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 "iov.h"
  15 #include "virtio.h"
  16 #include "net.h"
  17 #include "net/checksum.h"
  18 #include "net/tap.h"
  19 #include "qemu-error.h"
  20 #include "qemu-timer.h"
  21 #include "virtio-net.h"
  22 #include "vhost_net.h"
  23
  24 #define VIRTIO_NET_VM_VERSION    11
  25
  26 #define MAC_TABLE_ENTRIES    64
  27 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  28
  29 typedef struct VirtIONet
  30 {
  31     VirtIODevice vdev;
  32     uint8_t mac[ETH_ALEN];
  33     uint16_t status;
  34     VirtQueue *rx_vq;
  35     VirtQueue *tx_vq;
  36     VirtQueue *ctrl_vq;
  37     NICState *nic;
  38     QEMUTimer *tx_timer;
  39     int tx_timer_active;
  40     uint32_t has_vnet_hdr;
  41     uint8_t has_ufo;
  42     struct {
  43         VirtQueueElement elem;
  44         ssize_t len;
  45     } async_tx;
  46     int mergeable_rx_bufs;
  47     uint8_t promisc;
  48     uint8_t allmulti;
  49     uint8_t alluni;
  50     uint8_t nomulti;
  51     uint8_t nouni;
  52     uint8_t nobcast;
  53     uint8_t vhost_started;
  54     VMChangeStateEntry *vmstate;
  55     struct {
  56         int in_use;
  57         int first_multi;
  58         uint8_t multi_overflow;
  59         uint8_t uni_overflow;
  60         uint8_t *macs;
  61     } mac_table;
  62     uint32_t *vlans;
  63 } VirtIONet;
  64
  65 /* TODO
  66  * - we could suppress RX interrupt if we were so inclined.
  67  */
  68
  69 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  70 {
  71     return (VirtIONet *)vdev;
  72 }
  73
  74 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  75 {
  76     VirtIONet *n = to_virtio_net(vdev);
  77     struct virtio_net_config netcfg;
  78
  79     netcfg.status = n->status;
  80     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  81     memcpy(config, &netcfg, sizeof(netcfg));
  82 }
  83
  84 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  85 {
  86     VirtIONet *n = to_virtio_net(vdev);
  87     struct virtio_net_config netcfg;
  88
  89     memcpy(&netcfg, config, sizeof(netcfg));
  90
  91     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  92         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  93         qemu_format_nic_info_str(&n->nic->nc, n->mac);
  94     }
  95 }
  96
  97 static void virtio_net_set_link_status(VLANClientState *nc)
  98 {
  99     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 100     uint16_t old_status = n->status;
 101
 102     if (nc->link_down)
 103         n->status &= ~VIRTIO_NET_S_LINK_UP;
 104     else
 105         n->status |= VIRTIO_NET_S_LINK_UP;
 106
 107     if (n->status != old_status)
 108         virtio_notify_config(&n->vdev);
 109 }
 110
 111 static void virtio_net_reset(VirtIODevice *vdev)
 112 {
 113     VirtIONet *n = to_virtio_net(vdev);
 114
 115     /* Reset back to compatibility mode */
 116     n->promisc = 1;
 117     n->allmulti = 0;
 118     n->alluni = 0;
 119     n->nomulti = 0;
 120     n->nouni = 0;
 121     n->nobcast = 0;
 122     if (n->vhost_started) {
 123         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 124         n->vhost_started = 0;
 125     }
 126
 127     /* Flush any MAC and VLAN filter table state */
 128     n->mac_table.in_use = 0;
 129     n->mac_table.first_multi = 0;
 130     n->mac_table.multi_overflow = 0;
 131     n->mac_table.uni_overflow = 0;
 132     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 133     memset(n->vlans, 0, MAX_VLAN >> 3);
 134 }
 135
 136 static int peer_has_vnet_hdr(VirtIONet *n)
 137 {
 138     if (!n->nic->nc.peer)
 139         return 0;
 140
 141     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 142         return 0;
 143
 144     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 145
 146     return n->has_vnet_hdr;
 147 }
 148
 149 static int peer_has_ufo(VirtIONet *n)
 150 {
 151     if (!peer_has_vnet_hdr(n))
 152         return 0;
 153
 154     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 155
 156     return n->has_ufo;
 157 }
 158
 159 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 160 {
 161     VirtIONet *n = to_virtio_net(vdev);
 162
 163     features |= (1 << VIRTIO_NET_F_MAC);
 164
 165     if (peer_has_vnet_hdr(n)) {
 166         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 167     } else {
 168         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 169         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 170         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 171         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 172
 173         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 174         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 175         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 176         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 177     }
 178
 179     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 180         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 181         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 182     }
 183
 184     if (!n->nic->nc.peer ||
 185         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 186         return features;
 187     }
 188     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 189         return features;
 190     }
 191     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 192 }
 193
 194 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 195 {
 196     uint32_t features = 0;
 197
 198     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 199      * but also these: */
 200     features |= (1 << VIRTIO_NET_F_MAC);
 201     features |= (1 << VIRTIO_NET_F_CSUM);
 202     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 203     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 204     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 205
 206     return features;
 207 }
 208
 209 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 210 {
 211     VirtIONet *n = to_virtio_net(vdev);
 212
 213     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 214
 215     if (n->has_vnet_hdr) {
 216         tap_set_offload(n->nic->nc.peer,
 217                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 218                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 219                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 220                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 221                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 222     }
 223     if (!n->nic->nc.peer ||
