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                                (1ULL << VIRTIO_VSOCK_F_SEQPACKET)
  36 };
  37
  38 enum {
  39         VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
  40 };
  41
  42 /* Used to track all the vhost_vsock instances on the system. */
  43 static DEFINE_MUTEX(vhost_vsock_mutex);
  44 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
  45
  46 struct vhost_vsock {
  47         struct vhost_dev dev;
  48         struct vhost_virtqueue vqs[2];
  49
  50         /* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
  51         struct hlist_node hash;
  52
  53         struct vhost_work send_pkt_work;
  54         struct sk_buff_head send_pkt_queue; /* host->guest pending packets */
  55
  56         atomic_t queued_replies;
  57
  58         u32 guest_cid;
  59         bool seqpacket_allow;
  60 };
  61
  62 static u32 vhost_transport_get_local_cid(void)
  63 {
  64         return VHOST_VSOCK_DEFAULT_HOST_CID;
  65 }
  66
  67 /* Callers that dereference the return value must hold vhost_vsock_mutex or the
  68  * RCU read lock.
  69  */
  70 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
  71 {
  72         struct vhost_vsock *vsock;
  73
  74         hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
  75                 u32 other_cid = vsock->guest_cid;
  76
  77                 /* Skip instances that have no CID yet */
  78                 if (other_cid == 0)
  79                         continue;
  80
  81                 if (other_cid == guest_cid)
  82                         return vsock;
  83
  84         }
  85
  86         return NULL;
  87 }
  88
  89 static void
  90 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
  91                             struct vhost_virtqueue *vq)
  92 {
  93         struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
  94         int pkts = 0, total_len = 0;
  95         bool added = false;
  96         bool restart_tx = false;
  97
  98         mutex_lock(&vq->mutex);
  99
 100         if (!vhost_vq_get_backend(vq))
 101                 goto out;
 102
 103         if (!vq_meta_prefetch(vq))
 104                 goto out;
 105
 106         /* Avoid further vmexits, we're already processing the virtqueue */
 107         vhost_disable_notify(&vsock->dev, vq);
 108
 109         do {
 110                 struct virtio_vsock_hdr *hdr;
 111                 size_t iov_len, payload_len;
 112                 struct iov_iter iov_iter;
 113                 u32 flags_to_restore = 0;
 114                 struct sk_buff *skb;
 115                 unsigned out, in;
 116                 size_t nbytes;
 117                 int head;
 118
 119                 skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
 120
 121                 if (!skb) {
 122                         vhost_enable_notify(&vsock->dev, vq);
 123                         break;
 124                 }
 125
 126                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 127                                          &out, &in, NULL, NULL);
 128                 if (head < 0) {
 129                         virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
 130                         break;
 131                 }
 132
 133                 if (head == vq->num) {
 134                         virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
 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                         kfree_skb(skb);
 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(*hdr)) {
 153                         kfree_skb(skb);
 154                         vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
 155                         break;
 156                 }
 157
 158                 iov_iter_init(&iov_iter, ITER_DEST, &vq->iov[out], in, iov_len);
 159                 payload_len = skb->len;
 160                 hdr = virtio_vsock_hdr(skb);
 161
 162                 /* If the packet is greater than the space available in the
 163                  * buffer, we split it using multiple buffers.
 164                  */
 165                 if (payload_len > iov_len - sizeof(*hdr)) {
 166                         payload_len = iov_len - sizeof(*hdr);
 167
 168                         /* As we are copying pieces of large packet's buffer to
 169                          * small rx buffers, headers of packets in rx queue are
 170                          * created dynamically and are initialized with header
 171                          * of current packet(except length). But in case of
 172                          * SOCK_SEQPACKET, we also must clear message delimeter
 173                          * bit (VIRTIO_VSOCK_SEQ_EOM) and MSG_EOR bit
 174                          * (VIRTIO_VSOCK_SEQ_EOR) if set. Otherwise,
 175                          * there will be sequence of packets with these
 176                          * bits set. After initialized header will be copied to
 177                          * rx buffer, these required bits will be restored.
