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         u32 fwd_cnt_delta;
 402         bool low_rx_bytes;
 403         int err = -EFAULT;
 404         u32 free_space;
 405
 406         spin_lock_bh(&vvs->rx_lock);
 407
 408         if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes,
 409                       "rx_queue is empty, but rx_bytes is non-zero\n")) {
 410                 spin_unlock_bh(&vvs->rx_lock);
 411                 return err;
 412         }
 413
 414         while (total < len && !skb_queue_empty(&vvs->rx_queue)) {
 415                 skb = skb_peek(&vvs->rx_queue);
 416
 417                 bytes = len - total;
 418                 if (bytes > skb->len)
 419                         bytes = skb->len;
 420
 421                 /* sk_lock is held by caller so no one else can dequeue.
 422                  * Unlock rx_lock since memcpy_to_msg() may sleep.
 423                  */
 424                 spin_unlock_bh(&vvs->rx_lock);
 425
 426                 err = memcpy_to_msg(msg, skb->data, bytes);
 427                 if (err)
 428                         goto out;
 429
 430                 spin_lock_bh(&vvs->rx_lock);
 431
 432                 total += bytes;
 433                 skb_pull(skb, bytes);
 434
 435                 if (skb->len == 0) {
 436                         u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
 437
 438                         virtio_transport_dec_rx_pkt(vvs, pkt_len);
 439                         __skb_unlink(skb, &vvs->rx_queue);
 440                         consume_skb(skb);
 441                 }
 442         }
 443
 444         fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt;
 445         free_space = vvs->buf_alloc - fwd_cnt_delta;
 446         low_rx_bytes = (vvs->rx_bytes <
 447                         sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX));
 448
 449         spin_unlock_bh(&vvs->rx_lock);
 450
 451         /* To reduce the number of credit update messages,
 452          * don't update credits as long as lots of space is available.
 453          * Note: the limit chosen here is arbitrary. Setting the limit
 454          * too high causes extra messages. Too low causes transmitter
 455          * stalls. As stalls are in theory more expensive than extra
 456          * messages, we set the limit to a high value. TODO: experiment
 457          * with different values. Also send credit update message when
 458          * number of bytes in rx queue is not enough to wake up reader.
 459          */
 460         if (fwd_cnt_delta &&
 461             (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes))
 462                 virtio_transport_send_credit_update(vsk);
 463
 464         return total;
 465
 466 out:
 467         if (total)
 468                 err = total;
 469         return err;
 470 }
 471
 472 static ssize_t
 473 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk,
 474                                    struct msghdr *msg)
 475 {
 476         struct virtio_vsock_sock *vvs = vsk->trans;
 477         struct sk_buff *skb;
 478         size_t total, len;
 479
 480         spin_lock_bh(&vvs->rx_lock);
 481
 482         if (!vvs->msg_count) {
 483                 spin_unlock_bh(&vvs->rx_lock);
 484                 return 0;
 485         }
 486
 487         total = 0;
 488         len = msg_data_left(msg);
 489
 490         skb_queue_walk(&vvs->rx_queue, skb) {
 491                 struct virtio_vsock_hdr *hdr;
 492
 493                 if (total < len) {
 494                         size_t bytes;
 495                         int err;
 496
 497                         bytes = len - total;
 498                         if (bytes > skb->len)
 499                                 bytes = skb->len;
 500
 501                         spin_unlock_bh(&vvs->rx_lock);
 502
 503                         /* sk_lock is held by caller so no one else can dequeue.
 504                          * Unlock rx_lock since memcpy_to_msg() may sleep.
 505                          */
 506                         err = memcpy_to_msg(msg, skb->data, bytes);
 507                         if (err)
 508                                 return err;
 509
 510                         spin_lock_bh(&vvs->rx_lock);
 511                 }
 512
 513                 total += skb->len;
 514                 hdr = virtio_vsock_hdr(skb);
 515
 516                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
 517                         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
 518                                 msg->msg_flags |= MSG_EOR;
 519
 520                         break;
 521                 }
 522         }
 523
 524         spin_unlock_bh(&vvs->rx_lock);
 525
 526         return total;
 527 }
 528
 529 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
 530                                                  struct msghdr *msg,
 531                                                  int flags)
 532 {
 533         struct virtio_vsock_sock *vvs = vsk->trans;
 534         int dequeued_len = 0;
 535         size_t user_buf_len = msg_data_left(msg);
 536         bool msg_ready = false;
 537         struct sk_buff *skb;
 538
 539         spin_lock_bh(&vvs->rx_lock);
 540
 541         if (vvs->msg_count == 0) {
 542                 spin_unlock_bh(&vvs->rx_lock);
 543                 return 0;
 544         }
 545
 546         while (!msg_ready) {
 547                 struct virtio_vsock_hdr *hdr;
 548                 size_t pkt_len;
 549
 550                 skb = __skb_dequeue(&vvs->rx_queue);
 551                 if (!skb)
 552                         break;
 553                 hdr = virtio_vsock_hdr(skb);
 554                 pkt_len = (size_t)le32_to_cpu(hdr->len);
 555
 556                 if (dequeued_len >= 0) {
 557                         size_t bytes_to_copy;
 558
 559                         bytes_to_copy = min(user_buf_len, pkt_len);
 560
 561                         if (bytes_to_copy) {
 562                                 int err;
 563
 564                                 /* sk_lock is held by caller so no one else can dequeue.
