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 "qemu/iov.h"
  15 #include "virtio.h"
  16 #include "net/net.h"
  17 #include "net/checksum.h"
  18 #include "net/tap.h"
  19 #include "qemu/error-report.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     size_t host_hdr_len;
  45     size_t guest_hdr_len;
  46     uint8_t has_ufo;
  47     struct {
  48         VirtQueueElement elem;
  49         ssize_t len;
  50     } async_tx;
  51     int mergeable_rx_bufs;
  52     uint8_t promisc;
  53     uint8_t allmulti;
  54     uint8_t alluni;
  55     uint8_t nomulti;
  56     uint8_t nouni;
  57     uint8_t nobcast;
  58     uint8_t vhost_started;
  59     struct {
  60         int in_use;
  61         int first_multi;
  62         uint8_t multi_overflow;
  63         uint8_t uni_overflow;
  64         uint8_t *macs;
  65     } mac_table;
  66     uint32_t *vlans;
  67     DeviceState *qdev;
  68 } VirtIONet;
  69
  70 /* TODO
  71  * - we could suppress RX interrupt if we were so inclined.
  72  */
  73
  74 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  75 {
  76     return (VirtIONet *)vdev;
  77 }
  78
  79 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  80 {
  81     VirtIONet *n = to_virtio_net(vdev);
  82     struct virtio_net_config netcfg;
  83
  84     stw_p(&netcfg.status, n->status);
  85     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  86     memcpy(config, &netcfg, sizeof(netcfg));
  87 }
  88
  89 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  90 {
  91     VirtIONet *n = to_virtio_net(vdev);
  92     struct virtio_net_config netcfg;
  93
  94     memcpy(&netcfg, config, sizeof(netcfg));
  95
  96     if (!(n->vdev.guest_features >> VIRTIO_NET_F_CTRL_MAC_ADDR & 1) &&
  97         memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  98         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  99         qemu_format_nic_info_str(&n->nic->nc, n->mac);
 100     }
 101 }
 102
 103 static bool virtio_net_started(VirtIONet *n, uint8_t status)
 104 {
 105     return (status & VIRTIO_CONFIG_S_DRIVER_OK) &&
 106         (n->status & VIRTIO_NET_S_LINK_UP) && n->vdev.vm_running;
 107 }
 108
 109 static void virtio_net_vhost_status(VirtIONet *n, uint8_t status)
 110 {
 111     if (!n->nic->nc.peer) {
 112         return;
 113     }
 114     if (n->nic->nc.peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 115         return;
 116     }
 117
 118     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 119         return;
 120     }
 121     if (!!n->vhost_started == virtio_net_started(n, status) &&
 122                               !n->nic->nc.peer->link_down) {
 123         return;
 124     }
 125     if (!n->vhost_started) {
 126         int r;
 127         if (!vhost_net_query(tap_get_vhost_net(n->nic->nc.peer), &n->vdev)) {
 128             return;
 129         }
 130         n->vhost_started = 1;
 131         r = vhost_net_start(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 132         if (r < 0) {
 133             error_report("unable to start vhost net: %d: "
 134                          "falling back on userspace virtio", -r);
 135             n->vhost_started = 0;
 136         }
 137     } else {
 138         vhost_net_stop(tap_get_vhost_net(n->nic->nc.peer), &n->vdev);
 139         n->vhost_started = 0;
 140     }
 141 }
 142
 143 static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t status)
 144 {
 145     VirtIONet *n = to_virtio_net(vdev);
 146
 147     virtio_net_vhost_status(n, status);
 148
 149     if (!n->tx_waiting) {
 150         return;
 151     }
 152
 153     if (virtio_net_started(n, status) && !n->vhost_started) {
 154         if (n->tx_timer) {
 155             qemu_mod_timer(n->tx_timer,
 156                            qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 157         } else {
 158             qemu_bh_schedule(n->tx_bh);
 159         }
 160     } else {
 161         if (n->tx_timer) {
 162             qemu_del_timer(n->tx_timer);
 163         } else {
 164             qemu_bh_cancel(n->tx_bh);
 165         }
 166     }
 167 }
 168
 169 static void virtio_net_set_link_status(NetClientState *nc)
 170 {
 171     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 172     uint16_t old_status = n->status;
 173
 174     if (nc->link_down)
 175         n->status &= ~VIRTIO_NET_S_LINK_UP;
 176     else
 177         n->status |= VIRTIO_NET_S_LINK_UP;
 178
 179     if (n->status != old_status)
 180         virtio_notify_config(&n->vdev);
 181
 182     virtio_net_set_status(&n->vdev, n->vdev.status);
 183 }
 184
 185 static void virtio_net_reset(VirtIODevice *vdev)
 186 {
 187     VirtIONet *n = to_virtio_net(vdev);
 188
 189     /* Reset back to compatibility mode */
 190     n->promisc = 1;
 191     n->allmulti = 0;
 192     n->alluni = 0;
 193     n->nomulti = 0;
 194     n->nouni = 0;
 195     n->nobcast = 0;
 196
 197     /* Flush any MAC and VLAN filter table state */
 198     n->mac_table.in_use = 0;
 199     n->mac_table.first_multi = 0;
 200     n->mac_table.multi_overflow = 0;
 201     n->mac_table.uni_overflow = 0;
