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