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 "sysemu/replay.h" 44 #include "trace.h" 45 #include "monitor/qdev.h" 46 #include "monitor/monitor.h" 47 #include "hw/pci/pci_device.h" 48 #include "net_rx_pkt.h" 49 #include "hw/virtio/vhost.h" 50 #include "sysemu/qtest.h" 51 52 #define VIRTIO_NET_VM_VERSION 11 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 = n->has_vnet_hdr && 365 virtio_net_set_vnet_endian(vdev, n->nic->ncs, 366 queue_pairs, true); 367 } else if (virtio_net_started(n, vdev->status)) { 368 /* After using the device, we need to reset the network backend to 369 * the default (guest native endianness), otherwise the guest may 370 * lose network connectivity if it is rebooted into a different 371 * endianness. 372 */ 373 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queue_pairs, false); 374 } 375 } 376 377 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq) 378 { 379 unsigned int dropped = virtqueue_drop_all(vq); 380 if (dropped) { 381 virtio_notify(vdev, vq); 382 } 383 } 384 385 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status) 386 { 387 VirtIONet *n = VIRTIO_NET(vdev); 388 VirtIONetQueue *q; 389 int i; 390 uint8_t queue_status; 391 392 virtio_net_vnet_endian_status(n, status); 393 virtio_net_vhost_status(n, status); 394 395 for (i = 0; i < n->max_queue_pairs; i++) { 396 NetClientState *ncs = qemu_get_subqueue(n->nic, i); 397 bool queue_started; 398 q = &n->vqs[i]; 399 400 if ((!n->multiqueue && i != 0) || i >= n->curr_queue_pairs) { 401 queue_status = 0; 402 } else { 403 queue_status = status; 404 } 405 queue_started = 406 virtio_net_started(n, queue_status) && !n->vhost_started; 407 408 if (queue_started) { 409 qemu_flush_queued_packets(ncs); 410 } 411 412 if (!q->tx_waiting) { 413 continue; 414 } 415 416 if (queue_started) { 417 if (q->tx_timer) { 418 timer_mod(q->tx_timer, 419 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 420 } else { 421 replay_bh_schedule_event(q->tx_bh); 422 } 423 } else { 424 if (q->tx_timer) { 425 timer_del(q->tx_timer); 426 } else { 427 qemu_bh_cancel(q->tx_bh); 428 } 429 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 && 430 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) && 431 vdev->vm_running) { 432 /* if tx is waiting we are likely have some packets in tx queue 433 * and disabled notification */ 434 q->tx_waiting = 0; 435 virtio_queue_set_notification(q->tx_vq, 1); 436 virtio_net_drop_tx_queue_data(vdev, q->tx_vq); 437 } 438 } 439 } 440 } 441 442 static void virtio_net_set_link_status(NetClientState *nc) 443 { 444 VirtIONet *n = qemu_get_nic_opaque(nc); 445 VirtIODevice *vdev = VIRTIO_DEVICE(n); 446 uint16_t old_status = n->status; 447 448 if (nc->link_down) 449 n->status &= ~VIRTIO_NET_S_LINK_UP; 450 else 451 n->status |= VIRTIO_NET_S_LINK_UP; 452 453 if (n->status != old_status) 454 virtio_notify_config(vdev); 455 456 virtio_net_set_status(vdev, vdev->status); 457 } 458 459 static void rxfilter_notify(NetClientState *nc) 460 { 461 VirtIONet *n = qemu_get_nic_opaque(nc); 462 463 if (nc->rxfilter_notify_enabled) { 464 char *path = object_get_canonical_path(OBJECT(n->qdev)); 465 qapi_event_send_nic_rx_filter_changed(n->netclient_name, path); 466 g_free(path); 467 468 /* disable event notification to avoid events flooding */ 469 nc->rxfilter_notify_enabled = 0; 470 } 471 } 472 473 static intList *get_vlan_table(VirtIONet *n) 474 { 475 intList *list; 476 int i, j; 477 478 list = NULL; 479 for (i = 0; i < MAX_VLAN >> 5; i++) { 480 for (j = 0; n->vlans[i] && j <= 0x1f; j++) { 481 if (n->vlans[i] & (1U << j)) { 482 QAPI_LIST_PREPEND(list, (i << 5) + j); 483 } 484 } 485 } 486 487 return list; 488 } 489 490 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc) 491 { 492 VirtIONet *n = qemu_get_nic_opaque(nc); 493 VirtIODevice *vdev = VIRTIO_DEVICE(n); 494 RxFilterInfo *info; 495 strList *str_list; 496 int i; 497 498 info = g_malloc0(sizeof(*info)); 499 info->name = g_strdup(nc->name); 500 info->promiscuous = n->promisc; 501 502 if (n->nouni) { 503 info->unicast = RX_STATE_NONE; 504 } else if (n->alluni) { 505 info->unicast = RX_STATE_ALL; 506 } else { 507 info->unicast = RX_STATE_NORMAL; 508 } 509 510 if (n->nomulti) { 511 info->multicast = RX_STATE_NONE; 512 } else if (n->allmulti) { 513 info->multicast = RX_STATE_ALL; 514 } else { 515 info->multicast = RX_STATE_NORMAL; 516 } 517 518 info->broadcast_allowed = n->nobcast; 519 info->multicast_overflow = n->mac_table.multi_overflow; 520 info->unicast_overflow = n->mac_table.uni_overflow; 521 522 info->main_mac = qemu_mac_strdup_printf(n->mac); 523 524 str_list = NULL; 525 for (i = 0; i < n->mac_table.first_multi; i++) { 526 QAPI_LIST_PREPEND(str_list, 527 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN)); 528 } 529 info->unicast_table = str_list; 530 531 str_list = NULL; 532 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 533 QAPI_LIST_PREPEND(str_list, 534 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN)); 535 } 536 info->multicast_table = str_list; 537 info->vlan_table = get_vlan_table(n); 538 539 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) { 540 info->vlan = RX_STATE_ALL; 541 } else if (!info->vlan_table) { 542 info->vlan = RX_STATE_NONE; 543 } else { 544 info->vlan = RX_STATE_NORMAL; 545 } 546 547 /* enable event notification after query */ 548 nc->rxfilter_notify_enabled = 1; 549 550 return info; 551 } 552 553 static void virtio_net_queue_reset(VirtIODevice *vdev, uint32_t queue_index) 554 { 555 VirtIONet *n = VIRTIO_NET(vdev); 556 NetClientState *nc; 557 558 /* validate queue_index and skip for cvq */ 559 if (queue_index >= n->max_queue_pairs * 2) { 560 return; 561 } 562 563 nc = qemu_get_subqueue(n->nic, vq2q(queue_index)); 564 565 if (!nc->peer) { 566 return; 567 } 568 569 if (get_vhost_net(nc->peer) && 570 nc->peer->info->type == NET_CLIENT_DRIVER_TAP) { 571 vhost_net_virtqueue_reset(vdev, nc, queue_index); 572 } 573 574 flush_or_purge_queued_packets(nc); 575 } 576 577 static void virtio_net_queue_enable(VirtIODevice *vdev, uint32_t queue_index) 578 { 579 VirtIONet *n = VIRTIO_NET(vdev); 580 NetClientState *nc; 581 int r; 582 583 /* validate queue_index and skip for cvq */ 584 if (queue_index >= n->max_queue_pairs * 2) { 585 return; 586 } 587 588 nc = qemu_get_subqueue(n->nic, vq2q(queue_index)); 589 590 if (!nc->peer || !vdev->vhost_started) { 591 return; 592 } 593 594 if (get_vhost_net(nc->peer) && 595 nc->peer->info->type == NET_CLIENT_DRIVER_TAP) { 596 r = vhost_net_virtqueue_restart(vdev, nc, queue_index); 597 if (r < 0) { 598 error_report("unable to restart vhost net virtqueue: %d, " 599 "when resetting the queue", queue_index); 600 } 601 } 602 } 603 604 static void peer_test_vnet_hdr(VirtIONet *n) 605 { 606 NetClientState *nc = qemu_get_queue(n->nic); 607 if (!nc->peer) { 608 return; 609 } 610 611 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer); 612 } 613 614 static int peer_has_vnet_hdr(VirtIONet *n) 615 { 616 return n->has_vnet_hdr; 617 } 618 619 static int peer_has_ufo(VirtIONet *n) 620 { 621 if (!peer_has_vnet_hdr(n)) 622 return 0; 623 624 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer); 625 626 return n->has_ufo; 627 } 628 629 static int peer_has_uso(VirtIONet *n) 630 { 631 if (!peer_has_vnet_hdr(n)) { 632 return 0; 633 } 634 635 return qemu_has_uso(qemu_get_queue(n->nic)->peer); 636 } 637 638 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs, 639 int version_1, int hash_report) 640 { 641 int i; 642 NetClientState *nc; 643 644 n->mergeable_rx_bufs = mergeable_rx_bufs; 645 646 if (version_1) { 647 n->guest_hdr_len = hash_report ? 648 sizeof(struct virtio_net_hdr_v1_hash) : 649 sizeof(struct virtio_net_hdr_mrg_rxbuf); 650 n->rss_data.populate_hash = !!hash_report; 651 } else { 652 n->guest_hdr_len = n->mergeable_rx_bufs ? 653 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 654 sizeof(struct virtio_net_hdr); 655 n->rss_data.populate_hash = false; 656 } 657 658 for (i = 0; i < n->max_queue_pairs; i++) { 659 nc = qemu_get_subqueue(n->nic, i); 660 661 if (peer_has_vnet_hdr(n) && 662 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) { 663 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len); 664 n->host_hdr_len = n->guest_hdr_len; 665 } 666 } 667 } 668 669 static int virtio_net_max_tx_queue_size(VirtIONet *n) 670 { 671 NetClientState *peer = n->nic_conf.peers.ncs[0]; 672 673 /* 674 * Backends other than vhost-user or vhost-vdpa don't support max queue 675 * size. 676 */ 677 if (!peer) { 678 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 679 } 680 681 switch(peer->info->type) { 682 case NET_CLIENT_DRIVER_VHOST_USER: 683 case NET_CLIENT_DRIVER_VHOST_VDPA: 684 return VIRTQUEUE_MAX_SIZE; 685 default: 686 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 687 }; 688 } 689 690 static int peer_attach(VirtIONet *n, int index) 691 { 692 NetClientState *nc = qemu_get_subqueue(n->nic, index); 693 694 if (!nc->peer) { 695 return 0; 696 } 697 698 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 699 vhost_set_vring_enable(nc->peer, 1); 700 } 701 702 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 703 return 0; 704 } 705 706 if (n->max_queue_pairs == 1) { 707 return 0; 708 } 709 710 return tap_enable(nc->peer); 711 } 712 713 static int peer_detach(VirtIONet *n, int index) 714 { 715 NetClientState *nc = qemu_get_subqueue(n->nic, index); 716 717 if (!nc->peer) { 718 return 0; 719 } 720 721 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 722 vhost_set_vring_enable(nc->peer, 0); 723 } 724 725 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 726 return 0; 727 } 728 729 return tap_disable(nc->peer); 730 } 731 732 static void virtio_net_set_queue_pairs(VirtIONet *n) 733 { 734 int i; 735 int r; 736 737 if (n->nic->peer_deleted) { 738 return; 739 } 740 741 for (i = 0; i < n->max_queue_pairs; i++) { 742 if (i < n->curr_queue_pairs) { 743 r = peer_attach(n, i); 744 assert(!r); 745 } else { 746 r = peer_detach(n, i); 747 assert(!r); 748 } 749 } 750 } 751 752 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); 753 754 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features, 755 Error **errp) 756 { 757 VirtIONet *n = VIRTIO_NET(vdev); 758 NetClientState *nc = qemu_get_queue(n->nic); 759 760 /* Firstly sync all virtio-net possible supported features */ 761 features |= n->host_features; 762 763 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 764 765 if (!peer_has_vnet_hdr(n)) { 766 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM); 767 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4); 768 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6); 769 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN); 770 771 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM); 772 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4); 773 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6); 774 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN); 775 776 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_USO); 777 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO4); 778 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO6); 779 780 virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT); 781 } 782 783 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 784 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO); 785 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO); 786 } 787 788 if (!peer_has_uso(n)) { 789 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_USO); 790 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO4); 791 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO6); 792 } 793 794 if (!get_vhost_net(nc->peer)) { 795 return features; 796 } 797 798 if (!ebpf_rss_is_loaded(&n->ebpf_rss)) { 799 virtio_clear_feature(&features, VIRTIO_NET_F_RSS); 800 } 801 features = vhost_net_get_features(get_vhost_net(nc->peer), features); 802 vdev->backend_features = features; 803 804 if (n->mtu_bypass_backend && 805 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) { 806 features |= (1ULL << VIRTIO_NET_F_MTU); 807 } 808 809 /* 810 * Since GUEST_ANNOUNCE is emulated the feature bit could be set without 811 * enabled. This happens in the vDPA case. 812 * 813 * Make sure the feature set is not incoherent, as the driver could refuse 814 * to start. 