 178                          */
 179                         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
 180                                 hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
 181                                 flags_to_restore |= VIRTIO_VSOCK_SEQ_EOM;
 182
 183                                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) {
 184                                         hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
 185                                         flags_to_restore |= VIRTIO_VSOCK_SEQ_EOR;
 186                                 }
 187                         }
 188                 }
 189
 190                 /* Set the correct length in the header */
 191                 hdr->len = cpu_to_le32(payload_len);
 192
 193                 nbytes = copy_to_iter(hdr, sizeof(*hdr), &iov_iter);
 194                 if (nbytes != sizeof(*hdr)) {
 195                         kfree_skb(skb);
 196                         vq_err(vq, "Faulted on copying pkt hdr\n");
 197                         break;
 198                 }
 199
 200                 nbytes = copy_to_iter(skb->data, payload_len, &iov_iter);
 201                 if (nbytes != payload_len) {
 202                         kfree_skb(skb);
 203                         vq_err(vq, "Faulted on copying pkt buf\n");
 204                         break;
 205                 }
 206
 207                 /* Deliver to monitoring devices all packets that we
 208                  * will transmit.
 209                  */
 210                 virtio_transport_deliver_tap_pkt(skb);
 211
 212                 vhost_add_used(vq, head, sizeof(*hdr) + payload_len);
 213                 added = true;
 214
 215                 skb_pull(skb, payload_len);
 216                 total_len += payload_len;
 217
 218                 /* If we didn't send all the payload we can requeue the packet
 219                  * to send it with the next available buffer.
 220                  */
 221                 if (skb->len > 0) {
 222                         hdr->flags |= cpu_to_le32(flags_to_restore);
 223
 224                         /* We are queueing the same skb to handle
 225                          * the remaining bytes, and we want to deliver it
 226                          * to monitoring devices in the next iteration.
 227                          */
 228                         virtio_vsock_skb_clear_tap_delivered(skb);
 229                         virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
 230                 } else {
 231                         if (virtio_vsock_skb_reply(skb)) {
 232                                 int val;
 233
 234                                 val = atomic_dec_return(&vsock->queued_replies);
 235
 236                                 /* Do we have resources to resume tx
 237                                  * processing?
 238                                  */
 239                                 if (val + 1 == tx_vq->num)
 240                                         restart_tx = true;
 241                         }
 242
 243                         consume_skb(skb);
 244                 }
 245         } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
 246         if (added)
 247                 vhost_signal(&vsock->dev, vq);
 248
 249 out:
 250         mutex_unlock(&vq->mutex);
 251
 252         if (restart_tx)
 253                 vhost_poll_queue(&tx_vq->poll);
 254 }
 255
 256 static void vhost_transport_send_pkt_work(struct vhost_work *work)
 257 {
 258         struct vhost_virtqueue *vq;
 259         struct vhost_vsock *vsock;
 260
 261         vsock = container_of(work, struct vhost_vsock, send_pkt_work);
 262         vq = &vsock->vqs[VSOCK_VQ_RX];
 263
 264         vhost_transport_do_send_pkt(vsock, vq);
 265 }
 266
 267 static int
 268 vhost_transport_send_pkt(struct sk_buff *skb)
 269 {
 270         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
 271         struct vhost_vsock *vsock;
 272         int len = skb->len;
 273
 274         rcu_read_lock();
 275
 276         /* Find the vhost_vsock according to guest context id  */
 277         vsock = vhost_vsock_get(le64_to_cpu(hdr->dst_cid));
 278         if (!vsock) {
 279                 rcu_read_unlock();
 280                 kfree_skb(skb);
 281                 return -ENODEV;
 282         }
 283
 284         if (virtio_vsock_skb_reply(skb))
 285                 atomic_inc(&vsock->queued_replies);
 286
 287         virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb);
 288         vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work);
 289
 290         rcu_read_unlock();
 291         return len;
 292 }
 293
 294 static int
 295 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
 296 {
 297         struct vhost_vsock *vsock;
 298         int cnt = 0;
 299         int ret = -ENODEV;
 300
 301         rcu_read_lock();
 302
 303         /* Find the vhost_vsock according to guest context id  */
 304         vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
 305         if (!vsock)
 306                 goto out;
 307
 308         cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue);
 309
 310         if (cnt) {
 311                 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
 312                 int new_cnt;
 313
 314                 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
