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     QEMUBH *tx_bh;
  40     uint32_t tx_timeout;
  41     int32_t tx_burst;
  42     int tx_waiting;
  43     uint32_t has_vnet_hdr;
  44     uint8_t has_ufo;
  45     struct {
  46         VirtQueueElement elem;
  47         ssize_t len;
  48     } async_tx;
  49     int mergeable_rx_bufs;
  50     uint8_t promisc;
  51     uint8_t allmulti;
  52     uint8_t alluni;
  53     uint8_t nomulti;
  54     uint8_t nouni;
  55     uint8_t nobcast;
  56     uint8_t vhost_started;
  57     VMChangeStateEntry *vmstate;
  58     struct {
  59         int in_use;
  60         int first_multi;
  61         uint8_t multi_overflow;
  62         uint8_t uni_overflow;
  63         uint8_t *macs;
  64     } mac_table;
  65     uint32_t *vlans;
  66     DeviceState *qdev;
  67 } VirtIONet;
  68
  69 /* TODO
  70  * - we could suppress RX interrupt if we were so inclined.
  71  */
  72
  73 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  74 {
  75     return (VirtIONet *)vdev;
  76 }
  77
  78 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  79 {
  80     VirtIONet *n = to_virtio_net(vdev);
  81     struct virtio_net_config netcfg;
  82
  83     netcfg.status = n->status;
  84     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  85     memcpy(config, &netcfg, sizeof(netcfg));
  86 }
  87
  88 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  89 {
  90     VirtIONet *n = to_virtio_net(vdev);
  91     struct virtio_net_config netcfg;
  92
  93     memcpy(&netcfg, config, sizeof(netcfg));
  94
  95     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  96         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  97         qemu_format_nic_info_str(&n->nic->nc, n->mac);
  98     }
  99 }
 100
 101 static void virtio_net_set_link_status(VLANClientState *nc)
 102 {
 103     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 104     uint16_t old_status = n->status;
 105
 106     if (nc->link_down)
 107         n->status &= ~VIRTIO_NET_S_LINK_UP;
 108     else
 109         n->status |= VIRTIO_NET_S_LINK_UP;
 110
 111     if (n->status != old_status)
 112         virtio_notify_config(&n->vdev);
 113 }
 114
 115 static void virtio_net_reset(VirtIODevice *vdev)
 116 {
 117     VirtIONet *n = to_virtio_net(vdev);
 118
 119     /* Reset back to compatibility mode */
 120     n->promisc = 1;
 121     n->allmulti = 0;
 122     n->alluni = 0;
 123     n->nomulti = 0;
 124     n->nouni = 0;
 125     n->nobcast = 0;
 126     if (n->vhost_started) {
 127         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 128         n->vhost_started = 0;
 129     }
 130
 131     /* Flush any MAC and VLAN filter table state */
 132     n->mac_table.in_use = 0;
 133     n->mac_table.first_multi = 0;
 134     n->mac_table.multi_overflow = 0;
 135     n->mac_table.uni_overflow = 0;
 136     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 137     memset(n->vlans, 0, MAX_VLAN >> 3);
 138 }
 139
 140 static int peer_has_vnet_hdr(VirtIONet *n)
 141 {
 142     if (!n->nic->nc.peer)
 143         return 0;
 144
 145     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 146         return 0;
 147
 148     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 149
 150     return n->has_vnet_hdr;
 151 }
 152
 153 static int peer_has_ufo(VirtIONet *n)
 154 {
 155     if (!peer_has_vnet_hdr(n))
 156         return 0;
 157
 158     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 159
 160     return n->has_ufo;
 161 }
 162
 163 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 164 {
 165     VirtIONet *n = to_virtio_net(vdev);
 166
 167     features |= (1 << VIRTIO_NET_F_MAC);
 168
 169     if (peer_has_vnet_hdr(n)) {
 170         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 171     } else {
 172         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 173         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 174         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 175         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 176
 177         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 178         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 179         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 180         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 181     }
 182
 183     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 184         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 185         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 186     }
 187
 188     if (!n->nic->nc.peer ||
 189         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 190         return features;
 191     }
 192     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 193         return features;
 194     }
 195     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 196 }
 197
 198 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 199 {
 200     uint32_t features = 0;
 201
 202     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 203      * but also these: */
 204     features |= (1 << VIRTIO_NET_F_MAC);
 205     features |= (1 << VIRTIO_NET_F_CSUM);
 206     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 207     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 208     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 209
 210     return features;
 211 }
 212
 213 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 214 {
 215     VirtIONet *n = to_virtio_net(vdev);
 216
 217     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 218
 219     if (n->has_vnet_hdr) {
 220         tap_set_offload(n->nic->nc.peer,
 221                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 222                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 223                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 224                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 225                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 226     }
 227     if (!n->nic->nc.peer ||
 228         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 229         return;
 230     }
 231     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 232         return;
 233     }
 234     vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 235 }
 236
