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     DeviceState *qdev;
  64 } VirtIONet;
  65
  66 /* TODO
  67  * - we could suppress RX interrupt if we were so inclined.
  68  */
  69
  70 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  71 {
  72     return (VirtIONet *)vdev;
  73 }
  74
  75 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  76 {
  77     VirtIONet *n = to_virtio_net(vdev);
  78     struct virtio_net_config netcfg;
  79
  80     netcfg.status = n->status;
  81     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  82     memcpy(config, &netcfg, sizeof(netcfg));
  83 }
  84
  85 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  86 {
  87     VirtIONet *n = to_virtio_net(vdev);
  88     struct virtio_net_config netcfg;
  89
  90     memcpy(&netcfg, config, sizeof(netcfg));
  91
  92     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  93         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  94         qemu_format_nic_info_str(&n->nic->nc, n->mac);
  95     }
  96 }
  97
  98 static void virtio_net_set_link_status(VLANClientState *nc)
  99 {
 100     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 101     uint16_t old_status = n->status;
 102
 103     if (nc->link_down)
 104         n->status &= ~VIRTIO_NET_S_LINK_UP;
 105     else
 106         n->status |= VIRTIO_NET_S_LINK_UP;
 107
 108     if (n->status != old_status)
 109         virtio_notify_config(&n->vdev);
 110 }
 111
 112 static void virtio_net_reset(VirtIODevice *vdev)
 113 {
 114     VirtIONet *n = to_virtio_net(vdev);
 115
 116     /* Reset back to compatibility mode */
 117     n->promisc = 1;
 118     n->allmulti = 0;
 119     n->alluni = 0;
 120     n->nomulti = 0;
 121     n->nouni = 0;
 122     n->nobcast = 0;
 123     if (n->vhost_started) {
 124         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 125         n->vhost_started = 0;
 126     }
 127
 128     /* Flush any MAC and VLAN filter table state */
 129     n->mac_table.in_use = 0;
 130     n->mac_table.first_multi = 0;
 131     n->mac_table.multi_overflow = 0;
 132     n->mac_table.uni_overflow = 0;
 133     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 134     memset(n->vlans, 0, MAX_VLAN >> 3);
 135 }
 136
 137 static int peer_has_vnet_hdr(VirtIONet *n)
 138 {
 139     if (!n->nic->nc.peer)
 140         return 0;
 141
 142     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 143         return 0;
 144
 145     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 146
 147     return n->has_vnet_hdr;
 148 }
 149
 150 static int peer_has_ufo(VirtIONet *n)
 151 {
 152     if (!peer_has_vnet_hdr(n))
 153         return 0;
 154
 155     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 156
 157     return n->has_ufo;
 158 }
 159
 160 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 161 {
 162     VirtIONet *n = to_virtio_net(vdev);
 163
 164     features |= (1 << VIRTIO_NET_F_MAC);
 165
 166     if (peer_has_vnet_hdr(n)) {
 167         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 168     } else {
 169         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 170         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 171         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 172         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 173
 174         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 175         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 176         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 177         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 178     }
 179
 180     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 181         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 182         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 183     }
 184
 185     if (!n->nic->nc.peer ||
 186         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 187         return features;
 188     }
 189     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 190         return features;
 191     }
 192     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 193 }
 194
 195 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 196 {
 197     uint32_t features = 0;
 198
 199     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 200      * but also these: */
 201     features |= (1 << VIRTIO_NET_F_MAC);
 202     features |= (1 << VIRTIO_NET_F_CSUM);
 203     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 204     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 205     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 206
 207     return features;
 208 }
 209
 210 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 211 {
 212     VirtIONet *n = to_virtio_net(vdev);
 213
 214     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 215
 216     if (n->has_vnet_hdr) {
 217         tap_set_offload(n->nic->nc.peer,
 218                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 219                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 220                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 221                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 222                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 223     }
 224     if (!n->nic->nc.peer ||
 225         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 226         return;
 227     }
 228     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 229         return;
 230     }
 231     vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 232 }
 233
 234 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 235                                      VirtQueueElement *elem)
 236 {
 237     uint8_t on;
 238
 239     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 240         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 241         exit(1);
 242     }
 243
 244     on = ldub_p(elem->out_sg[1].iov_base);
 245
