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