net/vmw_vsock/virtio_transport_common.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * common code for virtio 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/spinlock.h>
  10 #include <linux/module.h>
  11 #include <linux/sched/signal.h>
  12 #include <linux/ctype.h>
  13 #include <linux/list.h>
  14 #include <linux/virtio_vsock.h>
  15 #include <uapi/linux/vsockmon.h>
  16
  17 #include <net/sock.h>
  18 #include <net/af_vsock.h>
  19
  20 #define CREATE_TRACE_POINTS
  21 #include <trace/events/vsock_virtio_transport_common.h>
  22
  23 /* How long to wait for graceful shutdown of a connection */
  24 #define VSOCK_CLOSE_TIMEOUT (8 * HZ)
  25
  26 /* Threshold for detecting small packets to copy */
  27 #define GOOD_COPY_LEN  128
  28
  29 static const struct virtio_transport *
  30 virtio_transport_get_ops(struct vsock_sock *vsk)
  31 {
  32         const struct vsock_transport *t = vsock_core_get_transport(vsk);
  33
  34         if (WARN_ON(!t))
  35                 return NULL;
  36
  37         return container_of(t, struct virtio_transport, transport);
  38 }
  39
  40 static bool virtio_transport_can_zcopy(const struct virtio_transport *t_ops,
  41                                        struct virtio_vsock_pkt_info *info,
  42                                        size_t pkt_len)
  43 {
  44         struct iov_iter *iov_iter;
  45
  46         if (!info->msg)
  47                 return false;
  48
  49         iov_iter = &info->msg->msg_iter;
  50
  51         if (iov_iter->iov_offset)
  52                 return false;
  53
  54         /* We can't send whole iov. */
  55         if (iov_iter->count > pkt_len)
  56                 return false;
  57
  58         /* Check that transport can send data in zerocopy mode. */
  59         t_ops = virtio_transport_get_ops(info->vsk);
  60
  61         if (t_ops->can_msgzerocopy) {
  62                 int pages_to_send = iov_iter_npages(iov_iter, MAX_SKB_FRAGS);
  63
  64                 /* +1 is for packet header. */
  65                 return t_ops->can_msgzerocopy(pages_to_send + 1);
  66         }
  67
  68         return true;
  69 }
  70
  71 static int virtio_transport_init_zcopy_skb(struct vsock_sock *vsk,
  72                                            struct sk_buff *skb,
  73                                            struct msghdr *msg,
  74                                            bool zerocopy)
  75 {
  76         struct ubuf_info *uarg;
  77
  78         if (msg->msg_ubuf) {
  79                 uarg = msg->msg_ubuf;
  80                 net_zcopy_get(uarg);
  81         } else {
  82                 struct iov_iter *iter = &msg->msg_iter;
  83                 struct ubuf_info_msgzc *uarg_zc;
  84
  85                 uarg = msg_zerocopy_realloc(sk_vsock(vsk),
  86                                             iter->count,
  87                                             NULL);
  88                 if (!uarg)
  89                         return -1;
  90
  91                 uarg_zc = uarg_to_msgzc(uarg);
  92                 uarg_zc->zerocopy = zerocopy ? 1 : 0;
  93         }
  94
  95         skb_zcopy_init(skb, uarg);
  96
  97         return 0;
  98 }
  99
 100 static int virtio_transport_fill_skb(struct sk_buff *skb,
 101                                      struct virtio_vsock_pkt_info *info,
 102                                      size_t len,
 103                                      bool zcopy)
 104 {
 105         if (zcopy)
 106                 return __zerocopy_sg_from_iter(info->msg, NULL, skb,
 107                                                &info->msg->msg_iter,
 108                                                len);
 109
 110         return memcpy_from_msg(skb_put(skb, len), info->msg, len);
 111 }
 112
 113 static void virtio_transport_init_hdr(struct sk_buff *skb,
 114                                       struct virtio_vsock_pkt_info *info,
 115                                       size_t payload_len,
 116                                       u32 src_cid,
 117                                       u32 src_port,
 118                                       u32 dst_cid,
 119                                       u32 dst_port)
 120 {
 121         struct virtio_vsock_hdr *hdr;
 122
 123         hdr = virtio_vsock_hdr(skb);
 124         hdr->type       = cpu_to_le16(info->type);
 125         hdr->op         = cpu_to_le16(info->op);
 126         hdr->src_cid    = cpu_to_le64(src_cid);
 127         hdr->dst_cid    = cpu_to_le64(dst_cid);
 128         hdr->src_port   = cpu_to_le32(src_port);
 129         hdr->dst_port   = cpu_to_le32(dst_port);
 130         hdr->flags      = cpu_to_le32(info->flags);
 131         hdr->len        = cpu_to_le32(payload_len);
 132         hdr->buf_alloc  = cpu_to_le32(0);
 133         hdr->fwd_cnt    = cpu_to_le32(0);
 134 }
 135
 136 static void virtio_transport_copy_nonlinear_skb(const struct sk_buff *skb,
 137                                                 void *dst,
 138                                                 size_t len)
 139 {
 140         struct iov_iter iov_iter = { 0 };
 141         struct kvec kvec;
 142         size_t to_copy;
 143
 144         kvec.iov_base = dst;
 145         kvec.iov_len = len;
 146
 147         iov_iter.iter_type = ITER_KVEC;
 148         iov_iter.kvec = &kvec;
 149         iov_iter.nr_segs = 1;
 150
 151         to_copy = min_t(size_t, len, skb->len);
 152
 153         skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset,
 154                                &iov_iter, to_copy);
 155 }
 156
 157 /* Packet capture */
 158 static struct sk_buff *virtio_transport_build_skb(void *opaque)
 159 {
 160         struct virtio_vsock_hdr *pkt_hdr;
 161         struct sk_buff *pkt = opaque;
 162         struct af_vsockmon_hdr *hdr;
 163         struct sk_buff *skb;
 164         size_t payload_len;
 165
 166         /* A packet could be split to fit the RX buffer, so we can retrieve
 167          * the payload length from the header and the buffer pointer taking
 168          * care of the offset in the original packet.
