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 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 124 {
 125     return net->dev.acked_features;
 126 }
 127
 128 static int vhost_net_get_fd(NetClientState *backend)
 129 {
 130     switch (backend->info->type) {
 131     case NET_CLIENT_OPTIONS_KIND_TAP:
 132         return tap_get_fd(backend);
 133     default:
 134         fprintf(stderr, "vhost-net requires tap backend\n");
 135         return -EBADFD;
 136     }
 137 }
 138
 139 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 140 {
 141     int r;
 142     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 143     struct vhost_net *net = g_malloc(sizeof *net);
 144     uint64_t features = 0;
 145
 146     if (!options->net_backend) {
 147         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 148         goto fail;
 149     }
 150     net->nc = options->net_backend;
 151
 152     net->dev.max_queues = 1;
 153     net->dev.nvqs = 2;
 154     net->dev.vqs = net->vqs;
 155
 156     if (backend_kernel) {
 157         r = vhost_net_get_fd(options->net_backend);
 158         if (r < 0) {
 159             goto fail;
 160         }
 161         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 162             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
 163         net->backend = r;
 164         net->dev.protocol_features = 0;
 165     } else {
 166         net->dev.backend_features = 0;
 167         net->dev.protocol_features = 0;
 168         net->backend = -1;
 169
 170         /* vhost-user needs vq_index to initiate a specific queue pair */
 171         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
 172     }
 173
 174     r = vhost_dev_init(&net->dev, options->opaque,
 175                        options->backend_type);
 176     if (r < 0) {
 177         goto fail;
 178     }
 179     if (backend_kernel) {
 180         if (!qemu_has_vnet_hdr_len(options->net_backend,
 181                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 182             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
 183         }
 184         if (~net->dev.features & net->dev.backend_features) {
 185             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 186                    " for backend\n",
 187                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 188             vhost_dev_cleanup(&net->dev);
 189             goto fail;
 190         }
 191     }
 192
 193     /* Set sane init value. Override when guest acks. */
 194     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
 195         features = vhost_user_get_acked_features(net->nc);
 196         if (~net->dev.features & features) {
 197             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 198                     " for backend\n",
 199                     (uint64_t)(~net->dev.features & features));
 200             vhost_dev_cleanup(&net->dev);
 201             goto fail;
 202         }
 203     }
 204
 205     vhost_net_ack_features(net, features);
 206
 207     return net;
 208 fail:
 209     g_free(net);
 210     return NULL;
 211 }
 212
 213 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 214 {
 215     net->dev.vq_index = vq_index;
 216 }
 217
 218 static int vhost_net_start_one(struct vhost_net *net,
 219                                VirtIODevice *dev)
 220 {
 221     struct vhost_vring_file file = { };
 222     int r;
 223
 224     net->dev.nvqs = 2;
 225     net->dev.vqs = net->vqs;
 226
 227     r = vhost_dev_enable_notifiers(&net->dev, dev);
 228     if (r < 0) {
 229         goto fail_notifiers;
 230     }
 231
 232     r = vhost_dev_start(&net->dev, dev);
 233     if (r < 0) {
 234         goto fail_start;
 235     }
 236
 237     if (net->nc->info->poll) {
 238         net->nc->info->poll(net->nc, false);
 239     }
 240
 241     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 242         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 243         file.fd = net->backend;
 244         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 245             const VhostOps *vhost_ops = net->dev.vhost_ops;
 246             r = vhost_ops->vhost_net_set_backend(&net->dev, &file);
 247             if (r < 0) {
 248                 r = -errno;
 249                 goto fail;
 250             }
 251         }
 252     }
 253     return 0;
 254 fail:
 255     file.fd = -1;
 256     if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP) {
 257         while (file.index-- > 0) {
 258             const VhostOps *vhost_ops = net->dev.vhost_ops;
 259             int r = vhost_ops->vhost_net_set_backend(&net->dev, &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_net_set_backend(&net->dev, &file);
 282             assert(r >= 0);
 283         }
 284     }
 285     if (net->nc->info->poll) {
 286         net->nc->info->poll(net->nc, true);
 287     }
 288     vhost_dev_stop(&net->dev, dev);
 289     vhost_dev_disable_notifiers(&net->dev, dev);
 290 }
 291
 292 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 293                     int total_queues)
 294 {
 295     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 296     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 297     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 298     int r, e, i;
 299
 300     if (!k->set_guest_notifiers) {
 301         error_report("binding does not support guest notifiers");
 302         return -ENOSYS;
 303     }
 304
 305     for (i = 0; i < total_queues; i++) {
 306         struct vhost_net *net;
 307
 308         net = get_vhost_net(ncs[i].peer);
 309         vhost_net_set_vq_index(net, i * 2);
 310
 311         /* Suppress the masking guest notifiers on vhost user
 312          * because vhost user doesn't interrupt masking/unmasking
 313          * properly.
