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 "net/net.h"
  17 #include "net/tap.h"
  18 #include "net/vhost-user.h"
  19
  20 #include "hw/virtio/virtio-net.h"
  21 #include "net/vhost_net.h"
  22 #include "qemu/error-report.h"
  23
  24 #include "config.h"
  25
  26 #ifdef CONFIG_VHOST_NET
  27 #include <linux/vhost.h>
  28 #include <sys/socket.h>
  29 #include <linux/kvm.h>
  30 #include <fcntl.h>
  31 #include <netpacket/packet.h>
  32 #include <net/ethernet.h>
  33 #include <net/if.h>
  34 #include <netinet/in.h>
  35
  36 #include <stdio.h>
  37
  38 #include "standard-headers/linux/virtio_ring.h"
  39 #include "hw/virtio/vhost.h"
  40 #include "hw/virtio/virtio-bus.h"
  41 #include "hw/virtio/virtio-access.h"
  42
  43 struct vhost_net {
  44     struct vhost_dev dev;
  45     struct vhost_virtqueue vqs[2];
  46     int backend;
  47     NetClientState *nc;
  48 };
  49
  50 /* Features supported by host kernel. */
  51 static const int kernel_feature_bits[] = {
  52     VIRTIO_F_NOTIFY_ON_EMPTY,
  53     VIRTIO_RING_F_INDIRECT_DESC,
  54     VIRTIO_RING_F_EVENT_IDX,
  55     VIRTIO_NET_F_MRG_RXBUF,
  56     VIRTIO_F_VERSION_1,
  57     VHOST_INVALID_FEATURE_BIT
  58 };
  59
  60 /* Features supported by others. */
  61 static const int user_feature_bits[] = {
  62     VIRTIO_F_NOTIFY_ON_EMPTY,
  63     VIRTIO_RING_F_INDIRECT_DESC,
  64     VIRTIO_RING_F_EVENT_IDX,
  65
  66     VIRTIO_F_ANY_LAYOUT,
  67     VIRTIO_F_VERSION_1,
  68     VIRTIO_NET_F_CSUM,
  69     VIRTIO_NET_F_GUEST_CSUM,
  70     VIRTIO_NET_F_GSO,
  71     VIRTIO_NET_F_GUEST_TSO4,
  72     VIRTIO_NET_F_GUEST_TSO6,
  73     VIRTIO_NET_F_GUEST_ECN,
  74     VIRTIO_NET_F_GUEST_UFO,
  75     VIRTIO_NET_F_HOST_TSO4,
  76     VIRTIO_NET_F_HOST_TSO6,
  77     VIRTIO_NET_F_HOST_ECN,
  78     VIRTIO_NET_F_HOST_UFO,
  79     VIRTIO_NET_F_MRG_RXBUF,
  80     VIRTIO_NET_F_STATUS,
  81     VIRTIO_NET_F_CTRL_VQ,
  82     VIRTIO_NET_F_CTRL_RX,
  83     VIRTIO_NET_F_CTRL_VLAN,
  84     VIRTIO_NET_F_CTRL_RX_EXTRA,
  85     VIRTIO_NET_F_CTRL_MAC_ADDR,
  86     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,
  87
  88     VIRTIO_NET_F_GUEST_ANNOUNCE,
  89
  90     VIRTIO_NET_F_MQ,
  91
  92     VHOST_INVALID_FEATURE_BIT
  93 };
  94
  95 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
  96 {
  97     const int *feature_bits = 0;
  98
  99     switch (net->nc->info->type) {
 100     case NET_CLIENT_OPTIONS_KIND_TAP:
 101         feature_bits = kernel_feature_bits;
 102         break;
 103     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
 104         feature_bits = user_feature_bits;
 105         break;
 106     default:
 107         error_report("Feature bits not defined for this type: %d",
 108                 net->nc->info->type);
 109         break;
 110     }
 111
 112     return feature_bits;
 113 }
 114
 115 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 116 {
 117     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
 118             features);
 119 }
 120
 121 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 122 {
 123     net->dev.acked_features = net->dev.backend_features;
 124     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 125 }
 126
 127 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 128 {
 129     return net->dev.max_queues;
 130 }
 131
 132 static int vhost_net_get_fd(NetClientState *backend)
 133 {
 134     switch (backend->info->type) {
 135     case NET_CLIENT_OPTIONS_KIND_TAP:
 136         return tap_get_fd(backend);
 137     default:
 138         fprintf(stderr, "vhost-net requires tap backend\n");
 139         return -EBADFD;
 140     }
 141 }
 142
 143 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 144 {
 145     int r;
 146     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 147     struct vhost_net *net = g_malloc(sizeof *net);
 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);
 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             vhost_dev_cleanup(&net->dev);
 192             goto fail;
 193         }
 194     }
 195     /* Set sane init value. Override when guest acks. */
 196     vhost_net_ack_features(net, 0);
 197     return net;
 198 fail:
 199     g_free(net);
 200     return NULL;
 201 }
 202
 203 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 204 {
 205     net->dev.vq_index = vq_index;
 206 }
 207
 208 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
