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 #include "hw/virtio/virtio-bus.h"
  24
  25 #define VIRTIO_NET_VM_VERSION    11
  26
  27 #define MAC_TABLE_ENTRIES    64
  28 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  29
  30 /*
  31  * Calculate the number of bytes up to and including the given 'field' of
  32  * 'container'.
  33  */
  34 #define endof(container, field) \
  35     (offsetof(container, field) + sizeof(((container *)0)->field))
  36
  37 typedef struct VirtIOFeature {
  38     uint32_t flags;
  39     size_t end;
  40 } VirtIOFeature;
  41
  42 static VirtIOFeature feature_sizes[] = {
  43     {.flags = 1 << VIRTIO_NET_F_MAC,
  44      .end = endof(struct virtio_net_config, mac)},
  45     {.flags = 1 << VIRTIO_NET_F_STATUS,
  46      .end = endof(struct virtio_net_config, status)},
  47     {.flags = 1 << VIRTIO_NET_F_MQ,
  48      .end = endof(struct virtio_net_config, max_virtqueue_pairs)},
  49     {}
  50 };
  51
  52 static VirtIONetQueue *virtio_net_get_subqueue(NetClientState *nc)
  53 {
  54     VirtIONet *n = qemu_get_nic_opaque(nc);
  55
  56     return &n->vqs[nc->queue_index];
  57 }
  58
  59 static int vq2q(int queue_index)
  60 {
  61     return queue_index / 2;
  62 }
  63
  64 /* TODO
  65  * - we could suppress RX interrupt if we were so inclined.
  66  */
  67
  68 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  69 {
  70     VirtIONet *n = VIRTIO_NET(vdev);
  71     struct virtio_net_config netcfg;
  72
  73     stw_p(&netcfg.status, n->status);
  74     stw_p(&netcfg.max_virtqueue_pairs, n->max_queues);
  75     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  76     memcpy(config, &netcfg, n->config_size);
  77 }
  78
  79 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  80 {
  81     VirtIONet *n = VIRTIO_NET(vdev);
  82     struct virtio_net_config netcfg = {};
  83
  84     memcpy(&netcfg, config, n->config_size);
  85
  86     if (!(vdev->guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) &&
  87         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  88         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  89         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
  90     }
  91 }
  92
  93 static bool virtio_net_started(VirtIONet *n, uint8_t status)
  94 {
  95     VirtIODevice *vdev = VIRTIO_DEVICE(n);
  96     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
  97         (n->status & VIRTIO_NET_S_LINK_UP) && vdev->vm_running;
  98 }
  99
 100 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
 101 {
 102     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 103     NetClientState *nc = qemu_get_queue(n->nic);
 104     int queues = n->multiqueue ? n->max_queues : 1;
 105
 106     if (!nc->peer) {
 107         return;
 108     }
 109     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 110         return;
 111     }
 112
 113     if (!tap_get_vhost_net(nc->peer)) {
 114         return;
 115     }
 116
 117     if (!!n->vhost_started == virtio_net_started(n, status) &&
 118                               !nc->peer->link_down) {
 119         return;
 120     }
 121     if (!n->vhost_started) {
 122         int r;
 123         if (!vhost_net_query(tap_get_vhost_net(nc->peer), vdev)) {
 124             return;
 125         }
 126         n->vhost_started = 1;
 127         r = vhost_net_start(vdev, n->nic->ncs, queues);
 128         if (r < 0) {
 129             error_report("unable to start vhost net: %d: "
 130                          "falling back on userspace virtio", -r);
 131             n->vhost_started = 0;
 132         }
 133     } else {
 134         vhost_net_stop(vdev, n->nic->ncs, queues);
 135         n->vhost_started = 0;
 136     }
 137 }
 138
 139 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 140 {
 141     VirtIONet *n = VIRTIO_NET(vdev);
 142     VirtIONetQueue *q;
 143     int i;
 144     uint8_t queue_status;
 145
 146     virtio_net_vhost_status(n, status);
 147
 148     for (i = 0; i < n->max_queues; i++) {
 149         q = &n->vqs[i];
 150
 151         if ((!n->multiqueue && i != 0) || i >= n->curr_queues) {
 152             queue_status = 0;
 153         } else {
 154             queue_status = status;
 155         }
 156
 157         if (!q->tx_waiting) {
 158             continue;
 159         }
 160
 161         if (virtio_net_started(n, queue_status) && !n->vhost_started) {
 162             if (q->tx_timer) {
 163                 qemu_mod_timer(q->tx_timer,
 164                                qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 165             } else {
 166                 qemu_bh_schedule(q->tx_bh);
 167             }
 168         } else {
 169             if (q->tx_timer) {
 170                 qemu_del_timer(q->tx_timer);
 171             } else {
 172                 qemu_bh_cancel(q->tx_bh);
 173             }
 174         }
 175     }
 176 }
 177
 178 static void virtio_net_set_link_status(NetClientState *nc)
 179 {
 180     VirtIONet *n = qemu_get_nic_opaque(nc);
 181     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 182     uint16_t old_status = n->status;
 183
 184     if (nc->link_down)
 185         n->status &= ~VIRTIO_NET_S_LINK_UP;
 186     else
 187         n->status |= VIRTIO_NET_S_LINK_UP;
 188
 189     if (n->status != old_status)
 190         virtio_notify_config(vdev);
 191
 192     virtio_net_set_status(vdev, vdev->status);
 193 }
 194
 195 static void virtio_net_reset(VirtIODevice *vdev)
 196 {
 197     VirtIONet *n = VIRTIO_NET(vdev);
 198
