hw/virtio-net.c

   1 /*
   2  * Virtio Network Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "qemu/iov.h"
  15 #include "hw/virtio.h"
  16 #include "net/net.h"
  17 #include "net/checksum.h"
  18 #include "net/tap.h"
  19 #include "qemu/error-report.h"
  20 #include "qemu/timer.h"
  21 #include "hw/virtio-net.h"
  22 #include "hw/vhost_net.h"
  23
  24 #define VIRTIO_NET_VM_VERSION    11
  25
  26 #define MAC_TABLE_ENTRIES    64
  27 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  28
  29 /*
  30  * Calculate the number of bytes up to and including the given 'field' of
  31  * 'container'.
  32  */
  33 #define endof(container, field) \
  34     (offsetof(container, field) + sizeof(((container *)0)->field))
  35
  36 typedef struct VirtIOFeature {
  37     uint32_t flags;
  38     size_t end;
  39 } VirtIOFeature;
  40
  41 static VirtIOFeature feature_sizes[] = {
  42     {.flags = 1 << VIRTIO_NET_F_MAC,
  43      .end = endof(struct virtio_net_config, mac)},
  44     {.flags = 1 << VIRTIO_NET_F_STATUS,
  45      .end = endof(struct virtio_net_config, status)},
  46     {.flags = 1 << VIRTIO_NET_F_MQ,
  47      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
  48     {}
  49 };
  50
  51 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
  52 {
  53     VirtIONet *n = qemu_get_nic_opaque(nc);
  54
  55     return &n->vqs[nc->queue_index];
  56 }
  57
  58 static int vq2q(int queue_index)
  59 {
  60     return queue_index / 2;
  61 }
  62
  63 /* TODO
  64  * - we could suppress RX interrupt if we were so inclined.
  65  */
  66
  67 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  68 {
  69     return (VirtIONet *)vdev;
  70 }
  71
  72 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  73 {
  74     VirtIONet *n = to_virtio_net(vdev);
  75     struct virtio_net_config netcfg;
  76
  77     stw_p(&netcfg.status, n->status);
  78     stw_p(&netcfg.max_virtqueue_pairs, n->max_queues);
  79     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  80     memcpy(config, &netcfg, n->config_size);
  81 }
  82
  83 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  84 {
  85     VirtIONet *n = to_virtio_net(vdev);
  86     struct virtio_net_config netcfg = {};
  87
  88     memcpy(&netcfg, config, n->config_size);
  89
  90     if (!(n->vdev.guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) &&
  91         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  92         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  93         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
  94     }
  95 }
  96
  97 static bool virtio_net_started(VirtIONet *n, uint8_t status)
  98 {
  99     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 100         (n->status & VIRTIO_NET_S_LINK_UP) && n->vdev.vm_running;
 101 }
 102
 103 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
 104 {
 105     NetClientState *nc = qemu_get_queue(n->nic);
 106     int queues = n->multiqueue ? n->max_queues : 1;
 107
 108     if (!nc->peer) {
 109         return;
 110     }
 111     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 112         return;
 113     }
 114
 115     if (!tap_get_vhost_net(nc->peer)) {
 116         return;
 117     }
 118
 119     if (!!n->vhost_started == virtio_net_started(n, status) &&
 120                               !nc->peer->link_down) {
 121         return;
 122     }
 123     if (!n->vhost_started) {
 124         int r;
 125         if (!vhost_net_query(tap_get_vhost_net(nc->peer), &n->vdev)) {
 126             return;
 127         }
 128         n->vhost_started = 1;
 129         r = vhost_net_start(&n->vdev, n->nic->ncs, queues);
 130         if (r < 0) {
 131             error_report("unable to start vhost net: %d: "
 132                          "falling back on userspace virtio", -r);
 133             n->vhost_started = 0;
 134         }
 135     } else {
 136         vhost_net_stop(&n->vdev, n->nic->ncs, queues);
 137         n->vhost_started = 0;
 138     }
 139 }
 140
 141 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 142 {
 143     VirtIONet *n = to_virtio_net(vdev);
 144     VirtIONetQueue *q;
 145     int i;
 146     uint8_t queue_status;
 147
 148     virtio_net_vhost_status(n, status);
 149
 150     for (i = 0; i < n->max_queues; i++) {
 151         q = &n->vqs[i];
 152
 153         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
 154             queue_status = 0;
 155         } else {
 156             queue_status = status;
 157         }
 158
 159         if (!q->tx_waiting) {
 160             continue;
 161         }
 162
 163         if (virtio_net_started(n, queue_status) && !n->vhost_started) {
 164             if (q->tx_timer) {
 165                 qemu_mod_timer(q->tx_timer,
 166                                qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 167             } else {
 168                 qemu_bh_schedule(q->tx_bh);
 169             }
 170         } else {
 171             if (q->tx_timer) {
 172                 qemu_del_timer(q->tx_timer);
 173             } else {
 174                 qemu_bh_cancel(q->tx_bh);
 175             }
 176         }
 177     }
 178 }
 179
 180 static void virtio_net_set_link_status(NetClientState *nc)
 181 {
 182     VirtIONet *n = qemu_get_nic_opaque(nc);
 183     uint16_t old_status = n->status;
 184
 185     if (nc->link_down)
 186         n->status &= ~VIRTIO_NET_S_LINK_UP;
 187     else
 188         n->status |= VIRTIO_NET_S_LINK_UP;
 189
 190     if (n->status != old_status)
 191         virtio_notify_config(&n->vdev);
 192
 193     virtio_net_set_status(&n->vdev, n->vdev.status);
 194 }
 195
 196 static void virtio_net_reset(VirtIODevice *vdev)
 197 {
 198     VirtIONet *n = to_virtio_net(vdev);
 199
 200     /* Reset back to compatibility mode */
 201     n->promisc = 1;
 202     n->allmulti = 0;
 203     n->alluni = 0;
 204     n->nomulti = 0;
 205     n->nouni = 0;
 206     n->nobcast = 0;
 207     /* multiqueue is disabled by default */