 237 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 238                                      VirtQueueElement *elem)
 239 {
 240     uint8_t on;
 241
 242     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 243         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 244         exit(1);
 245     }
 246
 247     on = ldub_p(elem->out_sg[1].iov_base);
 248
 249     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 250         n->promisc = on;
 251     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 252         n->allmulti = on;
 253     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 254         n->alluni = on;
 255     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 256         n->nomulti = on;
 257     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 258         n->nouni = on;
 259     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 260         n->nobcast = on;
 261     else
 262         return VIRTIO_NET_ERR;
 263
 264     return VIRTIO_NET_OK;
 265 }
 266
 267 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 268                                  VirtQueueElement *elem)
 269 {
 270     struct virtio_net_ctrl_mac mac_data;
 271
 272     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 273         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 274         elem->out_sg[2].iov_len < sizeof(mac_data))
 275         return VIRTIO_NET_ERR;
 276
 277     n->mac_table.in_use = 0;
 278     n->mac_table.first_multi = 0;
 279     n->mac_table.uni_overflow = 0;
 280     n->mac_table.multi_overflow = 0;
 281     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 282
 283     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 284
 285     if (sizeof(mac_data.entries) +
 286         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 287         return VIRTIO_NET_ERR;
 288
 289     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 290         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 291                mac_data.entries * ETH_ALEN);
 292         n->mac_table.in_use += mac_data.entries;
 293     } else {
 294         n->mac_table.uni_overflow = 1;
 295     }
 296
 297     n->mac_table.first_multi = n->mac_table.in_use;
 298
 299     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 300
 301     if (sizeof(mac_data.entries) +
 302         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 303         return VIRTIO_NET_ERR;
 304
 305     if (mac_data.entries) {
 306         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 307             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 308                    elem->out_sg[2].iov_base + sizeof(mac_data),
 309                    mac_data.entries * ETH_ALEN);
 310             n->mac_table.in_use += mac_data.entries;
 311         } else {
 312             n->mac_table.multi_overflow = 1;
 313         }
 314     }
 315
 316     return VIRTIO_NET_OK;
 317 }
 318
 319 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 320                                         VirtQueueElement *elem)
 321 {
 322     uint16_t vid;
 323
 324     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 325         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 326         return VIRTIO_NET_ERR;
 327     }
 328
 329     vid = lduw_le_p(elem->out_sg[1].iov_base);
 330
 331     if (vid >= MAX_VLAN)
 332         return VIRTIO_NET_ERR;
 333
 334     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 335         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 336     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 337         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 338     else
 339         return VIRTIO_NET_ERR;
 340
 341     return VIRTIO_NET_OK;
 342 }
 343
 344 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 345 {
 346     VirtIONet *n = to_virtio_net(vdev);
 347     struct virtio_net_ctrl_hdr ctrl;
 348     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 349     VirtQueueElement elem;
 350
 351     while (virtqueue_pop(vq, &elem)) {
 352         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 353             fprintf(stderr, "virtio-net ctrl missing headers\n");
 354             exit(1);
 355         }
 356
 357         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 358             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 359             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 360             exit(1);
 361         }
 362
 363         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 364         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 365
 366         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 367             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 368         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 369             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 370         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 371             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 372
 373         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 374
 375         virtqueue_push(vq, &elem, sizeof(status));
 376         virtio_notify(vdev, vq);
 377     }
 378 }
 379
 380 /* RX */
 381
 382 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 383 {
 384     VirtIONet *n = to_virtio_net(vdev);
 385
 386     qemu_flush_queued_packets(&n->nic->nc);
 387
 388     /* We now have RX buffers, signal to the IO thread to break out of the
 389      * select to re-poll the tap file descriptor */
 390     qemu_notify_event();
 391 }
 392
 393 static int virtio_net_can_receive(VLANClientState *nc)
 394 {
 395     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 396
 397     if (!virtio_queue_ready(n->rx_vq) ||
 398         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 399         return 0;
 400
 401     return 1;
 402 }
 403
 404 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 405 {
 406     if (virtio_queue_empty(n->rx_vq) ||
 407         (n->mergeable_rx_bufs &&
 408          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 409         virtio_queue_set_notification(n->rx_vq, 1);
 410
 411         /* To avoid a race condition where the guest has made some buffers
 412          * available after the above check but before notification was
 413          * enabled, check for available buffers again.
