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/osdep.h" 15 #include "qemu/atomic.h" 16 #include "qemu/iov.h" 17 #include "qemu/main-loop.h" 18 #include "qemu/module.h" 19 #include "hw/virtio/virtio.h" 20 #include "net/net.h" 21 #include "net/checksum.h" 22 #include "net/tap.h" 23 #include "qemu/error-report.h" 24 #include "qemu/timer.h" 25 #include "qemu/option.h" 26 #include "qemu/option_int.h" 27 #include "qemu/config-file.h" 28 #include "qapi/qmp/qdict.h" 29 #include "hw/virtio/virtio-net.h" 30 #include "net/vhost_net.h" 31 #include "net/announce.h" 32 #include "hw/virtio/virtio-bus.h" 33 #include "qapi/error.h" 34 #include "qapi/qapi-events-net.h" 35 #include "hw/qdev-properties.h" 36 #include "qapi/qapi-types-migration.h" 37 #include "qapi/qapi-events-migration.h" 38 #include "hw/virtio/virtio-access.h" 39 #include "migration/misc.h" 40 #include "standard-headers/linux/ethtool.h" 41 #include "sysemu/sysemu.h" 42 #include "trace.h" 43 #include "monitor/qdev.h" 44 #include "hw/pci/pci.h" 45 #include "net_rx_pkt.h" 46 #include "hw/virtio/vhost.h" 47 48 #define VIRTIO_NET_VM_VERSION 11 49 50 #define MAC_TABLE_ENTRIES 64 51 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */ 52 53 /* previously fixed value */ 54 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256 55 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256 56 57 /* for now, only allow larger queues; with virtio-1, guest can downsize */ 58 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 59 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 60 61 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */ 62 63 #define VIRTIO_NET_TCP_FLAG 0x3F 64 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000 65 66 /* IPv4 max payload, 16 bits in the header */ 67 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header)) 68 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535 69 70 /* header length value in ip header without option */ 71 #define VIRTIO_NET_IP4_HEADER_LENGTH 5 72 73 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */ 74 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD 75 76 /* Purge coalesced packets timer interval, This value affects the performance 77 a lot, and should be tuned carefully, '300000'(300us) is the recommended 78 value to pass the WHQL test, '50000' can gain 2x netperf throughput with 79 tso/gso/gro 'off'. */ 80 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000 81 82 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \ 83 VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \ 84 VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \ 85 VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \ 86 VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \ 87 VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \ 88 VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \ 89 VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \ 90 VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) 91 92 static VirtIOFeature feature_sizes[] = { 93 {.flags = 1ULL << VIRTIO_NET_F_MAC, 94 .end = endof(struct virtio_net_config, mac)}, 95 {.flags = 1ULL << VIRTIO_NET_F_STATUS, 96 .end = endof(struct virtio_net_config, status)}, 97 {.flags = 1ULL << VIRTIO_NET_F_MQ, 98 .end = endof(struct virtio_net_config, max_virtqueue_pairs)}, 99 {.flags = 1ULL << VIRTIO_NET_F_MTU, 100 .end = endof(struct virtio_net_config, mtu)}, 101 {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX, 102 .end = endof(struct virtio_net_config, duplex)}, 103 {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT), 104 .end = endof(struct virtio_net_config, supported_hash_types)}, 105 {} 106 }; 107 108 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc) 109 { 110 VirtIONet *n = qemu_get_nic_opaque(nc); 111 112 return &n->vqs[nc->queue_index]; 113 } 114 115 static int vq2q(int queue_index) 116 { 117 return queue_index / 2; 118 } 119 120 /* TODO 121 * - we could suppress RX interrupt if we were so inclined. 122 */ 123 124 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config) 125 { 126 VirtIONet *n = VIRTIO_NET(vdev); 127 struct virtio_net_config netcfg; 128 NetClientState *nc = qemu_get_queue(n->nic); 129 130 int ret = 0; 131 memset(&netcfg, 0 , sizeof(struct virtio_net_config)); 132 virtio_stw_p(vdev, &netcfg.status, n->status); 133 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queues); 134 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu); 135 memcpy(netcfg.mac, n->mac, ETH_ALEN); 136 virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed); 137 netcfg.duplex = n->net_conf.duplex; 138 netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE; 139 virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length, 140 virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ? 141 VIRTIO_NET_RSS_MAX_TABLE_LEN : 1); 142 virtio_stl_p(vdev, &netcfg.supported_hash_types, 143 VIRTIO_NET_RSS_SUPPORTED_HASHES); 144 memcpy(config, &netcfg, n->config_size); 145 146 /* 147 * Is this VDPA? No peer means not VDPA: there's no way to 148 * disconnect/reconnect a VDPA peer. 149 */ 150 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 151 ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg, 152 n->config_size); 153 if (ret != -1) { 154 memcpy(config, &netcfg, n->config_size); 155 } 156 } 157 } 158 159 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config) 160 { 161 VirtIONet *n = VIRTIO_NET(vdev); 162 struct virtio_net_config netcfg = {}; 163 NetClientState *nc = qemu_get_queue(n->nic); 164 165 memcpy(&netcfg, config, n->config_size); 166 167 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) && 168 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) && 169 memcmp(netcfg.mac, n->mac, ETH_ALEN)) { 170 memcpy(n->mac, netcfg.mac, ETH_ALEN); 171 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 172 } 173 174 /* 175 * Is this VDPA? No peer means not VDPA: there's no way to 176 * disconnect/reconnect a VDPA peer. 177 */ 178 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 179 vhost_net_set_config(get_vhost_net(nc->peer), 180 (uint8_t *)&netcfg, 0, n->config_size, 181 VHOST_SET_CONFIG_TYPE_MASTER); 182 } 183 } 184 185 static bool virtio_net_started(VirtIONet *n, uint8_t status) 186 { 187 VirtIODevice *vdev = VIRTIO_DEVICE(n); 188 return (status & VIRTIO_CONFIG_S_DRIVER_OK) && 189 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running; 190 } 191 192 static void virtio_net_announce_notify(VirtIONet *net) 193 { 194 VirtIODevice *vdev = VIRTIO_DEVICE(net); 195 trace_virtio_net_announce_notify(); 196 197 net->status |= VIRTIO_NET_S_ANNOUNCE; 198 virtio_notify_config(vdev); 199 } 200 201 static void virtio_net_announce_timer(void *opaque) 202 { 203 VirtIONet *n = opaque; 204 trace_virtio_net_announce_timer(n->announce_timer.round); 205 206 n->announce_timer.round--; 207 virtio_net_announce_notify(n); 208 } 209 210 static void virtio_net_announce(NetClientState *nc) 211 { 212 VirtIONet *n = qemu_get_nic_opaque(nc); 213 VirtIODevice *vdev = VIRTIO_DEVICE(n); 214 215 /* 216 * Make sure the virtio migration announcement timer isn't running 217 * If it is, let it trigger announcement so that we do not cause 218 * confusion. 219 */ 220 if (n->announce_timer.round) { 221 return; 222 } 223 224 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 225 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 226 virtio_net_announce_notify(n); 227 } 228 } 229 230 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status) 231 { 232 VirtIODevice *vdev = VIRTIO_DEVICE(n); 233 NetClientState *nc = qemu_get_queue(n->nic); 234 int queues = n->multiqueue ? n->max_queues : 1; 235 236 if (!get_vhost_net(nc->peer)) { 237 return; 238 } 239 240 if ((virtio_net_started(n, status) && !nc->peer->link_down) == 241 !!n->vhost_started) { 242 return; 243 } 244 if (!n->vhost_started) { 245 int r, i; 246 247 if (n->needs_vnet_hdr_swap) { 248 error_report("backend does not support %s vnet headers; " 249 "falling back on userspace virtio", 250 virtio_is_big_endian(vdev) ? "BE" : "LE"); 251 return; 252 } 253 254 /* Any packets outstanding? Purge them to avoid touching rings 255 * when vhost is running. 256 */ 257 for (i = 0; i < queues; i++) { 258 NetClientState *qnc = qemu_get_subqueue(n->nic, i); 259 260 /* Purge both directions: TX and RX. */ 261 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc); 262 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer); 263 } 264 265 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) { 266 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu); 267 if (r < 0) { 268 error_report("%uBytes MTU not supported by the backend", 269 n->net_conf.mtu); 270 271 return; 272 } 273 } 274 275 n->vhost_started = 1; 276 r = vhost_net_start(vdev, n->nic->ncs, queues); 277 if (r < 0) { 278 error_report("unable to start vhost net: %d: " 279 "falling back on userspace virtio", -r); 280 n->vhost_started = 0; 281 } 282 } else { 283 vhost_net_stop(vdev, n->nic->ncs, queues); 284 n->vhost_started = 0; 285 } 286 } 287 288 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev, 289 NetClientState *peer, 290 bool enable) 291 { 292 if (virtio_is_big_endian(vdev)) { 293 return qemu_set_vnet_be(peer, enable); 294 } else { 295 return qemu_set_vnet_le(peer, enable); 296 } 297 } 298 299 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs, 300 int queues, bool enable) 301 { 302 int i; 303 304 for (i = 0; i < queues; i++) { 305 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 && 306 enable) { 307 while (--i >= 0) { 308 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false); 309 } 310 311 return true; 312 } 313 } 314 315 return false; 316 } 317 318 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status) 319 { 320 VirtIODevice *vdev = VIRTIO_DEVICE(n); 321 int queues = n->multiqueue ? n->max_queues : 1; 322 323 if (virtio_net_started(n, status)) { 324 /* Before using the device, we tell the network backend about the 325 * endianness to use when parsing vnet headers. If the backend 326 * can't do it, we fallback onto fixing the headers in the core 327 * virtio-net code. 328 */ 329 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs, 330 queues, true); 331 } else if (virtio_net_started(n, vdev->status)) { 332 /* After using the device, we need to reset the network backend to 333 * the default (guest native endianness), otherwise the guest may 334 * lose network connectivity if it is rebooted into a different 335 * endianness. 336 */ 337 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queues, false); 338 } 339 } 340 341 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq) 342 { 343 unsigned int dropped = virtqueue_drop_all(vq); 344 if (dropped) { 345 virtio_notify(vdev, vq); 346 } 347 } 348 349 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status) 350 { 351 VirtIONet *n = VIRTIO_NET(vdev); 352 VirtIONetQueue *q; 353 int i; 354 uint8_t queue_status; 355 356 virtio_net_vnet_endian_status(n, status); 357 virtio_net_vhost_status(n, status); 358 359 for (i = 0; i < n->max_queues; i++) { 360 NetClientState *ncs = qemu_get_subqueue(n->nic, i); 361 bool queue_started; 362 q = &n->vqs[i]; 363 364 if ((!n->multiqueue && i != 0) || i >= n->curr_queues) { 365 queue_status = 0; 366 } else { 367 queue_status = status; 368 } 369 queue_started = 370 virtio_net_started(n, queue_status) && !n->vhost_started; 371 372 if (queue_started) { 373 qemu_flush_queued_packets(ncs); 374 } 375 376 if (!q->tx_waiting) { 377 continue; 378 } 379 380 if (queue_started) { 381 if (q->tx_timer) { 382 timer_mod(q->tx_timer, 383 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 384 } else { 385 qemu_bh_schedule(q->tx_bh); 386 } 387 } else { 388 if (q->tx_timer) { 389 timer_del(q->tx_timer); 390 } else { 391 qemu_bh_cancel(q->tx_bh); 392 } 393 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 && 394 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) && 395 vdev->vm_running) { 396 /* if tx is waiting we are likely have some packets in tx queue 397 * and disabled notification */ 398 q->tx_waiting = 0; 399 virtio_queue_set_notification(q->tx_vq, 1); 400 virtio_net_drop_tx_queue_data(vdev, q->tx_vq); 401 } 402 } 403 } 404 } 405 406 static void virtio_net_set_link_status(NetClientState *nc) 407 { 408 VirtIONet *n = qemu_get_nic_opaque(nc); 409 VirtIODevice *vdev = VIRTIO_DEVICE(n); 410 uint16_t old_status = n->status; 411 412 if (nc->link_down) 413 n->status &= ~VIRTIO_NET_S_LINK_UP; 414 else 415 n->status |= VIRTIO_NET_S_LINK_UP; 416 417 if (n->status != old_status) 418 virtio_notify_config(vdev); 419 420 virtio_net_set_status(vdev, vdev->status); 421 } 422 423 static void rxfilter_notify(NetClientState *nc) 424 { 425 VirtIONet *n = qemu_get_nic_opaque(nc); 426 427 if (nc->rxfilter_notify_enabled) { 428 char *path = object_get_canonical_path(OBJECT(n->qdev)); 429 qapi_event_send_nic_rx_filter_changed(!!n->netclient_name, 430 n->netclient_name, path); 431 g_free(path); 432 433 /* disable event notification to avoid events flooding */ 434 nc->rxfilter_notify_enabled = 0; 435 } 436 } 437 438 static intList *get_vlan_table(VirtIONet *n) 439 { 440 intList *list; 441 int i, j; 442 443 list = NULL; 444 for (i = 0; i < MAX_VLAN >> 5; i++) { 445 for (j = 0; n->vlans[i] && j <= 0x1f; j++) { 446 if (n->vlans[i] & (1U << j)) { 447 QAPI_LIST_PREPEND(list, (i << 5) + j); 448 } 449 } 450 } 451 452 return list; 453 } 454 455 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc) 456 { 457 VirtIONet *n = qemu_get_nic_opaque(nc); 458 VirtIODevice *vdev = VIRTIO_DEVICE(n); 459 RxFilterInfo *info; 460 strList *str_list; 461 int i; 462 463 info = g_malloc0(sizeof(*info)); 464 info->name = g_strdup(nc->name); 465 info->promiscuous = n->promisc; 466 467 if (n->nouni) { 468 info->unicast = RX_STATE_NONE; 469 } else if (n->alluni) { 470 info->unicast = RX_STATE_ALL; 471 } else { 472 info->unicast = RX_STATE_NORMAL; 473 } 474 475 if (n->nomulti) { 476 info->multicast = RX_STATE_NONE; 477 } else if (n->allmulti) { 478 info->multicast = RX_STATE_ALL; 479 } else { 480 info->multicast = RX_STATE_NORMAL; 481 } 482 483 info->broadcast_allowed = n->nobcast; 484 info->multicast_overflow = n->mac_table.multi_overflow; 485 info->unicast_overflow = n->mac_table.uni_overflow; 486 487 info->main_mac = qemu_mac_strdup_printf(n->mac); 488 489 str_list = NULL; 490 for (i = 0; i < n->mac_table.first_multi; i++) { 491 QAPI_LIST_PREPEND(str_list, 492 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN)); 493 } 494 info->unicast_table = str_list; 495 496 str_list = NULL; 497 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 498 QAPI_LIST_PREPEND(str_list, 499 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN)); 500 } 501 info->multicast_table = str_list; 502 info->vlan_table = get_vlan_table(n); 503 504 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) { 505 info->vlan = RX_STATE_ALL; 506 } else if (!info->vlan_table) { 507 info->vlan = RX_STATE_NONE; 508 } else { 509 info->vlan = RX_STATE_NORMAL; 510 } 511 512 /* enable event notification after query */ 513 nc->rxfilter_notify_enabled = 1; 514 515 return info; 516 } 517 518 static void virtio_net_reset(VirtIODevice *vdev) 519 { 520 VirtIONet *n = VIRTIO_NET(vdev); 521 int i; 522 523 /* Reset back to compatibility mode */ 524 n->promisc = 1; 525 n->allmulti = 0; 526 n->alluni = 0; 527 n->nomulti = 0; 528 n->nouni = 0; 529 n->nobcast = 0; 530 /* multiqueue is disabled by default */ 531 n->curr_queues = 1; 532 timer_del(n->announce_timer.tm); 533 n->announce_timer.round = 0; 534 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 535 536 /* Flush any MAC and VLAN filter table state */ 537 n->mac_table.in_use = 0; 538 n->mac_table.first_multi = 0; 539 n->mac_table.multi_overflow = 0; 540 n->mac_table.uni_overflow = 0; 541 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN); 542 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac)); 543 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 544 memset(n->vlans, 0, MAX_VLAN >> 3); 545 546 /* Flush any async TX */ 547 for (i = 0; i < n->max_queues; i++) { 548 NetClientState *nc = qemu_get_subqueue(n->nic, i); 549 550 if (nc->peer) { 551 qemu_flush_or_purge_queued_packets(nc->peer, true); 552 assert(!virtio_net_get_subqueue(nc)->async_tx.elem); 553 } 554 } 555 } 556 557 static void peer_test_vnet_hdr(VirtIONet *n) 558 { 559 NetClientState *nc = qemu_get_queue(n->nic); 560 if (!nc->peer) { 561 return; 562 } 563 564 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer); 565 } 566 567 static int peer_has_vnet_hdr(VirtIONet *n) 568 { 569 return n->has_vnet_hdr; 570 } 571 572 static int peer_has_ufo(VirtIONet *n) 573 { 574 if (!peer_has_vnet_hdr(n)) 575 return 0; 576 577 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer); 578 579 return n->has_ufo; 580 } 581 582 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs, 583 int version_1, int hash_report) 584 { 585 int i; 586 NetClientState *nc; 587 588 n->mergeable_rx_bufs = mergeable_rx_bufs; 589 590 if (version_1) { 591 n->guest_hdr_len = hash_report ? 592 sizeof(struct virtio_net_hdr_v1_hash) : 593 sizeof(struct virtio_net_hdr_mrg_rxbuf); 594 n->rss_data.populate_hash = !!hash_report; 595 } else { 596 n->guest_hdr_len = n->mergeable_rx_bufs ? 597 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 598 sizeof(struct virtio_net_hdr); 599 } 600 601 for (i = 0; i < n->max_queues; i++) { 602 nc = qemu_get_subqueue(n->nic, i); 603 604 if (peer_has_vnet_hdr(n) && 605 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) { 606 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len); 607 n->host_hdr_len = n->guest_hdr_len; 608 } 609 } 610 } 611 612 static int virtio_net_max_tx_queue_size(VirtIONet *n) 613 { 614 NetClientState *peer = n->nic_conf.peers.ncs[0]; 615 616 /* 617 * Backends other than vhost-user don't support max queue size. 618 */ 619 if (!peer) { 620 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 621 } 622 623 if (peer->info->type != NET_CLIENT_DRIVER_VHOST_USER) { 624 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 625 } 626 627 return VIRTQUEUE_MAX_SIZE; 628 } 629 630 static int peer_attach(VirtIONet *n, int index) 631 { 632 NetClientState *nc = qemu_get_subqueue(n->nic, index); 633 634 if (!nc->peer) { 635 return 0; 636 } 637 638 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 639 vhost_set_vring_enable(nc->peer, 1); 640 } 641 642 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 643 return 0; 644 } 645 646 if (n->max_queues == 1) { 647 return 0; 648 } 649 650 return tap_enable(nc->peer); 651 } 652 653 static int peer_detach(VirtIONet *n, int index) 654 { 655 NetClientState *nc = qemu_get_subqueue(n->nic, index); 656 657 if (!nc->peer) { 658 return 0; 659 } 660 661 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 662 vhost_set_vring_enable(nc->peer, 0); 663 } 664 665 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 666 return 0; 667 } 668 669 return tap_disable(nc->peer); 670 } 671 672 static void virtio_net_set_queues(VirtIONet *n) 673 { 674 int i; 675 int r; 676 677 if (n->nic->peer_deleted) { 678 return; 679 } 680 681 for (i = 0; i < n->max_queues; i++) { 682 if (i < n->curr_queues) { 683 r = peer_attach(n, i); 684 assert(!r); 685 } else { 686 r = peer_detach(n, i); 687 assert(!r); 688 } 689 } 690 } 691 692 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); 693 694 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features, 695 Error **errp) 696 { 697 VirtIONet *n = VIRTIO_NET(vdev); 698 NetClientState *nc = qemu_get_queue(n->nic); 699 700 /* Firstly sync all virtio-net possible supported features */ 701 features |= n->host_features; 702 703 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 704 705 if (!peer_has_vnet_hdr(n)) { 706 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM); 707 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4); 708 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6); 709 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN); 710 711 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM); 712 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4); 713 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6); 714 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN); 715 716 virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT); 717 } 718 719 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 720 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO); 721 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO); 722 } 723 724 if (!get_vhost_net(nc->peer)) { 725 return features; 726 } 727 728 virtio_clear_feature(&features, VIRTIO_NET_F_RSS); 729 virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT); 730 features = vhost_net_get_features(get_vhost_net(nc->peer), features); 731 vdev->backend_features = features; 732 733 if (n->mtu_bypass_backend && 734 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) { 735 features |= (1ULL << VIRTIO_NET_F_MTU); 736 } 737 738 return features; 739 } 740 741 static uint64_t virtio_net_bad_features(VirtIODevice *vdev) 742 { 743 uint64_t features = 0; 744 745 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 746 * but also these: */ 747 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 748 virtio_add_feature(&features, VIRTIO_NET_F_CSUM); 749 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4); 750 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6); 751 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN); 752 753 return features; 754 } 755 756 static void virtio_net_apply_guest_offloads(VirtIONet *n) 757 { 758 qemu_set_offload(qemu_get_queue(n->nic)->peer, 759 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 760 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 761 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 762 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 763 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO))); 764 } 765 766 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features) 767 { 768 static const uint64_t guest_offloads_mask = 769 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 770 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 771 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 772 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 773 (1ULL << VIRTIO_NET_F_GUEST_UFO); 774 775 return guest_offloads_mask & features; 776 } 777 778 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n) 779 { 780 VirtIODevice *vdev = VIRTIO_DEVICE(n); 781 return virtio_net_guest_offloads_by_features(vdev->guest_features); 782 } 783 784 typedef struct { 785 VirtIONet *n; 786 char *id; 787 } FailoverId; 788 789 /** 790 * Set the id of the failover primary device 791 * 792 * @opaque: FailoverId to setup 793 * @opts: opts for device we are handling 794 * @errp: returns an error if this function fails 795 */ 796 static int failover_set_primary(void *opaque, QemuOpts *opts, Error **errp) 797 { 798 FailoverId *fid = opaque; 799 const char *standby_id = qemu_opt_get(opts, "failover_pair_id"); 800 801 if (g_strcmp0(standby_id, fid->n->netclient_name) == 0) { 802 fid->id = g_strdup(opts->id); 803 return 1; 804 } 805 806 return 0; 807 } 808 809 /** 810 * Find the primary device id for this failover virtio-net 811 * 812 * @n: VirtIONet device 813 * @errp: returns an error if this function fails 814 */ 815 static char *failover_find_primary_device_id(VirtIONet *n) 816 { 817 Error *err = NULL; 818 FailoverId fid; 819 820 fid.n = n; 821 if (!qemu_opts_foreach(qemu_find_opts("device"), 822 failover_set_primary, &fid, &err)) { 823 return NULL; 824 } 825 return fid.id; 826 } 827 828 /** 829 * Find the primary device for this failover virtio-net 830 * 831 * @n: VirtIONet device 832 * @errp: returns an error if this function fails 833 */ 834 static DeviceState *failover_find_primary_device(VirtIONet *n) 835 { 836 char *id = failover_find_primary_device_id(n); 837 838 if (!id) { 839 return NULL; 840 } 841 842 return qdev_find_recursive(sysbus_get_default(), id); 843 } 844 845 static void failover_add_primary(VirtIONet *n, Error **errp) 846 { 847 Error *err = NULL; 848 QemuOpts *opts; 849 char *id; 850 DeviceState *dev = failover_find_primary_device(n); 851 852 if (dev) { 853 return; 854 } 855 856 id = failover_find_primary_device_id(n); 857 if (!id) { 858 return; 859 } 860 opts = qemu_opts_find(qemu_find_opts("device"), id); 861 if (opts) { 862 dev = qdev_device_add(opts, &err); 863 if (err) { 864 qemu_opts_del(opts); 865 } 866 } else { 867 error_setg(errp, "Primary device not found"); 868 error_append_hint(errp, "Virtio-net failover will not work. Make " 869 "sure primary device has parameter" 870 " failover_pair_id=<virtio-net-id>\n"); 871 } 872 error_propagate(errp, err); 873 } 874 875 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features) 876 { 877 VirtIONet *n = VIRTIO_NET(vdev); 878 Error *err = NULL; 879 int i; 880 881 if (n->mtu_bypass_backend && 882 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) { 883 features &= ~(1ULL << VIRTIO_NET_F_MTU); 884 } 885 886 virtio_net_set_multiqueue(n, 887 virtio_has_feature(features, VIRTIO_NET_F_RSS) || 888 virtio_has_feature(features, VIRTIO_NET_F_MQ)); 889 890 virtio_net_set_mrg_rx_bufs(n, 891 virtio_has_feature(features, 892 VIRTIO_NET_F_MRG_RXBUF), 893 virtio_has_feature(features, 894 VIRTIO_F_VERSION_1), 895 virtio_has_feature(features, 896 VIRTIO_NET_F_HASH_REPORT)); 897 898 n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) && 899 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4); 900 n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) && 901 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6); 902 n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS); 903 904 if (n->has_vnet_hdr) { 905 n->curr_guest_offloads = 906 virtio_net_guest_offloads_by_features(features); 907 virtio_net_apply_guest_offloads(n); 908 } 909 910 for (i = 0; i < n->max_queues; i++) { 911 NetClientState *nc = qemu_get_subqueue(n->nic, i); 912 913 if (!get_vhost_net(nc->peer)) { 914 continue; 915 } 916 vhost_net_ack_features(get_vhost_net(nc->peer), features); 917 } 918 919 if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) { 920 memset(n->vlans, 0, MAX_VLAN >> 3); 921 } else { 922 memset(n->vlans, 0xff, MAX_VLAN >> 3); 923 } 924 925 if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) { 926 qapi_event_send_failover_negotiated(n->netclient_name); 927 qatomic_set(&n->failover_primary_hidden, false); 928 failover_add_primary(n, &err); 929 if (err) { 930 warn_report_err(err); 931 } 932 } 933 } 934 935 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 936 struct iovec *iov, unsigned int iov_cnt) 937 { 938 uint8_t on; 939 size_t s; 940 NetClientState *nc = qemu_get_queue(n->nic); 941 942 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 943 if (s != sizeof(on)) { 944 return VIRTIO_NET_ERR; 945 } 946 947 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 948 n->promisc = on; 949 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 950 n->allmulti = on; 951 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 952 n->alluni = on; 953 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 954 n->nomulti = on; 955 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 956 n->nouni = on; 957 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 958 n->nobcast = on; 959 } else { 960 return VIRTIO_NET_ERR; 961 } 962 963 rxfilter_notify(nc); 964 965 return VIRTIO_NET_OK; 966 } 967 968 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 969 struct iovec *iov, unsigned int iov_cnt) 970 { 971 VirtIODevice *vdev = VIRTIO_DEVICE(n); 972 uint64_t offloads; 973 size_t s; 974 975 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 976 return VIRTIO_NET_ERR; 977 } 978 979 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 980 if (s != sizeof(offloads)) { 981 return VIRTIO_NET_ERR; 982 } 983 984 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 985 uint64_t supported_offloads; 986 987 offloads = virtio_ldq_p(vdev, &offloads); 988 989 if (!n->has_vnet_hdr) { 990 return VIRTIO_NET_ERR; 991 } 992 993 n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) && 994 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4); 995 n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) && 996 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6); 997 virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT); 998 999 supported_offloads = virtio_net_supported_guest_offloads(n); 1000 if (offloads & ~supported_offloads) { 1001 return VIRTIO_NET_ERR; 1002 } 1003 1004 n->curr_guest_offloads = offloads; 1005 virtio_net_apply_guest_offloads(n); 1006 1007 return VIRTIO_NET_OK; 1008 } else { 1009 return VIRTIO_NET_ERR; 1010 } 1011 } 1012 1013 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 1014 struct iovec *iov, unsigned int iov_cnt) 1015 { 1016 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1017 struct virtio_net_ctrl_mac mac_data; 1018 size_t s; 1019 NetClientState *nc = qemu_get_queue(n->nic); 1020 1021 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 1022 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 1023 return VIRTIO_NET_ERR; 1024 } 1025 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 1026 assert(s == sizeof(n->mac)); 1027 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 1028 rxfilter_notify(nc); 1029 1030 return VIRTIO_NET_OK; 1031 } 1032 1033 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 1034 return VIRTIO_NET_ERR; 1035 } 1036 1037 int in_use = 0; 1038 int first_multi = 0; 1039 uint8_t uni_overflow = 0; 1040 uint8_t multi_overflow = 0; 1041 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 1042 1043 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 1044 sizeof(mac_data.entries)); 1045 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 1046 if (s != sizeof(mac_data.entries)) { 1047 goto error; 1048 } 1049 iov_discard_front(&iov, &iov_cnt, s); 1050 1051 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 1052 goto error; 1053 } 1054 1055 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 1056 s = iov_to_buf(iov, iov_cnt, 0, macs, 1057 mac_data.entries * ETH_ALEN); 1058 if (s != mac_data.entries * ETH_ALEN) { 1059 goto error; 1060 } 1061 in_use += mac_data.entries; 1062 } else { 1063 uni_overflow = 1; 1064 } 1065 1066 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 1067 1068 first_multi = in_use; 1069 1070 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 1071 sizeof(mac_data.entries)); 1072 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 1073 if (s != sizeof(mac_data.entries)) { 1074 goto error; 1075 } 1076 1077 iov_discard_front(&iov, &iov_cnt, s); 1078 1079 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 1080 goto error; 1081 } 1082 1083 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) { 1084 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN], 1085 mac_data.entries * ETH_ALEN); 1086 if (s != mac_data.entries * ETH_ALEN) { 1087 goto error; 1088 } 1089 in_use += mac_data.entries; 1090 } else { 1091 multi_overflow = 1; 1092 } 1093 1094 n->mac_table.in_use = in_use; 1095 n->mac_table.first_multi = first_multi; 1096 n->mac_table.uni_overflow = uni_overflow; 1097 n->mac_table.multi_overflow = multi_overflow; 1098 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN); 1099 g_free(macs); 1100 rxfilter_notify(nc); 1101 1102 return VIRTIO_NET_OK; 1103 1104 error: 1105 g_free(macs); 1106 return VIRTIO_NET_ERR; 1107 } 1108 1109 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 1110 struct iovec *iov, unsigned int iov_cnt) 1111 { 1112 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1113 uint16_t vid; 1114 size_t s; 1115 NetClientState *nc = qemu_get_queue(n->nic); 1116 1117 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 1118 vid = virtio_lduw_p(vdev, &vid); 1119 if (s != sizeof(vid)) { 1120 return VIRTIO_NET_ERR; 1121 } 1122 1123 if (vid >= MAX_VLAN) 1124 return VIRTIO_NET_ERR; 1125 1126 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 1127 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 1128 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 1129 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 1130 else 1131 return VIRTIO_NET_ERR; 1132 1133 rxfilter_notify(nc); 1134 1135 return VIRTIO_NET_OK; 1136 } 1137 1138 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd, 1139 struct iovec *iov, unsigned int iov_cnt) 1140 { 1141 trace_virtio_net_handle_announce(n->announce_timer.round); 1142 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK && 1143 n->status & VIRTIO_NET_S_ANNOUNCE) { 1144 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 1145 if (n->announce_timer.round) { 1146 qemu_announce_timer_step(&n->announce_timer); 1147 } 1148 return VIRTIO_NET_OK; 1149 } else { 1150 return VIRTIO_NET_ERR; 1151 } 1152 } 1153 1154 static void virtio_net_disable_rss(VirtIONet *n) 1155 { 1156 if (n->rss_data.enabled) { 1157 trace_virtio_net_rss_disable(); 1158 } 1159 n->rss_data.enabled = false; 1160 } 1161 1162 static uint16_t virtio_net_handle_rss(VirtIONet *n, 1163 struct iovec *iov, 1164 unsigned int iov_cnt, 1165 bool do_rss) 1166 { 1167 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1168 struct virtio_net_rss_config cfg; 1169 size_t s, offset = 0, size_get; 1170 uint16_t queues, i; 1171 struct { 1172 uint16_t us; 1173 uint8_t b; 1174 } QEMU_PACKED temp; 1175 const char *err_msg = ""; 1176 uint32_t err_value = 0; 1177 1178 if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) { 1179 err_msg = "RSS is not negotiated"; 1180 goto error; 1181 } 1182 if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) { 1183 err_msg = "Hash report is not negotiated"; 1184 goto error; 1185 } 1186 size_get = offsetof(struct virtio_net_rss_config, indirection_table); 1187 s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get); 1188 if (s != size_get) { 1189 err_msg = "Short command buffer"; 1190 err_value = (uint32_t)s; 1191 goto error; 1192 } 1193 n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types); 1194 n->rss_data.indirections_len = 1195 virtio_lduw_p(vdev, &cfg.indirection_table_mask); 1196 n->rss_data.indirections_len++; 1197 if (!do_rss) { 1198 n->rss_data.indirections_len = 1; 1199 } 1200 if (!is_power_of_2(n->rss_data.indirections_len)) { 1201 err_msg = "Invalid size of indirection table"; 1202 err_value = n->rss_data.indirections_len; 1203 goto error; 1204 } 1205 if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) { 1206 err_msg = "Too large indirection table"; 1207 err_value = n->rss_data.indirections_len; 1208 goto error; 1209 } 1210 n->rss_data.default_queue = do_rss ? 1211 virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0; 1212 if (n->rss_data.default_queue >= n->max_queues) { 1213 err_msg = "Invalid default queue"; 1214 err_value = n->rss_data.default_queue; 1215 goto error; 1216 } 1217 offset += size_get; 1218 size_get = sizeof(uint16_t) * n->rss_data.indirections_len; 1219 g_free(n->rss_data.indirections_table); 1220 n->rss_data.indirections_table = g_malloc(size_get); 1221 if (!n->rss_data.indirections_table) { 1222 err_msg = "Can't allocate indirections table"; 1223 err_value = n->rss_data.indirections_len; 1224 goto error; 1225 } 1226 s = iov_to_buf(iov, iov_cnt, offset, 1227 n->rss_data.indirections_table, size_get); 1228 if (s != size_get) { 1229 err_msg = "Short indirection table buffer"; 1230 err_value = (uint32_t)s; 1231 goto error; 1232 } 1233 for (i = 0; i < n->rss_data.indirections_len; ++i) { 1234 uint16_t val = n->rss_data.indirections_table[i]; 1235 n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val); 1236 } 1237 offset += size_get; 1238 size_get = sizeof(temp); 1239 s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get); 1240 if (s != size_get) { 1241 err_msg = "Can't get queues"; 1242 err_value = (uint32_t)s; 1243 goto error; 1244 } 1245 queues = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queues; 1246 if (queues == 0 || queues > n->max_queues) { 1247 err_msg = "Invalid number of queues"; 1248 err_value = queues; 1249 goto error; 1250 } 1251 if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) { 1252 err_msg = "Invalid key size"; 1253 err_value = temp.b; 1254 goto error; 1255 } 1256 if (!temp.b && n->rss_data.hash_types) { 1257 err_msg = "No key provided"; 1258 err_value = 0; 1259 goto error; 1260 } 1261 if (!temp.b && !n->rss_data.hash_types) { 1262 virtio_net_disable_rss(n); 1263 return queues; 1264 } 1265 offset += size_get; 1266 size_get = temp.b; 1267 s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get); 1268 if (s != size_get) { 1269 err_msg = "Can get key buffer"; 1270 err_value = (uint32_t)s; 1271 goto error; 1272 } 1273 n->rss_data.enabled = true; 1274 trace_virtio_net_rss_enable(n->rss_data.hash_types, 1275 n->rss_data.indirections_len, 1276 temp.b); 1277 return queues; 1278 error: 1279 trace_virtio_net_rss_error(err_msg, err_value); 1280 virtio_net_disable_rss(n); 1281 return 0; 1282 } 1283 1284 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 1285 struct iovec *iov, unsigned int iov_cnt) 1286 { 1287 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1288 uint16_t queues; 1289 1290 virtio_net_disable_rss(n); 1291 if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) { 1292 queues = virtio_net_handle_rss(n, iov, iov_cnt, false); 1293 return queues ? VIRTIO_NET_OK : VIRTIO_NET_ERR; 1294 } 1295 if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) { 1296 queues = virtio_net_handle_rss(n, iov, iov_cnt, true); 1297 } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 1298 struct virtio_net_ctrl_mq mq; 1299 size_t s; 1300 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) { 1301 return VIRTIO_NET_ERR; 1302 } 1303 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 1304 if (s != sizeof(mq)) { 1305 return VIRTIO_NET_ERR; 1306 } 1307 queues = virtio_lduw_p(vdev, &mq.virtqueue_pairs); 1308 1309 } else { 1310 return VIRTIO_NET_ERR; 1311 } 1312 1313 if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 1314 queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 1315 queues > n->max_queues || 1316 !n->multiqueue) { 1317 return VIRTIO_NET_ERR; 1318 } 1319 1320 n->curr_queues = queues; 1321 /* stop the backend before changing the number of queues to avoid handling a 1322 * disabled queue */ 1323 virtio_net_set_status(vdev, vdev->status); 1324 virtio_net_set_queues(n); 1325 1326 return VIRTIO_NET_OK; 1327 } 1328 1329 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 1330 { 1331 VirtIONet *n = VIRTIO_NET(vdev); 1332 struct virtio_net_ctrl_hdr ctrl; 1333 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 1334 VirtQueueElement *elem; 1335 size_t s; 1336 struct iovec *iov, *iov2; 1337 unsigned int iov_cnt; 1338 1339 for (;;) { 1340 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 1341 if (!elem) { 1342 break; 1343 } 1344 if (iov_size(elem->in_sg, elem->in_num) < sizeof(status) || 1345 iov_size(elem->out_sg, elem->out_num) < sizeof(ctrl)) { 1346 virtio_error(vdev, "virtio-net ctrl missing headers"); 1347 virtqueue_detach_element(vq, elem, 0); 1348 g_free(elem); 1349 break; 1350 } 1351 1352 iov_cnt = elem->out_num; 1353 iov2 = iov = g_memdup(elem->out_sg, sizeof(struct iovec) * elem->out_num); 1354 s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl)); 1355 iov_discard_front(&iov, &iov_cnt, sizeof(ctrl)); 1356 if (s != sizeof(ctrl)) { 1357 status = VIRTIO_NET_ERR; 1358 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 1359 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt); 1360 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 1361 status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt); 1362 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 1363 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt); 1364 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) { 1365 status = virtio_net_handle_announce(n, ctrl.cmd, iov, iov_cnt); 1366 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 1367 status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt); 1368 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 1369 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt); 1370 } 1371 1372 s = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, sizeof(status)); 1373 assert(s == sizeof(status)); 1374 1375 virtqueue_push(vq, elem, sizeof(status)); 1376 virtio_notify(vdev, vq); 1377 g_free(iov2); 1378 g_free(elem); 1379 } 1380 } 1381 1382 /* RX */ 1383 1384 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 1385 { 1386 VirtIONet *n = VIRTIO_NET(vdev); 1387 int queue_index = vq2q(virtio_get_queue_index(vq)); 1388 1389 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 1390 } 1391 1392 static bool virtio_net_can_receive(NetClientState *nc) 1393 { 1394 VirtIONet *n = qemu_get_nic_opaque(nc); 1395 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1396 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1397 1398 if (!vdev->vm_running) { 1399 return false; 1400 } 1401 1402 if (nc->queue_index >= n->curr_queues) { 1403 return false; 1404 } 1405 1406 if (!virtio_queue_ready(q->rx_vq) || 1407 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1408 return false; 1409 } 1410 1411 return true; 1412 } 1413 1414 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 1415 { 1416 VirtIONet *n = q->n; 1417 if (virtio_queue_empty(q->rx_vq) || 1418 (n->mergeable_rx_bufs && 1419 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 1420 virtio_queue_set_notification(q->rx_vq, 1); 1421 1422 /* To avoid a race condition where the guest has made some buffers 1423 * available after the above check but before notification was 1424 * enabled, check for available buffers again. 