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 ==
 118         (virtio_net_started(n, status) && !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_apply_guest_offloads(VirtIONet *n)
 363 {
 364     tap_set_offload(qemu_get_subqueue(n->nic, 0)->peer,
 365             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_CSUM)),
 366             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO4)),
 367             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_TSO6)),
 368             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_ECN)),
 369             !!(n->curr_guest_offloads & (1ULL << VIRTIO_NET_F_GUEST_UFO)));
 370 }
 371
 372 static uint64_t virtio_net_guest_offloads_by_features(uint32_t features)
 373 {
 374     static const uint64_t guest_offloads_mask =
 375         (1ULL << VIRTIO_NET_F_GUEST_CSUM) |
 376         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
 377         (1ULL << VIRTIO_NET_F_GUEST_TSO6) |
 378         (1ULL << VIRTIO_NET_F_GUEST_ECN)  |
 379         (1ULL << VIRTIO_NET_F_GUEST_UFO);
 380
 381     return guest_offloads_mask & features;
 382 }
 383
 384 static inline uint64_t virtio_net_supported_guest_offloads(VirtIONet *n)
 385 {
 386     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 387     return virtio_net_guest_offloads_by_features(vdev->guest_features);
 388 }
 389
 390 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 391 {
 392     VirtIONet *n = VIRTIO_NET(vdev);
 393     int i;
 394
 395     virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
 396
 397     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
 398
 399     if (n->has_vnet_hdr) {
 400         n->curr_guest_offloads =
 401             virtio_net_guest_offloads_by_features(features);
 402         virtio_net_apply_guest_offloads(n);
 403     }
 404
 405     for (i = 0;  i < n->max_queues; i++) {
 406         NetClientState *nc = qemu_get_subqueue(n->nic, i);
 407
 408         if (!nc->peer || nc->peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 409             continue;
 410         }
 411         if (!tap_get_vhost_net(nc->peer)) {
 412             continue;
 413         }
 414         vhost_net_ack_features(tap_get_vhost_net(nc->peer), features);
 415     }
 416 }
 417
 418 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 419                                      struct iovec *iov, unsigned int iov_cnt)
 420 {
 421     uint8_t on;
 422     size_t s;
 423
 424     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
 425     if (s != sizeof(on)) {
 426         return VIRTIO_NET_ERR;
 427     }
 428
 429     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
 430         n->promisc = on;
 431     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
 432         n->allmulti = on;
 433     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
 434         n->alluni = on;
 435     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
 436         n->nomulti = on;
 437     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
 438         n->nouni = on;
 439     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
 440         n->nobcast = on;
 441     } else {
 442         return VIRTIO_NET_ERR;
 443     }
 444
 445     return VIRTIO_NET_OK;
 446 }
 447
 448 static int virtio_net_handle_offloads(VirtIONet *n, uint8_t cmd,
 449                                      struct iovec *iov, unsigned int iov_cnt)
 450 {
 451     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 452     uint64_t offloads;
 453     size_t s;
 454
 455     if (!((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features)) {
 456         return VIRTIO_NET_ERR;
 457     }
 458
 459     s = iov_to_buf(iov, iov_cnt, 0, &offloads, sizeof(offloads));
 460     if (s != sizeof(offloads)) {
 461         return VIRTIO_NET_ERR;
 462     }
 463
 464     if (cmd == VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET) {
 465         uint64_t supported_offloads;
 466
 467         if (!n->has_vnet_hdr) {
 468             return VIRTIO_NET_ERR;
 469         }
 470
 471         supported_offloads = virtio_net_supported_guest_offloads(n);
 472         if (offloads & ~supported_offloads) {
 473             return VIRTIO_NET_ERR;
 474         }
 475
 476         n->curr_guest_offloads = offloads;
 477         virtio_net_apply_guest_offloads(n);
 478
 479         return VIRTIO_NET_OK;
 480     } else {
 481         return VIRTIO_NET_ERR;
 482     }
 483 }
 484
 485 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 486                                  struct iovec *iov, unsigned int iov_cnt)
 487 {
 488     struct virtio_net_ctrl_mac mac_data;
 489     size_t s;
 490
 491     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
 492         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
 493             return VIRTIO_NET_ERR;
 494         }
 495         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
 496         assert(s == sizeof(n->mac));
 497         qemu_format_nic_info_str(qemu_get_queue(n->nic), n->mac);
 498         return VIRTIO_NET_OK;
 499     }
 500
 501     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
 502         return VIRTIO_NET_ERR;
 503     }
 504
 505     n->mac_table.in_use = 0;
 506     n->mac_table.first_multi = 0;
 507     n->mac_table.uni_overflow = 0;
 508     n->mac_table.multi_overflow = 0;
 509     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 510
 511     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 512                    sizeof(mac_data.entries));
 513     mac_data.entries = ldl_p(&mac_data.entries);
 514     if (s != sizeof(mac_data.entries)) {
 515         return VIRTIO_NET_ERR;
 516     }
 517     iov_discard_front(&iov, &iov_cnt, s);
 518
 519     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
 520         return VIRTIO_NET_ERR;
 521     }
 522
 523     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 524         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 525                        mac_data.entries * ETH_ALEN);
 526         if (s != mac_data.entries * ETH_ALEN) {
 527             return VIRTIO_NET_ERR;
 528         }
 529         n->mac_table.in_use += mac_data.entries;
 530     } else {
 531         n->mac_table.uni_overflow = 1;
 532     }