 414          */
 415         if (virtio_queue_empty(n->rx_vq) ||
 416             (n->mergeable_rx_bufs &&
 417              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 418             return 0;
 419     }
 420
 421     virtio_queue_set_notification(n->rx_vq, 0);
 422     return 1;
 423 }
 424
 425 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 426  * it never finds out that the packets don't have valid checksums.  This
 427  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 428  * fix this with Xen but it hasn't appeared in an upstream release of
 429  * dhclient yet.
 430  *
 431  * To avoid breaking existing guests, we catch udp packets and add
 432  * checksums.  This is terrible but it's better than hacking the guest
 433  * kernels.
 434  *
 435  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 436  * we should provide a mechanism to disable it to avoid polluting the host
 437  * cache.
 438  */
 439 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 440                                         const uint8_t *buf, size_t size)
 441 {
 442     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 443         (size > 27 && size < 1500) && /* normal sized MTU */
 444         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 445         (buf[23] == 17) && /* ip.protocol == UDP */
 446         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 447         /* FIXME this cast is evil */
 448         net_checksum_calculate((uint8_t *)buf, size);
 449         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 450     }
 451 }
 452
 453 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 454                           const void *buf, size_t size, size_t hdr_len)
 455 {
 456     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 457     int offset = 0;
 458
 459     hdr->flags = 0;
 460     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 461
 462     if (n->has_vnet_hdr) {
 463         memcpy(hdr, buf, sizeof(*hdr));
 464         offset = sizeof(*hdr);
 465         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 466     }
 467
 468     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 469      * but we may be passing along a larger header to the guest.
 470      */
 471     iov[0].iov_base += hdr_len;
 472     iov[0].iov_len  -= hdr_len;
 473
 474     return offset;
 475 }
 476
 477 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 478 {
 479     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 480     static const uint8_t vlan[] = {0x81, 0x00};
 481     uint8_t *ptr = (uint8_t *)buf;
 482     int i;
 483
 484     if (n->promisc)
 485         return 1;
 486
 487     if (n->has_vnet_hdr) {
 488         ptr += sizeof(struct virtio_net_hdr);
 489     }
 490
 491     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 492         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 493         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 494             return 0;
 495     }
 496
 497     if (ptr[0] & 1) { // multicast
 498         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 499             return !n->nobcast;
 500         } else if (n->nomulti) {
 501             return 0;
 502         } else if (n->allmulti || n->mac_table.multi_overflow) {
 503             return 1;
 504         }
 505
 506         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 507             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 508                 return 1;
 509             }
 510         }
 511     } else { // unicast
 512         if (n->nouni) {
 513             return 0;
 514         } else if (n->alluni || n->mac_table.uni_overflow) {
 515             return 1;
 516         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 517             return 1;
 518         }
 519
 520         for (i = 0; i < n->mac_table.first_multi; i++) {
 521             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 522                 return 1;
 523             }
 524         }
 525     }
 526
 527     return 0;
 528 }
 529
 530 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 531 {
 532     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 533     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 534     size_t guest_hdr_len, offset, i, host_hdr_len;
 535
 536     if (!virtio_net_can_receive(&n->nic->nc))
 537         return -1;
 538
 539     /* hdr_len refers to the header we supply to the guest */
 540     guest_hdr_len = n->mergeable_rx_bufs ?