 246     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 247         n->promisc = on;
 248     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 249         n->allmulti = on;
 250     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 251         n->alluni = on;
 252     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 253         n->nomulti = on;
 254     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 255         n->nouni = on;
 256     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 257         n->nobcast = on;
 258     else
 259         return VIRTIO_NET_ERR;
 260
 261     return VIRTIO_NET_OK;
 262 }
 263
 264 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 265                                  VirtQueueElement *elem)
 266 {
 267     struct virtio_net_ctrl_mac mac_data;
 268
 269     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 270         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 271         elem->out_sg[2].iov_len < sizeof(mac_data))
 272         return VIRTIO_NET_ERR;
 273
 274     n->mac_table.in_use = 0;
 275     n->mac_table.first_multi = 0;
 276     n->mac_table.uni_overflow = 0;
 277     n->mac_table.multi_overflow = 0;
 278     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 279
 280     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 281
 282     if (sizeof(mac_data.entries) +
 283         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 284         return VIRTIO_NET_ERR;
 285
 286     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 287         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 288                mac_data.entries * ETH_ALEN);
 289         n->mac_table.in_use += mac_data.entries;
 290     } else {
 291         n->mac_table.uni_overflow = 1;
 292     }
 293
 294     n->mac_table.first_multi = n->mac_table.in_use;
 295
 296     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 297
 298     if (sizeof(mac_data.entries) +
 299         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 300         return VIRTIO_NET_ERR;
 301
 302     if (mac_data.entries) {
 303         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 304             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 305                    elem->out_sg[2].iov_base + sizeof(mac_data),
 306                    mac_data.entries * ETH_ALEN);
 307             n->mac_table.in_use += mac_data.entries;
 308         } else {
 309             n->mac_table.multi_overflow = 1;
 310         }
 311     }
 312
 313     return VIRTIO_NET_OK;
 314 }
 315
 316 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 317                                         VirtQueueElement *elem)
 318 {
 319     uint16_t vid;
 320
 321     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 322         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 323         return VIRTIO_NET_ERR;
 324     }
 325
 326     vid = lduw_le_p(elem->out_sg[1].iov_base);
 327
 328     if (vid >= MAX_VLAN)
 329         return VIRTIO_NET_ERR;
 330
 331     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 332         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 333     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 334         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 335     else
 336         return VIRTIO_NET_ERR;
 337
 338     return VIRTIO_NET_OK;
 339 }
 340
 341 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 342 {
 343     VirtIONet *n = to_virtio_net(vdev);
 344     struct virtio_net_ctrl_hdr ctrl;
 345     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 346     VirtQueueElement elem;
 347
 348     while (virtqueue_pop(vq, &elem)) {
 349         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 350             fprintf(stderr, "virtio-net ctrl missing headers\n");
 351             exit(1);
 352         }
 353
 354         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 355             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 356             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 357             exit(1);
 358         }
 359
 360         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 361         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 362
 363         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 364             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 365         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 366             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 367         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 368             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 369
 370         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 371
 372         virtqueue_push(vq, &elem, sizeof(status));
 373         virtio_notify(vdev, vq);
 374     }
 375 }
 376
 377 /* RX */
 378
 379 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 380 {
 381     VirtIONet *n = to_virtio_net(vdev);
 382
 383     qemu_flush_queued_packets(&n->nic->nc);
 384
 385     /* We now have RX buffers, signal to the IO thread to break out of the
 386      * select to re-poll the tap file descriptor */
 387     qemu_notify_event();
 388 }
 389
 390 static int virtio_net_can_receive(VLANClientState *nc)
 391 {
 392     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 393
 394     if (!virtio_queue_ready(n->rx_vq) ||
 395         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 396         return 0;
 397
 398     return 1;
 399 }
 400
 401 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 402 {
 403     if (virtio_queue_empty(n->rx_vq) ||
 404         (n->mergeable_rx_bufs &&
 405          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 406         virtio_queue_set_notification(n->rx_vq, 1);
 407
 408         /* To avoid a race condition where the guest has made some buffers
 409          * available after the above check but before notification was
 410          * enabled, check for available buffers again.
