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/log.h" 18 #include "qemu/main-loop.h" 19 #include "qemu/module.h" 20 #include "hw/virtio/virtio.h" 21 #include "net/net.h" 22 #include "net/checksum.h" 23 #include "net/tap.h" 24 #include "qemu/error-report.h" 25 #include "qemu/timer.h" 26 #include "qemu/option.h" 27 #include "qemu/option_int.h" 28 #include "qemu/config-file.h" 29 #include "qapi/qmp/qdict.h" 30 #include "hw/virtio/virtio-net.h" 31 #include "net/vhost_net.h" 32 #include "net/announce.h" 33 #include "hw/virtio/virtio-bus.h" 34 #include "qapi/error.h" 35 #include "qapi/qapi-events-net.h" 36 #include "hw/qdev-properties.h" 37 #include "qapi/qapi-types-migration.h" 38 #include "qapi/qapi-events-migration.h" 39 #include "hw/virtio/virtio-access.h" 40 #include "migration/misc.h" 41 #include "standard-headers/linux/ethtool.h" 42 #include "sysemu/sysemu.h" 43 #include "trace.h" 44 #include "monitor/qdev.h" 45 #include "hw/pci/pci_device.h" 46 #include "net_rx_pkt.h" 47 #include "hw/virtio/vhost.h" 48 #include "sysemu/qtest.h" 49 50 #define VIRTIO_NET_VM_VERSION 11 51 52 #define MAX_VLAN (1 << 12) /* Per 802.1Q definition */ 53 54 /* previously fixed value */ 55 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256 56 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256 57 58 /* for now, only allow larger queue_pairs; with virtio-1, guest can downsize */ 59 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 60 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 61 62 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */ 63 64 #define VIRTIO_NET_TCP_FLAG 0x3F 65 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000 66 67 /* IPv4 max payload, 16 bits in the header */ 68 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header)) 69 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535 70 71 /* header length value in ip header without option */ 72 #define VIRTIO_NET_IP4_HEADER_LENGTH 5 73 74 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */ 75 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD 76 77 /* Purge coalesced packets timer interval, This value affects the performance 78 a lot, and should be tuned carefully, '300000'(300us) is the recommended 79 value to pass the WHQL test, '50000' can gain 2x netperf throughput with 80 tso/gso/gro 'off'. */ 81 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000 82 83 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \ 84 VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \ 85 VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \ 86 VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \ 87 VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \ 88 VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \ 89 VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \ 90 VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \ 91 VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) 92 93 static const VirtIOFeature feature_sizes[] = { 94 {.flags = 1ULL << VIRTIO_NET_F_MAC, 95 .end = endof(struct virtio_net_config, mac)}, 96 {.flags = 1ULL << VIRTIO_NET_F_STATUS, 97 .end = endof(struct virtio_net_config, status)}, 98 {.flags = 1ULL << VIRTIO_NET_F_MQ, 99 .end = endof(struct virtio_net_config, max_virtqueue_pairs)}, 100 {.flags = 1ULL << VIRTIO_NET_F_MTU, 101 .end = endof(struct virtio_net_config, mtu)}, 102 {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX, 103 .end = endof(struct virtio_net_config, duplex)}, 104 {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT), 105 .end = endof(struct virtio_net_config, supported_hash_types)}, 106 {} 107 }; 108 109 static const VirtIOConfigSizeParams cfg_size_params = { 110 .min_size = endof(struct virtio_net_config, mac), 111 .max_size = sizeof(struct virtio_net_config), 112 .feature_sizes = feature_sizes 113 }; 114 115 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc) 116 { 117 VirtIONet *n = qemu_get_nic_opaque(nc); 118 119 return &n->vqs[nc->queue_index]; 120 } 121 122 static int vq2q(int queue_index) 123 { 124 return queue_index / 2; 125 } 126 127 static void flush_or_purge_queued_packets(NetClientState *nc) 128 { 129 if (!nc->peer) { 130 return; 131 } 132 133 qemu_flush_or_purge_queued_packets(nc->peer, true); 134 assert(!virtio_net_get_subqueue(nc)->async_tx.elem); 135 } 136 137 /* TODO 138 * - we could suppress RX interrupt if we were so inclined. 139 */ 140 141 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config) 142 { 143 VirtIONet *n = VIRTIO_NET(vdev); 144 struct virtio_net_config netcfg; 145 NetClientState *nc = qemu_get_queue(n->nic); 146 static const MACAddr zero = { .a = { 0, 0, 0, 0, 0, 0 } }; 147 148 int ret = 0; 149 memset(&netcfg, 0 , sizeof(struct virtio_net_config)); 150 virtio_stw_p(vdev, &netcfg.status, n->status); 151 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queue_pairs); 152 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu); 153 memcpy(netcfg.mac, n->mac, ETH_ALEN); 154 virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed); 155 netcfg.duplex = n->net_conf.duplex; 156 netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE; 157 virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length, 158 virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ? 159 VIRTIO_NET_RSS_MAX_TABLE_LEN : 1); 160 virtio_stl_p(vdev, &netcfg.supported_hash_types, 161 VIRTIO_NET_RSS_SUPPORTED_HASHES); 162 memcpy(config, &netcfg, n->config_size); 163 164 /* 165 * Is this VDPA? No peer means not VDPA: there's no way to 166 * disconnect/reconnect a VDPA peer. 167 */ 168 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 169 ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg, 170 n->config_size); 171 if (ret == -1) { 172 return; 173 } 174 175 /* 176 * Some NIC/kernel combinations present 0 as the mac address. As that 177 * is not a legal address, try to proceed with the address from the 178 * QEMU command line in the hope that the address has been configured 179 * correctly elsewhere - just not reported by the device. 180 */ 181 if (memcmp(&netcfg.mac, &zero, sizeof(zero)) == 0) { 182 info_report("Zero hardware mac address detected. Ignoring."); 183 memcpy(netcfg.mac, n->mac, ETH_ALEN); 184 } 185 186 netcfg.status |= virtio_tswap16(vdev, 187 n->status & VIRTIO_NET_S_ANNOUNCE); 188 memcpy(config, &netcfg, n->config_size); 189 } 190 } 191 192 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config) 193 { 194 VirtIONet *n = VIRTIO_NET(vdev); 195 struct virtio_net_config netcfg = {}; 196 NetClientState *nc = qemu_get_queue(n->nic); 197 198 memcpy(&netcfg, config, n->config_size); 199 200 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) && 201 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) && 202 memcmp(netcfg.mac, n->mac, ETH_ALEN)) { 203 memcpy(n->mac, netcfg.mac, ETH_ALEN); 204 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 205 } 206 207 /* 208 * Is this VDPA? No peer means not VDPA: there's no way to 209 * disconnect/reconnect a VDPA peer. 210 */ 211 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 212 vhost_net_set_config(get_vhost_net(nc->peer), 213 (uint8_t *)&netcfg, 0, n->config_size, 214 VHOST_SET_CONFIG_TYPE_FRONTEND); 215 } 216 } 217 218 static bool virtio_net_started(VirtIONet *n, uint8_t status) 219 { 220 VirtIODevice *vdev = VIRTIO_DEVICE(n); 221 return (status & VIRTIO_CONFIG_S_DRIVER_OK) && 222 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running; 223 } 224 225 static void virtio_net_announce_notify(VirtIONet *net) 226 { 227 VirtIODevice *vdev = VIRTIO_DEVICE(net); 228 trace_virtio_net_announce_notify(); 229 230 net->status |= VIRTIO_NET_S_ANNOUNCE; 231 virtio_notify_config(vdev); 232 } 233 234 static void virtio_net_announce_timer(void *opaque) 235 { 236 VirtIONet *n = opaque; 237 trace_virtio_net_announce_timer(n->announce_timer.round); 238 239 n->announce_timer.round--; 240 virtio_net_announce_notify(n); 241 } 242 243 static void virtio_net_announce(NetClientState *nc) 244 { 245 VirtIONet *n = qemu_get_nic_opaque(nc); 246 VirtIODevice *vdev = VIRTIO_DEVICE(n); 247 248 /* 249 * Make sure the virtio migration announcement timer isn't running 250 * If it is, let it trigger announcement so that we do not cause 251 * confusion. 252 */ 253 if (n->announce_timer.round) { 254 return; 255 } 256 257 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 258 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 259 virtio_net_announce_notify(n); 260 } 261 } 262 263 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status) 264 { 265 VirtIODevice *vdev = VIRTIO_DEVICE(n); 266 NetClientState *nc = qemu_get_queue(n->nic); 267 int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 268 int cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ? 269 n->max_ncs - n->max_queue_pairs : 0; 270 271 if (!get_vhost_net(nc->peer)) { 272 return; 273 } 274 275 if ((virtio_net_started(n, status) && !nc->peer->link_down) == 276 !!n->vhost_started) { 277 return; 278 } 279 if (!n->vhost_started) { 280 int r, i; 281 282 if (n->needs_vnet_hdr_swap) { 283 error_report("backend does not support %s vnet headers; " 284 "falling back on userspace virtio", 285 virtio_is_big_endian(vdev) ? "BE" : "LE"); 286 return; 287 } 288 289 /* Any packets outstanding? Purge them to avoid touching rings 290 * when vhost is running. 291 */ 292 for (i = 0; i < queue_pairs; i++) { 293 NetClientState *qnc = qemu_get_subqueue(n->nic, i); 294 295 /* Purge both directions: TX and RX. */ 296 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc); 297 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer); 298 } 299 300 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) { 301 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu); 302 if (r < 0) { 303 error_report("%uBytes MTU not supported by the backend", 304 n->net_conf.mtu); 305 306 return; 307 } 308 } 309 310 n->vhost_started = 1; 311 r = vhost_net_start(vdev, n->nic->ncs, queue_pairs, cvq); 312 if (r < 0) { 313 error_report("unable to start vhost net: %d: " 314 "falling back on userspace virtio", -r); 315 n->vhost_started = 0; 316 } 317 } else { 318 vhost_net_stop(vdev, n->nic->ncs, queue_pairs, cvq); 319 n->vhost_started = 0; 320 } 321 } 322 323 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev, 324 NetClientState *peer, 325 bool enable) 326 { 327 if (virtio_is_big_endian(vdev)) { 328 return qemu_set_vnet_be(peer, enable); 329 } else { 330 return qemu_set_vnet_le(peer, enable); 331 } 332 } 333 334 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs, 335 int queue_pairs, bool enable) 336 { 337 int i; 338 339 for (i = 0; i < queue_pairs; i++) { 340 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 && 341 enable) { 342 while (--i >= 0) { 343 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false); 344 } 345 346 return true; 347 } 348 } 349 350 return false; 351 } 352 353 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status) 354 { 355 VirtIODevice *vdev = VIRTIO_DEVICE(n); 356 int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 357 358 if (virtio_net_started(n, status)) { 359 /* Before using the device, we tell the network backend about the 360 * endianness to use when parsing vnet headers. If the backend 361 * can't do it, we fallback onto fixing the headers in the core 362 * virtio-net code. 363 */ 364 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs, 365 queue_pairs, true); 366 } else if (virtio_net_started(n, vdev->status)) { 367 /* After using the device, we need to reset the network backend to 368 * the default (guest native endianness), otherwise the guest may 369 * lose network connectivity if it is rebooted into a different 370 * endianness. 371 */ 372 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queue_pairs, false); 373 } 374 } 375 376 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq) 377 { 378 unsigned int dropped = virtqueue_drop_all(vq); 379 if (dropped) { 380 virtio_notify(vdev, vq); 381 } 382 } 383 384 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status) 385 { 386 VirtIONet *n = VIRTIO_NET(vdev); 387 VirtIONetQueue *q; 388 int i; 389 uint8_t queue_status; 390 391 virtio_net_vnet_endian_status(n, status); 392 virtio_net_vhost_status(n, status); 393 394 for (i = 0; i < n->max_queue_pairs; i++) { 395 NetClientState *ncs = qemu_get_subqueue(n->nic, i); 396 bool queue_started; 397 q = &n->vqs[i]; 398 399 if ((!n->multiqueue && i != 0) || i >= n->curr_queue_pairs) { 400 queue_status = 0; 401 } else { 402 queue_status = status; 403 } 404 queue_started = 405 virtio_net_started(n, queue_status) && !n->vhost_started; 406 407 if (queue_started) { 408 qemu_flush_queued_packets(ncs); 409 } 410 411 if (!q->tx_waiting) { 412 continue; 413 } 414 415 if (queue_started) { 416 if (q->tx_timer) { 417 timer_mod(q->tx_timer, 418 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 419 } else { 420 qemu_bh_schedule(q->tx_bh); 421 } 422 } else { 423 if (q->tx_timer) { 424 timer_del(q->tx_timer); 425 } else { 426 qemu_bh_cancel(q->tx_bh); 427 } 428 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 && 429 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) && 430 vdev->vm_running) { 431 /* if tx is waiting we are likely have some packets in tx queue 432 * and disabled notification */ 433 q->tx_waiting = 0; 434 virtio_queue_set_notification(q->tx_vq, 1); 435 virtio_net_drop_tx_queue_data(vdev, q->tx_vq); 436 } 437 } 438 } 439 } 440 441 static void virtio_net_set_link_status(NetClientState *nc) 442 { 443 VirtIONet *n = qemu_get_nic_opaque(nc); 444 VirtIODevice *vdev = VIRTIO_DEVICE(n); 445 uint16_t old_status = n->status; 446 447 if (nc->link_down) 448 n->status &= ~VIRTIO_NET_S_LINK_UP; 449 else 450 n->status |= VIRTIO_NET_S_LINK_UP; 451 452 if (n->status != old_status) 453 virtio_notify_config(vdev); 454 455 virtio_net_set_status(vdev, vdev->status); 456 } 457 458 static void rxfilter_notify(NetClientState *nc) 459 { 460 VirtIONet *n = qemu_get_nic_opaque(nc); 461 462 if (nc->rxfilter_notify_enabled) { 463 char *path = object_get_canonical_path(OBJECT(n->qdev)); 464 qapi_event_send_nic_rx_filter_changed(n->netclient_name, path); 465 g_free(path); 466 467 /* disable event notification to avoid events flooding */ 468 nc->rxfilter_notify_enabled = 0; 469 } 470 } 471 472 static intList *get_vlan_table(VirtIONet *n) 473 { 474 intList *list; 475 int i, j; 476 477 list = NULL; 478 for (i = 0; i < MAX_VLAN >> 5; i++) { 479 for (j = 0; n->vlans[i] && j <= 0x1f; j++) { 480 if (n->vlans[i] & (1U << j)) { 481 QAPI_LIST_PREPEND(list, (i << 5) + j); 482 } 483 } 484 } 485 486 return list; 487 } 488 489 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc) 490 { 491 VirtIONet *n = qemu_get_nic_opaque(nc); 492 VirtIODevice *vdev = VIRTIO_DEVICE(n); 493 RxFilterInfo *info; 494 strList *str_list; 495 int i; 496 497 info = g_malloc0(sizeof(*info)); 498 info->name = g_strdup(nc->name); 499 info->promiscuous = n->promisc; 500 501 if (n->nouni) { 502 info->unicast = RX_STATE_NONE; 503 } else if (n->alluni) { 504 info->unicast = RX_STATE_ALL; 505 } else { 506 info->unicast = RX_STATE_NORMAL; 507 } 508 509 if (n->nomulti) { 510 info->multicast = RX_STATE_NONE; 511 } else if (n->allmulti) { 512 info->multicast = RX_STATE_ALL; 513 } else { 514 info->multicast = RX_STATE_NORMAL; 515 } 516 517 info->broadcast_allowed = n->nobcast; 518 info->multicast_overflow = n->mac_table.multi_overflow; 519 info->unicast_overflow = n->mac_table.uni_overflow; 520 521 info->main_mac = qemu_mac_strdup_printf(n->mac); 522 523 str_list = NULL; 524 for (i = 0; i < n->mac_table.first_multi; i++) { 525 QAPI_LIST_PREPEND(str_list, 526 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN)); 527 } 528 info->unicast_table = str_list; 529 530 str_list = NULL; 531 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 532 QAPI_LIST_PREPEND(str_list, 533 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN)); 534 } 535 info->multicast_table = str_list; 536 info->vlan_table = get_vlan_table(n); 537 538 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) { 539 info->vlan = RX_STATE_ALL; 540 } else if (!info->vlan_table) { 541 info->vlan = RX_STATE_NONE; 542 } else { 543 info->vlan = RX_STATE_NORMAL; 544 } 545 546 /* enable event notification after query */ 547 nc->rxfilter_notify_enabled = 1; 548 549 return info; 550 } 551 552 static void virtio_net_queue_reset(VirtIODevice *vdev, uint32_t queue_index) 553 { 554 VirtIONet *n = VIRTIO_NET(vdev); 555 NetClientState *nc; 556 557 /* validate queue_index and skip for cvq */ 558 if (queue_index >= n->max_queue_pairs * 2) { 559 return; 560 } 561 562 nc = qemu_get_subqueue(n->nic, vq2q(queue_index)); 563 564 if (!nc->peer) { 565 return; 566 } 567 568 if (get_vhost_net(nc->peer) && 569 nc->peer->info->type == NET_CLIENT_DRIVER_TAP) { 570 vhost_net_virtqueue_reset(vdev, nc, queue_index); 571 } 572 573 flush_or_purge_queued_packets(nc); 574 } 575 576 static void virtio_net_queue_enable(VirtIODevice *vdev, uint32_t queue_index) 577 { 578 VirtIONet *n = VIRTIO_NET(vdev); 579 NetClientState *nc; 580 int r; 581 582 /* validate queue_index and skip for cvq */ 583 if (queue_index >= n->max_queue_pairs * 2) { 584 return; 585 } 586 587 nc = qemu_get_subqueue(n->nic, vq2q(queue_index)); 588 589 if (!nc->peer || !vdev->vhost_started) { 590 return; 591 } 592 593 if (get_vhost_net(nc->peer) && 594 nc->peer->info->type == NET_CLIENT_DRIVER_TAP) { 595 r = vhost_net_virtqueue_restart(vdev, nc, queue_index); 596 if (r < 0) { 597 error_report("unable to restart vhost net virtqueue: %d, " 598 "when resetting the queue", queue_index); 599 } 600 } 601 } 602 603 static void virtio_net_reset(VirtIODevice *vdev) 604 { 605 VirtIONet *n = VIRTIO_NET(vdev); 606 int i; 607 608 /* Reset back to compatibility mode */ 609 n->promisc = 1; 610 n->allmulti = 0; 611 n->alluni = 0; 612 n->nomulti = 0; 613 n->nouni = 0; 614 n->nobcast = 0; 615 /* multiqueue is disabled by default */ 616 n->curr_queue_pairs = 1; 617 timer_del(n->announce_timer.tm); 618 n->announce_timer.round = 0; 619 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 620 621 /* Flush any MAC and VLAN filter table state */ 622 n->mac_table.in_use = 0; 623 n->mac_table.first_multi = 0; 624 n->mac_table.multi_overflow = 0; 625 n->mac_table.uni_overflow = 0; 626 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN); 627 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac)); 628 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 629 memset(n->vlans, 0, MAX_VLAN >> 3); 630 631 /* Flush any async TX */ 632 for (i = 0; i < n->max_queue_pairs; i++) { 633 flush_or_purge_queued_packets(qemu_get_subqueue(n->nic, i)); 634 } 635 } 636 637 static void peer_test_vnet_hdr(VirtIONet *n) 638 { 639 NetClientState *nc = qemu_get_queue(n->nic); 640 if (!nc->peer) { 641 return; 642 } 643 644 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer); 645 } 646 647 static int peer_has_vnet_hdr(VirtIONet *n) 648 { 649 return n->has_vnet_hdr; 650 } 651 652 static int peer_has_ufo(VirtIONet *n) 653 { 654 if (!peer_has_vnet_hdr(n)) 655 return 0; 656 657 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer); 658 659 return n->has_ufo; 660 } 661 662 static int peer_has_uso(VirtIONet *n) 663 { 664 if (!peer_has_vnet_hdr(n)) { 665 return 0; 666 } 667 668 return qemu_has_uso(qemu_get_queue(n->nic)->peer); 669 } 670 671 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs, 672 int version_1, int hash_report) 673 { 674 int i; 675 NetClientState *nc; 676 677 n->mergeable_rx_bufs = mergeable_rx_bufs; 678 679 if (version_1) { 680 n->guest_hdr_len = hash_report ? 681 sizeof(struct virtio_net_hdr_v1_hash) : 682 sizeof(struct virtio_net_hdr_mrg_rxbuf); 683 n->rss_data.populate_hash = !!hash_report; 684 } else { 685 n->guest_hdr_len = n->mergeable_rx_bufs ? 686 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 687 sizeof(struct virtio_net_hdr); 688 } 689 690 for (i = 0; i < n->max_queue_pairs; i++) { 691 nc = qemu_get_subqueue(n->nic, i); 692 693 if (peer_has_vnet_hdr(n) && 694 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) { 695 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len); 696 n->host_hdr_len = n->guest_hdr_len; 697 } 698 } 699 } 700 701 static int virtio_net_max_tx_queue_size(VirtIONet *n) 702 { 703 NetClientState *peer = n->nic_conf.peers.ncs[0]; 704 705 /* 706 * Backends other than vhost-user or vhost-vdpa don't support max queue 707 * size. 708 */ 709 if (!peer) { 710 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 711 } 712 713 switch(peer->info->type) { 714 case NET_CLIENT_DRIVER_VHOST_USER: 715 case NET_CLIENT_DRIVER_VHOST_VDPA: 716 return VIRTQUEUE_MAX_SIZE; 717 default: 718 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 719 }; 720 } 721 722 static int peer_attach(VirtIONet *n, int index) 723 { 724 NetClientState *nc = qemu_get_subqueue(n->nic, index); 725 726 if (!nc->peer) { 727 return 0; 728 } 729 730 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 731 vhost_set_vring_enable(nc->peer, 1); 732 } 733 734 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 735 return 0; 736 } 737 738 if (n->max_queue_pairs == 1) { 739 return 0; 740 } 741 742 return tap_enable(nc->peer); 743 } 744 745 static int peer_detach(VirtIONet *n, int index) 746 { 747 NetClientState *nc = qemu_get_subqueue(n->nic, index); 748 749 if (!nc->peer) { 750 return 0; 751 } 752 753 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 754 vhost_set_vring_enable(nc->peer, 0); 755 } 756 757 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 758 return 0; 759 } 760 761 return tap_disable(nc->peer); 762 } 763 764 static void virtio_net_set_queue_pairs(VirtIONet *n) 765 { 766 int i; 767 int r; 768 769 if (n->nic->peer_deleted) { 770 return; 771 } 772 773 for (i = 0; i < n->max_queue_pairs; i++) { 774 if (i < n->curr_queue_pairs) { 775 r = peer_attach(n, i); 776 assert(!r); 777 } else { 778 r = peer_detach(n, i); 779 assert(!r); 780 } 781 } 782 } 783 784 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); 785 786 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features, 787 Error **errp) 788 { 789 VirtIONet *n = VIRTIO_NET(vdev); 790 NetClientState *nc = qemu_get_queue(n->nic); 791 792 /* Firstly sync all virtio-net possible supported features */ 793 features |= n->host_features; 794 795 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 796 797 if (!peer_has_vnet_hdr(n)) { 798 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM); 799 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4); 800 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6); 801 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN); 802 803 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM); 804 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4); 805 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6); 806 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN); 807 808 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_USO); 809 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO4); 810 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO6); 811 812 virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT); 813 } 814 815 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 816 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO); 817 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO); 818 } 819 820 if (!peer_has_uso(n)) { 821 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_USO); 822 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO4); 823 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO6); 824 } 825 826 if (!get_vhost_net(nc->peer)) { 827 return features; 828 } 829 830 if (!ebpf_rss_is_loaded(&n->ebpf_rss)) { 831 virtio_clear_feature(&features, VIRTIO_NET_F_RSS); 832 } 833 features = vhost_net_get_features(get_vhost_net(nc->peer), features); 834 vdev->backend_features = features; 835 836 if (n->mtu_bypass_backend && 837 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) { 838 features |= (1ULL << VIRTIO_NET_F_MTU); 839 } 840 841 /* 842 * Since GUEST_ANNOUNCE is emulated the feature bit could be set without 843 * enabled. This happens in the vDPA case. 844 * 845 * Make sure the feature set is not incoherent, as the driver could refuse 846 * to start. 847 * 848 * TODO: QEMU is able to emulate a CVQ just for guest_announce purposes, 849 * helping guest to notify the new location with vDPA devices that does not 850 * support it. 851 */ 852 if (!virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_CTRL_VQ)) { 853 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ANNOUNCE); 854 } 855 856 return features; 857 } 858 859 static uint64_t virtio_net_bad_features(VirtIODevice *vdev) 860 { 861 uint64_t features = 0; 862 863 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 864 * but also these: */ 865 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 866 virtio_add_feature(&features, VIRTIO_NET_F_CSUM); 867 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4); 868 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6); 869 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN); 870 871 return features; 872 } 873 874 static void virtio_net_apply_guest_offloads(VirtIONet *n) 875 { 876 qemu_set_offload(qemu_get_queue(n->nic)->peer, 877 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 878 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 879 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 880 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 881 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)), 882 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_USO4)), 883 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_USO6))); 884 } 885 886 static uint64_t virtio_net_guest_offloads_by_features(uint64_t features) 887 { 888 static const uint64_t guest_offloads_mask = 889 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 890 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 891 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 892 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 893 (1ULL << VIRTIO_NET_F_GUEST_UFO) | 894 (1ULL << VIRTIO_NET_F_GUEST_USO4) | 895 (1ULL << VIRTIO_NET_F_GUEST_USO6); 896 897 return guest_offloads_mask & features; 898 } 899 900 uint64_t virtio_net_supported_guest_offloads(const VirtIONet *n) 901 { 902 VirtIODevice *vdev = VIRTIO_DEVICE(n); 903 return virtio_net_guest_offloads_by_features(vdev->guest_features); 904 } 905 906 typedef struct { 907 VirtIONet *n; 908 DeviceState *dev; 909 } FailoverDevice; 910 911 /** 912 * Set the failover primary device 913 * 914 * @opaque: FailoverId to setup 915 * @opts: opts for device we are handling 916 * @errp: returns an error if this function fails 917 */ 918 static int failover_set_primary(DeviceState *dev, void *opaque) 919 { 920 FailoverDevice *fdev = opaque; 921 PCIDevice *pci_dev = (PCIDevice *) 922 object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE); 923 924 if (!pci_dev) { 925 return 0; 926 } 927 928 if (!g_strcmp0(pci_dev->failover_pair_id, fdev->n->netclient_name)) { 929 fdev->dev = dev; 930 return 1; 931 } 932 933 return 0; 934 } 935 936 /** 937 * Find the primary device for this failover virtio-net 938 * 939 * @n: VirtIONet device 940 * @errp: returns an error if this function fails 941 */ 942 static DeviceState *failover_find_primary_device(VirtIONet *n) 943 { 944 FailoverDevice fdev = { 945 .n = n, 946 }; 947 948 qbus_walk_children(sysbus_get_default(), failover_set_primary, NULL, 949 NULL, NULL, &fdev); 950 return fdev.dev; 951 } 952 953 static void failover_add_primary(VirtIONet *n, Error **errp) 954 { 955 Error *err = NULL; 956 DeviceState *dev = failover_find_primary_device(n); 957 958 if (dev) { 959 return; 960 } 961 962 if (!n->primary_opts) { 963 error_setg(errp, "Primary device not found"); 964 error_append_hint(errp, "Virtio-net failover will not work. Make " 965 "sure primary device has parameter" 966 " failover_pair_id=%s\n", n->netclient_name); 967 return; 968 } 969 970 dev = qdev_device_add_from_qdict(n->primary_opts, 971 n->primary_opts_from_json, 972 &err); 973 if (err) { 974 qobject_unref(n->primary_opts); 975 n->primary_opts = NULL; 976 } else { 977 object_unref(OBJECT(dev)); 978 } 979 error_propagate(errp, err); 980 } 981 982 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features) 983 { 984 VirtIONet *n = VIRTIO_NET(vdev); 985 Error *err = NULL; 986 int i; 987 988 if (n->mtu_bypass_backend && 989 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) { 990 features &= ~(1ULL << VIRTIO_NET_F_MTU); 991 } 992 993 virtio_net_set_multiqueue(n, 994 virtio_has_feature(features, VIRTIO_NET_F_RSS) || 995 virtio_has_feature(features, VIRTIO_NET_F_MQ)); 996 997 virtio_net_set_mrg_rx_bufs(n, 998 virtio_has_feature(features, 999 VIRTIO_NET_F_MRG_RXBUF), 1000 virtio_has_feature(features, 1001 VIRTIO_F_VERSION_1), 1002 virtio_has_feature(features, 1003 VIRTIO_NET_F_HASH_REPORT)); 1004 1005 n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) && 1006 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4); 1007 n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) && 1008 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6); 1009 n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS); 1010 1011 if (n->has_vnet_hdr) { 1012 n->curr_guest_offloads = 1013 virtio_net_guest_offloads_by_features(features); 1014 virtio_net_apply_guest_offloads(n); 1015 } 1016 1017 for (i = 0; i < n->max_queue_pairs; i++) { 1018 NetClientState *nc = qemu_get_subqueue(n->nic, i); 1019 1020 if (!get_vhost_net(nc->peer)) { 1021 continue; 1022 } 1023 vhost_net_ack_features(get_vhost_net(nc->peer), features); 1024 1025 /* 1026 * keep acked_features in NetVhostUserState up-to-date so it 1027 * can't miss any features configured by guest virtio driver. 