 169          */
 170         pkt_hdr = virtio_vsock_hdr(pkt);
 171         payload_len = pkt->len;
 172
 173         skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len,
 174                         GFP_ATOMIC);
 175         if (!skb)
 176                 return NULL;
 177
 178         hdr = skb_put(skb, sizeof(*hdr));
 179
 180         /* pkt->hdr is little-endian so no need to byteswap here */
 181         hdr->src_cid = pkt_hdr->src_cid;
 182         hdr->src_port = pkt_hdr->src_port;
 183         hdr->dst_cid = pkt_hdr->dst_cid;
 184         hdr->dst_port = pkt_hdr->dst_port;
 185
 186         hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
 187         hdr->len = cpu_to_le16(sizeof(*pkt_hdr));
 188         memset(hdr->reserved, 0, sizeof(hdr->reserved));
 189
 190         switch (le16_to_cpu(pkt_hdr->op)) {
 191         case VIRTIO_VSOCK_OP_REQUEST:
 192         case VIRTIO_VSOCK_OP_RESPONSE:
 193                 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
 194                 break;
 195         case VIRTIO_VSOCK_OP_RST:
 196         case VIRTIO_VSOCK_OP_SHUTDOWN:
 197                 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
 198                 break;
 199         case VIRTIO_VSOCK_OP_RW:
 200                 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
 201                 break;
 202         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
 203         case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
 204                 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
 205                 break;
 206         default:
 207                 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
 208                 break;
 209         }
 210
 211         skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr));
 212
 213         if (payload_len) {
 214                 if (skb_is_nonlinear(pkt)) {
 215                         void *data = skb_put(skb, payload_len);
 216
 217                         virtio_transport_copy_nonlinear_skb(pkt, data, payload_len);
 218                 } else {
 219                         skb_put_data(skb, pkt->data, payload_len);
 220                 }
 221         }
 222
 223         return skb;
 224 }
 225
 226 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb)
 227 {
 228         if (virtio_vsock_skb_tap_delivered(skb))
 229                 return;
 230
 231         vsock_deliver_tap(virtio_transport_build_skb, skb);
 232         virtio_vsock_skb_set_tap_delivered(skb);
 233 }
 234 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
 235
 236 static u16 virtio_transport_get_type(struct sock *sk)
 237 {
 238         if (sk->sk_type == SOCK_STREAM)
 239                 return VIRTIO_VSOCK_TYPE_STREAM;
 240         else
 241                 return VIRTIO_VSOCK_TYPE_SEQPACKET;
 242 }
 243
 244 /* Returns new sk_buff on success, otherwise returns NULL. */
 245 static struct sk_buff *virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info,
 246                                                   size_t payload_len,
 247                                                   bool zcopy,
 248                                                   u32 src_cid,
 249                                                   u32 src_port,
 250                                                   u32 dst_cid,
 251                                                   u32 dst_port)
 252 {
 253         struct vsock_sock *vsk;
 254         struct sk_buff *skb;
 255         size_t skb_len;
 256
 257         skb_len = VIRTIO_VSOCK_SKB_HEADROOM;
 258
 259         if (!zcopy)
 260                 skb_len += payload_len;
 261
 262         skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL);
 263         if (!skb)
 264                 return NULL;
 265
 266         virtio_transport_init_hdr(skb, info, payload_len, src_cid, src_port,
 267                                   dst_cid, dst_port);
 268
 269         vsk = info->vsk;
 270
 271         /* If 'vsk' != NULL then payload is always present, so we
 272          * will never call '__zerocopy_sg_from_iter()' below without
 273          * setting skb owner in 'skb_set_owner_w()'. The only case
 274          * when 'vsk' == NULL is VIRTIO_VSOCK_OP_RST control message
 275          * without payload.
 276          */
 277         WARN_ON_ONCE(!(vsk && (info->msg && payload_len)) && zcopy);
 278
 279         /* Set owner here, because '__zerocopy_sg_from_iter()' uses
 280          * owner of skb without check to update 'sk_wmem_alloc'.
 281          */
 282         if (vsk)
 283                 skb_set_owner_w(skb, sk_vsock(vsk));
 284
 285         if (info->msg && payload_len > 0) {
 286                 int err;
 287
 288                 err = virtio_transport_fill_skb(skb, info, payload_len, zcopy);
 289                 if (err)
 290                         goto out;
 291
 292                 if (msg_data_left(info->msg) == 0 &&
 293                     info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) {
 294                         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
 295
 296                         hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
 297
 298                         if (info->msg->msg_flags & MSG_EOR)
 299                                 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
 300                 }
 301         }
 302
 303         if (info->reply)
 304                 virtio_vsock_skb_set_reply(skb);
 305
 306         trace_virtio_transport_alloc_pkt(src_cid, src_port,
 307                                          dst_cid, dst_port,
 308                                          payload_len,
 309                                          info->type,
 310                                          info->op,
 311                                          info->flags,
 312                                          zcopy);
 313
 314         return skb;
 315 out:
 316         kfree_skb(skb);
 317         return NULL;
 318 }
 319
 320 /* This function can only be used on connecting/connected sockets,
 321  * since a socket assigned to a transport is required.
 322  *
 323  * Do not use on listener sockets!
 324  */
 325 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
 326                                           struct virtio_vsock_pkt_info *info)
 327 {
 328         u32 max_skb_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
 329         u32 src_cid, src_port, dst_cid, dst_port;
 330         const struct virtio_transport *t_ops;
 331         struct virtio_vsock_sock *vvs;
 332         u32 pkt_len = info->pkt_len;
 333         bool can_zcopy = false;
 334         u32 rest_len;
 335         int ret;
 336
 337         info->type = virtio_transport_get_type(sk_vsock(vsk));
 338
 339         t_ops = virtio_transport_get_ops(vsk);
 340         if (unlikely(!t_ops))
 341                 return -EFAULT;
 342
 343         src_cid = t_ops->transport.get_local_cid();
 344         src_port = vsk->local_addr.svm_port;
 345         if (!info->remote_cid) {
 346                 dst_cid = vsk->remote_addr.svm_cid;
 347                 dst_port = vsk->remote_addr.svm_port;
 348         } else {
 349                 dst_cid = info->remote_cid;
 350                 dst_port = info->remote_port;
 351         }
 352
 353         vvs = vsk->trans;
 354
 355         /* virtio_transport_get_credit might return less than pkt_len credit */
 356         pkt_len = virtio_transport_get_credit(vvs, pkt_len);
 357
 358         /* Do not send zero length OP_RW pkt */
 359         if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
 360                 return pkt_len;
 361
 362         if (info->msg) {
 363                 /* If zerocopy is not enabled by 'setsockopt()', we behave as
 364                  * there is no MSG_ZEROCOPY flag set.
