drivers/vhost/vsock.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * vhost transport for vsock
   4  *
   5  * Copyright (C) 2013-2015 Red Hat, Inc.
   6  * Author: Asias He <asias@redhat.com>
   7  *         Stefan Hajnoczi <stefanha@redhat.com>
   8  */
   9 #include <linux/miscdevice.h>
  10 #include <linux/atomic.h>
  11 #include <linux/module.h>
  12 #include <linux/mutex.h>
  13 #include <linux/vmalloc.h>
  14 #include <net/sock.h>
  15 #include <linux/virtio_vsock.h>
  16 #include <linux/vhost.h>
  17 #include <linux/hashtable.h>
  18
  19 #include <net/af_vsock.h>
  20 #include "vhost.h"
  21
  22 #define VHOST_VSOCK_DEFAULT_HOST_CID    2
  23 /* Max number of bytes transferred before requeueing the job.
  24  * Using this limit prevents one virtqueue from starving others. */
  25 #define VHOST_VSOCK_WEIGHT 0x80000
  26 /* Max number of packets transferred before requeueing the job.
  27  * Using this limit prevents one virtqueue from starving others with
  28  * small pkts.
  29  */
  30 #define VHOST_VSOCK_PKT_WEIGHT 256
  31
  32 enum {
  33         VHOST_VSOCK_FEATURES = VHOST_FEATURES |
  34                                (1ULL << VIRTIO_F_ACCESS_PLATFORM)
  35 };
  36
  37 enum {
  38         VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
  39 };
  40
  41 /* Used to track all the vhost_vsock instances on the system. */
  42 static DEFINE_MUTEX(vhost_vsock_mutex);
  43 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
  44
  45 struct vhost_vsock {
  46         struct vhost_dev dev;
  47         struct vhost_virtqueue vqs[2];
  48
  49         /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
  50         struct hlist_node hash;
  51
  52         struct vhost_work send_pkt_work;
  53         spinlock_t send_pkt_list_lock;
  54         struct list_head send_pkt_list; /* host->guest pending packets */
  55
  56         atomic_t queued_replies;
  57
  58         u32 guest_cid;
  59 };
  60
  61 static u32 vhost_transport_get_local_cid(void)
  62 {
  63         return VHOST_VSOCK_DEFAULT_HOST_CID;
  64 }
  65
  66 /* Callers that dereference the return value must hold vhost_vsock_mutex or the
  67  * RCU read lock.
  68  */
  69 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
  70 {
  71         struct vhost_vsock *vsock;
  72
  73         hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
  74                 u32 other_cid = vsock->guest_cid;
  75
  76                 /* Skip instances that have no CID yet */
  77                 if (other_cid == 0)
  78                         continue;
  79
  80                 if (other_cid == guest_cid)
  81                         return vsock;
  82
  83         }
  84
  85         return NULL;
  86 }
  87
  88 static void
  89 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
  90                             struct vhost_virtqueue *vq)
  91 {
  92         struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
  93         int pkts = 0, total_len = 0;
  94         bool added = false;
  95         bool restart_tx = false;
  96
  97         mutex_lock(&vq->mutex);
  98
  99         if (!vhost_vq_get_backend(vq))
 100                 goto out;
 101
 102         if (!vq_meta_prefetch(vq))
 103                 goto out;
 104
 105         /* Avoid further vmexits, we're already processing the virtqueue */
 106         vhost_disable_notify(&vsock->dev, vq);
 107
 108         do {
 109                 struct virtio_vsock_pkt *pkt;
 110                 struct iov_iter iov_iter;
 111                 unsigned out, in;
 112                 size_t nbytes;
 113                 size_t iov_len, payload_len;
 114                 int head;
 115
 116                 spin_lock_bh(&vsock->send_pkt_list_lock);
 117                 if (list_empty(&vsock->send_pkt_list)) {
 118                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 119                         vhost_enable_notify(&vsock->dev, vq);
 120                         break;
 121                 }
 122
 123                 pkt = list_first_entry(&vsock->send_pkt_list,
 124                                        struct virtio_vsock_pkt, list);
 125                 list_del_init(&pkt->list);
 126                 spin_unlock_bh(&vsock->send_pkt_list_lock);
 127
 128                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 129                                          &out, &in, NULL, NULL);
 130                 if (head < 0) {
 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                         break;
 135                 }
 136
 137                 if (head == vq->num) {
 138                         spin_lock_bh(&vsock->send_pkt_list_lock);
 139                         list_add(&pkt->list, &vsock->send_pkt_list);
 140                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 141
 142                         /* We cannot finish yet if more buffers snuck in while
 143                          * re-enabling notify.
