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 #include "linux-headers/linux/vhost.h"
  38
  39
  40 /* Features supported by host kernel. */
  41 static const int kernel_feature_bits[] = {
  42     VIRTIO_F_NOTIFY_ON_EMPTY,
  43     VIRTIO_RING_F_INDIRECT_DESC,
  44     VIRTIO_RING_F_EVENT_IDX,
  45     VIRTIO_NET_F_MRG_RXBUF,
  46     VIRTIO_F_VERSION_1,
  47     VIRTIO_NET_F_MTU,
  48     VIRTIO_F_IOMMU_PLATFORM,
  49     VIRTIO_F_RING_PACKED,
  50     VIRTIO_F_RING_RESET,
  51     VIRTIO_NET_F_HASH_REPORT,
  52     VHOST_INVALID_FEATURE_BIT
  53 };
  54
  55 /* Features supported by others. */
  56 static const int user_feature_bits[] = {
  57     VIRTIO_F_NOTIFY_ON_EMPTY,
  58     VIRTIO_RING_F_INDIRECT_DESC,
  59     VIRTIO_RING_F_EVENT_IDX,
  60
  61     VIRTIO_F_ANY_LAYOUT,
  62     VIRTIO_F_VERSION_1,
  63     VIRTIO_NET_F_CSUM,
  64     VIRTIO_NET_F_GUEST_CSUM,
  65     VIRTIO_NET_F_GSO,
  66     VIRTIO_NET_F_GUEST_TSO4,
  67     VIRTIO_NET_F_GUEST_TSO6,
  68     VIRTIO_NET_F_GUEST_ECN,
  69     VIRTIO_NET_F_GUEST_UFO,
  70     VIRTIO_NET_F_HOST_TSO4,
  71     VIRTIO_NET_F_HOST_TSO6,
  72     VIRTIO_NET_F_HOST_ECN,
  73     VIRTIO_NET_F_HOST_UFO,
  74     VIRTIO_NET_F_MRG_RXBUF,
  75     VIRTIO_NET_F_MTU,
  76     VIRTIO_F_IOMMU_PLATFORM,
  77     VIRTIO_F_RING_PACKED,
  78     VIRTIO_F_RING_RESET,
  79     VIRTIO_NET_F_RSS,
  80     VIRTIO_NET_F_HASH_REPORT,
  81     VIRTIO_NET_F_GUEST_USO4,
  82     VIRTIO_NET_F_GUEST_USO6,
  83     VIRTIO_NET_F_HOST_USO,
  84
  85     /* This bit implies RARP isn't sent by QEMU out of band */
  86     VIRTIO_NET_F_GUEST_ANNOUNCE,
  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_DRIVER_TAP:
  99         feature_bits = kernel_feature_bits;
 100         break;
 101     case NET_CLIENT_DRIVER_VHOST_USER:
 102         feature_bits = user_feature_bits;
 103         break;
 104 #ifdef CONFIG_VHOST_NET_VDPA
 105     case NET_CLIENT_DRIVER_VHOST_VDPA:
 106         feature_bits = vdpa_feature_bits;
 107         break;
 108 #endif
 109     default:
 110         error_report("Feature bits not defined for this type: %d",
 111                 net->nc->info->type);
 112         break;
 113     }
 114
 115     return feature_bits;
 116 }
 117
 118 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 119 {
 120     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
 121             features);
 122 }
 123 int vhost_net_get_config(struct vhost_net *net,  uint8_t *config,
 124                          uint32_t config_len)
 125 {
 126     return vhost_dev_get_config(&net->dev, config, config_len, NULL);
 127 }
 128 int vhost_net_set_config(struct vhost_net *net, const uint8_t *data,
 129                          uint32_t offset, uint32_t size, uint32_t flags)
 130 {
 131     return vhost_dev_set_config(&net->dev, data, offset, size, flags);
 132 }
 133
 134 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 135 {
 136     net->dev.acked_features = net->dev.backend_features;
 137     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 138 }
 139
 140 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 141 {
 142     return net->dev.max_queues;
 143 }
 144
 145 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 146 {
 147     return net->dev.acked_features;
 148 }
 149
 150 void vhost_net_save_acked_features(NetClientState *nc)
 151 {
 152 #ifdef CONFIG_VHOST_NET_USER
 153     if (nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 154         vhost_user_save_acked_features(nc);
 155     }
 156 #endif
 157 }
 158
 159 static int vhost_net_get_fd(NetClientState *backend)
 160 {
 161     switch (backend->info->type) {
 162     case NET_CLIENT_DRIVER_TAP:
 163         return tap_get_fd(backend);
 164     default:
 165         fprintf(stderr, "vhost-net requires tap backend\n");
 166         return -ENOSYS;
 167     }
 168 }
 169
 170 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 171 {
 172     int r;
 173     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 174     struct vhost_net *net = g_new0(struct vhost_net, 1);
 175     uint64_t features = 0;
 176     Error *local_err = NULL;
 177
 178     if (!options->net_backend) {
 179         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 180         goto fail;
 181     }
 182     net->nc = options->net_backend;
 183     net->dev.nvqs = options->nvqs;
 184
 185     net->dev.max_queues = 1;
 186     net->dev.vqs = net->vqs;
 187
 188     if (backend_kernel) {
 189         r = vhost_net_get_fd(options->net_backend);
 190         if (r < 0) {
 191             goto fail;
 192         }
 193         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 194             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
 195         net->backend = r;
 196         net->dev.protocol_features = 0;
 197     } else {
 198         net->dev.backend_features = 0;
 199         net->dev.protocol_features = 0;
 200         net->backend = -1;
 201
 202         /* vhost-user needs vq_index to initiate a specific queue pair */
 203         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
 204     }
 205
 206     r = vhost_dev_init(&net->dev, options->opaque,
