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     struct {
  58         int in_use;
  59         int first_multi;
  60         uint8_t multi_overflow;
  61         uint8_t uni_overflow;
  62         uint8_t *macs;
  63     } mac_table;
  64     uint32_t *vlans;
  65     DeviceState *qdev;
  66 } VirtIONet;
  67
  68 /* TODO
  69  * - we could suppress RX interrupt if we were so inclined.
  70  */
  71
  72 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  73 {
  74     return (VirtIONet *)vdev;
  75 }
  76
  77 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  78 {
  79     VirtIONet *n = to_virtio_net(vdev);
  80     struct virtio_net_config netcfg;
  81
  82     stw_p(&netcfg.status, n->status);
  83     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  84     memcpy(config, &netcfg, sizeof(netcfg));
  85 }
  86
  87 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  88 {
  89     VirtIONet *n = to_virtio_net(vdev);
  90     struct virtio_net_config netcfg;
  91
  92     memcpy(&netcfg, config, sizeof(netcfg));
  93
  94     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  95         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  96         qemu_format_nic_info_str(&n->nic->nc, n->mac);
  97     }
  98 }
  99
 100 static bool virtio_net_started(VirtIONet *n, uint8_t status)
 101 {
 102     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 103         (n->status & VIRTIO_NET_S_LINK_UP) && n->vdev.vm_running;
 104 }
 105
 106 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
 107 {
 108     if (!n->nic->nc.peer) {
 109         return;
 110     }
 111     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 112         return;
 113     }
 114
 115     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 116         return;
 117     }
 118     if (!!n->vhost_started == virtio_net_started(n, status) &&
 119                               !n->nic->nc.peer->link_down) {
 120         return;
 121     }
 122     if (!n->vhost_started) {
 123         int r;
 124         if (!vhost_net_query(tap_get_vhost_net(n->nic->nc.peer), &n->vdev)) {
 125             return;
 126         }
 127         r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 128         if (r < 0) {
 129             error_report("unable to start vhost net: %d: "
 130                          "falling back on userspace virtio", -r);
 131         } else {
 132             n->vhost_started = 1;
 133         }
 134     } else {
 135         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 136         n->vhost_started = 0;
 137     }
 138 }
 139
 140 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 141 {
 142     VirtIONet *n = to_virtio_net(vdev);
 143
 144     virtio_net_vhost_status(n, status);
 145
 146     if (!n->tx_waiting) {
 147         return;
 148     }
 149
 150     if (virtio_net_started(n, status) && !n->vhost_started) {
 151         if (n->tx_timer) {
 152             qemu_mod_timer(n->tx_timer,
 153                            qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 154         } else {
 155             qemu_bh_schedule(n->tx_bh);
 156         }
 157     } else {
 158         if (n->tx_timer) {
 159             qemu_del_timer(n->tx_timer);
 160         } else {
 161             qemu_bh_cancel(n->tx_bh);
 162         }
 163     }
 164 }
 165
 166 static void virtio_net_set_link_status(VLANClientState *nc)
 167 {
 168     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 169     uint16_t old_status = n->status;
 170
 171     if (nc->link_down)
 172         n->status &= ~VIRTIO_NET_S_LINK_UP;
 173     else
 174         n->status |= VIRTIO_NET_S_LINK_UP;
 175
 176     if (n->status != old_status)
 177         virtio_notify_config(&n->vdev);
 178
 179     virtio_net_set_status(&n->vdev, n->vdev.status);
 180 }
 181
 182 static void virtio_net_reset(VirtIODevice *vdev)
 183 {
 184     VirtIONet *n = to_virtio_net(vdev);
 185
 186     /* Reset back to compatibility mode */
 187     n->promisc = 1;
 188     n->allmulti = 0;
 189     n->alluni = 0;
 190     n->nomulti = 0;
 191     n->nouni = 0;
 192     n->nobcast = 0;
 193
 194     /* Flush any MAC and VLAN filter table state */
 195     n->mac_table.in_use = 0;
 196     n->mac_table.first_multi = 0;
 197     n->mac_table.multi_overflow = 0;
 198     n->mac_table.uni_overflow = 0;
 199     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 200     memset(n->vlans, 0, MAX_VLAN >> 3);
 201 }
 202
 203 static int peer_has_vnet_hdr(VirtIONet *n)
 204 {
 205     if (!n->nic->nc.peer)
 206         return 0;
 207
 208     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 209         return 0;
 210
 211     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 212
 213     return n->has_vnet_hdr;
 214 }
 215
 216 static int peer_has_ufo(VirtIONet *n)
 217 {
 218     if (!peer_has_vnet_hdr(n))
 219         return 0;
 220
 221     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 222
 223     return n->has_ufo;
 224 }
 225
 226 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 227 {
 228     VirtIONet *n = to_virtio_net(vdev);
 229
 230     features |= (1 << VIRTIO_NET_F_MAC);
 231
 232     if (peer_has_vnet_hdr(n)) {
 233         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 234     } else {
 235         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 236         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 237         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 238         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 239