 202     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 203     memcpy(&n->mac[0], &n->nic->conf->macaddr, sizeof(n->mac));
 204     memset(n->vlans, 0, MAX_VLAN >> 3);
 205 }
 206
 207 static void peer_test_vnet_hdr(VirtIONet *n)
 208 {
 209     if (!n->nic->nc.peer)
 210         return;
 211
 212     if (n->nic->nc.peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP)
 213         return;
 214
 215     n->has_vnet_hdr = tap_has_vnet_hdr(n->nic->nc.peer);
 216 }
 217
 218 static int peer_has_vnet_hdr(VirtIONet *n)
 219 {
 220     return n->has_vnet_hdr;
 221 }
 222
 223 static int peer_has_ufo(VirtIONet *n)
 224 {
 225     if (!peer_has_vnet_hdr(n))
 226         return 0;
 227
 228     n->has_ufo = tap_has_ufo(n->nic->nc.peer);
 229
 230     return n->has_ufo;
 231 }
 232
 233 static void virtio_net_set_mrg_rx_bufs(VirtIONet *n, int mergeable_rx_bufs)
 234 {
 235     n->mergeable_rx_bufs = mergeable_rx_bufs;
 236
 237     n->guest_hdr_len = n->mergeable_rx_bufs ?
 238         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 239
 240     if (peer_has_vnet_hdr(n) &&
 241         tap_has_vnet_hdr_len(n->nic->nc.peer, n->guest_hdr_len)) {
 242         tap_set_vnet_hdr_len(n->nic->nc.peer, n->guest_hdr_len);
 243         n->host_hdr_len = n->guest_hdr_len;
 244     }
 245 }
 246
 247 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 248 {
 249     VirtIONet *n = to_virtio_net(vdev);
 250
 251     features |= (1 << VIRTIO_NET_F_MAC);
 252
 253     if (!peer_has_vnet_hdr(n)) {
 254         features &= ~(0x1 << VIRTIO_NET_F_CSUM);
 255         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO4);
 256         features &= ~(0x1 << VIRTIO_NET_F_HOST_TSO6);
 257         features &= ~(0x1 << VIRTIO_NET_F_HOST_ECN);
 258
 259         features &= ~(0x1 << VIRTIO_NET_F_GUEST_CSUM);
 260         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO4);
 261         features &= ~(0x1 << VIRTIO_NET_F_GUEST_TSO6);
 262         features &= ~(0x1 << VIRTIO_NET_F_GUEST_ECN);
 263     }
 264
 265     if (!peer_has_vnet_hdr(n) || !peer_has_ufo(n)) {
 266         features &= ~(0x1 << VIRTIO_NET_F_GUEST_UFO);
 267         features &= ~(0x1 << VIRTIO_NET_F_HOST_UFO);
 268     }
 269
 270     if (!n->nic->nc.peer ||
 271         n->nic->nc.peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 272         return features;
 273     }
 274     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 275         return features;
 276     }
 277     return vhost_net_get_features(tap_get_vhost_net(n->nic->nc.peer), features);
 278 }
 279
 280 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 281 {
 282     uint32_t features = 0;
 283
 284     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 285      * but also these: */
 286     features |= (1 << VIRTIO_NET_F_MAC);
 287     features |= (1 << VIRTIO_NET_F_CSUM);
 288     features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 289     features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 290     features |= (1 << VIRTIO_NET_F_HOST_ECN);
 291
 292     return features;
 293 }
 294
 295 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 296 {
 297     VirtIONet *n = to_virtio_net(vdev);
 298
 299     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
 300
 301     if (n->has_vnet_hdr) {
 302         tap_set_offload(n->nic->nc.peer,
 303                         (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 304                         (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 305                         (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 306                         (features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 307                         (features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 308     }
 309     if (!n->nic->nc.peer ||
 310         n->nic->nc.peer->info->type != NET_CLIENT_OPTIONS_KIND_TAP) {
 311         return;
 312     }
 313     if (!tap_get_vhost_net(n->nic->nc.peer)) {
 314         return;
 315     }
 316     vhost_net_ack_features(tap_get_vhost_net(n->nic->nc.peer), features);
 317 }
 318
 319 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 320                                      struct iovec *iov, unsigned int iov_cnt)
 321 {
 322     uint8_t on;
 323     size_t s;
 324
 325     s = iov_to_buf(iov, iov_cnt, 0, &on, sizeof(on));
 326     if (s != sizeof(on)) {
 327         return VIRTIO_NET_ERR;
 328     }
 329
 330     if (cmd == VIRTIO_NET_CTRL_RX_PROMISC) {
 331         n->promisc = on;
 332     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLMULTI) {
 333         n->allmulti = on;
 334     } else if (cmd == VIRTIO_NET_CTRL_RX_ALLUNI) {
 335         n->alluni = on;
 336     } else if (cmd == VIRTIO_NET_CTRL_RX_NOMULTI) {
 337         n->nomulti = on;
 338     } else if (cmd == VIRTIO_NET_CTRL_RX_NOUNI) {
 339         n->nouni = on;
 340     } else if (cmd == VIRTIO_NET_CTRL_RX_NOBCAST) {
 341         n->nobcast = on;
 342     } else {
 343         return VIRTIO_NET_ERR;
 344     }
 345
 346     return VIRTIO_NET_OK;
 347 }
 348