 209                                      bool set)
 210 {
 211     int r = 0;
 212
 213     if (virtio_vdev_has_feature(dev, VIRTIO_F_VERSION_1) ||
 214         (virtio_legacy_is_cross_endian(dev) && !virtio_is_big_endian(dev))) {
 215         r = qemu_set_vnet_le(peer, set);
 216         if (r) {
 217             error_report("backend does not support LE vnet headers");
 218         }
 219     } else if (virtio_legacy_is_cross_endian(dev)) {
 220         r = qemu_set_vnet_be(peer, set);
 221         if (r) {
 222             error_report("backend does not support BE vnet headers");
 223         }
 224     }
 225
 226     return r;
 227 }
 228
 229 static int vhost_net_start_one(struct vhost_net *net,
 230                                VirtIODevice *dev)
 231 {
 232     struct vhost_vring_file file = { };
 233     int r;
 234
 235     net->dev.nvqs = 2;
 236     net->dev.vqs = net->vqs;
 237
 238     r = vhost_dev_enable_notifiers(&net->dev, dev);
 239     if (r < 0) {
 240         goto fail_notifiers;
 241     }
 242
 243     r = vhost_dev_start(&net->dev, dev);
 244     if (r < 0) {
 245         goto fail_start;
 246     }
 247
 248     if (net->nc->info->poll) {
 249         net->nc->info->poll(net->nc, false);
 250     }
 251
 252     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 253         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 254         file.fd = net->backend;
 255         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 256             const VhostOps *vhost_ops = net->dev.vhost_ops;
 257             r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
 258             if (r < 0) {
 259                 r = -errno;
 260                 goto fail;
 261             }
 262         }
 263     }
 264     return 0;
 265 fail:
 266     file.fd = -1;
 267     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 268         while (file.index-- > 0) {
 269             const VhostOps *vhost_ops = net->dev.vhost_ops;
 270             int r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
 271             assert(r >= 0);
 272         }
 273     }
 274     if (net->nc->info->poll) {
 275         net->nc->info->poll(net->nc, true);
 276     }
 277     vhost_dev_stop(&net->dev, dev);
 278 fail_start:
 279     vhost_dev_disable_notifiers(&net->dev, dev);
 280 fail_notifiers:
 281     return r;
 282 }
 283
 284 static void vhost_net_stop_one(struct vhost_net *net,
 285                                VirtIODevice *dev)
 286 {
 287     struct vhost_vring_file file = { .fd = -1 };
 288
 289     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 290         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 291             const VhostOps *vhost_ops = net->dev.vhost_ops;
 292             int r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
 293             assert(r >= 0);
 294         }
 295     } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
 296         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 297             const VhostOps *vhost_ops = net->dev.vhost_ops;
 298             int r = vhost_ops->vhost_reset_device(&net->dev);
 299             assert(r >= 0);
 300         }
 301     }
 302     if (net->nc->info->poll) {
 303         net->nc->info->poll(net->nc, true);
 304     }
 305     vhost_dev_stop(&net->dev, dev);
 306     vhost_dev_disable_notifiers(&net->dev, dev);
 307 }
 308
 309 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 310                     int total_queues)
 311 {
 312     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 313     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 314     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 315     int r, e, i;
 316
 317     if (!k->set_guest_notifiers) {
 318         error_report("binding does not support guest notifiers");
 319         r = -ENOSYS;
 320         goto err;
 321     }
 322
 323     r = vhost_net_set_vnet_endian(dev, ncs[0].peer, true);
 324     if (r < 0) {
 325         goto err;
 326     }
 327
 328     for (i = 0; i < total_queues; i++) {
 329         vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
 330     }
 331
 332     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 333     if (r < 0) {
 334         error_report("Error binding guest notifier: %d", -r);
 335         goto err_endian;
 336     }
 337
 338     for (i = 0; i < total_queues; i++) {
