hw/virtio-net.c

   1 /*
   2  * Virtio Network Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "iov.h"
  15 #include "virtio.h"
  16 #include "net.h"
  17 #include "net/checksum.h"
  18 #include "net/tap.h"
  19 #include "qemu-error.h"
  20 #include "qemu-timer.h"
  21 #include "virtio-net.h"
  22 #include "vhost_net.h"
  23
  24 #define VIRTIO_NET_VM_VERSION    11
  25
  26 #define MAC_TABLE_ENTRIES    64
  27 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  28
  29 typedef struct VirtIONet
  30 {
  31     VirtIODevice vdev;
  32     uint8_t mac[ETH_ALEN];
  33     uint16_t status;
  34     VirtQueue *rx_vq;
  35     VirtQueue *tx_vq;
  36     VirtQueue *ctrl_vq;
  37     NICState *nic;
  38     QEMUTimer *tx_timer;
  39     uint32_t tx_timeout;
  40     int32_t tx_burst;
  41     int tx_waiting;
  42     uint32_t has_vnet_hdr;
  43     uint8_t has_ufo;
  44     struct {
  45         VirtQueueElement elem;
  46         ssize_t len;
  47     } async_tx;
  48     int mergeable_rx_bufs;
  49     uint8_t promisc;
  50     uint8_t allmulti;
  51     uint8_t alluni;
  52     uint8_t nomulti;
  53     uint8_t nouni;
  54     uint8_t nobcast;
  55     uint8_t vhost_started;
  56     VMChangeStateEntry *vmstate;
  57     struct {
  58         int in_use;
  59         int first_multi;
  60         uint8_t multi_overflow;
  61         uint8_t uni_overflow;
  62         uint8_t *macs;
  63     } mac_table;
  64     uint32_t *vlans;
  65     DeviceState *qdev;
  66 } VirtIONet;
  67
  68 /* TODO
  69  * - we could suppress RX interrupt if we were so inclined.
  70  */
  71
  72 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  73 {
  74     return (VirtIONet *)vdev;
  75 }
  76
  77 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  78 {
  79     VirtIONet *n = to_virtio_net(vdev);
  80     struct virtio_net_config netcfg;
  81
  82     netcfg.status = n->status;
  83     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  84     memcpy(config, &netcfg, sizeof(netcfg));
  85 }
  86
  87 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  88 {
  89     VirtIONet *n = to_virtio_net(vdev);
  90     struct virtio_net_config netcfg;
  91
  92     memcpy(&netcfg, config, sizeof(netcfg));
  93
  94     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  95         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  96         qemu_format_nic_info_str(&n->nic->nc, n->mac);
  97     }
  98 }
  99
 100 static void virtio_net_set_link_status(VLANClientState *nc)
 101 {
 102     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 103     uint16_t old_status = n->status;
 104
 105     if (nc->link_down)
 106         n->status &= ~VIRTIO_NET_S_LINK_UP;
 107     else
 108         n->status |= VIRTIO_NET_S_LINK_UP;
 109
 110     if (n->status != old_status)
 111         virtio_notify_config(&n->vdev);
 112 }
 113
 114 static void virtio_net_reset(VirtIODevice *vdev)
 115 {
 116     VirtIONet *n = to_virtio_net(vdev);
 117
 118     /* Reset back to compatibility mode */
 119     n->promisc = 1;
 120     n->allmulti = 0;
 121     n->alluni = 0;
 122     n->nomulti = 0;
 123     n->nouni = 0;
 124     n->nobcast = 0;
 125     if (n->vhost_started) {
 126         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 127         n->vhost_started = 0;
 128     }
 129
 130     /* Flush any MAC and VLAN filter table state */
 131     n->mac_table.in_use = 0;
 132     n->mac_table.first_multi = 0;
 133     n->mac_table.multi_overflow = 0;
 134     n->mac_table.uni_overflow = 0;
 135     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 136     memset(n->vlans, 0, MAX_VLAN >> 3);
 137 }
 138
 139 static int peer_has_vnet_hdr(VirtIONet *n)
 140 {
 141     if (!n->nic->nc.peer)
 142         return 0;
 143
 144     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 145         return 0;
 146
 147     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 148
 149     return n->has_vnet_hdr;
 150 }
 151
 152 static int peer_has_ufo(VirtIONet *n)
 153 {
 154     if (!peer_has_vnet_hdr(n))
 155         return 0;
 156
 157     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 158
 159     return n->has_ufo;
 160 }
 161
 162 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 163 {
 164     VirtIONet *n = to_virtio_net(vdev);
 165
 166     features |= (1 << VIRTIO_NET_F_MAC);
 167
 168     if (peer_has_vnet_hdr(n)) {
 169         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 170     } else {
 171         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 172         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 173         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 174         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 175
 176         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 177         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 178         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 179         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 180     }
 181
 182     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 183         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 184         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 185     }
