hw/virtio-net.c

   1 /*
   2  * Virtio Network Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "virtio.h"
  15 #include "net.h"
  16 #include "net/checksum.h"
  17 #include "net/tap.h"
  18 #include "qemu-timer.h"
  19 #include "virtio-net.h"
  20
  21 #define VIRTIO_NET_VM_VERSION    11
  22
  23 #define MAC_TABLE_ENTRIES    64
  24 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  25
  26 typedef struct VirtIONet
  27 {
  28     VirtIODevice vdev;
  29     uint8_t mac[ETH_ALEN];
  30     uint16_t status;
  31     VirtQueue *rx_vq;
  32     VirtQueue *tx_vq;
  33     VirtQueue *ctrl_vq;
  34     NICState *nic;
  35     QEMUTimer *tx_timer;
  36     int tx_timer_active;
  37     uint32_t has_vnet_hdr;
  38     uint8_t has_ufo;
  39     struct {
  40         VirtQueueElement elem;
  41         ssize_t len;
  42     } async_tx;
  43     int mergeable_rx_bufs;
  44     uint8_t promisc;
  45     uint8_t allmulti;
  46     uint8_t alluni;
  47     uint8_t nomulti;
  48     uint8_t nouni;
  49     uint8_t nobcast;
  50     struct {
  51         int in_use;
  52         int first_multi;
  53         uint8_t multi_overflow;
  54         uint8_t uni_overflow;
  55         uint8_t *macs;
  56     } mac_table;
  57     uint32_t *vlans;
  58 } VirtIONet;
  59
  60 /* TODO
  61  * - we could suppress RX interrupt if we were so inclined.
  62  */
  63
  64 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  65 {
  66     return (VirtIONet *)vdev;
  67 }
  68
  69 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  70 {
  71     VirtIONet *n = to_virtio_net(vdev);
  72     struct virtio_net_config netcfg;
  73
  74     netcfg.status = n->status;
  75     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  76     memcpy(config, &netcfg, sizeof(netcfg));
  77 }
  78
  79 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  80 {
  81     VirtIONet *n = to_virtio_net(vdev);
  82     struct virtio_net_config netcfg;
  83
  84     memcpy(&netcfg, config, sizeof(netcfg));
  85
  86     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  87         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  88         qemu_format_nic_info_str(&n->nic->nc, n->mac);
  89     }
  90 }
  91
  92 static void virtio_net_set_link_status(VLANClientState *nc)
  93 {
  94     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
  95     uint16_t old_status = n->status;
  96
  97     if (nc->link_down)
  98         n->status &= ~VIRTIO_NET_S_LINK_UP;
  99     else
 100         n->status |= VIRTIO_NET_S_LINK_UP;
 101
 102     if (n->status != old_status)
 103         virtio_notify_config(&n->vdev);
 104 }
 105
 106 static void virtio_net_reset(VirtIODevice *vdev)
 107 {
 108     VirtIONet *n = to_virtio_net(vdev);
 109
 110     /* Reset back to compatibility mode */
 111     n->promisc = 1;
 112     n->allmulti = 0;
 113     n->alluni = 0;
 114     n->nomulti = 0;
 115     n->nouni = 0;
 116     n->nobcast = 0;
 117
 118     /* Flush any MAC and VLAN filter table state */
 119     n->mac_table.in_use = 0;
 120     n->mac_table.first_multi = 0;
 121     n->mac_table.multi_overflow = 0;
 122     n->mac_table.uni_overflow = 0;
 123     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 124     memset(n->vlans, 0, MAX_VLAN >> 3);
 125 }
 126
 127 static int peer_has_vnet_hdr(VirtIONet *n)
 128 {
 129     if (!n->nic->nc.peer)
 130         return 0;
 131
 132     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 133         return 0;
 134
 135     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 136
 137     return n->has_vnet_hdr;
 138 }
 139
 140 static int peer_has_ufo(VirtIONet *n)
 141 {
 142     if (!peer_has_vnet_hdr(n))
 143         return 0;
 144
 145     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 146
 147     return n->has_ufo;
 148 }
 149
 150 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 151 {
 152     VirtIONet *n = to_virtio_net(vdev);
 153
 154     features |= (1 << VIRTIO_NET_F_MAC);
 155
 156     if (peer_has_vnet_hdr(n)) {
 157         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 158     } else {
 159         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 160         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 161         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 162         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 163
 164         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 165         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 166         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 167         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 168     }
 169
 170     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 171         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 172         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 173     }
 174
 175     return features;
 176 }
