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 "virtio.h"
  15 #include "net.h"
  16 #include "net/checksum.h"
  17 #include "net/tap.h"
  18 #include "qemu-timer.h"
  19 #include "virtio-net.h"
  20
  21 #define VIRTIO_NET_VM_VERSION    11
  22
  23 #define MAC_TABLE_ENTRIES    64
  24 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  25
  26 typedef struct VirtIONet
  27 {
  28     VirtIODevice vdev;
  29     uint8_t mac[ETH_ALEN];
  30     uint16_t status;
  31     VirtQueue *rx_vq;
  32     VirtQueue *tx_vq;
  33     VirtQueue *ctrl_vq;
  34     NICState *nic;
  35     QEMUTimer *tx_timer;
  36     int tx_timer_active;
  37     uint32_t has_vnet_hdr;
  38     uint8_t has_ufo;
  39     struct {
  40         VirtQueueElement elem;
  41         ssize_t len;
  42     } async_tx;
  43     int mergeable_rx_bufs;
  44     uint8_t promisc;
  45     uint8_t allmulti;
  46     uint8_t alluni;
  47     uint8_t nomulti;
  48     uint8_t nouni;
  49     uint8_t nobcast;
  50     struct {
  51         int in_use;
  52         int first_multi;
  53         uint8_t multi_overflow;
  54         uint8_t uni_overflow;
  55         uint8_t *macs;
  56     } mac_table;
  57     uint32_t *vlans;
  58 } VirtIONet;
  59
  60 /* TODO
  61  * - we could suppress RX interrupt if we were so inclined.
  62  */
  63
  64 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  65 {
  66     return (VirtIONet *)vdev;
  67 }
  68
  69 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  70 {
  71     VirtIONet *n = to_virtio_net(vdev);
  72     struct virtio_net_config netcfg;
  73
  74     netcfg.status = n->status;
  75     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  76     memcpy(config, &netcfg, sizeof(netcfg));
  77 }
  78
  79 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  80 {
  81     VirtIONet *n = to_virtio_net(vdev);
  82     struct virtio_net_config netcfg;
  83
  84     memcpy(&netcfg, config, sizeof(netcfg));
  85
  86     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  87         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  88         qemu_format_nic_info_str(&n->nic->nc, n->mac);
  89     }
  90 }
  91
  92 static void virtio_net_set_link_status(VLANClientState *nc)
  93 {
  94     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
  95     uint16_t old_status = n->status;
  96
  97     if (nc->link_down)
  98         n->status &= ~VIRTIO_NET_S_LINK_UP;
  99     else
 100         n->status |= VIRTIO_NET_S_LINK_UP;
 101
 102     if (n->status != old_status)
 103         virtio_notify_config(&n->vdev);
 104 }
 105
 106 static void virtio_net_reset(VirtIODevice *vdev)
 107 {
 108     VirtIONet *n = to_virtio_net(vdev);
 109
 110     /* Reset back to compatibility mode */
 111     n->promisc = 1;
 112     n->allmulti = 0;
 113     n->alluni = 0;
 114     n->nomulti = 0;
 115     n->nouni = 0;
 116     n->nobcast = 0;
 117
 118     /* Flush any MAC and VLAN filter table state */
 119     n->mac_table.in_use = 0;
 120     n->mac_table.first_multi = 0;
 121     n->mac_table.multi_overflow = 0;
 122     n->mac_table.uni_overflow = 0;
 123     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 124     memset(n->vlans, 0, MAX_VLAN >> 3);
 125 }
 126
 127 static int peer_has_vnet_hdr(VirtIONet *n)
 128 {
 129     if (!n->nic->nc.peer)
 130         return 0;
 131
 132     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 133         return 0;
 134
 135     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 136
 137     return n->has_vnet_hdr;
 138 }
 139
 140 static int peer_has_ufo(VirtIONet *n)
 141 {
 142     if (!peer_has_vnet_hdr(n))
 143         return 0;
 144
 145     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 146
 147     return n->has_ufo;
 148 }
 149
 150 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 151 {
 152     VirtIONet *n = to_virtio_net(vdev);
 153
 154     features |= (1 << VIRTIO_NET_F_MAC);
 155
 156     if (peer_has_vnet_hdr(n)) {
 157         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 158     } else {
 159         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 160         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 161         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 162         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 163
 164         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 165         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 166         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 167         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 168     }
 169
 170     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 171         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 172         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 173     }
 174
 175     return features;
 176 }
 177
 178 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 179 {
 180     uint32_t features = 0;
 181
 182     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 183      * but also these: */
