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 "qemu-timer.h"
  17 #include "virtio-net.h"
  18
  19 #define VIRTIO_NET_VM_VERSION    11
  20
  21 #define MAC_TABLE_ENTRIES    64
  22 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  23
  24 typedef struct VirtIONet
  25 {
  26     VirtIODevice vdev;
  27     uint8_t mac[ETH_ALEN];
  28     uint16_t status;
  29     VirtQueue *rx_vq;
  30     VirtQueue *tx_vq;
  31     VirtQueue *ctrl_vq;
  32     VLANClientState *vc;
  33     QEMUTimer *tx_timer;
  34     int tx_timer_active;
  35     uint32_t has_vnet_hdr;
  36     uint8_t has_ufo;
  37     struct {
  38         VirtQueueElement elem;
  39         ssize_t len;
  40     } async_tx;
  41     int mergeable_rx_bufs;
  42     uint8_t promisc;
  43     uint8_t allmulti;
  44     uint8_t alluni;
  45     uint8_t nomulti;
  46     uint8_t nouni;
  47     uint8_t nobcast;
  48     struct {
  49         int in_use;
  50         int first_multi;
  51         uint8_t multi_overflow;
  52         uint8_t uni_overflow;
  53         uint8_t *macs;
  54     } mac_table;
  55     uint32_t *vlans;
  56 } VirtIONet;
  57
  58 /* TODO
  59  * - we could suppress RX interrupt if we were so inclined.
  60  */
  61
  62 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  63 {
  64     return (VirtIONet *)vdev;
  65 }
  66
  67 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  68 {
  69     VirtIONet *n = to_virtio_net(vdev);
  70     struct virtio_net_config netcfg;
  71
  72     netcfg.status = n->status;
  73     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  74     memcpy(config, &netcfg, sizeof(netcfg));
  75 }
  76
  77 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  78 {
  79     VirtIONet *n = to_virtio_net(vdev);
  80     struct virtio_net_config netcfg;
  81
  82     memcpy(&netcfg, config, sizeof(netcfg));
  83
  84     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  85         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  86         qemu_format_nic_info_str(n->vc, n->mac);
  87     }
  88 }
  89
  90 static void virtio_net_set_link_status(VLANClientState *vc)
  91 {
  92     VirtIONet *n = vc->opaque;
  93     uint16_t old_status = n->status;
  94
  95     if (vc->link_down)
  96         n->status &= ~VIRTIO_NET_S_LINK_UP;
  97     else
  98         n->status |= VIRTIO_NET_S_LINK_UP;
  99
 100     if (n->status != old_status)
 101         virtio_notify_config(&n->vdev);
 102 }
 103
 104 static void virtio_net_reset(VirtIODevice *vdev)
 105 {
 106     VirtIONet *n = to_virtio_net(vdev);
 107
 108     /* Reset back to compatibility mode */
 109     n->promisc = 1;
 110     n->allmulti = 0;
 111     n->alluni = 0;
 112     n->nomulti = 0;
 113     n->nouni = 0;
 114     n->nobcast = 0;
 115
 116     /* Flush any MAC and VLAN filter table state */
 117     n->mac_table.in_use = 0;
 118     n->mac_table.first_multi = 0;
 119     n->mac_table.multi_overflow = 0;
 120     n->mac_table.uni_overflow = 0;
 121     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 122     memset(n->vlans, 0, MAX_VLAN >> 3);
 123 }
 124
 125 static int peer_has_vnet_hdr(VirtIONet *n)
 126 {
 127     if (!n->vc->peer)
 128         return 0;
 129
 130     if (n->vc->peer->type != NET_CLIENT_TYPE_TAP)
 131         return 0;
 132
 133     n->has_vnet_hdr = tap_has_vnet_hdr(n->vc->peer);
 134
 135     return n->has_vnet_hdr;
 136 }
 137
 138 static int peer_has_ufo(VirtIONet *n)
 139 {
 140     if (!peer_has_vnet_hdr(n))
 141         return 0;
 142
 143     n->has_ufo = tap_has_ufo(n->vc->peer);
 144
 145     return n->has_ufo;
 146 }
 147
 148 static uint32_t virtio_net_get_features(VirtIODevice *vdev)
 149 {
 150     VirtIONet *n = to_virtio_net(vdev);
 151     uint32_t features = (1 << VIRTIO_NET_F_MAC) |
 152                         (1 << VIRTIO_NET_F_MRG_RXBUF) |
 153                         (1 << VIRTIO_NET_F_STATUS) |
 154                         (1 << VIRTIO_NET_F_CTRL_VQ) |
 155                         (1 << VIRTIO_NET_F_CTRL_RX) |
 156                         (1 << VIRTIO_NET_F_CTRL_VLAN) |
 157                         (1 << VIRTIO_NET_F_CTRL_RX_EXTRA);
 158
 159     if (peer_has_vnet_hdr(n)) {
 160         tap_using_vnet_hdr(n->vc->peer, 1);
 161
 162         features |= (1 << VIRTIO_NET_F_CSUM);
 163         features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 164         features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 165         features |= (1 << VIRTIO_NET_F_HOST_ECN);
 166
 167         features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
 168         features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
 169         features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
 170         features |= (1 << VIRTIO_NET_F_GUEST_ECN);
 171
 172         if (peer_has_ufo(n)) {
 173             features |= (1 << VIRTIO_NET_F_GUEST_UFO);
 174             features |= (1 << VIRTIO_NET_F_HOST_UFO);
 175         }
 176     }
