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     QEMUBH *tx_bh;
  40     uint32_t tx_timeout;
  41     int32_t tx_burst;
  42     int tx_waiting;
  43     uint32_t has_vnet_hdr;
  44     uint8_t has_ufo;
  45     struct {
  46         VirtQueueElement elem;
  47         ssize_t len;
  48     } async_tx;
  49     int mergeable_rx_bufs;
  50     uint8_t promisc;
  51     uint8_t allmulti;
  52     uint8_t alluni;
  53     uint8_t nomulti;
  54     uint8_t nouni;
  55     uint8_t nobcast;
  56     uint8_t vhost_started;
  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 bool virtio_net_started(VirtIONet *n, uint8_t status)
 101 {
 102     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 103         (n->status & VIRTIO_NET_S_LINK_UP) && n->vdev.vm_running;
 104 }
 105
 106 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
 107 {
 108     if (!n->nic->nc.peer) {
 109         return;
 110     }
 111     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 112         return;
 113     }
 114
 115     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 116         return;
 117     }
 118     if (!!n->vhost_started == virtio_net_started(n, status)) {
 119         return;
 120     }
 121     if (!n->vhost_started) {
 122         int r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 123         if (r < 0) {
 124             error_report("unable to start vhost net: %d: "
 125                          "falling back on userspace virtio", -r);
 126         } else {
 127             n->vhost_started = 1;
 128         }
 129     } else {
 130         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 131         n->vhost_started = 0;
 132     }
 133 }
 134
 135 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 136 {
 137     VirtIONet *n = to_virtio_net(vdev);
 138
 139     virtio_net_vhost_status(n, status);
 140
 141     if (!n->tx_waiting) {
 142         return;
 143     }
 144
 145     if (virtio_net_started(n, status) && !n->vhost_started) {
 146         if (n->tx_timer) {
 147             qemu_mod_timer(n->tx_timer,
 148                            qemu_get_clock(vm_clock) + n->tx_timeout);
 149         } else {
 150             qemu_bh_schedule(n->tx_bh);
 151         }
 152     } else {
 153         if (n->tx_timer) {
 154             qemu_del_timer(n->tx_timer);
 155         } else {
 156             qemu_bh_cancel(n->tx_bh);
 157         }
 158     }
 159 }
 160
 161 static void virtio_net_set_link_status(VLANClientState *nc)
 162 {
 163     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 164     uint16_t old_status = n->status;
 165
 166     if (nc->link_down)
 167         n->status &= ~VIRTIO_NET_S_LINK_UP;
 168     else
 169         n->status |= VIRTIO_NET_S_LINK_UP;
 170
 171     if (n->status != old_status)
 172         virtio_notify_config(&n->vdev);
 173
 174     virtio_net_set_status(&n->vdev, n->vdev.status);
 175 }
 176
 177 static void virtio_net_reset(VirtIODevice *vdev)
 178 {
 179     VirtIONet *n = to_virtio_net(vdev);
 180
 181     /* Reset back to compatibility mode */
 182     n->promisc = 1;
 183     n->allmulti = 0;
 184     n->alluni = 0;
 185     n->nomulti = 0;
 186     n->nouni = 0;
 187     n->nobcast = 0;
 188
 189     /* Flush any MAC and VLAN filter table state */
 190     n->mac_table.in_use = 0;
 191     n->mac_table.first_multi = 0;
 192     n->mac_table.multi_overflow = 0;
 193     n->mac_table.uni_overflow = 0;
 194     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 195     memset(n->vlans, 0, MAX_VLAN >> 3);
 196 }
 197
 198 static int peer_has_vnet_hdr(VirtIONet *n)
 199 {
 200     if (!n->nic->nc.peer)
 201         return 0;
 202
 203     if (n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP)
 204         return 0;
 205
 206     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 207
 208     return n->has_vnet_hdr;
 209 }
 210
 211 static int peer_has_ufo(VirtIONet *n)
 212 {
 213     if (!peer_has_vnet_hdr(n))
 214         return 0;
 215
 216     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 217
 218     return n->has_ufo;
 219 }
 220
 221 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 222 {
 223     VirtIONet *n = to_virtio_net(vdev);
 224
 225     features |= (1 << VIRTIO_NET_F_MAC);
 226
 227     if (peer_has_vnet_hdr(n)) {
 228         tap_using_vnet_hdr(n->nic->nc.peer, 1);
 229     } else {
 230         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 231         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 232         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 233         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 234
 235         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 236         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 237         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 238         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 239     }
 240
 241     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 242         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 243         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 244     }
 245
 246     if (!n->nic->nc.peer ||