 186
 187     if (!n->nic->nc.peer ||
 188         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 189         return features;
 190     }
 191     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 192         return features;
 193     }
 194     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 195 }
 196
 197 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 198 {
 199     uint32_t features = 0;
 200
 201     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 202      * but also these: */
 203     features |= (1 << VIRTIO_NET_F_MAC);
 204     features |= (1 << VIRTIO_NET_F_CSUM);
 205     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 206     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 207     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 208
 209     return features;
 210 }
 211
 212 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 213 {
 214     VirtIONet *n = to_virtio_net(vdev);
 215
 216     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 217
 218     if (n->has_vnet_hdr) {
 219         tap_set_offload(n->nic->nc.peer,
 220                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 221                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 222                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 223                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 224                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 225     }
 226     if (!n->nic->nc.peer ||
 227         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 228         return;
 229     }
 230     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 231         return;
 232     }
 233     vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 234 }
 235
 236 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 237                                      VirtQueueElement *elem)
 238 {
 239     uint8_t on;
 240
 241     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 242         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 243         exit(1);
 244     }
 245
 246     on = ldub_p(elem->out_sg[1].iov_base);
 247
 248     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 249         n->promisc = on;
 250     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 251         n->allmulti = on;
 252     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 253         n->alluni = on;
 254     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 255         n->nomulti = on;
 256     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 257         n->nouni = on;
 258     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 259         n->nobcast = on;
 260     else
 261         return VIRTIO_NET_ERR;
 262
 263     return VIRTIO_NET_OK;
 264 }
 265
 266 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 267                                  VirtQueueElement *elem)
 268 {
 269     struct virtio_net_ctrl_mac mac_data;
 270
 271     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 272         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 273         elem->out_sg[2].iov_len < sizeof(mac_data))
 274         return VIRTIO_NET_ERR;
 275
 276     n->mac_table.in_use = 0;
 277     n->mac_table.first_multi = 0;
 278     n->mac_table.uni_overflow = 0;
 279     n->mac_table.multi_overflow = 0;
 280     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 281
 282     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 283
 284     if (sizeof(mac_data.entries) +
 285         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 286         return VIRTIO_NET_ERR;
 287
 288     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 289         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 290                mac_data.entries * ETH_ALEN);
 291         n->mac_table.in_use += mac_data.entries;
 292     } else {
 293         n->mac_table.uni_overflow = 1;
 294     }
 295
 296     n->mac_table.first_multi = n->mac_table.in_use;
 297
 298     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 299
 300     if (sizeof(mac_data.entries) +
 301         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 302         return VIRTIO_NET_ERR;
 303
 304     if (mac_data.entries) {
 305         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 306             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 307                    elem->out_sg[2].iov_base + sizeof(mac_data),
 308                    mac_data.entries * ETH_ALEN);
 309             n->mac_table.in_use += mac_data.entries;
 310         } else {
 311             n->mac_table.multi_overflow = 1;
 312         }
 313     }
 314
 315     return VIRTIO_NET_OK;
 316 }
 317
 318 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 319                                         VirtQueueElement *elem)
 320 {
 321     uint16_t vid;
 322
 323     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 324         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 325         return VIRTIO_NET_ERR;
 326     }
 327
 328     vid = lduw_le_p(elem->out_sg[1].iov_base);
 329
 330     if (vid >= MAX_VLAN)
 331         return VIRTIO_NET_ERR;
 332
