hw/virtio-net.c

   1 /*
   2  * Virtio Network Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "virtio.h"
  15 #include "net.h"
  16 #include "net/checksum.h"
  17 #include "net/tap.h"
  18 #include "qemu-error.h"
  19 #include "qemu-timer.h"
  20 #include "virtio-net.h"
  21 #include "vhost_net.h"
  22
  23 #define VIRTIO_NET_VM_VERSION    11
  24
  25 #define MAC_TABLE_ENTRIES    64
  26 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  27
  28 typedef struct VirtIONet
  29 {
  30     VirtIODevice vdev;
  31     uint8_t mac[ETH_ALEN];
  32     uint16_t status;
  33     VirtQueue *rx_vq;
  34     VirtQueue *tx_vq;
  35     VirtQueue *ctrl_vq;
  36     NICState *nic;
  37     QEMUTimer *tx_timer;
  38     int tx_timer_active;
  39     uint32_t has_vnet_hdr;
  40     uint8_t has_ufo;
  41     struct {
  42         VirtQueueElement elem;
  43         ssize_t len;
  44     } async_tx;
  45     int mergeable_rx_bufs;
  46     uint8_t promisc;
  47     uint8_t allmulti;
  48     uint8_t alluni;
  49     uint8_t nomulti;
  50     uint8_t nouni;
  51     uint8_t nobcast;
  52     uint8_t vhost_started;
  53     VMChangeStateEntry *vmstate;
  54     struct {
  55         int in_use;
  56         int first_multi;
  57         uint8_t multi_overflow;
  58         uint8_t uni_overflow;
  59         uint8_t *macs;
  60     } mac_table;
  61     uint32_t *vlans;
  62 } VirtIONet;
  63
  64 /* TODO
  65  * - we could suppress RX interrupt if we were so inclined.
  66  */
  67
  68 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  69 {
  70     return (VirtIONet *)vdev;
  71 }
  72
  73 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  74 {
  75     VirtIONet *n = to_virtio_net(vdev);
  76     struct virtio_net_config netcfg;
  77
  78     netcfg.status = n->status;
  79     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  80     memcpy(config, &netcfg, sizeof(netcfg));
  81 }
  82
  83 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  84 {
  85     VirtIONet *n = to_virtio_net(vdev);
  86     struct virtio_net_config netcfg;
  87
  88     memcpy(&netcfg, config, sizeof(netcfg));
  89
  90     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  91         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  92         qemu_format_nic_info_str(&n->nic->nc, n->mac);
  93     }
  94 }
  95
  96 static void virtio_net_set_link_status(VLANClientState *nc)
  97 {
  98     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
  99     uint16_t old_status = n->status;
 100
 101     if (nc->link_down)
 102         n->status &= ~VIRTIO_NET_S_LINK_UP;
 103     else
 104         n->status |= VIRTIO_NET_S_LINK_UP;
 105
 106     if (n->status != old_status)
 107         virtio_notify_config(&n->vdev);
 108 }
 109
 110 static void virtio_net_reset(VirtIODevice *vdev)
 111 {
 112     VirtIONet *n = to_virtio_net(vdev);
 113
 114     /* Reset back to compatibility mode */
 115     n->promisc = 1;
 116     n->allmulti = 0;
 117     n->alluni = 0;
 118     n->nomulti = 0;
 119     n->nouni = 0;
 120     n->nobcast = 0;
 121     if (n->vhost_started) {
 122         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 123         n->vhost_started = 0;
 124     }
 125
 126     /* Flush any MAC and VLAN filter table state */
 127     n->mac_table.in_use = 0;
 128     n->mac_table.first_multi = 0;
 129     n->mac_table.multi_overflow = 0;
 130     n->mac_table.uni_overflow = 0;
 131     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 132     memset(n->vlans, 0, MAX_VLAN >> 3);
 133 }
 134
 135 static int peer_has_vnet_hdr(VirtIONet *n)
 136 {
 137     if (!n->nic->nc.peer)
 138         return 0;
 139
 140     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 141         return 0;
 142
 143     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 144
 145     return n->has_vnet_hdr;
 146 }
 147
 148 static int peer_has_ufo(VirtIONet *n)
 149 {
 150     if (!peer_has_vnet_hdr(n))
 151         return 0;
 152
 153     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 154
 155     return n->has_ufo;
 156 }
 157
 158 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 159 {
 160     VirtIONet *n = to_virtio_net(vdev);
 161
 162     features |= (1 << VIRTIO_NET_F_MAC);
 163
 164     if (peer_has_vnet_hdr(n)) {
 165         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 166     } else {
 167         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 168         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 169         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 170         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 171
 172         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 173         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 174         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 175         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 176     }
 177
 178     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 179         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 180         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 181     }
 182
 183     if (!n->nic->nc.peer ||
 184         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 185         return features;
 186     }
 187     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 188         return features;
 189     }
 190     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 191 }
 192
 193 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 194 {
 195     uint32_t features = 0;
 196
 197     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 198      * but also these: */
 199     features |= (1 << VIRTIO_NET_F_MAC);
 200     features |= (1 << VIRTIO_NET_F_CSUM);
 201     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 202     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 203     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 204
 205     return features;
 206 }
 207
 208 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 209 {
 210     VirtIONet *n = to_virtio_net(vdev);
 211
 212     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 213
 214     if (n->has_vnet_hdr) {
 215         tap_set_offload(n->nic->nc.peer,
 216                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 217                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 218                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 219                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 220                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 221     }
 222     if (!n->nic->nc.peer ||