 565                                  * Unlock rx_lock since memcpy_to_msg() may sleep.
 566                                  */
 567                                 spin_unlock_bh(&vvs->rx_lock);
 568
 569                                 err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
 570                                 if (err) {
 571                                         /* Copy of message failed. Rest of
 572                                          * fragments will be freed without copy.
 573                                          */
 574                                         dequeued_len = err;
 575                                 } else {
 576                                         user_buf_len -= bytes_to_copy;
 577                                 }
 578
 579                                 spin_lock_bh(&vvs->rx_lock);
 580                         }
 581
 582                         if (dequeued_len >= 0)
 583                                 dequeued_len += pkt_len;
 584                 }
 585
 586                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
 587                         msg_ready = true;
 588                         vvs->msg_count--;
 589
 590                         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
 591                                 msg->msg_flags |= MSG_EOR;
 592                 }
 593
 594                 virtio_transport_dec_rx_pkt(vvs, pkt_len);
 595                 kfree_skb(skb);
 596         }
 597
 598         spin_unlock_bh(&vvs->rx_lock);
 599
 600         virtio_transport_send_credit_update(vsk);
 601
 602         return dequeued_len;
 603 }
 604
 605 ssize_t
 606 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
 607                                 struct msghdr *msg,
 608                                 size_t len, int flags)
 609 {
 610         if (flags & MSG_PEEK)
 611                 return virtio_transport_stream_do_peek(vsk, msg, len);
 612         else
 613                 return virtio_transport_stream_do_dequeue(vsk, msg, len);
 614 }
 615 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
 616
 617 ssize_t
 618 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk,
 619                                    struct msghdr *msg,
 620                                    int flags)
 621 {
 622         if (flags & MSG_PEEK)
 623                 return virtio_transport_seqpacket_do_peek(vsk, msg);
 624         else
 625                 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags);
 626 }
 627 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue);
 628
 629 int
 630 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk,
 631                                    struct msghdr *msg,
 632                                    size_t len)
 633 {
 634         struct virtio_vsock_sock *vvs = vsk->trans;
 635
 636         spin_lock_bh(&vvs->tx_lock);
 637
 638         if (len > vvs->peer_buf_alloc) {
 639                 spin_unlock_bh(&vvs->tx_lock);
 640                 return -EMSGSIZE;
 641         }
 642
 643         spin_unlock_bh(&vvs->tx_lock);
 644
 645         return virtio_transport_stream_enqueue(vsk, msg, len);
 646 }
 647 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue);
 648
 649 int
 650 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
 651                                struct msghdr *msg,
 652                                size_t len, int flags)
 653 {
 654         return -EOPNOTSUPP;
 655 }
 656 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
 657
 658 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
 659 {
 660         struct virtio_vsock_sock *vvs = vsk->trans;
 661         s64 bytes;
 662
 663         spin_lock_bh(&vvs->rx_lock);
 664         bytes = vvs->rx_bytes;
 665         spin_unlock_bh(&vvs->rx_lock);
 666
 667         return bytes;
 668 }
 669 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
 670
 671 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk)
 672 {
 673         struct virtio_vsock_sock *vvs = vsk->trans;
 674         u32 msg_count;
 675
 676         spin_lock_bh(&vvs->rx_lock);
 677         msg_count = vvs->msg_count;
 678         spin_unlock_bh(&vvs->rx_lock);
 679
 680         return msg_count;
 681 }
 682 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data);
 683
 684 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
 685 {
 686         struct virtio_vsock_sock *vvs = vsk->trans;
 687         s64 bytes;
 688
 689         bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 690         if (bytes < 0)
 691                 bytes = 0;
 692
 693         return bytes;
 694 }
 695
 696 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
 697 {
 698         struct virtio_vsock_sock *vvs = vsk->trans;
 699         s64 bytes;
 700
 701         spin_lock_bh(&vvs->tx_lock);
 702         bytes = virtio_transport_has_space(vsk);
 703         spin_unlock_bh(&vvs->tx_lock);
 704
 705         return bytes;
 706 }
 707 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
 708
 709 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
 710                                     struct vsock_sock *psk)
 711 {
 712         struct virtio_vsock_sock *vvs;
 713