815 * 816 * TODO: QEMU is able to emulate a CVQ just for guest_announce purposes, 817 * helping guest to notify the new location with vDPA devices that does not 818 * support it. 819 */ 820 if (!virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_CTRL_VQ)) { 821 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ANNOUNCE); 822 } 823 824 return features; 825 } 826 827 static uint64_t virtio_net_bad_features(VirtIODevice *vdev) 828 { 829 uint64_t features = 0; 830 831 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 832 * but also these: */ 833 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 834 virtio_add_feature(&features, VIRTIO_NET_F_CSUM); 835 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4); 836 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6); 837 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN); 838 839 return features; 840 } 841 842 static void virtio_net_apply_guest_offloads(VirtIONet *n) 843 { 844 qemu_set_offload(qemu_get_queue(n->nic)->peer, 845 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 846 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 847 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 848 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 849 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)), 850 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_USO4)), 851 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_USO6))); 852 } 853 854 static uint64_t virtio_net_guest_offloads_by_features(uint64_t features) 855 { 856 static const uint64_t guest_offloads_mask = 857 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 858 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 859 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 860 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 861 (1ULL << VIRTIO_NET_F_GUEST_UFO) | 862 (1ULL << VIRTIO_NET_F_GUEST_USO4) | 863 (1ULL << VIRTIO_NET_F_GUEST_USO6); 864 865 return guest_offloads_mask & features; 866 } 867 868 uint64_t virtio_net_supported_guest_offloads(const VirtIONet *n) 869 { 870 VirtIODevice *vdev = VIRTIO_DEVICE(n); 871 return virtio_net_guest_offloads_by_features(vdev->guest_features); 872 } 873 874 typedef struct { 875 VirtIONet *n; 876 DeviceState *dev; 877 } FailoverDevice; 878 879 /** 880 * Set the failover primary device 881 * 882 * @opaque: FailoverId to setup 883 * @opts: opts for device we are handling 884 * @errp: returns an error if this function fails 885 */ 886 static int failover_set_primary(DeviceState *dev, void *opaque) 887 { 888 FailoverDevice *fdev = opaque; 889 PCIDevice *pci_dev = (PCIDevice *) 890 object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE); 891 892 if (!pci_dev) { 893 return 0; 894 } 895 896 if (!g_strcmp0(pci_dev->failover_pair_id, fdev->n->netclient_name)) { 897 fdev->dev = dev; 898 return 1; 899 } 900 901 return 0; 902 } 903 904 /** 905 * Find the primary device for this failover virtio-net 906 * 907 * @n: VirtIONet device 908 * @errp: returns an error if this function fails 909 */ 910 static DeviceState *failover_find_primary_device(VirtIONet *n) 911 { 912 FailoverDevice fdev = { 913 .n = n, 914 }; 915 916 qbus_walk_children(sysbus_get_default(), failover_set_primary, NULL, 917 NULL, NULL, &fdev); 918 return fdev.dev; 919 } 920 921 static void failover_add_primary(VirtIONet *n, Error **errp) 922 { 923 Error *err = NULL; 924 DeviceState *dev = failover_find_primary_device(n); 925 926 if (dev) { 927 return; 928 } 929 930 if (!n->primary_opts) { 931 error_setg(errp, "Primary device not found"); 932 error_append_hint(errp, "Virtio-net failover will not work. Make " 933 "sure primary device has parameter" 934 " failover_pair_id=%s\n", n->netclient_name); 935 return; 936 } 937 938 dev = qdev_device_add_from_qdict(n->primary_opts, 939 n->primary_opts_from_json, 940 &err); 941 if (err) { 942 qobject_unref(n->primary_opts); 943 n->primary_opts = NULL; 944 } else { 945 object_unref(OBJECT(dev)); 946 } 947 error_propagate(errp, err); 948 } 949 950 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features) 951 { 952 VirtIONet *n = VIRTIO_NET(vdev); 953 Error *err = NULL; 954 int i; 955 956 if (n->mtu_bypass_backend && 957 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) { 958 features &= ~(1ULL << VIRTIO_NET_F_MTU); 959 } 960 961 virtio_net_set_multiqueue(n, 962 virtio_has_feature(features, VIRTIO_NET_F_RSS) || 963 virtio_has_feature(features, VIRTIO_NET_F_MQ)); 964 965 virtio_net_set_mrg_rx_bufs(n, 966 virtio_has_feature(features, 967 VIRTIO_NET_F_MRG_RXBUF), 968 virtio_has_feature(features, 969 VIRTIO_F_VERSION_1), 970 virtio_has_feature(features, 971 VIRTIO_NET_F_HASH_REPORT)); 972 973 n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) && 974 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4); 975 n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) && 976 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6); 977 n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS); 978 979 if (n->has_vnet_hdr) { 980 n->curr_guest_offloads = 981 virtio_net_guest_offloads_by_features(features); 982 virtio_net_apply_guest_offloads(n); 983 } 984 985 for (i = 0; i < n->max_queue_pairs; i++) { 986 NetClientState *nc = qemu_get_subqueue(n->nic, i); 987 988 if (!get_vhost_net(nc->peer)) { 989 continue; 990 } 991 vhost_net_ack_features(get_vhost_net(nc->peer), features); 992 993 /* 994 * keep acked_features in NetVhostUserState up-to-date so it 995 * can't miss any features configured by guest virtio driver. 996 */ 997 vhost_net_save_acked_features(nc->peer); 998 } 999 1000 if (!virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) { 1001 memset(n->vlans, 0xff, MAX_VLAN >> 3); 1002 } 1003 1004 if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) { 1005 qapi_event_send_failover_negotiated(n->netclient_name); 1006 qatomic_set(&n->failover_primary_hidden, false); 1007 failover_add_primary(n, &err); 1008 if (err) { 1009 if (!qtest_enabled()) { 1010 warn_report_err(err); 1011 } else { 1012 error_free(err); 1013 } 1014 } 1015 } 1016 } 1017 1018 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 1019 struct iovec *iov, unsigned int iov_cnt) 1020 { 1021 uint8_t on; 1022 size_t s; 1023 NetClientState *nc = qemu_get_queue(n->nic); 1024 1025 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 1026 if (s != sizeof(on)) { 1027 return VIRTIO_NET_ERR; 1028 } 1029 1030 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 1031 n->promisc = on; 1032 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 1033 n->allmulti = on; 1034 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 1035 n->alluni = on; 1036 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 1037 n->nomulti = on; 1038 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 1039 n->nouni = on; 1040 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 1041 n->nobcast = on; 1042 } else { 1043 return VIRTIO_NET_ERR; 1044 } 1045 1046 rxfilter_notify(nc); 1047 1048 return VIRTIO_NET_OK; 1049 } 1050 1051 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 1052 struct iovec *iov, unsigned int iov_cnt) 1053 { 1054 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1055 uint64_t offloads; 1056 size_t s; 1057 1058 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 1059 return VIRTIO_NET_ERR; 1060 } 1061 1062 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 1063 if (s != sizeof(offloads)) { 1064 return VIRTIO_NET_ERR; 1065 } 1066 1067 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 1068 uint64_t supported_offloads; 1069 1070 offloads = virtio_ldq_p(vdev, &offloads); 1071 1072 if (!n->has_vnet_hdr) { 1073 return VIRTIO_NET_ERR; 1074 } 1075 1076 n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) && 1077 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4); 1078 n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) && 1079 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6); 1080 virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT); 1081 1082 supported_offloads = virtio_net_supported_guest_offloads(n); 1083 if (offloads & ~supported_offloads) { 1084 return VIRTIO_NET_ERR; 1085 } 1086 1087 n->curr_guest_offloads = offloads; 1088 virtio_net_apply_guest_offloads(n); 1089 1090 return VIRTIO_NET_OK; 1091 } else { 1092 return VIRTIO_NET_ERR; 1093 } 1094 } 1095 1096 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 1097 struct iovec *iov, unsigned int iov_cnt) 1098 { 1099 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1100 struct virtio_net_ctrl_mac mac_data; 1101 size_t s; 1102 NetClientState *nc = qemu_get_queue(n->nic); 1103 1104 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 1105 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 1106 return VIRTIO_NET_ERR; 1107 } 1108 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 1109 assert(s == sizeof(n->mac)); 1110 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 1111 rxfilter_notify(nc); 1112 1113 return VIRTIO_NET_OK; 1114 } 1115 1116 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 1117 return VIRTIO_NET_ERR; 1118 } 1119 1120 int in_use = 0; 1121 int first_multi = 0; 1122 uint8_t uni_overflow = 0; 1123 uint8_t multi_overflow = 0; 1124 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 1125 1126 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 1127 sizeof(mac_data.entries)); 1128 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 1129 if (s != sizeof(mac_data.entries)) { 1130 goto error; 1131 } 1132 iov_discard_front(&iov, &iov_cnt, s); 1133 1134 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 1135 goto error; 1136 } 1137 1138 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 1139 s = iov_to_buf(iov, iov_cnt, 0, macs, 1140 mac_data.entries * ETH_ALEN); 1141 if (s != mac_data.entries * ETH_ALEN) { 1142 goto error; 1143 } 1144 in_use += mac_data.entries; 1145 } else { 1146 uni_overflow = 1; 1147 } 1148 1149 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 1150 1151 first_multi = in_use; 1152 1153 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 1154 sizeof(mac_data.entries)); 1155 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 1156 if (s != sizeof(mac_data.entries)) { 1157 goto error; 1158 } 1159 1160 iov_discard_front(&iov, &iov_cnt, s); 1161 1162 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 1163 goto error; 1164 } 1165 1166 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) { 1167 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN], 1168 mac_data.entries * ETH_ALEN); 1169 if (s != mac_data.entries * ETH_ALEN) { 1170 goto error; 1171 } 1172 in_use += mac_data.entries; 1173 } else { 1174 multi_overflow = 1; 1175 } 1176 1177 n->mac_table.in_use = in_use; 1178 n->mac_table.first_multi = first_multi; 1179 n->mac_table.uni_overflow = uni_overflow; 1180 n->mac_table.multi_overflow = multi_overflow; 1181 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN); 1182 g_free(macs); 1183 rxfilter_notify(nc); 1184 1185 return VIRTIO_NET_OK; 1186 1187 error: 1188 g_free(macs); 1189 return VIRTIO_NET_ERR; 1190 } 1191 1192 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 1193 struct iovec *iov, unsigned int iov_cnt) 1194 { 1195 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1196 uint16_t vid; 1197 size_t s; 1198 NetClientState *nc = qemu_get_queue(n->nic); 1199 1200 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 1201 vid = virtio_lduw_p(vdev, &vid); 1202 if (s != sizeof(vid)) { 1203 return VIRTIO_NET_ERR; 1204 } 1205 1206 if (vid >= MAX_VLAN) 1207 return VIRTIO_NET_ERR; 1208 1209 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 1210 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 1211 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 1212 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 1213 else 1214 return VIRTIO_NET_ERR; 1215 1216 rxfilter_notify(nc); 1217 1218 return VIRTIO_NET_OK; 1219 } 1220 1221 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd, 1222 struct iovec *iov, unsigned int iov_cnt) 1223 { 1224 trace_virtio_net_handle_announce(n->announce_timer.round); 1225 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK && 1226 n->status & VIRTIO_NET_S_ANNOUNCE) { 1227 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 1228 if (n->announce_timer.round) { 1229 qemu_announce_timer_step(&n->announce_timer); 1230 } 1231 return VIRTIO_NET_OK; 1232 } else { 1233 return VIRTIO_NET_ERR; 1234 } 1235 } 1236 1237 static bool virtio_net_attach_ebpf_to_backend(NICState *nic, int prog_fd) 1238 { 1239 NetClientState *nc = qemu_get_peer(qemu_get_queue(nic), 0); 1240 if (nc == NULL || nc->info->set_steering_ebpf == NULL) { 1241 return false; 1242 } 1243 1244 trace_virtio_net_rss_attach_ebpf(nic, prog_fd); 1245 return nc->info->set_steering_ebpf(nc, prog_fd); 1246 } 1247 1248 static void rss_data_to_rss_config(struct VirtioNetRssData *data, 1249 struct EBPFRSSConfig *config) 1250 { 1251 config->redirect = data->redirect; 1252 config->populate_hash = data->populate_hash; 1253 config->hash_types = data->hash_types; 1254 config->indirections_len = data->indirections_len; 1255 config->default_queue = data->default_queue; 1256 } 1257 1258 static bool virtio_net_attach_ebpf_rss(VirtIONet *n) 1259 { 1260 struct EBPFRSSConfig config = {}; 1261 1262 if (!ebpf_rss_is_loaded(&n->ebpf_rss)) { 1263 return false; 1264 } 1265 1266 rss_data_to_rss_config(&n->rss_data, &config); 1267 1268 if (!ebpf_rss_set_all(&n->ebpf_rss, &config, 1269 n->rss_data.indirections_table, n->rss_data.key, 1270 NULL)) { 1271 return false; 1272 } 1273 1274 if (!virtio_net_attach_ebpf_to_backend(n->nic, n->ebpf_rss.program_fd)) { 1275 return false; 1276 } 1277 1278 return true; 1279 } 1280 1281 static void virtio_net_detach_ebpf_rss(VirtIONet *n) 1282 { 1283 virtio_net_attach_ebpf_to_backend(n->nic, -1); 1284 } 1285 1286 static void virtio_net_commit_rss_config(VirtIONet *n) 1287 { 1288 if (n->rss_data.enabled) { 1289 n->rss_data.enabled_software_rss = n->rss_data.populate_hash; 1290 if (n->rss_data.populate_hash) { 1291 virtio_net_detach_ebpf_rss(n); 1292 } else if (!virtio_net_attach_ebpf_rss(n)) { 1293 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) { 1294 warn_report("Can't load eBPF RSS for vhost"); 1295 } else { 1296 warn_report("Can't load eBPF RSS - fallback to software RSS"); 1297 n->rss_data.enabled_software_rss = true; 1298 } 1299 } 1300 1301 trace_virtio_net_rss_enable(n, 1302 n->rss_data.hash_types, 1303 n->rss_data.indirections_len, 1304 sizeof(n->rss_data.key)); 1305 } else { 1306 virtio_net_detach_ebpf_rss(n); 1307 trace_virtio_net_rss_disable(n); 1308 } 1309 } 1310 1311 static void virtio_net_disable_rss(VirtIONet *n) 1312 { 1313 if (!n->rss_data.enabled) { 1314 return; 1315 } 1316 1317 n->rss_data.enabled = false; 1318 virtio_net_commit_rss_config(n); 1319 } 1320 1321 static bool virtio_net_load_ebpf_fds(VirtIONet *n, Error **errp) 1322 { 1323 int fds[EBPF_RSS_MAX_FDS] = { [0 ... EBPF_RSS_MAX_FDS - 1] = -1}; 1324 int ret = true; 1325 int i = 0; 1326 1327 if (n->nr_ebpf_rss_fds != EBPF_RSS_MAX_FDS) { 1328 error_setg(errp, "Expected %d file descriptors but got %d", 1329 EBPF_RSS_MAX_FDS, n->nr_ebpf_rss_fds); 1330 return false; 1331 } 1332 1333 for (i = 0; i < n->nr_ebpf_rss_fds; i++) { 1334 fds[i] = monitor_fd_param(monitor_cur(), n->ebpf_rss_fds[i], errp); 1335 if (fds[i] < 0) { 1336 ret = false; 1337 goto exit; 1338 } 1339 } 1340 1341 ret = ebpf_rss_load_fds(&n->ebpf_rss, fds[0], fds[1], fds[2], fds[3], errp); 1342 1343 exit: 1344 if (!ret) { 1345 for (i = 0; i < n->nr_ebpf_rss_fds && fds[i] != -1; i++) { 1346 close(fds[i]); 1347 } 1348 } 1349 1350 return ret; 1351 } 1352 1353 static bool virtio_net_load_ebpf(VirtIONet *n, Error **errp) 1354 { 1355 bool ret = false; 1356 1357 if (virtio_net_attach_ebpf_to_backend(n->nic, -1)) { 1358 trace_virtio_net_rss_load(n, n->nr_ebpf_rss_fds, n->ebpf_rss_fds); 1359 if (n->ebpf_rss_fds) { 1360 ret = virtio_net_load_ebpf_fds(n, errp); 1361 } else { 1362 ret = ebpf_rss_load(&n->ebpf_rss, errp); 1363 } 1364 } 1365 1366 return ret; 1367 } 1368 1369 static void virtio_net_unload_ebpf(VirtIONet *n) 1370 { 1371 virtio_net_attach_ebpf_to_backend(n->nic, -1); 1372 ebpf_rss_unload(&n->ebpf_rss); 1373 } 1374 1375 static uint16_t virtio_net_handle_rss(VirtIONet *n, 1376 struct iovec *iov, 1377 unsigned int iov_cnt, 1378 bool do_rss) 1379 { 1380 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1381 struct virtio_net_rss_config cfg; 1382 size_t s, offset = 0, size_get; 1383 uint16_t queue_pairs, i; 1384 struct { 1385 uint16_t us; 1386 uint8_t b; 1387 } QEMU_PACKED temp; 1388 const char *err_msg = ""; 1389 uint32_t err_value = 0; 1390 1391 if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) { 1392 err_msg = "RSS is not negotiated"; 1393 goto error; 1394 } 1395 if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) { 1396 err_msg = "Hash report is not negotiated"; 1397 goto error; 1398 } 1399 size_get = offsetof(struct virtio_net_rss_config, indirection_table); 1400 s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get); 1401 if (s != size_get) { 1402 err_msg = "Short command buffer"; 1403 err_value = (uint32_t)s; 1404 goto error; 1405 } 1406 n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types); 1407 n->rss_data.indirections_len = 1408 virtio_lduw_p(vdev, &cfg.indirection_table_mask); 1409 if (!do_rss) { 1410 n->rss_data.indirections_len = 0; 1411 } 1412 if (n->rss_data.indirections_len >= VIRTIO_NET_RSS_MAX_TABLE_LEN) { 1413 err_msg = "Too large indirection table"; 1414 err_value = n->rss_data.indirections_len; 1415 goto error; 1416 } 1417 n->rss_data.indirections_len++; 1418 if (!is_power_of_2(n->rss_data.indirections_len)) { 1419 err_msg = "Invalid size of indirection table"; 1420 err_value = n->rss_data.indirections_len; 1421 goto error; 1422 } 1423 n->rss_data.default_queue = do_rss ? 