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