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)
 151 {
 152     VirtIONet *n = to_virtio_net(vdev);
 153     uint32_t features = (1 << VIRTIO_NET_F_MAC) |
 154                         (1 << VIRTIO_NET_F_MRG_RXBUF) |
 155                         (1 << VIRTIO_NET_F_STATUS) |
 156                         (1 << VIRTIO_NET_F_CTRL_VQ) |
 157                         (1 << VIRTIO_NET_F_CTRL_RX) |
 158                         (1 << VIRTIO_NET_F_CTRL_VLAN) |
 159                         (1 << VIRTIO_NET_F_CTRL_RX_EXTRA);
 160
 161     if (peer_has_vnet_hdr(n)) {
 162         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 163
 164         features |= (1 << VIRTIO_NET_F_CSUM);
 165         features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 166         features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 167         features |= (1 << VIRTIO_NET_F_HOST_ECN);
 168
 169         features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
 170         features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
 171         features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
 172         features |= (1 << VIRTIO_NET_F_GUEST_ECN);
 173
 174         if (peer_has_ufo(n)) {
 175             features |= (1 << VIRTIO_NET_F_GUEST_UFO);
 176             features |= (1 << VIRTIO_NET_F_HOST_UFO);
 177         }
 178     }
 179
 180     return features;
 181 }
 182
 183 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 184 {
 185     uint32_t features = 0;
 186
 187     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 188      * but also these: */
 189     features |= (1 << VIRTIO_NET_F_MAC);
 190     features |= (1 << VIRTIO_NET_F_CSUM);
 191     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 192     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 193     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 194
 195     return features & virtio_net_get_features(vdev);
 196 }
 197
 198 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 199 {
 200     VirtIONet *n = to_virtio_net(vdev);
 201
 202     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 203
 204     if (n->has_vnet_hdr) {
 205         tap_set_offload(n->nic->nc.peer,
 206                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 207                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 208                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 209                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 210                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 211     }
 212 }
 213
 214 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 215                                      VirtQueueElement *elem)
 216 {
 217     uint8_t on;
 218
 219     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 220         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 221         exit(1);
 222     }
 223
 224     on = ldub_p(elem->out_sg[1].iov_base);
 225
 226     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 227         n->promisc = on;
 228     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 229         n->allmulti = on;
 230     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 231         n->alluni = on;
 232     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 233         n->nomulti = on;
 234     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 235         n->nouni = on;
 236     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 237         n->nobcast = on;
 238     else
 239         return VIRTIO_NET_ERR;
 240
 241     return VIRTIO_NET_OK;
 242 }
 243
 244 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 245                                  VirtQueueElement *elem)
 246 {
 247     struct virtio_net_ctrl_mac mac_data;
 248
 249     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 250         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 251         elem->out_sg[2].iov_len < sizeof(mac_data))
 252         return VIRTIO_NET_ERR;
 253
 254     n->mac_table.in_use = 0;
 255     n->mac_table.first_multi = 0;
 256     n->mac_table.uni_overflow = 0;
 257     n->mac_table.multi_overflow = 0;
 258     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 259
 260     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 261
 262     if (sizeof(mac_data.entries) +
 263         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 264         return VIRTIO_NET_ERR;
 265
 266     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 267         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 268                mac_data.entries * ETH_ALEN);
 269         n->mac_table.in_use += mac_data.entries;
 270     } else {
 271         n->mac_table.uni_overflow = 1;
 272     }
 273
 274     n->mac_table.first_multi = n->mac_table.in_use;
 275
 276     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 277
 278     if (sizeof(mac_data.entries) +
 279         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 280         return VIRTIO_NET_ERR;
 281
 282     if (mac_data.entries) {
 283         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 284             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 285                    elem->out_sg[2].iov_base + sizeof(mac_data),
 286                    mac_data.entries * ETH_ALEN);
 287             n->mac_table.in_use += mac_data.entries;
 288         } else {
 289             n->mac_table.multi_overflow = 1;
 290         }
 291     }
 292
 293     return VIRTIO_NET_OK;
 294 }
 295
 296 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 297                                         VirtQueueElement *elem)
 298 {
 299     uint16_t vid;
 300
 301     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 302         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 303         return VIRTIO_NET_ERR;
