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