 208     n->curr_queues = 1;
 209
 210     /* Flush any MAC and VLAN filter table state */
 211     n->mac_table.in_use = 0;
 212     n->mac_table.first_multi = 0;
 213     n->mac_table.multi_overflow = 0;
 214     n->mac_table.uni_overflow = 0;
 215     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 216     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
 217     memset(n->vlans, 0, MAX_VLAN >> 3);
 218 }
 219
 220 static void peer_test_vnet_hdr(VirtIONet *n)
 221 {
 222     NetClientState *nc = qemu_get_queue(n->nic);
 223     if (!nc->peer) {
 224         return;
 225     }
 226
 227     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 228         return;
 229     }
 230
 231     n->has_vnet_hdr = tap_has_vnet_hdr(nc->peer);
 232 }
 233
 234 static int peer_has_vnet_hdr(VirtIONet *n)
 235 {
 236     return n->has_vnet_hdr;
 237 }
 238
 239 static int peer_has_ufo(VirtIONet *n)
 240 {
 241     if (!peer_has_vnet_hdr(n))
 242         return 0;
 243
 244     n->has_ufo = tap_has_ufo(qemu_get_queue(n->nic)->peer);
 245
 246     return n->has_ufo;
 247 }
 248
 249 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
 250 {
 251     int i;
 252     NetClientState *nc;
 253
 254     n->mergeable_rx_bufs = mergeable_rx_bufs;
 255
 256     n->guest_hdr_len = n->mergeable_rx_bufs ?
 257         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 258
 259     for (i = 0; i < n->max_queues; i++) {
 260         nc = qemu_get_subqueue(n->nic, i);
 261
 262         if (peer_has_vnet_hdr(n) &&
 263             tap_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
 264             tap_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
 265             n->host_hdr_len = n->guest_hdr_len;
 266         }
 267     }
 268 }
 269
 270 static int peer_attach(VirtIONet *n, int index)
 271 {
 272     NetClientState *nc = qemu_get_subqueue(n->nic, index);
 273
 274     if (!nc->peer) {
 275         return 0;
 276     }
 277
 278     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 279         return 0;
 280     }
 281
 282     return tap_enable(nc->peer);
 283 }
 284
 285 static int peer_detach(VirtIONet *n, int index)
 286 {
 287     NetClientState *nc = qemu_get_subqueue(n->nic, index);
 288
 289     if (!nc->peer) {
 290         return 0;
 291     }
 292
 293     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
 294         return 0;
 295     }
 296
 297     return tap_disable(nc->peer);
 298 }
 299
 300 static void virtio_net_set_queues(VirtIONet *n)
 301 {
 302     int i;
 303
 304     for (i = 0; i < n->max_queues; i++) {
 305         if (i < n->curr_queues) {
 306             assert(!peer_attach(n, i));
 307         } else {
 308             assert(!peer_detach(n, i));
 309         }
 310     }
 311 }
 312
 313 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl);
 314
 315 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 316 {
 317     VirtIONet *n = to_virtio_net(vdev);
 318     NetClientState *nc = qemu_get_queue(n->nic);
 319
 320     features |= (1 << VIRTIO_NET_F_MAC);
 321
 322     if (!peer_has_vnet_hdr(n)) {
 323         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 324         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 325         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 326         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 327
 328         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 329         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 330         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 331         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 332     }
 333
 334     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 335         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 336         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 337     }
 338
 339     if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 340         return features;
 341     }
 342     if (!tap_get_vhost_net(nc->peer)) {
 343         return features;
 344     }
 345     return vhost_net_get_features(tap_get_vhost_net(nc->peer), features);
 346 }
 347
 348 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 349 {
 350     uint32_t features = 0;
 351
 352     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 353      * but also these: */
 354     features |= (1 << VIRTIO_NET_F_MAC);
 355     features |= (1 << VIRTIO_NET_F_CSUM);
 356     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 357     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 358     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 359
 360     return features;
 361 }
 362
 363 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 364 {
 365     VirtIONet *n = to_virtio_net(vdev);
 366     int i;
 367
 368     virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)),
 369                               !!(features & (1 << VIRTIO_NET_F_CTRL_VQ)));
 370
 371     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
 372
 373     if (n->has_vnet_hdr) {
 374         tap_set_offload(qemu_get_subqueue(n->nic, 0)->peer,
 375                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 376                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 377                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 378                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 379                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 380     }
 381
 382     for (i = 0;  i < n->max_queues; i++) {
 383         NetClientState *nc = qemu_get_subqueue(n->nic, i);
 384
 385         if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 386             continue;
 387         }
 388         if (!tap_get_vhost_net(nc->peer)) {
 389             continue;
 390         }
 391         vhost_net_ack_features(tap_get_vhost_net(nc->peer), features);
 392     }
 393 }
 394