 247         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 248         return features;
 249     }
 250     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 251         return features;
 252     }
 253     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 254 }
 255
 256 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 257 {
 258     uint32_t features = 0;
 259
 260     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 261      * but also these: */
 262     features |= (1 << VIRTIO_NET_F_MAC);
 263     features |= (1 << VIRTIO_NET_F_CSUM);
 264     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 265     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 266     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 267
 268     return features;
 269 }
 270
 271 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 272 {
 273     VirtIONet *n = to_virtio_net(vdev);
 274
 275     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 276
 277     if (n->has_vnet_hdr) {
 278         tap_set_offload(n->nic->nc.peer,
 279                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 280                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 281                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 282                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 283                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 284     }
 285     if (!n->nic->nc.peer ||
 286         n->nic->nc.peer->info->type != NET_CLIENT_TYPE_TAP) {
 287         return;
 288     }
 289     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 290         return;
 291     }
 292     vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 293 }
 294
 295 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 296                                      VirtQueueElement *elem)
 297 {
 298     uint8_t on;
 299
 300     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 301         error_report("virtio-net ctrl invalid rx mode command");
 302         exit(1);
 303     }
 304
 305     on = ldub_p(elem->out_sg[1].iov_base);
 306
 307     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 308         n->promisc = on;
 309     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 310         n->allmulti = on;
 311     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 312         n->alluni = on;
 313     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 314         n->nomulti = on;
 315     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 316         n->nouni = on;
 317     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 318         n->nobcast = on;
 319     else
 320         return VIRTIO_NET_ERR;
 321
 322     return VIRTIO_NET_OK;
 323 }
 324
 325 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 326                                  VirtQueueElement *elem)
 327 {
 328     struct virtio_net_ctrl_mac mac_data;
 329
 330     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 331         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 332         elem->out_sg[2].iov_len < sizeof(mac_data))
 333         return VIRTIO_NET_ERR;
 334
 335     n->mac_table.in_use = 0;
 336     n->mac_table.first_multi = 0;
 337     n->mac_table.uni_overflow = 0;
 338     n->mac_table.multi_overflow = 0;
 339     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 340
 341     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 342
 343     if (sizeof(mac_data.entries) +
 344         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 345         return VIRTIO_NET_ERR;
 346
 347     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 348         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 349                mac_data.entries * ETH_ALEN);
 350         n->mac_table.in_use += mac_data.entries;
 351     } else {
 352         n->mac_table.uni_overflow = 1;
 353     }
 354
 355     n->mac_table.first_multi = n->mac_table.in_use;
 356
 357     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 358
 359     if (sizeof(mac_data.entries) +
 360         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 361         return VIRTIO_NET_ERR;
 362
 363     if (mac_data.entries) {
 364         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 365             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 366                    elem->out_sg[2].iov_base + sizeof(mac_data),
 367                    mac_data.entries * ETH_ALEN);
 368             n->mac_table.in_use += mac_data.entries;
 369         } else {
 370             n->mac_table.multi_overflow = 1;
 371         }
 372     }
 373
 374     return VIRTIO_NET_OK;
 375 }
 376
 377 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 378                                         VirtQueueElement *elem)
 379 {
 380     uint16_t vid;
 381
 382     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 383         error_report("virtio-net ctrl invalid vlan command");
 384         return VIRTIO_NET_ERR;
 385     }
 386
 387     vid = lduw_le_p(elem->out_sg[1].iov_base);
 388
 389     if (vid >= MAX_VLAN)
 390         return VIRTIO_NET_ERR;
 391
