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 typedef struct VirtIONetQueue {
  30     VirtQueue *rx_vq;
  31     VirtQueue *tx_vq;
  32     QEMUTimer *tx_timer;
  33     QEMUBH *tx_bh;
  34     int tx_waiting;
  35     struct {
  36         VirtQueueElement elem;
  37         ssize_t len;
  38     } async_tx;
  39     struct VirtIONet *n;
  40 } VirtIONetQueue;
  41
  42 typedef struct VirtIONet
  43 {
  44     VirtIODevice vdev;
  45     uint8_t mac[ETH_ALEN];
  46     uint16_t status;
  47     VirtIONetQueue *vqs;
  48     VirtQueue *ctrl_vq;
  49     NICState *nic;
  50     uint32_t tx_timeout;
  51     int32_t tx_burst;
  52     uint32_t has_vnet_hdr;
  53     size_t host_hdr_len;
  54     size_t guest_hdr_len;
  55     uint8_t has_ufo;
  56     int mergeable_rx_bufs;
  57     uint8_t promisc;
  58     uint8_t allmulti;
  59     uint8_t alluni;
  60     uint8_t nomulti;
  61     uint8_t nouni;
  62     uint8_t nobcast;
  63     uint8_t vhost_started;
  64     struct {
  65         int in_use;
  66         int first_multi;
  67         uint8_t multi_overflow;
  68         uint8_t uni_overflow;
  69         uint8_t *macs;
  70     } mac_table;
  71     uint32_t *vlans;
  72     DeviceState *qdev;
  73     int multiqueue;
  74     uint16_t max_queues;
  75     uint16_t curr_queues;
  76     size_t config_size;
  77 } VirtIONet;
  78
  79 /*
  80  * Calculate the number of bytes up to and including the given 'field' of
  81  * 'container'.
  82  */
  83 #define endof(container, field) \
  84     (offsetof(container, field) + sizeof(((container *)0)->field))
  85
  86 typedef struct VirtIOFeature {
  87     uint32_t flags;
  88     size_t end;
  89 } VirtIOFeature;
  90
  91 static VirtIOFeature feature_sizes[] = {
  92     {.flags = 1 << VIRTIO_NET_F_MAC,
  93      .end = endof(struct virtio_net_config, mac)},
  94     {.flags = 1 << VIRTIO_NET_F_STATUS,
  95      .end = endof(struct virtio_net_config, status)},
  96     {.flags = 1 << VIRTIO_NET_F_MQ,
  97      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
  98     {}
  99 };
 100
 101 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
 102 {
 103     VirtIONet *n = qemu_get_nic_opaque(nc);
 104
 105     return &n->vqs[nc->queue_index];
 106 }
 107
 108 static int vq2q(int queue_index)
 109 {
 110     return queue_index / 2;
 111 }
 112
 113 /* TODO
 114  * - we could suppress RX interrupt if we were so inclined.
 115  */
 116
 117 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
 118 {
 119     return (VirtIONet *)vdev;
 120 }
 121
 122 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
 123 {
 124     VirtIONet *n = to_virtio_net(vdev);
 125     struct virtio_net_config netcfg;
 126
 127     stw_p(&netcfg.status, n->status);
 128     stw_p(&netcfg.max_virtqueue_pairs, n->max_queues);
 129     memcpy(netcfg.mac, n->mac, ETH_ALEN);
 130     memcpy(config, &netcfg, n->config_size);
 131 }
 132
 133 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
 134 {
 135     VirtIONet *n = to_virtio_net(vdev);
 136     struct virtio_net_config netcfg = {};
 137
 138     memcpy(&netcfg, config, n->config_size);
 139
 140     if (!(n->vdev.guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) &&
 141         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
 142         memcpy(n->mac, netcfg.mac, ETH_ALEN);
 143         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
 144     }
 145 }
 146
 147 static bool virtio_net_started(VirtIONet *n, uint8_t status)
 148 {
 149     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 150         (n->status & VIRTIO_NET_S_LINK_UP) && n->vdev.vm_running;
 151 }
 152
 153 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
 154 {
 155     NetClientState *nc = qemu_get_queue(n->nic);
 156     int queues = n->multiqueue ? n->max_queues : 1;
 157
 158     if (!nc->peer) {
 159         return;
 160     }
 161     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 162         return;
 163     }
 164
 165     if (!tap_get_vhost_net(nc->peer)) {
 166         return;
 167     }
 168
 169     if (!!n->vhost_started == virtio_net_started(n, status) &&
 170                               !nc->peer->link_down) {
 171         return;
 172     }
 173     if (!n->vhost_started) {
 174         int r;
 175         if (!vhost_net_query(tap_get_vhost_net(nc->peer), &n->vdev)) {
 176             return;
 177         }
 178         n->vhost_started = 1;
 179         r = vhost_net_start(&n->vdev, n->nic->ncs, queues);
 180         if (r < 0) {
 181             error_report("unable to start vhost net: %d: "
 182                          "falling back on userspace virtio", -r);
 183             n->vhost_started = 0;
 184         }
 185     } else {
 186         vhost_net_stop(&n->vdev, n->nic->ncs, queues);
 187         n->vhost_started = 0;
 188     }
 189 }
 190
 191 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 192 {
 193     VirtIONet *n = to_virtio_net(vdev);
 194     VirtIONetQueue *q;
 195     int i;
 196     uint8_t queue_status;
 197
 198     virtio_net_vhost_status(n, status);
 199
 200     for (i = 0; i < n->max_queues; i++) {
 201         q = &n->vqs[i];
 202
 203         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
 204             queue_status = 0;
 205         } else {
 206             queue_status = status;
 207         }
 208
 209         if (!q->tx_waiting) {
 210             continue;
 211         }
 212
 213         if (virtio_net_started(n, queue_status) && !n->vhost_started) {
 214             if (q->tx_timer) {
 215                 qemu_mod_timer(q->tx_timer,