 177
 178 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 179 {
 180     uint32_t features = 0;
 181
 182     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 183      * but also these: */
 184     features |= (1 << VIRTIO_NET_F_MAC);
 185     features |= (1 << VIRTIO_NET_F_CSUM);
 186     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 187     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 188     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 189
 190     return features;
 191 }
 192
 193 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 194 {
 195     VirtIONet *n = to_virtio_net(vdev);
 196
 197     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 198
 199     if (n->has_vnet_hdr) {
 200         tap_set_offload(n->nic->nc.peer,
 201                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 202                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 203                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 204                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 205                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 206     }
 207 }
 208
 209 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 210                                      VirtQueueElement *elem)
 211 {
 212     uint8_t on;
 213
 214     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 215         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 216         exit(1);
 217     }
 218
 219     on = ldub_p(elem->out_sg[1].iov_base);
 220
 221     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 222         n->promisc = on;
 223     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 224         n->allmulti = on;
 225     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 226         n->alluni = on;
 227     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 228         n->nomulti = on;
 229     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 230         n->nouni = on;
 231     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 232         n->nobcast = on;
 233     else
 234         return VIRTIO_NET_ERR;
 235
 236     return VIRTIO_NET_OK;
 237 }
 238
 239 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 240                                  VirtQueueElement *elem)
 241 {
 242     struct virtio_net_ctrl_mac mac_data;
 243
 244     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 245         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 246         elem->out_sg[2].iov_len < sizeof(mac_data))
 247         return VIRTIO_NET_ERR;
 248
 249     n->mac_table.in_use = 0;
 250     n->mac_table.first_multi = 0;
 251     n->mac_table.uni_overflow = 0;
 252     n->mac_table.multi_overflow = 0;
 253     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 254
 255     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 256
 257     if (sizeof(mac_data.entries) +
 258         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 259         return VIRTIO_NET_ERR;
 260
 261     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 262         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 263                mac_data.entries * ETH_ALEN);
 264         n->mac_table.in_use += mac_data.entries;
 265     } else {
 266         n->mac_table.uni_overflow = 1;
 267     }
 268
 269     n->mac_table.first_multi = n->mac_table.in_use;
 270
 271     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 272
 273     if (sizeof(mac_data.entries) +
 274         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 275         return VIRTIO_NET_ERR;
 276
 277     if (mac_data.entries) {
 278         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 279             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 280                    elem->out_sg[2].iov_base + sizeof(mac_data),
 281                    mac_data.entries * ETH_ALEN);
 282             n->mac_table.in_use += mac_data.entries;
 283         } else {
 284             n->mac_table.multi_overflow = 1;
 285         }
 286     }
 287
 288     return VIRTIO_NET_OK;
 289 }
 290
 291 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 292                                         VirtQueueElement *elem)
 293 {
 294     uint16_t vid;
 295
 296     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 297         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 298         return VIRTIO_NET_ERR;
 299     }
 300
 301     vid = lduw_le_p(elem->out_sg[1].iov_base);
 302
 303     if (vid >= MAX_VLAN)
 304         return VIRTIO_NET_ERR;
 305
 306     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 307         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 308     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 309         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 310     else
 311         return VIRTIO_NET_ERR;
 312
