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