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 bool virtio_net_started(VirtIONet *n, uint8_t status)
 103 {
 104     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 105         (n->status & VIRTIO_NET_S_LINK_UP) && n->vm_running;
 106 }
 107
 108 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
 109 {
 110     if (!n->nic->nc.peer) {
 111         return;
 112     }
 113     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 114         return;
 115     }
 116
 117     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 118         return;
 119     }
 120     if (!!n->vhost_started == virtio_net_started(n, status)) {
 121         return;
 122     }
 123     if (!n->vhost_started) {
 124         int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 125         if (r < 0) {
 126             error_report("unable to start vhost net: %d: "
 127                          "falling back on userspace virtio", -r);
 128         } else {
 129             n->vhost_started = 1;
 130         }
 131     } else {
 132         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 133         n->vhost_started = 0;
 134     }
 135 }
 136
 137 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 138 {
 139     VirtIONet *n = to_virtio_net(vdev);
 140
 141     virtio_net_vhost_status(n, status);
 142
 143     if (!n->tx_waiting) {
 144         return;
 145     }
 146
 147     if (virtio_net_started(n, status) && !n->vhost_started) {
 148         if (n->tx_timer) {
 149             qemu_mod_timer(n->tx_timer,
 150                            qemu_get_clock(vm_clock) + n->tx_timeout);
 151         } else {
 152             qemu_bh_schedule(n->tx_bh);
 153         }
 154     } else {
 155         if (n->tx_timer) {
 156             qemu_del_timer(n->tx_timer);
 157         } else {
 158             qemu_bh_cancel(n->tx_bh);
 159         }
 160     }
 161 }
 162
 163 static void virtio_net_set_link_status(VLANClientState *nc)
 164 {
 165     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 166     uint16_t old_status = n->status;
 167
 168     if (nc->link_down)
 169         n->status &= ~VIRTIO_NET_S_LINK_UP;
 170     else
 171         n->status |= VIRTIO_NET_S_LINK_UP;
 172
 173     if (n->status != old_status)
 174         virtio_notify_config(&n->vdev);
 175
 176     virtio_net_set_status(&n->vdev, n->vdev.status);
 177 }
 178
 179 static void virtio_net_reset(VirtIODevice *vdev)
 180 {
 181     VirtIONet *n = to_virtio_net(vdev);
 182
 183     /* Reset back to compatibility mode */
 184     n->promisc = 1;
 185     n->allmulti = 0;
 186     n->alluni = 0;
 187     n->nomulti = 0;
 188     n->nouni = 0;
 189     n->nobcast = 0;
 190
 191     /* Flush any MAC and VLAN filter table state */
 192     n->mac_table.in_use = 0;
 193     n->mac_table.first_multi = 0;
 194     n->mac_table.multi_overflow = 0;
 195     n->mac_table.uni_overflow = 0;
 196     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 197     memset(n->vlans, 0, MAX_VLAN >> 3);
 198 }
 199
 200 static int peer_has_vnet_hdr(VirtIONet *n)
 201 {
 202     if (!n->nic->nc.peer)
 203         return 0;
 204
 205     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 206         return 0;
 207
 208     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 209
 210     return n->has_vnet_hdr;
 211 }
 212
 213 static int peer_has_ufo(VirtIONet *n)
 214 {
 215     if (!peer_has_vnet_hdr(n))
 216         return 0;
 217
 218     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 219
 220     return n->has_ufo;
 221 }
 222
 223 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 224 {
 225     VirtIONet *n = to_virtio_net(vdev);
 226
 227     features |= (1 << VIRTIO_NET_F_MAC);
 228
 229     if (peer_has_vnet_hdr(n)) {
 230         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 231     } else {
 232         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 233         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 234         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 235         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 236
 237         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 238         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 239         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 240         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 241     }
 242
 243     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 244         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 245         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 246     }
 247
 248     if (!n->nic->nc.peer ||
 249         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 250         return features;
 251     }
 252     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 253         return features;
 254     }
 255     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 256 }
 257
 258 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 259 {
 260     uint32_t features = 0;
 261
 262     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 263      * but also these: */
 264     features |= (1 << VIRTIO_NET_F_MAC);
 265     features |= (1 << VIRTIO_NET_F_CSUM);
 266     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 267     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 268     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 269
 270     return features;
 271 }
 272
 273 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 274 {
 275     VirtIONet *n = to_virtio_net(vdev);
 276
 277     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 278
 279     if (n->has_vnet_hdr) {
 280         tap_set_offload(n->nic->nc.peer,
 281                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 282                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 283                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 284                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 285                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 286     }
 287     if (!n->nic->nc.peer ||
 288         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 289         return;
 290     }
 291     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 292         return;
 293     }
 294     vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 295 }
 296