 240         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 241         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 242         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 243         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 244     }
 245
 246     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 247         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 248         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 249     }
 250
 251     if (!n->nic->nc.peer ||
 252         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 253         return features;
 254     }
 255     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 256         return features;
 257     }
 258     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 259 }
 260
 261 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 262 {
 263     uint32_t features = 0;
 264
 265     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 266      * but also these: */
 267     features |= (1 << VIRTIO_NET_F_MAC);
 268     features |= (1 << VIRTIO_NET_F_CSUM);
 269     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 270     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 271     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 272
 273     return features;
 274 }
 275
 276 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 277 {
 278     VirtIONet *n = to_virtio_net(vdev);
 279
 280     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 281
 282     if (n->has_vnet_hdr) {
 283         tap_set_offload(n->nic->nc.peer,
 284                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 285                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 286                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 287                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 288                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 289     }
 290     if (!n->nic->nc.peer ||
 291         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 292         return;
 293     }
 294     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 295         return;
 296     }
 297     vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 298 }
 299
 300 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 301                                      VirtQueueElement *elem)
 302 {
 303     uint8_t on;
 304
 305     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 306         error_report("virtio-net ctrl invalid rx mode command");
 307         exit(1);
 308     }
 309
 310     on = ldub_p(elem->out_sg[1].iov_base);
 311
 312     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 313         n->promisc = on;
 314     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 315         n->allmulti = on;
 316     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 317         n->alluni = on;
 318     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 319         n->nomulti = on;
 320     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 321         n->nouni = on;
 322     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 323         n->nobcast = on;
 324     else
 325         return VIRTIO_NET_ERR;
 326
 327     return VIRTIO_NET_OK;
 328 }
 329
 330 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 331                                  VirtQueueElement *elem)
 332 {
 333     struct virtio_net_ctrl_mac mac_data;
 334
 335     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 336         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 337         elem->out_sg[2].iov_len < sizeof(mac_data))
 338         return VIRTIO_NET_ERR;
 339
 340     n->mac_table.in_use = 0;
 341     n->mac_table.first_multi = 0;
 342     n->mac_table.uni_overflow = 0;
 343     n->mac_table.multi_overflow = 0;
 344     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 345
 346     mac_data.entries = ldl_p(elem->out_sg[1].iov_base);
 347
 348     if (sizeof(mac_data.entries) +
 349         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 350         return VIRTIO_NET_ERR;
 351
 352     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 353         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 354                mac_data.entries * ETH_ALEN);
 355         n->mac_table.in_use += mac_data.entries;
 356     } else {
 357         n->mac_table.uni_overflow = 1;
 358     }
 359
 360     n->mac_table.first_multi = n->mac_table.in_use;
 361
 362     mac_data.entries = ldl_p(elem->out_sg[2].iov_base);
 363
 364     if (sizeof(mac_data.entries) +
 365         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 366         return VIRTIO_NET_ERR;
 367
 368     if (mac_data.entries) {
 369         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 370             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 371                    elem->out_sg[2].iov_base + sizeof(mac_data),
 372                    mac_data.entries * ETH_ALEN);
 373             n->mac_table.in_use += mac_data.entries;
 374         } else {
 375             n->mac_table.multi_overflow = 1;
 376         }
 377     }
 378
 379     return VIRTIO_NET_OK;
 380 }
 381
 382 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 383                                         VirtQueueElement *elem)
 384 {
 385     uint16_t vid;
 386
 387     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 388         error_report("virtio-net ctrl invalid vlan command");