 144                          */
 145                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 146                                 vhost_disable_notify(&vsock->dev, vq);
 147                                 continue;
 148                         }
 149                         break;
 150                 }
 151
 152                 if (out) {
 153                         virtio_transport_free_pkt(pkt);
 154                         vq_err(vq, "Expected 0 output buffers, got %u\n", out);
 155                         break;
 156                 }
 157
 158                 iov_len = iov_length(&vq->iov[out], in);
 159                 if (iov_len < sizeof(pkt->hdr)) {
 160                         virtio_transport_free_pkt(pkt);
 161                         vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
 162                         break;
 163                 }
 164
 165                 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len);
 166                 payload_len = pkt->len - pkt->off;
 167
 168                 /* If the packet is greater than the space available in the
 169                  * buffer, we split it using multiple buffers.
 170                  */
 171                 if (payload_len > iov_len - sizeof(pkt->hdr))
 172                         payload_len = iov_len - sizeof(pkt->hdr);
 173
 174                 /* Set the correct length in the header */
 175                 pkt->hdr.len = cpu_to_le32(payload_len);
 176
 177                 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
 178                 if (nbytes != sizeof(pkt->hdr)) {
 179                         virtio_transport_free_pkt(pkt);
 180                         vq_err(vq, "Faulted on copying pkt hdr\n");
 181                         break;
 182                 }
 183
 184                 nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len,
 185                                       &iov_iter);
 186                 if (nbytes != payload_len) {
 187                         virtio_transport_free_pkt(pkt);
 188                         vq_err(vq, "Faulted on copying pkt buf\n");
 189                         break;
 190                 }
 191
 192                 /* Deliver to monitoring devices all packets that we
 193                  * will transmit.
 194                  */
 195                 virtio_transport_deliver_tap_pkt(pkt);
 196
 197                 vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len);
 198                 added = true;
 199
 200                 pkt->off += payload_len;
 201                 total_len += payload_len;
 202
 203                 /* If we didn't send all the payload we can requeue the packet
 204                  * to send it with the next available buffer.
 205                  */
 206                 if (pkt->off < pkt->len) {
 207                         /* We are queueing the same virtio_vsock_pkt to handle
 208                          * the remaining bytes, and we want to deliver it
 209                          * to monitoring devices in the next iteration.
 210                          */
 211                         pkt->tap_delivered = false;
 212
 213                         spin_lock_bh(&vsock->send_pkt_list_lock);
 214                         list_add(&pkt->list, &vsock->send_pkt_list);
 215                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 216                 } else {
 217                         if (pkt->reply) {
 218                                 int val;
 219
 220                                 val = atomic_dec_return(&vsock->queued_replies);
 221
 222                                 /* Do we have resources to resume tx
 223                                  * processing?