 411          */
 412         if (virtio_queue_empty(n->rx_vq) ||
 413             (n->mergeable_rx_bufs &&
 414              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 415             return 0;
 416     }
 417
 418     virtio_queue_set_notification(n->rx_vq, 0);
 419     return 1;
 420 }
 421
 422 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 423  * it never finds out that the packets don't have valid checksums.  This
 424  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 425  * fix this with Xen but it hasn't appeared in an upstream release of
 426  * dhclient yet.
 427  *
 428  * To avoid breaking existing guests, we catch udp packets and add
 429  * checksums.  This is terrible but it's better than hacking the guest
 430  * kernels.
 431  *
 432  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 433  * we should provide a mechanism to disable it to avoid polluting the host
 434  * cache.
 435  */
 436 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 437                                         const uint8_t *buf, size_t size)
 438 {
 439     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 440         (size > 27 && size < 1500) && /* normal sized MTU */
 441         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 442         (buf[23] == 17) && /* ip.protocol == UDP */
 443         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 444         /* FIXME this cast is evil */
 445         net_checksum_calculate((uint8_t *)buf, size);
 446         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 447     }
 448 }
 449
 450 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 451                           const void *buf, size_t size, size_t hdr_len)
 452 {
 453     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 454     int offset = 0;
 455
 456     hdr->flags = 0;
 457     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 458
 459     if (n->has_vnet_hdr) {
 460         memcpy(hdr, buf, sizeof(*hdr));
 461         offset = sizeof(*hdr);
 462         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 463     }
 464
 465     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 466      * but we may be passing along a larger header to the guest.
 467      */
 468     iov[0].iov_base += hdr_len;
 469     iov[0].iov_len  -= hdr_len;
 470
 471     return offset;
 472 }
 473
 474 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 475 {
 476     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 477     static const uint8_t vlan[] = {0x81, 0x00};
 478     uint8_t *ptr = (uint8_t *)buf;
 479     int i;
 480
 481     if (n->promisc)
 482         return 1;
 483
 484     if (n->has_vnet_hdr) {
 485         ptr += sizeof(struct virtio_net_hdr);
 486     }
 487
 488     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 489         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 490         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 491             return 0;
 492     }
 493
 494     if (ptr[0] & 1) { // multicast
 495         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 496             return !n->nobcast;
 497         } else if (n->nomulti) {
 498             return 0;
 499         } else if (n->allmulti || n->mac_table.multi_overflow) {
 500             return 1;
 501         }
 502
 503         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 504             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 505                 return 1;
 506             }
 507         }
 508     } else { // unicast
 509         if (n->nouni) {
 510             return 0;
 511         } else if (n->alluni || n->mac_table.uni_overflow) {
 512             return 1;
 513         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 514             return 1;
 515         }
 516
 517         for (i = 0; i < n->mac_table.first_multi; i++) {
 518             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 519                 return 1;
 520             }
 521         }
 522     }
 523
 524     return 0;
 525 }
 526
 527 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 528 {
 529     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 530     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 531     size_t guest_hdr_len, offset, i, host_hdr_len;
 532
 533     if (!virtio_net_can_receive(&n->nic->nc))
 534         return -1;
 535
 536     /* hdr_len refers to the header we supply to the guest */
 537     guest_hdr_len = n->mergeable_rx_bufs ?
 538         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 539
 540
 541     host_hdr_len = n->has_vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 542     if (!virtio_net_has_buffers(n, size + guest_hdr_len - host_hdr_len))
 543         return 0;
 544
 545     if (!receive_filter(n, buf, size))
 546         return size;
 547
 548     offset = i = 0;
 549
 550     while (offset < size) {
 551         VirtQueueElement elem;
 552         int len, total;
 553         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 554
 555         total = 0;
 556
 557         if (virtqueue_pop(n->rx_vq, &elem) == 0) {
 558             if (i == 0)
 559                 return -1;
 560             fprintf(stderr, "virtio-net unexpected empty queue: "
 561                     "i %zd mergeable %d offset %zd, size %zd, "
 562                     "guest hdr len %zd, host hdr len %zd guest features 0x%x\n",
 563                     i, n->mergeable_rx_bufs, offset, size,
 564                     guest_hdr_len, host_hdr_len, n->vdev.guest_features);
 565             exit(1);
 566         }
 567
 568         if (elem.in_num < 1) {
 569             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 570             exit(1);
 571         }
 572
 573         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != guest_hdr_len) {
 574             fprintf(stderr, "virtio-net header not in first element\n");
 575             exit(1);
 576         }
 577
 578         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 579
 580         if (i == 0) {
 581             if (n->mergeable_rx_bufs)
 582                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 583
 584             offset += receive_header(n, sg, elem.in_num,
 585                                      buf + offset, size - offset, guest_hdr_len);
 586             total += guest_hdr_len;
 587         }
 588
 589         /* copy in packet.  ugh */
 590         len = iov_from_buf(sg, elem.in_num,
 591                            buf + offset, size - offset);
 592         total += len;
 593         offset += len;
 594         /* If buffers can't be merged, at this point we
 595          * must have consumed the complete packet.
