hw/net/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/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/virtio-net.h"
  22 #include "net/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                                 struct iovec *iov, unsigned int iov_cnt)
 532 {
 533     struct virtio_net_ctrl_mq mq;
 534     size_t s;
 535     uint16_t queues;
 536
 537     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
 538     if (s != sizeof(mq)) {
 539         return VIRTIO_NET_ERR;
 540     }
 541
 542     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 543         return VIRTIO_NET_ERR;
 544     }
 545
 546     queues = lduw_p(&mq.virtqueue_pairs);
 547
 548     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
 549         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
 550         queues > n->max_queues ||
 551         !n->multiqueue) {
 552         return VIRTIO_NET_ERR;
 553     }
 554
 555     n->curr_queues = queues;
 556     /* stop the backend before changing the number of queues to avoid handling a
 557      * disabled queue */
 558     virtio_net_set_status(&n->vdev, n->vdev.status);
 559     virtio_net_set_queues(n);
 560
 561     return VIRTIO_NET_OK;
 562 }
 563 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 564 {
 565     VirtIONet *n = to_virtio_net(vdev);
 566     struct virtio_net_ctrl_hdr ctrl;
 567     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 568     VirtQueueElement elem;
 569     size_t s;
 570     struct iovec *iov;
 571     unsigned int iov_cnt;
 572
 573     while (virtqueue_pop(vq, &elem)) {
 574         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
 575             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
 576             error_report("virtio-net ctrl missing headers");
 577             exit(1);
 578         }
 579
 580         iov = elem.out_sg;
 581         iov_cnt = elem.out_num;
 582         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
 583         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
 584         if (s != sizeof(ctrl)) {
 585             status = VIRTIO_NET_ERR;
 586         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
 587             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
 588         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
 589             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
 590         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
 591             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
 592         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
 593             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
 594         }
 595
 596         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
 597         assert(s == sizeof(status));
 598
 599         virtqueue_push(vq, &elem, sizeof(status));
 600         virtio_notify(vdev, vq);
 601     }
 602 }
 603
 604 /* RX */
 605
 606 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 607 {
 608     VirtIONet *n = to_virtio_net(vdev);
 609     int queue_index = vq2q(virtio_get_queue_index(vq));
 610
 611     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
 612 }
 613
 614 static int virtio_net_can_receive(NetClientState *nc)
 615 {
 616     VirtIONet *n = qemu_get_nic_opaque(nc);
 617     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 618
 619     if (!n->vdev.vm_running) {
 620         return 0;
 621     }
 622
 623     if (nc->queue_index >= n->curr_queues) {
 624         return 0;
 625     }
 626
 627     if (!virtio_queue_ready(q->rx_vq) ||
 628         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 629         return 0;
 630     }
 631
 632     return 1;
 633 }
 634
 635 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
 636 {
 637     VirtIONet *n = q->n;
 638     if (virtio_queue_empty(q->rx_vq) ||
 639         (n->mergeable_rx_bufs &&
 640          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
 641         virtio_queue_set_notification(q->rx_vq, 1);
 642
 643         /* To avoid a race condition where the guest has made some buffers
 644          * available after the above check but before notification was
 645          * enabled, check for available buffers again.
 646          */
 647         if (virtio_queue_empty(q->rx_vq) ||
 648             (n->mergeable_rx_bufs &&
 649              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
 650             return 0;
 651         }
 652     }
 653
 654     virtio_queue_set_notification(q->rx_vq, 0);
 655     return 1;
 656 }
 657
 658 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 659  * it never finds out that the packets don't have valid checksums.  This
 660  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 661  * fix this with Xen but it hasn't appeared in an upstream release of
 662  * dhclient yet.
 663  *
 664  * To avoid breaking existing guests, we catch udp packets and add
 665  * checksums.  This is terrible but it's better than hacking the guest
 666  * kernels.
 667  *
 668  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 669  * we should provide a mechanism to disable it to avoid polluting the host
 670  * cache.
