hw/net/vhost_net.c

   1 /*
   2  * vhost-net support
   3  *
   4  * Copyright Red Hat, Inc. 2010
   5  *
   6  * Authors:
   7  *  Michael S. Tsirkin <mst@redhat.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  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu/osdep.h"
  17 #include "net/net.h"
  18 #include "net/tap.h"
  19 #include "net/vhost-user.h"
  20
  21 #include "hw/virtio/virtio-net.h"
  22 #include "net/vhost_net.h"
  23 #include "qemu/error-report.h"
  24
  25
  26 #ifdef CONFIG_VHOST_NET
  27 #include <linux/vhost.h>
  28 #include <sys/socket.h>
  29 #include <linux/kvm.h>
  30 #include <netpacket/packet.h>
  31 #include <net/ethernet.h>
  32 #include <net/if.h>
  33 #include <netinet/in.h>
  34
  35
  36 #include "standard-headers/linux/virtio_ring.h"
  37 #include "hw/virtio/vhost.h"
  38 #include "hw/virtio/virtio-bus.h"
  39
  40 struct vhost_net {
  41     struct vhost_dev dev;
  42     struct vhost_virtqueue vqs[2];
  43     int backend;
  44     NetClientState *nc;
  45 };
  46
  47 /* Features supported by host kernel. */
  48 static const int kernel_feature_bits[] = {
  49     VIRTIO_F_NOTIFY_ON_EMPTY,
  50     VIRTIO_RING_F_INDIRECT_DESC,
  51     VIRTIO_RING_F_EVENT_IDX,
  52     VIRTIO_NET_F_MRG_RXBUF,
  53     VIRTIO_F_VERSION_1,
  54     VIRTIO_NET_F_MTU,
  55     VIRTIO_F_IOMMU_PLATFORM,
  56     VHOST_INVALID_FEATURE_BIT
  57 };
  58
  59 /* Features supported by others. */
  60 static const int user_feature_bits[] = {
  61     VIRTIO_F_NOTIFY_ON_EMPTY,
  62     VIRTIO_RING_F_INDIRECT_DESC,
  63     VIRTIO_RING_F_EVENT_IDX,
  64
  65     VIRTIO_F_ANY_LAYOUT,
  66     VIRTIO_F_VERSION_1,
  67     VIRTIO_NET_F_CSUM,
  68     VIRTIO_NET_F_GUEST_CSUM,
  69     VIRTIO_NET_F_GSO,
  70     VIRTIO_NET_F_GUEST_TSO4,
  71     VIRTIO_NET_F_GUEST_TSO6,
  72     VIRTIO_NET_F_GUEST_ECN,
  73     VIRTIO_NET_F_GUEST_UFO,
  74     VIRTIO_NET_F_HOST_TSO4,
  75     VIRTIO_NET_F_HOST_TSO6,
  76     VIRTIO_NET_F_HOST_ECN,
  77     VIRTIO_NET_F_HOST_UFO,
  78     VIRTIO_NET_F_MRG_RXBUF,
  79     VIRTIO_NET_F_MTU,
  80
  81     /* This bit implies RARP isn't sent by QEMU out of band */
  82     VIRTIO_NET_F_GUEST_ANNOUNCE,
  83
  84     VIRTIO_NET_F_MQ,
  85
  86     VHOST_INVALID_FEATURE_BIT
  87 };
  88
  89 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
  90 {
  91     const int *feature_bits = 0;
  92
  93     switch (net->nc->info->type) {
  94     case NET_CLIENT_DRIVER_TAP:
  95         feature_bits = kernel_feature_bits;
  96         break;
  97     case NET_CLIENT_DRIVER_VHOST_USER:
  98         feature_bits = user_feature_bits;
  99         break;
 100     default:
 101         error_report("Feature bits not defined for this type: %d",
 102                 net->nc->info->type);
 103         break;
 104     }
 105
 106     return feature_bits;
 107 }
 108
 109 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 110 {
 111     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
 112             features);
 113 }
 114
 115 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 116 {
 117     net->dev.acked_features = net->dev.backend_features;
 118     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 119 }
 120
 121 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 122 {
 123     return net->dev.max_queues;
 124 }
 125
 126 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 127 {
 128     return net->dev.acked_features;
 129 }
 130
 131 static int vhost_net_get_fd(NetClientState *backend)
 132 {
 133     switch (backend->info->type) {
 134     case NET_CLIENT_DRIVER_TAP:
 135         return tap_get_fd(backend);
 136     default:
 137         fprintf(stderr, "vhost-net requires tap backend\n");
 138         return -EBADFD;
 139     }
 140 }
 141
 142 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 143 {
 144     int r;
 145     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 146     struct vhost_net *net = g_new0(struct vhost_net, 1);
 147     uint64_t features = 0;
 148
 149     if (!options->net_backend) {
 150         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 151         goto fail;
 152     }
 153     net->nc = options->net_backend;
 154
 155     net->dev.max_queues = 1;
 156     net->dev.nvqs = 2;
 157     net->dev.vqs = net->vqs;
 158
 159     if (backend_kernel) {
 160         r = vhost_net_get_fd(options->net_backend);
 161         if (r < 0) {
 162             goto fail;
 163         }
 164         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 165             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
 166         net->backend = r;
 167         net->dev.protocol_features = 0;
 168     } else {
 169         net->dev.backend_features = 0;
 170         net->dev.protocol_features = 0;