 533
 534     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
 535
 536     n->mac_table.first_multi = n->mac_table.in_use;
 537
 538     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 539                    sizeof(mac_data.entries));
 540     mac_data.entries = ldl_p(&mac_data.entries);
 541     if (s != sizeof(mac_data.entries)) {
 542         return VIRTIO_NET_ERR;
 543     }
 544
 545     iov_discard_front(&iov, &iov_cnt, s);
 546
 547     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
 548         return VIRTIO_NET_ERR;
 549     }
 550
 551     if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 552         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 553                        mac_data.entries * ETH_ALEN);
 554         if (s != mac_data.entries * ETH_ALEN) {
 555             return VIRTIO_NET_ERR;
 556         }
 557         n->mac_table.in_use += mac_data.entries;
 558     } else {
 559         n->mac_table.multi_overflow = 1;
 560     }
 561
 562     return VIRTIO_NET_OK;
 563 }
 564
 565 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 566                                         struct iovec *iov, unsigned int iov_cnt)
 567 {
 568     uint16_t vid;
 569     size_t s;
 570
 571     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
 572     vid = lduw_p(&vid);
 573     if (s != sizeof(vid)) {
 574         return VIRTIO_NET_ERR;
 575     }
 576
 577     if (vid >= MAX_VLAN)
 578         return VIRTIO_NET_ERR;
 579
 580     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 581         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 582     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 583         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 584     else
 585         return VIRTIO_NET_ERR;
 586
 587     return VIRTIO_NET_OK;
 588 }
 589
 590 static int virtio_net_handle_mq(VirtIONet *n, uint8_t cmd,
 591                                 struct iovec *iov, unsigned int iov_cnt)
 592 {
 593     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 594     struct virtio_net_ctrl_mq mq;
 595     size_t s;
 596     uint16_t queues;
 597
 598     s = iov_to_buf(iov, iov_cnt, 0, &mq, sizeof(mq));
 599     if (s != sizeof(mq)) {
 600         return VIRTIO_NET_ERR;
 601     }
 602
 603     if (cmd != VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET) {
 604         return VIRTIO_NET_ERR;
 605     }
 606
 607     queues = lduw_p(&mq.virtqueue_pairs);
 608
 609     if (queues < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
 610         queues > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
 611         queues > n->max_queues ||
 612         !n->multiqueue) {
 613         return VIRTIO_NET_ERR;
 614     }
 615
 616     n->curr_queues = queues;
 617     /* stop the backend before changing the number of queues to avoid handling a
 618      * disabled queue */
 619     virtio_net_set_status(vdev, vdev->status);
 620     virtio_net_set_queues(n);
 621
 622     return VIRTIO_NET_OK;
 623 }
 624 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 625 {
 626     VirtIONet *n = VIRTIO_NET(vdev);
 627     struct virtio_net_ctrl_hdr ctrl;
 628     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 629     VirtQueueElement elem;
 630     size_t s;
 631     struct iovec *iov;
 632     unsigned int iov_cnt;
 633
 634     while (virtqueue_pop(vq, &elem)) {
 635         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
 636             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
 637             error_report("virtio-net ctrl missing headers");
 638             exit(1);
 639         }
 640
 641         iov = elem.out_sg;
 642         iov_cnt = elem.out_num;
 643         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
 644         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
 645         if (s != sizeof(ctrl)) {
 646             status = VIRTIO_NET_ERR;
 647         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
 648             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
 649         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
 650             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
 651         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
 652             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
 653         } else if (ctrl.class == VIRTIO_NET_CTRL_MQ) {
 654             status = virtio_net_handle_mq(n, ctrl.cmd, iov, iov_cnt);
 655         } else if (ctrl.class == VIRTIO_NET_CTRL_GUEST_OFFLOADS) {
 656             status = virtio_net_handle_offloads(n, ctrl.cmd, iov, iov_cnt);
 657         }
 658
 659         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
 660         assert(s == sizeof(status));
 661
 662         virtqueue_push(vq, &elem, sizeof(status));
 663         virtio_notify(vdev, vq);
 664     }
 665 }
 666
 667 /* RX */
 668
 669 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 670 {
 671     VirtIONet *n = VIRTIO_NET(vdev);
 672     int queue_index = vq2q(virtio_get_queue_index(vq));
 673
 674     qemu_flush_queued_packets(qemu_get_subqueue(n->nic, queue_index));
 675 }
 676
 677 static int virtio_net_can_receive(NetClientState *nc)
 678 {
 679     VirtIONet *n = qemu_get_nic_opaque(nc);
 680     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 681     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 682
 683     if (!vdev->vm_running) {
 684         return 0;
 685     }
 686
 687     if (nc->queue_index >= n->curr_queues) {
 688         return 0;
 689     }
 690
 691     if (!virtio_queue_ready(q->rx_vq) ||
 692         !(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 693         return 0;
 694     }
 695
 696     return 1;
 697 }
 698
 699 static int virtio_net_has_buffers(VirtIONetQueue *q, int bufsize)
 700 {
 701     VirtIONet *n = q->n;
 702     if (virtio_queue_empty(q->rx_vq) ||
 703         (n->mergeable_rx_bufs &&
 704          !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
 705         virtio_queue_set_notification(q->rx_vq, 1);
 706
 707         /* To avoid a race condition where the guest has made some buffers
 708          * available after the above check but before notification was
 709          * enabled, check for available buffers again.