 199     /* Reset back to compatibility mode */
 200     n->promisc = 1;
 201     n->allmulti = 0;
 202     n->alluni = 0;
 203     n->nomulti = 0;
 204     n->nouni = 0;
 205     n->nobcast = 0;
 206     /* multiqueue is disabled by default */
 207     n->curr_queues = 1;
 208
 209     /* Flush any MAC and VLAN filter table state */
 210     n->mac_table.in_use = 0;
 211     n->mac_table.first_multi = 0;
 212     n->mac_table.multi_overflow = 0;
 213     n->mac_table.uni_overflow = 0;
 214     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 215     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
 216     memset(n->vlans, 0, MAX_VLAN >> 3);
 217 }
 218
 219 static void peer_test_vnet_hdr(VirtIONet *n)
 220 {
 221     NetClientState *nc = qemu_get_queue(n->nic);
 222     if (!nc->peer) {
 223         return;
 224     }
 225
 226     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 227         return;
 228     }
 229
 230     n->has_vnet_hdr = tap_has_vnet_hdr(nc->peer);
 231 }
 232
 233 static int peer_has_vnet_hdr(VirtIONet *n)
 234 {
 235     return n->has_vnet_hdr;
 236 }
 237
 238 static int peer_has_ufo(VirtIONet *n)
 239 {
 240     if (!peer_has_vnet_hdr(n))
 241         return 0;
 242
 243     n->has_ufo = tap_has_ufo(qemu_get_queue(n->nic)->peer);
 244
 245     return n->has_ufo;
 246 }
 247
 248 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
 249 {
 250     int i;
 251     NetClientState *nc;
 252
 253     n->mergeable_rx_bufs = mergeable_rx_bufs;
 254
 255     n->guest_hdr_len = n->mergeable_rx_bufs ?
 256         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 257
 258     for (i = 0; i < n->max_queues; i++) {
 259         nc = qemu_get_subqueue(n->nic, i);
 260
 261         if (peer_has_vnet_hdr(n) &&
 262             tap_has_vnet_hdr_len(nc->peer, n->guest_hdr_len)) {
 263             tap_set_vnet_hdr_len(nc->peer, n->guest_hdr_len);
 264             n->host_hdr_len = n->guest_hdr_len;
 265         }
 266     }
 267 }
 268
 269 static int peer_attach(VirtIONet *n, int index)
 270 {
 271     NetClientState *nc = qemu_get_subqueue(n->nic, index);
 272
 273     if (!nc->peer) {
 274         return 0;
 275     }
 276
 277     if (nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 278         return 0;
 279     }
 280
 281     return tap_enable(nc->peer);
 282 }
 283
 284 static int peer_detach(VirtIONet *n, int index)
 285 {
 286     NetClientState *nc = qemu_get_subqueue(n->nic, index);
 287
 288     if (!nc->peer) {
 289         return 0;
 290     }
 291
 292     if (nc->peer->info->type !=  NET_CLIENT_OPTIONS_KIND_TAP) {
 293         return 0;
 294     }
 295
 296     return tap_disable(nc->peer);
 297 }
 298
 299 static void virtio_net_set_queues(VirtIONet *n)
 300 {
 301     int i;
 302
 303     for (i = 0; i < n->max_queues; i++) {
 304         if (i < n->curr_queues) {
 305             assert(!peer_attach(n, i));
 306         } else {
 307             assert(!peer_detach(n, i));
 308         }
 309     }
 310 }
 311
 312 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
 313
 314 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 315 {
 316     VirtIONet *n = VIRTIO_NET(vdev);
 317     NetClientState *nc = qemu_get_queue(n->nic);
 318
 319     features |= (1 << VIRTIO_NET_F_MAC);
 320
 321     if (!peer_has_vnet_hdr(n)) {
 322         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 323         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 324         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 325         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 326
 327         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 328         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 329         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 330         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 331     }
 332
 333     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 334         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 335         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 336     }
 337
 338     if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 339         return features;
 340     }
 341     if (!tap_get_vhost_net(nc->peer)) {
 342         return features;
 343     }
 344     return vhost_net_get_features(tap_get_vhost_net(nc->peer), features);
 345 }
 346
 347 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 348 {
 349     uint32_t features = 0;
 350
 351     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 352      * but also these: */
 353     features |= (1 << VIRTIO_NET_F_MAC);
 354     features |= (1 << VIRTIO_NET_F_CSUM);
 355     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 356     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 357     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 358
 359     return features;
 360 }
 361
 362 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 363 {
 364     VirtIONet *n = VIRTIO_NET(vdev);
 365     int i;
 366
 367     virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
 368
 369     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
 370
 371     if (n->has_vnet_hdr) {
 372         tap_set_offload(qemu_get_subqueue(n->nic, 0)->peer,
 373                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 374                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 375                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 376                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 377                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 378     }