 216                                qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 217             } else {
 218                 qemu_bh_schedule(q->tx_bh);
 219             }
 220         } else {
 221             if (q->tx_timer) {
 222                 qemu_del_timer(q->tx_timer);
 223             } else {
 224                 qemu_bh_cancel(q->tx_bh);
 225             }
 226         }
 227     }
 228 }
 229
 230 static void virtio_net_set_link_status(NetClientState *nc)
 231 {
 232     VirtIONet *n = qemu_get_nic_opaque(nc);
 233     uint16_t old_status = n->status;
 234
 235     if (nc->link_down)
 236         n->status &= ~VIRTIO_NET_S_LINK_UP;
 237     else
 238         n->status |= VIRTIO_NET_S_LINK_UP;
 239
 240     if (n->status != old_status)
 241         virtio_notify_config(&n->vdev);
 242
 243     virtio_net_set_status(&n->vdev, n->vdev.status);
 244 }
 245
 246 static void virtio_net_reset(VirtIODevice *vdev)
 247 {
 248     VirtIONet *n = to_virtio_net(vdev);
 249
 250     /* Reset back to compatibility mode */
 251     n->promisc = 1;
 252     n->allmulti = 0;
 253     n->alluni = 0;
 254     n->nomulti = 0;
 255     n->nouni = 0;
 256     n->nobcast = 0;
 257     /* multiqueue is disabled by default */
 258     n->curr_queues = 1;
 259
 260     /* Flush any MAC and VLAN filter table state */
 261     n->mac_table.in_use = 0;
 262     n->mac_table.first_multi = 0;
 263     n->mac_table.multi_overflow = 0;
 264     n->mac_table.uni_overflow = 0;
 265     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 266     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
 267     memset(n->vlans, 0, MAX_VLAN >> 3);
 268 }
 269
 270 static void peer_test_vnet_hdr(VirtIONet *n)
 271 {
 272     NetClientState *nc = qemu_get_queue(n->nic);
 273     if (!nc->peer) {
 274         return;
 275     }
 276
 277     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 278         return;
 279     }
 280
 281     n->has_vnet_hdr = tap_has_vnet_hdr(nc->peer);
 282 }
 283
 284 static int peer_has_vnet_hdr(VirtIONet *n)
 285 {
 286     return n->has_vnet_hdr;
 287 }
 288
 289 static int peer_has_ufo(VirtIONet *n)
 290 {
 291     if (!peer_has_vnet_hdr(n))
 292         return 0;
 293
 294     n->has_ufo = tap_has_ufo(qemu_get_queue(n->nic)->peer);
 295
 296     return n->has_ufo;
 297 }
 298
 299 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
 300 {
 301     int i;
 302     NetClientState *nc;
 303
 304     n->mergeable_rx_bufs = mergeable_rx_bufs;
 305
 306     n->guest_hdr_len = n->mergeable_rx_bufs ?
 307         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 308
 309     for (i = 0; i < n->max_queues; i++) {
 310         nc = qemu_get_subqueue(n->nic, i);
 311
 312         if (peer_has_vnet_hdr(n) &&
 313             tap_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
 314             tap_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
 315             n->host_hdr_len = n->guest_hdr_len;
 316         }
 317     }
 318 }
 319
 320 static int peer_attach(VirtIONet *n, int index)
 321 {
 322     NetClientState *nc = qemu_get_subqueue(n->nic, index);
 323
 324     if (!nc->peer) {
 325         return 0;
 326     }
 327
 328     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 329         return 0;
 330     }
 331
 332     return tap_enable(nc->peer);
 333 }
 334
 335 static int peer_detach(VirtIONet *n, int index)
 336 {
 337     NetClientState *nc = qemu_get_subqueue(n->nic, index);
 338
 339     if (!nc->peer) {
 340         return 0;
 341     }
 342
 343     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
 344         return 0;
 345     }
 346
 347     return tap_disable(nc->peer);
 348 }
 349
 350 static void virtio_net_set_queues(VirtIONet *n)
 351 {
 352     int i;
 353
 354     for (i = 0; i < n->max_queues; i++) {
 355         if (i < n->curr_queues) {
 356             assert(!peer_attach(n, i));
 357         } else {
 358             assert(!peer_detach(n, i));
 359         }
 360     }
 361 }
 362
 363 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl);
 364
 365 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 366 {
 367     VirtIONet *n = to_virtio_net(vdev);
 368     NetClientState *nc = qemu_get_queue(n->nic);
 369
 370     features |= (1 << VIRTIO_NET_F_MAC);
 371
 372     if (!peer_has_vnet_hdr(n)) {
 373         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 374         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 375         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 376         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 377
 378         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 379         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 380         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 381         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 382     }
 383
 384     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 385         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 386         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 387     }
 388
 389     if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 390         return features;
 391     }
 392     if (!tap_get_vhost_net(nc->peer)) {
 393         return features;
 394     }
 395     return vhost_net_get_features(tap_get_vhost_net(nc->peer), features);
 396 }
 397
 398 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 399 {
 400     uint32_t features = 0;
 401
 402     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 403      * but also these: */
 404     features |= (1 << VIRTIO_NET_F_MAC);
 405     features |= (1 << VIRTIO_NET_F_CSUM);