 365                  */
 366                 if (!sock_flag(sk_vsock(vsk), SOCK_ZEROCOPY))
 367                         info->msg->msg_flags &= ~MSG_ZEROCOPY;
 368
 369                 if (info->msg->msg_flags & MSG_ZEROCOPY)
 370                         can_zcopy = virtio_transport_can_zcopy(t_ops, info, pkt_len);
 371
 372                 if (can_zcopy)
 373                         max_skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE,
 374                                             (MAX_SKB_FRAGS * PAGE_SIZE));
 375         }
 376
 377         rest_len = pkt_len;
 378
 379         do {
 380                 struct sk_buff *skb;
 381                 size_t skb_len;
 382
 383                 skb_len = min(max_skb_len, rest_len);
 384
 385                 skb = virtio_transport_alloc_skb(info, skb_len, can_zcopy,
 386                                                  src_cid, src_port,
 387                                                  dst_cid, dst_port);
 388                 if (!skb) {
 389                         ret = -ENOMEM;
 390                         break;
 391                 }
 392
 393                 /* We process buffer part by part, allocating skb on
 394                  * each iteration. If this is last skb for this buffer
 395                  * and MSG_ZEROCOPY mode is in use - we must allocate
 396                  * completion for the current syscall.
 397                  */
 398                 if (info->msg && info->msg->msg_flags & MSG_ZEROCOPY &&
 399                     skb_len == rest_len && info->op == VIRTIO_VSOCK_OP_RW) {
 400                         if (virtio_transport_init_zcopy_skb(vsk, skb,
 401                                                             info->msg,
 402                                                             can_zcopy)) {
 403                                 ret = -ENOMEM;
 404                                 break;
 405                         }
 406                 }
 407
 408                 virtio_transport_inc_tx_pkt(vvs, skb);
 409
 410                 ret = t_ops->send_pkt(skb);
 411                 if (ret < 0)
 412                         break;
 413
 414                 /* Both virtio and vhost 'send_pkt()' returns 'skb_len',
 415                  * but for reliability use 'ret' instead of 'skb_len'.
 416                  * Also if partial send happens (e.g. 'ret' != 'skb_len')
 417                  * somehow, we break this loop, but account such returned
 418                  * value in 'virtio_transport_put_credit()'.
 419                  */
 420                 rest_len -= ret;
 421
 422                 if (WARN_ONCE(ret != skb_len,
 423                               "'send_pkt()' returns %i, but %zu expected\n",
 424                               ret, skb_len))
 425                         break;
 426         } while (rest_len);
 427
 428         virtio_transport_put_credit(vvs, rest_len);
 429
 430         /* Return number of bytes, if any data has been sent. */
 431         if (rest_len != pkt_len)
 432                 ret = pkt_len - rest_len;
 433
 434         return ret;
 435 }
 436
 437 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
 438                                         u32 len)
 439 {
 440         if (vvs->rx_bytes + len > vvs->buf_alloc)
 441                 return false;
 442
 443         vvs->rx_bytes += len;
 444         return true;
 445 }
 446
 447 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
 448                                         u32 len)
 449 {
 450         vvs->rx_bytes -= len;
 451         vvs->fwd_cnt += len;
 452 }
 453
 454 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb)
 455 {
 456         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
 457
 458         spin_lock_bh(&vvs->rx_lock);
 459         vvs->last_fwd_cnt = vvs->fwd_cnt;
 460         hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
 461         hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc);
 462         spin_unlock_bh(&vvs->rx_lock);
 463 }
 464 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
 465
 466 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
 467 {
 468         u32 ret;
 469
 470         if (!credit)
 471                 return 0;
 472
 473         spin_lock_bh(&vvs->tx_lock);
 474         ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 475         if (ret > credit)
 476                 ret = credit;
 477         vvs->tx_cnt += ret;
 478         spin_unlock_bh(&vvs->tx_lock);
 479
 480         return ret;
 481 }
 482 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
 483
 484 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
 485 {
 486         if (!credit)
 487                 return;
 488
 489         spin_lock_bh(&vvs->tx_lock);
 490         vvs->tx_cnt -= credit;
 491         spin_unlock_bh(&vvs->tx_lock);
 492 }
 493 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
 494
 495 static int virtio_transport_send_credit_update(struct vsock_sock *vsk)
 496 {
 497         struct virtio_vsock_pkt_info info = {
 498                 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
 499                 .vsk = vsk,
 500         };
 501
 502         return virtio_transport_send_pkt_info(vsk, &info);
 503 }
 504
 505 static ssize_t
 506 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
 507                                 struct msghdr *msg,
 508                                 size_t len)
 509 {
 510         struct virtio_vsock_sock *vvs = vsk->trans;
 511         struct sk_buff *skb;
 512         size_t total = 0;
 513         int err;
 514
 515         spin_lock_bh(&vvs->rx_lock);
 516
 517         skb_queue_walk(&vvs->rx_queue, skb) {
 518                 size_t bytes;
 519
 520                 bytes = len - total;
 521                 if (bytes > skb->len)
 522                         bytes = skb->len;
 523
 524                 spin_unlock_bh(&vvs->rx_lock);
 525
 526                 /* sk_lock is held by caller so no one else can dequeue.
 527                  * Unlock rx_lock since skb_copy_datagram_iter() may sleep.
 528                  */
 529                 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset,
 530                                              &msg->msg_iter, bytes);
 531                 if (err)
 532                         goto out;
 533
 534                 total += bytes;
 535
 536                 spin_lock_bh(&vvs->rx_lock);
 537
 538                 if (total == len)
 539                         break;
 540         }
 541
 542         spin_unlock_bh(&vvs->rx_lock);
 543
 544         return total;
 545
 546 out:
 547         if (total)
 548                 err = total;
 549         return err;
 550 }
 551
 552 static ssize_t
 553 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
 554                                    struct msghdr *msg,
 555                                    size_t len)
 556 {
 557         struct virtio_vsock_sock *vvs = vsk->trans;
 558         size_t bytes, total = 0;
 559         struct sk_buff *skb;
 560         int err = -EFAULT;
 561         u32 free_space;
 562
 563         spin_lock_bh(&vvs->rx_lock);
 564
 565         if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes,
 566                       "rx_queue is empty, but rx_bytes is non-zero\n")) {
 567                 spin_unlock_bh(&vvs->rx_lock);
 568                 return err;
 569         }
 570
 571         while (total < len && !skb_queue_empty(&vvs->rx_queue)) {
 572                 skb = skb_peek(&vvs->rx_queue);
 573
 574                 bytes = min_t(size_t, len - total,
 575                               skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset);
 576
 577                 /* sk_lock is held by caller so no one else can dequeue.
 578                  * Unlock rx_lock since skb_copy_datagram_iter() may sleep.
 579                  */
 580                 spin_unlock_bh(&vvs->rx_lock);
 581
 582                 err = skb_copy_datagram_iter(skb,
 583                                              VIRTIO_VSOCK_SKB_CB(skb)->offset,
 584                                              &msg->msg_iter, bytes);
 585                 if (err)
 586                         goto out;
 587
 588                 spin_lock_bh(&vvs->rx_lock);
 589
 590                 total += bytes;
 591
 592                 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes;
 593
 594                 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) {
 595                         u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
 596
 597                         virtio_transport_dec_rx_pkt(vvs, pkt_len);
 598                         __skb_unlink(skb, &vvs->rx_queue);
 599                         consume_skb(skb);
 600                 }
 601         }
 602
 603         free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
 604
 605         spin_unlock_bh(&vvs->rx_lock);
 606
 607         /* To reduce the number of credit update messages,
 608          * don't update credits as long as lots of space is available.