 395 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 396                                      struct iovec *iov, unsigned int iov_cnt)
 397 {
 398     uint8_t on;
 399     size_t s;
 400
 401     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
 402     if (s != sizeof(on)) {
 403         return VIRTIO_NET_ERR;
 404     }
 405
 406     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
 407         n->promisc = on;
 408     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
 409         n->allmulti = on;
 410     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
 411         n->alluni = on;
 412     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
 413         n->nomulti = on;
 414     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
 415         n->nouni = on;
 416     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
 417         n->nobcast = on;
 418     } else {
 419         return VIRTIO_NET_ERR;
 420     }
 421
 422     return VIRTIO_NET_OK;
 423 }
 424
 425 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 426                                  struct iovec *iov, unsigned int iov_cnt)
 427 {
 428     struct virtio_net_ctrl_mac mac_data;
 429     size_t s;
 430
 431     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
 432         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
 433             return VIRTIO_NET_ERR;
 434         }
 435         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
 436         assert(s == sizeof(n->mac));
 437         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
 438         return VIRTIO_NET_OK;
 439     }
 440
 441     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
 442         return VIRTIO_NET_ERR;
 443     }
 444
 445     n->mac_table.in_use = 0;
 446     n->mac_table.first_multi = 0;
 447     n->mac_table.uni_overflow = 0;
 448     n->mac_table.multi_overflow = 0;
 449     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 450
 451     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 452                    sizeof(mac_data.entries));
 453     mac_data.entries = ldl_p(&mac_data.entries);
 454     if (s != sizeof(mac_data.entries)) {
 455         return VIRTIO_NET_ERR;
 456     }
 457     iov_discard_front(&iov, &iov_cnt, s);
 458
 459     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
 460         return VIRTIO_NET_ERR;
 461     }
 462
 463     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 464         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 465                        mac_data.entries * ETH_ALEN);
 466         if (s != mac_data.entries * ETH_ALEN) {
 467             return VIRTIO_NET_ERR;
 468         }
 469         n->mac_table.in_use += mac_data.entries;
 470     } else {
 471         n->mac_table.uni_overflow = 1;
 472     }
 473
 474     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
 475
 476     n->mac_table.first_multi = n->mac_table.in_use;
 477
 478     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 479                    sizeof(mac_data.entries));
 480     mac_data.entries = ldl_p(&mac_data.entries);
 481     if (s != sizeof(mac_data.entries)) {
 482         return VIRTIO_NET_ERR;
 483     }
 484
 485     iov_discard_front(&iov, &iov_cnt, s);
 486
 487     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
 488         return VIRTIO_NET_ERR;
 489     }
 490
 491     if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 492         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 493                        mac_data.entries * ETH_ALEN);
 494         if (s != mac_data.entries * ETH_ALEN) {
 495             return VIRTIO_NET_ERR;
 496         }
 497         n->mac_table.in_use += mac_data.entries;
 498     } else {
 499         n->mac_table.multi_overflow = 1;
 500     }
 501
 502     return VIRTIO_NET_OK;
 503 }
 504
 505 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 506                                         struct iovec *iov, unsigned int iov_cnt)
 507 {
 508     uint16_t vid;
 509     size_t s;
 510
 511     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
 512     vid = lduw_p(&vid);
 513     if (s != sizeof(vid)) {
 514         return VIRTIO_NET_ERR;
 515     }
 516
 517     if (vid >= MAX_VLAN)
 518         return VIRTIO_NET_ERR;
 519
 520     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 521         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 522     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 523         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 524     else
 525         return VIRTIO_NET_ERR;
 526
 527     return VIRTIO_NET_OK;
 528 }
 529
 530 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
 531                                 VirtQueueElement *elem)
 532 {
 533     struct virtio_net_ctrl_mq s;
 534
 535     if (elem->out_num != 2 ||
 536         elem->out_sg[1].iov_len != sizeof(struct virtio_net_ctrl_mq)) {
 537         error_report("virtio-net ctrl invalid steering command");
 538         return VIRTIO_NET_ERR;
 539     }
 540
 541     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 542         return VIRTIO_NET_ERR;
 543     }
 544
 545     memcpy(&s, elem->out_sg[1].iov_base, sizeof(struct virtio_net_ctrl_mq));
 546
 547     if (s.virtqueue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
 548         s.virtqueue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
 549         s.virtqueue_pairs > n->max_queues ||
 550         !n->multiqueue) {
 551         return VIRTIO_NET_ERR;
 552     }
 553
 554     n->curr_queues = s.virtqueue_pairs;
 555     /* stop the backend before changing the number of queues to avoid handling a
 556      * disabled queue */
 557     virtio_net_set_status(&n->vdev, n->vdev.status);
 558     virtio_net_set_queues(n);
 559
 560     return VIRTIO_NET_OK;
 561 }
 562 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 563 {
 564     VirtIONet *n = to_virtio_net(vdev);
 565     struct virtio_net_ctrl_hdr ctrl;
