hw/virtio-net.c

   1 /*
   2  * Virtio Network Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "virtio.h"
  15 #include "net.h"
  16 #include "qemu-timer.h"
  17 #include "virtio-net.h"
  18
  19 #define VIRTIO_NET_VM_VERSION    6
  20
  21 #define MAC_TABLE_ENTRIES    32
  22 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  23
  24 typedef struct VirtIONet
  25 {
  26     VirtIODevice vdev;
  27     uint8_t mac[ETH_ALEN];
  28     uint16_t status;
  29     VirtQueue *rx_vq;
  30     VirtQueue *tx_vq;
  31     VirtQueue *ctrl_vq;
  32     VLANClientState *vc;
  33     QEMUTimer *tx_timer;
  34     int tx_timer_active;
  35     int mergeable_rx_bufs;
  36     int promisc;
  37     int allmulti;
  38     struct {
  39         int in_use;
  40         uint8_t *macs;
  41     } mac_table;
  42     uint32_t *vlans;
  43 } VirtIONet;
  44
  45 /* TODO
  46  * - we could suppress RX interrupt if we were so inclined.
  47  */
  48
  49 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  50 {
  51     return (VirtIONet *)vdev;
  52 }
  53
  54 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  55 {
  56     VirtIONet *n = to_virtio_net(vdev);
  57     struct virtio_net_config netcfg;
  58
  59     netcfg.status = n->status;
  60     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  61     memcpy(config, &netcfg, sizeof(netcfg));
  62 }
  63
  64 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  65 {
  66     VirtIONet *n = to_virtio_net(vdev);
  67     struct virtio_net_config netcfg;
  68
  69     memcpy(&netcfg, config, sizeof(netcfg));
  70
  71     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  72         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  73         qemu_format_nic_info_str(n->vc, n->mac);
  74     }
  75 }
  76
  77 static void virtio_net_set_link_status(VLANClientState *vc)
  78 {
  79     VirtIONet *n = vc->opaque;
  80     uint16_t old_status = n->status;
  81
  82     if (vc->link_down)
  83         n->status &= ~VIRTIO_NET_S_LINK_UP;
  84     else
  85         n->status |= VIRTIO_NET_S_LINK_UP;
  86
  87     if (n->status != old_status)
  88         virtio_notify_config(&n->vdev);
  89 }
  90
  91 static void virtio_net_reset(VirtIODevice *vdev)
  92 {
  93     VirtIONet *n = to_virtio_net(vdev);
  94
  95     /* Reset back to compatibility mode */
  96     n->promisc = 1;
  97     n->allmulti = 0;
  98
  99     /* Flush any MAC and VLAN filter table state */
 100     n->mac_table.in_use = 0;
 101     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 102     memset(n->vlans, 0, MAX_VLAN >> 3);
 103 }
 104
 105 static uint32_t virtio_net_get_features(VirtIODevice *vdev)
 106 {
 107     uint32_t features = (1 << VIRTIO_NET_F_MAC) |
 108                         (1 << VIRTIO_NET_F_STATUS) |
 109                         (1 << VIRTIO_NET_F_CTRL_VQ) |
 110                         (1 << VIRTIO_NET_F_CTRL_RX) |
 111                         (1 << VIRTIO_NET_F_CTRL_VLAN);
 112
 113     return features;
 114 }
 115
 116 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 117 {
 118     VirtIONet *n = to_virtio_net(vdev);
 119
 120     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 121 }
 122
 123 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 124                                      VirtQueueElement *elem)
 125 {
 126     uint8_t on;
 127
 128     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 129         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 130         exit(1);
 131     }
 132
 133     on = ldub_p(elem->out_sg[1].iov_base);
 134
 135     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 136         n->promisc = on;
 137     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 138         n->allmulti = on;
 139     else
 140         return VIRTIO_NET_ERR;
 141
 142     return VIRTIO_NET_OK;
 143 }
 144
 145 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 146                                  VirtQueueElement *elem)
 147 {
 148     struct virtio_net_ctrl_mac mac_data;
 149
 150     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 151         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 152         elem->out_sg[2].iov_len < sizeof(mac_data))
 153         return VIRTIO_NET_ERR;
 154
 155     n->mac_table.in_use = 0;
 156     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 157
 158     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 159
 160     if (sizeof(mac_data.entries) +
 161         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 162         return VIRTIO_NET_ERR;
 163
 164     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 165         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 166                mac_data.entries * ETH_ALEN);
 167         n->mac_table.in_use += mac_data.entries;
 168     } else {
 169         n->promisc = 1;
 170         return VIRTIO_NET_OK;
 171     }
 172
 173     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 174
 175     if (sizeof(mac_data.entries) +
 176         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 177         return VIRTIO_NET_ERR;
 178
 179     if (mac_data.entries) {
 180         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 181             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 182                    elem->out_sg[2].iov_base + sizeof(mac_data),
 183                    mac_data.entries * ETH_ALEN);
 184             n->mac_table.in_use += mac_data.entries;
 185         } else
 186             n->allmulti = 1;
 187     }
 188
 189     return VIRTIO_NET_OK;
 190 }
 191