 171         net->backend = -1;
 172
 173         /* vhost-user needs vq_index to initiate a specific queue pair */
 174         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
 175     }
 176
 177     r = vhost_dev_init(&net->dev, options->opaque,
 178                        options->backend_type, options->busyloop_timeout);
 179     if (r < 0) {
 180         goto fail;
 181     }
 182     if (backend_kernel) {
 183         if (!qemu_has_vnet_hdr_len(options->net_backend,
 184                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 185             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
 186         }
 187         if (~net->dev.features & net->dev.backend_features) {
 188             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 189                    " for backend\n",
 190                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 191             goto fail;
 192         }
 193     }
 194
 195     /* Set sane init value. Override when guest acks. */
 196     if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 197         features = vhost_user_get_acked_features(net->nc);
 198         if (~net->dev.features & features) {
 199             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 200                     " for backend\n",
 201                     (uint64_t)(~net->dev.features & features));
 202             goto fail;
 203         }
 204     }
 205
 206     vhost_net_ack_features(net, features);
 207
 208     return net;
 209
 210 fail:
 211     vhost_dev_cleanup(&net->dev);
 212     g_free(net);
 213     return NULL;
 214 }
 215
 216 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 217 {
 218     net->dev.vq_index = vq_index;
 219 }
 220
 221 static int vhost_net_start_one(struct vhost_net *net,
 222                                VirtIODevice *dev)
 223 {
 224     struct vhost_vring_file file = { };
 225     int r;
 226
 227     net->dev.nvqs = 2;
 228     net->dev.vqs = net->vqs;
 229
 230     r = vhost_dev_enable_notifiers(&net->dev, dev);
 231     if (r < 0) {
 232         goto fail_notifiers;
 233     }
 234
 235     r = vhost_dev_start(&net->dev, dev);
 236     if (r < 0) {
 237         goto fail_start;
 238     }
 239
 240     if (net->nc->info->poll) {
 241         net->nc->info->poll(net->nc, false);
 242     }
 243
 244     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 245         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 246         file.fd = net->backend;
 247         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 248             r = vhost_net_set_backend(&net->dev, &file);
 249             if (r < 0) {
 250                 r = -errno;
 251                 goto fail;
 252             }
 253         }
 254     }
 255     return 0;
 256 fail:
 257     file.fd = -1;
 258     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 259         while (file.index-- > 0) {
 260             int r = vhost_net_set_backend(&net->dev, &file);
 261             assert(r >= 0);
 262         }
 263     }
 264     if (net->nc->info->poll) {
 265         net->nc->info->poll(net->nc, true);
 266     }
 267     vhost_dev_stop(&net->dev, dev);
 268 fail_start:
 269     vhost_dev_disable_notifiers(&net->dev, dev);
 270 fail_notifiers:
 271     return r;
 272 }
 273
 274 static void vhost_net_stop_one(struct vhost_net *net,
 275                                VirtIODevice *dev)
 276 {
 277     struct vhost_vring_file file = { .fd = -1 };
 278
 279     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 280         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 281             int r = vhost_net_set_backend(&net->dev, &file);
 282             assert(r >= 0);
 283         }
 284     }
 285     if (net->nc->info->poll) {
 286         net->nc->info->poll(net->nc, true);
 287     }
 288     vhost_dev_stop(&net->dev, dev);
 289     vhost_dev_disable_notifiers(&net->dev, dev);
 290 }
 291
 292 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 293                     int total_queues)
 294 {
 295     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 296     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 297     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 298     int r, e, i;
 299
 300     if (!k->set_guest_notifiers) {
 301         error_report("binding does not support guest notifiers");
 302         return -ENOSYS;
 303     }
 304
 305     for (i = 0; i < total_queues; i++) {
 306         struct vhost_net *net;
 307
 308         net = get_vhost_net(ncs[i].peer);
 309         vhost_net_set_vq_index(net, i * 2);
 310
 311         /* Suppress the masking guest notifiers on vhost user
 312          * because vhost user doesn't interrupt masking/unmasking
 313          * properly.