 207                        options->backend_type, options->busyloop_timeout,
 208                        &local_err);
 209     if (r < 0) {
 210         error_report_err(local_err);
 211         goto fail;
 212     }
 213     if (backend_kernel) {
 214         if (!qemu_has_vnet_hdr_len(options->net_backend,
 215                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 216             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
 217         }
 218         if (~net->dev.features & net->dev.backend_features) {
 219             fprintf(stderr, "vhost lacks feature mask 0x%" PRIx64
 220                    " for backend\n",
 221                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 222             goto fail;
 223         }
 224     }
 225
 226     /* Set sane init value. Override when guest acks. */
 227 #ifdef CONFIG_VHOST_NET_USER
 228     if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 229         features = vhost_user_get_acked_features(net->nc);
 230         if (~net->dev.features & features) {
 231             fprintf(stderr, "vhost lacks feature mask 0x%" PRIx64
 232                     " for backend\n",
 233                     (uint64_t)(~net->dev.features & features));
 234             goto fail;
 235         }
 236     }
 237 #endif
 238
 239     vhost_net_ack_features(net, features);
 240
 241     return net;
 242
 243 fail:
 244     vhost_dev_cleanup(&net->dev);
 245     g_free(net);
 246     return NULL;
 247 }
 248
 249 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index,
 250                                    int vq_index_end)
 251 {
 252     net->dev.vq_index = vq_index;
 253     net->dev.vq_index_end = vq_index_end;
 254 }
 255
 256 static int vhost_net_start_one(struct vhost_net *net,
 257                                VirtIODevice *dev)
 258 {
 259     struct vhost_vring_file file = { };
 260     int r;
 261
 262     if (net->nc->info->start) {
 263         r = net->nc->info->start(net->nc);
 264         if (r < 0) {
 265             return r;
 266         }
 267     }
 268
 269     r = vhost_dev_enable_notifiers(&net->dev, dev);
 270     if (r < 0) {
 271         goto fail_notifiers;
 272     }
 273
 274     r = vhost_dev_start(&net->dev, dev, false);
 275     if (r < 0) {
 276         goto fail_start;
 277     }
 278
 279     if (net->nc->info->poll) {
 280         net->nc->info->poll(net->nc, false);
 281     }
 282
 283     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 284         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 285         file.fd = net->backend;
 286         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 287             if (!virtio_queue_enabled(dev, net->dev.vq_index +
 288                                       file.index)) {
 289                 /* Queue might not be ready for start */
 290                 continue;
 291             }
 292             r = vhost_net_set_backend(&net->dev, &file);
 293             if (r < 0) {
 294                 r = -errno;
 295                 goto fail;
 296             }
 297         }
 298     }
 299
 300     if (net->nc->info->load) {
 301         r = net->nc->info->load(net->nc);
 302         if (r < 0) {
 303             goto fail;
 304         }
 305     }
 306     return 0;
 307 fail:
 308     file.fd = -1;
 309     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 310         while (file.index-- > 0) {
 311             if (!virtio_queue_enabled(dev, net->dev.vq_index +
 312                                       file.index)) {
 313                 /* Queue might not be ready for start */
 314                 continue;
 315             }
 316             int ret = vhost_net_set_backend(&net->dev, &file);
 317             assert(ret >= 0);
 318         }
 319     }
 320     if (net->nc->info->poll) {
 321         net->nc->info->poll(net->nc, true);
 322     }
 323     vhost_dev_stop(&net->dev, dev, false);
 324 fail_start:
 325     vhost_dev_disable_notifiers(&net->dev, dev);
 326 fail_notifiers:
 327     return r;
 328 }
 329
 330 static void vhost_net_stop_one(struct vhost_net *net,
 331                                VirtIODevice *dev)
 332 {
 333     struct vhost_vring_file file = { .fd = -1 };
 334
 335     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 336         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 337             int r = vhost_net_set_backend(&net->dev, &file);
 338             assert(r >= 0);
 339         }
 340     }
 341     if (net->nc->info->poll) {
 342         net->nc->info->poll(net->nc, true);
 343     }
 344     vhost_dev_stop(&net->dev, dev, false);
 345     if (net->nc->info->stop) {
 346         net->nc->info->stop(net->nc);
 347     }
 348     vhost_dev_disable_notifiers(&net->dev, dev);
 349 }
 350
 351 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 352                     int data_queue_pairs, int cvq)
 353 {
 354     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 355     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 356     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 357     int total_notifiers = data_queue_pairs * 2 + cvq;