 297 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 298                                      VirtQueueElement *elem)
 299 {
 300     uint8_t on;
 301
 302     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 303         error_report("virtio-net ctrl invalid rx mode command");
 304         exit(1);
 305     }
 306
 307     on = ldub_p(elem->out_sg[1].iov_base);
 308
 309     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 310         n->promisc = on;
 311     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 312         n->allmulti = on;
 313     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 314         n->alluni = on;
 315     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 316         n->nomulti = on;
 317     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 318         n->nouni = on;
 319     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 320         n->nobcast = on;
 321     else
 322         return VIRTIO_NET_ERR;
 323
 324     return VIRTIO_NET_OK;
 325 }
 326
 327 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 328                                  VirtQueueElement *elem)
 329 {
 330     struct virtio_net_ctrl_mac mac_data;
 331
 332     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 333         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 334         elem->out_sg[2].iov_len < sizeof(mac_data))
 335         return VIRTIO_NET_ERR;
 336
 337     n->mac_table.in_use = 0;
 338     n->mac_table.first_multi = 0;
 339     n->mac_table.uni_overflow = 0;
 340     n->mac_table.multi_overflow = 0;
 341     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 342
 343     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 344
 345     if (sizeof(mac_data.entries) +
 346         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 347         return VIRTIO_NET_ERR;
 348
 349     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 350         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 351                mac_data.entries * ETH_ALEN);
 352         n->mac_table.in_use += mac_data.entries;
 353     } else {
 354         n->mac_table.uni_overflow = 1;
 355     }
 356
 357     n->mac_table.first_multi = n->mac_table.in_use;
 358
 359     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 360
 361     if (sizeof(mac_data.entries) +
 362         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 363         return VIRTIO_NET_ERR;
 364
 365     if (mac_data.entries) {
 366         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 367             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 368                    elem->out_sg[2].iov_base + sizeof(mac_data),
 369                    mac_data.entries * ETH_ALEN);
 370             n->mac_table.in_use += mac_data.entries;
 371         } else {
 372             n->mac_table.multi_overflow = 1;
 373         }
 374     }
 375
 376     return VIRTIO_NET_OK;
 377 }
 378
 379 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 380                                         VirtQueueElement *elem)
 381 {
 382     uint16_t vid;
 383
 384     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 385         error_report("virtio-net ctrl invalid vlan command");
 386         return VIRTIO_NET_ERR;
 387     }
 388
 389     vid = lduw_le_p(elem->out_sg[1].iov_base);
 390
 391     if (vid >= MAX_VLAN)
 392         return VIRTIO_NET_ERR;
 393
 394     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 395         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 396     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 397         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 398     else
 399         return VIRTIO_NET_ERR;
 400
 401     return VIRTIO_NET_OK;
 402 }
 403
 404 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 405 {
 406     VirtIONet *n = to_virtio_net(vdev);
 407     struct virtio_net_ctrl_hdr ctrl;
 408     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 409     VirtQueueElement elem;
 410
 411     while (virtqueue_pop(vq, &elem)) {
 412         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 413             error_report("virtio-net ctrl missing headers");
 414             exit(1);
 415         }
 416
 417         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 418             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 419             error_report("virtio-net ctrl header not in correct element");
 420             exit(1);
 421         }
 422
 423         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 424         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 425
 426         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 427             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 428         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 429             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 430         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 431             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 432
 433         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 434
 435         virtqueue_push(vq, &elem, sizeof(status));
 436         virtio_notify(vdev, vq);
 437     }
 438 }
 439
 440 /* RX */
 441
 442 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 443 {
 444     VirtIONet *n = to_virtio_net(vdev);
 445
 446     qemu_flush_queued_packets(&n->nic->nc);
 447
 448     /* We now have RX buffers, signal to the IO thread to break out of the
 449      * select to re-poll the tap file descriptor */
 450     qemu_notify_event();
 451 }
 452
 453 static int virtio_net_can_receive(VLANClientState *nc)
 454 {
 455     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 456     if (!n->vm_running) {
 457         return 0;
 458     }
 459
 460     if (!virtio_queue_ready(n->rx_vq) ||
 461         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 462         return 0;
 463
 464     return 1;
 465 }
 466
 467 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 468 {
 469     if (virtio_queue_empty(n->rx_vq) ||
 470         (n->mergeable_rx_bufs &&
 471          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 472         virtio_queue_set_notification(n->rx_vq, 1);
 473
 474         /* To avoid a race condition where the guest has made some buffers
 475          * available after the above check but before notification was
 476          * enabled, check for available buffers again.
