1 /* 2 * Virtio Network Device 3 * 4 * Copyright IBM, Corp. 2007 5 * 6 * Authors: 7 * Anthony Liguori <aliguori@us.ibm.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2. See 10 * the COPYING file in the top-level directory. 11 * 12 */ 13 14 #include "qemu/osdep.h" 15 #include "qemu/atomic.h" 16 #include "qemu/iov.h" 17 #include "qemu/log.h" 18 #include "qemu/main-loop.h" 19 #include "qemu/module.h" 20 #include "hw/virtio/virtio.h" 21 #include "net/net.h" 22 #include "net/checksum.h" 23 #include "net/tap.h" 24 #include "qemu/error-report.h" 25 #include "qemu/timer.h" 26 #include "qemu/option.h" 27 #include "qemu/option_int.h" 28 #include "qemu/config-file.h" 29 #include "qapi/qmp/qdict.h" 30 #include "hw/virtio/virtio-net.h" 31 #include "net/vhost_net.h" 32 #include "net/announce.h" 33 #include "hw/virtio/virtio-bus.h" 34 #include "qapi/error.h" 35 #include "qapi/qapi-events-net.h" 36 #include "hw/qdev-properties.h" 37 #include "qapi/qapi-types-migration.h" 38 #include "qapi/qapi-events-migration.h" 39 #include "hw/virtio/virtio-access.h" 40 #include "migration/misc.h" 41 #include "standard-headers/linux/ethtool.h" 42 #include "sysemu/sysemu.h" 43 #include "trace.h" 44 #include "monitor/qdev.h" 45 #include "monitor/monitor.h" 46 #include "hw/pci/pci_device.h" 47 #include "net_rx_pkt.h" 48 #include "hw/virtio/vhost.h" 49 #include "sysemu/qtest.h" 50 51 #define VIRTIO_NET_VM_VERSION 11 52 53 /* previously fixed value */ 54 #define VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 256 55 #define VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 256 56 57 /* for now, only allow larger queue_pairs; with virtio-1, guest can downsize */ 58 #define VIRTIO_NET_RX_QUEUE_MIN_SIZE VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE 59 #define VIRTIO_NET_TX_QUEUE_MIN_SIZE VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE 60 61 #define VIRTIO_NET_IP4_ADDR_SIZE 8 /* ipv4 saddr + daddr */ 62 63 #define VIRTIO_NET_TCP_FLAG 0x3F 64 #define VIRTIO_NET_TCP_HDR_LENGTH 0xF000 65 66 /* IPv4 max payload, 16 bits in the header */ 67 #define VIRTIO_NET_MAX_IP4_PAYLOAD (65535 - sizeof(struct ip_header)) 68 #define VIRTIO_NET_MAX_TCP_PAYLOAD 65535 69 70 /* header length value in ip header without option */ 71 #define VIRTIO_NET_IP4_HEADER_LENGTH 5 72 73 #define VIRTIO_NET_IP6_ADDR_SIZE 32 /* ipv6 saddr + daddr */ 74 #define VIRTIO_NET_MAX_IP6_PAYLOAD VIRTIO_NET_MAX_TCP_PAYLOAD 75 76 /* Purge coalesced packets timer interval, This value affects the performance 77 a lot, and should be tuned carefully, '300000'(300us) is the recommended 78 value to pass the WHQL test, '50000' can gain 2x netperf throughput with 79 tso/gso/gro 'off'. */ 80 #define VIRTIO_NET_RSC_DEFAULT_INTERVAL 300000 81 82 #define VIRTIO_NET_RSS_SUPPORTED_HASHES (VIRTIO_NET_RSS_HASH_TYPE_IPv4 | \ 83 VIRTIO_NET_RSS_HASH_TYPE_TCPv4 | \ 84 VIRTIO_NET_RSS_HASH_TYPE_UDPv4 | \ 85 VIRTIO_NET_RSS_HASH_TYPE_IPv6 | \ 86 VIRTIO_NET_RSS_HASH_TYPE_TCPv6 | \ 87 VIRTIO_NET_RSS_HASH_TYPE_UDPv6 | \ 88 VIRTIO_NET_RSS_HASH_TYPE_IP_EX | \ 89 VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | \ 90 VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) 91 92 static const VirtIOFeature feature_sizes[] = { 93 {.flags = 1ULL << VIRTIO_NET_F_MAC, 94 .end = endof(struct virtio_net_config, mac)}, 95 {.flags = 1ULL << VIRTIO_NET_F_STATUS, 96 .end = endof(struct virtio_net_config, status)}, 97 {.flags = 1ULL << VIRTIO_NET_F_MQ, 98 .end = endof(struct virtio_net_config, max_virtqueue_pairs)}, 99 {.flags = 1ULL << VIRTIO_NET_F_MTU, 100 .end = endof(struct virtio_net_config, mtu)}, 101 {.flags = 1ULL << VIRTIO_NET_F_SPEED_DUPLEX, 102 .end = endof(struct virtio_net_config, duplex)}, 103 {.flags = (1ULL << VIRTIO_NET_F_RSS) | (1ULL << VIRTIO_NET_F_HASH_REPORT), 104 .end = endof(struct virtio_net_config, supported_hash_types)}, 105 {} 106 }; 107 108 static const VirtIOConfigSizeParams cfg_size_params = { 109 .min_size = endof(struct virtio_net_config, mac), 110 .max_size = sizeof(struct virtio_net_config), 111 .feature_sizes = feature_sizes 112 }; 113 114 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc) 115 { 116 VirtIONet *n = qemu_get_nic_opaque(nc); 117 118 return &n->vqs[nc->queue_index]; 119 } 120 121 static int vq2q(int queue_index) 122 { 123 return queue_index / 2; 124 } 125 126 static void flush_or_purge_queued_packets(NetClientState *nc) 127 { 128 if (!nc->peer) { 129 return; 130 } 131 132 qemu_flush_or_purge_queued_packets(nc->peer, true); 133 assert(!virtio_net_get_subqueue(nc)->async_tx.elem); 134 } 135 136 /* TODO 137 * - we could suppress RX interrupt if we were so inclined. 138 */ 139 140 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config) 141 { 142 VirtIONet *n = VIRTIO_NET(vdev); 143 struct virtio_net_config netcfg; 144 NetClientState *nc = qemu_get_queue(n->nic); 145 static const MACAddr zero = { .a = { 0, 0, 0, 0, 0, 0 } }; 146 147 int ret = 0; 148 memset(&netcfg, 0 , sizeof(struct virtio_net_config)); 149 virtio_stw_p(vdev, &netcfg.status, n->status); 150 virtio_stw_p(vdev, &netcfg.max_virtqueue_pairs, n->max_queue_pairs); 151 virtio_stw_p(vdev, &netcfg.mtu, n->net_conf.mtu); 152 memcpy(netcfg.mac, n->mac, ETH_ALEN); 153 virtio_stl_p(vdev, &netcfg.speed, n->net_conf.speed); 154 netcfg.duplex = n->net_conf.duplex; 155 netcfg.rss_max_key_size = VIRTIO_NET_RSS_MAX_KEY_SIZE; 156 virtio_stw_p(vdev, &netcfg.rss_max_indirection_table_length, 157 virtio_host_has_feature(vdev, VIRTIO_NET_F_RSS) ? 158 VIRTIO_NET_RSS_MAX_TABLE_LEN : 1); 159 virtio_stl_p(vdev, &netcfg.supported_hash_types, 160 VIRTIO_NET_RSS_SUPPORTED_HASHES); 161 memcpy(config, &netcfg, n->config_size); 162 163 /* 164 * Is this VDPA? No peer means not VDPA: there's no way to 165 * disconnect/reconnect a VDPA peer. 166 */ 167 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 168 ret = vhost_net_get_config(get_vhost_net(nc->peer), (uint8_t *)&netcfg, 169 n->config_size); 170 if (ret == -1) { 171 return; 172 } 173 174 /* 175 * Some NIC/kernel combinations present 0 as the mac address. As that 176 * is not a legal address, try to proceed with the address from the 177 * QEMU command line in the hope that the address has been configured 178 * correctly elsewhere - just not reported by the device. 179 */ 180 if (memcmp(&netcfg.mac, &zero, sizeof(zero)) == 0) { 181 info_report("Zero hardware mac address detected. Ignoring."); 182 memcpy(netcfg.mac, n->mac, ETH_ALEN); 183 } 184 185 netcfg.status |= virtio_tswap16(vdev, 186 n->status & VIRTIO_NET_S_ANNOUNCE); 187 memcpy(config, &netcfg, n->config_size); 188 } 189 } 190 191 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config) 192 { 193 VirtIONet *n = VIRTIO_NET(vdev); 194 struct virtio_net_config netcfg = {}; 195 NetClientState *nc = qemu_get_queue(n->nic); 196 197 memcpy(&netcfg, config, n->config_size); 198 199 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR) && 200 !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1) && 201 memcmp(netcfg.mac, n->mac, ETH_ALEN)) { 202 memcpy(n->mac, netcfg.mac, ETH_ALEN); 203 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 204 } 205 206 /* 207 * Is this VDPA? No peer means not VDPA: there's no way to 208 * disconnect/reconnect a VDPA peer. 209 */ 210 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 211 vhost_net_set_config(get_vhost_net(nc->peer), 212 (uint8_t *)&netcfg, 0, n->config_size, 213 VHOST_SET_CONFIG_TYPE_FRONTEND); 214 } 215 } 216 217 static bool virtio_net_started(VirtIONet *n, uint8_t status) 218 { 219 VirtIODevice *vdev = VIRTIO_DEVICE(n); 220 return (status & VIRTIO_CONFIG_S_DRIVER_OK) && 221 (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running; 222 } 223 224 static void virtio_net_announce_notify(VirtIONet *net) 225 { 226 VirtIODevice *vdev = VIRTIO_DEVICE(net); 227 trace_virtio_net_announce_notify(); 228 229 net->status |= VIRTIO_NET_S_ANNOUNCE; 230 virtio_notify_config(vdev); 231 } 232 233 static void virtio_net_announce_timer(void *opaque) 234 { 235 VirtIONet *n = opaque; 236 trace_virtio_net_announce_timer(n->announce_timer.round); 237 238 n->announce_timer.round--; 239 virtio_net_announce_notify(n); 240 } 241 242 static void virtio_net_announce(NetClientState *nc) 243 { 244 VirtIONet *n = qemu_get_nic_opaque(nc); 245 VirtIODevice *vdev = VIRTIO_DEVICE(n); 246 247 /* 248 * Make sure the virtio migration announcement timer isn't running 249 * If it is, let it trigger announcement so that we do not cause 250 * confusion. 251 */ 252 if (n->announce_timer.round) { 253 return; 254 } 255 256 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 257 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 258 virtio_net_announce_notify(n); 259 } 260 } 261 262 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status) 263 { 264 VirtIODevice *vdev = VIRTIO_DEVICE(n); 265 NetClientState *nc = qemu_get_queue(n->nic); 266 int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 267 int cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ? 268 n->max_ncs - n->max_queue_pairs : 0; 269 270 if (!get_vhost_net(nc->peer)) { 271 return; 272 } 273 274 if ((virtio_net_started(n, status) && !nc->peer->link_down) == 275 !!n->vhost_started) { 276 return; 277 } 278 if (!n->vhost_started) { 279 int r, i; 280 281 if (n->needs_vnet_hdr_swap) { 282 error_report("backend does not support %s vnet headers; " 283 "falling back on userspace virtio", 284 virtio_is_big_endian(vdev) ? "BE" : "LE"); 285 return; 286 } 287 288 /* Any packets outstanding? Purge them to avoid touching rings 289 * when vhost is running. 290 */ 291 for (i = 0; i < queue_pairs; i++) { 292 NetClientState *qnc = qemu_get_subqueue(n->nic, i); 293 294 /* Purge both directions: TX and RX. */ 295 qemu_net_queue_purge(qnc->peer->incoming_queue, qnc); 296 qemu_net_queue_purge(qnc->incoming_queue, qnc->peer); 297 } 298 299 if (virtio_has_feature(vdev->guest_features, VIRTIO_NET_F_MTU)) { 300 r = vhost_net_set_mtu(get_vhost_net(nc->peer), n->net_conf.mtu); 301 if (r < 0) { 302 error_report("%uBytes MTU not supported by the backend", 303 n->net_conf.mtu); 304 305 return; 306 } 307 } 308 309 n->vhost_started = 1; 310 r = vhost_net_start(vdev, n->nic->ncs, queue_pairs, cvq); 311 if (r < 0) { 312 error_report("unable to start vhost net: %d: " 313 "falling back on userspace virtio", -r); 314 n->vhost_started = 0; 315 } 316 } else { 317 vhost_net_stop(vdev, n->nic->ncs, queue_pairs, cvq); 318 n->vhost_started = 0; 319 } 320 } 321 322 static int virtio_net_set_vnet_endian_one(VirtIODevice *vdev, 323 NetClientState *peer, 324 bool enable) 325 { 326 if (virtio_is_big_endian(vdev)) { 327 return qemu_set_vnet_be(peer, enable); 328 } else { 329 return qemu_set_vnet_le(peer, enable); 330 } 331 } 332 333 static bool virtio_net_set_vnet_endian(VirtIODevice *vdev, NetClientState *ncs, 334 int queue_pairs, bool enable) 335 { 336 int i; 337 338 for (i = 0; i < queue_pairs; i++) { 339 if (virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, enable) < 0 && 340 enable) { 341 while (--i >= 0) { 342 virtio_net_set_vnet_endian_one(vdev, ncs[i].peer, false); 343 } 344 345 return true; 346 } 347 } 348 349 return false; 350 } 351 352 static void virtio_net_vnet_endian_status(VirtIONet *n, uint8_t status) 353 { 354 VirtIODevice *vdev = VIRTIO_DEVICE(n); 355 int queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 356 357 if (virtio_net_started(n, status)) { 358 /* Before using the device, we tell the network backend about the 359 * endianness to use when parsing vnet headers. If the backend 360 * can't do it, we fallback onto fixing the headers in the core 361 * virtio-net code. 362 */ 363 n->needs_vnet_hdr_swap = virtio_net_set_vnet_endian(vdev, n->nic->ncs, 364 queue_pairs, true); 365 } else if (virtio_net_started(n, vdev->status)) { 366 /* After using the device, we need to reset the network backend to 367 * the default (guest native endianness), otherwise the guest may 368 * lose network connectivity if it is rebooted into a different 369 * endianness. 370 */ 371 virtio_net_set_vnet_endian(vdev, n->nic->ncs, queue_pairs, false); 372 } 373 } 374 375 static void virtio_net_drop_tx_queue_data(VirtIODevice *vdev, VirtQueue *vq) 376 { 377 unsigned int dropped = virtqueue_drop_all(vq); 378 if (dropped) { 379 virtio_notify(vdev, vq); 380 } 381 } 382 383 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status) 384 { 385 VirtIONet *n = VIRTIO_NET(vdev); 386 VirtIONetQueue *q; 387 int i; 388 uint8_t queue_status; 389 390 virtio_net_vnet_endian_status(n, status); 391 virtio_net_vhost_status(n, status); 392 393 for (i = 0; i < n->max_queue_pairs; i++) { 394 NetClientState *ncs = qemu_get_subqueue(n->nic, i); 395 bool queue_started; 396 q = &n->vqs[i]; 397 398 if ((!n->multiqueue && i != 0) || i >= n->curr_queue_pairs) { 399 queue_status = 0; 400 } else { 401 queue_status = status; 402 } 403 queue_started = 404 virtio_net_started(n, queue_status) && !n->vhost_started; 405 406 if (queue_started) { 407 qemu_flush_queued_packets(ncs); 408 } 409 410 if (!q->tx_waiting) { 411 continue; 412 } 413 414 if (queue_started) { 415 if (q->tx_timer) { 416 timer_mod(q->tx_timer, 417 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 418 } else { 419 qemu_bh_schedule(q->tx_bh); 420 } 421 } else { 422 if (q->tx_timer) { 423 timer_del(q->tx_timer); 424 } else { 425 qemu_bh_cancel(q->tx_bh); 426 } 427 if ((n->status & VIRTIO_NET_S_LINK_UP) == 0 && 428 (queue_status & VIRTIO_CONFIG_S_DRIVER_OK) && 429 vdev->vm_running) { 430 /* if tx is waiting we are likely have some packets in tx queue 431 * and disabled notification */ 432 q->tx_waiting = 0; 433 virtio_queue_set_notification(q->tx_vq, 1); 434 virtio_net_drop_tx_queue_data(vdev, q->tx_vq); 435 } 436 } 437 } 438 } 439 440 static void virtio_net_set_link_status(NetClientState *nc) 441 { 442 VirtIONet *n = qemu_get_nic_opaque(nc); 443 VirtIODevice *vdev = VIRTIO_DEVICE(n); 444 uint16_t old_status = n->status; 445 446 if (nc->link_down) 447 n->status &= ~VIRTIO_NET_S_LINK_UP; 448 else 449 n->status |= VIRTIO_NET_S_LINK_UP; 450 451 if (n->status != old_status) 452 virtio_notify_config(vdev); 453 454 virtio_net_set_status(vdev, vdev->status); 455 } 456 457 static void rxfilter_notify(NetClientState *nc) 458 { 459 VirtIONet *n = qemu_get_nic_opaque(nc); 460 461 if (nc->rxfilter_notify_enabled) { 462 char *path = object_get_canonical_path(OBJECT(n->qdev)); 463 qapi_event_send_nic_rx_filter_changed(n->netclient_name, path); 464 g_free(path); 465 466 /* disable event notification to avoid events flooding */ 467 nc->rxfilter_notify_enabled = 0; 468 } 469 } 470 471 static intList *get_vlan_table(VirtIONet *n) 472 { 473 intList *list; 474 int i, j; 475 476 list = NULL; 477 for (i = 0; i < MAX_VLAN >> 5; i++) { 478 for (j = 0; n->vlans[i] && j <= 0x1f; j++) { 479 if (n->vlans[i] & (1U << j)) { 480 QAPI_LIST_PREPEND(list, (i << 5) + j); 481 } 482 } 483 } 484 485 return list; 486 } 487 488 static RxFilterInfo *virtio_net_query_rxfilter(NetClientState *nc) 489 { 490 VirtIONet *n = qemu_get_nic_opaque(nc); 491 VirtIODevice *vdev = VIRTIO_DEVICE(n); 492 RxFilterInfo *info; 493 strList *str_list; 494 int i; 495 496 info = g_malloc0(sizeof(*info)); 497 info->name = g_strdup(nc->name); 498 info->promiscuous = n->promisc; 499 500 if (n->nouni) { 501 info->unicast = RX_STATE_NONE; 502 } else if (n->alluni) { 503 info->unicast = RX_STATE_ALL; 504 } else { 505 info->unicast = RX_STATE_NORMAL; 506 } 507 508 if (n->nomulti) { 509 info->multicast = RX_STATE_NONE; 510 } else if (n->allmulti) { 511 info->multicast = RX_STATE_ALL; 512 } else { 513 info->multicast = RX_STATE_NORMAL; 514 } 515 516 info->broadcast_allowed = n->nobcast; 517 info->multicast_overflow = n->mac_table.multi_overflow; 518 info->unicast_overflow = n->mac_table.uni_overflow; 519 520 info->main_mac = qemu_mac_strdup_printf(n->mac); 521 522 str_list = NULL; 523 for (i = 0; i < n->mac_table.first_multi; i++) { 524 QAPI_LIST_PREPEND(str_list, 525 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN)); 526 } 527 info->unicast_table = str_list; 528 529 str_list = NULL; 530 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 531 QAPI_LIST_PREPEND(str_list, 532 qemu_mac_strdup_printf(n->mac_table.macs + i * ETH_ALEN)); 533 } 534 info->multicast_table = str_list; 535 info->vlan_table = get_vlan_table(n); 536 537 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VLAN)) { 538 info->vlan = RX_STATE_ALL; 539 } else if (!info->vlan_table) { 540 info->vlan = RX_STATE_NONE; 541 } else { 542 info->vlan = RX_STATE_NORMAL; 543 } 544 545 /* enable event notification after query */ 546 nc->rxfilter_notify_enabled = 1; 547 548 return info; 549 } 550 551 static void virtio_net_queue_reset(VirtIODevice *vdev, uint32_t queue_index) 552 { 553 VirtIONet *n = VIRTIO_NET(vdev); 554 NetClientState *nc; 555 556 /* validate queue_index and skip for cvq */ 557 if (queue_index >= n->max_queue_pairs * 2) { 558 return; 559 } 560 561 nc = qemu_get_subqueue(n->nic, vq2q(queue_index)); 562 563 if (!nc->peer) { 564 return; 565 } 566 567 if (get_vhost_net(nc->peer) && 568 nc->peer->info->type == NET_CLIENT_DRIVER_TAP) { 569 vhost_net_virtqueue_reset(vdev, nc, queue_index); 570 } 571 572 flush_or_purge_queued_packets(nc); 573 } 574 575 static void virtio_net_queue_enable(VirtIODevice *vdev, uint32_t queue_index) 576 { 577 VirtIONet *n = VIRTIO_NET(vdev); 578 NetClientState *nc; 579 int r; 580 581 /* validate queue_index and skip for cvq */ 582 if (queue_index >= n->max_queue_pairs * 2) { 583 return; 584 } 585 586 nc = qemu_get_subqueue(n->nic, vq2q(queue_index)); 587 588 if (!nc->peer || !vdev->vhost_started) { 589 return; 590 } 591 592 if (get_vhost_net(nc->peer) && 593 nc->peer->info->type == NET_CLIENT_DRIVER_TAP) { 594 r = vhost_net_virtqueue_restart(vdev, nc, queue_index); 595 if (r < 0) { 596 error_report("unable to restart vhost net virtqueue: %d, " 597 "when resetting the queue", queue_index); 598 } 599 } 600 } 601 602 static void virtio_net_reset(VirtIODevice *vdev) 603 { 604 VirtIONet *n = VIRTIO_NET(vdev); 605 int i; 606 607 /* Reset back to compatibility mode */ 608 n->promisc = 1; 609 n->allmulti = 0; 610 n->alluni = 0; 611 n->nomulti = 0; 612 n->nouni = 0; 613 n->nobcast = 0; 614 /* multiqueue is disabled by default */ 615 n->curr_queue_pairs = 1; 616 timer_del(n->announce_timer.tm); 617 n->announce_timer.round = 0; 618 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 619 620 /* Flush any MAC and VLAN filter table state */ 621 n->mac_table.in_use = 0; 622 n->mac_table.first_multi = 0; 623 n->mac_table.multi_overflow = 0; 624 n->mac_table.uni_overflow = 0; 625 memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN); 626 memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac)); 627 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 628 memset(n->vlans, 0, MAX_VLAN >> 3); 629 630 /* Flush any async TX */ 631 for (i = 0; i < n->max_queue_pairs; i++) { 632 flush_or_purge_queued_packets(qemu_get_subqueue(n->nic, i)); 633 } 634 } 635 636 static void peer_test_vnet_hdr(VirtIONet *n) 637 { 638 NetClientState *nc = qemu_get_queue(n->nic); 639 if (!nc->peer) { 640 return; 641 } 642 643 n->has_vnet_hdr = qemu_has_vnet_hdr(nc->peer); 644 } 645 646 static int peer_has_vnet_hdr(VirtIONet *n) 647 { 648 return n->has_vnet_hdr; 649 } 650 651 static int peer_has_ufo(VirtIONet *n) 652 { 653 if (!peer_has_vnet_hdr(n)) 654 return 0; 655 656 n->has_ufo = qemu_has_ufo(qemu_get_queue(n->nic)->peer); 657 658 return n->has_ufo; 659 } 660 661 static int peer_has_uso(VirtIONet *n) 662 { 663 if (!peer_has_vnet_hdr(n)) { 664 return 0; 665 } 666 667 return qemu_has_uso(qemu_get_queue(n->nic)->peer); 668 } 669 670 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs, 671 int version_1, int hash_report) 672 { 673 int i; 674 NetClientState *nc; 675 676 n->mergeable_rx_bufs = mergeable_rx_bufs; 677 678 /* 679 * Note: when extending the vnet header, please make sure to 680 * change the vnet header copying logic in virtio_net_flush_tx() 681 * as well. 682 */ 683 if (version_1) { 684 n->guest_hdr_len = hash_report ? 685 sizeof(struct virtio_net_hdr_v1_hash) : 686 sizeof(struct virtio_net_hdr_mrg_rxbuf); 687 n->rss_data.populate_hash = !!hash_report; 688 } else { 689 n->guest_hdr_len = n->mergeable_rx_bufs ? 690 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 691 sizeof(struct virtio_net_hdr); 692 } 693 694 for (i = 0; i < n->max_queue_pairs; i++) { 695 nc = qemu_get_subqueue(n->nic, i); 696 697 if (peer_has_vnet_hdr(n) && 698 qemu_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) { 699 qemu_set_vnet_hdr_len(nc->peer, n->guest_hdr_len); 700 n->host_hdr_len = n->guest_hdr_len; 701 } 702 } 703 } 704 705 static int virtio_net_max_tx_queue_size(VirtIONet *n) 706 { 707 NetClientState *peer = n->nic_conf.peers.ncs[0]; 708 709 /* 710 * Backends other than vhost-user or vhost-vdpa don't support max queue 711 * size. 712 */ 713 if (!peer) { 714 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 715 } 716 717 switch(peer->info->type) { 718 case NET_CLIENT_DRIVER_VHOST_USER: 719 case NET_CLIENT_DRIVER_VHOST_VDPA: 720 return VIRTQUEUE_MAX_SIZE; 721 default: 722 return VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE; 723 }; 724 } 725 726 static int peer_attach(VirtIONet *n, int index) 727 { 728 NetClientState *nc = qemu_get_subqueue(n->nic, index); 729 730 if (!nc->peer) { 731 return 0; 732 } 733 734 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 735 vhost_set_vring_enable(nc->peer, 1); 736 } 737 738 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 739 return 0; 740 } 741 742 if (n->max_queue_pairs == 1) { 743 return 0; 744 } 745 746 return tap_enable(nc->peer); 747 } 748 749 static int peer_detach(VirtIONet *n, int index) 750 { 751 NetClientState *nc = qemu_get_subqueue(n->nic, index); 752 753 if (!nc->peer) { 754 return 0; 755 } 756 757 if (nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_USER) { 758 vhost_set_vring_enable(nc->peer, 0); 759 } 760 761 if (nc->peer->info->type != NET_CLIENT_DRIVER_TAP) { 762 return 0; 763 } 764 765 return tap_disable(nc->peer); 766 } 767 768 static void virtio_net_set_queue_pairs(VirtIONet *n) 769 { 770 int i; 771 int r; 772 773 if (n->nic->peer_deleted) { 774 return; 775 } 776 777 for (i = 0; i < n->max_queue_pairs; i++) { 778 if (i < n->curr_queue_pairs) { 779 r = peer_attach(n, i); 780 assert(!r); 781 } else { 782 r = peer_detach(n, i); 783 assert(!r); 784 } 785 } 786 } 787 788 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue); 789 790 static uint64_t virtio_net_get_features(VirtIODevice *vdev, uint64_t features, 791 Error **errp) 792 { 793 VirtIONet *n = VIRTIO_NET(vdev); 794 NetClientState *nc = qemu_get_queue(n->nic); 795 796 /* Firstly sync all virtio-net possible supported features */ 797 features |= n->host_features; 798 799 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 800 801 if (!peer_has_vnet_hdr(n)) { 802 virtio_clear_feature(&features, VIRTIO_NET_F_CSUM); 803 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO4); 804 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_TSO6); 805 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_ECN); 806 807 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_CSUM); 808 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO4); 809 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_TSO6); 810 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ECN); 811 812 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_USO); 813 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO4); 814 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO6); 815 816 virtio_clear_feature(&features, VIRTIO_NET_F_HASH_REPORT); 817 } 818 819 if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) { 820 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_UFO); 821 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_UFO); 822 } 823 824 if (!peer_has_uso(n)) { 825 virtio_clear_feature(&features, VIRTIO_NET_F_HOST_USO); 826 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO4); 827 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_USO6); 828 } 829 830 if (!get_vhost_net(nc->peer)) { 831 return features; 832 } 833 834 if (!ebpf_rss_is_loaded(&n->ebpf_rss)) { 835 virtio_clear_feature(&features, VIRTIO_NET_F_RSS); 836 } 837 features = vhost_net_get_features(get_vhost_net(nc->peer), features); 838 vdev->backend_features = features; 839 840 if (n->mtu_bypass_backend && 841 (n->host_features & 1ULL << VIRTIO_NET_F_MTU)) { 842 features |= (1ULL << VIRTIO_NET_F_MTU); 843 } 844 845 /* 846 * Since GUEST_ANNOUNCE is emulated the feature bit could be set without 847 * enabled. This happens in the vDPA case. 848 * 849 * Make sure the feature set is not incoherent, as the driver could refuse 850 * to start. 851 * 852 * TODO: QEMU is able to emulate a CVQ just for guest_announce purposes, 853 * helping guest to notify the new location with vDPA devices that does not 854 * support it. 855 */ 856 if (!virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_CTRL_VQ)) { 857 virtio_clear_feature(&features, VIRTIO_NET_F_GUEST_ANNOUNCE); 858 } 859 860 return features; 861 } 862 863 static uint64_t virtio_net_bad_features(VirtIODevice *vdev) 864 { 865 uint64_t features = 0; 866 867 /* Linux kernel 2.6.25. It understood MAC (as everyone must), 868 * but also these: */ 869 virtio_add_feature(&features, VIRTIO_NET_F_MAC); 870 virtio_add_feature(&features, VIRTIO_NET_F_CSUM); 871 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO4); 872 virtio_add_feature(&features, VIRTIO_NET_F_HOST_TSO6); 873 virtio_add_feature(&features, VIRTIO_NET_F_HOST_ECN); 874 875 return features; 876 } 877 878 static void virtio_net_apply_guest_offloads(VirtIONet *n) 879 { 880 qemu_set_offload(qemu_get_queue(n->nic)->peer, 881 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)), 882 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)), 883 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)), 884 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)), 885 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)), 886 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_USO4)), 887 !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_USO6))); 888 } 889 890 static uint64_t virtio_net_guest_offloads_by_features(uint64_t features) 891 { 892 static const uint64_t guest_offloads_mask = 893 (1ULL << VIRTIO_NET_F_GUEST_CSUM) | 894 (1ULL << VIRTIO_NET_F_GUEST_TSO4) | 895 (1ULL << VIRTIO_NET_F_GUEST_TSO6) | 896 (1ULL << VIRTIO_NET_F_GUEST_ECN) | 897 (1ULL << VIRTIO_NET_F_GUEST_UFO) | 898 (1ULL << VIRTIO_NET_F_GUEST_USO4) | 899 (1ULL << VIRTIO_NET_F_GUEST_USO6); 900 901 return guest_offloads_mask & features; 902 } 903 904 uint64_t virtio_net_supported_guest_offloads(const VirtIONet *n) 905 { 906 VirtIODevice *vdev = VIRTIO_DEVICE(n); 907 return virtio_net_guest_offloads_by_features(vdev->guest_features); 908 } 909 910 typedef struct { 911 VirtIONet *n; 912 DeviceState *dev; 913 } FailoverDevice; 914 915 /** 916 * Set the failover primary device 917 * 918 * @opaque: FailoverId to setup 919 * @opts: opts for device we are handling 920 * @errp: returns an error if this function fails 921 */ 922 static int failover_set_primary(DeviceState *dev, void *opaque) 923 { 924 FailoverDevice *fdev = opaque; 925 PCIDevice *pci_dev = (PCIDevice *) 926 object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE); 927 928 if (!pci_dev) { 929 return 0; 930 } 931 932 if (!g_strcmp0(pci_dev->failover_pair_id, fdev->n->netclient_name)) { 933 fdev->dev = dev; 934 return 1; 935 } 936 937 return 0; 938 } 939 940 /** 941 * Find the primary device for this failover virtio-net 942 * 943 * @n: VirtIONet device 944 * @errp: returns an error if this function fails 945 */ 946 static DeviceState *failover_find_primary_device(VirtIONet *n) 947 { 948 FailoverDevice fdev = { 949 .n = n, 950 }; 951 952 qbus_walk_children(sysbus_get_default(), failover_set_primary, NULL, 953 NULL, NULL, &fdev); 954 return fdev.dev; 955 } 956 957 static void failover_add_primary(VirtIONet *n, Error **errp) 958 { 959 Error *err = NULL; 960 DeviceState *dev = failover_find_primary_device(n); 961 962 if (dev) { 963 return; 964 } 965 966 if (!n->primary_opts) { 967 error_setg(errp, "Primary device not found"); 968 error_append_hint(errp, "Virtio-net failover will not work. Make " 969 "sure primary device has parameter" 970 " failover_pair_id=%s\n", n->netclient_name); 971 return; 972 } 973 974 dev = qdev_device_add_from_qdict(n->primary_opts, 975 n->primary_opts_from_json, 976 &err); 977 if (err) { 978 qobject_unref(n->primary_opts); 979 n->primary_opts = NULL; 980 } else { 981 object_unref(OBJECT(dev)); 982 } 983 error_propagate(errp, err); 984 } 985 986 static void virtio_net_set_features(VirtIODevice *vdev, uint64_t features) 987 { 988 VirtIONet *n = VIRTIO_NET(vdev); 989 Error *err = NULL; 990 int i; 991 992 if (n->mtu_bypass_backend && 993 !virtio_has_feature(vdev->backend_features, VIRTIO_NET_F_MTU)) { 994 features &= ~(1ULL << VIRTIO_NET_F_MTU); 995 } 996 997 virtio_net_set_multiqueue(n, 998 virtio_has_feature(features, VIRTIO_NET_F_RSS) || 999 virtio_has_feature(features, VIRTIO_NET_F_MQ)); 1000 1001 virtio_net_set_mrg_rx_bufs(n, 1002 virtio_has_feature(features, 1003 VIRTIO_NET_F_MRG_RXBUF), 1004 virtio_has_feature(features, 1005 VIRTIO_F_VERSION_1), 1006 virtio_has_feature(features, 1007 VIRTIO_NET_F_HASH_REPORT)); 1008 1009 n->rsc4_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) && 1010 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO4); 1011 n->rsc6_enabled = virtio_has_feature(features, VIRTIO_NET_F_RSC_EXT) && 1012 virtio_has_feature(features, VIRTIO_NET_F_GUEST_TSO6); 1013 n->rss_data.redirect = virtio_has_feature(features, VIRTIO_NET_F_RSS); 1014 1015 if (n->has_vnet_hdr) { 1016 n->curr_guest_offloads = 1017 virtio_net_guest_offloads_by_features(features); 1018 virtio_net_apply_guest_offloads(n); 1019 } 1020 1021 for (i = 0; i < n->max_queue_pairs; i++) { 1022 NetClientState *nc = qemu_get_subqueue(n->nic, i); 1023 1024 if (!get_vhost_net(nc->peer)) { 1025 continue; 1026 } 1027 vhost_net_ack_features(get_vhost_net(nc->peer), features); 1028 1029 /* 1030 * keep acked_features in NetVhostUserState up-to-date so it 1031 * can't miss any features configured by guest virtio driver. 