 392     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 393         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 394     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 395         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 396     else
 397         return VIRTIO_NET_ERR;
 398
 399     return VIRTIO_NET_OK;
 400 }
 401
 402 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 403 {
 404     VirtIONet *n = to_virtio_net(vdev);
 405     struct virtio_net_ctrl_hdr ctrl;
 406     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 407     VirtQueueElement elem;
 408
 409     while (virtqueue_pop(vq, &elem)) {
 410         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 411             error_report("virtio-net ctrl missing headers");
 412             exit(1);
 413         }
 414
 415         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 416             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 417             error_report("virtio-net ctrl header not in correct element");
 418             exit(1);
 419         }
 420
 421         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 422         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 423
 424         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 425             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 426         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 427             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 428         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 429             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 430
 431         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 432
 433         virtqueue_push(vq, &elem, sizeof(status));
 434         virtio_notify(vdev, vq);
 435     }
 436 }
 437
 438 /* RX */
 439
 440 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 441 {
 442     VirtIONet *n = to_virtio_net(vdev);
 443
 444     qemu_flush_queued_packets(&n->nic->nc);
 445
 446     /* We now have RX buffers, signal to the IO thread to break out of the
 447      * select to re-poll the tap file descriptor */
 448     qemu_notify_event();
 449 }
 450
 451 static int virtio_net_can_receive(VLANClientState *nc)
 452 {
 453     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 454     if (!n->vdev.vm_running) {
 455         return 0;
 456     }
 457
 458     if (!virtio_queue_ready(n->rx_vq) ||
 459         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 460         return 0;
 461
 462     return 1;
 463 }
 464
 465 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 466 {
 467     if (virtio_queue_empty(n->rx_vq) ||
 468         (n->mergeable_rx_bufs &&
 469          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 470         virtio_queue_set_notification(n->rx_vq, 1);
 471
 472         /* To avoid a race condition where the guest has made some buffers
 473          * available after the above check but before notification was
 474          * enabled, check for available buffers again.
 475          */
 476         if (virtio_queue_empty(n->rx_vq) ||
 477             (n->mergeable_rx_bufs &&
 478              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 479             return 0;
 480     }
 481
 482     virtio_queue_set_notification(n->rx_vq, 0);
 483     return 1;
 484 }
 485
 486 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 487  * it never finds out that the packets don't have valid checksums.  This
 488  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 489  * fix this with Xen but it hasn't appeared in an upstream release of
 490  * dhclient yet.
 491  *
 492  * To avoid breaking existing guests, we catch udp packets and add
 493  * checksums.  This is terrible but it's better than hacking the guest
 494  * kernels.
 495  *
 496  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 497  * we should provide a mechanism to disable it to avoid polluting the host
 498  * cache.
 499  */
 500 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 501                                         const uint8_t *buf, size_t size)
 502 {
 503     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 504         (size > 27 && size < 1500) && /* normal sized MTU */
 505         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 506         (buf[23] == 17) && /* ip.protocol == UDP */
 507         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 508         /* FIXME this cast is evil */
 509         net_checksum_calculate((uint8_t *)buf, size);
 510         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 511     }
 512 }
 513
 514 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 515                           const void *buf, size_t size, size_t hdr_len)
 516 {
 517     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 518     int offset = 0;
 519
 520     hdr->flags = 0;
 521     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 522
 523     if (n->has_vnet_hdr) {
 524         memcpy(hdr, buf, sizeof(*hdr));
 525         offset = sizeof(*hdr);
 526         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 527     }
 528
 529     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 530      * but we may be passing along a larger header to the guest.
