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 <[email protected]>
   7  *         Stefan Hajnoczi <[email protected]>
   8  */
   9 #include <linux/spinlock.h>
  10 #include <linux/module.h>
  11 #include <linux/sched/signal.h>
  12 #include <linux/ctype.h>
  13 #include <linux/list.h>
  14 #include <linux/virtio_vsock.h>
  15 #include <uapi/linux/vsockmon.h>
  16
  17 #include <net/sock.h>
  18 #include <net/af_vsock.h>
  19
  20 #define CREATE_TRACE_POINTS
  21 #include <trace/events/vsock_virtio_transport_common.h>
  22
  23 /* How long to wait for graceful shutdown of a connection */
  24 #define VSOCK_CLOSE_TIMEOUT (8 * HZ)
  25
  26 /* Threshold for detecting small packets to copy */
  27 #define GOOD_COPY_LEN  128
  28
  29 static const struct virtio_transport *
  30 virtio_transport_get_ops(struct vsock_sock *vsk)
  31 {
  32         const struct vsock_transport *t = vsock_core_get_transport(vsk);
  33
  34         if (WARN_ON(!t))
  35                 return NULL;
  36
  37         return container_of(t, struct virtio_transport, transport);
  38 }
  39
  40 static bool virtio_transport_can_zcopy(const struct virtio_transport *t_ops,
  41                                        struct virtio_vsock_pkt_info *info,
  42                                        size_t pkt_len)
  43 {
  44         struct iov_iter *iov_iter;
  45
  46         if (!info->msg)
  47                 return false;
  48
  49         iov_iter = &info->msg->msg_iter;
  50
  51         if (iov_iter->iov_offset)
  52                 return false;
  53
  54         /* We can't send whole iov. */
  55         if (iov_iter->count > pkt_len)
  56                 return false;
  57
  58         /* Check that transport can send data in zerocopy mode. */
  59         t_ops = virtio_transport_get_ops(info->vsk);
  60
  61         if (t_ops->can_msgzerocopy) {
  62                 int pages_to_send = iov_iter_npages(iov_iter, MAX_SKB_FRAGS);
  63
  64                 /* +1 is for packet header. */
  65                 return t_ops->can_msgzerocopy(pages_to_send + 1);
  66         }
  67
  68         return true;
  69 }
  70
  71 static int virtio_transport_init_zcopy_skb(struct vsock_sock *vsk,
  72                                            struct sk_buff *skb,
  73                                            struct msghdr *msg,
  74                                            bool zerocopy)
  75 {
  76         struct ubuf_info *uarg;
  77
  78         if (msg->msg_ubuf) {
  79                 uarg = msg->msg_ubuf;
  80                 net_zcopy_get(uarg);
  81         } else {
  82                 struct iov_iter *iter = &msg->msg_iter;
  83                 struct ubuf_info_msgzc *uarg_zc;
  84
  85                 uarg = msg_zerocopy_realloc(sk_vsock(vsk),
  86                                             iter->count,
  87                                             NULL);
  88                 if (!uarg)
  89                         return -1;
  90
  91                 uarg_zc = uarg_to_msgzc(uarg);
  92                 uarg_zc->zerocopy = zerocopy ? 1 : 0;
  93         }
  94
  95         skb_zcopy_init(skb, uarg);
  96
  97         return 0;
  98 }
  99
 100 static int virtio_transport_fill_skb(struct sk_buff *skb,
 101                                      struct virtio_vsock_pkt_info *info,
 102                                      size_t len,
 103                                      bool zcopy)
 104 {
 105         if (zcopy)
 106                 return __zerocopy_sg_from_iter(info->msg, NULL, skb,
 107                                                &info->msg->msg_iter,
 108                                                len);
 109
 110         return memcpy_from_msg(skb_put(skb, len), info->msg, len);
 111 }
 112
 113 static void virtio_transport_init_hdr(struct sk_buff *skb,
 114                                       struct virtio_vsock_pkt_info *info,
 115                                       size_t payload_len,
 116                                       u32 src_cid,
 117                                       u32 src_port,
