drivers/vhost/vsock.c

   1 /*
   2  * vhost transport for vsock
   3  *
   4  * Copyright (C) 2013-2015 Red Hat, Inc.
   5  * Author: Asias He <asias@redhat.com>
   6  *         Stefan Hajnoczi <stefanha@redhat.com>
   7  *
   8  * This work is licensed under the terms of the GNU GPL, version 2.
   9  */
  10 #include <linux/miscdevice.h>
  11 #include <linux/atomic.h>
  12 #include <linux/module.h>
  13 #include <linux/mutex.h>
  14 #include <linux/vmalloc.h>
  15 #include <net/sock.h>
  16 #include <linux/virtio_vsock.h>
  17 #include <linux/vhost.h>
  18 #include <linux/hashtable.h>
  19
  20 #include <net/af_vsock.h>
  21 #include "vhost.h"
  22
  23 #define VHOST_VSOCK_DEFAULT_HOST_CID    2
  24
  25 enum {
  26         VHOST_VSOCK_FEATURES = VHOST_FEATURES,
  27 };
  28
  29 /* Used to track all the vhost_vsock instances on the system. */
  30 static DEFINE_MUTEX(vhost_vsock_mutex);
  31 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
  32
  33 struct vhost_vsock {
  34         struct vhost_dev dev;
  35         struct vhost_virtqueue vqs[2];
  36
  37         /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
  38         struct hlist_node hash;
  39
  40         struct vhost_work send_pkt_work;
  41         spinlock_t send_pkt_list_lock;
  42         struct list_head send_pkt_list; /* host->guest pending packets */
  43
  44         atomic_t queued_replies;
  45
  46         u32 guest_cid;
  47 };
  48
  49 static u32 vhost_transport_get_local_cid(void)
  50 {
  51         return VHOST_VSOCK_DEFAULT_HOST_CID;
  52 }
  53
  54 /* Callers that dereference the return value must hold vhost_vsock_mutex or the
  55  * RCU read lock.
  56  */
  57 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
  58 {
  59         struct vhost_vsock *vsock;
  60
  61         hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
  62                 u32 other_cid = vsock->guest_cid;
  63
  64                 /* Skip instances that have no CID yet */
  65                 if (other_cid == 0)
  66                         continue;
  67
  68                 if (other_cid == guest_cid)
  69                         return vsock;
  70
  71         }
  72
  73         return NULL;
  74 }
  75
  76 static void
  77 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
  78                             struct vhost_virtqueue *vq)
  79 {
  80         struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
  81         bool added = false;
  82         bool restart_tx = false;
  83
  84         mutex_lock(&vq->mutex);
  85
  86         if (!vq->private_data)
  87                 goto out;
  88
  89         /* Avoid further vmexits, we're already processing the virtqueue */
  90         vhost_disable_notify(&vsock->dev, vq);
  91
  92         for (;;) {
  93                 struct virtio_vsock_pkt *pkt;
  94                 struct iov_iter iov_iter;
  95                 unsigned out, in;
  96                 size_t nbytes;
  97                 size_t len;
  98                 int head;
  99
 100                 spin_lock_bh(&vsock->send_pkt_list_lock);
 101                 if (list_empty(&vsock->send_pkt_list)) {
 102                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 103                         vhost_enable_notify(&vsock->dev, vq);
 104                         break;
 105                 }
 106
 107                 pkt = list_first_entry(&vsock->send_pkt_list,
 108                                        struct virtio_vsock_pkt, list);
 109                 list_del_init(&pkt->list);
 110                 spin_unlock_bh(&vsock->send_pkt_list_lock);
 111
 112                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 113                                          &out, &in, NULL, NULL);
 114                 if (head < 0) {
 115                         spin_lock_bh(&vsock->send_pkt_list_lock);
 116                         list_add(&pkt->list, &vsock->send_pkt_list);
 117                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 118                         break;
 119                 }
 120
 121                 if (head == vq->num) {
 122                         spin_lock_bh(&vsock->send_pkt_list_lock);
 123                         list_add(&pkt->list, &vsock->send_pkt_list);
 124                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 125
 126                         /* We cannot finish yet if more buffers snuck in while
 127                          * re-enabling notify.
