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 /* Max number of bytes transferred before requeueing the job.
  25  * Using this limit prevents one virtqueue from starving others. */
  26 #define VHOST_VSOCK_WEIGHT 0x80000
  27 /* Max number of packets transferred before requeueing the job.
  28  * Using this limit prevents one virtqueue from starving others with
  29  * small pkts.
  30  */
  31 #define VHOST_VSOCK_PKT_WEIGHT 256
  32
  33 enum {
  34         VHOST_VSOCK_FEATURES = VHOST_FEATURES,
  35 };
  36
  37 /* Used to track all the vhost_vsock instances on the system. */
  38 static DEFINE_SPINLOCK(vhost_vsock_lock);
  39 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
  40
  41 struct vhost_vsock {
  42         struct vhost_dev dev;
  43         struct vhost_virtqueue vqs[2];
  44
  45         /* Link to global vhost_vsock_hash, writes use vhost_vsock_lock */
  46         struct hlist_node hash;
  47
  48         struct vhost_work send_pkt_work;
  49         spinlock_t send_pkt_list_lock;
  50         struct list_head send_pkt_list; /* host->guest pending packets */
  51
  52         atomic_t queued_replies;
  53
  54         u32 guest_cid;
  55 };
  56
  57 static u32 vhost_transport_get_local_cid(void)
  58 {
  59         return VHOST_VSOCK_DEFAULT_HOST_CID;
  60 }
  61
  62 /* Callers that dereference the return value must hold vhost_vsock_lock or the
  63  * RCU read lock.
  64  */
  65 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
  66 {
  67         struct vhost_vsock *vsock;
  68
  69         hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
  70                 u32 other_cid = vsock->guest_cid;
  71
  72                 /* Skip instances that have no CID yet */
  73                 if (other_cid == 0)
  74                         continue;
  75
  76                 if (other_cid == guest_cid)
  77                         return vsock;
  78
  79         }
  80
  81         return NULL;
  82 }
  83
  84 static void
  85 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
  86                             struct vhost_virtqueue *vq)
  87 {
  88         struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
  89         int pkts = 0, total_len = 0;
  90         bool added = false;
  91         bool restart_tx = false;
  92
  93         mutex_lock(&vq->mutex);
  94
  95         if (!vq->private_data)
  96                 goto out;
  97
  98         /* Avoid further vmexits, we're already processing the virtqueue */
  99         vhost_disable_notify(&vsock->dev, vq);
 100
 101         do {
 102                 struct virtio_vsock_pkt *pkt;
 103                 struct iov_iter iov_iter;
 104                 unsigned out, in;
 105                 size_t nbytes;
 106                 size_t iov_len, payload_len;
 107                 int head;
 108
 109                 spin_lock_bh(&vsock->send_pkt_list_lock);
 110                 if (list_empty(&vsock->send_pkt_list)) {
 111                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 112                         vhost_enable_notify(&vsock->dev, vq);
 113                         break;
 114                 }
 115
 116                 pkt = list_first_entry(&vsock->send_pkt_list,
 117                                        struct virtio_vsock_pkt, list);
 118                 list_del_init(&pkt->list);
 119                 spin_unlock_bh(&vsock->send_pkt_list_lock);
 120
 121                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 122                                          &out, &in, NULL, NULL);
 123                 if (head < 0) {
 124                         spin_lock_bh(&vsock->send_pkt_list_lock);
 125                         list_add(&pkt->list, &vsock->send_pkt_list);
 126                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 127                         break;
 128                 }
 129
 130                 if (head == vq->num) {
 131                         spin_lock_bh(&vsock->send_pkt_list_lock);
 132                         list_add(&pkt->list, &vsock->send_pkt_list);
 133                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 134
 135                         /* We cannot finish yet if more buffers snuck in while
 136                          * re-enabling notify.