 304     }
 305
 306     vid = lduw_le_p(elem->out_sg[1].iov_base);
 307
 308     if (vid >= MAX_VLAN)
 309         return VIRTIO_NET_ERR;
 310
 311     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 312         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 313     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 314         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 315     else
 316         return VIRTIO_NET_ERR;
 317
 318     return VIRTIO_NET_OK;
 319 }
 320
 321 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 322 {
 323     VirtIONet *n = to_virtio_net(vdev);
 324     struct virtio_net_ctrl_hdr ctrl;
 325     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 326     VirtQueueElement elem;
 327
 328     while (virtqueue_pop(vq, &elem)) {
 329         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 330             fprintf(stderr, "virtio-net ctrl missing headers\n");
 331             exit(1);
 332         }
 333
 334         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 335             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 336             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 337             exit(1);
 338         }
 339
 340         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 341         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 342
 343         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 344             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 345         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 346             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 347         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 348             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 349
 350         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 351
 352         virtqueue_push(vq, &elem, sizeof(status));
 353         virtio_notify(vdev, vq);
 354     }
 355 }
 356
 357 /* RX */
 358
 359 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 360 {
 361     VirtIONet *n = to_virtio_net(vdev);
 362
 363     qemu_flush_queued_packets(&n->nic->nc);
 364
 365     /* We now have RX buffers, signal to the IO thread to break out of the
 366      * select to re-poll the tap file descriptor */
 367     qemu_notify_event();
 368 }
 369
 370 static int virtio_net_can_receive(VLANClientState *nc)
 371 {
 372     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 373
 374     if (!virtio_queue_ready(n->rx_vq) ||
 375         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 376         return 0;
 377
 378     return 1;
 379 }
 380
 381 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 382 {
 383     if (virtio_queue_empty(n->rx_vq) ||
 384         (n->mergeable_rx_bufs &&
 385          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 386         virtio_queue_set_notification(n->rx_vq, 1);
 387
 388         /* To avoid a race condition where the guest has made some buffers
 389          * available after the above check but before notification was
 390          * enabled, check for available buffers again.
 391          */
 392         if (virtio_queue_empty(n->rx_vq) ||
 393             (n->mergeable_rx_bufs &&
 394              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 395             return 0;
 396     }
 397
 398     virtio_queue_set_notification(n->rx_vq, 0);
 399     return 1;
 400 }
 401
 402 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 403  * it never finds out that the packets don't have valid checksums.  This
 404  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 405  * fix this with Xen but it hasn't appeared in an upstream release of
 406  * dhclient yet.
 407  *
 408  * To avoid breaking existing guests, we catch udp packets and add
 409  * checksums.  This is terrible but it's better than hacking the guest
 410  * kernels.
 411  *
 412  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 413  * we should provide a mechanism to disable it to avoid polluting the host
 414  * cache.
 415  */
 416 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 417                                         const uint8_t *buf, size_t size)
 418 {
 419     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 420         (size > 27 && size < 1500) && /* normal sized MTU */
 421         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 422         (buf[23] == 17) && /* ip.protocol == UDP */
 423         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 424         /* FIXME this cast is evil */
 425         net_checksum_calculate((uint8_t *)buf, size);
 426         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 427     }
 428 }
 429
 430 static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count)
 431 {
 432     int offset, i;
 433
 434     offset = i = 0;
 435     while (offset < count && i < iovcnt) {
 436         int len = MIN(iov[i].iov_len, count - offset);
 437         memcpy(iov[i].iov_base, buf + offset, len);
 438         offset += len;
 439         i++;
 440     }
 441
 442     return offset;
 443 }
 444
 445 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 446                           const void *buf, size_t size, size_t hdr_len)
 447 {
 448     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 449     int offset = 0;
 450
 451     hdr->flags = 0;
 452     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 453
 454     if (n->has_vnet_hdr) {
 455         memcpy(hdr, buf, sizeof(*hdr));
 456         offset = sizeof(*hdr);
 457         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 458     }
 459
 460     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 461      * but we may be passing along a larger header to the guest.