 566     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 567     VirtQueueElement elem;
 568     size_t s;
 569     struct iovec *iov;
 570     unsigned int iov_cnt;
 571
 572     while (virtqueue_pop(vq, &elem)) {
 573         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
 574             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
 575             error_report("virtio-net ctrl missing headers");
 576             exit(1);
 577         }
 578
 579         iov = elem.out_sg;
 580         iov_cnt = elem.out_num;
 581         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
 582         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
 583         if (s != sizeof(ctrl)) {
 584             status = VIRTIO_NET_ERR;
 585         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
 586             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
 587         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
 588             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
 589         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
 590             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
 591         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
 592             status = virtio_net_handle_mq(n, ctrl.cmd, &elem);
 593         }
 594
 595         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
 596         assert(s == sizeof(status));
 597
 598         virtqueue_push(vq, &elem, sizeof(status));
 599         virtio_notify(vdev, vq);
 600     }
 601 }
 602
 603 /* RX */
 604
 605 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 606 {
 607     VirtIONet *n = to_virtio_net(vdev);
 608     int queue_index = vq2q(virtio_get_queue_index(vq));
 609
 610     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
 611 }
 612
 613 static int virtio_net_can_receive(NetClientState *nc)
 614 {
 615     VirtIONet *n = qemu_get_nic_opaque(nc);
 616     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 617
 618     if (!n->vdev.vm_running) {
 619         return 0;
 620     }
 621
 622     if (nc->queue_index >= n->curr_queues) {
 623         return 0;
 624     }
 625
 626     if (!virtio_queue_ready(q->rx_vq) ||
 627         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 628         return 0;
 629     }
 630
 631     return 1;
 632 }
 633
 634 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
 635 {
 636     VirtIONet *n = q->n;
 637     if (virtio_queue_empty(q->rx_vq) ||
 638         (n->mergeable_rx_bufs &&
 639          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
 640         virtio_queue_set_notification(q->rx_vq, 1);
 641
 642         /* To avoid a race condition where the guest has made some buffers
 643          * available after the above check but before notification was
 644          * enabled, check for available buffers again.
 645          */
 646         if (virtio_queue_empty(q->rx_vq) ||
 647             (n->mergeable_rx_bufs &&
 648              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
 649             return 0;
 650         }
 651     }
 652
 653     virtio_queue_set_notification(q->rx_vq, 0);
 654     return 1;
 655 }
 656
 657 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 658  * it never finds out that the packets don't have valid checksums.  This
 659  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 660  * fix this with Xen but it hasn't appeared in an upstream release of
 661  * dhclient yet.
 662  *
 663  * To avoid breaking existing guests, we catch udp packets and add
 664  * checksums.  This is terrible but it's better than hacking the guest
 665  * kernels.
 666  *
 667  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 668  * we should provide a mechanism to disable it to avoid polluting the host
 669  * cache.
 670  */
 671 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 672                                         uint8_t *buf, size_t size)
 673 {
 674     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 675         (size > 27 && size < 1500) && /* normal sized MTU */
 676         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 677         (buf[23] == 17) && /* ip.protocol == UDP */
 678         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 679         net_checksum_calculate(buf, size);
 680         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 681     }
 682 }
 683
 684 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
 685                            const void *buf, size_t size)
 686 {
 687     if (n->has_vnet_hdr) {
 688         /* FIXME this cast is evil */
 689         void *wbuf = (void *)buf;
 690         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
 691                                     size - n->host_hdr_len);
 692         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
 693     } else {
 694         struct virtio_net_hdr hdr = {
 695             .flags = 0,
 696             .gso_type = VIRTIO_NET_HDR_GSO_NONE
 697         };
 698         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
 699     }
 700 }
 701
 702 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 703 {
 704     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 705     static const uint8_t vlan[] = {0x81, 0x00};
 706     uint8_t *ptr = (uint8_t *)buf;
 707     int i;
 708
 709     if (n->promisc)
 710         return 1;
 711
 712     ptr += n->host_hdr_len;
 713
 714     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 715         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 716         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 717             return 0;
 718     }
 719
 720     if (ptr[0] & 1) { // multicast
 721         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 722             return !n->nobcast;
 723         } else if (n->nomulti) {
 724             return 0;
 725         } else if (n->allmulti || n->mac_table.multi_overflow) {
 726             return 1;
 727         }
 728