 349 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 350                                  struct iovec *iov, unsigned int iov_cnt)
 351 {
 352     struct virtio_net_ctrl_mac mac_data;
 353     size_t s;
 354
 355     if (cmd == VIRTIO_NET_CTRL_MAC_ADDR_SET) {
 356         if (iov_size(iov, iov_cnt) != sizeof(n->mac)) {
 357             return VIRTIO_NET_ERR;
 358         }
 359         s = iov_to_buf(iov, iov_cnt, 0, &n->mac, sizeof(n->mac));
 360         assert(s == sizeof(n->mac));
 361         qemu_format_nic_info_str(&n->nic->nc, n->mac);
 362         return VIRTIO_NET_OK;
 363     }
 364
 365     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET) {
 366         return VIRTIO_NET_ERR;
 367     }
 368
 369     n->mac_table.in_use = 0;
 370     n->mac_table.first_multi = 0;
 371     n->mac_table.uni_overflow = 0;
 372     n->mac_table.multi_overflow = 0;
 373     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 374
 375     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 376                    sizeof(mac_data.entries));
 377     mac_data.entries = ldl_p(&mac_data.entries);
 378     if (s != sizeof(mac_data.entries)) {
 379         return VIRTIO_NET_ERR;
 380     }
 381     iov_discard_front(&iov, &iov_cnt, s);
 382
 383     if (mac_data.entries * ETH_ALEN > iov_size(iov, iov_cnt)) {
 384         return VIRTIO_NET_ERR;
 385     }
 386
 387     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 388         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 389                        mac_data.entries * ETH_ALEN);
 390         if (s != mac_data.entries * ETH_ALEN) {
 391             return VIRTIO_NET_ERR;
 392         }
 393         n->mac_table.in_use += mac_data.entries;
 394     } else {
 395         n->mac_table.uni_overflow = 1;
 396     }
 397
 398     iov_discard_front(&iov, &iov_cnt, mac_data.entries * ETH_ALEN);
 399
 400     n->mac_table.first_multi = n->mac_table.in_use;
 401
 402     s = iov_to_buf(iov, iov_cnt, 0, &mac_data.entries,
 403                    sizeof(mac_data.entries));
 404     mac_data.entries = ldl_p(&mac_data.entries);
 405     if (s != sizeof(mac_data.entries)) {
 406         return VIRTIO_NET_ERR;
 407     }
 408
 409     iov_discard_front(&iov, &iov_cnt, s);
 410
 411     if (mac_data.entries * ETH_ALEN != iov_size(iov, iov_cnt)) {
 412         return VIRTIO_NET_ERR;
 413     }
 414
 415     if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 416         s = iov_to_buf(iov, iov_cnt, 0, n->mac_table.macs,
 417                        mac_data.entries * ETH_ALEN);
 418         if (s != mac_data.entries * ETH_ALEN) {
 419             return VIRTIO_NET_ERR;
 420         }
 421         n->mac_table.in_use += mac_data.entries;
 422     } else {
 423         n->mac_table.multi_overflow = 1;
 424     }
 425
 426     return VIRTIO_NET_OK;
 427 }
 428
 429 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 430                                         struct iovec *iov, unsigned int iov_cnt)
 431 {
 432     uint16_t vid;
 433     size_t s;
 434
 435     s = iov_to_buf(iov, iov_cnt, 0, &vid, sizeof(vid));
 436     vid = lduw_p(&vid);
 437     if (s != sizeof(vid)) {
 438         return VIRTIO_NET_ERR;
 439     }
 440
 441     if (vid >= MAX_VLAN)
 442         return VIRTIO_NET_ERR;
 443
 444     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 445         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 446     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 447         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 448     else
 449         return VIRTIO_NET_ERR;
 450
 451     return VIRTIO_NET_OK;
 452 }
 453
 454 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 455 {
 456     VirtIONet *n = to_virtio_net(vdev);
 457     struct virtio_net_ctrl_hdr ctrl;
 458     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 459     VirtQueueElement elem;
 460     size_t s;
 461     struct iovec *iov;
 462     unsigned int iov_cnt;
 463
 464     while (virtqueue_pop(vq, &elem)) {
 465         if (iov_size(elem.in_sg, elem.in_num) < sizeof(status) ||
 466             iov_size(elem.out_sg, elem.out_num) < sizeof(ctrl)) {
 467             error_report("virtio-net ctrl missing headers");
 468             exit(1);
 469         }
 470
 471         iov = elem.out_sg;
 472         iov_cnt = elem.out_num;
 473         s = iov_to_buf(iov, iov_cnt, 0, &ctrl, sizeof(ctrl));
 474         iov_discard_front(&iov, &iov_cnt, sizeof(ctrl));
 475         if (s != sizeof(ctrl)) {
 476             status = VIRTIO_NET_ERR;
 477         } else if (ctrl.class == VIRTIO_NET_CTRL_RX) {
 478             status = virtio_net_handle_rx_mode(n, ctrl.cmd, iov, iov_cnt);
 479         } else if (ctrl.class == VIRTIO_NET_CTRL_MAC) {
 480             status = virtio_net_handle_mac(n, ctrl.cmd, iov, iov_cnt);
 481         } else if (ctrl.class == VIRTIO_NET_CTRL_VLAN) {
 482             status = virtio_net_handle_vlan_table(n, ctrl.cmd, iov, iov_cnt);
 483         }
 484
 485         s = iov_from_buf(elem.in_sg, elem.in_num, 0, &status, sizeof(status));
 486         assert(s == sizeof(status));