1028 */ 1029 vhost_net_save_acked_features(nc->peer); 1030 } 1031 1032 if (virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) { 1033 memset(n->vlans, 0, MAX_VLAN >> 3); 1034 } else { 1035 memset(n->vlans, 0xff, MAX_VLAN >> 3); 1036 } 1037 1038 if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) { 1039 qapi_event_send_failover_negotiated(n->netclient_name); 1040 qatomic_set(&n->failover_primary_hidden, false); 1041 failover_add_primary(n, &err); 1042 if (err) { 1043 if (!qtest_enabled()) { 1044 warn_report_err(err); 1045 } else { 1046 error_free(err); 1047 } 1048 } 1049 } 1050 } 1051 1052 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 1053 struct iovec *iov, unsigned int iov_cnt) 1054 { 1055 uint8_t on; 1056 size_t s; 1057 NetClientState *nc = qemu_get_queue(n->nic); 1058 1059 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 1060 if (s != sizeof(on)) { 1061 return VIRTIO_NET_ERR; 1062 } 1063 1064 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 1065 n->promisc = on; 1066 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 1067 n->allmulti = on; 1068 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 1069 n->alluni = on; 1070 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 1071 n->nomulti = on; 1072 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 1073 n->nouni = on; 1074 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 1075 n->nobcast = on; 1076 } else { 1077 return VIRTIO_NET_ERR; 1078 } 1079 1080 rxfilter_notify(nc); 1081 1082 return VIRTIO_NET_OK; 1083 } 1084 1085 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 1086 struct iovec *iov, unsigned int iov_cnt) 1087 { 1088 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1089 uint64_t offloads; 1090 size_t s; 1091 1092 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 1093 return VIRTIO_NET_ERR; 1094 } 1095 1096 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 1097 if (s != sizeof(offloads)) { 1098 return VIRTIO_NET_ERR; 1099 } 1100 1101 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 1102 uint64_t supported_offloads; 1103 1104 offloads = virtio_ldq_p(vdev, &offloads); 1105 1106 if (!n->has_vnet_hdr) { 1107 return VIRTIO_NET_ERR; 1108 } 1109 1110 n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) && 1111 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4); 1112 n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) && 1113 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6); 1114 virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT); 1115 1116 supported_offloads = virtio_net_supported_guest_offloads(n); 1117 if (offloads & ~supported_offloads) { 1118 return VIRTIO_NET_ERR; 1119 } 1120 1121 n->curr_guest_offloads = offloads; 1122 virtio_net_apply_guest_offloads(n); 1123 1124 return VIRTIO_NET_OK; 1125 } else { 1126 return VIRTIO_NET_ERR; 1127 } 1128 } 1129 1130 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 1131 struct iovec *iov, unsigned int iov_cnt) 1132 { 1133 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1134 struct virtio_net_ctrl_mac mac_data; 1135 size_t s; 1136 NetClientState *nc = qemu_get_queue(n->nic); 1137 1138 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 1139 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 1140 return VIRTIO_NET_ERR; 1141 } 1142 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 1143 assert(s == sizeof(n->mac)); 1144 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 1145 rxfilter_notify(nc); 1146 1147 return VIRTIO_NET_OK; 1148 } 1149 1150 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 1151 return VIRTIO_NET_ERR; 1152 } 1153 1154 int in_use = 0; 1155 int first_multi = 0; 1156 uint8_t uni_overflow = 0; 1157 uint8_t multi_overflow = 0; 1158 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 1159 1160 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 1161 sizeof(mac_data.entries)); 1162 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 1163 if (s != sizeof(mac_data.entries)) { 1164 goto error; 1165 } 1166 iov_discard_front(&iov, &iov_cnt, s); 1167 1168 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 1169 goto error; 1170 } 1171 1172 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 1173 s = iov_to_buf(iov, iov_cnt, 0, macs, 1174 mac_data.entries * ETH_ALEN); 1175 if (s != mac_data.entries * ETH_ALEN) { 1176 goto error; 1177 } 1178 in_use += mac_data.entries; 1179 } else { 1180 uni_overflow = 1; 1181 } 1182 1183 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 1184 1185 first_multi = in_use; 1186 1187 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 1188 sizeof(mac_data.entries)); 1189 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 1190 if (s != sizeof(mac_data.entries)) { 1191 goto error; 1192 } 1193 1194 iov_discard_front(&iov, &iov_cnt, s); 1195 1196 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 1197 goto error; 1198 } 1199 1200 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) { 1201 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN], 1202 mac_data.entries * ETH_ALEN); 1203 if (s != mac_data.entries * ETH_ALEN) { 1204 goto error; 1205 } 1206 in_use += mac_data.entries; 1207 } else { 1208 multi_overflow = 1; 1209 } 1210 1211 n->mac_table.in_use = in_use; 1212 n->mac_table.first_multi = first_multi; 1213 n->mac_table.uni_overflow = uni_overflow; 1214 n->mac_table.multi_overflow = multi_overflow; 1215 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN); 1216 g_free(macs); 1217 rxfilter_notify(nc); 1218 1219 return VIRTIO_NET_OK; 1220 1221 error: 1222 g_free(macs); 1223 return VIRTIO_NET_ERR; 1224 } 1225 1226 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 1227 struct iovec *iov, unsigned int iov_cnt) 1228 { 1229 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1230 uint16_t vid; 1231 size_t s; 1232 NetClientState *nc = qemu_get_queue(n->nic); 1233 1234 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 1235 vid = virtio_lduw_p(vdev, &vid); 1236 if (s != sizeof(vid)) { 1237 return VIRTIO_NET_ERR; 1238 } 1239 1240 if (vid >= MAX_VLAN) 1241 return VIRTIO_NET_ERR; 1242 1243 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 1244 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 1245 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 1246 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 1247 else 1248 return VIRTIO_NET_ERR; 1249 1250 rxfilter_notify(nc); 1251 1252 return VIRTIO_NET_OK; 1253 } 1254 1255 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd, 1256 struct iovec *iov, unsigned int iov_cnt) 1257 { 1258 trace_virtio_net_handle_announce(n->announce_timer.round); 1259 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK && 1260 n->status & VIRTIO_NET_S_ANNOUNCE) { 1261 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 1262 if (n->announce_timer.round) { 1263 qemu_announce_timer_step(&n->announce_timer); 1264 } 1265 return VIRTIO_NET_OK; 1266 } else { 1267 return VIRTIO_NET_ERR; 1268 } 1269 } 1270 1271 static void virtio_net_detach_epbf_rss(VirtIONet *n); 1272 1273 static void virtio_net_disable_rss(VirtIONet *n) 1274 { 1275 if (n->rss_data.enabled) { 1276 trace_virtio_net_rss_disable(); 1277 } 1278 n->rss_data.enabled = false; 1279 1280 virtio_net_detach_epbf_rss(n); 1281 } 1282 1283 static bool virtio_net_attach_ebpf_to_backend(NICState *nic, int prog_fd) 1284 { 1285 NetClientState *nc = qemu_get_peer(qemu_get_queue(nic), 0); 1286 if (nc == NULL || nc->info->set_steering_ebpf == NULL) { 1287 return false; 1288 } 1289 1290 return nc->info->set_steering_ebpf(nc, prog_fd); 1291 } 1292 1293 static void rss_data_to_rss_config(struct VirtioNetRssData *data, 1294 struct EBPFRSSConfig *config) 1295 { 1296 config->redirect = data->redirect; 1297 config->populate_hash = data->populate_hash; 1298 config->hash_types = data->hash_types; 1299 config->indirections_len = data->indirections_len; 1300 config->default_queue = data->default_queue; 1301 } 1302 1303 static bool virtio_net_attach_epbf_rss(VirtIONet *n) 1304 { 1305 struct EBPFRSSConfig config = {}; 1306 1307 if (!ebpf_rss_is_loaded(&n->ebpf_rss)) { 1308 return false; 1309 } 1310 1311 rss_data_to_rss_config(&n->rss_data, &config); 1312 1313 if (!ebpf_rss_set_all(&n->ebpf_rss, &config, 1314 n->rss_data.indirections_table, n->rss_data.key)) { 1315 return false; 1316 } 1317 1318 if (!virtio_net_attach_ebpf_to_backend(n->nic, n->ebpf_rss.program_fd)) { 1319 return false; 1320 } 1321 1322 return true; 1323 } 1324 1325 static void virtio_net_detach_epbf_rss(VirtIONet *n) 1326 { 1327 virtio_net_attach_ebpf_to_backend(n->nic, -1); 1328 } 1329 1330 static bool virtio_net_load_ebpf(VirtIONet *n) 1331 { 1332 if (!virtio_net_attach_ebpf_to_backend(n->nic, -1)) { 1333 /* backend doesn't support steering ebpf */ 1334 return false; 1335 } 1336 1337 return ebpf_rss_load(&n->ebpf_rss); 1338 } 1339 1340 static void virtio_net_unload_ebpf(VirtIONet *n) 1341 { 1342 virtio_net_attach_ebpf_to_backend(n->nic, -1); 1343 ebpf_rss_unload(&n->ebpf_rss); 1344 } 1345 1346 static uint16_t virtio_net_handle_rss(VirtIONet *n, 1347 struct iovec *iov, 1348 unsigned int iov_cnt, 1349 bool do_rss) 1350 { 1351 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1352 struct virtio_net_rss_config cfg; 1353 size_t s, offset = 0, size_get; 1354 uint16_t queue_pairs, i; 1355 struct { 1356 uint16_t us; 1357 uint8_t b; 1358 } QEMU_PACKED temp; 1359 const char *err_msg = ""; 1360 uint32_t err_value = 0; 1361 1362 if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) { 1363 err_msg = "RSS is not negotiated"; 1364 goto error; 1365 } 1366 if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) { 1367 err_msg = "Hash report is not negotiated"; 1368 goto error; 1369 } 1370 size_get = offsetof(struct virtio_net_rss_config, indirection_table); 1371 s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get); 1372 if (s != size_get) { 1373 err_msg = "Short command buffer"; 1374 err_value = (uint32_t)s; 1375 goto error; 1376 } 1377 n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types); 1378 n->rss_data.indirections_len = 1379 virtio_lduw_p(vdev, &cfg.indirection_table_mask); 1380 n->rss_data.indirections_len++; 1381 if (!do_rss) { 1382 n->rss_data.indirections_len = 1; 1383 } 1384 if (!is_power_of_2(n->rss_data.indirections_len)) { 1385 err_msg = "Invalid size of indirection table"; 1386 err_value = n->rss_data.indirections_len; 1387 goto error; 1388 } 1389 if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) { 1390 err_msg = "Too large indirection table"; 1391 err_value = n->rss_data.indirections_len; 1392 goto error; 1393 } 1394 n->rss_data.default_queue = do_rss ? 1395 virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0; 1396 if (n->rss_data.default_queue >= n->max_queue_pairs) { 1397 err_msg = "Invalid default queue"; 1398 err_value = n->rss_data.default_queue; 1399 goto error; 1400 } 1401 offset += size_get; 1402 size_get = sizeof(uint16_t) * n->rss_data.indirections_len; 1403 g_free(n->rss_data.indirections_table); 1404 n->rss_data.indirections_table = g_malloc(size_get); 1405 if (!n->rss_data.indirections_table) { 1406 err_msg = "Can't allocate indirections table"; 1407 err_value = n->rss_data.indirections_len; 1408 goto error; 1409 } 1410 s = iov_to_buf(iov, iov_cnt, offset, 1411 n->rss_data.indirections_table, size_get); 1412 if (s != size_get) { 1413 err_msg = "Short indirection table buffer"; 1414 err_value = (uint32_t)s; 1415 goto error; 1416 } 1417 for (i = 0; i < n->rss_data.indirections_len; ++i) { 1418 uint16_t val = n->rss_data.indirections_table[i]; 1419 n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val); 1420 } 1421 offset += size_get; 1422 size_get = sizeof(temp); 1423 s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get); 1424 if (s != size_get) { 1425 err_msg = "Can't get queue_pairs"; 1426 err_value = (uint32_t)s; 1427 goto error; 1428 } 1429 queue_pairs = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queue_pairs; 1430 if (queue_pairs == 0 || queue_pairs > n->max_queue_pairs) { 1431 err_msg = "Invalid number of queue_pairs"; 1432 err_value = queue_pairs; 1433 goto error; 1434 } 1435 if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) { 1436 err_msg = "Invalid key size"; 1437 err_value = temp.b; 1438 goto error; 1439 } 1440 if (!temp.b && n->rss_data.hash_types) { 1441 err_msg = "No key provided"; 1442 err_value = 0; 1443 goto error; 1444 } 1445 if (!temp.b && !n->rss_data.hash_types) { 1446 virtio_net_disable_rss(n); 1447 return queue_pairs; 1448 } 1449 offset += size_get; 1450 size_get = temp.b; 1451 s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get); 1452 if (s != size_get) { 1453 err_msg = "Can get key buffer"; 1454 err_value = (uint32_t)s; 1455 goto error; 1456 } 1457 n->rss_data.enabled = true; 1458 1459 if (!n->rss_data.populate_hash) { 1460 if (!virtio_net_attach_epbf_rss(n)) { 1461 /* EBPF must be loaded for vhost */ 1462 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) { 1463 warn_report("Can't load eBPF RSS for vhost"); 1464 goto error; 1465 } 1466 /* fallback to software RSS */ 1467 warn_report("Can't load eBPF RSS - fallback to software RSS"); 1468 n->rss_data.enabled_software_rss = true; 1469 } 1470 } else { 1471 /* use software RSS for hash populating */ 1472 /* and detach eBPF if was loaded before */ 1473 virtio_net_detach_epbf_rss(n); 1474 n->rss_data.enabled_software_rss = true; 1475 } 1476 1477 trace_virtio_net_rss_enable(n->rss_data.hash_types, 1478 n->rss_data.indirections_len, 1479 temp.b); 1480 return queue_pairs; 1481 error: 1482 trace_virtio_net_rss_error(err_msg, err_value); 1483 virtio_net_disable_rss(n); 1484 return 0; 1485 } 1486 1487 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 1488 struct iovec *iov, unsigned int iov_cnt) 1489 { 1490 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1491 uint16_t queue_pairs; 1492 NetClientState *nc = qemu_get_queue(n->nic); 1493 1494 virtio_net_disable_rss(n); 1495 if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) { 1496 queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, false); 1497 return queue_pairs ? VIRTIO_NET_OK : VIRTIO_NET_ERR; 1498 } 1499 if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) { 1500 queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, true); 1501 } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 1502 struct virtio_net_ctrl_mq mq; 1503 size_t s; 1504 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) { 1505 return VIRTIO_NET_ERR; 1506 } 1507 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 1508 if (s != sizeof(mq)) { 1509 return VIRTIO_NET_ERR; 1510 } 1511 queue_pairs = virtio_lduw_p(vdev, &mq.virtqueue_pairs); 1512 1513 } else { 1514 return VIRTIO_NET_ERR; 1515 } 1516 1517 if (queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 1518 queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 1519 queue_pairs > n->max_queue_pairs || 1520 !n->multiqueue) { 1521 return VIRTIO_NET_ERR; 1522 } 1523 1524 n->curr_queue_pairs = queue_pairs; 1525 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 1526 /* 1527 * Avoid updating the backend for a vdpa device: We're only interested 1528 * in updating the device model queues. 