 315                 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
 316                         vhost_poll_queue(&tx_vq->poll);
 317         }
 318
 319         ret = 0;
 320 out:
 321         rcu_read_unlock();
 322         return ret;
 323 }
 324
 325 static struct sk_buff *
 326 vhost_vsock_alloc_skb(struct vhost_virtqueue *vq,
 327                       unsigned int out, unsigned int in)
 328 {
 329         struct virtio_vsock_hdr *hdr;
 330         struct iov_iter iov_iter;
 331         struct sk_buff *skb;
 332         size_t payload_len;
 333         size_t nbytes;
 334         size_t len;
 335
 336         if (in != 0) {
 337                 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
 338                 return NULL;
 339         }
 340
 341         len = iov_length(vq->iov, out);
 342
 343         /* len contains both payload and hdr */
 344         skb = virtio_vsock_alloc_skb(len, GFP_KERNEL);
 345         if (!skb)
 346                 return NULL;
 347
 348         iov_iter_init(&iov_iter, ITER_SOURCE, vq->iov, out, len);
 349
 350         hdr = virtio_vsock_hdr(skb);
 351         nbytes = copy_from_iter(hdr, sizeof(*hdr), &iov_iter);
 352         if (nbytes != sizeof(*hdr)) {
 353                 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
 354                        sizeof(*hdr), nbytes);
 355                 kfree_skb(skb);
 356                 return NULL;
 357         }
 358
 359         payload_len = le32_to_cpu(hdr->len);
 360
 361         /* No payload */
 362         if (!payload_len)
 363                 return skb;
 364
 365         /* The pkt is too big or the length in the header is invalid */
 366         if (payload_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE ||
 367             payload_len + sizeof(*hdr) > len) {
 368                 kfree_skb(skb);
 369                 return NULL;
 370         }
 371
 372         virtio_vsock_skb_rx_put(skb);
 373
 374         nbytes = copy_from_iter(skb->data, payload_len, &iov_iter);
 375         if (nbytes != payload_len) {
 376                 vq_err(vq, "Expected %zu byte payload, got %zu bytes\n",
 377                        payload_len, nbytes);
 378                 kfree_skb(skb);
 379                 return NULL;
 380         }
 381
 382         return skb;
 383 }
 384
 385 /* Is there space left for replies to rx packets? */
 386 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
 387 {
 388         struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
 389         int val;
 390
 391         smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
 392         val = atomic_read(&vsock->queued_replies);
 393
 394         return val < vq->num;
 395 }
 396
 397 static bool vhost_transport_seqpacket_allow(u32 remote_cid);
 398
 399 static struct virtio_transport vhost_transport = {
 400         .transport = {
 401                 .module                   = THIS_MODULE,
 402
 403                 .get_local_cid            = vhost_transport_get_local_cid,
 404
 405                 .init                     = virtio_transport_do_socket_init,
 406                 .destruct                 = virtio_transport_destruct,
 407                 .release                  = virtio_transport_release,
 408                 .connect                  = virtio_transport_connect,
 409                 .shutdown                 = virtio_transport_shutdown,
 410                 .cancel_pkt               = vhost_transport_cancel_pkt,
 411
 412                 .dgram_enqueue            = virtio_transport_dgram_enqueue,
 413                 .dgram_dequeue            = virtio_transport_dgram_dequeue,
 414                 .dgram_bind               = virtio_transport_dgram_bind,
 415                 .dgram_allow              = virtio_transport_dgram_allow,
 416
 417                 .stream_enqueue           = virtio_transport_stream_enqueue,
 418                 .stream_dequeue           = virtio_transport_stream_dequeue,
 419                 .stream_has_data          = virtio_transport_stream_has_data,
 420                 .stream_has_space         = virtio_transport_stream_has_space,
 421                 .stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
 422                 .stream_is_active         = virtio_transport_stream_is_active,
 423                 .stream_allow             = virtio_transport_stream_allow,
 424
 425                 .seqpacket_dequeue        = virtio_transport_seqpacket_dequeue,
 426                 .seqpacket_enqueue        = virtio_transport_seqpacket_enqueue,
 427                 .seqpacket_allow          = vhost_transport_seqpacket_allow,
 428                 .seqpacket_has_data       = virtio_transport_seqpacket_has_data,
 429
 430                 .notify_poll_in           = virtio_transport_notify_poll_in,
 431                 .notify_poll_out          = virtio_transport_notify_poll_out,
 432                 .notify_recv_init         = virtio_transport_notify_recv_init,
 433                 .notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
 434                 .notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
 435                 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
 436                 .notify_send_init         = virtio_transport_notify_send_init,
 437                 .notify_send_pre_block    = virtio_transport_notify_send_pre_block,
