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 #include "net/vhost-vdpa.h"
  21
  22 #include "standard-headers/linux/vhost_types.h"
  23 #include "hw/virtio/virtio-net.h"
  24 #include "net/vhost_net.h"
  25 #include "qapi/error.h"
  26 #include "qemu/error-report.h"
  27 #include "qemu/main-loop.h"
  28
  29 #include <sys/socket.h>
  30 #include <net/if.h>
  31 #include <netinet/in.h>
  32
  33
  34 #include "standard-headers/linux/virtio_ring.h"
  35 #include "hw/virtio/vhost.h"
  36 #include "hw/virtio/virtio-bus.h"
  37
  38
  39 /* Features supported by host kernel. */
  40 static const int kernel_feature_bits[] = {
  41     VIRTIO_F_NOTIFY_ON_EMPTY,
  42     VIRTIO_RING_F_INDIRECT_DESC,
  43     VIRTIO_RING_F_EVENT_IDX,
  44     VIRTIO_NET_F_MRG_RXBUF,
  45     VIRTIO_F_VERSION_1,
  46     VIRTIO_NET_F_MTU,
  47     VIRTIO_F_IOMMU_PLATFORM,
  48     VIRTIO_F_RING_PACKED,
  49     VIRTIO_NET_F_HASH_REPORT,
  50     VHOST_INVALID_FEATURE_BIT
  51 };
  52
  53 /* Features supported by others. */
  54 static const int user_feature_bits[] = {
  55     VIRTIO_F_NOTIFY_ON_EMPTY,
  56     VIRTIO_RING_F_INDIRECT_DESC,
  57     VIRTIO_RING_F_EVENT_IDX,
  58
  59     VIRTIO_F_ANY_LAYOUT,
  60     VIRTIO_F_VERSION_1,
  61     VIRTIO_NET_F_CSUM,
  62     VIRTIO_NET_F_GUEST_CSUM,
  63     VIRTIO_NET_F_GSO,
  64     VIRTIO_NET_F_GUEST_TSO4,
  65     VIRTIO_NET_F_GUEST_TSO6,
  66     VIRTIO_NET_F_GUEST_ECN,
  67     VIRTIO_NET_F_GUEST_UFO,
  68     VIRTIO_NET_F_HOST_TSO4,
  69     VIRTIO_NET_F_HOST_TSO6,
  70     VIRTIO_NET_F_HOST_ECN,
  71     VIRTIO_NET_F_HOST_UFO,
  72     VIRTIO_NET_F_MRG_RXBUF,
  73     VIRTIO_NET_F_MTU,
  74     VIRTIO_F_IOMMU_PLATFORM,
  75     VIRTIO_F_RING_PACKED,
  76     VIRTIO_NET_F_RSS,
  77     VIRTIO_NET_F_HASH_REPORT,
  78
  79     /* This bit implies RARP isn't sent by QEMU out of band */
  80     VIRTIO_NET_F_GUEST_ANNOUNCE,
  81
  82     VIRTIO_NET_F_MQ,
  83
  84     VHOST_INVALID_FEATURE_BIT
  85 };
  86
  87 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
  88 {
  89     const int *feature_bits = 0;
  90
  91     switch (net->nc->info->type) {
  92     case NET_CLIENT_DRIVER_TAP:
  93         feature_bits = kernel_feature_bits;
  94         break;
  95     case NET_CLIENT_DRIVER_VHOST_USER:
  96         feature_bits = user_feature_bits;
  97         break;
  98 #ifdef CONFIG_VHOST_NET_VDPA
  99     case NET_CLIENT_DRIVER_VHOST_VDPA:
 100         feature_bits = vdpa_feature_bits;
 101         break;
 102 #endif
 103     default:
 104         error_report("Feature bits not defined for this type: %d",
 105                 net->nc->info->type);
 106         break;
 107     }
 108
 109     return feature_bits;
 110 }
 111
 112 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 113 {
 114     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
 115             features);
 116 }
 117 int vhost_net_get_config(struct vhost_net *net,  uint8_t *config,
 118                          uint32_t config_len)
 119 {
 120     return vhost_dev_get_config(&net->dev, config, config_len, NULL);
 121 }
 122 int vhost_net_set_config(struct vhost_net *net, const uint8_t *data,
 123                          uint32_t offset, uint32_t size, uint32_t flags)
 124 {
 125     return vhost_dev_set_config(&net->dev, data, offset, size, flags);
 126 }
 127
 128 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 129 {
 130     net->dev.acked_features = net->dev.backend_features;
 131     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 132 }
 133
 134 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 135 {
 136     return net->dev.max_queues;
 137 }
 138
 139 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 140 {
 141     return net->dev.acked_features;
 142 }
 143
 144 static int vhost_net_get_fd(NetClientState *backend)
 145 {
 146     switch (backend->info->type) {
 147     case NET_CLIENT_DRIVER_TAP:
 148         return tap_get_fd(backend);
 149     default:
 150         fprintf(stderr, "vhost-net requires tap backend\n");
 151         return -ENOSYS;
 152     }
 153 }
 154
 155 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 156 {
 157     int r;
 158     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 159     struct vhost_net *net = g_new0(struct vhost_net, 1);
 160     uint64_t features = 0;
 161     Error *local_err = NULL;
 162
 163     if (!options->net_backend) {
 164         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 165         goto fail;
 166     }
 167     net->nc = options->net_backend;
 168
 169     net->dev.max_queues = 1;
 170     net->dev.nvqs = 2;
 171     net->dev.vqs = net->vqs;
 172
 173     if (backend_kernel) {
 174         r = vhost_net_get_fd(options->net_backend);
 175         if (r < 0) {
 176             goto fail;
 177         }
 178         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 179             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
 180         net->backend = r;
 181         net->dev.protocol_features = 0;
 182     } else {
 183         net->dev.backend_features = 0;
 184         net->dev.protocol_features = 0;
 185         net->backend = -1;
 186
 187         /* vhost-user needs vq_index to initiate a specific queue pair */