 223         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 224         return;
 225     }
 226     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 227         return;
 228     }
 229     return vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 230 }
 231
 232 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 233                                      VirtQueueElement *elem)
 234 {
 235     uint8_t on;
 236
 237     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 238         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 239         exit(1);
 240     }
 241
 242     on = ldub_p(elem->out_sg[1].iov_base);
 243
 244     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 245         n->promisc = on;
 246     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 247         n->allmulti = on;
 248     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 249         n->alluni = on;
 250     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 251         n->nomulti = on;
 252     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 253         n->nouni = on;
 254     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 255         n->nobcast = on;
 256     else
 257         return VIRTIO_NET_ERR;
 258
 259     return VIRTIO_NET_OK;
 260 }
 261
 262 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 263                                  VirtQueueElement *elem)
 264 {
 265     struct virtio_net_ctrl_mac mac_data;
 266
 267     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 268         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 269         elem->out_sg[2].iov_len < sizeof(mac_data))
 270         return VIRTIO_NET_ERR;
 271
 272     n->mac_table.in_use = 0;
 273     n->mac_table.first_multi = 0;
 274     n->mac_table.uni_overflow = 0;
 275     n->mac_table.multi_overflow = 0;
 276     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 277
 278     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 279
 280     if (sizeof(mac_data.entries) +
 281         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 282         return VIRTIO_NET_ERR;
 283
 284     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 285         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 286                mac_data.entries * ETH_ALEN);
 287         n->mac_table.in_use += mac_data.entries;
 288     } else {
 289         n->mac_table.uni_overflow = 1;
 290     }
 291
 292     n->mac_table.first_multi = n->mac_table.in_use;
 293
 294     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 295
 296     if (sizeof(mac_data.entries) +
 297         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 298         return VIRTIO_NET_ERR;
 299
 300     if (mac_data.entries) {
 301         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 302             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 303                    elem->out_sg[2].iov_base + sizeof(mac_data),
 304                    mac_data.entries * ETH_ALEN);
 305             n->mac_table.in_use += mac_data.entries;
 306         } else {
 307             n->mac_table.multi_overflow = 1;
 308         }
 309     }
 310
 311     return VIRTIO_NET_OK;
 312 }
 313
 314 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 315                                         VirtQueueElement *elem)
 316 {
 317     uint16_t vid;
 318
 319     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 320         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 321         return VIRTIO_NET_ERR;
 322     }
 323
 324     vid = lduw_le_p(elem->out_sg[1].iov_base);
 325
 326     if (vid >= MAX_VLAN)
 327         return VIRTIO_NET_ERR;
 328
 329     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 330         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 331     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 332         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 333     else
 334         return VIRTIO_NET_ERR;
 335
 336     return VIRTIO_NET_OK;
 337 }
 338
 339 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 340 {
 341     VirtIONet *n = to_virtio_net(vdev);
 342     struct virtio_net_ctrl_hdr ctrl;
 343     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 344     VirtQueueElement elem;
 345
 346     while (virtqueue_pop(vq, &elem)) {
 347         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 348             fprintf(stderr, "virtio-net ctrl missing headers\n");
 349             exit(1);
 350         }
 351
 352         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 353             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 354             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 355             exit(1);
 356         }
 357
 358         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 359         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 360
 361         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 362             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 363         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 364             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 365         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 366             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 367
 368         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 369
 370         virtqueue_push(vq, &elem, sizeof(status));
 371         virtio_notify(vdev, vq);
 372     }
 373 }
 374
 375 /* RX */
 376
 377 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 378 {
 379     VirtIONet *n = to_virtio_net(vdev);
 380
 381     qemu_flush_queued_packets(&n->nic->nc);
 382
 383     /* We now have RX buffers, signal to the IO thread to break out of the
 384      * select to re-poll the tap file descriptor */
 385     qemu_notify_event();
 386 }
 387
 388 static int virtio_net_can_receive(VLANClientState *nc)
 389 {
 390     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 391
 392     if (!virtio_queue_ready(n->rx_vq) ||
 393         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 394         return 0;
 395
 396     return 1;
 397 }
 398
 399 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 400 {
 401     if (virtio_queue_empty(n->rx_vq) ||
 402         (n->mergeable_rx_bufs &&
 403          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 404         virtio_queue_set_notification(n->rx_vq, 1);
 405
 406         /* To avoid a race condition where the guest has made some buffers
 407          * available after the above check but before notification was
 408          * enabled, check for available buffers again.