 729         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 730             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 731                 return 1;
 732             }
 733         }
 734     } else { // unicast
 735         if (n->nouni) {
 736             return 0;
 737         } else if (n->alluni || n->mac_table.uni_overflow) {
 738             return 1;
 739         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 740             return 1;
 741         }
 742
 743         for (i = 0; i < n->mac_table.first_multi; i++) {
 744             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 745                 return 1;
 746             }
 747         }
 748     }
 749
 750     return 0;
 751 }
 752
 753 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 754 {
 755     VirtIONet *n = qemu_get_nic_opaque(nc);
 756     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 757     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
 758     struct virtio_net_hdr_mrg_rxbuf mhdr;
 759     unsigned mhdr_cnt = 0;
 760     size_t offset, i, guest_offset;
 761
 762     if (!virtio_net_can_receive(nc)) {
 763         return -1;
 764     }
 765
 766     /* hdr_len refers to the header we supply to the guest */
 767     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
 768         return 0;
 769     }
 770
 771     if (!receive_filter(n, buf, size))
 772         return size;
 773
 774     offset = i = 0;
 775
 776     while (offset < size) {
 777         VirtQueueElement elem;
 778         int len, total;
 779         const struct iovec *sg = elem.in_sg;
 780
 781         total = 0;
 782
 783         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
 784             if (i == 0)
 785                 return -1;
 786             error_report("virtio-net unexpected empty queue: "
 787                     "i %zd mergeable %d offset %zd, size %zd, "
 788                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 789                     i, n->mergeable_rx_bufs, offset, size,
 790                     n->guest_hdr_len, n->host_hdr_len, n->vdev.guest_features);
 791             exit(1);
 792         }
 793
 794         if (elem.in_num < 1) {
 795             error_report("virtio-net receive queue contains no in buffers");
 796             exit(1);
 797         }
 798
 799         if (i == 0) {
 800             assert(offset == 0);
 801             if (n->mergeable_rx_bufs) {
 802                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
 803                                     sg, elem.in_num,
 804                                     offsetof(typeof(mhdr), num_buffers),
 805                                     sizeof(mhdr.num_buffers));
 806             }
 807
 808             receive_header(n, sg, elem.in_num, buf, size);
 809             offset = n->host_hdr_len;
 810             total += n->guest_hdr_len;
 811             guest_offset = n->guest_hdr_len;
 812         } else {
 813             guest_offset = 0;
 814         }
 815
 816         /* copy in packet.  ugh */
 817         len = iov_from_buf(sg, elem.in_num, guest_offset,
 818                            buf + offset, size - offset);
 819         total += len;
 820         offset += len;
 821         /* If buffers can't be merged, at this point we
 822          * must have consumed the complete packet.
 823          * Otherwise, drop it. */
 824         if (!n->mergeable_rx_bufs && offset < size) {
 825 #if 0
 826             error_report("virtio-net truncated non-mergeable packet: "
 827                          "i %zd mergeable %d offset %zd, size %zd, "
 828                          "guest hdr len %zd, host hdr len %zd",
 829                          i, n->mergeable_rx_bufs,
 830                          offset, size, n->guest_hdr_len, n->host_hdr_len);
 831 #endif
 832             return size;
 833         }
 834
 835         /* signal other side */
 836         virtqueue_fill(q->rx_vq, &elem, total, i++);
 837     }
 838
 839     if (mhdr_cnt) {
 840         stw_p(&mhdr.num_buffers, i);
 841         iov_from_buf(mhdr_sg, mhdr_cnt,
 842                      0,
 843                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
 844     }
 845
 846     virtqueue_flush(q->rx_vq, i);
 847     virtio_notify(&n->vdev, q->rx_vq);
 848
 849     return size;
 850 }
 851
 852 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
 853
 854 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
 855 {
 856     VirtIONet *n = qemu_get_nic_opaque(nc);
 857     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 858
 859     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
 860     virtio_notify(&n->vdev, q->tx_vq);
 861
 862     q->async_tx.elem.out_num = q->async_tx.len = 0;
 863
 864     virtio_queue_set_notification(q->tx_vq, 1);
 865     virtio_net_flush_tx(q);
 866 }
 867
 868 /* TX */
 869 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
 870 {
 871     VirtIONet *n = q->n;
 872     VirtQueueElement elem;
 873     int32_t num_packets = 0;
 874     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
 875     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 876         return num_packets;
 877     }
 878
 879     assert(n->vdev.vm_running);
 880
 881     if (q->async_tx.elem.out_num) {
 882         virtio_queue_set_notification(q->tx_vq, 0);
 883         return num_packets;
 884     }
 885
 886     while (virtqueue_pop(q->tx_vq, &elem)) {
 887         ssize_t ret, len;
 888         unsigned int out_num = elem.out_num;
 889         struct iovec *out_sg = &elem.out_sg[0];
 890         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 891
 892         if (out_num < 1) {
 893             error_report("virtio-net header not in first element");
 894             exit(1);
 895         }
 896
 897         /*
 898          * If host wants to see the guest header as is, we can
 899          * pass it on unchanged. Otherwise, copy just the parts
 900          * that host is interested in.