 477          */
 478         if (virtio_queue_empty(n->rx_vq) ||
 479             (n->mergeable_rx_bufs &&
 480              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 481             return 0;
 482     }
 483
 484     virtio_queue_set_notification(n->rx_vq, 0);
 485     return 1;
 486 }
 487
 488 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 489  * it never finds out that the packets don't have valid checksums.  This
 490  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 491  * fix this with Xen but it hasn't appeared in an upstream release of
 492  * dhclient yet.
 493  *
 494  * To avoid breaking existing guests, we catch udp packets and add
 495  * checksums.  This is terrible but it's better than hacking the guest
 496  * kernels.
 497  *
 498  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 499  * we should provide a mechanism to disable it to avoid polluting the host
 500  * cache.
 501  */
 502 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 503                                         const uint8_t *buf, size_t size)
 504 {
 505     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 506         (size > 27 && size < 1500) && /* normal sized MTU */
 507         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 508         (buf[23] == 17) && /* ip.protocol == UDP */
 509         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 510         /* FIXME this cast is evil */
 511         net_checksum_calculate((uint8_t *)buf, size);
 512         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 513     }
 514 }
 515
 516 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 517                           const void *buf, size_t size, size_t hdr_len)
 518 {
 519     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 520     int offset = 0;
 521
 522     hdr->flags = 0;
 523     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 524
 525     if (n->has_vnet_hdr) {
 526         memcpy(hdr, buf, sizeof(*hdr));
 527         offset = sizeof(*hdr);
 528         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 529     }
 530
 531     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 532      * but we may be passing along a larger header to the guest.
 533      */
 534     iov[0].iov_base += hdr_len;
 535     iov[0].iov_len  -= hdr_len;
 536
 537     return offset;
 538 }
 539
 540 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 541 {
 542     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 543     static const uint8_t vlan[] = {0x81, 0x00};
 544     uint8_t *ptr = (uint8_t *)buf;
 545     int i;
 546
 547     if (n->promisc)
 548         return 1;
 549
 550     if (n->has_vnet_hdr) {
 551         ptr += sizeof(struct virtio_net_hdr);
 552     }
 553
 554     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 555         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 556         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 557             return 0;
 558     }
 559
 560     if (ptr[0] & 1) { // multicast
 561         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 562             return !n->nobcast;
 563         } else if (n->nomulti) {
 564             return 0;
 565         } else if (n->allmulti || n->mac_table.multi_overflow) {
 566             return 1;
 567         }
 568
 569         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 570             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 571                 return 1;
 572             }
 573         }
 574     } else { // unicast
 575         if (n->nouni) {
 576             return 0;
 577         } else if (n->alluni || n->mac_table.uni_overflow) {
 578             return 1;
 579         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 580             return 1;
 581         }
 582
 583         for (i = 0; i < n->mac_table.first_multi; i++) {
 584             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 585                 return 1;
 586             }
 587         }
 588     }
 589
 590     return 0;
 591 }
 592
 593 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 594 {
 595     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 596     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 597     size_t guest_hdr_len, offset, i, host_hdr_len;
 598
 599     if (!virtio_net_can_receive(&n->nic->nc))
 600         return -1;
 601
 602     /* hdr_len refers to the header we supply to the guest */
 603     guest_hdr_len = n->mergeable_rx_bufs ?
