 972 int len, total; 973 const struct iovec *sg = elem.in_sg; 974 975 total = 0; 976 977 if (virtqueue_pop(q->rx_vq, &elem) == 0) { 978 if (i == 0) 979 return -1; 980 error_report("virtio-net unexpected empty queue: " 981 "i %zd mergeable %d offset %zd, size %zd, " 982 "guest hdr len %zd, host hdr len %zd guest features 0x%x", 983 i, n->mergeable_rx_bufs, offset, size, 984 n->guest_hdr_len, n->host_hdr_len, vdev->guest_features); 985 exit(1); 986 } 987 988 if (elem.in_num < 1) { 989 error_report("virtio-net receive queue contains no in buffers"); 990 exit(1); 991 } 992 993 if (i == 0) { 994 assert(offset == 0); 995 if (n->mergeable_rx_bufs) { 996 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 997 sg, elem.in_num, 998 offsetof(typeof(mhdr), num_buffers), 999 sizeof(mhdr.num_buffers)); 1000 } 1001 1002 receive_header(n, sg, elem.in_num, buf, size); 1003 offset = n->host_hdr_len; 1004 total += n->guest_hdr_len; 1005 guest_offset = n->guest_hdr_len; 1006 } else { 1007 guest_offset = 0; 1008 } 1009 1010 /* copy in packet. ugh */ 1011 len = iov_from_buf(sg, elem.in_num, guest_offset, 1012 buf + offset, size - offset); 1013 total += len; 1014 offset += len; 1015 /* If buffers can't be merged, at this point we 1016 * must have consumed the complete packet. 1017 * Otherwise, drop it. */ 1018 if (!n->mergeable_rx_bufs && offset < size) { 1019 #if 0 1020 error_report("virtio-net truncated non-mergeable packet: " 1021 "i %zd mergeable %d offset %zd, size %zd, " 1022 "guest hdr len %zd, host hdr len %zd", 1023 i, n->mergeable_rx_bufs, 1024 offset, size, n->guest_hdr_len, n->host_hdr_len); 1025 #endif 1026 return size; 1027 } 1028 1029 /* signal other side */ 1030 virtqueue_fill(q->rx_vq, &elem, total, i++); 1031 } 1032 1033 if (mhdr_cnt) { 1034 stw_p(&mhdr.num_buffers, i); 1035 iov_from_buf(mhdr_sg, mhdr_cnt, 1036 0, 1037 &mhdr.num_buffers, sizeof mhdr.num_buffers); 1038 } 1039 1040 virtqueue_flush(q->rx_vq, i); 1041 virtio_notify(vdev, q->rx_vq); 1042 1043 return size; 1044 } 1045 1046 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 1047 1048 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 1049 { 1050 VirtIONet *n = qemu_get_nic_opaque(nc); 1051 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1052 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1053 1054 virtqueue_push(q->tx_vq, &q->async_tx.elem, 0); 1055 virtio_notify(vdev, q->tx_vq); 1056 1057 q->async_tx.elem.out_num = q->async_tx.len = 0; 1058 1059 virtio_queue_set_notification(q->tx_vq, 1); 1060 virtio_net_flush_tx(q); 1061 } 1062 1063 /* TX */ 1064 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 1065 { 1066 VirtIONet *n = q->n; 1067 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1068 VirtQueueElement elem; 1069 int32_t num_packets = 0; 1070 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 1071 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1072 return num_packets; 1073 } 1074 1075 assert(vdev->vm_running); 1076 1077 if (q->async_tx.elem.out_num) { 1078 virtio_queue_set_notification(q->tx_vq, 0); 1079 return num_packets; 1080 } 1081 1082 while (virtqueue_pop(q->tx_vq, &elem)) { 1083 ssize_t ret, len; 1084 unsigned int out_num = elem.out_num; 1085 struct iovec *out_sg = &elem.out_sg[0]; 1086 struct iovec sg[VIRTQUEUE_MAX_SIZE]; 1087 1088 if (out_num < 1) { 1089 error_report("virtio-net header not in first element"); 1090 exit(1); 1091 } 1092 1093 /* 1094 * If host wants to see the guest header as is, we can 1095 * pass it on unchanged. Otherwise, copy just the parts 1096 * that host is interested in. 1097 */ 1098 assert(n->host_hdr_len <= n->guest_hdr_len); 1099 if (n->host_hdr_len != n->guest_hdr_len) { 1100 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 1101 out_sg, out_num, 1102 0, n->host_hdr_len); 1103 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 1104 out_sg, out_num, 1105 n->guest_hdr_len, -1); 1106 out_num = sg_num; 1107 out_sg = sg; 1108 } 1109 1110 len = n->guest_hdr_len; 1111 1112 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 1113 out_sg, out_num, virtio_net_tx_complete); 1114 if (ret == 0) { 1115 virtio_queue_set_notification(q->tx_vq, 0); 1116 q->async_tx.elem = elem; 1117 q->async_tx.len = len; 1118 return -EBUSY; 1119 } 1120 1121 len += ret; 1122 1123 virtqueue_push(q->tx_vq, &elem, 0); 1124 virtio_notify(vdev, q->tx_vq); 1125 1126 if (++num_packets >= n->tx_burst) { 1127 break; 1128 } 1129 } 1130 return num_packets; 1131 } 1132 1133 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 1134 { 1135 VirtIONet *n = VIRTIO_NET(vdev); 1136 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1137 1138 /* This happens when device was stopped but VCPU wasn't. */ 1139 if (!vdev->vm_running) { 1140 q->tx_waiting = 1; 1141 return; 1142 } 1143 1144 if (q->tx_waiting) { 1145 virtio_queue_set_notification(vq, 1); 1146 timer_del(q->tx_timer); 1147 q->tx_waiting = 0; 1148 virtio_net_flush_tx(q); 1149 } else { 1150 timer_mod(q->tx_timer, 1151 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 1152 q->tx_waiting = 1; 1153 virtio_queue_set_notification(vq, 0); 1154 } 1155 } 1156 1157 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 1158 { 1159 VirtIONet *n = VIRTIO_NET(vdev); 1160 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 1161 1162 if (unlikely(q->tx_waiting)) { 1163 return; 1164 } 1165 q->tx_waiting = 1; 1166 /* This happens when device was stopped but VCPU wasn't. */ 1167 if (!vdev->vm_running) { 1168 return; 1169 } 1170 virtio_queue_set_notification(vq, 0); 1171 qemu_bh_schedule(q->tx_bh); 1172 } 1173 1174 static void virtio_net_tx_timer(void *opaque) 1175 { 1176 VirtIONetQueue *q = opaque; 1177 VirtIONet *n = q->n; 1178 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1179 assert(vdev->vm_running); 1180 1181 q->tx_waiting = 0; 1182 1183 /* Just in case the driver is not ready on more */ 1184 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1185 return; 1186 } 1187 1188 virtio_queue_set_notification(q->tx_vq, 1); 1189 virtio_net_flush_tx(q); 1190 } 1191 1192 static void virtio_net_tx_bh(void *opaque) 1193 { 1194 VirtIONetQueue *q = opaque; 1195 VirtIONet *n = q->n; 1196 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1197 int32_t ret; 1198 1199 assert(vdev->vm_running); 1200 1201 q->tx_waiting = 0; 1202 1203 /* Just in case the driver is not ready on more */ 1204 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 1205 return; 1206 } 1207 1208 ret = virtio_net_flush_tx(q); 1209 if (ret == -EBUSY) { 1210 return; /* Notification re-enable handled by tx_complete */ 1211 } 1212 1213 /* If we flush a full burst of packets, assume there are 1214 * more coming and immediately reschedule */ 1215 if (ret >= n->tx_burst) { 1216 qemu_bh_schedule(q->tx_bh); 1217 q->tx_waiting = 1; 1218 return; 1219 } 1220 1221 /* If less than a full burst, re-enable notification and flush 1222 * anything that may have come in while we weren't looking. If 1223 * we find something, assume the guest is still active and reschedule */ 1224 virtio_queue_set_notification(q->tx_vq, 1); 1225 if (virtio_net_flush_tx(q) > 0) { 1226 virtio_queue_set_notification(q->tx_vq, 0); 1227 qemu_bh_schedule(q->tx_bh); 1228 q->tx_waiting = 1; 1229 } 1230 } 1231 1232 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 1233 { 1234 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1235 int i, max = multiqueue ? n->max_queues : 1; 1236 1237 n->multiqueue = multiqueue; 1238 1239 for (i = 2; i <= n->max_queues * 2 + 1; i++) { 1240 virtio_del_queue(vdev, i); 1241 } 1242 1243 for (i = 1; i < max; i++) { 1244 n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx); 1245 if (n->vqs[i].tx_timer) { 1246 n->vqs[i].tx_vq = 1247 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer); 1248 n->vqs[i].