1424 virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0; 1425 if (n->rss_data.default_queue >= n->max_queue_pairs) { 1426 err_msg = "Invalid default queue"; 1427 err_value = n->rss_data.default_queue; 1428 goto error; 1429 } 1430 offset += size_get; 1431 size_get = sizeof(uint16_t) * n->rss_data.indirections_len; 1432 g_free(n->rss_data.indirections_table); 1433 n->rss_data.indirections_table = g_malloc(size_get); 1434 if (!n->rss_data.indirections_table) { 1435 err_msg = "Can't allocate indirections table"; 1436 err_value = n->rss_data.indirections_len; 1437 goto error; 1438 } 1439 s = iov_to_buf(iov, iov_cnt, offset, 1440 n->rss_data.indirections_table, size_get); 1441 if (s != size_get) { 1442 err_msg = "Short indirection table buffer"; 1443 err_value = (uint32_t)s; 1444 goto error; 1445 } 1446 for (i = 0; i < n->rss_data.indirections_len; ++i) { 1447 uint16_t val = n->rss_data.indirections_table[i]; 1448 n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val); 1449 } 1450 offset += size_get; 1451 size_get = sizeof(temp); 1452 s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get); 1453 if (s != size_get) { 1454 err_msg = "Can't get queue_pairs"; 1455 err_value = (uint32_t)s; 1456 goto error; 1457 } 1458 queue_pairs = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queue_pairs; 1459 if (queue_pairs == 0 || queue_pairs > n->max_queue_pairs) { 1460 err_msg = "Invalid number of queue_pairs"; 1461 err_value = queue_pairs; 1462 goto error; 1463 } 1464 if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) { 1465 err_msg = "Invalid key size"; 1466 err_value = temp.b; 1467 goto error; 1468 } 1469 if (!temp.b && n->rss_data.hash_types) { 1470 err_msg = "No key provided"; 1471 err_value = 0; 1472 goto error; 1473 } 1474 if (!temp.b && !n->rss_data.hash_types) { 1475 virtio_net_disable_rss(n); 1476 return queue_pairs; 1477 } 1478 offset += size_get; 1479 size_get = temp.b; 1480 s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get); 1481 if (s != size_get) { 1482 err_msg = "Can get key buffer"; 1483 err_value = (uint32_t)s; 1484 goto error; 1485 } 1486 n->rss_data.enabled = true; 1487 virtio_net_commit_rss_config(n); 1488 return queue_pairs; 1489 error: 1490 trace_virtio_net_rss_error(n, err_msg, err_value); 1491 virtio_net_disable_rss(n); 1492 return 0; 1493 } 1494 1495 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 1496 struct iovec *iov, unsigned int iov_cnt) 1497 { 1498 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1499 uint16_t queue_pairs; 1500 NetClientState *nc = qemu_get_queue(n->nic); 1501 1502 virtio_net_disable_rss(n); 1503 if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) { 1504 queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, false); 1505 return queue_pairs ? VIRTIO_NET_OK : VIRTIO_NET_ERR; 1506 } 1507 if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) { 1508 queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, true); 1509 } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 1510 struct virtio_net_ctrl_mq mq; 1511 size_t s; 1512 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) { 1513 return VIRTIO_NET_ERR; 1514 } 1515 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 1516 if (s != sizeof(mq)) { 1517 return VIRTIO_NET_ERR; 1518 } 1519 queue_pairs = virtio_lduw_p(vdev, &mq.virtqueue_pairs); 1520 1521 } else { 1522 return VIRTIO_NET_ERR; 1523 } 1524 1525 if (queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 1526 queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 1527 queue_pairs > n->max_queue_pairs || 1528 !n->multiqueue) { 1529 return VIRTIO_NET_ERR; 1530 } 1531 1532 n->curr_queue_pairs = queue_pairs; 1533 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 1534 /* 1535 * Avoid updating the backend for a vdpa device: We're only interested 1536 * in updating the device model queues. 1537 */ 1538 return VIRTIO_NET_OK; 1539 } 1540 /* stop the backend before changing the number of queue_pairs to avoid handling a 1541 * disabled queue */ 1542 virtio_net_set_status(vdev, vdev->status); 1543 virtio_net_set_queue_pairs(n); 1544 1545 return VIRTIO_NET_OK; 1546 } 1547 1548 size_t virtio_net_handle_ctrl_iov(VirtIODevice *vdev, 1549 const struct iovec *in_sg, unsigned in_num, 1550 const struct iovec *out_sg, 1551 unsigned out_num) 1552 { 1553 VirtIONet *n = VIRTIO_NET(vdev); 1554 struct virtio_net_ctrl_hdr ctrl; 1555 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 1556 size_t s; 1557 struct iovec *iov, *iov2; 1558 1559 if (iov_size(in_sg, in_num) < sizeof(status) || 1560 iov_size(out_sg, out_num) < sizeof(ctrl)) { 1561 virtio_error(vdev, "virtio-net ctrl missing headers"); 1562 return 0; 1563 } 1564 1565 iov2 = iov = g_memdup2(out_sg, sizeof(struct iovec) * out_num); 1566 s = iov_to_buf(iov, out_num, 0, &ctrl, sizeof(ctrl)); 1567 iov_discard_front(&iov, &out_num, sizeof(ctrl)); 1568 if (s != sizeof(ctrl)) { 1569 status = VIRTIO_NET_ERR; 1570 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 1571 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, out_num); 1572 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 1573 status = virtio_net_handle_mac(n, ctrl.cmd, iov, out_num); 1574 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 1575 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, out_num); 1576 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) { 1577 status = virtio_net_handle_announce(n, ctrl.cmd, iov, out_num); 1578 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 1579 status = virtio_net_handle_mq(n, ctrl.cmd, iov, out_num); 1580 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 1581 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, out_num); 1582 } 1583 1584 s = iov_from_buf(in_sg, in_num, 0, &status, sizeof(status)); 1585 assert(s == sizeof(status)); 1586 1587 g_free(iov2); 1588 return sizeof(status); 1589 } 1590 1591 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 1592 { 1593 VirtQueueElement *elem; 1594 1595 for (;;) { 1596 size_t written; 1597 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 1598 if (!elem) { 1599 break; 1600 } 1601 1602 written = virtio_net_handle_ctrl_iov(vdev, elem->in_sg, elem->in_num, 1603 elem->out_sg, elem->out_num); 1604 if (written > 0) { 1605 virtqueue_push(vq, elem, written); 1606 virtio_notify(vdev, vq); 1607 g_free(elem); 1608 } else { 1609 virtqueue_detach_element(vq, elem, 0); 1610 g_free(elem); 1611 break; 1612 } 1613 } 1614 } 1615 1616 /* RX */ 1617 1618 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 1619 { 1620 VirtIONet *n = VIRTIO_NET(vdev); 1621 int queue_index = vq2q(virtio_get_queue_index(vq)); 1622 1623 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 1624 } 1625 1626 static bool virtio_net_can_receive(NetClientState *nc) 1627 { 1628 VirtIONet *n = qemu_get_nic_opaque(nc); 1629 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1630 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1631 1632 if (!vdev->vm_running) { 1633 return false; 1634 } 1635 1636 if (nc->queue_index >= n->curr_queue_pairs) { 1637 return false; 1638 } 1639 1640 if (!virtio_queue_ready(q->rx_vq) || 1641 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1642 return false; 1643 } 1644 1645 return true; 1646 } 1647 1648 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 1649 { 1650 int opaque; 1651 unsigned int in_bytes; 1652 VirtIONet *n = q->n; 1653 1654 while (virtio_queue_empty(q->rx_vq) || n->mergeable_rx_bufs) { 1655 opaque = virtqueue_get_avail_bytes(q->rx_vq, &in_bytes, NULL, 1656 bufsize, 0); 1657 /* Buffer is enough, disable notifiaction */ 1658 if (bufsize <= in_bytes) { 1659 break; 1660 } 1661 1662 if (virtio_queue_enable_notification_and_check(q->rx_vq, opaque)) { 1663 /* Guest has added some buffers, try again */ 1664 continue; 1665 } else { 1666 return 0; 1667 } 1668 } 1669 1670 virtio_queue_set_notification(q->rx_vq, 0); 1671 1672 return 1; 1673 } 1674 1675 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr) 1676 { 1677 virtio_tswap16s(vdev, &hdr->hdr_len); 1678 virtio_tswap16s(vdev, &hdr->gso_size); 1679 virtio_tswap16s(vdev, &hdr->csum_start); 1680 virtio_tswap16s(vdev, &hdr->csum_offset); 1681 } 1682 1683 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 1684 * it never finds out that the packets don't have valid checksums. This 1685 * causes dhclient to get upset. Fedora's carried a patch for ages to 1686 * fix this with Xen but it hasn't appeared in an upstream release of 1687 * dhclient yet. 1688 * 1689 * To avoid breaking existing guests, we catch udp packets and add 1690 * checksums. This is terrible but it's better than hacking the guest 1691 * kernels. 1692 * 1693 * N.B. if we introduce a zero-copy API, this operation is no longer free so 1694 * we should provide a mechanism to disable it to avoid polluting the host 1695 * cache. 1696 */ 1697 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 1698 size_t *hdr_len, const uint8_t *buf, 1699 size_t buf_size, size_t *buf_offset) 1700 { 1701 size_t csum_size = ETH_HLEN + sizeof(struct ip_header) + 1702 sizeof(struct udp_header); 1703 1704 buf += *buf_offset; 1705 buf_size -= *buf_offset; 1706 1707 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 1708 (buf_size >= csum_size && buf_size < 1500) && /* normal sized MTU */ 1709 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 1710 (buf[23] == 17) && /* ip.protocol == UDP */ 1711 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 1712 memcpy((uint8_t *)hdr + *hdr_len, buf, csum_size); 1713 net_checksum_calculate((uint8_t *)hdr + *hdr_len, csum_size, CSUM_UDP); 1714 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 1715 *hdr_len += csum_size; 1716 *buf_offset += csum_size; 1717 } 1718 } 1719 1720 static size_t receive_header(VirtIONet *n, struct virtio_net_hdr *hdr, 1721 const void *buf, size_t buf_size, 1722 size_t *buf_offset) 1723 { 1724 size_t hdr_len = n->guest_hdr_len; 1725 1726 memcpy(hdr, buf, sizeof(struct virtio_net_hdr)); 1727 1728 *buf_offset = n->host_hdr_len; 1729 work_around_broken_dhclient(hdr, &hdr_len, buf, buf_size, buf_offset); 1730 1731 if (n->needs_vnet_hdr_swap) { 1732 virtio_net_hdr_swap(VIRTIO_DEVICE(n), hdr); 1733 } 1734 1735 return hdr_len; 1736 } 1737 1738 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 1739 { 1740 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 1741 static const uint8_t vlan[] = {0x81, 0x00}; 1742 uint8_t *ptr = (uint8_t *)buf; 1743 int i; 1744 1745 if (n->promisc) 1746 return 1; 1747 1748 ptr += n->host_hdr_len; 1749 1750 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 1751 int vid = lduw_be_p(ptr + 14) & 0xfff; 1752 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 1753 return 0; 1754 } 1755 1756 if (ptr[0] & 1) { // multicast 1757 if (!memcmp(ptr, bcast, sizeof(bcast))) { 1758 return !n->nobcast; 1759 } else if (n->nomulti) { 1760 return 0; 1761 } else if (n->allmulti || n->mac_table.multi_overflow) { 1762 return 1; 1763 } 1764 1765 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 1766 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1767 return 1; 1768 } 1769 } 1770 } else { // unicast 1771 if (n->nouni) { 1772 return 0; 1773 } else if (n->alluni || n->mac_table.uni_overflow) { 1774 return 1; 1775 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 1776 return 1; 1777 } 1778 1779 for (i = 0; i < n->mac_table.first_multi; i++) { 1780 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1781 return 1; 1782 } 1783 } 1784 } 1785 1786 return 0; 1787 } 1788 1789 static uint8_t virtio_net_get_hash_type(bool hasip4, 1790 bool hasip6, 1791 EthL4HdrProto l4hdr_proto, 1792 uint32_t types) 1793 { 1794 if (hasip4) { 1795 switch (l4hdr_proto) { 1796 case ETH_L4_HDR_PROTO_TCP: 1797 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) { 1798 return NetPktRssIpV4Tcp; 1799 } 1800 break; 1801 1802 case ETH_L4_HDR_PROTO_UDP: 1803 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) { 1804 return NetPktRssIpV4Udp; 1805 } 1806 break; 1807 1808 default: 1809 break; 1810 } 1811 1812 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) { 1813 return NetPktRssIpV4; 1814 } 1815 } else if (hasip6) { 1816 switch (l4hdr_proto) { 1817 case ETH_L4_HDR_PROTO_TCP: 1818 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) { 1819 return NetPktRssIpV6TcpEx; 1820 } 1821 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) { 1822 return NetPktRssIpV6Tcp; 