1425 */ 1426 if (virtio_queue_empty(q->rx_vq) || 1427 (n->mergeable_rx_bufs && 1428 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 1429 return 0; 1430 } 1431 } 1432 1433 virtio_queue_set_notification(q->rx_vq, 0); 1434 return 1; 1435 } 1436 1437 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr) 1438 { 1439 virtio_tswap16s(vdev, &hdr->hdr_len); 1440 virtio_tswap16s(vdev, &hdr->gso_size); 1441 virtio_tswap16s(vdev, &hdr->csum_start); 1442 virtio_tswap16s(vdev, &hdr->csum_offset); 1443 } 1444 1445 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 1446 * it never finds out that the packets don't have valid checksums. This 1447 * causes dhclient to get upset. Fedora's carried a patch for ages to 1448 * fix this with Xen but it hasn't appeared in an upstream release of 1449 * dhclient yet. 1450 * 1451 * To avoid breaking existing guests, we catch udp packets and add 1452 * checksums. This is terrible but it's better than hacking the guest 1453 * kernels. 1454 * 1455 * N.B. if we introduce a zero-copy API, this operation is no longer free so 1456 * we should provide a mechanism to disable it to avoid polluting the host 1457 * cache. 1458 */ 1459 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 1460 uint8_t *buf, size_t size) 1461 { 1462 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 1463 (size > 27 && size < 1500) && /* normal sized MTU */ 1464 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 1465 (buf[23] == 17) && /* ip.protocol == UDP */ 1466 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 1467 net_checksum_calculate(buf, size); 1468 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 1469 } 1470 } 1471 1472 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt, 1473 const void *buf, size_t size) 1474 { 1475 if (n->has_vnet_hdr) { 1476 /* FIXME this cast is evil */ 1477 void *wbuf = (void *)buf; 1478 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, 1479 size - n->host_hdr_len); 1480 1481 if (n->needs_vnet_hdr_swap) { 1482 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf); 1483 } 1484 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr)); 1485 } else { 1486 struct virtio_net_hdr hdr = { 1487 .flags = 0, 1488 .gso_type = VIRTIO_NET_HDR_GSO_NONE 1489 }; 1490 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr); 1491 } 1492 } 1493 1494 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 1495 { 1496 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 1497 static const uint8_t vlan[] = {0x81, 0x00}; 1498 uint8_t *ptr = (uint8_t *)buf; 1499 int i; 1500 1501 if (n->promisc) 1502 return 1; 1503 1504 ptr += n->host_hdr_len; 1505 1506 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 1507 int vid = lduw_be_p(ptr + 14) & 0xfff; 1508 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 1509 return 0; 1510 } 1511 1512 if (ptr[0] & 1) { // multicast 1513 if (!memcmp(ptr, bcast, sizeof(bcast))) { 1514 return !n->nobcast; 1515 } else if (n->nomulti) { 1516 return 0; 1517 } else if (n->allmulti || n->mac_table.multi_overflow) { 1518 return 1; 1519 } 1520 1521 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 1522 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1523 return 1; 1524 } 1525 } 1526 } else { // unicast 1527 if (n->nouni) { 1528 return 0; 1529 } else if (n->alluni || n->mac_table.uni_overflow) { 1530 return 1; 1531 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 1532 return 1; 1533 } 1534 1535 for (i = 0; i < n->mac_table.first_multi; i++) { 1536 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1537 return 1; 1538 } 1539 } 1540 } 1541 1542 return 0; 1543 } 1544 1545 static uint8_t virtio_net_get_hash_type(bool isip4, 1546 bool isip6, 1547 bool isudp, 1548 bool istcp, 1549 uint32_t types) 1550 { 1551 if (isip4) { 1552 if (istcp && (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4)) { 1553 return NetPktRssIpV4Tcp; 1554 } 1555 if (isudp && (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4)) { 1556 return NetPktRssIpV4Udp; 1557 } 1558 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) { 1559 return NetPktRssIpV4; 1560 } 1561 } else if (isip6) { 1562 uint32_t mask = VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | 1563 VIRTIO_NET_RSS_HASH_TYPE_TCPv6; 1564 1565 if (istcp && (types & mask)) { 1566 return (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) ? 1567 NetPktRssIpV6TcpEx : NetPktRssIpV6Tcp; 1568 } 1569 mask = VIRTIO_NET_RSS_HASH_TYPE_UDP_EX | VIRTIO_NET_RSS_HASH_TYPE_UDPv6; 1570 if (isudp && (types & mask)) { 1571 return (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) ? 1572 NetPktRssIpV6UdpEx : NetPktRssIpV6Udp; 1573 } 1574 mask = VIRTIO_NET_RSS_HASH_TYPE_IP_EX | VIRTIO_NET_RSS_HASH_TYPE_IPv6; 1575 if (types & mask) { 1576 return (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) ? 1577 NetPktRssIpV6Ex : NetPktRssIpV6; 1578 } 1579 } 1580 return 0xff; 1581 } 1582 1583 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report, 1584 uint32_t hash) 1585 { 1586 struct virtio_net_hdr_v1_hash *hdr = (void *)buf; 1587 hdr->hash_value = hash; 1588 hdr->hash_report = report; 1589 } 1590 1591 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf, 1592 size_t size) 1593 { 1594 VirtIONet *n = qemu_get_nic_opaque(nc); 1595 unsigned int index = nc->queue_index, new_index = index; 1596 struct NetRxPkt *pkt = n->rx_pkt; 1597 uint8_t net_hash_type; 1598 uint32_t hash; 1599 bool isip4, isip6, isudp, istcp; 1600 static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = { 1601 VIRTIO_NET_HASH_REPORT_IPv4, 1602 VIRTIO_NET_HASH_REPORT_TCPv4, 1603 VIRTIO_NET_HASH_REPORT_TCPv6, 1604 VIRTIO_NET_HASH_REPORT_IPv6, 1605 VIRTIO_NET_HASH_REPORT_IPv6_EX, 1606 VIRTIO_NET_HASH_REPORT_TCPv6_EX, 1607 VIRTIO_NET_HASH_REPORT_UDPv4, 1608 VIRTIO_NET_HASH_REPORT_UDPv6, 1609 VIRTIO_NET_HASH_REPORT_UDPv6_EX 1610 }; 1611 1612 net_rx_pkt_set_protocols(pkt, buf + n->host_hdr_len, 1613 size - n->host_hdr_len); 1614 net_rx_pkt_get_protocols(pkt, &isip4, &isip6, &isudp, &istcp); 1615 if (isip4 && (net_rx_pkt_get_ip4_info(pkt)->fragment)) { 1616 istcp = isudp = false; 1617 } 1618 if (isip6 && (net_rx_pkt_get_ip6_info(pkt)->fragment)) { 1619 istcp = isudp = false; 1620 } 1621 net_hash_type = virtio_net_get_hash_type(isip4, isip6, isudp, istcp, 1622 n->rss_data.hash_types); 1623 if (net_hash_type > NetPktRssIpV6UdpEx) { 1624 if (n->rss_data.populate_hash) { 1625 virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0); 1626 } 1627 return n->rss_data.redirect ? n->rss_data.default_queue : -1; 1628 } 1629 1630 hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key); 1631 1632 if (n->rss_data.populate_hash) { 1633 virtio_set_packet_hash(buf, reports[net_hash_type], hash); 1634 } 1635 1636 if (n->rss_data.redirect) { 1637 new_index = hash & (n->rss_data.indirections_len - 1); 1638 new_index = n->rss_data.indirections_table[new_index]; 1639 } 1640 1641 return (index == new_index) ? -1 : new_index; 1642 } 1643 1644 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, 1645 size_t size, bool no_rss) 1646 { 1647 VirtIONet *n = qemu_get_nic_opaque(nc); 1648 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1649 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1650 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 1651 struct virtio_net_hdr_mrg_rxbuf mhdr; 1652 unsigned mhdr_cnt = 0; 1653 size_t offset, i, guest_offset; 1654 1655 if (!virtio_net_can_receive(nc)) { 1656 return -1; 1657 } 1658 1659 if (!no_rss && n->rss_data.enabled) { 1660 int index = virtio_net_process_rss(nc, buf, size); 1661 if (index >= 0) { 1662 NetClientState *nc2 = qemu_get_subqueue(n->nic, index); 1663 return virtio_net_receive_rcu(nc2, buf, size, true); 1664 } 1665 } 1666 1667 /* hdr_len refers to the header we supply to the guest */ 1668 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 1669 return 0; 1670 } 1671 1672 if (!receive_filter(n, buf, size)) 1673 return size; 1674 1675 offset = i = 0; 1676 1677 while (offset < size) { 1678 VirtQueueElement *elem; 1679 int len, total; 1680 const struct iovec *sg; 1681 1682 total = 0; 1683 1684 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement)); 1685 if (!elem) { 1686 if (i) { 1687 virtio_error(vdev, "virtio-net unexpected empty queue: " 1688 "i %zd mergeable %d offset %zd, size %zd, " 1689 "guest hdr len %zd, host hdr len %zd " 1690 "guest features 0x%" PRIx64, 1691 i, n->mergeable_rx_bufs, offset, size, 1692 n->guest_hdr_len, n->host_hdr_len, 1693 vdev->guest_features); 1694 } 1695 return -1; 1696 } 1697 1698 if (elem->in_num < 1) { 1699 virtio_error(vdev, 1700 "virtio-net receive queue contains no in buffers"); 1701 virtqueue_detach_element(q->rx_vq, elem, 0); 1702 g_free(elem); 1703 return -1; 1704 } 1705 1706 sg = elem->in_sg; 1707 if (i == 0) { 1708 assert(offset == 0); 1709 if (n->mergeable_rx_bufs) { 1710 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 1711 sg, elem->in_num, 1712 offsetof(typeof(mhdr), num_buffers), 1713 sizeof(mhdr.num_buffers)); 1714 } 1715 1716 receive_header(n, sg, elem->in_num, buf, size); 1717 if (n->rss_data.populate_hash) { 1718 offset = sizeof(mhdr); 1719 iov_from_buf(sg, elem->in_num, offset, 1720 buf + offset, n->host_hdr_len - sizeof(mhdr)); 1721 } 1722 offset = n->host_hdr_len; 1723 total += n->guest_hdr_len; 1724 guest_offset = n->guest_hdr_len; 1725 } else { 1726 guest_offset = 0; 1727 } 1728 1729 /* copy in packet. ugh */ 1730 len = iov_from_buf(sg, elem->in_num, guest_offset, 1731 buf + offset, size - offset); 1732 total += len; 1733 offset += len; 1734 /* If buffers can't be merged, at this point we 1735 * must have consumed the complete packet. 1736 * Otherwise, drop it. */ 1737 if (!n->mergeable_rx_bufs && offset < size) { 1738 virtqueue_unpop(q->rx_vq, elem, total); 1739 g_free(elem); 1740 return size; 1741 } 1742 1743 /* signal other side */ 1744 virtqueue_fill(q->rx_vq, elem, total, i++); 1745 g_free(elem); 1746 } 1747 1748 if (mhdr_cnt) { 1749 virtio_stw_p(vdev, &mhdr.num_buffers, i); 1750 iov_from_buf(mhdr_sg, mhdr_cnt, 1751 0, 1752 &mhdr.num_buffers, sizeof mhdr.num_buffers); 1753 } 1754 1755 virtqueue_flush(q->rx_vq, i); 1756 virtio_notify(vdev, q->rx_vq); 1757 1758 return size; 1759 } 1760 1761 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf, 1762 size_t size) 1763 { 1764 RCU_READ_LOCK_GUARD(); 1765 1766 return virtio_net_receive_rcu(nc, buf, size, false); 1767 } 1768 1769 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain, 1770 const uint8_t *buf, 1771 VirtioNetRscUnit *unit) 1772 { 1773 uint16_t ip_hdrlen; 1774 struct ip_header *ip; 1775 1776 ip = (struct ip_header *)(buf + chain->n->guest_hdr_len 1777 + sizeof(struct eth_header)); 1778 unit->ip = (void *)ip; 1779 ip_hdrlen = (ip->ip_ver_len & 0xF) << 2; 1780 unit->ip_plen = &ip->ip_len; 1781 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen); 1782 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10; 1783 unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen; 1784 } 1785 1786 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain, 1787 const uint8_t *buf, 1788 VirtioNetRscUnit *unit) 1789 { 1790 struct ip6_header *ip6; 1791 1792 ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len 1793 + sizeof(struct eth_header)); 1794 unit->ip = ip6; 1795 unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen); 1796 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) 1797 + sizeof(struct ip6_header)); 1798 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10; 1799 1800 /* There is a difference between payload lenght in ipv4 and v6, 1801 ip header is excluded in ipv6 */ 1802 unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen; 1803 } 1804 1805 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain, 1806 VirtioNetRscSeg *seg) 1807 { 1808 int ret; 1809 struct virtio_net_hdr_v1 *h; 1810 1811 h = (struct virtio_net_hdr_v1 *)seg->buf; 1812 h->flags = 0; 1813 h->gso_type = VIRTIO_NET_HDR_GSO_NONE; 1814 1815 if (seg->is_coalesced) { 1816 h->rsc.segments = seg->packets; 1817 h->rsc.dup_acks = seg->dup_ack; 1818 h->flags = VIRTIO_NET_HDR_F_RSC_INFO; 1819 if (chain->proto == ETH_P_IP) { 1820 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 1821 } else { 1822 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 1823 } 1824 } 1825 1826 ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size); 1827 QTAILQ_REMOVE(&chain->buffers, seg, next); 