 604         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 605
 606
 607     host_hdr_len = n->has_vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 608     if (!virtio_net_has_buffers(n, size + guest_hdr_len - host_hdr_len))
 609         return 0;
 610
 611     if (!receive_filter(n, buf, size))
 612         return size;
 613
 614     offset = i = 0;
 615
 616     while (offset < size) {
 617         VirtQueueElement elem;
 618         int len, total;
 619         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 620
 621         total = 0;
 622
 623         if (virtqueue_pop(n->rx_vq, &elem) == 0) {
 624             if (i == 0)
 625                 return -1;
 626             error_report("virtio-net unexpected empty queue: "
 627                     "i %zd mergeable %d offset %zd, size %zd, "
 628                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 629                     i, n->mergeable_rx_bufs, offset, size,
 630                     guest_hdr_len, host_hdr_len, n->vdev.guest_features);
 631             exit(1);
 632         }
 633
 634         if (elem.in_num < 1) {
 635             error_report("virtio-net receive queue contains no in buffers");
 636             exit(1);
 637         }
 638
 639         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != guest_hdr_len) {
 640             error_report("virtio-net header not in first element");
 641             exit(1);
 642         }
 643
 644         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 645
 646         if (i == 0) {
 647             if (n->mergeable_rx_bufs)
 648                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 649
 650             offset += receive_header(n, sg, elem.in_num,
 651                                      buf + offset, size - offset, guest_hdr_len);
 652             total += guest_hdr_len;
 653         }
 654
 655         /* copy in packet.  ugh */
 656         len = iov_from_buf(sg, elem.in_num,
 657                            buf + offset, size - offset);
 658         total += len;
 659         offset += len;
 660         /* If buffers can't be merged, at this point we
 661          * must have consumed the complete packet.
 662          * Otherwise, drop it. */
 663         if (!n->mergeable_rx_bufs && offset < size) {
 664 #if 0
 665             error_report("virtio-net truncated non-mergeable packet: "
 666                          "i %zd mergeable %d offset %zd, size %zd, "
 667                          "guest hdr len %zd, host hdr len %zd",
 668                          i, n->mergeable_rx_bufs,
 669                          offset, size, guest_hdr_len, host_hdr_len);
 670 #endif
 671             return size;
 672         }
 673
 674         /* signal other side */
 675         virtqueue_fill(n->rx_vq, &elem, total, i++);
 676     }
 677
 678     if (mhdr)
 679         mhdr->num_buffers = i;
 680
 681     virtqueue_flush(n->rx_vq, i);
 682     virtio_notify(&n->vdev, n->rx_vq);
 683
 684     return size;
 685 }
 686
 687 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 688
 689 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 690 {
 691     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 692
 693     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 694     virtio_notify(&n->vdev, n->tx_vq);
 695
 696     n->async_tx.elem.out_num = n->async_tx.len = 0;
 697
 698     virtio_queue_set_notification(n->tx_vq, 1);
 699     virtio_net_flush_tx(n, n->tx_vq);
 700 }
 701
 702 /* TX */
 703 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 704 {
 705     VirtQueueElement elem;
 706     int32_t num_packets = 0;
 707     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 708         return num_packets;
 709     }
 710
 711     assert(n->vm_running);
 712
 713     if (n->async_tx.elem.out_num) {
 714         virtio_queue_set_notification(n->tx_vq, 0);
 715         return num_packets;
 716     }
 717
 718     while (virtqueue_pop(vq, &elem)) {
 719         ssize_t ret, len = 0;
 720         unsigned int out_num = elem.out_num;
 721         struct iovec *out_sg = &elem.out_sg[0];
 722         unsigned hdr_len;
 723
 724         /* hdr_len refers to the header received from the guest */
 725         hdr_len = n->mergeable_rx_bufs ?