 531      */
 532     iov[0].iov_base += hdr_len;
 533     iov[0].iov_len  -= hdr_len;
 534
 535     return offset;
 536 }
 537
 538 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 539 {
 540     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 541     static const uint8_t vlan[] = {0x81, 0x00};
 542     uint8_t *ptr = (uint8_t *)buf;
 543     int i;
 544
 545     if (n->promisc)
 546         return 1;
 547
 548     if (n->has_vnet_hdr) {
 549         ptr += sizeof(struct virtio_net_hdr);
 550     }
 551
 552     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 553         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 554         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 555             return 0;
 556     }
 557
 558     if (ptr[0] & 1) { // multicast
 559         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 560             return !n->nobcast;
 561         } else if (n->nomulti) {
 562             return 0;
 563         } else if (n->allmulti || n->mac_table.multi_overflow) {
 564             return 1;
 565         }
 566
 567         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 568             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 569                 return 1;
 570             }
 571         }
 572     } else { // unicast
 573         if (n->nouni) {
 574             return 0;
 575         } else if (n->alluni || n->mac_table.uni_overflow) {
 576             return 1;
 577         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 578             return 1;
 579         }
 580
 581         for (i = 0; i < n->mac_table.first_multi; i++) {
 582             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 583                 return 1;
 584             }
 585         }
 586     }
 587
 588     return 0;
 589 }
 590
 591 static ssize_t virtio_net_receive(VLANClientState *nc, const uint8_t *buf, size_t size)
 592 {
 593     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 594     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 595     size_t guest_hdr_len, offset, i, host_hdr_len;
 596
 597     if (!virtio_net_can_receive(&n->nic->nc))
 598         return -1;
 599
 600     /* hdr_len refers to the header we supply to the guest */
 601     guest_hdr_len = n->mergeable_rx_bufs ?
 602         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 603
 604
 605     host_hdr_len = n->has_vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 606     if (!virtio_net_has_buffers(n, size + guest_hdr_len - host_hdr_len))
 607         return 0;
 608
 609     if (!receive_filter(n, buf, size))
 610         return size;
 611
 612     offset = i = 0;
 613
 614     while (offset < size) {
 615         VirtQueueElement elem;
 616         int len, total;
 617         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 618
 619         total = 0;
 620
 621         if (virtqueue_pop(n->rx_vq, &elem) == 0) {
 622             if (i == 0)
 623                 return -1;
 624             error_report("virtio-net unexpected empty queue: "
 625                     "i %zd mergeable %d offset %zd, size %zd, "
 626                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 627                     i, n->mergeable_rx_bufs, offset, size,
 628                     guest_hdr_len, host_hdr_len, n->vdev.guest_features);
 629             exit(1);
 630         }
 631
 632         if (elem.in_num < 1) {
 633             error_report("virtio-net receive queue contains no in buffers");
 634             exit(1);
 635         }
 636
 637         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != guest_hdr_len) {
 638             error_report("virtio-net header not in first element");
 639             exit(1);
 640         }
 641
 642         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 643
 644         if (i == 0) {
 645             if (n->mergeable_rx_bufs)
 646                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 647
 648             offset += receive_header(n, sg, elem.in_num,
 649                                      buf + offset, size - offset, guest_hdr_len);
 650             total += guest_hdr_len;
 651         }
 652
 653         /* copy in packet.  ugh */
 654         len = iov_from_buf(sg, elem.in_num,
 655                            buf + offset, size - offset);
 656         total += len;
 657         offset += len;
 658         /* If buffers can't be merged, at this point we
 659          * must have consumed the complete packet.
 660          * Otherwise, drop it. */
 661         if (!n->mergeable_rx_bufs && offset < size) {
 662 #if 0
 663             error_report("virtio-net truncated non-mergeable packet: "
 664                          "i %zd mergeable %d offset %zd, size %zd, "
 665                          "guest hdr len %zd, host hdr len %zd",
 666                          i, n->mergeable_rx_bufs,
 667                          offset, size, guest_hdr_len, host_hdr_len);
 668 #endif
 669             return size;
 670         }
 671
 672         /* signal other side */
 673         virtqueue_fill(n->rx_vq, &elem, total, i++);
 674     }
 675
 676     if (mhdr)
 677         mhdr->num_buffers = i;
 678
 679     virtqueue_flush(n->rx_vq, i);
 680     virtio_notify(&n->vdev, n->rx_vq);
 681
 682     return size;
 683 }
 684
 685 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 686
 687 static void virtio_net_tx_complete(VLANClientState *nc, ssize_t len)
 688 {
 689     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 690
 691     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 692     virtio_notify(&n->vdev, n->tx_vq);
 693
 694     n->async_tx.elem.out_num = n->async_tx.len = 0;
 695
 696     virtio_queue_set_notification(n->tx_vq, 1);
 697     virtio_net_flush_tx(n, n->tx_vq);
 698 }
 699
 700 /* TX */
 701 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 702 {
 703     VirtQueueElement elem;
 704     int32_t num_packets = 0;
 705     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 706         return num_packets;
 707     }
 708
 709     assert(n->vdev.vm_running);
 710
 711     if (n->async_tx.elem.out_num) {
 712         virtio_queue_set_notification(n->tx_vq, 0);
 713         return num_packets;
 714     }
 715
 716     while (virtqueue_pop(vq, &elem)) {
 717         ssize_t ret, len = 0;
 718         unsigned int out_num = elem.out_num;
 719         struct iovec *out_sg = &elem.out_sg[0];
 720         unsigned hdr_len;
 721
 722         /* hdr_len refers to the header received from the guest */
 723         hdr_len = n->mergeable_rx_bufs ?