 710          */
 711         if (virtio_queue_empty(q->rx_vq) ||
 712             (n->mergeable_rx_bufs &&
 713              !virtqueue_avail_bytes(q->rx_vq, bufsize, 0))) {
 714             return 0;
 715         }
 716     }
 717
 718     virtio_queue_set_notification(q->rx_vq, 0);
 719     return 1;
 720 }
 721
 722 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 723  * it never finds out that the packets don't have valid checksums.  This
 724  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 725  * fix this with Xen but it hasn't appeared in an upstream release of
 726  * dhclient yet.
 727  *
 728  * To avoid breaking existing guests, we catch udp packets and add
 729  * checksums.  This is terrible but it's better than hacking the guest
 730  * kernels.
 731  *
 732  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 733  * we should provide a mechanism to disable it to avoid polluting the host
 734  * cache.
 735  */
 736 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 737                                         uint8_t *buf, size_t size)
 738 {
 739     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 740         (size > 27 && size < 1500) && /* normal sized MTU */
 741         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 742         (buf[23] == 17) && /* ip.protocol == UDP */
 743         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 744         net_checksum_calculate(buf, size);
 745         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 746     }
 747 }
 748
 749 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
 750                            const void *buf, size_t size)
 751 {
 752     if (n->has_vnet_hdr) {
 753         /* FIXME this cast is evil */
 754         void *wbuf = (void *)buf;
 755         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
 756                                     size - n->host_hdr_len);
 757         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
 758     } else {
 759         struct virtio_net_hdr hdr = {
 760             .flags = 0,
 761             .gso_type = VIRTIO_NET_HDR_GSO_NONE
 762         };
 763         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
 764     }
 765 }
 766
 767 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 768 {
 769     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 770     static const uint8_t vlan[] = {0x81, 0x00};
 771     uint8_t *ptr = (uint8_t *)buf;
 772     int i;
 773
 774     if (n->promisc)
 775         return 1;
 776
 777     ptr += n->host_hdr_len;
 778
 779     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 780         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 781         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 782             return 0;
 783     }
 784
 785     if (ptr[0] & 1) { // multicast
 786         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 787             return !n->nobcast;
 788         } else if (n->nomulti) {
 789             return 0;
 790         } else if (n->allmulti || n->mac_table.multi_overflow) {
 791             return 1;
 792         }
 793
 794         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 795             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 796                 return 1;
 797             }
 798         }
 799     } else { // unicast
 800         if (n->nouni) {
 801             return 0;
 802         } else if (n->alluni || n->mac_table.uni_overflow) {
 803             return 1;
 804         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 805             return 1;
 806         }
 807
 808         for (i = 0; i < n->mac_table.first_multi; i++) {
 809             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 810                 return 1;
 811             }
 812         }
 813     }
 814
 815     return 0;
 816 }
 817
 818 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 819 {
 820     VirtIONet *n = qemu_get_nic_opaque(nc);
 821     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 822     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 823     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
 824     struct virtio_net_hdr_mrg_rxbuf mhdr;
 825     unsigned mhdr_cnt = 0;
 826     size_t offset, i, guest_offset;
 827
 828     if (!virtio_net_can_receive(nc)) {
 829         return -1;
 830     }
 831
 832     /* hdr_len refers to the header we supply to the guest */
 833     if (!virtio_net_has_buffers(q, size + n->guest_hdr_len - n->host_hdr_len)) {
 834         return 0;
 835     }
 836
 837     if (!receive_filter(n, buf, size))
 838         return size;
 839
 840     offset = i = 0;
 841
 842     while (offset < size) {
 843         VirtQueueElement elem;
 844         int len, total;
 845         const struct iovec *sg = elem.in_sg;
 846
 847         total = 0;
 848
 849         if (virtqueue_pop(q->rx_vq, &elem) == 0) {
 850             if (i == 0)
 851                 return -1;
 852             error_report("virtio-net unexpected empty queue: "
 853                     "i %zd mergeable %d offset %zd, size %zd, "
 854                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 855                     i, n->mergeable_rx_bufs, offset, size,
 856                     n->guest_hdr_len, n->host_hdr_len, vdev->guest_features);
 857             exit(1);
 858         }
 859
 860         if (elem.in_num < 1) {
 861             error_report("virtio-net receive queue contains no in buffers");
 862             exit(1);
 863         }
 864
 865         if (i == 0) {
 866             assert(offset == 0);
 867             if (n->mergeable_rx_bufs) {
 868                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
 869                                     sg, elem.in_num,
 870                                     offsetof(typeof(mhdr), num_buffers),
 871                                     sizeof(mhdr.num_buffers));
 872             }
 873
 874             receive_header(n, sg, elem.in_num, buf, size);
 875             offset = n->host_hdr_len;
 876             total += n->guest_hdr_len;
 877             guest_offset = n->guest_hdr_len;
 878         } else {
 879             guest_offset = 0;
 880         }
 881
 882         /* copy in packet.  ugh */
 883         len = iov_from_buf(sg, elem.in_num, guest_offset,
 884                            buf + offset, size - offset);
 885         total += len;
 886         offset += len;
 887         /* If buffers can't be merged, at this point we
 888          * must have consumed the complete packet.