 541         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 542
 543
 544     host_hdr_len = n->has_vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 545     if (!virtio_net_has_buffers(n, size + guest_hdr_len - host_hdr_len))
 546         return 0;
 547
 548     if (!receive_filter(n, buf, size))
 549         return size;
 550
 551     offset = i = 0;
 552
 553     while (offset < size) {
 554         VirtQueueElement elem;
 555         int len, total;
 556         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 557
 558         total = 0;
 559
 560         if (virtqueue_pop(n->rx_vq, &elem) == 0) {
 561             if (i == 0)
 562                 return -1;
 563             fprintf(stderr, "virtio-net unexpected empty queue: "
 564                     "i %zd mergeable %d offset %zd, size %zd, "
 565                     "guest hdr len %zd, host hdr len %zd guest features 0x%x\n",
 566                     i, n->mergeable_rx_bufs, offset, size,
 567                     guest_hdr_len, host_hdr_len, n->vdev.guest_features);
 568             exit(1);
 569         }
 570
 571         if (elem.in_num < 1) {
 572             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 573             exit(1);
 574         }
 575
 576         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != guest_hdr_len) {
 577             fprintf(stderr, "virtio-net header not in first element\n");
 578             exit(1);
 579         }
 580
 581         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 582
 583         if (i == 0) {
 584             if (n->mergeable_rx_bufs)
 585                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 586
 587             offset += receive_header(n, sg, elem.in_num,
 588                                      buf + offset, size - offset, guest_hdr_len);
 589             total += guest_hdr_len;
 590         }
 591
 592         /* copy in packet.  ugh */
 593         len = iov_from_buf(sg, elem.in_num,
 594                            buf + offset, size - offset);
 595         total += len;
 596         offset += len;
 597         /* If buffers can't be merged, at this point we
 598          * must have consumed the complete packet.
 599          * Otherwise, drop it. */
 600         if (!n->mergeable_rx_bufs && offset < size) {
 601 #if 0
 602             fprintf(stderr, "virtio-net truncated non-mergeable packet: "
 603
 604                     "i %zd mergeable %d offset %zd, size %zd, "
 605                     "guest hdr len %zd, host hdr len %zd\n",
 606                     i, n->mergeable_rx_bufs,
 607                     offset, size, guest_hdr_len, host_hdr_len);
 608 #endif
 609             return size;
 610         }
 611
 612         /* signal other side */
 613         virtqueue_fill(n->rx_vq, &elem, total, i++);
 614     }
 615
 616     if (mhdr)
 617         mhdr->num_buffers = i;
 618
 619     virtqueue_flush(n->rx_vq, i);
 620     virtio_notify(&n->vdev, n->rx_vq);
 621
 622     return size;
 623 }
 624
 625 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 626
 627 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 628 {
 629     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 630
 631     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 632     virtio_notify(&n->vdev, n->tx_vq);
 633
 634     n->async_tx.elem.out_num = n->async_tx.len = 0;
 635
 636     virtio_queue_set_notification(n->tx_vq, 1);
 637     virtio_net_flush_tx(n, n->tx_vq);
 638 }
 639
 640 /* TX */
 641 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 642 {
 643     VirtQueueElement elem;
 644     int32_t num_packets = 0;
 645
 646     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 647         return num_packets;
 648     }
 649
 650     if (n->async_tx.elem.out_num) {
 651         virtio_queue_set_notification(n->tx_vq, 0);
 652         return num_packets;
 653     }
 654
 655     while (virtqueue_pop(vq, &elem)) {
 656         ssize_t ret, len = 0;
 657         unsigned int out_num = elem.out_num;
 658         struct iovec *out_sg = &elem.out_sg[0];
 659         unsigned hdr_len;
 660
 661         /* hdr_len refers to the header received from the guest */
 662         hdr_len = n->mergeable_rx_bufs ?