 671  */
 672 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 673                                         uint8_t *buf, size_t size)
 674 {
 675     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 676         (size > 27 && size < 1500) && /* normal sized MTU */
 677         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 678         (buf[23] == 17) && /* ip.protocol == UDP */
 679         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 680         net_checksum_calculate(buf, size);
 681         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 682     }
 683 }
 684
 685 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
 686                            const void *buf, size_t size)
 687 {
 688     if (n->has_vnet_hdr) {
 689         /* FIXME this cast is evil */
 690         void *wbuf = (void *)buf;
 691         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
 692                                     size - n->host_hdr_len);
 693         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
 694     } else {
 695         struct virtio_net_hdr hdr = {
 696             .flags = 0,
 697             .gso_type = VIRTIO_NET_HDR_GSO_NONE
 698         };
 699         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
 700     }
 701 }
 702
 703 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 704 {
 705     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 706     static const uint8_t vlan[] = {0x81, 0x00};
 707     uint8_t *ptr = (uint8_t *)buf;
 708     int i;
 709
 710     if (n->promisc)
 711         return 1;
 712
 713     ptr += n->host_hdr_len;
 714
 715     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 716         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 717         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 718             return 0;
 719     }
 720
 721     if (ptr[0] & 1) { // multicast
 722         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 723             return !n->nobcast;
 724         } else if (n->nomulti) {
 725             return 0;
 726         } else if (n->allmulti || n->mac_table.multi_overflow) {
 727             return 1;
 728         }
 729
 730         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 731             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 732                 return 1;
 733             }
 734         }
 735     } else { // unicast
 736         if (n->nouni) {
 737             return 0;
 738         } else if (n->alluni || n->mac_table.uni_overflow) {
 739             return 1;
 740         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 741             return 1;
 742         }
 743
 744         for (i = 0; i < n->mac_table.first_multi; i++) {
 745             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 746                 return 1;
 747             }
 748         }
 749     }
 750
 751     return 0;
 752 }
 753
 754 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 755 {
 756     VirtIONet *n = qemu_get_nic_opaque(nc);
 757     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 758     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
 759     struct virtio_net_hdr_mrg_rxbuf mhdr;
 760     unsigned mhdr_cnt = 0;
 761     size_t offset, i, guest_offset;
 762
 763     if (!virtio_net_can_receive(nc)) {
 764         return -1;
 765     }
 766
 767     /* hdr_len refers to the header we supply to the guest */
 768     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
 769         return 0;
 770     }
 771
 772     if (!receive_filter(n, buf, size))
 773         return size;
 774
 775     offset = i = 0;
 776
 777     while (offset < size) {
 778         VirtQueueElement elem;
 779         int len, total;
 780         const struct iovec *sg = elem.in_sg;
 781
 782         total = 0;
 783
 784         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
 785             if (i == 0)
 786                 return -1;
 787             error_report("virtio-net unexpected empty queue: "
 788                     "i %zd mergeable %d offset %zd, size %zd, "
 789                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 790                     i, n->mergeable_rx_bufs, offset, size,
 791                     n->guest_hdr_len, n->host_hdr_len, n->vdev.guest_features);
 792             exit(1);
 793         }
 794
 795         if (elem.in_num < 1) {
 796             error_report("virtio-net receive queue contains no in buffers");
 797             exit(1);
 798         }
 799
 800         if (i == 0) {
 801             assert(offset == 0);
 802             if (n->mergeable_rx_bufs) {
 803                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
 804                                     sg, elem.in_num,
 805                                     offsetof(typeof(mhdr), num_buffers),
 806                                     sizeof(mhdr.num_buffers));
 807             }
 808
 809             receive_header(n, sg, elem.in_num, buf, size);
 810             offset = n->host_hdr_len;
 811             total += n->guest_hdr_len;
 812             guest_offset = n->guest_hdr_len;
 813         } else {
 814             guest_offset = 0;
 815         }
 816
 817         /* copy in packet.  ugh */
 818         len = iov_from_buf(sg, elem.in_num, guest_offset,
 819                            buf + offset, size - offset);
 820         total += len;
 821         offset += len;
 822         /* If buffers can't be merged, at this point we
 823          * must have consumed the complete packet.