1529 */ 1530 return VIRTIO_NET_OK; 1531 } 1532 /* stop the backend before changing the number of queue_pairs to avoid handling a 1533 * disabled queue */ 1534 virtio_net_set_status(vdev, vdev->status); 1535 virtio_net_set_queue_pairs(n); 1536 1537 return VIRTIO_NET_OK; 1538 } 1539 1540 size_t virtio_net_handle_ctrl_iov(VirtIODevice *vdev, 1541 const struct iovec *in_sg, unsigned in_num, 1542 const struct iovec *out_sg, 1543 unsigned out_num) 1544 { 1545 VirtIONet *n = VIRTIO_NET(vdev); 1546 struct virtio_net_ctrl_hdr ctrl; 1547 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 1548 size_t s; 1549 struct iovec *iov, *iov2; 1550 1551 if (iov_size(in_sg, in_num) < sizeof(status) || 1552 iov_size(out_sg, out_num) < sizeof(ctrl)) { 1553 virtio_error(vdev, "virtio-net ctrl missing headers"); 1554 return 0; 1555 } 1556 1557 iov2 = iov = g_memdup2(out_sg, sizeof(struct iovec) * out_num); 1558 s = iov_to_buf(iov, out_num, 0, &ctrl, sizeof(ctrl)); 1559 iov_discard_front(&iov, &out_num, sizeof(ctrl)); 1560 if (s != sizeof(ctrl)) { 1561 status = VIRTIO_NET_ERR; 1562 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 1563 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, out_num); 1564 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 1565 status = virtio_net_handle_mac(n, ctrl.cmd, iov, out_num); 1566 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 1567 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, out_num); 1568 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) { 1569 status = virtio_net_handle_announce(n, ctrl.cmd, iov, out_num); 1570 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 1571 status = virtio_net_handle_mq(n, ctrl.cmd, iov, out_num); 1572 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 1573 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, out_num); 1574 } 1575 1576 s = iov_from_buf(in_sg, in_num, 0, &status, sizeof(status)); 1577 assert(s == sizeof(status)); 1578 1579 g_free(iov2); 1580 return sizeof(status); 1581 } 1582 1583 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 1584 { 1585 VirtQueueElement *elem; 1586 1587 for (;;) { 1588 size_t written; 1589 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 1590 if (!elem) { 1591 break; 1592 } 1593 1594 written = virtio_net_handle_ctrl_iov(vdev, elem->in_sg, elem->in_num, 1595 elem->out_sg, elem->out_num); 1596 if (written > 0) { 1597 virtqueue_push(vq, elem, written); 1598 virtio_notify(vdev, vq); 1599 g_free(elem); 1600 } else { 1601 virtqueue_detach_element(vq, elem, 0); 1602 g_free(elem); 1603 break; 1604 } 1605 } 1606 } 1607 1608 /* RX */ 1609 1610 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 1611 { 1612 VirtIONet *n = VIRTIO_NET(vdev); 1613 int queue_index = vq2q(virtio_get_queue_index(vq)); 1614 1615 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 1616 } 1617 1618 static bool virtio_net_can_receive(NetClientState *nc) 1619 { 1620 VirtIONet *n = qemu_get_nic_opaque(nc); 1621 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1622 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1623 1624 if (!vdev->vm_running) { 1625 return false; 1626 } 1627 1628 if (nc->queue_index >= n->curr_queue_pairs) { 1629 return false; 1630 } 1631 1632 if (!virtio_queue_ready(q->rx_vq) || 1633 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1634 return false; 1635 } 1636 1637 return true; 1638 } 1639 1640 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 1641 { 1642 VirtIONet *n = q->n; 1643 if (virtio_queue_empty(q->rx_vq) || 1644 (n->mergeable_rx_bufs && 1645 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 1646 virtio_queue_set_notification(q->rx_vq, 1); 1647 1648 /* To avoid a race condition where the guest has made some buffers 1649 * available after the above check but before notification was 1650 * enabled, check for available buffers again. 1651 */ 1652 if (virtio_queue_empty(q->rx_vq) || 1653 (n->mergeable_rx_bufs && 1654 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 1655 return 0; 1656 } 1657 } 1658 1659 virtio_queue_set_notification(q->rx_vq, 0); 1660 return 1; 1661 } 1662 1663 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr) 1664 { 1665 virtio_tswap16s(vdev, &hdr->hdr_len); 1666 virtio_tswap16s(vdev, &hdr->gso_size); 1667 virtio_tswap16s(vdev, &hdr->csum_start); 1668 virtio_tswap16s(vdev, &hdr->csum_offset); 1669 } 1670 1671 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 1672 * it never finds out that the packets don't have valid checksums. This 1673 * causes dhclient to get upset. Fedora's carried a patch for ages to 1674 * fix this with Xen but it hasn't appeared in an upstream release of 1675 * dhclient yet. 1676 * 1677 * To avoid breaking existing guests, we catch udp packets and add 1678 * checksums. This is terrible but it's better than hacking the guest 1679 * kernels. 1680 * 1681 * N.B. if we introduce a zero-copy API, this operation is no longer free so 1682 * we should provide a mechanism to disable it to avoid polluting the host 1683 * cache. 1684 */ 1685 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 1686 uint8_t *buf, size_t size) 1687 { 1688 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 1689 (size > 27 && size < 1500) && /* normal sized MTU */ 1690 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 1691 (buf[23] == 17) && /* ip.protocol == UDP */ 1692 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 1693 net_checksum_calculate(buf, size, CSUM_UDP); 1694 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 1695 } 1696 } 1697 1698 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt, 1699 const void *buf, size_t size) 1700 { 1701 if (n->has_vnet_hdr) { 1702 /* FIXME this cast is evil */ 1703 void *wbuf = (void *)buf; 1704 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, 1705 size - n->host_hdr_len); 1706 1707 if (n->needs_vnet_hdr_swap) { 1708 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf); 1709 } 1710 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr)); 1711 } else { 1712 struct virtio_net_hdr hdr = { 1713 .flags = 0, 1714 .gso_type = VIRTIO_NET_HDR_GSO_NONE 1715 }; 1716 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr); 1717 } 1718 } 1719 1720 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 1721 { 1722 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 1723 static const uint8_t vlan[] = {0x81, 0x00}; 1724 uint8_t *ptr = (uint8_t *)buf; 1725 int i; 1726 1727 if (n->promisc) 1728 return 1; 1729 1730 ptr += n->host_hdr_len; 1731 1732 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 1733 int vid = lduw_be_p(ptr + 14) & 0xfff; 1734 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 1735 return 0; 1736 } 1737 1738 if (ptr[0] & 1) { // multicast 1739 if (!memcmp(ptr, bcast, sizeof(bcast))) { 1740 return !n->nobcast; 1741 } else if (n->nomulti) { 1742 return 0; 1743 } else if (n->allmulti || n->mac_table.multi_overflow) { 1744 return 1; 1745 } 1746 1747 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 1748 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1749 return 1; 1750 } 1751 } 1752 } else { // unicast 1753 if (n->nouni) { 1754 return 0; 1755 } else if (n->alluni || n->mac_table.uni_overflow) { 1756 return 1; 1757 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 1758 return 1; 1759 } 1760 1761 for (i = 0; i < n->mac_table.first_multi; i++) { 1762 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1763 return 1; 1764 } 1765 } 1766 } 1767 1768 return 0; 1769 } 1770 1771 static uint8_t virtio_net_get_hash_type(bool hasip4, 1772 bool hasip6, 1773 EthL4HdrProto l4hdr_proto, 1774 uint32_t types) 1775 { 1776 if (hasip4) { 1777 switch (l4hdr_proto) { 1778 case ETH_L4_HDR_PROTO_TCP: 1779 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) { 1780 return NetPktRssIpV4Tcp; 1781 } 1782 break; 1783 1784 case ETH_L4_HDR_PROTO_UDP: 1785 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) { 1786 return NetPktRssIpV4Udp; 1787 } 1788 break; 1789 1790 default: 1791 break; 1792 } 1793 1794 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) { 1795 return NetPktRssIpV4; 1796 } 1797 } else if (hasip6) { 1798 switch (l4hdr_proto) { 1799 case ETH_L4_HDR_PROTO_TCP: 1800 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) { 1801 return NetPktRssIpV6TcpEx; 1802 } 1803 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) { 1804 return NetPktRssIpV6Tcp; 1805 } 1806 break; 1807 1808 case ETH_L4_HDR_PROTO_UDP: 1809 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) { 1810 return NetPktRssIpV6UdpEx; 1811 } 1812 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) { 1813 return NetPktRssIpV6Udp; 1814 } 1815 break; 1816 1817 default: 1818 break; 1819 } 1820 1821 if (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) { 1822 return NetPktRssIpV6Ex; 1823 } 1824 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv6) { 1825 return NetPktRssIpV6; 1826 } 1827 } 1828 return 0xff; 1829 } 1830 1831 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report, 1832 uint32_t hash) 1833 { 1834 struct virtio_net_hdr_v1_hash *hdr = (void *)buf; 1835 hdr->hash_value = hash; 1836 hdr->hash_report = report; 1837 } 1838 1839 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf, 1840 size_t size) 1841 { 1842 VirtIONet *n = qemu_get_nic_opaque(nc); 1843 unsigned int index = nc->queue_index, new_index = index; 1844 struct NetRxPkt *pkt = n->rx_pkt; 1845 uint8_t net_hash_type; 1846 uint32_t hash; 1847 bool hasip4, hasip6; 1848 EthL4HdrProto l4hdr_proto; 1849 static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = { 1850 VIRTIO_NET_HASH_REPORT_IPv4, 1851 VIRTIO_NET_HASH_REPORT_TCPv4, 1852 VIRTIO_NET_HASH_REPORT_TCPv6, 1853 VIRTIO_NET_HASH_REPORT_IPv6, 1854 VIRTIO_NET_HASH_REPORT_IPv6_EX, 1855 VIRTIO_NET_HASH_REPORT_TCPv6_EX, 1856 VIRTIO_NET_HASH_REPORT_UDPv4, 1857 VIRTIO_NET_HASH_REPORT_UDPv6, 1858 VIRTIO_NET_HASH_REPORT_UDPv6_EX 1859 }; 1860 struct iovec iov = { 1861 .iov_base = (void *)buf, 1862 .iov_len = size 1863 }; 1864 1865 net_rx_pkt_set_protocols(pkt, &iov, 1, n->host_hdr_len); 1866 net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto); 1867 net_hash_type = virtio_net_get_hash_type(hasip4, hasip6, l4hdr_proto, 1868 n->rss_data.hash_types); 1869 if (net_hash_type > NetPktRssIpV6UdpEx) { 1870 if (n->rss_data.populate_hash) { 1871 virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0); 1872 } 1873 return n->rss_data.redirect ? n->rss_data.default_queue : -1; 1874 } 1875 1876 hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key); 1877 1878 if (n->rss_data.populate_hash) { 1879 virtio_set_packet_hash(buf, reports[net_hash_type], hash); 1880 } 1881 1882 if (n->rss_data.redirect) { 1883 new_index = hash & (n->rss_data.indirections_len - 1); 1884 new_index = n->rss_data.indirections_table[new_index]; 1885 } 1886 1887 return (index == new_index) ? -1 : new_index; 1888 } 1889 1890 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, 1891 size_t size, bool no_rss) 1892 { 1893 VirtIONet *n = qemu_get_nic_opaque(nc); 1894 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1895 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1896 VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE]; 1897 size_t lens[VIRTQUEUE_MAX_SIZE]; 1898 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 1899 struct virtio_net_hdr_mrg_rxbuf mhdr; 1900 unsigned mhdr_cnt = 0; 1901 size_t offset, i, guest_offset, j; 1902 ssize_t err; 1903 1904 if (!virtio_net_can_receive(nc)) { 1905 return -1; 1906 } 1907 1908 if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) { 1909 int index = virtio_net_process_rss(nc, buf, size); 1910 if (index >= 0) { 1911 NetClientState *nc2 = qemu_get_subqueue(n->nic, index); 1912 return virtio_net_receive_rcu(nc2, buf, size, true); 1913 } 1914 } 1915 1916 /* hdr_len refers to the header we supply to the guest */ 1917 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 1918 return 0; 1919 } 1920 1921 if (!receive_filter(n, buf, size)) 1922 return size; 1923 1924 offset = i = 0; 1925 1926 while (offset < size) { 1927 VirtQueueElement *elem; 1928 int len, total; 1929 const struct iovec *sg; 1930 1931 total = 0; 1932 1933 if (i == VIRTQUEUE_MAX_SIZE) { 1934 virtio_error(vdev, "virtio-net unexpected long buffer chain"); 1935 err = size; 1936 goto err; 1937 } 1938 1939 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement)); 1940 if (!elem) { 1941 if (i) { 1942 virtio_error(vdev, "virtio-net unexpected empty queue: " 1943 "i %zd mergeable %d offset %zd, size %zd, " 1944 "guest hdr len %zd, host hdr len %zd " 1945 "guest features 0x%" PRIx64, 1946 i, n->mergeable_rx_bufs, offset, size, 1947 n->guest_hdr_len, n->host_hdr_len, 1948 vdev->guest_features); 1949 } 1950 err = -1; 1951 goto err; 1952 } 1953 1954 if (elem->in_num < 1) { 1955 virtio_error(vdev, 1956 "virtio-net receive queue contains no in buffers"); 1957 virtqueue_detach_element(q->rx_vq, elem, 0); 1958 g_free(elem); 1959 err = -1; 1960 goto err; 1961 } 1962 1963 sg = elem->in_sg; 1964 if (i == 0) { 1965 assert(offset == 0); 1966 if (n->mergeable_rx_bufs) { 1967 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 1968 sg, elem->in_num, 1969 offsetof(typeof(mhdr), num_buffers), 1970 sizeof(mhdr.num_buffers)); 1971 } 1972 1973 receive_header(n, sg, elem->in_num, buf, size); 1974 if (n->rss_data.populate_hash) { 1975 offset = sizeof(mhdr); 1976 iov_from_buf(sg, elem->in_num, offset, 1977 buf + offset, n->host_hdr_len - sizeof(mhdr)); 1978 } 1979 offset = n->host_hdr_len; 1980 total += n->guest_hdr_len; 1981 guest_offset = n->guest_hdr_len; 1982 } else { 1983 guest_offset = 0; 1984 } 1985 1986 /* copy in packet. ugh */ 1987 len = iov_from_buf(sg, elem->in_num, guest_offset, 1988 buf + offset, size - offset); 1989 total += len; 1990 offset += len; 1991 /* If buffers can't be merged, at this point we 1992 * must have consumed the complete packet. 