 714         vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
 715         if (!vvs)
 716                 return -ENOMEM;
 717
 718         vsk->trans = vvs;
 719         vvs->vsk = vsk;
 720         if (psk && psk->trans) {
 721                 struct virtio_vsock_sock *ptrans = psk->trans;
 722
 723                 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
 724         }
 725
 726         if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
 727                 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
 728
 729         vvs->buf_alloc = vsk->buffer_size;
 730
 731         spin_lock_init(&vvs->rx_lock);
 732         spin_lock_init(&vvs->tx_lock);
 733         skb_queue_head_init(&vvs->rx_queue);
 734
 735         return 0;
 736 }
 737 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
 738
 739 /* sk_lock held by the caller */
 740 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
 741 {
 742         struct virtio_vsock_sock *vvs = vsk->trans;
 743
 744         if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
 745                 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
 746
 747         vvs->buf_alloc = *val;
 748
 749         virtio_transport_send_credit_update(vsk);
 750 }
 751 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
 752
 753 int
 754 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
 755                                 size_t target,
 756                                 bool *data_ready_now)
 757 {
 758         *data_ready_now = vsock_stream_has_data(vsk) >= target;
 759
 760         return 0;
 761 }
 762 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
 763
 764 int
 765 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
 766                                  size_t target,
 767                                  bool *space_avail_now)
 768 {
 769         s64 free_space;
 770
 771         free_space = vsock_stream_has_space(vsk);
 772         if (free_space > 0)
 773                 *space_avail_now = true;
 774         else if (free_space == 0)
 775                 *space_avail_now = false;
 776
 777         return 0;
 778 }
 779 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
 780
 781 int virtio_transport_notify_recv_init(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_init);
 787
 788 int virtio_transport_notify_recv_pre_block(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_block);
 794
 795 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
 796         size_t target, struct vsock_transport_recv_notify_data *data)
 797 {
 798         return 0;
 799 }
 800 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
 801
 802 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
 803         size_t target, ssize_t copied, bool data_read,
 804         struct vsock_transport_recv_notify_data *data)
 805 {
 806         return 0;
 807 }
 808 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
 809
 810 int virtio_transport_notify_send_init(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_init);
 816
 817 int virtio_transport_notify_send_pre_block(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_block);
 823
 824 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
 825         struct vsock_transport_send_notify_data *data)
 826 {
 827         return 0;
 828 }
 829 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
 830
 831 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
 832         ssize_t written, struct vsock_transport_send_notify_data *data)
 833 {
 834         return 0;
 835 }
 836 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
 837
 838 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
 839 {
 840         return vsk->buffer_size;
 841 }
 842 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
 843
 844 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
 845 {
 846         return true;
 847 }
 848 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
 849
 850 bool virtio_transport_stream_allow(u32 cid, u32 port)
 851 {
 852         return true;
 853 }
 854 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
 855
 856 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
 857                                 struct sockaddr_vm *addr)
 858 {
 859         return -EOPNOTSUPP;
 860 }
 861 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
 862
 863 bool virtio_transport_dgram_allow(u32 cid, u32 port)
 864 {
 865         return false;
 866 }
 867 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
 868
 869 int virtio_transport_connect(struct vsock_sock *vsk)
 870 {
 871         struct virtio_vsock_pkt_info info = {
 872                 .op = VIRTIO_VSOCK_OP_REQUEST,
 873                 .vsk = vsk,
 874         };
 875
 876         return virtio_transport_send_pkt_info(vsk, &info);
 877 }
 878 EXPORT_SYMBOL_GPL(virtio_transport_connect);
 879
 880 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
 881 {
 882         struct virtio_vsock_pkt_info info = {
 883                 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
 884                 .flags = (mode & RCV_SHUTDOWN ?
 885                           VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
 886                          (mode & SEND_SHUTDOWN ?