 137                          */
 138                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 139                                 vhost_disable_notify(&vsock->dev, vq);
 140                                 continue;
 141                         }
 142                         break;
 143                 }
 144
 145                 if (out) {
 146                         virtio_transport_free_pkt(pkt);
 147                         vq_err(vq, "Expected 0 output buffers, got %u\n", out);
 148                         break;
 149                 }
 150
 151                 iov_len = iov_length(&vq->iov[out], in);
 152                 if (iov_len < sizeof(pkt->hdr)) {
 153                         virtio_transport_free_pkt(pkt);
 154                         vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
 155                         break;
 156                 }
 157
 158                 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len);
 159                 payload_len = pkt->len - pkt->off;
 160
 161                 /* If the packet is greater than the space available in the
 162                  * buffer, we split it using multiple buffers.
 163                  */
 164                 if (payload_len > iov_len - sizeof(pkt->hdr))
 165                         payload_len = iov_len - sizeof(pkt->hdr);
 166
 167                 /* Set the correct length in the header */
 168                 pkt->hdr.len = cpu_to_le32(payload_len);
 169
 170                 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
 171                 if (nbytes != sizeof(pkt->hdr)) {
 172                         virtio_transport_free_pkt(pkt);
 173                         vq_err(vq, "Faulted on copying pkt hdr\n");
 174                         break;
 175                 }
 176
 177                 nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len,
 178                                       &iov_iter);
 179                 if (nbytes != payload_len) {
 180                         virtio_transport_free_pkt(pkt);
 181                         vq_err(vq, "Faulted on copying pkt buf\n");
 182                         break;
 183                 }
 184
 185                 /* Deliver to monitoring devices all packets that we
 186                  * will transmit.
 187                  */
 188                 virtio_transport_deliver_tap_pkt(pkt);
 189
 190                 vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len);
 191                 added = true;
 192
 193                 pkt->off += payload_len;
 194                 total_len += payload_len;
 195
 196                 /* If we didn't send all the payload we can requeue the packet
 197                  * to send it with the next available buffer.
 198                  */
 199                 if (pkt->off < pkt->len) {
 200                         spin_lock_bh(&vsock->send_pkt_list_lock);
 201                         list_add(&pkt->list, &vsock->send_pkt_list);
 202                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 203                 } else {
 204                         if (pkt->reply) {
 205                                 int val;
 206
 207                                 val = atomic_dec_return(&vsock->queued_replies);
 208
 209                                 /* Do we have resources to resume tx
 210                                  * processing?
 211                                  */
 212                                 if (val + 1 == tx_vq->num)
 213                                         restart_tx = true;
 214                         }
 215
 216                         virtio_transport_free_pkt(pkt);
 217                 }
 218         } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
 219         if (added)
 220                 vhost_signal(&vsock->dev, vq);
 221
 222 out:
 223         mutex_unlock(&vq->mutex);
 224
 225         if (restart_tx)
 226                 vhost_poll_queue(&tx_vq->poll);
 227 }
 228
 229 static void vhost_transport_send_pkt_work(struct vhost_work *work)
 230 {
 231         struct vhost_virtqueue *vq;
 232         struct vhost_vsock *vsock;
 233
 234         vsock = container_of(work, struct vhost_vsock, send_pkt_work);
 235         vq = &vsock->vqs[VSOCK_VQ_RX];
 236
 237         vhost_transport_do_send_pkt(vsock, vq);
 238 }
 239
 240 static int
 241 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
 242 {
 243         struct vhost_vsock *vsock;
 244         int len = pkt->len;
 245
 246         rcu_read_lock();
 247
 248         /* Find the vhost_vsock according to guest context id  */
 249         vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
 250         if (!vsock) {
 251                 rcu_read_unlock();
 252                 virtio_transport_free_pkt(pkt);
 253                 return -ENODEV;
 254         }
 255
 256         if (pkt->reply)
 257                 atomic_inc(&vsock->queued_replies);
 258
 259         spin_lock_bh(&vsock->send_pkt_list_lock);
 260         list_add_tail(&pkt->list, &vsock->send_pkt_list);
 261         spin_unlock_bh(&vsock->send_pkt_list_lock);
 262
 263         vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
 264
 265         rcu_read_unlock();
 266         return len;
 267 }
 268
 269 static int
 270 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
 271 {
 272         struct vhost_vsock *vsock;
 273         struct virtio_vsock_pkt *pkt, *n;
 274         int cnt = 0;
 275         int ret = -ENODEV;
 276         LIST_HEAD(freeme);
 277