1993 * Otherwise, drop it. */ 1994 if (!n->mergeable_rx_bufs && offset < size) { 1995 virtqueue_unpop(q->rx_vq, elem, total); 1996 g_free(elem); 1997 err = size; 1998 goto err; 1999 } 2000 2001 elems[i] = elem; 2002 lens[i] = total; 2003 i++; 2004 } 2005 2006 if (mhdr_cnt) { 2007 virtio_stw_p(vdev, &mhdr.num_buffers, i); 2008 iov_from_buf(mhdr_sg, mhdr_cnt, 2009 0, 2010 &mhdr.num_buffers, sizeof mhdr.num_buffers); 2011 } 2012 2013 for (j = 0; j < i; j++) { 2014 /* signal other side */ 2015 virtqueue_fill(q->rx_vq, elems[j], lens[j], j); 2016 g_free(elems[j]); 2017 } 2018 2019 virtqueue_flush(q->rx_vq, i); 2020 virtio_notify(vdev, q->rx_vq); 2021 2022 return size; 2023 2024 err: 2025 for (j = 0; j < i; j++) { 2026 virtqueue_detach_element(q->rx_vq, elems[j], lens[j]); 2027 g_free(elems[j]); 2028 } 2029 2030 return err; 2031 } 2032 2033 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf, 2034 size_t size) 2035 { 2036 RCU_READ_LOCK_GUARD(); 2037 2038 return virtio_net_receive_rcu(nc, buf, size, false); 2039 } 2040 2041 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain, 2042 const uint8_t *buf, 2043 VirtioNetRscUnit *unit) 2044 { 2045 uint16_t ip_hdrlen; 2046 struct ip_header *ip; 2047 2048 ip = (struct ip_header *)(buf + chain->n->guest_hdr_len 2049 + sizeof(struct eth_header)); 2050 unit->ip = (void *)ip; 2051 ip_hdrlen = (ip->ip_ver_len & 0xF) << 2; 2052 unit->ip_plen = &ip->ip_len; 2053 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen); 2054 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10; 2055 unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen; 2056 } 2057 2058 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain, 2059 const uint8_t *buf, 2060 VirtioNetRscUnit *unit) 2061 { 2062 struct ip6_header *ip6; 2063 2064 ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len 2065 + sizeof(struct eth_header)); 2066 unit->ip = ip6; 2067 unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen); 2068 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) 2069 + sizeof(struct ip6_header)); 2070 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10; 2071 2072 /* There is a difference between payload length in ipv4 and v6, 2073 ip header is excluded in ipv6 */ 2074 unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen; 2075 } 2076 2077 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain, 2078 VirtioNetRscSeg *seg) 2079 { 2080 int ret; 2081 struct virtio_net_hdr_v1 *h; 2082 2083 h = (struct virtio_net_hdr_v1 *)seg->buf; 2084 h->flags = 0; 2085 h->gso_type = VIRTIO_NET_HDR_GSO_NONE; 2086 2087 if (seg->is_coalesced) { 2088 h->rsc.segments = seg->packets; 2089 h->rsc.dup_acks = seg->dup_ack; 2090 h->flags = VIRTIO_NET_HDR_F_RSC_INFO; 2091 if (chain->proto == ETH_P_IP) { 2092 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2093 } else { 2094 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2095 } 2096 } 2097 2098 ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size); 2099 QTAILQ_REMOVE(&chain->buffers, seg, next); 2100 g_free(seg->buf); 2101 g_free(seg); 2102 2103 return ret; 2104 } 2105 2106 static void virtio_net_rsc_purge(void *opq) 2107 { 2108 VirtioNetRscSeg *seg, *rn; 2109 VirtioNetRscChain *chain = (VirtioNetRscChain *)opq; 2110 2111 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) { 2112 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2113 chain->stat.purge_failed++; 2114 continue; 2115 } 2116 } 2117 2118 chain->stat.timer++; 2119 if (!QTAILQ_EMPTY(&chain->buffers)) { 2120 timer_mod(chain->drain_timer, 2121 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout); 2122 } 2123 } 2124 2125 static void virtio_net_rsc_cleanup(VirtIONet *n) 2126 { 2127 VirtioNetRscChain *chain, *rn_chain; 2128 VirtioNetRscSeg *seg, *rn_seg; 2129 2130 QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) { 2131 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) { 2132 QTAILQ_REMOVE(&chain->buffers, seg, next); 2133 g_free(seg->buf); 2134 g_free(seg); 2135 } 2136 2137 timer_free(chain->drain_timer); 2138 QTAILQ_REMOVE(&n->rsc_chains, chain, next); 2139 g_free(chain); 2140 } 2141 } 2142 2143 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain, 2144 NetClientState *nc, 2145 const uint8_t *buf, size_t size) 2146 { 2147 uint16_t hdr_len; 2148 VirtioNetRscSeg *seg; 2149 2150 hdr_len = chain->n->guest_hdr_len; 2151 seg = g_new(VirtioNetRscSeg, 1); 2152 seg->buf = g_malloc(hdr_len + sizeof(struct eth_header) 2153 + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD); 2154 memcpy(seg->buf, buf, size); 2155 seg->size = size; 2156 seg->packets = 1; 2157 seg->dup_ack = 0; 2158 seg->is_coalesced = 0; 2159 seg->nc = nc; 2160 2161 QTAILQ_INSERT_TAIL(&chain->buffers, seg, next); 2162 chain->stat.cache++; 2163 2164 switch (chain->proto) { 2165 case ETH_P_IP: 2166 virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit); 2167 break; 2168 case ETH_P_IPV6: 2169 virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit); 2170 break; 2171 default: 2172 g_assert_not_reached(); 2173 } 2174 } 2175 2176 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain, 2177 VirtioNetRscSeg *seg, 2178 const uint8_t *buf, 2179 struct tcp_header *n_tcp, 2180 struct tcp_header *o_tcp) 2181 { 2182 uint32_t nack, oack; 2183 uint16_t nwin, owin; 2184 2185 nack = htonl(n_tcp->th_ack); 2186 nwin = htons(n_tcp->th_win); 2187 oack = htonl(o_tcp->th_ack); 2188 owin = htons(o_tcp->th_win); 2189 2190 if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) { 2191 chain->stat.ack_out_of_win++; 2192 return RSC_FINAL; 2193 } else if (nack == oack) { 2194 /* duplicated ack or window probe */ 2195 if (nwin == owin) { 2196 /* duplicated ack, add dup ack count due to whql test up to 1 */ 2197 chain->stat.dup_ack++; 2198 return RSC_FINAL; 2199 } else { 2200 /* Coalesce window update */ 2201 o_tcp->th_win = n_tcp->th_win; 2202 chain->stat.win_update++; 2203 return RSC_COALESCE; 2204 } 2205 } else { 2206 /* pure ack, go to 'C', finalize*/ 2207 chain->stat.pure_ack++; 2208 return RSC_FINAL; 2209 } 2210 } 2211 2212 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain, 2213 VirtioNetRscSeg *seg, 2214 const uint8_t *buf, 2215 VirtioNetRscUnit *n_unit) 2216 { 2217 void *data; 2218 uint16_t o_ip_len; 2219 uint32_t nseq, oseq; 2220 VirtioNetRscUnit *o_unit; 2221 2222 o_unit = &seg->unit; 2223 o_ip_len = htons(*o_unit->ip_plen); 2224 nseq = htonl(n_unit->tcp->th_seq); 2225 oseq = htonl(o_unit->tcp->th_seq); 2226 2227 /* out of order or retransmitted. */ 2228 if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) { 2229 chain->stat.data_out_of_win++; 2230 return RSC_FINAL; 2231 } 2232 2233 data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen; 2234 if (nseq == oseq) { 2235 if ((o_unit->payload == 0) && n_unit->payload) { 2236 /* From no payload to payload, normal case, not a dup ack or etc */ 2237 chain->stat.data_after_pure_ack++; 2238 goto coalesce; 2239 } else { 2240 return virtio_net_rsc_handle_ack(chain, seg, buf, 2241 n_unit->tcp, o_unit->tcp); 2242 } 2243 } else if ((nseq - oseq) != o_unit->payload) { 2244 /* Not a consistent packet, out of order */ 2245 chain->stat.data_out_of_order++; 2246 return RSC_FINAL; 2247 } else { 2248 coalesce: 2249 if ((o_ip_len + n_unit->payload) > chain->max_payload) { 2250 chain->stat.over_size++; 2251 return RSC_FINAL; 2252 } 2253 2254 /* Here comes the right data, the payload length in v4/v6 is different, 2255 so use the field value to update and record the new data len */ 2256 o_unit->payload += n_unit->payload; /* update new data len */ 2257 2258 /* update field in ip header */ 2259 *o_unit->ip_plen = htons(o_ip_len + n_unit->payload); 2260 2261 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced 2262 for windows guest, while this may change the behavior for linux 2263 guest (only if it uses RSC feature). */ 2264 o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags; 2265 2266 o_unit->tcp->th_ack = n_unit->tcp->th_ack; 2267 o_unit->tcp->th_win = n_unit->tcp->th_win; 2268 2269 memmove(seg->buf + seg->size, data, n_unit->payload); 2270 seg->size += n_unit->payload; 2271 seg->packets++; 2272 chain->stat.coalesced++; 2273 return RSC_COALESCE; 2274 } 2275 } 2276 2277 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain, 2278 VirtioNetRscSeg *seg, 2279 const uint8_t *buf, size_t size, 2280 VirtioNetRscUnit *unit) 2281 { 2282 struct ip_header *ip1, *ip2; 2283 2284 ip1 = (struct ip_header *)(unit->ip); 2285 ip2 = (struct ip_header *)(seg->unit.ip); 2286 if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst) 2287 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport) 2288 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) { 2289 chain->stat.no_match++; 2290 return RSC_NO_MATCH; 2291 } 2292 2293 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit); 2294 } 2295 2296 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain, 2297 VirtioNetRscSeg *seg, 2298 const uint8_t *buf, size_t size, 2299 VirtioNetRscUnit *unit) 2300 { 2301 struct ip6_header *ip1, *ip2; 2302 2303 ip1 = (struct ip6_header *)(unit->ip); 2304 ip2 = (struct ip6_header *)(seg->unit.ip); 2305 if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address)) 2306 || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address)) 2307 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport) 2308 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) { 2309 chain->stat.no_match++; 2310 return RSC_NO_MATCH; 2311 } 2312 2313 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit); 2314 } 2315 2316 /* Packets with 'SYN' should bypass, other flag should be sent after drain 2317 * to prevent out of order */ 2318 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain, 2319 struct tcp_header *tcp) 2320 { 2321 uint16_t tcp_hdr; 2322 uint16_t tcp_flag; 2323 2324 tcp_flag = htons(tcp->th_offset_flags); 2325 tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10; 2326 tcp_flag &= VIRTIO_NET_TCP_FLAG; 2327 if (tcp_flag & TH_SYN) { 2328 chain->stat.tcp_syn++; 2329 return RSC_BYPASS; 2330 } 2331 2332 if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) { 2333 chain->stat.tcp_ctrl_drain++; 2334 return RSC_FINAL; 2335 } 2336 2337 if (tcp_hdr > sizeof(struct tcp_header)) { 2338 chain->stat.tcp_all_opt++; 2339 return RSC_FINAL; 2340 } 2341 2342 return RSC_CANDIDATE; 2343 } 2344 2345 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain, 2346 NetClientState *nc, 2347 const uint8_t *buf, size_t size, 2348 VirtioNetRscUnit *unit) 2349 { 2350 int ret; 2351 VirtioNetRscSeg *seg, *nseg; 2352 2353 if (QTAILQ_EMPTY(&chain->buffers)) { 2354 chain->stat.empty_cache++; 2355 virtio_net_rsc_cache_buf(chain, nc, buf, size); 2356 timer_mod(chain->drain_timer, 2357 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout); 2358 return size; 2359 } 2360 2361 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) { 2362 if (chain->proto == ETH_P_IP) { 2363 ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit); 2364 } else { 2365 ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit); 2366 } 2367 2368 if (ret == RSC_FINAL) { 2369 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2370 /* Send failed */ 2371 chain->stat.final_failed++; 2372 return 0; 2373 } 2374 2375 /* Send current packet */ 2376 return virtio_net_do_receive(nc, buf, size); 2377 } else if (ret == RSC_NO_MATCH) { 2378 continue; 2379 } else { 2380 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */ 2381 seg->is_coalesced = 1; 2382 return size; 2383 } 2384 } 2385 2386 chain->stat.no_match_cache++; 2387 virtio_net_rsc_cache_buf(chain, nc, buf, size); 2388 return size; 2389 } 2390 2391 /* Drain a connection data, this is to avoid out of order segments */ 2392 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain, 2393 NetClientState *nc, 2394 const uint8_t *buf, size_t size, 2395 uint16_t ip_start, uint16_t ip_size, 2396 uint16_t tcp_port) 2397 { 2398 VirtioNetRscSeg *seg, *nseg; 2399 uint32_t ppair1, ppair2; 2400 2401 ppair1 = *(uint32_t *)(buf + tcp_port); 2402 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) { 2403 ppair2 = *(uint32_t *)(seg->buf + tcp_port); 2404 if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size) 2405 || (ppair1 != ppair2)) { 2406 continue; 2407 } 2408 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2409 chain->stat.drain_failed++; 2410 } 2411 2412 break; 2413 } 2414 2415 return virtio_net_do_receive(nc, buf, size); 2416 } 2417 2418 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain, 2419 struct ip_header *ip, 2420 const uint8_t *buf, size_t size) 2421 { 2422 uint16_t ip_len; 2423 2424 /* Not an ipv4 packet */ 2425 if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) { 2426 chain->stat.ip_option++; 2427 return RSC_BYPASS; 2428 } 2429 2430 /* Don't handle packets with ip option */ 2431 if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) { 2432 chain->stat.ip_option++; 2433 return RSC_BYPASS; 2434 } 2435 2436 if (ip->ip_p != IPPROTO_TCP) { 2437 chain->stat.bypass_not_tcp++; 2438 return RSC_BYPASS; 2439 } 2440 2441 /* Don't handle packets with ip fragment */ 2442 if (!(htons(ip->ip_off) & IP_DF)) { 2443 chain->stat.ip_frag++; 2444 return RSC_BYPASS; 2445 } 2446 2447 /* Don't handle packets with ecn flag */ 2448 if (IPTOS_ECN(ip->ip_tos)) { 2449 chain->stat.ip_ecn++; 2450 return RSC_BYPASS; 2451 } 2452 2453 ip_len = htons(ip->ip_len); 2454 if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header)) 2455 || ip_len > (size - chain->n->guest_hdr_len - 2456 sizeof(struct eth_header))) { 2457 chain->stat.ip_hacked++; 2458 return RSC_BYPASS; 2459 } 2460 2461 return RSC_CANDIDATE; 2462 } 2463 2464 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain, 2465 NetClientState *nc, 2466 const uint8_t *buf, size_t size) 2467 { 2468 int32_t ret; 2469 uint16_t hdr_len; 2470 VirtioNetRscUnit unit; 2471 2472 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len; 2473 2474 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header) 2475 + sizeof(struct tcp_header))) { 2476 chain->stat.bypass_not_tcp++; 2477 return virtio_net_do_receive(nc, buf, size); 2478 } 2479 2480 virtio_net_rsc_extract_unit4(chain, buf, &unit); 2481 if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size) 2482 != RSC_CANDIDATE) { 2483 return virtio_net_do_receive(nc, buf, size); 2484 } 2485 2486 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp); 2487 if (ret == RSC_BYPASS) { 2488 return virtio_net_do_receive(nc, buf, size); 2489 } else if (ret == RSC_FINAL) { 2490 return virtio_net_rsc_drain_flow(chain, nc, buf, size, 2491 ((hdr_len + sizeof(struct eth_header)) + 12), 2492 VIRTIO_NET_IP4_ADDR_SIZE, 2493 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)); 2494 } 2495 2496 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit); 2497 } 2498 2499 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain, 2500 struct ip6_header *ip6, 2501 const uint8_t *buf, size_t size) 2502 { 2503 uint16_t ip_len; 2504 2505 if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4) 2506 != IP_HEADER_VERSION_6) { 2507 return RSC_BYPASS; 2508 } 2509 2510 /* Both option and protocol is checked in this */ 2511 if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) { 2512 chain->stat.bypass_not_tcp++; 2513 return RSC_BYPASS; 2514 } 2515 2516 ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen); 2517 if (ip_len < sizeof(struct tcp_header) || 2518 ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header) 2519 - sizeof(struct ip6_header))) { 2520 chain->stat.ip_hacked++; 2521 return RSC_BYPASS; 2522 } 2523 2524 /* Don't handle packets with ecn flag */ 2525 if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) { 2526 chain->stat.ip_ecn++; 2527 return RSC_BYPASS; 2528 } 2529 2530 return RSC_CANDIDATE; 2531 } 2532 2533 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc, 2534 const uint8_t *buf, size_t size) 2535 { 2536 int32_t ret; 2537 uint16_t hdr_len; 2538 VirtioNetRscChain *chain; 2539 VirtioNetRscUnit unit; 2540 2541 chain = opq; 2542 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len; 2543 2544 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header) 2545 + sizeof(tcp_header))) { 2546 return virtio_net_do_receive(nc, buf, size); 2547 } 2548 2549 virtio_net_rsc_extract_unit6(chain, buf, &unit); 2550 if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain, 2551 unit.ip, buf, size)) { 2552 return virtio_net_do_receive(nc, buf, size); 2553 } 2554 2555 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp); 2556 if (ret == RSC_BYPASS) { 2557 return virtio_net_do_receive(nc, buf, size); 2558 } else if (ret == RSC_FINAL) { 2559 return virtio_net_rsc_drain_flow(chain, nc, buf, size, 2560 ((hdr_len + sizeof(struct eth_header)) + 8), 2561 VIRTIO_NET_IP6_ADDR_SIZE, 2562 hdr_len + sizeof(struct eth_header) 2563 + sizeof(struct ip6_header)); 2564 } 2565 2566 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit); 2567 } 2568 2569 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n, 2570 NetClientState *nc, 2571 uint16_t proto) 2572 { 2573 VirtioNetRscChain *chain; 2574 2575 if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) { 2576 return NULL; 2577 } 2578 2579 QTAILQ_FOREACH(chain, &n->rsc_chains, next) { 2580 if (chain->proto == proto) { 2581 return chain; 2582 } 2583 } 2584 2585 chain = g_malloc(sizeof(*chain)); 2586 chain->n = n; 2587 chain->proto = proto; 2588 if (proto == (uint16_t)ETH_P_IP) { 2589 chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD; 2590 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2591 } else { 