 128                          */
 129                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 130                                 vhost_disable_notify(&vsock->dev, vq);
 131                                 continue;
 132                         }
 133                         break;
 134                 }
 135
 136                 if (out) {
 137                         virtio_transport_free_pkt(pkt);
 138                         vq_err(vq, "Expected 0 output buffers, got %u\n", out);
 139                         break;
 140                 }
 141
 142                 len = iov_length(&vq->iov[out], in);
 143                 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len);
 144
 145                 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
 146                 if (nbytes != sizeof(pkt->hdr)) {
 147                         virtio_transport_free_pkt(pkt);
 148                         vq_err(vq, "Faulted on copying pkt hdr\n");
 149                         break;
 150                 }
 151
 152                 nbytes = copy_to_iter(pkt->buf, pkt->len, &iov_iter);
 153                 if (nbytes != pkt->len) {
 154                         virtio_transport_free_pkt(pkt);
 155                         vq_err(vq, "Faulted on copying pkt buf\n");
 156                         break;
 157                 }
 158
 159                 vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len);
 160                 added = true;
 161
 162                 if (pkt->reply) {
 163                         int val;
 164
 165                         val = atomic_dec_return(&vsock->queued_replies);
 166
 167                         /* Do we have resources to resume tx processing? */
 168                         if (val + 1 == tx_vq->num)
 169                                 restart_tx = true;
 170                 }
 171
 172                 /* Deliver to monitoring devices all correctly transmitted
 173                  * packets.
 174                  */
 175                 virtio_transport_deliver_tap_pkt(pkt);
 176
 177                 virtio_transport_free_pkt(pkt);
 178         }
 179         if (added)
 180                 vhost_signal(&vsock->dev, vq);
 181
 182 out:
 183         mutex_unlock(&vq->mutex);
 184
 185         if (restart_tx)
 186                 vhost_poll_queue(&tx_vq->poll);
 187 }
 188
 189 static void vhost_transport_send_pkt_work(struct vhost_work *work)
 190 {
 191         struct vhost_virtqueue *vq;
 192         struct vhost_vsock *vsock;
 193
 194         vsock = container_of(work, struct vhost_vsock, send_pkt_work);
 195         vq = &vsock->vqs[VSOCK_VQ_RX];
 196
 197         vhost_transport_do_send_pkt(vsock, vq);
 198 }
 199
 200 static int
 201 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
 202 {
 203         struct vhost_vsock *vsock;
 204         int len = pkt->len;
 205
 206         rcu_read_lock();
 207
 208         /* Find the vhost_vsock according to guest context id  */
 209         vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
 210         if (!vsock) {
 211                 rcu_read_unlock();
 212                 virtio_transport_free_pkt(pkt);
 213                 return -ENODEV;
 214         }
 215
 216         if (pkt->reply)
 217                 atomic_inc(&vsock->queued_replies);
 218
 219         spin_lock_bh(&vsock->send_pkt_list_lock);
 220         list_add_tail(&pkt->list, &vsock->send_pkt_list);
 221         spin_unlock_bh(&vsock->send_pkt_list_lock);
 222
 223         vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
 224
 225         rcu_read_unlock();
 226         return len;
 227 }
 228
 229 static int
 230 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
 231 {
 232         struct vhost_vsock *vsock;
 233         struct virtio_vsock_pkt *pkt, *n;
 234         int cnt = 0;
 235         int ret = -ENODEV;
 236         LIST_HEAD(freeme);
 237
 238         rcu_read_lock();
 239
 240         /* Find the vhost_vsock according to guest context id  */
 241         vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
 242         if (!vsock)
 243                 goto out;
 244
 245         spin_lock_bh(&vsock->send_pkt_list_lock);
 246         list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
 247                 if (pkt->vsk != vsk)
 248                         continue;
 249                 list_move(&pkt->list, &freeme);
 250         }
 251         spin_unlock_bh(&vsock->send_pkt_list_lock);
 252
 253         list_for_each_entry_safe(pkt, n, &freeme, list) {
 254                 if (pkt->reply)
 255                         cnt++;
 256                 list_del(&pkt->list);
 257                 virtio_transport_free_pkt(pkt);
 258         }
 259
 260         if (cnt) {
 261                 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
 262                 int new_cnt;
 263
 264                 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
 265                 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
 266                         vhost_poll_queue(&tx_vq->poll);
 267         }
 268
 269         ret = 0;