 609          * Note: the limit chosen here is arbitrary. Setting the limit
 610          * too high causes extra messages. Too low causes transmitter
 611          * stalls. As stalls are in theory more expensive than extra
 612          * messages, we set the limit to a high value. TODO: experiment
 613          * with different values.
 614          */
 615         if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
 616                 virtio_transport_send_credit_update(vsk);
 617
 618         return total;
 619
 620 out:
 621         if (total)
 622                 err = total;
 623         return err;
 624 }
 625
 626 static ssize_t
 627 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk,
 628                                    struct msghdr *msg)
 629 {
 630         struct virtio_vsock_sock *vvs = vsk->trans;
 631         struct sk_buff *skb;
 632         size_t total, len;
 633
 634         spin_lock_bh(&vvs->rx_lock);
 635
 636         if (!vvs->msg_count) {
 637                 spin_unlock_bh(&vvs->rx_lock);
 638                 return 0;
 639         }
 640
 641         total = 0;
 642         len = msg_data_left(msg);
 643
 644         skb_queue_walk(&vvs->rx_queue, skb) {
 645                 struct virtio_vsock_hdr *hdr;
 646
 647                 if (total < len) {
 648                         size_t bytes;
 649                         int err;
 650
 651                         bytes = len - total;
 652                         if (bytes > skb->len)
 653                                 bytes = skb->len;
 654
 655                         spin_unlock_bh(&vvs->rx_lock);
 656
 657                         /* sk_lock is held by caller so no one else can dequeue.
 658                          * Unlock rx_lock since skb_copy_datagram_iter() may sleep.
 659                          */
 660                         err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset,
 661                                                      &msg->msg_iter, bytes);
 662                         if (err)
 663                                 return err;
 664
 665                         spin_lock_bh(&vvs->rx_lock);
 666                 }
 667
 668                 total += skb->len;
 669                 hdr = virtio_vsock_hdr(skb);
 670
 671                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
 672                         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
 673                                 msg->msg_flags |= MSG_EOR;
 674
 675                         break;
 676                 }
 677         }
 678
 679         spin_unlock_bh(&vvs->rx_lock);
 680
 681         return total;
 682 }
 683
 684 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
 685                                                  struct msghdr *msg,
 686                                                  int flags)
 687 {
 688         struct virtio_vsock_sock *vvs = vsk->trans;
 689         int dequeued_len = 0;
 690         size_t user_buf_len = msg_data_left(msg);
 691         bool msg_ready = false;
 692         struct sk_buff *skb;
 693
 694         spin_lock_bh(&vvs->rx_lock);
 695
 696         if (vvs->msg_count == 0) {
 697                 spin_unlock_bh(&vvs->rx_lock);
 698                 return 0;
 699         }
 700
 701         while (!msg_ready) {
 702                 struct virtio_vsock_hdr *hdr;
 703                 size_t pkt_len;
 704
 705                 skb = __skb_dequeue(&vvs->rx_queue);
 706                 if (!skb)
 707                         break;
 708                 hdr = virtio_vsock_hdr(skb);
 709                 pkt_len = (size_t)le32_to_cpu(hdr->len);
 710
 711                 if (dequeued_len >= 0) {
 712                         size_t bytes_to_copy;
 713
 714                         bytes_to_copy = min(user_buf_len, pkt_len);
 715
 716                         if (bytes_to_copy) {
 717                                 int err;
 718
 719                                 /* sk_lock is held by caller so no one else can dequeue.
 720                                  * Unlock rx_lock since skb_copy_datagram_iter() may sleep.
 721                                  */
 722                                 spin_unlock_bh(&vvs->rx_lock);
 723
 724                                 err = skb_copy_datagram_iter(skb, 0,
 725                                                              &msg->msg_iter,
 726                                                              bytes_to_copy);
 727                                 if (err) {
 728                                         /* Copy of message failed. Rest of
 729                                          * fragments will be freed without copy.
 730                                          */
 731                                         dequeued_len = err;
 732                                 } else {
 733                                         user_buf_len -= bytes_to_copy;
 734                                 }
 735
 736                                 spin_lock_bh(&vvs->rx_lock);
 737                         }
 738
 739                         if (dequeued_len >= 0)
 740                                 dequeued_len += pkt_len;
 741                 }
 742
 743                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
 744                         msg_ready = true;
 745                         vvs->msg_count--;
 746
 747                         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
 748                                 msg->msg_flags |= MSG_EOR;
 749                 }
 750
 751                 virtio_transport_dec_rx_pkt(vvs, pkt_len);
 752                 kfree_skb(skb);
 753         }
 754
 755         spin_unlock_bh(&vvs->rx_lock);
 756
 757         virtio_transport_send_credit_update(vsk);
 758
 759         return dequeued_len;
 760 }
 761
 762 ssize_t
 763 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
 764                                 struct msghdr *msg,
 765                                 size_t len, int flags)
 766 {
 767         if (flags & MSG_PEEK)
 768                 return virtio_transport_stream_do_peek(vsk, msg, len);
 769         else
 770                 return virtio_transport_stream_do_dequeue(vsk, msg, len);
 771 }
 772 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
 773
 774 ssize_t
 775 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk,
 776                                    struct msghdr *msg,
 777                                    int flags)
 778 {
 779         if (flags & MSG_PEEK)
 780                 return virtio_transport_seqpacket_do_peek(vsk, msg);
 781         else
 782                 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags);
 783 }
 784 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue);
 785
 786 int
 787 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk,
 788                                    struct msghdr *msg,
 789                                    size_t len)
 790 {
 791         struct virtio_vsock_sock *vvs = vsk->trans;
 792
 793         spin_lock_bh(&vvs->tx_lock);
 794
 795         if (len > vvs->peer_buf_alloc) {
 796                 spin_unlock_bh(&vvs->tx_lock);
 797                 return -EMSGSIZE;
 798         }
 799
 800         spin_unlock_bh(&vvs->tx_lock);
 801
 802         return virtio_transport_stream_enqueue(vsk, msg, len);
 803 }
 804 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue);
 805
 806 int
 807 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
 808                                struct msghdr *msg,
 809                                size_t len, int flags)
 810 {
 811         return -EOPNOTSUPP;
 812 }
 813 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
 814
 815 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
 816 {
 817         struct virtio_vsock_sock *vvs = vsk->trans;
 818         s64 bytes;
 819
 820         spin_lock_bh(&vvs->rx_lock);
 821         bytes = vvs->rx_bytes;
 822         spin_unlock_bh(&vvs->rx_lock);
 823
 824         return bytes;
 825 }
 826 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
 827
 828 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk)
 829 {
 830         struct virtio_vsock_sock *vvs = vsk->trans;
 831         u32 msg_count;
 832
 833         spin_lock_bh(&vvs->rx_lock);
 834         msg_count = vvs->msg_count;
 835         spin_unlock_bh(&vvs->rx_lock);
 836
 837         return msg_count;
 838 }
 839 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data);
 840
 841 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
 842 {
 843         struct virtio_vsock_sock *vvs = vsk->trans;
