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_OPTIONS_KIND_TAP:
  92         feature_bits = kernel_feature_bits;
  93         break;
  94     case NET_CLIENT_OPTIONS_KIND_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 static int vhost_net_get_fd(NetClientState *backend)
 124 {
 125     switch (backend->info->type) {
 126     case NET_CLIENT_OPTIONS_KIND_TAP:
 127         return tap_get_fd(backend);
 128     default:
 129         fprintf(stderr, "vhost-net requires tap backend\n");
 130         return -EBADFD;
 131     }
 132 }
 133
 134 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 135 {
 136     int r;
 137     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 138     struct vhost_net *net = g_malloc(sizeof *net);
 139
 140     if (!options->net_backend) {
 141         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 142         goto fail;
 143     }
 144     net->nc = options->net_backend;
 145
 146     net->dev.max_queues = 1;
 147     net->dev.nvqs = 2;
 148     net->dev.vqs = net->vqs;
 149
 150     if (backend_kernel) {
 151         r = vhost_net_get_fd(options->net_backend);
 152         if (r < 0) {
 153             goto fail;
 154         }
 155         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 156             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
 157         net->backend = r;
 158         net->dev.protocol_features = 0;
 159     } else {
 160         net->dev.backend_features = 0;
 161         net->dev.protocol_features = 0;
 162         net->backend = -1;
 163
 164         /* vhost-user needs vq_index to initiate a specific queue pair */
 165         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
 166     }
 167
 168     r = vhost_dev_init(&net->dev, options->opaque,
 169                        options->backend_type);
 170     if (r < 0) {
 171         goto fail;
 172     }
 173     if (backend_kernel) {
 174         if (!qemu_has_vnet_hdr_len(options->net_backend,
 175                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 176             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
 177         }
 178         if (~net->dev.features & net->dev.backend_features) {
 179             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 180                    " for backend\n",
 181                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 182             vhost_dev_cleanup(&net->dev);
 183             goto fail;
 184         }
 185     }
 186     /* Set sane init value. Override when guest acks. */
 187     vhost_net_ack_features(net, 0);
 188     return net;
 189 fail:
 190     g_free(net);
 191     return NULL;
 192 }
 193
 194 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 195 {
 196     net->dev.vq_index = vq_index;
 197 }
 198
 199 static int vhost_net_start_one(struct vhost_net *net,
 200                                VirtIODevice *dev)
 201 {
 202     struct vhost_vring_file file = { };
 203     int r;
 204
 205     net->dev.nvqs = 2;
 206     net->dev.vqs = net->vqs;
 207
 208     r = vhost_dev_enable_notifiers(&net->dev, dev);
 209     if (r < 0) {
 210         goto fail_notifiers;
 211     }
 212
 213     r = vhost_dev_start(&net->dev, dev);
 214     if (r < 0) {
 215         goto fail_start;
 216     }
 217
 218     if (net->nc->info->poll) {
 219         net->nc->info->poll(net->nc, false);
 220     }
 221
 222     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 223         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 224         file.fd = net->backend;
 225         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 226             const VhostOps *vhost_ops = net->dev.vhost_ops;
 227             r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
 228             if (r < 0) {
 229                 r = -errno;
 230                 goto fail;
 231             }
 232         }
 233     }
 234     return 0;
 235 fail:
 236     file.fd = -1;
 237     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 238         while (file.index-- > 0) {
 239             const VhostOps *vhost_ops = net->dev.vhost_ops;
 240             int r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
 241             assert(r >= 0);
 242         }
 243     }
 244     if (net->nc->info->poll) {
 245         net->nc->info->poll(net->nc, true);
 246     }
 247     vhost_dev_stop(&net->dev, dev);
 248 fail_start:
 249     vhost_dev_disable_notifiers(&net->dev, dev);
 250 fail_notifiers:
 251     return r;
 252 }
 253
 254 static void vhost_net_stop_one(struct vhost_net *net,
 255                                VirtIODevice *dev)
 256 {
 257     struct vhost_vring_file file = { .fd = -1 };
 258
 259     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 260         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 261             const VhostOps *vhost_ops = net->dev.vhost_ops;
 262             int r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
 263             assert(r >= 0);
 264         }
 265     }
 266     if (net->nc->info->poll) {
 267         net->nc->info->poll(net->nc, true);
 268     }
 269     vhost_dev_stop(&net->dev, dev);
 270     vhost_dev_disable_notifiers(&net->dev, dev);
 271 }
 272
 273 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 274                     int total_queues)
 275 {
 276     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 277     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 278     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 279     int r, e, i;
 280
 281     if (!k->set_guest_notifiers) {
 282         error_report("binding does not support guest notifiers");
 283         return -ENOSYS;
 284     }
 285
 286     for (i = 0; i < total_queues; i++) {
 287         struct vhost_net *net;
 288
 289         net = get_vhost_net(ncs[i].peer);
 290         vhost_net_set_vq_index(net, i * 2);
 291
 292         /* Suppress the masking guest notifiers on vhost user
 293          * because vhost user doesn't interrupt masking/unmasking
 294          * properly.