 889          * Otherwise, drop it. */
 890         if (!n->mergeable_rx_bufs && offset < size) {
 891 #if 0
 892             error_report("virtio-net truncated non-mergeable packet: "
 893                          "i %zd mergeable %d offset %zd, size %zd, "
 894                          "guest hdr len %zd, host hdr len %zd",
 895                          i, n->mergeable_rx_bufs,
 896                          offset, size, n->guest_hdr_len, n->host_hdr_len);
 897 #endif
 898             return size;
 899         }
 900
 901         /* signal other side */
 902         virtqueue_fill(q->rx_vq, &elem, total, i++);
 903     }
 904
 905     if (mhdr_cnt) {
 906         stw_p(&mhdr.num_buffers, i);
 907         iov_from_buf(mhdr_sg, mhdr_cnt,
 908                      0,
 909                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
 910     }
 911
 912     virtqueue_flush(q->rx_vq, i);
 913     virtio_notify(vdev, q->rx_vq);
 914
 915     return size;
 916 }
 917
 918 static int32_t virtio_net_flush_tx(VirtIONetQueue *q);
 919
 920 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
 921 {
 922     VirtIONet *n = qemu_get_nic_opaque(nc);
 923     VirtIONetQueue *q = virtio_net_get_subqueue(nc);
 924     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 925
 926     virtqueue_push(q->tx_vq, &q->async_tx.elem, 0);
 927     virtio_notify(vdev, q->tx_vq);
 928
 929     q->async_tx.elem.out_num = q->async_tx.len = 0;
 930
 931     virtio_queue_set_notification(q->tx_vq, 1);
 932     virtio_net_flush_tx(q);
 933 }
 934
 935 /* TX */
 936 static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
 937 {
 938     VirtIONet *n = q->n;
 939     VirtIODevice *vdev = VIRTIO_DEVICE(n);
 940     VirtQueueElement elem;
 941     int32_t num_packets = 0;
 942     int queue_index = vq2q(virtio_get_queue_index(q->tx_vq));
 943     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 944         return num_packets;
 945     }
 946
 947     assert(vdev->vm_running);
 948
 949     if (q->async_tx.elem.out_num) {
 950         virtio_queue_set_notification(q->tx_vq, 0);
 951         return num_packets;
 952     }
 953
 954     while (virtqueue_pop(q->tx_vq, &elem)) {
 955         ssize_t ret, len;
 956         unsigned int out_num = elem.out_num;
 957         struct iovec *out_sg = &elem.out_sg[0];
 958         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 959
 960         if (out_num < 1) {
 961             error_report("virtio-net header not in first element");
 962             exit(1);
 963         }
 964
 965         /*
 966          * If host wants to see the guest header as is, we can
 967          * pass it on unchanged. Otherwise, copy just the parts
 968          * that host is interested in.