 462      */
 463     iov[0].iov_base += hdr_len;
 464     iov[0].iov_len  -= hdr_len;
 465
 466     return offset;
 467 }
 468
 469 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 470 {
 471     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 472     static const uint8_t vlan[] = {0x81, 0x00};
 473     uint8_t *ptr = (uint8_t *)buf;
 474     int i;
 475
 476     if (n->promisc)
 477         return 1;
 478
 479     if (n->has_vnet_hdr) {
 480         ptr += sizeof(struct virtio_net_hdr);
 481     }
 482
 483     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 484         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 485         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 486             return 0;
 487     }
 488
 489     if (ptr[0] & 1) { // multicast
 490         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 491             return !n->nobcast;
 492         } else if (n->nomulti) {
 493             return 0;
 494         } else if (n->allmulti || n->mac_table.multi_overflow) {
 495             return 1;
 496         }
 497
 498         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 499             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 500                 return 1;
 501             }
 502         }
 503     } else { // unicast
 504         if (n->nouni) {
 505             return 0;
 506         } else if (n->alluni || n->mac_table.uni_overflow) {
 507             return 1;
 508         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 509             return 1;
 510         }
 511
 512         for (i = 0; i < n->mac_table.first_multi; i++) {
 513             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 514                 return 1;
 515             }
 516         }
 517     }
 518
 519     return 0;
 520 }
 521
 522 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 523 {
 524     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 525     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 526     size_t hdr_len, offset, i;
 527
 528     if (!virtio_net_can_receive(&n->nic->nc))
 529         return -1;
 530
 531     if (!virtio_net_has_buffers(n, size))
 532         return 0;
 533
 534     if (!receive_filter(n, buf, size))
 535         return size;
 536
 537     /* hdr_len refers to the header we supply to the guest */
 538     hdr_len = n->mergeable_rx_bufs ?
 539         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 540
 541     offset = i = 0;
 542
 543     while (offset < size) {
 544         VirtQueueElement elem;
 545         int len, total;
 546         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 547
 548         len = total = 0;
 549
 550         if ((i != 0 && !n->mergeable_rx_bufs) ||
 551             virtqueue_pop(n->rx_vq, &elem) == 0) {
 552             if (i == 0)
 553                 return -1;
 554             fprintf(stderr, "virtio-net truncating packet\n");
 555             exit(1);
 556         }
 557
 558         if (elem.in_num < 1) {
 559             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 560             exit(1);
 561         }
 562
 563         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
 564             fprintf(stderr, "virtio-net header not in first element\n");
 565             exit(1);
 566         }
 567
 568         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 569
 570         if (i == 0) {
 571             if (n->mergeable_rx_bufs)
 572                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 573
 574             offset += receive_header(n, sg, elem.in_num,
 575                                      buf + offset, size - offset, hdr_len);
 576             total += hdr_len;
 577         }
 578
 579         /* copy in packet.  ugh */
 580         len = iov_fill(sg, elem.in_num,
 581                        buf + offset, size - offset);
 582         total += len;
 583
 584         /* signal other side */
 585         virtqueue_fill(n->rx_vq, &elem, total, i++);
 586
 587         offset += len;
 588     }
 589
 590     if (mhdr)
 591         mhdr->num_buffers = i;
 592
 593     virtqueue_flush(n->rx_vq, i);
 594     virtio_notify(&n->vdev, n->rx_vq);
 595
 596     return size;
 597 }
 598
 599 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 600
 601 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 602 {
 603     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 604
 605     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 606     virtio_notify(&n->vdev, n->tx_vq);
 607
 608     n->async_tx.elem.out_num = n->async_tx.len = 0;
 609
 610     virtio_queue_set_notification(n->tx_vq, 1);
 611     virtio_net_flush_tx(n, n->tx_vq);
 612 }
 613
 614 /* TX */
 615 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 616 {
 617     VirtQueueElement elem;
 618
 619     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 620         return;
 621
 622     if (n->async_tx.elem.out_num) {
 623         virtio_queue_set_notification(n->tx_vq, 0);
 624         return;
 625     }
 626
 627     while (virtqueue_pop(vq, &elem)) {
 628         ssize_t ret, len = 0;
 629         unsigned int out_num = elem.out_num;
 630         struct iovec *out_sg = &elem.out_sg[0];
 631         unsigned hdr_len;
 632
 633         /* hdr_len refers to the header received from the guest */
 634         hdr_len = n->mergeable_rx_bufs ?