 406     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 407     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 408     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 409
 410     return features;
 411 }
 412
 413 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 414 {
 415     VirtIONet *n = to_virtio_net(vdev);
 416     int i;
 417
 418     virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)),
 419                               !!(features & (1 << VIRTIO_NET_F_CTRL_VQ)));
 420
 421     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
 422
 423     if (n->has_vnet_hdr) {
 424         tap_set_offload(qemu_get_subqueue(n->nic, 0)->peer,
 425                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 426                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 427                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 428                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 429                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 430     }
 431
 432     for (i = 0;  i < n->max_queues; i++) {
 433         NetClientState *nc = qemu_get_subqueue(n->nic, i);
 434
 435         if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 436             continue;
 437         }
 438         if (!tap_get_vhost_net(nc->peer)) {
 439             continue;
 440         }
 441         vhost_net_ack_features(tap_get_vhost_net(nc->peer), features);
 442     }
 443 }
 444
 445 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 446                                      struct iovec *iov, unsigned int iov_cnt)
 447 {
 448     uint8_t on;
 449     size_t s;
 450
 451     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
 452     if (s != sizeof(on)) {
 453         return VIRTIO_NET_ERR;
 454     }
 455
 456     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
 457         n->promisc = on;
 458     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
 459         n->allmulti = on;
 460     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
 461         n->alluni = on;
 462     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
 463         n->nomulti = on;
 464     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
 465         n->nouni = on;
 466     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
 467         n->nobcast = on;
 468     } else {
 469         return VIRTIO_NET_ERR;
 470     }
 471
 472     return VIRTIO_NET_OK;
 473 }
 474
 475 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 476                                  struct iovec *iov, unsigned int iov_cnt)
 477 {
 478     struct virtio_net_ctrl_mac mac_data;
 479     size_t s;
 480
 481     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
 482         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
 483             return VIRTIO_NET_ERR;
 484         }
 485         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
 486         assert(s == sizeof(n->mac));
 487         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
 488         return VIRTIO_NET_OK;
 489     }
 490
 491     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
 492         return VIRTIO_NET_ERR;
 493     }
 494
 495     n->mac_table.in_use = 0;
 496     n->mac_table.first_multi = 0;
 497     n->mac_table.uni_overflow = 0;
 498     n->mac_table.multi_overflow = 0;
 499     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 500
 501     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 502                    sizeof(mac_data.entries));
 503     mac_data.entries = ldl_p(&mac_data.entries);
 504     if (s != sizeof(mac_data.entries)) {
 505         return VIRTIO_NET_ERR;
 506     }
 507     iov_discard_front(&iov, &iov_cnt, s);
 508
 509     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
 510         return VIRTIO_NET_ERR;
 511     }
 512
 513     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 514         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 515                        mac_data.entries * ETH_ALEN);
 516         if (s != mac_data.entries * ETH_ALEN) {
 517             return VIRTIO_NET_ERR;
 518         }
 519         n->mac_table.in_use += mac_data.entries;
 520     } else {
 521         n->mac_table.uni_overflow = 1;
 522     }
 523
 524     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
 525
 526     n->mac_table.first_multi = n->mac_table.in_use;
 527
 528     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 529                    sizeof(mac_data.entries));
 530     mac_data.entries = ldl_p(&mac_data.entries);
 531     if (s != sizeof(mac_data.entries)) {
 532         return VIRTIO_NET_ERR;
 533     }
 534
 535     iov_discard_front(&iov, &iov_cnt, s);
 536
 537     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
 538         return VIRTIO_NET_ERR;
 539     }
 540
 541     if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 542         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 543                        mac_data.entries * ETH_ALEN);
 544         if (s != mac_data.entries * ETH_ALEN) {
 545             return VIRTIO_NET_ERR;
 546         }
 547         n->mac_table.in_use += mac_data.entries;
 548     } else {
 549         n->mac_table.multi_overflow = 1;
 550     }
 551
 552     return VIRTIO_NET_OK;
 553 }
 554
 555 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 556                                         struct iovec *iov, unsigned int iov_cnt)
 557 {
 558     uint16_t vid;
 559     size_t s;
 560
 561     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
 562     vid = lduw_p(&vid);
 563     if (s != sizeof(vid)) {