1032 */ 1033 vhost_net_save_acked_features(nc->peer); 1034 } 1035 1036 if (!virtio_has_feature(features, VIRTIO_NET_F_CTRL_VLAN)) { 1037 memset(n->vlans, 0xff, MAX_VLAN >> 3); 1038 } 1039 1040 if (virtio_has_feature(features, VIRTIO_NET_F_STANDBY)) { 1041 qapi_event_send_failover_negotiated(n->netclient_name); 1042 qatomic_set(&n->failover_primary_hidden, false); 1043 failover_add_primary(n, &err); 1044 if (err) { 1045 if (!qtest_enabled()) { 1046 warn_report_err(err); 1047 } else { 1048 error_free(err); 1049 } 1050 } 1051 } 1052 } 1053 1054 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd, 1055 struct iovec *iov, unsigned int iov_cnt) 1056 { 1057 uint8_t on; 1058 size_t s; 1059 NetClientState *nc = qemu_get_queue(n->nic); 1060 1061 s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on)); 1062 if (s != sizeof(on)) { 1063 return VIRTIO_NET_ERR; 1064 } 1065 1066 if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) { 1067 n->promisc = on; 1068 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) { 1069 n->allmulti = on; 1070 } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) { 1071 n->alluni = on; 1072 } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) { 1073 n->nomulti = on; 1074 } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) { 1075 n->nouni = on; 1076 } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) { 1077 n->nobcast = on; 1078 } else { 1079 return VIRTIO_NET_ERR; 1080 } 1081 1082 rxfilter_notify(nc); 1083 1084 return VIRTIO_NET_OK; 1085 } 1086 1087 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd, 1088 struct iovec *iov, unsigned int iov_cnt) 1089 { 1090 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1091 uint64_t offloads; 1092 size_t s; 1093 1094 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 1095 return VIRTIO_NET_ERR; 1096 } 1097 1098 s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads)); 1099 if (s != sizeof(offloads)) { 1100 return VIRTIO_NET_ERR; 1101 } 1102 1103 if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) { 1104 uint64_t supported_offloads; 1105 1106 offloads = virtio_ldq_p(vdev, &offloads); 1107 1108 if (!n->has_vnet_hdr) { 1109 return VIRTIO_NET_ERR; 1110 } 1111 1112 n->rsc4_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) && 1113 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO4); 1114 n->rsc6_enabled = virtio_has_feature(offloads, VIRTIO_NET_F_RSC_EXT) && 1115 virtio_has_feature(offloads, VIRTIO_NET_F_GUEST_TSO6); 1116 virtio_clear_feature(&offloads, VIRTIO_NET_F_RSC_EXT); 1117 1118 supported_offloads = virtio_net_supported_guest_offloads(n); 1119 if (offloads & ~supported_offloads) { 1120 return VIRTIO_NET_ERR; 1121 } 1122 1123 n->curr_guest_offloads = offloads; 1124 virtio_net_apply_guest_offloads(n); 1125 1126 return VIRTIO_NET_OK; 1127 } else { 1128 return VIRTIO_NET_ERR; 1129 } 1130 } 1131 1132 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd, 1133 struct iovec *iov, unsigned int iov_cnt) 1134 { 1135 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1136 struct virtio_net_ctrl_mac mac_data; 1137 size_t s; 1138 NetClientState *nc = qemu_get_queue(n->nic); 1139 1140 if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) { 1141 if (iov_size(iov, iov_cnt) != sizeof(n->mac)) { 1142 return VIRTIO_NET_ERR; 1143 } 1144 s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac)); 1145 assert(s == sizeof(n->mac)); 1146 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac); 1147 rxfilter_notify(nc); 1148 1149 return VIRTIO_NET_OK; 1150 } 1151 1152 if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) { 1153 return VIRTIO_NET_ERR; 1154 } 1155 1156 int in_use = 0; 1157 int first_multi = 0; 1158 uint8_t uni_overflow = 0; 1159 uint8_t multi_overflow = 0; 1160 uint8_t *macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 1161 1162 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 1163 sizeof(mac_data.entries)); 1164 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 1165 if (s != sizeof(mac_data.entries)) { 1166 goto error; 1167 } 1168 iov_discard_front(&iov, &iov_cnt, s); 1169 1170 if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) { 1171 goto error; 1172 } 1173 1174 if (mac_data.entries <= MAC_TABLE_ENTRIES) { 1175 s = iov_to_buf(iov, iov_cnt, 0, macs, 1176 mac_data.entries * ETH_ALEN); 1177 if (s != mac_data.entries * ETH_ALEN) { 1178 goto error; 1179 } 1180 in_use += mac_data.entries; 1181 } else { 1182 uni_overflow = 1; 1183 } 1184 1185 iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN); 1186 1187 first_multi = in_use; 1188 1189 s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries, 1190 sizeof(mac_data.entries)); 1191 mac_data.entries = virtio_ldl_p(vdev, &mac_data.entries); 1192 if (s != sizeof(mac_data.entries)) { 1193 goto error; 1194 } 1195 1196 iov_discard_front(&iov, &iov_cnt, s); 1197 1198 if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) { 1199 goto error; 1200 } 1201 1202 if (mac_data.entries <= MAC_TABLE_ENTRIES - in_use) { 1203 s = iov_to_buf(iov, iov_cnt, 0, &macs[in_use * ETH_ALEN], 1204 mac_data.entries * ETH_ALEN); 1205 if (s != mac_data.entries * ETH_ALEN) { 1206 goto error; 1207 } 1208 in_use += mac_data.entries; 1209 } else { 1210 multi_overflow = 1; 1211 } 1212 1213 n->mac_table.in_use = in_use; 1214 n->mac_table.first_multi = first_multi; 1215 n->mac_table.uni_overflow = uni_overflow; 1216 n->mac_table.multi_overflow = multi_overflow; 1217 memcpy(n->mac_table.macs, macs, MAC_TABLE_ENTRIES * ETH_ALEN); 1218 g_free(macs); 1219 rxfilter_notify(nc); 1220 1221 return VIRTIO_NET_OK; 1222 1223 error: 1224 g_free(macs); 1225 return VIRTIO_NET_ERR; 1226 } 1227 1228 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd, 1229 struct iovec *iov, unsigned int iov_cnt) 1230 { 1231 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1232 uint16_t vid; 1233 size_t s; 1234 NetClientState *nc = qemu_get_queue(n->nic); 1235 1236 s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid)); 1237 vid = virtio_lduw_p(vdev, &vid); 1238 if (s != sizeof(vid)) { 1239 return VIRTIO_NET_ERR; 1240 } 1241 1242 if (vid >= MAX_VLAN) 1243 return VIRTIO_NET_ERR; 1244 1245 if (cmd == VIRTIO_NET_CTRL_VLAN_ADD) 1246 n->vlans[vid >> 5] |= (1U << (vid & 0x1f)); 1247 else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL) 1248 n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f)); 1249 else 1250 return VIRTIO_NET_ERR; 1251 1252 rxfilter_notify(nc); 1253 1254 return VIRTIO_NET_OK; 1255 } 1256 1257 static int virtio_net_handle_announce(VirtIONet *n, uint8_t cmd, 1258 struct iovec *iov, unsigned int iov_cnt) 1259 { 1260 trace_virtio_net_handle_announce(n->announce_timer.round); 1261 if (cmd == VIRTIO_NET_CTRL_ANNOUNCE_ACK && 1262 n->status & VIRTIO_NET_S_ANNOUNCE) { 1263 n->status &= ~VIRTIO_NET_S_ANNOUNCE; 1264 if (n->announce_timer.round) { 1265 qemu_announce_timer_step(&n->announce_timer); 1266 } 1267 return VIRTIO_NET_OK; 1268 } else { 1269 return VIRTIO_NET_ERR; 1270 } 1271 } 1272 1273 static void virtio_net_detach_epbf_rss(VirtIONet *n); 1274 1275 static void virtio_net_disable_rss(VirtIONet *n) 1276 { 1277 if (n->rss_data.enabled) { 1278 trace_virtio_net_rss_disable(); 1279 } 1280 n->rss_data.enabled = false; 1281 1282 virtio_net_detach_epbf_rss(n); 1283 } 1284 1285 static bool virtio_net_attach_ebpf_to_backend(NICState *nic, int prog_fd) 1286 { 1287 NetClientState *nc = qemu_get_peer(qemu_get_queue(nic), 0); 1288 if (nc == NULL || nc->info->set_steering_ebpf == NULL) { 1289 return false; 1290 } 1291 1292 return nc->info->set_steering_ebpf(nc, prog_fd); 1293 } 1294 1295 static void rss_data_to_rss_config(struct VirtioNetRssData *data, 1296 struct EBPFRSSConfig *config) 1297 { 1298 config->redirect = data->redirect; 1299 config->populate_hash = data->populate_hash; 1300 config->hash_types = data->hash_types; 1301 config->indirections_len = data->indirections_len; 1302 config->default_queue = data->default_queue; 1303 } 1304 1305 static bool virtio_net_attach_epbf_rss(VirtIONet *n) 1306 { 1307 struct EBPFRSSConfig config = {}; 1308 1309 if (!ebpf_rss_is_loaded(&n->ebpf_rss)) { 1310 return false; 1311 } 1312 1313 rss_data_to_rss_config(&n->rss_data, &config); 1314 1315 if (!ebpf_rss_set_all(&n->ebpf_rss, &config, 1316 n->rss_data.indirections_table, n->rss_data.key)) { 1317 return false; 1318 } 1319 1320 if (!virtio_net_attach_ebpf_to_backend(n->nic, n->ebpf_rss.program_fd)) { 1321 return false; 1322 } 1323 1324 return true; 1325 } 1326 1327 static void virtio_net_detach_epbf_rss(VirtIONet *n) 1328 { 1329 virtio_net_attach_ebpf_to_backend(n->nic, -1); 1330 } 1331 1332 static bool virtio_net_load_ebpf_fds(VirtIONet *n, Error **errp) 1333 { 1334 int fds[EBPF_RSS_MAX_FDS] = { [0 ... EBPF_RSS_MAX_FDS - 1] = -1}; 1335 int ret = true; 1336 int i = 0; 1337 1338 ERRP_GUARD(); 1339 1340 if (n->nr_ebpf_rss_fds != EBPF_RSS_MAX_FDS) { 1341 error_setg(errp, 1342 "Expected %d file descriptors but got %d", 1343 EBPF_RSS_MAX_FDS, n->nr_ebpf_rss_fds); 1344 return false; 1345 } 1346 1347 for (i = 0; i < n->nr_ebpf_rss_fds; i++) { 1348 fds[i] = monitor_fd_param(monitor_cur(), n->ebpf_rss_fds[i], errp); 1349 if (*errp) { 1350 ret = false; 1351 goto exit; 1352 } 1353 } 1354 1355 ret = ebpf_rss_load_fds(&n->ebpf_rss, fds[0], fds[1], fds[2], fds[3]); 1356 1357 exit: 1358 if (!ret || *errp) { 1359 for (i = 0; i < n->nr_ebpf_rss_fds && fds[i] != -1; i++) { 1360 close(fds[i]); 1361 } 1362 } 1363 1364 return ret; 1365 } 1366 1367 static bool virtio_net_load_ebpf(VirtIONet *n, Error **errp) 1368 { 1369 bool ret = false; 1370 1371 if (virtio_net_attach_ebpf_to_backend(n->nic, -1)) { 1372 if (!(n->ebpf_rss_fds 1373 && virtio_net_load_ebpf_fds(n, errp))) { 1374 ret = ebpf_rss_load(&n->ebpf_rss); 1375 } 1376 } 1377 1378 return ret; 1379 } 1380 1381 static void virtio_net_unload_ebpf(VirtIONet *n) 1382 { 1383 virtio_net_attach_ebpf_to_backend(n->nic, -1); 1384 ebpf_rss_unload(&n->ebpf_rss); 1385 } 1386 1387 static uint16_t virtio_net_handle_rss(VirtIONet *n, 1388 struct iovec *iov, 1389 unsigned int iov_cnt, 1390 bool do_rss) 1391 { 1392 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1393 struct virtio_net_rss_config cfg; 1394 size_t s, offset = 0, size_get; 1395 uint16_t queue_pairs, i; 1396 struct { 1397 uint16_t us; 1398 uint8_t b; 1399 } QEMU_PACKED temp; 1400 const char *err_msg = ""; 1401 uint32_t err_value = 0; 1402 1403 if (do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_RSS)) { 1404 err_msg = "RSS is not negotiated"; 1405 goto error; 1406 } 1407 if (!do_rss && !virtio_vdev_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) { 1408 err_msg = "Hash report is not negotiated"; 1409 goto error; 1410 } 1411 size_get = offsetof(struct virtio_net_rss_config, indirection_table); 1412 s = iov_to_buf(iov, iov_cnt, offset, &cfg, size_get); 1413 if (s != size_get) { 1414 err_msg = "Short command buffer"; 1415 err_value = (uint32_t)s; 1416 goto error; 1417 } 1418 n->rss_data.hash_types = virtio_ldl_p(vdev, &cfg.hash_types); 1419 n->rss_data.indirections_len = 1420 virtio_lduw_p(vdev, &cfg.indirection_table_mask); 1421 n->rss_data.indirections_len++; 1422 if (!do_rss) { 1423 n->rss_data.indirections_len = 1; 1424 } 1425 if (!is_power_of_2(n->rss_data.indirections_len)) { 1426 err_msg = "Invalid size of indirection table"; 1427 err_value = n->rss_data.indirections_len; 1428 goto error; 1429 } 1430 if (n->rss_data.indirections_len > VIRTIO_NET_RSS_MAX_TABLE_LEN) { 1431 err_msg = "Too large indirection table"; 1432 err_value = n->rss_data.indirections_len; 1433 goto error; 1434 } 1435 n->rss_data.default_queue = do_rss ? 1436 virtio_lduw_p(vdev, &cfg.unclassified_queue) : 0; 1437 if (n->rss_data.default_queue >= n->max_queue_pairs) { 1438 err_msg = "Invalid default queue"; 1439 err_value = n->rss_data.default_queue; 1440 goto error; 1441 } 1442 offset += size_get; 1443 size_get = sizeof(uint16_t) * n->rss_data.indirections_len; 1444 g_free(n->rss_data.indirections_table); 1445 n->rss_data.indirections_table = g_malloc(size_get); 1446 if (!n->rss_data.indirections_table) { 1447 err_msg = "Can't allocate indirections table"; 1448 err_value = n->rss_data.indirections_len; 1449 goto error; 1450 } 1451 s = iov_to_buf(iov, iov_cnt, offset, 1452 n->rss_data.indirections_table, size_get); 1453 if (s != size_get) { 1454 err_msg = "Short indirection table buffer"; 1455 err_value = (uint32_t)s; 1456 goto error; 1457 } 1458 for (i = 0; i < n->rss_data.indirections_len; ++i) { 1459 uint16_t val = n->rss_data.indirections_table[i]; 1460 n->rss_data.indirections_table[i] = virtio_lduw_p(vdev, &val); 1461 } 1462 offset += size_get; 1463 size_get = sizeof(temp); 1464 s = iov_to_buf(iov, iov_cnt, offset, &temp, size_get); 1465 if (s != size_get) { 1466 err_msg = "Can't get queue_pairs"; 1467 err_value = (uint32_t)s; 1468 goto error; 1469 } 1470 queue_pairs = do_rss ? virtio_lduw_p(vdev, &temp.us) : n->curr_queue_pairs; 1471 if (queue_pairs == 0 || queue_pairs > n->max_queue_pairs) { 1472 err_msg = "Invalid number of queue_pairs"; 1473 err_value = queue_pairs; 1474 goto error; 1475 } 1476 if (temp.b > VIRTIO_NET_RSS_MAX_KEY_SIZE) { 1477 err_msg = "Invalid key size"; 1478 err_value = temp.b; 1479 goto error; 1480 } 1481 if (!temp.b && n->rss_data.hash_types) { 1482 err_msg = "No key provided"; 1483 err_value = 0; 1484 goto error; 1485 } 1486 if (!temp.b && !n->rss_data.hash_types) { 1487 virtio_net_disable_rss(n); 1488 return queue_pairs; 1489 } 1490 offset += size_get; 1491 size_get = temp.b; 1492 s = iov_to_buf(iov, iov_cnt, offset, n->rss_data.key, size_get); 1493 if (s != size_get) { 1494 err_msg = "Can get key buffer"; 1495 err_value = (uint32_t)s; 1496 goto error; 1497 } 1498 n->rss_data.enabled = true; 1499 1500 if (!n->rss_data.populate_hash) { 1501 if (!virtio_net_attach_epbf_rss(n)) { 1502 /* EBPF must be loaded for vhost */ 1503 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) { 1504 warn_report("Can't load eBPF RSS for vhost"); 1505 goto error; 1506 } 1507 /* fallback to software RSS */ 1508 warn_report("Can't load eBPF RSS - fallback to software RSS"); 1509 n->rss_data.enabled_software_rss = true; 1510 } 1511 } else { 1512 /* use software RSS for hash populating */ 1513 /* and detach eBPF if was loaded before */ 1514 virtio_net_detach_epbf_rss(n); 1515 n->rss_data.enabled_software_rss = true; 1516 } 1517 1518 trace_virtio_net_rss_enable(n->rss_data.hash_types, 1519 n->rss_data.indirections_len, 1520 temp.b); 1521 return queue_pairs; 1522 error: 1523 trace_virtio_net_rss_error(err_msg, err_value); 1524 virtio_net_disable_rss(n); 1525 return 0; 1526 } 1527 1528 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd, 1529 struct iovec *iov, unsigned int iov_cnt) 1530 { 1531 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1532 uint16_t queue_pairs; 1533 NetClientState *nc = qemu_get_queue(n->nic); 1534 1535 virtio_net_disable_rss(n); 1536 if (cmd == VIRTIO_NET_CTRL_MQ_HASH_CONFIG) { 1537 queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, false); 1538 return queue_pairs ? VIRTIO_NET_OK : VIRTIO_NET_ERR; 1539 } 1540 if (cmd == VIRTIO_NET_CTRL_MQ_RSS_CONFIG) { 1541 queue_pairs = virtio_net_handle_rss(n, iov, iov_cnt, true); 1542 } else if (cmd == VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) { 1543 struct virtio_net_ctrl_mq mq; 1544 size_t s; 1545 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_MQ)) { 1546 return VIRTIO_NET_ERR; 1547 } 1548 s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq)); 1549 if (s != sizeof(mq)) { 1550 return VIRTIO_NET_ERR; 1551 } 1552 queue_pairs = virtio_lduw_p(vdev, &mq.virtqueue_pairs); 1553 1554 } else { 1555 return VIRTIO_NET_ERR; 1556 } 1557 1558 if (queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 1559 queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || 1560 queue_pairs > n->max_queue_pairs || 1561 !n->multiqueue) { 1562 return VIRTIO_NET_ERR; 1563 } 1564 1565 n->curr_queue_pairs = queue_pairs; 1566 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 1567 /* 1568 * Avoid updating the backend for a vdpa device: We're only interested 1569 * in updating the device model queues. 