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