 887                           VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
 888                 .vsk = vsk,
 889         };
 890
 891         return virtio_transport_send_pkt_info(vsk, &info);
 892 }
 893 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
 894
 895 int
 896 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
 897                                struct sockaddr_vm *remote_addr,
 898                                struct msghdr *msg,
 899                                size_t dgram_len)
 900 {
 901         return -EOPNOTSUPP;
 902 }
 903 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
 904
 905 ssize_t
 906 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
 907                                 struct msghdr *msg,
 908                                 size_t len)
 909 {
 910         struct virtio_vsock_pkt_info info = {
 911                 .op = VIRTIO_VSOCK_OP_RW,
 912                 .msg = msg,
 913                 .pkt_len = len,
 914                 .vsk = vsk,
 915         };
 916
 917         return virtio_transport_send_pkt_info(vsk, &info);
 918 }
 919 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
 920
 921 void virtio_transport_destruct(struct vsock_sock *vsk)
 922 {
 923         struct virtio_vsock_sock *vvs = vsk->trans;
 924
 925         kfree(vvs);
 926 }
 927 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
 928
 929 static int virtio_transport_reset(struct vsock_sock *vsk,
 930                                   struct sk_buff *skb)
 931 {
 932         struct virtio_vsock_pkt_info info = {
 933                 .op = VIRTIO_VSOCK_OP_RST,
 934                 .reply = !!skb,
 935                 .vsk = vsk,
 936         };
 937
 938         /* Send RST only if the original pkt is not a RST pkt */
 939         if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST)
 940                 return 0;
 941
 942         return virtio_transport_send_pkt_info(vsk, &info);
 943 }
 944
 945 /* Normally packets are associated with a socket.  There may be no socket if an
 946  * attempt was made to connect to a socket that does not exist.
 947  */
 948 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
 949                                           struct sk_buff *skb)
 950 {
 951         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
 952         struct virtio_vsock_pkt_info info = {
 953                 .op = VIRTIO_VSOCK_OP_RST,
 954                 .type = le16_to_cpu(hdr->type),
 955                 .reply = true,
 956         };
 957         struct sk_buff *reply;
 958
 959         /* Send RST only if the original pkt is not a RST pkt */
 960         if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
 961                 return 0;
 962
 963         if (!t)
 964                 return -ENOTCONN;
 965
 966         reply = virtio_transport_alloc_skb(&info, 0,
 967                                            le64_to_cpu(hdr->dst_cid),
 968                                            le32_to_cpu(hdr->dst_port),
 969                                            le64_to_cpu(hdr->src_cid),
 970                                            le32_to_cpu(hdr->src_port));
 971         if (!reply)
 972                 return -ENOMEM;
 973
 974         return t->send_pkt(reply);
 975 }
 976
 977 /* This function should be called with sk_lock held and SOCK_DONE set */
 978 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
 979 {
 980         struct virtio_vsock_sock *vvs = vsk->trans;
 981
 982         /* We don't need to take rx_lock, as the socket is closing and we are
 983          * removing it.
 984          */
 985         __skb_queue_purge(&vvs->rx_queue);
 986         vsock_remove_sock(vsk);
 987 }
 988
 989 static void virtio_transport_wait_close(struct sock *sk, long timeout)
 990 {
 991         if (timeout) {
 992                 DEFINE_WAIT_FUNC(wait, woken_wake_function);
 993
 994                 add_wait_queue(sk_sleep(sk), &wait);
 995
 996                 do {
 997                         if (sk_wait_event(sk, &timeout,
 998                                           sock_flag(sk, SOCK_DONE), &wait))
 999                                 break;
1000                 } while (!signal_pending(current) && timeout);
1001
1002                 remove_wait_queue(sk_sleep(sk), &wait);
1003         }
1004 }
1005
1006 static void virtio_transport_do_close(struct vsock_sock *vsk,
1007                                       bool cancel_timeout)
1008 {
1009         struct sock *sk = sk_vsock(vsk);
1010
1011         sock_set_flag(sk, SOCK_DONE);
1012         vsk->peer_shutdown = SHUTDOWN_MASK;
1013         if (vsock_stream_has_data(vsk) <= 0)
1014                 sk->sk_state = TCP_CLOSING;
1015         sk->sk_state_change(sk);
1016
1017         if (vsk->close_work_scheduled &&