 844         s64 bytes;
 845
 846         bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 847         if (bytes < 0)
 848                 bytes = 0;
 849
 850         return bytes;
 851 }
 852
 853 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
 854 {
 855         struct virtio_vsock_sock *vvs = vsk->trans;
 856         s64 bytes;
 857
 858         spin_lock_bh(&vvs->tx_lock);
 859         bytes = virtio_transport_has_space(vsk);
 860         spin_unlock_bh(&vvs->tx_lock);
 861
 862         return bytes;
 863 }
 864 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
 865
 866 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
 867                                     struct vsock_sock *psk)
 868 {
 869         struct virtio_vsock_sock *vvs;
 870
 871         vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
 872         if (!vvs)
 873                 return -ENOMEM;
 874
 875         vsk->trans = vvs;
 876         vvs->vsk = vsk;
 877         if (psk && psk->trans) {
 878                 struct virtio_vsock_sock *ptrans = psk->trans;
 879
 880                 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
 881         }
 882
 883         if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
 884                 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
 885
 886         vvs->buf_alloc = vsk->buffer_size;
 887
 888         spin_lock_init(&vvs->rx_lock);
 889         spin_lock_init(&vvs->tx_lock);
 890         skb_queue_head_init(&vvs->rx_queue);
 891
 892         return 0;
 893 }
 894 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
 895
 896 /* sk_lock held by the caller */
 897 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
 898 {
 899         struct virtio_vsock_sock *vvs = vsk->trans;
 900
 901         if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
 902                 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
 903
 904         vvs->buf_alloc = *val;
 905
 906         virtio_transport_send_credit_update(vsk);
 907 }
 908 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
 909
 910 int
 911 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
 912                                 size_t target,
 913                                 bool *data_ready_now)
 914 {
 915         *data_ready_now = vsock_stream_has_data(vsk) >= target;
 916
 917         return 0;
 918 }
 919 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
 920
 921 int
 922 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
 923                                  size_t target,
 924                                  bool *space_avail_now)
 925 {
 926         s64 free_space;
 927
 928         free_space = vsock_stream_has_space(vsk);
 929         if (free_space > 0)
 930                 *space_avail_now = true;
 931         else if (free_space == 0)
 932                 *space_avail_now = false;
 933
 934         return 0;
 935 }
 936 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
 937
 938 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
 939         size_t target, struct vsock_transport_recv_notify_data *data)
 940 {
 941         return 0;
 942 }
 943 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
 944
 945 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
 946         size_t target, struct vsock_transport_recv_notify_data *data)
 947 {
 948         return 0;
 949 }
 950 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
 951
 952 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
 953         size_t target, struct vsock_transport_recv_notify_data *data)
 954 {
 955         return 0;
 956 }
 957 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
 958
 959 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
 960         size_t target, ssize_t copied, bool data_read,
 961         struct vsock_transport_recv_notify_data *data)
 962 {
 963         return 0;
 964 }
 965 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
 966
 967 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
 968         struct vsock_transport_send_notify_data *data)
 969 {
 970         return 0;
 971 }
 972 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
 973
 974 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
 975         struct vsock_transport_send_notify_data *data)
 976 {
 977         return 0;
 978 }
 979 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
 980
 981 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
 982         struct vsock_transport_send_notify_data *data)
 983 {
 984         return 0;
 985 }
 986 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
 987
 988 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
 989         ssize_t written, struct vsock_transport_send_notify_data *data)
 990 {
 991         return 0;
 992 }
 993 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
 994
 995 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
 996 {
 997         return vsk->buffer_size;
 998 }
 999 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
1000
1001 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
1002 {
1003         return true;
1004 }
1005 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
1006
1007 bool virtio_transport_stream_allow(u32 cid, u32 port)
1008 {
1009         return true;
1010 }
1011 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
1012
1013 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
1014                                 struct sockaddr_vm *addr)
1015 {
1016         return -EOPNOTSUPP;
1017 }
1018 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
1019
1020 bool virtio_transport_dgram_allow(u32 cid, u32 port)
1021 {
1022         return false;
1023 }
1024 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
1025
1026 int virtio_transport_connect(struct vsock_sock *vsk)
1027 {
1028         struct virtio_vsock_pkt_info info = {
1029                 .op = VIRTIO_VSOCK_OP_REQUEST,
1030                 .vsk = vsk,
1031         };
1032
1033         return virtio_transport_send_pkt_info(vsk, &info);
1034 }
1035 EXPORT_SYMBOL_GPL(virtio_transport_connect);
1036
1037 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
1038 {
1039         struct virtio_vsock_pkt_info info = {
1040                 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
1041                 .flags = (mode & RCV_SHUTDOWN ?
1042                           VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
1043                          (mode & SEND_SHUTDOWN ?
1044                           VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
1045                 .vsk = vsk,
1046         };
1047
1048         return virtio_transport_send_pkt_info(vsk, &info);
1049 }
1050 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
1051
1052 int
1053 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
1054                                struct sockaddr_vm *remote_addr,
1055                                struct msghdr *msg,
1056                                size_t dgram_len)
1057 {
1058         return -EOPNOTSUPP;
1059 }
1060 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
1061
1062 ssize_t
1063 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
1064                                 struct msghdr *msg,
1065                                 size_t len)
1066 {
1067         struct virtio_vsock_pkt_info info = {
1068                 .op = VIRTIO_VSOCK_OP_RW,
1069                 .msg = msg,
1070                 .pkt_len = len,
1071                 .vsk = vsk,
1072         };
1073
1074         return virtio_transport_send_pkt_info(vsk, &info);
1075 }
1076 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
1077
1078 void virtio_transport_destruct(struct vsock_sock *vsk)
1079 {
1080         struct virtio_vsock_sock *vvs = vsk->trans;
1081
1082         kfree(vvs);
1083 }
1084 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
1085
1086 static int virtio_transport_reset(struct vsock_sock *vsk,
1087                                   struct sk_buff *skb)
1088 {
1089         struct virtio_vsock_pkt_info info = {
1090                 .op = VIRTIO_VSOCK_OP_RST,
1091                 .reply = !!skb,
1092                 .vsk = vsk,
1093         };
1094
1095         /* Send RST only if the original pkt is not a RST pkt */
1096         if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST)
1097                 return 0;
1098
1099         return virtio_transport_send_pkt_info(vsk, &info);
1100 }
1101
1102 /* Normally packets are associated with a socket.  There may be no socket if an
1103  * attempt was made to connect to a socket that does not exist.