 270 out:
 271         rcu_read_unlock();
 272         return ret;
 273 }
 274
 275 static struct virtio_vsock_pkt *
 276 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
 277                       unsigned int out, unsigned int in)
 278 {
 279         struct virtio_vsock_pkt *pkt;
 280         struct iov_iter iov_iter;
 281         size_t nbytes;
 282         size_t len;
 283
 284         if (in != 0) {
 285                 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
 286                 return NULL;
 287         }
 288
 289         pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
 290         if (!pkt)
 291                 return NULL;
 292
 293         len = iov_length(vq->iov, out);
 294         iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
 295
 296         nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
 297         if (nbytes != sizeof(pkt->hdr)) {
 298                 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
 299                        sizeof(pkt->hdr), nbytes);
 300                 kfree(pkt);
 301                 return NULL;
 302         }
 303
 304         if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
 305                 pkt->len = le32_to_cpu(pkt->hdr.len);
 306
 307         /* No payload */
 308         if (!pkt->len)
 309                 return pkt;
 310
 311         /* The pkt is too big */
 312         if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
 313                 kfree(pkt);
 314                 return NULL;
 315         }
 316
 317         pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
 318         if (!pkt->buf) {
 319                 kfree(pkt);
 320                 return NULL;
 321         }
 322
 323         nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
 324         if (nbytes != pkt->len) {
 325                 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
 326                        pkt->len, nbytes);
 327                 virtio_transport_free_pkt(pkt);
 328                 return NULL;
 329         }
 330
 331         return pkt;
 332 }
 333
 334 /* Is there space left for replies to rx packets? */
 335 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
 336 {
 337         struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
 338         int val;
 339
 340         smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
 341         val = atomic_read(&vsock->queued_replies);
 342
 343         return val < vq->num;
 344 }
 345
 346 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
 347 {
 348         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 349                                                   poll.work);
 350         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 351                                                  dev);
 352         struct virtio_vsock_pkt *pkt;
 353         int head;
 354         unsigned int out, in;
 355         bool added = false;
 356
 357         mutex_lock(&vq->mutex);
 358
 359         if (!vq->private_data)
 360                 goto out;
 361
 362         vhost_disable_notify(&vsock->dev, vq);
 363         for (;;) {
 364                 u32 len;
 365
 366                 if (!vhost_vsock_more_replies(vsock)) {
 367                         /* Stop tx until the device processes already
 368                          * pending replies.  Leave tx virtqueue
 369                          * callbacks disabled.
 370                          */
 371                         goto no_more_replies;
 372                 }
 373
 374                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 375                                          &out, &in, NULL, NULL);
 376                 if (head < 0)
 377                         break;
 378
 379                 if (head == vq->num) {
 380                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 381                                 vhost_disable_notify(&vsock->dev, vq);
 382                                 continue;
 383                         }
 384                         break;
 385                 }
 386
 387                 pkt = vhost_vsock_alloc_pkt(vq, out, in);
 388                 if (!pkt) {
 389                         vq_err(vq, "Faulted on pkt\n");
 390                         continue;
 391                 }
 392
 393                 len = pkt->len;
 394
 395                 /* Deliver to monitoring devices all received packets */
 396                 virtio_transport_deliver_tap_pkt(pkt);
 397
 398                 /* Only accept correctly addressed packets */
 399                 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid)
 400                         virtio_transport_recv_pkt(pkt);
 401                 else
 402                         virtio_transport_free_pkt(pkt);
 403
 404                 vhost_add_used(vq, head, sizeof(pkt->hdr) + len);
 405                 added = true;
 406         }
 407
 408 no_more_replies:
 409         if (added)
 410                 vhost_signal(&vsock->dev, vq);
 411
 412 out:
 413         mutex_unlock(&vq->mutex);
 414 }
 415
