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 "qemu-timer.h"
  17 #include "virtio-net.h"
  18 #ifdef CONFIG_KVM
  19 #include "qemu-kvm.h"
  20 #endif
  21
  22 #define TAP_VNET_HDR
  23
  24 #define VIRTIO_NET_VM_VERSION    10
  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     VLANClientState *vc;
  38     QEMUTimer *tx_timer;
  39     int tx_timer_active;
  40     struct {
  41         VirtQueueElement elem;
  42         ssize_t len;
  43     } async_tx;
  44     int mergeable_rx_bufs;
  45     uint8_t promisc;
  46     uint8_t allmulti;
  47     uint8_t alluni;
  48     uint8_t nomulti;
  49     uint8_t nouni;
  50     uint8_t nobcast;
  51     struct {
  52         int in_use;
  53         int first_multi;
  54         uint8_t multi_overflow;
  55         uint8_t uni_overflow;
  56         uint8_t *macs;
  57     } mac_table;
  58     uint32_t *vlans;
  59 } VirtIONet;
  60
  61 /* TODO
  62  * - we could suppress RX interrupt if we were so inclined.
  63  */
  64
  65 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  66 {
  67     return (VirtIONet *)vdev;
  68 }
  69
  70 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  71 {
  72     VirtIONet *n = to_virtio_net(vdev);
  73     struct virtio_net_config netcfg;
  74
  75     netcfg.status = n->status;
  76     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  77     memcpy(config, &netcfg, sizeof(netcfg));
  78 }
  79
  80 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  81 {
  82     VirtIONet *n = to_virtio_net(vdev);
  83     struct virtio_net_config netcfg;
  84
  85     memcpy(&netcfg, config, sizeof(netcfg));
  86
  87     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  88         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  89         qemu_format_nic_info_str(n->vc, n->mac);
  90     }
  91 }
  92
  93 static void virtio_net_set_link_status(VLANClientState *vc)
  94 {
  95     VirtIONet *n = vc->opaque;
  96     uint16_t old_status = n->status;
  97
  98     if (vc->link_down)
  99         n->status &= ~VIRTIO_NET_S_LINK_UP;
 100     else
 101         n->status |= VIRTIO_NET_S_LINK_UP;
 102
 103     if (n->status != old_status)
 104         virtio_notify_config(&n->vdev);
 105 }
 106
 107 static void virtio_net_reset(VirtIODevice *vdev)
 108 {
 109     VirtIONet *n = to_virtio_net(vdev);
 110
 111     /* Reset back to compatibility mode */
 112     n->promisc = 1;
 113     n->allmulti = 0;
 114     n->alluni = 0;
 115     n->nomulti = 0;
 116     n->nouni = 0;
 117     n->nobcast = 0;
 118
 119     /* Flush any MAC and VLAN filter table state */
 120     n->mac_table.in_use = 0;
 121     n->mac_table.first_multi = 0;
 122     n->mac_table.multi_overflow = 0;
 123     n->mac_table.uni_overflow = 0;
 124     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 125     memset(n->vlans, 0, MAX_VLAN >> 3);
 126 }
 127
 128 static uint32_t virtio_net_get_features(VirtIODevice *vdev)
 129 {
 130     uint32_t features = (1 << VIRTIO_NET_F_MAC) |
 131                         (1 << VIRTIO_NET_F_MRG_RXBUF) |
 132                         (1 << VIRTIO_NET_F_STATUS) |
 133                         (1 << VIRTIO_NET_F_CTRL_VQ) |
 134                         (1 << VIRTIO_NET_F_CTRL_RX) |
 135                         (1 << VIRTIO_NET_F_CTRL_VLAN) |
 136                         (1 << VIRTIO_NET_F_CTRL_RX_EXTRA);
 137
 138 #ifdef TAP_VNET_HDR
 139     VirtIONet *n = to_virtio_net(vdev);
 140     VLANClientState *host = QTAILQ_FIRST(&n->vc->vlan->clients);
 141
 142     if (tap_has_vnet_hdr(host)) {
 143         tap_using_vnet_hdr(host, 1);
 144         features |= (1 << VIRTIO_NET_F_CSUM);
 145         features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
 146         features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
 147         features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
 148         features |= (1 << VIRTIO_NET_F_GUEST_ECN);
 149         features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 150         features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 151         features |= (1 << VIRTIO_NET_F_HOST_ECN);
 152         features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
 153         if (tap_has_ufo(host)) {
 154             features |= (1 << VIRTIO_NET_F_GUEST_UFO);
 155             features |= (1 << VIRTIO_NET_F_HOST_UFO);
 156         }
 157     }
 158 #endif
 159
 160     return features;
 161 }
 162
 163 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 164 {
 165     uint32_t features = 0;
 166
 167     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 168      * but also these: */
 169     features |= (1 << VIRTIO_NET_F_MAC);
 170     features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
 171     features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
 172     features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
 173     features |= (1 << VIRTIO_NET_F_GUEST_ECN);