 487
 488         virtqueue_push(vq, &elem, sizeof(status));
 489         virtio_notify(vdev, vq);
 490     }
 491 }
 492
 493 /* RX */
 494
 495 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 496 {
 497     VirtIONet *n = to_virtio_net(vdev);
 498
 499     qemu_flush_queued_packets(&n->nic->nc);
 500 }
 501
 502 static int virtio_net_can_receive(NetClientState *nc)
 503 {
 504     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 505     if (!n->vdev.vm_running) {
 506         return 0;
 507     }
 508
 509     if (!virtio_queue_ready(n->rx_vq) ||
 510         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 511         return 0;
 512
 513     return 1;
 514 }
 515
 516 static int virtio_net_has_buffers(VirtIONet *n, int bufsize)
 517 {
 518     if (virtio_queue_empty(n->rx_vq) ||
 519         (n->mergeable_rx_bufs &&
 520          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 521         virtio_queue_set_notification(n->rx_vq, 1);
 522
 523         /* To avoid a race condition where the guest has made some buffers
 524          * available after the above check but before notification was
 525          * enabled, check for available buffers again.
 526          */
 527         if (virtio_queue_empty(n->rx_vq) ||
 528             (n->mergeable_rx_bufs &&
 529              !virtqueue_avail_bytes(n->rx_vq, bufsize, 0)))
 530             return 0;
 531     }
 532
 533     virtio_queue_set_notification(n->rx_vq, 0);
 534     return 1;
 535 }
 536
 537 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 538  * it never finds out that the packets don't have valid checksums.  This
 539  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 540  * fix this with Xen but it hasn't appeared in an upstream release of
 541  * dhclient yet.
 542  *
 543  * To avoid breaking existing guests, we catch udp packets and add
 544  * checksums.  This is terrible but it's better than hacking the guest
 545  * kernels.
 546  *
 547  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 548  * we should provide a mechanism to disable it to avoid polluting the host
 549  * cache.
 550  */
 551 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 552                                         uint8_t *buf, size_t size)
 553 {
 554     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 555         (size > 27 && size < 1500) && /* normal sized MTU */
 556         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 557         (buf[23] == 17) && /* ip.protocol == UDP */
 558         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 559         net_checksum_calculate(buf, size);
 560         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 561     }
 562 }
 563
 564 static void receive_header(VirtIONet *n, const struct iovec *iov, int iov_cnt,
 565                            const void *buf, size_t size)
 566 {
 567     if (n->has_vnet_hdr) {
 568         /* FIXME this cast is evil */
 569         void *wbuf = (void *)buf;
 570         work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len,
 571                                     size - n->host_hdr_len);
 572         iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct virtio_net_hdr));
 573     } else {
 574         struct virtio_net_hdr hdr = {
 575             .flags = 0,
 576             .gso_type = VIRTIO_NET_HDR_GSO_NONE
 577         };
 578         iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
 579     }
 580 }
 581
 582 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 583 {
 584     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 585     static const uint8_t vlan[] = {0x81, 0x00};
 586     uint8_t *ptr = (uint8_t *)buf;
 587     int i;
 588
 589     if (n->promisc)
 590         return 1;
 591
 592     ptr += n->host_hdr_len;
 593
 594     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 595         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 596         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 597             return 0;
 598     }
 599
 600     if (ptr[0] & 1) { // multicast
 601         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 602             return !n->nobcast;
 603         } else if (n->nomulti) {
 604             return 0;
 605         } else if (n->allmulti || n->mac_table.multi_overflow) {
 606             return 1;
 607         }
 608
 609         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 610             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 611                 return 1;
 612             }
 613         }
 614     } else { // unicast
 615         if (n->nouni) {
 616             return 0;
 617         } else if (n->alluni || n->mac_table.uni_overflow) {
 618             return 1;
 619         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 620             return 1;
 621         }
 622
 623         for (i = 0; i < n->mac_table.first_multi; i++) {
 624             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 625                 return 1;
 626             }
 627         }
 628     }
 629
 630     return 0;
 631 }
 632