 333     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 334         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 335     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 336         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 337     else
 338         return VIRTIO_NET_ERR;
 339
 340     return VIRTIO_NET_OK;
 341 }
 342
 343 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 344 {
 345     VirtIONet *n = to_virtio_net(vdev);
 346     struct virtio_net_ctrl_hdr ctrl;
 347     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 348     VirtQueueElement elem;
 349
 350     while (virtqueue_pop(vq, &elem)) {
 351         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 352             fprintf(stderr, "virtio-net ctrl missing headers\n");
 353             exit(1);
 354         }
 355
 356         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 357             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 358             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 359             exit(1);
 360         }
 361
 362         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 363         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 364
 365         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 366             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 367         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 368             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 369         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 370             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 371
 372         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 373
 374         virtqueue_push(vq, &elem, sizeof(status));
 375         virtio_notify(vdev, vq);
 376     }
 377 }
 378
 379 /* RX */
 380
 381 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 382 {
 383     VirtIONet *n = to_virtio_net(vdev);
 384
 385     qemu_flush_queued_packets(&n->nic->nc);
 386
 387     /* We now have RX buffers, signal to the IO thread to break out of the
 388      * select to re-poll the tap file descriptor */
 389     qemu_notify_event();
 390 }
 391
 392 static int virtio_net_can_receive(VLANClientState *nc)
 393 {
 394     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 395
 396     if (!virtio_queue_ready(n->rx_vq) ||
 397         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 398         return 0;
 399
 400     return 1;
 401 }
 402
 403 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 404 {
 405     if (virtio_queue_empty(n->rx_vq) ||
 406         (n->mergeable_rx_bufs &&
 407          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 408         virtio_queue_set_notification(n->rx_vq, 1);
 409
 410         /* To avoid a race condition where the guest has made some buffers
 411          * available after the above check but before notification was
 412          * enabled, check for available buffers again.
 413          */
 414         if (virtio_queue_empty(n->rx_vq) ||
 415             (n->mergeable_rx_bufs &&
 416              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 417             return 0;
 418     }
 419
 420     virtio_queue_set_notification(n->rx_vq, 0);
 421     return 1;
 422 }
 423
 424 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 425  * it never finds out that the packets don't have valid checksums.  This
 426  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 427  * fix this with Xen but it hasn't appeared in an upstream release of
 428  * dhclient yet.
 429  *
 430  * To avoid breaking existing guests, we catch udp packets and add
 431  * checksums.  This is terrible but it's better than hacking the guest
 432  * kernels.
 433  *
 434  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 435  * we should provide a mechanism to disable it to avoid polluting the host
 436  * cache.
 437  */
 438 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 439                                         const uint8_t *buf, size_t size)
 440 {
 441     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 442         (size > 27 && size < 1500) && /* normal sized MTU */
 443         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 444         (buf[23] == 17) && /* ip.protocol == UDP */
 445         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 446         /* FIXME this cast is evil */
 447         net_checksum_calculate((uint8_t *)buf, size);
 448         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 449     }
 450 }
 451
 452 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 453                           const void *buf, size_t size, size_t hdr_len)
 454 {
 455     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 456     int offset = 0;
 457
 458     hdr->flags = 0;
 459     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 460
 461     if (n->has_vnet_hdr) {
 462         memcpy(hdr, buf, sizeof(*hdr));
 463         offset = sizeof(*hdr);
 464         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 465     }
 466
 467     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 468      * but we may be passing along a larger header to the guest.
