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_MQ,
  89
  90     VHOST_INVALID_FEATURE_BIT
  91 };
  92
  93 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
  94 {
  95     const int *feature_bits = 0;
  96
  97     switch (net->nc->info->type) {
  98     case NET_CLIENT_OPTIONS_KIND_TAP:
  99         feature_bits = kernel_feature_bits;
 100         break;
 101     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
 102         feature_bits = user_feature_bits;
 103         break;
 104     default:
 105         error_report("Feature bits not defined for this type: %d",
 106                 net->nc->info->type);
 107         break;
 108     }
 109
 110     return feature_bits;
 111 }
 112
 113 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 114 {
 115     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
 116             features);
 117 }
 118
 119 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 120 {
 121     net->dev.acked_features = net->dev.backend_features;
 122     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 123 }
 124
 125 static int vhost_net_get_fd(NetClientState *backend)
 126 {
 127     switch (backend->info->type) {
 128     case NET_CLIENT_OPTIONS_KIND_TAP:
 129         return tap_get_fd(backend);
 130     default:
 131         fprintf(stderr, "vhost-net requires tap backend\n");
 132         return -EBADFD;
 133     }
 134 }
 135
 136 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 137 {
 138     int r;
 139     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 140     struct vhost_net *net = g_malloc(sizeof *net);
 141
 142     if (!options->net_backend) {
 143         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 144         goto fail;
 145     }
 146
 147     if (backend_kernel) {
 148         r = vhost_net_get_fd(options->net_backend);
 149         if (r < 0) {
 150             goto fail;
 151         }
 152         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 153             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
 154         net->backend = r;
 155     } else {
 156         net->dev.backend_features = 0;
 157         net->backend = -1;
 158     }
 159     net->nc = options->net_backend;
 160
 161     net->dev.nvqs = 2;
 162     net->dev.vqs = net->vqs;
 163
 164     r = vhost_dev_init(&net->dev, options->opaque,
 165                        options->backend_type);
 166     if (r < 0) {
 167         goto fail;
 168     }
 169     if (backend_kernel) {
 170         if (!qemu_has_vnet_hdr_len(options->net_backend,
 171                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 172             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
 173         }
 174         if (~net->dev.features & net->dev.backend_features) {
 175             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 176                    " for backend\n",
 177                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 178             vhost_dev_cleanup(&net->dev);
 179             goto fail;
 180         }
 181     }
 182     /* Set sane init value. Override when guest acks. */
 183     vhost_net_ack_features(net, 0);
 184     return net;
 185 fail:
 186     g_free(net);
 187     return NULL;
 188 }
 189
 190 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 191 {
 192     net->dev.vq_index = vq_index;
 193 }
 194
 195 static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
 196                                      bool set)
 197 {
 198     int r = 0;
 199
 200     if (virtio_has_feature(dev, VIRTIO_F_VERSION_1) ||
 201         (virtio_legacy_is_cross_endian(dev) && !virtio_is_big_endian(dev))) {
 202         r = qemu_set_vnet_le(peer, set);
 203         if (r) {
 204             error_report("backend does not support LE vnet headers");
 205         }
 206     } else if (virtio_legacy_is_cross_endian(dev)) {
 207         r = qemu_set_vnet_be(peer, set);
 208         if (r) {
 209             error_report("backend does not support BE vnet headers");
 210         }
 211     }
 212
 213     return r;
 214 }
 215
 216 static int vhost_net_start_one(struct vhost_net *net,
 217                                VirtIODevice *dev)
 218 {
 219     struct vhost_vring_file file = { };
 220     int r;
 221
 222     net->dev.nvqs = 2;
 223     net->dev.vqs = net->vqs;
 224
 225     r = vhost_dev_enable_notifiers(&net->dev, dev);
 226     if (r < 0) {
 227         goto fail_notifiers;
 228     }
 229
 230     r = vhost_dev_start(&net->dev, dev);
 231     if (r < 0) {
 232         goto fail_start;
 233     }
 234
 235     if (net->nc->info->poll) {
 236         net->nc->info->poll(net->nc, false);
 237     }
 238
 239     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 240         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 241         file.fd = net->backend;
 242         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 243             const VhostOps *vhost_ops = net->dev.vhost_ops;
 244             r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
 245                                       &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_OPTIONS_KIND_TAP) {
 256         while (file.index-- > 0) {
 257             const VhostOps *vhost_ops = net->dev.vhost_ops;
 258             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
 259                                           &file);
 260             assert(r >= 0);
 261         }
 262     }
 263     if (net->nc->info->poll) {
 264         net->nc->info->poll(net->nc, true);
 265     }
 266     vhost_dev_stop(&net->dev, dev);