1823 } 1824 break; 1825 1826 case ETH_L4_HDR_PROTO_UDP: 1827 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) { 1828 return NetPktRssIpV6UdpEx; 1829 } 1830 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) { 1831 return NetPktRssIpV6Udp; 1832 } 1833 break; 1834 1835 default: 1836 break; 1837 } 1838 1839 if (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) { 1840 return NetPktRssIpV6Ex; 1841 } 1842 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv6) { 1843 return NetPktRssIpV6; 1844 } 1845 } 1846 return 0xff; 1847 } 1848 1849 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf, 1850 size_t size, 1851 struct virtio_net_hdr_v1_hash *hdr) 1852 { 1853 VirtIONet *n = qemu_get_nic_opaque(nc); 1854 unsigned int index = nc->queue_index, new_index = index; 1855 struct NetRxPkt *pkt = n->rx_pkt; 1856 uint8_t net_hash_type; 1857 uint32_t hash; 1858 bool hasip4, hasip6; 1859 EthL4HdrProto l4hdr_proto; 1860 static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = { 1861 VIRTIO_NET_HASH_REPORT_IPv4, 1862 VIRTIO_NET_HASH_REPORT_TCPv4, 1863 VIRTIO_NET_HASH_REPORT_TCPv6, 1864 VIRTIO_NET_HASH_REPORT_IPv6, 1865 VIRTIO_NET_HASH_REPORT_IPv6_EX, 1866 VIRTIO_NET_HASH_REPORT_TCPv6_EX, 1867 VIRTIO_NET_HASH_REPORT_UDPv4, 1868 VIRTIO_NET_HASH_REPORT_UDPv6, 1869 VIRTIO_NET_HASH_REPORT_UDPv6_EX 1870 }; 1871 struct iovec iov = { 1872 .iov_base = (void *)buf, 1873 .iov_len = size 1874 }; 1875 1876 net_rx_pkt_set_protocols(pkt, &iov, 1, n->host_hdr_len); 1877 net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto); 1878 net_hash_type = virtio_net_get_hash_type(hasip4, hasip6, l4hdr_proto, 1879 n->rss_data.hash_types); 1880 if (net_hash_type > NetPktRssIpV6UdpEx) { 1881 if (n->rss_data.populate_hash) { 1882 hdr->hash_value = VIRTIO_NET_HASH_REPORT_NONE; 1883 hdr->hash_report = 0; 1884 } 1885 return n->rss_data.redirect ? n->rss_data.default_queue : -1; 1886 } 1887 1888 hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key); 1889 1890 if (n->rss_data.populate_hash) { 1891 hdr->hash_value = hash; 1892 hdr->hash_report = reports[net_hash_type]; 1893 } 1894 1895 if (n->rss_data.redirect) { 1896 new_index = hash & (n->rss_data.indirections_len - 1); 1897 new_index = n->rss_data.indirections_table[new_index]; 1898 } 1899 1900 return (index == new_index) ? -1 : new_index; 1901 } 1902 1903 typedef struct Header { 1904 struct virtio_net_hdr_v1_hash virtio_net; 1905 struct eth_header eth; 1906 struct ip_header ip; 1907 struct udp_header udp; 1908 } Header; 1909 1910 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, 1911 size_t size) 1912 { 1913 VirtIONet *n = qemu_get_nic_opaque(nc); 1914 VirtIONetQueue *q; 1915 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1916 VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE]; 1917 size_t lens[VIRTQUEUE_MAX_SIZE]; 1918 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 1919 Header hdr; 1920 unsigned mhdr_cnt = 0; 1921 size_t offset, i, guest_offset, j; 1922 ssize_t err; 1923 1924 memset(&hdr.virtio_net, 0, sizeof(hdr.virtio_net)); 1925 1926 if (n->rss_data.enabled && n->rss_data.enabled_software_rss) { 1927 int index = virtio_net_process_rss(nc, buf, size, &hdr.virtio_net); 1928 if (index >= 0) { 1929 nc = qemu_get_subqueue(n->nic, index % n->curr_queue_pairs); 1930 } 1931 } 1932 1933 if (!virtio_net_can_receive(nc)) { 1934 return -1; 1935 } 1936 1937 q = virtio_net_get_subqueue(nc); 1938 1939 /* hdr_len refers to the header we supply to the guest */ 1940 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 1941 return 0; 1942 } 1943 1944 if (!receive_filter(n, buf, size)) 1945 return size; 1946 1947 offset = i = 0; 1948 1949 while (offset < size) { 1950 VirtQueueElement *elem; 1951 int len, total; 1952 const struct iovec *sg; 1953 1954 total = 0; 1955 1956 if (i == VIRTQUEUE_MAX_SIZE) { 1957 virtio_error(vdev, "virtio-net unexpected long buffer chain"); 1958 err = size; 1959 goto err; 1960 } 1961 1962 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement)); 1963 if (!elem) { 1964 if (i) { 1965 virtio_error(vdev, "virtio-net unexpected empty queue: " 1966 "i %zd mergeable %d offset %zd, size %zd, " 1967 "guest hdr len %zd, host hdr len %zd " 1968 "guest features 0x%" PRIx64, 1969 i, n->mergeable_rx_bufs, offset, size, 1970 n->guest_hdr_len, n->host_hdr_len, 1971 vdev->guest_features); 1972 } 1973 err = -1; 1974 goto err; 1975 } 1976 1977 if (elem->in_num < 1) { 1978 virtio_error(vdev, 1979 "virtio-net receive queue contains no in buffers"); 1980 virtqueue_detach_element(q->rx_vq, elem, 0); 1981 g_free(elem); 1982 err = -1; 1983 goto err; 1984 } 1985 1986 sg = elem->in_sg; 1987 if (i == 0) { 1988 assert(offset == 0); 1989 if (n->mergeable_rx_bufs) { 1990 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 1991 sg, elem->in_num, 1992 offsetof(typeof(hdr), 1993 virtio_net.hdr.num_buffers), 1994 sizeof(hdr.virtio_net.hdr.num_buffers)); 1995 } 1996 1997 guest_offset = n->has_vnet_hdr ? 1998 receive_header(n, (struct virtio_net_hdr *)&hdr, 1999 buf, size, &offset) : 2000 n->guest_hdr_len; 2001 2002 iov_from_buf(sg, elem->in_num, 0, &hdr, guest_offset); 2003 total += guest_offset; 2004 } else { 2005 guest_offset = 0; 2006 } 2007 2008 /* copy in packet. ugh */ 2009 len = iov_from_buf(sg, elem->in_num, guest_offset, 2010 buf + offset, size - offset); 2011 total += len; 2012 offset += len; 2013 /* If buffers can't be merged, at this point we 2014 * must have consumed the complete packet. 2015 * Otherwise, drop it. */ 2016 if (!n->mergeable_rx_bufs && offset < size) { 2017 virtqueue_unpop(q->rx_vq, elem, total); 2018 g_free(elem); 2019 err = size; 2020 goto err; 2021 } 2022 2023 elems[i] = elem; 2024 lens[i] = total; 2025 i++; 2026 } 2027 2028 if (mhdr_cnt) { 2029 virtio_stw_p(vdev, &hdr.virtio_net.hdr.num_buffers, i); 2030 iov_from_buf(mhdr_sg, mhdr_cnt, 2031 0, 2032 &hdr.virtio_net.hdr.num_buffers, 2033 sizeof hdr.virtio_net.hdr.num_buffers); 2034 } 2035 2036 for (j = 0; j < i; j++) { 2037 /* signal other side */ 2038 virtqueue_fill(q->rx_vq, elems[j], lens[j], j); 2039 g_free(elems[j]); 2040 } 2041 2042 virtqueue_flush(q->rx_vq, i); 2043 virtio_notify(vdev, q->rx_vq); 2044 2045 return size; 2046 2047 err: 2048 for (j = 0; j < i; j++) { 2049 virtqueue_detach_element(q->rx_vq, elems[j], lens[j]); 2050 g_free(elems[j]); 2051 } 2052 2053 return err; 2054 } 2055 2056 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf, 2057 size_t size) 2058 { 2059 RCU_READ_LOCK_GUARD(); 2060 2061 return virtio_net_receive_rcu(nc, buf, size); 2062 } 2063 2064 /* 2065 * Accessors to read and write the IP packet data length field. This 2066 * is a potentially unaligned network-byte-order 16 bit unsigned integer 2067 * pointed to by unit->ip_len. 2068 */ 2069 static uint16_t read_unit_ip_len(VirtioNetRscUnit *unit) 2070 { 2071 return lduw_be_p(unit->ip_plen); 2072 } 2073 2074 static void write_unit_ip_len(VirtioNetRscUnit *unit, uint16_t l) 2075 { 2076 stw_be_p(unit->ip_plen, l); 2077 } 2078 2079 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain, 2080 const uint8_t *buf, 2081 VirtioNetRscUnit *unit) 2082 { 2083 uint16_t ip_hdrlen; 2084 struct ip_header *ip; 2085 2086 ip = (struct ip_header *)(buf + chain->n->guest_hdr_len 2087 + sizeof(struct eth_header)); 2088 unit->ip = (void *)ip; 2089 ip_hdrlen = (ip->ip_ver_len & 0xF) << 2; 2090 unit->ip_plen = &ip->ip_len; 2091 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen); 2092 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10; 2093 unit->payload = read_unit_ip_len(unit) - ip_hdrlen - unit->tcp_hdrlen; 2094 } 2095 2096 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain, 2097 const uint8_t *buf, 2098 VirtioNetRscUnit *unit) 2099 { 2100 struct ip6_header *ip6; 2101 2102 ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len 2103 + sizeof(struct eth_header)); 2104 unit->ip = ip6; 2105 unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen); 2106 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) 2107 + sizeof(struct ip6_header)); 2108 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10; 2109 2110 /* There is a difference between payload length in ipv4 and v6, 2111 ip header is excluded in ipv6 */ 2112 unit->payload = read_unit_ip_len(unit) - unit->tcp_hdrlen; 2113 } 2114 2115 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain, 2116 VirtioNetRscSeg *seg) 2117 { 2118 int ret; 2119 struct virtio_net_hdr_v1 *h; 2120 2121 h = (struct virtio_net_hdr_v1 *)seg->buf; 2122 h->flags = 0; 2123 h->gso_type = VIRTIO_NET_HDR_GSO_NONE; 2124 2125 if (seg->is_coalesced) { 2126 h->rsc.segments = seg->packets; 2127 h->rsc.dup_acks = seg->dup_ack; 2128 h->flags = VIRTIO_NET_HDR_F_RSC_INFO; 2129 if (chain->proto == ETH_P_IP) { 2130 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2131 } else { 2132 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2133 } 2134 } 2135 2136 ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size); 2137 QTAILQ_REMOVE(&chain->buffers, seg, next); 2138 g_free(seg->buf); 2139 g_free(seg); 2140 2141 return ret; 2142 } 2143 2144 static void virtio_net_rsc_purge(void *opq) 2145 { 2146 VirtioNetRscSeg *seg, *rn; 2147 VirtioNetRscChain *chain = (VirtioNetRscChain *)opq; 2148 2149 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) { 2150 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2151 chain->stat.purge_failed++; 2152 continue; 2153 } 2154 } 2155 2156 chain->stat.timer++; 2157 if (!QTAILQ_EMPTY(&chain->buffers)) { 2158 timer_mod(chain->drain_timer, 2159 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + chain->n->rsc_timeout); 2160 } 2161 } 2162 2163 static void virtio_net_rsc_cleanup(VirtIONet *n) 2164 { 2165 VirtioNetRscChain *chain, *rn_chain; 2166 VirtioNetRscSeg *seg, *rn_seg; 2167 2168 QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) { 2169 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) { 2170 QTAILQ_REMOVE(&chain->buffers, seg, next); 2171 g_free(seg->buf); 2172 g_free(seg); 2173 } 2174 2175 timer_free(chain->drain_timer); 2176 QTAILQ_REMOVE(&n->rsc_chains, chain, next); 2177 g_free(chain); 2178 } 2179 } 2180 2181 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain, 2182 NetClientState *nc, 2183 const uint8_t *buf, size_t size) 2184 { 2185 uint16_t hdr_len; 2186 VirtioNetRscSeg *seg; 2187 2188 hdr_len = chain->n->guest_hdr_len; 2189 seg = g_new(VirtioNetRscSeg, 1); 2190 seg->buf = g_malloc(hdr_len + sizeof(struct eth_header) 2191 + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD); 2192 memcpy(seg->buf, buf, size); 2193 seg->size = size; 2194 seg->packets = 1; 2195 seg->dup_ack = 0; 2196 seg->is_coalesced = 0; 2197 seg->nc = nc; 2198 2199 QTAILQ_INSERT_TAIL(&chain->buffers, seg, next); 2200 chain->stat.cache++; 2201 2202 switch (chain->proto) { 2203 case ETH_P_IP: 2204 virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit); 2205 break; 2206 case ETH_P_IPV6: 2207 virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit); 2208 break; 2209 default: 2210 g_assert_not_reached(); 2211 } 2212 } 2213 2214 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain, 2215 VirtioNetRscSeg *seg, 2216 const uint8_t *buf, 2217 struct tcp_header *n_tcp, 2218 struct tcp_header *o_tcp) 2219 { 2220 uint32_t nack, oack; 2221 uint16_t nwin, owin; 2222 2223 nack = htonl(n_tcp->th_ack); 2224 nwin = htons(n_tcp->th_win); 2225 oack = htonl(o_tcp->th_ack); 2226 owin = htons(o_tcp->th_win); 2227 2228 if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) { 2229 chain->stat.ack_out_of_win++; 2230 return RSC_FINAL; 2231 } else if (nack == oack) { 2232 /* duplicated ack or window probe */ 2233 if (nwin == owin) { 2234 /* duplicated ack, add dup ack count due to whql test up to 1 */ 2235 chain->stat.dup_ack++; 2236 return RSC_FINAL; 2237 } else { 2238 /* Coalesce window update */ 2239 o_tcp->th_win = n_tcp->th_win; 2240 chain->stat.win_update++; 2241 return RSC_COALESCE; 2242 } 2243 } else { 2244 /* pure ack, go to 'C', finalize*/ 2245 chain->stat.pure_ack++; 2246 return RSC_FINAL; 2247 } 2248 } 2249 2250 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain, 2251 VirtioNetRscSeg *seg, 2252 const uint8_t *buf, 2253 VirtioNetRscUnit *n_unit) 2254 { 2255 void *data; 2256 uint16_t o_ip_len; 2257 uint32_t nseq, oseq; 2258 VirtioNetRscUnit *o_unit; 2259 2260 o_unit = &seg->unit; 2261 o_ip_len = read_unit_ip_len(o_unit); 2262 nseq = htonl(n_unit->tcp->th_seq); 2263 oseq = htonl(o_unit->tcp->th_seq); 2264 2265 /* out of order or retransmitted. */ 2266 if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) { 2267 chain->stat.data_out_of_win++; 2268 return RSC_FINAL; 2269 } 2270 2271 data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen; 2272 if (nseq == oseq) { 2273 if ((o_unit->payload == 0) && n_unit->payload) { 2274 /* From no payload to payload, normal case, not a dup ack or etc */ 2275 chain->stat.data_after_pure_ack++; 2276 goto coalesce; 2277 } else { 2278 return virtio_net_rsc_handle_ack(chain, seg, buf, 2279 n_unit->tcp, o_unit->tcp); 2280 } 2281 } else if ((nseq - oseq) != o_unit->payload) { 2282 /* Not a consistent packet, out of order */ 2283 chain->stat.data_out_of_order++; 2284 return RSC_FINAL; 2285 } else { 2286 coalesce: 2287 if ((o_ip_len + n_unit->payload) > chain->max_payload) { 2288 chain->stat.over_size++; 2289 return RSC_FINAL; 2290 } 2291 2292 /* Here comes the right data, the payload length in v4/v6 is different, 2293 so use the field value to update and record the new data len */ 2294 o_unit->payload += n_unit->payload; /* update new data len */ 2295 2296 /* update field in ip header */ 2297 write_unit_ip_len(o_unit, o_ip_len + n_unit->payload); 2298 2299 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced 2300 for windows guest, while this may change the behavior for linux 2301 guest (only if it uses RSC feature). */ 2302 o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags; 2303 2304 o_unit->tcp->th_ack = n_unit->tcp->th_ack; 2305 o_unit->tcp->th_win = n_unit->tcp->th_win; 2306 2307 memmove(seg->buf + seg->size, data, n_unit->payload); 2308 seg->size += n_unit->payload; 2309 seg->packets++; 2310 chain->stat.coalesced++; 2311 return RSC_COALESCE; 2312 } 2313 } 2314 2315 