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