 184     features |= (1 << VIRTIO_NET_F_MAC);
 185     features |= (1 << VIRTIO_NET_F_CSUM);
 186     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 187     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 188     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 189
 190     return features;
 191 }
 192
 193 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 194 {
 195     VirtIONet *n = to_virtio_net(vdev);
 196
 197     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 198
 199     if (n->has_vnet_hdr) {
 200         tap_set_offload(n->nic->nc.peer,
 201                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 202                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 203                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 204                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 205                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 206     }
 207 }
 208
 209 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 210                                      VirtQueueElement *elem)
 211 {
 212     uint8_t on;
 213
 214     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 215         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 216         exit(1);
 217     }
 218
 219     on = ldub_p(elem->out_sg[1].iov_base);
 220
 221     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 222         n->promisc = on;
 223     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 224         n->allmulti = on;
 225     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 226         n->alluni = on;
 227     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 228         n->nomulti = on;
 229     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 230         n->nouni = on;
 231     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 232         n->nobcast = on;
 233     else
 234         return VIRTIO_NET_ERR;
 235
 236     return VIRTIO_NET_OK;
 237 }
 238
 239 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 240                                  VirtQueueElement *elem)
 241 {
 242     struct virtio_net_ctrl_mac mac_data;
 243
 244     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 245         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 246         elem->out_sg[2].iov_len < sizeof(mac_data))
 247         return VIRTIO_NET_ERR;
 248
 249     n->mac_table.in_use = 0;
 250     n->mac_table.first_multi = 0;
 251     n->mac_table.uni_overflow = 0;
 252     n->mac_table.multi_overflow = 0;
 253     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 254
 255     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 256
 257     if (sizeof(mac_data.entries) +
 258         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 259         return VIRTIO_NET_ERR;
 260
 261     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 262         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 263                mac_data.entries * ETH_ALEN);
 264         n->mac_table.in_use += mac_data.entries;
 265     } else {
 266         n->mac_table.uni_overflow = 1;
 267     }
 268
 269     n->mac_table.first_multi = n->mac_table.in_use;
 270
 271     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 272
 273     if (sizeof(mac_data.entries) +
 274         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 275         return VIRTIO_NET_ERR;
 276
 277     if (mac_data.entries) {
 278         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 279             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 280                    elem->out_sg[2].iov_base + sizeof(mac_data),
 281                    mac_data.entries * ETH_ALEN);
 282             n->mac_table.in_use += mac_data.entries;
 283         } else {
 284             n->mac_table.multi_overflow = 1;
 285         }
 286     }
 287
 288     return VIRTIO_NET_OK;
 289 }
 290
 291 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 292                                         VirtQueueElement *elem)
 293 {
 294     uint16_t vid;
 295
 296     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 297         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 298         return VIRTIO_NET_ERR;
 299     }
 300
 301     vid = lduw_le_p(elem->out_sg[1].iov_base);
 302
 303     if (vid >= MAX_VLAN)
 304         return VIRTIO_NET_ERR;
 305
 306     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 307         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 308     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 309         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 310     else
 311         return VIRTIO_NET_ERR;
 312
 313     return VIRTIO_NET_OK;
 314 }
 315
 316 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 317 {
 318     VirtIONet *n = to_virtio_net(vdev);
 319     struct virtio_net_ctrl_hdr ctrl;
 320     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 321     VirtQueueElement elem;
 322