 313     return VIRTIO_NET_OK;
 314 }
 315
 316 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 317 {
 318     VirtIONet *n = to_virtio_net(vdev);
 319     struct virtio_net_ctrl_hdr ctrl;
 320     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 321     VirtQueueElement elem;
 322
 323     while (virtqueue_pop(vq, &elem)) {
 324         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 325             fprintf(stderr, "virtio-net ctrl missing headers\n");
 326             exit(1);
 327         }
 328
 329         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 330             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 331             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 332             exit(1);
 333         }
 334
 335         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 336         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 337
 338         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 339             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 340         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 341             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 342         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 343             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 344
 345         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 346
 347         virtqueue_push(vq, &elem, sizeof(status));
 348         virtio_notify(vdev, vq);
 349     }
 350 }
 351
 352 /* RX */
 353
 354 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 355 {
 356     VirtIONet *n = to_virtio_net(vdev);
 357
 358     qemu_flush_queued_packets(&n->nic->nc);
 359
 360     /* We now have RX buffers, signal to the IO thread to break out of the
 361      * select to re-poll the tap file descriptor */
 362     qemu_notify_event();
 363 }
 364
 365 static int virtio_net_can_receive(VLANClientState *nc)
 366 {
 367     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 368
 369     if (!virtio_queue_ready(n->rx_vq) ||
 370         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 371         return 0;
 372
 373     return 1;
 374 }
 375
 376 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 377 {
 378     if (virtio_queue_empty(n->rx_vq) ||
 379         (n->mergeable_rx_bufs &&
 380          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 381         virtio_queue_set_notification(n->rx_vq, 1);
 382         return 0;
 383     }
 384
 385     virtio_queue_set_notification(n->rx_vq, 0);
 386     return 1;
 387 }
 388
 389 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 390  * it never finds out that the packets don't have valid checksums.  This
 391  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 392  * fix this with Xen but it hasn't appeared in an upstream release of
 393  * dhclient yet.
 394  *
 395  * To avoid breaking existing guests, we catch udp packets and add
 396  * checksums.  This is terrible but it's better than hacking the guest
 397  * kernels.
 398  *
 399  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 400  * we should provide a mechanism to disable it to avoid polluting the host
 401  * cache.
 402  */
 403 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 404                                         const uint8_t *buf, size_t size)
 405 {
 406     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 407         (size > 27 && size < 1500) && /* normal sized MTU */
 408         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 409         (buf[23] == 17) && /* ip.protocol == UDP */
 410         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 411         /* FIXME this cast is evil */
 412         net_checksum_calculate((uint8_t *)buf, size);
 413         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 414     }
 415 }
 416
 417 static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count)
 418 {
 419     int offset, i;
 420
 421     offset = i = 0;
 422     while (offset < count && i < iovcnt) {
 423         int len = MIN(iov[i].iov_len, count - offset);
 424         memcpy(iov[i].iov_base, buf + offset, len);
 425         offset += len;
 426         i++;
 427     }
 428
 429     return offset;
 430 }
 431
 432 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 433                           const void *buf, size_t size, size_t hdr_len)
 434 {
 435     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 436     int offset = 0;
 437
 438     hdr->flags = 0;
 439     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 440
 441     if (n->has_vnet_hdr) {
 442         memcpy(hdr, buf, sizeof(*hdr));
 443         offset = sizeof(*hdr);
 444         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 445     }
 446
 447     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 448      * but we may be passing along a larger header to the guest.