 969          */
 970         assert(n->host_hdr_len <= n->guest_hdr_len);
 971         if (n->host_hdr_len != n->guest_hdr_len) {
 972             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
 973                                        out_sg, out_num,
 974                                        0, n->host_hdr_len);
 975             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
 976                              out_sg, out_num,
 977                              n->guest_hdr_len, -1);
 978             out_num = sg_num;
 979             out_sg = sg;
 980         }
 981
 982         len = n->guest_hdr_len;
 983
 984         ret = qemu_sendv_packet_async(qemu_get_subqueue(n->nic, queue_index),
 985                                       out_sg, out_num, virtio_net_tx_complete);
 986         if (ret == 0) {
 987             virtio_queue_set_notification(q->tx_vq, 0);
 988             q->async_tx.elem = elem;
 989             q->async_tx.len  = len;
 990             return -EBUSY;
 991         }
 992
 993         len += ret;
 994
 995         virtqueue_push(q->tx_vq, &elem, 0);
 996         virtio_notify(vdev, q->tx_vq);
 997
 998         if (++num_packets >= n->tx_burst) {
 999             break;
1000         }
1001     }
1002     return num_packets;
1003 }
1004
1005 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
1006 {
1007     VirtIONet *n = VIRTIO_NET(vdev);
1008     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1009
1010     /* This happens when device was stopped but VCPU wasn't. */
1011     if (!vdev->vm_running) {
1012         q->tx_waiting = 1;
1013         return;
1014     }
1015
1016     if (q->tx_waiting) {
1017         virtio_queue_set_notification(vq, 1);
1018         qemu_del_timer(q->tx_timer);
1019         q->tx_waiting = 0;
1020         virtio_net_flush_tx(q);
1021     } else {
1022         qemu_mod_timer(q->tx_timer,
1023                        qemu_get_clock_ns(vm_clock) + n->tx_timeout);
1024         q->tx_waiting = 1;
1025         virtio_queue_set_notification(vq, 0);
1026     }
1027 }
1028
1029 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
1030 {
1031     VirtIONet *n = VIRTIO_NET(vdev);
1032     VirtIONetQueue *q = &n->vqs[vq2q(virtio_get_queue_index(vq))];
1033
1034     if (unlikely(q->tx_waiting)) {
1035         return;
1036     }
1037     q->tx_waiting = 1;
1038     /* This happens when device was stopped but VCPU wasn't. */
1039     if (!vdev->vm_running) {
1040         return;
1041     }
1042     virtio_queue_set_notification(vq, 0);
1043     qemu_bh_schedule(q->tx_bh);
1044 }
1045
1046 static void virtio_net_tx_timer(void *opaque)
1047 {
1048     VirtIONetQueue *q = opaque;
1049     VirtIONet *n = q->n;
1050     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1051     assert(vdev->vm_running);
1052
1053     q->tx_waiting = 0;
1054
1055     /* Just in case the driver is not ready on more */
1056     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1057         return;
1058     }
1059
1060     virtio_queue_set_notification(q->tx_vq, 1);
1061     virtio_net_flush_tx(q);
1062 }
1063
1064 static void virtio_net_tx_bh(void *opaque)
1065 {
1066     VirtIONetQueue *q = opaque;
1067     VirtIONet *n = q->n;
1068     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1069     int32_t ret;
1070
1071     assert(vdev->vm_running);
1072
1073     q->tx_waiting = 0;
1074
1075     /* Just in case the driver is not ready on more */
1076     if (unlikely(!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))) {
1077         return;
1078     }
1079
1080     ret = virtio_net_flush_tx(q);
1081     if (ret == -EBUSY) {
1082         return; /* Notification re-enable handled by tx_complete */
1083     }
1084
1085     /* If we flush a full burst of packets, assume there are
1086      * more coming and immediately reschedule */
1087     if (ret >= n->tx_burst) {
1088         qemu_bh_schedule(q->tx_bh);
1089         q->tx_waiting = 1;
1090         return;
1091     }
1092
1093     /* If less than a full burst, re-enable notification and flush
1094      * anything that may have come in while we weren't looking.  If
1095      * we find something, assume the guest is still active and reschedule */
1096     virtio_queue_set_notification(q->tx_vq, 1);
1097     if (virtio_net_flush_tx(q) > 0) {
1098         virtio_queue_set_notification(q->tx_vq, 0);
1099         qemu_bh_schedule(q->tx_bh);
1100         q->tx_waiting = 1;
1101     }
1102 }
1103
1104 static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
1105 {
1106     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1107     int i, max = multiqueue ? n->max_queues : 1;
1108
1109     n->multiqueue = multiqueue;
1110
1111     for (i = 2; i <= n->max_queues * 2 + 1; i++) {
1112         virtio_del_queue(vdev, i);
1113     }
1114
1115     for (i = 1; i < max; i++) {
1116         n->vqs[i].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1117         if (n->vqs[i].tx_timer) {
1118             n->vqs[i].tx_vq =
1119                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_timer);
1120             n->vqs[i].tx_timer = qemu_new_timer_ns(vm_clock,
1121                                                    virtio_net_tx_timer,
1122                                                    &n->vqs[i]);
1123         } else {
1124             n->vqs[i].tx_vq =
1125                 virtio_add_queue(vdev, 256, virtio_net_handle_tx_bh);
1126             n->vqs[i].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[i]);
1127         }
1128
1129         n->vqs[i].tx_waiting = 0;
1130         n->vqs[i].n = n;
1131     }
1132
1133     /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
1134      * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
1135      * breaking them.