2592 chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD; 2593 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2594 } 2595 chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST, 2596 virtio_net_rsc_purge, chain); 2597 memset(&chain->stat, 0, sizeof(chain->stat)); 2598 2599 QTAILQ_INIT(&chain->buffers); 2600 QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next); 2601 2602 return chain; 2603 } 2604 2605 static ssize_t virtio_net_rsc_receive(NetClientState *nc, 2606 const uint8_t *buf, 2607 size_t size) 2608 { 2609 uint16_t proto; 2610 VirtioNetRscChain *chain; 2611 struct eth_header *eth; 2612 VirtIONet *n; 2613 2614 n = qemu_get_nic_opaque(nc); 2615 if (size < (n->host_hdr_len + sizeof(struct eth_header))) { 2616 return virtio_net_do_receive(nc, buf, size); 2617 } 2618 2619 eth = (struct eth_header *)(buf + n->guest_hdr_len); 2620 proto = htons(eth->h_proto); 2621 2622 chain = virtio_net_rsc_lookup_chain(n, nc, proto); 2623 if (chain) { 2624 chain->stat.received++; 2625 if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) { 2626 return virtio_net_rsc_receive4(chain, nc, buf, size); 2627 } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) { 2628 return virtio_net_rsc_receive6(chain, nc, buf, size); 2629 } 2630 } 2631 return virtio_net_do_receive(nc, buf, size); 2632 } 2633 2634 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, 2635 size_t size) 2636 { 2637 VirtIONet *n = qemu_get_nic_opaque(nc); 2638 if ((n->rsc4_enabled || n->rsc6_enabled)) { 2639 return virtio_net_rsc_receive(nc, buf, size); 2640 } else { 2641 return virtio_net_do_receive(nc, buf, size); 2642 } 2643 } 2644 2645 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 2646 2647 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 2648 { 2649 VirtIONet *n = qemu_get_nic_opaque(nc); 2650 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 2651 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2652 int ret; 2653 2654 virtqueue_push(q->tx_vq, q->async_tx.elem, 0); 2655 virtio_notify(vdev, q->tx_vq); 2656 2657 g_free(q->async_tx.elem); 2658 q->async_tx.elem = NULL; 2659 2660 virtio_queue_set_notification(q->tx_vq, 1); 2661 ret = virtio_net_flush_tx(q); 2662 if (ret >= n->tx_burst) { 2663 /* 2664 * the flush has been stopped by tx_burst 2665 * we will not receive notification for the 2666 * remainining part, so re-schedule 2667 */ 2668 virtio_queue_set_notification(q->tx_vq, 0); 2669 if (q->tx_bh) { 2670 qemu_bh_schedule(q->tx_bh); 2671 } else { 2672 timer_mod(q->tx_timer, 2673 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2674 } 2675 q->tx_waiting = 1; 2676 } 2677 } 2678 2679 /* TX */ 2680 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 2681 { 2682 VirtIONet *n = q->n; 2683 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2684 VirtQueueElement *elem; 2685 int32_t num_packets = 0; 2686 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 2687 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 2688 return num_packets; 2689 } 2690 2691 if (q->async_tx.elem) { 2692 virtio_queue_set_notification(q->tx_vq, 0); 2693 return num_packets; 2694 } 2695 2696 for (;;) { 2697 ssize_t ret; 2698 unsigned int out_num; 2699 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg; 2700 struct virtio_net_hdr_mrg_rxbuf mhdr; 2701 2702 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement)); 2703 if (!elem) { 2704 break; 2705 } 2706 2707 out_num = elem->out_num; 2708 out_sg = elem->out_sg; 2709 if (out_num < 1) { 2710 virtio_error(vdev, "virtio-net header not in first element"); 2711 virtqueue_detach_element(q->tx_vq, elem, 0); 2712 g_free(elem); 2713 return -EINVAL; 2714 } 2715 2716 if (n->has_vnet_hdr) { 2717 if (iov_to_buf(out_sg, out_num, 0, &mhdr, n->guest_hdr_len) < 2718 n->guest_hdr_len) { 2719 virtio_error(vdev, "virtio-net header incorrect"); 2720 virtqueue_detach_element(q->tx_vq, elem, 0); 2721 g_free(elem); 2722 return -EINVAL; 2723 } 2724 if (n->needs_vnet_hdr_swap) { 2725 virtio_net_hdr_swap(vdev, (void *) &mhdr); 2726 sg2[0].iov_base = &mhdr; 2727 sg2[0].iov_len = n->guest_hdr_len; 2728 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1, 2729 out_sg, out_num, 2730 n->guest_hdr_len, -1); 2731 if (out_num == VIRTQUEUE_MAX_SIZE) { 2732 goto drop; 2733 } 2734 out_num += 1; 2735 out_sg = sg2; 2736 } 2737 } 2738 /* 2739 * If host wants to see the guest header as is, we can 2740 * pass it on unchanged. Otherwise, copy just the parts 2741 * that host is interested in. 2742 */ 2743 assert(n->host_hdr_len <= n->guest_hdr_len); 2744 if (n->host_hdr_len != n->guest_hdr_len) { 2745 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 2746 out_sg, out_num, 2747 0, n->host_hdr_len); 2748 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 2749 out_sg, out_num, 2750 n->guest_hdr_len, -1); 2751 out_num = sg_num; 2752 out_sg = sg; 2753 } 2754 2755 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 2756 out_sg, out_num, virtio_net_tx_complete); 2757 if (ret == 0) { 2758 virtio_queue_set_notification(q->tx_vq, 0); 2759 q->async_tx.elem = elem; 2760 return -EBUSY; 2761 } 2762 2763 drop: 2764 virtqueue_push(q->tx_vq, elem, 0); 2765 virtio_notify(vdev, q->tx_vq); 2766 g_free(elem); 2767 2768 if (++num_packets >= n->tx_burst) { 2769 break; 2770 } 2771 } 2772 return num_packets; 2773 } 2774 2775 static void virtio_net_tx_timer(void *opaque); 2776 2777 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 2778 { 2779 VirtIONet *n = VIRTIO_NET(vdev); 2780 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 2781 2782 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 2783 virtio_net_drop_tx_queue_data(vdev, vq); 2784 return; 2785 } 2786 2787 /* This happens when device was stopped but VCPU wasn't. */ 2788 if (!vdev->vm_running) { 2789 q->tx_waiting = 1; 2790 return; 2791 } 2792 2793 if (q->tx_waiting) { 2794 /* We already have queued packets, immediately flush */ 2795 timer_del(q->tx_timer); 2796 virtio_net_tx_timer(q); 2797 } else { 2798 /* re-arm timer to flush it (and more) on next tick */ 2799 timer_mod(q->tx_timer, 2800 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2801 q->tx_waiting = 1; 2802 virtio_queue_set_notification(vq, 0); 2803 } 2804 } 2805 2806 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 2807 { 2808 VirtIONet *n = VIRTIO_NET(vdev); 2809 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 2810 2811 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 2812 virtio_net_drop_tx_queue_data(vdev, vq); 2813 return; 2814 } 2815 2816 if (unlikely(q->tx_waiting)) { 2817 return; 2818 } 2819 q->tx_waiting = 1; 2820 /* This happens when device was stopped but VCPU wasn't. */ 2821 if (!vdev->vm_running) { 2822 return; 2823 } 2824 virtio_queue_set_notification(vq, 0); 2825 qemu_bh_schedule(q->tx_bh); 2826 } 2827 2828 static void virtio_net_tx_timer(void *opaque) 2829 { 2830 VirtIONetQueue *q = opaque; 2831 VirtIONet *n = q->n; 2832 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2833 int ret; 2834 2835 /* This happens when device was stopped but BH wasn't. */ 2836 if (!vdev->vm_running) { 2837 /* Make sure tx waiting is set, so we'll run when restarted. */ 2838 assert(q->tx_waiting); 2839 return; 2840 } 2841 2842 q->tx_waiting = 0; 2843 2844 /* Just in case the driver is not ready on more */ 2845 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 2846 return; 2847 } 2848 2849 ret = virtio_net_flush_tx(q); 2850 if (ret == -EBUSY || ret == -EINVAL) { 2851 return; 2852 } 2853 /* 2854 * If we flush a full burst of packets, assume there are 2855 * more coming and immediately rearm 2856 */ 2857 if (ret >= n->tx_burst) { 2858 q->tx_waiting = 1; 2859 timer_mod(q->tx_timer, 2860 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2861 return; 2862 } 2863 /* 2864 * If less than a full burst, re-enable notification and flush 2865 * anything that may have come in while we weren't looking. If 2866 * we find something, assume the guest is still active and rearm 2867 */ 2868 virtio_queue_set_notification(q->tx_vq, 1); 2869 ret = virtio_net_flush_tx(q); 2870 if (ret > 0) { 2871 virtio_queue_set_notification(q->tx_vq, 0); 2872 q->tx_waiting = 1; 2873 timer_mod(q->tx_timer, 2874 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2875 } 2876 } 2877 2878 static void virtio_net_tx_bh(void *opaque) 2879 { 2880 VirtIONetQueue *q = opaque; 2881 VirtIONet *n = q->n; 2882 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2883 int32_t ret; 2884 2885 /* This happens when device was stopped but BH wasn't. */ 2886 if (!vdev->vm_running) { 2887 /* Make sure tx waiting is set, so we'll run when restarted. */ 2888 assert(q->tx_waiting); 2889 return; 2890 } 2891 2892 q->tx_waiting = 0; 2893 2894 /* Just in case the driver is not ready on more */ 2895 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 2896 return; 2897 } 2898 2899 ret = virtio_net_flush_tx(q); 2900 if (ret == -EBUSY || ret == -EINVAL) { 2901 return; /* Notification re-enable handled by tx_complete or device 2902 * broken */ 2903 } 2904 2905 /* If we flush a full burst of packets, assume there are 2906 * more coming and immediately reschedule */ 2907 if (ret >= n->tx_burst) { 2908 qemu_bh_schedule(q->tx_bh); 2909 q->tx_waiting = 1; 2910 return; 2911 } 2912 2913 /* If less than a full burst, re-enable notification and flush 2914 * anything that may have come in while we weren't looking. If 2915 * we find something, assume the guest is still active and reschedule */ 2916 virtio_queue_set_notification(q->tx_vq, 1); 2917 ret = virtio_net_flush_tx(q); 2918 if (ret == -EINVAL) { 2919 return; 2920 } else if (ret > 0) { 2921 virtio_queue_set_notification(q->tx_vq, 0); 2922 qemu_bh_schedule(q->tx_bh); 2923 q->tx_waiting = 1; 2924 } 2925 } 2926 2927 static void virtio_net_add_queue(VirtIONet *n, int index) 2928 { 2929 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2930 2931 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size, 2932 virtio_net_handle_rx); 2933 2934 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 2935 n->vqs[index].tx_vq = 2936 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 2937 virtio_net_handle_tx_timer); 2938 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 2939 virtio_net_tx_timer, 2940 &n->vqs[index]); 2941 } else { 2942 n->vqs[index].tx_vq = 2943 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 2944 virtio_net_handle_tx_bh); 2945 n->vqs[index].tx_bh = qemu_bh_new_guarded(virtio_net_tx_bh, &n->vqs[index], 2946 &DEVICE(vdev)->mem_reentrancy_guard); 2947 } 2948 2949 n->vqs[index].tx_waiting = 0; 2950 n->vqs[index].n = n; 2951 } 2952 2953 static void virtio_net_del_queue(VirtIONet *n, int index) 2954 { 2955 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2956 VirtIONetQueue *q = &n->vqs[index]; 2957 NetClientState *nc = qemu_get_subqueue(n->nic, index); 2958 2959 qemu_purge_queued_packets(nc); 2960 2961 virtio_del_queue(vdev, index * 2); 2962 if (q->tx_timer) { 2963 timer_free(q->tx_timer); 2964 q->tx_timer = NULL; 2965 } else { 2966 qemu_bh_delete(q->tx_bh); 2967 q->tx_bh = NULL; 2968 } 2969 q->tx_waiting = 0; 2970 virtio_del_queue(vdev, index * 2 + 1); 2971 } 2972 2973 static void virtio_net_change_num_queue_pairs(VirtIONet *n, int new_max_queue_pairs) 2974 { 2975 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2976 int old_num_queues = virtio_get_num_queues(vdev); 2977 int new_num_queues = new_max_queue_pairs * 2 + 1; 2978 int i; 2979 2980 assert(old_num_queues >= 3); 2981 assert(old_num_queues % 2 == 1); 2982 2983 if (old_num_queues == new_num_queues) { 2984 return; 2985 } 2986 2987 /* 2988 * We always need to remove and add ctrl vq if 2989 * old_num_queues != new_num_queues. Remove ctrl_vq first, 2990 * and then we only enter one of the following two loops. 2991 */ 2992 virtio_del_queue(vdev, old_num_queues - 1); 2993 2994 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) { 2995 /* new_num_queues < old_num_queues */ 2996 virtio_net_del_queue(n, i / 2); 2997 } 2998 2999 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) { 3000 /* new_num_queues > old_num_queues */ 3001 virtio_net_add_queue(n, i / 2); 3002 } 3003 3004 /* add ctrl_vq last */ 3005 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 3006 } 3007 3008 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 3009 { 3010 int max = multiqueue ? n->max_queue_pairs : 1; 3011 3012 n->multiqueue = multiqueue; 3013 virtio_net_change_num_queue_pairs(n, max); 3014 3015 virtio_net_set_queue_pairs(n); 3016 } 3017 3018 static int virtio_net_post_load_device(void *opaque, int version_id) 3019 { 3020 VirtIONet *n = opaque; 3021 VirtIODevice *vdev = VIRTIO_DEVICE(n); 3022 int i, link_down; 3023 3024 trace_virtio_net_post_load_device(); 3025 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs, 3026 virtio_vdev_has_feature(vdev, 3027 VIRTIO_F_VERSION_1), 3028 virtio_vdev_has_feature(vdev, 3029 VIRTIO_NET_F_HASH_REPORT)); 3030 3031 /* MAC_TABLE_ENTRIES may be different from the saved image */ 3032 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) { 3033 n->mac_table.in_use = 0; 3034 } 3035 3036 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 3037 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 3038 } 3039 3040 /* 3041 * curr_guest_offloads will be later overwritten by the 3042 * virtio_set_features_nocheck call done from the virtio_load. 3043 * Here we make sure it is preserved and restored accordingly 3044 * in the virtio_net_post_load_virtio callback. 3045 */ 3046 n->saved_guest_offloads = n->curr_guest_offloads; 3047 3048 virtio_net_set_queue_pairs(n); 3049 3050 /* Find the first multicast entry in the saved MAC filter */ 3051 for (i = 0; i < n->mac_table.in_use; i++) { 3052 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 3053 break; 3054 } 3055 } 3056 n->mac_table.first_multi = i; 3057 3058 /* nc.link_down can't be migrated, so infer link_down according 3059 * to link status bit in n->status */ 3060 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 3061 for (i = 0; i < n->max_queue_pairs; i++) { 3062 qemu_get_subqueue(n->nic, i)->link_down = link_down; 3063 } 3064 3065 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 3066 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3067 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(), 3068 QEMU_CLOCK_VIRTUAL, 3069 virtio_net_announce_timer, n); 3070 if (n->announce_timer.round) { 3071 timer_mod(n->announce_timer.tm, 3072 qemu_clock_get_ms(n->announce_timer.type)); 3073 } else { 3074 qemu_announce_timer_del(&n->announce_timer, false); 3075 } 3076 } 3077 3078 if (n->rss_data.enabled) { 3079 n->rss_data.enabled_software_rss = n->rss_data.populate_hash; 3080 if (!n->rss_data.populate_hash) { 3081 if (!virtio_net_attach_epbf_rss(n)) { 3082 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) { 3083 warn_report("Can't post-load eBPF RSS for vhost"); 3084 } else { 3085 warn_report("Can't post-load eBPF RSS - " 3086 "fallback to software RSS"); 3087 n->rss_data.enabled_software_rss = true; 3088 } 3089 } 3090 } 3091 3092 trace_virtio_net_rss_enable(n->rss_data.hash_types, 3093 n->rss_data.indirections_len, 3094 sizeof(n->rss_data.key)); 3095 } else { 3096 trace_virtio_net_rss_disable(); 3097 } 3098 return 0; 3099 } 3100 3101 static int virtio_net_post_load_virtio(VirtIODevice *vdev) 3102 { 3103 VirtIONet *n = VIRTIO_NET(vdev); 3104 /* 3105 * The actual needed state is now in saved_guest_offloads, 3106 * see virtio_net_post_load_device for detail. 