 449      */
 450     iov[0].iov_base += hdr_len;
 451     iov[0].iov_len  -= hdr_len;
 452
 453     return offset;
 454 }
 455
 456 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 457 {
 458     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 459     static const uint8_t vlan[] = {0x81, 0x00};
 460     uint8_t *ptr = (uint8_t *)buf;
 461     int i;
 462
 463     if (n->promisc)
 464         return 1;
 465
 466     if (n->has_vnet_hdr) {
 467         ptr += sizeof(struct virtio_net_hdr);
 468     }
 469
 470     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 471         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 472         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 473             return 0;
 474     }
 475
 476     if (ptr[0] & 1) { // multicast
 477         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 478             return !n->nobcast;
 479         } else if (n->nomulti) {
 480             return 0;
 481         } else if (n->allmulti || n->mac_table.multi_overflow) {
 482             return 1;
 483         }
 484
 485         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 486             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 487                 return 1;
 488             }
 489         }
 490     } else { // unicast
 491         if (n->nouni) {
 492             return 0;
 493         } else if (n->alluni || n->mac_table.uni_overflow) {
 494             return 1;
 495         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 496             return 1;
 497         }
 498
 499         for (i = 0; i < n->mac_table.first_multi; i++) {
 500             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 501                 return 1;
 502             }
 503         }
 504     }
 505
 506     return 0;
 507 }
 508
 509 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 510 {
 511     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 512     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 513     size_t hdr_len, offset, i;
 514
 515     if (!virtio_net_can_receive(&n->nic->nc))
 516         return -1;
 517
 518     if (!virtio_net_has_buffers(n, size))
 519         return 0;
 520
 521     if (!receive_filter(n, buf, size))
 522         return size;
 523
 524     /* hdr_len refers to the header we supply to the guest */
 525     hdr_len = n->mergeable_rx_bufs ?
 526         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 527
 528     offset = i = 0;
 529
 530     while (offset < size) {
 531         VirtQueueElement elem;
 532         int len, total;
 533         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 534
 535         total = 0;
 536
 537         if ((i != 0 && !n->mergeable_rx_bufs) ||
 538             virtqueue_pop(n->rx_vq, &elem) == 0) {
 539             if (i == 0)
 540                 return -1;
 541             fprintf(stderr, "virtio-net truncating packet\n");
 542             exit(1);
 543         }
 544
 545         if (elem.in_num < 1) {
 546             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 547             exit(1);
 548         }
 549
 550         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
 551             fprintf(stderr, "virtio-net header not in first element\n");
 552             exit(1);
 553         }
 554
 555         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 556
 557         if (i == 0) {
 558             if (n->mergeable_rx_bufs)
 559                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 560
 561             offset += receive_header(n, sg, elem.in_num,
 562                                      buf + offset, size - offset, hdr_len);
 563             total += hdr_len;
 564         }
 565
 566         /* copy in packet.  ugh */
 567         len = iov_fill(sg, elem.in_num,
 568                        buf + offset, size - offset);
 569         total += len;
 570
 571         /* signal other side */
 572         virtqueue_fill(n->rx_vq, &elem, total, i++);
 573
 574         offset += len;
 575     }
 576
 577     if (mhdr)
 578         mhdr->num_buffers = i;
 579
 580     virtqueue_flush(n->rx_vq, i);
 581     virtio_notify(&n->vdev, n->rx_vq);
 582
 583     return size;
 584 }
 585
 586 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 587
 588 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 589 {
 590     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 591
 592     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 593     virtio_notify(&n->vdev, n->tx_vq);
 594
 595     n->async_tx.elem.out_num = n->async_tx.len = 0;
 596
 597     virtio_queue_set_notification(n->tx_vq, 1);
 598     virtio_net_flush_tx(n, n->tx_vq);
 599 }
 600
 601 /* TX */
 602 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 603 {
 604     VirtQueueElement elem;
 605
 606     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 607         return;
 608
 609     if (n->async_tx.elem.out_num) {
 610         virtio_queue_set_notification(n->tx_vq, 0);
 611         return;
 612     }
 613
 614     while (virtqueue_pop(vq, &elem)) {
 615         ssize_t ret, len = 0;
 616         unsigned int out_num = elem.out_num;
 617         struct iovec *out_sg = &elem.out_sg[0];
 618         unsigned hdr_len;
 619
 620         /* hdr_len refers to the header received from the guest */
 621         hdr_len = n->mergeable_rx_bufs ?