 188         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
 189     }
 190
 191     r = vhost_dev_init(&net->dev, options->opaque,
 192                        options->backend_type, options->busyloop_timeout,
 193                        &local_err);
 194     if (r < 0) {
 195         error_report_err(local_err);
 196         goto fail;
 197     }
 198     if (backend_kernel) {
 199         if (!qemu_has_vnet_hdr_len(options->net_backend,
 200                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 201             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
 202         }
 203         if (~net->dev.features & net->dev.backend_features) {
 204             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 205                    " for backend\n",
 206                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 207             goto fail;
 208         }
 209     }
 210
 211     /* Set sane init value. Override when guest acks. */
 212 #ifdef CONFIG_VHOST_NET_USER
 213     if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 214         features = vhost_user_get_acked_features(net->nc);
 215         if (~net->dev.features & features) {
 216             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 217                     " for backend\n",
 218                     (uint64_t)(~net->dev.features & features));
 219             goto fail;
 220         }
 221     }
 222 #endif
 223
 224     vhost_net_ack_features(net, features);
 225
 226     return net;
 227
 228 fail:
 229     vhost_dev_cleanup(&net->dev);
 230     g_free(net);
 231     return NULL;
 232 }
 233
 234 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 235 {
 236     net->dev.vq_index = vq_index;
 237 }
 238
 239 static int vhost_net_start_one(struct vhost_net *net,
 240                                VirtIODevice *dev)
 241 {
 242     struct vhost_vring_file file = { };
 243     int r;
 244
 245     r = vhost_dev_enable_notifiers(&net->dev, dev);
 246     if (r < 0) {
 247         goto fail_notifiers;
 248     }
 249
 250     r = vhost_dev_start(&net->dev, dev);
 251     if (r < 0) {
 252         goto fail_start;
 253     }
 254
 255     if (net->nc->info->poll) {
 256         net->nc->info->poll(net->nc, false);
 257     }
 258
 259     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 260         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 261         file.fd = net->backend;
 262         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 263             if (!virtio_queue_enabled(dev, net->dev.vq_index +
 264                                       file.index)) {
 265                 /* Queue might not be ready for start */
 266                 continue;
 267             }
 268             r = vhost_net_set_backend(&net->dev, &file);
 269             if (r < 0) {
 270                 r = -errno;
 271                 goto fail;
 272             }
 273         }
 274     }
 275     return 0;
 276 fail:
 277     file.fd = -1;
 278     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 279         while (file.index-- > 0) {
 280             if (!virtio_queue_enabled(dev, net->dev.vq_index +
 281                                       file.index)) {
 282                 /* Queue might not be ready for start */
 283                 continue;
 284             }
 285             int r = vhost_net_set_backend(&net->dev, &file);
 286             assert(r >= 0);
 287         }
 288     }
 289     if (net->nc->info->poll) {
 290         net->nc->info->poll(net->nc, true);
 291     }
 292     vhost_dev_stop(&net->dev, dev);
 293 fail_start:
 294     vhost_dev_disable_notifiers(&net->dev, dev);
 295 fail_notifiers:
 296     return r;
 297 }
 298
 299 static void vhost_net_stop_one(struct vhost_net *net,
 300                                VirtIODevice *dev)
 301 {
 302     struct vhost_vring_file file = { .fd = -1 };
 303
 304     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 305         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 306             int r = vhost_net_set_backend(&net->dev, &file);
 307             assert(r >= 0);
 308         }
 309     }
 310     if (net->nc->info->poll) {
 311         net->nc->info->poll(net->nc, true);
 312     }
 313     vhost_dev_stop(&net->dev, dev);
 314     vhost_dev_disable_notifiers(&net->dev, dev);
 315 }
 316
 317 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 318                     int total_queues)
 319 {
 320     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 321     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 322     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 323     struct vhost_net *net;
 324     int r, e, i;
 325     NetClientState *peer;
 326
 327     if (!k->set_guest_notifiers) {
 328         error_report("binding does not support guest notifiers");
 329         return -ENOSYS;
 330     }
 331
 332     for (i = 0; i < total_queues; i++) {
 333
 334         peer = qemu_get_peer(ncs, i);
 335         net = get_vhost_net(peer);
 336         vhost_net_set_vq_index(net, i * 2);
 337
 338         /* Suppress the masking guest notifiers on vhost user
 339          * because vhost user doesn't interrupt masking/unmasking
 340          * properly.