 192 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 193                                         VirtQueueElement *elem)
 194 {
 195     uint16_t vid;
 196
 197     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 198         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 199         return VIRTIO_NET_ERR;
 200     }
 201
 202     vid = lduw_le_p(elem->out_sg[1].iov_base);
 203
 204     if (vid >= MAX_VLAN)
 205         return VIRTIO_NET_ERR;
 206
 207     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 208         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 209     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 210         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 211     else
 212         return VIRTIO_NET_ERR;
 213
 214     return VIRTIO_NET_OK;
 215 }
 216
 217 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 218 {
 219     VirtIONet *n = to_virtio_net(vdev);
 220     struct virtio_net_ctrl_hdr ctrl;
 221     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 222     VirtQueueElement elem;
 223
 224     while (virtqueue_pop(vq, &elem)) {
 225         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 226             fprintf(stderr, "virtio-net ctrl missing headers\n");
 227             exit(1);
 228         }
 229
 230         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 231             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 232             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 233             exit(1);
 234         }
 235
 236         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 237         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 238
 239         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 240             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 241         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 242             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 243         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 244             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 245
 246         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 247
 248         virtqueue_push(vq, &elem, sizeof(status));
 249         virtio_notify(vdev, vq);
 250     }
 251 }
 252
 253 /* RX */
 254
 255 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 256 {
 257 }
 258
 259 static int do_virtio_net_can_receive(VirtIONet *n, int bufsize)
 260 {
 261     if (!virtio_queue_ready(n->rx_vq) ||
 262         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 263         return 0;
 264
 265     if (virtio_queue_empty(n->rx_vq) ||
 266         (n->mergeable_rx_bufs &&
 267          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 268         virtio_queue_set_notification(n->rx_vq, 1);
 269         return 0;
 270     }
 271
 272     virtio_queue_set_notification(n->rx_vq, 0);
 273     return 1;
 274 }
 275
 276 static int virtio_net_can_receive(void *opaque)
 277 {
 278     VirtIONet *n = opaque;
 279
 280     return do_virtio_net_can_receive(n, VIRTIO_NET_MAX_BUFSIZE);
 281 }
 282
 283 static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count)
 284 {
 285     int offset, i;
 286
 287     offset = i = 0;
 288     while (offset < count && i < iovcnt) {
 289         int len = MIN(iov[i].iov_len, count - offset);
 290         memcpy(iov[i].iov_base, buf + offset, len);
 291         offset += len;
 292         i++;
 293     }
 294
 295     return offset;
 296 }
 297
 298 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 299                           const void *buf, size_t size, size_t hdr_len)
 300 {
 301     struct virtio_net_hdr *hdr = iov[0].iov_base;
 302     int offset = 0;
 303
 304     hdr->flags = 0;
 305     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 306
 307     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 308      * but we may be passing along a larger header to the guest.
 309      */
 310     iov[0].iov_base += hdr_len;
 311     iov[0].iov_len  -= hdr_len;
 312
 313     return offset;
 314 }
 315
 316 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 317 {
 318     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 319     static const uint8_t vlan[] = {0x81, 0x00};
 320     uint8_t *ptr = (uint8_t *)buf;
 321     int i;
 322
 323     if (n->promisc)
 324         return 1;
 325
 326 #ifdef TAP_VNET_HDR
 327     if (tap_has_vnet_hdr(n->vc->vlan->first_client))
 328         ptr += sizeof(struct virtio_net_hdr);
 329 #endif
 330
 331     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 332         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 333         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 334             return 0;
 335     }
 336
 337     if ((ptr[0] & 1) && n->allmulti)
 338         return 1;
 339
 340     if (!memcmp(ptr, bcast, sizeof(bcast)))
 341         return 1;
 342
 343     if (!memcmp(ptr, n->mac, ETH_ALEN))
 344         return 1;
 345
 346     for (i = 0; i < n->mac_table.in_use; i++) {
 347         if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN))
 348             return 1;
 349     }
 350
 351     return 0;
 352 }
 353
 354 static void virtio_net_receive(void *opaque, const uint8_t *buf, int size)
 355 {
 356     VirtIONet *n = opaque;
 357     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 358     size_t hdr_len, offset, i;
 359
 360     if (!do_virtio_net_can_receive(n, size))
 361         return;
 362
 363     if (!receive_filter(n, buf, size))
 364         return;
 365
 366     /* hdr_len refers to the header we supply to the guest */
 367     hdr_len = n->mergeable_rx_bufs ?