 358     VirtIONet *n = VIRTIO_NET(dev);
 359     int nvhosts = data_queue_pairs + cvq;
 360     struct vhost_net *net;
 361     int r, e, i, index_end = data_queue_pairs * 2;
 362     NetClientState *peer;
 363
 364     if (cvq) {
 365         index_end += 1;
 366     }
 367
 368     if (!k->set_guest_notifiers) {
 369         error_report("binding does not support guest notifiers");
 370         return -ENOSYS;
 371     }
 372
 373     for (i = 0; i < nvhosts; i++) {
 374
 375         if (i < data_queue_pairs) {
 376             peer = qemu_get_peer(ncs, i);
 377         } else { /* Control Virtqueue */
 378             peer = qemu_get_peer(ncs, n->max_queue_pairs);
 379         }
 380
 381         net = get_vhost_net(peer);
 382         vhost_net_set_vq_index(net, i * 2, index_end);
 383
 384         /* Suppress the masking guest notifiers on vhost user
 385          * because vhost user doesn't interrupt masking/unmasking
 386          * properly.
 387          */
 388         if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 389             dev->use_guest_notifier_mask = false;
 390         }
 391      }
 392
 393     r = k->set_guest_notifiers(qbus->parent, total_notifiers, true);
 394     if (r < 0) {
 395         error_report("Error binding guest notifier: %d", -r);
 396         goto err;
 397     }
 398
 399     for (i = 0; i < nvhosts; i++) {
 400         if (i < data_queue_pairs) {
 401             peer = qemu_get_peer(ncs, i);
 402         } else {
 403             peer = qemu_get_peer(ncs, n->max_queue_pairs);
 404         }
 405
 406         if (peer->vring_enable) {
 407             /* restore vring enable state */
 408             r = vhost_set_vring_enable(peer, peer->vring_enable);
 409
 410             if (r < 0) {
 411                 goto err_start;
 412             }
 413         }
 414
 415         r = vhost_net_start_one(get_vhost_net(peer), dev);
 416         if (r < 0) {
 417             goto err_start;
 418         }
 419     }
 420
 421     return 0;
 422
 423 err_start:
 424     while (--i >= 0) {
 425         peer = qemu_get_peer(ncs, i < data_queue_pairs ?
 426                                   i : n->max_queue_pairs);
 427         vhost_net_stop_one(get_vhost_net(peer), dev);
 428     }
 429     e = k->set_guest_notifiers(qbus->parent, total_notifiers, false);
 430     if (e < 0) {
 431         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 432         fflush(stderr);
 433     }
 434 err:
 435     return r;
 436 }
 437
 438 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 439                     int data_queue_pairs, int cvq)
 440 {
 441     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 442     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 443     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 444     VirtIONet *n = VIRTIO_NET(dev);
 445     NetClientState *peer;
 446     int total_notifiers = data_queue_pairs * 2 + cvq;
 447     int nvhosts = data_queue_pairs + cvq;
 448     int i, r;
 449
 450     for (i = 0; i < nvhosts; i++) {
 451         if (i < data_queue_pairs) {
 452             peer = qemu_get_peer(ncs, i);
 453         } else {
 454             peer = qemu_get_peer(ncs, n->max_queue_pairs);
 455         }
 456         vhost_net_stop_one(get_vhost_net(peer), dev);
 457     }
 458
 459     r = k->set_guest_notifiers(qbus->parent, total_notifiers, false);
 460     if (r < 0) {
 461         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 462         fflush(stderr);
 463     }
 464     assert(r >= 0);
 465 }
 466
 467 void vhost_net_cleanup(struct vhost_net *net)
 468 {
 469     vhost_dev_cleanup(&net->dev);
 470 }
 471
 472 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 473 {
 474     const VhostOps *vhost_ops = net->dev.vhost_ops;
 475
 476     assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 477     assert(vhost_ops->vhost_migration_done);
 478
 479     return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 480 }
 481
 482 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 483 {
 484     return vhost_virtqueue_pending(&net->dev, idx);
 485 }
 486
 487 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 488                               int idx, bool mask)
 489 {
 490     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 491 }
 492
 493 bool vhost_net_config_pending(VHostNetState *net)
 494 {
 495     return vhost_config_pending(&net->dev);
 496 }
 497
 498 void vhost_net_config_mask(VHostNetState *net, VirtIODevice *dev, bool mask)
 499 {
 500     vhost_config_mask(&net->dev, dev, mask);
 501 }
 502 VHostNetState *get_vhost_net(NetClientState *nc)
 503 {
 504     VHostNetState *vhost_net = 0;
 505
 506     if (!nc) {
 507         return 0;
 508     }
 509
 510     switch (nc->info->type) {
 511     case NET_CLIENT_DRIVER_TAP:
 512         vhost_net = tap_get_vhost_net(nc);
 513         /*
 514          * tap_get_vhost_net() can return NULL if a tap net-device backend is
 515          * created with 'vhost=off' option, 'vhostforce=off' or no vhost or
 516          * vhostforce or vhostfd options at all. Please see net_init_tap_one().