1104  */
1105 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
1106                                           struct sk_buff *skb)
1107 {
1108         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1109         struct virtio_vsock_pkt_info info = {
1110                 .op = VIRTIO_VSOCK_OP_RST,
1111                 .type = le16_to_cpu(hdr->type),
1112                 .reply = true,
1113         };
1114         struct sk_buff *reply;
1115
1116         /* Send RST only if the original pkt is not a RST pkt */
1117         if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1118                 return 0;
1119
1120         if (!t)
1121                 return -ENOTCONN;
1122
1123         reply = virtio_transport_alloc_skb(&info, 0, false,
1124                                            le64_to_cpu(hdr->dst_cid),
1125                                            le32_to_cpu(hdr->dst_port),
1126                                            le64_to_cpu(hdr->src_cid),
1127                                            le32_to_cpu(hdr->src_port));
1128         if (!reply)
1129                 return -ENOMEM;
1130
1131         return t->send_pkt(reply);
1132 }
1133
1134 /* This function should be called with sk_lock held and SOCK_DONE set */
1135 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
1136 {
1137         struct virtio_vsock_sock *vvs = vsk->trans;
1138
1139         /* We don't need to take rx_lock, as the socket is closing and we are
1140          * removing it.
1141          */
1142         __skb_queue_purge(&vvs->rx_queue);
1143         vsock_remove_sock(vsk);
1144 }
1145
1146 static void virtio_transport_wait_close(struct sock *sk, long timeout)
1147 {
1148         if (timeout) {
1149                 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1150
1151                 add_wait_queue(sk_sleep(sk), &wait);
1152
1153                 do {
1154                         if (sk_wait_event(sk, &timeout,
1155                                           sock_flag(sk, SOCK_DONE), &wait))
1156                                 break;
1157                 } while (!signal_pending(current) && timeout);
1158
1159                 remove_wait_queue(sk_sleep(sk), &wait);
1160         }
1161 }
1162
1163 static void virtio_transport_do_close(struct vsock_sock *vsk,
1164                                       bool cancel_timeout)
1165 {
1166         struct sock *sk = sk_vsock(vsk);
1167
1168         sock_set_flag(sk, SOCK_DONE);
1169         vsk->peer_shutdown = SHUTDOWN_MASK;
1170         if (vsock_stream_has_data(vsk) <= 0)
1171                 sk->sk_state = TCP_CLOSING;
1172         sk->sk_state_change(sk);
1173
1174         if (vsk->close_work_scheduled &&
1175             (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
1176                 vsk->close_work_scheduled = false;
1177
1178                 virtio_transport_remove_sock(vsk);
1179
1180                 /* Release refcnt obtained when we scheduled the timeout */
1181                 sock_put(sk);
1182         }
1183 }
1184
1185 static void virtio_transport_close_timeout(struct work_struct *work)
1186 {
1187         struct vsock_sock *vsk =
1188                 container_of(work, struct vsock_sock, close_work.work);
1189         struct sock *sk = sk_vsock(vsk);
1190
1191         sock_hold(sk);
1192         lock_sock(sk);
1193
1194         if (!sock_flag(sk, SOCK_DONE)) {
1195                 (void)virtio_transport_reset(vsk, NULL);
1196
1197                 virtio_transport_do_close(vsk, false);
1198         }
1199
1200         vsk->close_work_scheduled = false;
1201
1202         release_sock(sk);
1203         sock_put(sk);
1204 }
1205
1206 /* User context, vsk->sk is locked */
1207 static bool virtio_transport_close(struct vsock_sock *vsk)
1208 {
1209         struct sock *sk = &vsk->sk;
1210
1211         if (!(sk->sk_state == TCP_ESTABLISHED ||
1212               sk->sk_state == TCP_CLOSING))
1213                 return true;
1214
1215         /* Already received SHUTDOWN from peer, reply with RST */
1216         if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
1217                 (void)virtio_transport_reset(vsk, NULL);
1218                 return true;
1219         }
1220
1221         if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
1222                 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
1223
1224         if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
1225                 virtio_transport_wait_close(sk, sk->sk_lingertime);
1226
1227         if (sock_flag(sk, SOCK_DONE)) {
1228                 return true;
1229         }
1230
1231         sock_hold(sk);
1232         INIT_DELAYED_WORK(&vsk->close_work,
1233                           virtio_transport_close_timeout);
1234         vsk->close_work_scheduled = true;
1235         schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
1236         return false;
1237 }
1238
1239 void virtio_transport_release(struct vsock_sock *vsk)
1240 {
1241         struct sock *sk = &vsk->sk;
1242         bool remove_sock = true;
1243
1244         if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)
1245                 remove_sock = virtio_transport_close(vsk);
1246
1247         if (remove_sock) {
1248                 sock_set_flag(sk, SOCK_DONE);
1249                 virtio_transport_remove_sock(vsk);
1250         }
1251 }
1252 EXPORT_SYMBOL_GPL(virtio_transport_release);
1253
1254 static int
1255 virtio_transport_recv_connecting(struct sock *sk,
1256                                  struct sk_buff *skb)
1257 {
1258         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1259         struct vsock_sock *vsk = vsock_sk(sk);
1260         int skerr;
1261         int err;
1262
1263         switch (le16_to_cpu(hdr->op)) {
1264         case VIRTIO_VSOCK_OP_RESPONSE:
1265                 sk->sk_state = TCP_ESTABLISHED;
1266                 sk->sk_socket->state = SS_CONNECTED;
1267                 vsock_insert_connected(vsk);
1268                 sk->sk_state_change(sk);
1269                 break;
1270         case VIRTIO_VSOCK_OP_INVALID:
1271                 break;
1272         case VIRTIO_VSOCK_OP_RST:
1273                 skerr = ECONNRESET;
1274                 err = 0;
1275                 goto destroy;
1276         default:
1277                 skerr = EPROTO;
1278                 err = -EINVAL;
1279                 goto destroy;
1280         }
1281         return 0;
1282
1283 destroy:
1284         virtio_transport_reset(vsk, skb);
1285         sk->sk_state = TCP_CLOSE;
1286         sk->sk_err = skerr;
1287         sk_error_report(sk);
1288         return err;
1289 }
1290
1291 static void
1292 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
1293                               struct sk_buff *skb)
1294 {
1295         struct virtio_vsock_sock *vvs = vsk->trans;
1296         bool can_enqueue, free_pkt = false;
1297         struct virtio_vsock_hdr *hdr;
1298         u32 len;
1299
1300         hdr = virtio_vsock_hdr(skb);
1301         len = le32_to_cpu(hdr->len);
1302
1303         spin_lock_bh(&vvs->rx_lock);
1304
1305         can_enqueue = virtio_transport_inc_rx_pkt(vvs, len);
1306         if (!can_enqueue) {
1307                 free_pkt = true;
1308                 goto out;
1309         }
1310
1311         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1312                 vvs->msg_count++;
1313
1314         /* Try to copy small packets into the buffer of last packet queued,
1315          * to avoid wasting memory queueing the entire buffer with a small
1316          * payload.