 469      */
 470     iov[0].iov_base += hdr_len;
 471     iov[0].iov_len  -= hdr_len;
 472
 473     return offset;
 474 }
 475
 476 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 477 {
 478     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 479     static const uint8_t vlan[] = {0x81, 0x00};
 480     uint8_t *ptr = (uint8_t *)buf;
 481     int i;
 482
 483     if (n->promisc)
 484         return 1;
 485
 486     if (n->has_vnet_hdr) {
 487         ptr += sizeof(struct virtio_net_hdr);
 488     }
 489
 490     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 491         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 492         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 493             return 0;
 494     }
 495
 496     if (ptr[0] & 1) { // multicast
 497         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 498             return !n->nobcast;
 499         } else if (n->nomulti) {
 500             return 0;
 501         } else if (n->allmulti || n->mac_table.multi_overflow) {
 502             return 1;
 503         }
 504
 505         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 506             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 507                 return 1;
 508             }
 509         }
 510     } else { // unicast
 511         if (n->nouni) {
 512             return 0;
 513         } else if (n->alluni || n->mac_table.uni_overflow) {
 514             return 1;
 515         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 516             return 1;
 517         }
 518
 519         for (i = 0; i < n->mac_table.first_multi; i++) {
 520             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 521                 return 1;
 522             }
 523         }
 524     }
 525
 526     return 0;
 527 }
 528
 529 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 530 {
 531     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 532     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 533     size_t guest_hdr_len, offset, i, host_hdr_len;
 534
 535     if (!virtio_net_can_receive(&n->nic->nc))
 536         return -1;
 537
 538     /* hdr_len refers to the header we supply to the guest */
 539     guest_hdr_len = n->mergeable_rx_bufs ?
 540         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 541
 542
 543     host_hdr_len = n->has_vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 544     if (!virtio_net_has_buffers(n, size + guest_hdr_len - host_hdr_len))
 545         return 0;
 546
 547     if (!receive_filter(n, buf, size))
 548         return size;
 549
 550     offset = i = 0;
 551
 552     while (offset < size) {
 553         VirtQueueElement elem;
 554         int len, total;
 555         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 556
 557         total = 0;
 558
 559         if (virtqueue_pop(n->rx_vq, &elem) == 0) {
 560             if (i == 0)
 561                 return -1;
 562             fprintf(stderr, "virtio-net unexpected empty queue: "
 563                     "i %zd mergeable %d offset %zd, size %zd, "
 564                     "guest hdr len %zd, host hdr len %zd guest features 0x%x\n",
 565                     i, n->mergeable_rx_bufs, offset, size,
 566                     guest_hdr_len, host_hdr_len, n->vdev.guest_features);
 567             exit(1);
 568         }
 569
 570         if (elem.in_num < 1) {
 571             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 572             exit(1);
 573         }
 574
 575         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != guest_hdr_len) {
 576             fprintf(stderr, "virtio-net header not in first element\n");
 577             exit(1);
 578         }
 579
 580         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 581
 582         if (i == 0) {
 583             if (n->mergeable_rx_bufs)
 584                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 585
 586             offset += receive_header(n, sg, elem.in_num,
 587                                      buf + offset, size - offset, guest_hdr_len);
 588             total += guest_hdr_len;
 589         }
 590
 591         /* copy in packet.  ugh */
 592         len = iov_from_buf(sg, elem.in_num,
 593                            buf + offset, size - offset);
 594         total += len;
 595         offset += len;
 596         /* If buffers can't be merged, at this point we
 597          * must have consumed the complete packet.