1828 g_free(seg->buf); 1829 g_free(seg); 1830 1831 return ret; 1832 } 1833 1834 static void virtio_net_rsc_purge(void *opq) 1835 { 1836 VirtioNetRscSeg *seg, *rn; 1837 VirtioNetRscChain *chain = (VirtioNetRscChain *)opq; 1838 1839 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) { 1840 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 1841 chain->stat.purge_failed++; 1842 continue; 1843 } 1844 } 1845 1846 chain->stat.timer++; 1847 if (!QTAILQ_EMPTY(&chain->buffers)) { 1848 timer_mod(chain->drain_timer, 1849 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout); 1850 } 1851 } 1852 1853 static void virtio_net_rsc_cleanup(VirtIONet *n) 1854 { 1855 VirtioNetRscChain *chain, *rn_chain; 1856 VirtioNetRscSeg *seg, *rn_seg; 1857 1858 QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) { 1859 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) { 1860 QTAILQ_REMOVE(&chain->buffers, seg, next); 1861 g_free(seg->buf); 1862 g_free(seg); 1863 } 1864 1865 timer_del(chain->drain_timer); 1866 timer_free(chain->drain_timer); 1867 QTAILQ_REMOVE(&n->rsc_chains, chain, next); 1868 g_free(chain); 1869 } 1870 } 1871 1872 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain, 1873 NetClientState *nc, 1874 const uint8_t *buf, size_t size) 1875 { 1876 uint16_t hdr_len; 1877 VirtioNetRscSeg *seg; 1878 1879 hdr_len = chain->n->guest_hdr_len; 1880 seg = g_malloc(sizeof(VirtioNetRscSeg)); 1881 seg->buf = g_malloc(hdr_len + sizeof(struct eth_header) 1882 + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD); 1883 memcpy(seg->buf, buf, size); 1884 seg->size = size; 1885 seg->packets = 1; 1886 seg->dup_ack = 0; 1887 seg->is_coalesced = 0; 1888 seg->nc = nc; 1889 1890 QTAILQ_INSERT_TAIL(&chain->buffers, seg, next); 1891 chain->stat.cache++; 1892 1893 switch (chain->proto) { 1894 case ETH_P_IP: 1895 virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit); 1896 break; 1897 case ETH_P_IPV6: 1898 virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit); 1899 break; 1900 default: 1901 g_assert_not_reached(); 1902 } 1903 } 1904 1905 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain, 1906 VirtioNetRscSeg *seg, 1907 const uint8_t *buf, 1908 struct tcp_header *n_tcp, 1909 struct tcp_header *o_tcp) 1910 { 1911 uint32_t nack, oack; 1912 uint16_t nwin, owin; 1913 1914 nack = htonl(n_tcp->th_ack); 1915 nwin = htons(n_tcp->th_win); 1916 oack = htonl(o_tcp->th_ack); 1917 owin = htons(o_tcp->th_win); 1918 1919 if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) { 1920 chain->stat.ack_out_of_win++; 1921 return RSC_FINAL; 1922 } else if (nack == oack) { 1923 /* duplicated ack or window probe */ 1924 if (nwin == owin) { 1925 /* duplicated ack, add dup ack count due to whql test up to 1 */ 1926 chain->stat.dup_ack++; 1927 return RSC_FINAL; 1928 } else { 1929 /* Coalesce window update */ 1930 o_tcp->th_win = n_tcp->th_win; 1931 chain->stat.win_update++; 1932 return RSC_COALESCE; 1933 } 1934 } else { 1935 /* pure ack, go to 'C', finalize*/ 1936 chain->stat.pure_ack++; 1937 return RSC_FINAL; 1938 } 1939 } 1940 1941 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain, 1942 VirtioNetRscSeg *seg, 1943 const uint8_t *buf, 1944 VirtioNetRscUnit *n_unit) 1945 { 1946 void *data; 1947 uint16_t o_ip_len; 1948 uint32_t nseq, oseq; 1949 VirtioNetRscUnit *o_unit; 1950 1951 o_unit = &seg->unit; 1952 o_ip_len = htons(*o_unit->ip_plen); 1953 nseq = htonl(n_unit->tcp->th_seq); 1954 oseq = htonl(o_unit->tcp->th_seq); 1955 1956 /* out of order or retransmitted. */ 1957 if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) { 1958 chain->stat.data_out_of_win++; 1959 return RSC_FINAL; 1960 } 1961 1962 data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen; 1963 if (nseq == oseq) { 1964 if ((o_unit->payload == 0) && n_unit->payload) { 1965 /* From no payload to payload, normal case, not a dup ack or etc */ 1966 chain->stat.data_after_pure_ack++; 1967 goto coalesce; 1968 } else { 1969 return virtio_net_rsc_handle_ack(chain, seg, buf, 1970 n_unit->tcp, o_unit->tcp); 1971 } 1972 } else if ((nseq - oseq) != o_unit->payload) { 1973 /* Not a consistent packet, out of order */ 1974 chain->stat.data_out_of_order++; 1975 return RSC_FINAL; 1976 } else { 1977 coalesce: 1978 if ((o_ip_len + n_unit->payload) > chain->max_payload) { 1979 chain->stat.over_size++; 1980 return RSC_FINAL; 1981 } 1982 1983 /* Here comes the right data, the payload length in v4/v6 is different, 1984 so use the field value to update and record the new data len */ 1985 o_unit->payload += n_unit->payload; /* update new data len */ 1986 1987 /* update field in ip header */ 1988 *o_unit->ip_plen = htons(o_ip_len + n_unit->payload); 1989 1990 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced 1991 for windows guest, while this may change the behavior for linux 1992 guest (only if it uses RSC feature). */ 1993 o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags; 1994 1995 o_unit->tcp->th_ack = n_unit->tcp->th_ack; 1996 o_unit->tcp->th_win = n_unit->tcp->th_win; 1997 1998 memmove(seg->buf + seg->size, data, n_unit->payload); 1999 seg->size += n_unit->payload; 2000 seg->packets++; 2001 chain->stat.coalesced++; 2002 return RSC_COALESCE; 2003 } 2004 } 2005 2006 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain, 2007 VirtioNetRscSeg *seg, 2008 const uint8_t *buf, size_t size, 2009 VirtioNetRscUnit *unit) 2010 { 2011 struct ip_header *ip1, *ip2; 2012 2013 ip1 = (struct ip_header *)(unit->ip); 2014 ip2 = (struct ip_header *)(seg->unit.ip); 2015 if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst) 2016 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport) 2017 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) { 2018 chain->stat.no_match++; 2019 return RSC_NO_MATCH; 2020 } 2021 2022 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit); 2023 } 2024 2025 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain, 2026 VirtioNetRscSeg *seg, 2027 const uint8_t *buf, size_t size, 2028 VirtioNetRscUnit *unit) 2029 { 2030 struct ip6_header *ip1, *ip2; 2031 2032 ip1 = (struct ip6_header *)(unit->ip); 2033 ip2 = (struct ip6_header *)(seg->unit.ip); 2034 if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address)) 2035 || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address)) 2036 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport) 2037 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) { 2038 chain->stat.no_match++; 2039 return RSC_NO_MATCH; 2040 } 2041 2042 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit); 2043 } 2044 2045 /* Packets with 'SYN' should bypass, other flag should be sent after drain 2046 * to prevent out of order */ 2047 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain, 2048 struct tcp_header *tcp) 2049 { 2050 uint16_t tcp_hdr; 2051 uint16_t tcp_flag; 2052 2053 tcp_flag = htons(tcp->th_offset_flags); 2054 tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10; 2055 tcp_flag &= VIRTIO_NET_TCP_FLAG; 2056 if (tcp_flag & TH_SYN) { 2057 chain->stat.tcp_syn++; 2058 return RSC_BYPASS; 2059 } 2060 2061 if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) { 2062 chain->stat.tcp_ctrl_drain++; 2063 return RSC_FINAL; 2064 } 2065 2066 if (tcp_hdr > sizeof(struct tcp_header)) { 2067 chain->stat.tcp_all_opt++; 2068 return RSC_FINAL; 2069 } 2070 2071 return RSC_CANDIDATE; 2072 } 2073 2074 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain, 2075 NetClientState *nc, 2076 const uint8_t *buf, size_t size, 2077 VirtioNetRscUnit *unit) 2078 { 2079 int ret; 2080 VirtioNetRscSeg *seg, *nseg; 2081 2082 if (QTAILQ_EMPTY(&chain->buffers)) { 2083 chain->stat.empty_cache++; 2084 virtio_net_rsc_cache_buf(chain, nc, buf, size); 2085 timer_mod(chain->drain_timer, 2086 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout); 2087 return size; 2088 } 2089 2090 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) { 2091 if (chain->proto == ETH_P_IP) { 2092 ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit); 2093 } else { 2094 ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit); 2095 } 2096 2097 if (ret == RSC_FINAL) { 2098 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2099 /* Send failed */ 2100 chain->stat.final_failed++; 2101 return 0; 2102 } 2103 2104 /* Send current packet */ 2105 return virtio_net_do_receive(nc, buf, size); 2106 } else if (ret == RSC_NO_MATCH) { 2107 continue; 2108 } else { 2109 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */ 2110 seg->is_coalesced = 1; 2111 return size; 2112 } 2113 } 2114 2115 chain->stat.no_match_cache++; 2116 virtio_net_rsc_cache_buf(chain, nc, buf, size); 2117 return size; 2118 } 2119 2120 /* Drain a connection data, this is to avoid out of order segments */ 2121 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain, 2122 NetClientState *nc, 2123 const uint8_t *buf, size_t size, 2124 uint16_t ip_start, uint16_t ip_size, 2125 uint16_t tcp_port) 2126 { 2127 VirtioNetRscSeg *seg, *nseg; 2128 uint32_t ppair1, ppair2; 2129 2130 ppair1 = *(uint32_t *)(buf + tcp_port); 2131 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) { 2132 ppair2 = *(uint32_t *)(seg->buf + tcp_port); 2133 if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size) 2134 || (ppair1 != ppair2)) { 2135 continue; 2136 } 2137 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2138 chain->stat.drain_failed++; 2139 } 2140 2141 break; 2142 } 2143 2144 return virtio_net_do_receive(nc, buf, size); 2145 } 2146 2147 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain, 2148 struct ip_header *ip, 2149 const uint8_t *buf, size_t size) 2150 { 2151 uint16_t ip_len; 2152 2153 /* Not an ipv4 packet */ 2154 if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) { 2155 chain->stat.ip_option++; 2156 return RSC_BYPASS; 2157 } 2158 2159 /* Don't handle packets with ip option */ 2160 if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) { 2161 chain->stat.ip_option++; 2162 return RSC_BYPASS; 2163 } 2164 2165 if (ip->ip_p != IPPROTO_TCP) { 2166 chain->stat.bypass_not_tcp++; 2167 return RSC_BYPASS; 2168 } 2169 2170 /* Don't handle packets with ip fragment */ 2171 if (!(htons(ip->ip_off) & IP_DF)) { 2172 chain->stat.ip_frag++; 2173 return RSC_BYPASS; 2174 } 2175 2176 /* Don't handle packets with ecn flag */ 2177 if (IPTOS_ECN(ip->ip_tos)) { 2178 chain->stat.ip_ecn++; 2179 return RSC_BYPASS; 2180 } 2181 2182 ip_len = htons(ip->ip_len); 2183 if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header)) 2184 || ip_len > (size - chain->n->guest_hdr_len - 2185 sizeof(struct eth_header))) { 2186 chain->stat.ip_hacked++; 2187 return RSC_BYPASS; 2188 } 2189 2190 return RSC_CANDIDATE; 2191 } 2192 2193 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain, 2194 NetClientState *nc, 2195 const uint8_t *buf, size_t size) 2196 { 2197 int32_t ret; 2198 uint16_t hdr_len; 2199 VirtioNetRscUnit unit; 2200 2201 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len; 2202 2203 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header) 2204 + sizeof(struct tcp_header))) { 2205 chain->stat.bypass_not_tcp++; 2206 return virtio_net_do_receive(nc, buf, size); 2207 } 2208 2209 virtio_net_rsc_extract_unit4(chain, buf, &unit); 2210 if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size) 2211 != RSC_CANDIDATE) { 2212 return virtio_net_do_receive(nc, buf, size); 2213 } 2214 2215 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp); 2216 if (ret == RSC_BYPASS) { 2217 return virtio_net_do_receive(nc, buf, size); 2218 } else if (ret == RSC_FINAL) { 2219 return virtio_net_rsc_drain_flow(chain, nc, buf, size, 2220 ((hdr_len + sizeof(struct eth_header)) + 12), 2221 VIRTIO_NET_IP4_ADDR_SIZE, 2222 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)); 2223 } 2224 2225 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit); 2226 } 2227 2228 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain, 2229 struct ip6_header *ip6, 2230 const uint8_t *buf, size_t size) 2231 { 2232 uint16_t ip_len; 2233 2234 if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4) 2235 != IP_HEADER_VERSION_6) { 2236 return RSC_BYPASS; 2237 } 2238 2239 /* Both option and protocol is checked in this */ 2240 if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) { 2241 chain->stat.bypass_not_tcp++; 2242 return RSC_BYPASS; 2243 } 2244 2245 ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen); 2246 if (ip_len < sizeof(struct tcp_header) || 2247 ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header) 2248 - sizeof(struct ip6_header))) { 2249 chain->stat.ip_hacked++; 2250 return RSC_BYPASS; 2251 } 2252 2253 /* Don't handle packets with ecn flag */ 2254 if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) { 2255 chain->stat.ip_ecn++; 2256 return RSC_BYPASS; 2257 } 2258 2259 return RSC_CANDIDATE; 2260 } 2261 2262 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc, 2263 const uint8_t *buf, size_t size) 2264 { 2265 int32_t ret; 2266 uint16_t hdr_len; 2267 VirtioNetRscChain *chain; 2268 VirtioNetRscUnit unit; 2269 2270 chain = (VirtioNetRscChain *)opq; 2271 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len; 2272 2273 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header) 2274 + sizeof(tcp_header))) { 2275 return virtio_net_do_receive(nc, buf, size); 2276 } 2277 2278 virtio_net_rsc_extract_unit6(chain, buf, &unit); 2279 if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain, 2280 unit.ip, buf, size)) { 2281 return virtio_net_do_receive(nc, buf, size); 2282 } 2283 2284 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp); 