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain, 2316 VirtioNetRscSeg *seg, 2317 const uint8_t *buf, size_t size, 2318 VirtioNetRscUnit *unit) 2319 { 2320 struct ip_header *ip1, *ip2; 2321 2322 ip1 = (struct ip_header *)(unit->ip); 2323 ip2 = (struct ip_header *)(seg->unit.ip); 2324 if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst) 2325 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport) 2326 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) { 2327 chain->stat.no_match++; 2328 return RSC_NO_MATCH; 2329 } 2330 2331 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit); 2332 } 2333 2334 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain, 2335 VirtioNetRscSeg *seg, 2336 const uint8_t *buf, size_t size, 2337 VirtioNetRscUnit *unit) 2338 { 2339 struct ip6_header *ip1, *ip2; 2340 2341 ip1 = (struct ip6_header *)(unit->ip); 2342 ip2 = (struct ip6_header *)(seg->unit.ip); 2343 if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address)) 2344 || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address)) 2345 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport) 2346 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) { 2347 chain->stat.no_match++; 2348 return RSC_NO_MATCH; 2349 } 2350 2351 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit); 2352 } 2353 2354 /* Packets with 'SYN' should bypass, other flag should be sent after drain 2355 * to prevent out of order */ 2356 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain, 2357 struct tcp_header *tcp) 2358 { 2359 uint16_t tcp_hdr; 2360 uint16_t tcp_flag; 2361 2362 tcp_flag = htons(tcp->th_offset_flags); 2363 tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10; 2364 tcp_flag &= VIRTIO_NET_TCP_FLAG; 2365 if (tcp_flag & TH_SYN) { 2366 chain->stat.tcp_syn++; 2367 return RSC_BYPASS; 2368 } 2369 2370 if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) { 2371 chain->stat.tcp_ctrl_drain++; 2372 return RSC_FINAL; 2373 } 2374 2375 if (tcp_hdr > sizeof(struct tcp_header)) { 2376 chain->stat.tcp_all_opt++; 2377 return RSC_FINAL; 2378 } 2379 2380 return RSC_CANDIDATE; 2381 } 2382 2383 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain, 2384 NetClientState *nc, 2385 const uint8_t *buf, size_t size, 2386 VirtioNetRscUnit *unit) 2387 { 2388 int ret; 2389 VirtioNetRscSeg *seg, *nseg; 2390 2391 if (QTAILQ_EMPTY(&chain->buffers)) { 2392 chain->stat.empty_cache++; 2393 virtio_net_rsc_cache_buf(chain, nc, buf, size); 2394 timer_mod(chain->drain_timer, 2395 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + chain->n->rsc_timeout); 2396 return size; 2397 } 2398 2399 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) { 2400 if (chain->proto == ETH_P_IP) { 2401 ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit); 2402 } else { 2403 ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit); 2404 } 2405 2406 if (ret == RSC_FINAL) { 2407 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2408 /* Send failed */ 2409 chain->stat.final_failed++; 2410 return 0; 2411 } 2412 2413 /* Send current packet */ 2414 return virtio_net_do_receive(nc, buf, size); 2415 } else if (ret == RSC_NO_MATCH) { 2416 continue; 2417 } else { 2418 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */ 2419 seg->is_coalesced = 1; 2420 return size; 2421 } 2422 } 2423 2424 chain->stat.no_match_cache++; 2425 virtio_net_rsc_cache_buf(chain, nc, buf, size); 2426 return size; 2427 } 2428 2429 /* Drain a connection data, this is to avoid out of order segments */ 2430 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain, 2431 NetClientState *nc, 2432 const uint8_t *buf, size_t size, 2433 uint16_t ip_start, uint16_t ip_size, 2434 uint16_t tcp_port) 2435 { 2436 VirtioNetRscSeg *seg, *nseg; 2437 uint32_t ppair1, ppair2; 2438 2439 ppair1 = *(uint32_t *)(buf + tcp_port); 2440 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) { 2441 ppair2 = *(uint32_t *)(seg->buf + tcp_port); 2442 if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size) 2443 || (ppair1 != ppair2)) { 2444 continue; 2445 } 2446 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2447 chain->stat.drain_failed++; 2448 } 2449 2450 break; 2451 } 2452 2453 return virtio_net_do_receive(nc, buf, size); 2454 } 2455 2456 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain, 2457 struct ip_header *ip, 2458 const uint8_t *buf, size_t size) 2459 { 2460 uint16_t ip_len; 2461 2462 /* Not an ipv4 packet */ 2463 if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) { 2464 chain->stat.ip_option++; 2465 return RSC_BYPASS; 2466 } 2467 2468 /* Don't handle packets with ip option */ 2469 if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) { 2470 chain->stat.ip_option++; 2471 return RSC_BYPASS; 2472 } 2473 2474 if (ip->ip_p != IPPROTO_TCP) { 2475 chain->stat.bypass_not_tcp++; 2476 return RSC_BYPASS; 2477 } 2478 2479 /* Don't handle packets with ip fragment */ 2480 if (!(htons(ip->ip_off) & IP_DF)) { 2481 chain->stat.ip_frag++; 2482 return RSC_BYPASS; 2483 } 2484 2485 /* Don't handle packets with ecn flag */ 2486 if (IPTOS_ECN(ip->ip_tos)) { 2487 chain->stat.ip_ecn++; 2488 return RSC_BYPASS; 2489 } 2490 2491 ip_len = htons(ip->ip_len); 2492 if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header)) 2493 || ip_len > (size - chain->n->guest_hdr_len - 2494 sizeof(struct eth_header))) { 2495 chain->stat.ip_hacked++; 2496 return RSC_BYPASS; 2497 } 2498 2499 return RSC_CANDIDATE; 2500 } 2501 2502 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain, 2503 NetClientState *nc, 2504 const uint8_t *buf, size_t size) 2505 { 2506 int32_t ret; 2507 uint16_t hdr_len; 2508 VirtioNetRscUnit unit; 2509 2510 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len; 2511 2512 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header) 2513 + sizeof(struct tcp_header))) { 2514 chain->stat.bypass_not_tcp++; 2515 return virtio_net_do_receive(nc, buf, size); 2516 } 2517 2518 virtio_net_rsc_extract_unit4(chain, buf, &unit); 2519 if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size) 2520 != RSC_CANDIDATE) { 2521 return virtio_net_do_receive(nc, buf, size); 2522 } 2523 2524 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp); 2525 if (ret == RSC_BYPASS) { 2526 return virtio_net_do_receive(nc, buf, size); 2527 } else if (ret == RSC_FINAL) { 2528 return virtio_net_rsc_drain_flow(chain, nc, buf, size, 2529 ((hdr_len + sizeof(struct eth_header)) + 12), 2530 VIRTIO_NET_IP4_ADDR_SIZE, 2531 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)); 2532 } 2533 2534 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit); 2535 } 2536 2537 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain, 2538 struct ip6_header *ip6, 2539 const uint8_t *buf, size_t size) 2540 { 2541 uint16_t ip_len; 2542 2543 if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4) 2544 != IP_HEADER_VERSION_6) { 2545 return RSC_BYPASS; 2546 } 2547 2548 /* Both option and protocol is checked in this */ 2549 if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) { 2550 chain->stat.bypass_not_tcp++; 2551 return RSC_BYPASS; 2552 } 2553 2554 ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen); 2555 if (ip_len < sizeof(struct tcp_header) || 2556 ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header) 2557 - sizeof(struct ip6_header))) { 2558 chain->stat.ip_hacked++; 2559 return RSC_BYPASS; 2560 } 2561 2562 /* Don't handle packets with ecn flag */ 2563 if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) { 2564 chain->stat.ip_ecn++; 2565 return RSC_BYPASS; 2566 } 2567 2568 return RSC_CANDIDATE; 2569 } 2570 2571 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc, 2572 const uint8_t *buf, size_t size) 2573 { 2574 int32_t ret; 2575 uint16_t hdr_len; 2576 VirtioNetRscChain *chain; 2577 VirtioNetRscUnit unit; 2578 2579 chain = opq; 2580 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len; 2581 2582 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header) 2583 + sizeof(tcp_header))) { 2584 return virtio_net_do_receive(nc, buf, size); 2585 } 2586 2587 virtio_net_rsc_extract_unit6(chain, buf, &unit); 2588 if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain, 2589 unit.ip, buf, size)) { 2590 return virtio_net_do_receive(nc, buf, size); 2591 } 2592 2593 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp); 2594 if (ret == RSC_BYPASS) { 2595 return virtio_net_do_receive(nc, buf, size); 2596 } else if (ret == RSC_FINAL) { 2597 return virtio_net_rsc_drain_flow(chain, nc, buf, size, 2598 ((hdr_len + sizeof(struct eth_header)) + 8), 2599 VIRTIO_NET_IP6_ADDR_SIZE, 2600 hdr_len + sizeof(struct eth_header) 2601 + sizeof(struct ip6_header)); 2602 } 2603 2604 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit); 2605 } 2606 2607 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n, 2608 NetClientState *nc, 2609 uint16_t proto) 2610 { 2611 VirtioNetRscChain *chain; 2612 2613 if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) { 2614 return NULL; 2615 } 2616 2617 QTAILQ_FOREACH(chain, &n->rsc_chains, next) { 2618 if (chain->proto == proto) { 2619 return chain; 2620 } 2621 } 2622 2623 chain = g_malloc(sizeof(*chain)); 2624 chain->n = n; 2625 chain->proto = proto; 2626 if (proto == (uint16_t)ETH_P_IP) { 2627 chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD; 2628 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2629 } else { 2630 chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD; 2631 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2632 } 2633 chain->drain_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 2634 virtio_net_rsc_purge, chain); 2635 memset(&chain->stat, 0, sizeof(chain->stat)); 2636 2637 QTAILQ_INIT(&chain->buffers); 2638 QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next); 2639 2640 return chain; 2641 } 2642 2643 static ssize_t virtio_net_rsc_receive(NetClientState *nc, 2644 const uint8_t *buf, 2645 size_t size) 2646 { 2647 uint16_t proto; 2648 VirtioNetRscChain *chain; 2649 struct eth_header *eth; 2650 VirtIONet *n; 2651 2652 n = qemu_get_nic_opaque(nc); 2653 if (size < (n->host_hdr_len + sizeof(struct eth_header))) { 2654 return virtio_net_do_receive(nc, buf, size); 2655 } 2656 2657 eth = (struct eth_header *)(buf + n->guest_hdr_len); 2658 proto = htons(eth->h_proto); 2659 2660 chain = virtio_net_rsc_lookup_chain(n, nc, proto); 2661 if (chain) { 2662 chain->stat.received++; 2663 if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) { 2664 return virtio_net_rsc_receive4(chain, nc, buf, size); 2665 } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) { 2666 return virtio_net_rsc_receive6(chain, nc, buf, size); 2667 } 2668 } 2669 return virtio_net_do_receive(nc, buf, size); 2670 } 2671 2672 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, 2673 size_t size) 2674 { 2675 VirtIONet *n = qemu_get_nic_opaque(nc); 2676 if ((n->rsc4_enabled || n->rsc6_enabled)) { 2677 return virtio_net_rsc_receive(nc, buf, size); 2678 } else { 2679 return virtio_net_do_receive(nc, buf, size); 2680 } 2681 } 2682 2683 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 2684 2685 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 2686 { 2687 VirtIONet *n = qemu_get_nic_opaque(nc); 2688 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 2689 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2690 int ret; 2691 2692 virtqueue_push(q->tx_vq, q->async_tx.elem, 0); 2693 virtio_notify(vdev, q->tx_vq); 2694 2695 g_free(q->async_tx.elem); 2696 q->async_tx.elem = NULL; 2697 2698 virtio_queue_set_notification(q->tx_vq, 1); 2699 ret = virtio_net_flush_tx(q); 2700 if (ret >= n->tx_burst) { 2701 /* 2702 * the flush has been stopped by tx_burst 2703 * we will not receive notification for the 2704 * remainining part, so re-schedule 2705 */ 2706 virtio_queue_set_notification(q->tx_vq, 0); 2707 if (q->tx_bh) { 2708 replay_bh_schedule_event(q->tx_bh); 2709 } else { 2710 timer_mod(q->tx_timer, 2711 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2712 } 2713 q->tx_waiting = 1; 2714 } 2715 } 2716 2717 /* TX */ 2718 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 2719 { 2720 VirtIONet *n = q->n; 2721 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2722 VirtQueueElement *elem; 2723 int32_t num_packets = 0; 2724 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 2725 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 2726 return num_packets; 2727 } 2728 2729 if (q->async_tx.elem) { 2730 virtio_queue_set_notification(q->tx_vq, 0); 2731 return num_packets; 2732 } 2733 2734 for (;;) { 2735 ssize_t ret; 2736 unsigned int out_num; 2737 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg; 2738 struct virtio_net_hdr vhdr; 2739 2740 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement)); 2741 if (!elem) { 2742 break; 2743 } 2744 2745 out_num = elem->out_num; 2746 out_sg = elem->out_sg; 2747 if (out_num < 1) { 2748 virtio_error(vdev, "virtio-net header not in first element"); 2749 goto detach; 2750 } 2751 2752 if (n->needs_vnet_hdr_swap) { 2753 if (iov_to_buf(out_sg, out_num, 0, &vhdr, sizeof(vhdr)) < 2754 sizeof(vhdr)) { 2755 virtio_error(vdev, "virtio-net header incorrect"); 2756 goto detach; 2757 } 2758 virtio_net_hdr_swap(vdev, &vhdr); 2759 sg2[0].iov_base = &vhdr; 2760 sg2[0].iov_len = sizeof(vhdr); 2761 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1, out_sg, out_num, 2762 sizeof(vhdr), -1); 2763 if (out_num == VIRTQUEUE_MAX_SIZE) { 2764 goto drop; 2765 } 2766 out_num += 1; 2767 out_sg = sg2; 2768 } 2769 /* 2770 * If host wants to see the guest header as is, we can 2771 * pass it on unchanged. Otherwise, copy just the parts 2772 * that host is interested in. 2773 */ 2774 assert(n->host_hdr_len <= n->guest_hdr_len); 2775 if (n->host_hdr_len != n->guest_hdr_len) { 2776 if (iov_size(out_sg, out_num) < n->guest_hdr_len) { 2777 virtio_error(vdev, "virtio-net header is invalid"); 2778 goto detach; 2779 } 2780 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 2781 out_sg, out_num, 2782 0, n->host_hdr_len); 2783 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 2784 out_sg, out_num, 2785 n->guest_hdr_len, -1); 2786 out_num = sg_num; 2787 out_sg = sg; 2788 2789 if (out_num < 1) { 2790 