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     net->dev.nvqs = options->nvqs;
 169
 170     net->dev.max_queues = 1;
 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                                    int vq_index_end)
 236 {
 237     net->dev.vq_index = vq_index;
 238     net->dev.vq_index_end = vq_index_end;
 239 }
 240
 241 static int vhost_net_start_one(struct vhost_net *net,
 242                                VirtIODevice *dev)
 243 {
 244     struct vhost_vring_file file = { };
 245     int r;
 246
 247     r = vhost_dev_enable_notifiers(&net->dev, dev);
 248     if (r < 0) {
 249         goto fail_notifiers;
 250     }
 251
 252     r = vhost_dev_start(&net->dev, dev);
 253     if (r < 0) {
 254         goto fail_start;
 255     }
 256
 257     if (net->nc->info->poll) {
 258         net->nc->info->poll(net->nc, false);
 259     }
 260
 261     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 262         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 263         file.fd = net->backend;
 264         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 265             if (!virtio_queue_enabled(dev, net->dev.vq_index +
 266                                       file.index)) {
 267                 /* Queue might not be ready for start */
 268                 continue;
 269             }
 270             r = vhost_net_set_backend(&net->dev, &file);
 271             if (r < 0) {
 272                 r = -errno;
 273                 goto fail;
 274             }
 275         }
 276     }
 277     return 0;
 278 fail:
 279     file.fd = -1;
 280     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 281         while (file.index-- > 0) {
 282             if (!virtio_queue_enabled(dev, net->dev.vq_index +
 283                                       file.index)) {
 284                 /* Queue might not be ready for start */
 285                 continue;
 286             }
 287             int r = vhost_net_set_backend(&net->dev, &file);
 288             assert(r >= 0);
 289         }
 290     }
 291     if (net->nc->info->poll) {
 292         net->nc->info->poll(net->nc, true);
 293     }
 294     vhost_dev_stop(&net->dev, dev);
 295 fail_start:
 296     vhost_dev_disable_notifiers(&net->dev, dev);
 297 fail_notifiers:
 298     return r;
 299 }
 300
 301 static void vhost_net_stop_one(struct vhost_net *net,
 302                                VirtIODevice *dev)
 303 {
 304     struct vhost_vring_file file = { .fd = -1 };
 305
 306     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 307         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 308             int r = vhost_net_set_backend(&net->dev, &file);
 309             assert(r >= 0);
 310         }
 311     }
 312     if (net->nc->info->poll) {
 313         net->nc->info->poll(net->nc, true);
 314     }
 315     vhost_dev_stop(&net->dev, dev);
 316     vhost_dev_disable_notifiers(&net->dev, dev);
 317 }
 318
 319 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 320                     int data_queue_pairs, int cvq)
 321 {
 322     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 323     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 324     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 325     int total_notifiers = data_queue_pairs * 2 + cvq;
 326     VirtIONet *n = VIRTIO_NET(dev);
 327     int nvhosts = data_queue_pairs + cvq;
 328     struct vhost_net *net;
 329     int r, e, i, index_end = data_queue_pairs * 2;
 330     NetClientState *peer;
 331
 332     if (cvq) {
 333         index_end += 1;
 334     }
 335
 336     if (!k->set_guest_notifiers) {
 337         error_report("binding does not support guest notifiers");
 338         return -ENOSYS;
 339     }
 340
 341     for (i = 0; i < nvhosts; i++) {
 342
 343         if (i < data_queue_pairs) {
 344             peer = qemu_get_peer(ncs, i);
 345         } else { /* Control Virtqueue */
 346             peer = qemu_get_peer(ncs, n->max_queue_pairs);
 347         }
 348
 349         net = get_vhost_net(peer);
 350         vhost_net_set_vq_index(net, i * 2, index_end);
 351
 352         /* Suppress the masking guest notifiers on vhost user
 353          * because vhost user doesn't interrupt masking/unmasking
 354          * properly.