3107 * Restore it back and apply the desired offloads. 3108 */ 3109 n->curr_guest_offloads = n->saved_guest_offloads; 3110 if (peer_has_vnet_hdr(n)) { 3111 virtio_net_apply_guest_offloads(n); 3112 } 3113 3114 return 0; 3115 } 3116 3117 /* tx_waiting field of a VirtIONetQueue */ 3118 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = { 3119 .name = "virtio-net-queue-tx_waiting", 3120 .fields = (VMStateField[]) { 3121 VMSTATE_UINT32(tx_waiting, VirtIONetQueue), 3122 VMSTATE_END_OF_LIST() 3123 }, 3124 }; 3125 3126 static bool max_queue_pairs_gt_1(void *opaque, int version_id) 3127 { 3128 return VIRTIO_NET(opaque)->max_queue_pairs > 1; 3129 } 3130 3131 static bool has_ctrl_guest_offloads(void *opaque, int version_id) 3132 { 3133 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque), 3134 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS); 3135 } 3136 3137 static bool mac_table_fits(void *opaque, int version_id) 3138 { 3139 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES; 3140 } 3141 3142 static bool mac_table_doesnt_fit(void *opaque, int version_id) 3143 { 3144 return !mac_table_fits(opaque, version_id); 3145 } 3146 3147 /* This temporary type is shared by all the WITH_TMP methods 3148 * although only some fields are used by each. 3149 */ 3150 struct VirtIONetMigTmp { 3151 VirtIONet *parent; 3152 VirtIONetQueue *vqs_1; 3153 uint16_t curr_queue_pairs_1; 3154 uint8_t has_ufo; 3155 uint32_t has_vnet_hdr; 3156 }; 3157 3158 /* The 2nd and subsequent tx_waiting flags are loaded later than 3159 * the 1st entry in the queue_pairs and only if there's more than one 3160 * entry. We use the tmp mechanism to calculate a temporary 3161 * pointer and count and also validate the count. 3162 */ 3163 3164 static int virtio_net_tx_waiting_pre_save(void *opaque) 3165 { 3166 struct VirtIONetMigTmp *tmp = opaque; 3167 3168 tmp->vqs_1 = tmp->parent->vqs + 1; 3169 tmp->curr_queue_pairs_1 = tmp->parent->curr_queue_pairs - 1; 3170 if (tmp->parent->curr_queue_pairs == 0) { 3171 tmp->curr_queue_pairs_1 = 0; 3172 } 3173 3174 return 0; 3175 } 3176 3177 static int virtio_net_tx_waiting_pre_load(void *opaque) 3178 { 3179 struct VirtIONetMigTmp *tmp = opaque; 3180 3181 /* Reuse the pointer setup from save */ 3182 virtio_net_tx_waiting_pre_save(opaque); 3183 3184 if (tmp->parent->curr_queue_pairs > tmp->parent->max_queue_pairs) { 3185 error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x", 3186 tmp->parent->curr_queue_pairs, tmp->parent->max_queue_pairs); 3187 3188 return -EINVAL; 3189 } 3190 3191 return 0; /* all good */ 3192 } 3193 3194 static const VMStateDescription vmstate_virtio_net_tx_waiting = { 3195 .name = "virtio-net-tx_waiting", 3196 .pre_load = virtio_net_tx_waiting_pre_load, 3197 .pre_save = virtio_net_tx_waiting_pre_save, 3198 .fields = (VMStateField[]) { 3199 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp, 3200 curr_queue_pairs_1, 3201 vmstate_virtio_net_queue_tx_waiting, 3202 struct VirtIONetQueue), 3203 VMSTATE_END_OF_LIST() 3204 }, 3205 }; 3206 3207 /* the 'has_ufo' flag is just tested; if the incoming stream has the 3208 * flag set we need to check that we have it 3209 */ 3210 static int virtio_net_ufo_post_load(void *opaque, int version_id) 3211 { 3212 struct VirtIONetMigTmp *tmp = opaque; 3213 3214 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) { 3215 error_report("virtio-net: saved image requires TUN_F_UFO support"); 3216 return -EINVAL; 3217 } 3218 3219 return 0; 3220 } 3221 3222 static int virtio_net_ufo_pre_save(void *opaque) 3223 { 3224 struct VirtIONetMigTmp *tmp = opaque; 3225 3226 tmp->has_ufo = tmp->parent->has_ufo; 3227 3228 return 0; 3229 } 3230 3231 static const VMStateDescription vmstate_virtio_net_has_ufo = { 3232 .name = "virtio-net-ufo", 3233 .post_load = virtio_net_ufo_post_load, 3234 .pre_save = virtio_net_ufo_pre_save, 3235 .fields = (VMStateField[]) { 3236 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp), 3237 VMSTATE_END_OF_LIST() 3238 }, 3239 }; 3240 3241 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the 3242 * flag set we need to check that we have it 3243 */ 3244 static int virtio_net_vnet_post_load(void *opaque, int version_id) 3245 { 3246 struct VirtIONetMigTmp *tmp = opaque; 3247 3248 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) { 3249 error_report("virtio-net: saved image requires vnet_hdr=on"); 3250 return -EINVAL; 3251 } 3252 3253 return 0; 3254 } 3255 3256 static int virtio_net_vnet_pre_save(void *opaque) 3257 { 3258 struct VirtIONetMigTmp *tmp = opaque; 3259 3260 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr; 3261 3262 return 0; 3263 } 3264 3265 static const VMStateDescription vmstate_virtio_net_has_vnet = { 3266 .name = "virtio-net-vnet", 3267 .post_load = virtio_net_vnet_post_load, 3268 .pre_save = virtio_net_vnet_pre_save, 3269 .fields = (VMStateField[]) { 3270 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp), 3271 VMSTATE_END_OF_LIST() 3272 }, 3273 }; 3274 3275 static bool virtio_net_rss_needed(void *opaque) 3276 { 3277 return VIRTIO_NET(opaque)->rss_data.enabled; 3278 } 3279 3280 static const VMStateDescription vmstate_virtio_net_rss = { 3281 .name = "virtio-net-device/rss", 3282 .version_id = 1, 3283 .minimum_version_id = 1, 3284 .needed = virtio_net_rss_needed, 3285 .fields = (VMStateField[]) { 3286 VMSTATE_BOOL(rss_data.enabled, VirtIONet), 3287 VMSTATE_BOOL(rss_data.redirect, VirtIONet), 3288 VMSTATE_BOOL(rss_data.populate_hash, VirtIONet), 3289 VMSTATE_UINT32(rss_data.hash_types, VirtIONet), 3290 VMSTATE_UINT16(rss_data.indirections_len, VirtIONet), 3291 VMSTATE_UINT16(rss_data.default_queue, VirtIONet), 3292 VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet, 3293 VIRTIO_NET_RSS_MAX_KEY_SIZE), 3294 VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet, 3295 rss_data.indirections_len, 0, 3296 vmstate_info_uint16, uint16_t), 3297 VMSTATE_END_OF_LIST() 3298 }, 3299 }; 3300 3301 static const VMStateDescription vmstate_virtio_net_device = { 3302 .name = "virtio-net-device", 3303 .version_id = VIRTIO_NET_VM_VERSION, 3304 .minimum_version_id = VIRTIO_NET_VM_VERSION, 3305 .post_load = virtio_net_post_load_device, 3306 .fields = (VMStateField[]) { 3307 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN), 3308 VMSTATE_STRUCT_POINTER(vqs, VirtIONet, 3309 vmstate_virtio_net_queue_tx_waiting, 3310 VirtIONetQueue), 3311 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet), 3312 VMSTATE_UINT16(status, VirtIONet), 3313 VMSTATE_UINT8(promisc, VirtIONet), 3314 VMSTATE_UINT8(allmulti, VirtIONet), 3315 VMSTATE_UINT32(mac_table.in_use, VirtIONet), 3316 3317 /* Guarded pair: If it fits we load it, else we throw it away 3318 * - can happen if source has a larger MAC table.; post-load 3319 * sets flags in this case. 3320 */ 3321 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet, 3322 0, mac_table_fits, mac_table.in_use, 3323 ETH_ALEN), 3324 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0, 3325 mac_table.in_use, ETH_ALEN), 3326 3327 /* Note: This is an array of uint32's that's always been saved as a 3328 * buffer; hold onto your endiannesses; it's actually used as a bitmap 3329 * but based on the uint. 3330 */ 3331 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3), 3332 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3333 vmstate_virtio_net_has_vnet), 3334 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet), 3335 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet), 3336 VMSTATE_UINT8(alluni, VirtIONet), 3337 VMSTATE_UINT8(nomulti, VirtIONet), 3338 VMSTATE_UINT8(nouni, VirtIONet), 3339 VMSTATE_UINT8(nobcast, VirtIONet), 3340 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3341 vmstate_virtio_net_has_ufo), 3342 VMSTATE_SINGLE_TEST(max_queue_pairs, VirtIONet, max_queue_pairs_gt_1, 0, 3343 vmstate_info_uint16_equal, uint16_t), 3344 VMSTATE_UINT16_TEST(curr_queue_pairs, VirtIONet, max_queue_pairs_gt_1), 3345 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3346 vmstate_virtio_net_tx_waiting), 3347 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet, 3348 has_ctrl_guest_offloads), 3349 VMSTATE_END_OF_LIST() 3350 }, 3351 .subsections = (const VMStateDescription * []) { 3352 &vmstate_virtio_net_rss, 3353 NULL 3354 } 3355 }; 3356 3357 static NetClientInfo net_virtio_info = { 3358 .type = NET_CLIENT_DRIVER_NIC, 3359 .size = sizeof(NICState), 3360 .can_receive = virtio_net_can_receive, 3361 .receive = virtio_net_receive, 3362 .link_status_changed = virtio_net_set_link_status, 3363 .query_rx_filter = virtio_net_query_rxfilter, 3364 .announce = virtio_net_announce, 3365 }; 3366 3367 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 3368 { 3369 VirtIONet *n = VIRTIO_NET(vdev); 3370 NetClientState *nc; 3371 assert(n->vhost_started); 3372 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ) && idx == 2) { 3373 /* Must guard against invalid features and bogus queue index 3374 * from being set by malicious guest, or penetrated through 3375 * buggy migration stream. 3376 */ 3377 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3378 qemu_log_mask(LOG_GUEST_ERROR, 3379 "%s: bogus vq index ignored\n", __func__); 3380 return false; 3381 } 3382 nc = qemu_get_subqueue(n->nic, n->max_queue_pairs); 3383 } else { 3384 nc = qemu_get_subqueue(n->nic, vq2q(idx)); 3385 } 3386 /* 3387 * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1 3388 * as the macro of configure interrupt's IDX, If this driver does not 3389 * support, the function will return false 3390 */ 3391 3392 if (idx == VIRTIO_CONFIG_IRQ_IDX) { 3393 return vhost_net_config_pending(get_vhost_net(nc->peer)); 3394 } 3395 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx); 3396 } 3397 3398 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 3399 bool mask) 3400 { 3401 VirtIONet *n = VIRTIO_NET(vdev); 3402 NetClientState *nc; 3403 assert(n->vhost_started); 3404 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ) && idx == 2) { 3405 /* Must guard against invalid features and bogus queue index 3406 * from being set by malicious guest, or penetrated through 3407 * buggy migration stream. 3408 */ 3409 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3410 qemu_log_mask(LOG_GUEST_ERROR, 3411 "%s: bogus vq index ignored\n", __func__); 3412 return; 3413 } 3414 nc = qemu_get_subqueue(n->nic, n->max_queue_pairs); 3415 } else { 3416 nc = qemu_get_subqueue(n->nic, vq2q(idx)); 3417 } 3418 /* 3419 *Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1 3420 * as the macro of configure interrupt's IDX, If this driver does not 3421 * support, the function will return 3422 */ 3423 3424 if (idx == VIRTIO_CONFIG_IRQ_IDX) { 3425 vhost_net_config_mask(get_vhost_net(nc->peer), vdev, mask); 3426 return; 3427 } 3428 vhost_net_virtqueue_mask(get_vhost_net(nc->peer), vdev, idx, mask); 3429 } 3430 3431 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features) 3432 { 3433 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC); 3434 3435 n->config_size = virtio_get_config_size(&cfg_size_params, host_features); 3436 } 3437 3438 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 3439 const char *type) 3440 { 3441 /* 3442 * The name can be NULL, the netclient name will be type.x. 3443 */ 3444 assert(type != NULL); 3445 3446 g_free(n->netclient_name); 3447 g_free(n->netclient_type); 3448 n->netclient_name = g_strdup(name); 3449 n->netclient_type = g_strdup(type); 3450 } 3451 3452 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev) 3453 { 3454 HotplugHandler *hotplug_ctrl; 3455 PCIDevice *pci_dev; 3456 Error *err = NULL; 3457 3458 hotplug_ctrl = qdev_get_hotplug_handler(dev); 3459 if (hotplug_ctrl) { 3460 pci_dev = PCI_DEVICE(dev); 3461 pci_dev->partially_hotplugged = true; 3462 hotplug_handler_unplug_request(hotplug_ctrl, dev, &err); 3463 if (err) { 3464 error_report_err(err); 3465 return false; 3466 } 3467 } else { 3468 return false; 3469 } 3470 return true; 3471 } 3472 3473 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev, 3474 Error **errp) 3475 { 3476 Error *err = NULL; 3477 HotplugHandler *hotplug_ctrl; 3478 PCIDevice *pdev = PCI_DEVICE(dev); 3479 BusState *primary_bus; 3480 3481 if (!pdev->partially_hotplugged) { 3482 return true; 3483 } 3484 primary_bus = dev->parent_bus; 3485 if (!primary_bus) { 3486 error_setg(errp, "virtio_net: couldn't find primary bus"); 3487 return false; 3488 } 3489 qdev_set_parent_bus(dev, primary_bus, &error_abort); 3490 qatomic_set(&n->failover_primary_hidden, false); 3491 hotplug_ctrl = qdev_get_hotplug_handler(dev); 3492 if (hotplug_ctrl) { 3493 hotplug_handler_pre_plug(hotplug_ctrl, dev, &err); 3494 if (err) { 3495 goto out; 3496 } 3497 hotplug_handler_plug(hotplug_ctrl, dev, &err); 3498 } 3499 pdev->partially_hotplugged = false; 3500 3501 out: 3502 error_propagate(errp, err); 3503 return !err; 3504 } 3505 3506 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationState *s) 3507 { 3508 bool should_be_hidden; 3509 Error *err = NULL; 3510 DeviceState *dev = failover_find_primary_device(n); 3511 3512 if (!dev) { 3513 return; 3514 } 3515 3516 should_be_hidden = qatomic_read(&n->failover_primary_hidden); 3517 3518 if (migration_in_setup(s) && !should_be_hidden) { 3519 if (failover_unplug_primary(n, dev)) { 3520 vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev); 3521 qapi_event_send_unplug_primary(dev->id); 3522 qatomic_set(&n->failover_primary_hidden, true); 3523 } else { 3524 warn_report("couldn't unplug primary device"); 3525 } 3526 } else if (migration_has_failed(s)) { 3527 /* We already unplugged the device let's plug it back */ 3528 if (!failover_replug_primary(n, dev, &err)) { 3529 if (err) { 3530 error_report_err(err); 3531 } 3532 } 3533 } 3534 } 3535 3536 static void virtio_net_migration_state_notifier(Notifier *notifier, void *data) 3537 { 3538 MigrationState *s = data; 3539 VirtIONet *n = container_of(notifier, VirtIONet, migration_state); 3540 virtio_net_handle_migration_primary(n, s); 3541 } 3542 3543 static bool failover_hide_primary_device(DeviceListener *listener, 3544 const QDict *device_opts, 3545 bool from_json, 3546 Error **errp) 3547 { 3548 VirtIONet *n = container_of(listener, VirtIONet, primary_listener); 3549 const char *standby_id; 3550 3551 if (!device_opts) { 3552 return false; 3553 } 3554 3555 if (!qdict_haskey(device_opts, "failover_pair_id")) { 3556 return false; 3557 } 3558 3559 if (!qdict_haskey(device_opts, "id")) { 3560 error_setg(errp, "Device with failover_pair_id needs to have id"); 3561 return false; 3562 } 3563 3564 standby_id = qdict_get_str(device_opts, "failover_pair_id"); 3565 if (g_strcmp0(standby_id, n->netclient_name) != 0) { 3566 return false; 3567 } 3568 3569 /* 3570 * The hide helper can be called several times for a given device. 3571 * Check there is only one primary for a virtio-net device but 3572 * don't duplicate the qdict several times if it's called for the same 3573 * device. 