 174
 175     return features & virtio_net_get_features(vdev);
 176 }
 177
 178 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 179 {
 180     VirtIONet *n = to_virtio_net(vdev);
 181 #ifdef TAP_VNET_HDR
 182     VLANClientState *host = QTAILQ_FIRST(&n->vc->vlan->clients);
 183 #endif
 184
 185     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 186
 187 #ifdef TAP_VNET_HDR
 188     if (!tap_has_vnet_hdr(host) || !host->set_offload)
 189         return;
 190
 191     host->set_offload(host,
 192                       (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 193                       (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 194                       (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 195                       (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 196                       (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 197 #endif
 198 }
 199
 200 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 201                                      VirtQueueElement *elem)
 202 {
 203     uint8_t on;
 204
 205     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 206         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 207         exit(1);
 208     }
 209
 210     on = ldub_p(elem->out_sg[1].iov_base);
 211
 212     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 213         n->promisc = on;
 214     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 215         n->allmulti = on;
 216     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 217         n->alluni = on;
 218     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 219         n->nomulti = on;
 220     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 221         n->nouni = on;
 222     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 223         n->nobcast = on;
 224     else
 225         return VIRTIO_NET_ERR;
 226
 227     return VIRTIO_NET_OK;
 228 }
 229
 230 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 231                                  VirtQueueElement *elem)
 232 {
 233     struct virtio_net_ctrl_mac mac_data;
 234
 235     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 236         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 237         elem->out_sg[2].iov_len < sizeof(mac_data))
 238         return VIRTIO_NET_ERR;
 239
 240     n->mac_table.in_use = 0;
 241     n->mac_table.first_multi = 0;
 242     n->mac_table.uni_overflow = 0;
 243     n->mac_table.multi_overflow = 0;
 244     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 245
 246     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 247
 248     if (sizeof(mac_data.entries) +
 249         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 250         return VIRTIO_NET_ERR;
 251
 252     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 253         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 254                mac_data.entries * ETH_ALEN);
 255         n->mac_table.in_use += mac_data.entries;
 256     } else {
 257         n->mac_table.uni_overflow = 1;
 258     }
 259
 260     n->mac_table.first_multi = n->mac_table.in_use;
 261
 262     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 263
 264     if (sizeof(mac_data.entries) +
 265         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 266         return VIRTIO_NET_ERR;
 267
 268     if (mac_data.entries) {
 269         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 270             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 271                    elem->out_sg[2].iov_base + sizeof(mac_data),
 272                    mac_data.entries * ETH_ALEN);
 273             n->mac_table.in_use += mac_data.entries;
 274         } else {
 275             n->mac_table.multi_overflow = 1;
 276         }
 277     }
 278
 279     return VIRTIO_NET_OK;
 280 }
 281
 282 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 283                                         VirtQueueElement *elem)
 284 {
 285     uint16_t vid;
 286
 287     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 288         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 289         return VIRTIO_NET_ERR;
 290     }
 291
 292     vid = lduw_le_p(elem->out_sg[1].iov_base);
 293
 294     if (vid >= MAX_VLAN)
 295         return VIRTIO_NET_ERR;
 296
 297     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 298         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 299     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 300         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 301     else
 302         return VIRTIO_NET_ERR;
 303
 304     return VIRTIO_NET_OK;
 305 }
 306
 307 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 308 {
 309     VirtIONet *n = to_virtio_net(vdev);
 310     struct virtio_net_ctrl_hdr ctrl;
 311     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 312     VirtQueueElement elem;