1018             (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
1019                 vsk->close_work_scheduled = false;
1020
1021                 virtio_transport_remove_sock(vsk);
1022
1023                 /* Release refcnt obtained when we scheduled the timeout */
1024                 sock_put(sk);
1025         }
1026 }
1027
1028 static void virtio_transport_close_timeout(struct work_struct *work)
1029 {
1030         struct vsock_sock *vsk =
1031                 container_of(work, struct vsock_sock, close_work.work);
1032         struct sock *sk = sk_vsock(vsk);
1033
1034         sock_hold(sk);
1035         lock_sock(sk);
1036
1037         if (!sock_flag(sk, SOCK_DONE)) {
1038                 (void)virtio_transport_reset(vsk, NULL);
1039
1040                 virtio_transport_do_close(vsk, false);
1041         }
1042
1043         vsk->close_work_scheduled = false;
1044
1045         release_sock(sk);
1046         sock_put(sk);
1047 }
1048
1049 /* User context, vsk->sk is locked */
1050 static bool virtio_transport_close(struct vsock_sock *vsk)
1051 {
1052         struct sock *sk = &vsk->sk;
1053
1054         if (!(sk->sk_state == TCP_ESTABLISHED ||
1055               sk->sk_state == TCP_CLOSING))
1056                 return true;
1057
1058         /* Already received SHUTDOWN from peer, reply with RST */
1059         if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
1060                 (void)virtio_transport_reset(vsk, NULL);
1061                 return true;
1062         }
1063
1064         if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
1065                 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
1066
1067         if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
1068                 virtio_transport_wait_close(sk, sk->sk_lingertime);
1069
1070         if (sock_flag(sk, SOCK_DONE)) {
1071                 return true;
1072         }
1073
1074         sock_hold(sk);
1075         INIT_DELAYED_WORK(&vsk->close_work,
1076                           virtio_transport_close_timeout);
1077         vsk->close_work_scheduled = true;
1078         schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
1079         return false;
1080 }
1081
1082 void virtio_transport_release(struct vsock_sock *vsk)
1083 {
1084         struct sock *sk = &vsk->sk;
1085         bool remove_sock = true;
1086
1087         if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)
1088                 remove_sock = virtio_transport_close(vsk);
1089
1090         if (remove_sock) {
1091                 sock_set_flag(sk, SOCK_DONE);
1092                 virtio_transport_remove_sock(vsk);
1093         }
1094 }
1095 EXPORT_SYMBOL_GPL(virtio_transport_release);
1096
1097 static int
1098 virtio_transport_recv_connecting(struct sock *sk,
1099                                  struct sk_buff *skb)
1100 {
1101         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1102         struct vsock_sock *vsk = vsock_sk(sk);
1103         int skerr;
1104         int err;
1105
1106         switch (le16_to_cpu(hdr->op)) {
1107         case VIRTIO_VSOCK_OP_RESPONSE:
1108                 sk->sk_state = TCP_ESTABLISHED;
1109                 sk->sk_socket->state = SS_CONNECTED;
1110                 vsock_insert_connected(vsk);
1111                 sk->sk_state_change(sk);
1112                 break;
1113         case VIRTIO_VSOCK_OP_INVALID:
1114                 break;
1115         case VIRTIO_VSOCK_OP_RST:
1116                 skerr = ECONNRESET;
1117                 err = 0;
1118                 goto destroy;
1119         default:
1120                 skerr = EPROTO;
1121                 err = -EINVAL;
1122                 goto destroy;
1123         }
1124         return 0;
1125
1126 destroy:
1127         virtio_transport_reset(vsk, skb);
1128         sk->sk_state = TCP_CLOSE;
1129         sk->sk_err = skerr;
1130         sk_error_report(sk);
1131         return err;
1132 }
1133
1134 static void
1135 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
1136                               struct sk_buff *skb)
1137 {
1138         struct virtio_vsock_sock *vvs = vsk->trans;
1139         bool can_enqueue, free_pkt = false;
1140         struct virtio_vsock_hdr *hdr;
1141         u32 len;
1142
1143         hdr = virtio_vsock_hdr(skb);
1144         len = le32_to_cpu(hdr->len);
1145
1146         spin_lock_bh(&vvs->rx_lock);
1147
1148         can_enqueue = virtio_transport_inc_rx_pkt(vvs, len);
1149         if (!can_enqueue) {
1150                 free_pkt = true;
1151                 goto out;
1152         }
1153
1154         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1155                 vvs->msg_count++;
1156
1157         /* Try to copy small packets into the buffer of last packet queued,
1158          * to avoid wasting memory queueing the entire buffer with a small
1159          * payload.