 564         return VIRTIO_NET_ERR;
 565     }
 566
 567     if (vid >= MAX_VLAN)
 568         return VIRTIO_NET_ERR;
 569
 570     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 571         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 572     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 573         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 574     else
 575         return VIRTIO_NET_ERR;
 576
 577     return VIRTIO_NET_OK;
 578 }
 579
 580 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
 581                                 VirtQueueElement *elem)
 582 {
 583     struct virtio_net_ctrl_mq s;
 584
 585     if (elem->out_num != 2 ||
 586         elem->out_sg[1].iov_len != sizeof(struct virtio_net_ctrl_mq)) {
 587         error_report("virtio-net ctrl invalid steering command");
 588         return VIRTIO_NET_ERR;
 589     }
 590
 591     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 592         return VIRTIO_NET_ERR;
 593     }
 594
 595     memcpy(&s, elem->out_sg[1].iov_base, sizeof(struct virtio_net_ctrl_mq));
 596
 597     if (s.virtqueue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
 598         s.virtqueue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
 599         s.virtqueue_pairs > n->max_queues ||
 600         !n->multiqueue) {
 601         return VIRTIO_NET_ERR;
 602     }
 603
 604     n->curr_queues = s.virtqueue_pairs;
 605     /* stop the backend before changing the number of queues to avoid handling a
 606      * disabled queue */
 607     virtio_net_set_status(&n->vdev, n->vdev.status);
 608     virtio_net_set_queues(n);
 609
 610     return VIRTIO_NET_OK;
 611 }
 612 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 613 {
 614     VirtIONet *n = to_virtio_net(vdev);
 615     struct virtio_net_ctrl_hdr ctrl;
 616     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 617     VirtQueueElement elem;
 618     size_t s;
 619     struct iovec *iov;
 620     unsigned int iov_cnt;
 621
 622     while (virtqueue_pop(vq, &elem)) {
 623         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
 624             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
 625             error_report("virtio-net ctrl missing headers");
 626             exit(1);
 627         }
 628
 629         iov = elem.out_sg;
 630         iov_cnt = elem.out_num;
 631         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
 632         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
 633         if (s != sizeof(ctrl)) {
 634             status = VIRTIO_NET_ERR;
 635         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
 636             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
 637         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
 638             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
 639         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
 640             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
 641         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
 642             status = virtio_net_handle_mq(n, ctrl.cmd, &elem);
 643         }
 644
 645         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
 646         assert(s == sizeof(status));
 647
 648         virtqueue_push(vq, &elem, sizeof(status));
 649         virtio_notify(vdev, vq);
 650     }
 651 }
 652
 653 /* RX */
 654
 655 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 656 {
 657     VirtIONet *n = to_virtio_net(vdev);
 658     int queue_index = vq2q(virtio_get_queue_index(vq));
 659
 660     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
 661 }
 662
 663 static int virtio_net_can_receive(NetClientState *nc)
 664 {
 665     VirtIONet *n = qemu_get_nic_opaque(nc);
 666     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 667
 668     if (!n->vdev.vm_running) {
 669         return 0;
 670     }
 671
 672     if (nc->queue_index >= n->curr_queues) {
 673         return 0;
 674     }
 675
 676     if (!virtio_queue_ready(q->rx_vq) ||
 677         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 678         return 0;
 679     }
 680
 681     return 1;
 682 }
 683
 684 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
 685 {
 686     VirtIONet *n = q->n;
 687     if (virtio_queue_empty(q->rx_vq) ||
 688         (n->mergeable_rx_bufs &&
 689          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
 690         virtio_queue_set_notification(q->rx_vq, 1);
 691
 692         /* To avoid a race condition where the guest has made some buffers
 693          * available after the above check but before notification was
 694          * enabled, check for available buffers again.
 695          */
 696         if (virtio_queue_empty(q->rx_vq) ||
 697             (n->mergeable_rx_bufs &&
 698              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
 699             return 0;
 700         }
 701     }
 702
 703     virtio_queue_set_notification(q->rx_vq, 0);
 704     return 1;
 705 }
 706
 707 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 708  * it never finds out that the packets don't have valid checksums.  This
 709  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 710  * fix this with Xen but it hasn't appeared in an upstream release of
 711  * dhclient yet.
 712  *
 713  * To avoid breaking existing guests, we catch udp packets and add
 714  * checksums.  This is terrible but it's better than hacking the guest
 715  * kernels.
 716  *
 717  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 718  * we should provide a mechanism to disable it to avoid polluting the host
 719  * cache.