 596          * Otherwise, drop it. */
 597         if (!n->mergeable_rx_bufs && offset < size) {
 598 #if 0
 599             fprintf(stderr, "virtio-net truncated non-mergeable packet: "
 600
 601                     "i %zd mergeable %d offset %zd, size %zd, "
 602                     "guest hdr len %zd, host hdr len %zd\n",
 603                     i, n->mergeable_rx_bufs,
 604                     offset, size, guest_hdr_len, host_hdr_len);
 605 #endif
 606             return size;
 607         }
 608
 609         /* signal other side */
 610         virtqueue_fill(n->rx_vq, &elem, total, i++);
 611     }
 612
 613     if (mhdr)
 614         mhdr->num_buffers = i;
 615
 616     virtqueue_flush(n->rx_vq, i);
 617     virtio_notify(&n->vdev, n->rx_vq);
 618
 619     return size;
 620 }
 621
 622 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 623
 624 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 625 {
 626     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 627
 628     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 629     virtio_notify(&n->vdev, n->tx_vq);
 630
 631     n->async_tx.elem.out_num = n->async_tx.len = 0;
 632
 633     virtio_queue_set_notification(n->tx_vq, 1);
 634     virtio_net_flush_tx(n, n->tx_vq);
 635 }
 636
 637 /* TX */
 638 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 639 {
 640     VirtQueueElement elem;
 641
 642     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 643         return;
 644
 645     if (n->async_tx.elem.out_num) {
 646         virtio_queue_set_notification(n->tx_vq, 0);
 647         return;
 648     }
 649
 650     while (virtqueue_pop(vq, &elem)) {
 651         ssize_t ret, len = 0;
 652         unsigned int out_num = elem.out_num;
 653         struct iovec *out_sg = &elem.out_sg[0];
 654         unsigned hdr_len;
 655
 656         /* hdr_len refers to the header received from the guest */
 657         hdr_len = n->mergeable_rx_bufs ?
 658             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 659             sizeof(struct virtio_net_hdr);
 660
 661         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 662             fprintf(stderr, "virtio-net header not in first element\n");
 663             exit(1);
 664         }
 665
 666         /* ignore the header if GSO is not supported */
 667         if (!n->has_vnet_hdr) {
 668             out_num--;
 669             out_sg++;
 670             len += hdr_len;
 671         } else if (n->mergeable_rx_bufs) {
 672             /* tapfd expects a struct virtio_net_hdr */
 673             hdr_len -= sizeof(struct virtio_net_hdr);
 674             out_sg->iov_len -= hdr_len;
 675             len += hdr_len;
 676         }
 677
 678         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 679                                       virtio_net_tx_complete);
 680         if (ret == 0) {
 681             virtio_queue_set_notification(n->tx_vq, 0);
 682             n->async_tx.elem = elem;
 683             n->async_tx.len  = len;
 684             return;
 685         }
 686
 687         len += ret;
 688
 689         virtqueue_push(vq, &elem, len);
 690         virtio_notify(&n->vdev, vq);
 691     }
 692 }
 693
 694 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 695 {
 696     VirtIONet *n = to_virtio_net(vdev);
 697
 698     if (n->tx_timer_active) {
 699         virtio_queue_set_notification(vq, 1);
 700         qemu_del_timer(n->tx_timer);
 701         n->tx_timer_active = 0;
 702         virtio_net_flush_tx(n, vq);
 703     } else {
 704         qemu_mod_timer(n->tx_timer,
 705                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 706         n->tx_timer_active = 1;
 707         virtio_queue_set_notification(vq, 0);
 708     }
 709 }
 710
 711 static void virtio_net_tx_timer(void *opaque)
 712 {
 713     VirtIONet *n = opaque;
 714
 715     n->tx_timer_active = 0;
 716
 717     /* Just in case the driver is not ready on more */
 718     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 719         return;
 720
 721     virtio_queue_set_notification(n->tx_vq, 1);
 722     virtio_net_flush_tx(n, n->tx_vq);
 723 }
 724
 725 static void virtio_net_save(QEMUFile *f, void *opaque)
 726 {
 727     VirtIONet *n = opaque;
 728
 729     if (n->vhost_started) {
 730         /* TODO: should we really stop the backend?