 278         rcu_read_lock();
 279
 280         /* Find the vhost_vsock according to guest context id  */
 281         vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
 282         if (!vsock)
 283                 goto out;
 284
 285         spin_lock_bh(&vsock->send_pkt_list_lock);
 286         list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
 287                 if (pkt->vsk != vsk)
 288                         continue;
 289                 list_move(&pkt->list, &freeme);
 290         }
 291         spin_unlock_bh(&vsock->send_pkt_list_lock);
 292
 293         list_for_each_entry_safe(pkt, n, &freeme, list) {
 294                 if (pkt->reply)
 295                         cnt++;
 296                 list_del(&pkt->list);
 297                 virtio_transport_free_pkt(pkt);
 298         }
 299
 300         if (cnt) {
 301                 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
 302                 int new_cnt;
 303
 304                 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
 305                 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
 306                         vhost_poll_queue(&tx_vq->poll);
 307         }
 308
 309         ret = 0;
 310 out:
 311         rcu_read_unlock();
 312         return ret;
 313 }
 314
 315 static struct virtio_vsock_pkt *
 316 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
 317                       unsigned int out, unsigned int in)
 318 {
 319         struct virtio_vsock_pkt *pkt;
 320         struct iov_iter iov_iter;
 321         size_t nbytes;
 322         size_t len;
 323
 324         if (in != 0) {
 325                 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
 326                 return NULL;
 327         }
 328
 329         pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
 330         if (!pkt)
 331                 return NULL;
 332
 333         len = iov_length(vq->iov, out);
 334         iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
 335
 336         nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
 337         if (nbytes != sizeof(pkt->hdr)) {
 338                 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
 339                        sizeof(pkt->hdr), nbytes);
 340                 kfree(pkt);
 341                 return NULL;
 342         }
 343
 344         if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
 345                 pkt->len = le32_to_cpu(pkt->hdr.len);
 346
 347         /* No payload */
 348         if (!pkt->len)
 349                 return pkt;
 350
 351         /* The pkt is too big */
 352         if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
 353                 kfree(pkt);
 354                 return NULL;
 355         }
 356
 357         pkt->buf = kvmalloc(pkt->len, GFP_KERNEL);
 358         if (!pkt->buf) {
 359                 kfree(pkt);
 360                 return NULL;
 361         }
 362
 363         nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
 364         if (nbytes != pkt->len) {
 365                 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
 366                        pkt->len, nbytes);
 367                 virtio_transport_free_pkt(pkt);
 368                 return NULL;
 369         }
 370
 371         return pkt;
 372 }
 373
 374 /* Is there space left for replies to rx packets? */
 375 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
 376 {
 377         struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
 378         int val;
 379
 380         smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
 381         val = atomic_read(&vsock->queued_replies);
 382
 383         return val < vq->num;
 384 }
 385
 386 static struct virtio_transport vhost_transport = {
 387         .transport = {
 388                 .get_local_cid            = vhost_transport_get_local_cid,
 389
 390                 .init                     = virtio_transport_do_socket_init,
 391                 .destruct                 = virtio_transport_destruct,
 392                 .release                  = virtio_transport_release,
 393                 .connect                  = virtio_transport_connect,
 394                 .shutdown                 = virtio_transport_shutdown,
 395                 .cancel_pkt               = vhost_transport_cancel_pkt,
 396
 397                 .dgram_enqueue            = virtio_transport_dgram_enqueue,
 398                 .dgram_dequeue            = virtio_transport_dgram_dequeue,
 399                 .dgram_bind               = virtio_transport_dgram_bind,
 400                 .dgram_allow              = virtio_transport_dgram_allow,
 401
 402                 .stream_enqueue           = virtio_transport_stream_enqueue,
 403                 .stream_dequeue           = virtio_transport_stream_dequeue,
 404                 .stream_has_data          = virtio_transport_stream_has_data,
 405                 .stream_has_space         = virtio_transport_stream_has_space,
 406                 .stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
 407                 .stream_is_active         = virtio_transport_stream_is_active,
 408                 .stream_allow             = virtio_transport_stream_allow,
 409
 410                 .notify_poll_in           = virtio_transport_notify_poll_in,
 411                 .notify_poll_out          = virtio_transport_notify_poll_out,
 412                 .notify_recv_init         = virtio_transport_notify_recv_init,
 413                 .notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