 724             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 725             sizeof(struct virtio_net_hdr);
 726
 727         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 728             error_report("virtio-net header not in first element");
 729             exit(1);
 730         }
 731
 732         /* ignore the header if GSO is not supported */
 733         if (!n->has_vnet_hdr) {
 734             out_num--;
 735             out_sg++;
 736             len += hdr_len;
 737         } else if (n->mergeable_rx_bufs) {
 738             /* tapfd expects a struct virtio_net_hdr */
 739             hdr_len -= sizeof(struct virtio_net_hdr);
 740             out_sg->iov_len -= hdr_len;
 741             len += hdr_len;
 742         }
 743
 744         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 745                                       virtio_net_tx_complete);
 746         if (ret == 0) {
 747             virtio_queue_set_notification(n->tx_vq, 0);
 748             n->async_tx.elem = elem;
 749             n->async_tx.len  = len;
 750             return -EBUSY;
 751         }
 752
 753         len += ret;
 754
 755         virtqueue_push(vq, &elem, len);
 756         virtio_notify(&n->vdev, vq);
 757
 758         if (++num_packets >= n->tx_burst) {
 759             break;
 760         }
 761     }
 762     return num_packets;
 763 }
 764
 765 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 766 {
 767     VirtIONet *n = to_virtio_net(vdev);
 768
 769     /* This happens when device was stopped but VCPU wasn't. */
 770     if (!n->vdev.vm_running) {
 771         n->tx_waiting = 1;
 772         return;
 773     }
 774
 775     if (n->tx_waiting) {
 776         virtio_queue_set_notification(vq, 1);
 777         qemu_del_timer(n->tx_timer);
 778         n->tx_waiting = 0;
 779         virtio_net_flush_tx(n, vq);
 780     } else {
 781         qemu_mod_timer(n->tx_timer,
 782                        qemu_get_clock(vm_clock) + n->tx_timeout);
 783         n->tx_waiting = 1;
 784         virtio_queue_set_notification(vq, 0);
 785     }
 786 }
 787
 788 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
 789 {
 790     VirtIONet *n = to_virtio_net(vdev);
 791
 792     if (unlikely(n->tx_waiting)) {
 793         return;
 794     }
 795     n->tx_waiting = 1;
 796     /* This happens when device was stopped but VCPU wasn't. */
 797     if (!n->vdev.vm_running) {
 798         return;
 799     }
 800     virtio_queue_set_notification(vq, 0);
 801     qemu_bh_schedule(n->tx_bh);
 802 }
 803
 804 static void virtio_net_tx_timer(void *opaque)
 805 {
 806     VirtIONet *n = opaque;
 807     assert(n->vdev.vm_running);
 808
 809     n->tx_waiting = 0;
 810
 811     /* Just in case the driver is not ready on more */
 812     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 813         return;
 814
 815     virtio_queue_set_notification(n->tx_vq, 1);
 816     virtio_net_flush_tx(n, n->tx_vq);
 817 }
 818
 819 static void virtio_net_tx_bh(void *opaque)
 820 {
 821     VirtIONet *n = opaque;
 822     int32_t ret;
 823
 824     assert(n->vdev.vm_running);
 825
 826     n->tx_waiting = 0;
 827
 828     /* Just in case the driver is not ready on more */
 829     if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
 830         return;
 831
 832     ret = virtio_net_flush_tx(n, n->tx_vq);
 833     if (ret == -EBUSY) {
 834         return; /* Notification re-enable handled by tx_complete */
 835     }
 836
 837     /* If we flush a full burst of packets, assume there are
 838      * more coming and immediately reschedule */
 839     if (ret >= n->tx_burst) {
 840         qemu_bh_schedule(n->tx_bh);
 841         n->tx_waiting = 1;
 842         return;
 843     }
 844
 845     /* If less than a full burst, re-enable notification and flush
 846      * anything that may have come in while we weren't looking.  If
 847      * we find something, assume the guest is still active and reschedule */
 848     virtio_queue_set_notification(n->tx_vq, 1);