 379
 380     for (i = 0;  i < n->max_queues; i++) {
 381         NetClientState *nc = qemu_get_subqueue(n->nic, i);
 382
 383         if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 384             continue;
 385         }
 386         if (!tap_get_vhost_net(nc->peer)) {
 387             continue;
 388         }
 389         vhost_net_ack_features(tap_get_vhost_net(nc->peer), features);
 390     }
 391 }
 392
 393 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 394                                      struct iovec *iov, unsigned int iov_cnt)
 395 {
 396     uint8_t on;
 397     size_t s;
 398
 399     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
 400     if (s != sizeof(on)) {
 401         return VIRTIO_NET_ERR;
 402     }
 403
 404     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
 405         n->promisc = on;
 406     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
 407         n->allmulti = on;
 408     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
 409         n->alluni = on;
 410     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
 411         n->nomulti = on;
 412     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
 413         n->nouni = on;
 414     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
 415         n->nobcast = on;
 416     } else {
 417         return VIRTIO_NET_ERR;
 418     }
 419
 420     return VIRTIO_NET_OK;
 421 }
 422
 423 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 424                                  struct iovec *iov, unsigned int iov_cnt)
 425 {
 426     struct virtio_net_ctrl_mac mac_data;
 427     size_t s;
 428
 429     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
 430         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
 431             return VIRTIO_NET_ERR;
 432         }
 433         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
 434         assert(s == sizeof(n->mac));
 435         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
 436         return VIRTIO_NET_OK;
 437     }
 438
 439     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
 440         return VIRTIO_NET_ERR;
 441     }
 442
 443     n->mac_table.in_use = 0;
 444     n->mac_table.first_multi = 0;
 445     n->mac_table.uni_overflow = 0;
 446     n->mac_table.multi_overflow = 0;
 447     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 448
 449     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 450                    sizeof(mac_data.entries));
 451     mac_data.entries = ldl_p(&mac_data.entries);
 452     if (s != sizeof(mac_data.entries)) {
 453         return VIRTIO_NET_ERR;
 454     }
 455     iov_discard_front(&iov, &iov_cnt, s);
 456
 457     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
 458         return VIRTIO_NET_ERR;
 459     }
 460
 461     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 462         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 463                        mac_data.entries * ETH_ALEN);
 464         if (s != mac_data.entries * ETH_ALEN) {
 465             return VIRTIO_NET_ERR;
 466         }
 467         n->mac_table.in_use += mac_data.entries;
 468     } else {
 469         n->mac_table.uni_overflow = 1;
 470     }
 471
 472     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
 473
 474     n->mac_table.first_multi = n->mac_table.in_use;
 475
 476     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 477                    sizeof(mac_data.entries));
 478     mac_data.entries = ldl_p(&mac_data.entries);
 479     if (s != sizeof(mac_data.entries)) {
 480         return VIRTIO_NET_ERR;
 481     }
 482
 483     iov_discard_front(&iov, &iov_cnt, s);
 484
 485     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
 486         return VIRTIO_NET_ERR;
 487     }
 488
 489     if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 490         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 491                        mac_data.entries * ETH_ALEN);
 492         if (s != mac_data.entries * ETH_ALEN) {
 493             return VIRTIO_NET_ERR;
 494         }
 495         n->mac_table.in_use += mac_data.entries;
 496     } else {
 497         n->mac_table.multi_overflow = 1;
 498     }
 499
 500     return VIRTIO_NET_OK;
 501 }
 502
 503 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 504                                         struct iovec *iov, unsigned int iov_cnt)
 505 {
 506     uint16_t vid;
 507     size_t s;
 508
 509     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
 510     vid = lduw_p(&vid);
 511     if (s != sizeof(vid)) {
 512         return VIRTIO_NET_ERR;
 513     }
 514
 515     if (vid >= MAX_VLAN)
 516         return VIRTIO_NET_ERR;
 517
 518     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 519         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 520     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 521         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 522     else
 523         return VIRTIO_NET_ERR;
 524
 525     return VIRTIO_NET_OK;
 526 }
 527
 528 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
 529                                 struct iovec *iov, unsigned int iov_cnt)
 530 {
 531     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 532     struct virtio_net_ctrl_mq mq;
 533     size_t s;
 534     uint16_t queues;
 535
 536     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
 537     if (s != sizeof(mq)) {
 538         return VIRTIO_NET_ERR;