 409          */
 410         if (virtio_queue_empty(n->rx_vq) ||
 411             (n->mergeable_rx_bufs &&
 412              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 413             return 0;
 414     }
 415
 416     virtio_queue_set_notification(n->rx_vq, 0);
 417     return 1;
 418 }
 419
 420 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 421  * it never finds out that the packets don't have valid checksums.  This
 422  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 423  * fix this with Xen but it hasn't appeared in an upstream release of
 424  * dhclient yet.
 425  *
 426  * To avoid breaking existing guests, we catch udp packets and add
 427  * checksums.  This is terrible but it's better than hacking the guest
 428  * kernels.
 429  *
 430  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 431  * we should provide a mechanism to disable it to avoid polluting the host
 432  * cache.
 433  */
 434 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 435                                         const uint8_t *buf, size_t size)
 436 {
 437     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 438         (size > 27 && size < 1500) && /* normal sized MTU */
 439         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 440         (buf[23] == 17) && /* ip.protocol == UDP */
 441         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 442         /* FIXME this cast is evil */
 443         net_checksum_calculate((uint8_t *)buf, size);
 444         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 445     }
 446 }
 447
 448 static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count)
 449 {
 450     int offset, i;
 451
 452     offset = i = 0;
 453     while (offset < count && i < iovcnt) {
 454         int len = MIN(iov[i].iov_len, count - offset);
 455         memcpy(iov[i].iov_base, buf + offset, len);
 456         offset += len;
 457         i++;
 458     }
 459
 460     return offset;
 461 }
 462
 463 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 464                           const void *buf, size_t size, size_t hdr_len)
 465 {
 466     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 467     int offset = 0;
 468
 469     hdr->flags = 0;
 470     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 471
 472     if (n->has_vnet_hdr) {
 473         memcpy(hdr, buf, sizeof(*hdr));
 474         offset = sizeof(*hdr);
 475         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 476     }
 477
 478     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 479      * but we may be passing along a larger header to the guest.
 480      */
 481     iov[0].iov_base += hdr_len;
 482     iov[0].iov_len  -= hdr_len;
 483
 484     return offset;
 485 }
 486
 487 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 488 {
 489     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 490     static const uint8_t vlan[] = {0x81, 0x00};
 491     uint8_t *ptr = (uint8_t *)buf;
 492     int i;
 493
 494     if (n->promisc)
 495         return 1;
 496
 497     if (n->has_vnet_hdr) {
 498         ptr += sizeof(struct virtio_net_hdr);
 499     }
 500
 501     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 502         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 503         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 504             return 0;
 505     }
 506
 507     if (ptr[0] & 1) { // multicast
 508         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 509             return !n->nobcast;
 510         } else if (n->nomulti) {
 511             return 0;
 512         } else if (n->allmulti || n->mac_table.multi_overflow) {
 513             return 1;
 514         }
 515
 516         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 517             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 518                 return 1;
 519             }
 520         }
 521     } else { // unicast
 522         if (n->nouni) {
 523             return 0;
 524         } else if (n->alluni || n->mac_table.uni_overflow) {
 525             return 1;
 526         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 527             return 1;
 528         }
 529
 530         for (i = 0; i < n->mac_table.first_multi; i++) {
 531             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 532                 return 1;
 533             }
 534         }
 535     }
 536
 537     return 0;
 538 }
 539
 540 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 541 {
 542     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 543     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 544     size_t hdr_len, offset, i;
 545
 546     if (!virtio_net_can_receive(&n->nic->nc))
 547         return -1;
 548
 549     if (!virtio_net_has_buffers(n, size))
 550         return 0;
 551
 552     if (!receive_filter(n, buf, size))
 553         return size;
 554
 555     /* hdr_len refers to the header we supply to the guest */
 556     hdr_len = n->mergeable_rx_bufs ?