 224                                  */
 225                                 if (val + 1 == tx_vq->num)
 226                                         restart_tx = true;
 227                         }
 228
 229                         virtio_transport_free_pkt(pkt);
 230                 }
 231         } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
 232         if (added)
 233                 vhost_signal(&vsock->dev, vq);
 234
 235 out:
 236         mutex_unlock(&vq->mutex);
 237
 238         if (restart_tx)
 239                 vhost_poll_queue(&tx_vq->poll);
 240 }
 241
 242 static void vhost_transport_send_pkt_work(struct vhost_work *work)
 243 {
 244         struct vhost_virtqueue *vq;
 245         struct vhost_vsock *vsock;
 246
 247         vsock = container_of(work, struct vhost_vsock, send_pkt_work);
 248         vq = &vsock->vqs[VSOCK_VQ_RX];
 249
 250         vhost_transport_do_send_pkt(vsock, vq);
 251 }
 252
 253 static int
 254 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
 255 {
 256         struct vhost_vsock *vsock;
 257         int len = pkt->len;
 258
 259         rcu_read_lock();
 260
 261         /* Find the vhost_vsock according to guest context id  */
 262         vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
 263         if (!vsock) {
 264                 rcu_read_unlock();
 265                 virtio_transport_free_pkt(pkt);
 266                 return -ENODEV;
 267         }
 268
 269         if (pkt->reply)
 270                 atomic_inc(&vsock->queued_replies);
 271
 272         spin_lock_bh(&vsock->send_pkt_list_lock);
 273         list_add_tail(&pkt->list, &vsock->send_pkt_list);
 274         spin_unlock_bh(&vsock->send_pkt_list_lock);
 275
 276         vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
 277
 278         rcu_read_unlock();
 279         return len;
 280 }
 281
 282 static int
 283 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
 284 {
 285         struct vhost_vsock *vsock;
 286         struct virtio_vsock_pkt *pkt, *n;
 287         int cnt = 0;
 288         int ret = -ENODEV;
 289         LIST_HEAD(freeme);
 290
 291         rcu_read_lock();
 292
 293         /* Find the vhost_vsock according to guest context id  */
 294         vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
 295         if (!vsock)
 296                 goto out;
 297
 298         spin_lock_bh(&vsock->send_pkt_list_lock);
 299         list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
 300                 if (pkt->vsk != vsk)
 301                         continue;
 302                 list_move(&pkt->list, &freeme);
 303         }
 304         spin_unlock_bh(&vsock->send_pkt_list_lock);
 305
 306         list_for_each_entry_safe(pkt, n, &freeme, list) {
 307                 if (pkt->reply)
 308                         cnt++;
 309                 list_del(&pkt->list);
 310                 virtio_transport_free_pkt(pkt);
 311         }
 312
 313         if (cnt) {
 314                 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
 315                 int new_cnt;
 316
 317                 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
 318                 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
 319                         vhost_poll_queue(&tx_vq->poll);
 320         }
 321
 322         ret = 0;
 323 out:
 324         rcu_read_unlock();
 325         return ret;
 326 }
 327
 328 static struct virtio_vsock_pkt *
 329 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
 330                       unsigned int out, unsigned int in)
 331 {
 332         struct virtio_vsock_pkt *pkt;
 333         struct iov_iter iov_iter;
 334         size_t nbytes;
 335         size_t len;
 336
 337         if (in != 0) {
 338                 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
 339                 return NULL;
 340         }
 341
 342         pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
 343         if (!pkt)
 344                 return NULL;
 345
 346         len = iov_length(vq->iov, out);
 347         iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
 348
 349         nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
 350         if (nbytes != sizeof(pkt->hdr)) {
 351                 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
 352                        sizeof(pkt->hdr), nbytes);
 353                 kfree(pkt);
 354                 return NULL;
 355         }
 356
 357         if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
 358                 pkt->len = le32_to_cpu(pkt->hdr.len);
 359
 360         /* No payload */
 361         if (!pkt->len)
 362                 return pkt;
 363
 364         /* The pkt is too big */
 365         if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
 366                 kfree(pkt);
 367                 return NULL;
 368         }
 369
 370         pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
 371         if (!pkt->buf) {
 372                 kfree(pkt);
 373                 return NULL;
 374         }
 375
 376         pkt->buf_len = pkt->len;
 377
 378         nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
 379         if (nbytes != pkt->len) {
 380                 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
 381                        pkt->len, nbytes);
 382                 virtio_transport_free_pkt(pkt);
 383                 return NULL;
 384         }
 385
 386         return pkt;
 387 }
 388
 389 /* Is there space left for replies to rx packets? */
 390 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
 391 {
 392         struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
 393         int val;
 394
 395         smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
 396         val = atomic_read(&vsock->queued_replies);
 397
 398         return val < vq->num;
 399 }
 400
 401 static struct virtio_transport vhost_transport = {
 402         .transport = {
 403                 .module                   = THIS_MODULE,
 404
 405                 .get_local_cid            = vhost_transport_get_local_cid,
 406
 407                 .init                     = virtio_transport_do_socket_init,
 408                 .destruct                 = virtio_transport_destruct,
 409                 .release                  = virtio_transport_release,
 410                 .connect                  = virtio_transport_connect,
 411                 .shutdown                 = virtio_transport_shutdown,