 313
 314     while (virtqueue_pop(vq, &elem)) {
 315         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 316             fprintf(stderr, "virtio-net ctrl missing headers\n");
 317             exit(1);
 318         }
 319
 320         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 321             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 322             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 323             exit(1);
 324         }
 325
 326         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 327         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 328
 329         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 330             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 331         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 332             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 333         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 334             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 335
 336         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 337
 338         virtqueue_push(vq, &elem, sizeof(status));
 339         virtio_notify(vdev, vq);
 340     }
 341 }
 342
 343 /* RX */
 344
 345 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 346 {
 347     VirtIONet *n = to_virtio_net(vdev);
 348
 349     qemu_flush_queued_packets(n->vc);
 350
 351     /* We now have RX buffers, signal to the IO thread to break out of the
 352      * select to re-poll the tap file descriptor */
 353     qemu_notify_event();
 354 }
 355
 356 static int do_virtio_net_can_receive(VirtIONet *n, int bufsize)
 357 {
 358     if (!virtio_queue_ready(n->rx_vq) ||
 359         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 360         return 0;
 361
 362     if (virtio_queue_empty(n->rx_vq) ||
 363         (n->mergeable_rx_bufs &&
 364          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 365         virtio_queue_set_notification(n->rx_vq, 1);
 366         return 0;
 367     }
 368
 369     virtio_queue_set_notification(n->rx_vq, 0);
 370     return 1;
 371 }
 372
 373 static int virtio_net_can_receive(VLANClientState *vc)
 374 {
 375     VirtIONet *n = vc->opaque;
 376
 377     return do_virtio_net_can_receive(n, VIRTIO_NET_MAX_BUFSIZE);
 378 }
 379
 380 #ifdef TAP_VNET_HDR
 381 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 382  * it never finds out that the packets don't have valid checksums.  This
 383  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 384  * fix this with Xen but it hasn't appeared in an upstream release of
 385  * dhclient yet.
 386  *
 387  * To avoid breaking existing guests, we catch udp packets and add
 388  * checksums.  This is terrible but it's better than hacking the guest
 389  * kernels.
 390  *
 391  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 392  * we should provide a mechanism to disable it to avoid polluting the host
 393  * cache.
 394  */
 395 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 396                                         const uint8_t *buf, size_t size)
 397 {
 398     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 399         (size > 27 && size < 1500) && /* normal sized MTU */
 400         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 401         (buf[23] == 17) && /* ip.protocol == UDP */
 402         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 403         /* FIXME this cast is evil */
 404         net_checksum_calculate((uint8_t *)buf, size);
 405         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 406     }
 407 }
 408 #endif
 409
 410 static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count)
 411 {
 412     int offset, i;
 413
 414     offset = i = 0;
 415     while (offset < count && i < iovcnt) {
 416         int len = MIN(iov[i].iov_len, count - offset);
 417         memcpy(iov[i].iov_base, buf + offset, len);
 418         offset += len;
 419         i++;
 420     }
 421
 422     return offset;
 423 }
 424
 425 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 426                           const void *buf, size_t size, size_t hdr_len, int raw)
 427 {
 428     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 429     int offset = 0;
 430
 431     hdr->flags = 0;
 432     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 433
 434 #ifdef TAP_VNET_HDR
 435     if (tap_has_vnet_hdr(QTAILQ_FIRST(&n->vc->vlan->clients))) {
 436         if (!raw) {
 437             memcpy(hdr, buf, sizeof(*hdr));
 438         } else {
 439             memset(hdr, 0, sizeof(*hdr));
 440         }
 441         offset = sizeof(*hdr);
 442         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 443     }
 444 #endif
 445
 446     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 447      * but we may be passing along a larger header to the guest.