 414                 .notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
 415                 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
 416                 .notify_send_init         = virtio_transport_notify_send_init,
 417                 .notify_send_pre_block    = virtio_transport_notify_send_pre_block,
 418                 .notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
 419                 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
 420
 421                 .set_buffer_size          = virtio_transport_set_buffer_size,
 422                 .set_min_buffer_size      = virtio_transport_set_min_buffer_size,
 423                 .set_max_buffer_size      = virtio_transport_set_max_buffer_size,
 424                 .get_buffer_size          = virtio_transport_get_buffer_size,
 425                 .get_min_buffer_size      = virtio_transport_get_min_buffer_size,
 426                 .get_max_buffer_size      = virtio_transport_get_max_buffer_size,
 427         },
 428
 429         .send_pkt = vhost_transport_send_pkt,
 430 };
 431
 432 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
 433 {
 434         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 435                                                   poll.work);
 436         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 437                                                  dev);
 438         struct virtio_vsock_pkt *pkt;
 439         int head, pkts = 0, total_len = 0;
 440         unsigned int out, in;
 441         bool added = false;
 442
 443         mutex_lock(&vq->mutex);
 444
 445         if (!vq->private_data)
 446                 goto out;
 447
 448         vhost_disable_notify(&vsock->dev, vq);
 449         do {
 450                 u32 len;
 451
 452                 if (!vhost_vsock_more_replies(vsock)) {
 453                         /* Stop tx until the device processes already
 454                          * pending replies.  Leave tx virtqueue
 455                          * callbacks disabled.
 456                          */
 457                         goto no_more_replies;
 458                 }
 459
 460                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 461                                          &out, &in, NULL, NULL);
 462                 if (head < 0)
 463                         break;
 464
 465                 if (head == vq->num) {
 466                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 467                                 vhost_disable_notify(&vsock->dev, vq);
 468                                 continue;
 469                         }
 470                         break;
 471                 }
 472
 473                 pkt = vhost_vsock_alloc_pkt(vq, out, in);
 474                 if (!pkt) {
 475                         vq_err(vq, "Faulted on pkt\n");
 476                         continue;
 477                 }
 478
 479                 len = pkt->len;
 480
 481                 /* Deliver to monitoring devices all received packets */
 482                 virtio_transport_deliver_tap_pkt(pkt);
 483
 484                 /* Only accept correctly addressed packets */
 485                 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid &&
 486                     le64_to_cpu(pkt->hdr.dst_cid) ==
 487                     vhost_transport_get_local_cid())
 488                         virtio_transport_recv_pkt(&vhost_transport, pkt);
 489                 else
 490                         virtio_transport_free_pkt(pkt);
 491
 492                 len += sizeof(pkt->hdr);
 493                 vhost_add_used(vq, head, 0);
 494                 total_len += len;
 495                 added = true;
 496         } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
 497
 498 no_more_replies:
 499         if (added)
 500                 vhost_signal(&vsock->dev, vq);
 501
 502 out:
 503         mutex_unlock(&vq->mutex);
 504 }
 505
 506 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
 507 {
 508         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 509                                                 poll.work);
 510         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 511                                                  dev);
 512
 513         vhost_transport_do_send_pkt(vsock, vq);
 514 }
 515
 516 static int vhost_vsock_start(struct vhost_vsock *vsock)
 517 {
 518         struct vhost_virtqueue *vq;
 519         size_t i;
 520         int ret;
 521
 522         mutex_lock(&vsock->dev.mutex);
 523
 524         ret = vhost_dev_check_owner(&vsock->dev);
 525         if (ret)
 526                 goto err;
 527
 528         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 529                 vq = &vsock->vqs[i];
 530
 531                 mutex_lock(&vq->mutex);
 532
 533                 if (!vhost_vq_access_ok(vq)) {
 534                         ret = -EFAULT;
 535                         goto err_vq;
 536                 }
 537
 538                 if (!vq->private_data) {
 539                         vq->private_data = vsock;
 540                         ret = vhost_vq_init_access(vq);
 541                         if (ret)
 542                                 goto err_vq;
 543                 }
 544
 545                 mutex_unlock(&vq->mutex);
 546         }
 547
 548         /* Some packets may have been queued before the device was started,
 549          * let's kick the send worker to send them.
