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