virtio_error(vdev, "virtio-net nothing to send"); 2791 goto detach; 2792 } 2793 } 2794 2795 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 2796 out_sg, out_num, virtio_net_tx_complete); 2797 if (ret == 0) { 2798 virtio_queue_set_notification(q->tx_vq, 0); 2799 q->async_tx.elem = elem; 2800 return -EBUSY; 2801 } 2802 2803 drop: 2804 virtqueue_push(q->tx_vq, elem, 0); 2805 virtio_notify(vdev, q->tx_vq); 2806 g_free(elem); 2807 2808 if (++num_packets >= n->tx_burst) { 2809 break; 2810 } 2811 } 2812 return num_packets; 2813 2814 detach: 2815 virtqueue_detach_element(q->tx_vq, elem, 0); 2816 g_free(elem); 2817 return -EINVAL; 2818 } 2819 2820 static void virtio_net_tx_timer(void *opaque); 2821 2822 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 2823 { 2824 VirtIONet *n = VIRTIO_NET(vdev); 2825 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 2826 2827 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 2828 virtio_net_drop_tx_queue_data(vdev, vq); 2829 return; 2830 } 2831 2832 /* This happens when device was stopped but VCPU wasn't. */ 2833 if (!vdev->vm_running) { 2834 q->tx_waiting = 1; 2835 return; 2836 } 2837 2838 if (q->tx_waiting) { 2839 /* We already have queued packets, immediately flush */ 2840 timer_del(q->tx_timer); 2841 virtio_net_tx_timer(q); 2842 } else { 2843 /* re-arm timer to flush it (and more) on next tick */ 2844 timer_mod(q->tx_timer, 2845 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2846 q->tx_waiting = 1; 2847 virtio_queue_set_notification(vq, 0); 2848 } 2849 } 2850 2851 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 2852 { 2853 VirtIONet *n = VIRTIO_NET(vdev); 2854 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 2855 2856 if (unlikely(n->vhost_started)) { 2857 return; 2858 } 2859 2860 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 2861 virtio_net_drop_tx_queue_data(vdev, vq); 2862 return; 2863 } 2864 2865 if (unlikely(q->tx_waiting)) { 2866 return; 2867 } 2868 q->tx_waiting = 1; 2869 /* This happens when device was stopped but VCPU wasn't. */ 2870 if (!vdev->vm_running) { 2871 return; 2872 } 2873 virtio_queue_set_notification(vq, 0); 2874 replay_bh_schedule_event(q->tx_bh); 2875 } 2876 2877 static void virtio_net_tx_timer(void *opaque) 2878 { 2879 VirtIONetQueue *q = opaque; 2880 VirtIONet *n = q->n; 2881 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2882 int ret; 2883 2884 /* This happens when device was stopped but BH wasn't. */ 2885 if (!vdev->vm_running) { 2886 /* Make sure tx waiting is set, so we'll run when restarted. */ 2887 assert(q->tx_waiting); 2888 return; 2889 } 2890 2891 q->tx_waiting = 0; 2892 2893 /* Just in case the driver is not ready on more */ 2894 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 2895 return; 2896 } 2897 2898 ret = virtio_net_flush_tx(q); 2899 if (ret == -EBUSY || ret == -EINVAL) { 2900 return; 2901 } 2902 /* 2903 * If we flush a full burst of packets, assume there are 2904 * more coming and immediately rearm 2905 */ 2906 if (ret >= n->tx_burst) { 2907 q->tx_waiting = 1; 2908 timer_mod(q->tx_timer, 2909 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 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 rearm 2916 */ 2917 virtio_queue_set_notification(q->tx_vq, 1); 2918 ret = virtio_net_flush_tx(q); 2919 if (ret > 0) { 2920 virtio_queue_set_notification(q->tx_vq, 0); 2921 q->tx_waiting = 1; 2922 timer_mod(q->tx_timer, 2923 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2924 } 2925 } 2926 2927 static void virtio_net_tx_bh(void *opaque) 2928 { 2929 VirtIONetQueue *q = opaque; 2930 VirtIONet *n = q->n; 2931 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2932 int32_t ret; 2933 2934 /* This happens when device was stopped but BH wasn't. */ 2935 if (!vdev->vm_running) { 2936 /* Make sure tx waiting is set, so we'll run when restarted. */ 2937 assert(q->tx_waiting); 2938 return; 2939 } 2940 2941 q->tx_waiting = 0; 2942 2943 /* Just in case the driver is not ready on more */ 2944 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 2945 return; 2946 } 2947 2948 ret = virtio_net_flush_tx(q); 2949 if (ret == -EBUSY || ret == -EINVAL) { 2950 return; /* Notification re-enable handled by tx_complete or device 2951 * broken */ 2952 } 2953 2954 /* If we flush a full burst of packets, assume there are 2955 * more coming and immediately reschedule */ 2956 if (ret >= n->tx_burst) { 2957 replay_bh_schedule_event(q->tx_bh); 2958 q->tx_waiting = 1; 2959 return; 2960 } 2961 2962 /* If less than a full burst, re-enable notification and flush 2963 * anything that may have come in while we weren't looking. If 2964 * we find something, assume the guest is still active and reschedule */ 2965 virtio_queue_set_notification(q->tx_vq, 1); 2966 ret = virtio_net_flush_tx(q); 2967 if (ret == -EINVAL) { 2968 return; 2969 } else if (ret > 0) { 2970 virtio_queue_set_notification(q->tx_vq, 0); 2971 replay_bh_schedule_event(q->tx_bh); 2972 q->tx_waiting = 1; 2973 } 2974 } 2975 2976 static void virtio_net_add_queue(VirtIONet *n, int index) 2977 { 2978 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2979 2980 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size, 2981 virtio_net_handle_rx); 2982 2983 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 2984 n->vqs[index].tx_vq = 2985 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 2986 virtio_net_handle_tx_timer); 2987 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 2988 virtio_net_tx_timer, 2989 &n->vqs[index]); 2990 } else { 2991 n->vqs[index].tx_vq = 2992 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 2993 virtio_net_handle_tx_bh); 2994 n->vqs[index].tx_bh = qemu_bh_new_guarded(virtio_net_tx_bh, &n->vqs[index], 2995 &DEVICE(vdev)->mem_reentrancy_guard); 2996 } 2997 2998 n->vqs[index].tx_waiting = 0; 2999 n->vqs[index].n = n; 3000 } 3001 3002 static void virtio_net_del_queue(VirtIONet *n, int index) 3003 { 3004 VirtIODevice *vdev = VIRTIO_DEVICE(n); 3005 VirtIONetQueue *q = &n->vqs[index]; 3006 NetClientState *nc = qemu_get_subqueue(n->nic, index); 3007 3008 qemu_purge_queued_packets(nc); 3009 3010 virtio_del_queue(vdev, index * 2); 3011 if (q->tx_timer) { 3012 timer_free(q->tx_timer); 3013 q->tx_timer = NULL; 3014 } else { 3015 qemu_bh_delete(q->tx_bh); 3016 q->tx_bh = NULL; 3017 } 3018 q->tx_waiting = 0; 3019 virtio_del_queue(vdev, index * 2 + 1); 3020 } 3021 3022 static void virtio_net_change_num_queue_pairs(VirtIONet *n, int new_max_queue_pairs) 3023 { 3024 VirtIODevice *vdev = VIRTIO_DEVICE(n); 3025 int old_num_queues = virtio_get_num_queues(vdev); 3026 int new_num_queues = new_max_queue_pairs * 2 + 1; 3027 int i; 3028 3029 assert(old_num_queues >= 3); 3030 assert(old_num_queues % 2 == 1); 3031 3032 if (old_num_queues == new_num_queues) { 3033 return; 3034 } 3035 3036 /* 3037 * We always need to remove and add ctrl vq if 3038 * old_num_queues != new_num_queues. Remove ctrl_vq first, 3039 * and then we only enter one of the following two loops. 3040 */ 3041 virtio_del_queue(vdev, old_num_queues - 1); 3042 3043 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) { 3044 /* new_num_queues < old_num_queues */ 3045 virtio_net_del_queue(n, i / 2); 3046 } 3047 3048 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) { 3049 /* new_num_queues > old_num_queues */ 3050 virtio_net_add_queue(n, i / 2); 3051 } 3052 3053 /* add ctrl_vq last */ 3054 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 3055 } 3056 3057 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 3058 { 3059 int max = multiqueue ? n->max_queue_pairs : 1; 3060 3061 n->multiqueue = multiqueue; 3062 virtio_net_change_num_queue_pairs(n, max); 3063 3064 virtio_net_set_queue_pairs(n); 3065 } 3066 3067 static int virtio_net_pre_load_queues(VirtIODevice *vdev) 3068 { 3069 virtio_net_set_multiqueue(VIRTIO_NET(vdev), 3070 virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_RSS) || 3071 virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MQ)); 3072 3073 return 0; 3074 } 3075 3076 static int virtio_net_post_load_device(void *opaque, int version_id) 3077 { 3078 VirtIONet *n = opaque; 3079 VirtIODevice *vdev = VIRTIO_DEVICE(n); 3080 int i, link_down; 3081 3082 trace_virtio_net_post_load_device(); 3083 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs, 3084 virtio_vdev_has_feature(vdev, 3085 VIRTIO_F_VERSION_1), 3086 virtio_vdev_has_feature(vdev, 3087 VIRTIO_NET_F_HASH_REPORT)); 3088 3089 /* MAC_TABLE_ENTRIES may be different from the saved image */ 3090 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) { 3091 n->mac_table.in_use = 0; 3092 } 3093 3094 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 3095 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 3096 } 3097 3098 /* 3099 * curr_guest_offloads will be later overwritten by the 3100 * virtio_set_features_nocheck call done from the virtio_load. 3101 * Here we make sure it is preserved and restored accordingly 3102 * in the virtio_net_post_load_virtio callback. 3103 */ 3104 n->saved_guest_offloads = n->curr_guest_offloads; 3105 3106 virtio_net_set_queue_pairs(n); 3107 3108 /* Find the first multicast entry in the saved MAC filter */ 3109 for (i = 0; i < n->mac_table.in_use; i++) { 3110 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 3111 break; 3112 } 3113 } 3114 n->mac_table.first_multi = i; 3115 3116 /* nc.link_down can't be migrated, so infer link_down according 3117 * to link status bit in n->status */ 3118 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 3119 for (i = 0; i < n->max_queue_pairs; i++) { 3120 qemu_get_subqueue(n->nic, i)->link_down = link_down; 3121 } 3122 3123 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 3124 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3125 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(), 3126 QEMU_CLOCK_VIRTUAL, 3127 virtio_net_announce_timer, n); 3128 if (n->announce_timer.round) { 3129 timer_mod(n->announce_timer.tm, 3130 qemu_clock_get_ms(n->announce_timer.type)); 3131 } else { 3132 qemu_announce_timer_del(&n->announce_timer, false); 3133 } 3134 } 3135 3136 virtio_net_commit_rss_config(n); 3137 return 0; 3138 } 3139 3140 static int virtio_net_post_load_virtio(VirtIODevice *vdev) 3141 { 3142 VirtIONet *n = VIRTIO_NET(vdev); 3143 /* 3144 * The actual needed state is now in saved_guest_offloads, 3145 * see virtio_net_post_load_device for detail. 3146 * Restore it back and apply the desired offloads. 3147 */ 3148 n->curr_guest_offloads = n->saved_guest_offloads; 3149 if (peer_has_vnet_hdr(n)) { 3150 virtio_net_apply_guest_offloads(n); 3151 } 3152 3153 return 0; 3154 } 3155 3156 /* tx_waiting field of a VirtIONetQueue */ 3157 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = { 3158 .name = "virtio-net-queue-tx_waiting", 3159 .fields = (const VMStateField[]) { 3160 VMSTATE_UINT32(tx_waiting, VirtIONetQueue), 3161 VMSTATE_END_OF_LIST() 3162 }, 3163 }; 3164 3165 static bool max_queue_pairs_gt_1(void *opaque, int version_id) 3166 { 3167 return VIRTIO_NET(opaque)->max_queue_pairs > 1; 3168 } 3169 3170 static bool has_ctrl_guest_offloads(void *opaque, int version_id) 3171 { 3172 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque), 3173 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS); 3174 } 3175 3176 static bool mac_table_fits(void *opaque, int version_id) 3177 { 3178 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES; 3179 } 3180 3181 static bool mac_table_doesnt_fit(void *opaque, int version_id) 3182 { 3183 return !mac_table_fits(opaque, version_id); 3184 } 3185 3186 /* This temporary type is shared by all the WITH_TMP methods 3187 * although only some fields are used by each. 3188 */ 3189 struct VirtIONetMigTmp { 3190 VirtIONet *parent; 3191 VirtIONetQueue *vqs_1; 3192 uint16_t curr_queue_pairs_1; 3193 uint8_t has_ufo; 3194 uint32_t has_vnet_hdr; 3195 }; 3196 3197 /* The 2nd and subsequent tx_waiting flags are loaded later than 3198 * the 1st entry in the queue_pairs and only if there's more than one 3199 * entry. We use the tmp mechanism to calculate a temporary 3200 * pointer and count and also validate the count. 