 177
 178     return features;
 179 }
 180
 181 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 182 {
 183     uint32_t features = 0;
 184
 185     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 186      * but also these: */
 187     features |= (1 << VIRTIO_NET_F_MAC);
 188     features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
 189     features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
 190     features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
 191     features |= (1 << VIRTIO_NET_F_GUEST_ECN);
 192
 193     return features & virtio_net_get_features(vdev);
 194 }
 195
 196 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 197 {
 198     VirtIONet *n = to_virtio_net(vdev);
 199
 200     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 201
 202     if (n->has_vnet_hdr) {
 203         tap_set_offload(n->vc->peer,
 204                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 205                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 206                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 207                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 208                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 209     }
 210 }
 211
 212 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 213                                      VirtQueueElement *elem)
 214 {
 215     uint8_t on;
 216
 217     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 218         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 219         exit(1);
 220     }
 221
 222     on = ldub_p(elem->out_sg[1].iov_base);
 223
 224     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 225         n->promisc = on;
 226     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 227         n->allmulti = on;
 228     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 229         n->alluni = on;
 230     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 231         n->nomulti = on;
 232     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 233         n->nouni = on;
 234     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 235         n->nobcast = on;
 236     else
 237         return VIRTIO_NET_ERR;
 238
 239     return VIRTIO_NET_OK;
 240 }
 241
 242 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 243                                  VirtQueueElement *elem)
 244 {
 245     struct virtio_net_ctrl_mac mac_data;
 246
 247     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 248         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 249         elem->out_sg[2].iov_len < sizeof(mac_data))
 250         return VIRTIO_NET_ERR;
 251
 252     n->mac_table.in_use = 0;
 253     n->mac_table.first_multi = 0;
 254     n->mac_table.uni_overflow = 0;
 255     n->mac_table.multi_overflow = 0;
 256     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 257
 258     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 259
 260     if (sizeof(mac_data.entries) +
 261         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 262         return VIRTIO_NET_ERR;
 263
 264     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 265         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 266                mac_data.entries * ETH_ALEN);
 267         n->mac_table.in_use += mac_data.entries;
 268     } else {
 269         n->mac_table.uni_overflow = 1;
 270     }
 271
 272     n->mac_table.first_multi = n->mac_table.in_use;
 273
 274     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 275
 276     if (sizeof(mac_data.entries) +
 277         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 278         return VIRTIO_NET_ERR;
 279
 280     if (mac_data.entries) {
 281         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 282             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 283                    elem->out_sg[2].iov_base + sizeof(mac_data),
 284                    mac_data.entries * ETH_ALEN);
 285             n->mac_table.in_use += mac_data.entries;
 286         } else {
 287             n->mac_table.multi_overflow = 1;
 288         }
 289     }
 290
 291     return VIRTIO_NET_OK;
 292 }
 293
 294 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 295                                         VirtQueueElement *elem)
 296 {
 297     uint16_t vid;
 298
 299     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 300         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 301         return VIRTIO_NET_ERR;
 302     }
 303
 304     vid = lduw_le_p(elem->out_sg[1].iov_base);
 305
 306     if (vid >= MAX_VLAN)
 307         return VIRTIO_NET_ERR;
 308
 309     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 310         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 311     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 312         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 313     else