 726             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 727             sizeof(struct virtio_net_hdr);
 728
 729         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 730             error_report("virtio-net header not in first element");
 731             exit(1);
 732         }
 733
 734         /* ignore the header if GSO is not supported */
 735         if (!n->has_vnet_hdr) {
 736             out_num--;
 737             out_sg++;
 738             len += hdr_len;
 739         } else if (n->mergeable_rx_bufs) {
 740             /* tapfd expects a struct virtio_net_hdr */
 741             hdr_len -= sizeof(struct virtio_net_hdr);
 742             out_sg->iov_len -= hdr_len;
 743             len += hdr_len;
 744         }
 745
 746         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 747                                       virtio_net_tx_complete);
 748         if (ret == 0) {
 749             virtio_queue_set_notification(n->tx_vq, 0);
 750             n->async_tx.elem = elem;
 751             n->async_tx.len  = len;
 752             return -EBUSY;
 753         }
 754
 755         len += ret;
 756
 757         virtqueue_push(vq, &elem, len);
 758         virtio_notify(&n->vdev, vq);
 759
 760         if (++num_packets >= n->tx_burst) {
 761             break;
 762         }
 763     }
 764     return num_packets;
 765 }
 766
 767 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 768 {
 769     VirtIONet *n = to_virtio_net(vdev);
 770
 771     /* This happens when device was stopped but VCPU wasn't. */
 772     if (!n->vm_running) {
 773         n->tx_waiting = 1;
 774         return;
 775     }
 776
 777     if (n->tx_waiting) {
 778         virtio_queue_set_notification(vq, 1);
 779         qemu_del_timer(n->tx_timer);
 780         n->tx_waiting = 0;
 781         virtio_net_flush_tx(n, vq);
 782     } else {
 783         qemu_mod_timer(n->tx_timer,
 784                        qemu_get_clock(vm_clock) + n->tx_timeout);
 785         n->tx_waiting = 1;
 786         virtio_queue_set_notification(vq, 0);
 787     }
 788 }
 789
 790 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
 791 {
 792     VirtIONet *n = to_virtio_net(vdev);
 793
 794     if (unlikely(n->tx_waiting)) {
 795         return;
 796     }
 797     n->tx_waiting = 1;
 798     /* This happens when device was stopped but VCPU wasn't. */
 799     if (!n->vm_running) {
 800         return;
 801     }
 802     virtio_queue_set_notification(vq, 0);
 803     qemu_bh_schedule(n->tx_bh);
 804 }
 805
 806 static void virtio_net_tx_timer(void *opaque)
 807 {
 808     VirtIONet *n = opaque;
 809     assert(n->vm_running);
 810
 811     n->tx_waiting = 0;
 812
 813     /* Just in case the driver is not ready on more */
 814     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 815         return;
 816
 817     virtio_queue_set_notification(n->tx_vq, 1);
 818     virtio_net_flush_tx(n, n->tx_vq);
 819 }
 820
 821 static void virtio_net_tx_bh(void *opaque)
 822 {
 823     VirtIONet *n = opaque;
 824     int32_t ret;
 825
 826     assert(n->vm_running);
 827
 828     n->tx_waiting = 0;
 829
 830     /* Just in case the driver is not ready on more */
 831     if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
 832         return;
 833
 834     ret = virtio_net_flush_tx(n, n->tx_vq);
 835     if (ret == -EBUSY) {
 836         return; /* Notification re-enable handled by tx_complete */
 837     }
 838
 839     /* If we flush a full burst of packets, assume there are
 840      * more coming and immediately reschedule */
 841     if (ret >= n->tx_burst) {
 842         qemu_bh_schedule(n->tx_bh);
 843         n->tx_waiting = 1;
 844         return;
 845     }
 846
 847     /* If less than a full burst, re-enable notification and flush
 848      * anything that may have come in while we weren't looking.  If
 849      * we find something, assume the guest is still active and reschedule */
 850     virtio_queue_set_notification(n->tx_vq, 1);
 851     if (virtio_net_flush_tx(n, n->tx_vq) > 0) {
 852         virtio_queue_set_notification(n->tx_vq, 0);
 853         qemu_bh_schedule(n->tx_bh);
 854         n->tx_waiting = 1;
 855     }