 550          */
 551         vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
 552
 553         mutex_unlock(&vsock->dev.mutex);
 554         return 0;
 555
 556 err_vq:
 557         vq->private_data = NULL;
 558         mutex_unlock(&vq->mutex);
 559
 560         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 561                 vq = &vsock->vqs[i];
 562
 563                 mutex_lock(&vq->mutex);
 564                 vq->private_data = NULL;
 565                 mutex_unlock(&vq->mutex);
 566         }
 567 err:
 568         mutex_unlock(&vsock->dev.mutex);
 569         return ret;
 570 }
 571
 572 static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner)
 573 {
 574         size_t i;
 575         int ret = 0;
 576
 577         mutex_lock(&vsock->dev.mutex);
 578
 579         if (check_owner) {
 580                 ret = vhost_dev_check_owner(&vsock->dev);
 581                 if (ret)
 582                         goto err;
 583         }
 584
 585         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 586                 struct vhost_virtqueue *vq = &vsock->vqs[i];
 587
 588                 mutex_lock(&vq->mutex);
 589                 vq->private_data = NULL;
 590                 mutex_unlock(&vq->mutex);
 591         }
 592
 593 err:
 594         mutex_unlock(&vsock->dev.mutex);
 595         return ret;
 596 }
 597
 598 static void vhost_vsock_free(struct vhost_vsock *vsock)
 599 {
 600         kvfree(vsock);
 601 }
 602
 603 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
 604 {
 605         struct vhost_virtqueue **vqs;
 606         struct vhost_vsock *vsock;
 607         int ret;
 608
 609         /* This struct is large and allocation could fail, fall back to vmalloc
 610          * if there is no other way.
 611          */
 612         vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
 613         if (!vsock)
 614                 return -ENOMEM;
 615
 616         vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
 617         if (!vqs) {
 618                 ret = -ENOMEM;
 619                 goto out;
 620         }
 621
 622         vsock->guest_cid = 0; /* no CID assigned yet */
 623
 624         atomic_set(&vsock->queued_replies, 0);
 625
 626         vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
 627         vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
 628         vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
 629         vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
 630
 631         vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
 632                        UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
 633                        VHOST_VSOCK_WEIGHT);
 634
 635         file->private_data = vsock;
 636         spin_lock_init(&vsock->send_pkt_list_lock);
 637         INIT_LIST_HEAD(&vsock->send_pkt_list);
 638         vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
 639         return 0;
 640
 641 out:
 642         vhost_vsock_free(vsock);
 643         return ret;
 644 }
 645
 646 static void vhost_vsock_flush(struct vhost_vsock *vsock)
 647 {
 648         int i;
 649
 650         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
 651                 if (vsock->vqs[i].handle_kick)
 652                         vhost_poll_flush(&vsock->vqs[i].poll);
 653         vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
 654 }
 655
 656 static void vhost_vsock_reset_orphans(struct sock *sk)
 657 {
 658         struct vsock_sock *vsk = vsock_sk(sk);
 659
 660         /* vmci_transport.c doesn't take sk_lock here either.  At least we're
 661          * under vsock_table_lock so the sock cannot disappear while we're
 662          * executing.