 438                 .notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
 439                 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
 440                 .notify_buffer_size       = virtio_transport_notify_buffer_size,
 441                 .notify_set_rcvlowat      = virtio_transport_notify_set_rcvlowat,
 442
 443                 .read_skb = virtio_transport_read_skb,
 444         },
 445
 446         .send_pkt = vhost_transport_send_pkt,
 447 };
 448
 449 static bool vhost_transport_seqpacket_allow(u32 remote_cid)
 450 {
 451         struct vhost_vsock *vsock;
 452         bool seqpacket_allow = false;
 453
 454         rcu_read_lock();
 455         vsock = vhost_vsock_get(remote_cid);
 456
 457         if (vsock)
 458                 seqpacket_allow = vsock->seqpacket_allow;
 459
 460         rcu_read_unlock();
 461
 462         return seqpacket_allow;
 463 }
 464
 465 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
 466 {
 467         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 468                                                   poll.work);
 469         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 470                                                  dev);
 471         int head, pkts = 0, total_len = 0;
 472         unsigned int out, in;
 473         struct sk_buff *skb;
 474         bool added = false;
 475
 476         mutex_lock(&vq->mutex);
 477
 478         if (!vhost_vq_get_backend(vq))
 479                 goto out;
 480
 481         if (!vq_meta_prefetch(vq))
 482                 goto out;
 483
 484         vhost_disable_notify(&vsock->dev, vq);
 485         do {
 486                 struct virtio_vsock_hdr *hdr;
 487
 488                 if (!vhost_vsock_more_replies(vsock)) {
 489                         /* Stop tx until the device processes already
 490                          * pending replies.  Leave tx virtqueue
 491                          * callbacks disabled.
 492                          */
 493                         goto no_more_replies;
 494                 }
 495
 496                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 497                                          &out, &in, NULL, NULL);
 498                 if (head < 0)
 499                         break;
 500
 501                 if (head == vq->num) {
 502                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 503                                 vhost_disable_notify(&vsock->dev, vq);
 504                                 continue;
 505                         }
 506                         break;
 507                 }
 508
 509                 skb = vhost_vsock_alloc_skb(vq, out, in);
 510                 if (!skb) {
 511                         vq_err(vq, "Faulted on pkt\n");
 512                         continue;
 513                 }
 514
 515                 total_len += sizeof(*hdr) + skb->len;
 516
 517                 /* Deliver to monitoring devices all received packets */
 518                 virtio_transport_deliver_tap_pkt(skb);
 519
 520                 hdr = virtio_vsock_hdr(skb);
 521
 522                 /* Only accept correctly addressed packets */
 523                 if (le64_to_cpu(hdr->src_cid) == vsock->guest_cid &&
 524                     le64_to_cpu(hdr->dst_cid) ==
 525                     vhost_transport_get_local_cid())
 526                         virtio_transport_recv_pkt(&vhost_transport, skb);
 527                 else
 528                         kfree_skb(skb);
 529
 530                 vhost_add_used(vq, head, 0);
 531                 added = true;
 532         } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
 533
 534 no_more_replies:
 535         if (added)
 536                 vhost_signal(&vsock->dev, vq);
 537
 538 out:
 539         mutex_unlock(&vq->mutex);
 540 }
 541
 542 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
 543 {
 544         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 545                                                 poll.work);
 546         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 547                                                  dev);
 548
 549         vhost_transport_do_send_pkt(vsock, vq);
 550 }
 551
 552 static int vhost_vsock_start(struct vhost_vsock *vsock)
 553 {
 554         struct vhost_virtqueue *vq;
 555         size_t i;
 556         int ret;
 557
 558         mutex_lock(&vsock->dev.mutex);
 559
 560         ret = vhost_dev_check_owner(&vsock->dev);
 561         if (ret)
 562                 goto err;
 563
 564         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 565                 vq = &vsock->vqs[i];
 566
 567                 mutex_lock(&vq->mutex);
 568
 569                 if (!vhost_vq_access_ok(vq)) {
 570                         ret = -EFAULT;
 571                         goto err_vq;
 572                 }
 573
 574                 if (!vhost_vq_get_backend(vq)) {
 575                         vhost_vq_set_backend(vq, vsock);
 576                         ret = vhost_vq_init_access(vq);
 577                         if (ret)
 578                                 goto err_vq;
 579                 }
 580
 581                 mutex_unlock(&vq->mutex);
 582         }
 583
 584         /* Some packets may have been queued before the device was started,
 585          * let's kick the send worker to send them.