 856 }
 857
 858 static void virtio_net_save(QEMUFile *f, void *opaque)
 859 {
 860     VirtIONet *n = opaque;
 861
 862     /* At this point, backend must be stopped, otherwise
 863      * it might keep writing to memory. */
 864     assert(!n->vhost_started);
 865     virtio_save(&n->vdev, f);
 866
 867     qemu_put_buffer(f, n->mac, ETH_ALEN);
 868     qemu_put_be32(f, n->tx_waiting);
 869     qemu_put_be32(f, n->mergeable_rx_bufs);
 870     qemu_put_be16(f, n->status);
 871     qemu_put_byte(f, n->promisc);
 872     qemu_put_byte(f, n->allmulti);
 873     qemu_put_be32(f, n->mac_table.in_use);
 874     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 875     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 876     qemu_put_be32(f, n->has_vnet_hdr);
 877     qemu_put_byte(f, n->mac_table.multi_overflow);
 878     qemu_put_byte(f, n->mac_table.uni_overflow);
 879     qemu_put_byte(f, n->alluni);
 880     qemu_put_byte(f, n->nomulti);
 881     qemu_put_byte(f, n->nouni);
 882     qemu_put_byte(f, n->nobcast);
 883     qemu_put_byte(f, n->has_ufo);
 884 }
 885
 886 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 887 {
 888     VirtIONet *n = opaque;
 889     int i;
 890
 891     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 892         return -EINVAL;
 893
 894     virtio_load(&n->vdev, f);
 895
 896     qemu_get_buffer(f, n->mac, ETH_ALEN);
 897     n->tx_waiting = qemu_get_be32(f);
 898     n->mergeable_rx_bufs = qemu_get_be32(f);
 899
 900     if (version_id >= 3)
 901         n->status = qemu_get_be16(f);
 902
 903     if (version_id >= 4) {
 904         if (version_id < 8) {
 905             n->promisc = qemu_get_be32(f);
 906             n->allmulti = qemu_get_be32(f);
 907         } else {
 908             n->promisc = qemu_get_byte(f);
 909             n->allmulti = qemu_get_byte(f);
 910         }
 911     }
 912
 913     if (version_id >= 5) {
 914         n->mac_table.in_use = qemu_get_be32(f);
 915         /* MAC_TABLE_ENTRIES may be different from the saved image */
 916         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 917             qemu_get_buffer(f, n->mac_table.macs,
 918                             n->mac_table.in_use * ETH_ALEN);
 919         } else if (n->mac_table.in_use) {
 920             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 921             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 922             n->mac_table.in_use = 0;
 923         }
 924     }
 925
 926     if (version_id >= 6)
 927         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 928
 929     if (version_id >= 7) {
 930         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 931             error_report("virtio-net: saved image requires vnet_hdr=on");
 932             return -1;
 933         }
 934
 935         if (n->has_vnet_hdr) {
 936             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 937             tap_set_offload(n->nic->nc.peer,
 938                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 939                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 940                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 941                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 942                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 943         }
 944     }
 945
 946     if (version_id >= 9) {
 947         n->mac_table.multi_overflow = qemu_get_byte(f);
 948         n->mac_table.uni_overflow = qemu_get_byte(f);
 949     }
 950
 951     if (version_id >= 10) {
 952         n->alluni = qemu_get_byte(f);
 953         n->nomulti = qemu_get_byte(f);
 954         n->nouni = qemu_get_byte(f);
 955         n->nobcast = qemu_get_byte(f);
 956     }
 957
 958     if (version_id >= 11) {
 959         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 960             error_report("virtio-net: saved image requires TUN_F_UFO support");
 961             return -1;
 962         }
 963     }
 964
 965     /* Find the first multicast entry in the saved MAC filter */
 966     for (i = 0; i < n->mac_table.in_use; i++) {
 967         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 968             break;