2285 if (ret == RSC_BYPASS) { 2286 return virtio_net_do_receive(nc, buf, size); 2287 } else if (ret == RSC_FINAL) { 2288 return virtio_net_rsc_drain_flow(chain, nc, buf, size, 2289 ((hdr_len + sizeof(struct eth_header)) + 8), 2290 VIRTIO_NET_IP6_ADDR_SIZE, 2291 hdr_len + sizeof(struct eth_header) 2292 + sizeof(struct ip6_header)); 2293 } 2294 2295 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit); 2296 } 2297 2298 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n, 2299 NetClientState *nc, 2300 uint16_t proto) 2301 { 2302 VirtioNetRscChain *chain; 2303 2304 if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) { 2305 return NULL; 2306 } 2307 2308 QTAILQ_FOREACH(chain, &n->rsc_chains, next) { 2309 if (chain->proto == proto) { 2310 return chain; 2311 } 2312 } 2313 2314 chain = g_malloc(sizeof(*chain)); 2315 chain->n = n; 2316 chain->proto = proto; 2317 if (proto == (uint16_t)ETH_P_IP) { 2318 chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD; 2319 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2320 } else { 2321 chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD; 2322 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2323 } 2324 chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST, 2325 virtio_net_rsc_purge, chain); 2326 memset(&chain->stat, 0, sizeof(chain->stat)); 2327 2328 QTAILQ_INIT(&chain->buffers); 2329 QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next); 2330 2331 return chain; 2332 } 2333 2334 static ssize_t virtio_net_rsc_receive(NetClientState *nc, 2335 const uint8_t *buf, 2336 size_t size) 2337 { 2338 uint16_t proto; 2339 VirtioNetRscChain *chain; 2340 struct eth_header *eth; 2341 VirtIONet *n; 2342 2343 n = qemu_get_nic_opaque(nc); 2344 if (size < (n->host_hdr_len + sizeof(struct eth_header))) { 2345 return virtio_net_do_receive(nc, buf, size); 2346 } 2347 2348 eth = (struct eth_header *)(buf + n->guest_hdr_len); 2349 proto = htons(eth->h_proto); 2350 2351 chain = virtio_net_rsc_lookup_chain(n, nc, proto); 2352 if (chain) { 2353 chain->stat.received++; 2354 if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) { 2355 return virtio_net_rsc_receive4(chain, nc, buf, size); 2356 } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) { 2357 return virtio_net_rsc_receive6(chain, nc, buf, size); 2358 } 2359 } 2360 return virtio_net_do_receive(nc, buf, size); 2361 } 2362 2363 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, 2364 size_t size) 2365 { 2366 VirtIONet *n = qemu_get_nic_opaque(nc); 2367 if ((n->rsc4_enabled || n->rsc6_enabled)) { 2368 return virtio_net_rsc_receive(nc, buf, size); 2369 } else { 2370 return virtio_net_do_receive(nc, buf, size); 2371 } 2372 } 2373 2374 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 2375 2376 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 2377 { 2378 VirtIONet *n = qemu_get_nic_opaque(nc); 2379 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 2380 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2381 2382 virtqueue_push(q->tx_vq, q->async_tx.elem, 0); 2383 virtio_notify(vdev, q->tx_vq); 2384 2385 g_free(q->async_tx.elem); 2386 q->async_tx.elem = NULL; 2387 2388 virtio_queue_set_notification(q->tx_vq, 1); 2389 virtio_net_flush_tx(q); 2390 } 2391 2392 /* TX */ 2393 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 2394 { 2395 VirtIONet *n = q->n; 2396 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2397 VirtQueueElement *elem; 2398 int32_t num_packets = 0; 2399 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 2400 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 2401 return num_packets; 2402 } 2403 2404 if (q->async_tx.elem) { 2405 virtio_queue_set_notification(q->tx_vq, 0); 2406 return num_packets; 2407 } 2408 2409 for (;;) { 2410 ssize_t ret; 2411 unsigned int out_num; 2412 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg; 2413 struct virtio_net_hdr_mrg_rxbuf mhdr; 2414 2415 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement)); 2416 if (!elem) { 2417 break; 2418 } 2419 2420 out_num = elem->out_num; 2421 out_sg = elem->out_sg; 2422 if (out_num < 1) { 2423 virtio_error(vdev, "virtio-net header not in first element"); 2424 virtqueue_detach_element(q->tx_vq, elem, 0); 2425 g_free(elem); 2426 return -EINVAL; 2427 } 2428 2429 if (n->has_vnet_hdr) { 2430 if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) < 2431 n->guest_hdr_len) { 2432 virtio_error(vdev, "virtio-net header incorrect"); 2433 virtqueue_detach_element(q->tx_vq, elem, 0); 2434 g_free(elem); 2435 return -EINVAL; 2436 } 2437 if (n->needs_vnet_hdr_swap) { 2438 virtio_net_hdr_swap(vdev, (void *) &mhdr); 2439 sg2[0].iov_base = &mhdr; 2440 sg2[0].iov_len = n->guest_hdr_len; 2441 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1, 2442 out_sg, out_num, 2443 n->guest_hdr_len, -1); 2444 if (out_num == VIRTQUEUE_MAX_SIZE) { 2445 goto drop; 2446 } 2447 out_num += 1; 2448 out_sg = sg2; 2449 } 2450 } 2451 /* 2452 * If host wants to see the guest header as is, we can 2453 * pass it on unchanged. Otherwise, copy just the parts 2454 * that host is interested in. 2455 */ 2456 assert(n->host_hdr_len <= n->guest_hdr_len); 2457 if (n->host_hdr_len != n->guest_hdr_len) { 2458 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 2459 out_sg, out_num, 2460 0, n->host_hdr_len); 2461 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 2462 out_sg, out_num, 2463 n->guest_hdr_len, -1); 2464 out_num = sg_num; 2465 out_sg = sg; 2466 } 2467 2468 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 2469 out_sg, out_num, virtio_net_tx_complete); 2470 if (ret == 0) { 2471 virtio_queue_set_notification(q->tx_vq, 0); 2472 q->async_tx.elem = elem; 2473 return -EBUSY; 2474 } 2475 2476 drop: 2477 virtqueue_push(q->tx_vq, elem, 0); 2478 virtio_notify(vdev, q->tx_vq); 2479 g_free(elem); 2480 2481 if (++num_packets >= n->tx_burst) { 2482 break; 2483 } 2484 } 2485 return num_packets; 2486 } 2487 2488 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 2489 { 2490 VirtIONet *n = VIRTIO_NET(vdev); 2491 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 2492 2493 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 2494 virtio_net_drop_tx_queue_data(vdev, vq); 2495 return; 2496 } 2497 2498 /* This happens when device was stopped but VCPU wasn't. */ 2499 if (!vdev->vm_running) { 2500 q->tx_waiting = 1; 2501 return; 2502 } 2503 2504 if (q->tx_waiting) { 2505 virtio_queue_set_notification(vq, 1); 2506 timer_del(q->tx_timer); 2507 q->tx_waiting = 0; 2508 if (virtio_net_flush_tx(q) == -EINVAL) { 2509 return; 2510 } 2511 } else { 2512 timer_mod(q->tx_timer, 2513 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2514 q->tx_waiting = 1; 2515 virtio_queue_set_notification(vq, 0); 2516 } 2517 } 2518 2519 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 2520 { 2521 VirtIONet *n = VIRTIO_NET(vdev); 2522 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 2523 2524 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 2525 virtio_net_drop_tx_queue_data(vdev, vq); 2526 return; 2527 } 2528 2529 if (unlikely(q->tx_waiting)) { 2530 return; 2531 } 2532 q->tx_waiting = 1; 2533 /* This happens when device was stopped but VCPU wasn't. */ 2534 if (!vdev->vm_running) { 2535 return; 2536 } 2537 virtio_queue_set_notification(vq, 0); 2538 qemu_bh_schedule(q->tx_bh); 2539 } 2540 2541 static void virtio_net_tx_timer(void *opaque) 2542 { 2543 VirtIONetQueue *q = opaque; 2544 VirtIONet *n = q->n; 2545 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2546 /* This happens when device was stopped but BH wasn't. */ 2547 if (!vdev->vm_running) { 2548 /* Make sure tx waiting is set, so we'll run when restarted. */ 2549 assert(q->tx_waiting); 2550 return; 2551 } 2552 2553 q->tx_waiting = 0; 2554 2555 /* Just in case the driver is not ready on more */ 2556 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 2557 return; 2558 } 2559 2560 virtio_queue_set_notification(q->tx_vq, 1); 2561 virtio_net_flush_tx(q); 2562 } 2563 2564 static void virtio_net_tx_bh(void *opaque) 2565 { 2566 VirtIONetQueue *q = opaque; 2567 VirtIONet *n = q->n; 2568 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2569 int32_t ret; 2570 2571 /* This happens when device was stopped but BH wasn't. */ 2572 if (!vdev->vm_running) { 2573 /* Make sure tx waiting is set, so we'll run when restarted. */ 2574 assert(q->tx_waiting); 2575 return; 2576 } 2577 2578 q->tx_waiting = 0; 2579 2580 /* Just in case the driver is not ready on more */ 2581 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 2582 return; 2583 } 2584 2585 ret = virtio_net_flush_tx(q); 2586 if (ret == -EBUSY || ret == -EINVAL) { 2587 return; /* Notification re-enable handled by tx_complete or device 2588 * broken */ 2589 } 2590 2591 /* If we flush a full burst of packets, assume there are 2592 * more coming and immediately reschedule */ 2593 if (ret >= n->tx_burst) { 2594 qemu_bh_schedule(q->tx_bh); 2595 q->tx_waiting = 1; 2596 return; 2597 } 2598 2599 /* If less than a full burst, re-enable notification and flush 2600 * anything that may have come in while we weren't looking. If 2601 * we find something, assume the guest is still active and reschedule */ 2602 virtio_queue_set_notification(q->tx_vq, 1); 2603 ret = virtio_net_flush_tx(q); 2604 if (ret == -EINVAL) { 2605 return; 2606 } else if (ret > 0) { 2607 virtio_queue_set_notification(q->tx_vq, 0); 2608 qemu_bh_schedule(q->tx_bh); 2609 q->tx_waiting = 1; 2610 } 2611 } 2612 2613 static void virtio_net_add_queue(VirtIONet *n, int index) 2614 { 2615 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2616 2617 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size, 2618 virtio_net_handle_rx); 2619 2620 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 2621 n->vqs[index].tx_vq = 2622 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 2623 virtio_net_handle_tx_timer); 2624 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 2625 virtio_net_tx_timer, 2626 &n->vqs[index]); 2627 } else { 2628 n->vqs[index].tx_vq = 2629 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 2630 virtio_net_handle_tx_bh); 2631 n->vqs[index].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[index]); 2632 } 2633 2634 n->vqs[index].tx_waiting = 0; 2635 n->vqs[index].n = n; 2636 } 2637 2638 static void virtio_net_del_queue(VirtIONet *n, int index) 2639 { 2640 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2641 VirtIONetQueue *q = &n->vqs[index]; 2642 NetClientState *nc = qemu_get_subqueue(n->nic, index); 2643 2644 qemu_purge_queued_packets(nc); 2645 2646 virtio_del_queue(vdev, index * 2); 2647 if (q->tx_timer) { 2648 timer_del(q->tx_timer); 2649 timer_free(q->tx_timer); 2650 q->tx_timer = NULL; 2651 } else { 2652 qemu_bh_delete(q->tx_bh); 2653 q->tx_bh = NULL; 2654 } 2655 q->tx_waiting = 0; 2656 virtio_del_queue(vdev, index * 2 + 1); 2657 } 2658 2659 static void virtio_net_change_num_queues(VirtIONet *n, int new_max_queues) 2660 { 2661 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2662 int old_num_queues = virtio_get_num_queues(vdev); 2663 int new_num_queues = new_max_queues * 2 + 1; 2664 int i; 2665 2666 assert(old_num_queues >= 3); 2667 assert(old_num_queues % 2 == 1); 2668 2669 if (old_num_queues == new_num_queues) { 2670 return; 2671 } 2672 2673 /* 2674 * We always need to remove and add ctrl vq if 2675 * old_num_queues != new_num_queues. Remove ctrl_vq first, 2676 * and then we only enter one of the following two loops. 2677 */ 2678 virtio_del_queue(vdev, old_num_queues - 1); 2679 2680 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) { 2681 /* new_num_queues < old_num_queues */ 2682 virtio_net_del_queue(n, i / 2); 2683 } 2684 2685 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) { 2686 /* new_num_queues > old_num_queues */ 2687 virtio_net_add_queue(n, i / 2); 2688 } 2689 2690 /* add ctrl_vq last */ 2691 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 2692 } 2693 2694 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 2695 { 2696 int max = multiqueue ? n->max_queues : 1; 2697 2698 n->multiqueue = multiqueue; 2699 virtio_net_change_num_queues(n, max); 2700 2701 virtio_net_set_queues(n); 2702 } 2703 2704 static int virtio_net_post_load_device(void *opaque, int version_id) 2705 { 2706 VirtIONet *n = opaque; 2707 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2708 int i, link_down; 2709 2710 trace_virtio_net_post_load_device(); 2711 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs, 2712 virtio_vdev_has_feature(vdev, 2713 VIRTIO_F_VERSION_1), 2714 virtio_vdev_has_feature(vdev, 2715 VIRTIO_NET_F_HASH_REPORT)); 2716 2717 /* MAC_TABLE_ENTRIES may be different from the saved image */ 2718 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) { 2719 n->mac_table.in_use = 0; 2720 } 2721 2722 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 2723 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 2724 } 2725 2726 /* 2727 * curr_guest_offloads will be later overwritten by the 2728 * virtio_set_features_nocheck call done from the virtio_load. 2729 * Here we make sure it is preserved and restored accordingly 2730 * in the virtio_net_post_load_virtio callback. 