 368         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 369
 370     offset = i = 0;
 371
 372     while (offset < size) {
 373         VirtQueueElement elem;
 374         int len, total;
 375         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 376
 377         len = total = 0;
 378
 379         if ((i != 0 && !n->mergeable_rx_bufs) ||
 380             virtqueue_pop(n->rx_vq, &elem) == 0) {
 381             if (i == 0)
 382                 return;
 383             fprintf(stderr, "virtio-net truncating packet\n");
 384             exit(1);
 385         }
 386
 387         if (elem.in_num < 1) {
 388             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 389             exit(1);
 390         }
 391
 392         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
 393             fprintf(stderr, "virtio-net header not in first element\n");
 394             exit(1);
 395         }
 396
 397         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 398
 399         if (i == 0) {
 400             if (n->mergeable_rx_bufs)
 401                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 402
 403             offset += receive_header(n, sg, elem.in_num,
 404                                      buf + offset, size - offset, hdr_len);
 405             total += hdr_len;
 406         }
 407
 408         /* copy in packet.  ugh */
 409         len = iov_fill(sg, elem.in_num,
 410                        buf + offset, size - offset);
 411         total += len;
 412
 413         /* signal other side */
 414         virtqueue_fill(n->rx_vq, &elem, total, i++);
 415
 416         offset += len;
 417     }
 418
 419     if (mhdr)
 420         mhdr->num_buffers = i;
 421
 422     virtqueue_flush(n->rx_vq, i);
 423     virtio_notify(&n->vdev, n->rx_vq);
 424 }
 425
 426 /* TX */
 427 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 428 {
 429     VirtQueueElement elem;
 430     int has_vnet_hdr = 0;
 431
 432     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 433         return;
 434
 435     while (virtqueue_pop(vq, &elem)) {
 436         ssize_t len = 0;
 437         unsigned int out_num = elem.out_num;
 438         struct iovec *out_sg = &elem.out_sg[0];
 439         unsigned hdr_len;
 440
 441         /* hdr_len refers to the header received from the guest */
 442         hdr_len = n->mergeable_rx_bufs ?
 443             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 444             sizeof(struct virtio_net_hdr);
 445
 446         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 447             fprintf(stderr, "virtio-net header not in first element\n");
 448             exit(1);
 449         }
 450
 451         /* ignore the header if GSO is not supported */
 452         if (!has_vnet_hdr) {
 453             out_num--;
 454             out_sg++;
 455             len += hdr_len;
 456         } else if (n->mergeable_rx_bufs) {
 457             /* tapfd expects a struct virtio_net_hdr */
 458             hdr_len -= sizeof(struct virtio_net_hdr);
 459             out_sg->iov_len -= hdr_len;
 460             len += hdr_len;
 461         }
 462
 463         len += qemu_sendv_packet(n->vc, out_sg, out_num);
 464
 465         virtqueue_push(vq, &elem, len);
 466         virtio_notify(&n->vdev, vq);
 467     }
 468 }
 469
 470 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 471 {
 472     VirtIONet *n = to_virtio_net(vdev);
 473
 474     if (n->tx_timer_active) {
 475         virtio_queue_set_notification(vq, 1);
 476         qemu_del_timer(n->tx_timer);
 477         n->tx_timer_active = 0;
 478         virtio_net_flush_tx(n, vq);
 479     } else {
 480         qemu_mod_timer(n->tx_timer,
 481                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 482         n->tx_timer_active = 1;
 483         virtio_queue_set_notification(vq, 0);
 484     }
 485 }
 486
 487 static void virtio_net_tx_timer(void *opaque)
 488 {
 489     VirtIONet *n = opaque;
 490
 491     n->tx_timer_active = 0;
 492
 493     /* Just in case the driver is not ready on more */
 494     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 495         return;
 496