 633 static ssize_t virtio_net_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 634 {
 635     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 636     struct iovec mhdr_sg[VIRTQUEUE_MAX_SIZE];
 637     struct virtio_net_hdr_mrg_rxbuf mhdr;
 638     unsigned mhdr_cnt = 0;
 639     size_t offset, i, guest_offset;
 640
 641     if (!virtio_net_can_receive(&n->nic->nc))
 642         return -1;
 643
 644     /* hdr_len refers to the header we supply to the guest */
 645     if (!virtio_net_has_buffers(n, size + n->guest_hdr_len - n->host_hdr_len))
 646         return 0;
 647
 648     if (!receive_filter(n, buf, size))
 649         return size;
 650
 651     offset = i = 0;
 652
 653     while (offset < size) {
 654         VirtQueueElement elem;
 655         int len, total;
 656         const struct iovec *sg = elem.in_sg;
 657
 658         total = 0;
 659
 660         if (virtqueue_pop(n->rx_vq, &elem) == 0) {
 661             if (i == 0)
 662                 return -1;
 663             error_report("virtio-net unexpected empty queue: "
 664                     "i %zd mergeable %d offset %zd, size %zd, "
 665                     "guest hdr len %zd, host hdr len %zd guest features 0x%x",
 666                     i, n->mergeable_rx_bufs, offset, size,
 667                     n->guest_hdr_len, n->host_hdr_len, n->vdev.guest_features);
 668             exit(1);
 669         }
 670
 671         if (elem.in_num < 1) {
 672             error_report("virtio-net receive queue contains no in buffers");
 673             exit(1);
 674         }
 675
 676         if (i == 0) {
 677             assert(offset == 0);
 678             if (n->mergeable_rx_bufs) {
 679                 mhdr_cnt = iov_copy(mhdr_sg, ARRAY_SIZE(mhdr_sg),
 680                                     sg, elem.in_num,
 681                                     offsetof(typeof(mhdr), num_buffers),
 682                                     sizeof(mhdr.num_buffers));
 683             }
 684
 685             receive_header(n, sg, elem.in_num, buf, size);
 686             offset = n->host_hdr_len;
 687             total += n->guest_hdr_len;
 688             guest_offset = n->guest_hdr_len;
 689         } else {
 690             guest_offset = 0;
 691         }
 692
 693         /* copy in packet.  ugh */
 694         len = iov_from_buf(sg, elem.in_num, guest_offset,
 695                            buf + offset, size - offset);
 696         total += len;
 697         offset += len;
 698         /* If buffers can't be merged, at this point we
 699          * must have consumed the complete packet.
 700          * Otherwise, drop it. */
 701         if (!n->mergeable_rx_bufs && offset < size) {
 702 #if 0
 703             error_report("virtio-net truncated non-mergeable packet: "
 704                          "i %zd mergeable %d offset %zd, size %zd, "
 705                          "guest hdr len %zd, host hdr len %zd",
 706                          i, n->mergeable_rx_bufs,
 707                          offset, size, n->guest_hdr_len, n->host_hdr_len);
 708 #endif
 709             return size;
 710         }
 711
 712         /* signal other side */
 713         virtqueue_fill(n->rx_vq, &elem, total, i++);
 714     }
 715
 716     if (mhdr_cnt) {
 717         stw_p(&mhdr.num_buffers, i);
 718         iov_from_buf(mhdr_sg, mhdr_cnt,
 719                      0,
 720                      &mhdr.num_buffers, sizeof mhdr.num_buffers);
 721     }
 722
 723     virtqueue_flush(n->rx_vq, i);
 724     virtio_notify(&n->vdev, n->rx_vq);
 725
 726     return size;
 727 }
 728
 729 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 730
 731 static void virtio_net_tx_complete(NetClientState *nc, ssize_t len)
 732 {
 733     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
 734
 735     virtqueue_push(n->tx_vq, &n->async_tx.elem, 0);
 736     virtio_notify(&n->vdev, n->tx_vq);
 737
 738     n->async_tx.elem.out_num = n->async_tx.len = 0;
 739
 740     virtio_queue_set_notification(n->tx_vq, 1);
 741     virtio_net_flush_tx(n, n->tx_vq);
 742 }
 743
 744 /* TX */
 745 static int32_t virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 746 {
 747     VirtQueueElement elem;
 748     int32_t num_packets = 0;
 749     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)) {
 750         return num_packets;
 751     }
 752
 753     assert(n->vdev.vm_running);
 754
 755     if (n->async_tx.elem.out_num) {
 756         virtio_queue_set_notification(n->tx_vq, 0);
 757         return num_packets;
 758     }
 759
 760     while (virtqueue_pop(vq, &elem)) {
 761         ssize_t ret, len;
 762         unsigned int out_num = elem.out_num;
 763         struct iovec *out_sg = &elem.out_sg[0];
 764         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 765
 766         if (out_num < 1) {
 767             error_report("virtio-net header not in first element");
 768             exit(1);
 769         }
 770
 771         /*
 772          * If host wants to see the guest header as is, we can
 773          * pass it on unchanged. Otherwise, copy just the parts
 774          * that host is interested in.
