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 "standard-headers/linux/vhost_types.h"
  22 #include "hw/virtio/virtio-net.h"
  23 #include "net/vhost_net.h"
  24 #include "qemu/error-report.h"
  25 #include "qemu/main-loop.h"
  26
  27 #include <sys/socket.h>
  28 #include <net/if.h>
  29 #include <netinet/in.h>
  30
  31
  32 #include "standard-headers/linux/virtio_ring.h"
  33 #include "hw/virtio/vhost.h"
  34 #include "hw/virtio/virtio-bus.h"
  35
  36 struct vhost_net {
  37     struct vhost_dev dev;
  38     struct vhost_virtqueue vqs[2];
  39     int backend;
  40     NetClientState *nc;
  41 };
  42
  43 /* Features supported by host kernel. */
  44 static const int kernel_feature_bits[] = {
  45     VIRTIO_F_NOTIFY_ON_EMPTY,
  46     VIRTIO_RING_F_INDIRECT_DESC,
  47     VIRTIO_RING_F_EVENT_IDX,
  48     VIRTIO_NET_F_MRG_RXBUF,
  49     VIRTIO_F_VERSION_1,
  50     VIRTIO_NET_F_MTU,
  51     VIRTIO_F_IOMMU_PLATFORM,
  52     VIRTIO_F_RING_PACKED,
  53     VHOST_INVALID_FEATURE_BIT
  54 };
  55
  56 /* Features supported by others. */
  57 static const int user_feature_bits[] = {
  58     VIRTIO_F_NOTIFY_ON_EMPTY,
  59     VIRTIO_RING_F_INDIRECT_DESC,
  60     VIRTIO_RING_F_EVENT_IDX,
  61
  62     VIRTIO_F_ANY_LAYOUT,
  63     VIRTIO_F_VERSION_1,
  64     VIRTIO_NET_F_CSUM,
  65     VIRTIO_NET_F_GUEST_CSUM,
  66     VIRTIO_NET_F_GSO,
  67     VIRTIO_NET_F_GUEST_TSO4,
  68     VIRTIO_NET_F_GUEST_TSO6,
  69     VIRTIO_NET_F_GUEST_ECN,
  70     VIRTIO_NET_F_GUEST_UFO,
  71     VIRTIO_NET_F_HOST_TSO4,
  72     VIRTIO_NET_F_HOST_TSO6,
  73     VIRTIO_NET_F_HOST_ECN,
  74     VIRTIO_NET_F_HOST_UFO,
  75     VIRTIO_NET_F_MRG_RXBUF,
  76     VIRTIO_NET_F_MTU,
  77     VIRTIO_F_IOMMU_PLATFORM,
  78     VIRTIO_F_RING_PACKED,
  79
  80     /* This bit implies RARP isn't sent by QEMU out of band */
  81     VIRTIO_NET_F_GUEST_ANNOUNCE,
  82
  83     VIRTIO_NET_F_MQ,
  84
  85     VHOST_INVALID_FEATURE_BIT
  86 };
  87
  88 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
  89 {
  90     const int *feature_bits = 0;
  91
  92     switch (net->nc->info->type) {
  93     case NET_CLIENT_DRIVER_TAP:
  94         feature_bits = kernel_feature_bits;
  95         break;
  96     case NET_CLIENT_DRIVER_VHOST_USER:
  97         feature_bits = user_feature_bits;
  98         break;
  99     default:
 100         error_report("Feature bits not defined for this type: %d",
 101                 net->nc->info->type);
 102         break;
 103     }
 104
 105     return feature_bits;
 106 }
 107
 108 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 109 {
 110     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
 111             features);
 112 }
 113
 114 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 115 {
 116     net->dev.acked_features = net->dev.backend_features;
 117     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 118 }
 119
 120 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 121 {
 122     return net->dev.max_queues;
 123 }
 124
 125 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 126 {
 127     return net->dev.acked_features;
 128 }
 129
 130 static int vhost_net_get_fd(NetClientState *backend)
 131 {
 132     switch (backend->info->type) {
 133     case NET_CLIENT_DRIVER_TAP:
 134         return tap_get_fd(backend);
 135     default:
 136         fprintf(stderr, "vhost-net requires tap backend\n");
 137         return -ENOSYS;
 138     }
 139 }
 140
 141 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 142 {
 143     int r;
 144     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 145     struct vhost_net *net = g_new0(struct vhost_net, 1);
 146     uint64_t features = 0;
 147
 148     if (!options->net_backend) {
 149         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 150         goto fail;
 151     }
 152     net->nc = options->net_backend;
 153
 154     net->dev.max_queues = 1;
 155     net->dev.nvqs = 2;
 156     net->dev.vqs = net->vqs;
 157
 158     if (backend_kernel) {
 159         r = vhost_net_get_fd(options->net_backend);
 160         if (r < 0) {
 161             goto fail;
 162         }
 163         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 164             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
 165         net->backend = r;
 166         net->dev.protocol_features = 0;
 167     } else {
 168         net->dev.backend_features = 0;
 169         net->dev.protocol_features = 0;
 170         net->backend = -1;
 171
 172         /* vhost-user needs vq_index to initiate a specific queue pair */
 173         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
 174     }
 175
 176     r = vhost_dev_init(&net->dev, options->opaque,
 177                        options->backend_type, options->busyloop_timeout);