 224         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 225         return;
 226     }
 227     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 228         return;
 229     }
 230     vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 231 }
 232
 233 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 234                                      VirtQueueElement *elem)
 235 {
 236     uint8_t on;
 237
 238     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 239         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 240         exit(1);
 241     }
 242
 243     on = ldub_p(elem->out_sg[1].iov_base);
 244
 245     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 246         n->promisc = on;
 247     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 248         n->allmulti = on;
 249     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 250         n->alluni = on;
 251     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 252         n->nomulti = on;
 253     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 254         n->nouni = on;
 255     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 256         n->nobcast = on;
 257     else
 258         return VIRTIO_NET_ERR;
 259
 260     return VIRTIO_NET_OK;
 261 }
 262
 263 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 264                                  VirtQueueElement *elem)
 265 {
 266     struct virtio_net_ctrl_mac mac_data;
 267
 268     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 269         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 270         elem->out_sg[2].iov_len < sizeof(mac_data))
 271         return VIRTIO_NET_ERR;
 272
 273     n->mac_table.in_use = 0;
 274     n->mac_table.first_multi = 0;
 275     n->mac_table.uni_overflow = 0;
 276     n->mac_table.multi_overflow = 0;
 277     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 278
 279     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 280
 281     if (sizeof(mac_data.entries) +
 282         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 283         return VIRTIO_NET_ERR;
 284
 285     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 286         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 287                mac_data.entries * ETH_ALEN);
 288         n->mac_table.in_use += mac_data.entries;
 289     } else {
 290         n->mac_table.uni_overflow = 1;
 291     }
 292
 293     n->mac_table.first_multi = n->mac_table.in_use;
 294
 295     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 296
 297     if (sizeof(mac_data.entries) +
 298         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 299         return VIRTIO_NET_ERR;
 300
 301     if (mac_data.entries) {
 302         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 303             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 304                    elem->out_sg[2].iov_base + sizeof(mac_data),
 305                    mac_data.entries * ETH_ALEN);
 306             n->mac_table.in_use += mac_data.entries;
 307         } else {
 308             n->mac_table.multi_overflow = 1;
 309         }
 310     }
 311
 312     return VIRTIO_NET_OK;
 313 }
 314
 315 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 316                                         VirtQueueElement *elem)
 317 {
 318     uint16_t vid;
 319
 320     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 321         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 322         return VIRTIO_NET_ERR;
 323     }
 324
 325     vid = lduw_le_p(elem->out_sg[1].iov_base);
 326
 327     if (vid >= MAX_VLAN)
 328         return VIRTIO_NET_ERR;
 329
 330     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 331         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 332     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 333         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 334     else
 335         return VIRTIO_NET_ERR;
 336
 337     return VIRTIO_NET_OK;
 338 }
 339
 340 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 341 {
 342     VirtIONet *n = to_virtio_net(vdev);
 343     struct virtio_net_ctrl_hdr ctrl;
 344     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 345     VirtQueueElement elem;
 346
 347     while (virtqueue_pop(vq, &elem)) {
 348         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 349             fprintf(stderr, "virtio-net ctrl missing headers\n");
 350             exit(1);
 351         }
 352
 353         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 354             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 355             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 356             exit(1);
 357         }
 358
 359         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 360         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 361
 362         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 363             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 364         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 365             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 366         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 367             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 368
 369         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 370
 371         virtqueue_push(vq, &elem, sizeof(status));
 372         virtio_notify(vdev, vq);
 373     }
 374 }
 375
 376 /* RX */
 377
 378 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 379 {
 380     VirtIONet *n = to_virtio_net(vdev);
 381
 382     qemu_flush_queued_packets(&n->nic->nc);
 383
 384     /* We now have RX buffers, signal to the IO thread to break out of the
 385      * select to re-poll the tap file descriptor */
 386     qemu_notify_event();
 387 }
 388
 389 static int virtio_net_can_receive(VLANClientState *nc)
 390 {
 391     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 392
 393     if (!virtio_queue_ready(n->rx_vq) ||
 394         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 395         return 0;
 396
 397     return 1;
 398 }
 399
 400 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 401 {
 402     if (virtio_queue_empty(n->rx_vq) ||
 403         (n->mergeable_rx_bufs &&
 404          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 405         virtio_queue_set_notification(n->rx_vq, 1);
 406
 407         /* To avoid a race condition where the guest has made some buffers
 408          * available after the above check but before notification was
 409          * enabled, check for available buffers again.