1136      */
1137     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1138
1139     virtio_net_set_queues(n);
1140 }
1141
1142 static void virtio_net_save(QEMUFile *f, void *opaque)
1143 {
1144     int i;
1145     VirtIONet *n = opaque;
1146     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1147
1148     /* At this point, backend must be stopped, otherwise
1149      * it might keep writing to memory. */
1150     assert(!n->vhost_started);
1151     virtio_save(vdev, f);
1152
1153     qemu_put_buffer(f, n->mac, ETH_ALEN);
1154     qemu_put_be32(f, n->vqs[0].tx_waiting);
1155     qemu_put_be32(f, n->mergeable_rx_bufs);
1156     qemu_put_be16(f, n->status);
1157     qemu_put_byte(f, n->promisc);
1158     qemu_put_byte(f, n->allmulti);
1159     qemu_put_be32(f, n->mac_table.in_use);
1160     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
1161     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1162     qemu_put_be32(f, n->has_vnet_hdr);
1163     qemu_put_byte(f, n->mac_table.multi_overflow);
1164     qemu_put_byte(f, n->mac_table.uni_overflow);
1165     qemu_put_byte(f, n->alluni);
1166     qemu_put_byte(f, n->nomulti);
1167     qemu_put_byte(f, n->nouni);
1168     qemu_put_byte(f, n->nobcast);
1169     qemu_put_byte(f, n->has_ufo);
1170     if (n->max_queues > 1) {
1171         qemu_put_be16(f, n->max_queues);
1172         qemu_put_be16(f, n->curr_queues);
1173         for (i = 1; i < n->curr_queues; i++) {
1174             qemu_put_be32(f, n->vqs[i].tx_waiting);
1175         }
1176     }
1177
1178     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1179         qemu_put_be64(f, n->curr_guest_offloads);
1180     }
1181 }
1182
1183 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
1184 {
1185     VirtIONet *n = opaque;
1186     VirtIODevice *vdev = VIRTIO_DEVICE(n);
1187     int ret, i, link_down;
1188
1189     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
1190         return -EINVAL;
1191
1192     ret = virtio_load(vdev, f);
1193     if (ret) {
1194         return ret;
1195     }
1196
1197     qemu_get_buffer(f, n->mac, ETH_ALEN);
1198     n->vqs[0].tx_waiting = qemu_get_be32(f);
1199
1200     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
1201
1202     if (version_id >= 3)
1203         n->status = qemu_get_be16(f);
1204
1205     if (version_id >= 4) {
1206         if (version_id < 8) {
1207             n->promisc = qemu_get_be32(f);
1208             n->allmulti = qemu_get_be32(f);
1209         } else {
1210             n->promisc = qemu_get_byte(f);
1211             n->allmulti = qemu_get_byte(f);
1212         }
1213     }
1214
1215     if (version_id >= 5) {
1216         n->mac_table.in_use = qemu_get_be32(f);
1217         /* MAC_TABLE_ENTRIES may be different from the saved image */
1218         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
1219             qemu_get_buffer(f, n->mac_table.macs,
1220                             n->mac_table.in_use * ETH_ALEN);
1221         } else if (n->mac_table.in_use) {
1222             uint8_t *buf = g_malloc0(n->mac_table.in_use);
1223             qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
1224             g_free(buf);
1225             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
1226             n->mac_table.in_use = 0;
1227         }
1228     }
1229
1230     if (version_id >= 6)
1231         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
1232
1233     if (version_id >= 7) {
1234         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
1235             error_report("virtio-net: saved image requires vnet_hdr=on");
1236             return -1;
1237         }
1238     }
1239
1240     if (version_id >= 9) {
1241         n->mac_table.multi_overflow = qemu_get_byte(f);
1242         n->mac_table.uni_overflow = qemu_get_byte(f);
1243     }
1244
1245     if (version_id >= 10) {
1246         n->alluni = qemu_get_byte(f);
1247         n->nomulti = qemu_get_byte(f);
1248         n->nouni = qemu_get_byte(f);
1249         n->nobcast = qemu_get_byte(f);
1250     }
1251
1252     if (version_id >= 11) {
1253         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1254             error_report("virtio-net: saved image requires TUN_F_UFO support");
1255             return -1;
1256         }
1257     }
1258
1259     if (n->max_queues > 1) {
1260         if (n->max_queues != qemu_get_be16(f)) {
1261             error_report("virtio-net: different max_queues ");
1262             return -1;
1263         }
1264
1265         n->curr_queues = qemu_get_be16(f);
1266         for (i = 1; i < n->curr_queues; i++) {
1267             n->vqs[i].tx_waiting = qemu_get_be32(f);
1268         }
1269     }
1270
1271     if ((1 << VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) & vdev->guest_features) {
1272         n->curr_guest_offloads = qemu_get_be64(f);
1273     } else {
1274         n->curr_guest_offloads = virtio_net_supported_guest_offloads(n);
1275     }
1276