 314         return VIRTIO_NET_ERR;
 315
 316     return VIRTIO_NET_OK;
 317 }
 318
 319 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 320 {
 321     VirtIONet *n = to_virtio_net(vdev);
 322     struct virtio_net_ctrl_hdr ctrl;
 323     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 324     VirtQueueElement elem;
 325
 326     while (virtqueue_pop(vq, &elem)) {
 327         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 328             fprintf(stderr, "virtio-net ctrl missing headers\n");
 329             exit(1);
 330         }
 331
 332         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 333             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 334             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 335             exit(1);
 336         }
 337
 338         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 339         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 340
 341         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 342             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 343         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 344             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 345         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 346             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 347
 348         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 349
 350         virtqueue_push(vq, &elem, sizeof(status));
 351         virtio_notify(vdev, vq);
 352     }
 353 }
 354
 355 /* RX */
 356
 357 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 358 {
 359     VirtIONet *n = to_virtio_net(vdev);
 360
 361     qemu_flush_queued_packets(n->vc);
 362
 363     /* We now have RX buffers, signal to the IO thread to break out of the
 364      * select to re-poll the tap file descriptor */
 365     qemu_notify_event();
 366 }
 367
 368 static int do_virtio_net_can_receive(VirtIONet *n, int bufsize)
 369 {
 370     if (!virtio_queue_ready(n->rx_vq) ||
 371         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 372         return 0;
 373
 374     if (virtio_queue_empty(n->rx_vq) ||
 375         (n->mergeable_rx_bufs &&
 376          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 377         virtio_queue_set_notification(n->rx_vq, 1);
 378         return 0;
 379     }
 380
 381     virtio_queue_set_notification(n->rx_vq, 0);
 382     return 1;
 383 }
 384
 385 static int virtio_net_can_receive(VLANClientState *vc)
 386 {
 387     VirtIONet *n = vc->opaque;
 388
 389     return do_virtio_net_can_receive(n, VIRTIO_NET_MAX_BUFSIZE);
 390 }
 391
 392 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 393  * it never finds out that the packets don't have valid checksums.  This
 394  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 395  * fix this with Xen but it hasn't appeared in an upstream release of
 396  * dhclient yet.
 397  *
 398  * To avoid breaking existing guests, we catch udp packets and add
 399  * checksums.  This is terrible but it's better than hacking the guest
 400  * kernels.
 401  *
 402  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 403  * we should provide a mechanism to disable it to avoid polluting the host
 404  * cache.
 405  */
 406 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 407                                         const uint8_t *buf, size_t size)
 408 {
 409     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 410         (size > 27 && size < 1500) && /* normal sized MTU */
 411         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 412         (buf[23] == 17) && /* ip.protocol == UDP */
 413         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 414         /* FIXME this cast is evil */
 415         net_checksum_calculate((uint8_t *)buf, size);
 416         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 417     }
 418 }
 419
 420 static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count)
 421 {
 422     int offset, i;
 423
 424     offset = i = 0;
 425     while (offset < count && i < iovcnt) {
 426         int len = MIN(iov[i].iov_len, count - offset);
 427         memcpy(iov[i].iov_base, buf + offset, len);
 428         offset += len;
 429         i++;
 430     }
 431
 432     return offset;
 433 }
 434
 435 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 436                           const void *buf, size_t size, size_t hdr_len)
 437 {
 438     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 439     int offset = 0;
 440
 441     hdr->flags = 0;
 442     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 443
 444     if (n->has_vnet_hdr) {
 445         memcpy(hdr, buf, sizeof(*hdr));
 446         offset = sizeof(*hdr);
 447         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 448     }
 449
 450     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 451      * but we may be passing along a larger header to the guest.