 389         return VIRTIO_NET_ERR;
 390     }
 391
 392     vid = lduw_p(elem->out_sg[1].iov_base);
 393
 394     if (vid >= MAX_VLAN)
 395         return VIRTIO_NET_ERR;
 396
 397     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 398         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 399     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 400         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 401     else
 402         return VIRTIO_NET_ERR;
 403
 404     return VIRTIO_NET_OK;
 405 }
 406
 407 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 408 {
 409     VirtIONet *n = to_virtio_net(vdev);
 410     struct virtio_net_ctrl_hdr ctrl;
 411     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 412     VirtQueueElement elem;
 413
 414     while (virtqueue_pop(vq, &elem)) {
 415         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 416             error_report("virtio-net ctrl missing headers");
 417             exit(1);
 418         }
 419
 420         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 421             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 422             error_report("virtio-net ctrl header not in correct element");
 423             exit(1);
 424         }
 425
 426         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 427         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 428
 429         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 430             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 431         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 432             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 433         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 434             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 435
 436         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 437
 438         virtqueue_push(vq, &elem, sizeof(status));
 439         virtio_notify(vdev, vq);
 440     }
 441 }
 442
 443 /* RX */
 444
 445 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 446 {
 447     VirtIONet *n = to_virtio_net(vdev);
 448
 449     qemu_flush_queued_packets(&n->nic->nc);
 450
 451     /* We now have RX buffers, signal to the IO thread to break out of the
 452      * select to re-poll the tap file descriptor */
 453     qemu_notify_event();
 454 }
 455
 456 static int virtio_net_can_receive(VLANClientState *nc)
 457 {
 458     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 459     if (!n->vdev.vm_running) {
 460         return 0;
 461     }
 462
 463     if (!virtio_queue_ready(n->rx_vq) ||
 464         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 465         return 0;
 466
 467     return 1;
 468 }
 469
 470 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 471 {
 472     if (virtio_queue_empty(n->rx_vq) ||
 473         (n->mergeable_rx_bufs &&
 474          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 475         virtio_queue_set_notification(n->rx_vq, 1);
 476
 477         /* To avoid a race condition where the guest has made some buffers
 478          * available after the above check but before notification was
 479          * enabled, check for available buffers again.
 480          */
 481         if (virtio_queue_empty(n->rx_vq) ||
 482             (n->mergeable_rx_bufs &&
 483              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 484             return 0;
 485     }
 486
 487     virtio_queue_set_notification(n->rx_vq, 0);
 488     return 1;
 489 }
 490
 491 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 492  * it never finds out that the packets don't have valid checksums.  This
 493  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 494  * fix this with Xen but it hasn't appeared in an upstream release of
 495  * dhclient yet.
 496  *
 497  * To avoid breaking existing guests, we catch udp packets and add
 498  * checksums.  This is terrible but it's better than hacking the guest
 499  * kernels.
 500  *
 501  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 502  * we should provide a mechanism to disable it to avoid polluting the host
 503  * cache.
 504  */
 505 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 506                                         const uint8_t *buf, size_t size)
 507 {
 508     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 509         (size > 27 && size < 1500) && /* normal sized MTU */
 510         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 511         (buf[23] == 17) && /* ip.protocol == UDP */
 512         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 513         /* FIXME this cast is evil */
 514         net_checksum_calculate((uint8_t *)buf, size);
 515         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 516     }
 517 }
 518
 519 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 520                           const void *buf, size_t size, size_t hdr_len)
 521 {
 522     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 523     int offset = 0;
 524
 525     hdr->flags = 0;
 526     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 527
 528     if (n->has_vnet_hdr) {
 529         memcpy(hdr, buf, sizeof(*hdr));
 530         offset = sizeof(*hdr);
 531         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 532     }
 533
 534     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 535      * but we may be passing along a larger header to the guest.