 598          * Otherwise, drop it. */
 599         if (!n->mergeable_rx_bufs && offset < size) {
 600 #if 0
 601             fprintf(stderr, "virtio-net truncated non-mergeable packet: "
 602
 603                     "i %zd mergeable %d offset %zd, size %zd, "
 604                     "guest hdr len %zd, host hdr len %zd\n",
 605                     i, n->mergeable_rx_bufs,
 606                     offset, size, guest_hdr_len, host_hdr_len);
 607 #endif
 608             return size;
 609         }
 610
 611         /* signal other side */
 612         virtqueue_fill(n->rx_vq, &elem, total, i++);
 613     }
 614
 615     if (mhdr)
 616         mhdr->num_buffers = i;
 617
 618     virtqueue_flush(n->rx_vq, i);
 619     virtio_notify(&n->vdev, n->rx_vq);
 620
 621     return size;
 622 }
 623
 624 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 625
 626 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 627 {
 628     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 629
 630     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 631     virtio_notify(&n->vdev, n->tx_vq);
 632
 633     n->async_tx.elem.out_num = n->async_tx.len = 0;
 634
 635     virtio_queue_set_notification(n->tx_vq, 1);
 636     virtio_net_flush_tx(n, n->tx_vq);
 637 }
 638
 639 /* TX */
 640 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 641 {
 642     VirtQueueElement elem;
 643     int32_t num_packets = 0;
 644
 645     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 646         return num_packets;
 647     }
 648
 649     if (n->async_tx.elem.out_num) {
 650         virtio_queue_set_notification(n->tx_vq, 0);
 651         return num_packets;
 652     }
 653
 654     while (virtqueue_pop(vq, &elem)) {
 655         ssize_t ret, len = 0;
 656         unsigned int out_num = elem.out_num;
 657         struct iovec *out_sg = &elem.out_sg[0];
 658         unsigned hdr_len;
 659
 660         /* hdr_len refers to the header received from the guest */
 661         hdr_len = n->mergeable_rx_bufs ?
 662             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 663             sizeof(struct virtio_net_hdr);
 664
 665         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 666             fprintf(stderr, "virtio-net header not in first element\n");
 667             exit(1);
 668         }
 669
 670         /* ignore the header if GSO is not supported */
 671         if (!n->has_vnet_hdr) {
 672             out_num--;
 673             out_sg++;
 674             len += hdr_len;
 675         } else if (n->mergeable_rx_bufs) {
 676             /* tapfd expects a struct virtio_net_hdr */
 677             hdr_len -= sizeof(struct virtio_net_hdr);
 678             out_sg->iov_len -= hdr_len;
 679             len += hdr_len;
 680         }
 681
 682         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 683                                       virtio_net_tx_complete);
 684         if (ret == 0) {
 685             virtio_queue_set_notification(n->tx_vq, 0);
 686             n->async_tx.elem = elem;
 687             n->async_tx.len  = len;
 688             return -EBUSY;
 689         }
 690
 691         len += ret;
 692
 693         virtqueue_push(vq, &elem, len);
 694         virtio_notify(&n->vdev, vq);
 695
 696         if (++num_packets >= n->tx_burst) {
 697             break;
 698         }
 699     }
 700     return num_packets;
 701 }
 702
 703 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 704 {
 705     VirtIONet *n = to_virtio_net(vdev);
 706
 707     if (n->tx_waiting) {
 708         virtio_queue_set_notification(vq, 1);
 709         qemu_del_timer(n->tx_timer);
 710         n->tx_waiting = 0;
 711         virtio_net_flush_tx(n, vq);
 712     } else {
 713         qemu_mod_timer(n->tx_timer,
 714                        qemu_get_clock(vm_clock) + n->tx_timeout);
 715         n->tx_waiting = 1;
 716         virtio_queue_set_notification(vq, 0);
 717     }
 718 }
 719
 720 static void virtio_net_tx_timer(void *opaque)
 721 {
 722     VirtIONet *n = opaque;
 723
 724     n->tx_waiting = 0;
 725
 726     /* Just in case the driver is not ready on more */
 727     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 728         return;
 729
 730     virtio_queue_set_notification(n->tx_vq, 1);
 731     virtio_net_flush_tx(n, n->tx_vq);
 732 }
 733
 734 static void virtio_net_save(QEMUFile *f, void *opaque)
 735 {
 736     VirtIONet *n = opaque;
 737
 738     if (n->vhost_started) {
 739         /* TODO: should we really stop the backend?