 410          */
 411         if (virtio_queue_empty(n->rx_vq) ||
 412             (n->mergeable_rx_bufs &&
 413              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 414             return 0;
 415     }
 416
 417     virtio_queue_set_notification(n->rx_vq, 0);
 418     return 1;
 419 }
 420
 421 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 422  * it never finds out that the packets don't have valid checksums.  This
 423  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 424  * fix this with Xen but it hasn't appeared in an upstream release of
 425  * dhclient yet.
 426  *
 427  * To avoid breaking existing guests, we catch udp packets and add
 428  * checksums.  This is terrible but it's better than hacking the guest
 429  * kernels.
 430  *
 431  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 432  * we should provide a mechanism to disable it to avoid polluting the host
 433  * cache.
 434  */
 435 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 436                                         const uint8_t *buf, size_t size)
 437 {
 438     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 439         (size > 27 && size < 1500) && /* normal sized MTU */
 440         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 441         (buf[23] == 17) && /* ip.protocol == UDP */
 442         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 443         /* FIXME this cast is evil */
 444         net_checksum_calculate((uint8_t *)buf, size);
 445         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 446     }
 447 }
 448
 449 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 450                           const void *buf, size_t size, size_t hdr_len)
 451 {
 452     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 453     int offset = 0;
 454
 455     hdr->flags = 0;
 456     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 457
 458     if (n->has_vnet_hdr) {
 459         memcpy(hdr, buf, sizeof(*hdr));
 460         offset = sizeof(*hdr);
 461         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 462     }
 463
 464     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 465      * but we may be passing along a larger header to the guest.
 466      */
 467     iov[0].iov_base += hdr_len;
 468     iov[0].iov_len  -= hdr_len;
 469
 470     return offset;
 471 }
 472
 473 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 474 {
 475     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 476     static const uint8_t vlan[] = {0x81, 0x00};
 477     uint8_t *ptr = (uint8_t *)buf;
 478     int i;
 479
 480     if (n->promisc)
 481         return 1;
 482
 483     if (n->has_vnet_hdr) {
 484         ptr += sizeof(struct virtio_net_hdr);
 485     }
 486
 487     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 488         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 489         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 490             return 0;
 491     }
 492
 493     if (ptr[0] & 1) { // multicast
 494         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 495             return !n->nobcast;
 496         } else if (n->nomulti) {
 497             return 0;
 498         } else if (n->allmulti || n->mac_table.multi_overflow) {
 499             return 1;
 500         }
 501
 502         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 503             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 504                 return 1;
 505             }
 506         }
 507     } else { // unicast
 508         if (n->nouni) {
 509             return 0;
 510         } else if (n->alluni || n->mac_table.uni_overflow) {
 511             return 1;
 512         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 513             return 1;
 514         }
 515
 516         for (i = 0; i < n->mac_table.first_multi; i++) {
 517             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 518                 return 1;
 519             }
 520         }
 521     }
 522
 523     return 0;
 524 }
 525
 526 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 527 {
 528     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 529     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 530     size_t hdr_len, offset, i;
 531
 532     if (!virtio_net_can_receive(&n->nic->nc))
 533         return -1;
 534
 535     /* hdr_len refers to the header we supply to the guest */
 536     hdr_len = n->mergeable_rx_bufs ?