 824          * Otherwise, drop it. */
 825         if (!n->mergeable_rx_bufs && offset < size) {
 826 #if 0
 827             error_report("virtio-net truncated non-mergeable packet: "
 828                          "i %zd mergeable %d offset %zd, size %zd, "
 829                          "guest hdr len %zd, host hdr len %zd",
 830                          i, n->mergeable_rx_bufs,
 831                          offset, size, n->guest_hdr_len, n->host_hdr_len);
 832 #endif
 833             return size;
 834         }
 835
 836         /* signal other side */
 837         virtqueue_fill(q->rx_vq, &elem, total, i++);
 838     }
 839
 840     if (mhdr_cnt) {
 841         stw_p(&mhdr.num_buffers, i);
 842         iov_from_buf(mhdr_sg, mhdr_cnt,
 843                      0,
 844                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
 845     }
 846
 847     virtqueue_flush(q->rx_vq, i);
 848     virtio_notify(&n->vdev, q->rx_vq);
 849
 850     return size;
 851 }
 852
 853 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
 854
 855 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
 856 {
 857     VirtIONet *n = qemu_get_nic_opaque(nc);
 858     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 859
 860     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
 861     virtio_notify(&n->vdev, q->tx_vq);
 862
 863     q->async_tx.elem.out_num = q->async_tx.len = 0;
 864
 865     virtio_queue_set_notification(q->tx_vq, 1);
 866     virtio_net_flush_tx(q);
 867 }
 868
 869 /* TX */
 870 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
 871 {
 872     VirtIONet *n = q->n;
 873     VirtQueueElement elem;
 874     int32_t num_packets = 0;
 875     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
 876     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 877         return num_packets;
 878     }
 879
 880     assert(n->vdev.vm_running);
 881
 882     if (q->async_tx.elem.out_num) {
 883         virtio_queue_set_notification(q->tx_vq, 0);
 884         return num_packets;
 885     }
 886
 887     while (virtqueue_pop(q->tx_vq, &elem)) {
 888         ssize_t ret, len;
 889         unsigned int out_num = elem.out_num;
 890         struct iovec *out_sg = &elem.out_sg[0];
 891         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 892
 893         if (out_num < 1) {
 894             error_report("virtio-net header not in first element");
 895             exit(1);
 896         }
 897
 898         /*
 899          * If host wants to see the guest header as is, we can
 900          * pass it on unchanged. Otherwise, copy just the parts
 901          * that host is interested in.
 902          */
 903         assert(n->host_hdr_len <= n->guest_hdr_len);
 904         if (n->host_hdr_len != n->guest_hdr_len) {
 905             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
 906                                        out_sg, out_num,
 907                                        0, n->host_hdr_len);
 908             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
 909                              out_sg, out_num,
 910                              n->guest_hdr_len, -1);
 911             out_num = sg_num;
 912             out_sg = sg;
 913         }
 914
 915         len = n->guest_hdr_len;
 916
 917         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
 918                                       out_sg, out_num, virtio_net_tx_complete);
 919         if (ret == 0) {
 920             virtio_queue_set_notification(q->tx_vq, 0);
 921             q->async_tx.elem = elem;
 922             q->async_tx.len  = len;
 923             return -EBUSY;
 924         }
 925
 926         len += ret;
 927
 928         virtqueue_push(q->tx_vq, &elem, 0);
 929         virtio_notify(&n->vdev, q->tx_vq);
 930
 931         if (++num_packets >= n->tx_burst) {
 932             break;
 933         }
 934     }
 935     return num_packets;
 936 }
 937
 938 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 939 {
 940     VirtIONet *n = to_virtio_net(vdev);
 941     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
 942
 943     /* This happens when device was stopped but VCPU wasn't. */
 944     if (!n->vdev.vm_running) {
 945         q->tx_waiting = 1;
 946         return;
 947     }
 948
 949     if (q->tx_waiting) {
 950         virtio_queue_set_notification(vq, 1);
 951         qemu_del_timer(q->tx_timer);
 952         q->tx_waiting = 0;
 953         virtio_net_flush_tx(q);
 954     } else {
 955         qemu_mod_timer(q->tx_timer,
 956                        qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 957         q->tx_waiting = 1;
 958         virtio_queue_set_notification(vq, 0);
 959     }
 960 }
 961
 962 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
 963 {
 964     VirtIONet *n = to_virtio_net(vdev);