 267 fail_start:
 268     vhost_dev_disable_notifiers(&net->dev, dev);
 269 fail_notifiers:
 270     return r;
 271 }
 272
 273 static void vhost_net_stop_one(struct vhost_net *net,
 274                                VirtIODevice *dev)
 275 {
 276     struct vhost_vring_file file = { .fd = -1 };
 277
 278     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 279         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 280             const VhostOps *vhost_ops = net->dev.vhost_ops;
 281             int r = vhost_ops->vhost_call(&net->dev, VHOST_NET_SET_BACKEND,
 282                                           &file);
 283             assert(r >= 0);
 284         }
 285     } else if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
 286         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 287             const VhostOps *vhost_ops = net->dev.vhost_ops;
 288             int r = vhost_ops->vhost_call(&net->dev, VHOST_RESET_OWNER,
 289                                           NULL);
 290             assert(r >= 0);
 291         }
 292     }
 293     if (net->nc->info->poll) {
 294         net->nc->info->poll(net->nc, true);
 295     }
 296     vhost_dev_stop(&net->dev, dev);
 297     vhost_dev_disable_notifiers(&net->dev, dev);
 298 }
 299
 300 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 301                     int total_queues)
 302 {
 303     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 304     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 305     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 306     int r, e, i;
 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     r = vhost_net_set_vnet_endian(dev, ncs[0].peer, true);
 315     if (r < 0) {
 316         goto err;
 317     }
 318
 319     for (i = 0; i < total_queues; i++) {
 320         vhost_net_set_vq_index(get_vhost_net(ncs[i].peer), i * 2);
 321     }
 322
 323     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 324     if (r < 0) {
 325         error_report("Error binding guest notifier: %d", -r);
 326         goto err_endian;
 327     }
 328
 329     for (i = 0; i < total_queues; i++) {
 330         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 331
 332         if (r < 0) {
 333             goto err_start;
 334         }
 335     }
 336
 337     return 0;
 338
 339 err_start:
 340     while (--i >= 0) {
 341         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 342     }
 343     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 344     if (e < 0) {
 345         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 346         fflush(stderr);
 347     }
 348 err_endian:
 349     vhost_net_set_vnet_endian(dev, ncs[0].peer, false);
 350 err:
 351     return r;
 352 }
 353
 354 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 355                     int total_queues)
 356 {
 357     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 358     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 359     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 360     int i, r;
 361
 362     for (i = 0; i < total_queues; i++) {
 363         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 364     }
 365
 366     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 367     if (r < 0) {
 368         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 369         fflush(stderr);
 370     }
 371     assert(r >= 0);
 372
 373     assert(vhost_net_set_vnet_endian(dev, ncs[0].peer, false) >= 0);
 374 }
 375
 376 void vhost_net_cleanup(struct vhost_net *net)
 377 {
 378     vhost_dev_cleanup(&net->dev);
 379     g_free(net);
 380 }
 381
 382 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 383 {
 384     return vhost_virtqueue_pending(&net->dev, idx);
 385 }
 386
 387 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 388                               int idx, bool mask)
 389 {
 390     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 391 }
 392
 393 VHostNetState *get_vhost_net(NetClientState *nc)
 394 {
 395     VHostNetState *vhost_net = 0;
 396
 397     if (!nc) {
 398         return 0;
 399     }
 400
 401     switch (nc->info->type) {
 402     case NET_CLIENT_OPTIONS_KIND_TAP:
 403         vhost_net = tap_get_vhost_net(nc);
 404         break;
 405     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
 406         vhost_net = vhost_user_get_vhost_net(nc);
 407         break;
 408     default:
 409         break;
 410     }
 411
 412     return vhost_net;
 413 }
 414 #else
 415 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 416 {
 417     error_report("vhost-net support is not compiled in");
 418     return NULL;
 419 }
 420
 421 int vhost_net_start(VirtIODevice *dev,
 422                     NetClientState *ncs,
 423                     int total_queues)
 424 {
 425     return -ENOSYS;
 426 }
 427 void vhost_net_stop(VirtIODevice *dev,
 428                     NetClientState *ncs,
 429                     int total_queues)
 430 {
 431 }
 432
 433 void vhost_net_cleanup(struct vhost_net *net)
 434 {
 435 }
 436
 437 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 438 {
 439     return features;
 440 }
 441 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 442 {
 443 }
 444
 445 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 446 {
 447     return false;
 448 }
 449
 450 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 451                               int idx, bool mask)
 452 {
 453 }
 454
 455 VHostNetState *get_vhost_net(NetClientState *nc)
 456 {
 457     return 0;
 458 }
 459 #endif