 557         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 558
 559     offset = i = 0;
 560
 561     while (offset < size) {
 562         VirtQueueElement elem;
 563         int len, total;
 564         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 565
 566         total = 0;
 567
 568         if ((i != 0 && !n->mergeable_rx_bufs) ||
 569             virtqueue_pop(n->rx_vq, &elem) == 0) {
 570             if (i == 0)
 571                 return -1;
 572             fprintf(stderr, "virtio-net truncating packet\n");
 573             exit(1);
 574         }
 575
 576         if (elem.in_num < 1) {
 577             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 578             exit(1);
 579         }
 580
 581         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
 582             fprintf(stderr, "virtio-net header not in first element\n");
 583             exit(1);
 584         }
 585
 586         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 587
 588         if (i == 0) {
 589             if (n->mergeable_rx_bufs)
 590                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 591
 592             offset += receive_header(n, sg, elem.in_num,
 593                                      buf + offset, size - offset, hdr_len);
 594             total += hdr_len;
 595         }
 596
 597         /* copy in packet.  ugh */
 598         len = iov_fill(sg, elem.in_num,
 599                        buf + offset, size - offset);
 600         total += len;
 601
 602         /* signal other side */
 603         virtqueue_fill(n->rx_vq, &elem, total, i++);
 604
 605         offset += len;
 606     }
 607
 608     if (mhdr)
 609         mhdr->num_buffers = i;
 610
 611     virtqueue_flush(n->rx_vq, i);
 612     virtio_notify(&n->vdev, n->rx_vq);
 613
 614     return size;
 615 }
 616
 617 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 618
 619 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 620 {
 621     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 622
 623     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 624     virtio_notify(&n->vdev, n->tx_vq);
 625
 626     n->async_tx.elem.out_num = n->async_tx.len = 0;
 627
 628     virtio_queue_set_notification(n->tx_vq, 1);
 629     virtio_net_flush_tx(n, n->tx_vq);
 630 }
 631
 632 /* TX */
 633 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 634 {
 635     VirtQueueElement elem;
 636
 637     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 638         return;
 639
 640     if (n->async_tx.elem.out_num) {
 641         virtio_queue_set_notification(n->tx_vq, 0);
 642         return;
 643     }
 644
 645     while (virtqueue_pop(vq, &elem)) {
 646         ssize_t ret, len = 0;
 647         unsigned int out_num = elem.out_num;
 648         struct iovec *out_sg = &elem.out_sg[0];
 649         unsigned hdr_len;
 650
 651         /* hdr_len refers to the header received from the guest */
 652         hdr_len = n->mergeable_rx_bufs ?