 118                                       u32 dst_cid,
 119                                       u32 dst_port)
 120 {
 121         struct virtio_vsock_hdr *hdr;
 122
 123         hdr = virtio_vsock_hdr(skb);
 124         hdr->type       = cpu_to_le16(info->type);
 125         hdr->op         = cpu_to_le16(info->op);
 126         hdr->src_cid    = cpu_to_le64(src_cid);
 127         hdr->dst_cid    = cpu_to_le64(dst_cid);
 128         hdr->src_port   = cpu_to_le32(src_port);
 129         hdr->dst_port   = cpu_to_le32(dst_port);
 130         hdr->flags      = cpu_to_le32(info->flags);
 131         hdr->len        = cpu_to_le32(payload_len);
 132         hdr->buf_alloc  = cpu_to_le32(0);
 133         hdr->fwd_cnt    = cpu_to_le32(0);
 134 }
 135
 136 static void virtio_transport_copy_nonlinear_skb(const struct sk_buff *skb,
 137                                                 void *dst,
 138                                                 size_t len)
 139 {
 140         struct iov_iter iov_iter = { 0 };
 141         struct kvec kvec;
 142         size_t to_copy;
 143
 144         kvec.iov_base = dst;
 145         kvec.iov_len = len;
 146
 147         iov_iter.iter_type = ITER_KVEC;
 148         iov_iter.kvec = &kvec;
 149         iov_iter.nr_segs = 1;
 150
 151         to_copy = min_t(size_t, len, skb->len);
 152
 153         skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset,
 154                                &iov_iter, to_copy);
 155 }
 156
 157 /* Packet capture */
 158 static struct sk_buff *virtio_transport_build_skb(void *opaque)
 159 {
 160         struct virtio_vsock_hdr *pkt_hdr;
 161         struct sk_buff *pkt = opaque;
 162         struct af_vsockmon_hdr *hdr;
 163         struct sk_buff *skb;
 164         size_t payload_len;
 165
 166         /* A packet could be split to fit the RX buffer, so we can retrieve
 167          * the payload length from the header and the buffer pointer taking
 168          * care of the offset in the original packet.
 169          */
 170         pkt_hdr = virtio_vsock_hdr(pkt);
 171         payload_len = pkt->len;
 172
 173         skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len,
 174                         GFP_ATOMIC);
 175         if (!skb)
 176                 return NULL;
 177
 178         hdr = skb_put(skb, sizeof(*hdr));
 179
 180         /* pkt->hdr is little-endian so no need to byteswap here */
 181         hdr->src_cid = pkt_hdr->src_cid;
 182         hdr->src_port = pkt_hdr->src_port;
 183         hdr->dst_cid = pkt_hdr->dst_cid;
 184         hdr->dst_port = pkt_hdr->dst_port;
 185
 186         hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
 187         hdr->len = cpu_to_le16(sizeof(*pkt_hdr));
 188         memset(hdr->reserved, 0, sizeof(hdr->reserved));
 189
 190         switch (le16_to_cpu(pkt_hdr->op)) {
 191         case VIRTIO_VSOCK_OP_REQUEST:
 192         case VIRTIO_VSOCK_OP_RESPONSE:
 193                 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
 194                 break;
 195         case VIRTIO_VSOCK_OP_RST:
 196         case VIRTIO_VSOCK_OP_SHUTDOWN:
 197                 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
 198                 break;
 199         case VIRTIO_VSOCK_OP_RW:
 200                 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
 201                 break;
 202         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
 203         case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
 204                 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
 205                 break;
 206         default:
 207                 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
 208                 break;
 209         }
 210
 211         skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr));
 212
 213         if (payload_len) {
 214                 if (skb_is_nonlinear(pkt)) {
 215                         void *data = skb_put(skb, payload_len);
 216
 217                         virtio_transport_copy_nonlinear_skb(pkt, data, payload_len);
 218                 } else {