 663          */
 664
 665         /* If the peer is still valid, no need to reset connection */
 666         if (vhost_vsock_get(vsk->remote_addr.svm_cid))
 667                 return;
 668
 669         /* If the close timeout is pending, let it expire.  This avoids races
 670          * with the timeout callback.
 671          */
 672         if (vsk->close_work_scheduled)
 673                 return;
 674
 675         sock_set_flag(sk, SOCK_DONE);
 676         vsk->peer_shutdown = SHUTDOWN_MASK;
 677         sk->sk_state = SS_UNCONNECTED;
 678         sk->sk_err = ECONNRESET;
 679         sk->sk_error_report(sk);
 680 }
 681
 682 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
 683 {
 684         struct vhost_vsock *vsock = file->private_data;
 685
 686         spin_lock_bh(&vhost_vsock_lock);
 687         if (vsock->guest_cid)
 688                 hash_del_rcu(&vsock->hash);
 689         spin_unlock_bh(&vhost_vsock_lock);
 690
 691         /* Wait for other CPUs to finish using vsock */
 692         synchronize_rcu();
 693
 694         /* Iterating over all connections for all CIDs to find orphans is
 695          * inefficient.  Room for improvement here. */
 696         vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
 697
 698         /* Don't check the owner, because we are in the release path, so we
 699          * need to stop the vsock device in any case.
 700          * vhost_vsock_stop() can not fail in this case, so we don't need to
 701          * check the return code.
 702          */
 703         vhost_vsock_stop(vsock, false);
 704         vhost_vsock_flush(vsock);
 705         vhost_dev_stop(&vsock->dev);
 706
 707         spin_lock_bh(&vsock->send_pkt_list_lock);
 708         while (!list_empty(&vsock->send_pkt_list)) {
 709                 struct virtio_vsock_pkt *pkt;
 710
 711                 pkt = list_first_entry(&vsock->send_pkt_list,
 712                                 struct virtio_vsock_pkt, list);
 713                 list_del_init(&pkt->list);
 714                 virtio_transport_free_pkt(pkt);
 715         }
 716         spin_unlock_bh(&vsock->send_pkt_list_lock);
 717
 718         vhost_dev_cleanup(&vsock->dev);
 719         kfree(vsock->dev.vqs);
 720         vhost_vsock_free(vsock);
 721         return 0;
 722 }
 723
 724 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
 725 {
 726         struct vhost_vsock *other;
 727
 728         /* Refuse reserved CIDs */
 729         if (guest_cid <= VMADDR_CID_HOST ||
 730             guest_cid == U32_MAX)
 731                 return -EINVAL;
 732
 733         /* 64-bit CIDs are not yet supported */
 734         if (guest_cid > U32_MAX)
 735                 return -EINVAL;
 736
 737         /* Refuse if CID is already in use */
 738         spin_lock_bh(&vhost_vsock_lock);
 739         other = vhost_vsock_get(guest_cid);
 740         if (other && other != vsock) {
 741                 spin_unlock_bh(&vhost_vsock_lock);
 742                 return -EADDRINUSE;
 743         }
 744
 745         if (vsock->guest_cid)
 746                 hash_del_rcu(&vsock->hash);
 747
 748         vsock->guest_cid = guest_cid;
 749         hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
 750         spin_unlock_bh(&vhost_vsock_lock);
 751
 752         return 0;
 753 }
 754
 755 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
 756 {
 757         struct vhost_virtqueue *vq;
 758         int i;
 759
 760         if (features & ~VHOST_VSOCK_FEATURES)
 761                 return -EOPNOTSUPP;
 762
 763         mutex_lock(&vsock->dev.mutex);
 764         if ((features & (1 << VHOST_F_LOG_ALL)) &&
 765             !vhost_log_access_ok(&vsock->dev)) {
 766                 mutex_unlock(&vsock->dev.mutex);
 767                 return -EFAULT;
 768         }
 769