 622             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 623             sizeof(struct virtio_net_hdr);
 624
 625         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 626             fprintf(stderr, "virtio-net header not in first element\n");
 627             exit(1);
 628         }
 629
 630         /* ignore the header if GSO is not supported */
 631         if (!n->has_vnet_hdr) {
 632             out_num--;
 633             out_sg++;
 634             len += hdr_len;
 635         } else if (n->mergeable_rx_bufs) {
 636             /* tapfd expects a struct virtio_net_hdr */
 637             hdr_len -= sizeof(struct virtio_net_hdr);
 638             out_sg->iov_len -= hdr_len;
 639             len += hdr_len;
 640         }
 641
 642         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 643                                       virtio_net_tx_complete);
 644         if (ret == 0) {
 645             virtio_queue_set_notification(n->tx_vq, 0);
 646             n->async_tx.elem = elem;
 647             n->async_tx.len  = len;
 648             return;
 649         }
 650
 651         len += ret;
 652
 653         virtqueue_push(vq, &elem, len);
 654         virtio_notify(&n->vdev, vq);
 655     }
 656 }
 657
 658 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 659 {
 660     VirtIONet *n = to_virtio_net(vdev);
 661
 662     if (n->tx_timer_active) {
 663         virtio_queue_set_notification(vq, 1);
 664         qemu_del_timer(n->tx_timer);
 665         n->tx_timer_active = 0;
 666         virtio_net_flush_tx(n, vq);
 667     } else {
 668         qemu_mod_timer(n->tx_timer,
 669                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 670         n->tx_timer_active = 1;
 671         virtio_queue_set_notification(vq, 0);
 672     }
 673 }
 674
 675 static void virtio_net_tx_timer(void *opaque)
 676 {
 677     VirtIONet *n = opaque;
 678
 679     n->tx_timer_active = 0;
 680
 681     /* Just in case the driver is not ready on more */
 682     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 683         return;
 684
 685     virtio_queue_set_notification(n->tx_vq, 1);
 686     virtio_net_flush_tx(n, n->tx_vq);
 687 }
 688
 689 static void virtio_net_save(QEMUFile *f, void *opaque)
 690 {
 691     VirtIONet *n = opaque;
 692
 693     virtio_save(&n->vdev, f);
 694
 695     qemu_put_buffer(f, n->mac, ETH_ALEN);
 696     qemu_put_be32(f, n->tx_timer_active);
 697     qemu_put_be32(f, n->mergeable_rx_bufs);
 698     qemu_put_be16(f, n->status);
 699     qemu_put_byte(f, n->promisc);
 700     qemu_put_byte(f, n->allmulti);
 701     qemu_put_be32(f, n->mac_table.in_use);
 702     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 703     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 704     qemu_put_be32(f, n->has_vnet_hdr);
 705     qemu_put_byte(f, n->mac_table.multi_overflow);
 706     qemu_put_byte(f, n->mac_table.uni_overflow);
 707     qemu_put_byte(f, n->alluni);
 708     qemu_put_byte(f, n->nomulti);
 709     qemu_put_byte(f, n->nouni);
 710     qemu_put_byte(f, n->nobcast);
 711     qemu_put_byte(f, n->has_ufo);
 712 }
 713
 714 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 715 {
 716     VirtIONet *n = opaque;
 717     int i;
 718
 719     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 720         return -EINVAL;
 721
 722     virtio_load(&n->vdev, f);
 723
 724     qemu_get_buffer(f, n->mac, ETH_ALEN);
 725     n->tx_timer_active = qemu_get_be32(f);
 726     n->mergeable_rx_bufs = qemu_get_be32(f);
 727
 728     if (version_id >= 3)
 729         n->status = qemu_get_be16(f);
 730
 731     if (version_id >= 4) {
 732         if (version_id < 8) {
 733             n->promisc = qemu_get_be32(f);
 734             n->allmulti = qemu_get_be32(f);
 735         } else {
 736             n->promisc = qemu_get_byte(f);
 737             n->allmulti = qemu_get_byte(f);
 738         }
 739     }
 740
 741     if (version_id >= 5) {
 742         n->mac_table.in_use = qemu_get_be32(f);
 743         /* MAC_TABLE_ENTRIES may be different from the saved image */
 744         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 745             qemu_get_buffer(f, n->mac_table.macs,
 746                             n->mac_table.in_use * ETH_ALEN);
 747         } else if (n->mac_table.in_use) {
 748             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 749             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 750             n->mac_table.in_use = 0;
 751         }
 752     }
 753
 754     if (version_id >= 6)