 219                         skb_put_data(skb, pkt->data, payload_len);
 220                 }
 221         }
 222
 223         return skb;
 224 }
 225
 226 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb)
 227 {
 228         if (virtio_vsock_skb_tap_delivered(skb))
 229                 return;
 230
 231         vsock_deliver_tap(virtio_transport_build_skb, skb);
 232         virtio_vsock_skb_set_tap_delivered(skb);
 233 }
 234 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
 235
 236 static u16 virtio_transport_get_type(struct sock *sk)
 237 {
 238         if (sk->sk_type == SOCK_STREAM)
 239                 return VIRTIO_VSOCK_TYPE_STREAM;
 240         else
 241                 return VIRTIO_VSOCK_TYPE_SEQPACKET;
 242 }
 243
 244 /* Returns new sk_buff on success, otherwise returns NULL. */
 245 static struct sk_buff *virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info,
 246                                                   size_t payload_len,
 247                                                   bool zcopy,
 248                                                   u32 src_cid,
 249                                                   u32 src_port,
 250                                                   u32 dst_cid,
 251                                                   u32 dst_port)
 252 {
 253         struct vsock_sock *vsk;
 254         struct sk_buff *skb;
 255         size_t skb_len;
 256
 257         skb_len = VIRTIO_VSOCK_SKB_HEADROOM;
 258
 259         if (!zcopy)
 260                 skb_len += payload_len;
 261
 262         skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL);
 263         if (!skb)
 264                 return NULL;
 265
 266         virtio_transport_init_hdr(skb, info, payload_len, src_cid, src_port,
 267                                   dst_cid, dst_port);
 268
 269         vsk = info->vsk;
 270
 271         /* If 'vsk' != NULL then payload is always present, so we
 272          * will never call '__zerocopy_sg_from_iter()' below without
 273          * setting skb owner in 'skb_set_owner_w()'. The only case
 274          * when 'vsk' == NULL is VIRTIO_VSOCK_OP_RST control message
 275          * without payload.
 276          */
 277         WARN_ON_ONCE(!(vsk && (info->msg && payload_len)) && zcopy);
 278
 279         /* Set owner here, because '__zerocopy_sg_from_iter()' uses
 280          * owner of skb without check to update 'sk_wmem_alloc'.
 281          */
 282         if (vsk)
 283                 skb_set_owner_w(skb, sk_vsock(vsk));
 284
 285         if (info->msg && payload_len > 0) {
 286                 int err;
 287
 288                 err = virtio_transport_fill_skb(skb, info, payload_len, zcopy);
 289                 if (err)
 290                         goto out;
 291
 292                 if (msg_data_left(info->msg) == 0 &&
 293                     info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) {
 294                         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
 295
 296                         hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
 297
 298                         if (info->msg->msg_flags & MSG_EOR)
 299                                 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
 300                 }
 301         }
 302
 303         if (info->reply)
 304                 virtio_vsock_skb_set_reply(skb);
 305
 306         trace_virtio_transport_alloc_pkt(src_cid, src_port,
 307                                          dst_cid, dst_port,
 308                                          payload_len,
 309                                          info->type,
 310                                          info->op,
 311                                          info->flags,
 312                                          zcopy);
 313
 314         return skb;
 315 out:
 316         kfree_skb(skb);
 317         return NULL;
 318 }
 319
 320 /* This function can only be used on connecting/connected sockets,
 321  * since a socket assigned to a transport is required.
 322  *
 323  * Do not use on listener sockets!