 965     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
 966
 967     if (unlikely(q->tx_waiting)) {
 968         return;
 969     }
 970     q->tx_waiting = 1;
 971     /* This happens when device was stopped but VCPU wasn't. */
 972     if (!n->vdev.vm_running) {
 973         return;
 974     }
 975     virtio_queue_set_notification(vq, 0);
 976     qemu_bh_schedule(q->tx_bh);
 977 }
 978
 979 static void virtio_net_tx_timer(void *opaque)
 980 {
 981     VirtIONetQueue *q = opaque;
 982     VirtIONet *n = q->n;
 983     assert(n->vdev.vm_running);
 984
 985     q->tx_waiting = 0;
 986
 987     /* Just in case the driver is not ready on more */
 988     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 989         return;
 990
 991     virtio_queue_set_notification(q->tx_vq, 1);
 992     virtio_net_flush_tx(q);
 993 }
 994
 995 static void virtio_net_tx_bh(void *opaque)
 996 {
 997     VirtIONetQueue *q = opaque;
 998     VirtIONet *n = q->n;
 999     int32_t ret;
1000
1001     assert(n->vdev.vm_running);
1002
1003     q->tx_waiting = 0;
1004
1005     /* Just in case the driver is not ready on more */
1006     if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
1007         return;
1008
1009     ret = virtio_net_flush_tx(q);
1010     if (ret == -EBUSY) {
1011         return; /* Notification re-enable handled by tx_complete */
1012     }
1013
1014     /* If we flush a full burst of packets, assume there are
1015      * more coming and immediately reschedule */
1016     if (ret >= n->tx_burst) {
1017         qemu_bh_schedule(q->tx_bh);
1018         q->tx_waiting = 1;
1019         return;
1020     }
1021
1022     /* If less than a full burst, re-enable notification and flush
1023      * anything that may have come in while we weren't looking.  If
1024      * we find something, assume the guest is still active and reschedule */
1025     virtio_queue_set_notification(q->tx_vq, 1);
1026     if (virtio_net_flush_tx(q) > 0) {
1027         virtio_queue_set_notification(q->tx_vq, 0);
1028         qemu_bh_schedule(q->tx_bh);
1029         q->tx_waiting = 1;
1030     }
1031 }
1032
1033 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
1034 {
1035     VirtIODevice *vdev = &n->vdev;
1036     int i, max = multiqueue ? n->max_queues : 1;
1037
1038     n->multiqueue = multiqueue;
1039
1040     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1041         virtio_del_queue(vdev, i);
1042     }
1043
1044     for (i = 1; i < max; i++) {
1045         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1046         if (n->vqs[i].tx_timer) {
1047             n->vqs[i].tx_vq =
1048                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1049             n->vqs[i].tx_timer = qemu_new_timer_ns(vm_clock,
1050                                                    virtio_net_tx_timer,
1051                                                    &n->vqs[i]);
1052         } else {
1053             n->vqs[i].tx_vq =
1054                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1055             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1056         }
1057
1058         n->vqs[i].tx_waiting = 0;
1059         n->vqs[i].n = n;
1060     }
1061
1062     if (ctrl) {
1063         n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1064     }
1065
1066     virtio_net_set_queues(n);
1067 }
1068
1069 static void virtio_net_save(QEMUFile *f, void *opaque)
1070 {
1071     int i;
1072     VirtIONet *n = opaque;
1073
1074     /* At this point, backend must be stopped, otherwise
1075      * it might keep writing to memory. */
1076     assert(!n->vhost_started);
1077     virtio_save(&n->vdev, f);
1078
1079     qemu_put_buffer(f, n->mac, ETH_ALEN);
1080     qemu_put_be32(f, n->vqs[0].tx_waiting);
1081     qemu_put_be32(f, n->mergeable_rx_bufs);
1082     qemu_put_be16(f, n->status);
1083     qemu_put_byte(f, n->promisc);
1084     qemu_put_byte(f, n->allmulti);
1085     qemu_put_be32(f, n->mac_table.in_use);
1086     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1087     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1088     qemu_put_be32(f, n->has_vnet_hdr);
1089     qemu_put_byte(f, n->mac_table.multi_overflow);
1090     qemu_put_byte(f, n->mac_table.uni_overflow);
1091     qemu_put_byte(f, n->alluni);
1092     qemu_put_byte(f, n->nomulti);
1093     qemu_put_byte(f, n->nouni);
1094     qemu_put_byte(f, n->nobcast);
1095     qemu_put_byte(f, n->has_ufo);
1096     if (n->max_queues > 1) {
1097         qemu_put_be16(f, n->max_queues);
1098         qemu_put_be16(f, n->curr_queues);
1099         for (i = 1; i < n->curr_queues; i++) {
1100             qemu_put_be32(f, n->vqs[i].tx_waiting);
1101         }
1102     }
1103 }
1104
1105 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1106 {
1107     VirtIONet *n = opaque;