 341          */
 342         if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 343             dev->use_guest_notifier_mask = false;
 344         }
 345      }
 346
 347     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 348     if (r < 0) {
 349         error_report("Error binding guest notifier: %d", -r);
 350         goto err;
 351     }
 352
 353     for (i = 0; i < total_queues; i++) {
 354         peer = qemu_get_peer(ncs, i);
 355         r = vhost_net_start_one(get_vhost_net(peer), dev);
 356
 357         if (r < 0) {
 358             goto err_start;
 359         }
 360
 361         if (peer->vring_enable) {
 362             /* restore vring enable state */
 363             r = vhost_set_vring_enable(peer, peer->vring_enable);
 364
 365             if (r < 0) {
 366                 goto err_start;
 367             }
 368         }
 369     }
 370
 371     return 0;
 372
 373 err_start:
 374     while (--i >= 0) {
 375         peer = qemu_get_peer(ncs , i);
 376         vhost_net_stop_one(get_vhost_net(peer), dev);
 377     }
 378     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 379     if (e < 0) {
 380         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 381         fflush(stderr);
 382     }
 383 err:
 384     return r;
 385 }
 386
 387 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 388                     int total_queues)
 389 {
 390     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 391     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 392     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 393     int i, r;
 394
 395     for (i = 0; i < total_queues; i++) {
 396         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 397     }
 398
 399     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 400     if (r < 0) {
 401         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 402         fflush(stderr);
 403     }
 404     assert(r >= 0);
 405 }
 406
 407 void vhost_net_cleanup(struct vhost_net *net)
 408 {
 409     vhost_dev_cleanup(&net->dev);
 410 }
 411
 412 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 413 {
 414     const VhostOps *vhost_ops = net->dev.vhost_ops;
 415
 416     assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 417     assert(vhost_ops->vhost_migration_done);
 418
 419     return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 420 }
 421
 422 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 423 {
 424     return vhost_virtqueue_pending(&net->dev, idx);
 425 }
 426
 427 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 428                               int idx, bool mask)
 429 {
 430     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 431 }
 432
 433 VHostNetState *get_vhost_net(NetClientState *nc)
 434 {
 435     VHostNetState *vhost_net = 0;
 436
 437     if (!nc) {
 438         return 0;
 439     }
 440
 441     switch (nc->info->type) {
 442     case NET_CLIENT_DRIVER_TAP:
 443         vhost_net = tap_get_vhost_net(nc);
 444         break;
 445 #ifdef CONFIG_VHOST_NET_USER
 446     case NET_CLIENT_DRIVER_VHOST_USER:
 447         vhost_net = vhost_user_get_vhost_net(nc);
 448         assert(vhost_net);
 449         break;
 450 #endif
 451 #ifdef CONFIG_VHOST_NET_VDPA
 452     case NET_CLIENT_DRIVER_VHOST_VDPA:
 453         vhost_net = vhost_vdpa_get_vhost_net(nc);
 454         assert(vhost_net);
 455         break;
 456 #endif
 457     default:
 458         break;
 459     }
 460
 461     return vhost_net;
 462 }
 463
 464 int vhost_set_vring_enable(NetClientState *nc, int enable)
 465 {
 466     VHostNetState *net = get_vhost_net(nc);
 467     const VhostOps *vhost_ops = net->dev.vhost_ops;
 468
 469     nc->vring_enable = enable;
 470
 471     if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
 472         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 473     }
 474
 475     return 0;
 476 }
 477
 478 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
 479 {
 480     const VhostOps *vhost_ops = net->dev.vhost_ops;
 481
 482     if (!vhost_ops->vhost_net_set_mtu) {
 483         return 0;
 484     }
 485
 486     return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
 487 }