 536      */
 537     iov[0].iov_base += hdr_len;
 538     iov[0].iov_len  -= hdr_len;
 539
 540     return offset;
 541 }
 542
 543 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 544 {
 545     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 546     static const uint8_t vlan[] = {0x81, 0x00};
 547     uint8_t *ptr = (uint8_t *)buf;
 548     int i;
 549
 550     if (n->promisc)
 551         return 1;
 552
 553     if (n->has_vnet_hdr) {
 554         ptr += sizeof(struct virtio_net_hdr);
 555     }
 556
 557     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 558         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 559         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 560             return 0;
 561     }
 562
 563     if (ptr[0] & 1) { // multicast
 564         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 565             return !n->nobcast;
 566         } else if (n->nomulti) {
 567             return 0;
 568         } else if (n->allmulti || n->mac_table.multi_overflow) {
 569             return 1;
 570         }
 571
 572         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 573             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 574                 return 1;
 575             }
 576         }
 577     } else { // unicast
 578         if (n->nouni) {
 579             return 0;
 580         } else if (n->alluni || n->mac_table.uni_overflow) {
 581             return 1;
 582         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 583             return 1;
 584         }
 585
 586         for (i = 0; i < n->mac_table.first_multi; i++) {
 587             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 588                 return 1;
 589             }
 590         }
 591     }
 592
 593     return 0;
 594 }
 595
 596 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 597 {
 598     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 599     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 600     size_t guest_hdr_len, offset, i, host_hdr_len;
 601
 602     if (!virtio_net_can_receive(&n->nic->nc))
 603         return -1;
 604
 605     /* hdr_len refers to the header we supply to the guest */
 606     guest_hdr_len = n->mergeable_rx_bufs ?
 607         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 608
 609
 610     host_hdr_len = n->has_vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 611     if (!virtio_net_has_buffers(n, size + guest_hdr_len - host_hdr_len))
 612         return 0;
 613
 614     if (!receive_filter(n, buf, size))
 615         return size;
 616
 617     offset = i = 0;
 618
 619     while (offset < size) {
 620         VirtQueueElement elem;
 621         int len, total;
 622         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 623
 624         total = 0;
 625
 626         if (virtqueue_pop(n->rx_vq, &elem) == 0) {
 627             if (i == 0)
 628                 return -1;
 629             error_report("virtio-net unexpected empty queue: "
 630                     "i %zd mergeable %d offset %zd, size %zd, "
 631                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 632                     i, n->mergeable_rx_bufs, offset, size,
 633                     guest_hdr_len, host_hdr_len, n->vdev.guest_features);
 634             exit(1);
 635         }
 636
 637         if (elem.in_num < 1) {
 638             error_report("virtio-net receive queue contains no in buffers");
 639             exit(1);
 640         }
 641
 642         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != guest_hdr_len) {
 643             error_report("virtio-net header not in first element");
 644             exit(1);
 645         }
 646
 647         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 648
 649         if (i == 0) {
 650             if (n->mergeable_rx_bufs)
 651                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 652
 653             offset += receive_header(n, sg, elem.in_num,
 654                                      buf + offset, size - offset, guest_hdr_len);
 655             total += guest_hdr_len;
 656         }
 657
 658         /* copy in packet.  ugh */
 659         len = iov_from_buf(sg, elem.in_num,
 660                            buf + offset, 0, size - offset);
 661         total += len;
 662         offset += len;
 663         /* If buffers can't be merged, at this point we
 664          * must have consumed the complete packet.