 537         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 538
 539
 540     if (!virtio_net_has_buffers(n, size + hdr_len))
 541         return 0;
 542
 543     if (!receive_filter(n, buf, size))
 544         return size;
 545
 546     offset = i = 0;
 547
 548     while (offset < size) {
 549         VirtQueueElement elem;
 550         int len, total;
 551         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 552
 553         total = 0;
 554
 555         if ((i != 0 && !n->mergeable_rx_bufs) ||
 556             virtqueue_pop(n->rx_vq, &elem) == 0) {
 557             if (i == 0)
 558                 return -1;
 559             fprintf(stderr, "virtio-net truncating packet: "
 560                     "offset %zd, size %zd, hdr_len %zd\n",
 561                     offset, size, hdr_len);
 562             exit(1);
 563         }
 564
 565         if (elem.in_num < 1) {
 566             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 567             exit(1);
 568         }
 569
 570         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
 571             fprintf(stderr, "virtio-net header not in first element\n");
 572             exit(1);
 573         }
 574
 575         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 576
 577         if (i == 0) {
 578             if (n->mergeable_rx_bufs)
 579                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 580
 581             offset += receive_header(n, sg, elem.in_num,
 582                                      buf + offset, size - offset, hdr_len);
 583             total += hdr_len;
 584         }
 585
 586         /* copy in packet.  ugh */
 587         len = iov_from_buf(sg, elem.in_num,
 588                            buf + offset, size - offset);
 589         total += len;
 590
 591         /* signal other side */
 592         virtqueue_fill(n->rx_vq, &elem, total, i++);
 593
 594         offset += len;
 595     }
 596
 597     if (mhdr)
 598         mhdr->num_buffers = i;
 599
 600     virtqueue_flush(n->rx_vq, i);
 601     virtio_notify(&n->vdev, n->rx_vq);
 602
 603     return size;
 604 }
 605
 606 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 607
 608 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 609 {
 610     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 611
 612     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 613     virtio_notify(&n->vdev, n->tx_vq);
 614
 615     n->async_tx.elem.out_num = n->async_tx.len = 0;
 616
 617     virtio_queue_set_notification(n->tx_vq, 1);
 618     virtio_net_flush_tx(n, n->tx_vq);
 619 }
 620
 621 /* TX */
 622 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 623 {
 624     VirtQueueElement elem;
 625
 626     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 627         return;
 628
 629     if (n->async_tx.elem.out_num) {
 630         virtio_queue_set_notification(n->tx_vq, 0);
 631         return;
 632     }
 633
 634     while (virtqueue_pop(vq, &elem)) {
 635         ssize_t ret, len = 0;
 636         unsigned int out_num = elem.out_num;
 637         struct iovec *out_sg = &elem.out_sg[0];
 638         unsigned hdr_len;
 639
 640         /* hdr_len refers to the header received from the guest */
 641         hdr_len = n->mergeable_rx_bufs ?
 642             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 643             sizeof(struct virtio_net_hdr);
 644
 645         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 646             fprintf(stderr, "virtio-net header not in first element\n");
 647             exit(1);
 648         }
 649
 650         /* ignore the header if GSO is not supported */
 651         if (!n->has_vnet_hdr) {
 652             out_num--;
 653             out_sg++;
 654             len += hdr_len;
 655         } else if (n->mergeable_rx_bufs) {
 656             /* tapfd expects a struct virtio_net_hdr */
 657             hdr_len -= sizeof(struct virtio_net_hdr);
 658             out_sg->iov_len -= hdr_len;
 659             len += hdr_len;
 660         }
 661
 662         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 663                                       virtio_net_tx_complete);
 664         if (ret == 0) {
 665             virtio_queue_set_notification(n->tx_vq, 0);
 666             n->async_tx.elem = elem;
 667             n->async_tx.len  = len;
 668             return;
 669         }
 670
 671         len += ret;
 672
 673         virtqueue_push(vq, &elem, len);
 674         virtio_notify(&n->vdev, vq);
 675     }
 676 }
 677
 678 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 679 {
 680     VirtIONet *n = to_virtio_net(vdev);
 681
 682     if (n->tx_timer_active) {
 683         virtio_queue_set_notification(vq, 1);
 684         qemu_del_timer(n->tx_timer);
 685         n->tx_timer_active = 0;
 686         virtio_net_flush_tx(n, vq);
 687     } else {
 688         qemu_mod_timer(n->tx_timer,
 689                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 690         n->tx_timer_active = 1;
 691         virtio_queue_set_notification(vq, 0);
 692     }
 693 }
 694
 695 static void virtio_net_tx_timer(void *opaque)
 696 {
 697     VirtIONet *n = opaque;
 698
 699     n->tx_timer_active = 0;
 700
 701     /* Just in case the driver is not ready on more */
 702     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 703         return;
 704
 705     virtio_queue_set_notification(n->tx_vq, 1);
 706     virtio_net_flush_tx(n, n->tx_vq);
 707 }
 708
 709 static void virtio_net_save(QEMUFile *f, void *opaque)
 710 {
 711     VirtIONet *n = opaque;
 712
 713     if (n->vhost_started) {
 714         /* TODO: should we really stop the backend?