1570 */ 1571 return VIRTIO_NET_OK; 1572 } 1573 /* stop the backend before changing the number of queue_pairs to avoid handling a 1574 * disabled queue */ 1575 virtio_net_set_status(vdev, vdev->status); 1576 virtio_net_set_queue_pairs(n); 1577 1578 return VIRTIO_NET_OK; 1579 } 1580 1581 size_t virtio_net_handle_ctrl_iov(VirtIODevice *vdev, 1582 const struct iovec *in_sg, unsigned in_num, 1583 const struct iovec *out_sg, 1584 unsigned out_num) 1585 { 1586 VirtIONet *n = VIRTIO_NET(vdev); 1587 struct virtio_net_ctrl_hdr ctrl; 1588 virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 1589 size_t s; 1590 struct iovec *iov, *iov2; 1591 1592 if (iov_size(in_sg, in_num) < sizeof(status) || 1593 iov_size(out_sg, out_num) < sizeof(ctrl)) { 1594 virtio_error(vdev, "virtio-net ctrl missing headers"); 1595 return 0; 1596 } 1597 1598 iov2 = iov = g_memdup2(out_sg, sizeof(struct iovec) * out_num); 1599 s = iov_to_buf(iov, out_num, 0, &ctrl, sizeof(ctrl)); 1600 iov_discard_front(&iov, &out_num, sizeof(ctrl)); 1601 if (s != sizeof(ctrl)) { 1602 status = VIRTIO_NET_ERR; 1603 } else if (ctrl.class == VIRTIO_NET_CTRL_RX) { 1604 status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, out_num); 1605 } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) { 1606 status = virtio_net_handle_mac(n, ctrl.cmd, iov, out_num); 1607 } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) { 1608 status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, out_num); 1609 } else if (ctrl.class == VIRTIO_NET_CTRL_ANNOUNCE) { 1610 status = virtio_net_handle_announce(n, ctrl.cmd, iov, out_num); 1611 } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) { 1612 status = virtio_net_handle_mq(n, ctrl.cmd, iov, out_num); 1613 } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) { 1614 status = virtio_net_handle_offloads(n, ctrl.cmd, iov, out_num); 1615 } 1616 1617 s = iov_from_buf(in_sg, in_num, 0, &status, sizeof(status)); 1618 assert(s == sizeof(status)); 1619 1620 g_free(iov2); 1621 return sizeof(status); 1622 } 1623 1624 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq) 1625 { 1626 VirtQueueElement *elem; 1627 1628 for (;;) { 1629 size_t written; 1630 elem = virtqueue_pop(vq, sizeof(VirtQueueElement)); 1631 if (!elem) { 1632 break; 1633 } 1634 1635 written = virtio_net_handle_ctrl_iov(vdev, elem->in_sg, elem->in_num, 1636 elem->out_sg, elem->out_num); 1637 if (written > 0) { 1638 virtqueue_push(vq, elem, written); 1639 virtio_notify(vdev, vq); 1640 g_free(elem); 1641 } else { 1642 virtqueue_detach_element(vq, elem, 0); 1643 g_free(elem); 1644 break; 1645 } 1646 } 1647 } 1648 1649 /* RX */ 1650 1651 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq) 1652 { 1653 VirtIONet *n = VIRTIO_NET(vdev); 1654 int queue_index = vq2q(virtio_get_queue_index(vq)); 1655 1656 qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index)); 1657 } 1658 1659 static bool virtio_net_can_receive(NetClientState *nc) 1660 { 1661 VirtIONet *n = qemu_get_nic_opaque(nc); 1662 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1663 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1664 1665 if (!vdev->vm_running) { 1666 return false; 1667 } 1668 1669 if (nc->queue_index >= n->curr_queue_pairs) { 1670 return false; 1671 } 1672 1673 if (!virtio_queue_ready(q->rx_vq) || 1674 !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 1675 return false; 1676 } 1677 1678 return true; 1679 } 1680 1681 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize) 1682 { 1683 VirtIONet *n = q->n; 1684 if (virtio_queue_empty(q->rx_vq) || 1685 (n->mergeable_rx_bufs && 1686 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 1687 virtio_queue_set_notification(q->rx_vq, 1); 1688 1689 /* To avoid a race condition where the guest has made some buffers 1690 * available after the above check but before notification was 1691 * enabled, check for available buffers again. 1692 */ 1693 if (virtio_queue_empty(q->rx_vq) || 1694 (n->mergeable_rx_bufs && 1695 !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) { 1696 return 0; 1697 } 1698 } 1699 1700 virtio_queue_set_notification(q->rx_vq, 0); 1701 return 1; 1702 } 1703 1704 static void virtio_net_hdr_swap(VirtIODevice *vdev, struct virtio_net_hdr *hdr) 1705 { 1706 virtio_tswap16s(vdev, &hdr->hdr_len); 1707 virtio_tswap16s(vdev, &hdr->gso_size); 1708 virtio_tswap16s(vdev, &hdr->csum_start); 1709 virtio_tswap16s(vdev, &hdr->csum_offset); 1710 } 1711 1712 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so 1713 * it never finds out that the packets don't have valid checksums. This 1714 * causes dhclient to get upset. Fedora's carried a patch for ages to 1715 * fix this with Xen but it hasn't appeared in an upstream release of 1716 * dhclient yet. 1717 * 1718 * To avoid breaking existing guests, we catch udp packets and add 1719 * checksums. This is terrible but it's better than hacking the guest 1720 * kernels. 1721 * 1722 * N.B. if we introduce a zero-copy API, this operation is no longer free so 1723 * we should provide a mechanism to disable it to avoid polluting the host 1724 * cache. 1725 */ 1726 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr, 1727 uint8_t *buf, size_t size) 1728 { 1729 if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */ 1730 (size > 27 && size < 1500) && /* normal sized MTU */ 1731 (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */ 1732 (buf[23] == 17) && /* ip.protocol == UDP */ 1733 (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */ 1734 net_checksum_calculate(buf, size, CSUM_UDP); 1735 hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM; 1736 } 1737 } 1738 1739 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt, 1740 const void *buf, size_t size) 1741 { 1742 if (n->has_vnet_hdr) { 1743 /* FIXME this cast is evil */ 1744 void *wbuf = (void *)buf; 1745 work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, 1746 size - n->host_hdr_len); 1747 1748 if (n->needs_vnet_hdr_swap) { 1749 virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf); 1750 } 1751 iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr)); 1752 } else { 1753 struct virtio_net_hdr hdr = { 1754 .flags = 0, 1755 .gso_type = VIRTIO_NET_HDR_GSO_NONE 1756 }; 1757 iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr); 1758 } 1759 } 1760 1761 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size) 1762 { 1763 static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff}; 1764 static const uint8_t vlan[] = {0x81, 0x00}; 1765 uint8_t *ptr = (uint8_t *)buf; 1766 int i; 1767 1768 if (n->promisc) 1769 return 1; 1770 1771 ptr += n->host_hdr_len; 1772 1773 if (!memcmp(&ptr[12], vlan, sizeof(vlan))) { 1774 int vid = lduw_be_p(ptr + 14) & 0xfff; 1775 if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f)))) 1776 return 0; 1777 } 1778 1779 if (ptr[0] & 1) { // multicast 1780 if (!memcmp(ptr, bcast, sizeof(bcast))) { 1781 return !n->nobcast; 1782 } else if (n->nomulti) { 1783 return 0; 1784 } else if (n->allmulti || n->mac_table.multi_overflow) { 1785 return 1; 1786 } 1787 1788 for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) { 1789 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1790 return 1; 1791 } 1792 } 1793 } else { // unicast 1794 if (n->nouni) { 1795 return 0; 1796 } else if (n->alluni || n->mac_table.uni_overflow) { 1797 return 1; 1798 } else if (!memcmp(ptr, n->mac, ETH_ALEN)) { 1799 return 1; 1800 } 1801 1802 for (i = 0; i < n->mac_table.first_multi; i++) { 1803 if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) { 1804 return 1; 1805 } 1806 } 1807 } 1808 1809 return 0; 1810 } 1811 1812 static uint8_t virtio_net_get_hash_type(bool hasip4, 1813 bool hasip6, 1814 EthL4HdrProto l4hdr_proto, 1815 uint32_t types) 1816 { 1817 if (hasip4) { 1818 switch (l4hdr_proto) { 1819 case ETH_L4_HDR_PROTO_TCP: 1820 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) { 1821 return NetPktRssIpV4Tcp; 1822 } 1823 break; 1824 1825 case ETH_L4_HDR_PROTO_UDP: 1826 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) { 1827 return NetPktRssIpV4Udp; 1828 } 1829 break; 1830 1831 default: 1832 break; 1833 } 1834 1835 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv4) { 1836 return NetPktRssIpV4; 1837 } 1838 } else if (hasip6) { 1839 switch (l4hdr_proto) { 1840 case ETH_L4_HDR_PROTO_TCP: 1841 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCP_EX) { 1842 return NetPktRssIpV6TcpEx; 1843 } 1844 if (types & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) { 1845 return NetPktRssIpV6Tcp; 1846 } 1847 break; 1848 1849 case ETH_L4_HDR_PROTO_UDP: 1850 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDP_EX) { 1851 return NetPktRssIpV6UdpEx; 1852 } 1853 if (types & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) { 1854 return NetPktRssIpV6Udp; 1855 } 1856 break; 1857 1858 default: 1859 break; 1860 } 1861 1862 if (types & VIRTIO_NET_RSS_HASH_TYPE_IP_EX) { 1863 return NetPktRssIpV6Ex; 1864 } 1865 if (types & VIRTIO_NET_RSS_HASH_TYPE_IPv6) { 1866 return NetPktRssIpV6; 1867 } 1868 } 1869 return 0xff; 1870 } 1871 1872 static void virtio_set_packet_hash(const uint8_t *buf, uint8_t report, 1873 uint32_t hash) 1874 { 1875 struct virtio_net_hdr_v1_hash *hdr = (void *)buf; 1876 hdr->hash_value = hash; 1877 hdr->hash_report = report; 1878 } 1879 1880 static int virtio_net_process_rss(NetClientState *nc, const uint8_t *buf, 1881 size_t size) 1882 { 1883 VirtIONet *n = qemu_get_nic_opaque(nc); 1884 unsigned int index = nc->queue_index, new_index = index; 1885 struct NetRxPkt *pkt = n->rx_pkt; 1886 uint8_t net_hash_type; 1887 uint32_t hash; 1888 bool hasip4, hasip6; 1889 EthL4HdrProto l4hdr_proto; 1890 static const uint8_t reports[NetPktRssIpV6UdpEx + 1] = { 1891 VIRTIO_NET_HASH_REPORT_IPv4, 1892 VIRTIO_NET_HASH_REPORT_TCPv4, 1893 VIRTIO_NET_HASH_REPORT_TCPv6, 1894 VIRTIO_NET_HASH_REPORT_IPv6, 1895 VIRTIO_NET_HASH_REPORT_IPv6_EX, 1896 VIRTIO_NET_HASH_REPORT_TCPv6_EX, 1897 VIRTIO_NET_HASH_REPORT_UDPv4, 1898 VIRTIO_NET_HASH_REPORT_UDPv6, 1899 VIRTIO_NET_HASH_REPORT_UDPv6_EX 1900 }; 1901 struct iovec iov = { 1902 .iov_base = (void *)buf, 1903 .iov_len = size 1904 }; 1905 1906 net_rx_pkt_set_protocols(pkt, &iov, 1, n->host_hdr_len); 1907 net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto); 1908 net_hash_type = virtio_net_get_hash_type(hasip4, hasip6, l4hdr_proto, 1909 n->rss_data.hash_types); 1910 if (net_hash_type > NetPktRssIpV6UdpEx) { 1911 if (n->rss_data.populate_hash) { 1912 virtio_set_packet_hash(buf, VIRTIO_NET_HASH_REPORT_NONE, 0); 1913 } 1914 return n->rss_data.redirect ? n->rss_data.default_queue : -1; 1915 } 1916 1917 hash = net_rx_pkt_calc_rss_hash(pkt, net_hash_type, n->rss_data.key); 1918 1919 if (n->rss_data.populate_hash) { 1920 virtio_set_packet_hash(buf, reports[net_hash_type], hash); 1921 } 1922 1923 if (n->rss_data.redirect) { 1924 new_index = hash & (n->rss_data.indirections_len - 1); 1925 new_index = n->rss_data.indirections_table[new_index]; 1926 } 1927 1928 return (index == new_index) ? -1 : new_index; 1929 } 1930 1931 static ssize_t virtio_net_receive_rcu(NetClientState *nc, const uint8_t *buf, 1932 size_t size, bool no_rss) 1933 { 1934 VirtIONet *n = qemu_get_nic_opaque(nc); 1935 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 1936 VirtIODevice *vdev = VIRTIO_DEVICE(n); 1937 VirtQueueElement *elems[VIRTQUEUE_MAX_SIZE]; 1938 size_t lens[VIRTQUEUE_MAX_SIZE]; 1939 struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE]; 1940 struct virtio_net_hdr_mrg_rxbuf mhdr; 1941 unsigned mhdr_cnt = 0; 1942 size_t offset, i, guest_offset, j; 1943 ssize_t err; 1944 1945 if (!virtio_net_can_receive(nc)) { 1946 return -1; 1947 } 1948 1949 if (!no_rss && n->rss_data.enabled && n->rss_data.enabled_software_rss) { 1950 int index = virtio_net_process_rss(nc, buf, size); 1951 if (index >= 0) { 1952 NetClientState *nc2 = qemu_get_subqueue(n->nic, index); 1953 return virtio_net_receive_rcu(nc2, buf, size, true); 1954 } 1955 } 1956 1957 /* hdr_len refers to the header we supply to the guest */ 1958 if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) { 1959 return 0; 1960 } 1961 1962 if (!receive_filter(n, buf, size)) 1963 return size; 1964 1965 offset = i = 0; 1966 1967 while (offset < size) { 1968 VirtQueueElement *elem; 1969 int len, total; 1970 const struct iovec *sg; 1971 1972 total = 0; 1973 1974 if (i == VIRTQUEUE_MAX_SIZE) { 1975 virtio_error(vdev, "virtio-net unexpected long buffer chain"); 1976 err = size; 1977 goto err; 1978 } 1979 1980 elem = virtqueue_pop(q->rx_vq, sizeof(VirtQueueElement)); 1981 if (!elem) { 1982 if (i) { 1983 virtio_error(vdev, "virtio-net unexpected empty queue: " 1984 "i %zd mergeable %d offset %zd, size %zd, " 1985 "guest hdr len %zd, host hdr len %zd " 1986 "guest features 0x%" PRIx64, 1987 i, n->mergeable_rx_bufs, offset, size, 1988 n->guest_hdr_len, n->host_hdr_len, 1989 vdev->guest_features); 1990 } 1991 err = -1; 1992 goto err; 1993 } 1994 1995 if (elem->in_num < 1) { 1996 virtio_error(vdev, 1997 "virtio-net receive queue contains no in buffers"); 1998 virtqueue_detach_element(q->rx_vq, elem, 0); 1999 g_free(elem); 2000 err = -1; 2001 goto err; 2002 } 2003 2004 sg = elem->in_sg; 2005 if (i == 0) { 2006 assert(offset == 0); 2007 if (n->mergeable_rx_bufs) { 2008 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg), 2009 sg, elem->in_num, 2010 offsetof(typeof(mhdr), num_buffers), 2011 sizeof(mhdr.num_buffers)); 2012 } 2013 2014 receive_header(n, sg, elem->in_num, buf, size); 2015 if (n->rss_data.populate_hash) { 2016 offset = sizeof(mhdr); 2017 iov_from_buf(sg, elem->in_num, offset, 2018 buf + offset, n->host_hdr_len - sizeof(mhdr)); 2019 } 2020 offset = n->host_hdr_len; 2021 total += n->guest_hdr_len; 2022 guest_offset = n->guest_hdr_len; 2023 } else { 2024 guest_offset = 0; 2025 } 2026 2027 /* copy in packet. ugh */ 2028 len = iov_from_buf(sg, elem->in_num, guest_offset, 2029 buf + offset, size - offset); 2030 total += len; 2031 offset += len; 2032 /* If buffers can't be merged, at this point we 2033 * must have consumed the complete packet. 