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 1249 virtio_net_tx_timer, 1250 &n->vqs[i]); 1251 } else { 1252 n->vqs[i].tx_vq = 1253 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh); 1254 n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]); 1255 } 1256 1257 n->vqs[i].tx_waiting = 0; 1258 n->vqs[i].n = n; 1259 } 1260 1261 /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack 1262 * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid 1263 * breaking them. 1264 */ 1265 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1266 1267 virtio_net_set_queues(n); 1268 } 1269 1270 static void virtio_net_save(QEMUFile *f, void *opaque) 1271 { 1272 int i; 1273 VirtIONet *n = opaque; 1274 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1275 1276 /* At this point, backend must be stopped, otherwise 1277 * it might keep writing to memory. */ 1278 assert(!n->vhost_started); 1279 virtio_save(vdev, f); 1280 1281 qemu_put_buffer(f, n->mac, ETH_ALEN); 1282 qemu_put_be32(f, n->vqs[0].tx_waiting); 1283 qemu_put_be32(f, n->mergeable_rx_bufs); 1284 qemu_put_be16(f, n->status); 1285 qemu_put_byte(f, n->promisc); 1286 qemu_put_byte(f, n->allmulti); 1287 qemu_put_be32(f, n->mac_table.in_use); 1288 qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN); 1289 qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3); 1290 qemu_put_be32(f, n->has_vnet_hdr); 1291 qemu_put_byte(f, n->mac_table.multi_overflow); 1292 qemu_put_byte(f, n->mac_table.uni_overflow); 1293 qemu_put_byte(f, n->alluni); 1294 qemu_put_byte(f, n->nomulti); 1295 qemu_put_byte(f, n->nouni); 1296 qemu_put_byte(f, n->nobcast); 1297 qemu_put_byte(f, n->has_ufo); 1298 if (n->max_queues > 1) { 1299 qemu_put_be16(f, n->max_queues); 1300 qemu_put_be16(f, n->curr_queues); 1301 for (i = 1; i < n->curr_queues; i++) { 1302 qemu_put_be32(f, n->vqs[i].tx_waiting); 1303 } 1304 } 1305 1306 if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) { 1307 qemu_put_be64(f, n->curr_guest_offloads); 1308 } 1309 } 1310 1311 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id) 1312 { 1313 VirtIONet *n = opaque; 1314 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1315 int ret, i, link_down; 1316 1317 if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION) 1318 return -EINVAL; 1319 1320 ret = virtio_load(vdev, f); 1321 if (ret) { 1322 return ret; 1323 } 1324 1325 qemu_get_buffer(f, n->mac, ETH_ALEN); 1326 n->vqs[0].tx_waiting = qemu_get_be32(f); 1327 1328 virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f)); 1329 1330 if (version_id >= 3) 1331 n->status = qemu_get_be16(f); 1332 1333 if (version_id >= 4) { 1334 if (version_id < 8) { 1335 n->promisc = qemu_get_be32(f); 1336 n->allmulti = qemu_get_be32(f); 1337 } else { 1338 n->promisc = qemu_get_byte(f); 1339 n->allmulti = qemu_get_byte(f); 1340 } 1341 } 1342 1343 if (version_id >= 5) { 1344 n->mac_table.in_use = qemu_get_be32(f); 1345 /* MAC_TABLE_ENTRIES may be different from the saved image */ 1346 if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) { 1347 qemu_get_buffer(f, n->mac_table.macs, 1348 n->mac_table.in_use * ETH_ALEN); 1349 } else if (n->mac_table.in_use) { 1350 uint8_t *buf = g_malloc0(n->mac_table.in_use); 1351 qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN); 1352 g_free(buf); 1353 n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1; 1354 n->mac_table.in_use = 0; 1355 } 1356 } 1357 1358 if (version_id >= 6) 1359 qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3); 1360 1361 if (version_id >= 7) { 1362 if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) { 1363 error_report("virtio-net: saved image requires vnet_hdr=on"); 1364 return -1; 1365 } 1366 } 1367 1368 if (version_id >= 9) { 1369 n->mac_table.multi_overflow = qemu_get_byte(f); 1370 n->mac_table.uni_overflow = qemu_get_byte(f); 1371 } 1372 1373 if (version_id >= 10) { 1374 n->alluni = qemu_get_byte(f); 1375 n->nomulti = qemu_get_byte(f); 1376 n->nouni = qemu_get_byte(f); 1377 n->nobcast = qemu_get_byte(f); 1378 } 1379 1380 if (version_id >= 11) { 1381 if (qemu_get_byte(f) && !peer_has_ufo(n)) { 1382 error_report("virtio-net: saved image requires TUN_F_UFO support"); 1383 return -1; 1384 } 1385 } 1386 1387 if (n->max_queues > 1) { 1388 if (n->max_queues != qemu_get_be16(f)) { 1389 error_report("virtio-net: different max_queues "); 1390 return -1; 1391 } 1392 1393 n->curr_queues = qemu_get_be16(f); 1394 for (i = 1; i < n->curr_queues; i++) { 1395 n->vqs[i].tx_waiting = qemu_get_be32(f); 1396 } 1397 } 1398 1399 if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) { 1400 n->curr_guest_offloads = qemu_get_be64(f); 1401 } else { 1402 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 1403 } 1404 1405 if (peer_has_vnet_hdr(n)) { 1406 