 339         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 340
 341         if (r < 0) {
 342             goto err_start;
 343         }
 344     }
 345
 346     return 0;
 347
 348 err_start:
 349     while (--i >= 0) {
 350         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 351     }
 352     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 353     if (e < 0) {
 354         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 355         fflush(stderr);
 356     }
 357 err_endian:
 358     vhost_net_set_vnet_endian(dev, ncs[0].peer, false);
 359 err:
 360     return r;
 361 }
 362
 363 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 364                     int total_queues)
 365 {
 366     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 367     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 368     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 369     int i, r;
 370
 371     for (i = 0; i < total_queues; i++) {
 372         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 373     }
 374
 375     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 376     if (r < 0) {
 377         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 378         fflush(stderr);
 379     }
 380     assert(r >= 0);
 381
 382     assert(vhost_net_set_vnet_endian(dev, ncs[0].peer, false) >= 0);
 383 }
 384
 385 void vhost_net_cleanup(struct vhost_net *net)
 386 {
 387     vhost_dev_cleanup(&net->dev);
 388     g_free(net);
 389 }
 390
 391 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 392 {
 393     const VhostOps *vhost_ops = net->dev.vhost_ops;
 394     int r = -1;
 395
 396     if (vhost_ops->vhost_migration_done) {
 397         r = vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 398     }
 399
 400     return r;
 401 }
 402
 403 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 404 {
 405     return vhost_virtqueue_pending(&net->dev, idx);
 406 }
 407
 408 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 409                               int idx, bool mask)
 410 {
 411     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 412 }
 413
 414 VHostNetState *get_vhost_net(NetClientState *nc)
 415 {
 416     VHostNetState *vhost_net = 0;
 417
 418     if (!nc) {
 419         return 0;
 420     }
 421
 422     switch (nc->info->type) {
 423     case NET_CLIENT_OPTIONS_KIND_TAP:
 424         vhost_net = tap_get_vhost_net(nc);
 425         break;
 426     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
 427         vhost_net = vhost_user_get_vhost_net(nc);
 428         break;
 429     default:
 430         break;
 431     }
 432
 433     return vhost_net;
 434 }
 435
 436 int vhost_set_vring_enable(NetClientState *nc, int enable)
 437 {
 438     VHostNetState *net = get_vhost_net(nc);
 439     const VhostOps *vhost_ops = net->dev.vhost_ops;
 440
 441     if (vhost_ops->vhost_set_vring_enable) {
 442         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 443     }
 444
 445     return 0;
 446 }
 447
 448 #else
 449 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 450 {
 451     return 1;
 452 }
 453
 454 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 455 {
 456     error_report("vhost-net support is not compiled in");
 457     return NULL;
 458 }
 459
 460 int vhost_net_start(VirtIODevice *dev,
 461                     NetClientState *ncs,
 462                     int total_queues)
 463 {
 464     return -ENOSYS;
 465 }
 466 void vhost_net_stop(VirtIODevice *dev,
 467                     NetClientState *ncs,
 468                     int total_queues)
 469 {
 470 }
 471
 472 void vhost_net_cleanup(struct vhost_net *net)
 473 {
 474 }
 475
 476 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 477 {
 478     return features;
 479 }
 480 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 481 {
 482 }
 483
 484 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 485 {
 486     return false;
 487 }
 488
 489 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 490                               int idx, bool mask)
 491 {
 492 }
 493
 494 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 495 {
 496     return -1;
 497 }
 498
 499 VHostNetState *get_vhost_net(NetClientState *nc)
 500 {
 501     return 0;
 502 }
 503
 504 int vhost_set_vring_enable(NetClientState *nc, int enable)
 505 {
 506     return 0;
 507 }
 508 #endif