 665          * Otherwise, drop it. */
 666         if (!n->mergeable_rx_bufs && offset < size) {
 667 #if 0
 668             error_report("virtio-net truncated non-mergeable packet: "
 669                          "i %zd mergeable %d offset %zd, size %zd, "
 670                          "guest hdr len %zd, host hdr len %zd",
 671                          i, n->mergeable_rx_bufs,
 672                          offset, size, guest_hdr_len, host_hdr_len);
 673 #endif
 674             return size;
 675         }
 676
 677         /* signal other side */
 678         virtqueue_fill(n->rx_vq, &elem, total, i++);
 679     }
 680
 681     if (mhdr) {
 682         stw_p(&mhdr->num_buffers, i);
 683     }
 684
 685     virtqueue_flush(n->rx_vq, i);
 686     virtio_notify(&n->vdev, n->rx_vq);
 687
 688     return size;
 689 }
 690
 691 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 692
 693 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 694 {
 695     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 696
 697     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 698     virtio_notify(&n->vdev, n->tx_vq);
 699
 700     n->async_tx.elem.out_num = n->async_tx.len = 0;
 701
 702     virtio_queue_set_notification(n->tx_vq, 1);
 703     virtio_net_flush_tx(n, n->tx_vq);
 704 }
 705
 706 /* TX */
 707 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 708 {
 709     VirtQueueElement elem;
 710     int32_t num_packets = 0;
 711     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 712         return num_packets;
 713     }
 714
 715     assert(n->vdev.vm_running);
 716
 717     if (n->async_tx.elem.out_num) {
 718         virtio_queue_set_notification(n->tx_vq, 0);
 719         return num_packets;
 720     }
 721
 722     while (virtqueue_pop(vq, &elem)) {
 723         ssize_t ret, len = 0;
 724         unsigned int out_num = elem.out_num;
 725         struct iovec *out_sg = &elem.out_sg[0];
 726         unsigned hdr_len;
 727
 728         /* hdr_len refers to the header received from the guest */
 729         hdr_len = n->mergeable_rx_bufs ?
 730             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 731             sizeof(struct virtio_net_hdr);
 732
 733         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 734             error_report("virtio-net header not in first element");
 735             exit(1);
 736         }
 737
 738         /* ignore the header if GSO is not supported */
 739         if (!n->has_vnet_hdr) {
 740             out_num--;
 741             out_sg++;
 742             len += hdr_len;
 743         } else if (n->mergeable_rx_bufs) {
 744             /* tapfd expects a struct virtio_net_hdr */
 745             hdr_len -= sizeof(struct virtio_net_hdr);
 746             out_sg->iov_len -= hdr_len;
 747             len += hdr_len;
 748         }
 749
 750         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 751                                       virtio_net_tx_complete);
 752         if (ret == 0) {
 753             virtio_queue_set_notification(n->tx_vq, 0);
 754             n->async_tx.elem = elem;
 755             n->async_tx.len  = len;
 756             return -EBUSY;
 757         }
 758
 759         len += ret;
 760
 761         virtqueue_push(vq, &elem, len);
 762         virtio_notify(&n->vdev, vq);
 763
 764         if (++num_packets >= n->tx_burst) {
 765             break;
 766         }
 767     }
 768     return num_packets;
 769 }
 770
 771 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 772 {
 773     VirtIONet *n = to_virtio_net(vdev);
 774
 775     /* This happens when device was stopped but VCPU wasn't. */
 776     if (!n->vdev.vm_running) {
 777         n->tx_waiting = 1;
 778         return;
 779     }
 780
 781     if (n->tx_waiting) {
 782         virtio_queue_set_notification(vq, 1);
 783         qemu_del_timer(n->tx_timer);
 784         n->tx_waiting = 0;
 785         virtio_net_flush_tx(n, vq);
 786     } else {
 787         qemu_mod_timer(n->tx_timer,
 788                        qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 789         n->tx_waiting = 1;
 790         virtio_queue_set_notification(vq, 0);
 791     }
 792 }
 793
 794 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
 795 {
 796     VirtIONet *n = to_virtio_net(vdev);
 797
 798     if (unlikely(n->tx_waiting)) {
 799         return;
 800     }
 801     n->tx_waiting = 1;
 802     /* This happens when device was stopped but VCPU wasn't. */
 803     if (!n->vdev.vm_running) {
 804         return;
 805     }
 806     virtio_queue_set_notification(vq, 0);
 807     qemu_bh_schedule(n->tx_bh);
 808 }
 809
 810 static void virtio_net_tx_timer(void *opaque)
 811 {
 812     VirtIONet *n = opaque;