2034 * Otherwise, drop it. */ 2035 if (!n->mergeable_rx_bufs && offset < size) { 2036 virtqueue_unpop(q->rx_vq, elem, total); 2037 g_free(elem); 2038 err = size; 2039 goto err; 2040 } 2041 2042 elems[i] = elem; 2043 lens[i] = total; 2044 i++; 2045 } 2046 2047 if (mhdr_cnt) { 2048 virtio_stw_p(vdev, &mhdr.num_buffers, i); 2049 iov_from_buf(mhdr_sg, mhdr_cnt, 2050 0, 2051 &mhdr.num_buffers, sizeof mhdr.num_buffers); 2052 } 2053 2054 for (j = 0; j < i; j++) { 2055 /* signal other side */ 2056 virtqueue_fill(q->rx_vq, elems[j], lens[j], j); 2057 g_free(elems[j]); 2058 } 2059 2060 virtqueue_flush(q->rx_vq, i); 2061 virtio_notify(vdev, q->rx_vq); 2062 2063 return size; 2064 2065 err: 2066 for (j = 0; j < i; j++) { 2067 virtqueue_detach_element(q->rx_vq, elems[j], lens[j]); 2068 g_free(elems[j]); 2069 } 2070 2071 return err; 2072 } 2073 2074 static ssize_t virtio_net_do_receive(NetClientState *nc, const uint8_t *buf, 2075 size_t size) 2076 { 2077 RCU_READ_LOCK_GUARD(); 2078 2079 return virtio_net_receive_rcu(nc, buf, size, false); 2080 } 2081 2082 static void virtio_net_rsc_extract_unit4(VirtioNetRscChain *chain, 2083 const uint8_t *buf, 2084 VirtioNetRscUnit *unit) 2085 { 2086 uint16_t ip_hdrlen; 2087 struct ip_header *ip; 2088 2089 ip = (struct ip_header *)(buf + chain->n->guest_hdr_len 2090 + sizeof(struct eth_header)); 2091 unit->ip = (void *)ip; 2092 ip_hdrlen = (ip->ip_ver_len & 0xF) << 2; 2093 unit->ip_plen = &ip->ip_len; 2094 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) + ip_hdrlen); 2095 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10; 2096 unit->payload = htons(*unit->ip_plen) - ip_hdrlen - unit->tcp_hdrlen; 2097 } 2098 2099 static void virtio_net_rsc_extract_unit6(VirtioNetRscChain *chain, 2100 const uint8_t *buf, 2101 VirtioNetRscUnit *unit) 2102 { 2103 struct ip6_header *ip6; 2104 2105 ip6 = (struct ip6_header *)(buf + chain->n->guest_hdr_len 2106 + sizeof(struct eth_header)); 2107 unit->ip = ip6; 2108 unit->ip_plen = &(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen); 2109 unit->tcp = (struct tcp_header *)(((uint8_t *)unit->ip) 2110 + sizeof(struct ip6_header)); 2111 unit->tcp_hdrlen = (htons(unit->tcp->th_offset_flags) & 0xF000) >> 10; 2112 2113 /* There is a difference between payload length in ipv4 and v6, 2114 ip header is excluded in ipv6 */ 2115 unit->payload = htons(*unit->ip_plen) - unit->tcp_hdrlen; 2116 } 2117 2118 static size_t virtio_net_rsc_drain_seg(VirtioNetRscChain *chain, 2119 VirtioNetRscSeg *seg) 2120 { 2121 int ret; 2122 struct virtio_net_hdr_v1 *h; 2123 2124 h = (struct virtio_net_hdr_v1 *)seg->buf; 2125 h->flags = 0; 2126 h->gso_type = VIRTIO_NET_HDR_GSO_NONE; 2127 2128 if (seg->is_coalesced) { 2129 h->rsc.segments = seg->packets; 2130 h->rsc.dup_acks = seg->dup_ack; 2131 h->flags = VIRTIO_NET_HDR_F_RSC_INFO; 2132 if (chain->proto == ETH_P_IP) { 2133 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2134 } else { 2135 h->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2136 } 2137 } 2138 2139 ret = virtio_net_do_receive(seg->nc, seg->buf, seg->size); 2140 QTAILQ_REMOVE(&chain->buffers, seg, next); 2141 g_free(seg->buf); 2142 g_free(seg); 2143 2144 return ret; 2145 } 2146 2147 static void virtio_net_rsc_purge(void *opq) 2148 { 2149 VirtioNetRscSeg *seg, *rn; 2150 VirtioNetRscChain *chain = (VirtioNetRscChain *)opq; 2151 2152 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn) { 2153 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2154 chain->stat.purge_failed++; 2155 continue; 2156 } 2157 } 2158 2159 chain->stat.timer++; 2160 if (!QTAILQ_EMPTY(&chain->buffers)) { 2161 timer_mod(chain->drain_timer, 2162 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout); 2163 } 2164 } 2165 2166 static void virtio_net_rsc_cleanup(VirtIONet *n) 2167 { 2168 VirtioNetRscChain *chain, *rn_chain; 2169 VirtioNetRscSeg *seg, *rn_seg; 2170 2171 QTAILQ_FOREACH_SAFE(chain, &n->rsc_chains, next, rn_chain) { 2172 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, rn_seg) { 2173 QTAILQ_REMOVE(&chain->buffers, seg, next); 2174 g_free(seg->buf); 2175 g_free(seg); 2176 } 2177 2178 timer_free(chain->drain_timer); 2179 QTAILQ_REMOVE(&n->rsc_chains, chain, next); 2180 g_free(chain); 2181 } 2182 } 2183 2184 static void virtio_net_rsc_cache_buf(VirtioNetRscChain *chain, 2185 NetClientState *nc, 2186 const uint8_t *buf, size_t size) 2187 { 2188 uint16_t hdr_len; 2189 VirtioNetRscSeg *seg; 2190 2191 hdr_len = chain->n->guest_hdr_len; 2192 seg = g_new(VirtioNetRscSeg, 1); 2193 seg->buf = g_malloc(hdr_len + sizeof(struct eth_header) 2194 + sizeof(struct ip6_header) + VIRTIO_NET_MAX_TCP_PAYLOAD); 2195 memcpy(seg->buf, buf, size); 2196 seg->size = size; 2197 seg->packets = 1; 2198 seg->dup_ack = 0; 2199 seg->is_coalesced = 0; 2200 seg->nc = nc; 2201 2202 QTAILQ_INSERT_TAIL(&chain->buffers, seg, next); 2203 chain->stat.cache++; 2204 2205 switch (chain->proto) { 2206 case ETH_P_IP: 2207 virtio_net_rsc_extract_unit4(chain, seg->buf, &seg->unit); 2208 break; 2209 case ETH_P_IPV6: 2210 virtio_net_rsc_extract_unit6(chain, seg->buf, &seg->unit); 2211 break; 2212 default: 2213 g_assert_not_reached(); 2214 } 2215 } 2216 2217 static int32_t virtio_net_rsc_handle_ack(VirtioNetRscChain *chain, 2218 VirtioNetRscSeg *seg, 2219 const uint8_t *buf, 2220 struct tcp_header *n_tcp, 2221 struct tcp_header *o_tcp) 2222 { 2223 uint32_t nack, oack; 2224 uint16_t nwin, owin; 2225 2226 nack = htonl(n_tcp->th_ack); 2227 nwin = htons(n_tcp->th_win); 2228 oack = htonl(o_tcp->th_ack); 2229 owin = htons(o_tcp->th_win); 2230 2231 if ((nack - oack) >= VIRTIO_NET_MAX_TCP_PAYLOAD) { 2232 chain->stat.ack_out_of_win++; 2233 return RSC_FINAL; 2234 } else if (nack == oack) { 2235 /* duplicated ack or window probe */ 2236 if (nwin == owin) { 2237 /* duplicated ack, add dup ack count due to whql test up to 1 */ 2238 chain->stat.dup_ack++; 2239 return RSC_FINAL; 2240 } else { 2241 /* Coalesce window update */ 2242 o_tcp->th_win = n_tcp->th_win; 2243 chain->stat.win_update++; 2244 return RSC_COALESCE; 2245 } 2246 } else { 2247 /* pure ack, go to 'C', finalize*/ 2248 chain->stat.pure_ack++; 2249 return RSC_FINAL; 2250 } 2251 } 2252 2253 static int32_t virtio_net_rsc_coalesce_data(VirtioNetRscChain *chain, 2254 VirtioNetRscSeg *seg, 2255 const uint8_t *buf, 2256 VirtioNetRscUnit *n_unit) 2257 { 2258 void *data; 2259 uint16_t o_ip_len; 2260 uint32_t nseq, oseq; 2261 VirtioNetRscUnit *o_unit; 2262 2263 o_unit = &seg->unit; 2264 o_ip_len = htons(*o_unit->ip_plen); 2265 nseq = htonl(n_unit->tcp->th_seq); 2266 oseq = htonl(o_unit->tcp->th_seq); 2267 2268 /* out of order or retransmitted. */ 2269 if ((nseq - oseq) > VIRTIO_NET_MAX_TCP_PAYLOAD) { 2270 chain->stat.data_out_of_win++; 2271 return RSC_FINAL; 2272 } 2273 2274 data = ((uint8_t *)n_unit->tcp) + n_unit->tcp_hdrlen; 2275 if (nseq == oseq) { 2276 if ((o_unit->payload == 0) && n_unit->payload) { 2277 /* From no payload to payload, normal case, not a dup ack or etc */ 2278 chain->stat.data_after_pure_ack++; 2279 goto coalesce; 2280 } else { 2281 return virtio_net_rsc_handle_ack(chain, seg, buf, 2282 n_unit->tcp, o_unit->tcp); 2283 } 2284 } else if ((nseq - oseq) != o_unit->payload) { 2285 /* Not a consistent packet, out of order */ 2286 chain->stat.data_out_of_order++; 2287 return RSC_FINAL; 2288 } else { 2289 coalesce: 2290 if ((o_ip_len + n_unit->payload) > chain->max_payload) { 2291 chain->stat.over_size++; 2292 return RSC_FINAL; 2293 } 2294 2295 /* Here comes the right data, the payload length in v4/v6 is different, 2296 so use the field value to update and record the new data len */ 2297 o_unit->payload += n_unit->payload; /* update new data len */ 2298 2299 /* update field in ip header */ 2300 *o_unit->ip_plen = htons(o_ip_len + n_unit->payload); 2301 2302 /* Bring 'PUSH' big, the whql test guide says 'PUSH' can be coalesced 2303 for windows guest, while this may change the behavior for linux 2304 guest (only if it uses RSC feature). */ 2305 o_unit->tcp->th_offset_flags = n_unit->tcp->th_offset_flags; 2306 2307 o_unit->tcp->th_ack = n_unit->tcp->th_ack; 2308 o_unit->tcp->th_win = n_unit->tcp->th_win; 2309 2310 memmove(seg->buf + seg->size, data, n_unit->payload); 2311 seg->size += n_unit->payload; 2312 seg->packets++; 2313 chain->stat.coalesced++; 2314 return RSC_COALESCE; 2315 } 2316 } 2317 2318 static int32_t virtio_net_rsc_coalesce4(VirtioNetRscChain *chain, 2319 VirtioNetRscSeg *seg, 2320 const uint8_t *buf, size_t size, 2321 VirtioNetRscUnit *unit) 2322 { 2323 struct ip_header *ip1, *ip2; 2324 2325 ip1 = (struct ip_header *)(unit->ip); 2326 ip2 = (struct ip_header *)(seg->unit.ip); 2327 if ((ip1->ip_src ^ ip2->ip_src) || (ip1->ip_dst ^ ip2->ip_dst) 2328 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport) 2329 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) { 2330 chain->stat.no_match++; 2331 return RSC_NO_MATCH; 2332 } 2333 2334 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit); 2335 } 2336 2337 static int32_t virtio_net_rsc_coalesce6(VirtioNetRscChain *chain, 2338 VirtioNetRscSeg *seg, 2339 const uint8_t *buf, size_t size, 2340 VirtioNetRscUnit *unit) 2341 { 2342 struct ip6_header *ip1, *ip2; 2343 2344 ip1 = (struct ip6_header *)(unit->ip); 2345 ip2 = (struct ip6_header *)(seg->unit.ip); 2346 if (memcmp(&ip1->ip6_src, &ip2->ip6_src, sizeof(struct in6_address)) 2347 || memcmp(&ip1->ip6_dst, &ip2->ip6_dst, sizeof(struct in6_address)) 2348 || (unit->tcp->th_sport ^ seg->unit.tcp->th_sport) 2349 || (unit->tcp->th_dport ^ seg->unit.tcp->th_dport)) { 2350 chain->stat.no_match++; 2351 return RSC_NO_MATCH; 2352 } 2353 2354 return virtio_net_rsc_coalesce_data(chain, seg, buf, unit); 2355 } 2356 2357 /* Packets with 'SYN' should bypass, other flag should be sent after drain 2358 * to prevent out of order */ 2359 static int virtio_net_rsc_tcp_ctrl_check(VirtioNetRscChain *chain, 2360 struct tcp_header *tcp) 2361 { 2362 uint16_t tcp_hdr; 2363 uint16_t tcp_flag; 2364 2365 tcp_flag = htons(tcp->th_offset_flags); 2366 tcp_hdr = (tcp_flag & VIRTIO_NET_TCP_HDR_LENGTH) >> 10; 2367 tcp_flag &= VIRTIO_NET_TCP_FLAG; 2368 if (tcp_flag & TH_SYN) { 2369 chain->stat.tcp_syn++; 2370 return RSC_BYPASS; 2371 } 2372 2373 if (tcp_flag & (TH_FIN | TH_URG | TH_RST | TH_ECE | TH_CWR)) { 2374 chain->stat.tcp_ctrl_drain++; 2375 return RSC_FINAL; 2376 } 2377 2378 if (tcp_hdr > sizeof(struct tcp_header)) { 2379 chain->stat.tcp_all_opt++; 2380 return RSC_FINAL; 2381 } 2382 2383 return RSC_CANDIDATE; 2384 } 2385 2386 static size_t virtio_net_rsc_do_coalesce(VirtioNetRscChain *chain, 2387 NetClientState *nc, 2388 const uint8_t *buf, size_t size, 2389 VirtioNetRscUnit *unit) 2390 { 2391 int ret; 2392 VirtioNetRscSeg *seg, *nseg; 2393 2394 if (QTAILQ_EMPTY(&chain->buffers)) { 2395 chain->stat.empty_cache++; 2396 virtio_net_rsc_cache_buf(chain, nc, buf, size); 2397 timer_mod(chain->drain_timer, 2398 qemu_clock_get_ns(QEMU_CLOCK_HOST) + chain->n->rsc_timeout); 2399 return size; 2400 } 2401 2402 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) { 2403 if (chain->proto == ETH_P_IP) { 2404 ret = virtio_net_rsc_coalesce4(chain, seg, buf, size, unit); 2405 } else { 2406 ret = virtio_net_rsc_coalesce6(chain, seg, buf, size, unit); 2407 } 2408 2409 if (ret == RSC_FINAL) { 2410 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2411 /* Send failed */ 2412 chain->stat.final_failed++; 2413 return 0; 2414 } 2415 2416 /* Send current packet */ 2417 return virtio_net_do_receive(nc, buf, size); 2418 } else if (ret == RSC_NO_MATCH) { 2419 continue; 2420 } else { 2421 /* Coalesced, mark coalesced flag to tell calc cksum for ipv4 */ 2422 seg->is_coalesced = 1; 2423 return size; 2424 } 2425 } 2426 2427 chain->stat.no_match_cache++; 2428 virtio_net_rsc_cache_buf(chain, nc, buf, size); 2429 return size; 2430 } 2431 2432 /* Drain a connection data, this is to avoid out of order segments */ 2433 static size_t virtio_net_rsc_drain_flow(VirtioNetRscChain *chain, 2434 NetClientState *nc, 2435 const uint8_t *buf, size_t size, 2436 uint16_t ip_start, uint16_t ip_size, 2437 uint16_t tcp_port) 2438 { 2439 VirtioNetRscSeg *seg, *nseg; 2440 uint32_t ppair1, ppair2; 2441 2442 ppair1 = *(uint32_t *)(buf + tcp_port); 2443 QTAILQ_FOREACH_SAFE(seg, &chain->buffers, next, nseg) { 2444 ppair2 = *(uint32_t *)(seg->buf + tcp_port); 2445 if (memcmp(buf + ip_start, seg->buf + ip_start, ip_size) 2446 || (ppair1 != ppair2)) { 2447 continue; 2448 } 2449 if (virtio_net_rsc_drain_seg(chain, seg) == 0) { 2450 chain->stat.drain_failed++; 2451 } 2452 2453 break; 2454 } 2455 2456 return virtio_net_do_receive(nc, buf, size); 2457 } 2458 2459 static int32_t virtio_net_rsc_sanity_check4(VirtioNetRscChain *chain, 2460 struct ip_header *ip, 2461 const uint8_t *buf, size_t size) 2462 { 2463 uint16_t ip_len; 2464 2465 /* Not an ipv4 packet */ 2466 if (((ip->ip_ver_len & 0xF0) >> 4) != IP_HEADER_VERSION_4) { 2467 chain->stat.ip_option++; 2468 return RSC_BYPASS; 2469 } 2470 2471 /* Don't handle packets with ip option */ 2472 if ((ip->ip_ver_len & 0xF) != VIRTIO_NET_IP4_HEADER_LENGTH) { 2473 chain->stat.ip_option++; 2474 return RSC_BYPASS; 2475 } 2476 2477 if (ip->ip_p != IPPROTO_TCP) { 2478 chain->stat.bypass_not_tcp++; 2479 return RSC_BYPASS; 2480 } 2481 2482 /* Don't handle packets with ip fragment */ 2483 if (!(htons(ip->ip_off) & IP_DF)) { 2484 chain->stat.ip_frag++; 2485 return RSC_BYPASS; 2486 } 2487 2488 /* Don't handle packets with ecn flag */ 2489 if (IPTOS_ECN(ip->ip_tos)) { 2490 chain->stat.ip_ecn++; 2491 return RSC_BYPASS; 2492 } 2493 2494 ip_len = htons(ip->ip_len); 2495 if (ip_len < (sizeof(struct ip_header) + sizeof(struct tcp_header)) 2496 || ip_len > (size - chain->n->guest_hdr_len - 2497 sizeof(struct eth_header))) { 2498 chain->stat.ip_hacked++; 2499 return RSC_BYPASS; 2500 } 2501 2502 return RSC_CANDIDATE; 2503 } 2504 2505 static size_t virtio_net_rsc_receive4(VirtioNetRscChain *chain, 2506 NetClientState *nc, 2507 const uint8_t *buf, size_t size) 2508 { 2509 int32_t ret; 2510 uint16_t hdr_len; 2511 VirtioNetRscUnit unit; 2512 2513 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len; 2514 2515 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header) 2516 + sizeof(struct tcp_header))) { 2517 chain->stat.bypass_not_tcp++; 2518 return virtio_net_do_receive(nc, buf, size); 2519 } 2520 2521 virtio_net_rsc_extract_unit4(chain, buf, &unit); 2522 if (virtio_net_rsc_sanity_check4(chain, unit.ip, buf, size) 2523 != RSC_CANDIDATE) { 2524 return virtio_net_do_receive(nc, buf, size); 2525 } 2526 2527 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp); 2528 if (ret == RSC_BYPASS) { 2529 return virtio_net_do_receive(nc, buf, size); 2530 } else if (ret == RSC_FINAL) { 2531 return virtio_net_rsc_drain_flow(chain, nc, buf, size, 2532 ((hdr_len + sizeof(struct eth_header)) + 12), 2533 VIRTIO_NET_IP4_ADDR_SIZE, 2534 hdr_len + sizeof(struct eth_header) + sizeof(struct ip_header)); 2535 } 2536 2537 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit); 2538 } 2539 2540 static int32_t virtio_net_rsc_sanity_check6(VirtioNetRscChain *chain, 2541 struct ip6_header *ip6, 2542 const uint8_t *buf, size_t size) 2543 { 2544 uint16_t ip_len; 2545 2546 if (((ip6->ip6_ctlun.ip6_un1.ip6_un1_flow & 0xF0) >> 4) 2547 != IP_HEADER_VERSION_6) { 2548 return RSC_BYPASS; 2549 } 2550 2551 /* Both option and protocol is checked in this */ 2552 if (ip6->ip6_ctlun.ip6_un1.ip6_un1_nxt != IPPROTO_TCP) { 2553 chain->stat.bypass_not_tcp++; 2554 return RSC_BYPASS; 2555 } 2556 2557 ip_len = htons(ip6->ip6_ctlun.ip6_un1.ip6_un1_plen); 2558 if (ip_len < sizeof(struct tcp_header) || 2559 ip_len > (size - chain->n->guest_hdr_len - sizeof(struct eth_header) 2560 - sizeof(struct ip6_header))) { 2561 chain->stat.ip_hacked++; 2562 return RSC_BYPASS; 2563 } 2564 2565 /* Don't handle packets with ecn flag */ 2566 if (IP6_ECN(ip6->ip6_ctlun.ip6_un3.ip6_un3_ecn)) { 2567 chain->stat.ip_ecn++; 2568 return RSC_BYPASS; 2569 } 2570 2571 return RSC_CANDIDATE; 2572 } 2573 2574 static size_t virtio_net_rsc_receive6(void *opq, NetClientState *nc, 2575 const uint8_t *buf, size_t size) 2576 { 2577 int32_t ret; 2578 uint16_t hdr_len; 2579 VirtioNetRscChain *chain; 2580 VirtioNetRscUnit unit; 2581 2582 chain = opq; 2583 hdr_len = ((VirtIONet *)(chain->n))->guest_hdr_len; 2584 2585 if (size < (hdr_len + sizeof(struct eth_header) + sizeof(struct ip6_header) 2586 + sizeof(tcp_header))) { 2587 return virtio_net_do_receive(nc, buf, size); 2588 } 2589 2590 virtio_net_rsc_extract_unit6(chain, buf, &unit); 2591 if (RSC_CANDIDATE != virtio_net_rsc_sanity_check6(chain, 2592 unit.ip, buf, size)) { 2593 return virtio_net_do_receive(nc, buf, size); 2594 } 2595 2596 ret = virtio_net_rsc_tcp_ctrl_check(chain, unit.tcp); 2597 if (ret == RSC_BYPASS) { 2598 return virtio_net_do_receive(nc, buf, size); 2599 } else if (ret == RSC_FINAL) { 2600 return virtio_net_rsc_drain_flow(chain, nc, buf, size, 2601 ((hdr_len + sizeof(struct eth_header)) + 8), 2602 VIRTIO_NET_IP6_ADDR_SIZE, 2603 hdr_len + sizeof(struct eth_header) 2604 + sizeof(struct ip6_header)); 2605 } 2606 2607 return virtio_net_rsc_do_coalesce(chain, nc, buf, size, &unit); 2608 } 2609 2610 static VirtioNetRscChain *virtio_net_rsc_lookup_chain(VirtIONet *n, 2611 NetClientState *nc, 2612 uint16_t proto) 2613 { 2614 VirtioNetRscChain *chain; 2615 2616 if ((proto != (uint16_t)ETH_P_IP) && (proto != (uint16_t)ETH_P_IPV6)) { 2617 return NULL; 2618 } 2619 2620 QTAILQ_FOREACH(chain, &n->rsc_chains, next) { 2621 if (chain->proto == proto) { 2622 return chain; 2623 } 2624 } 2625 2626 chain = g_malloc(sizeof(*chain)); 2627 chain->n = n; 2628 chain->proto = proto; 2629 if (proto == (uint16_t)ETH_P_IP) { 2630 chain->max_payload = VIRTIO_NET_MAX_IP4_PAYLOAD; 2631 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 2632 } else { 2633 chain->max_payload = VIRTIO_NET_MAX_IP6_PAYLOAD; 2634 chain->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 2635 } 2636 chain->drain_timer = timer_new_ns(QEMU_CLOCK_HOST, 2637 virtio_net_rsc_purge, chain); 2638 memset(&chain->stat, 0, sizeof(chain->stat)); 2639 2640 QTAILQ_INIT(&chain->buffers); 2641 QTAILQ_INSERT_TAIL(&n->rsc_chains, chain, next); 2642 2643 return chain; 2644 } 2645 2646 static ssize_t virtio_net_rsc_receive(NetClientState *nc, 2647 const uint8_t *buf, 2648 size_t size) 2649 { 2650 uint16_t proto; 2651 VirtioNetRscChain *chain; 2652 struct eth_header *eth; 2653 VirtIONet *n; 2654 2655 n = qemu_get_nic_opaque(nc); 2656 if (size < (n->host_hdr_len + sizeof(struct eth_header))) { 2657 return virtio_net_do_receive(nc, buf, size); 2658 } 2659 2660 eth = (struct eth_header *)(buf + n->guest_hdr_len); 2661 proto = htons(eth->h_proto); 2662 2663 chain = virtio_net_rsc_lookup_chain(n, nc, proto); 2664 if (chain) { 2665 chain->stat.received++; 2666 if (proto == (uint16_t)ETH_P_IP && n->rsc4_enabled) { 2667 return virtio_net_rsc_receive4(chain, nc, buf, size); 2668 } else if (proto == (uint16_t)ETH_P_IPV6 && n->rsc6_enabled) { 2669 return virtio_net_rsc_receive6(chain, nc, buf, size); 2670 } 2671 } 2672 return virtio_net_do_receive(nc, buf, size); 2673 } 2674 2675 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, 2676 size_t size) 2677 { 2678 VirtIONet *n = qemu_get_nic_opaque(nc); 2679 if ((n->rsc4_enabled || n->rsc6_enabled)) { 2680 return virtio_net_rsc_receive(nc, buf, size); 2681 } else { 2682 return virtio_net_do_receive(nc, buf, size); 2683 } 2684 } 2685 2686 static int32_t virtio_net_flush_tx(VirtIONetQueue *q); 2687 2688 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len) 2689 { 2690 VirtIONet *n = qemu_get_nic_opaque(nc); 2691 VirtIONetQueue *q = virtio_net_get_subqueue(nc); 2692 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2693 int ret; 2694 2695 virtqueue_push(q->tx_vq, q->async_tx.elem, 0); 2696 virtio_notify(vdev, q->tx_vq); 2697 2698 g_free(q->async_tx.elem); 2699 q->async_tx.elem = NULL; 2700 2701 virtio_queue_set_notification(q->tx_vq, 1); 2702 ret = virtio_net_flush_tx(q); 2703 if (ret >= n->tx_burst) { 2704 /* 2705 * the flush has been stopped by tx_burst 2706 * we will not receive notification for the 2707 * remainining part, so re-schedule 2708 */ 2709 virtio_queue_set_notification(q->tx_vq, 0); 2710 if (q->tx_bh) { 2711 qemu_bh_schedule(q->tx_bh); 2712 } else { 2713 timer_mod(q->tx_timer, 2714 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2715 } 2716 q->tx_waiting = 1; 2717 } 2718 } 2719 2720 /* TX */ 2721 static int32_t virtio_net_flush_tx(VirtIONetQueue *q) 2722 { 2723 VirtIONet *n = q->n; 2724 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2725 VirtQueueElement *elem; 2726 int32_t num_packets = 0; 2727 int queue_index = vq2q(virtio_get_queue_index(q->tx_vq)); 2728 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 2729 return num_packets; 2730 } 2731 2732 if (q->async_tx.elem) { 2733 virtio_queue_set_notification(q->tx_vq, 0); 2734 return num_packets; 2735 } 2736 2737 for (;;) { 2738 ssize_t ret; 2739 unsigned int out_num; 2740 struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE + 1], *out_sg; 2741 struct virtio_net_hdr_v1_hash vhdr; 2742 2743 elem = virtqueue_pop(q->tx_vq, sizeof(VirtQueueElement)); 2744 if (!elem) { 2745 break; 2746 } 2747 2748 out_num = elem->out_num; 2749 out_sg = elem->out_sg; 2750 if (out_num < 1) { 2751 virtio_error(vdev, "virtio-net header not in first element"); 2752 virtqueue_detach_element(q->tx_vq, elem, 0); 2753 g_free(elem); 2754 return -EINVAL; 2755 } 2756 2757 if (n->has_vnet_hdr) { 2758 if (iov_to_buf(out_sg, out_num, 0, &vhdr, n->guest_hdr_len) < 2759 n->guest_hdr_len) { 2760 virtio_error(vdev, "virtio-net header incorrect"); 2761 virtqueue_detach_element(q->tx_vq, elem, 0); 2762 g_free(elem); 2763 return -EINVAL; 2764 } 2765 if (n->needs_vnet_hdr_swap) { 2766 virtio_net_hdr_swap(vdev, (void *) &vhdr); 2767 sg2[0].iov_base = &vhdr; 2768 sg2[0].iov_len = n->guest_hdr_len; 2769 out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1, 2770 out_sg, out_num, 2771 n->guest_hdr_len, -1); 2772 if (out_num == VIRTQUEUE_MAX_SIZE) { 2773 goto drop; 2774 } 2775 out_num += 1; 2776 out_sg = sg2; 2777 } 2778 } 2779 /* 2780 * If host wants to see the guest header as is, we can 2781 * pass it on unchanged. Otherwise, copy just the parts 2782 * that host is interested in. 2783 */ 2784 assert(n->host_hdr_len <= n->guest_hdr_len); 2785 if (n->host_hdr_len != n->guest_hdr_len) { 2786 unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg), 2787 out_sg, out_num, 2788 0, n->host_hdr_len); 2789 sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num, 2790 out_sg, out_num, 2791 n->guest_hdr_len, -1); 2792 out_num = sg_num; 2793 out_sg = sg; 2794 } 2795 2796 ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index), 2797 out_sg, out_num, virtio_net_tx_complete); 2798 if (ret == 0) { 2799 virtio_queue_set_notification(q->tx_vq, 0); 2800 q->async_tx.elem = elem; 2801 return -EBUSY; 2802 } 2803 2804 drop: 2805 virtqueue_push(q->tx_vq, elem, 0); 2806 virtio_notify(vdev, q->tx_vq); 2807 g_free(elem); 2808 2809 if (++num_packets >= n->tx_burst) { 2810 break; 2811 } 2812 } 2813 return num_packets; 2814 } 2815 2816 static void virtio_net_tx_timer(void *opaque); 2817 2818 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq) 2819 { 2820 VirtIONet *n = VIRTIO_NET(vdev); 2821 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 2822 2823 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 2824 virtio_net_drop_tx_queue_data(vdev, vq); 2825 return; 2826 } 2827 2828 /* This happens when device was stopped but VCPU wasn't. */ 2829 if (!vdev->vm_running) { 2830 q->tx_waiting = 1; 2831 return; 2832 } 2833 2834 if (q->tx_waiting) { 2835 /* We already have queued packets, immediately flush */ 2836 timer_del(q->tx_timer); 2837 virtio_net_tx_timer(q); 2838 } else { 2839 /* re-arm timer to flush it (and more) on next tick */ 2840 timer_mod(q->tx_timer, 2841 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2842 q->tx_waiting = 1; 2843 virtio_queue_set_notification(vq, 0); 2844 } 2845 } 2846 2847 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq) 2848 { 2849 VirtIONet *n = VIRTIO_NET(vdev); 2850 VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))]; 2851 2852 if (unlikely(n->vhost_started)) { 2853 return; 2854 } 2855 2856 if (unlikely((n->status & VIRTIO_NET_S_LINK_UP) == 0)) { 2857 virtio_net_drop_tx_queue_data(vdev, vq); 2858 return; 2859 } 2860 2861 if (unlikely(q->tx_waiting)) { 2862 return; 2863 } 2864 q->tx_waiting = 1; 2865 /* This happens when device was stopped but VCPU wasn't. */ 2866 if (!vdev->vm_running) { 2867 return; 2868 } 2869 virtio_queue_set_notification(vq, 0); 2870 qemu_bh_schedule(q->tx_bh); 2871 } 2872 2873 static void virtio_net_tx_timer(void *opaque) 2874 { 2875 VirtIONetQueue *q = opaque; 2876 VirtIONet *n = q->n; 2877 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2878 int ret; 2879 2880 /* This happens when device was stopped but BH wasn't. */ 2881 if (!vdev->vm_running) { 2882 /* Make sure tx waiting is set, so we'll run when restarted. */ 2883 assert(q->tx_waiting); 2884 return; 2885 } 2886 2887 q->tx_waiting = 0; 2888 2889 /* Just in case the driver is not ready on more */ 2890 if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) { 2891 return; 2892 } 2893 2894 ret = virtio_net_flush_tx(q); 2895 if (ret == -EBUSY || ret == -EINVAL) { 2896 return; 2897 } 2898 /* 2899 * If we flush a full burst of packets, assume there are 2900 * more coming and immediately rearm 2901 */ 2902 if (ret >= n->tx_burst) { 2903 q->tx_waiting = 1; 2904 timer_mod(q->tx_timer, 2905 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2906 return; 2907 } 2908 /* 2909 * If less than a full burst, re-enable notification and flush 2910 * anything that may have come in while we weren't looking. If 2911 * we find something, assume the guest is still active and rearm 2912 */ 2913 virtio_queue_set_notification(q->tx_vq, 1); 2914 ret = virtio_net_flush_tx(q); 2915 if (ret > 0) { 2916 virtio_queue_set_notification(q->tx_vq, 0); 2917 q->tx_waiting = 1; 2918 timer_mod(q->tx_timer, 2919 qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + n->tx_timeout); 2920 } 2921 } 2922 2923 static void virtio_net_tx_bh(void *opaque) 2924 { 2925 VirtIONetQueue *q = opaque; 2926 VirtIONet *n = q->n; 2927 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2928 int32_t ret; 2929 2930 /* This happens when device was stopped but BH wasn't. */ 2931 if (!vdev->vm_running) { 2932 /* Make sure tx waiting is set, so we'll run when restarted. */ 2933 assert(q->tx_waiting); 2934 return; 2935 } 2936 2937 q->tx_waiting = 0; 2938 2939 /* Just in case the driver is not ready on more */ 2940 if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) { 2941 return; 2942 } 2943 2944 ret = virtio_net_flush_tx(q); 2945 if (ret == -EBUSY || ret == -EINVAL) { 2946 return; /* Notification re-enable handled by tx_complete or device 2947 * broken */ 2948 } 2949 2950 /* If we flush a full burst of packets, assume there are 2951 * more coming and immediately reschedule */ 2952 if (ret >= n->tx_burst) { 2953 qemu_bh_schedule(q->tx_bh); 2954 q->tx_waiting = 1; 2955 return; 2956 } 2957 2958 /* If less than a full burst, re-enable notification and flush 2959 * anything that may have come in while we weren't looking. If 2960 * we find something, assume the guest is still active and reschedule */ 2961 virtio_queue_set_notification(q->tx_vq, 1); 2962 ret = virtio_net_flush_tx(q); 2963 if (ret == -EINVAL) { 2964 return; 2965 } else if (ret > 0) { 2966 virtio_queue_set_notification(q->tx_vq, 0); 2967 qemu_bh_schedule(q->tx_bh); 2968 q->tx_waiting = 1; 2969 } 2970 } 2971 2972 static void virtio_net_add_queue(VirtIONet *n, int index) 2973 { 2974 VirtIODevice *vdev = VIRTIO_DEVICE(n); 2975 2976 n->vqs[index].rx_vq = virtio_add_queue(vdev, n->net_conf.rx_queue_size, 2977 virtio_net_handle_rx); 2978 2979 if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) { 2980 n->vqs[index].tx_vq = 2981 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 2982 virtio_net_handle_tx_timer); 2983 n->vqs[index].tx_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, 2984 virtio_net_tx_timer, 2985 &n->vqs[index]); 2986 } else { 2987 n->vqs[index].tx_vq = 2988 virtio_add_queue(vdev, n->net_conf.tx_queue_size, 2989 virtio_net_handle_tx_bh); 2990 n->vqs[index].tx_bh = qemu_bh_new_guarded(virtio_net_tx_bh, &n->vqs[index], 2991 &DEVICE(vdev)->mem_reentrancy_guard); 2992 } 2993 2994 n->vqs[index].tx_waiting = 0; 2995 n->vqs[index].n = n; 2996 } 2997 2998 static void virtio_net_del_queue(VirtIONet *n, int index) 2999 { 3000 VirtIODevice *vdev = VIRTIO_DEVICE(n); 3001 VirtIONetQueue *q = &n->vqs[index]; 3002 NetClientState *nc = qemu_get_subqueue(n->nic, index); 3003 3004 qemu_purge_queued_packets(nc); 3005 3006 virtio_del_queue(vdev, index * 2); 3007 if (q->tx_timer) { 3008 timer_free(q->tx_timer); 3009 q->tx_timer = NULL; 3010 } else { 3011 qemu_bh_delete(q->tx_bh); 3012 q->tx_bh = NULL; 3013 } 3014 q->tx_waiting = 0; 3015 virtio_del_queue(vdev, index * 2 + 1); 3016 } 3017 3018 static void virtio_net_change_num_queue_pairs(VirtIONet *n, int new_max_queue_pairs) 3019 { 3020 VirtIODevice *vdev = VIRTIO_DEVICE(n); 3021 int old_num_queues = virtio_get_num_queues(vdev); 3022 int new_num_queues = new_max_queue_pairs * 2 + 1; 3023 int i; 3024 3025 assert(old_num_queues >= 3); 3026 assert(old_num_queues % 2 == 1); 3027 3028 if (old_num_queues == new_num_queues) { 3029 return; 3030 } 3031 3032 /* 3033 * We always need to remove and add ctrl vq if 3034 * old_num_queues != new_num_queues. Remove ctrl_vq first, 3035 * and then we only enter one of the following two loops. 3036 */ 3037 virtio_del_queue(vdev, old_num_queues - 1); 3038 3039 for (i = new_num_queues - 1; i < old_num_queues - 1; i += 2) { 3040 /* new_num_queues < old_num_queues */ 3041 virtio_net_del_queue(n, i / 2); 3042 } 3043 3044 for (i = old_num_queues - 1; i < new_num_queues - 1; i += 2) { 3045 /* new_num_queues > old_num_queues */ 3046 virtio_net_add_queue(n, i / 2); 3047 } 3048 3049 /* add ctrl_vq last */ 3050 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 3051 } 3052 3053 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue) 3054 { 3055 int max = multiqueue ? n->max_queue_pairs : 1; 3056 3057 n->multiqueue = multiqueue; 3058 virtio_net_change_num_queue_pairs(n, max); 3059 3060 virtio_net_set_queue_pairs(n); 3061 } 3062 3063 static int virtio_net_post_load_device(void *opaque, int version_id) 3064 { 3065 VirtIONet *n = opaque; 3066 VirtIODevice *vdev = VIRTIO_DEVICE(n); 3067 int i, link_down; 3068 3069 trace_virtio_net_post_load_device(); 3070 virtio_net_set_mrg_rx_bufs(n, n->mergeable_rx_bufs, 3071 virtio_vdev_has_feature(vdev, 3072 VIRTIO_F_VERSION_1), 3073 virtio_vdev_has_feature(vdev, 3074 VIRTIO_NET_F_HASH_REPORT)); 3075 3076 /* MAC_TABLE_ENTRIES may be different from the saved image */ 3077 if (n->mac_table.in_use > MAC_TABLE_ENTRIES) { 3078 n->mac_table.in_use = 0; 3079 } 3080 3081 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) { 3082 n->curr_guest_offloads = virtio_net_supported_guest_offloads(n); 3083 } 3084 3085 /* 3086 * curr_guest_offloads will be later overwritten by the 3087 * virtio_set_features_nocheck call done from the virtio_load. 3088 * Here we make sure it is preserved and restored accordingly 3089 * in the virtio_net_post_load_virtio callback. 3090 */ 3091 n->saved_guest_offloads = n->curr_guest_offloads; 3092 3093 virtio_net_set_queue_pairs(n); 3094 3095 /* Find the first multicast entry in the saved MAC filter */ 3096 for (i = 0; i < n->mac_table.in_use; i++) { 3097 if (n->mac_table.macs[i * ETH_ALEN] & 1) { 3098 break; 3099 } 3100 } 3101 n->mac_table.first_multi = i; 3102 3103 /* nc.link_down can't be migrated, so infer link_down according 3104 * to link status bit in n->status */ 3105 link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0; 3106 for (i = 0; i < n->max_queue_pairs; i++) { 3107 qemu_get_subqueue(n->nic, i)->link_down = link_down; 3108 } 3109 3110 if (virtio_vdev_has_feature(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE) && 3111 virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3112 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(), 3113 QEMU_CLOCK_VIRTUAL, 3114 virtio_net_announce_timer, n); 3115 if (n->announce_timer.round) { 3116 timer_mod(n->announce_timer.tm, 3117 qemu_clock_get_ms(n->announce_timer.type)); 3118 } else { 3119 qemu_announce_timer_del(&n->announce_timer, false); 3120 } 3121 } 3122 3123 if (n->rss_data.enabled) { 3124 n->rss_data.enabled_software_rss = n->rss_data.populate_hash; 3125 if (!n->rss_data.populate_hash) { 3126 if (!virtio_net_attach_epbf_rss(n)) { 3127 if (get_vhost_net(qemu_get_queue(n->nic)->peer)) { 3128 warn_report("Can't post-load eBPF RSS for vhost"); 3129 } else { 3130 warn_report("Can't post-load eBPF RSS - " 3131 "fallback to software RSS"); 3132 n->rss_data.enabled_software_rss = true; 3133 } 3134 } 3135 } 3136 3137 trace_virtio_net_rss_enable(n->rss_data.hash_types, 3138 n->rss_data.indirections_len, 3139 sizeof(n->rss_data.key)); 3140 } else { 3141 trace_virtio_net_rss_disable(); 3142 } 3143 return 0; 3144 } 3145 3146 static int virtio_net_post_load_virtio(VirtIODevice *vdev) 3147 { 3148 VirtIONet *n = VIRTIO_NET(vdev); 3149 /* 3150 * The actual needed state is now in saved_guest_offloads, 3151 * see virtio_net_post_load_device for detail. 