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_work_queue(&vsock->dev, &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
 442                 .read_skb = virtio_transport_read_skb,
 443         },
 444
 445         .send_pkt = vhost_transport_send_pkt,
 446 };
 447
 448 static bool vhost_transport_seqpacket_allow(u32 remote_cid)
 449 {
 450         struct vhost_vsock *vsock;
 451         bool seqpacket_allow = false;
 452
 453         rcu_read_lock();
 454         vsock = vhost_vsock_get(remote_cid);
 455
 456         if (vsock)
 457                 seqpacket_allow = vsock->seqpacket_allow;
 458
 459         rcu_read_unlock();
 460
 461         return seqpacket_allow;
 462 }
 463
 464 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
 465 {
 466         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 467                                                   poll.work);
 468         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 469                                                  dev);
 470         int head, pkts = 0, total_len = 0;
 471         unsigned int out, in;
 472         struct sk_buff *skb;
 473         bool added = false;
 474
 475         mutex_lock(&vq->mutex);
 476
 477         if (!vhost_vq_get_backend(vq))
 478                 goto out;
 479
 480         if (!vq_meta_prefetch(vq))
 481                 goto out;
 482
 483         vhost_disable_notify(&vsock->dev, vq);
 484         do {
 485                 struct virtio_vsock_hdr *hdr;
 486
 487                 if (!vhost_vsock_more_replies(vsock)) {
 488                         /* Stop tx until the device processes already
 489                          * pending replies.  Leave tx virtqueue
 490                          * callbacks disabled.
 491                          */
 492                         goto no_more_replies;
 493                 }
 494
 495                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 496                                          &out, &in, NULL, NULL);
 497                 if (head < 0)
 498                         break;
 499
 500                 if (head == vq->num) {
 501                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 502                                 vhost_disable_notify(&vsock->dev, vq);
 503                                 continue;
 504                         }
 505                         break;
 506                 }
 507
 508                 skb = vhost_vsock_alloc_skb(vq, out, in);
 509                 if (!skb) {
 510                         vq_err(vq, "Faulted on pkt\n");
 511                         continue;
 512                 }
 513
 514                 total_len += sizeof(*hdr) + skb->len;
 515
 516                 /* Deliver to monitoring devices all received packets */
 517                 virtio_transport_deliver_tap_pkt(skb);
 518
 519                 hdr = virtio_vsock_hdr(skb);
 520
 521                 /* Only accept correctly addressed packets */
 522                 if (le64_to_cpu(hdr->src_cid) == vsock->guest_cid &&
 523                     le64_to_cpu(hdr->dst_cid) ==
 524                     vhost_transport_get_local_cid())
 525                         virtio_transport_recv_pkt(&vhost_transport, skb);
 526                 else
 527                         kfree_skb(skb);
 528
 529                 vhost_add_used(vq, head, 0);
 530                 added = true;
 531         } while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
 532
 533 no_more_replies:
 534         if (added)
 535                 vhost_signal(&vsock->dev, vq);
 536
 537 out:
 538         mutex_unlock(&vq->mutex);
 539 }
 540
 541 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
 542 {
 543         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 544                                                 poll.work);
 545         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 546                                                  dev);
 547
 548         vhost_transport_do_send_pkt(vsock, vq);
 549 }
 550
 551 static int vhost_vsock_start(struct vhost_vsock *vsock)
 552 {
 553         struct vhost_virtqueue *vq;
 554         size_t i;
 555         int ret;
 556
 557         mutex_lock(&vsock->dev.mutex);
 558
 559         ret = vhost_dev_check_owner(&vsock->dev);
 560         if (ret)
 561                 goto err;
 562
 563         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 564                 vq = &vsock->vqs[i];
 565
 566                 mutex_lock(&vq->mutex);
 567
 568                 if (!vhost_vq_access_ok(vq)) {
 569                         ret = -EFAULT;
 570                         goto err_vq;
 571                 }
 572
 573                 if (!vhost_vq_get_backend(vq)) {
 574                         vhost_vq_set_backend(vq, vsock);
 575                         ret = vhost_vq_init_access(vq);
 576                         if (ret)
 577                                 goto err_vq;
 578                 }
 579
 580                 mutex_unlock(&vq->mutex);
 581         }
 582
 583         /* Some packets may have been queued before the device was started,
 584          * let's kick the send worker to send them.