 539     }
 540
 541     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 542         return VIRTIO_NET_ERR;
 543     }
 544
 545     queues = lduw_p(&mq.virtqueue_pairs);
 546
 547     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
 548         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
 549         queues > n->max_queues ||
 550         !n->multiqueue) {
 551         return VIRTIO_NET_ERR;
 552     }
 553
 554     n->curr_queues = queues;
 555     /* stop the backend before changing the number of queues to avoid handling a
 556      * disabled queue */
 557     virtio_net_set_status(vdev, vdev->status);
 558     virtio_net_set_queues(n);
 559
 560     return VIRTIO_NET_OK;
 561 }
 562 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 563 {
 564     VirtIONet *n = VIRTIO_NET(vdev);
 565     struct virtio_net_ctrl_hdr ctrl;
 566     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 567     VirtQueueElement elem;
 568     size_t s;
 569     struct iovec *iov;
 570     unsigned int iov_cnt;
 571
 572     while (virtqueue_pop(vq, &elem)) {
 573         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
 574             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
 575             error_report("virtio-net ctrl missing headers");
 576             exit(1);
 577         }
 578
 579         iov = elem.out_sg;
 580         iov_cnt = elem.out_num;
 581         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
 582         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
 583         if (s != sizeof(ctrl)) {
 584             status = VIRTIO_NET_ERR;
 585         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
 586             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
 587         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
 588             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
 589         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
 590             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
 591         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
 592             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
 593         }
 594
 595         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
 596         assert(s == sizeof(status));
 597
 598         virtqueue_push(vq, &elem, sizeof(status));
 599         virtio_notify(vdev, vq);
 600     }
 601 }
 602
 603 /* RX */
 604
 605 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 606 {
 607     VirtIONet *n = VIRTIO_NET(vdev);
 608     int queue_index = vq2q(virtio_get_queue_index(vq));
 609
 610     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
 611 }
 612
 613 static int virtio_net_can_receive(NetClientState *nc)
 614 {
 615     VirtIONet *n = qemu_get_nic_opaque(nc);
 616     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 617     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 618
 619     if (!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         !(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     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 759     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
 760     struct virtio_net_hdr_mrg_rxbuf mhdr;
 761     unsigned mhdr_cnt = 0;
 762     size_t offset, i, guest_offset;
 763
 764     if (!virtio_net_can_receive(nc)) {
 765         return -1;
 766     }
 767
 768     /* hdr_len refers to the header we supply to the guest */
 769     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
 770         return 0;
 771     }
 772
 773     if (!receive_filter(n, buf, size))
 774         return size;
 775
 776     offset = i = 0;
 777
 778     while (offset < size) {
 779         VirtQueueElement elem;
 780         int len, total;
 781         const struct iovec *sg = elem.in_sg;
 782
 783         total = 0;
 784
 785         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
 786             if (i == 0)
 787                 return -1;
 788             error_report("virtio-net unexpected empty queue: "
 789                     "i %zd mergeable %d offset %zd, size %zd, "
 790                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 791                     i, n->mergeable_rx_bufs, offset, size,
 792                     n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
 793             exit(1);
 794         }
 795
 796         if (elem.in_num < 1) {
 797             error_report("virtio-net receive queue contains no in buffers");
 798             exit(1);
 799         }
 800
 801         if (i == 0) {
 802             assert(offset == 0);
 803             if (n->mergeable_rx_bufs) {
 804                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
 805                                     sg, elem.in_num,
 806                                     offsetof(typeof(mhdr), num_buffers),
 807                                     sizeof(mhdr.num_buffers));
 808             }
 809
 810             receive_header(n, sg, elem.in_num, buf, size);
 811             offset = n->host_hdr_len;
 812             total += n->guest_hdr_len;
 813             guest_offset = n->guest_hdr_len;
 814         } else {
 815             guest_offset = 0;
 816         }
 817
 818         /* copy in packet.  ugh */
 819         len = iov_from_buf(sg, elem.in_num, guest_offset,
 820                            buf + offset, size - offset);
 821         total += len;
 822         offset += len;
 823         /* If buffers can't be merged, at this point we
 824          * must have consumed the complete packet.