 755         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 756
 757     if (version_id >= 7) {
 758         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 759             qemu_error("virtio-net: saved image requires vnet_hdr=on\n");
 760             return -1;
 761         }
 762
 763         if (n->has_vnet_hdr) {
 764             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 765             tap_set_offload(n->nic->nc.peer,
 766                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 767                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 768                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 769                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 770                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 771         }
 772     }
 773
 774     if (version_id >= 9) {
 775         n->mac_table.multi_overflow = qemu_get_byte(f);
 776         n->mac_table.uni_overflow = qemu_get_byte(f);
 777     }
 778
 779     if (version_id >= 10) {
 780         n->alluni = qemu_get_byte(f);
 781         n->nomulti = qemu_get_byte(f);
 782         n->nouni = qemu_get_byte(f);
 783         n->nobcast = qemu_get_byte(f);
 784     }
 785
 786     if (version_id >= 11) {
 787         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 788             qemu_error("virtio-net: saved image requires TUN_F_UFO support\n");
 789             return -1;
 790         }
 791     }
 792
 793     /* Find the first multicast entry in the saved MAC filter */
 794     for (i = 0; i < n->mac_table.in_use; i++) {
 795         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 796             break;
 797         }
 798     }
 799     n->mac_table.first_multi = i;
 800
 801     if (n->tx_timer_active) {
 802         qemu_mod_timer(n->tx_timer,
 803                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 804     }
 805
 806     return 0;
 807 }
 808
 809 static void virtio_net_cleanup(VLANClientState *nc)
 810 {
 811     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 812
 813     n->nic = NULL;
 814 }
 815
 816 static NetClientInfo net_virtio_info = {
 817     .type = NET_CLIENT_TYPE_NIC,
 818     .size = sizeof(NICState),
 819     .can_receive = virtio_net_can_receive,
 820     .receive = virtio_net_receive,
 821         .cleanup = virtio_net_cleanup,
 822     .link_status_changed = virtio_net_set_link_status,
 823 };
 824
 825 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
 826 {
 827     VirtIONet *n;
 828     static int virtio_net_id;
 829
 830     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 831                                         sizeof(struct virtio_net_config),
 832                                         sizeof(VirtIONet));
 833
 834     n->vdev.get_config = virtio_net_get_config;
 835     n->vdev.set_config = virtio_net_set_config;
 836     n->vdev.get_features = virtio_net_get_features;
 837     n->vdev.set_features = virtio_net_set_features;
 838     n->vdev.bad_features = virtio_net_bad_features;
 839     n->vdev.reset = virtio_net_reset;
 840     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 841     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 842     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 843     qemu_macaddr_default_if_unset(&conf->macaddr);
 844     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
 845     n->status = VIRTIO_NET_S_LINK_UP;
 846
 847     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
 848
 849     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
 850
 851     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 852     n->tx_timer_active = 0;
 853     n->mergeable_rx_bufs = 0;
 854     n->promisc = 1; /* for compatibility */
 855
 856     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 857
 858     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 859
 860     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 861                     virtio_net_save, virtio_net_load, n);
 862
 863     return &n->vdev;
 864 }
 865
 866 void virtio_net_exit(VirtIODevice *vdev)
 867 {
 868     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 869
 870     qemu_purge_queued_packets(&n->nic->nc);
 871
 872     unregister_savevm("virtio-net", n);
 873
 874     qemu_free(n->mac_table.macs);
 875     qemu_free(n->vlans);
 876
 877     qemu_del_timer(n->tx_timer);
 878     qemu_free_timer(n->tx_timer);
 879
 880     virtio_cleanup(&n->vdev);
 881     qemu_del_vlan_client(&n->nic->nc);
 882 }