 901          */
 902         assert(n->host_hdr_len <= n->guest_hdr_len);
 903         if (n->host_hdr_len != n->guest_hdr_len) {
 904             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
 905                                        out_sg, out_num,
 906                                        0, n->host_hdr_len);
 907             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
 908                              out_sg, out_num,
 909                              n->guest_hdr_len, -1);
 910             out_num = sg_num;
 911             out_sg = sg;
 912         }
 913
 914         len = n->guest_hdr_len;
 915
 916         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
 917                                       out_sg, out_num, virtio_net_tx_complete);
 918         if (ret == 0) {
 919             virtio_queue_set_notification(q->tx_vq, 0);
 920             q->async_tx.elem = elem;
 921             q->async_tx.len  = len;
 922             return -EBUSY;
 923         }
 924
 925         len += ret;
 926
 927         virtqueue_push(q->tx_vq, &elem, 0);
 928         virtio_notify(&n->vdev, q->tx_vq);
 929
 930         if (++num_packets >= n->tx_burst) {
 931             break;
 932         }
 933     }
 934     return num_packets;
 935 }
 936
 937 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 938 {
 939     VirtIONet *n = to_virtio_net(vdev);
 940     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
 941
 942     /* This happens when device was stopped but VCPU wasn't. */
 943     if (!n->vdev.vm_running) {
 944         q->tx_waiting = 1;
 945         return;
 946     }
 947
 948     if (q->tx_waiting) {
 949         virtio_queue_set_notification(vq, 1);
 950         qemu_del_timer(q->tx_timer);
 951         q->tx_waiting = 0;
 952         virtio_net_flush_tx(q);
 953     } else {
 954         qemu_mod_timer(q->tx_timer,
 955                        qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 956         q->tx_waiting = 1;
 957         virtio_queue_set_notification(vq, 0);
 958     }
 959 }
 960
 961 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
 962 {
 963     VirtIONet *n = to_virtio_net(vdev);
 964     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
 965
 966     if (unlikely(q->tx_waiting)) {
 967         return;
 968     }
 969     q->tx_waiting = 1;
 970     /* This happens when device was stopped but VCPU wasn't. */
 971     if (!n->vdev.vm_running) {
 972         return;
 973     }
 974     virtio_queue_set_notification(vq, 0);
 975     qemu_bh_schedule(q->tx_bh);
 976 }
 977
 978 static void virtio_net_tx_timer(void *opaque)
 979 {
 980     VirtIONetQueue *q = opaque;
 981     VirtIONet *n = q->n;
 982     assert(n->vdev.vm_running);
 983
 984     q->tx_waiting = 0;
 985
 986     /* Just in case the driver is not ready on more */
 987     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 988         return;
 989
 990     virtio_queue_set_notification(q->tx_vq, 1);
 991     virtio_net_flush_tx(q);
 992 }
 993
 994 static void virtio_net_tx_bh(void *opaque)
 995 {
 996     VirtIONetQueue *q = opaque;
 997     VirtIONet *n = q->n;
 998     int32_t ret;
 999
1000     assert(n->vdev.vm_running);
1001
1002     q->tx_waiting = 0;
1003
1004     /* Just in case the driver is not ready on more */
1005     if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
1006         return;
1007
1008     ret = virtio_net_flush_tx(q);
1009     if (ret == -EBUSY) {
1010         return; /* Notification re-enable handled by tx_complete */
1011     }
1012
1013     /* If we flush a full burst of packets, assume there are
1014      * more coming and immediately reschedule */
1015     if (ret >= n->tx_burst) {
1016         qemu_bh_schedule(q->tx_bh);
1017         q->tx_waiting = 1;
1018         return;
1019     }
1020
1021     /* If less than a full burst, re-enable notification and flush
1022      * anything that may have come in while we weren't looking.  If
1023      * we find something, assume the guest is still active and reschedule */
1024     virtio_queue_set_notification(q->tx_vq, 1);
1025     if (virtio_net_flush_tx(q) > 0) {
1026         virtio_queue_set_notification(q->tx_vq, 0);
1027         qemu_bh_schedule(q->tx_bh);
1028         q->tx_waiting = 1;
1029     }
1030 }
1031
1032 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
1033 {
1034     VirtIODevice *vdev = &n->vdev;
1035     int i, max = multiqueue ? n->max_queues : 1;
1036
1037     n->multiqueue = multiqueue;
1038
1039     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1040         virtio_del_queue(vdev, i);
1041     }
1042
1043     for (i = 1; i < max; i++) {
1044         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1045         if (n->vqs[i].tx_timer) {
1046             n->vqs[i].tx_vq =
1047                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1048             n->vqs[i].tx_timer = qemu_new_timer_ns(vm_clock,
1049                                                    virtio_net_tx_timer,
1050                                                    &n->vqs[i]);
1051         } else {
1052             n->vqs[i].tx_vq =
1053                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1054             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1055         }
1056
1057         n->vqs[i].tx_waiting = 0;
1058         n->vqs[i].n = n;
1059     }
1060
1061     if (ctrl) {
1062         n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1063     }
1064
1065     virtio_net_set_queues(n);
1066 }
1067