virtio_net_apply_guest_offloads(n); 1407 } 1408 1409 virtio_net_set_queues(n); 1410 1411 /* Find the first multicast entry in the saved MAC filter */ 1412 for (i = 0; i < n->mac_table.in_use; i++) { 1413 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 1414 break; 1415 } 1416 } 1417 n->mac_table.first_multi = i; 1418 1419 /* nc.link_down can't be migrated, so infer link_down according 1420 * to link status bit in n->status */ 1421 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 1422 for (i = 0; i < n->max_queues; i++) { 1423 qemu_get_subqueue(n->nic, i)->link_down = link_down; 1424 } 1425 1426 return 0; 1427 } 1428 1429 static void virtio_net_cleanup(NetClientState *nc) 1430 { 1431 VirtIONet *n = qemu_get_nic_opaque(nc); 1432 1433 n->nic = NULL; 1434 } 1435 1436 static NetClientInfo net_virtio_info = { 1437 .type = NET_CLIENT_OPTIONS_KIND_NIC, 1438 .size = sizeof(NICState), 1439 .can_receive = virtio_net_can_receive, 1440 .receive = virtio_net_receive, 1441 .cleanup = virtio_net_cleanup, 1442 .link_status_changed = virtio_net_set_link_status, 1443 .query_rx_filter = virtio_net_query_rxfilter, 1444 }; 1445 1446 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 1447 { 1448 VirtIONet *n = VIRTIO_NET(vdev); 1449 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1450 assert(n->vhost_started); 1451 return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx); 1452 } 1453 1454 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 1455 bool mask) 1456 { 1457 VirtIONet *n = VIRTIO_NET(vdev); 1458 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 1459 assert(n->vhost_started); 1460 vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer), 1461 vdev, idx, mask); 1462 } 1463 1464 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features) 1465 { 1466 int i, config_size = 0; 1467 host_features |= (1 << VIRTIO_NET_F_MAC); 1468 for (i = 0; feature_sizes[i].flags != 0; i++) { 1469 if (host_features & feature_sizes[i].flags) { 1470 config_size = MAX(feature_sizes[i].end, config_size); 1471 } 1472 } 1473 n->config_size = config_size; 1474 } 1475 1476 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 1477 const char *type) 1478 { 1479 /* 1480 * The name can be NULL, the netclient name will be type.x. 1481 */ 1482 assert(type != NULL); 1483 1484 if (n->netclient_name) { 1485 g_free(n->netclient_name); 1486 n->netclient_name = NULL; 1487 } 1488 if (n->netclient_type) { 1489 g_free(n->netclient_type); 1490 n->netclient_type = NULL; 1491 } 1492 1493 if (name != NULL) { 1494 n->netclient_name = g_strdup(name); 1495 } 1496 n->netclient_type = g_strdup(type); 1497 } 1498 1499 static void virtio_net_device_realize(DeviceState *dev, Error **errp) 1500 { 1501 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1502 VirtIONet *n = VIRTIO_NET(dev); 1503 NetClientState *nc; 1504 int i; 1505 1506 virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size); 1507 1508 n->max_queues = MAX(n->nic_conf.queues, 1); 1509 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues); 1510 n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx); 1511 n->curr_queues = 1; 1512 n->vqs[0].n = n; 1513 n->tx_timeout = n->net_conf.txtimer; 1514 1515 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 1516 && strcmp(n->net_conf.tx, "bh")) { 1517 error_report("virtio-net: " 1518 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 1519 n->net_conf.tx); 1520 error_report("Defaulting to \"bh\""); 1521 } 1522 1523 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 1524 n->vqs[0].tx_vq = virtio_add_queue(vdev, 256, 1525 virtio_net_handle_tx_timer); 1526 n->vqs[0].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, virtio_net_tx_timer, 1527 &n->vqs[0]); 1528 } else { 1529 n->vqs[0].tx_vq = virtio_add_queue(vdev, 256, 1530 virtio_net_handle_tx_bh); 1531 n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]); 1532 } 1533 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 1534 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 1535 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 1536 n->status = VIRTIO_NET_S_LINK_UP; 1537 1538 if (n->netclient_type) { 1539 /* 1540 * Happen when virtio_net_set_netclient_name has been called. 