 715          * If we don't, it might keep writing to memory. */
 716         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 717         n->vhost_started = 0;
 718     }
 719     virtio_save(&n->vdev, f);
 720
 721     qemu_put_buffer(f, n->mac, ETH_ALEN);
 722     qemu_put_be32(f, n->tx_timer_active);
 723     qemu_put_be32(f, n->mergeable_rx_bufs);
 724     qemu_put_be16(f, n->status);
 725     qemu_put_byte(f, n->promisc);
 726     qemu_put_byte(f, n->allmulti);
 727     qemu_put_be32(f, n->mac_table.in_use);
 728     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 729     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 730     qemu_put_be32(f, n->has_vnet_hdr);
 731     qemu_put_byte(f, n->mac_table.multi_overflow);
 732     qemu_put_byte(f, n->mac_table.uni_overflow);
 733     qemu_put_byte(f, n->alluni);
 734     qemu_put_byte(f, n->nomulti);
 735     qemu_put_byte(f, n->nouni);
 736     qemu_put_byte(f, n->nobcast);
 737     qemu_put_byte(f, n->has_ufo);
 738 }
 739
 740 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 741 {
 742     VirtIONet *n = opaque;
 743     int i;
 744
 745     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 746         return -EINVAL;
 747
 748     virtio_load(&n->vdev, f);
 749
 750     qemu_get_buffer(f, n->mac, ETH_ALEN);
 751     n->tx_timer_active = qemu_get_be32(f);
 752     n->mergeable_rx_bufs = qemu_get_be32(f);
 753
 754     if (version_id >= 3)
 755         n->status = qemu_get_be16(f);
 756
 757     if (version_id >= 4) {
 758         if (version_id < 8) {
 759             n->promisc = qemu_get_be32(f);
 760             n->allmulti = qemu_get_be32(f);
 761         } else {
 762             n->promisc = qemu_get_byte(f);
 763             n->allmulti = qemu_get_byte(f);
 764         }
 765     }
 766
 767     if (version_id >= 5) {
 768         n->mac_table.in_use = qemu_get_be32(f);
 769         /* MAC_TABLE_ENTRIES may be different from the saved image */
 770         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 771             qemu_get_buffer(f, n->mac_table.macs,
 772                             n->mac_table.in_use * ETH_ALEN);
 773         } else if (n->mac_table.in_use) {
 774             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 775             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 776             n->mac_table.in_use = 0;
 777         }
 778     }
 779
 780     if (version_id >= 6)
 781         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 782
 783     if (version_id >= 7) {
 784         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 785             error_report("virtio-net: saved image requires vnet_hdr=on");
 786             return -1;
 787         }
 788
 789         if (n->has_vnet_hdr) {
 790             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 791             tap_set_offload(n->nic->nc.peer,
 792                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 793                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 794                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 795                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 796                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 797         }
 798     }
 799
 800     if (version_id >= 9) {
 801         n->mac_table.multi_overflow = qemu_get_byte(f);
 802         n->mac_table.uni_overflow = qemu_get_byte(f);
 803     }
 804
 805     if (version_id >= 10) {
 806         n->alluni = qemu_get_byte(f);
 807         n->nomulti = qemu_get_byte(f);
 808         n->nouni = qemu_get_byte(f);
 809         n->nobcast = qemu_get_byte(f);
 810     }
 811
 812     if (version_id >= 11) {
 813         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 814             error_report("virtio-net: saved image requires TUN_F_UFO support");
 815             return -1;
 816         }
 817     }
 818
 819     /* Find the first multicast entry in the saved MAC filter */
 820     for (i = 0; i < n->mac_table.in_use; i++) {
 821         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 822             break;
 823         }
 824     }
 825     n->mac_table.first_multi = i;
 826
 827     if (n->tx_timer_active) {
 828         qemu_mod_timer(n->tx_timer,
 829                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 830     }