 663             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 664             sizeof(struct virtio_net_hdr);
 665
 666         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 667             fprintf(stderr, "virtio-net header not in first element\n");
 668             exit(1);
 669         }
 670
 671         /* ignore the header if GSO is not supported */
 672         if (!n->has_vnet_hdr) {
 673             out_num--;
 674             out_sg++;
 675             len += hdr_len;
 676         } else if (n->mergeable_rx_bufs) {
 677             /* tapfd expects a struct virtio_net_hdr */
 678             hdr_len -= sizeof(struct virtio_net_hdr);
 679             out_sg->iov_len -= hdr_len;
 680             len += hdr_len;
 681         }
 682
 683         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 684                                       virtio_net_tx_complete);
 685         if (ret == 0) {
 686             virtio_queue_set_notification(n->tx_vq, 0);
 687             n->async_tx.elem = elem;
 688             n->async_tx.len  = len;
 689             return -EBUSY;
 690         }
 691
 692         len += ret;
 693
 694         virtqueue_push(vq, &elem, len);
 695         virtio_notify(&n->vdev, vq);
 696
 697         if (++num_packets >= n->tx_burst) {
 698             break;
 699         }
 700     }
 701     return num_packets;
 702 }
 703
 704 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 705 {
 706     VirtIONet *n = to_virtio_net(vdev);
 707
 708     if (n->tx_waiting) {
 709         virtio_queue_set_notification(vq, 1);
 710         qemu_del_timer(n->tx_timer);
 711         n->tx_waiting = 0;
 712         virtio_net_flush_tx(n, vq);
 713     } else {
 714         qemu_mod_timer(n->tx_timer,
 715                        qemu_get_clock(vm_clock) + n->tx_timeout);
 716         n->tx_waiting = 1;
 717         virtio_queue_set_notification(vq, 0);
 718     }
 719 }
 720
 721 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
 722 {
 723     VirtIONet *n = to_virtio_net(vdev);
 724
 725     if (unlikely(n->tx_waiting)) {
 726         return;
 727     }
 728     virtio_queue_set_notification(vq, 0);
 729     qemu_bh_schedule(n->tx_bh);
 730     n->tx_waiting = 1;
 731 }
 732
 733 static void virtio_net_tx_timer(void *opaque)
 734 {
 735     VirtIONet *n = opaque;
 736
 737     n->tx_waiting = 0;
 738
 739     /* Just in case the driver is not ready on more */
 740     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 741         return;
 742
 743     virtio_queue_set_notification(n->tx_vq, 1);
 744     virtio_net_flush_tx(n, n->tx_vq);
 745 }
 746
 747 static void virtio_net_tx_bh(void *opaque)
 748 {
 749     VirtIONet *n = opaque;
 750     int32_t ret;
 751
 752     n->tx_waiting = 0;
 753
 754     /* Just in case the driver is not ready on more */
 755     if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
 756         return;
 757
 758     ret = virtio_net_flush_tx(n, n->tx_vq);
 759     if (ret == -EBUSY) {
 760         return; /* Notification re-enable handled by tx_complete */
 761     }
 762
 763     /* If we flush a full burst of packets, assume there are
 764      * more coming and immediately reschedule */
 765     if (ret >= n->tx_burst) {
 766         qemu_bh_schedule(n->tx_bh);
 767         n->tx_waiting = 1;
 768         return;
 769     }
 770
 771     /* If less than a full burst, re-enable notification and flush
 772      * anything that may have come in while we weren't looking.  If
 773      * we find something, assume the guest is still active and reschedule */
 774     virtio_queue_set_notification(n->tx_vq, 1);
 775     if (virtio_net_flush_tx(n, n->tx_vq) > 0) {
 776         virtio_queue_set_notification(n->tx_vq, 0);
 777         qemu_bh_schedule(n->tx_bh);
 778         n->tx_waiting = 1;
 779     }
 780 }
 781
 782 static void virtio_net_save(QEMUFile *f, void *opaque)
 783 {
 784     VirtIONet *n = opaque;
 785
 786     if (n->vhost_started) {
 787         /* TODO: should we really stop the backend?