1277     if (peer_has_vnet_hdr(n)) {
1278         virtio_net_apply_guest_offloads(n);
1279     }
1280
1281     virtio_net_set_queues(n);
1282
1283     /* Find the first multicast entry in the saved MAC filter */
1284     for (i = 0; i < n->mac_table.in_use; i++) {
1285         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1286             break;
1287         }
1288     }
1289     n->mac_table.first_multi = i;
1290
1291     /* nc.link_down can't be migrated, so infer link_down according
1292      * to link status bit in n->status */
1293     link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1294     for (i = 0; i < n->max_queues; i++) {
1295         qemu_get_subqueue(n->nic, i)->link_down = link_down;
1296     }
1297
1298     return 0;
1299 }
1300
1301 static void virtio_net_cleanup(NetClientState *nc)
1302 {
1303     VirtIONet *n = qemu_get_nic_opaque(nc);
1304
1305     n->nic = NULL;
1306 }
1307
1308 static NetClientInfo net_virtio_info = {
1309     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1310     .size = sizeof(NICState),
1311     .can_receive = virtio_net_can_receive,
1312     .receive = virtio_net_receive,
1313         .cleanup = virtio_net_cleanup,
1314     .link_status_changed = virtio_net_set_link_status,
1315 };
1316
1317 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1318 {
1319     VirtIONet *n = VIRTIO_NET(vdev);
1320     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1321     assert(n->vhost_started);
1322     return vhost_net_virtqueue_pending(tap_get_vhost_net(nc->peer), idx);
1323 }
1324
1325 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1326                                            bool mask)
1327 {
1328     VirtIONet *n = VIRTIO_NET(vdev);
1329     NetClientState *nc = qemu_get_subqueue(n->nic, vq2q(idx));
1330     assert(n->vhost_started);
1331     vhost_net_virtqueue_mask(tap_get_vhost_net(nc->peer),
1332                              vdev, idx, mask);
1333 }
1334
1335 void virtio_net_set_config_size(VirtIONet *n, uint32_t host_features)
1336 {
1337     int i, config_size = 0;
1338     host_features |= (1 << VIRTIO_NET_F_MAC);
1339     for (i = 0; feature_sizes[i].flags != 0; i++) {
1340         if (host_features & feature_sizes[i].flags) {
1341             config_size = MAX(feature_sizes[i].end, config_size);
1342         }
1343     }
1344     n->config_size = config_size;
1345 }
1346
1347 void virtio_net_set_netclient_name(VirtIONet *n, const char *name,
1348                                    const char *type)
1349 {
1350     /*
1351      * The name can be NULL, the netclient name will be type.x.
1352      */
1353     assert(type != NULL);
1354
1355     if (n->netclient_name) {
1356         g_free(n->netclient_name);
1357         n->netclient_name = NULL;
1358     }
1359     if (n->netclient_type) {
1360         g_free(n->netclient_type);
1361         n->netclient_type = NULL;
1362     }
1363
1364     if (name != NULL) {
1365         n->netclient_name = g_strdup(name);
1366     }
1367     n->netclient_type = g_strdup(type);
1368 }
1369
1370 static int virtio_net_device_init(VirtIODevice *vdev)
1371 {
1372     int i;
1373
1374     DeviceState *qdev = DEVICE(vdev);
1375     VirtIONet *n = VIRTIO_NET(vdev);
1376
1377     virtio_init(VIRTIO_DEVICE(n), "virtio-net", VIRTIO_ID_NET,
1378                                   n->config_size);
1379
1380     n->max_queues = MAX(n->nic_conf.queues, 1);
1381     n->vqs = g_malloc0(sizeof(VirtIONetQueue) * n->max_queues);
1382     n->vqs[0].rx_vq = virtio_add_queue(vdev, 256, virtio_net_handle_rx);
1383     n->curr_queues = 1;
1384     n->vqs[0].n = n;
1385     n->tx_timeout = n->net_conf.txtimer;
1386
1387     if (n->net_conf.tx && strcmp(n->net_conf.tx, "timer")
1388                        && strcmp(n->net_conf.tx, "bh")) {
1389         error_report("virtio-net: "
1390                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1391                      n->net_conf.tx);
1392         error_report("Defaulting to \"bh\"");
1393     }
1394
1395     if (n->net_conf.tx && !strcmp(n->net_conf.tx, "timer")) {
1396         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1397                                            virtio_net_handle_tx_timer);
1398         n->vqs[0].tx_timer = qemu_new_timer_ns(vm_clock, virtio_net_tx_timer,
1399                                                &n->vqs[0]);
1400     } else {
1401         n->vqs[0].tx_vq = virtio_add_queue(vdev, 256,
1402                                            virtio_net_handle_tx_bh);
1403         n->vqs[0].tx_bh = qemu_bh_new(virtio_net_tx_bh, &n->vqs[0]);
1404     }
1405     n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
1406     qemu_macaddr_default_if_unset(&n->nic_conf.macaddr);
1407     memcpy(&n->mac[0], &n->nic_conf.macaddr, sizeof(n->mac));
1408     n->status = VIRTIO_NET_S_LINK_UP;
1409
1410     if (n->netclient_type) {
1411         /*
1412          * Happen when virtio_net_set_netclient_name has been called.