 412                 .cancel_pkt               = vhost_transport_cancel_pkt,
 413
 414                 .dgram_enqueue            = virtio_transport_dgram_enqueue,
 415                 .dgram_dequeue            = virtio_transport_dgram_dequeue,
 416                 .dgram_bind               = virtio_transport_dgram_bind,
 417                 .dgram_allow              = virtio_transport_dgram_allow,
 418
 419                 .stream_enqueue           = virtio_transport_stream_enqueue,
 420                 .stream_dequeue           = virtio_transport_stream_dequeue,
 421                 .stream_has_data          = virtio_transport_stream_has_data,
 422                 .stream_has_space         = virtio_transport_stream_has_space,
 423                 .stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
 424                 .stream_is_active         = virtio_transport_stream_is_active,
 425                 .stream_allow             = virtio_transport_stream_allow,
 426
 427                 .notify_poll_in           = virtio_transport_notify_poll_in,
 428                 .notify_poll_out          = virtio_transport_notify_poll_out,
 429                 .notify_recv_init         = virtio_transport_notify_recv_init,
 430                 .notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
 431                 .notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
 432                 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
 433                 .notify_send_init         = virtio_transport_notify_send_init,
 434                 .notify_send_pre_block    = virtio_transport_notify_send_pre_block,
 435                 .notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
 436                 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
 437                 .notify_buffer_size       = virtio_transport_notify_buffer_size,
 438
 439         },
 440
 441         .send_pkt = vhost_transport_send_pkt,
 442 };
 443
 444 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
 445 {
 446         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 447                                                   poll.work);
 448         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 449                                                  dev);
 450         struct virtio_vsock_pkt *pkt;
 451         int head, pkts = 0, total_len = 0;
 452         unsigned int out, in;
 453         bool added = false;
 454
 455         mutex_lock(&vq->mutex);
 456
 457         if (!vhost_vq_get_backend(vq))
 458                 goto out;
 459
 460         if (!vq_meta_prefetch(vq))
 461                 goto out;
 462
 463         vhost_disable_notify(&vsock->dev, vq);
 464         do {
 465                 u32 len;
 466
 467                 if (!vhost_vsock_more_replies(vsock)) {
 468                         /* Stop tx until the device processes already
 469                          * pending replies.  Leave tx virtqueue
 470                          * callbacks disabled.
 471                          */
 472                         goto no_more_replies;
 473                 }
 474
 475                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 476                                          &out, &in, NULL, NULL);
 477                 if (head < 0)
 478                         break;
 479
 480                 if (head == vq->num) {
 481                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 482                                 vhost_disable_notify(&vsock->dev, vq);
 483                                 continue;
 484                         }
 485                         break;
 486                 }
 487
 488                 pkt = vhost_vsock_alloc_pkt(vq, out, in);
 489                 if (!pkt) {
 490                         vq_err(vq, "Faulted on pkt\n");
 491                         continue;
 492                 }
 493
 494                 len = pkt->len;
 495
 496                 /* Deliver to monitoring devices all received packets */
 497                 virtio_transport_deliver_tap_pkt(pkt);
 498
 499                 /* Only accept correctly addressed packets */
 500                 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid &&
 501                     le64_to_cpu(pkt->hdr.dst_cid) ==
 502                     vhost_transport_get_local_cid())
 503                         virtio_transport_recv_pkt(&vhost_transport, pkt);
 504                 else
 505                         virtio_transport_free_pkt(pkt);
 506
 507                 len += sizeof(pkt->hdr);
 508                 vhost_add_used(vq, head, len);
 509                 total_len += len;
 510                 added = true;
 511         } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
 512
 513 no_more_replies:
 514         if (added)
 515                 vhost_signal(&vsock->dev, vq);
 516
 517 out:
 518         mutex_unlock(&vq->mutex);
 519 }
 520
 521 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
 522 {
 523         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 524                                                 poll.work);
 525         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 526                                                  dev);
 527
 528         vhost_transport_do_send_pkt(vsock, vq);
 529 }
 530
 531 static int vhost_vsock_start(struct vhost_vsock *vsock)
 532 {
 533         struct vhost_virtqueue *vq;
 534         size_t i;
 535         int ret;
 536
 537         mutex_lock(&vsock->dev.mutex);
 538
 539         ret = vhost_dev_check_owner(&vsock->dev);
 540         if (ret)
 541                 goto err;
 542
 543         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 544                 vq = &vsock->vqs[i];
 545
 546                 mutex_lock(&vq->mutex);
 547
 548                 if (!vhost_vq_access_ok(vq)) {
 549                         ret = -EFAULT;
 550                         goto err_vq;
 551                 }
 552
 553                 if (!vhost_vq_get_backend(vq)) {
 554                         vhost_vq_set_backend(vq, vsock);
 555                         ret = vhost_vq_init_access(vq);
 556                         if (ret)
 557                                 goto err_vq;
 558                 }
 559
 560                 mutex_unlock(&vq->mutex);
 561         }
 562
 563         /* Some packets may have been queued before the device was started,
 564          * let's kick the send worker to send them.