 586          */
 587         vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work);
 588
 589         mutex_unlock(&vsock->dev.mutex);
 590         return 0;
 591
 592 err_vq:
 593         vhost_vq_set_backend(vq, NULL);
 594         mutex_unlock(&vq->mutex);
 595
 596         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 597                 vq = &vsock->vqs[i];
 598
 599                 mutex_lock(&vq->mutex);
 600                 vhost_vq_set_backend(vq, NULL);
 601                 mutex_unlock(&vq->mutex);
 602         }
 603 err:
 604         mutex_unlock(&vsock->dev.mutex);
 605         return ret;
 606 }
 607
 608 static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner)
 609 {
 610         size_t i;
 611         int ret = 0;
 612
 613         mutex_lock(&vsock->dev.mutex);
 614
 615         if (check_owner) {
 616                 ret = vhost_dev_check_owner(&vsock->dev);
 617                 if (ret)
 618                         goto err;
 619         }
 620
 621         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 622                 struct vhost_virtqueue *vq = &vsock->vqs[i];
 623
 624                 mutex_lock(&vq->mutex);
 625                 vhost_vq_set_backend(vq, NULL);
 626                 mutex_unlock(&vq->mutex);
 627         }
 628
 629 err:
 630         mutex_unlock(&vsock->dev.mutex);
 631         return ret;
 632 }
 633
 634 static void vhost_vsock_free(struct vhost_vsock *vsock)
 635 {
 636         kvfree(vsock);
 637 }
 638
 639 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
 640 {
 641         struct vhost_virtqueue **vqs;
 642         struct vhost_vsock *vsock;
 643         int ret;
 644
 645         /* This struct is large and allocation could fail, fall back to vmalloc
 646          * if there is no other way.
 647          */
 648         vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
 649         if (!vsock)
 650                 return -ENOMEM;
 651
 652         vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
 653         if (!vqs) {
 654                 ret = -ENOMEM;
 655                 goto out;
 656         }
 657
 658         vsock->guest_cid = 0; /* no CID assigned yet */
 659
 660         atomic_set(&vsock->queued_replies, 0);
 661
 662         vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
 663         vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
 664         vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
 665         vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
 666
 667         vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
 668                        UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
 669                        VHOST_VSOCK_WEIGHT, true, NULL);
 670
 671         file->private_data = vsock;
 672         skb_queue_head_init(&vsock->send_pkt_queue);
 673         vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
 674         return 0;
 675
 676 out:
 677         vhost_vsock_free(vsock);
 678         return ret;
 679 }
 680
 681 static void vhost_vsock_flush(struct vhost_vsock *vsock)
 682 {
 683         vhost_dev_flush(&vsock->dev);
 684 }
 685
 686 static void vhost_vsock_reset_orphans(struct sock *sk)
 687 {
 688         struct vsock_sock *vsk = vsock_sk(sk);
 689
 690         /* vmci_transport.c doesn't take sk_lock here either.  At least we're
 691          * under vsock_table_lock so the sock cannot disappear while we're
 692          * executing.
 693          */
 694
 695         /* If the peer is still valid, no need to reset connection */
 696         if (vhost_vsock_get(vsk->remote_addr.svm_cid))
 697                 return;
 698
 699         /* If the close timeout is pending, let it expire.  This avoids races
 700          * with the timeout callback.