 497     virtio_queue_set_notification(n->tx_vq, 1);
 498     virtio_net_flush_tx(n, n->tx_vq);
 499 }
 500
 501 static void virtio_net_save(QEMUFile *f, void *opaque)
 502 {
 503     VirtIONet *n = opaque;
 504
 505     virtio_save(&n->vdev, f);
 506
 507     qemu_put_buffer(f, n->mac, ETH_ALEN);
 508     qemu_put_be32(f, n->tx_timer_active);
 509     qemu_put_be32(f, n->mergeable_rx_bufs);
 510     qemu_put_be16(f, n->status);
 511     qemu_put_be32(f, n->promisc);
 512     qemu_put_be32(f, n->allmulti);
 513     qemu_put_be32(f, n->mac_table.in_use);
 514     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 515     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 516 }
 517
 518 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 519 {
 520     VirtIONet *n = opaque;
 521
 522     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 523         return -EINVAL;
 524
 525     virtio_load(&n->vdev, f);
 526
 527     qemu_get_buffer(f, n->mac, ETH_ALEN);
 528     n->tx_timer_active = qemu_get_be32(f);
 529     n->mergeable_rx_bufs = qemu_get_be32(f);
 530
 531     if (version_id >= 3)
 532         n->status = qemu_get_be16(f);
 533
 534     if (version_id >= 4) {
 535         n->promisc = qemu_get_be32(f);
 536         n->allmulti = qemu_get_be32(f);
 537     }
 538
 539     if (version_id >= 5) {
 540         n->mac_table.in_use = qemu_get_be32(f);
 541         /* MAC_TABLE_ENTRIES may be different from the saved image */
 542         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 543             qemu_get_buffer(f, n->mac_table.macs,
 544                             n->mac_table.in_use * ETH_ALEN);
 545         } else if (n->mac_table.in_use) {
 546             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 547             n->promisc = 1;
 548             n->mac_table.in_use = 0;
 549         }
 550     }
 551
 552     if (version_id >= 6)
 553         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 554
 555     if (n->tx_timer_active) {
 556         qemu_mod_timer(n->tx_timer,
 557                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 558     }
 559
 560     return 0;
 561 }
 562
 563 PCIDevice *virtio_net_init(PCIBus *bus, NICInfo *nd, int devfn)
 564 {
 565     VirtIONet *n;
 566     static int virtio_net_id;
 567
 568     n = (VirtIONet *)virtio_init_pci(bus, "virtio-net",
 569                                      PCI_VENDOR_ID_REDHAT_QUMRANET,
 570                                      PCI_DEVICE_ID_VIRTIO_NET,
 571                                      PCI_VENDOR_ID_REDHAT_QUMRANET,
 572                                      VIRTIO_ID_NET,
 573                                      PCI_CLASS_NETWORK_ETHERNET, 0x00,
 574                                      sizeof(struct virtio_net_config),
 575                                      sizeof(VirtIONet));
 576     if (!n)
 577         return NULL;
 578
 579     n->vdev.get_config = virtio_net_get_config;
 580     n->vdev.set_config = virtio_net_set_config;
 581     n->vdev.get_features = virtio_net_get_features;
 582     n->vdev.set_features = virtio_net_set_features;
 583     n->vdev.reset = virtio_net_reset;
 584     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 585     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 586     n->ctrl_vq = virtio_add_queue(&n->vdev, 16, virtio_net_handle_ctrl);
 587     memcpy(n->mac, nd->macaddr, ETH_ALEN);
 588     n->status = VIRTIO_NET_S_LINK_UP;
 589     n->vc = qemu_new_vlan_client(nd->vlan, nd->model, nd->name,
 590                                  virtio_net_receive, virtio_net_can_receive, n);
 591     n->vc->link_status_changed = virtio_net_set_link_status;
 592
 593     qemu_format_nic_info_str(n->vc, n->mac);
 594
 595     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 596     n->tx_timer_active = 0;
 597     n->mergeable_rx_bufs = 0;
 598     n->promisc = 1; /* for compatibility */
 599
 600     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 601
 602     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 603
 604     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 605                     virtio_net_save, virtio_net_load, n);
 606     return (PCIDevice *)n;
 607 }