 295          */
 296         if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
 297                 dev->use_guest_notifier_mask = false;
 298         }
 299      }
 300
 301     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 302     if (r < 0) {
 303         error_report("Error binding guest notifier: %d", -r);
 304         goto err;
 305     }
 306
 307     for (i = 0; i < total_queues; i++) {
 308         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 309
 310         if (r < 0) {
 311             goto err_start;
 312         }
 313     }
 314
 315     return 0;
 316
 317 err_start:
 318     while (--i >= 0) {
 319         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 320     }
 321     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 322     if (e < 0) {
 323         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 324         fflush(stderr);
 325     }
 326 err:
 327     return r;
 328 }
 329
 330 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 331                     int total_queues)
 332 {
 333     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 334     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 335     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 336     int i, r;
 337
 338     for (i = 0; i < total_queues; i++) {
 339         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 340     }
 341
 342     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 343     if (r < 0) {
 344         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 345         fflush(stderr);
 346     }
 347     assert(r >= 0);
 348 }
 349
 350 void vhost_net_cleanup(struct vhost_net *net)
 351 {
 352     vhost_dev_cleanup(&net->dev);
 353     g_free(net);
 354 }
 355
 356 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 357 {
 358     const VhostOps *vhost_ops = net->dev.vhost_ops;
 359     int r = -1;
 360
 361     if (vhost_ops->vhost_migration_done) {
 362         r = vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 363     }
 364
 365     return r;
 366 }
 367
 368 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 369 {
 370     return vhost_virtqueue_pending(&net->dev, idx);
 371 }
 372
 373 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 374                               int idx, bool mask)
 375 {
 376     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 377 }
 378
 379 VHostNetState *get_vhost_net(NetClientState *nc)
 380 {
 381     VHostNetState *vhost_net = 0;
 382
 383     if (!nc) {
 384         return 0;
 385     }
 386
 387     switch (nc->info->type) {
 388     case NET_CLIENT_OPTIONS_KIND_TAP:
 389         vhost_net = tap_get_vhost_net(nc);
 390         break;
 391     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
 392         vhost_net = vhost_user_get_vhost_net(nc);
 393         break;
 394     default:
 395         break;
 396     }
 397
 398     return vhost_net;
 399 }
 400
 401 int vhost_set_vring_enable(NetClientState *nc, int enable)
 402 {
 403     VHostNetState *net = get_vhost_net(nc);
 404     const VhostOps *vhost_ops = net->dev.vhost_ops;
 405
 406     if (vhost_ops->vhost_set_vring_enable) {
 407         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 408     }
 409
 410     return 0;
 411 }
 412
 413 #else
 414 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 415 {
 416     return 1;
 417 }
 418
 419 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 420 {
 421     error_report("vhost-net support is not compiled in");
 422     return NULL;
 423 }
 424
 425 int vhost_net_start(VirtIODevice *dev,
 426                     NetClientState *ncs,
 427                     int total_queues)
 428 {
 429     return -ENOSYS;
 430 }
 431 void vhost_net_stop(VirtIODevice *dev,
 432                     NetClientState *ncs,
 433                     int total_queues)
 434 {
 435 }
 436
 437 void vhost_net_cleanup(struct vhost_net *net)
 438 {
 439 }
 440
 441 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 442 {
 443     return features;
 444 }
 445 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 446 {
 447 }
 448
 449 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 450 {
 451     return false;
 452 }
 453
 454 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 455                               int idx, bool mask)
 456 {
 457 }
 458
 459 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 460 {
 461     return -1;
 462 }
 463
 464 VHostNetState *get_vhost_net(NetClientState *nc)
 465 {
 466     return 0;
 467 }
 468
 469 int vhost_set_vring_enable(NetClientState *nc, int enable)
 470 {
 471     return 0;
 472 }
 473 #endif