1068 static void virtio_net_save(QEMUFile *f, void *opaque)
1069 {
1070     int i;
1071     VirtIONet *n = opaque;
1072
1073     /* At this point, backend must be stopped, otherwise
1074      * it might keep writing to memory. */
1075     assert(!n->vhost_started);
1076     virtio_save(&n->vdev, f);
1077
1078     qemu_put_buffer(f, n->mac, ETH_ALEN);
1079     qemu_put_be32(f, n->vqs[0].tx_waiting);
1080     qemu_put_be32(f, n->mergeable_rx_bufs);
1081     qemu_put_be16(f, n->status);
1082     qemu_put_byte(f, n->promisc);
1083     qemu_put_byte(f, n->allmulti);
1084     qemu_put_be32(f, n->mac_table.in_use);
1085     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1086     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1087     qemu_put_be32(f, n->has_vnet_hdr);
1088     qemu_put_byte(f, n->mac_table.multi_overflow);
1089     qemu_put_byte(f, n->mac_table.uni_overflow);
1090     qemu_put_byte(f, n->alluni);
1091     qemu_put_byte(f, n->nomulti);
1092     qemu_put_byte(f, n->nouni);
1093     qemu_put_byte(f, n->nobcast);
1094     qemu_put_byte(f, n->has_ufo);
1095     if (n->max_queues > 1) {
1096         qemu_put_be16(f, n->max_queues);
1097         qemu_put_be16(f, n->curr_queues);
1098         for (i = 1; i < n->curr_queues; i++) {
1099             qemu_put_be32(f, n->vqs[i].tx_waiting);
1100         }
1101     }
1102 }
1103
1104 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1105 {
1106     VirtIONet *n = opaque;
1107     int ret, i, link_down;
1108
1109     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1110         return -EINVAL;
1111
1112     ret = virtio_load(&n->vdev, f);
1113     if (ret) {
1114         return ret;
1115     }
1116
1117     qemu_get_buffer(f, n->mac, ETH_ALEN);
1118     n->vqs[0].tx_waiting = qemu_get_be32(f);
1119
1120     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1121
1122     if (version_id >= 3)
1123         n->status = qemu_get_be16(f);
1124
1125     if (version_id >= 4) {
1126         if (version_id < 8) {
1127             n->promisc = qemu_get_be32(f);
1128             n->allmulti = qemu_get_be32(f);
1129         } else {
1130             n->promisc = qemu_get_byte(f);
1131             n->allmulti = qemu_get_byte(f);
1132         }
1133     }
1134
1135     if (version_id >= 5) {
1136         n->mac_table.in_use = qemu_get_be32(f);
1137         /* MAC_TABLE_ENTRIES may be different from the saved image */
1138         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1139             qemu_get_buffer(f, n->mac_table.macs,
1140                             n->mac_table.in_use * ETH_ALEN);
1141         } else if (n->mac_table.in_use) {
1142             uint8_t *buf = g_malloc0(n->mac_table.in_use);
1143             qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
1144             g_free(buf);
1145             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1146             n->mac_table.in_use = 0;
1147         }
1148     }
1149
1150     if (version_id >= 6)
1151         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1152
1153     if (version_id >= 7) {
1154         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1155             error_report("virtio-net: saved image requires vnet_hdr=on");
1156             return -1;
1157         }
1158
1159         if (n->has_vnet_hdr) {
1160             tap_set_offload(qemu_get_queue(n->nic)->peer,
1161                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
1162                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
1163                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
1164                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
1165                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
1166         }
1167     }
1168
1169     if (version_id >= 9) {
1170         n->mac_table.multi_overflow = qemu_get_byte(f);
1171         n->mac_table.uni_overflow = qemu_get_byte(f);
1172     }
1173
1174     if (version_id >= 10) {
1175         n->alluni = qemu_get_byte(f);
1176         n->nomulti = qemu_get_byte(f);
1177         n->nouni = qemu_get_byte(f);
1178         n->nobcast = qemu_get_byte(f);
1179     }
1180
1181     if (version_id >= 11) {
1182         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1183             error_report("virtio-net: saved image requires TUN_F_UFO support");
1184             return -1;
1185         }
1186     }
1187
1188     if (n->max_queues > 1) {
1189         if (n->max_queues != qemu_get_be16(f)) {
1190             error_report("virtio-net: different max_queues ");
1191             return -1;
1192         }
1193
1194         n->curr_queues = qemu_get_be16(f);
1195         for (i = 1; i < n->curr_queues; i++) {
1196             n->vqs[i].tx_waiting = qemu_get_be32(f);
1197         }
1198     }
1199
1200     virtio_net_set_queues(n);
1201
1202     /* Find the first multicast entry in the saved MAC filter */
1203     for (i = 0; i < n->mac_table.in_use; i++) {
1204         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1205             break;
1206         }
1207     }
1208     n->mac_table.first_multi = i;
1209
1210     /* nc.link_down can't be migrated, so infer link_down according
1211      * to link status bit in n->status */
1212     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1213     for (i = 0; i < n->max_queues; i++) {
1214         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1215     }
1216
1217     return 0;
1218 }
1219
1220 static void virtio_net_cleanup(NetClientState *nc)
1221 {
1222     VirtIONet *n = qemu_get_nic_opaque(nc);