 585          */
 586         vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
 587
 588         mutex_unlock(&vsock->dev.mutex);
 589         return 0;
 590
 591 err_vq:
 592         vhost_vq_set_backend(vq, NULL);
 593         mutex_unlock(&vq->mutex);
 594
 595         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 596                 vq = &vsock->vqs[i];
 597
 598                 mutex_lock(&vq->mutex);
 599                 vhost_vq_set_backend(vq, NULL);
 600                 mutex_unlock(&vq->mutex);
 601         }
 602 err:
 603         mutex_unlock(&vsock->dev.mutex);
 604         return ret;
 605 }
 606
 607 static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner)
 608 {
 609         size_t i;
 610         int ret = 0;
 611
 612         mutex_lock(&vsock->dev.mutex);
 613
 614         if (check_owner) {
 615                 ret = vhost_dev_check_owner(&vsock->dev);
 616                 if (ret)
 617                         goto err;
 618         }
 619
 620         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 621                 struct vhost_virtqueue *vq = &vsock->vqs[i];
 622
 623                 mutex_lock(&vq->mutex);
 624                 vhost_vq_set_backend(vq, NULL);
 625                 mutex_unlock(&vq->mutex);
 626         }
 627
 628 err:
 629         mutex_unlock(&vsock->dev.mutex);
 630         return ret;
 631 }
 632
 633 static void vhost_vsock_free(struct vhost_vsock *vsock)
 634 {
 635         kvfree(vsock);
 636 }
 637
 638 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
 639 {
 640         struct vhost_virtqueue **vqs;
 641         struct vhost_vsock *vsock;
 642         int ret;
 643
 644         /* This struct is large and allocation could fail, fall back to vmalloc
 645          * if there is no other way.
 646          */
 647         vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
 648         if (!vsock)
 649                 return -ENOMEM;
 650
 651         vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
 652         if (!vqs) {
 653                 ret = -ENOMEM;
 654                 goto out;
 655         }
 656
 657         vsock->guest_cid = 0; /* no CID assigned yet */
 658
 659         atomic_set(&vsock->queued_replies, 0);
 660
 661         vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
 662         vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
 663         vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
 664         vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
 665
 666         vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
 667                        UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
 668                        VHOST_VSOCK_WEIGHT, true, NULL);
 669
 670         file->private_data = vsock;
 671         skb_queue_head_init(&vsock->send_pkt_queue);
 672         vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
 673         return 0;
 674
 675 out:
 676         vhost_vsock_free(vsock);
 677         return ret;
 678 }
 679
 680 static void vhost_vsock_flush(struct vhost_vsock *vsock)
 681 {
 682         vhost_dev_flush(&vsock->dev);
 683 }
 684
 685 static void vhost_vsock_reset_orphans(struct sock *sk)
 686 {
 687         struct vsock_sock *vsk = vsock_sk(sk);
 688
 689         /* vmci_transport.c doesn't take sk_lock here either.  At least we're
 690          * under vsock_table_lock so the sock cannot disappear while we're
 691          * executing.
 692          */
 693
 694         /* If the peer is still valid, no need to reset connection */
 695         if (vhost_vsock_get(vsk->remote_addr.svm_cid))
 696                 return;
 697
 698         /* If the close timeout is pending, let it expire.  This avoids races
 699          * with the timeout callback.
 700          */
 701         if (vsk->close_work_scheduled)
 702                 return;
 703
 704         sock_set_flag(sk, SOCK_DONE);
 705         vsk->peer_shutdown = SHUTDOWN_MASK;
 706         sk->sk_state = SS_UNCONNECTED;
 707         sk->sk_err = ECONNRESET;
 708         sk_error_report(sk);
 709 }
 710
 711 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
 712 {
 713         struct vhost_vsock *vsock = file->private_data;
 714
 715         mutex_lock(&vhost_vsock_mutex);
 716         if (vsock->guest_cid)
 717                 hash_del_rcu(&vsock->hash);
 718         mutex_unlock(&vhost_vsock_mutex);
 719
 720         /* Wait for other CPUs to finish using vsock */
 721         synchronize_rcu();
 722
 723         /* Iterating over all connections for all CIDs to find orphans is
 724          * inefficient.  Room for improvement here. */
 725         vsock_for_each_connected_socket(&vhost_transport.transport,
 726                                         vhost_vsock_reset_orphans);
 727
 728         /* Don't check the owner, because we are in the release path, so we
 729          * need to stop the vsock device in any case.