 701          */
 702         if (vsk->close_work_scheduled)
 703                 return;
 704
 705         sock_set_flag(sk, SOCK_DONE);
 706         vsk->peer_shutdown = SHUTDOWN_MASK;
 707         sk->sk_state = SS_UNCONNECTED;
 708         sk->sk_err = ECONNRESET;
 709         sk_error_report(sk);
 710 }
 711
 712 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
 713 {
 714         struct vhost_vsock *vsock = file->private_data;
 715
 716         mutex_lock(&vhost_vsock_mutex);
 717         if (vsock->guest_cid)
 718                 hash_del_rcu(&vsock->hash);
 719         mutex_unlock(&vhost_vsock_mutex);
 720
 721         /* Wait for other CPUs to finish using vsock */
 722         synchronize_rcu();
 723
 724         /* Iterating over all connections for all CIDs to find orphans is
 725          * inefficient.  Room for improvement here. */
 726         vsock_for_each_connected_socket(&vhost_transport.transport,
 727                                         vhost_vsock_reset_orphans);
 728
 729         /* Don't check the owner, because we are in the release path, so we
 730          * need to stop the vsock device in any case.
 731          * vhost_vsock_stop() can not fail in this case, so we don't need to
 732          * check the return code.
 733          */
 734         vhost_vsock_stop(vsock, false);
 735         vhost_vsock_flush(vsock);
 736         vhost_dev_stop(&vsock->dev);
 737
 738         virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
 739
 740         vhost_dev_cleanup(&vsock->dev);
 741         kfree(vsock->dev.vqs);
 742         vhost_vsock_free(vsock);
 743         return 0;
 744 }
 745
 746 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
 747 {
 748         struct vhost_vsock *other;
 749
 750         /* Refuse reserved CIDs */
 751         if (guest_cid <= VMADDR_CID_HOST ||
 752             guest_cid == U32_MAX)
 753                 return -EINVAL;
 754
 755         /* 64-bit CIDs are not yet supported */
 756         if (guest_cid > U32_MAX)
 757                 return -EINVAL;
 758
 759         /* Refuse if CID is assigned to the guest->host transport (i.e. nested
 760          * VM), to make the loopback work.
 761          */
 762         if (vsock_find_cid(guest_cid))
 763                 return -EADDRINUSE;
 764
 765         /* Refuse if CID is already in use */
 766         mutex_lock(&vhost_vsock_mutex);
 767         other = vhost_vsock_get(guest_cid);
 768         if (other && other != vsock) {
 769                 mutex_unlock(&vhost_vsock_mutex);
 770                 return -EADDRINUSE;
 771         }
 772
 773         if (vsock->guest_cid)
 774                 hash_del_rcu(&vsock->hash);
 775
 776         vsock->guest_cid = guest_cid;
 777         hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
 778         mutex_unlock(&vhost_vsock_mutex);
 779
 780         return 0;
 781 }
 782
 783 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
 784 {
 785         struct vhost_virtqueue *vq;
 786         int i;
 787
 788         if (features & ~VHOST_VSOCK_FEATURES)
 789                 return -EOPNOTSUPP;
 790
 791         mutex_lock(&vsock->dev.mutex);
 792         if ((features & (1 << VHOST_F_LOG_ALL)) &&
 793             !vhost_log_access_ok(&vsock->dev)) {
 794                 goto err;
 795         }
 796
 797         if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
 798                 if (vhost_init_device_iotlb(&vsock->dev))
 799                         goto err;
 800         }
 801
 802         if (features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET))
 803                 vsock->seqpacket_allow = true;
 804
 805         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 806                 vq = &vsock->vqs[i];
 807                 mutex_lock(&vq->mutex);
 808                 vq->acked_features = features;
 809                 mutex_unlock(&vq->mutex);
 810         }
 811         mutex_unlock(&vsock->dev.mutex);
 812         return 0;
 813
 814 err:
 815         mutex_unlock(&vsock->dev.mutex);
 816         return -EFAULT;
 817 }
 818
 819 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
 820                                   unsigned long arg)
 821 {
 822         struct vhost_vsock *vsock = f->private_data;
 823         void __user *argp = (void __user *)arg;
 824         u64 guest_cid;
 825         u64 features;
 826         int start;
 827         int r;
 828
 829         switch (ioctl) {
 830         case VHOST_VSOCK_SET_GUEST_CID:
 831                 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
 832                         return -EFAULT;