 653             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 654             sizeof(struct virtio_net_hdr);
 655
 656         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 657             fprintf(stderr, "virtio-net header not in first element\n");
 658             exit(1);
 659         }
 660
 661         /* ignore the header if GSO is not supported */
 662         if (!n->has_vnet_hdr) {
 663             out_num--;
 664             out_sg++;
 665             len += hdr_len;
 666         } else if (n->mergeable_rx_bufs) {
 667             /* tapfd expects a struct virtio_net_hdr */
 668             hdr_len -= sizeof(struct virtio_net_hdr);
 669             out_sg->iov_len -= hdr_len;
 670             len += hdr_len;
 671         }
 672
 673         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 674                                       virtio_net_tx_complete);
 675         if (ret == 0) {
 676             virtio_queue_set_notification(n->tx_vq, 0);
 677             n->async_tx.elem = elem;
 678             n->async_tx.len  = len;
 679             return;
 680         }
 681
 682         len += ret;
 683
 684         virtqueue_push(vq, &elem, len);
 685         virtio_notify(&n->vdev, vq);
 686     }
 687 }
 688
 689 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 690 {
 691     VirtIONet *n = to_virtio_net(vdev);
 692
 693     if (n->tx_timer_active) {
 694         virtio_queue_set_notification(vq, 1);
 695         qemu_del_timer(n->tx_timer);
 696         n->tx_timer_active = 0;
 697         virtio_net_flush_tx(n, vq);
 698     } else {
 699         qemu_mod_timer(n->tx_timer,
 700                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 701         n->tx_timer_active = 1;
 702         virtio_queue_set_notification(vq, 0);
 703     }
 704 }
 705
 706 static void virtio_net_tx_timer(void *opaque)
 707 {
 708     VirtIONet *n = opaque;
 709
 710     n->tx_timer_active = 0;
 711
 712     /* Just in case the driver is not ready on more */
 713     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 714         return;
 715
 716     virtio_queue_set_notification(n->tx_vq, 1);
 717     virtio_net_flush_tx(n, n->tx_vq);
 718 }
 719
 720 static void virtio_net_save(QEMUFile *f, void *opaque)
 721 {
 722     VirtIONet *n = opaque;
 723
 724     if (n->vhost_started) {
 725         /* TODO: should we really stop the backend?
 726          * If we don't, it might keep writing to memory. */
 727         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 728         n->vhost_started = 0;
 729     }
 730     virtio_save(&n->vdev, f);
 731
 732     qemu_put_buffer(f, n->mac, ETH_ALEN);
 733     qemu_put_be32(f, n->tx_timer_active);
 734     qemu_put_be32(f, n->mergeable_rx_bufs);
 735     qemu_put_be16(f, n->status);
 736     qemu_put_byte(f, n->promisc);
 737     qemu_put_byte(f, n->allmulti);
 738     qemu_put_be32(f, n->mac_table.in_use);
 739     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 740     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 741     qemu_put_be32(f, n->has_vnet_hdr);
 742     qemu_put_byte(f, n->mac_table.multi_overflow);
 743     qemu_put_byte(f, n->mac_table.uni_overflow);
 744     qemu_put_byte(f, n->alluni);
 745     qemu_put_byte(f, n->nomulti);
 746     qemu_put_byte(f, n->nouni);
 747     qemu_put_byte(f, n->nobcast);
 748     qemu_put_byte(f, n->has_ufo);
 749 }
 750
 751 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 752 {
 753     VirtIONet *n = opaque;
 754     int i;
 755
 756     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 757         return -EINVAL;
 758
 759     virtio_load(&n->vdev, f);
 760
 761     qemu_get_buffer(f, n->mac, ETH_ALEN);
 762     n->tx_timer_active = qemu_get_be32(f);
 763     n->mergeable_rx_bufs = qemu_get_be32(f);
 764
 765     if (version_id >= 3)
 766         n->status = qemu_get_be16(f);
 767
 768     if (version_id >= 4) {
 769         if (version_id < 8) {
 770             n->promisc = qemu_get_be32(f);
 771             n->allmulti = qemu_get_be32(f);
 772         } else {
 773             n->promisc = qemu_get_byte(f);
 774             n->allmulti = qemu_get_byte(f);
 775         }
 776     }
 777
 778     if (version_id >= 5) {
 779         n->mac_table.in_use = qemu_get_be32(f);
 780         /* MAC_TABLE_ENTRIES may be different from the saved image */
 781         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 782             qemu_get_buffer(f, n->mac_table.macs,
 783                             n->mac_table.in_use * ETH_ALEN);
 784         } else if (n->mac_table.in_use) {
 785             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 786             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 787             n->mac_table.in_use = 0;
 788         }
 789     }
 790
 791     if (version_id >= 6)
 792         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 793
 794     if (version_id >= 7) {
 795         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 796             error_report("virtio-net: saved image requires vnet_hdr=on");
 797             return -1;
 798         }
 799
 800         if (n->has_vnet_hdr) {
 801             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 802             tap_set_offload(n->nic->nc.peer,
 803                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 804                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 805                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 806                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 807                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 808         }