 448      */
 449     iov[0].iov_base += hdr_len;
 450     iov[0].iov_len  -= hdr_len;
 451
 452     return offset;
 453 }
 454
 455 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 456 {
 457     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 458     static const uint8_t vlan[] = {0x81, 0x00};
 459     uint8_t *ptr = (uint8_t *)buf;
 460     int i;
 461
 462     if (n->promisc)
 463         return 1;
 464
 465 #ifdef TAP_VNET_HDR
 466     if (tap_has_vnet_hdr(QTAILQ_FIRST(&n->vc->vlan->clients)))
 467         ptr += sizeof(struct virtio_net_hdr);
 468 #endif
 469
 470     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 471         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 472         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 473             return 0;
 474     }
 475
 476     if (ptr[0] & 1) { // multicast
 477         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 478             return !n->nobcast;
 479         } else if (n->nomulti) {
 480             return 0;
 481         } else if (n->allmulti || n->mac_table.multi_overflow) {
 482             return 1;
 483         }
 484
 485         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 486             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 487                 return 1;
 488             }
 489         }
 490     } else { // unicast
 491         if (n->nouni) {
 492             return 0;
 493         } else if (n->alluni || n->mac_table.uni_overflow) {
 494             return 1;
 495         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 496             return 1;
 497         }
 498
 499         for (i = 0; i < n->mac_table.first_multi; i++) {
 500             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 501                 return 1;
 502             }
 503         }
 504     }
 505
 506     return 0;
 507 }
 508
 509 static ssize_t virtio_net_receive2(VLANClientState *vc, const uint8_t *buf, size_t size, int raw)
 510 {
 511     VirtIONet *n = vc->opaque;
 512     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 513     size_t hdr_len, offset, i;
 514
 515     if (!do_virtio_net_can_receive(n, size))
 516         return 0;
 517
 518     if (!receive_filter(n, buf, size))
 519         return size;
 520
 521     /* hdr_len refers to the header we supply to the guest */
 522     hdr_len = n->mergeable_rx_bufs ?
 523         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 524
 525     offset = i = 0;
 526
 527     while (offset < size) {
 528         VirtQueueElement elem;
 529         int len, total;
 530         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 531
 532         len = total = 0;
 533
 534         if ((i != 0 && !n->mergeable_rx_bufs) ||
 535             virtqueue_pop(n->rx_vq, &elem) == 0) {
 536             if (i == 0)
 537                 return -1;
 538             fprintf(stderr, "virtio-net truncating packet\n");
 539             exit(1);
 540         }
 541
 542         if (elem.in_num < 1) {
 543             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 544             exit(1);
 545         }
 546
 547         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
 548             fprintf(stderr, "virtio-net header not in first element\n");
 549             exit(1);
 550         }
 551
 552         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 553
 554         if (i == 0) {
 555             if (n->mergeable_rx_bufs)
 556                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 557
 558             offset += receive_header(n, sg, elem.in_num,
 559                                      buf + offset, size - offset, hdr_len, raw);
 560             total += hdr_len;
 561         }
 562
 563         /* copy in packet.  ugh */
 564         len = iov_fill(sg, elem.in_num,
 565                        buf + offset, size - offset);
 566         total += len;
 567
 568         /* signal other side */
 569         virtqueue_fill(n->rx_vq, &elem, total, i++);
 570
 571         offset += len;
 572     }
 573
 574     if (mhdr)
 575         mhdr->num_buffers = i;
 576
 577     virtqueue_flush(n->rx_vq, i);
 578     virtio_notify(&n->vdev, n->rx_vq);
 579
 580     return size;
 581 }
 582
 583 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 584
 585 static void virtio_net_tx_complete(VLANClientState *vc, ssize_t len)
 586 {
 587     VirtIONet *n = vc->opaque;
 588
 589     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 590     virtio_notify(&n->vdev, n->tx_vq);
 591
 592     n->async_tx.elem.out_num = n->async_tx.len = 0;
 593
 594     virtio_queue_set_notification(n->tx_vq, 1);
 595     virtio_net_flush_tx(n, n->tx_vq);
 596 }
 597
 598 static ssize_t virtio_net_receive(VLANClientState *vc, const uint8_t *buf, size_t size)
 599 {
 600     return virtio_net_receive2(vc, buf, size, 0);
 601 }
 602
 603 static ssize_t virtio_net_receive_raw(VLANClientState *vc, const uint8_t *buf, size_t size)
 604 {
 605     return virtio_net_receive2(vc, buf, size, 1);
 606 }
 607
 608 /* TX */
 609 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 610 {
 611     VirtQueueElement elem;
 612 #ifdef TAP_VNET_HDR
 613     int has_vnet_hdr = tap_has_vnet_hdr(QTAILQ_FIRST(&n->vc->vlan->clients));
 614 #else
 615     int has_vnet_hdr = 0;
 616 #endif
 617
 618     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 619         return;
 620
 621     if (n->async_tx.elem.out_num) {
 622         virtio_queue_set_notification(n->tx_vq, 0);
 623         return;
 624     }
 625
 626     while (virtqueue_pop(vq, &elem)) {
 627         ssize_t ret, len = 0;
 628         unsigned int out_num = elem.out_num;
 629         struct iovec *out_sg = &elem.out_sg[0];
 630         unsigned hdr_len;
 631
 632         /* hdr_len refers to the header received from the guest */
 633         hdr_len = n->mergeable_rx_bufs ?