1317          */
1318         if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
1319                 struct virtio_vsock_hdr *last_hdr;
1320                 struct sk_buff *last_skb;
1321
1322                 last_skb = skb_peek_tail(&vvs->rx_queue);
1323                 last_hdr = virtio_vsock_hdr(last_skb);
1324
1325                 /* If there is space in the last packet queued, we copy the
1326                  * new packet in its buffer. We avoid this if the last packet
1327                  * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is
1328                  * delimiter of SEQPACKET message, so 'pkt' is the first packet
1329                  * of a new message.
1330                  */
1331                 if (skb->len < skb_tailroom(last_skb) &&
1332                     !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) {
1333                         memcpy(skb_put(last_skb, skb->len), skb->data, skb->len);
1334                         free_pkt = true;
1335                         last_hdr->flags |= hdr->flags;
1336                         le32_add_cpu(&last_hdr->len, len);
1337                         goto out;
1338                 }
1339         }
1340
1341         __skb_queue_tail(&vvs->rx_queue, skb);
1342
1343 out:
1344         spin_unlock_bh(&vvs->rx_lock);
1345         if (free_pkt)
1346                 kfree_skb(skb);
1347 }
1348
1349 static int
1350 virtio_transport_recv_connected(struct sock *sk,
1351                                 struct sk_buff *skb)
1352 {
1353         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1354         struct vsock_sock *vsk = vsock_sk(sk);
1355         int err = 0;
1356
1357         switch (le16_to_cpu(hdr->op)) {
1358         case VIRTIO_VSOCK_OP_RW:
1359                 virtio_transport_recv_enqueue(vsk, skb);
1360                 vsock_data_ready(sk);
1361                 return err;
1362         case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
1363                 virtio_transport_send_credit_update(vsk);
1364                 break;
1365         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
1366                 sk->sk_write_space(sk);
1367                 break;
1368         case VIRTIO_VSOCK_OP_SHUTDOWN:
1369                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
1370                         vsk->peer_shutdown |= RCV_SHUTDOWN;
1371                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
1372                         vsk->peer_shutdown |= SEND_SHUTDOWN;
1373                 if (vsk->peer_shutdown == SHUTDOWN_MASK) {
1374                         if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) {
1375                                 (void)virtio_transport_reset(vsk, NULL);
1376                                 virtio_transport_do_close(vsk, true);
1377                         }
1378                         /* Remove this socket anyway because the remote peer sent
1379                          * the shutdown. This way a new connection will succeed
1380                          * if the remote peer uses the same source port,
1381                          * even if the old socket is still unreleased, but now disconnected.
1382                          */
1383                         vsock_remove_sock(vsk);
1384                 }
1385                 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags))
1386                         sk->sk_state_change(sk);
1387                 break;
1388         case VIRTIO_VSOCK_OP_RST:
1389                 virtio_transport_do_close(vsk, true);
1390                 break;
1391         default:
1392                 err = -EINVAL;
1393                 break;
1394         }
1395
1396         kfree_skb(skb);
1397         return err;
1398 }
1399
1400 static void
1401 virtio_transport_recv_disconnecting(struct sock *sk,
1402                                     struct sk_buff *skb)
1403 {
1404         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1405         struct vsock_sock *vsk = vsock_sk(sk);
1406
1407         if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1408                 virtio_transport_do_close(vsk, true);
1409 }
1410
1411 static int
1412 virtio_transport_send_response(struct vsock_sock *vsk,
1413                                struct sk_buff *skb)
1414 {
1415         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1416         struct virtio_vsock_pkt_info info = {
1417                 .op = VIRTIO_VSOCK_OP_RESPONSE,
1418                 .remote_cid = le64_to_cpu(hdr->src_cid),
1419                 .remote_port = le32_to_cpu(hdr->src_port),
1420                 .reply = true,
1421                 .vsk = vsk,
1422         };
1423
1424         return virtio_transport_send_pkt_info(vsk, &info);
1425 }
1426
1427 static bool virtio_transport_space_update(struct sock *sk,
1428                                           struct sk_buff *skb)
1429 {
1430         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1431         struct vsock_sock *vsk = vsock_sk(sk);
1432         struct virtio_vsock_sock *vvs = vsk->trans;
1433         bool space_available;
1434
1435         /* Listener sockets are not associated with any transport, so we are
1436          * not able to take the state to see if there is space available in the
1437          * remote peer, but since they are only used to receive requests, we
1438          * can assume that there is always space available in the other peer.
1439          */
1440         if (!vvs)
1441                 return true;
1442
1443         /* buf_alloc and fwd_cnt is always included in the hdr */
1444         spin_lock_bh(&vvs->tx_lock);
1445         vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc);
1446         vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt);
1447         space_available = virtio_transport_has_space(vsk);
1448         spin_unlock_bh(&vvs->tx_lock);
1449         return space_available;
1450 }
1451
1452 /* Handle server socket */
1453 static int
1454 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb,
1455                              struct virtio_transport *t)
1456 {
1457         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1458         struct vsock_sock *vsk = vsock_sk(sk);
1459         struct vsock_sock *vchild;
1460         struct sock *child;
1461         int ret;
1462
1463         if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) {
1464                 virtio_transport_reset_no_sock(t, skb);
1465                 return -EINVAL;
1466         }
1467
1468         if (sk_acceptq_is_full(sk)) {
1469                 virtio_transport_reset_no_sock(t, skb);
1470                 return -ENOMEM;
1471         }
1472
1473         child = vsock_create_connected(sk);
1474         if (!child) {
1475                 virtio_transport_reset_no_sock(t, skb);
1476                 return -ENOMEM;
1477         }
1478
1479         sk_acceptq_added(sk);
1480
1481         lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1482
1483         child->sk_state = TCP_ESTABLISHED;
1484
1485         vchild = vsock_sk(child);
1486         vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid),
1487                         le32_to_cpu(hdr->dst_port));
1488         vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid),
1489                         le32_to_cpu(hdr->src_port));
1490
1491         ret = vsock_assign_transport(vchild, vsk);
1492         /* Transport assigned (looking at remote_addr) must be the same
1493          * where we received the request.