 825          * Otherwise, drop it. */
 826         if (!n->mergeable_rx_bufs && offset < size) {
 827 #if 0
 828             error_report("virtio-net truncated non-mergeable packet: "
 829                          "i %zd mergeable %d offset %zd, size %zd, "
 830                          "guest hdr len %zd, host hdr len %zd",
 831                          i, n->mergeable_rx_bufs,
 832                          offset, size, n->guest_hdr_len, n->host_hdr_len);
 833 #endif
 834             return size;
 835         }
 836
 837         /* signal other side */
 838         virtqueue_fill(q->rx_vq, &elem, total, i++);
 839     }
 840
 841     if (mhdr_cnt) {
 842         stw_p(&mhdr.num_buffers, i);
 843         iov_from_buf(mhdr_sg, mhdr_cnt,
 844                      0,
 845                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
 846     }
 847
 848     virtqueue_flush(q->rx_vq, i);
 849     virtio_notify(vdev, q->rx_vq);
 850
 851     return size;
 852 }
 853
 854 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
 855
 856 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
 857 {
 858     VirtIONet *n = qemu_get_nic_opaque(nc);
 859     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 860     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 861
 862     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
 863     virtio_notify(vdev, q->tx_vq);
 864
 865     q->async_tx.elem.out_num = q->async_tx.len = 0;
 866
 867     virtio_queue_set_notification(q->tx_vq, 1);
 868     virtio_net_flush_tx(q);
 869 }
 870
 871 /* TX */
 872 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
 873 {
 874     VirtIONet *n = q->n;
 875     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 876     VirtQueueElement elem;
 877     int32_t num_packets = 0;
 878     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
 879     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 880         return num_packets;
 881     }
 882
 883     assert(vdev->vm_running);
 884
 885     if (q->async_tx.elem.out_num) {
 886         virtio_queue_set_notification(q->tx_vq, 0);
 887         return num_packets;
 888     }
 889
 890     while (virtqueue_pop(q->tx_vq, &elem)) {
 891         ssize_t ret, len;
 892         unsigned int out_num = elem.out_num;
 893         struct iovec *out_sg = &elem.out_sg[0];
 894         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 895
 896         if (out_num < 1) {
 897             error_report("virtio-net header not in first element");
 898             exit(1);
 899         }
 900
 901         /*
 902          * If host wants to see the guest header as is, we can
 903          * pass it on unchanged. Otherwise, copy just the parts
 904          * that host is interested in.
 905          */
 906         assert(n->host_hdr_len <= n->guest_hdr_len);
 907         if (n->host_hdr_len != n->guest_hdr_len) {
 908             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
 909                                        out_sg, out_num,
 910                                        0, n->host_hdr_len);
 911             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
 912                              out_sg, out_num,
 913                              n->guest_hdr_len, -1);
 914             out_num = sg_num;
 915             out_sg = sg;
 916         }
 917
 918         len = n->guest_hdr_len;
 919
 920         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
 921                                       out_sg, out_num, virtio_net_tx_complete);
 922         if (ret == 0) {
 923             virtio_queue_set_notification(q->tx_vq, 0);
 924             q->async_tx.elem = elem;
 925             q->async_tx.len  = len;
 926             return -EBUSY;
 927         }
 928
 929         len += ret;
 930
 931         virtqueue_push(q->tx_vq, &elem, 0);
 932         virtio_notify(vdev, q->tx_vq);
 933
 934         if (++num_packets >= n->tx_burst) {
 935             break;
 936         }
 937     }
 938     return num_packets;
 939 }
 940
 941 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 942 {
 943     VirtIONet *n = VIRTIO_NET(vdev);
 944     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
 945
 946     /* This happens when device was stopped but VCPU wasn't. */
 947     if (!vdev->vm_running) {
 948         q->tx_waiting = 1;
 949         return;
 950     }
 951
 952     if (q->tx_waiting) {
 953         virtio_queue_set_notification(vq, 1);
 954         qemu_del_timer(q->tx_timer);
 955         q->tx_waiting = 0;
 956         virtio_net_flush_tx(q);
 957     } else {
 958         qemu_mod_timer(q->tx_timer,
 959                        qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 960         q->tx_waiting = 1;
 961         virtio_queue_set_notification(vq, 0);
 962     }
 963 }
 964
 965 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
 966 {
 967     VirtIONet *n = VIRTIO_NET(vdev);
 968     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
 969
 970     if (unlikely(q->tx_waiting)) {
 971         return;
 972     }
 973     q->tx_waiting = 1;
 974     /* This happens when device was stopped but VCPU wasn't. */
 975     if (!vdev->vm_running) {
 976         return;
 977     }
 978     virtio_queue_set_notification(vq, 0);
 979     qemu_bh_schedule(q->tx_bh);
 980 }
 981
 982 static void virtio_net_tx_timer(void *opaque)
 983 {
 984     VirtIONetQueue *q = opaque;