 314          */
 315         if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 316             dev->use_guest_notifier_mask = false;
 317         }
 318      }
 319
 320     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 321     if (r < 0) {
 322         error_report("Error binding guest notifier: %d", -r);
 323         goto err;
 324     }
 325
 326     for (i = 0; i < total_queues; i++) {
 327         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 328
 329         if (r < 0) {
 330             goto err_start;
 331         }
 332
 333         if (ncs[i].peer->vring_enable) {
 334             /* restore vring enable state */
 335             r = vhost_set_vring_enable(ncs[i].peer, ncs[i].peer->vring_enable);
 336
 337             if (r < 0) {
 338                 goto err_start;
 339             }
 340         }
 341     }
 342
 343     return 0;
 344
 345 err_start:
 346     while (--i >= 0) {
 347         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 348     }
 349     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 350     if (e < 0) {
 351         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 352         fflush(stderr);
 353     }
 354 err:
 355     return r;
 356 }
 357
 358 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 359                     int total_queues)
 360 {
 361     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 362     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 363     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 364     int i, r;
 365
 366     for (i = 0; i < total_queues; i++) {
 367         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 368     }
 369
 370     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 371     if (r < 0) {
 372         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 373         fflush(stderr);
 374     }
 375     assert(r >= 0);
 376 }
 377
 378 void vhost_net_cleanup(struct vhost_net *net)
 379 {
 380     vhost_dev_cleanup(&net->dev);
 381 }
 382
 383 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 384 {
 385     const VhostOps *vhost_ops = net->dev.vhost_ops;
 386
 387     assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 388     assert(vhost_ops->vhost_migration_done);
 389
 390     return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 391 }
 392
 393 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 394 {
 395     return vhost_virtqueue_pending(&net->dev, idx);
 396 }
 397
 398 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 399                               int idx, bool mask)
 400 {
 401     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 402 }
 403
 404 VHostNetState *get_vhost_net(NetClientState *nc)
 405 {
 406     VHostNetState *vhost_net = 0;
 407
 408     if (!nc) {
 409         return 0;
 410     }
 411
 412     switch (nc->info->type) {
 413     case NET_CLIENT_DRIVER_TAP:
 414         vhost_net = tap_get_vhost_net(nc);
 415         break;
 416     case NET_CLIENT_DRIVER_VHOST_USER:
 417         vhost_net = vhost_user_get_vhost_net(nc);
 418         assert(vhost_net);
 419         break;
 420     default:
 421         break;
 422     }
 423
 424     return vhost_net;
 425 }
 426
 427 int vhost_set_vring_enable(NetClientState *nc, int enable)
 428 {
 429     VHostNetState *net = get_vhost_net(nc);
 430     const VhostOps *vhost_ops = net->dev.vhost_ops;
 431
 432     nc->vring_enable = enable;
 433
 434     if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
 435         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 436     }
 437
 438     return 0;
 439 }
 440
 441 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
 442 {
 443     const VhostOps *vhost_ops = net->dev.vhost_ops;
 444
 445     if (!vhost_ops->vhost_net_set_mtu) {
 446         return 0;
 447     }
 448
 449     return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
 450 }
 451
 452 #else
 453 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 454 {
 455     return 1;
 456 }
 457
 458 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 459 {
 460     error_report("vhost-net support is not compiled in");
 461     return NULL;
 462 }
 463
 464 int vhost_net_start(VirtIODevice *dev,
 465                     NetClientState *ncs,
 466                     int total_queues)
 467 {
 468     return -ENOSYS;
 469 }
 470 void vhost_net_stop(VirtIODevice *dev,
 471                     NetClientState *ncs,
 472                     int total_queues)
 473 {
 474 }
 475
 476 void vhost_net_cleanup(struct vhost_net *net)
 477 {
 478 }
 479
 480 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 481 {
 482     return features;
 483 }
 484
 485 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 486 {
 487 }
 488
 489 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 490 {
 491     return 0;
 492 }
 493
 494 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 495 {
 496     return false;
 497 }
 498
 499 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 500                               int idx, bool mask)
 501 {
 502 }
 503
 504 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 505 {
 506     return -1;
 507 }
 508
 509 VHostNetState *get_vhost_net(NetClientState *nc)
 510 {
 511     return 0;
 512 }
 513
 514 int vhost_set_vring_enable(NetClientState *nc, int enable)
 515 {
 516     return 0;
 517 }
 518
 519 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
 520 {
 521     return 0;
 522 }
 523 #endif