1494          */
1495         if (ret || vchild->transport != &t->transport) {
1496                 release_sock(child);
1497                 virtio_transport_reset_no_sock(t, skb);
1498                 sock_put(child);
1499                 return ret;
1500         }
1501
1502         if (virtio_transport_space_update(child, skb))
1503                 child->sk_write_space(child);
1504
1505         vsock_insert_connected(vchild);
1506         vsock_enqueue_accept(sk, child);
1507         virtio_transport_send_response(vchild, skb);
1508
1509         release_sock(child);
1510
1511         sk->sk_data_ready(sk);
1512         return 0;
1513 }
1514
1515 static bool virtio_transport_valid_type(u16 type)
1516 {
1517         return (type == VIRTIO_VSOCK_TYPE_STREAM) ||
1518                (type == VIRTIO_VSOCK_TYPE_SEQPACKET);
1519 }
1520
1521 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1522  * lock.
1523  */
1524 void virtio_transport_recv_pkt(struct virtio_transport *t,
1525                                struct sk_buff *skb)
1526 {
1527         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1528         struct sockaddr_vm src, dst;
1529         struct vsock_sock *vsk;
1530         struct sock *sk;
1531         bool space_available;
1532
1533         vsock_addr_init(&src, le64_to_cpu(hdr->src_cid),
1534                         le32_to_cpu(hdr->src_port));
1535         vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid),
1536                         le32_to_cpu(hdr->dst_port));
1537
1538         trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1539                                         dst.svm_cid, dst.svm_port,
1540                                         le32_to_cpu(hdr->len),
1541                                         le16_to_cpu(hdr->type),
1542                                         le16_to_cpu(hdr->op),
1543                                         le32_to_cpu(hdr->flags),
1544                                         le32_to_cpu(hdr->buf_alloc),
1545                                         le32_to_cpu(hdr->fwd_cnt));
1546
1547         if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) {
1548                 (void)virtio_transport_reset_no_sock(t, skb);
1549                 goto free_pkt;
1550         }
1551
1552         /* The socket must be in connected or bound table
1553          * otherwise send reset back
1554          */
1555         sk = vsock_find_connected_socket(&src, &dst);
1556         if (!sk) {
1557                 sk = vsock_find_bound_socket(&dst);
1558                 if (!sk) {
1559                         (void)virtio_transport_reset_no_sock(t, skb);
1560                         goto free_pkt;
1561                 }
1562         }
1563
1564         if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) {
1565                 (void)virtio_transport_reset_no_sock(t, skb);
1566                 sock_put(sk);
1567                 goto free_pkt;
1568         }
1569
1570         if (!skb_set_owner_sk_safe(skb, sk)) {
1571                 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
1572                 goto free_pkt;
1573         }
1574
1575         vsk = vsock_sk(sk);
1576
1577         lock_sock(sk);
1578
1579         /* Check if sk has been closed before lock_sock */
1580         if (sock_flag(sk, SOCK_DONE)) {
1581                 (void)virtio_transport_reset_no_sock(t, skb);
1582                 release_sock(sk);
1583                 sock_put(sk);
1584                 goto free_pkt;
1585         }
1586
1587         space_available = virtio_transport_space_update(sk, skb);
1588
1589         /* Update CID in case it has changed after a transport reset event */
1590         if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1591                 vsk->local_addr.svm_cid = dst.svm_cid;
1592
1593         if (space_available)
1594                 sk->sk_write_space(sk);
1595
1596         switch (sk->sk_state) {
1597         case TCP_LISTEN:
1598                 virtio_transport_recv_listen(sk, skb, t);
1599                 kfree_skb(skb);
1600                 break;
1601         case TCP_SYN_SENT:
1602                 virtio_transport_recv_connecting(sk, skb);
1603                 kfree_skb(skb);
1604                 break;
1605         case TCP_ESTABLISHED:
1606                 virtio_transport_recv_connected(sk, skb);
1607                 break;
1608         case TCP_CLOSING:
1609                 virtio_transport_recv_disconnecting(sk, skb);
1610                 kfree_skb(skb);
1611                 break;
1612         default:
1613                 (void)virtio_transport_reset_no_sock(t, skb);
1614                 kfree_skb(skb);
1615                 break;
1616         }
1617
1618         release_sock(sk);
1619
1620         /* Release refcnt obtained when we fetched this socket out of the
1621          * bound or connected list.
1622          */
1623         sock_put(sk);
1624         return;
1625
1626 free_pkt:
1627         kfree_skb(skb);
1628 }
1629 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1630
1631 /* Remove skbs found in a queue that have a vsk that matches.
1632  *
1633  * Each skb is freed.
1634  *
1635  * Returns the count of skbs that were reply packets.
1636  */
1637 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue)
1638 {
1639         struct sk_buff_head freeme;
1640         struct sk_buff *skb, *tmp;
1641         int cnt = 0;
1642
1643         skb_queue_head_init(&freeme);
1644
1645         spin_lock_bh(&queue->lock);
1646         skb_queue_walk_safe(queue, skb, tmp) {
1647                 if (vsock_sk(skb->sk) != vsk)
1648                         continue;
1649
1650                 __skb_unlink(skb, queue);
1651                 __skb_queue_tail(&freeme, skb);
1652
1653                 if (virtio_vsock_skb_reply(skb))
1654                         cnt++;
1655         }
1656         spin_unlock_bh(&queue->lock);
1657
1658         __skb_queue_purge(&freeme);
1659
1660         return cnt;
1661 }
1662 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs);
1663
1664 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor)
1665 {
1666         struct virtio_vsock_sock *vvs = vsk->trans;
1667         struct sock *sk = sk_vsock(vsk);
1668         struct sk_buff *skb;
1669         int off = 0;
1670         int err;
1671
1672         spin_lock_bh(&vvs->rx_lock);
1673         /* Use __skb_recv_datagram() for race-free handling of the receive. It
1674          * works for types other than dgrams.
1675          */
1676         skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err);
1677         spin_unlock_bh(&vvs->rx_lock);
1678
1679         if (!skb)
1680                 return err;
1681
1682         return recv_actor(sk, skb);
1683 }
1684 EXPORT_SYMBOL_GPL(virtio_transport_read_skb);
1685
1686 MODULE_LICENSE("GPL v2");
1687 MODULE_AUTHOR("Asias He");
1688 MODULE_DESCRIPTION("common code for virtio vsock");