 788          * If we don't, it might keep writing to memory. */
 789         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 790         n->vhost_started = 0;
 791     }
 792     virtio_save(&n->vdev, f);
 793
 794     qemu_put_buffer(f, n->mac, ETH_ALEN);
 795     qemu_put_be32(f, n->tx_waiting);
 796     qemu_put_be32(f, n->mergeable_rx_bufs);
 797     qemu_put_be16(f, n->status);
 798     qemu_put_byte(f, n->promisc);
 799     qemu_put_byte(f, n->allmulti);
 800     qemu_put_be32(f, n->mac_table.in_use);
 801     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 802     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 803     qemu_put_be32(f, n->has_vnet_hdr);
 804     qemu_put_byte(f, n->mac_table.multi_overflow);
 805     qemu_put_byte(f, n->mac_table.uni_overflow);
 806     qemu_put_byte(f, n->alluni);
 807     qemu_put_byte(f, n->nomulti);
 808     qemu_put_byte(f, n->nouni);
 809     qemu_put_byte(f, n->nobcast);
 810     qemu_put_byte(f, n->has_ufo);
 811 }
 812
 813 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 814 {
 815     VirtIONet *n = opaque;
 816     int i;
 817
 818     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 819         return -EINVAL;
 820
 821     virtio_load(&n->vdev, f);
 822
 823     qemu_get_buffer(f, n->mac, ETH_ALEN);
 824     n->tx_waiting = qemu_get_be32(f);
 825     n->mergeable_rx_bufs = qemu_get_be32(f);
 826
 827     if (version_id >= 3)
 828         n->status = qemu_get_be16(f);
 829
 830     if (version_id >= 4) {
 831         if (version_id < 8) {
 832             n->promisc = qemu_get_be32(f);
 833             n->allmulti = qemu_get_be32(f);
 834         } else {
 835             n->promisc = qemu_get_byte(f);
 836             n->allmulti = qemu_get_byte(f);
 837         }
 838     }
 839
 840     if (version_id >= 5) {
 841         n->mac_table.in_use = qemu_get_be32(f);
 842         /* MAC_TABLE_ENTRIES may be different from the saved image */
 843         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 844             qemu_get_buffer(f, n->mac_table.macs,
 845                             n->mac_table.in_use * ETH_ALEN);
 846         } else if (n->mac_table.in_use) {
 847             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 848             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 849             n->mac_table.in_use = 0;
 850         }
 851     }
 852
 853     if (version_id >= 6)
 854         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 855
 856     if (version_id >= 7) {
 857         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 858             error_report("virtio-net: saved image requires vnet_hdr=on");
 859             return -1;
 860         }
 861
 862         if (n->has_vnet_hdr) {
 863             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 864             tap_set_offload(n->nic->nc.peer,
 865                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 866                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 867                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 868                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 869                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 870         }
 871     }
 872
 873     if (version_id >= 9) {
 874         n->mac_table.multi_overflow = qemu_get_byte(f);
 875         n->mac_table.uni_overflow = qemu_get_byte(f);
 876     }
 877
 878     if (version_id >= 10) {
 879         n->alluni = qemu_get_byte(f);
 880         n->nomulti = qemu_get_byte(f);
 881         n->nouni = qemu_get_byte(f);
 882         n->nobcast = qemu_get_byte(f);
 883     }
 884
 885     if (version_id >= 11) {
 886         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 887             error_report("virtio-net: saved image requires TUN_F_UFO support");
 888             return -1;
 889         }
 890     }
 891
 892     /* Find the first multicast entry in the saved MAC filter */
 893     for (i = 0; i < n->mac_table.in_use; i++) {
 894         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 895             break;
 896         }
 897     }
 898     n->mac_table.first_multi = i;
 899
 900     if (n->tx_waiting) {
 901         if (n->tx_timer) {
 902             qemu_mod_timer(n->tx_timer,
 903                            qemu_get_clock(vm_clock) + n->tx_timeout);
 904         } else {
 905             qemu_bh_schedule(n->tx_bh);
 906         }
 907     }
 908     return 0;
 909 }
 910
 911 static void virtio_net_cleanup(VLANClientState *nc)
 912 {
 913     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 914
 915     n->nic = NULL;
 916 }
 917
 918 static NetClientInfo net_virtio_info = {
 919     .type = NET_CLIENT_TYPE_NIC,
 920     .size = sizeof(NICState),