 634             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 635             sizeof(struct virtio_net_hdr);
 636
 637         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 638             fprintf(stderr, "virtio-net header not in first element\n");
 639             exit(1);
 640         }
 641
 642         /* ignore the header if GSO is not supported */
 643         if (!has_vnet_hdr) {
 644             out_num--;
 645             out_sg++;
 646             len += hdr_len;
 647         } else if (n->mergeable_rx_bufs) {
 648             /* tapfd expects a struct virtio_net_hdr */
 649             hdr_len -= sizeof(struct virtio_net_hdr);
 650             out_sg->iov_len -= hdr_len;
 651             len += hdr_len;
 652         }
 653
 654         ret = qemu_sendv_packet_async(n->vc, out_sg, out_num,
 655                                       virtio_net_tx_complete);
 656         if (ret == 0) {
 657             virtio_queue_set_notification(n->tx_vq, 0);
 658             n->async_tx.elem = elem;
 659             n->async_tx.len  = len;
 660             return;
 661         }
 662
 663         len += ret;
 664
 665         virtqueue_push(vq, &elem, len);
 666         virtio_notify(&n->vdev, vq);
 667     }
 668 }
 669
 670 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 671 {
 672     VirtIONet *n = to_virtio_net(vdev);
 673
 674     if (n->tx_timer_active) {
 675         virtio_queue_set_notification(vq, 1);
 676         qemu_del_timer(n->tx_timer);
 677         n->tx_timer_active = 0;
 678         virtio_net_flush_tx(n, vq);
 679     } else {
 680         qemu_mod_timer(n->tx_timer,
 681                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 682         n->tx_timer_active = 1;
 683         virtio_queue_set_notification(vq, 0);
 684     }
 685 }
 686
 687 static void virtio_net_tx_timer(void *opaque)
 688 {
 689     VirtIONet *n = opaque;
 690
 691     n->tx_timer_active = 0;
 692
 693     /* Just in case the driver is not ready on more */
 694     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 695         return;
 696
 697     virtio_queue_set_notification(n->tx_vq, 1);
 698     virtio_net_flush_tx(n, n->tx_vq);
 699 }
 700
 701 static void virtio_net_save(QEMUFile *f, void *opaque)
 702 {
 703     VirtIONet *n = opaque;
 704
 705     virtio_save(&n->vdev, f);
 706
 707     qemu_put_buffer(f, n->mac, ETH_ALEN);
 708     qemu_put_be32(f, n->tx_timer_active);
 709     qemu_put_be32(f, n->mergeable_rx_bufs);
 710     qemu_put_be16(f, n->status);
 711     qemu_put_byte(f, n->promisc);
 712     qemu_put_byte(f, n->allmulti);
 713     qemu_put_be32(f, n->mac_table.in_use);
 714     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 715     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 716
 717 #ifdef TAP_VNET_HDR
 718     qemu_put_be32(f, tap_has_vnet_hdr(QTAILQ_FIRST(&n->vc->vlan->clients)));
 719 #else
 720     qemu_put_be32(f, 0);
 721 #endif
 722
 723     qemu_put_byte(f, n->mac_table.multi_overflow);
 724     qemu_put_byte(f, n->mac_table.uni_overflow);
 725     qemu_put_byte(f, n->alluni);
 726     qemu_put_byte(f, n->nomulti);
 727     qemu_put_byte(f, n->nouni);
 728     qemu_put_byte(f, n->nobcast);
 729 }
 730
 731 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 732 {
 733     VirtIONet *n = opaque;
 734     int i;
 735
 736     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 737         return -EINVAL;
 738
 739     virtio_load(&n->vdev, f);
 740
 741     qemu_get_buffer(f, n->mac, ETH_ALEN);
 742     n->tx_timer_active = qemu_get_be32(f);
 743     n->mergeable_rx_bufs = qemu_get_be32(f);
 744
 745     if (version_id >= 3)
 746         n->status = qemu_get_be16(f);
 747
 748     if (version_id >= 4) {
 749         if (version_id < 8) {
 750             n->promisc = qemu_get_be32(f);
 751             n->allmulti = qemu_get_be32(f);
 752         } else {
 753             n->promisc = qemu_get_byte(f);
 754             n->allmulti = qemu_get_byte(f);
 755         }
 756     }
 757
 758     if (version_id >= 5) {
 759         n->mac_table.in_use = qemu_get_be32(f);
 760         /* MAC_TABLE_ENTRIES may be different from the saved image */