 740          * If we don't, it might keep writing to memory. */
 741         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 742         n->vhost_started = 0;
 743     }
 744     virtio_save(&n->vdev, f);
 745
 746     qemu_put_buffer(f, n->mac, ETH_ALEN);
 747     qemu_put_be32(f, n->tx_waiting);
 748     qemu_put_be32(f, n->mergeable_rx_bufs);
 749     qemu_put_be16(f, n->status);
 750     qemu_put_byte(f, n->promisc);
 751     qemu_put_byte(f, n->allmulti);
 752     qemu_put_be32(f, n->mac_table.in_use);
 753     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 754     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 755     qemu_put_be32(f, n->has_vnet_hdr);
 756     qemu_put_byte(f, n->mac_table.multi_overflow);
 757     qemu_put_byte(f, n->mac_table.uni_overflow);
 758     qemu_put_byte(f, n->alluni);
 759     qemu_put_byte(f, n->nomulti);
 760     qemu_put_byte(f, n->nouni);
 761     qemu_put_byte(f, n->nobcast);
 762     qemu_put_byte(f, n->has_ufo);
 763 }
 764
 765 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 766 {
 767     VirtIONet *n = opaque;
 768     int i;
 769
 770     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 771         return -EINVAL;
 772
 773     virtio_load(&n->vdev, f);
 774
 775     qemu_get_buffer(f, n->mac, ETH_ALEN);
 776     n->tx_waiting = qemu_get_be32(f);
 777     n->mergeable_rx_bufs = qemu_get_be32(f);
 778
 779     if (version_id >= 3)
 780         n->status = qemu_get_be16(f);
 781
 782     if (version_id >= 4) {
 783         if (version_id < 8) {
 784             n->promisc = qemu_get_be32(f);
 785             n->allmulti = qemu_get_be32(f);
 786         } else {
 787             n->promisc = qemu_get_byte(f);
 788             n->allmulti = qemu_get_byte(f);
 789         }
 790     }
 791
 792     if (version_id >= 5) {
 793         n->mac_table.in_use = qemu_get_be32(f);
 794         /* MAC_TABLE_ENTRIES may be different from the saved image */
 795         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 796             qemu_get_buffer(f, n->mac_table.macs,
 797                             n->mac_table.in_use * ETH_ALEN);
 798         } else if (n->mac_table.in_use) {
 799             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 800             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 801             n->mac_table.in_use = 0;
 802         }
 803     }
 804
 805     if (version_id >= 6)
 806         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 807
 808     if (version_id >= 7) {
 809         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 810             error_report("virtio-net: saved image requires vnet_hdr=on");
 811             return -1;
 812         }
 813
 814         if (n->has_vnet_hdr) {
 815             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 816             tap_set_offload(n->nic->nc.peer,
 817                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 818                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 819                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 820                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 821                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 822         }
 823     }
 824
 825     if (version_id >= 9) {
 826         n->mac_table.multi_overflow = qemu_get_byte(f);
 827         n->mac_table.uni_overflow = qemu_get_byte(f);
 828     }
 829
 830     if (version_id >= 10) {
 831         n->alluni = qemu_get_byte(f);
 832         n->nomulti = qemu_get_byte(f);
 833         n->nouni = qemu_get_byte(f);
 834         n->nobcast = qemu_get_byte(f);
 835     }
 836
 837     if (version_id >= 11) {
 838         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 839             error_report("virtio-net: saved image requires TUN_F_UFO support");
 840             return -1;
 841         }
 842     }
 843
 844     /* Find the first multicast entry in the saved MAC filter */
 845     for (i = 0; i < n->mac_table.in_use; i++) {
 846         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 847             break;
 848         }
 849     }
 850     n->mac_table.first_multi = i;
 851
 852     if (n->tx_waiting) {
 853         qemu_mod_timer(n->tx_timer,
 854                        qemu_get_clock(vm_clock) + n->tx_timeout);
 855     }
 856     return 0;
 857 }
 858
 859 static void virtio_net_cleanup(VLANClientState *nc)