 775          */
 776         assert(n->host_hdr_len <= n->guest_hdr_len);
 777         if (n->host_hdr_len != n->guest_hdr_len) {
 778             unsigned sg_num = iov_copy(sg, ARRAY_SIZE(sg),
 779                                        out_sg, out_num,
 780                                        0, n->host_hdr_len);
 781             sg_num += iov_copy(sg + sg_num, ARRAY_SIZE(sg) - sg_num,
 782                              out_sg, out_num,
 783                              n->guest_hdr_len, -1);
 784             out_num = sg_num;
 785             out_sg = sg;
 786         }
 787
 788         len = n->guest_hdr_len;
 789
 790         ret = qemu_sendv_packet_async(&n->nic->nc, out_sg, out_num,
 791                                       virtio_net_tx_complete);
 792         if (ret == 0) {
 793             virtio_queue_set_notification(n->tx_vq, 0);
 794             n->async_tx.elem = elem;
 795             n->async_tx.len  = len;
 796             return -EBUSY;
 797         }
 798
 799         len += ret;
 800
 801         virtqueue_push(vq, &elem, 0);
 802         virtio_notify(&n->vdev, vq);
 803
 804         if (++num_packets >= n->tx_burst) {
 805             break;
 806         }
 807     }
 808     return num_packets;
 809 }
 810
 811 static void virtio_net_handle_tx_timer(VirtIODevice *vdev, VirtQueue *vq)
 812 {
 813     VirtIONet *n = to_virtio_net(vdev);
 814
 815     /* This happens when device was stopped but VCPU wasn't. */
 816     if (!n->vdev.vm_running) {
 817         n->tx_waiting = 1;
 818         return;
 819     }
 820
 821     if (n->tx_waiting) {
 822         virtio_queue_set_notification(vq, 1);
 823         qemu_del_timer(n->tx_timer);
 824         n->tx_waiting = 0;
 825         virtio_net_flush_tx(n, vq);
 826     } else {
 827         qemu_mod_timer(n->tx_timer,
 828                        qemu_get_clock_ns(vm_clock) + n->tx_timeout);
 829         n->tx_waiting = 1;
 830         virtio_queue_set_notification(vq, 0);
 831     }
 832 }
 833
 834 static void virtio_net_handle_tx_bh(VirtIODevice *vdev, VirtQueue *vq)
 835 {
 836     VirtIONet *n = to_virtio_net(vdev);
 837
 838     if (unlikely(n->tx_waiting)) {
 839         return;
 840     }
 841     n->tx_waiting = 1;
 842     /* This happens when device was stopped but VCPU wasn't. */
 843     if (!n->vdev.vm_running) {
 844         return;
 845     }
 846     virtio_queue_set_notification(vq, 0);
 847     qemu_bh_schedule(n->tx_bh);
 848 }
 849
 850 static void virtio_net_tx_timer(void *opaque)
 851 {
 852     VirtIONet *n = opaque;
 853     assert(n->vdev.vm_running);
 854
 855     n->tx_waiting = 0;
 856
 857     /* Just in case the driver is not ready on more */
 858     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 859         return;
 860
 861     virtio_queue_set_notification(n->tx_vq, 1);
 862     virtio_net_flush_tx(n, n->tx_vq);
 863 }
 864
 865 static void virtio_net_tx_bh(void *opaque)
 866 {
 867     VirtIONet *n = opaque;
 868     int32_t ret;
 869
 870     assert(n->vdev.vm_running);
 871
 872     n->tx_waiting = 0;
 873
 874     /* Just in case the driver is not ready on more */
 875     if (unlikely(!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK)))
 876         return;
 877
 878     ret = virtio_net_flush_tx(n, n->tx_vq);
 879     if (ret == -EBUSY) {
 880         return; /* Notification re-enable handled by tx_complete */
 881     }
 882
 883     /* If we flush a full burst of packets, assume there are
 884      * more coming and immediately reschedule */
 885     if (ret >= n->tx_burst) {
 886         qemu_bh_schedule(n->tx_bh);
 887         n->tx_waiting = 1;
 888         return;
 889     }
 890
 891     /* If less than a full burst, re-enable notification and flush
 892      * anything that may have come in while we weren't looking.  If
 893      * we find something, assume the guest is still active and reschedule */
 894     virtio_queue_set_notification(n->tx_vq, 1);
 895     if (virtio_net_flush_tx(n, n->tx_vq) > 0) {
 896         virtio_queue_set_notification(n->tx_vq, 0);
 897         qemu_bh_schedule(n->tx_bh);
 898         n->tx_waiting = 1;
 899     }
 900 }
 901
 902 static void virtio_net_save(QEMUFile *f, void *opaque)
 903 {
 904     VirtIONet *n = opaque;
 905
 906     /* At this point, backend must be stopped, otherwise
 907      * it might keep writing to memory. */
 908     assert(!n->vhost_started);
 909     virtio_save(&n->vdev, f);
 910
 911     qemu_put_buffer(f, n->mac, ETH_ALEN);