 452      */
 453     iov[0].iov_base += hdr_len;
 454     iov[0].iov_len  -= hdr_len;
 455
 456     return offset;
 457 }
 458
 459 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 460 {
 461     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 462     static const uint8_t vlan[] = {0x81, 0x00};
 463     uint8_t *ptr = (uint8_t *)buf;
 464     int i;
 465
 466     if (n->promisc)
 467         return 1;
 468
 469     if (n->has_vnet_hdr) {
 470         ptr += sizeof(struct virtio_net_hdr);
 471     }
 472
 473     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 474         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 475         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 476             return 0;
 477     }
 478
 479     if (ptr[0] & 1) { // multicast
 480         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 481             return !n->nobcast;
 482         } else if (n->nomulti) {
 483             return 0;
 484         } else if (n->allmulti || n->mac_table.multi_overflow) {
 485             return 1;
 486         }
 487
 488         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 489             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 490                 return 1;
 491             }
 492         }
 493     } else { // unicast
 494         if (n->nouni) {
 495             return 0;
 496         } else if (n->alluni || n->mac_table.uni_overflow) {
 497             return 1;
 498         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 499             return 1;
 500         }
 501
 502         for (i = 0; i < n->mac_table.first_multi; i++) {
 503             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 504                 return 1;
 505             }
 506         }
 507     }
 508
 509     return 0;
 510 }
 511
 512 static ssize_t virtio_net_receive(VLANClientState *vc, const uint8_t *buf, size_t size)
 513 {
 514     VirtIONet *n = vc->opaque;
 515     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 516     size_t hdr_len, offset, i;
 517
 518     if (!do_virtio_net_can_receive(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         len = 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 *vc, ssize_t len)
 589 {
 590     VirtIONet *n = vc->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->vc, 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->vc->peer, 1);
 765             tap_set_offload(n->vc->peer,
 766                             (n->vdev.features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 767                             (n->vdev.features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 768                             (n->vdev.features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 769                             (n->vdev.features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 770                             (n->vdev.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 *vc)
 810 {
 811     VirtIONet *n = vc->opaque;
 812
 813     n->vc = NULL;
 814 }
 815
 816 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
 817 {
 818     VirtIONet *n;
 819     static int virtio_net_id;
 820
 821     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 822                                         sizeof(struct virtio_net_config),
 823                                         sizeof(VirtIONet));
 824
 825     n->vdev.get_config = virtio_net_get_config;
 826     n->vdev.set_config = virtio_net_set_config;
 827     n->vdev.get_features = virtio_net_get_features;
 828     n->vdev.set_features = virtio_net_set_features;
 829     n->vdev.bad_features = virtio_net_bad_features;
 830     n->vdev.reset = virtio_net_reset;
 831     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 832     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 833     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 834     qemu_macaddr_default_if_unset(&conf->macaddr);
 835     n->status = VIRTIO_NET_S_LINK_UP;
 836     n->vc = qemu_new_vlan_client(NET_CLIENT_TYPE_NIC, conf->vlan, conf->peer,
 837                                  dev->info->name, dev->id,
 838                                  virtio_net_can_receive,
 839                                  virtio_net_receive, NULL, NULL,
 840                                  virtio_net_cleanup, n);
 841     n->vc->link_status_changed = virtio_net_set_link_status;
 842
 843     qemu_format_nic_info_str(n->vc, conf->macaddr.a);
 844
 845     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 846     n->tx_timer_active = 0;
 847     n->mergeable_rx_bufs = 0;
 848     n->promisc = 1; /* for compatibility */
 849
 850     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 851
 852     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 853
 854     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 855                     virtio_net_save, virtio_net_load, n);
 856
 857     return &n->vdev;
 858 }
 859
 860 void virtio_net_exit(VirtIODevice *vdev)
 861 {
 862     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 863
 864     qemu_purge_queued_packets(n->vc);
 865
 866     unregister_savevm("virtio-net", n);
 867
 868     qemu_free(n->mac_table.macs);
 869     qemu_free(n->vlans);
 870
 871     qemu_del_timer(n->tx_timer);
 872     qemu_free_timer(n->tx_timer);
 873
 874     virtio_cleanup(&n->vdev);
 875     qemu_del_vlan_client(n->vc);
 876 }