 324  */
 325 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
 326                                           struct virtio_vsock_pkt_info *info)
 327 {
 328         u32 max_skb_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
 329         u32 src_cid, src_port, dst_cid, dst_port;
 330         const struct virtio_transport *t_ops;
 331         struct virtio_vsock_sock *vvs;
 332         u32 pkt_len = info->pkt_len;
 333         bool can_zcopy = false;
 334         u32 rest_len;
 335         int ret;
 336
 337         info->type = virtio_transport_get_type(sk_vsock(vsk));
 338
 339         t_ops = virtio_transport_get_ops(vsk);
 340         if (unlikely(!t_ops))
 341                 return -EFAULT;
 342
 343         src_cid = t_ops->transport.get_local_cid();
 344         src_port = vsk->local_addr.svm_port;
 345         if (!info->remote_cid) {
 346                 dst_cid = vsk->remote_addr.svm_cid;
 347                 dst_port = vsk->remote_addr.svm_port;
 348         } else {
 349                 dst_cid = info->remote_cid;
 350                 dst_port = info->remote_port;
 351         }
 352
 353         vvs = vsk->trans;
 354
 355         /* virtio_transport_get_credit might return less than pkt_len credit */
 356         pkt_len = virtio_transport_get_credit(vvs, pkt_len);
 357
 358         /* Do not send zero length OP_RW pkt */
 359         if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
 360                 return pkt_len;
 361
 362         if (info->msg) {
 363                 /* If zerocopy is not enabled by 'setsockopt()', we behave as
 364                  * there is no MSG_ZEROCOPY flag set.
 365                  */
 366                 if (!sock_flag(sk_vsock(vsk), SOCK_ZEROCOPY))
 367                         info->msg->msg_flags &= ~MSG_ZEROCOPY;
 368
 369                 if (info->msg->msg_flags & MSG_ZEROCOPY)
 370                         can_zcopy = virtio_transport_can_zcopy(t_ops, info, pkt_len);
 371
 372                 if (can_zcopy)
 373                         max_skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE,
 374                                             (MAX_SKB_FRAGS * PAGE_SIZE));
 375         }
 376
 377         rest_len = pkt_len;
 378
 379         do {
 380                 struct sk_buff *skb;
 381                 size_t skb_len;
 382
 383                 skb_len = min(max_skb_len, rest_len);
 384
 385                 skb = virtio_transport_alloc_skb(info, skb_len, can_zcopy,
 386                                                  src_cid, src_port,
 387                                                  dst_cid, dst_port);
 388                 if (!skb) {
 389                         ret = -ENOMEM;
 390                         break;
 391                 }
 392
 393                 /* We process buffer part by part, allocating skb on
 394                  * each iteration. If this is last skb for this buffer
 395                  * and MSG_ZEROCOPY mode is in use - we must allocate
 396                  * completion for the current syscall.
 397                  */
 398                 if (info->msg && info->msg->msg_flags & MSG_ZEROCOPY &&
 399                     skb_len == rest_len && info->op == VIRTIO_VSOCK_OP_RW) {
 400                         if (virtio_transport_init_zcopy_skb(vsk, skb,
 401                                                             info->msg,
 402                                                             can_zcopy)) {
 403                                 kfree_skb(skb);
 404                                 ret = -ENOMEM;
 405                                 break;
 406                         }
 407                 }
 408
 409                 virtio_transport_inc_tx_pkt(vvs, skb);
 410
 411                 ret = t_ops->send_pkt(skb);
 412                 if (ret < 0)
 413                         break;
 414
 415                 /* Both virtio and vhost 'send_pkt()' returns 'skb_len',
 416                  * but for reliability use 'ret' instead of 'skb_len'.
 417                  * Also if partial send happens (e.g. 'ret' != 'skb_len')
 418                  * somehow, we break this loop, but account such returned
 419                  * value in 'virtio_transport_put_credit()'.