 912     qemu_put_be32(f, n->tx_waiting);
 913     qemu_put_be32(f, n->mergeable_rx_bufs);
 914     qemu_put_be16(f, n->status);
 915     qemu_put_byte(f, n->promisc);
 916     qemu_put_byte(f, n->allmulti);
 917     qemu_put_be32(f, n->mac_table.in_use);
 918     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 919     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 920     qemu_put_be32(f, n->has_vnet_hdr);
 921     qemu_put_byte(f, n->mac_table.multi_overflow);
 922     qemu_put_byte(f, n->mac_table.uni_overflow);
 923     qemu_put_byte(f, n->alluni);
 924     qemu_put_byte(f, n->nomulti);
 925     qemu_put_byte(f, n->nouni);
 926     qemu_put_byte(f, n->nobcast);
 927     qemu_put_byte(f, n->has_ufo);
 928 }
 929
 930 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 931 {
 932     VirtIONet *n = opaque;
 933     int i;
 934     int ret;
 935
 936     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 937         return -EINVAL;
 938
 939     ret = virtio_load(&n->vdev, f);
 940     if (ret) {
 941         return ret;
 942     }
 943
 944     qemu_get_buffer(f, n->mac, ETH_ALEN);
 945     n->tx_waiting = qemu_get_be32(f);
 946
 947     virtio_net_set_mrg_rx_bufs(n, qemu_get_be32(f));
 948
 949     if (version_id >= 3)
 950         n->status = qemu_get_be16(f);
 951
 952     if (version_id >= 4) {
 953         if (version_id < 8) {
 954             n->promisc = qemu_get_be32(f);
 955             n->allmulti = qemu_get_be32(f);
 956         } else {
 957             n->promisc = qemu_get_byte(f);
 958             n->allmulti = qemu_get_byte(f);
 959         }
 960     }
 961
 962     if (version_id >= 5) {
 963         n->mac_table.in_use = qemu_get_be32(f);
 964         /* MAC_TABLE_ENTRIES may be different from the saved image */
 965         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 966             qemu_get_buffer(f, n->mac_table.macs,
 967                             n->mac_table.in_use * ETH_ALEN);
 968         } else if (n->mac_table.in_use) {
 969             uint8_t *buf = g_malloc0(n->mac_table.in_use);
 970             qemu_get_buffer(f, buf, n->mac_table.in_use * ETH_ALEN);
 971             g_free(buf);
 972             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 973             n->mac_table.in_use = 0;
 974         }
 975     }
 976
 977     if (version_id >= 6)
 978         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 979
 980     if (version_id >= 7) {
 981         if (qemu_get_be32(f) && !peer_has_vnet_hdr(n)) {
 982             error_report("virtio-net: saved image requires vnet_hdr=on");
 983             return -1;
 984         }
 985
 986         if (n->has_vnet_hdr) {
 987             tap_set_offload(n->nic->nc.peer,
 988                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 989                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 990                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 991                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_ECN)  & 1,
 992                     (n->vdev.guest_features >> VIRTIO_NET_F_GUEST_UFO)  & 1);
 993         }
 994     }
 995
 996     if (version_id >= 9) {
 997         n->mac_table.multi_overflow = qemu_get_byte(f);
 998         n->mac_table.uni_overflow = qemu_get_byte(f);
 999     }
1000
1001     if (version_id >= 10) {
1002         n->alluni = qemu_get_byte(f);
1003         n->nomulti = qemu_get_byte(f);
1004         n->nouni = qemu_get_byte(f);
1005         n->nobcast = qemu_get_byte(f);
1006     }
1007
1008     if (version_id >= 11) {
1009         if (qemu_get_byte(f) && !peer_has_ufo(n)) {
1010             error_report("virtio-net: saved image requires TUN_F_UFO support");
1011             return -1;
1012         }
1013     }
1014
1015     /* Find the first multicast entry in the saved MAC filter */
1016     for (i = 0; i < n->mac_table.in_use; i++) {
1017         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
1018             break;
1019         }
1020     }
1021     n->mac_table.first_multi = i;
1022
1023     /* nc.link_down can't be migrated, so infer link_down according
1024      * to link status bit in n->status */
1025     n->nic->nc.link_down = (n->status & VIRTIO_NET_S_LINK_UP) == 0;
1026
1027     return 0;
1028 }
1029
1030 static void virtio_net_cleanup(NetClientState *nc)
1031 {
1032     VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
1033
1034     n->nic = NULL;
1035 }
1036
1037 static NetClientInfo net_virtio_info = {