 833                 return vhost_vsock_set_cid(vsock, guest_cid);
 834         case VHOST_VSOCK_SET_RUNNING:
 835                 if (copy_from_user(&start, argp, sizeof(start)))
 836                         return -EFAULT;
 837                 if (start)
 838                         return vhost_vsock_start(vsock);
 839                 else
 840                         return vhost_vsock_stop(vsock, true);
 841         case VHOST_GET_FEATURES:
 842                 features = VHOST_VSOCK_FEATURES;
 843                 if (copy_to_user(argp, &features, sizeof(features)))
 844                         return -EFAULT;
 845                 return 0;
 846         case VHOST_SET_FEATURES:
 847                 if (copy_from_user(&features, argp, sizeof(features)))
 848                         return -EFAULT;
 849                 return vhost_vsock_set_features(vsock, features);
 850         case VHOST_GET_BACKEND_FEATURES:
 851                 features = VHOST_VSOCK_BACKEND_FEATURES;
 852                 if (copy_to_user(argp, &features, sizeof(features)))
 853                         return -EFAULT;
 854                 return 0;
 855         case VHOST_SET_BACKEND_FEATURES:
 856                 if (copy_from_user(&features, argp, sizeof(features)))
 857                         return -EFAULT;
 858                 if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
 859                         return -EOPNOTSUPP;
 860                 vhost_set_backend_features(&vsock->dev, features);
 861                 return 0;
 862         default:
 863                 mutex_lock(&vsock->dev.mutex);
 864                 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
 865                 if (r == -ENOIOCTLCMD)
 866                         r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
 867                 else
 868                         vhost_vsock_flush(vsock);
 869                 mutex_unlock(&vsock->dev.mutex);
 870                 return r;
 871         }
 872 }
 873
 874 static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
 875 {
 876         struct file *file = iocb->ki_filp;
 877         struct vhost_vsock *vsock = file->private_data;
 878         struct vhost_dev *dev = &vsock->dev;
 879         int noblock = file->f_flags & O_NONBLOCK;
 880
 881         return vhost_chr_read_iter(dev, to, noblock);
 882 }
 883
 884 static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
 885                                         struct iov_iter *from)
 886 {
 887         struct file *file = iocb->ki_filp;
 888         struct vhost_vsock *vsock = file->private_data;
 889         struct vhost_dev *dev = &vsock->dev;
 890
 891         return vhost_chr_write_iter(dev, from);
 892 }
 893
 894 static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
 895 {
 896         struct vhost_vsock *vsock = file->private_data;
 897         struct vhost_dev *dev = &vsock->dev;
 898
 899         return vhost_chr_poll(file, dev, wait);
 900 }
 901
 902 static const struct file_operations vhost_vsock_fops = {
 903         .owner          = THIS_MODULE,
 904         .open           = vhost_vsock_dev_open,
 905         .release        = vhost_vsock_dev_release,
 906         .llseek         = noop_llseek,
 907         .unlocked_ioctl = vhost_vsock_dev_ioctl,
 908         .compat_ioctl   = compat_ptr_ioctl,
 909         .read_iter      = vhost_vsock_chr_read_iter,
 910         .write_iter     = vhost_vsock_chr_write_iter,
 911         .poll           = vhost_vsock_chr_poll,
 912 };
 913
 914 static struct miscdevice vhost_vsock_misc = {
 915         .minor = VHOST_VSOCK_MINOR,
 916         .name = "vhost-vsock",
 917         .fops = &vhost_vsock_fops,
 918 };
 919
 920 static int __init vhost_vsock_init(void)
 921 {
 922         int ret;
 923
 924         ret = vsock_core_register(&vhost_transport.transport,
 925                                   VSOCK_TRANSPORT_F_H2G);
 926         if (ret < 0)
 927                 return ret;
 928
 929         ret = misc_register(&vhost_vsock_misc);
 930         if (ret) {
 931                 vsock_core_unregister(&vhost_transport.transport);
 932                 return ret;
 933         }
 934
 935         return 0;
 936 };
 937
 938 static void __exit vhost_vsock_exit(void)
 939 {
 940         misc_deregister(&vhost_vsock_misc);
 941         vsock_core_unregister(&vhost_transport.transport);
 942 };
 943
 944 module_init(vhost_vsock_init);
 945 module_exit(vhost_vsock_exit);
 946 MODULE_LICENSE("GPL v2");
 947 MODULE_AUTHOR("Asias He");
 948 MODULE_DESCRIPTION("vhost transport for vsock ");
 949 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
 950 MODULE_ALIAS("devname:vhost-vsock");