 420                  */
 421                 rest_len -= ret;
 422
 423                 if (WARN_ONCE(ret != skb_len,
 424                               "'send_pkt()' returns %i, but %zu expected\n",
 425                               ret, skb_len))
 426                         break;
 427         } while (rest_len);
 428
 429         virtio_transport_put_credit(vvs, rest_len);
 430
 431         /* Return number of bytes, if any data has been sent. */
 432         if (rest_len != pkt_len)
 433                 ret = pkt_len - rest_len;
 434
 435         return ret;
 436 }
 437
 438 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
 439                                         u32 len)
 440 {
 441         if (vvs->rx_bytes + len > vvs->buf_alloc)
 442                 return false;
 443
 444         vvs->rx_bytes += len;
 445         return true;
 446 }
 447
 448 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
 449                                         u32 len)
 450 {
 451         vvs->rx_bytes -= len;
 452         vvs->fwd_cnt += len;
 453 }
 454
 455 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb)
 456 {
 457         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
 458
 459         spin_lock_bh(&vvs->rx_lock);
 460         vvs->last_fwd_cnt = vvs->fwd_cnt;
 461         hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
 462         hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc);
 463         spin_unlock_bh(&vvs->rx_lock);
 464 }
 465 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
 466
 467 void virtio_transport_consume_skb_sent(struct sk_buff *skb, bool consume)
 468 {
 469         struct sock *s = skb->sk;
 470
 471         if (s && skb->len) {
 472                 struct vsock_sock *vs = vsock_sk(s);
 473                 struct virtio_vsock_sock *vvs;
 474
 475                 vvs = vs->trans;
 476
 477                 spin_lock_bh(&vvs->tx_lock);
 478                 vvs->bytes_unsent -= skb->len;
 479                 spin_unlock_bh(&vvs->tx_lock);
 480         }
 481
 482         if (consume)
 483                 consume_skb(skb);
 484 }
 485 EXPORT_SYMBOL_GPL(virtio_transport_consume_skb_sent);
 486
 487 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
 488 {
 489         u32 ret;
 490
 491         if (!credit)
 492                 return 0;
 493
 494         spin_lock_bh(&vvs->tx_lock);
 495         ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 496         if (ret > credit)
 497                 ret = credit;
 498         vvs->tx_cnt += ret;
 499         vvs->bytes_unsent += ret;
 500         spin_unlock_bh(&vvs->tx_lock);
 501
 502         return ret;
 503 }
 504 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
 505
 506 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
 507 {
 508         if (!credit)
 509                 return;
 510
 511         spin_lock_bh(&vvs->tx_lock);
 512         vvs->tx_cnt -= credit;
 513         vvs->bytes_unsent -= credit;
 514         spin_unlock_bh(&vvs->tx_lock);
 515 }
 516 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
 517
 518 static int virtio_transport_send_credit_update(struct vsock_sock *vsk)
 519 {
 520         struct virtio_vsock_pkt_info info = {
 521                 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
 522                 .vsk = vsk,
 523         };
 524
 525         return virtio_transport_send_pkt_info(vsk, &info);
 526 }
 527
 528 static ssize_t
 529 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
 530                                 struct msghdr *msg,
 531                                 size_t len)
 532 {
 533         struct virtio_vsock_sock *vvs = vsk->trans;
 534         struct sk_buff *skb;
 535         size_t total = 0;
 536         int err;
 537
 538         spin_lock_bh(&vvs->rx_lock);
 539
 540         skb_queue_walk(&vvs->rx_queue, skb) {
 541                 size_t bytes;
 542
 543                 bytes = len - total;
 544                 if (bytes > skb->len)
 545                         bytes = skb->len;
 546
 547                 spin_unlock_bh(&vvs->rx_lock);
 548
 549                 /* sk_lock is held by caller so no one else can dequeue.
 550                  * Unlock rx_lock since skb_copy_datagram_iter() may sleep.