 720  */
 721 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 722                                         uint8_t *buf, size_t size)
 723 {
 724     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 725         (size > 27 && size < 1500) && /* normal sized MTU */
 726         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 727         (buf[23] == 17) && /* ip.protocol == UDP */
 728         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 729         net_checksum_calculate(buf, size);
 730         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 731     }
 732 }
 733
 734 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
 735                            const void *buf, size_t size)
 736 {
 737     if (n->has_vnet_hdr) {
 738         /* FIXME this cast is evil */
 739         void *wbuf = (void *)buf;
 740         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
 741                                     size - n->host_hdr_len);
 742         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
 743     } else {
 744         struct virtio_net_hdr hdr = {
 745             .flags = 0,
 746             .gso_type = VIRTIO_NET_HDR_GSO_NONE
 747         };
 748         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
 749     }
 750 }
 751
 752 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 753 {
 754     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 755     static const uint8_t vlan[] = {0x81, 0x00};
 756     uint8_t *ptr = (uint8_t *)buf;
 757     int i;
 758
 759     if (n->promisc)
 760         return 1;
 761
 762     ptr += n->host_hdr_len;
 763
 764     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 765         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 766         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 767             return 0;
 768     }
 769
 770     if (ptr[0] & 1) { // multicast
 771         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 772             return !n->nobcast;
 773         } else if (n->nomulti) {
 774             return 0;
 775         } else if (n->allmulti || n->mac_table.multi_overflow) {
 776             return 1;
 777         }
 778
 779         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 780             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 781                 return 1;
 782             }
 783         }
 784     } else { // unicast
 785         if (n->nouni) {
 786             return 0;
 787         } else if (n->alluni || n->mac_table.uni_overflow) {
 788             return 1;
 789         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 790             return 1;
 791         }
 792
 793         for (i = 0; i < n->mac_table.first_multi; i++) {
 794             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 795                 return 1;
 796             }
 797         }
 798     }
 799
 800     return 0;
 801 }
 802
 803 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 804 {
 805     VirtIONet *n = qemu_get_nic_opaque(nc);
 806     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 807     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
 808     struct virtio_net_hdr_mrg_rxbuf mhdr;
 809     unsigned mhdr_cnt = 0;
 810     size_t offset, i, guest_offset;
 811
 812     if (!virtio_net_can_receive(nc)) {
 813         return -1;
 814     }
 815
 816     /* hdr_len refers to the header we supply to the guest */
 817     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
 818         return 0;
 819     }
 820
 821     if (!receive_filter(n, buf, size))
 822         return size;
 823
 824     offset = i = 0;
 825
 826     while (offset < size) {
 827         VirtQueueElement elem;
 828         int len, total;
 829         const struct iovec *sg = elem.in_sg;
 830
 831         total = 0;
 832
 833         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
 834             if (i == 0)
 835                 return -1;
 836             error_report("virtio-net unexpected empty queue: "
 837                     "i %zd mergeable %d offset %zd, size %zd, "
 838                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 839                     i, n->mergeable_rx_bufs, offset, size,
 840                     n->guest_hdr_len, n->host_hdr_len, n->vdev.guest_features);
 841             exit(1);
 842         }
 843
 844         if (elem.in_num < 1) {
 845             error_report("virtio-net receive queue contains no in buffers");
 846             exit(1);
 847         }
 848
 849         if (i == 0) {
 850             assert(offset == 0);
 851             if (n->mergeable_rx_bufs) {
 852                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
 853                                     sg, elem.in_num,
 854                                     offsetof(typeof(mhdr), num_buffers),
 855                                     sizeof(mhdr.num_buffers));
 856             }
 857
 858             receive_header(n, sg, elem.in_num, buf, size);
 859             offset = n->host_hdr_len;
 860             total += n->guest_hdr_len;
 861             guest_offset = n->guest_hdr_len;
 862         } else {
 863             guest_offset = 0;
 864         }
 865
 866         /* copy in packet.  ugh */
 867         len = iov_from_buf(sg, elem.in_num, guest_offset,
 868                            buf + offset, size - offset);
 869         total += len;
 870         offset += len;
 871         /* If buffers can't be merged, at this point we
 872          * must have consumed the complete packet.
 873          * Otherwise, drop it. */
 874         if (!n->mergeable_rx_bufs && offset < size) {
 875 #if 0
 876             error_report("virtio-net truncated non-mergeable packet: "
 877                          "i %zd mergeable %d offset %zd, size %zd, "
 878                          "guest hdr len %zd, host hdr len %zd",
 879                          i, n->mergeable_rx_bufs,
 880                          offset, size, n->guest_hdr_len, n->host_hdr_len);
 881 #endif
 882             return size;
 883         }
 884
 885         /* signal other side */
 886         virtqueue_fill(q->rx_vq, &elem, total, i++);
 887     }
 888
 889     if (mhdr_cnt) {
 890         stw_p(&mhdr.num_buffers, i);
 891         iov_from_buf(mhdr_sg, mhdr_cnt,
 892                      0,
 893                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
 894     }
 895
 896     virtqueue_flush(q->rx_vq, i);
 897     virtio_notify(&n->vdev, q->rx_vq);
 898
 899     return size;
 900 }
 901
 902 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
 903
 904 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
 905 {
 906     VirtIONet *n = qemu_get_nic_opaque(nc);
 907     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 908
 909     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
 910     virtio_notify(&n->vdev, q->tx_vq);
 911
 912     q->async_tx.elem.out_num = q->async_tx.len = 0;
 913
 914     virtio_queue_set_notification(q->tx_vq, 1);
 915     virtio_net_flush_tx(q);
 916 }
 917
 918 /* TX */
 919 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
 920 {
 921     VirtIONet *n = q->n;
 922     VirtQueueElement elem;
 923     int32_t num_packets = 0;
 924     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
 925     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 926         return num_packets;
 927     }
 928
 929     assert(n->vdev.vm_running);
 930
 931     if (q->async_tx.elem.out_num) {
 932         virtio_queue_set_notification(q->tx_vq, 0);
 933         return num_packets;
 934     }
 935
 936     while (virtqueue_pop(q->tx_vq, &elem)) {
 937         ssize_t ret, len;
 938         unsigned int out_num = elem.out_num;
 939         struct iovec *out_sg = &elem.out_sg[0];
 940         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 941
 942         if (out_num < 1) {
 943             error_report("virtio-net header not in first element");
 944             exit(1);
 945         }
 946
 947         /*
 948          * If host wants to see the guest header as is, we can
 949          * pass it on unchanged. Otherwise, copy just the parts
 950          * that host is interested in.