 985     VirtIONet *n = q->n;
 986     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 987     assert(vdev->vm_running);
 988
 989     q->tx_waiting = 0;
 990
 991     /* Just in case the driver is not ready on more */
 992     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 993         return;
 994     }
 995
 996     virtio_queue_set_notification(q->tx_vq, 1);
 997     virtio_net_flush_tx(q);
 998 }
 999
1000 static void virtio_net_tx_bh(void *opaque)
1001 {
1002     VirtIONetQueue *q = opaque;
1003     VirtIONet *n = q->n;
1004     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1005     int32_t ret;
1006
1007     assert(vdev->vm_running);
1008
1009     q->tx_waiting = 0;
1010
1011     /* Just in case the driver is not ready on more */
1012     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1013         return;
1014     }
1015
1016     ret = virtio_net_flush_tx(q);
1017     if (ret == -EBUSY) {
1018         return; /* Notification re-enable handled by tx_complete */
1019     }
1020
1021     /* If we flush a full burst of packets, assume there are
1022      * more coming and immediately reschedule */
1023     if (ret >= n->tx_burst) {
1024         qemu_bh_schedule(q->tx_bh);
1025         q->tx_waiting = 1;
1026         return;
1027     }
1028
1029     /* If less than a full burst, re-enable notification and flush
1030      * anything that may have come in while we weren't looking.  If
1031      * we find something, assume the guest is still active and reschedule */
1032     virtio_queue_set_notification(q->tx_vq, 1);
1033     if (virtio_net_flush_tx(q) > 0) {
1034         virtio_queue_set_notification(q->tx_vq, 0);
1035         qemu_bh_schedule(q->tx_bh);
1036         q->tx_waiting = 1;
1037     }
1038 }
1039
1040 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1041 {
1042     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1043     int i, max = multiqueue ? n->max_queues : 1;
1044
1045     n->multiqueue = multiqueue;
1046
1047     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1048         virtio_del_queue(vdev, i);
1049     }
1050
1051     for (i = 1; i < max; i++) {
1052         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1053         if (n->vqs[i].tx_timer) {
1054             n->vqs[i].tx_vq =
1055                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1056             n->vqs[i].tx_timer = qemu_new_timer_ns(vm_clock,
1057                                                    virtio_net_tx_timer,
1058                                                    &n->vqs[i]);
1059         } else {
1060             n->vqs[i].tx_vq =
1061                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1062             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1063         }
1064
1065         n->vqs[i].tx_waiting = 0;
1066         n->vqs[i].n = n;
1067     }
1068
1069     /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
1070      * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
1071      * breaking them.
1072      */
1073     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1074
1075     virtio_net_set_queues(n);
1076 }
1077
1078 static void virtio_net_save(QEMUFile *f, void *opaque)
1079 {
1080     int i;
1081     VirtIONet *n = opaque;
1082     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1083
1084     /* At this point, backend must be stopped, otherwise
1085      * it might keep writing to memory. */
1086     assert(!n->vhost_started);
1087     virtio_save(vdev, f);
1088
1089     qemu_put_buffer(f, n->mac, ETH_ALEN);
1090     qemu_put_be32(f, n->vqs[0].tx_waiting);
1091     qemu_put_be32(f, n->mergeable_rx_bufs);
1092     qemu_put_be16(f, n->status);
1093     qemu_put_byte(f, n->promisc);
1094     qemu_put_byte(f, n->allmulti);
1095     qemu_put_be32(f, n->mac_table.in_use);
1096     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1097     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1098     qemu_put_be32(f, n->has_vnet_hdr);
1099     qemu_put_byte(f, n->mac_table.multi_overflow);
1100     qemu_put_byte(f, n->mac_table.uni_overflow);
1101     qemu_put_byte(f, n->alluni);
1102     qemu_put_byte(f, n->nomulti);
1103     qemu_put_byte(f, n->nouni);
1104     qemu_put_byte(f, n->nobcast);
1105     qemu_put_byte(f, n->has_ufo);
1106     if (n->max_queues > 1) {
1107         qemu_put_be16(f, n->max_queues);
1108         qemu_put_be16(f, n->curr_queues);
1109         for (i = 1; i < n->curr_queues; i++) {
1110             qemu_put_be32(f, n->vqs[i].tx_waiting);
1111         }
1112     }
1113 }
1114
1115 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1116 {
1117     VirtIONet *n = opaque;
1118     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1119     int ret, i, link_down;
1120
1121     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1122         return -EINVAL;
1123
1124     ret = virtio_load(vdev, f);
1125     if (ret) {
1126         return ret;
1127     }
1128
1129     qemu_get_buffer(f, n->mac, ETH_ALEN);
1130     n->vqs[0].tx_waiting = qemu_get_be32(f);
1131
1132     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1133
1134     if (version_id >= 3)
1135         n->status = qemu_get_be16(f);
1136
1137     if (version_id >= 4) {
1138         if (version_id < 8) {
1139             n->promisc = qemu_get_be32(f);
1140             n->allmulti = qemu_get_be32(f);
1141         } else {