 314          */
 315         if (net->nc->info->type == NET_CLIENT_OPTIONS_KIND_VHOST_USER) {
 316                 dev->use_guest_notifier_mask = false;
 317         }
 318      }
 319
 320     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 321     if (r < 0) {
 322         error_report("Error binding guest notifier: %d", -r);
 323         goto err;
 324     }
 325
 326     for (i = 0; i < total_queues; i++) {
 327         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 328
 329         if (r < 0) {
 330             goto err_start;
 331         }
 332
 333         if (ncs[i].peer->vring_enable) {
 334             /* restore vring enable state */
 335             r = vhost_set_vring_enable(ncs[i].peer, ncs[i].peer->vring_enable);
 336
 337             if (r < 0) {
 338                 goto err_start;
 339             }
 340         }
 341     }
 342
 343     return 0;
 344
 345 err_start:
 346     while (--i >= 0) {
 347         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 348     }
 349     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 350     if (e < 0) {
 351         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 352         fflush(stderr);
 353     }
 354 err:
 355     return r;
 356 }
 357
 358 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 359                     int total_queues)
 360 {
 361     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 362     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 363     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 364     int i, r;
 365
 366     for (i = 0; i < total_queues; i++) {
 367         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 368     }
 369
 370     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 371     if (r < 0) {
 372         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 373         fflush(stderr);
 374     }
 375     assert(r >= 0);
 376 }
 377
 378 void vhost_net_cleanup(struct vhost_net *net)
 379 {
 380     vhost_dev_cleanup(&net->dev);
 381     g_free(net);
 382 }
 383
 384 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 385 {
 386     const VhostOps *vhost_ops = net->dev.vhost_ops;
 387     int r = -1;
 388
 389     if (vhost_ops->vhost_migration_done) {
 390         r = vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 391     }
 392
 393     return r;
 394 }
 395
 396 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 397 {
 398     return vhost_virtqueue_pending(&net->dev, idx);
 399 }
 400
 401 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 402                               int idx, bool mask)
 403 {
 404     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 405 }
 406
 407 VHostNetState *get_vhost_net(NetClientState *nc)
 408 {
 409     VHostNetState *vhost_net = 0;
 410
 411     if (!nc) {
 412         return 0;
 413     }
 414
 415     switch (nc->info->type) {
 416     case NET_CLIENT_OPTIONS_KIND_TAP:
 417         vhost_net = tap_get_vhost_net(nc);
 418         break;
 419     case NET_CLIENT_OPTIONS_KIND_VHOST_USER:
 420         vhost_net = vhost_user_get_vhost_net(nc);
 421         break;
 422     default:
 423         break;
 424     }
 425
 426     return vhost_net;
 427 }
 428
 429 int vhost_set_vring_enable(NetClientState *nc, int enable)
 430 {
 431     VHostNetState *net = get_vhost_net(nc);
 432     const VhostOps *vhost_ops;
 433
 434     nc->vring_enable = enable;
 435
 436     if (!net) {
 437         return 0;
 438     }
 439
 440     vhost_ops = net->dev.vhost_ops;
 441     if (vhost_ops->vhost_set_vring_enable) {
 442         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 443     }
 444
 445     return 0;
 446 }
 447
 448 #else
 449 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 450 {
 451     return 1;
 452 }
 453
 454 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 455 {
 456     error_report("vhost-net support is not compiled in");
 457     return NULL;
 458 }
 459
 460 int vhost_net_start(VirtIODevice *dev,
 461                     NetClientState *ncs,
 462                     int total_queues)
 463 {
 464     return -ENOSYS;
 465 }
 466 void vhost_net_stop(VirtIODevice *dev,
 467                     NetClientState *ncs,
 468                     int total_queues)
 469 {
 470 }
 471
 472 void vhost_net_cleanup(struct vhost_net *net)
 473 {
 474 }
 475
 476 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 477 {
 478     return features;
 479 }
 480
 481 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 482 {
 483 }
 484
 485 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 486 {
 487     return 0;
 488 }
 489
 490 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 491 {
 492     return false;
 493 }
 494
 495 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 496                               int idx, bool mask)
 497 {
 498 }
 499
 500 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 501 {
 502     return -1;
 503 }
 504
 505 VHostNetState *get_vhost_net(NetClientState *nc)
 506 {
 507     return 0;
 508 }
 509
 510 int vhost_set_vring_enable(NetClientState *nc, int enable)
 511 {
 512     return 0;
 513 }
 514 #endif