 761         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 762             qemu_get_buffer(f, n->mac_table.macs,
 763                             n->mac_table.in_use * ETH_ALEN);
 764         } else if (n->mac_table.in_use) {
 765             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 766             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 767             n->mac_table.in_use = 0;
 768         }
 769     }
 770
 771     if (version_id >= 6)
 772         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 773
 774     if (version_id >= 7 && qemu_get_be32(f)) {
 775 #ifdef TAP_VNET_HDR
 776         tap_using_vnet_hdr(QTAILQ_FIRST(&n->vc->vlan->clients), 1);
 777 #else
 778         fprintf(stderr,
 779                 "virtio-net: saved image requires vnet header support\n");
 780         exit(1);
 781 #endif
 782     }
 783
 784     if (version_id >= 9) {
 785         n->mac_table.multi_overflow = qemu_get_byte(f);
 786         n->mac_table.uni_overflow = qemu_get_byte(f);
 787     }
 788
 789     if (version_id >= 10) {
 790         n->alluni = qemu_get_byte(f);
 791         n->nomulti = qemu_get_byte(f);
 792         n->nouni = qemu_get_byte(f);
 793         n->nobcast = qemu_get_byte(f);
 794     }
 795
 796     /* Find the first multicast entry in the saved MAC filter */
 797     for (i = 0; i < n->mac_table.in_use; i++) {
 798         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 799             break;
 800         }
 801     }
 802     n->mac_table.first_multi = i;
 803
 804     if (n->tx_timer_active) {
 805         qemu_mod_timer(n->tx_timer,
 806                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 807     }
 808
 809     return 0;
 810 }
 811
 812 static void virtio_net_cleanup(VLANClientState *vc)
 813 {
 814     VirtIONet *n = vc->opaque;
 815
 816     n->vc = NULL;
 817 }
 818
 819 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf)
 820 {
 821     VirtIONet *n;
 822     static int virtio_net_id;
 823
 824     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 825                                         sizeof(struct virtio_net_config),
 826                                         sizeof(VirtIONet));
 827
 828     n->vdev.get_config = virtio_net_get_config;
 829     n->vdev.set_config = virtio_net_set_config;
 830     n->vdev.get_features = virtio_net_get_features;
 831     n->vdev.set_features = virtio_net_set_features;
 832     n->vdev.bad_features = virtio_net_bad_features;
 833     n->vdev.reset = virtio_net_reset;
 834     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 835     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 836     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 837     qemu_macaddr_default_if_unset(&conf->macaddr);
 838     n->status = VIRTIO_NET_S_LINK_UP;
 839     n->vc = qemu_new_vlan_client(NET_CLIENT_TYPE_NIC, conf->vlan, conf->peer,
 840                                  dev->info->name, dev->id,
 841                                  virtio_net_can_receive,
 842                                  virtio_net_receive, NULL,
 843                                  virtio_net_cleanup, n);
 844     n->vc->link_status_changed = virtio_net_set_link_status;
 845     n->vc->receive_raw = virtio_net_receive_raw;
 846
 847     qemu_format_nic_info_str(n->vc, conf->macaddr.a);
 848
 849     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 850     n->tx_timer_active = 0;
 851     n->mergeable_rx_bufs = 0;
 852     n->promisc = 1; /* for compatibility */
 853
 854     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 855
 856     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 857
 858     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 859                     virtio_net_save, virtio_net_load, n);
 860
 861     return &n->vdev;
 862 }
 863
 864 void virtio_net_exit(VirtIODevice *vdev)
 865 {
 866     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 867
 868     qemu_purge_queued_packets(n->vc);
 869
 870     unregister_savevm("virtio-net", n);
 871
 872     qemu_free(n->mac_table.macs);
 873     qemu_free(n->vlans);
 874
 875     qemu_del_timer(n->tx_timer);
 876     qemu_free_timer(n->tx_timer);
 877
 878     virtio_cleanup(&n->vdev);
 879     qemu_del_vlan_client(n->vc);
 880 }