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