 969         }
 970     }
 971     n->mac_table.first_multi = i;
 972     return 0;
 973 }
 974
 975 static void virtio_net_cleanup(VLANClientState *nc)
 976 {
 977     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 978
 979     n->nic = NULL;
 980 }
 981
 982 static NetClientInfo net_virtio_info = {
 983     .type = NET_CLIENT_TYPE_NIC,
 984     .size = sizeof(NICState),
 985     .can_receive = virtio_net_can_receive,
 986     .receive = virtio_net_receive,
 987         .cleanup = virtio_net_cleanup,
 988     .link_status_changed = virtio_net_set_link_status,
 989 };
 990
 991 static void virtio_net_vmstate_change(void *opaque, int running, int reason)
 992 {
 993     VirtIONet *n = opaque;
 994     n->vm_running = running;
 995     /* This is called when vm is started/stopped,
 996      * it will start/stop vhost backend if appropriate
 997      * e.g. after migration. */
 998     virtio_net_set_status(&n->vdev, n->vdev.status);
 999 }
1000
1001 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
1002                               virtio_net_conf *net)
1003 {
1004     VirtIONet *n;
1005
1006     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
1007                                         sizeof(struct virtio_net_config),
1008                                         sizeof(VirtIONet));
1009
1010     n->vdev.get_config = virtio_net_get_config;
1011     n->vdev.set_config = virtio_net_set_config;
1012     n->vdev.get_features = virtio_net_get_features;
1013     n->vdev.set_features = virtio_net_set_features;
1014     n->vdev.bad_features = virtio_net_bad_features;
1015     n->vdev.reset = virtio_net_reset;
1016     n->vdev.set_status = virtio_net_set_status;
1017     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
1018
1019     if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
1020         error_report("virtio-net: "
1021                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1022                      net->tx);
1023         error_report("Defaulting to \"bh\"");
1024     }
1025
1026     if (net->tx && !strcmp(net->tx, "timer")) {
1027         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_timer);
1028         n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
1029         n->tx_timeout = net->txtimer;
1030     } else {
1031         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_bh);
1032         n->tx_bh = qemu_bh_new(virtio_net_tx_bh, n);
1033     }
1034     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
1035     qemu_macaddr_default_if_unset(&conf->macaddr);
1036     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
1037     n->status = VIRTIO_NET_S_LINK_UP;
1038
1039     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
1040
1041     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
1042
1043     n->tx_waiting = 0;
1044     n->tx_burst = net->txburst;
1045     n->mergeable_rx_bufs = 0;
1046     n->promisc = 1; /* for compatibility */
1047
1048     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
1049
1050     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
1051
1052     n->qdev = dev;
1053     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1054                     virtio_net_save, virtio_net_load, n);
1055     n->vmstate = qemu_add_vm_change_state_handler(virtio_net_vmstate_change, n);
1056
1057     add_boot_device_path(conf->bootindex, dev, "/ethernet-phy@0");
1058
1059     return &n->vdev;
1060 }
1061
1062 void virtio_net_exit(VirtIODevice *vdev)
1063 {
1064     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
1065     qemu_del_vm_change_state_handler(n->vmstate);
1066
1067     /* This will stop vhost backend if appropriate. */
1068     virtio_net_set_status(vdev, 0);
1069
1070     qemu_purge_queued_packets(&n->nic->nc);
1071
1072     unregister_savevm(n->qdev, "virtio-net", n);
1073
1074     qemu_free(n->mac_table.macs);
1075     qemu_free(n->vlans);
1076
1077     if (n->tx_timer) {
1078         qemu_del_timer(n->tx_timer);
1079         qemu_free_timer(n->tx_timer);
1080     } else {
1081         qemu_bh_delete(n->tx_bh);
1082     }
1083
1084     virtio_cleanup(&n->vdev);
1085     qemu_del_vlan_client(&n->nic->nc);
1086 }