 323     while (virtqueue_pop(vq, &elem)) {
 324         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 325             fprintf(stderr, "virtio-net ctrl missing headers\n");
 326             exit(1);
 327         }
 328
 329         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 330             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 331             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 332             exit(1);
 333         }
 334
 335         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 336         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 337
 338         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 339             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 340         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 341             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 342         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 343             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 344
 345         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 346
 347         virtqueue_push(vq, &elem, sizeof(status));
 348         virtio_notify(vdev, vq);
 349     }
 350 }
 351
 352 /* RX */
 353
 354 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 355 {
 356     VirtIONet *n = to_virtio_net(vdev);
 357
 358     qemu_flush_queued_packets(&n->nic->nc);
 359
 360     /* We now have RX buffers, signal to the IO thread to break out of the
 361      * select to re-poll the tap file descriptor */
 362     qemu_notify_event();
 363 }
 364
 365 static int virtio_net_can_receive(VLANClientState *nc)
 366 {
 367     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 368
 369     if (!virtio_queue_ready(n->rx_vq) ||
 370         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 371         return 0;
 372
 373     return 1;
 374 }
 375
 376 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 377 {
 378     if (virtio_queue_empty(n->rx_vq) ||
 379         (n->mergeable_rx_bufs &&
 380          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 381         virtio_queue_set_notification(n->rx_vq, 1);
 382
 383         /* To avoid a race condition where the guest has made some buffers
 384          * available after the above check but before notification was
 385          * enabled, check for available buffers again.
 386          */
 387         if (virtio_queue_empty(n->rx_vq) ||
 388             (n->mergeable_rx_bufs &&
 389              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 390             return 0;
 391     }
 392
 393     virtio_queue_set_notification(n->rx_vq, 0);
 394     return 1;
 395 }
 396
 397 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 398  * it never finds out that the packets don't have valid checksums.  This
 399  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 400  * fix this with Xen but it hasn't appeared in an upstream release of
 401  * dhclient yet.
 402  *
 403  * To avoid breaking existing guests, we catch udp packets and add
 404  * checksums.  This is terrible but it's better than hacking the guest
 405  * kernels.
 406  *
 407  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 408  * we should provide a mechanism to disable it to avoid polluting the host
 409  * cache.
 410  */
 411 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 412                                         const uint8_t *buf, size_t size)
 413 {
 414     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 415         (size > 27 && size < 1500) && /* normal sized MTU */
 416         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 417         (buf[23] == 17) && /* ip.protocol == UDP */
 418         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 419         /* FIXME this cast is evil */
 420         net_checksum_calculate((uint8_t *)buf, size);
 421         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 422     }
 423 }
 424
 425 static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count)
 426 {
 427     int offset, i;
 428
 429     offset = i = 0;
 430     while (offset < count && i < iovcnt) {
 431         int len = MIN(iov[i].iov_len, count - offset);
 432         memcpy(iov[i].iov_base, buf + offset, len);
 433         offset += len;
 434         i++;
 435     }
 436
 437     return offset;
 438 }
 439
 440 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 441                           const void *buf, size_t size, size_t hdr_len)
 442 {
 443     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 444     int offset = 0;
 445
 446     hdr->flags = 0;
 447     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 448
 449     if (n->has_vnet_hdr) {
 450         memcpy(hdr, buf, sizeof(*hdr));
 451         offset = sizeof(*hdr);
 452         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 453     }
 454
 455     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 456      * but we may be passing along a larger header to the guest.