1413          */
1414         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1415                               n->netclient_type, n->netclient_name, n);
1416     } else {
1417         n->nic = qemu_new_nic(&net_virtio_info, &n->nic_conf,
1418                               object_get_typename(OBJECT(qdev)), qdev->id, n);
1419     }
1420
1421     peer_test_vnet_hdr(n);
1422     if (peer_has_vnet_hdr(n)) {
1423         for (i = 0; i < n->max_queues; i++) {
1424             tap_using_vnet_hdr(qemu_get_subqueue(n->nic, i)->peer, true);
1425         }
1426         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1427     } else {
1428         n->host_hdr_len = 0;
1429     }
1430
1431     qemu_format_nic_info_str(qemu_get_queue(n->nic), n->nic_conf.macaddr.a);
1432
1433     n->vqs[0].tx_waiting = 0;
1434     n->tx_burst = n->net_conf.txburst;
1435     virtio_net_set_mrg_rx_bufs(n, 0);
1436     n->promisc = 1; /* for compatibility */
1437
1438     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1439
1440     n->vlans = g_malloc0(MAX_VLAN >> 3);
1441
1442     n->qdev = qdev;
1443     register_savevm(qdev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1444                     virtio_net_save, virtio_net_load, n);
1445
1446     add_boot_device_path(n->nic_conf.bootindex, qdev, "/ethernet-phy@0");
1447     return 0;
1448 }
1449
1450 static int virtio_net_device_exit(DeviceState *qdev)
1451 {
1452     VirtIONet *n = VIRTIO_NET(qdev);
1453     VirtIODevice *vdev = VIRTIO_DEVICE(qdev);
1454     int i;
1455
1456     /* This will stop vhost backend if appropriate. */
1457     virtio_net_set_status(vdev, 0);
1458
1459     unregister_savevm(qdev, "virtio-net", n);
1460
1461     if (n->netclient_name) {
1462         g_free(n->netclient_name);
1463         n->netclient_name = NULL;
1464     }
1465     if (n->netclient_type) {
1466         g_free(n->netclient_type);
1467         n->netclient_type = NULL;
1468     }
1469
1470     g_free(n->mac_table.macs);
1471     g_free(n->vlans);
1472
1473     for (i = 0; i < n->max_queues; i++) {
1474         VirtIONetQueue *q = &n->vqs[i];
1475         NetClientState *nc = qemu_get_subqueue(n->nic, i);
1476
1477         qemu_purge_queued_packets(nc);
1478
1479         if (q->tx_timer) {
1480             qemu_del_timer(q->tx_timer);
1481             qemu_free_timer(q->tx_timer);
1482         } else {
1483             qemu_bh_delete(q->tx_bh);
1484         }
1485     }
1486
1487     g_free(n->vqs);
1488     qemu_del_nic(n->nic);
1489     virtio_cleanup(vdev);
1490
1491     return 0;
1492 }
1493
1494 static void virtio_net_instance_init(Object *obj)
1495 {
1496     VirtIONet *n = VIRTIO_NET(obj);
1497
1498     /*
1499      * The default config_size is sizeof(struct virtio_net_config).
1500      * Can be overriden with virtio_net_set_config_size.
1501      */
1502     n->config_size = sizeof(struct virtio_net_config);
1503 }
1504
1505 static Property virtio_net_properties[] = {
1506     DEFINE_NIC_PROPERTIES(VirtIONet, nic_conf),
1507     DEFINE_PROP_UINT32("x-txtimer", VirtIONet, net_conf.txtimer,
1508                                                TX_TIMER_INTERVAL),
1509     DEFINE_PROP_INT32("x-txburst", VirtIONet, net_conf.txburst, TX_BURST),
1510     DEFINE_PROP_STRING("tx", VirtIONet, net_conf.tx),
1511     DEFINE_PROP_END_OF_LIST(),
1512 };
1513
1514 static void virtio_net_class_init(ObjectClass *klass, void *data)
1515 {
1516     DeviceClass *dc = DEVICE_CLASS(klass);
1517     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1518     dc->exit = virtio_net_device_exit;
1519     dc->props = virtio_net_properties;
1520     vdc->init = virtio_net_device_init;
1521     vdc->get_config = virtio_net_get_config;
1522     vdc->set_config = virtio_net_set_config;
1523     vdc->get_features = virtio_net_get_features;
1524     vdc->set_features = virtio_net_set_features;
1525     vdc->bad_features = virtio_net_bad_features;
1526     vdc->reset = virtio_net_reset;
1527     vdc->set_status = virtio_net_set_status;
1528     vdc->guest_notifier_mask = virtio_net_guest_notifier_mask;
1529     vdc->guest_notifier_pending = virtio_net_guest_notifier_pending;
1530 }
1531
1532 static const TypeInfo virtio_net_info = {
1533     .name = TYPE_VIRTIO_NET,
1534     .parent = TYPE_VIRTIO_DEVICE,
1535     .instance_size = sizeof(VirtIONet),
1536     .instance_init = virtio_net_instance_init,
1537     .class_init = virtio_net_class_init,
1538 };
1539
1540 static void virtio_register_types(void)
1541 {
1542     type_register_static(&virtio_net_info);
1543 }
1544
1545 type_init(virtio_register_types)