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