 457      */
 458     iov[0].iov_base += hdr_len;
 459     iov[0].iov_len  -= hdr_len;
 460
 461     return offset;
 462 }
 463
 464 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 465 {
 466     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 467     static const uint8_t vlan[] = {0x81, 0x00};
 468     uint8_t *ptr = (uint8_t *)buf;
 469     int i;
 470
 471     if (n->promisc)
 472         return 1;
 473
 474     if (n->has_vnet_hdr) {
 475         ptr += sizeof(struct virtio_net_hdr);
 476     }
 477
 478     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 479         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 480         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 481             return 0;
 482     }
 483
 484     if (ptr[0] & 1) { // multicast
 485         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 486             return !n->nobcast;
 487         } else if (n->nomulti) {
 488             return 0;
 489         } else if (n->allmulti || n->mac_table.multi_overflow) {
 490             return 1;
 491         }
 492
 493         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 494             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 495                 return 1;
 496             }
 497         }
 498     } else { // unicast
 499         if (n->nouni) {
 500             return 0;
 501         } else if (n->alluni || n->mac_table.uni_overflow) {
 502             return 1;
 503         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 504             return 1;
 505         }
 506
 507         for (i = 0; i < n->mac_table.first_multi; i++) {
 508             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 509                 return 1;
 510             }
 511         }
 512     }
 513
 514     return 0;
 515 }
 516
 517 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 518 {
 519     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 520     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 521     size_t hdr_len, offset, i;
 522
 523     if (!virtio_net_can_receive(&n->nic->nc))
 524         return -1;
 525
 526     if (!virtio_net_has_buffers(n, size))
 527         return 0;
 528
 529     if (!receive_filter(n, buf, size))
 530         return size;
 531
 532     /* hdr_len refers to the header we supply to the guest */
 533     hdr_len = n->mergeable_rx_bufs ?
 534         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 535
 536     offset = i = 0;
 537
 538     while (offset < size) {
 539         VirtQueueElement elem;
 540         int len, total;
 541         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 542
 543         total = 0;
 544
 545         if ((i != 0 && !n->mergeable_rx_bufs) ||
 546             virtqueue_pop(n->rx_vq, &elem) == 0) {
 547             if (i == 0)
 548                 return -1;
 549             fprintf(stderr, "virtio-net truncating packet\n");
 550             exit(1);
 551         }
 552
 553         if (elem.in_num < 1) {
 554             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 555             exit(1);
 556         }
 557
 558         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
 559             fprintf(stderr, "virtio-net header not in first element\n");
 560             exit(1);
 561         }
 562
 563         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 564
 565         if (i == 0) {
 566             if (n->mergeable_rx_bufs)
 567                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 568
 569             offset += receive_header(n, sg, elem.in_num,
 570                                      buf + offset, size - offset, hdr_len);
 571             total += hdr_len;
 572         }
 573
 574         /* copy in packet.  ugh */
 575         len = iov_fill(sg, elem.in_num,
 576                        buf + offset, size - offset);
 577         total += len;
 578
 579         /* signal other side */
 580         virtqueue_fill(n->rx_vq, &elem, total, i++);
 581
 582         offset += len;
 583     }
 584
 585     if (mhdr)
 586         mhdr->num_buffers = i;
 587
 588     virtqueue_flush(n->rx_vq, i);
 589     virtio_notify(&n->vdev, n->rx_vq);
 590
 591     return size;
 592 }
 593
 594 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 595
 596 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 597 {
 598     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 599
 600     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 601     virtio_notify(&n->vdev, n->tx_vq);
 602
 603     n->async_tx.elem.out_num = n->async_tx.len = 0;
 604
 605     virtio_queue_set_notification(n->tx_vq, 1);
 606     virtio_net_flush_tx(n, n->tx_vq);
 607 }
 608
 609 /* TX */
 610 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 611 {
 612     VirtQueueElement elem;
 613
 614     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 615         return;
 616
 617     if (n->async_tx.elem.out_num) {
 618         virtio_queue_set_notification(n->tx_vq, 0);
 619         return;
 620     }
 621
 622     while (virtqueue_pop(vq, &elem)) {
 623         ssize_t ret, len = 0;
 624         unsigned int out_num = elem.out_num;
 625         struct iovec *out_sg = &elem.out_sg[0];
 626         unsigned hdr_len;
 627
 628         /* hdr_len refers to the header received from the guest */
 629         hdr_len = n->mergeable_rx_bufs ?