 565          */
 566         vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
 567
 568         mutex_unlock(&vsock->dev.mutex);
 569         return 0;
 570
 571 err_vq:
 572         vhost_vq_set_backend(vq, NULL);
 573         mutex_unlock(&vq->mutex);
 574
 575         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 576                 vq = &vsock->vqs[i];
 577
 578                 mutex_lock(&vq->mutex);
 579                 vhost_vq_set_backend(vq, NULL);
 580                 mutex_unlock(&vq->mutex);
 581         }
 582 err:
 583         mutex_unlock(&vsock->dev.mutex);
 584         return ret;
 585 }
 586
 587 static int vhost_vsock_stop(struct vhost_vsock *vsock)
 588 {
 589         size_t i;
 590         int ret;
 591
 592         mutex_lock(&vsock->dev.mutex);
 593
 594         ret = vhost_dev_check_owner(&vsock->dev);
 595         if (ret)
 596                 goto err;
 597
 598         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 599                 struct vhost_virtqueue *vq = &vsock->vqs[i];
 600
 601                 mutex_lock(&vq->mutex);
 602                 vhost_vq_set_backend(vq, NULL);
 603                 mutex_unlock(&vq->mutex);
 604         }
 605
 606 err:
 607         mutex_unlock(&vsock->dev.mutex);
 608         return ret;
 609 }
 610
 611 static void vhost_vsock_free(struct vhost_vsock *vsock)
 612 {
 613         kvfree(vsock);
 614 }
 615
 616 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
 617 {
 618         struct vhost_virtqueue **vqs;
 619         struct vhost_vsock *vsock;
 620         int ret;
 621
 622         /* This struct is large and allocation could fail, fall back to vmalloc
 623          * if there is no other way.
 624          */
 625         vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
 626         if (!vsock)
 627                 return -ENOMEM;
 628
 629         vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
 630         if (!vqs) {
 631                 ret = -ENOMEM;
 632                 goto out;
 633         }
 634
 635         vsock->guest_cid = 0; /* no CID assigned yet */
 636
 637         atomic_set(&vsock->queued_replies, 0);
 638
 639         vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
 640         vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
 641         vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
 642         vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
 643
 644         vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
 645                        UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
 646                        VHOST_VSOCK_WEIGHT, true, NULL);
 647
 648         file->private_data = vsock;
 649         spin_lock_init(&vsock->send_pkt_list_lock);
 650         INIT_LIST_HEAD(&vsock->send_pkt_list);
 651         vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
 652         return 0;
 653
 654 out:
 655         vhost_vsock_free(vsock);
 656         return ret;
 657 }
 658
 659 static void vhost_vsock_flush(struct vhost_vsock *vsock)
 660 {
 661         int i;
 662
 663         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
 664                 if (vsock->vqs[i].handle_kick)
 665                         vhost_poll_flush(&vsock->vqs[i].poll);
 666         vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
 667 }
 668
 669 static void vhost_vsock_reset_orphans(struct sock *sk)
 670 {
 671         struct vsock_sock *vsk = vsock_sk(sk);
 672
 673         /* vmci_transport.c doesn't take sk_lock here either.  At least we're
 674          * under vsock_table_lock so the sock cannot disappear while we're
 675          * executing.
 676          */
 677
 678         /* If the peer is still valid, no need to reset connection */
 679         if (vhost_vsock_get(vsk->remote_addr.svm_cid))
 680                 return;
 681
 682         /* If the close timeout is pending, let it expire.  This avoids races
 683          * with the timeout callback.