 416 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
 417 {
 418         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 419                                                 poll.work);
 420         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 421                                                  dev);
 422
 423         vhost_transport_do_send_pkt(vsock, vq);
 424 }
 425
 426 static int vhost_vsock_start(struct vhost_vsock *vsock)
 427 {
 428         struct vhost_virtqueue *vq;
 429         size_t i;
 430         int ret;
 431
 432         mutex_lock(&vsock->dev.mutex);
 433
 434         ret = vhost_dev_check_owner(&vsock->dev);
 435         if (ret)
 436                 goto err;
 437
 438         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 439                 vq = &vsock->vqs[i];
 440
 441                 mutex_lock(&vq->mutex);
 442
 443                 if (!vhost_vq_access_ok(vq)) {
 444                         ret = -EFAULT;
 445                         goto err_vq;
 446                 }
 447
 448                 if (!vq->private_data) {
 449                         vq->private_data = vsock;
 450                         ret = vhost_vq_init_access(vq);
 451                         if (ret)
 452                                 goto err_vq;
 453                 }
 454
 455                 mutex_unlock(&vq->mutex);
 456         }
 457
 458         mutex_unlock(&vsock->dev.mutex);
 459         return 0;
 460
 461 err_vq:
 462         vq->private_data = NULL;
 463         mutex_unlock(&vq->mutex);
 464
 465         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 466                 vq = &vsock->vqs[i];
 467
 468                 mutex_lock(&vq->mutex);
 469                 vq->private_data = NULL;
 470                 mutex_unlock(&vq->mutex);
 471         }
 472 err:
 473         mutex_unlock(&vsock->dev.mutex);
 474         return ret;
 475 }
 476
 477 static int vhost_vsock_stop(struct vhost_vsock *vsock)
 478 {
 479         size_t i;
 480         int ret;
 481
 482         mutex_lock(&vsock->dev.mutex);
 483
 484         ret = vhost_dev_check_owner(&vsock->dev);
 485         if (ret)
 486                 goto err;
 487
 488         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 489                 struct vhost_virtqueue *vq = &vsock->vqs[i];
 490
 491                 mutex_lock(&vq->mutex);
 492                 vq->private_data = NULL;
 493                 mutex_unlock(&vq->mutex);
 494         }
 495
 496 err:
 497         mutex_unlock(&vsock->dev.mutex);
 498         return ret;
 499 }
 500
 501 static void vhost_vsock_free(struct vhost_vsock *vsock)
 502 {
 503         kvfree(vsock);
 504 }
 505
 506 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
 507 {
 508         struct vhost_virtqueue **vqs;
 509         struct vhost_vsock *vsock;
 510         int ret;
 511
 512         /* This struct is large and allocation could fail, fall back to vmalloc
 513          * if there is no other way.
 514          */
 515         vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
 516         if (!vsock)
 517                 return -ENOMEM;
 518
 519         vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
 520         if (!vqs) {
 521                 ret = -ENOMEM;
 522                 goto out;
 523         }
 524
 525         vsock->guest_cid = 0; /* no CID assigned yet */
 526
 527         atomic_set(&vsock->queued_replies, 0);
 528
 529         vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
 530         vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
 531         vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
 532         vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
 533
 534         vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs), UIO_MAXIOV);
 535
 536         file->private_data = vsock;
 537         spin_lock_init(&vsock->send_pkt_list_lock);
 538         INIT_LIST_HEAD(&vsock->send_pkt_list);
 539         vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
 540         return 0;
 541
 542 out:
 543         vhost_vsock_free(vsock);
 544         return ret;
 545 }
 546
 547 static void vhost_vsock_flush(struct vhost_vsock *vsock)
 548 {
 549         int i;
 550
 551         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
 552                 if (vsock->vqs[i].handle_kick)
 553                         vhost_poll_flush(&vsock->vqs[i].poll);
 554         vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
 555 }
 556
 557 static void vhost_vsock_reset_orphans(struct sock *sk)
 558 {
 559         struct vsock_sock *vsk = vsock_sk(sk);
 560
 561         /* vmci_transport.c doesn't take sk_lock here either.  At least we're
 562          * under vsock_table_lock so the sock cannot disappear while we're
 563          * executing.
 564          */
 565
 566         /* If the peer is still valid, no need to reset connection */
 567         if (vhost_vsock_get(vsk->remote_addr.svm_cid))
 568                 return;
 569
 570         /* If the close timeout is pending, let it expire.  This avoids races
 571          * with the timeout callback.