 731          * If we don't, it might keep writing to memory. */
 732         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 733         n->vhost_started = 0;
 734     }
 735     virtio_save(&n->vdev, f);
 736
 737     qemu_put_buffer(f, n->mac, ETH_ALEN);
 738     qemu_put_be32(f, n->tx_timer_active);
 739     qemu_put_be32(f, n->mergeable_rx_bufs);
 740     qemu_put_be16(f, n->status);
 741     qemu_put_byte(f, n->promisc);
 742     qemu_put_byte(f, n->allmulti);
 743     qemu_put_be32(f, n->mac_table.in_use);
 744     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 745     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 746     qemu_put_be32(f, n->has_vnet_hdr);
 747     qemu_put_byte(f, n->mac_table.multi_overflow);
 748     qemu_put_byte(f, n->mac_table.uni_overflow);
 749     qemu_put_byte(f, n->alluni);
 750     qemu_put_byte(f, n->nomulti);
 751     qemu_put_byte(f, n->nouni);
 752     qemu_put_byte(f, n->nobcast);
 753     qemu_put_byte(f, n->has_ufo);
 754 }
 755
 756 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 757 {
 758     VirtIONet *n = opaque;
 759     int i;
 760
 761     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 762         return -EINVAL;
 763
 764     virtio_load(&n->vdev, f);
 765
 766     qemu_get_buffer(f, n->mac, ETH_ALEN);
 767     n->tx_timer_active = qemu_get_be32(f);
 768     n->mergeable_rx_bufs = qemu_get_be32(f);
 769
 770     if (version_id >= 3)
 771         n->status = qemu_get_be16(f);
 772
 773     if (version_id >= 4) {
 774         if (version_id < 8) {
 775             n->promisc = qemu_get_be32(f);
 776             n->allmulti = qemu_get_be32(f);
 777         } else {
 778             n->promisc = qemu_get_byte(f);
 779             n->allmulti = qemu_get_byte(f);
 780         }
 781     }
 782
 783     if (version_id >= 5) {
 784         n->mac_table.in_use = qemu_get_be32(f);
 785         /* MAC_TABLE_ENTRIES may be different from the saved image */
 786         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 787             qemu_get_buffer(f, n->mac_table.macs,
 788                             n->mac_table.in_use * ETH_ALEN);
 789         } else if (n->mac_table.in_use) {
 790             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 791             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 792             n->mac_table.in_use = 0;
 793         }
 794     }
 795
 796     if (version_id >= 6)
 797         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 798
 799     if (version_id >= 7) {
 800         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 801             error_report("virtio-net: saved image requires vnet_hdr=on");
 802             return -1;
 803         }
 804
 805         if (n->has_vnet_hdr) {
 806             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 807             tap_set_offload(n->nic->nc.peer,
 808                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 809                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 810                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 811                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 812                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 813         }
 814     }
 815
 816     if (version_id >= 9) {
 817         n->mac_table.multi_overflow = qemu_get_byte(f);
 818         n->mac_table.uni_overflow = qemu_get_byte(f);
 819     }
 820
 821     if (version_id >= 10) {
 822         n->alluni = qemu_get_byte(f);
 823         n->nomulti = qemu_get_byte(f);
 824         n->nouni = qemu_get_byte(f);
 825         n->nobcast = qemu_get_byte(f);
 826     }
 827
 828     if (version_id >= 11) {
 829         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 830             error_report("virtio-net: saved image requires TUN_F_UFO support");
 831             return -1;
 832         }
 833     }
 834
 835     /* Find the first multicast entry in the saved MAC filter */
 836     for (i = 0; i < n->mac_table.in_use; i++) {
 837         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 838             break;
 839         }
 840     }
 841     n->mac_table.first_multi = i;
 842
 843     if (n->tx_timer_active) {
 844         qemu_mod_timer(n->tx_timer,
 845                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 846     }