 517          * Hence, we omit the assertion here.
 518          */
 519         break;
 520 #ifdef CONFIG_VHOST_NET_USER
 521     case NET_CLIENT_DRIVER_VHOST_USER:
 522         vhost_net = vhost_user_get_vhost_net(nc);
 523         assert(vhost_net);
 524         break;
 525 #endif
 526 #ifdef CONFIG_VHOST_NET_VDPA
 527     case NET_CLIENT_DRIVER_VHOST_VDPA:
 528         vhost_net = vhost_vdpa_get_vhost_net(nc);
 529         assert(vhost_net);
 530         break;
 531 #endif
 532     default:
 533         break;
 534     }
 535
 536     return vhost_net;
 537 }
 538
 539 int vhost_set_vring_enable(NetClientState *nc, int enable)
 540 {
 541     VHostNetState *net = get_vhost_net(nc);
 542     const VhostOps *vhost_ops = net->dev.vhost_ops;
 543
 544     nc->vring_enable = enable;
 545
 546     if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
 547         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 548     }
 549
 550     return 0;
 551 }
 552
 553 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
 554 {
 555     const VhostOps *vhost_ops = net->dev.vhost_ops;
 556
 557     if (!vhost_ops->vhost_net_set_mtu) {
 558         return 0;
 559     }
 560
 561     return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
 562 }
 563
 564 void vhost_net_virtqueue_reset(VirtIODevice *vdev, NetClientState *nc,
 565                                int vq_index)
 566 {
 567     VHostNetState *net = get_vhost_net(nc->peer);
 568     const VhostOps *vhost_ops = net->dev.vhost_ops;
 569     struct vhost_vring_file file = { .fd = -1 };
 570     int idx;
 571
 572     /* should only be called after backend is connected */
 573     assert(vhost_ops);
 574
 575     idx = vhost_ops->vhost_get_vq_index(&net->dev, vq_index);
 576
 577     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 578         file.index = idx;
 579         int r = vhost_net_set_backend(&net->dev, &file);
 580         assert(r >= 0);
 581     }
 582
 583     vhost_virtqueue_stop(&net->dev,
 584                          vdev,
 585                          net->dev.vqs + idx,
 586                          net->dev.vq_index + idx);
 587 }
 588
 589 int vhost_net_virtqueue_restart(VirtIODevice *vdev, NetClientState *nc,
 590                                 int vq_index)
 591 {
 592     VHostNetState *net = get_vhost_net(nc->peer);
 593     const VhostOps *vhost_ops = net->dev.vhost_ops;
 594     struct vhost_vring_file file = { };
 595     int idx, r;
 596
 597     if (!net->dev.started) {
 598         return -EBUSY;
 599     }
 600
 601     /* should only be called after backend is connected */
 602     assert(vhost_ops);
 603
 604     idx = vhost_ops->vhost_get_vq_index(&net->dev, vq_index);
 605
 606     r = vhost_virtqueue_start(&net->dev,
 607                               vdev,
 608                               net->dev.vqs + idx,
 609                               net->dev.vq_index + idx);
 610     if (r < 0) {
 611         goto err_start;
 612     }
 613
 614     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 615         file.index = idx;
 616         file.fd = net->backend;
 617         r = vhost_net_set_backend(&net->dev, &file);
 618         if (r < 0) {
 619             r = -errno;
 620             goto err_start;
 621         }
 622     }
 623
 624     return 0;
 625
 626 err_start:
 627     error_report("Error when restarting the queue.");
 628
 629     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 630         file.fd = VHOST_FILE_UNBIND;
 631         file.index = idx;
 632         int ret = vhost_net_set_backend(&net->dev, &file);
 633         assert(ret >= 0);
 634     }
 635
 636     vhost_dev_stop(&net->dev, vdev, false);
 637
 638     return r;
 639 }