 831     return 0;
 832 }
 833
 834 static void virtio_net_cleanup(VLANClientState *nc)
 835 {
 836     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 837
 838     n->nic = NULL;
 839 }
 840
 841 static NetClientInfo net_virtio_info = {
 842     .type = NET_CLIENT_TYPE_NIC,
 843     .size = sizeof(NICState),
 844     .can_receive = virtio_net_can_receive,
 845     .receive = virtio_net_receive,
 846         .cleanup = virtio_net_cleanup,
 847     .link_status_changed = virtio_net_set_link_status,
 848 };
 849
 850 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 851 {
 852     VirtIONet *n = to_virtio_net(vdev);
 853     if (!n->nic->nc.peer) {
 854         return;
 855     }
 856     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 857         return;
 858     }
 859
 860     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 861         return;
 862     }
 863     if (!!n->vhost_started == !!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 864         return;
 865     }
 866     if (status & VIRTIO_CONFIG_S_DRIVER_OK) {
 867         int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), vdev);
 868         if (r < 0) {
 869             fprintf(stderr, "unable to start vhost net: %d: "
 870                     "falling back on userspace virtio\n", -r);
 871         } else {
 872             n->vhost_started = 1;
 873         }
 874     } else {
 875         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 876         n->vhost_started = 0;
 877     }
 878 }
 879
 880 static void virtio_net_vmstate_change(void *opaque, int running, int reason)
 881 {
 882     VirtIONet *n = opaque;
 883     uint8_t status = running ? VIRTIO_CONFIG_S_DRIVER_OK : 0;
 884     /* This is called when vm is started/stopped,
 885      * it will start/stop vhost backend if * appropriate
 886      * e.g. after migration. */
 887     virtio_net_set_status(&n->vdev, n->vdev.status & status);
 888 }
 889
 890 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
 891 {
 892     VirtIONet *n;
 893     static int virtio_net_id;
 894
 895     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 896                                         sizeof(struct virtio_net_config),
 897                                         sizeof(VirtIONet));
 898
 899     n->vdev.get_config = virtio_net_get_config;
 900     n->vdev.set_config = virtio_net_set_config;
 901     n->vdev.get_features = virtio_net_get_features;
 902     n->vdev.set_features = virtio_net_set_features;
 903     n->vdev.bad_features = virtio_net_bad_features;
 904     n->vdev.reset = virtio_net_reset;
 905     n->vdev.set_status = virtio_net_set_status;
 906     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 907     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 908     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 909     qemu_macaddr_default_if_unset(&conf->macaddr);
 910     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
 911     n->status = VIRTIO_NET_S_LINK_UP;
 912
 913     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
 914
 915     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
 916
 917     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 918     n->tx_timer_active = 0;
 919     n->mergeable_rx_bufs = 0;
 920     n->promisc = 1; /* for compatibility */
 921
 922     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 923
 924     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 925
 926     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 927                     virtio_net_save, virtio_net_load, n);
 928     n->vmstate = qemu_add_vm_change_state_handler(virtio_net_vmstate_change, n);
 929
 930     return &n->vdev;
 931 }
 932
 933 void virtio_net_exit(VirtIODevice *vdev)
 934 {
 935     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 936     qemu_del_vm_change_state_handler(n->vmstate);
 937
 938     if (n->vhost_started) {
 939         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 940     }
 941
 942     qemu_purge_queued_packets(&n->nic->nc);
 943
 944     unregister_savevm("virtio-net", n);
 945
 946     qemu_free(n->mac_table.macs);
 947     qemu_free(n->vlans);
 948
 949     qemu_del_timer(n->tx_timer);
 950     qemu_free_timer(n->tx_timer);
 951
 952     virtio_cleanup(&n->vdev);
 953     qemu_del_vlan_client(&n->nic->nc);
 954 }