 355          */
 356         if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 357             dev->use_guest_notifier_mask = false;
 358         }
 359      }
 360
 361     r = k->set_guest_notifiers(qbus->parent, total_notifiers, true);
 362     if (r < 0) {
 363         error_report("Error binding guest notifier: %d", -r);
 364         goto err;
 365     }
 366
 367     for (i = 0; i < nvhosts; i++) {
 368         if (i < data_queue_pairs) {
 369             peer = qemu_get_peer(ncs, i);
 370         } else {
 371             peer = qemu_get_peer(ncs, n->max_queue_pairs);
 372         }
 373         r = vhost_net_start_one(get_vhost_net(peer), dev);
 374
 375         if (r < 0) {
 376             goto err_start;
 377         }
 378
 379         if (peer->vring_enable) {
 380             /* restore vring enable state */
 381             r = vhost_set_vring_enable(peer, peer->vring_enable);
 382
 383             if (r < 0) {
 384                 goto err_start;
 385             }
 386         }
 387     }
 388
 389     return 0;
 390
 391 err_start:
 392     while (--i >= 0) {
 393         peer = qemu_get_peer(ncs , i);
 394         vhost_net_stop_one(get_vhost_net(peer), dev);
 395     }
 396     e = k->set_guest_notifiers(qbus->parent, total_notifiers, false);
 397     if (e < 0) {
 398         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 399         fflush(stderr);
 400     }
 401 err:
 402     return r;
 403 }
 404
 405 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 406                     int data_queue_pairs, int cvq)
 407 {
 408     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 409     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 410     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 411     VirtIONet *n = VIRTIO_NET(dev);
 412     NetClientState *peer;
 413     int total_notifiers = data_queue_pairs * 2 + cvq;
 414     int nvhosts = data_queue_pairs + cvq;
 415     int i, r;
 416
 417     for (i = 0; i < nvhosts; i++) {
 418         if (i < data_queue_pairs) {
 419             peer = qemu_get_peer(ncs, i);
 420         } else {
 421             peer = qemu_get_peer(ncs, n->max_queue_pairs);
 422         }
 423         vhost_net_stop_one(get_vhost_net(peer), dev);
 424     }
 425
 426     r = k->set_guest_notifiers(qbus->parent, total_notifiers, false);
 427     if (r < 0) {
 428         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 429         fflush(stderr);
 430     }
 431     assert(r >= 0);
 432 }
 433
 434 void vhost_net_cleanup(struct vhost_net *net)
 435 {
 436     vhost_dev_cleanup(&net->dev);
 437 }
 438
 439 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 440 {
 441     const VhostOps *vhost_ops = net->dev.vhost_ops;
 442
 443     assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 444     assert(vhost_ops->vhost_migration_done);
 445
 446     return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 447 }
 448
 449 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 450 {
 451     return vhost_virtqueue_pending(&net->dev, idx);
 452 }
 453
 454 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 455                               int idx, bool mask)
 456 {
 457     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 458 }
 459
 460 VHostNetState *get_vhost_net(NetClientState *nc)
 461 {
 462     VHostNetState *vhost_net = 0;
 463
 464     if (!nc) {
 465         return 0;
 466     }
 467
 468     switch (nc->info->type) {
 469     case NET_CLIENT_DRIVER_TAP:
 470         vhost_net = tap_get_vhost_net(nc);
 471         break;
 472 #ifdef CONFIG_VHOST_NET_USER
 473     case NET_CLIENT_DRIVER_VHOST_USER:
 474         vhost_net = vhost_user_get_vhost_net(nc);
 475         assert(vhost_net);
 476         break;
 477 #endif
 478 #ifdef CONFIG_VHOST_NET_VDPA
 479     case NET_CLIENT_DRIVER_VHOST_VDPA:
 480         vhost_net = vhost_vdpa_get_vhost_net(nc);
 481         assert(vhost_net);
 482         break;
 483 #endif
 484     default:
 485         break;
 486     }
 487
 488     return vhost_net;
 489 }
 490
 491 int vhost_set_vring_enable(NetClientState *nc, int enable)
 492 {
 493     VHostNetState *net = get_vhost_net(nc);
 494     const VhostOps *vhost_ops = net->dev.vhost_ops;
 495
 496     nc->vring_enable = enable;
 497
 498     if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
 499         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 500     }
 501
 502     return 0;
 503 }
 504
 505 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
 506 {
 507     const VhostOps *vhost_ops = net->dev.vhost_ops;
 508
 509     if (!vhost_ops->vhost_net_set_mtu) {
 510         return 0;
 511     }
 512
 513     return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
 514 }