 572          */
 573         if (vsk->close_work_scheduled)
 574                 return;
 575
 576         sock_set_flag(sk, SOCK_DONE);
 577         vsk->peer_shutdown = SHUTDOWN_MASK;
 578         sk->sk_state = SS_UNCONNECTED;
 579         sk->sk_err = ECONNRESET;
 580         sk->sk_error_report(sk);
 581 }
 582
 583 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
 584 {
 585         struct vhost_vsock *vsock = file->private_data;
 586
 587         mutex_lock(&vhost_vsock_mutex);
 588         if (vsock->guest_cid)
 589                 hash_del_rcu(&vsock->hash);
 590         mutex_unlock(&vhost_vsock_mutex);
 591
 592         /* Wait for other CPUs to finish using vsock */
 593         synchronize_rcu();
 594
 595         /* Iterating over all connections for all CIDs to find orphans is
 596          * inefficient.  Room for improvement here. */
 597         vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
 598
 599         vhost_vsock_stop(vsock);
 600         vhost_vsock_flush(vsock);
 601         vhost_dev_stop(&vsock->dev);
 602
 603         spin_lock_bh(&vsock->send_pkt_list_lock);
 604         while (!list_empty(&vsock->send_pkt_list)) {
 605                 struct virtio_vsock_pkt *pkt;
 606
 607                 pkt = list_first_entry(&vsock->send_pkt_list,
 608                                 struct virtio_vsock_pkt, list);
 609                 list_del_init(&pkt->list);
 610                 virtio_transport_free_pkt(pkt);
 611         }
 612         spin_unlock_bh(&vsock->send_pkt_list_lock);
 613
 614         vhost_dev_cleanup(&vsock->dev);
 615         kfree(vsock->dev.vqs);
 616         vhost_vsock_free(vsock);
 617         return 0;
 618 }
 619
 620 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
 621 {
 622         struct vhost_vsock *other;
 623
 624         /* Refuse reserved CIDs */
 625         if (guest_cid <= VMADDR_CID_HOST ||
 626             guest_cid == U32_MAX)
 627                 return -EINVAL;
 628
 629         /* 64-bit CIDs are not yet supported */
 630         if (guest_cid > U32_MAX)
 631                 return -EINVAL;
 632
 633         /* Refuse if CID is already in use */
 634         mutex_lock(&vhost_vsock_mutex);
 635         other = vhost_vsock_get(guest_cid);
 636         if (other && other != vsock) {
 637                 mutex_unlock(&vhost_vsock_mutex);
 638                 return -EADDRINUSE;
 639         }
 640
 641         if (vsock->guest_cid)
 642                 hash_del_rcu(&vsock->hash);
 643
 644         vsock->guest_cid = guest_cid;
 645         hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
 646         mutex_unlock(&vhost_vsock_mutex);
 647
 648         return 0;
 649 }
 650
 651 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
 652 {
 653         struct vhost_virtqueue *vq;
 654         int i;
 655
 656         if (features & ~VHOST_VSOCK_FEATURES)
 657                 return -EOPNOTSUPP;
 658
 659         mutex_lock(&vsock->dev.mutex);
 660         if ((features & (1 << VHOST_F_LOG_ALL)) &&
 661             !vhost_log_access_ok(&vsock->dev)) {
 662                 mutex_unlock(&vsock->dev.mutex);
 663                 return -EFAULT;
 664         }
 665
 666         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 667                 vq = &vsock->vqs[i];
 668                 mutex_lock(&vq->mutex);
 669                 vq->acked_features = features;
 670                 mutex_unlock(&vq->mutex);
 671         }
 672         mutex_unlock(&vsock->dev.mutex);
 673         return 0;
 674 }
 675
 676 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
 677                                   unsigned long arg)
 678 {
 679         struct vhost_vsock *vsock = f->private_data;
 680         void __user *argp = (void __user *)arg;
 681         u64 guest_cid;
 682         u64 features;
 683         int start;
 684         int r;
 685
 686         switch (ioctl) {
 687         case VHOST_VSOCK_SET_GUEST_CID:
 688                 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
 689                         return -EFAULT;
 690                 return vhost_vsock_set_cid(vsock, guest_cid);
 691         case VHOST_VSOCK_SET_RUNNING:
 692                 if (copy_from_user(&start, argp, sizeof(start)))
 693                         return -EFAULT;
 694                 if (start)
 695                         return vhost_vsock_start(vsock);
 696                 else
 697                         return vhost_vsock_stop(vsock);
 698         case VHOST_GET_FEATURES:
 699                 features = VHOST_VSOCK_FEATURES;
 700                 if (copy_to_user(argp, &features, sizeof(features)))
 701                         return -EFAULT;
 702                 return 0;
 703         case VHOST_SET_FEATURES:
 704                 if (copy_from_user(&features, argp, sizeof(features)))
 705                         return -EFAULT;
 706                 return vhost_vsock_set_features(vsock, features);
 707         default:
 708                 mutex_lock(&vsock->dev.mutex);
 709                 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
 710                 if (r == -ENOIOCTLCMD)
 711                         r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
 712                 else
 713                         vhost_vsock_flush(vsock);
 714                 mutex_unlock(&vsock->dev.mutex);
 715                 return r;
 716         }
 717 }
 718
 719 #ifdef CONFIG_COMPAT
 720 static long vhost_vsock_dev_compat_ioctl(struct file *f, unsigned int ioctl,
 721                                          unsigned long arg)
 722 {
 723         return vhost_vsock_dev_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
 724 }
 725 #endif
 726
 727 static const struct file_operations vhost_vsock_fops = {
 728         .owner          = THIS_MODULE,
 729         .open           = vhost_vsock_dev_open,
 730         .release        = vhost_vsock_dev_release,
 731         .llseek         = noop_llseek,
 732         .unlocked_ioctl = vhost_vsock_dev_ioctl,
 733 #ifdef CONFIG_COMPAT
 734         .compat_ioctl   = vhost_vsock_dev_compat_ioctl,
 735 #endif
 736 };
 737
 738 static struct miscdevice vhost_vsock_misc = {
 739         .minor = VHOST_VSOCK_MINOR,
 740         .name = "vhost-vsock",
 741         .fops = &vhost_vsock_fops,
 742 };
 743
 744 static struct virtio_transport vhost_transport = {
 745         .transport = {
 746                 .get_local_cid            = vhost_transport_get_local_cid,
 747
 748                 .init                     = virtio_transport_do_socket_init,
 749                 .destruct                 = virtio_transport_destruct,
 750                 .release                  = virtio_transport_release,
 751                 .connect                  = virtio_transport_connect,
 752                 .shutdown                 = virtio_transport_shutdown,
 753                 .cancel_pkt               = vhost_transport_cancel_pkt,
 754
 755                 .dgram_enqueue            = virtio_transport_dgram_enqueue,
 756                 .dgram_dequeue            = virtio_transport_dgram_dequeue,
 757                 .dgram_bind               = virtio_transport_dgram_bind,
 758                 .dgram_allow              = virtio_transport_dgram_allow,
 759
 760                 .stream_enqueue           = virtio_transport_stream_enqueue,
 761                 .stream_dequeue           = virtio_transport_stream_dequeue,
 762                 .stream_has_data          = virtio_transport_stream_has_data,
 763                 .stream_has_space         = virtio_transport_stream_has_space,
 764                 .stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
 765                 .stream_is_active         = virtio_transport_stream_is_active,
 766                 .stream_allow             = virtio_transport_stream_allow,
 767
 768                 .notify_poll_in           = virtio_transport_notify_poll_in,
 769                 .notify_poll_out          = virtio_transport_notify_poll_out,
 770                 .notify_recv_init         = virtio_transport_notify_recv_init,
 771                 .notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
 772                 .notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
 773                 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
 774                 .notify_send_init         = virtio_transport_notify_send_init,
 775                 .notify_send_pre_block    = virtio_transport_notify_send_pre_block,
 776                 .notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
 777                 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
 778
 779                 .set_buffer_size          = virtio_transport_set_buffer_size,
 780                 .set_min_buffer_size      = virtio_transport_set_min_buffer_size,
 781                 .set_max_buffer_size      = virtio_transport_set_max_buffer_size,
 782                 .get_buffer_size          = virtio_transport_get_buffer_size,
 783                 .get_min_buffer_size      = virtio_transport_get_min_buffer_size,
 784                 .get_max_buffer_size      = virtio_transport_get_max_buffer_size,
 785         },
 786
 787         .send_pkt = vhost_transport_send_pkt,
 788 };
 789
 790 static int __init vhost_vsock_init(void)
 791 {
 792         int ret;
 793
 794         ret = vsock_core_init(&vhost_transport.transport);
 795         if (ret < 0)
 796                 return ret;
 797         return misc_register(&vhost_vsock_misc);
 798 };
 799
 800 static void __exit vhost_vsock_exit(void)
 801 {
 802         misc_deregister(&vhost_vsock_misc);
 803         vsock_core_exit();
 804 };
 805
 806 module_init(vhost_vsock_init);
 807 module_exit(vhost_vsock_exit);
 808 MODULE_LICENSE("GPL v2");
 809 MODULE_AUTHOR("Asias He");
 810 MODULE_DESCRIPTION("vhost transport for vsock ");
 811 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
 812 MODULE_ALIAS("devname:vhost-vsock");