 684          */
 685         if (vsk->close_work_scheduled)
 686                 return;
 687
 688         sock_set_flag(sk, SOCK_DONE);
 689         vsk->peer_shutdown = SHUTDOWN_MASK;
 690         sk->sk_state = SS_UNCONNECTED;
 691         sk->sk_err = ECONNRESET;
 692         sk->sk_error_report(sk);
 693 }
 694
 695 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
 696 {
 697         struct vhost_vsock *vsock = file->private_data;
 698
 699         mutex_lock(&vhost_vsock_mutex);
 700         if (vsock->guest_cid)
 701                 hash_del_rcu(&vsock->hash);
 702         mutex_unlock(&vhost_vsock_mutex);
 703
 704         /* Wait for other CPUs to finish using vsock */
 705         synchronize_rcu();
 706
 707         /* Iterating over all connections for all CIDs to find orphans is
 708          * inefficient.  Room for improvement here. */
 709         vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
 710
 711         vhost_vsock_stop(vsock);
 712         vhost_vsock_flush(vsock);
 713         vhost_dev_stop(&vsock->dev);
 714
 715         spin_lock_bh(&vsock->send_pkt_list_lock);
 716         while (!list_empty(&vsock->send_pkt_list)) {
 717                 struct virtio_vsock_pkt *pkt;
 718
 719                 pkt = list_first_entry(&vsock->send_pkt_list,
 720                                 struct virtio_vsock_pkt, list);
 721                 list_del_init(&pkt->list);
 722                 virtio_transport_free_pkt(pkt);
 723         }
 724         spin_unlock_bh(&vsock->send_pkt_list_lock);
 725
 726         vhost_dev_cleanup(&vsock->dev);
 727         kfree(vsock->dev.vqs);
 728         vhost_vsock_free(vsock);
 729         return 0;
 730 }
 731
 732 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
 733 {
 734         struct vhost_vsock *other;
 735
 736         /* Refuse reserved CIDs */
 737         if (guest_cid <= VMADDR_CID_HOST ||
 738             guest_cid == U32_MAX)
 739                 return -EINVAL;
 740
 741         /* 64-bit CIDs are not yet supported */
 742         if (guest_cid > U32_MAX)
 743                 return -EINVAL;
 744
 745         /* Refuse if CID is assigned to the guest->host transport (i.e. nested
 746          * VM), to make the loopback work.
 747          */
 748         if (vsock_find_cid(guest_cid))
 749                 return -EADDRINUSE;
 750
 751         /* Refuse if CID is already in use */
 752         mutex_lock(&vhost_vsock_mutex);
 753         other = vhost_vsock_get(guest_cid);
 754         if (other && other != vsock) {
 755                 mutex_unlock(&vhost_vsock_mutex);
 756                 return -EADDRINUSE;
 757         }
 758
 759         if (vsock->guest_cid)
 760                 hash_del_rcu(&vsock->hash);
 761
 762         vsock->guest_cid = guest_cid;
 763         hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
 764         mutex_unlock(&vhost_vsock_mutex);
 765
 766         return 0;
 767 }
 768
 769 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
 770 {
 771         struct vhost_virtqueue *vq;
 772         int i;
 773
 774         if (features & ~VHOST_VSOCK_FEATURES)
 775                 return -EOPNOTSUPP;
 776
 777         mutex_lock(&vsock->dev.mutex);
 778         if ((features & (1 << VHOST_F_LOG_ALL)) &&
 779             !vhost_log_access_ok(&vsock->dev)) {
 780                 goto err;
 781         }
 782
 783         if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
 784                 if (vhost_init_device_iotlb(&vsock->dev, true))
 785                         goto err;
 786         }
 787
 788         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 789                 vq = &vsock->vqs[i];
 790                 mutex_lock(&vq->mutex);
 791                 vq->acked_features = features;
 792                 mutex_unlock(&vq->mutex);
 793         }
 794         mutex_unlock(&vsock->dev.mutex);
 795         return 0;
 796
 797 err:
 798         mutex_unlock(&vsock->dev.mutex);
 799         return -EFAULT;
 800 }
 801
 802 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
 803                                   unsigned long arg)
 804 {
 805         struct vhost_vsock *vsock = f->private_data;
 806         void __user *argp = (void __user *)arg;
 807         u64 guest_cid;
 808         u64 features;
 809         int start;
 810         int r;
 811
 812         switch (ioctl) {
 813         case VHOST_VSOCK_SET_GUEST_CID:
 814                 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
 815                         return -EFAULT;