 813     assert(n->vdev.vm_running);
 814
 815     n->tx_waiting = 0;
 816
 817     /* Just in case the driver is not ready on more */
 818     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 819         return;
 820
 821     virtio_queue_set_notification(n->tx_vq, 1);
 822     virtio_net_flush_tx(n, n->tx_vq);
 823 }
 824
 825 static void virtio_net_tx_bh(void *opaque)
 826 {
 827     VirtIONet *n = opaque;
 828     int32_t ret;
 829
 830     assert(n->vdev.vm_running);
 831
 832     n->tx_waiting = 0;
 833
 834     /* Just in case the driver is not ready on more */
 835     if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
 836         return;
 837
 838     ret = virtio_net_flush_tx(n, n->tx_vq);
 839     if (ret == -EBUSY) {
 840         return; /* Notification re-enable handled by tx_complete */
 841     }
 842
 843     /* If we flush a full burst of packets, assume there are
 844      * more coming and immediately reschedule */
 845     if (ret >= n->tx_burst) {
 846         qemu_bh_schedule(n->tx_bh);
 847         n->tx_waiting = 1;
 848         return;
 849     }
 850
 851     /* If less than a full burst, re-enable notification and flush
 852      * anything that may have come in while we weren't looking.  If
 853      * we find something, assume the guest is still active and reschedule */
 854     virtio_queue_set_notification(n->tx_vq, 1);
 855     if (virtio_net_flush_tx(n, n->tx_vq) > 0) {
 856         virtio_queue_set_notification(n->tx_vq, 0);
 857         qemu_bh_schedule(n->tx_bh);
 858         n->tx_waiting = 1;
 859     }
 860 }
 861
 862 static void virtio_net_save(QEMUFile *f, void *opaque)
 863 {
 864     VirtIONet *n = opaque;
 865
 866     /* At this point, backend must be stopped, otherwise
 867      * it might keep writing to memory. */
 868     assert(!n->vhost_started);
 869     virtio_save(&n->vdev, f);
 870
 871     qemu_put_buffer(f, n->mac, ETH_ALEN);
 872     qemu_put_be32(f, n->tx_waiting);
 873     qemu_put_be32(f, n->mergeable_rx_bufs);
 874     qemu_put_be16(f, n->status);
 875     qemu_put_byte(f, n->promisc);
 876     qemu_put_byte(f, n->allmulti);
 877     qemu_put_be32(f, n->mac_table.in_use);
 878     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 879     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 880     qemu_put_be32(f, n->has_vnet_hdr);
 881     qemu_put_byte(f, n->mac_table.multi_overflow);
 882     qemu_put_byte(f, n->mac_table.uni_overflow);
 883     qemu_put_byte(f, n->alluni);
 884     qemu_put_byte(f, n->nomulti);
 885     qemu_put_byte(f, n->nouni);
 886     qemu_put_byte(f, n->nobcast);
 887     qemu_put_byte(f, n->has_ufo);
 888 }
 889
 890 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 891 {
 892     VirtIONet *n = opaque;
 893     int i;
 894
 895     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 896         return -EINVAL;
 897
 898     virtio_load(&n->vdev, f);
 899
 900     qemu_get_buffer(f, n->mac, ETH_ALEN);
 901     n->tx_waiting = qemu_get_be32(f);
 902     n->mergeable_rx_bufs = qemu_get_be32(f);
 903
 904     if (version_id >= 3)
 905         n->status = qemu_get_be16(f);
 906
 907     if (version_id >= 4) {
 908         if (version_id < 8) {
 909             n->promisc = qemu_get_be32(f);
 910             n->allmulti = qemu_get_be32(f);
 911         } else {
 912             n->promisc = qemu_get_byte(f);
 913             n->allmulti = qemu_get_byte(f);
 914         }
 915     }
 916
 917     if (version_id >= 5) {
 918         n->mac_table.in_use = qemu_get_be32(f);
 919         /* MAC_TABLE_ENTRIES may be different from the saved image */
 920         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 921             qemu_get_buffer(f, n->mac_table.macs,
 922                             n->mac_table.in_use * ETH_ALEN);
 923         } else if (n->mac_table.in_use) {
 924             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 925             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 926             n->mac_table.in_use = 0;
 927         }
 928     }
 929
 930     if (version_id >= 6)
 931         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 932
 933     if (version_id >= 7) {
 934         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 935             error_report("virtio-net: saved image requires vnet_hdr=on");
 936             return -1;
 937         }
 938
 939         if (n->has_vnet_hdr) {
 940             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 941             tap_set_offload(n->nic->nc.peer,
 942                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 943                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 944                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 945                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 946                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 947         }