1108     int ret, i, link_down;
1109
1110     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1111         return -EINVAL;
1112
1113     ret = virtio_load(&n->vdev, f);
1114     if (ret) {
1115         return ret;
1116     }
1117
1118     qemu_get_buffer(f, n->mac, ETH_ALEN);
1119     n->vqs[0].tx_waiting = qemu_get_be32(f);
1120
1121     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1122
1123     if (version_id >= 3)
1124         n->status = qemu_get_be16(f);
1125
1126     if (version_id >= 4) {
1127         if (version_id < 8) {
1128             n->promisc = qemu_get_be32(f);
1129             n->allmulti = qemu_get_be32(f);
1130         } else {
1131             n->promisc = qemu_get_byte(f);
1132             n->allmulti = qemu_get_byte(f);
1133         }
1134     }
1135
1136     if (version_id >= 5) {
1137         n->mac_table.in_use = qemu_get_be32(f);
1138         /* MAC_TABLE_ENTRIES may be different from the saved image */
1139         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1140             qemu_get_buffer(f, n->mac_table.macs,
1141                             n->mac_table.in_use * ETH_ALEN);
1142         } else if (n->mac_table.in_use) {
1143             uint8_t *buf = g_malloc0(n->mac_table.in_use);
1144             qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
1145             g_free(buf);
1146             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1147             n->mac_table.in_use = 0;
1148         }
1149     }
1150
1151     if (version_id >= 6)
1152         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1153
1154     if (version_id >= 7) {
1155         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1156             error_report("virtio-net: saved image requires vnet_hdr=on");
1157             return -1;
1158         }
1159
1160         if (n->has_vnet_hdr) {
1161             tap_set_offload(qemu_get_queue(n->nic)->peer,
1162                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
1163                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
1164                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
1165                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
1166                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
1167         }
1168     }
1169
1170     if (version_id >= 9) {
1171         n->mac_table.multi_overflow = qemu_get_byte(f);
1172         n->mac_table.uni_overflow = qemu_get_byte(f);
1173     }
1174
1175     if (version_id >= 10) {
1176         n->alluni = qemu_get_byte(f);
1177         n->nomulti = qemu_get_byte(f);
1178         n->nouni = qemu_get_byte(f);
1179         n->nobcast = qemu_get_byte(f);
1180     }
1181
1182     if (version_id >= 11) {
1183         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1184             error_report("virtio-net: saved image requires TUN_F_UFO support");
1185             return -1;
1186         }
1187     }
1188
1189     if (n->max_queues > 1) {
1190         if (n->max_queues != qemu_get_be16(f)) {
1191             error_report("virtio-net: different max_queues ");
1192             return -1;
1193         }
1194
1195         n->curr_queues = qemu_get_be16(f);
1196         for (i = 1; i < n->curr_queues; i++) {
1197             n->vqs[i].tx_waiting = qemu_get_be32(f);
1198         }
1199     }
1200
1201     virtio_net_set_queues(n);
1202
1203     /* Find the first multicast entry in the saved MAC filter */
1204     for (i = 0; i < n->mac_table.in_use; i++) {
1205         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1206             break;
1207         }
1208     }
1209     n->mac_table.first_multi = i;
1210
1211     /* nc.link_down can't be migrated, so infer link_down according
1212      * to link status bit in n->status */
1213     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1214     for (i = 0; i < n->max_queues; i++) {
1215         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1216     }
1217
1218     return 0;
1219 }
1220
1221 static void virtio_net_cleanup(NetClientState *nc)
1222 {
1223     VirtIONet *n = qemu_get_nic_opaque(nc);
1224
1225     n->nic = NULL;
1226 }
1227
1228 static NetClientInfo net_virtio_info = {
1229     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1230     .size = sizeof(NICState),
1231     .can_receive = virtio_net_can_receive,
1232     .receive = virtio_net_receive,
1233         .cleanup = virtio_net_cleanup,
1234     .link_status_changed = virtio_net_set_link_status,
1235 };
1236
1237 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1238 {
1239     VirtIONet *n = to_virtio_net(vdev);
1240     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1241     assert(n->vhost_started);
1242     return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1243 }
1244