 860 {
 861     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 862
 863     n->nic = NULL;
 864 }
 865
 866 static NetClientInfo net_virtio_info = {
 867     .type = NET_CLIENT_TYPE_NIC,
 868     .size = sizeof(NICState),
 869     .can_receive = virtio_net_can_receive,
 870     .receive = virtio_net_receive,
 871         .cleanup = virtio_net_cleanup,
 872     .link_status_changed = virtio_net_set_link_status,
 873 };
 874
 875 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 876 {
 877     VirtIONet *n = to_virtio_net(vdev);
 878     if (!n->nic->nc.peer) {
 879         return;
 880     }
 881     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 882         return;
 883     }
 884
 885     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 886         return;
 887     }
 888     if (!!n->vhost_started == !!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 889         return;
 890     }
 891     if (status & VIRTIO_CONFIG_S_DRIVER_OK) {
 892         int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), vdev);
 893         if (r < 0) {
 894             fprintf(stderr, "unable to start vhost net: %d: "
 895                     "falling back on userspace virtio\n", -r);
 896         } else {
 897             n->vhost_started = 1;
 898         }
 899     } else {
 900         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 901         n->vhost_started = 0;
 902     }
 903 }
 904
 905 static void virtio_net_vmstate_change(void *opaque, int running, int reason)
 906 {
 907     VirtIONet *n = opaque;
 908     uint8_t status = running ? VIRTIO_CONFIG_S_DRIVER_OK : 0;
 909     /* This is called when vm is started/stopped,
 910      * it will start/stop vhost backend if * appropriate
 911      * e.g. after migration. */
 912     virtio_net_set_status(&n->vdev, n->vdev.status & status);
 913 }
 914
 915 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
 916                               virtio_net_conf *net)
 917 {
 918     VirtIONet *n;
 919
 920     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 921                                         sizeof(struct virtio_net_config),
 922                                         sizeof(VirtIONet));
 923
 924     n->vdev.get_config = virtio_net_get_config;
 925     n->vdev.set_config = virtio_net_set_config;
 926     n->vdev.get_features = virtio_net_get_features;
 927     n->vdev.set_features = virtio_net_set_features;
 928     n->vdev.bad_features = virtio_net_bad_features;
 929     n->vdev.reset = virtio_net_reset;
 930     n->vdev.set_status = virtio_net_set_status;
 931     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 932     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 933     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 934     qemu_macaddr_default_if_unset(&conf->macaddr);
 935     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
 936     n->status = VIRTIO_NET_S_LINK_UP;
 937
 938     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
 939
 940     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
 941
 942     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 943     n->tx_waiting = 0;
 944     n->tx_timeout = net->txtimer;
 945     n->tx_burst = net->txburst;
 946     n->mergeable_rx_bufs = 0;
 947     n->promisc = 1; /* for compatibility */
 948
 949     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 950
 951     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 952
 953     n->qdev = dev;
 954     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
 955                     virtio_net_save, virtio_net_load, n);
 956     n->vmstate = qemu_add_vm_change_state_handler(virtio_net_vmstate_change, n);
 957
 958     return &n->vdev;
 959 }
 960
 961 void virtio_net_exit(VirtIODevice *vdev)
 962 {
 963     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
 964     qemu_del_vm_change_state_handler(n->vmstate);
 965
 966     if (n->vhost_started) {
 967         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), vdev);
 968     }
 969
 970     qemu_purge_queued_packets(&n->nic->nc);
 971
 972     unregister_savevm(n->qdev, "virtio-net", n);
 973
 974     qemu_free(n->mac_table.macs);
 975     qemu_free(n->vlans);
 976
 977     qemu_del_timer(n->tx_timer);
 978     qemu_free_timer(n->tx_timer);
 979
 980     virtio_cleanup(&n->vdev);
 981     qemu_del_vlan_client(&n->nic->nc);
 982 }