 849     if (virtio_net_flush_tx(n, n->tx_vq) > 0) {
 850         virtio_queue_set_notification(n->tx_vq, 0);
 851         qemu_bh_schedule(n->tx_bh);
 852         n->tx_waiting = 1;
 853     }
 854 }
 855
 856 static void virtio_net_save(QEMUFile *f, void *opaque)
 857 {
 858     VirtIONet *n = opaque;
 859
 860     /* At this point, backend must be stopped, otherwise
 861      * it might keep writing to memory. */
 862     assert(!n->vhost_started);
 863     virtio_save(&n->vdev, f);
 864
 865     qemu_put_buffer(f, n->mac, ETH_ALEN);
 866     qemu_put_be32(f, n->tx_waiting);
 867     qemu_put_be32(f, n->mergeable_rx_bufs);
 868     qemu_put_be16(f, n->status);
 869     qemu_put_byte(f, n->promisc);
 870     qemu_put_byte(f, n->allmulti);
 871     qemu_put_be32(f, n->mac_table.in_use);
 872     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 873     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 874     qemu_put_be32(f, n->has_vnet_hdr);
 875     qemu_put_byte(f, n->mac_table.multi_overflow);
 876     qemu_put_byte(f, n->mac_table.uni_overflow);
 877     qemu_put_byte(f, n->alluni);
 878     qemu_put_byte(f, n->nomulti);
 879     qemu_put_byte(f, n->nouni);
 880     qemu_put_byte(f, n->nobcast);
 881     qemu_put_byte(f, n->has_ufo);
 882 }
 883
 884 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 885 {
 886     VirtIONet *n = opaque;
 887     int i;
 888
 889     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 890         return -EINVAL;
 891
 892     virtio_load(&n->vdev, f);
 893
 894     qemu_get_buffer(f, n->mac, ETH_ALEN);
 895     n->tx_waiting = qemu_get_be32(f);
 896     n->mergeable_rx_bufs = qemu_get_be32(f);
 897
 898     if (version_id >= 3)
 899         n->status = qemu_get_be16(f);
 900
 901     if (version_id >= 4) {
 902         if (version_id < 8) {
 903             n->promisc = qemu_get_be32(f);
 904             n->allmulti = qemu_get_be32(f);
 905         } else {
 906             n->promisc = qemu_get_byte(f);
 907             n->allmulti = qemu_get_byte(f);
 908         }
 909     }
 910
 911     if (version_id >= 5) {
 912         n->mac_table.in_use = qemu_get_be32(f);
 913         /* MAC_TABLE_ENTRIES may be different from the saved image */
 914         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 915             qemu_get_buffer(f, n->mac_table.macs,
 916                             n->mac_table.in_use * ETH_ALEN);
 917         } else if (n->mac_table.in_use) {
 918             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 919             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 920             n->mac_table.in_use = 0;
 921         }
 922     }
 923
 924     if (version_id >= 6)
 925         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 926
 927     if (version_id >= 7) {
 928         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 929             error_report("virtio-net: saved image requires vnet_hdr=on");
 930             return -1;
 931         }
 932
 933         if (n->has_vnet_hdr) {
 934             tap_using_vnet_hdr(n->nic->nc.peer, 1);
 935             tap_set_offload(n->nic->nc.peer,
 936                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 937                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 938                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 939                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 940                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 941         }
 942     }
 943
 944     if (version_id >= 9) {
 945         n->mac_table.multi_overflow = qemu_get_byte(f);
 946         n->mac_table.uni_overflow = qemu_get_byte(f);
 947     }
 948
 949     if (version_id >= 10) {
 950         n->alluni = qemu_get_byte(f);
 951         n->nomulti = qemu_get_byte(f);
 952         n->nouni = qemu_get_byte(f);
 953         n->nobcast = qemu_get_byte(f);
 954     }
 955
 956     if (version_id >= 11) {
 957         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