3574 */ 3575 if (n->primary_opts) { 3576 const char *old, *new; 3577 /* devices with failover_pair_id always have an id */ 3578 old = qdict_get_str(n->primary_opts, "id"); 3579 new = qdict_get_str(device_opts, "id"); 3580 if (strcmp(old, new) != 0) { 3581 error_setg(errp, "Cannot attach more than one primary device to " 3582 "'%s': '%s' and '%s'", n->netclient_name, old, new); 3583 return false; 3584 } 3585 } else { 3586 n->primary_opts = qdict_clone_shallow(device_opts); 3587 n->primary_opts_from_json = from_json; 3588 } 3589 3590 /* failover_primary_hidden is set during feature negotiation */ 3591 return qatomic_read(&n->failover_primary_hidden); 3592 } 3593 3594 static void virtio_net_device_realize(DeviceState *dev, Error **errp) 3595 { 3596 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3597 VirtIONet *n = VIRTIO_NET(dev); 3598 NetClientState *nc; 3599 int i; 3600 3601 if (n->net_conf.mtu) { 3602 n->host_features |= (1ULL << VIRTIO_NET_F_MTU); 3603 } 3604 3605 if (n->net_conf.duplex_str) { 3606 if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) { 3607 n->net_conf.duplex = DUPLEX_HALF; 3608 } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) { 3609 n->net_conf.duplex = DUPLEX_FULL; 3610 } else { 3611 error_setg(errp, "'duplex' must be 'half' or 'full'"); 3612 return; 3613 } 3614 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX); 3615 } else { 3616 n->net_conf.duplex = DUPLEX_UNKNOWN; 3617 } 3618 3619 if (n->net_conf.speed < SPEED_UNKNOWN) { 3620 error_setg(errp, "'speed' must be between 0 and INT_MAX"); 3621 return; 3622 } 3623 if (n->net_conf.speed >= 0) { 3624 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX); 3625 } 3626 3627 if (n->failover) { 3628 n->primary_listener.hide_device = failover_hide_primary_device; 3629 qatomic_set(&n->failover_primary_hidden, true); 3630 device_listener_register(&n->primary_listener); 3631 n->migration_state.notify = virtio_net_migration_state_notifier; 3632 add_migration_state_change_notifier(&n->migration_state); 3633 n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY); 3634 } 3635 3636 virtio_net_set_config_size(n, n->host_features); 3637 virtio_init(vdev, VIRTIO_ID_NET, n->config_size); 3638 3639 /* 3640 * We set a lower limit on RX queue size to what it always was. 3641 * Guests that want a smaller ring can always resize it without 3642 * help from us (using virtio 1 and up). 3643 */ 3644 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE || 3645 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE || 3646 !is_power_of_2(n->net_conf.rx_queue_size)) { 3647 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), " 3648 "must be a power of 2 between %d and %d.", 3649 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE, 3650 VIRTQUEUE_MAX_SIZE); 3651 virtio_cleanup(vdev); 3652 return; 3653 } 3654 3655 if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE || 3656 n->net_conf.tx_queue_size > virtio_net_max_tx_queue_size(n) || 3657 !is_power_of_2(n->net_conf.tx_queue_size)) { 3658 error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), " 3659 "must be a power of 2 between %d and %d", 3660 n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE, 3661 virtio_net_max_tx_queue_size(n)); 3662 virtio_cleanup(vdev); 3663 return; 3664 } 3665 3666 n->max_ncs = MAX(n->nic_conf.peers.queues, 1); 3667 3668 /* 3669 * Figure out the datapath queue pairs since the backend could 3670 * provide control queue via peers as well. 3671 */ 3672 if (n->nic_conf.peers.queues) { 3673 for (i = 0; i < n->max_ncs; i++) { 3674 if (n->nic_conf.peers.ncs[i]->is_datapath) { 3675 ++n->max_queue_pairs; 3676 } 3677 } 3678 } 3679 n->max_queue_pairs = MAX(n->max_queue_pairs, 1); 3680 3681 if (n->max_queue_pairs * 2 + 1 > VIRTIO_QUEUE_MAX) { 3682 error_setg(errp, "Invalid number of queue pairs (= %" PRIu32 "), " 3683 "must be a positive integer less than %d.", 3684 n->max_queue_pairs, (VIRTIO_QUEUE_MAX - 1) / 2); 3685 virtio_cleanup(vdev); 3686 return; 3687 } 3688 n->vqs = g_new0(VirtIONetQueue, n->max_queue_pairs); 3689 n->curr_queue_pairs = 1; 3690 n->tx_timeout = n->net_conf.txtimer; 3691 3692 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 3693 && strcmp(n->net_conf.tx, "bh")) { 3694 warn_report("virtio-net: " 3695 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 3696 n->net_conf.tx); 3697 error_printf("Defaulting to \"bh\""); 3698 } 3699 3700 n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n), 3701 n->net_conf.tx_queue_size); 3702 3703 for (i = 0; i < n->max_queue_pairs; i++) { 3704 virtio_net_add_queue(n, i); 3705 } 3706 3707 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 3708 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 3709 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 3710 n->status = VIRTIO_NET_S_LINK_UP; 3711 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(), 3712 QEMU_CLOCK_VIRTUAL, 3713 virtio_net_announce_timer, n); 3714 n->announce_timer.round = 0; 3715 3716 if (n->netclient_type) { 3717 /* 3718 * Happen when virtio_net_set_netclient_name has been called. 3719 */ 3720 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 3721 n->netclient_type, n->netclient_name, n); 3722 } else { 3723 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 3724 object_get_typename(OBJECT(dev)), dev->id, n); 3725 } 3726 3727 for (i = 0; i < n->max_queue_pairs; i++) { 3728 n->nic->ncs[i].do_not_pad = true; 3729 } 3730 3731 peer_test_vnet_hdr(n); 3732 if (peer_has_vnet_hdr(n)) { 3733 for (i = 0; i < n->max_queue_pairs; i++) { 3734 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true); 3735 } 3736 n->host_hdr_len = sizeof(struct virtio_net_hdr); 3737 } else { 3738 n->host_hdr_len = 0; 3739 } 3740 3741 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 3742 3743 n->vqs[0].tx_waiting = 0; 3744 n->tx_burst = n->net_conf.txburst; 3745 virtio_net_set_mrg_rx_bufs(n, 0, 0, 0); 3746 n->promisc = 1; /* for compatibility */ 3747 3748 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 3749 3750 n->vlans = g_malloc0(MAX_VLAN >> 3); 3751 3752 nc = qemu_get_queue(n->nic); 3753 nc->rxfilter_notify_enabled = 1; 3754 3755 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 3756 struct virtio_net_config netcfg = {}; 3757 memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN); 3758 vhost_net_set_config(get_vhost_net(nc->peer), 3759 (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_FRONTEND); 3760 } 3761 QTAILQ_INIT(&n->rsc_chains); 3762 n->qdev = dev; 3763 3764 net_rx_pkt_init(&n->rx_pkt); 3765 3766 if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) { 3767 virtio_net_load_ebpf(n); 3768 } 3769 } 3770 3771 static void virtio_net_device_unrealize(DeviceState *dev) 3772 { 3773 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3774 VirtIONet *n = VIRTIO_NET(dev); 3775 int i, max_queue_pairs; 3776 3777 if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) { 3778 virtio_net_unload_ebpf(n); 3779 } 3780 3781 /* This will stop vhost backend if appropriate. */ 3782 virtio_net_set_status(vdev, 0); 3783 3784 g_free(n->netclient_name); 3785 n->netclient_name = NULL; 3786 g_free(n->netclient_type); 3787 n->netclient_type = NULL; 3788 3789 g_free(n->mac_table.macs); 3790 g_free(n->vlans); 3791 3792 if (n->failover) { 3793 qobject_unref(n->primary_opts); 3794 device_listener_unregister(&n->primary_listener); 3795 remove_migration_state_change_notifier(&n->migration_state); 3796 } else { 3797 assert(n->primary_opts == NULL); 3798 } 3799 3800 max_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 3801 for (i = 0; i < max_queue_pairs; i++) { 3802 virtio_net_del_queue(n, i); 3803 } 3804 /* delete also control vq */ 3805 virtio_del_queue(vdev, max_queue_pairs * 2); 3806 qemu_announce_timer_del(&n->announce_timer, false); 3807 g_free(n->vqs); 3808 qemu_del_nic(n->nic); 3809 virtio_net_rsc_cleanup(n); 3810 g_free(n->rss_data.indirections_table); 3811 net_rx_pkt_uninit(n->rx_pkt); 3812 virtio_cleanup(vdev); 3813 } 3814 3815 static void virtio_net_instance_init(Object *obj) 3816 { 3817 VirtIONet *n = VIRTIO_NET(obj); 3818 3819 /* 3820 * The default config_size is sizeof(struct virtio_net_config). 3821 * Can be overridden with virtio_net_set_config_size. 3822 */ 3823 n->config_size = sizeof(struct virtio_net_config); 3824 device_add_bootindex_property(obj, &n->nic_conf.bootindex, 3825 "bootindex", "/ethernet-phy@0", 3826 DEVICE(n)); 3827 3828 ebpf_rss_init(&n->ebpf_rss); 3829 } 3830 3831 static int virtio_net_pre_save(void *opaque) 3832 { 3833 VirtIONet *n = opaque; 3834 3835 /* At this point, backend must be stopped, otherwise 3836 * it might keep writing to memory. */ 3837 assert(!n->vhost_started); 3838 3839 return 0; 3840 } 3841 3842 static bool primary_unplug_pending(void *opaque) 3843 { 3844 DeviceState *dev = opaque; 3845 DeviceState *primary; 3846 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3847 VirtIONet *n = VIRTIO_NET(vdev); 3848 3849 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) { 3850 return false; 3851 } 3852 primary = failover_find_primary_device(n); 3853 return primary ? primary->pending_deleted_event : false; 3854 } 3855 3856 static bool dev_unplug_pending(void *opaque) 3857 { 3858 DeviceState *dev = opaque; 3859 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); 3860 3861 return vdc->primary_unplug_pending(dev); 3862 } 3863 3864 static struct vhost_dev *virtio_net_get_vhost(VirtIODevice *vdev) 3865 { 3866 VirtIONet *n = VIRTIO_NET(vdev); 3867 NetClientState *nc = qemu_get_queue(n->nic); 3868 struct vhost_net *net = get_vhost_net(nc->peer); 3869 return &net->dev; 3870 } 3871 3872 static const VMStateDescription vmstate_virtio_net = { 3873 .name = "virtio-net", 3874 .minimum_version_id = VIRTIO_NET_VM_VERSION, 3875 .version_id = VIRTIO_NET_VM_VERSION, 3876 .fields = (VMStateField[]) { 3877 VMSTATE_VIRTIO_DEVICE, 3878 VMSTATE_END_OF_LIST() 3879 }, 3880 .pre_save = virtio_net_pre_save, 3881 .dev_unplug_pending = dev_unplug_pending, 3882 }; 3883 3884 static Property virtio_net_properties[] = { 3885 DEFINE_PROP_BIT64("csum", VirtIONet, host_features, 3886 VIRTIO_NET_F_CSUM, true), 3887 DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features, 3888 VIRTIO_NET_F_GUEST_CSUM, true), 3889 DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true), 3890 DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features, 3891 VIRTIO_NET_F_GUEST_TSO4, true), 3892 DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features, 3893 VIRTIO_NET_F_GUEST_TSO6, true), 3894 DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features, 3895 VIRTIO_NET_F_GUEST_ECN, true), 3896 DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features, 3897 VIRTIO_NET_F_GUEST_UFO, true), 3898 DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features, 3899 VIRTIO_NET_F_GUEST_ANNOUNCE, true), 3900 DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features, 3901 VIRTIO_NET_F_HOST_TSO4, true), 3902 DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features, 3903 VIRTIO_NET_F_HOST_TSO6, true), 3904 DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features, 3905 VIRTIO_NET_F_HOST_ECN, true), 3906 DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features, 3907 VIRTIO_NET_F_HOST_UFO, true), 3908 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features, 3909 VIRTIO_NET_F_MRG_RXBUF, true), 3910 DEFINE_PROP_BIT64("status", VirtIONet, host_features, 3911 VIRTIO_NET_F_STATUS, true), 3912 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features, 3913 VIRTIO_NET_F_CTRL_VQ, true), 3914 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features, 3915 VIRTIO_NET_F_CTRL_RX, true), 3916 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features, 3917 VIRTIO_NET_F_CTRL_VLAN, true), 3918 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features, 3919 VIRTIO_NET_F_CTRL_RX_EXTRA, true), 3920 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features, 3921 VIRTIO_NET_F_CTRL_MAC_ADDR, true), 3922 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features, 3923 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true), 3924 DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false), 3925 DEFINE_PROP_BIT64("rss", VirtIONet, host_features, 3926 VIRTIO_NET_F_RSS, false), 3927 DEFINE_PROP_BIT64("hash", VirtIONet, host_features, 3928 VIRTIO_NET_F_HASH_REPORT, false), 3929 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features, 3930 VIRTIO_NET_F_RSC_EXT, false), 3931 DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout, 3932 VIRTIO_NET_RSC_DEFAULT_INTERVAL), 3933 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 3934 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 3935 TX_TIMER_INTERVAL), 3936 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 3937 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 3938 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size, 3939 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE), 3940 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size, 3941 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE), 3942 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0), 3943 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend, 3944 true), 3945 DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN), 3946 DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str), 3947 DEFINE_PROP_BOOL("failover", VirtIONet, failover, false), 3948 DEFINE_PROP_BIT64("guest_uso4", VirtIONet, host_features, 3949 VIRTIO_NET_F_GUEST_USO4, true), 3950 DEFINE_PROP_BIT64("guest_uso6", VirtIONet, host_features, 3951 VIRTIO_NET_F_GUEST_USO6, true), 3952 DEFINE_PROP_BIT64("host_uso", VirtIONet, host_features, 3953 VIRTIO_NET_F_HOST_USO, true), 3954 DEFINE_PROP_END_OF_LIST(), 3955 }; 3956 3957 static void virtio_net_class_init(ObjectClass *klass, void *data) 3958 { 3959 DeviceClass *dc = DEVICE_CLASS(klass); 3960 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 3961 3962 device_class_set_props(dc, virtio_net_properties); 3963 dc->vmsd = &vmstate_virtio_net; 3964 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 3965 vdc->realize = virtio_net_device_realize; 3966 vdc->unrealize = virtio_net_device_unrealize; 3967 vdc->get_config = virtio_net_get_config; 3968 vdc->set_config = virtio_net_set_config; 3969 vdc->get_features = virtio_net_get_features; 3970 vdc->set_features = virtio_net_set_features; 3971 vdc->bad_features = virtio_net_bad_features; 3972 vdc->reset = virtio_net_reset; 3973 vdc->queue_reset = virtio_net_queue_reset; 3974 vdc->queue_enable = virtio_net_queue_enable; 3975 vdc->set_status = virtio_net_set_status; 3976 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 3977 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 3978 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO); 3979 vdc->post_load = virtio_net_post_load_virtio; 3980 vdc->vmsd = &vmstate_virtio_net_device; 3981 vdc->primary_unplug_pending = primary_unplug_pending; 3982 vdc->get_vhost = virtio_net_get_vhost; 3983 vdc->toggle_device_iotlb = vhost_toggle_device_iotlb; 3984 } 3985 3986 static const TypeInfo virtio_net_info = { 3987 .name = TYPE_VIRTIO_NET, 3988 .parent = TYPE_VIRTIO_DEVICE, 3989 .instance_size = sizeof(VirtIONet), 3990 .instance_init = virtio_net_instance_init, 3991 .class_init = virtio_net_class_init, 3992 }; 3993 3994 static void virtio_register_types(void) 3995 { 3996 type_register_static(&virtio_net_info); 3997 } 3998 3999 type_init(virtio_register_types) 4000