 635             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 636             sizeof(struct virtio_net_hdr);
 637
 638         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 639             fprintf(stderr, "virtio-net header not in first element\n");
 640             exit(1);
 641         }
 642
 643         /* ignore the header if GSO is not supported */
 644         if (!n->has_vnet_hdr) {
 645             out_num--;
 646             out_sg++;
 647             len += hdr_len;
 648         } else if (n->mergeable_rx_bufs) {
 649             /* tapfd expects a struct virtio_net_hdr */
 650             hdr_len -= sizeof(struct virtio_net_hdr);
 651             out_sg->iov_len -= hdr_len;
 652             len += hdr_len;
 653         }
 654
 655         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 656                                       virtio_net_tx_complete);
 657         if (ret == 0) {
 658             virtio_queue_set_notification(n->tx_vq, 0);
 659             n->async_tx.elem = elem;
 660             n->async_tx.len  = len;
 661             return;
 662         }
 663
 664         len += ret;
 665
 666         virtqueue_push(vq, &elem, len);
 667         virtio_notify(&n->vdev, vq);
 668     }
 669 }
 670
 671 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 672 {
 673     VirtIONet *n = to_virtio_net(vdev);
 674
 675     if (n->tx_timer_active) {
 676         virtio_queue_set_notification(vq, 1);
 677         qemu_del_timer(n->tx_timer);
 678         n->tx_timer_active = 0;
 679         virtio_net_flush_tx(n, vq);
 680     } else {
 681         qemu_mod_timer(n->tx_timer,
 682                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 683         n->tx_timer_active = 1;
 684         virtio_queue_set_notification(vq, 0);
 685     }
 686 }
 687
 688 static void virtio_net_tx_timer(void *opaque)
 689 {
 690     VirtIONet *n = opaque;
 691
 692     n->tx_timer_active = 0;
 693
 694     /* Just in case the driver is not ready on more */
 695     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 696         return;
 697
 698     virtio_queue_set_notification(n->tx_vq, 1);
 699     virtio_net_flush_tx(n, n->tx_vq);
 700 }
 701
 702 static void virtio_net_save(QEMUFile *f, void *opaque)
 703 {
 704     VirtIONet *n = opaque;
 705
 706     virtio_save(&n->vdev, f);
 707
 708     qemu_put_buffer(f, n->mac, ETH_ALEN);
 709     qemu_put_be32(f, n->tx_timer_active);
 710     qemu_put_be32(f, n->mergeable_rx_bufs);
 711     qemu_put_be16(f, n->status);
 712     qemu_put_byte(f, n->promisc);
 713     qemu_put_byte(f, n->allmulti);
 714     qemu_put_be32(f, n->mac_table.in_use);
 715     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 716     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 717     qemu_put_be32(f, n->has_vnet_hdr);
 718     qemu_put_byte(f, n->mac_table.multi_overflow);
 719     qemu_put_byte(f, n->mac_table.uni_overflow);
 720     qemu_put_byte(f, n->alluni);
 721     qemu_put_byte(f, n->nomulti);
 722     qemu_put_byte(f, n->nouni);
 723     qemu_put_byte(f, n->nobcast);
 724     qemu_put_byte(f, n->has_ufo);
 725 }
 726
 727 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 728 {
 729     VirtIONet *n = opaque;
 730     int i;
 731
 732     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 733         return -EINVAL;
 734
 735     virtio_load(&n->vdev, f);
 736
 737     qemu_get_buffer(f, n->mac, ETH_ALEN);
 738     n->tx_timer_active = qemu_get_be32(f);
 739     n->mergeable_rx_bufs = qemu_get_be32(f);
 740
 741     if (version_id >= 3)
 742         n->status = qemu_get_be16(f);
 743
 744     if (version_id >= 4) {
 745         if (version_id < 8) {
 746             n->promisc = qemu_get_be32(f);
 747             n->allmulti = qemu_get_be32(f);
 748         } else {
 749             n->promisc = qemu_get_byte(f);
 750             n->allmulti = qemu_get_byte(f);
 751         }
 752     }
 753
 754     if (version_id >= 5) {
 755         n->mac_table.in_use = qemu_get_be32(f);
 756         /* MAC_TABLE_ENTRIES may be different from the saved image */
 757         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 758             qemu_get_buffer(f, n->mac_table.macs,
 759                             n->mac_table.in_use * ETH_ALEN);
 760         } else if (n->mac_table.in_use) {
 761             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 762             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 763             n->mac_table.in_use = 0;
 764         }
 765     }
 766
 767     if (version_id >= 6)
 768         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 769