 178     if (r < 0) {
 179         goto fail;
 180     }
 181     if (backend_kernel) {
 182         if (!qemu_has_vnet_hdr_len(options->net_backend,
 183                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 184             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
 185         }
 186         if (~net->dev.features & net->dev.backend_features) {
 187             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 188                    " for backend\n",
 189                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 190             goto fail;
 191         }
 192     }
 193
 194     /* Set sane init value. Override when guest acks. */
 195 #ifdef CONFIG_VHOST_NET_USER
 196     if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 197         features = vhost_user_get_acked_features(net->nc);
 198         if (~net->dev.features & features) {
 199             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 200                     " for backend\n",
 201                     (uint64_t)(~net->dev.features & features));
 202             goto fail;
 203         }
 204     }
 205 #endif
 206
 207     vhost_net_ack_features(net, features);
 208
 209     return net;
 210
 211 fail:
 212     vhost_dev_cleanup(&net->dev);
 213     g_free(net);
 214     return NULL;
 215 }
 216
 217 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 218 {
 219     net->dev.vq_index = vq_index;
 220 }
 221
 222 static int vhost_net_start_one(struct vhost_net *net,
 223                                VirtIODevice *dev)
 224 {
 225     struct vhost_vring_file file = { };
 226     int r;
 227
 228     net->dev.nvqs = 2;
 229     net->dev.vqs = net->vqs;
 230
 231     r = vhost_dev_enable_notifiers(&net->dev, dev);
 232     if (r < 0) {
 233         goto fail_notifiers;
 234     }
 235
 236     r = vhost_dev_start(&net->dev, dev);
 237     if (r < 0) {
 238         goto fail_start;
 239     }
 240
 241     if (net->nc->info->poll) {
 242         net->nc->info->poll(net->nc, false);
 243     }
 244
 245     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 246         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 247         file.fd = net->backend;
 248         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 249             if (!virtio_queue_enabled(dev, net->dev.vq_index +
 250                                       file.index)) {
 251                 /* Queue might not be ready for start */
 252                 continue;
 253             }
 254             r = vhost_net_set_backend(&net->dev, &file);
 255             if (r < 0) {
 256                 r = -errno;
 257                 goto fail;
 258             }
 259         }
 260     }
 261     return 0;
 262 fail:
 263     file.fd = -1;
 264     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 265         while (file.index-- > 0) {
 266             if (!virtio_queue_enabled(dev, net->dev.vq_index +
 267                                       file.index)) {
 268                 /* Queue might not be ready for start */
 269                 continue;
 270             }
 271             int r = vhost_net_set_backend(&net->dev, &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 fail_start:
 280     vhost_dev_disable_notifiers(&net->dev, dev);
 281 fail_notifiers:
 282     return r;
 283 }
 284
 285 static void vhost_net_stop_one(struct vhost_net *net,
 286                                VirtIODevice *dev)
 287 {
 288     struct vhost_vring_file file = { .fd = -1 };
 289
 290     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 291         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 292             int r = vhost_net_set_backend(&net->dev, &file);
 293             assert(r >= 0);
 294         }
 295     }
 296     if (net->nc->info->poll) {
 297         net->nc->info->poll(net->nc, true);
 298     }
 299     vhost_dev_stop(&net->dev, dev);
 300     vhost_dev_disable_notifiers(&net->dev, dev);
 301 }
 302
 303 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 304                     int total_queues)
 305 {
 306     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 307     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 308     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 309     int r, e, i;
 310
 311     if (!k->set_guest_notifiers) {
 312         error_report("binding does not support guest notifiers");
 313         return -ENOSYS;
 314     }
 315
 316     for (i = 0; i < total_queues; i++) {
 317         struct vhost_net *net;
 318
 319         net = get_vhost_net(ncs[i].peer);
 320         vhost_net_set_vq_index(net, i * 2);
 321
 322         /* Suppress the masking guest notifiers on vhost user
 323          * because vhost user doesn't interrupt masking/unmasking
 324          * properly.