2731 */ 2732 n->saved_guest_offloads = n->curr_guest_offloads; 2733 2734 virtio_net_set_queues(n); 2735 2736 /* Find the first multicast entry in the saved MAC filter */ 2737 for (i = 0; i < n->mac_table.in_use; i++) { 2738 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 2739 break; 2740 } 2741 } 2742 n->mac_table.first_multi = i; 2743 2744 /* nc.link_down can't be migrated, so infer link_down according 2745 * to link status bit in n->status */ 2746 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 2747 for (i = 0; i < n->max_queues; i++) { 2748 qemu_get_subqueue(n->nic, i)->link_down = link_down; 2749 } 2750 2751 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 2752 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 2753 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(), 2754 QEMU_CLOCK_VIRTUAL, 2755 virtio_net_announce_timer, n); 2756 if (n->announce_timer.round) { 2757 timer_mod(n->announce_timer.tm, 2758 qemu_clock_get_ms(n->announce_timer.type)); 2759 } else { 2760 qemu_announce_timer_del(&n->announce_timer, false); 2761 } 2762 } 2763 2764 if (n->rss_data.enabled) { 2765 trace_virtio_net_rss_enable(n->rss_data.hash_types, 2766 n->rss_data.indirections_len, 2767 sizeof(n->rss_data.key)); 2768 } else { 2769 trace_virtio_net_rss_disable(); 2770 } 2771 return 0; 2772 } 2773 2774 static int virtio_net_post_load_virtio(VirtIODevice *vdev) 2775 { 2776 VirtIONet *n = VIRTIO_NET(vdev); 2777 /* 2778 * The actual needed state is now in saved_guest_offloads, 2779 * see virtio_net_post_load_device for detail. 2780 * Restore it back and apply the desired offloads. 2781 */ 2782 n->curr_guest_offloads = n->saved_guest_offloads; 2783 if (peer_has_vnet_hdr(n)) { 2784 virtio_net_apply_guest_offloads(n); 2785 } 2786 2787 return 0; 2788 } 2789 2790 /* tx_waiting field of a VirtIONetQueue */ 2791 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = { 2792 .name = "virtio-net-queue-tx_waiting", 2793 .fields = (VMStateField[]) { 2794 VMSTATE_UINT32(tx_waiting, VirtIONetQueue), 2795 VMSTATE_END_OF_LIST() 2796 }, 2797 }; 2798 2799 static bool max_queues_gt_1(void *opaque, int version_id) 2800 { 2801 return VIRTIO_NET(opaque)->max_queues > 1; 2802 } 2803 2804 static bool has_ctrl_guest_offloads(void *opaque, int version_id) 2805 { 2806 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque), 2807 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS); 2808 } 2809 2810 static bool mac_table_fits(void *opaque, int version_id) 2811 { 2812 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES; 2813 } 2814 2815 static bool mac_table_doesnt_fit(void *opaque, int version_id) 2816 { 2817 return !mac_table_fits(opaque, version_id); 2818 } 2819 2820 /* This temporary type is shared by all the WITH_TMP methods 2821 * although only some fields are used by each. 2822 */ 2823 struct VirtIONetMigTmp { 2824 VirtIONet *parent; 2825 VirtIONetQueue *vqs_1; 2826 uint16_t curr_queues_1; 2827 uint8_t has_ufo; 2828 uint32_t has_vnet_hdr; 2829 }; 2830 2831 /* The 2nd and subsequent tx_waiting flags are loaded later than 2832 * the 1st entry in the queues and only if there's more than one 2833 * entry. We use the tmp mechanism to calculate a temporary 2834 * pointer and count and also validate the count. 2835 */ 2836 2837 static int virtio_net_tx_waiting_pre_save(void *opaque) 2838 { 2839 struct VirtIONetMigTmp *tmp = opaque; 2840 2841 tmp->vqs_1 = tmp->parent->vqs + 1; 2842 tmp->curr_queues_1 = tmp->parent->curr_queues - 1; 2843 if (tmp->parent->curr_queues == 0) { 2844 tmp->curr_queues_1 = 0; 2845 } 2846 2847 return 0; 2848 } 2849 2850 static int virtio_net_tx_waiting_pre_load(void *opaque) 2851 { 2852 struct VirtIONetMigTmp *tmp = opaque; 2853 2854 /* Reuse the pointer setup from save */ 2855 virtio_net_tx_waiting_pre_save(opaque); 2856 2857 if (tmp->parent->curr_queues > tmp->parent->max_queues) { 2858 error_report("virtio-net: curr_queues %x > max_queues %x", 2859 tmp->parent->curr_queues, tmp->parent->max_queues); 2860 2861 return -EINVAL; 2862 } 2863 2864 return 0; /* all good */ 2865 } 2866 2867 static const VMStateDescription vmstate_virtio_net_tx_waiting = { 2868 .name = "virtio-net-tx_waiting", 2869 .pre_load = virtio_net_tx_waiting_pre_load, 2870 .pre_save = virtio_net_tx_waiting_pre_save, 2871 .fields = (VMStateField[]) { 2872 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp, 2873 curr_queues_1, 2874 vmstate_virtio_net_queue_tx_waiting, 2875 struct VirtIONetQueue), 2876 VMSTATE_END_OF_LIST() 2877 }, 2878 }; 2879 2880 /* the 'has_ufo' flag is just tested; if the incoming stream has the 2881 * flag set we need to check that we have it 2882 */ 2883 static int virtio_net_ufo_post_load(void *opaque, int version_id) 2884 { 2885 struct VirtIONetMigTmp *tmp = opaque; 2886 2887 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) { 2888 error_report("virtio-net: saved image requires TUN_F_UFO support"); 2889 return -EINVAL; 2890 } 2891 2892 return 0; 2893 } 2894 2895 static int virtio_net_ufo_pre_save(void *opaque) 2896 { 2897 struct VirtIONetMigTmp *tmp = opaque; 2898 2899 tmp->has_ufo = tmp->parent->has_ufo; 2900 2901 return 0; 2902 } 2903 2904 static const VMStateDescription vmstate_virtio_net_has_ufo = { 2905 .name = "virtio-net-ufo", 2906 .post_load = virtio_net_ufo_post_load, 2907 .pre_save = virtio_net_ufo_pre_save, 2908 .fields = (VMStateField[]) { 2909 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp), 2910 VMSTATE_END_OF_LIST() 2911 }, 2912 }; 2913 2914 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the 2915 * flag set we need to check that we have it 2916 */ 2917 static int virtio_net_vnet_post_load(void *opaque, int version_id) 2918 { 2919 struct VirtIONetMigTmp *tmp = opaque; 2920 2921 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) { 2922 error_report("virtio-net: saved image requires vnet_hdr=on"); 2923 return -EINVAL; 2924 } 2925 2926 return 0; 2927 } 2928 2929 static int virtio_net_vnet_pre_save(void *opaque) 2930 { 2931 struct VirtIONetMigTmp *tmp = opaque; 2932 2933 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr; 2934 2935 return 0; 2936 } 2937 2938 static const VMStateDescription vmstate_virtio_net_has_vnet = { 2939 .name = "virtio-net-vnet", 2940 .post_load = virtio_net_vnet_post_load, 2941 .pre_save = virtio_net_vnet_pre_save, 2942 .fields = (VMStateField[]) { 2943 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp), 2944 VMSTATE_END_OF_LIST() 2945 }, 2946 }; 2947 2948 static bool virtio_net_rss_needed(void *opaque) 2949 { 2950 return VIRTIO_NET(opaque)->rss_data.enabled; 2951 } 2952 2953 static const VMStateDescription vmstate_virtio_net_rss = { 2954 .name = "virtio-net-device/rss", 2955 .version_id = 1, 2956 .minimum_version_id = 1, 2957 .needed = virtio_net_rss_needed, 2958 .fields = (VMStateField[]) { 2959 VMSTATE_BOOL(rss_data.enabled, VirtIONet), 2960 VMSTATE_BOOL(rss_data.redirect, VirtIONet), 2961 VMSTATE_BOOL(rss_data.populate_hash, VirtIONet), 2962 VMSTATE_UINT32(rss_data.hash_types, VirtIONet), 2963 VMSTATE_UINT16(rss_data.indirections_len, VirtIONet), 2964 VMSTATE_UINT16(rss_data.default_queue, VirtIONet), 2965 VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet, 2966 VIRTIO_NET_RSS_MAX_KEY_SIZE), 2967 VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet, 2968 rss_data.indirections_len, 0, 2969 vmstate_info_uint16, uint16_t), 2970 VMSTATE_END_OF_LIST() 2971 }, 2972 }; 2973 2974 static const VMStateDescription vmstate_virtio_net_device = { 2975 .name = "virtio-net-device", 2976 .version_id = VIRTIO_NET_VM_VERSION, 2977 .minimum_version_id = VIRTIO_NET_VM_VERSION, 2978 .post_load = virtio_net_post_load_device, 2979 .fields = (VMStateField[]) { 2980 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN), 2981 VMSTATE_STRUCT_POINTER(vqs, VirtIONet, 2982 vmstate_virtio_net_queue_tx_waiting, 2983 VirtIONetQueue), 2984 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet), 2985 VMSTATE_UINT16(status, VirtIONet), 2986 VMSTATE_UINT8(promisc, VirtIONet), 2987 VMSTATE_UINT8(allmulti, VirtIONet), 2988 VMSTATE_UINT32(mac_table.in_use, VirtIONet), 2989 2990 /* Guarded pair: If it fits we load it, else we throw it away 2991 * - can happen if source has a larger MAC table.; post-load 2992 * sets flags in this case. 2993 */ 2994 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet, 2995 0, mac_table_fits, mac_table.in_use, 2996 ETH_ALEN), 2997 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0, 2998 mac_table.in_use, ETH_ALEN), 2999 3000 /* Note: This is an array of uint32's that's always been saved as a 3001 * buffer; hold onto your endiannesses; it's actually used as a bitmap 3002 * but based on the uint. 3003 */ 3004 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3), 3005 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3006 vmstate_virtio_net_has_vnet), 3007 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet), 3008 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet), 3009 VMSTATE_UINT8(alluni, VirtIONet), 3010 VMSTATE_UINT8(nomulti, VirtIONet), 3011 VMSTATE_UINT8(nouni, VirtIONet), 3012 VMSTATE_UINT8(nobcast, VirtIONet), 3013 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3014 vmstate_virtio_net_has_ufo), 3015 VMSTATE_SINGLE_TEST(max_queues, VirtIONet, max_queues_gt_1, 0, 3016 vmstate_info_uint16_equal, uint16_t), 3017 VMSTATE_UINT16_TEST(curr_queues, VirtIONet, max_queues_gt_1), 3018 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3019 vmstate_virtio_net_tx_waiting), 3020 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet, 3021 has_ctrl_guest_offloads), 3022 VMSTATE_END_OF_LIST() 3023 }, 3024 .subsections = (const VMStateDescription * []) { 3025 &vmstate_virtio_net_rss, 3026 NULL 3027 } 3028 }; 3029 3030 static NetClientInfo net_virtio_info = { 3031 .type = NET_CLIENT_DRIVER_NIC, 3032 .size = sizeof(NICState), 3033 .can_receive = virtio_net_can_receive, 3034 .receive = virtio_net_receive, 3035 .link_status_changed = virtio_net_set_link_status, 3036 .query_rx_filter = virtio_net_query_rxfilter, 3037 .announce = virtio_net_announce, 3038 }; 3039 3040 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 3041 { 3042 VirtIONet *n = VIRTIO_NET(vdev); 3043 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 3044 assert(n->vhost_started); 3045 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx); 3046 } 3047 3048 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 3049 bool mask) 3050 { 3051 VirtIONet *n = VIRTIO_NET(vdev); 3052 NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx)); 3053 assert(n->vhost_started); 3054 vhost_net_virtqueue_mask(get_vhost_net(nc->peer), 3055 vdev, idx, mask); 3056 } 3057 3058 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features) 3059 { 3060 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC); 3061 3062 n->config_size = virtio_feature_get_config_size(feature_sizes, 3063 host_features); 3064 } 3065 3066 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 3067 const char *type) 3068 { 3069 /* 3070 * The name can be NULL, the netclient name will be type.x. 3071 */ 3072 assert(type != NULL); 3073 3074 g_free(n->netclient_name); 3075 g_free(n->netclient_type); 3076 n->netclient_name = g_strdup(name); 3077 n->netclient_type = g_strdup(type); 3078 } 3079 3080 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev) 3081 { 3082 HotplugHandler *hotplug_ctrl; 3083 PCIDevice *pci_dev; 3084 Error *err = NULL; 3085 3086 hotplug_ctrl = qdev_get_hotplug_handler(dev); 3087 if (hotplug_ctrl) { 3088 pci_dev = PCI_DEVICE(dev); 3089 pci_dev->partially_hotplugged = true; 3090 hotplug_handler_unplug_request(hotplug_ctrl, dev, &err); 3091 if (err) { 3092 error_report_err(err); 3093 return false; 3094 } 3095 } else { 3096 return false; 3097 } 3098 return true; 3099 } 3100 3101 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev, 3102 Error **errp) 3103 { 3104 Error *err = NULL; 3105 HotplugHandler *hotplug_ctrl; 3106 PCIDevice *pdev = PCI_DEVICE(dev); 3107 BusState *primary_bus; 3108 3109 if (!pdev->partially_hotplugged) { 3110 return true; 3111 } 3112 primary_bus = dev->parent_bus; 3113 if (!primary_bus) { 3114 error_setg(errp, "virtio_net: couldn't find primary bus"); 3115 return false; 3116 } 3117 qdev_set_parent_bus(dev, primary_bus, &error_abort); 3118 qatomic_set(&n->failover_primary_hidden, false); 3119 hotplug_ctrl = qdev_get_hotplug_handler(dev); 3120 if (hotplug_ctrl) { 3121 hotplug_handler_pre_plug(hotplug_ctrl, dev, &err); 3122 if (err) { 3123 goto out; 3124 } 3125 hotplug_handler_plug(hotplug_ctrl, dev, &err); 3126 } 3127 3128 out: 3129 error_propagate(errp, err); 3130 return !err; 3131 } 3132 3133 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationState *s) 3134 { 3135 bool should_be_hidden; 3136 Error *err = NULL; 3137 DeviceState *dev = failover_find_primary_device(n); 3138 3139 if (!dev) { 3140 return; 3141 } 3142 3143 should_be_hidden = qatomic_read(&n->failover_primary_hidden); 3144 3145 if (migration_in_setup(s) && !should_be_hidden) { 3146 if (failover_unplug_primary(n, dev)) { 3147 vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev); 3148 qapi_event_send_unplug_primary(dev->id); 3149 qatomic_set(&n->failover_primary_hidden, true); 3150 } else { 3151 warn_report("couldn't unplug primary device"); 3152 } 3153 } else if (migration_has_failed(s)) { 3154 /* We already unplugged the device let's plug it back */ 3155 if (!failover_replug_primary(n, dev, &err)) { 3156 if (err) { 3157 error_report_err(err); 3158 } 3159 } 3160 } 3161 } 3162 3163 static void virtio_net_migration_state_notifier(Notifier *notifier, void *data) 3164 { 3165 MigrationState *s = data; 3166 VirtIONet *n = container_of(notifier, VirtIONet, migration_state); 3167 virtio_net_handle_migration_primary(n, s); 3168 } 3169 3170 static bool failover_hide_primary_device(DeviceListener *listener, 3171 QemuOpts *device_opts) 3172 { 3173 VirtIONet *n = container_of(listener, VirtIONet, primary_listener); 3174 const char *standby_id; 3175 3176 if (!device_opts) { 3177 return false; 3178 } 3179 standby_id = qemu_opt_get(device_opts, "failover_pair_id"); 3180 if (g_strcmp0(standby_id, n->netclient_name) != 0) { 3181 return false; 3182 } 3183 3184 /* failover_primary_hidden is set during feature negotiation */ 3185 return qatomic_read(&n->failover_primary_hidden); 3186 } 3187 3188 static void virtio_net_device_realize(DeviceState *dev, Error **errp) 3189 { 3190 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3191 VirtIONet *n = VIRTIO_NET(dev); 3192 NetClientState *nc; 3193 int i; 3194 3195 if (n->net_conf.mtu) { 3196 n->host_features |= (1ULL << VIRTIO_NET_F_MTU); 3197 } 3198 3199 if (n->net_conf.duplex_str) { 3200 if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) { 3201 n->net_conf.duplex = DUPLEX_HALF; 3202 } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) { 3203 n->net_conf.duplex = DUPLEX_FULL; 3204 } else { 3205 error_setg(errp, "'duplex' must be 'half' or 'full'"); 3206 return; 3207 } 3208 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX); 3209 } else { 3210 n->net_conf.duplex = DUPLEX_UNKNOWN; 3211 } 3212 3213 if (n->net_conf.speed < SPEED_UNKNOWN) { 3214 error_setg(errp, "'speed' must be between 0 and INT_MAX"); 3215 return; 3216 } 3217 if (n->net_conf.speed >= 0) { 3218 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX); 3219 } 3220 3221 if (n->failover) { 3222 n->primary_listener.hide_device = failover_hide_primary_device; 3223 qatomic_set(&n->failover_primary_hidden, true); 3224 device_listener_register(&n->primary_listener); 3225 n->migration_state.notify = virtio_net_migration_state_notifier; 3226 add_migration_state_change_notifier(&n->migration_state); 3227 n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY); 3228 } 3229 3230 virtio_net_set_config_size(n, n->host_features); 3231 virtio_init(vdev, "virtio-net", VIRTIO_ID_NET, n->config_size); 3232 3233 /* 3234 * We set a lower limit on RX queue size to what it always was. 3235 * Guests that want a smaller ring can always resize it without 3236 * help from us (using virtio 1 and up). 3237 */ 3238 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE || 3239 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE || 3240 !is_power_of_2(n->net_conf.rx_queue_size)) { 3241 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), " 3242 "must be a power of 2 between %d and %d.", 3243 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE, 3244 VIRTQUEUE_MAX_SIZE); 3245 virtio_cleanup(vdev); 3246 return; 3247 } 3248 3249 if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE || 3250 n->net_conf.tx_queue_size > VIRTQUEUE_MAX_SIZE || 3251 !is_power_of_2(n->net_conf.tx_queue_size)) { 3252 error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), " 3253 "must be a power of 2 between %d and %d", 3254 n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE, 3255 VIRTQUEUE_MAX_SIZE); 3256 virtio_cleanup(vdev); 3257 return; 3258 } 3259 3260 n->max_queues = MAX(n->nic_conf.peers.queues, 1); 3261 if (n->max_queues * 2 + 1 > VIRTIO_QUEUE_MAX) { 3262 error_setg(errp, "Invalid number of queues (= %" PRIu32 "), " 3263 "must be a positive integer less than %d.", 3264 n->max_queues, (VIRTIO_QUEUE_MAX - 1) / 2); 3265 virtio_cleanup(vdev); 3266 return; 3267 } 3268 n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues); 3269 n->curr_queues = 1; 3270 n->tx_timeout = n->net_conf.txtimer; 3271 3272 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 3273 && strcmp(n->net_conf.tx, "bh")) { 3274 warn_report("virtio-net: " 3275 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 3276 n->net_conf.tx); 3277 error_printf("Defaulting to \"bh\""); 3278 } 3279 3280 n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n), 3281 n->net_conf.tx_queue_size); 3282 3283 for (i = 0; i < n->max_queues; i++) { 3284 virtio_net_add_queue(n, i); 3285 } 3286 3287 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 3288 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 3289 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 3290 n->status = VIRTIO_NET_S_LINK_UP; 3291 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(), 3292 QEMU_CLOCK_VIRTUAL, 3293 virtio_net_announce_timer, n); 3294 n->announce_timer.round = 0; 3295 3296 if (n->netclient_type) { 3297 /* 3298 * Happen when virtio_net_set_netclient_name has been called. 3299 */ 3300 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 3301 n->netclient_type, n->netclient_name, n); 3302 } else { 3303 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 3304 object_get_typename(OBJECT(dev)), dev->id, n); 3305 } 3306 3307 peer_test_vnet_hdr(n); 3308 if (peer_has_vnet_hdr(n)) { 3309 for (i = 0; i < n->max_queues; i++) { 3310 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true); 3311 } 3312 n->host_hdr_len = sizeof(struct virtio_net_hdr); 3313 } else { 3314 n->host_hdr_len = 0; 3315 } 3316 3317 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 3318 3319 n->vqs[0].tx_waiting = 0; 3320 n->tx_burst = n->net_conf.txburst; 3321 virtio_net_set_mrg_rx_bufs(n, 0, 0, 0); 3322 n->promisc = 1; /* for compatibility */ 3323 3324 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 3325 3326 n->vlans = g_malloc0(MAX_VLAN >> 3); 3327 3328 nc = qemu_get_queue(n->nic); 3329 nc->rxfilter_notify_enabled = 1; 3330 3331 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 3332 struct virtio_net_config netcfg = {}; 3333 memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN); 3334 vhost_net_set_config(get_vhost_net(nc->peer), 3335 (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_MASTER); 3336 } 3337 QTAILQ_INIT(&n->rsc_chains); 3338 n->qdev = dev; 3339 3340 net_rx_pkt_init(&n->rx_pkt, false); 3341 } 3342 3343 static void virtio_net_device_unrealize(DeviceState *dev) 3344 { 3345 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3346 VirtIONet *n = VIRTIO_NET(dev); 3347 int i, max_queues; 3348 3349 /* This will stop vhost backend if appropriate. */ 3350 virtio_net_set_status(vdev, 0); 3351 3352 g_free(n->netclient_name); 3353 n->netclient_name = NULL; 3354 g_free(n->netclient_type); 3355 n->netclient_type = NULL; 3356 3357 g_free(n->mac_table.macs); 3358 g_free(n->vlans); 3359 3360 if (n->failover) { 3361 device_listener_unregister(&n->primary_listener); 3362 } 3363 3364 max_queues = n->multiqueue ? n->max_queues : 1; 3365 for (i = 0; i < max_queues; i++) { 3366 virtio_net_del_queue(n, i); 3367 } 3368 /* delete also control vq */ 3369 virtio_del_queue(vdev, max_queues * 2); 3370 qemu_announce_timer_del(&n->announce_timer, false); 3371 g_free(n->vqs); 3372 qemu_del_nic(n->nic); 3373 virtio_net_rsc_cleanup(n); 3374 g_free(n->rss_data.indirections_table); 3375 net_rx_pkt_uninit(n->rx_pkt); 3376 virtio_cleanup(vdev); 3377 } 3378 3379 static void virtio_net_instance_init(Object *obj) 3380 { 3381 VirtIONet *n = VIRTIO_NET(obj); 3382 3383 /* 3384 * The default config_size is sizeof(struct virtio_net_config). 3385 * Can be overriden with virtio_net_set_config_size. 3386 */ 3387 n->config_size = sizeof(struct virtio_net_config); 3388 device_add_bootindex_property(obj, &n->nic_conf.bootindex, 3389 "bootindex", "/ethernet-phy@0", 3390 DEVICE(n)); 3391 } 3392 3393 static int virtio_net_pre_save(void *opaque) 3394 { 3395 VirtIONet *n = opaque; 3396 3397 /* At this point, backend must be stopped, otherwise 3398 * it might keep writing to memory. */ 3399 assert(!n->vhost_started); 3400 3401 return 0; 3402 } 3403 3404 static bool primary_unplug_pending(void *opaque) 3405 { 3406 DeviceState *dev = opaque; 3407 DeviceState *primary; 3408 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3409 VirtIONet *n = VIRTIO_NET(vdev); 3410 3411 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) { 3412 return false; 3413 } 3414 primary = failover_find_primary_device(n); 3415 return primary ? primary->pending_deleted_event : false; 3416 } 3417 3418 static bool dev_unplug_pending(void *opaque) 3419 { 3420 DeviceState *dev = opaque; 3421 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); 3422 3423 return vdc->primary_unplug_pending(dev); 3424 } 3425 3426 static const VMStateDescription vmstate_virtio_net = { 3427 .name = "virtio-net", 3428 .minimum_version_id = VIRTIO_NET_VM_VERSION, 3429 .version_id = VIRTIO_NET_VM_VERSION, 3430 .fields = (VMStateField[]) { 3431 VMSTATE_VIRTIO_DEVICE, 3432 VMSTATE_END_OF_LIST() 3433 }, 3434 .pre_save = virtio_net_pre_save, 3435 .dev_unplug_pending = dev_unplug_pending, 3436 }; 3437 3438 static Property virtio_net_properties[] = { 3439 DEFINE_PROP_BIT64("csum", VirtIONet, host_features, 3440 VIRTIO_NET_F_CSUM, true), 3441 DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features, 3442 VIRTIO_NET_F_GUEST_CSUM, true), 3443 DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true), 3444 DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features, 3445 VIRTIO_NET_F_GUEST_TSO4, true), 3446 DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features, 3447 VIRTIO_NET_F_GUEST_TSO6, true), 3448 DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features, 3449 VIRTIO_NET_F_GUEST_ECN, true), 3450 DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features, 3451 VIRTIO_NET_F_GUEST_UFO, true), 3452 DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features, 3453 VIRTIO_NET_F_GUEST_ANNOUNCE, true), 3454 DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features, 3455 VIRTIO_NET_F_HOST_TSO4, true), 3456 DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features, 3457 VIRTIO_NET_F_HOST_TSO6, true), 3458 DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features, 3459 VIRTIO_NET_F_HOST_ECN, true), 3460 DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features, 3461 VIRTIO_NET_F_HOST_UFO, true), 3462 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features, 3463 VIRTIO_NET_F_MRG_RXBUF, true), 3464 DEFINE_PROP_BIT64("status", VirtIONet, host_features, 3465 VIRTIO_NET_F_STATUS, true), 3466 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features, 3467 VIRTIO_NET_F_CTRL_VQ, true), 3468 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features, 3469 VIRTIO_NET_F_CTRL_RX, true), 3470 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features, 3471 VIRTIO_NET_F_CTRL_VLAN, true), 3472 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features, 3473 VIRTIO_NET_F_CTRL_RX_EXTRA, true), 3474 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features, 3475 VIRTIO_NET_F_CTRL_MAC_ADDR, true), 3476 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features, 3477 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true), 3478 DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false), 3479 DEFINE_PROP_BIT64("rss", VirtIONet, host_features, 3480 VIRTIO_NET_F_RSS, false), 3481 DEFINE_PROP_BIT64("hash", VirtIONet, host_features, 3482 VIRTIO_NET_F_HASH_REPORT, false), 3483 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features, 3484 VIRTIO_NET_F_RSC_EXT, false), 3485 DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout, 3486 VIRTIO_NET_RSC_DEFAULT_INTERVAL), 3487 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 3488 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 3489 TX_TIMER_INTERVAL), 3490 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 3491 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 3492 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size, 3493 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE), 3494 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size, 3495 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE), 3496 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0), 3497 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend, 3498 true), 3499 DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN), 3500 DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str), 3501 DEFINE_PROP_BOOL("failover", VirtIONet, failover, false), 3502 DEFINE_PROP_END_OF_LIST(), 3503 }; 3504 3505 static void virtio_net_class_init(ObjectClass *klass, void *data) 3506 { 3507 DeviceClass *dc = DEVICE_CLASS(klass); 3508 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 3509 3510 device_class_set_props(dc, virtio_net_properties); 3511 dc->vmsd = &vmstate_virtio_net; 3512 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 3513 vdc->realize = virtio_net_device_realize; 3514 vdc->unrealize = virtio_net_device_unrealize; 3515 vdc->get_config = virtio_net_get_config; 3516 vdc->set_config = virtio_net_set_config; 3517 vdc->get_features = virtio_net_get_features; 3518 vdc->set_features = virtio_net_set_features; 3519 vdc->bad_features = virtio_net_bad_features; 3520 vdc->reset = virtio_net_reset; 3521 vdc->set_status = virtio_net_set_status; 3522 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 3523 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 3524 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO); 3525 vdc->post_load = virtio_net_post_load_virtio; 3526 vdc->vmsd = &vmstate_virtio_net_device; 3527 vdc->primary_unplug_pending = primary_unplug_pending; 3528 } 3529 3530 static const TypeInfo virtio_net_info = { 3531 .name = TYPE_VIRTIO_NET, 3532 .parent = TYPE_VIRTIO_DEVICE, 3533 .instance_size = sizeof(VirtIONet), 3534 .instance_init = virtio_net_instance_init, 3535 .class_init = virtio_net_class_init, 3536 }; 3537 3538 static void virtio_register_types(void) 3539 { 3540 type_register_static(&virtio_net_info); 3541 } 3542 3543 type_init(virtio_register_types) 3544