1223
1224     n->nic = NULL;
1225 }
1226
1227 static NetClientInfo net_virtio_info = {
1228     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1229     .size = sizeof(NICState),
1230     .can_receive = virtio_net_can_receive,
1231     .receive = virtio_net_receive,
1232         .cleanup = virtio_net_cleanup,
1233     .link_status_changed = virtio_net_set_link_status,
1234 };
1235
1236 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1237 {
1238     VirtIONet *n = to_virtio_net(vdev);
1239     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1240     assert(n->vhost_started);
1241     return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1242 }
1243
1244 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1245                                            bool mask)
1246 {
1247     VirtIONet *n = to_virtio_net(vdev);
1248     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1249     assert(n->vhost_started);
1250     vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1251                              vdev, idx, mask);
1252 }
1253
1254 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
1255                               virtio_net_conf *net, uint32_t host_features)
1256 {
1257     VirtIONet *n;
1258     int i, config_size = 0;
1259
1260     for (i = 0; feature_sizes[i].flags != 0; i++) {
1261         if (host_features & feature_sizes[i].flags) {
1262             config_size = MAX(feature_sizes[i].end, config_size);
1263         }
1264     }
1265
1266     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
1267                                         config_size, sizeof(VirtIONet));
1268
1269     n->config_size = config_size;
1270     n->vdev.get_config = virtio_net_get_config;
1271     n->vdev.set_config = virtio_net_set_config;
1272     n->vdev.get_features = virtio_net_get_features;
1273     n->vdev.set_features = virtio_net_set_features;
1274     n->vdev.bad_features = virtio_net_bad_features;
1275     n->vdev.reset = virtio_net_reset;
1276     n->vdev.set_status = virtio_net_set_status;
1277     n->vdev.guest_notifier_mask = virtio_net_guest_notifier_mask;
1278     n->vdev.guest_notifier_pending = virtio_net_guest_notifier_pending;
1279     n->max_queues = MAX(conf->queues, 1);
1280     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1281     n->vqs[0].rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
1282     n->curr_queues = 1;
1283     n->vqs[0].n = n;
1284     n->tx_timeout = net->txtimer;
1285
1286     if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
1287         error_report("virtio-net: "
1288                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1289                      net->tx);
1290         error_report("Defaulting to \"bh\"");
1291     }
1292
1293     if (net->tx && !strcmp(net->tx, "timer")) {
1294         n->vqs[0].tx_vq = virtio_add_queue(&n->vdev, 256,
1295                                            virtio_net_handle_tx_timer);
1296         n->vqs[0].tx_timer = qemu_new_timer_ns(vm_clock, virtio_net_tx_timer,
1297                                                &n->vqs[0]);
1298     } else {
1299         n->vqs[0].tx_vq = virtio_add_queue(&n->vdev, 256,
1300                                            virtio_net_handle_tx_bh);
1301         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1302     }
1303     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
1304     qemu_macaddr_default_if_unset(&conf->macaddr);
1305     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
1306     n->status = VIRTIO_NET_S_LINK_UP;
1307
1308     n->nic = qemu_new_nic(&net_virtio_info, conf, object_get_typename(OBJECT(dev)), dev->id, n);
1309     peer_test_vnet_hdr(n);
1310     if (peer_has_vnet_hdr(n)) {
1311         for (i = 0; i < n->max_queues; i++) {
1312             tap_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1313         }
1314         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1315     } else {
1316         n->host_hdr_len = 0;
1317     }
1318
1319     qemu_format_nic_info_str(qemu_get_queue(n->nic), conf->macaddr.a);
1320
1321     n->vqs[0].tx_waiting = 0;
1322     n->tx_burst = net->txburst;
1323     virtio_net_set_mrg_rx_bufs(n, 0);
1324     n->promisc = 1; /* for compatibility */
1325
1326     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1327
1328     n->vlans = g_malloc0(MAX_VLAN >> 3);
1329
1330     n->qdev = dev;
1331     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1332                     virtio_net_save, virtio_net_load, n);
1333
1334     add_boot_device_path(conf->bootindex, dev, "/ethernet-phy@0");
1335
1336     return &n->vdev;
1337 }
1338
1339 void virtio_net_exit(VirtIODevice *vdev)
1340 {
1341     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
1342     int i;
1343
1344     /* This will stop vhost backend if appropriate. */
1345     virtio_net_set_status(vdev, 0);
1346
1347     unregister_savevm(n->qdev, "virtio-net", n);
1348
1349     g_free(n->mac_table.macs);
1350     g_free(n->vlans);
1351
1352     for (i = 0; i < n->max_queues; i++) {
1353         VirtIONetQueue *q = &n->vqs[i];
1354         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1355
1356         qemu_purge_queued_packets(nc);
1357
1358         if (q->tx_timer) {
1359             qemu_del_timer(q->tx_timer);
1360             qemu_free_timer(q->tx_timer);
1361         } else {
1362             qemu_bh_delete(q->tx_bh);
1363         }
1364     }
1365
1366     g_free(n->vqs);
1367     qemu_del_nic(n->nic);
1368     virtio_cleanup(&n->vdev);
1369 }