 951          */
 952         assert(n->host_hdr_len <= n->guest_hdr_len);
 953         if (n->host_hdr_len != n->guest_hdr_len) {
 954             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
 955                                        out_sg, out_num,
 956                                        0, n->host_hdr_len);
 957             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
 958                              out_sg, out_num,
 959                              n->guest_hdr_len, -1);
 960             out_num = sg_num;
 961             out_sg = sg;
 962         }
 963
 964         len = n->guest_hdr_len;
 965
 966         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
 967                                       out_sg, out_num, virtio_net_tx_complete);
 968         if (ret == 0) {
 969             virtio_queue_set_notification(q->tx_vq, 0);
 970             q->async_tx.elem = elem;
 971             q->async_tx.len  = len;
 972             return -EBUSY;
 973         }
 974
 975         len += ret;
 976
 977         virtqueue_push(q->tx_vq, &elem, 0);
 978         virtio_notify(&n->vdev, q->tx_vq);
 979
 980         if (++num_packets >= n->tx_burst) {
 981             break;
 982         }
 983     }
 984     return num_packets;
 985 }
 986
 987 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 988 {
 989     VirtIONet *n = to_virtio_net(vdev);
 990     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
 991
 992     /* This happens when device was stopped but VCPU wasn't. */
 993     if (!n->vdev.vm_running) {
 994         q->tx_waiting = 1;
 995         return;
 996     }
 997
 998     if (q->tx_waiting) {
 999         virtio_queue_set_notification(vq, 1);
1000         qemu_del_timer(q->tx_timer);
1001         q->tx_waiting = 0;
1002         virtio_net_flush_tx(q);
1003     } else {
1004         qemu_mod_timer(q->tx_timer,
1005                        qemu_get_clock_ns(vm_clock) + n->tx_timeout);
1006         q->tx_waiting = 1;
1007         virtio_queue_set_notification(vq, 0);
1008     }
1009 }
1010
1011 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1012 {
1013     VirtIONet *n = to_virtio_net(vdev);
1014     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1015
1016     if (unlikely(q->tx_waiting)) {
1017         return;
1018     }
1019     q->tx_waiting = 1;
1020     /* This happens when device was stopped but VCPU wasn't. */
1021     if (!n->vdev.vm_running) {
1022         return;
1023     }
1024     virtio_queue_set_notification(vq, 0);
1025     qemu_bh_schedule(q->tx_bh);
1026 }
1027
1028 static void virtio_net_tx_timer(void *opaque)
1029 {
1030     VirtIONetQueue *q = opaque;
1031     VirtIONet *n = q->n;
1032     assert(n->vdev.vm_running);
1033
1034     q->tx_waiting = 0;
1035
1036     /* Just in case the driver is not ready on more */
1037     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
1038         return;
1039
1040     virtio_queue_set_notification(q->tx_vq, 1);
1041     virtio_net_flush_tx(q);
1042 }
1043
1044 static void virtio_net_tx_bh(void *opaque)
1045 {
1046     VirtIONetQueue *q = opaque;
1047     VirtIONet *n = q->n;
1048     int32_t ret;
1049
1050     assert(n->vdev.vm_running);
1051
1052     q->tx_waiting = 0;
1053
1054     /* Just in case the driver is not ready on more */
1055     if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
1056         return;
1057
1058     ret = virtio_net_flush_tx(q);
1059     if (ret == -EBUSY) {
1060         return; /* Notification re-enable handled by tx_complete */
1061     }
1062
1063     /* If we flush a full burst of packets, assume there are
1064      * more coming and immediately reschedule */
1065     if (ret >= n->tx_burst) {
1066         qemu_bh_schedule(q->tx_bh);
1067         q->tx_waiting = 1;
1068         return;
1069     }
1070
1071     /* If less than a full burst, re-enable notification and flush
1072      * anything that may have come in while we weren't looking.  If
1073      * we find something, assume the guest is still active and reschedule */
1074     virtio_queue_set_notification(q->tx_vq, 1);
1075     if (virtio_net_flush_tx(q) > 0) {
1076         virtio_queue_set_notification(q->tx_vq, 0);
1077         qemu_bh_schedule(q->tx_bh);
1078         q->tx_waiting = 1;
1079     }
1080 }
1081
1082 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
1083 {
1084     VirtIODevice *vdev = &n->vdev;
1085     int i, max = multiqueue ? n->max_queues : 1;
1086
1087     n->multiqueue = multiqueue;
1088
1089     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1090         virtio_del_queue(vdev, i);
1091     }
1092
1093     for (i = 1; i < max; i++) {
1094         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1095         if (n->vqs[i].tx_timer) {
1096             n->vqs[i].tx_vq =
1097                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1098             n->vqs[i].tx_timer = qemu_new_timer_ns(vm_clock,
1099                                                    virtio_net_tx_timer,
1100                                                    &n->vqs[i]);
1101         } else {
1102             n->vqs[i].tx_vq =
1103                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1104             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1105         }
1106
1107         n->vqs[i].tx_waiting = 0;
1108         n->vqs[i].n = n;
1109     }
1110
1111     if (ctrl) {
1112         n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1113     }
1114
1115     virtio_net_set_queues(n);
1116 }
1117