1142             n->promisc = qemu_get_byte(f);
1143             n->allmulti = qemu_get_byte(f);
1144         }
1145     }
1146
1147     if (version_id >= 5) {
1148         n->mac_table.in_use = qemu_get_be32(f);
1149         /* MAC_TABLE_ENTRIES may be different from the saved image */
1150         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1151             qemu_get_buffer(f, n->mac_table.macs,
1152                             n->mac_table.in_use * ETH_ALEN);
1153         } else if (n->mac_table.in_use) {
1154             uint8_t *buf = g_malloc0(n->mac_table.in_use);
1155             qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
1156             g_free(buf);
1157             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1158             n->mac_table.in_use = 0;
1159         }
1160     }
1161
1162     if (version_id >= 6)
1163         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1164
1165     if (version_id >= 7) {
1166         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1167             error_report("virtio-net: saved image requires vnet_hdr=on");
1168             return -1;
1169         }
1170
1171         if (n->has_vnet_hdr) {
1172             tap_set_offload(qemu_get_queue(n->nic)->peer,
1173                     (vdev->guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
1174                     (vdev->guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
1175                     (vdev->guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
1176                     (vdev->guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
1177                     (vdev->guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
1178         }
1179     }
1180
1181     if (version_id >= 9) {
1182         n->mac_table.multi_overflow = qemu_get_byte(f);
1183         n->mac_table.uni_overflow = qemu_get_byte(f);
1184     }
1185
1186     if (version_id >= 10) {
1187         n->alluni = qemu_get_byte(f);
1188         n->nomulti = qemu_get_byte(f);
1189         n->nouni = qemu_get_byte(f);
1190         n->nobcast = qemu_get_byte(f);
1191     }
1192
1193     if (version_id >= 11) {
1194         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1195             error_report("virtio-net: saved image requires TUN_F_UFO support");
1196             return -1;
1197         }
1198     }
1199
1200     if (n->max_queues > 1) {
1201         if (n->max_queues != qemu_get_be16(f)) {
1202             error_report("virtio-net: different max_queues ");
1203             return -1;
1204         }
1205
1206         n->curr_queues = qemu_get_be16(f);
1207         for (i = 1; i < n->curr_queues; i++) {
1208             n->vqs[i].tx_waiting = qemu_get_be32(f);
1209         }
1210     }
1211
1212     virtio_net_set_queues(n);
1213
1214     /* Find the first multicast entry in the saved MAC filter */
1215     for (i = 0; i < n->mac_table.in_use; i++) {
1216         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1217             break;
1218         }
1219     }
1220     n->mac_table.first_multi = i;
1221
1222     /* nc.link_down can't be migrated, so infer link_down according
1223      * to link status bit in n->status */
1224     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1225     for (i = 0; i < n->max_queues; i++) {
1226         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1227     }
1228
1229     return 0;
1230 }
1231
1232 static void virtio_net_cleanup(NetClientState *nc)
1233 {
1234     VirtIONet *n = qemu_get_nic_opaque(nc);
1235
1236     n->nic = NULL;
1237 }
1238
1239 static NetClientInfo net_virtio_info = {
1240     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1241     .size = sizeof(NICState),
1242     .can_receive = virtio_net_can_receive,
1243     .receive = virtio_net_receive,
1244         .cleanup = virtio_net_cleanup,
1245     .link_status_changed = virtio_net_set_link_status,
1246 };
1247
1248 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1249 {
1250     VirtIONet *n = VIRTIO_NET(vdev);
1251     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1252     assert(n->vhost_started);
1253     return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1254 }
1255
1256 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1257                                            bool mask)
1258 {
1259     VirtIONet *n = VIRTIO_NET(vdev);
1260     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1261     assert(n->vhost_started);
1262     vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1263                              vdev, idx, mask);
1264 }
1265
1266 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1267 {
1268     int i, config_size = 0;
1269     host_features |= (1 << VIRTIO_NET_F_MAC);
1270     for (i = 0; feature_sizes[i].flags != 0; i++) {
1271         if (host_features & feature_sizes[i].flags) {
1272             config_size = MAX(feature_sizes[i].end, config_size);
1273         }
1274     }
1275     n->config_size = config_size;
1276 }
1277
1278 static int virtio_net_device_init(VirtIODevice *vdev)
1279 {
1280     int i;
1281
1282     DeviceState *qdev = DEVICE(vdev);
1283     VirtIONet *n = VIRTIO_NET(vdev);
1284
1285     virtio_init(VIRTIO_DEVICE(n), "virtio-net", VIRTIO_ID_NET,
1286                                   n->config_size);
1287
1288     n->max_queues = MAX(n->nic_conf.queues, 1);