1245 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1246                                            bool mask)
1247 {
1248     VirtIONet *n = to_virtio_net(vdev);
1249     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1250     assert(n->vhost_started);
1251     vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1252                              vdev, idx, mask);
1253 }
1254
1255 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
1256                               virtio_net_conf *net, uint32_t host_features)
1257 {
1258     VirtIONet *n;
1259     int i, config_size = 0;
1260
1261     for (i = 0; feature_sizes[i].flags != 0; i++) {
1262         if (host_features & feature_sizes[i].flags) {
1263             config_size = MAX(feature_sizes[i].end, config_size);
1264         }
1265     }
1266
1267     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
1268                                         config_size, sizeof(VirtIONet));
1269
1270     n->config_size = config_size;
1271     n->vdev.get_config = virtio_net_get_config;
1272     n->vdev.set_config = virtio_net_set_config;
1273     n->vdev.get_features = virtio_net_get_features;
1274     n->vdev.set_features = virtio_net_set_features;
1275     n->vdev.bad_features = virtio_net_bad_features;
1276     n->vdev.reset = virtio_net_reset;
1277     n->vdev.set_status = virtio_net_set_status;
1278     n->vdev.guest_notifier_mask = virtio_net_guest_notifier_mask;
1279     n->vdev.guest_notifier_pending = virtio_net_guest_notifier_pending;
1280     n->max_queues = MAX(conf->queues, 1);
1281     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1282     n->vqs[0].rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
1283     n->curr_queues = 1;
1284     n->vqs[0].n = n;
1285     n->tx_timeout = net->txtimer;
1286
1287     if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
1288         error_report("virtio-net: "
1289                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1290                      net->tx);
1291         error_report("Defaulting to \"bh\"");
1292     }
1293
1294     if (net->tx && !strcmp(net->tx, "timer")) {
1295         n->vqs[0].tx_vq = virtio_add_queue(&n->vdev, 256,
1296                                            virtio_net_handle_tx_timer);
1297         n->vqs[0].tx_timer = qemu_new_timer_ns(vm_clock, virtio_net_tx_timer,
1298                                                &n->vqs[0]);
1299     } else {
1300         n->vqs[0].tx_vq = virtio_add_queue(&n->vdev, 256,
1301                                            virtio_net_handle_tx_bh);
1302         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1303     }
1304     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
1305     qemu_macaddr_default_if_unset(&conf->macaddr);
1306     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
1307     n->status = VIRTIO_NET_S_LINK_UP;
1308
1309     n->nic = qemu_new_nic(&net_virtio_info, conf, object_get_typename(OBJECT(dev)), dev->id, n);
1310     peer_test_vnet_hdr(n);
1311     if (peer_has_vnet_hdr(n)) {
1312         for (i = 0; i < n->max_queues; i++) {
1313             tap_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1314         }
1315         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1316     } else {
1317         n->host_hdr_len = 0;
1318     }
1319
1320     qemu_format_nic_info_str(qemu_get_queue(n->nic), conf->macaddr.a);
1321
1322     n->vqs[0].tx_waiting = 0;
1323     n->tx_burst = net->txburst;
1324     virtio_net_set_mrg_rx_bufs(n, 0);
1325     n->promisc = 1; /* for compatibility */
1326
1327     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1328
1329     n->vlans = g_malloc0(MAX_VLAN >> 3);
1330
1331     n->qdev = dev;
1332     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1333                     virtio_net_save, virtio_net_load, n);
1334
1335     add_boot_device_path(conf->bootindex, dev, "/ethernet-phy@0");
1336
1337     return &n->vdev;
1338 }
1339
1340 void virtio_net_exit(VirtIODevice *vdev)
1341 {
1342     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
1343     int i;
1344
1345     /* This will stop vhost backend if appropriate. */
1346     virtio_net_set_status(vdev, 0);
1347
1348     unregister_savevm(n->qdev, "virtio-net", n);
1349
1350     g_free(n->mac_table.macs);
1351     g_free(n->vlans);
1352
1353     for (i = 0; i < n->max_queues; i++) {
1354         VirtIONetQueue *q = &n->vqs[i];
1355         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1356
1357         qemu_purge_queued_packets(nc);
1358
1359         if (q->tx_timer) {
1360             qemu_del_timer(q->tx_timer);
1361             qemu_free_timer(q->tx_timer);
1362         } else {
1363             qemu_bh_delete(q->tx_bh);
1364         }
1365     }
1366
1367     g_free(n->vqs);
1368     qemu_del_nic(n->nic);
1369     virtio_cleanup(&n->vdev);
1370 }