3201 */ 3202 3203 static int virtio_net_tx_waiting_pre_save(void *opaque) 3204 { 3205 struct VirtIONetMigTmp *tmp = opaque; 3206 3207 tmp->vqs_1 = tmp->parent->vqs + 1; 3208 tmp->curr_queue_pairs_1 = tmp->parent->curr_queue_pairs - 1; 3209 if (tmp->parent->curr_queue_pairs == 0) { 3210 tmp->curr_queue_pairs_1 = 0; 3211 } 3212 3213 return 0; 3214 } 3215 3216 static int virtio_net_tx_waiting_pre_load(void *opaque) 3217 { 3218 struct VirtIONetMigTmp *tmp = opaque; 3219 3220 /* Reuse the pointer setup from save */ 3221 virtio_net_tx_waiting_pre_save(opaque); 3222 3223 if (tmp->parent->curr_queue_pairs > tmp->parent->max_queue_pairs) { 3224 error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x", 3225 tmp->parent->curr_queue_pairs, tmp->parent->max_queue_pairs); 3226 3227 return -EINVAL; 3228 } 3229 3230 return 0; /* all good */ 3231 } 3232 3233 static const VMStateDescription vmstate_virtio_net_tx_waiting = { 3234 .name = "virtio-net-tx_waiting", 3235 .pre_load = virtio_net_tx_waiting_pre_load, 3236 .pre_save = virtio_net_tx_waiting_pre_save, 3237 .fields = (const VMStateField[]) { 3238 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp, 3239 curr_queue_pairs_1, 3240 vmstate_virtio_net_queue_tx_waiting, 3241 struct VirtIONetQueue), 3242 VMSTATE_END_OF_LIST() 3243 }, 3244 }; 3245 3246 /* the 'has_ufo' flag is just tested; if the incoming stream has the 3247 * flag set we need to check that we have it 3248 */ 3249 static int virtio_net_ufo_post_load(void *opaque, int version_id) 3250 { 3251 struct VirtIONetMigTmp *tmp = opaque; 3252 3253 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) { 3254 error_report("virtio-net: saved image requires TUN_F_UFO support"); 3255 return -EINVAL; 3256 } 3257 3258 return 0; 3259 } 3260 3261 static int virtio_net_ufo_pre_save(void *opaque) 3262 { 3263 struct VirtIONetMigTmp *tmp = opaque; 3264 3265 tmp->has_ufo = tmp->parent->has_ufo; 3266 3267 return 0; 3268 } 3269 3270 static const VMStateDescription vmstate_virtio_net_has_ufo = { 3271 .name = "virtio-net-ufo", 3272 .post_load = virtio_net_ufo_post_load, 3273 .pre_save = virtio_net_ufo_pre_save, 3274 .fields = (const VMStateField[]) { 3275 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp), 3276 VMSTATE_END_OF_LIST() 3277 }, 3278 }; 3279 3280 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the 3281 * flag set we need to check that we have it 3282 */ 3283 static int virtio_net_vnet_post_load(void *opaque, int version_id) 3284 { 3285 struct VirtIONetMigTmp *tmp = opaque; 3286 3287 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) { 3288 error_report("virtio-net: saved image requires vnet_hdr=on"); 3289 return -EINVAL; 3290 } 3291 3292 return 0; 3293 } 3294 3295 static int virtio_net_vnet_pre_save(void *opaque) 3296 { 3297 struct VirtIONetMigTmp *tmp = opaque; 3298 3299 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr; 3300 3301 return 0; 3302 } 3303 3304 static const VMStateDescription vmstate_virtio_net_has_vnet = { 3305 .name = "virtio-net-vnet", 3306 .post_load = virtio_net_vnet_post_load, 3307 .pre_save = virtio_net_vnet_pre_save, 3308 .fields = (const VMStateField[]) { 3309 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp), 3310 VMSTATE_END_OF_LIST() 3311 }, 3312 }; 3313 3314 static bool virtio_net_rss_needed(void *opaque) 3315 { 3316 return VIRTIO_NET(opaque)->rss_data.enabled; 3317 } 3318 3319 static const VMStateDescription vmstate_virtio_net_rss = { 3320 .name = "virtio-net-device/rss", 3321 .version_id = 1, 3322 .minimum_version_id = 1, 3323 .needed = virtio_net_rss_needed, 3324 .fields = (const VMStateField[]) { 3325 VMSTATE_BOOL(rss_data.enabled, VirtIONet), 3326 VMSTATE_BOOL(rss_data.redirect, VirtIONet), 3327 VMSTATE_BOOL(rss_data.populate_hash, VirtIONet), 3328 VMSTATE_UINT32(rss_data.hash_types, VirtIONet), 3329 VMSTATE_UINT16(rss_data.indirections_len, VirtIONet), 3330 VMSTATE_UINT16(rss_data.default_queue, VirtIONet), 3331 VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet, 3332 VIRTIO_NET_RSS_MAX_KEY_SIZE), 3333 VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet, 3334 rss_data.indirections_len, 0, 3335 vmstate_info_uint16, uint16_t), 3336 VMSTATE_END_OF_LIST() 3337 }, 3338 }; 3339 3340 static const VMStateDescription vmstate_virtio_net_device = { 3341 .name = "virtio-net-device", 3342 .version_id = VIRTIO_NET_VM_VERSION, 3343 .minimum_version_id = VIRTIO_NET_VM_VERSION, 3344 .post_load = virtio_net_post_load_device, 3345 .fields = (const VMStateField[]) { 3346 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN), 3347 VMSTATE_STRUCT_POINTER(vqs, VirtIONet, 3348 vmstate_virtio_net_queue_tx_waiting, 3349 VirtIONetQueue), 3350 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet), 3351 VMSTATE_UINT16(status, VirtIONet), 3352 VMSTATE_UINT8(promisc, VirtIONet), 3353 VMSTATE_UINT8(allmulti, VirtIONet), 3354 VMSTATE_UINT32(mac_table.in_use, VirtIONet), 3355 3356 /* Guarded pair: If it fits we load it, else we throw it away 3357 * - can happen if source has a larger MAC table.; post-load 3358 * sets flags in this case. 3359 */ 3360 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet, 3361 0, mac_table_fits, mac_table.in_use, 3362 ETH_ALEN), 3363 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0, 3364 mac_table.in_use, ETH_ALEN), 3365 3366 /* Note: This is an array of uint32's that's always been saved as a 3367 * buffer; hold onto your endiannesses; it's actually used as a bitmap 3368 * but based on the uint. 3369 */ 3370 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3), 3371 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3372 vmstate_virtio_net_has_vnet), 3373 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet), 3374 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet), 3375 VMSTATE_UINT8(alluni, VirtIONet), 3376 VMSTATE_UINT8(nomulti, VirtIONet), 3377 VMSTATE_UINT8(nouni, VirtIONet), 3378 VMSTATE_UINT8(nobcast, VirtIONet), 3379 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3380 vmstate_virtio_net_has_ufo), 3381 VMSTATE_SINGLE_TEST(max_queue_pairs, VirtIONet, max_queue_pairs_gt_1, 0, 3382 vmstate_info_uint16_equal, uint16_t), 3383 VMSTATE_UINT16_TEST(curr_queue_pairs, VirtIONet, max_queue_pairs_gt_1), 3384 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3385 vmstate_virtio_net_tx_waiting), 3386 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet, 3387 has_ctrl_guest_offloads), 3388 VMSTATE_END_OF_LIST() 3389 }, 3390 .subsections = (const VMStateDescription * const []) { 3391 &vmstate_virtio_net_rss, 3392 NULL 3393 } 3394 }; 3395 3396 static NetClientInfo net_virtio_info = { 3397 .type = NET_CLIENT_DRIVER_NIC, 3398 .size = sizeof(NICState), 3399 .can_receive = virtio_net_can_receive, 3400 .receive = virtio_net_receive, 3401 .link_status_changed = virtio_net_set_link_status, 3402 .query_rx_filter = virtio_net_query_rxfilter, 3403 .announce = virtio_net_announce, 3404 }; 3405 3406 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 3407 { 3408 VirtIONet *n = VIRTIO_NET(vdev); 3409 NetClientState *nc; 3410 assert(n->vhost_started); 3411 if (!n->multiqueue && idx == 2) { 3412 /* Must guard against invalid features and bogus queue index 3413 * from being set by malicious guest, or penetrated through 3414 * buggy migration stream. 3415 */ 3416 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3417 qemu_log_mask(LOG_GUEST_ERROR, 3418 "%s: bogus vq index ignored\n", __func__); 3419 return false; 3420 } 3421 nc = qemu_get_subqueue(n->nic, n->max_queue_pairs); 3422 } else { 3423 nc = qemu_get_subqueue(n->nic, vq2q(idx)); 3424 } 3425 /* 3426 * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1 3427 * as the macro of configure interrupt's IDX, If this driver does not 3428 * support, the function will return false 3429 */ 3430 3431 if (idx == VIRTIO_CONFIG_IRQ_IDX) { 3432 return vhost_net_config_pending(get_vhost_net(nc->peer)); 3433 } 3434 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx); 3435 } 3436 3437 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 3438 bool mask) 3439 { 3440 VirtIONet *n = VIRTIO_NET(vdev); 3441 NetClientState *nc; 3442 assert(n->vhost_started); 3443 if (!n->multiqueue && idx == 2) { 3444 /* Must guard against invalid features and bogus queue index 3445 * from being set by malicious guest, or penetrated through 3446 * buggy migration stream. 3447 */ 3448 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3449 qemu_log_mask(LOG_GUEST_ERROR, 3450 "%s: bogus vq index ignored\n", __func__); 3451 return; 3452 } 3453 nc = qemu_get_subqueue(n->nic, n->max_queue_pairs); 3454 } else { 3455 nc = qemu_get_subqueue(n->nic, vq2q(idx)); 3456 } 3457 /* 3458 *Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1 3459 * as the macro of configure interrupt's IDX, If this driver does not 3460 * support, the function will return 3461 */ 3462 3463 if (idx == VIRTIO_CONFIG_IRQ_IDX) { 3464 vhost_net_config_mask(get_vhost_net(nc->peer), vdev, mask); 3465 return; 3466 } 3467 vhost_net_virtqueue_mask(get_vhost_net(nc->peer), vdev, idx, mask); 3468 } 3469 3470 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features) 3471 { 3472 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC); 3473 3474 n->config_size = virtio_get_config_size(&cfg_size_params, host_features); 3475 } 3476 3477 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 3478 const char *type) 3479 { 3480 /* 3481 * The name can be NULL, the netclient name will be type.x. 3482 */ 3483 assert(type != NULL); 3484 3485 g_free(n->netclient_name); 3486 g_free(n->netclient_type); 3487 n->netclient_name = g_strdup(name); 3488 n->netclient_type = g_strdup(type); 3489 } 3490 3491 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev) 3492 { 3493 HotplugHandler *hotplug_ctrl; 3494 PCIDevice *pci_dev; 3495 Error *err = NULL; 3496 3497 hotplug_ctrl = qdev_get_hotplug_handler(dev); 3498 if (hotplug_ctrl) { 3499 pci_dev = PCI_DEVICE(dev); 3500 pci_dev->partially_hotplugged = true; 3501 hotplug_handler_unplug_request(hotplug_ctrl, dev, &err); 3502 if (err) { 3503 error_report_err(err); 3504 return false; 3505 } 3506 } else { 3507 return false; 3508 } 3509 return true; 3510 } 3511 3512 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev, 3513 Error **errp) 3514 { 3515 Error *err = NULL; 3516 HotplugHandler *hotplug_ctrl; 3517 PCIDevice *pdev = PCI_DEVICE(dev); 3518 BusState *primary_bus; 3519 3520 if (!pdev->partially_hotplugged) { 3521 return true; 3522 } 3523 primary_bus = dev->parent_bus; 3524 if (!primary_bus) { 3525 error_setg(errp, "virtio_net: couldn't find primary bus"); 3526 return false; 3527 } 3528 qdev_set_parent_bus(dev, primary_bus, &error_abort); 3529 qatomic_set(&n->failover_primary_hidden, false); 3530 hotplug_ctrl = qdev_get_hotplug_handler(dev); 3531 if (hotplug_ctrl) { 3532 hotplug_handler_pre_plug(hotplug_ctrl, dev, &err); 3533 if (err) { 3534 goto out; 3535 } 3536 hotplug_handler_plug(hotplug_ctrl, dev, &err); 3537 } 3538 pdev->partially_hotplugged = false; 3539 3540 out: 3541 error_propagate(errp, err); 3542 return !err; 3543 } 3544 3545 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationEvent *e) 3546 { 3547 bool should_be_hidden; 3548 Error *err = NULL; 3549 DeviceState *dev = failover_find_primary_device(n); 3550 3551 if (!dev) { 3552 return; 3553 } 3554 3555 should_be_hidden = qatomic_read(&n->failover_primary_hidden); 3556 3557 if (e->type == MIG_EVENT_PRECOPY_SETUP && !should_be_hidden) { 3558 if (failover_unplug_primary(n, dev)) { 3559 vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev); 3560 qapi_event_send_unplug_primary(dev->id); 3561 qatomic_set(&n->failover_primary_hidden, true); 3562 } else { 3563 warn_report("couldn't unplug primary device"); 3564 } 3565 } else if (e->type == MIG_EVENT_PRECOPY_FAILED) { 3566 /* We already unplugged the device let's plug it back */ 3567 if (!failover_replug_primary(n, dev, &err)) { 3568 if (err) { 3569 error_report_err(err); 3570 } 3571 } 3572 } 3573 } 3574 3575 static int virtio_net_migration_state_notifier(NotifierWithReturn *notifier, 3576 MigrationEvent *e, Error **errp) 3577 { 3578 VirtIONet *n = container_of(notifier, VirtIONet, migration_state); 3579 virtio_net_handle_migration_primary(n, e); 3580 return 0; 3581 } 3582 3583 static bool failover_hide_primary_device(DeviceListener *listener, 3584 const QDict *device_opts, 3585 bool from_json, 3586 Error **errp) 3587 { 3588 VirtIONet *n = container_of(listener, VirtIONet, primary_listener); 3589 const char *standby_id; 3590 3591 if (!device_opts) { 3592 return false; 3593 } 3594 3595 if (!qdict_haskey(device_opts, "failover_pair_id")) { 3596 return false; 3597 } 3598 3599 if (!qdict_haskey(device_opts, "id")) { 3600 error_setg(errp, "Device with failover_pair_id needs to have id"); 3601 return false; 3602 } 3603 3604 standby_id = qdict_get_str(device_opts, "failover_pair_id"); 3605 if (g_strcmp0(standby_id, n->netclient_name) != 0) { 3606 return false; 3607 } 3608 3609 /* 3610 * The hide helper can be called several times for a given device. 3611 * Check there is only one primary for a virtio-net device but 3612 * don't duplicate the qdict several times if it's called for the same 3613 * device. 3614 */ 3615 if (n->primary_opts) { 3616 const char *old, *new; 3617 /* devices with failover_pair_id always have an id */ 3618 old = qdict_get_str(n->primary_opts, "id"); 3619 new = qdict_get_str(device_opts, "id"); 3620 if (strcmp(old, new) != 0) { 3621 error_setg(errp, "Cannot attach more than one primary device to " 3622 "'%s': '%s' and '%s'", n->netclient_name, old, new); 3623 return false; 3624 } 3625 } else { 3626 n->primary_opts = qdict_clone_shallow(device_opts); 3627 n->primary_opts_from_json = from_json; 3628 } 3629 3630 /* failover_primary_hidden is set during feature negotiation */ 3631 return qatomic_read(&n->failover_primary_hidden); 3632 } 3633 3634 static void virtio_net_device_realize(DeviceState *dev, Error **errp) 3635 { 3636 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3637 VirtIONet *n = VIRTIO_NET(dev); 3638 NetClientState *nc; 3639 int i; 3640 3641 if (n->net_conf.mtu) { 3642 n->host_features |= (1ULL << VIRTIO_NET_F_MTU); 3643 } 3644 3645 if (n->net_conf.duplex_str) { 3646 if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) { 3647 n->net_conf.duplex = DUPLEX_HALF; 3648 } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) { 3649 n->net_conf.duplex = DUPLEX_FULL; 3650 } else { 3651 error_setg(errp, "'duplex' must be 'half' or 'full'"); 3652 return; 3653 } 3654 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX); 3655 } else { 3656 n->net_conf.duplex = DUPLEX_UNKNOWN; 3657 } 3658 3659 if (n->net_conf.speed < SPEED_UNKNOWN) { 3660 error_setg(errp, "'speed' must be between 0 and INT_MAX"); 3661 return; 3662 } 3663 if (n->net_conf.speed >= 0) { 3664 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX); 3665 } 3666 3667 if (n->failover) { 3668 n->primary_listener.hide_device = failover_hide_primary_device; 3669 qatomic_set(&n->failover_primary_hidden, true); 3670 device_listener_register(&n->primary_listener); 3671 migration_add_notifier(&n->migration_state, 3672 virtio_net_migration_state_notifier); 3673 n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY); 3674 } 3675 3676 virtio_net_set_config_size(n, n->host_features); 3677 virtio_init(vdev, VIRTIO_ID_NET, n->config_size); 3678 3679 /* 3680 * We set a lower limit on RX queue size to what it always was. 3681 * Guests that want a smaller ring can always resize it without 3682 * help from us (using virtio 1 and up). 