 770         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 771                 vq = &vsock->vqs[i];
 772                 mutex_lock(&vq->mutex);
 773                 vq->acked_features = features;
 774                 mutex_unlock(&vq->mutex);
 775         }
 776         mutex_unlock(&vsock->dev.mutex);
 777         return 0;
 778 }
 779
 780 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
 781                                   unsigned long arg)
 782 {
 783         struct vhost_vsock *vsock = f->private_data;
 784         void __user *argp = (void __user *)arg;
 785         u64 guest_cid;
 786         u64 features;
 787         int start;
 788         int r;
 789
 790         switch (ioctl) {
 791         case VHOST_VSOCK_SET_GUEST_CID:
 792                 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
 793                         return -EFAULT;
 794                 return vhost_vsock_set_cid(vsock, guest_cid);
 795         case VHOST_VSOCK_SET_RUNNING:
 796                 if (copy_from_user(&start, argp, sizeof(start)))
 797                         return -EFAULT;
 798                 if (start)
 799                         return vhost_vsock_start(vsock);
 800                 else
 801                         return vhost_vsock_stop(vsock, true);
 802         case VHOST_GET_FEATURES:
 803                 features = VHOST_VSOCK_FEATURES;
 804                 if (copy_to_user(argp, &features, sizeof(features)))
 805                         return -EFAULT;
 806                 return 0;
 807         case VHOST_SET_FEATURES:
 808                 if (copy_from_user(&features, argp, sizeof(features)))
 809                         return -EFAULT;
 810                 return vhost_vsock_set_features(vsock, features);
 811         default:
 812                 mutex_lock(&vsock->dev.mutex);
 813                 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
 814                 if (r == -ENOIOCTLCMD)
 815                         r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
 816                 else
 817                         vhost_vsock_flush(vsock);
 818                 mutex_unlock(&vsock->dev.mutex);
 819                 return r;
 820         }
 821 }
 822
 823 #ifdef CONFIG_COMPAT
 824 static long vhost_vsock_dev_compat_ioctl(struct file *f, unsigned int ioctl,
 825                                          unsigned long arg)
 826 {
 827         return vhost_vsock_dev_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
 828 }
 829 #endif
 830
 831 static const struct file_operations vhost_vsock_fops = {
 832         .owner          = THIS_MODULE,
 833         .open           = vhost_vsock_dev_open,
 834         .release        = vhost_vsock_dev_release,
 835         .llseek         = noop_llseek,
 836         .unlocked_ioctl = vhost_vsock_dev_ioctl,
 837 #ifdef CONFIG_COMPAT
 838         .compat_ioctl   = vhost_vsock_dev_compat_ioctl,
 839 #endif
 840 };
 841
 842 static struct miscdevice vhost_vsock_misc = {
 843         .minor = VHOST_VSOCK_MINOR,
 844         .name = "vhost-vsock",
 845         .fops = &vhost_vsock_fops,
 846 };
 847
 848 static int __init vhost_vsock_init(void)
 849 {
 850         int ret;
 851
 852         ret = vsock_core_init(&vhost_transport.transport);
 853         if (ret < 0)
 854                 return ret;
 855         return misc_register(&vhost_vsock_misc);
 856 };
 857
 858 static void __exit vhost_vsock_exit(void)
 859 {
 860         misc_deregister(&vhost_vsock_misc);
 861         vsock_core_exit();
 862 };
 863
 864 module_init(vhost_vsock_init);
 865 module_exit(vhost_vsock_exit);
 866 MODULE_LICENSE("GPL v2");
 867 MODULE_AUTHOR("Asias He");
 868 MODULE_DESCRIPTION("vhost transport for vsock ");
 869 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
 870 MODULE_ALIAS("devname:vhost-vsock");