 730          * vhost_vsock_stop() can not fail in this case, so we don't need to
 731          * check the return code.
 732          */
 733         vhost_vsock_stop(vsock, false);
 734         vhost_vsock_flush(vsock);
 735         vhost_dev_stop(&vsock->dev);
 736
 737         virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
 738
 739         vhost_dev_cleanup(&vsock->dev);
 740         kfree(vsock->dev.vqs);
 741         vhost_vsock_free(vsock);
 742         return 0;
 743 }
 744
 745 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
 746 {
 747         struct vhost_vsock *other;
 748
 749         /* Refuse reserved CIDs */
 750         if (guest_cid <= VMADDR_CID_HOST ||
 751             guest_cid == U32_MAX)
 752                 return -EINVAL;
 753
 754         /* 64-bit CIDs are not yet supported */
 755         if (guest_cid > U32_MAX)
 756                 return -EINVAL;
 757
 758         /* Refuse if CID is assigned to the guest->host transport (i.e. nested
 759          * VM), to make the loopback work.
 760          */
 761         if (vsock_find_cid(guest_cid))
 762                 return -EADDRINUSE;
 763
 764         /* Refuse if CID is already in use */
 765         mutex_lock(&vhost_vsock_mutex);
 766         other = vhost_vsock_get(guest_cid);
 767         if (other && other != vsock) {
 768                 mutex_unlock(&vhost_vsock_mutex);
 769                 return -EADDRINUSE;
 770         }
 771
 772         if (vsock->guest_cid)
 773                 hash_del_rcu(&vsock->hash);
 774
 775         vsock->guest_cid = guest_cid;
 776         hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
 777         mutex_unlock(&vhost_vsock_mutex);
 778
 779         return 0;
 780 }
 781
 782 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
 783 {
 784         struct vhost_virtqueue *vq;
 785         int i;
 786
 787         if (features & ~VHOST_VSOCK_FEATURES)
 788                 return -EOPNOTSUPP;
 789
 790         mutex_lock(&vsock->dev.mutex);
 791         if ((features & (1 << VHOST_F_LOG_ALL)) &&
 792             !vhost_log_access_ok(&vsock->dev)) {
 793                 goto err;
 794         }
 795
 796         if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
 797                 if (vhost_init_device_iotlb(&vsock->dev))
 798                         goto err;
 799         }
 800
 801         if (features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET))
 802                 vsock->seqpacket_allow = true;
 803
 804         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 805                 vq = &vsock->vqs[i];
 806                 mutex_lock(&vq->mutex);
 807                 vq->acked_features = features;
 808                 mutex_unlock(&vq->mutex);
 809         }
 810         mutex_unlock(&vsock->dev.mutex);
 811         return 0;
 812
 813 err:
 814         mutex_unlock(&vsock->dev.mutex);
 815         return -EFAULT;
 816 }
 817
 818 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
 819                                   unsigned long arg)
 820 {
 821         struct vhost_vsock *vsock = f->private_data;
 822         void __user *argp = (void __user *)arg;
 823         u64 guest_cid;
 824         u64 features;
 825         int start;
 826         int r;
 827
 828         switch (ioctl) {
 829         case VHOST_VSOCK_SET_GUEST_CID:
 830                 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
 831                         return -EFAULT;
 832                 return vhost_vsock_set_cid(vsock, guest_cid);
 833         case VHOST_VSOCK_SET_RUNNING:
 834                 if (copy_from_user(&start, argp, sizeof(start)))
 835                         return -EFAULT;
 836                 if (start)
 837                         return vhost_vsock_start(vsock);
 838                 else
 839                         return vhost_vsock_stop(vsock, true);
 840         case VHOST_GET_FEATURES:
 841                 features = VHOST_VSOCK_FEATURES;
 842                 if (copy_to_user(argp, &features, sizeof(features)))
 843                         return -EFAULT;
