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
  42 struct vhost_net {
  43     struct vhost_dev dev;
  44     struct vhost_virtqueue vqs[2];
  45     int backend;
  46     NetClientState *nc;
  47 };
  48
  49 /* Features supported by host kernel. */
  50 static const int kernel_feature_bits[] = {
  51     VIRTIO_F_NOTIFY_ON_EMPTY,
  52     VIRTIO_RING_F_INDIRECT_DESC,
  53     VIRTIO_RING_F_EVENT_IDX,
  54     VIRTIO_NET_F_MRG_RXBUF,
  55     VHOST_INVALID_FEATURE_BIT
  56 };
  57
  58 /* Features supported by others. */
  59 static const int user_feature_bits[] = {
  60     VIRTIO_F_NOTIFY_ON_EMPTY,
  61     VIRTIO_RING_F_INDIRECT_DESC,
  62     VIRTIO_RING_F_EVENT_IDX,
  63
  64     VIRTIO_F_ANY_LAYOUT,
  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     VIRTIO_NET_F_STATUS,
  78     VIRTIO_NET_F_CTRL_VQ,
  79     VIRTIO_NET_F_CTRL_RX,
  80     VIRTIO_NET_F_CTRL_VLAN,
  81     VIRTIO_NET_F_CTRL_RX_EXTRA,
  82     VIRTIO_NET_F_CTRL_MAC_ADDR,
  83     VIRTIO_NET_F_CTRL_GUEST_OFFLOADS,
  84
  85     VIRTIO_NET_F_MQ,
  86
  87     VHOST_INVALID_FEATURE_BIT
  88 };
  89
  90 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
  91 {
  92     const int *feature_bits = 0;
  93
  94     switch (net->nc->info->type) {
  95     case NET_CLIENT_OPTIONS_KIND_TAP:
  96         feature_bits = kernel_feature_bits;
  97         break;
  98     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
  99         feature_bits = user_feature_bits;
 100         break;
 101     default:
 102         error_report("Feature bits not defined for this type: %d",
 103                 net->nc->info->type);
 104         break;
 105     }
 106
 107     return feature_bits;
 108 }
 109
 110 unsigned vhost_net_get_features(struct vhost_net *net, unsigned features)
 111 {
 112     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
 113             features);
 114 }
 115
 116 void vhost_net_ack_features(struct vhost_net *net, unsigned features)
 117 {
 118     net->dev.acked_features = net->dev.backend_features;
 119     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 120 }
 121
 122 static int vhost_net_get_fd(NetClientState *backend)
 123 {
 124     switch (backend->info->type) {
 125     case NET_CLIENT_OPTIONS_KIND_TAP:
 126         return tap_get_fd(backend);
 127     default:
 128         fprintf(stderr, "vhost-net requires tap backend\n");
 129         return -EBADFD;
 130     }
 131 }
 132
 133 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 134 {
 135     int r;
 136     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 137     struct vhost_net *net = g_malloc(sizeof *net);
 138
 139     if (!options->net_backend) {
 140         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 141         goto fail;
 142     }
 143
 144     if (backend_kernel) {
 145         r = vhost_net_get_fd(options->net_backend);
 146         if (r < 0) {
 147             goto fail;
 148         }
 149         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 150             ? 0 : (1 << VHOST_NET_F_VIRTIO_NET_HDR);
 151         net->backend = r;
 152     } else {
 153         net->dev.backend_features = 0;
 154         net->backend = -1;
 155     }
 156     net->nc = options->net_backend;
 157
 158     net->dev.nvqs = 2;
 159     net->dev.vqs = net->vqs;
 160
 161     r = vhost_dev_init(&net->dev, options->opaque,
 162                        options->backend_type, options->force);
 163     if (r < 0) {
 164         goto fail;
 165     }
 166     if (backend_kernel) {
 167         if (!qemu_has_vnet_hdr_len(options->net_backend,
 168                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 169             net->dev.features &= ~(1 << VIRTIO_NET_F_MRG_RXBUF);
 170         }
 171         if (~net->dev.features & net->dev.backend_features) {
 172             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 173                    " for backend\n",
 174                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 175             vhost_dev_cleanup(&net->dev);
 176             goto fail;
 177         }
 178     }
 179     /* Set sane init value. Override when guest acks. */
 180     vhost_net_ack_features(net, 0);
 181     return net;
 182 fail:
 183     g_free(net);
 184     return NULL;
 185 }
 186
 187 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev)
 188 {
 189     return vhost_dev_query(&net->dev, dev);
 190 }
 191
 192 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 193 {
 194     net->dev.vq_index = vq_index;
 195 }
 196
 197 static int vhost_net_start_one(struct vhost_net *net,
 198                                VirtIODevice *dev)
 199 {
 200     struct vhost_vring_file file = { };
 201     int r;
 202
 203     net->dev.nvqs = 2;
 204     net->dev.vqs = net->vqs;
 205
 206     r = vhost_dev_enable_notifiers(&net->dev, dev);
 207     if (r < 0) {
 208         goto fail_notifiers;
 209     }
 210
 211     r = vhost_dev_start(&net->dev, dev);
 212     if (r < 0) {
 213         goto fail_start;
 214     }
 215
 216     if (net->nc->info->poll) {
 217         net->nc->info->poll(net->nc, false);
 218     }
 219
 220     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 221         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 222         file.fd = net->backend;
 223         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 224             const VhostOps *vhost_ops = net->dev.vhost_ops;
 225             r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
 226                                       &file);
 227             if (r < 0) {
 228                 r = -errno;
 229                 goto fail;
 230             }
 231         }
 232     }
 233     return 0;
 234 fail:
 235     file.fd = -1;
 236     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 237         while (file.index-- > 0) {
 238             const VhostOps *vhost_ops = net->dev.vhost_ops;
 239             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
 240                                           &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_call(&net->dev, VHOST_NET_SET_BACKEND,
 263                                           &file);
 264             assert(r >= 0);
 265         }
 266     } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
 267         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 268             const VhostOps *vhost_ops = net->dev.vhost_ops;
 269             int r = vhost_ops->vhost_call(&net->dev, VHOST_RESET_OWNER,
 270                                           NULL);
 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     vhost_dev_disable_notifiers(&net->dev, dev);
 279 }
 280
 281 static bool vhost_net_device_endian_ok(VirtIODevice *vdev)
 282 {
 283 #ifdef TARGET_IS_BIENDIAN
 284 #ifdef HOST_WORDS_BIGENDIAN
 285     return virtio_is_big_endian(vdev);
 286 #else
 287     return !virtio_is_big_endian(vdev);
 288 #endif
 289 #else
 290     return true;
 291 #endif
 292 }
 293
 294 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 295                     int total_queues)
 296 {
 297     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 298     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 299     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 300     int r, e, i;
 301
 302     if (!vhost_net_device_endian_ok(dev)) {
 303         error_report("vhost-net does not support cross-endian");
 304         r = -ENOSYS;
 305         goto err;
 306     }
 307
 308     if (!k->set_guest_notifiers) {
 309         error_report("binding does not support guest notifiers");
 310         r = -ENOSYS;
 311         goto err;
 312     }
 313
 314     for (i = 0; i < total_queues; i++) {
 315         vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
 316     }
 317
 318     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 319     if (r < 0) {
 320         error_report("Error binding guest notifier: %d", -r);
 321         goto err;
 322     }
 323
 324     for (i = 0; i < total_queues; i++) {
 325         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 326
 327         if (r < 0) {
 328             goto err_start;
 329         }
 330     }
 331
 332     return 0;
 333
 334 err_start:
 335     while (--i >= 0) {
 336         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 337     }
 338     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 339     if (e < 0) {
 340         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 341         fflush(stderr);
 342     }
 343 err:
 344     return r;
 345 }
 346
 347 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 348                     int total_queues)
 349 {
 350     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 351     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 352     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 353     int i, r;
 354
 355     for (i = 0; i < total_queues; i++) {
 356         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 357     }
 358
 359     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 360     if (r < 0) {
 361         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 362         fflush(stderr);
 363     }
 364     assert(r >= 0);
 365 }
 366
 367 void vhost_net_cleanup(struct vhost_net *net)
 368 {
 369     vhost_dev_cleanup(&net->dev);
 370     g_free(net);
 371 }
 372
 373 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 374 {
 375     return vhost_virtqueue_pending(&net->dev, idx);
 376 }
 377
 378 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 379                               int idx, bool mask)
 380 {
 381     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 382 }
 383
 384 VHostNetState *get_vhost_net(NetClientState *nc)
 385 {
 386     VHostNetState *vhost_net = 0;
 387
 388     if (!nc) {
 389         return 0;
 390     }
 391
 392     switch (nc->info->type) {
 393     case NET_CLIENT_OPTIONS_KIND_TAP:
 394         vhost_net = tap_get_vhost_net(nc);
 395         break;
 396     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
 397         vhost_net = vhost_user_get_vhost_net(nc);
 398         break;
 399     default:
 400         break;
 401     }
 402
 403     return vhost_net;
 404 }
 405 #else
 406 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 407 {
 408     error_report("vhost-net support is not compiled in");
 409     return NULL;
 410 }
 411
 412 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev)
 413 {
 414     return false;
 415 }
 416
 417 int vhost_net_start(VirtIODevice *dev,
 418                     NetClientState *ncs,
 419                     int total_queues)
 420 {
 421     return -ENOSYS;
 422 }
 423 void vhost_net_stop(VirtIODevice *dev,
 424                     NetClientState *ncs,
 425                     int total_queues)
 426 {
 427 }
 428
 429 void vhost_net_cleanup(struct vhost_net *net)
 430 {
 431 }
 432
 433 unsigned vhost_net_get_features(struct vhost_net *net, unsigned features)
 434 {
 435     return features;
 436 }
 437 void vhost_net_ack_features(struct vhost_net *net, unsigned features)
 438 {
 439 }
 440
 441 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 442 {
 443     return false;
 444 }
 445
 446 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 447                               int idx, bool mask)
 448 {
 449 }
 450
 451 VHostNetState *get_vhost_net(NetClientState *nc)
 452 {
 453     return 0;
 454 }
 455 #endif