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