 844                 return 0;
 845         case VHOST_SET_FEATURES:
 846                 if (copy_from_user(&features, argp, sizeof(features)))
 847                         return -EFAULT;
 848                 return vhost_vsock_set_features(vsock, features);
 849         case VHOST_GET_BACKEND_FEATURES:
 850                 features = VHOST_VSOCK_BACKEND_FEATURES;
 851                 if (copy_to_user(argp, &features, sizeof(features)))
 852                         return -EFAULT;
 853                 return 0;
 854         case VHOST_SET_BACKEND_FEATURES:
 855                 if (copy_from_user(&features, argp, sizeof(features)))
 856                         return -EFAULT;
 857                 if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
 858                         return -EOPNOTSUPP;
 859                 vhost_set_backend_features(&vsock->dev, features);
 860                 return 0;
 861         default:
 862                 mutex_lock(&vsock->dev.mutex);
 863                 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
 864                 if (r == -ENOIOCTLCMD)
 865                         r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
 866                 else
 867                         vhost_vsock_flush(vsock);
 868                 mutex_unlock(&vsock->dev.mutex);
 869                 return r;
 870         }
 871 }
 872
 873 static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
 874 {
 875         struct file *file = iocb->ki_filp;
 876         struct vhost_vsock *vsock = file->private_data;
 877         struct vhost_dev *dev = &vsock->dev;
 878         int noblock = file->f_flags & O_NONBLOCK;
 879
 880         return vhost_chr_read_iter(dev, to, noblock);
 881 }
 882
 883 static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
 884                                         struct iov_iter *from)
 885 {
 886         struct file *file = iocb->ki_filp;
 887         struct vhost_vsock *vsock = file->private_data;
 888         struct vhost_dev *dev = &vsock->dev;
 889
 890         return vhost_chr_write_iter(dev, from);
 891 }
 892
 893 static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
 894 {
 895         struct vhost_vsock *vsock = file->private_data;
 896         struct vhost_dev *dev = &vsock->dev;
 897
 898         return vhost_chr_poll(file, dev, wait);
 899 }
 900
 901 static const struct file_operations vhost_vsock_fops = {
 902         .owner          = THIS_MODULE,
 903         .open           = vhost_vsock_dev_open,
 904         .release        = vhost_vsock_dev_release,
 905         .llseek         = noop_llseek,
 906         .unlocked_ioctl = vhost_vsock_dev_ioctl,
 907         .compat_ioctl   = compat_ptr_ioctl,
 908         .read_iter      = vhost_vsock_chr_read_iter,
 909         .write_iter     = vhost_vsock_chr_write_iter,
 910         .poll           = vhost_vsock_chr_poll,
 911 };
 912
 913 static struct miscdevice vhost_vsock_misc = {
 914         .minor = VHOST_VSOCK_MINOR,
 915         .name = "vhost-vsock",
 916         .fops = &vhost_vsock_fops,
 917 };
 918
 919 static int __init vhost_vsock_init(void)
 920 {
 921         int ret;
 922
 923         ret = vsock_core_register(&vhost_transport.transport,
 924                                   VSOCK_TRANSPORT_F_H2G);
 925         if (ret < 0)
 926                 return ret;
 927
 928         ret = misc_register(&vhost_vsock_misc);
 929         if (ret) {
 930                 vsock_core_unregister(&vhost_transport.transport);
 931                 return ret;
 932         }
 933
 934         return 0;
 935 };
 936
 937 static void __exit vhost_vsock_exit(void)
 938 {
 939         misc_deregister(&vhost_vsock_misc);
 940         vsock_core_unregister(&vhost_transport.transport);
 941 };
 942
 943 module_init(vhost_vsock_init);
 944 module_exit(vhost_vsock_exit);
 945 MODULE_LICENSE("GPL v2");
 946 MODULE_AUTHOR("Asias He");
 947 MODULE_DESCRIPTION("vhost transport for vsock ");
 948 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
 949 MODULE_ALIAS("devname:vhost-vsock");