 958             error_report("virtio-net: saved image requires TUN_F_UFO support");
 959             return -1;
 960         }
 961     }
 962
 963     /* Find the first multicast entry in the saved MAC filter */
 964     for (i = 0; i < n->mac_table.in_use; i++) {
 965         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 966             break;
 967         }
 968     }
 969     n->mac_table.first_multi = i;
 970     return 0;
 971 }
 972
 973 static void virtio_net_cleanup(VLANClientState *nc)
 974 {
 975     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 976
 977     n->nic = NULL;
 978 }
 979
 980 static NetClientInfo net_virtio_info = {
 981     .type = NET_CLIENT_TYPE_NIC,
 982     .size = sizeof(NICState),
 983     .can_receive = virtio_net_can_receive,
 984     .receive = virtio_net_receive,
 985         .cleanup = virtio_net_cleanup,
 986     .link_status_changed = virtio_net_set_link_status,
 987 };
 988
 989 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
 990                               virtio_net_conf *net)
 991 {
 992     VirtIONet *n;
 993
 994     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 995                                         sizeof(struct virtio_net_config),
 996                                         sizeof(VirtIONet));
 997
 998     n->vdev.get_config = virtio_net_get_config;
 999     n->vdev.set_config = virtio_net_set_config;
1000     n->vdev.get_features = virtio_net_get_features;
1001     n->vdev.set_features = virtio_net_set_features;
1002     n->vdev.bad_features = virtio_net_bad_features;
1003     n->vdev.reset = virtio_net_reset;
1004     n->vdev.set_status = virtio_net_set_status;
1005     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
1006
1007     if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
1008         error_report("virtio-net: "
1009                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1010                      net->tx);
1011         error_report("Defaulting to \"bh\"");
1012     }
1013
1014     if (net->tx && !strcmp(net->tx, "timer")) {
1015         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_timer);
1016         n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
1017         n->tx_timeout = net->txtimer;
1018     } else {
1019         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_bh);
1020         n->tx_bh = qemu_bh_new(virtio_net_tx_bh, n);
1021     }
1022     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
1023     qemu_macaddr_default_if_unset(&conf->macaddr);
1024     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
1025     n->status = VIRTIO_NET_S_LINK_UP;
1026
1027     n->nic = qemu_new_nic(&net_virtio_info, conf, dev->info->name, dev->id, n);
1028
1029     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
1030
1031     n->tx_waiting = 0;
1032     n->tx_burst = net->txburst;
1033     n->mergeable_rx_bufs = 0;
1034     n->promisc = 1; /* for compatibility */
1035
1036     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
1037
1038     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
1039
1040     n->qdev = dev;
1041     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1042                     virtio_net_save, virtio_net_load, n);
1043
1044     add_boot_device_path(conf->bootindex, dev, "/ethernet-phy@0");
1045
1046     return &n->vdev;
1047 }
1048
1049 void virtio_net_exit(VirtIODevice *vdev)
1050 {
1051     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
1052
1053     /* This will stop vhost backend if appropriate. */
1054     virtio_net_set_status(vdev, 0);
1055
1056     qemu_purge_queued_packets(&n->nic->nc);
1057
1058     unregister_savevm(n->qdev, "virtio-net", n);
1059
1060     qemu_free(n->mac_table.macs);
1061     qemu_free(n->vlans);
1062
1063     if (n->tx_timer) {
1064         qemu_del_timer(n->tx_timer);
1065         qemu_free_timer(n->tx_timer);
1066     } else {
1067         qemu_bh_delete(n->tx_bh);
1068     }
1069
1070     virtio_cleanup(&n->vdev);
1071     qemu_del_vlan_client(&n->nic->nc);
1072 }