 630             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 631             sizeof(struct virtio_net_hdr);
 632
 633         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 634             fprintf(stderr, "virtio-net header not in first element\n");
 635             exit(1);
 636         }
 637
 638         /* ignore the header if GSO is not supported */
 639         if (!n->has_vnet_hdr) {
 640             out_num--;
 641             out_sg++;
 642             len += hdr_len;
 643         } else if (n->mergeable_rx_bufs) {
 644             /* tapfd expects a struct virtio_net_hdr */
 645             hdr_len -= sizeof(struct virtio_net_hdr);
 646             out_sg->iov_len -= hdr_len;
 647             len += hdr_len;
 648         }
 649
 650         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 651                                       virtio_net_tx_complete);
 652         if (ret == 0) {
 653             virtio_queue_set_notification(n->tx_vq, 0);
 654             n->async_tx.elem = elem;
 655             n->async_tx.len  = len;
 656             return;
 657         }
 658
 659         len += ret;
 660
 661         virtqueue_push(vq, &elem, len);
 662         virtio_notify(&n->vdev, vq);
 663     }
 664 }
 665
 666 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 667 {
 668     VirtIONet *n = to_virtio_net(vdev);
 669
 670     if (n->tx_timer_active) {
 671         virtio_queue_set_notification(vq, 1);
 672         qemu_del_timer(n->tx_timer);
 673         n->tx_timer_active = 0;
 674         virtio_net_flush_tx(n, vq);
 675     } else {
 676         qemu_mod_timer(n->tx_timer,
 677                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 678         n->tx_timer_active = 1;
 679         virtio_queue_set_notification(vq, 0);
 680     }
 681 }
 682
 683 static void virtio_net_tx_timer(void *opaque)
 684 {
 685     VirtIONet *n = opaque;
 686
 687     n->tx_timer_active = 0;
 688
 689     /* Just in case the driver is not ready on more */
 690     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 691         return;
 692
 693     virtio_queue_set_notification(n->tx_vq, 1);
 694     virtio_net_flush_tx(n, n->tx_vq);
 695 }
 696
 697 static void virtio_net_save(QEMUFile *f, void *opaque)
 698 {
 699     VirtIONet *n = opaque;
 700
 701     virtio_save(&n->vdev, f);
 702
 703     qemu_put_buffer(f, n->mac, ETH_ALEN);
 704     qemu_put_be32(f, n->tx_timer_active);
 705     qemu_put_be32(f, n->mergeable_rx_bufs);
 706     qemu_put_be16(f, n->status);
 707     qemu_put_byte(f, n->promisc);
 708     qemu_put_byte(f, n->allmulti);
 709     qemu_put_be32(f, n->mac_table.in_use);
 710     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 711     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 712     qemu_put_be32(f, n->has_vnet_hdr);
 713     qemu_put_byte(f, n->mac_table.multi_overflow);
 714     qemu_put_byte(f, n->mac_table.uni_overflow);
 715     qemu_put_byte(f, n->alluni);
 716     qemu_put_byte(f, n->nomulti);
 717     qemu_put_byte(f, n->nouni);
 718     qemu_put_byte(f, n->nobcast);
 719     qemu_put_byte(f, n->has_ufo);
 720 }
 721
 722 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 723 {
 724     VirtIONet *n = opaque;
 725     int i;
 726
 727     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 728         return -EINVAL;
 729
 730     virtio_load(&n->vdev, f);
 731
 732     qemu_get_buffer(f, n->mac, ETH_ALEN);
 733     n->tx_timer_active = qemu_get_be32(f);
 734     n->mergeable_rx_bufs = qemu_get_be32(f);
 735
 736     if (version_id >= 3)
 737         n->status = qemu_get_be16(f);
 738
 739     if (version_id >= 4) {
 740         if (version_id < 8) {
 741             n->promisc = qemu_get_be32(f);
 742             n->allmulti = qemu_get_be32(f);
 743         } else {
 744             n->promisc = qemu_get_byte(f);
 745             n->allmulti = qemu_get_byte(f);
 746         }
 747     }
 748
 749     if (version_id >= 5) {
 750         n->mac_table.in_use = qemu_get_be32(f);
 751         /* MAC_TABLE_ENTRIES may be different from the saved image */
 752         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 753             qemu_get_buffer(f, n->mac_table.macs,
 754                             n->mac_table.in_use * ETH_ALEN);
 755         } else if (n->mac_table.in_use) {
 756             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 757             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 758             n->mac_table.in_use = 0;
 759         }
 760     }
 761
 762     if (version_id >= 6)