1160          */
1161         if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
1162                 struct virtio_vsock_hdr *last_hdr;
1163                 struct sk_buff *last_skb;
1164
1165                 last_skb = skb_peek_tail(&vvs->rx_queue);
1166                 last_hdr = virtio_vsock_hdr(last_skb);
1167
1168                 /* If there is space in the last packet queued, we copy the
1169                  * new packet in its buffer. We avoid this if the last packet
1170                  * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is
1171                  * delimiter of SEQPACKET message, so 'pkt' is the first packet
1172                  * of a new message.
1173                  */
1174                 if (skb->len < skb_tailroom(last_skb) &&
1175                     !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) {
1176                         memcpy(skb_put(last_skb, skb->len), skb->data, skb->len);
1177                         free_pkt = true;
1178                         last_hdr->flags |= hdr->flags;
1179                         le32_add_cpu(&last_hdr->len, len);
1180                         goto out;
1181                 }
1182         }
1183
1184         __skb_queue_tail(&vvs->rx_queue, skb);
1185
1186 out:
1187         spin_unlock_bh(&vvs->rx_lock);
1188         if (free_pkt)
1189                 kfree_skb(skb);
1190 }
1191
1192 static int
1193 virtio_transport_recv_connected(struct sock *sk,
1194                                 struct sk_buff *skb)
1195 {
1196         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1197         struct vsock_sock *vsk = vsock_sk(sk);
1198         int err = 0;
1199
1200         switch (le16_to_cpu(hdr->op)) {
1201         case VIRTIO_VSOCK_OP_RW:
1202                 virtio_transport_recv_enqueue(vsk, skb);
1203                 vsock_data_ready(sk);
1204                 return err;
1205         case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
1206                 virtio_transport_send_credit_update(vsk);
1207                 break;
1208         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
1209                 sk->sk_write_space(sk);
1210                 break;
1211         case VIRTIO_VSOCK_OP_SHUTDOWN:
1212                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
1213                         vsk->peer_shutdown |= RCV_SHUTDOWN;
1214                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
1215                         vsk->peer_shutdown |= SEND_SHUTDOWN;
1216                 if (vsk->peer_shutdown == SHUTDOWN_MASK) {
1217                         if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) {
1218                                 (void)virtio_transport_reset(vsk, NULL);
1219                                 virtio_transport_do_close(vsk, true);
1220                         }
1221                         /* Remove this socket anyway because the remote peer sent
1222                          * the shutdown. This way a new connection will succeed
1223                          * if the remote peer uses the same source port,
1224                          * even if the old socket is still unreleased, but now disconnected.
1225                          */
1226                         vsock_remove_sock(vsk);
1227                 }
1228                 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags))
1229                         sk->sk_state_change(sk);
1230                 break;
1231         case VIRTIO_VSOCK_OP_RST:
1232                 virtio_transport_do_close(vsk, true);
1233                 break;
1234         default:
1235                 err = -EINVAL;
1236                 break;
1237         }
1238
1239         kfree_skb(skb);
1240         return err;
1241 }
1242
1243 static void
1244 virtio_transport_recv_disconnecting(struct sock *sk,
1245                                     struct sk_buff *skb)
1246 {
1247         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1248         struct vsock_sock *vsk = vsock_sk(sk);
1249
1250         if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1251                 virtio_transport_do_close(vsk, true);
1252 }
1253
1254 static int
1255 virtio_transport_send_response(struct vsock_sock *vsk,
1256                                struct sk_buff *skb)
1257 {
1258         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1259         struct virtio_vsock_pkt_info info = {
1260                 .op = VIRTIO_VSOCK_OP_RESPONSE,
1261                 .remote_cid = le64_to_cpu(hdr->src_cid),
1262                 .remote_port = le32_to_cpu(hdr->src_port),
1263                 .reply = true,
1264                 .vsk = vsk,
1265         };
1266
1267         return virtio_transport_send_pkt_info(vsk, &info);
1268 }
1269
1270 static bool virtio_transport_space_update(struct sock *sk,
1271                                           struct sk_buff *skb)
1272 {
1273         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1274         struct vsock_sock *vsk = vsock_sk(sk);
1275         struct virtio_vsock_sock *vvs = vsk->trans;
1276         bool space_available;
1277
1278         /* Listener sockets are not associated with any transport, so we are
1279          * not able to take the state to see if there is space available in the
1280          * remote peer, but since they are only used to receive requests, we
1281          * can assume that there is always space available in the other peer.