1118 static void virtio_net_save(QEMUFile *f, void *opaque)
1119 {
1120     int i;
1121     VirtIONet *n = opaque;
1122
1123     /* At this point, backend must be stopped, otherwise
1124      * it might keep writing to memory. */
1125     assert(!n->vhost_started);
1126     virtio_save(&n->vdev, f);
1127
1128     qemu_put_buffer(f, n->mac, ETH_ALEN);
1129     qemu_put_be32(f, n->vqs[0].tx_waiting);
1130     qemu_put_be32(f, n->mergeable_rx_bufs);
1131     qemu_put_be16(f, n->status);
1132     qemu_put_byte(f, n->promisc);
1133     qemu_put_byte(f, n->allmulti);
1134     qemu_put_be32(f, n->mac_table.in_use);
1135     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1136     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1137     qemu_put_be32(f, n->has_vnet_hdr);
1138     qemu_put_byte(f, n->mac_table.multi_overflow);
1139     qemu_put_byte(f, n->mac_table.uni_overflow);
1140     qemu_put_byte(f, n->alluni);
1141     qemu_put_byte(f, n->nomulti);
1142     qemu_put_byte(f, n->nouni);
1143     qemu_put_byte(f, n->nobcast);
1144     qemu_put_byte(f, n->has_ufo);
1145     if (n->max_queues > 1) {
1146         qemu_put_be16(f, n->max_queues);
1147         qemu_put_be16(f, n->curr_queues);
1148         for (i = 1; i < n->curr_queues; i++) {
1149             qemu_put_be32(f, n->vqs[i].tx_waiting);
1150         }
1151     }
1152 }
1153
1154 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1155 {
1156     VirtIONet *n = opaque;
1157     int ret, i, link_down;
1158
1159     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1160         return -EINVAL;
1161
1162     ret = virtio_load(&n->vdev, f);
1163     if (ret) {
1164         return ret;
1165     }
1166
1167     qemu_get_buffer(f, n->mac, ETH_ALEN);
1168     n->vqs[0].tx_waiting = qemu_get_be32(f);
1169
1170     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1171
1172     if (version_id >= 3)
1173         n->status = qemu_get_be16(f);
1174
1175     if (version_id >= 4) {
1176         if (version_id < 8) {
1177             n->promisc = qemu_get_be32(f);
1178             n->allmulti = qemu_get_be32(f);
1179         } else {
1180             n->promisc = qemu_get_byte(f);
1181             n->allmulti = qemu_get_byte(f);
1182         }
1183     }
1184
1185     if (version_id >= 5) {
1186         n->mac_table.in_use = qemu_get_be32(f);
1187         /* MAC_TABLE_ENTRIES may be different from the saved image */
1188         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1189             qemu_get_buffer(f, n->mac_table.macs,
1190                             n->mac_table.in_use * ETH_ALEN);
1191         } else if (n->mac_table.in_use) {
1192             uint8_t *buf = g_malloc0(n->mac_table.in_use);
1193             qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
1194             g_free(buf);
1195             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1196             n->mac_table.in_use = 0;
1197         }
1198     }
1199
1200     if (version_id >= 6)
1201         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1202
1203     if (version_id >= 7) {
1204         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1205             error_report("virtio-net: saved image requires vnet_hdr=on");
1206             return -1;
1207         }
1208
1209         if (n->has_vnet_hdr) {
1210             tap_set_offload(qemu_get_queue(n->nic)->peer,
1211                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
1212                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
1213                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
1214                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
1215                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
1216         }
1217     }
1218
1219     if (version_id >= 9) {
1220         n->mac_table.multi_overflow = qemu_get_byte(f);
1221         n->mac_table.uni_overflow = qemu_get_byte(f);
1222     }
1223
1224     if (version_id >= 10) {
1225         n->alluni = qemu_get_byte(f);
1226         n->nomulti = qemu_get_byte(f);
1227         n->nouni = qemu_get_byte(f);
1228         n->nobcast = qemu_get_byte(f);
1229     }
1230
1231     if (version_id >= 11) {
1232         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1233             error_report("virtio-net: saved image requires TUN_F_UFO support");
1234             return -1;
1235         }
1236     }
1237
1238     if (n->max_queues > 1) {
1239         if (n->max_queues != qemu_get_be16(f)) {
1240             error_report("virtio-net: different max_queues ");
1241             return -1;
1242         }
1243
1244         n->curr_queues = qemu_get_be16(f);
1245         for (i = 1; i < n->curr_queues; i++) {
1246             n->vqs[i].tx_waiting = qemu_get_be32(f);
1247         }
1248     }
1249
1250     virtio_net_set_queues(n);
1251
1252     /* Find the first multicast entry in the saved MAC filter */
1253     for (i = 0; i < n->mac_table.in_use; i++) {
1254         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1255             break;
1256         }
1257     }
1258     n->mac_table.first_multi = i;
1259
1260     /* nc.link_down can't be migrated, so infer link_down according
1261      * to link status bit in n->status */
1262     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1263     for (i = 0; i < n->max_queues; i++) {
1264         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1265     }
1266
1267     return 0;
1268 }
1269
1270 static void virtio_net_cleanup(NetClientState *nc)
1271 {
1272     VirtIONet *n = qemu_get_nic_opaque(nc);