 847     return 0;
 848 }
 849
 850 static void virtio_net_cleanup(VLANClientState *nc)
 851 {
 852     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 853
 854     n->nic = NULL;
 855 }
 856
 857 static NetClientInfo net_virtio_info = {
 858     .type = NET_CLIENT_TYPE_NIC,
 859     .size = sizeof(NICState),
 860     .can_receive = virtio_net_can_receive,
 861     .receive = virtio_net_receive,
 862         .cleanup = virtio_net_cleanup,
 863     .link_status_changed = virtio_net_set_link_status,
 864 };
 865
 866 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 867 {
 868     VirtIONet *n = to_virtio_net(vdev);
 869     if (!n->nic->nc.peer) {
 870         return;
 871     }
 872     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 873         return;
 874     }
 875
 876     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 877         return;
 878     }
 879     if (!!n->vhost_started == !!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 880         return;
 881     }
 882     if (status & VIRTIO_CONFIG_S_DRIVER_OK) {
 883         int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), vdev);
 884         if (r < 0) {
 885             fprintf(stderr, "unable to start vhost net: %d: "
 886                     "falling back on userspace virtio\n", -r);
 887         } else {
 888             n->vhost_started = 1;
 889         }
 890     } else {
 891         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 892         n->vhost_started = 0;
 893     }
 894 }
 895
 896 static void virtio_net_vmstate_change(void *opaque, int running, int reason)
 897 {
 898     VirtIONet *n = opaque;
 899     uint8_t status = running ? VIRTIO_CONFIG_S_DRIVER_OK : 0;
 900     /* This is called when vm is started/stopped,
 901      * it will start/stop vhost backend if * appropriate
 902      * e.g. after migration. */
 903     virtio_net_set_status(&n->vdev, n->vdev.status & status);
 904 }
 905
 906 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
 907 {
 908     VirtIONet *n;
 909
 910     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 911                                         sizeof(struct virtio_net_config),
 912                                         sizeof(VirtIONet));
 913
 914     n->vdev.get_config = virtio_net_get_config;
 915     n->vdev.set_config = virtio_net_set_config;
 916     n->vdev.get_features = virtio_net_get_features;
 917     n->vdev.set_features = virtio_net_set_features;
 918     n->vdev.bad_features = virtio_net_bad_features;
 919     n->vdev.reset = virtio_net_reset;
 920     n->vdev.set_status = virtio_net_set_status;
 921     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 922     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 923     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 924     qemu_macaddr_default_if_unset(&conf->macaddr);
 925     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
 926     n->status = VIRTIO_NET_S_LINK_UP;
 927
 928     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
 929
 930     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
 931
 932     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 933     n->tx_timer_active = 0;
 934     n->mergeable_rx_bufs = 0;
 935     n->promisc = 1; /* for compatibility */
 936
 937     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 938
 939     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 940
 941     n->qdev = dev;
 942     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
 943                     virtio_net_save, virtio_net_load, n);
 944     n->vmstate = qemu_add_vm_change_state_handler(virtio_net_vmstate_change, n);
 945
 946     return &n->vdev;
 947 }
 948
 949 void virtio_net_exit(VirtIODevice *vdev)
 950 {
 951     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 952     qemu_del_vm_change_state_handler(n->vmstate);
 953
 954     if (n->vhost_started) {
 955         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 956     }
 957
 958     qemu_purge_queued_packets(&n->nic->nc);
 959
 960     unregister_savevm(n->qdev, "virtio-net", n);
 961
 962     qemu_free(n->mac_table.macs);
 963     qemu_free(n->vlans);
 964
 965     qemu_del_timer(n->tx_timer);
 966     qemu_free_timer(n->tx_timer);
 967
 968     virtio_cleanup(&n->vdev);
 969     qemu_del_vlan_client(&n->nic->nc);
 970 }