 816                 return vhost_vsock_set_cid(vsock, guest_cid);
 817         case VHOST_VSOCK_SET_RUNNING:
 818                 if (copy_from_user(&start, argp, sizeof(start)))
 819                         return -EFAULT;
 820                 if (start)
 821                         return vhost_vsock_start(vsock);
 822                 else
 823                         return vhost_vsock_stop(vsock);
 824         case VHOST_GET_FEATURES:
 825                 features = VHOST_VSOCK_FEATURES;
 826                 if (copy_to_user(argp, &features, sizeof(features)))
 827                         return -EFAULT;
 828                 return 0;
 829         case VHOST_SET_FEATURES:
 830                 if (copy_from_user(&features, argp, sizeof(features)))
 831                         return -EFAULT;
 832                 return vhost_vsock_set_features(vsock, features);
 833         case VHOST_GET_BACKEND_FEATURES:
 834                 features = VHOST_VSOCK_BACKEND_FEATURES;
 835                 if (copy_to_user(argp, &features, sizeof(features)))
 836                         return -EFAULT;
 837                 return 0;
 838         case VHOST_SET_BACKEND_FEATURES:
 839                 if (copy_from_user(&features, argp, sizeof(features)))
 840                         return -EFAULT;
 841                 if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
 842                         return -EOPNOTSUPP;
 843                 vhost_set_backend_features(&vsock->dev, features);
 844                 return 0;
 845         default:
 846                 mutex_lock(&vsock->dev.mutex);
 847                 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
 848                 if (r == -ENOIOCTLCMD)
 849                         r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
 850                 else
 851                         vhost_vsock_flush(vsock);
 852                 mutex_unlock(&vsock->dev.mutex);
 853                 return r;
 854         }
 855 }
 856
 857 static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
 858 {
 859         struct file *file = iocb->ki_filp;
 860         struct vhost_vsock *vsock = file->private_data;
 861         struct vhost_dev *dev = &vsock->dev;
 862         int noblock = file->f_flags & O_NONBLOCK;
 863
 864         return vhost_chr_read_iter(dev, to, noblock);
 865 }
 866
 867 static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
 868                                         struct iov_iter *from)
 869 {
 870         struct file *file = iocb->ki_filp;
 871         struct vhost_vsock *vsock = file->private_data;
 872         struct vhost_dev *dev = &vsock->dev;
 873
 874         return vhost_chr_write_iter(dev, from);
 875 }
 876
 877 static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
 878 {
 879         struct vhost_vsock *vsock = file->private_data;
 880         struct vhost_dev *dev = &vsock->dev;
 881
 882         return vhost_chr_poll(file, dev, wait);
 883 }
 884
 885 static const struct file_operations vhost_vsock_fops = {
 886         .owner          = THIS_MODULE,
 887         .open           = vhost_vsock_dev_open,
 888         .release        = vhost_vsock_dev_release,
 889         .llseek         = noop_llseek,
 890         .unlocked_ioctl = vhost_vsock_dev_ioctl,
 891         .compat_ioctl   = compat_ptr_ioctl,
 892         .read_iter      = vhost_vsock_chr_read_iter,
 893         .write_iter     = vhost_vsock_chr_write_iter,
 894         .poll           = vhost_vsock_chr_poll,
 895 };
 896
 897 static struct miscdevice vhost_vsock_misc = {
 898         .minor = VHOST_VSOCK_MINOR,
 899         .name = "vhost-vsock",
 900         .fops = &vhost_vsock_fops,
 901 };
 902
 903 static int __init vhost_vsock_init(void)
 904 {
 905         int ret;
 906
 907         ret = vsock_core_register(&vhost_transport.transport,
 908                                   VSOCK_TRANSPORT_F_H2G);
 909         if (ret < 0)
 910                 return ret;
 911         return misc_register(&vhost_vsock_misc);
 912 };
 913
 914 static void __exit vhost_vsock_exit(void)
 915 {
 916         misc_deregister(&vhost_vsock_misc);
 917         vsock_core_unregister(&vhost_transport.transport);
 918 };
 919
 920 module_init(vhost_vsock_init);
 921 module_exit(vhost_vsock_exit);
 922 MODULE_LICENSE("GPL v2");
 923 MODULE_AUTHOR("Asias He");
 924 MODULE_DESCRIPTION("vhost transport for vsock ");
 925 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
 926 MODULE_ALIAS("devname:vhost-vsock");