1541 */ 1542 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 1543 n->netclient_type, n->netclient_name, n); 1544 } else { 1545 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 1546 object_get_typename(OBJECT(dev)), dev->id, n); 1547 } 1548 1549 peer_test_vnet_hdr(n); 1550 if (peer_has_vnet_hdr(n)) { 1551 for (i = 0; i < n->max_queues; i++) { 1552 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true); 1553 } 1554 n->host_hdr_len = sizeof(struct virtio_net_hdr); 1555 } else { 1556 n->host_hdr_len = 0; 1557 } 1558 1559 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 1560 1561 n->vqs[0].tx_waiting = 0; 1562 n->tx_burst = n->net_conf.txburst; 1563 virtio_net_set_mrg_rx_bufs(n, 0); 1564 n->promisc = 1; /* for compatibility */ 1565 1566 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 1567 1568 n->vlans = g_malloc0(MAX_VLAN >> 3); 1569 1570 nc = qemu_get_queue(n->nic); 1571 nc->rxfilter_notify_enabled = 1; 1572 1573 n->qdev = dev; 1574 register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION, 1575 virtio_net_save, virtio_net_load, n); 1576 1577 add_boot_device_path(n->nic_conf.bootindex, dev, "/ethernet-phy@0"); 1578 } 1579 1580 static void virtio_net_device_unrealize(DeviceState *dev, Error **errp) 1581 { 1582 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 1583 VirtIONet *n = VIRTIO_NET(dev); 1584 int i; 1585 1586 /* This will stop vhost backend if appropriate. */ 1587 virtio_net_set_status(vdev, 0); 1588 1589 unregister_savevm(dev, "virtio-net", n); 1590 1591 if (n->netclient_name) { 1592 g_free(n->netclient_name); 1593 n->netclient_name = NULL; 1594 } 1595 if (n->netclient_type) { 1596 g_free(n->netclient_type); 1597 n->netclient_type = NULL; 1598 } 1599 1600 g_free(n->mac_table.macs); 1601 g_free(n->vlans); 1602 1603 for (i = 0; i < n->max_queues; i++) { 1604 VirtIONetQueue *q = &n->vqs[i]; 1605 NetClientState *nc = qemu_get_subqueue(n->nic, i); 1606 1607 qemu_purge_queued_packets(nc); 1608 1609 if (q->tx_timer) { 1610 timer_del(q->tx_timer); 1611 timer_free(q->tx_timer); 1612 } else if (q->tx_bh) { 1613 qemu_bh_delete(q->tx_bh); 1614 } 1615 } 1616 1617 g_free(n->vqs); 1618 qemu_del_nic(n->nic); 1619 virtio_cleanup(vdev); 1620 } 1621 1622 static void virtio_net_instance_init(Object *obj) 1623 { 1624 VirtIONet *n = VIRTIO_NET(obj); 1625 1626 /* 1627 * The default config_size is sizeof(struct virtio_net_config). 1628 * Can be overriden with virtio_net_set_config_size. 1629 */ 1630 n->config_size = sizeof(struct virtio_net_config); 1631 } 1632 1633 static Property virtio_net_properties[] = { 1634 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 1635 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 1636 TX_TIMER_INTERVAL), 1637 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 1638 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 1639 DEFINE_PROP_END_OF_LIST(), 1640 }; 1641 1642 static void virtio_net_class_init(ObjectClass *klass, void *data) 1643 { 1644 DeviceClass *dc = DEVICE_CLASS(klass); 1645 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 1646 1647 dc->props = virtio_net_properties; 1648 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 1649 vdc->realize = virtio_net_device_realize; 1650 vdc->unrealize = virtio_net_device_unrealize; 1651 vdc->get_config = virtio_net_get_config; 1652 vdc->set_config = virtio_net_set_config; 1653 vdc->get_features = virtio_net_get_features; 1654 vdc->set_features = virtio_net_set_features; 1655 vdc->bad_features = virtio_net_bad_features; 1656 vdc->reset = virtio_net_reset; 1657 vdc->set_status = virtio_net_set_status; 1658 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 1659 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 1660 } 1661 1662 static const TypeInfo virtio_net_info = { 1663 .name = TYPE_VIRTIO_NET, 1664 .parent = TYPE_VIRTIO_DEVICE, 1665 .instance_size = sizeof(VirtIONet), 1666 .instance_init = virtio_net_instance_init, 1667 .class_init = virtio_net_class_init, 1668 }; 1669 1670 static void virtio_register_types(void) 1671 { 1672 type_register_static(&virtio_net_info); 1673 } 1674 1675 type_init(virtio_register_types) 1676