1273
1274     n->nic = NULL;
1275 }
1276
1277 static NetClientInfo net_virtio_info = {
1278     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1279     .size = sizeof(NICState),
1280     .can_receive = virtio_net_can_receive,
1281     .receive = virtio_net_receive,
1282         .cleanup = virtio_net_cleanup,
1283     .link_status_changed = virtio_net_set_link_status,
1284 };
1285
1286 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1287 {
1288     VirtIONet *n = to_virtio_net(vdev);
1289     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1290     assert(n->vhost_started);
1291     return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1292 }
1293
1294 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1295                                            bool mask)
1296 {
1297     VirtIONet *n = to_virtio_net(vdev);
1298     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1299     assert(n->vhost_started);
1300     vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1301                              vdev, idx, mask);
1302 }
1303
1304 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
1305                               virtio_net_conf *net, uint32_t host_features)
1306 {
1307     VirtIONet *n;
1308     int i, config_size = 0;
1309
1310     for (i = 0; feature_sizes[i].flags != 0; i++) {
1311         if (host_features & feature_sizes[i].flags) {
1312             config_size = MAX(feature_sizes[i].end, config_size);
1313         }
1314     }
1315
1316     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
1317                                         config_size, sizeof(VirtIONet));
1318
1319     n->config_size = config_size;
1320     n->vdev.get_config = virtio_net_get_config;
1321     n->vdev.set_config = virtio_net_set_config;
1322     n->vdev.get_features = virtio_net_get_features;
1323     n->vdev.set_features = virtio_net_set_features;
1324     n->vdev.bad_features = virtio_net_bad_features;
1325     n->vdev.reset = virtio_net_reset;
1326     n->vdev.set_status = virtio_net_set_status;
1327     n->vdev.guest_notifier_mask = virtio_net_guest_notifier_mask;
1328     n->vdev.guest_notifier_pending = virtio_net_guest_notifier_pending;
1329     n->max_queues = MAX(conf->queues, 1);
1330     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1331     n->vqs[0].rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
1332     n->curr_queues = 1;
1333     n->vqs[0].n = n;
1334     n->tx_timeout = net->txtimer;
1335
1336     if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
1337         error_report("virtio-net: "
1338                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1339                      net->tx);
1340         error_report("Defaulting to \"bh\"");
1341     }
1342
1343     if (net->tx && !strcmp(net->tx, "timer")) {
1344         n->vqs[0].tx_vq = virtio_add_queue(&n->vdev, 256,
1345                                            virtio_net_handle_tx_timer);
1346         n->vqs[0].tx_timer = qemu_new_timer_ns(vm_clock, virtio_net_tx_timer,
1347                                                &n->vqs[0]);
1348     } else {
1349         n->vqs[0].tx_vq = virtio_add_queue(&n->vdev, 256,
1350                                            virtio_net_handle_tx_bh);
1351         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1352     }
1353     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
1354     qemu_macaddr_default_if_unset(&conf->macaddr);
1355     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
1356     n->status = VIRTIO_NET_S_LINK_UP;
1357
1358     n->nic = qemu_new_nic(&net_virtio_info, conf, object_get_typename(OBJECT(dev)), dev->id, n);
1359     peer_test_vnet_hdr(n);
1360     if (peer_has_vnet_hdr(n)) {
1361         for (i = 0; i < n->max_queues; i++) {
1362             tap_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1363         }
1364         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1365     } else {
1366         n->host_hdr_len = 0;
1367     }
1368
1369     qemu_format_nic_info_str(qemu_get_queue(n->nic), conf->macaddr.a);
1370
1371     n->vqs[0].tx_waiting = 0;
1372     n->tx_burst = net->txburst;
1373     virtio_net_set_mrg_rx_bufs(n, 0);
1374     n->promisc = 1; /* for compatibility */
1375
1376     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1377
1378     n->vlans = g_malloc0(MAX_VLAN >> 3);
1379
1380     n->qdev = dev;
1381     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1382                     virtio_net_save, virtio_net_load, n);
1383
1384     add_boot_device_path(conf->bootindex, dev, "/ethernet-phy@0");
1385
1386     return &n->vdev;
1387 }
1388
1389 void virtio_net_exit(VirtIODevice *vdev)
1390 {
1391     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
1392     int i;
1393
1394     /* This will stop vhost backend if appropriate. */
1395     virtio_net_set_status(vdev, 0);
1396
1397     unregister_savevm(n->qdev, "virtio-net", n);
1398
1399     g_free(n->mac_table.macs);
1400     g_free(n->vlans);
1401
1402     for (i = 0; i < n->max_queues; i++) {
1403         VirtIONetQueue *q = &n->vqs[i];
1404         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1405
1406         qemu_purge_queued_packets(nc);
1407
1408         if (q->tx_timer) {
1409             qemu_del_timer(q->tx_timer);
1410             qemu_free_timer(q->tx_timer);
1411         } else {
1412             qemu_bh_delete(q->tx_bh);
1413         }
1414     }
1415
1416     g_free(n->vqs);
1417     qemu_del_nic(n->nic);
1418     virtio_cleanup(&n->vdev);
1419 }