 809     }
 810
 811     if (version_id >= 9) {
 812         n->mac_table.multi_overflow = qemu_get_byte(f);
 813         n->mac_table.uni_overflow = qemu_get_byte(f);
 814     }
 815
 816     if (version_id >= 10) {
 817         n->alluni = qemu_get_byte(f);
 818         n->nomulti = qemu_get_byte(f);
 819         n->nouni = qemu_get_byte(f);
 820         n->nobcast = qemu_get_byte(f);
 821     }
 822
 823     if (version_id >= 11) {
 824         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 825             error_report("virtio-net: saved image requires TUN_F_UFO support");
 826             return -1;
 827         }
 828     }
 829
 830     /* Find the first multicast entry in the saved MAC filter */
 831     for (i = 0; i < n->mac_table.in_use; i++) {
 832         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 833             break;
 834         }
 835     }
 836     n->mac_table.first_multi = i;
 837
 838     if (n->tx_timer_active) {
 839         qemu_mod_timer(n->tx_timer,
 840                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 841     }
 842     return 0;
 843 }
 844
 845 static void virtio_net_cleanup(VLANClientState *nc)
 846 {
 847     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 848
 849     n->nic = NULL;
 850 }
 851
 852 static NetClientInfo net_virtio_info = {
 853     .type = NET_CLIENT_TYPE_NIC,
 854     .size = sizeof(NICState),
 855     .can_receive = virtio_net_can_receive,
 856     .receive = virtio_net_receive,
 857         .cleanup = virtio_net_cleanup,
 858     .link_status_changed = virtio_net_set_link_status,
 859 };
 860
 861 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 862 {
 863     VirtIONet *n = to_virtio_net(vdev);
 864     if (!n->nic->nc.peer) {
 865         return;
 866     }
 867     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 868         return;
 869     }
 870
 871     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 872         return;
 873     }
 874     if (!!n->vhost_started == !!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 875         return;
 876     }
 877     if (status & VIRTIO_CONFIG_S_DRIVER_OK) {
 878         int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), vdev);
 879         if (r < 0) {
 880             fprintf(stderr, "unable to start vhost net: %d: "
 881                     "falling back on userspace virtio\n", -r);
 882         } else {
 883             n->vhost_started = 1;
 884         }
 885     } else {
 886         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 887         n->vhost_started = 0;
 888     }
 889 }
 890
 891 static void virtio_net_vmstate_change(void *opaque, int running, int reason)
 892 {
 893     VirtIONet *n = opaque;
 894     if (!running) {
 895         return;
 896     }
 897     /* This is called when vm is started, it will start vhost backend if
 898      * appropriate e.g. after migration. */
 899     virtio_net_set_status(&n->vdev, n->vdev.status);
 900 }
 901
 902 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
 903 {
 904     VirtIONet *n;
 905     static int virtio_net_id;
 906
 907     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 908                                         sizeof(struct virtio_net_config),
 909                                         sizeof(VirtIONet));
 910
 911     n->vdev.get_config = virtio_net_get_config;
 912     n->vdev.set_config = virtio_net_set_config;
 913     n->vdev.get_features = virtio_net_get_features;
 914     n->vdev.set_features = virtio_net_set_features;
 915     n->vdev.bad_features = virtio_net_bad_features;
 916     n->vdev.reset = virtio_net_reset;
 917     n->vdev.set_status = virtio_net_set_status;
 918     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 919     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 920     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 921     qemu_macaddr_default_if_unset(&conf->macaddr);
 922     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
 923     n->status = VIRTIO_NET_S_LINK_UP;
 924
 925     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
 926
 927     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
 928
 929     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 930     n->tx_timer_active = 0;
 931     n->mergeable_rx_bufs = 0;
 932     n->promisc = 1; /* for compatibility */
 933
 934     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 935
 936     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 937
 938     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 939                     virtio_net_save, virtio_net_load, n);
 940     n->vmstate = qemu_add_vm_change_state_handler(virtio_net_vmstate_change, n);
 941
 942     return &n->vdev;
 943 }
 944
 945 void virtio_net_exit(VirtIODevice *vdev)
 946 {
 947     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 948     qemu_del_vm_change_state_handler(n->vmstate);
 949
 950     if (n->vhost_started) {
 951         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 952     }
 953
 954     qemu_purge_queued_packets(&n->nic->nc);
 955
 956     unregister_savevm("virtio-net", n);
 957
 958     qemu_free(n->mac_table.macs);
 959     qemu_free(n->vlans);
 960
 961     qemu_del_timer(n->tx_timer);
 962     qemu_free_timer(n->tx_timer);
 963
 964     virtio_cleanup(&n->vdev);
 965     qemu_del_vlan_client(&n->nic->nc);
 966 }