3152 * Restore it back and apply the desired offloads. 3153 */ 3154 n->curr_guest_offloads = n->saved_guest_offloads; 3155 if (peer_has_vnet_hdr(n)) { 3156 virtio_net_apply_guest_offloads(n); 3157 } 3158 3159 return 0; 3160 } 3161 3162 /* tx_waiting field of a VirtIONetQueue */ 3163 static const VMStateDescription vmstate_virtio_net_queue_tx_waiting = { 3164 .name = "virtio-net-queue-tx_waiting", 3165 .fields = (const VMStateField[]) { 3166 VMSTATE_UINT32(tx_waiting, VirtIONetQueue), 3167 VMSTATE_END_OF_LIST() 3168 }, 3169 }; 3170 3171 static bool max_queue_pairs_gt_1(void *opaque, int version_id) 3172 { 3173 return VIRTIO_NET(opaque)->max_queue_pairs > 1; 3174 } 3175 3176 static bool has_ctrl_guest_offloads(void *opaque, int version_id) 3177 { 3178 return virtio_vdev_has_feature(VIRTIO_DEVICE(opaque), 3179 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS); 3180 } 3181 3182 static bool mac_table_fits(void *opaque, int version_id) 3183 { 3184 return VIRTIO_NET(opaque)->mac_table.in_use <= MAC_TABLE_ENTRIES; 3185 } 3186 3187 static bool mac_table_doesnt_fit(void *opaque, int version_id) 3188 { 3189 return !mac_table_fits(opaque, version_id); 3190 } 3191 3192 /* This temporary type is shared by all the WITH_TMP methods 3193 * although only some fields are used by each. 3194 */ 3195 struct VirtIONetMigTmp { 3196 VirtIONet *parent; 3197 VirtIONetQueue *vqs_1; 3198 uint16_t curr_queue_pairs_1; 3199 uint8_t has_ufo; 3200 uint32_t has_vnet_hdr; 3201 }; 3202 3203 /* The 2nd and subsequent tx_waiting flags are loaded later than 3204 * the 1st entry in the queue_pairs and only if there's more than one 3205 * entry. We use the tmp mechanism to calculate a temporary 3206 * pointer and count and also validate the count. 3207 */ 3208 3209 static int virtio_net_tx_waiting_pre_save(void *opaque) 3210 { 3211 struct VirtIONetMigTmp *tmp = opaque; 3212 3213 tmp->vqs_1 = tmp->parent->vqs + 1; 3214 tmp->curr_queue_pairs_1 = tmp->parent->curr_queue_pairs - 1; 3215 if (tmp->parent->curr_queue_pairs == 0) { 3216 tmp->curr_queue_pairs_1 = 0; 3217 } 3218 3219 return 0; 3220 } 3221 3222 static int virtio_net_tx_waiting_pre_load(void *opaque) 3223 { 3224 struct VirtIONetMigTmp *tmp = opaque; 3225 3226 /* Reuse the pointer setup from save */ 3227 virtio_net_tx_waiting_pre_save(opaque); 3228 3229 if (tmp->parent->curr_queue_pairs > tmp->parent->max_queue_pairs) { 3230 error_report("virtio-net: curr_queue_pairs %x > max_queue_pairs %x", 3231 tmp->parent->curr_queue_pairs, tmp->parent->max_queue_pairs); 3232 3233 return -EINVAL; 3234 } 3235 3236 return 0; /* all good */ 3237 } 3238 3239 static const VMStateDescription vmstate_virtio_net_tx_waiting = { 3240 .name = "virtio-net-tx_waiting", 3241 .pre_load = virtio_net_tx_waiting_pre_load, 3242 .pre_save = virtio_net_tx_waiting_pre_save, 3243 .fields = (const VMStateField[]) { 3244 VMSTATE_STRUCT_VARRAY_POINTER_UINT16(vqs_1, struct VirtIONetMigTmp, 3245 curr_queue_pairs_1, 3246 vmstate_virtio_net_queue_tx_waiting, 3247 struct VirtIONetQueue), 3248 VMSTATE_END_OF_LIST() 3249 }, 3250 }; 3251 3252 /* the 'has_ufo' flag is just tested; if the incoming stream has the 3253 * flag set we need to check that we have it 3254 */ 3255 static int virtio_net_ufo_post_load(void *opaque, int version_id) 3256 { 3257 struct VirtIONetMigTmp *tmp = opaque; 3258 3259 if (tmp->has_ufo && !peer_has_ufo(tmp->parent)) { 3260 error_report("virtio-net: saved image requires TUN_F_UFO support"); 3261 return -EINVAL; 3262 } 3263 3264 return 0; 3265 } 3266 3267 static int virtio_net_ufo_pre_save(void *opaque) 3268 { 3269 struct VirtIONetMigTmp *tmp = opaque; 3270 3271 tmp->has_ufo = tmp->parent->has_ufo; 3272 3273 return 0; 3274 } 3275 3276 static const VMStateDescription vmstate_virtio_net_has_ufo = { 3277 .name = "virtio-net-ufo", 3278 .post_load = virtio_net_ufo_post_load, 3279 .pre_save = virtio_net_ufo_pre_save, 3280 .fields = (const VMStateField[]) { 3281 VMSTATE_UINT8(has_ufo, struct VirtIONetMigTmp), 3282 VMSTATE_END_OF_LIST() 3283 }, 3284 }; 3285 3286 /* the 'has_vnet_hdr' flag is just tested; if the incoming stream has the 3287 * flag set we need to check that we have it 3288 */ 3289 static int virtio_net_vnet_post_load(void *opaque, int version_id) 3290 { 3291 struct VirtIONetMigTmp *tmp = opaque; 3292 3293 if (tmp->has_vnet_hdr && !peer_has_vnet_hdr(tmp->parent)) { 3294 error_report("virtio-net: saved image requires vnet_hdr=on"); 3295 return -EINVAL; 3296 } 3297 3298 return 0; 3299 } 3300 3301 static int virtio_net_vnet_pre_save(void *opaque) 3302 { 3303 struct VirtIONetMigTmp *tmp = opaque; 3304 3305 tmp->has_vnet_hdr = tmp->parent->has_vnet_hdr; 3306 3307 return 0; 3308 } 3309 3310 static const VMStateDescription vmstate_virtio_net_has_vnet = { 3311 .name = "virtio-net-vnet", 3312 .post_load = virtio_net_vnet_post_load, 3313 .pre_save = virtio_net_vnet_pre_save, 3314 .fields = (const VMStateField[]) { 3315 VMSTATE_UINT32(has_vnet_hdr, struct VirtIONetMigTmp), 3316 VMSTATE_END_OF_LIST() 3317 }, 3318 }; 3319 3320 static bool virtio_net_rss_needed(void *opaque) 3321 { 3322 return VIRTIO_NET(opaque)->rss_data.enabled; 3323 } 3324 3325 static const VMStateDescription vmstate_virtio_net_rss = { 3326 .name = "virtio-net-device/rss", 3327 .version_id = 1, 3328 .minimum_version_id = 1, 3329 .needed = virtio_net_rss_needed, 3330 .fields = (const VMStateField[]) { 3331 VMSTATE_BOOL(rss_data.enabled, VirtIONet), 3332 VMSTATE_BOOL(rss_data.redirect, VirtIONet), 3333 VMSTATE_BOOL(rss_data.populate_hash, VirtIONet), 3334 VMSTATE_UINT32(rss_data.hash_types, VirtIONet), 3335 VMSTATE_UINT16(rss_data.indirections_len, VirtIONet), 3336 VMSTATE_UINT16(rss_data.default_queue, VirtIONet), 3337 VMSTATE_UINT8_ARRAY(rss_data.key, VirtIONet, 3338 VIRTIO_NET_RSS_MAX_KEY_SIZE), 3339 VMSTATE_VARRAY_UINT16_ALLOC(rss_data.indirections_table, VirtIONet, 3340 rss_data.indirections_len, 0, 3341 vmstate_info_uint16, uint16_t), 3342 VMSTATE_END_OF_LIST() 3343 }, 3344 }; 3345 3346 static const VMStateDescription vmstate_virtio_net_device = { 3347 .name = "virtio-net-device", 3348 .version_id = VIRTIO_NET_VM_VERSION, 3349 .minimum_version_id = VIRTIO_NET_VM_VERSION, 3350 .post_load = virtio_net_post_load_device, 3351 .fields = (const VMStateField[]) { 3352 VMSTATE_UINT8_ARRAY(mac, VirtIONet, ETH_ALEN), 3353 VMSTATE_STRUCT_POINTER(vqs, VirtIONet, 3354 vmstate_virtio_net_queue_tx_waiting, 3355 VirtIONetQueue), 3356 VMSTATE_UINT32(mergeable_rx_bufs, VirtIONet), 3357 VMSTATE_UINT16(status, VirtIONet), 3358 VMSTATE_UINT8(promisc, VirtIONet), 3359 VMSTATE_UINT8(allmulti, VirtIONet), 3360 VMSTATE_UINT32(mac_table.in_use, VirtIONet), 3361 3362 /* Guarded pair: If it fits we load it, else we throw it away 3363 * - can happen if source has a larger MAC table.; post-load 3364 * sets flags in this case. 3365 */ 3366 VMSTATE_VBUFFER_MULTIPLY(mac_table.macs, VirtIONet, 3367 0, mac_table_fits, mac_table.in_use, 3368 ETH_ALEN), 3369 VMSTATE_UNUSED_VARRAY_UINT32(VirtIONet, mac_table_doesnt_fit, 0, 3370 mac_table.in_use, ETH_ALEN), 3371 3372 /* Note: This is an array of uint32's that's always been saved as a 3373 * buffer; hold onto your endiannesses; it's actually used as a bitmap 3374 * but based on the uint. 3375 */ 3376 VMSTATE_BUFFER_POINTER_UNSAFE(vlans, VirtIONet, 0, MAX_VLAN >> 3), 3377 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3378 vmstate_virtio_net_has_vnet), 3379 VMSTATE_UINT8(mac_table.multi_overflow, VirtIONet), 3380 VMSTATE_UINT8(mac_table.uni_overflow, VirtIONet), 3381 VMSTATE_UINT8(alluni, VirtIONet), 3382 VMSTATE_UINT8(nomulti, VirtIONet), 3383 VMSTATE_UINT8(nouni, VirtIONet), 3384 VMSTATE_UINT8(nobcast, VirtIONet), 3385 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3386 vmstate_virtio_net_has_ufo), 3387 VMSTATE_SINGLE_TEST(max_queue_pairs, VirtIONet, max_queue_pairs_gt_1, 0, 3388 vmstate_info_uint16_equal, uint16_t), 3389 VMSTATE_UINT16_TEST(curr_queue_pairs, VirtIONet, max_queue_pairs_gt_1), 3390 VMSTATE_WITH_TMP(VirtIONet, struct VirtIONetMigTmp, 3391 vmstate_virtio_net_tx_waiting), 3392 VMSTATE_UINT64_TEST(curr_guest_offloads, VirtIONet, 3393 has_ctrl_guest_offloads), 3394 VMSTATE_END_OF_LIST() 3395 }, 3396 .subsections = (const VMStateDescription * const []) { 3397 &vmstate_virtio_net_rss, 3398 NULL 3399 } 3400 }; 3401 3402 static NetClientInfo net_virtio_info = { 3403 .type = NET_CLIENT_DRIVER_NIC, 3404 .size = sizeof(NICState), 3405 .can_receive = virtio_net_can_receive, 3406 .receive = virtio_net_receive, 3407 .link_status_changed = virtio_net_set_link_status, 3408 .query_rx_filter = virtio_net_query_rxfilter, 3409 .announce = virtio_net_announce, 3410 }; 3411 3412 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx) 3413 { 3414 VirtIONet *n = VIRTIO_NET(vdev); 3415 NetClientState *nc; 3416 assert(n->vhost_started); 3417 if (!n->multiqueue && idx == 2) { 3418 /* Must guard against invalid features and bogus queue index 3419 * from being set by malicious guest, or penetrated through 3420 * buggy migration stream. 3421 */ 3422 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3423 qemu_log_mask(LOG_GUEST_ERROR, 3424 "%s: bogus vq index ignored\n", __func__); 3425 return false; 3426 } 3427 nc = qemu_get_subqueue(n->nic, n->max_queue_pairs); 3428 } else { 3429 nc = qemu_get_subqueue(n->nic, vq2q(idx)); 3430 } 3431 /* 3432 * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1 3433 * as the macro of configure interrupt's IDX, If this driver does not 3434 * support, the function will return false 3435 */ 3436 3437 if (idx == VIRTIO_CONFIG_IRQ_IDX) { 3438 return vhost_net_config_pending(get_vhost_net(nc->peer)); 3439 } 3440 return vhost_net_virtqueue_pending(get_vhost_net(nc->peer), idx); 3441 } 3442 3443 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx, 3444 bool mask) 3445 { 3446 VirtIONet *n = VIRTIO_NET(vdev); 3447 NetClientState *nc; 3448 assert(n->vhost_started); 3449 if (!n->multiqueue && idx == 2) { 3450 /* Must guard against invalid features and bogus queue index 3451 * from being set by malicious guest, or penetrated through 3452 * buggy migration stream. 3453 */ 3454 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) { 3455 qemu_log_mask(LOG_GUEST_ERROR, 3456 "%s: bogus vq index ignored\n", __func__); 3457 return; 3458 } 3459 nc = qemu_get_subqueue(n->nic, n->max_queue_pairs); 3460 } else { 3461 nc = qemu_get_subqueue(n->nic, vq2q(idx)); 3462 } 3463 /* 3464 *Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1 3465 * as the macro of configure interrupt's IDX, If this driver does not 3466 * support, the function will return 3467 */ 3468 3469 if (idx == VIRTIO_CONFIG_IRQ_IDX) { 3470 vhost_net_config_mask(get_vhost_net(nc->peer), vdev, mask); 3471 return; 3472 } 3473 vhost_net_virtqueue_mask(get_vhost_net(nc->peer), vdev, idx, mask); 3474 } 3475 3476 static void virtio_net_set_config_size(VirtIONet *n, uint64_t host_features) 3477 { 3478 virtio_add_feature(&host_features, VIRTIO_NET_F_MAC); 3479 3480 n->config_size = virtio_get_config_size(&cfg_size_params, host_features); 3481 } 3482 3483 void virtio_net_set_netclient_name(VirtIONet *n, const char *name, 3484 const char *type) 3485 { 3486 /* 3487 * The name can be NULL, the netclient name will be type.x. 3488 */ 3489 assert(type != NULL); 3490 3491 g_free(n->netclient_name); 3492 g_free(n->netclient_type); 3493 n->netclient_name = g_strdup(name); 3494 n->netclient_type = g_strdup(type); 3495 } 3496 3497 static bool failover_unplug_primary(VirtIONet *n, DeviceState *dev) 3498 { 3499 HotplugHandler *hotplug_ctrl; 3500 PCIDevice *pci_dev; 3501 Error *err = NULL; 3502 3503 hotplug_ctrl = qdev_get_hotplug_handler(dev); 3504 if (hotplug_ctrl) { 3505 pci_dev = PCI_DEVICE(dev); 3506 pci_dev->partially_hotplugged = true; 3507 hotplug_handler_unplug_request(hotplug_ctrl, dev, &err); 3508 if (err) { 3509 error_report_err(err); 3510 return false; 3511 } 3512 } else { 3513 return false; 3514 } 3515 return true; 3516 } 3517 3518 static bool failover_replug_primary(VirtIONet *n, DeviceState *dev, 3519 Error **errp) 3520 { 3521 Error *err = NULL; 3522 HotplugHandler *hotplug_ctrl; 3523 PCIDevice *pdev = PCI_DEVICE(dev); 3524 BusState *primary_bus; 3525 3526 if (!pdev->partially_hotplugged) { 3527 return true; 3528 } 3529 primary_bus = dev->parent_bus; 3530 if (!primary_bus) { 3531 error_setg(errp, "virtio_net: couldn't find primary bus"); 3532 return false; 3533 } 3534 qdev_set_parent_bus(dev, primary_bus, &error_abort); 3535 qatomic_set(&n->failover_primary_hidden, false); 3536 hotplug_ctrl = qdev_get_hotplug_handler(dev); 3537 if (hotplug_ctrl) { 3538 hotplug_handler_pre_plug(hotplug_ctrl, dev, &err); 3539 if (err) { 3540 goto out; 3541 } 3542 hotplug_handler_plug(hotplug_ctrl, dev, &err); 3543 } 3544 pdev->partially_hotplugged = false; 3545 3546 out: 3547 error_propagate(errp, err); 3548 return !err; 3549 } 3550 3551 static void virtio_net_handle_migration_primary(VirtIONet *n, MigrationEvent *e) 3552 { 3553 bool should_be_hidden; 3554 Error *err = NULL; 3555 DeviceState *dev = failover_find_primary_device(n); 3556 3557 if (!dev) { 3558 return; 3559 } 3560 3561 should_be_hidden = qatomic_read(&n->failover_primary_hidden); 3562 3563 if (e->type == MIG_EVENT_PRECOPY_SETUP && !should_be_hidden) { 3564 if (failover_unplug_primary(n, dev)) { 3565 vmstate_unregister(VMSTATE_IF(dev), qdev_get_vmsd(dev), dev); 3566 qapi_event_send_unplug_primary(dev->id); 3567 qatomic_set(&n->failover_primary_hidden, true); 3568 } else { 3569 warn_report("couldn't unplug primary device"); 3570 } 3571 } else if (e->type == MIG_EVENT_PRECOPY_FAILED) { 3572 /* We already unplugged the device let's plug it back */ 3573 if (!failover_replug_primary(n, dev, &err)) { 3574 if (err) { 3575 error_report_err(err); 3576 } 3577 } 3578 } 3579 } 3580 3581 static int virtio_net_migration_state_notifier(NotifierWithReturn *notifier, 3582 MigrationEvent *e, Error **errp) 3583 { 3584 VirtIONet *n = container_of(notifier, VirtIONet, migration_state); 3585 virtio_net_handle_migration_primary(n, e); 3586 return 0; 3587 } 3588 3589 static bool failover_hide_primary_device(DeviceListener *listener, 3590 const QDict *device_opts, 3591 bool from_json, 3592 Error **errp) 3593 { 3594 VirtIONet *n = container_of(listener, VirtIONet, primary_listener); 3595 const char *standby_id; 3596 3597 if (!device_opts) { 3598 return false; 3599 } 3600 3601 if (!qdict_haskey(device_opts, "failover_pair_id")) { 3602 return false; 3603 } 3604 3605 if (!qdict_haskey(device_opts, "id")) { 3606 error_setg(errp, "Device with failover_pair_id needs to have id"); 3607 return false; 3608 } 3609 3610 standby_id = qdict_get_str(device_opts, "failover_pair_id"); 3611 if (g_strcmp0(standby_id, n->netclient_name) != 0) { 3612 return false; 3613 } 3614 3615 /* 3616 * The hide helper can be called several times for a given device. 3617 * Check there is only one primary for a virtio-net device but 3618 * don't duplicate the qdict several times if it's called for the same 3619 * device. 