 948     }
 949
 950     if (version_id >= 9) {
 951         n->mac_table.multi_overflow = qemu_get_byte(f);
 952         n->mac_table.uni_overflow = qemu_get_byte(f);
 953     }
 954
 955     if (version_id >= 10) {
 956         n->alluni = qemu_get_byte(f);
 957         n->nomulti = qemu_get_byte(f);
 958         n->nouni = qemu_get_byte(f);
 959         n->nobcast = qemu_get_byte(f);
 960     }
 961
 962     if (version_id >= 11) {
 963         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 964             error_report("virtio-net: saved image requires TUN_F_UFO support");
 965             return -1;
 966         }
 967     }
 968
 969     /* Find the first multicast entry in the saved MAC filter */
 970     for (i = 0; i < n->mac_table.in_use; i++) {
 971         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 972             break;
 973         }
 974     }
 975     n->mac_table.first_multi = i;
 976     return 0;
 977 }
 978
 979 static void virtio_net_cleanup(VLANClientState *nc)
 980 {
 981     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 982
 983     n->nic = NULL;
 984 }
 985
 986 static NetClientInfo net_virtio_info = {
 987     .type = NET_CLIENT_TYPE_NIC,
 988     .size = sizeof(NICState),
 989     .can_receive = virtio_net_can_receive,
 990     .receive = virtio_net_receive,
 991         .cleanup = virtio_net_cleanup,
 992     .link_status_changed = virtio_net_set_link_status,
 993 };
 994
 995 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
 996                               virtio_net_conf *net)
 997 {
 998     VirtIONet *n;
 999
1000     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
1001                                         sizeof(struct virtio_net_config),
1002                                         sizeof(VirtIONet));
1003
1004     n->vdev.get_config = virtio_net_get_config;
1005     n->vdev.set_config = virtio_net_set_config;
1006     n->vdev.get_features = virtio_net_get_features;
1007     n->vdev.set_features = virtio_net_set_features;
1008     n->vdev.bad_features = virtio_net_bad_features;
1009     n->vdev.reset = virtio_net_reset;
1010     n->vdev.set_status = virtio_net_set_status;
1011     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
1012
1013     if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
1014         error_report("virtio-net: "
1015                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1016                      net->tx);
1017         error_report("Defaulting to \"bh\"");
1018     }
1019
1020     if (net->tx && !strcmp(net->tx, "timer")) {
1021         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_timer);
1022         n->tx_timer = qemu_new_timer_ns(vm_clock, virtio_net_tx_timer, n);
1023         n->tx_timeout = net->txtimer;
1024     } else {
1025         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_bh);
1026         n->tx_bh = qemu_bh_new(virtio_net_tx_bh, n);
1027     }
1028     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
1029     qemu_macaddr_default_if_unset(&conf->macaddr);
1030     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
1031     n->status = VIRTIO_NET_S_LINK_UP;
1032
1033     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
1034
1035     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
1036
1037     n->tx_waiting = 0;
1038     n->tx_burst = net->txburst;
1039     n->mergeable_rx_bufs = 0;
1040     n->promisc = 1; /* for compatibility */
1041
1042     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
1043
1044     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
1045
1046     n->qdev = dev;
1047     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1048                     virtio_net_save, virtio_net_load, n);
1049
1050     add_boot_device_path(conf->bootindex, dev, "/ethernet-phy@0");
1051
1052     return &n->vdev;
1053 }
1054
1055 void virtio_net_exit(VirtIODevice *vdev)
1056 {
1057     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
1058
1059     /* This will stop vhost backend if appropriate. */
1060     virtio_net_set_status(vdev, 0);
1061
1062     qemu_purge_queued_packets(&n->nic->nc);
1063
1064     unregister_savevm(n->qdev, "virtio-net", n);
1065
1066     qemu_free(n->mac_table.macs);
1067     qemu_free(n->vlans);
1068
1069     if (n->tx_timer) {
1070         qemu_del_timer(n->tx_timer);
1071         qemu_free_timer(n->tx_timer);
1072     } else {
1073         qemu_bh_delete(n->tx_bh);
1074     }
1075
1076     qemu_del_vlan_client(&n->nic->nc);
1077     virtio_cleanup(&n->vdev);
1078 }