 770     if (version_id >= 7) {
 771         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 772             qemu_error("virtio-net: saved image requires vnet_hdr=on\n");
 773             return -1;
 774         }
 775
 776         if (n->has_vnet_hdr) {
 777             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 778             tap_set_offload(n->nic->nc.peer,
 779                             (n->vdev.features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 780                             (n->vdev.features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 781                             (n->vdev.features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 782                             (n->vdev.features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 783                             (n->vdev.features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 784         }
 785     }
 786
 787     if (version_id >= 9) {
 788         n->mac_table.multi_overflow = qemu_get_byte(f);
 789         n->mac_table.uni_overflow = qemu_get_byte(f);
 790     }
 791
 792     if (version_id >= 10) {
 793         n->alluni = qemu_get_byte(f);
 794         n->nomulti = qemu_get_byte(f);
 795         n->nouni = qemu_get_byte(f);
 796         n->nobcast = qemu_get_byte(f);
 797     }
 798
 799     if (version_id >= 11) {
 800         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 801             qemu_error("virtio-net: saved image requires TUN_F_UFO support\n");
 802             return -1;
 803         }
 804     }
 805
 806     /* Find the first multicast entry in the saved MAC filter */
 807     for (i = 0; i < n->mac_table.in_use; i++) {
 808         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 809             break;
 810         }
 811     }
 812     n->mac_table.first_multi = i;
 813
 814     if (n->tx_timer_active) {
 815         qemu_mod_timer(n->tx_timer,
 816                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 817     }
 818
 819     return 0;
 820 }
 821
 822 static void virtio_net_cleanup(VLANClientState *nc)
 823 {
 824     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 825
 826     n->nic = NULL;
 827 }
 828
 829 static NetClientInfo net_virtio_info = {
 830     .type = NET_CLIENT_TYPE_NIC,
 831     .size = sizeof(NICState),
 832     .can_receive = virtio_net_can_receive,
 833     .receive = virtio_net_receive,
 834         .cleanup = virtio_net_cleanup,
 835     .link_status_changed = virtio_net_set_link_status,
 836 };
 837
 838 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
 839 {
 840     VirtIONet *n;
 841     static int virtio_net_id;
 842
 843     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 844                                         sizeof(struct virtio_net_config),
 845                                         sizeof(VirtIONet));
 846
 847     n->vdev.get_config = virtio_net_get_config;
 848     n->vdev.set_config = virtio_net_set_config;
 849     n->vdev.get_features = virtio_net_get_features;
 850     n->vdev.set_features = virtio_net_set_features;
 851     n->vdev.bad_features = virtio_net_bad_features;
 852     n->vdev.reset = virtio_net_reset;
 853     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 854     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 855     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 856     qemu_macaddr_default_if_unset(&conf->macaddr);
 857     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
 858     n->status = VIRTIO_NET_S_LINK_UP;
 859
 860     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
 861
 862     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
 863
 864     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 865     n->tx_timer_active = 0;
 866     n->mergeable_rx_bufs = 0;
 867     n->promisc = 1; /* for compatibility */
 868
 869     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 870
 871     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 872
 873     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 874                     virtio_net_save, virtio_net_load, n);
 875
 876     return &n->vdev;
 877 }
 878
 879 void virtio_net_exit(VirtIODevice *vdev)
 880 {
 881     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 882
 883     qemu_purge_queued_packets(&n->nic->nc);
 884
 885     unregister_savevm("virtio-net", n);
 886
 887     qemu_free(n->mac_table.macs);
 888     qemu_free(n->vlans);
 889
 890     qemu_del_timer(n->tx_timer);
 891     qemu_free_timer(n->tx_timer);
 892
 893     virtio_cleanup(&n->vdev);
 894     qemu_del_vlan_client(&n->nic->nc);
 895 }