3620 */ 3621 if (n->primary_opts) { 3622 const char *old, *new; 3623 /* devices with failover_pair_id always have an id */ 3624 old = qdict_get_str(n->primary_opts, "id"); 3625 new = qdict_get_str(device_opts, "id"); 3626 if (strcmp(old, new) != 0) { 3627 error_setg(errp, "Cannot attach more than one primary device to " 3628 "'%s': '%s' and '%s'", n->netclient_name, old, new); 3629 return false; 3630 } 3631 } else { 3632 n->primary_opts = qdict_clone_shallow(device_opts); 3633 n->primary_opts_from_json = from_json; 3634 } 3635 3636 /* failover_primary_hidden is set during feature negotiation */ 3637 return qatomic_read(&n->failover_primary_hidden); 3638 } 3639 3640 static void virtio_net_device_realize(DeviceState *dev, Error **errp) 3641 { 3642 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3643 VirtIONet *n = VIRTIO_NET(dev); 3644 NetClientState *nc; 3645 int i; 3646 3647 if (n->net_conf.mtu) { 3648 n->host_features |= (1ULL << VIRTIO_NET_F_MTU); 3649 } 3650 3651 if (n->net_conf.duplex_str) { 3652 if (strncmp(n->net_conf.duplex_str, "half", 5) == 0) { 3653 n->net_conf.duplex = DUPLEX_HALF; 3654 } else if (strncmp(n->net_conf.duplex_str, "full", 5) == 0) { 3655 n->net_conf.duplex = DUPLEX_FULL; 3656 } else { 3657 error_setg(errp, "'duplex' must be 'half' or 'full'"); 3658 return; 3659 } 3660 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX); 3661 } else { 3662 n->net_conf.duplex = DUPLEX_UNKNOWN; 3663 } 3664 3665 if (n->net_conf.speed < SPEED_UNKNOWN) { 3666 error_setg(errp, "'speed' must be between 0 and INT_MAX"); 3667 return; 3668 } 3669 if (n->net_conf.speed >= 0) { 3670 n->host_features |= (1ULL << VIRTIO_NET_F_SPEED_DUPLEX); 3671 } 3672 3673 if (n->failover) { 3674 n->primary_listener.hide_device = failover_hide_primary_device; 3675 qatomic_set(&n->failover_primary_hidden, true); 3676 device_listener_register(&n->primary_listener); 3677 migration_add_notifier(&n->migration_state, 3678 virtio_net_migration_state_notifier); 3679 n->host_features |= (1ULL << VIRTIO_NET_F_STANDBY); 3680 } 3681 3682 virtio_net_set_config_size(n, n->host_features); 3683 virtio_init(vdev, VIRTIO_ID_NET, n->config_size); 3684 3685 /* 3686 * We set a lower limit on RX queue size to what it always was. 3687 * Guests that want a smaller ring can always resize it without 3688 * help from us (using virtio 1 and up). 3689 */ 3690 if (n->net_conf.rx_queue_size < VIRTIO_NET_RX_QUEUE_MIN_SIZE || 3691 n->net_conf.rx_queue_size > VIRTQUEUE_MAX_SIZE || 3692 !is_power_of_2(n->net_conf.rx_queue_size)) { 3693 error_setg(errp, "Invalid rx_queue_size (= %" PRIu16 "), " 3694 "must be a power of 2 between %d and %d.", 3695 n->net_conf.rx_queue_size, VIRTIO_NET_RX_QUEUE_MIN_SIZE, 3696 VIRTQUEUE_MAX_SIZE); 3697 virtio_cleanup(vdev); 3698 return; 3699 } 3700 3701 if (n->net_conf.tx_queue_size < VIRTIO_NET_TX_QUEUE_MIN_SIZE || 3702 n->net_conf.tx_queue_size > virtio_net_max_tx_queue_size(n) || 3703 !is_power_of_2(n->net_conf.tx_queue_size)) { 3704 error_setg(errp, "Invalid tx_queue_size (= %" PRIu16 "), " 3705 "must be a power of 2 between %d and %d", 3706 n->net_conf.tx_queue_size, VIRTIO_NET_TX_QUEUE_MIN_SIZE, 3707 virtio_net_max_tx_queue_size(n)); 3708 virtio_cleanup(vdev); 3709 return; 3710 } 3711 3712 n->max_ncs = MAX(n->nic_conf.peers.queues, 1); 3713 3714 /* 3715 * Figure out the datapath queue pairs since the backend could 3716 * provide control queue via peers as well. 3717 */ 3718 if (n->nic_conf.peers.queues) { 3719 for (i = 0; i < n->max_ncs; i++) { 3720 if (n->nic_conf.peers.ncs[i]->is_datapath) { 3721 ++n->max_queue_pairs; 3722 } 3723 } 3724 } 3725 n->max_queue_pairs = MAX(n->max_queue_pairs, 1); 3726 3727 if (n->max_queue_pairs * 2 + 1 > VIRTIO_QUEUE_MAX) { 3728 error_setg(errp, "Invalid number of queue pairs (= %" PRIu32 "), " 3729 "must be a positive integer less than %d.", 3730 n->max_queue_pairs, (VIRTIO_QUEUE_MAX - 1) / 2); 3731 virtio_cleanup(vdev); 3732 return; 3733 } 3734 n->vqs = g_new0(VirtIONetQueue, n->max_queue_pairs); 3735 n->curr_queue_pairs = 1; 3736 n->tx_timeout = n->net_conf.txtimer; 3737 3738 if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer") 3739 && strcmp(n->net_conf.tx, "bh")) { 3740 warn_report("virtio-net: " 3741 "Unknown option tx=%s, valid options: \"timer\" \"bh\"", 3742 n->net_conf.tx); 3743 error_printf("Defaulting to \"bh\""); 3744 } 3745 3746 n->net_conf.tx_queue_size = MIN(virtio_net_max_tx_queue_size(n), 3747 n->net_conf.tx_queue_size); 3748 3749 for (i = 0; i < n->max_queue_pairs; i++) { 3750 virtio_net_add_queue(n, i); 3751 } 3752 3753 n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl); 3754 qemu_macaddr_default_if_unset(&n->nic_conf.macaddr); 3755 memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac)); 3756 n->status = VIRTIO_NET_S_LINK_UP; 3757 qemu_announce_timer_reset(&n->announce_timer, migrate_announce_params(), 3758 QEMU_CLOCK_VIRTUAL, 3759 virtio_net_announce_timer, n); 3760 n->announce_timer.round = 0; 3761 3762 if (n->netclient_type) { 3763 /* 3764 * Happen when virtio_net_set_netclient_name has been called. 3765 */ 3766 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 3767 n->netclient_type, n->netclient_name, 3768 &dev->mem_reentrancy_guard, n); 3769 } else { 3770 n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf, 3771 object_get_typename(OBJECT(dev)), dev->id, 3772 &dev->mem_reentrancy_guard, n); 3773 } 3774 3775 for (i = 0; i < n->max_queue_pairs; i++) { 3776 n->nic->ncs[i].do_not_pad = true; 3777 } 3778 3779 peer_test_vnet_hdr(n); 3780 if (peer_has_vnet_hdr(n)) { 3781 for (i = 0; i < n->max_queue_pairs; i++) { 3782 qemu_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true); 3783 } 3784 n->host_hdr_len = sizeof(struct virtio_net_hdr); 3785 } else { 3786 n->host_hdr_len = 0; 3787 } 3788 3789 qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a); 3790 3791 n->vqs[0].tx_waiting = 0; 3792 n->tx_burst = n->net_conf.txburst; 3793 virtio_net_set_mrg_rx_bufs(n, 0, 0, 0); 3794 n->promisc = 1; /* for compatibility */ 3795 3796 n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN); 3797 3798 n->vlans = g_malloc0(MAX_VLAN >> 3); 3799 3800 nc = qemu_get_queue(n->nic); 3801 nc->rxfilter_notify_enabled = 1; 3802 3803 if (nc->peer && nc->peer->info->type == NET_CLIENT_DRIVER_VHOST_VDPA) { 3804 struct virtio_net_config netcfg = {}; 3805 memcpy(&netcfg.mac, &n->nic_conf.macaddr, ETH_ALEN); 3806 vhost_net_set_config(get_vhost_net(nc->peer), 3807 (uint8_t *)&netcfg, 0, ETH_ALEN, VHOST_SET_CONFIG_TYPE_FRONTEND); 3808 } 3809 QTAILQ_INIT(&n->rsc_chains); 3810 n->qdev = dev; 3811 3812 net_rx_pkt_init(&n->rx_pkt); 3813 3814 if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) { 3815 virtio_net_load_ebpf(n, errp); 3816 } 3817 } 3818 3819 static void virtio_net_device_unrealize(DeviceState *dev) 3820 { 3821 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3822 VirtIONet *n = VIRTIO_NET(dev); 3823 int i, max_queue_pairs; 3824 3825 if (virtio_has_feature(n->host_features, VIRTIO_NET_F_RSS)) { 3826 virtio_net_unload_ebpf(n); 3827 } 3828 3829 /* This will stop vhost backend if appropriate. */ 3830 virtio_net_set_status(vdev, 0); 3831 3832 g_free(n->netclient_name); 3833 n->netclient_name = NULL; 3834 g_free(n->netclient_type); 3835 n->netclient_type = NULL; 3836 3837 g_free(n->mac_table.macs); 3838 g_free(n->vlans); 3839 3840 if (n->failover) { 3841 qobject_unref(n->primary_opts); 3842 device_listener_unregister(&n->primary_listener); 3843 migration_remove_notifier(&n->migration_state); 3844 } else { 3845 assert(n->primary_opts == NULL); 3846 } 3847 3848 max_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 3849 for (i = 0; i < max_queue_pairs; i++) { 3850 virtio_net_del_queue(n, i); 3851 } 3852 /* delete also control vq */ 3853 virtio_del_queue(vdev, max_queue_pairs * 2); 3854 qemu_announce_timer_del(&n->announce_timer, false); 3855 g_free(n->vqs); 3856 qemu_del_nic(n->nic); 3857 virtio_net_rsc_cleanup(n); 3858 g_free(n->rss_data.indirections_table); 3859 net_rx_pkt_uninit(n->rx_pkt); 3860 virtio_cleanup(vdev); 3861 } 3862 3863 static void virtio_net_instance_init(Object *obj) 3864 { 3865 VirtIONet *n = VIRTIO_NET(obj); 3866 3867 /* 3868 * The default config_size is sizeof(struct virtio_net_config). 3869 * Can be overridden with virtio_net_set_config_size. 3870 */ 3871 n->config_size = sizeof(struct virtio_net_config); 3872 device_add_bootindex_property(obj, &n->nic_conf.bootindex, 3873 "bootindex", "/ethernet-phy@0", 3874 DEVICE(n)); 3875 3876 ebpf_rss_init(&n->ebpf_rss); 3877 } 3878 3879 static int virtio_net_pre_save(void *opaque) 3880 { 3881 VirtIONet *n = opaque; 3882 3883 /* At this point, backend must be stopped, otherwise 3884 * it might keep writing to memory. */ 3885 assert(!n->vhost_started); 3886 3887 return 0; 3888 } 3889 3890 static bool primary_unplug_pending(void *opaque) 3891 { 3892 DeviceState *dev = opaque; 3893 DeviceState *primary; 3894 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 3895 VirtIONet *n = VIRTIO_NET(vdev); 3896 3897 if (!virtio_vdev_has_feature(vdev, VIRTIO_NET_F_STANDBY)) { 3898 return false; 3899 } 3900 primary = failover_find_primary_device(n); 3901 return primary ? primary->pending_deleted_event : false; 3902 } 3903 3904 static bool dev_unplug_pending(void *opaque) 3905 { 3906 DeviceState *dev = opaque; 3907 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev); 3908 3909 return vdc->primary_unplug_pending(dev); 3910 } 3911 3912 static struct vhost_dev *virtio_net_get_vhost(VirtIODevice *vdev) 3913 { 3914 VirtIONet *n = VIRTIO_NET(vdev); 3915 NetClientState *nc = qemu_get_queue(n->nic); 3916 struct vhost_net *net = get_vhost_net(nc->peer); 3917 return &net->dev; 3918 } 3919 3920 static const VMStateDescription vmstate_virtio_net = { 3921 .name = "virtio-net", 3922 .minimum_version_id = VIRTIO_NET_VM_VERSION, 3923 .version_id = VIRTIO_NET_VM_VERSION, 3924 .fields = (const VMStateField[]) { 3925 VMSTATE_VIRTIO_DEVICE, 3926 VMSTATE_END_OF_LIST() 3927 }, 3928 .pre_save = virtio_net_pre_save, 3929 .dev_unplug_pending = dev_unplug_pending, 3930 }; 3931 3932 static Property virtio_net_properties[] = { 3933 DEFINE_PROP_BIT64("csum", VirtIONet, host_features, 3934 VIRTIO_NET_F_CSUM, true), 3935 DEFINE_PROP_BIT64("guest_csum", VirtIONet, host_features, 3936 VIRTIO_NET_F_GUEST_CSUM, true), 3937 DEFINE_PROP_BIT64("gso", VirtIONet, host_features, VIRTIO_NET_F_GSO, true), 3938 DEFINE_PROP_BIT64("guest_tso4", VirtIONet, host_features, 3939 VIRTIO_NET_F_GUEST_TSO4, true), 3940 DEFINE_PROP_BIT64("guest_tso6", VirtIONet, host_features, 3941 VIRTIO_NET_F_GUEST_TSO6, true), 3942 DEFINE_PROP_BIT64("guest_ecn", VirtIONet, host_features, 3943 VIRTIO_NET_F_GUEST_ECN, true), 3944 DEFINE_PROP_BIT64("guest_ufo", VirtIONet, host_features, 3945 VIRTIO_NET_F_GUEST_UFO, true), 3946 DEFINE_PROP_BIT64("guest_announce", VirtIONet, host_features, 3947 VIRTIO_NET_F_GUEST_ANNOUNCE, true), 3948 DEFINE_PROP_BIT64("host_tso4", VirtIONet, host_features, 3949 VIRTIO_NET_F_HOST_TSO4, true), 3950 DEFINE_PROP_BIT64("host_tso6", VirtIONet, host_features, 3951 VIRTIO_NET_F_HOST_TSO6, true), 3952 DEFINE_PROP_BIT64("host_ecn", VirtIONet, host_features, 3953 VIRTIO_NET_F_HOST_ECN, true), 3954 DEFINE_PROP_BIT64("host_ufo", VirtIONet, host_features, 3955 VIRTIO_NET_F_HOST_UFO, true), 3956 DEFINE_PROP_BIT64("mrg_rxbuf", VirtIONet, host_features, 3957 VIRTIO_NET_F_MRG_RXBUF, true), 3958 DEFINE_PROP_BIT64("status", VirtIONet, host_features, 3959 VIRTIO_NET_F_STATUS, true), 3960 DEFINE_PROP_BIT64("ctrl_vq", VirtIONet, host_features, 3961 VIRTIO_NET_F_CTRL_VQ, true), 3962 DEFINE_PROP_BIT64("ctrl_rx", VirtIONet, host_features, 3963 VIRTIO_NET_F_CTRL_RX, true), 3964 DEFINE_PROP_BIT64("ctrl_vlan", VirtIONet, host_features, 3965 VIRTIO_NET_F_CTRL_VLAN, true), 3966 DEFINE_PROP_BIT64("ctrl_rx_extra", VirtIONet, host_features, 3967 VIRTIO_NET_F_CTRL_RX_EXTRA, true), 3968 DEFINE_PROP_BIT64("ctrl_mac_addr", VirtIONet, host_features, 3969 VIRTIO_NET_F_CTRL_MAC_ADDR, true), 3970 DEFINE_PROP_BIT64("ctrl_guest_offloads", VirtIONet, host_features, 3971 VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, true), 3972 DEFINE_PROP_BIT64("mq", VirtIONet, host_features, VIRTIO_NET_F_MQ, false), 3973 DEFINE_PROP_BIT64("rss", VirtIONet, host_features, 3974 VIRTIO_NET_F_RSS, false), 3975 DEFINE_PROP_BIT64("hash", VirtIONet, host_features, 3976 VIRTIO_NET_F_HASH_REPORT, false), 3977 DEFINE_PROP_ARRAY("ebpf-rss-fds", VirtIONet, nr_ebpf_rss_fds, 3978 ebpf_rss_fds, qdev_prop_string, char*), 3979 DEFINE_PROP_BIT64("guest_rsc_ext", VirtIONet, host_features, 3980 VIRTIO_NET_F_RSC_EXT, false), 3981 DEFINE_PROP_UINT32("rsc_interval", VirtIONet, rsc_timeout, 3982 VIRTIO_NET_RSC_DEFAULT_INTERVAL), 3983 DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf), 3984 DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer, 3985 TX_TIMER_INTERVAL), 3986 DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST), 3987 DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx), 3988 DEFINE_PROP_UINT16("rx_queue_size", VirtIONet, net_conf.rx_queue_size, 3989 VIRTIO_NET_RX_QUEUE_DEFAULT_SIZE), 3990 DEFINE_PROP_UINT16("tx_queue_size", VirtIONet, net_conf.tx_queue_size, 3991 VIRTIO_NET_TX_QUEUE_DEFAULT_SIZE), 3992 DEFINE_PROP_UINT16("host_mtu", VirtIONet, net_conf.mtu, 0), 3993 DEFINE_PROP_BOOL("x-mtu-bypass-backend", VirtIONet, mtu_bypass_backend, 3994 true), 3995 DEFINE_PROP_INT32("speed", VirtIONet, net_conf.speed, SPEED_UNKNOWN), 3996 DEFINE_PROP_STRING("duplex", VirtIONet, net_conf.duplex_str), 3997 DEFINE_PROP_BOOL("failover", VirtIONet, failover, false), 3998 DEFINE_PROP_BIT64("guest_uso4", VirtIONet, host_features, 3999 VIRTIO_NET_F_GUEST_USO4, true), 4000 DEFINE_PROP_BIT64("guest_uso6", VirtIONet, host_features, 4001 VIRTIO_NET_F_GUEST_USO6, true), 4002 DEFINE_PROP_BIT64("host_uso", VirtIONet, host_features, 4003 VIRTIO_NET_F_HOST_USO, true), 4004 DEFINE_PROP_END_OF_LIST(), 4005 }; 4006 4007 static void virtio_net_class_init(ObjectClass *klass, void *data) 4008 { 4009 DeviceClass *dc = DEVICE_CLASS(klass); 4010 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 4011 4012 device_class_set_props(dc, virtio_net_properties); 4013 dc->vmsd = &vmstate_virtio_net; 4014 set_bit(DEVICE_CATEGORY_NETWORK, dc->categories); 4015 vdc->realize = virtio_net_device_realize; 4016 vdc->unrealize = virtio_net_device_unrealize; 4017 vdc->get_config = virtio_net_get_config; 4018 vdc->set_config = virtio_net_set_config; 4019 vdc->get_features = virtio_net_get_features; 4020 vdc->set_features = virtio_net_set_features; 4021 vdc->bad_features = virtio_net_bad_features; 4022 vdc->reset = virtio_net_reset; 4023 vdc->queue_reset = virtio_net_queue_reset; 4024 vdc->queue_enable = virtio_net_queue_enable; 4025 vdc->set_status = virtio_net_set_status; 4026 vdc->guest_notifier_mask = virtio_net_guest_notifier_mask; 4027 vdc->guest_notifier_pending = virtio_net_guest_notifier_pending; 4028 vdc->legacy_features |= (0x1 << VIRTIO_NET_F_GSO); 4029 vdc->post_load = virtio_net_post_load_virtio; 4030 vdc->vmsd = &vmstate_virtio_net_device; 4031 vdc->primary_unplug_pending = primary_unplug_pending; 4032 vdc->get_vhost = virtio_net_get_vhost; 4033 vdc->toggle_device_iotlb = vhost_toggle_device_iotlb; 4034 } 4035 4036 static const TypeInfo virtio_net_info = { 4037 .name = TYPE_VIRTIO_NET, 4038 .parent = TYPE_VIRTIO_DEVICE, 4039 .instance_size = sizeof(VirtIONet), 4040 .instance_init = virtio_net_instance_init, 4041 .class_init = virtio_net_class_init, 4042 }; 4043 4044 static void virtio_register_types(void) 4045 { 4046 type_register_static(&virtio_net_info); 4047 } 4048 4049 type_init(virtio_register_types) 4050