1282          */
1283         if (!vvs)
1284                 return true;
1285
1286         /* buf_alloc and fwd_cnt is always included in the hdr */
1287         spin_lock_bh(&vvs->tx_lock);
1288         vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc);
1289         vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt);
1290         space_available = virtio_transport_has_space(vsk);
1291         spin_unlock_bh(&vvs->tx_lock);
1292         return space_available;
1293 }
1294
1295 /* Handle server socket */
1296 static int
1297 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb,
1298                              struct virtio_transport *t)
1299 {
1300         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1301         struct vsock_sock *vsk = vsock_sk(sk);
1302         struct vsock_sock *vchild;
1303         struct sock *child;
1304         int ret;
1305
1306         if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) {
1307                 virtio_transport_reset_no_sock(t, skb);
1308                 return -EINVAL;
1309         }
1310
1311         if (sk_acceptq_is_full(sk)) {
1312                 virtio_transport_reset_no_sock(t, skb);
1313                 return -ENOMEM;
1314         }
1315
1316         child = vsock_create_connected(sk);
1317         if (!child) {
1318                 virtio_transport_reset_no_sock(t, skb);
1319                 return -ENOMEM;
1320         }
1321
1322         sk_acceptq_added(sk);
1323
1324         lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1325
1326         child->sk_state = TCP_ESTABLISHED;
1327
1328         vchild = vsock_sk(child);
1329         vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid),
1330                         le32_to_cpu(hdr->dst_port));
1331         vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid),
1332                         le32_to_cpu(hdr->src_port));
1333
1334         ret = vsock_assign_transport(vchild, vsk);
1335         /* Transport assigned (looking at remote_addr) must be the same
1336          * where we received the request.
1337          */
1338         if (ret || vchild->transport != &t->transport) {
1339                 release_sock(child);
1340                 virtio_transport_reset_no_sock(t, skb);
1341                 sock_put(child);
1342                 return ret;
1343         }
1344
1345         if (virtio_transport_space_update(child, skb))
1346                 child->sk_write_space(child);
1347
1348         vsock_insert_connected(vchild);
1349         vsock_enqueue_accept(sk, child);
1350         virtio_transport_send_response(vchild, skb);
1351
1352         release_sock(child);
1353
1354         sk->sk_data_ready(sk);
1355         return 0;
1356 }
1357
1358 static bool virtio_transport_valid_type(u16 type)
1359 {
1360         return (type == VIRTIO_VSOCK_TYPE_STREAM) ||
1361                (type == VIRTIO_VSOCK_TYPE_SEQPACKET);
1362 }
1363
1364 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1365  * lock.
1366  */
1367 void virtio_transport_recv_pkt(struct virtio_transport *t,
1368                                struct sk_buff *skb)
1369 {
1370         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1371         struct sockaddr_vm src, dst;
1372         struct vsock_sock *vsk;
1373         struct sock *sk;
1374         bool space_available;
1375
1376         vsock_addr_init(&src, le64_to_cpu(hdr->src_cid),
1377                         le32_to_cpu(hdr->src_port));
1378         vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid),
1379                         le32_to_cpu(hdr->dst_port));
1380
1381         trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1382                                         dst.svm_cid, dst.svm_port,
1383                                         le32_to_cpu(hdr->len),
1384                                         le16_to_cpu(hdr->type),
1385                                         le16_to_cpu(hdr->op),
1386                                         le32_to_cpu(hdr->flags),
1387                                         le32_to_cpu(hdr->buf_alloc),
1388                                         le32_to_cpu(hdr->fwd_cnt));
1389
1390         if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) {
1391                 (void)virtio_transport_reset_no_sock(t, skb);
1392                 goto free_pkt;
1393         }
1394
1395         /* The socket must be in connected or bound table
1396          * otherwise send reset back
1397          */
1398         sk = vsock_find_connected_socket(&src, &dst);
1399         if (!sk) {
1400                 sk = vsock_find_bound_socket(&dst);
1401                 if (!sk) {
1402                         (void)virtio_transport_reset_no_sock(t, skb);
1403                         goto free_pkt;
1404                 }
1405         }
1406
1407         if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) {
1408                 (void)virtio_transport_reset_no_sock(t, skb);
1409                 sock_put(sk);
1410                 goto free_pkt;
1411         }
