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     net->dev.vq_index = net->nc->queue_index;
 161
 162     r = vhost_dev_init(&net->dev, options->opaque,
 163                        options->backend_type, options->force);
 164     if (r < 0) {
 165         goto fail;
 166     }
 167     if (backend_kernel) {
 168         if (!qemu_has_vnet_hdr_len(options->net_backend,
 169                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 170             net->dev.features &= ~(1 << VIRTIO_NET_F_MRG_RXBUF);
 171         }
 172         if (~net->dev.features & net->dev.backend_features) {
 173             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 174                    " for backend\n",
 175                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 176             vhost_dev_cleanup(&net->dev);
 177             goto fail;
 178         }
 179     }
 180     /* Set sane init value. Override when guest acks. */
 181     vhost_net_ack_features(net, 0);
 182     return net;
 183 fail:
 184     g_free(net);
 185     return NULL;
 186 }
 187
 188 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev)
 189 {
 190     return vhost_dev_query(&net->dev, dev);
 191 }
 192
 193 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 194 {
 195     net->dev.vq_index = vq_index;
 196 }
 197
 198 static int vhost_net_start_one(struct vhost_net *net,
 199                                VirtIODevice *dev)
 200 {
 201     struct vhost_vring_file file = { };
 202     int r;
 203
 204     net->dev.nvqs = 2;
 205     net->dev.vqs = net->vqs;
 206
 207     r = vhost_dev_enable_notifiers(&net->dev, dev);
 208     if (r < 0) {
 209         goto fail_notifiers;
 210     }
 211
 212     r = vhost_dev_start(&net->dev, dev);
 213     if (r < 0) {
 214         goto fail_start;
 215     }
 216
 217     if (net->nc->info->poll) {
 218         net->nc->info->poll(net->nc, false);
 219     }
 220
 221     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 222         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 223         file.fd = net->backend;
 224         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 225             const VhostOps *vhost_ops = net->dev.vhost_ops;
 226             r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
 227                                       &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_call(&net->dev, VHOST_NET_SET_BACKEND,
 241                                           &file);
 242             assert(r >= 0);
 243         }
 244     }
 245     if (net->nc->info->poll) {
 246         net->nc->info->poll(net->nc, true);
 247     }
 248     vhost_dev_stop(&net->dev, dev);
 249 fail_start:
 250     vhost_dev_disable_notifiers(&net->dev, dev);
 251 fail_notifiers:
 252     return r;
 253 }
 254
 255 static void vhost_net_stop_one(struct vhost_net *net,
 256                                VirtIODevice *dev)
 257 {
 258     struct vhost_vring_file file = { .fd = -1 };
 259
 260     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 261         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 262             const VhostOps *vhost_ops = net->dev.vhost_ops;
 263             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
 264                                           &file);
 265             assert(r >= 0);
 266         }
 267     } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
 268         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 269             const VhostOps *vhost_ops = net->dev.vhost_ops;
 270             int r = vhost_ops->vhost_call(&net->dev, VHOST_RESET_OWNER,
 271                                           &file);
 272             assert(r >= 0);
 273         }
 274     }
 275     if (net->nc->info->poll) {
 276         net->nc->info->poll(net->nc, true);
 277     }
 278     vhost_dev_stop(&net->dev, dev);
 279     vhost_dev_disable_notifiers(&net->dev, dev);
 280 }
 281
 282 static bool vhost_net_device_endian_ok(VirtIODevice *vdev)
 283 {
 284 #ifdef TARGET_IS_BIENDIAN
 285 #ifdef HOST_WORDS_BIGENDIAN
 286     return virtio_is_big_endian(vdev);
 287 #else
 288     return !virtio_is_big_endian(vdev);
 289 #endif
 290 #else
 291     return true;
 292 #endif
 293 }
 294
 295 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 296                     int total_queues)
 297 {
 298     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 299     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 300     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 301     int r, e, i;
 302
 303     if (!vhost_net_device_endian_ok(dev)) {
 304         error_report("vhost-net does not support cross-endian");
 305         r = -ENOSYS;
 306         goto err;
 307     }
 308
 309     if (!k->set_guest_notifiers) {
 310         error_report("binding does not support guest notifiers");
 311         r = -ENOSYS;
 312         goto err;
 313     }
 314
 315     for (i = 0; i < total_queues; i++) {
 316         vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
 317     }
 318
 319     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 320     if (r < 0) {
 321         error_report("Error binding guest notifier: %d", -r);
 322         goto err;
 323     }
 324
 325     for (i = 0; i < total_queues; i++) {
 326         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 327
 328         if (r < 0) {
 329             goto err_start;
 330         }
 331     }
 332
 333     return 0;
 334
 335 err_start:
 336     while (--i >= 0) {
 337         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 338     }
 339     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 340     if (e < 0) {
 341         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 342         fflush(stderr);
 343     }
 344 err:
 345     return r;
 346 }
 347
 348 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 349                     int total_queues)
 350 {
 351     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 352     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 353     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 354     int i, r;
 355
 356     for (i = 0; i < total_queues; i++) {
 357         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 358     }
 359
 360     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 361     if (r < 0) {
 362         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 363         fflush(stderr);
 364     }
 365     assert(r >= 0);
 366 }
 367
 368 void vhost_net_cleanup(struct vhost_net *net)
 369 {
 370     vhost_dev_cleanup(&net->dev);
 371     g_free(net);
 372 }
 373
 374 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 375 {
 376     return vhost_virtqueue_pending(&net->dev, idx);
 377 }
 378
 379 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 380                               int idx, bool mask)
 381 {
 382     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 383 }
 384
 385 VHostNetState *get_vhost_net(NetClientState *nc)
 386 {
 387     VHostNetState *vhost_net = 0;
 388
 389     if (!nc) {
 390         return 0;
 391     }
 392
 393     switch (nc->info->type) {
 394     case NET_CLIENT_OPTIONS_KIND_TAP:
 395         vhost_net = tap_get_vhost_net(nc);
 396         break;
 397     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
 398         vhost_net = vhost_user_get_vhost_net(nc);
 399         break;
 400     default:
 401         break;
 402     }
 403
 404     return vhost_net;
 405 }
 406 #else
 407 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 408 {
 409     error_report("vhost-net support is not compiled in");
 410     return NULL;
 411 }
 412
 413 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev)
 414 {
 415     return false;
 416 }
 417
 418 int vhost_net_start(VirtIODevice *dev,
 419                     NetClientState *ncs,
 420                     int total_queues)
 421 {
 422     return -ENOSYS;
 423 }
 424 void vhost_net_stop(VirtIODevice *dev,
 425                     NetClientState *ncs,
 426                     int total_queues)
 427 {
 428 }
 429
 430 void vhost_net_cleanup(struct vhost_net *net)
 431 {
 432 }
 433
 434 unsigned vhost_net_get_features(struct vhost_net *net, unsigned features)
 435 {
 436     return features;
 437 }
 438 void vhost_net_ack_features(struct vhost_net *net, unsigned features)
 439 {
 440 }
 441
 442 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 443 {
 444     return false;
 445 }
 446
 447 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 448                               int idx, bool mask)
 449 {
 450 }
 451
 452 VHostNetState *get_vhost_net(NetClientState *nc)
 453 {
 454     return 0;
 455 }
 456 #endif