 921     .can_receive = virtio_net_can_receive,
 922     .receive = virtio_net_receive,
 923         .cleanup = virtio_net_cleanup,
 924     .link_status_changed = virtio_net_set_link_status,
 925 };
 926
 927 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 928 {
 929     VirtIONet *n = to_virtio_net(vdev);
 930     if (!n->nic->nc.peer) {
 931         return;
 932     }
 933     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 934         return;
 935     }
 936
 937     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 938         return;
 939     }
 940     if (!!n->vhost_started == !!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 941         return;
 942     }
 943     if (status & VIRTIO_CONFIG_S_DRIVER_OK) {
 944         int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), vdev);
 945         if (r < 0) {
 946             fprintf(stderr, "unable to start vhost net: %d: "
 947                     "falling back on userspace virtio\n", -r);
 948         } else {
 949             n->vhost_started = 1;
 950         }
 951     } else {
 952         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 953         n->vhost_started = 0;
 954     }
 955 }
 956
 957 static void virtio_net_vmstate_change(void *opaque, int running, int reason)
 958 {
 959     VirtIONet *n = opaque;
 960     uint8_t status = running ? VIRTIO_CONFIG_S_DRIVER_OK : 0;
 961     /* This is called when vm is started/stopped,
 962      * it will start/stop vhost backend if * appropriate
 963      * e.g. after migration. */
 964     virtio_net_set_status(&n->vdev, n->vdev.status & status);
 965 }
 966
 967 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
 968                               virtio_net_conf *net)
 969 {
 970     VirtIONet *n;
 971
 972     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 973                                         sizeof(struct virtio_net_config),
 974                                         sizeof(VirtIONet));
 975
 976     n->vdev.get_config = virtio_net_get_config;
 977     n->vdev.set_config = virtio_net_set_config;
 978     n->vdev.get_features = virtio_net_get_features;
 979     n->vdev.set_features = virtio_net_set_features;
 980     n->vdev.bad_features = virtio_net_bad_features;
 981     n->vdev.reset = virtio_net_reset;
 982     n->vdev.set_status = virtio_net_set_status;
 983     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 984
 985     if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
 986         fprintf(stderr, "virtio-net: "
 987                 "Unknown option tx=%s, valid options: \"timer\" \"bh\"\n",
 988                 net->tx);
 989         fprintf(stderr, "Defaulting to \"bh\"\n");
 990     }
 991
 992     if (net->tx && !strcmp(net->tx, "timer")) {
 993         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_timer);
 994         n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 995         n->tx_timeout = net->txtimer;
 996     } else {
 997         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_bh);
 998         n->tx_bh = qemu_bh_new(virtio_net_tx_bh, n);
 999     }
1000     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
1001     qemu_macaddr_default_if_unset(&conf->macaddr);
1002     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
1003     n->status = VIRTIO_NET_S_LINK_UP;
1004
1005     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
1006
1007     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
1008
1009     n->tx_waiting = 0;
1010     n->tx_burst = net->txburst;
1011     n->mergeable_rx_bufs = 0;
1012     n->promisc = 1; /* for compatibility */
1013
1014     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
1015
1016     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
1017
1018     n->qdev = dev;
1019     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1020                     virtio_net_save, virtio_net_load, n);
1021     n->vmstate = qemu_add_vm_change_state_handler(virtio_net_vmstate_change, n);
1022
1023     return &n->vdev;
1024 }
1025
1026 void virtio_net_exit(VirtIODevice *vdev)
1027 {
1028     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
1029     qemu_del_vm_change_state_handler(n->vmstate);
1030
1031     if (n->vhost_started) {
1032         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
1033     }
1034
1035     qemu_purge_queued_packets(&n->nic->nc);
1036
1037     unregister_savevm(n->qdev, "virtio-net", n);
1038
1039     qemu_free(n->mac_table.macs);
1040     qemu_free(n->vlans);
1041
1042     if (n->tx_timer) {
1043         qemu_del_timer(n->tx_timer);
1044         qemu_free_timer(n->tx_timer);
1045     } else {
1046         qemu_bh_delete(n->tx_bh);
1047     }
1048
1049     virtio_cleanup(&n->vdev);
1050     qemu_del_vlan_client(&n->nic->nc);
1051 }