3683 */ 3684 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE || 3685 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE || 3686 !is_power_of_2(n->net_conf.rx_queue_size)) { 3687 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), " 3688 "must be a power of 2 between %d and %d.", 3689 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE, 3690 VIRTQUEUE_MAX_SIZE); 3691 virtio_cleanup(vdev); 3692 return; 3693 } 3694 3695 if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE || 3696 n->net_conf.tx_queue_size > virtio_net_max_tx_queue_size(n) || 3697 !is_power_of_2(n->net_conf.tx_queue_size)) { 3698 error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), " 3699 "must be a power of 2 between %d and %d", 3700 n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE, 3701 virtio_net_max_tx_queue_size(n)); 3702 virtio_cleanup(vdev); 3703 return; 3704 } 3705 3706 n->max_ncs = MAX(n->nic_conf.peers.queues, 1); 3707 3708 /* 3709 * Figure out the datapath queue pairs since the backend could 3710 * provide control queue via peers as well. 3711 */ 3712 if (n->nic_conf.peers.queues) { 3713 for (i = 0; i < n->max_ncs; i++) { 3714 if (n->nic_conf.peers.ncs[i]->is_datapath) { 3715 ++n->max_queue_pairs; 3716 } 3717 } 3718 } 3719 n->max_queue_pairs = MAX(n->max_queue_pairs, 1); 3720 3721 if (n->max_queue_pairs * 2 + 1 > VIRTIO_QUEUE_MAX) { 3722 error_setg(errp, "Invalid number of queue pairs (= %" PRIu32 "), " 3723 "must be a positive integer less than %d.", 3724 n->max_queue_pairs, (VIRTIO_QUEUE_MAX - 1) / 2); 3725 virtio_cleanup(vdev); 3726 return; 3727 } 3728 n->vqs = g_new0(VirtIONetQueue, n->max_queue_pairs); 3729 n->curr_queue_pairs = 1; 3730 n->tx_timeout = n->net_conf.txtimer; 3731 3732 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 3733 && strcmp(n->net_conf.tx, "bh")) { 3734 warn_report("virtio-net: " 3735 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 3736 n->net_conf.tx); 3737 error_printf("Defaulting to \"bh\""); 3738 } 3739 3740 n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n), 3741 n->net_conf.tx_queue_size); 3742 3743 virtio_net_add_queue(n, 0); 3744 3745 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 3746 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 3747 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 3748 n->status = VIRTIO_NET_S_LINK_UP; 3749 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(), 3750 QEMU_CLOCK_VIRTUAL, 3751 virtio_net_announce_timer, n); 3752 n->announce_timer.round = 0; 3753 3754 if (n->netclient_type) { 3755 /* 3756 * Happen when virtio_net_set_netclient_name has been called. 3757 */ 3758 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 3759 n->netclient_type, n->netclient_name, 3760 &dev->mem_reentrancy_guard, n); 3761 } else { 3762 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 3763 object_get_typename(OBJECT(dev)), dev->id, 3764 &dev->mem_reentrancy_guard, n); 3765 } 3766 3767 for (i = 0; i < n->max_queue_pairs; i++) { 3768 n->nic->ncs[i].do_not_pad = true; 3769 } 3770 3771 peer_test_vnet_hdr(n); 3772 if (peer_has_vnet_hdr(n)) { 3773 n->host_hdr_len = sizeof(struct virtio_net_hdr); 3774 } else { 3775 n->host_hdr_len = 0; 3776 } 3777 3778 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 3779 3780 n->vqs[0].tx_waiting = 0; 3781 n->tx_burst = n->net_conf.txburst; 3782 virtio_net_set_mrg_rx_bufs(n, 0, 0, 0); 3783 n->promisc = 1; /* for compatibility */ 3784 3785 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 3786 3787 n->vlans = g_malloc0(MAX_VLAN >> 3); 3788 3789 nc = qemu_get_queue(n->nic); 3790 nc->rxfilter_notify_enabled = 1; 3791 3792 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 3793 struct virtio_net_config netcfg = {}; 3794 memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN); 3795 vhost_net_set_config(get_vhost_net(nc->peer), 3796 (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_FRONTEND); 3797 } 3798 QTAILQ_INIT(&n->rsc_chains); 3799 n->qdev = dev; 3800 3801 net_rx_pkt_init(&n->rx_pkt); 3802 3803 if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) { 3804 Error *err = NULL; 3805 if (!virtio_net_load_ebpf(n, &err)) { 3806 /* 3807 * If user explicitly gave QEMU RSS FDs to use, then 3808 * failing to use them must be considered a fatal 3809 * error. If no RSS FDs were provided, QEMU is trying 3810 * eBPF on a "best effort" basis only, so report a 3811 * warning and allow fallback to software RSS. 3812 */ 3813 if (n->ebpf_rss_fds) { 3814 error_propagate(errp, err); 3815 } else { 3816 warn_report("unable to load eBPF RSS: %s", 3817 error_get_pretty(err)); 3818 error_free(err); 3819 } 3820 } 3821 } 3822 } 3823 3824 static void virtio_net_device_unrealize(DeviceState *dev) 3825 { 3826 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3827 VirtIONet *n = VIRTIO_NET(dev); 3828 int i, max_queue_pairs; 3829 3830 if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) { 3831 virtio_net_unload_ebpf(n); 3832 } 3833 3834 /* This will stop vhost backend if appropriate. */ 3835 virtio_net_set_status(vdev, 0); 3836 3837 g_free(n->netclient_name); 3838 n->netclient_name = NULL; 3839 g_free(n->netclient_type); 3840 n->netclient_type = NULL; 3841 3842 g_free(n->mac_table.macs); 3843 g_free(n->vlans); 3844 3845 if (n->failover) { 3846 qobject_unref(n->primary_opts); 3847 device_listener_unregister(&n->primary_listener); 3848 migration_remove_notifier(&n->migration_state); 3849 } else { 3850 assert(n->primary_opts == NULL); 3851 } 3852 3853 max_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 3854 for (i = 0; i < max_queue_pairs; i++) { 3855 virtio_net_del_queue(n, i); 3856 } 3857 /* delete also control vq */ 3858 virtio_del_queue(vdev, max_queue_pairs * 2); 3859 qemu_announce_timer_del(&n->announce_timer, false); 3860 g_free(n->vqs); 3861 qemu_del_nic(n->nic); 3862 virtio_net_rsc_cleanup(n); 3863 g_free(n->rss_data.indirections_table); 3864 net_rx_pkt_uninit(n->rx_pkt); 3865 virtio_cleanup(vdev); 3866 } 3867 3868 static void virtio_net_reset(VirtIODevice *vdev) 3869 { 3870 VirtIONet *n = VIRTIO_NET(vdev); 3871 int i; 3872 3873 /* Reset back to compatibility mode */ 3874 n->promisc = 1; 3875 n->allmulti = 0; 3876 n->alluni = 0; 3877 n->nomulti = 0; 3878 n->nouni = 0; 3879 n->nobcast = 0; 3880 /* multiqueue is disabled by default */ 3881 n->curr_queue_pairs = 1; 3882 timer_del(n->announce_timer.tm); 3883 n->announce_timer.round = 0; 3884 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 3885 3886 /* Flush any MAC and VLAN filter table state */ 3887 n->mac_table.in_use = 0; 3888 n->mac_table.first_multi = 0; 3889 n->mac_table.multi_overflow = 0; 3890 n->mac_table.uni_overflow = 0; 3891 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN); 3892 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac)); 3893 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 3894 memset(n->vlans, 0, MAX_VLAN >> 3); 3895 3896 /* Flush any async TX */ 3897 for (i = 0; i < n->max_queue_pairs; i++) { 3898 flush_or_purge_queued_packets(qemu_get_subqueue(n->nic, i)); 3899 } 3900 3901 virtio_net_disable_rss(n); 3902 } 3903 3904 static void virtio_net_instance_init(Object *obj) 3905 { 3906 VirtIONet *n = VIRTIO_NET(obj); 3907 3908 /* 3909 * The default config_size is sizeof(struct virtio_net_config). 3910 * Can be overridden with virtio_net_set_config_size. 3911 */ 3912 n->config_size = sizeof(struct virtio_net_config); 3913 device_add_bootindex_property(obj, &n->nic_conf.bootindex, 3914 "bootindex", "/ethernet-phy@0", 3915 DEVICE(n)); 3916 3917 ebpf_rss_init(&n->ebpf_rss); 3918 } 3919 3920 static int virtio_net_pre_save(void *opaque) 3921 { 3922 VirtIONet *n = opaque; 3923 3924 /* At this point, backend must be stopped, otherwise 3925 * it might keep writing to memory. */ 3926 assert(!n->vhost_started); 3927 3928 return 0; 3929 } 3930 3931 static bool primary_unplug_pending(void *opaque) 3932 { 3933 DeviceState *dev = opaque; 3934 DeviceState *primary; 3935 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3936 VirtIONet *n = VIRTIO_NET(vdev); 3937 3938 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) { 3939 return false; 3940 } 3941 primary = failover_find_primary_device(n); 3942 return primary ? primary->pending_deleted_event : false; 3943 } 3944 3945 static bool dev_unplug_pending(void *opaque) 3946 { 3947 DeviceState *dev = opaque; 3948 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); 3949 3950 return vdc->primary_unplug_pending(dev); 3951 } 3952 3953 static struct vhost_dev *virtio_net_get_vhost(VirtIODevice *vdev) 3954 { 3955 VirtIONet *n = VIRTIO_NET(vdev); 3956 NetClientState *nc; 3957 struct vhost_net *net; 3958 3959 if (!n->nic) { 3960 return NULL; 3961 } 3962 3963 nc = qemu_get_queue(n->nic); 3964 if (!nc) { 3965 return NULL; 3966 } 3967 3968 net = get_vhost_net(nc->peer); 3969 if (!net) { 3970 return NULL; 3971 } 3972 3973 return &net->dev; 3974 } 3975 3976 static const VMStateDescription vmstate_virtio_net = { 3977 .name = "virtio-net", 3978 .minimum_version_id = VIRTIO_NET_VM_VERSION, 3979 .version_id = VIRTIO_NET_VM_VERSION, 3980 .fields = (const VMStateField[]) { 3981 VMSTATE_VIRTIO_DEVICE, 3982 VMSTATE_END_OF_LIST() 3983 }, 3984 .pre_save = virtio_net_pre_save, 3985 .dev_unplug_pending = dev_unplug_pending, 3986 }; 3987 3988 static Property virtio_net_properties[] = { 3989 DEFINE_PROP_BIT64("csum", VirtIONet, host_features, 3990 VIRTIO_NET_F_CSUM, true), 3991 DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features, 3992 VIRTIO_NET_F_GUEST_CSUM, true), 3993 DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true), 3994 DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features, 3995 VIRTIO_NET_F_GUEST_TSO4, true), 3996 DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features, 3997 VIRTIO_NET_F_GUEST_TSO6, true), 3998 DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features, 3999 VIRTIO_NET_F_GUEST_ECN, true), 4000 DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features, 4001 VIRTIO_NET_F_GUEST_UFO, true), 4002 DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features, 4003 VIRTIO_NET_F_GUEST_ANNOUNCE, true), 4004 DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features, 4005 VIRTIO_NET_F_HOST_TSO4, true), 4006 DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features, 4007 VIRTIO_NET_F_HOST_TSO6, true), 4008 DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features, 4009 VIRTIO_NET_F_HOST_ECN, true), 4010 DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features, 4011 VIRTIO_NET_F_HOST_UFO, true), 4012 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features, 4013 VIRTIO_NET_F_MRG_RXBUF, true), 4014 DEFINE_PROP_BIT64("status", VirtIONet, host_features, 4015 VIRTIO_NET_F_STATUS, true), 4016 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features, 4017 VIRTIO_NET_F_CTRL_VQ, true), 4018 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features, 4019 VIRTIO_NET_F_CTRL_RX, true), 4020 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features, 4021 VIRTIO_NET_F_CTRL_VLAN, true), 4022 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features, 4023 VIRTIO_NET_F_CTRL_RX_EXTRA, true), 4024 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features, 4025 VIRTIO_NET_F_CTRL_MAC_ADDR, true), 4026 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features, 4027 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true), 4028 DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false), 4029 DEFINE_PROP_BIT64("rss", VirtIONet, host_features, 4030 VIRTIO_NET_F_RSS, false), 4031 DEFINE_PROP_BIT64("hash", VirtIONet, host_features, 4032 VIRTIO_NET_F_HASH_REPORT, false), 4033 DEFINE_PROP_ARRAY("ebpf-rss-fds", VirtIONet, nr_ebpf_rss_fds, 4034 ebpf_rss_fds, qdev_prop_string, char*), 4035 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features, 4036 VIRTIO_NET_F_RSC_EXT, false), 4037 DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout, 4038 VIRTIO_NET_RSC_DEFAULT_INTERVAL), 4039 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 4040 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 4041 TX_TIMER_INTERVAL), 4042 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 4043 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 4044 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size, 4045 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE), 4046 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size, 4047 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE), 4048 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0), 4049 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend, 4050 true), 4051 DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN), 4052 DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str), 4053 DEFINE_PROP_BOOL("failover", VirtIONet, failover, false), 4054 DEFINE_PROP_BIT64("guest_uso4", VirtIONet, host_features, 4055 VIRTIO_NET_F_GUEST_USO4, true), 4056 DEFINE_PROP_BIT64("guest_uso6", VirtIONet, host_features, 4057 VIRTIO_NET_F_GUEST_USO6, true), 4058 DEFINE_PROP_BIT64("host_uso", VirtIONet, host_features, 4059 VIRTIO_NET_F_HOST_USO, true), 4060 DEFINE_PROP_END_OF_LIST(), 4061 }; 4062 4063 static void virtio_net_class_init(ObjectClass *klass, void *data) 4064 { 4065 DeviceClass *dc = DEVICE_CLASS(klass); 4066 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 4067 4068 device_class_set_props(dc, virtio_net_properties); 4069 dc->vmsd = &vmstate_virtio_net; 4070 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 4071 vdc->realize = virtio_net_device_realize; 4072 vdc->unrealize = virtio_net_device_unrealize; 4073 vdc->get_config = virtio_net_get_config; 4074 vdc->set_config = virtio_net_set_config; 4075 vdc->get_features = virtio_net_get_features; 4076 vdc->set_features = virtio_net_set_features; 4077 vdc->bad_features = virtio_net_bad_features; 4078 vdc->reset = virtio_net_reset; 4079 vdc->queue_reset = virtio_net_queue_reset; 4080 vdc->queue_enable = virtio_net_queue_enable; 4081 vdc->set_status = virtio_net_set_status; 4082 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 4083 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 4084 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO); 4085 vdc->pre_load_queues = virtio_net_pre_load_queues; 4086 vdc->post_load = virtio_net_post_load_virtio; 4087 vdc->vmsd = &vmstate_virtio_net_device; 4088 vdc->primary_unplug_pending = primary_unplug_pending; 4089 vdc->get_vhost = virtio_net_get_vhost; 4090 vdc->toggle_device_iotlb = vhost_toggle_device_iotlb; 4091 } 4092 4093 static const TypeInfo virtio_net_info = { 4094 .name = TYPE_VIRTIO_NET, 4095 .parent = TYPE_VIRTIO_DEVICE, 4096 .instance_size = sizeof(VirtIONet), 4097 .instance_init = virtio_net_instance_init, 4098 .class_init = virtio_net_class_init, 4099 }; 4100 4101 static void virtio_register_types(void) 4102 { 4103 type_register_static(&virtio_net_info); 4104 } 4105 4106 type_init(virtio_register_types) 4107