 763         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 764
 765     if (version_id >= 7) {
 766         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 767             qemu_error("virtio-net: saved image requires vnet_hdr=on\n");
 768             return -1;
 769         }
 770
 771         if (n->has_vnet_hdr) {
 772             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 773             tap_set_offload(n->nic->nc.peer,
 774                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 775                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 776                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 777                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 778                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 779         }
 780     }
 781
 782     if (version_id >= 9) {
 783         n->mac_table.multi_overflow = qemu_get_byte(f);
 784         n->mac_table.uni_overflow = qemu_get_byte(f);
 785     }
 786
 787     if (version_id >= 10) {
 788         n->alluni = qemu_get_byte(f);
 789         n->nomulti = qemu_get_byte(f);
 790         n->nouni = qemu_get_byte(f);
 791         n->nobcast = qemu_get_byte(f);
 792     }
 793
 794     if (version_id >= 11) {
 795         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 796             qemu_error("virtio-net: saved image requires TUN_F_UFO support\n");
 797             return -1;
 798         }
 799     }
 800
 801     /* Find the first multicast entry in the saved MAC filter */
 802     for (i = 0; i < n->mac_table.in_use; i++) {
 803         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 804             break;
 805         }
 806     }
 807     n->mac_table.first_multi = i;
 808
 809     if (n->tx_timer_active) {
 810         qemu_mod_timer(n->tx_timer,
 811                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 812     }
 813
 814     return 0;
 815 }
 816
 817 static void virtio_net_cleanup(VLANClientState *nc)
 818 {
 819     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 820
 821     n->nic = NULL;
 822 }
 823
 824 static NetClientInfo net_virtio_info = {
 825     .type = NET_CLIENT_TYPE_NIC,
 826     .size = sizeof(NICState),
 827     .can_receive = virtio_net_can_receive,
 828     .receive = virtio_net_receive,
 829         .cleanup = virtio_net_cleanup,
 830     .link_status_changed = virtio_net_set_link_status,
 831 };
 832
 833 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
 834 {
 835     VirtIONet *n;
 836     static int virtio_net_id;
 837
 838     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 839                                         sizeof(struct virtio_net_config),
 840                                         sizeof(VirtIONet));
 841
 842     n->vdev.get_config = virtio_net_get_config;
 843     n->vdev.set_config = virtio_net_set_config;
 844     n->vdev.get_features = virtio_net_get_features;
 845     n->vdev.set_features = virtio_net_set_features;
 846     n->vdev.bad_features = virtio_net_bad_features;
 847     n->vdev.reset = virtio_net_reset;
 848     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 849     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 850     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 851     qemu_macaddr_default_if_unset(&conf->macaddr);
 852     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
 853     n->status = VIRTIO_NET_S_LINK_UP;
 854
 855     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
 856
 857     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
 858
 859     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 860     n->tx_timer_active = 0;
 861     n->mergeable_rx_bufs = 0;
 862     n->promisc = 1; /* for compatibility */
 863
 864     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 865
 866     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 867
 868     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 869                     virtio_net_save, virtio_net_load, n);
 870
 871     return &n->vdev;
 872 }
 873
 874 void virtio_net_exit(VirtIODevice *vdev)
 875 {
 876     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 877
 878     qemu_purge_queued_packets(&n->nic->nc);
 879
 880     unregister_savevm("virtio-net", n);
 881
 882     qemu_free(n->mac_table.macs);
 883     qemu_free(n->vlans);
 884
 885     qemu_del_timer(n->tx_timer);
 886     qemu_free_timer(n->tx_timer);
 887
 888     virtio_cleanup(&n->vdev);
 889     qemu_del_vlan_client(&n->nic->nc);
 890 }