1038     .type = NET_CLIENT_OPTIONS_KIND_NIC,
1039     .size = sizeof(NICState),
1040     .can_receive = virtio_net_can_receive,
1041     .receive = virtio_net_receive,
1042         .cleanup = virtio_net_cleanup,
1043     .link_status_changed = virtio_net_set_link_status,
1044 };
1045
1046 static bool virtio_net_guest_notifier_pending(VirtIODevice *vdev, int idx)
1047 {
1048     VirtIONet *n = to_virtio_net(vdev);
1049     assert(n->vhost_started);
1050     return vhost_net_virtqueue_pending(tap_get_vhost_net(n->nic->nc.peer), idx);
1051 }
1052
1053 static void virtio_net_guest_notifier_mask(VirtIODevice *vdev, int idx,
1054                                            bool mask)
1055 {
1056     VirtIONet *n = to_virtio_net(vdev);
1057     assert(n->vhost_started);
1058     vhost_net_virtqueue_mask(tap_get_vhost_net(n->nic->nc.peer),
1059                              vdev, idx, mask);
1060 }
1061
1062 VirtIODevice *virtio_net_init(DeviceState *dev, NICConf *conf,
1063                               virtio_net_conf *net)
1064 {
1065     VirtIONet *n;
1066
1067     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
1068                                         sizeof(struct virtio_net_config),
1069                                         sizeof(VirtIONet));
1070
1071     n->vdev.get_config = virtio_net_get_config;
1072     n->vdev.set_config = virtio_net_set_config;
1073     n->vdev.get_features = virtio_net_get_features;
1074     n->vdev.set_features = virtio_net_set_features;
1075     n->vdev.bad_features = virtio_net_bad_features;
1076     n->vdev.reset = virtio_net_reset;
1077     n->vdev.set_status = virtio_net_set_status;
1078     n->vdev.guest_notifier_mask = virtio_net_guest_notifier_mask;
1079     n->vdev.guest_notifier_pending = virtio_net_guest_notifier_pending;
1080     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
1081
1082     if (net->tx && strcmp(net->tx, "timer") && strcmp(net->tx, "bh")) {
1083         error_report("virtio-net: "
1084                      "Unknown option tx=%s, valid options: \"timer\" \"bh\"",
1085                      net->tx);
1086         error_report("Defaulting to \"bh\"");
1087     }
1088
1089     if (net->tx && !strcmp(net->tx, "timer")) {
1090         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_timer);
1091         n->tx_timer = qemu_new_timer_ns(vm_clock, virtio_net_tx_timer, n);
1092         n->tx_timeout = net->txtimer;
1093     } else {
1094         n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx_bh);
1095         n->tx_bh = qemu_bh_new(virtio_net_tx_bh, n);
1096     }
1097     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
1098     qemu_macaddr_default_if_unset(&conf->macaddr);
1099     memcpy(&n->mac[0], &conf->macaddr, sizeof(n->mac));
1100     n->status = VIRTIO_NET_S_LINK_UP;
1101
1102     n->nic = qemu_new_nic(&net_virtio_info, conf, object_get_typename(OBJECT(dev)), dev->id, n);
1103     peer_test_vnet_hdr(n);
1104     if (peer_has_vnet_hdr(n)) {
1105         tap_using_vnet_hdr(n->nic->nc.peer, 1);
1106         n->host_hdr_len = sizeof(struct virtio_net_hdr);
1107     } else {
1108         n->host_hdr_len = 0;
1109     }
1110
1111     qemu_format_nic_info_str(&n->nic->nc, conf->macaddr.a);
1112
1113     n->tx_waiting = 0;
1114     n->tx_burst = net->txburst;
1115     virtio_net_set_mrg_rx_bufs(n, 0);
1116     n->promisc = 1; /* for compatibility */
1117
1118     n->mac_table.macs = g_malloc0(MAC_TABLE_ENTRIES * ETH_ALEN);
1119
1120     n->vlans = g_malloc0(MAX_VLAN >> 3);
1121
1122     n->qdev = dev;
1123     register_savevm(dev, "virtio-net", -1, VIRTIO_NET_VM_VERSION,
1124                     virtio_net_save, virtio_net_load, n);
1125
1126     add_boot_device_path(conf->bootindex, dev, "/ethernet-phy@0");
1127
1128     return &n->vdev;
1129 }
1130
1131 void virtio_net_exit(VirtIODevice *vdev)
1132 {
1133     VirtIONet *n = DO_UPCAST(VirtIONet, vdev, vdev);
1134
1135     /* This will stop vhost backend if appropriate. */
1136     virtio_net_set_status(vdev, 0);
1137
1138     qemu_purge_queued_packets(&n->nic->nc);
1139
1140     unregister_savevm(n->qdev, "virtio-net", n);
1141
1142     g_free(n->mac_table.macs);
1143     g_free(n->vlans);
1144
1145     if (n->tx_timer) {
1146         qemu_del_timer(n->tx_timer);
1147         qemu_free_timer(n->tx_timer);
1148     } else {
1149         qemu_bh_delete(n->tx_bh);
1150     }
1151
1152     qemu_del_net_client(&n->nic->nc);
1153     virtio_cleanup(&n->vdev);
1154 }