1289     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1290     n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1291     n->curr_queues = 1;
1292     n->vqs[0].n = n;
1293     n->tx_timeout = n->net_conf.txtimer;
1294
1295     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1296                        && strcmp(n->net_conf.tx, "bh")) {
1297         error_report("virtio-net: "
1298                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1299                      n->net_conf.tx);
1300         error_report("Defaulting to \"bh\"");
1301     }
1302
1303     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1304         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1305                                            virtio_net_handle_tx_timer);
1306         n->vqs[0].tx_timer = qemu_new_timer_ns(vm_clock, virtio_net_tx_timer,
1307                                                &n->vqs[0]);
1308     } else {
1309         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1310                                            virtio_net_handle_tx_bh);
1311         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1312     }
1313     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1314     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1315     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1316     n->status = VIRTIO_NET_S_LINK_UP;
1317
1318     n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1319                           object_get_typename(OBJECT(qdev)), qdev->id, n);
1320     peer_test_vnet_hdr(n);
1321     if (peer_has_vnet_hdr(n)) {
1322         for (i = 0; i < n->max_queues; i++) {
1323             tap_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1324         }
1325         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1326     } else {
1327         n->host_hdr_len = 0;
1328     }
1329
1330     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1331
1332     n->vqs[0].tx_waiting = 0;
1333     n->tx_burst = n->net_conf.txburst;
1334     virtio_net_set_mrg_rx_bufs(n, 0);
1335     n->promisc = 1; /* for compatibility */
1336
1337     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1338
1339     n->vlans = g_malloc0(MAX_VLAN >> 3);
1340
1341     n->qdev = qdev;
1342     register_savevm(qdev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1343                     virtio_net_save, virtio_net_load, n);
1344
1345     add_boot_device_path(n->nic_conf.bootindex, qdev, "/ethernet-phy@0");
1346     return 0;
1347 }
1348
1349 static int virtio_net_device_exit(DeviceState *qdev)
1350 {
1351     VirtIONet *n = VIRTIO_NET(qdev);
1352     VirtIODevice *vdev = VIRTIO_DEVICE(qdev);
1353     int i;
1354
1355     /* This will stop vhost backend if appropriate. */
1356     virtio_net_set_status(vdev, 0);
1357
1358     unregister_savevm(qdev, "virtio-net", n);
1359
1360     g_free(n->mac_table.macs);
1361     g_free(n->vlans);
1362
1363     for (i = 0; i < n->max_queues; i++) {
1364         VirtIONetQueue *q = &n->vqs[i];
1365         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1366
1367         qemu_purge_queued_packets(nc);
1368
1369         if (q->tx_timer) {
1370             qemu_del_timer(q->tx_timer);
1371             qemu_free_timer(q->tx_timer);
1372         } else {
1373             qemu_bh_delete(q->tx_bh);
1374         }
1375     }
1376
1377     g_free(n->vqs);
1378     qemu_del_nic(n->nic);
1379     virtio_cleanup(vdev);
1380
1381     return 0;
1382 }
1383
1384 static void virtio_net_instance_init(Object *obj)
1385 {
1386     VirtIONet *n = VIRTIO_NET(obj);
1387
1388     /*
1389      * The default config_size is sizeof(struct virtio_net_config).
1390      * Can be overriden with virtio_net_set_config_size.
1391      */
1392     n->config_size = sizeof(struct virtio_net_config);
1393 }
1394
1395 static Property virtio_net_properties[] = {
1396     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1397     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1398                                                TX_TIMER_INTERVAL),
1399     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1400     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1401     DEFINE_PROP_END_OF_LIST(),
1402 };
1403
1404 static void virtio_net_class_init(ObjectClass *klass, void *data)
1405 {
1406     DeviceClass *dc = DEVICE_CLASS(klass);
1407     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1408     dc->exit = virtio_net_device_exit;
1409     dc->props = virtio_net_properties;
1410     vdc->init = virtio_net_device_init;
1411     vdc->get_config = virtio_net_get_config;
1412     vdc->set_config = virtio_net_set_config;
1413     vdc->get_features = virtio_net_get_features;
1414     vdc->set_features = virtio_net_set_features;
1415     vdc->bad_features = virtio_net_bad_features;
1416     vdc->reset = virtio_net_reset;
1417     vdc->set_status = virtio_net_set_status;
1418     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1419     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1420 }
1421
1422 static const TypeInfo virtio_net_info = {
1423     .name = TYPE_VIRTIO_NET,
1424     .parent = TYPE_VIRTIO_DEVICE,
1425     .instance_size = sizeof(VirtIONet),
1426     .instance_init = virtio_net_instance_init,
1427     .class_init = virtio_net_class_init,
1428 };
1429
1430 static void virtio_register_types(void)
1431 {
1432     type_register_static(&virtio_net_info);
1433 }
1434
1435 type_init(virtio_register_types)