 325          */
 326         if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 327             dev->use_guest_notifier_mask = false;
 328         }
 329      }
 330
 331     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 332     if (r < 0) {
 333         error_report("Error binding guest notifier: %d", -r);
 334         goto err;
 335     }
 336
 337     for (i = 0; i < total_queues; i++) {
 338         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 339
 340         if (r < 0) {
 341             goto err_start;
 342         }
 343
 344         if (ncs[i].peer->vring_enable) {
 345             /* restore vring enable state */
 346             r = vhost_set_vring_enable(ncs[i].peer, ncs[i].peer->vring_enable);
 347
 348             if (r < 0) {
 349                 goto err_start;
 350             }
 351         }
 352     }
 353
 354     return 0;
 355
 356 err_start:
 357     while (--i >= 0) {
 358         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 359     }
 360     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 361     if (e < 0) {
 362         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 363         fflush(stderr);
 364     }
 365 err:
 366     return r;
 367 }
 368
 369 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 370                     int total_queues)
 371 {
 372     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 373     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 374     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 375     int i, r;
 376
 377     for (i = 0; i < total_queues; i++) {
 378         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 379     }
 380
 381     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 382     if (r < 0) {
 383         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 384         fflush(stderr);
 385     }
 386     assert(r >= 0);
 387 }
 388
 389 void vhost_net_cleanup(struct vhost_net *net)
 390 {
 391     vhost_dev_cleanup(&net->dev);
 392 }
 393
 394 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 395 {
 396     const VhostOps *vhost_ops = net->dev.vhost_ops;
 397
 398     assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 399     assert(vhost_ops->vhost_migration_done);
 400
 401     return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 402 }
 403
 404 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 405 {
 406     return vhost_virtqueue_pending(&net->dev, idx);
 407 }
 408
 409 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 410                               int idx, bool mask)
 411 {
 412     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 413 }
 414
 415 VHostNetState *get_vhost_net(NetClientState *nc)
 416 {
 417     VHostNetState *vhost_net = 0;
 418
 419     if (!nc) {
 420         return 0;
 421     }
 422
 423     switch (nc->info->type) {
 424     case NET_CLIENT_DRIVER_TAP:
 425         vhost_net = tap_get_vhost_net(nc);
 426         break;
 427 #ifdef CONFIG_VHOST_NET_USER
 428     case NET_CLIENT_DRIVER_VHOST_USER:
 429         vhost_net = vhost_user_get_vhost_net(nc);
 430         assert(vhost_net);
 431         break;
 432 #endif
 433     default:
 434         break;
 435     }
 436
 437     return vhost_net;
 438 }
 439
 440 int vhost_set_vring_enable(NetClientState *nc, int enable)
 441 {
 442     VHostNetState *net = get_vhost_net(nc);
 443     const VhostOps *vhost_ops = net->dev.vhost_ops;
 444
 445     nc->vring_enable = enable;
 446
 447     if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
 448         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 449     }
 450
 451     return 0;
 452 }
 453
 454 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
 455 {
 456     const VhostOps *vhost_ops = net->dev.vhost_ops;
 457
 458     if (!vhost_ops->vhost_net_set_mtu) {
 459         return 0;
 460     }
 461
 462     return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
 463 }