1412
1413         if (!skb_set_owner_sk_safe(skb, sk)) {
1414                 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
1415                 goto free_pkt;
1416         }
1417
1418         vsk = vsock_sk(sk);
1419
1420         lock_sock(sk);
1421
1422         /* Check if sk has been closed before lock_sock */
1423         if (sock_flag(sk, SOCK_DONE)) {
1424                 (void)virtio_transport_reset_no_sock(t, skb);
1425                 release_sock(sk);
1426                 sock_put(sk);
1427                 goto free_pkt;
1428         }
1429
1430         space_available = virtio_transport_space_update(sk, skb);
1431
1432         /* Update CID in case it has changed after a transport reset event */
1433         if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1434                 vsk->local_addr.svm_cid = dst.svm_cid;
1435
1436         if (space_available)
1437                 sk->sk_write_space(sk);
1438
1439         switch (sk->sk_state) {
1440         case TCP_LISTEN:
1441                 virtio_transport_recv_listen(sk, skb, t);
1442                 kfree_skb(skb);
1443                 break;
1444         case TCP_SYN_SENT:
1445                 virtio_transport_recv_connecting(sk, skb);
1446                 kfree_skb(skb);
1447                 break;
1448         case TCP_ESTABLISHED:
1449                 virtio_transport_recv_connected(sk, skb);
1450                 break;
1451         case TCP_CLOSING:
1452                 virtio_transport_recv_disconnecting(sk, skb);
1453                 kfree_skb(skb);
1454                 break;
1455         default:
1456                 (void)virtio_transport_reset_no_sock(t, skb);
1457                 kfree_skb(skb);
1458                 break;
1459         }
1460
1461         release_sock(sk);
1462
1463         /* Release refcnt obtained when we fetched this socket out of the
1464          * bound or connected list.
1465          */
1466         sock_put(sk);
1467         return;
1468
1469 free_pkt:
1470         kfree_skb(skb);
1471 }
1472 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1473
1474 /* Remove skbs found in a queue that have a vsk that matches.
1475  *
1476  * Each skb is freed.
1477  *
1478  * Returns the count of skbs that were reply packets.
1479  */
1480 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue)
1481 {
1482         struct sk_buff_head freeme;
1483         struct sk_buff *skb, *tmp;
1484         int cnt = 0;
1485
1486         skb_queue_head_init(&freeme);
1487
1488         spin_lock_bh(&queue->lock);
1489         skb_queue_walk_safe(queue, skb, tmp) {
1490                 if (vsock_sk(skb->sk) != vsk)
1491                         continue;
1492
1493                 __skb_unlink(skb, queue);
1494                 __skb_queue_tail(&freeme, skb);
1495
1496                 if (virtio_vsock_skb_reply(skb))
1497                         cnt++;
1498         }
1499         spin_unlock_bh(&queue->lock);
1500
1501         __skb_queue_purge(&freeme);
1502
1503         return cnt;
1504 }
1505 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs);
1506
1507 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor)
1508 {
1509         struct virtio_vsock_sock *vvs = vsk->trans;
1510         struct sock *sk = sk_vsock(vsk);
1511         struct sk_buff *skb;
1512         int off = 0;
1513         int err;
1514
1515         spin_lock_bh(&vvs->rx_lock);
1516         /* Use __skb_recv_datagram() for race-free handling of the receive. It
1517          * works for types other than dgrams.
1518          */
1519         skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err);
1520         spin_unlock_bh(&vvs->rx_lock);
1521
1522         if (!skb)
1523                 return err;
1524
1525         return recv_actor(sk, skb);
1526 }
1527 EXPORT_SYMBOL_GPL(virtio_transport_read_skb);
1528
1529 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val)
1530 {
1531         struct virtio_vsock_sock *vvs = vsk->trans;
1532         bool send_update;
1533
1534         spin_lock_bh(&vvs->rx_lock);
1535
1536         /* If number of available bytes is less than new SO_RCVLOWAT value,
1537          * kick sender to send more data, because sender may sleep in its
1538          * 'send()' syscall waiting for enough space at our side. Also
1539          * don't send credit update when peer already knows actual value -
1540          * such transmission will be useless.
1541          */
1542         send_update = (vvs->rx_bytes < val) &&
1543                       (vvs->fwd_cnt != vvs->last_fwd_cnt);
1544
1545         spin_unlock_bh(&vvs->rx_lock);
1546
1547         if (send_update) {
1548                 int err;
1549
1550                 err = virtio_transport_send_credit_update(vsk);
1551                 if (err < 0)
1552                         return err;
1553         }
1554
1555         return 0;
1556 }
1557 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat);
1558
1559 MODULE_LICENSE("GPL v2");
1560 MODULE_AUTHOR("Asias He");
1561 MODULE_DESCRIPTION("common code for virtio vsock");