 551                  */
 552                 err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset,
 553                                              &msg->msg_iter, bytes);
 554                 if (err)
 555                         goto out;
 556
 557                 total += bytes;
 558
 559                 spin_lock_bh(&vvs->rx_lock);
 560
 561                 if (total == len)
 562                         break;
 563         }
 564
 565         spin_unlock_bh(&vvs->rx_lock);
 566
 567         return total;
 568
 569 out:
 570         if (total)
 571                 err = total;
 572         return err;
 573 }
 574
 575 static ssize_t
 576 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
 577                                    struct msghdr *msg,
 578                                    size_t len)
 579 {
 580         struct virtio_vsock_sock *vvs = vsk->trans;
 581         size_t bytes, total = 0;
 582         struct sk_buff *skb;
 583         u32 fwd_cnt_delta;
 584         bool low_rx_bytes;
 585         int err = -EFAULT;
 586         u32 free_space;
 587
 588         spin_lock_bh(&vvs->rx_lock);
 589
 590         if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes,
 591                       "rx_queue is empty, but rx_bytes is non-zero\n")) {
 592                 spin_unlock_bh(&vvs->rx_lock);
 593                 return err;
 594         }
 595
 596         while (total < len && !skb_queue_empty(&vvs->rx_queue)) {
 597                 skb = skb_peek(&vvs->rx_queue);
 598
 599                 bytes = min_t(size_t, len - total,
 600                               skb->len - VIRTIO_VSOCK_SKB_CB(skb)->offset);
 601
 602                 /* sk_lock is held by caller so no one else can dequeue.
 603                  * Unlock rx_lock since skb_copy_datagram_iter() may sleep.
 604                  */
 605                 spin_unlock_bh(&vvs->rx_lock);
 606
 607                 err = skb_copy_datagram_iter(skb,
 608                                              VIRTIO_VSOCK_SKB_CB(skb)->offset,
 609                                              &msg->msg_iter, bytes);
 610                 if (err)
 611                         goto out;
 612
 613                 spin_lock_bh(&vvs->rx_lock);
 614
 615                 total += bytes;
 616
 617                 VIRTIO_VSOCK_SKB_CB(skb)->offset += bytes;
 618
 619                 if (skb->len == VIRTIO_VSOCK_SKB_CB(skb)->offset) {
 620                         u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
 621
 622                         virtio_transport_dec_rx_pkt(vvs, pkt_len);
 623                         __skb_unlink(skb, &vvs->rx_queue);
 624                         consume_skb(skb);
 625                 }
 626         }
 627
 628         fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt;
 629         free_space = vvs->buf_alloc - fwd_cnt_delta;
 630         low_rx_bytes = (vvs->rx_bytes <
 631                         sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX));
 632
 633         spin_unlock_bh(&vvs->rx_lock);
 634
 635         /* To reduce the number of credit update messages,
 636          * don't update credits as long as lots of space is available.
 637          * Note: the limit chosen here is arbitrary. Setting the limit
 638          * too high causes extra messages. Too low causes transmitter
 639          * stalls. As stalls are in theory more expensive than extra
 640          * messages, we set the limit to a high value. TODO: experiment
 641          * with different values. Also send credit update message when
 642          * number of bytes in rx queue is not enough to wake up reader.
 643          */
 644         if (fwd_cnt_delta &&
 645             (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes))
 646                 virtio_transport_send_credit_update(vsk);
 647
 648         return total;
 649
 650 out:
 651         if (total)
 652                 err = total;
 653         return err;
 654 }
 655
 656 static ssize_t
 657 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk,
 658                                    struct msghdr *msg)
 659 {
 660         struct virtio_vsock_sock *vvs = vsk->trans;
 661         struct sk_buff *skb;
 662         size_t total, len;
 663
 664         spin_lock_bh(&vvs->rx_lock);
 665
 666         if (!vvs->msg_count) {
 667                 spin_unlock_bh(&vvs->rx_lock);
 668                 return 0;
 669         }
 670
 671         total = 0;
 672         len = msg_data_left(msg);
 673
 674         skb_queue_walk(&vvs->rx_queue, skb) {
 675                 struct virtio_vsock_hdr *hdr;
 676
 677                 if (total < len) {
 678                         size_t bytes;
 679                         int err;
 680
 681                         bytes = len - total;
 682                         if (bytes > skb->len)
 683                                 bytes = skb->len;
 684
 685                         spin_unlock_bh(&vvs->rx_lock);
 686
 687                         /* sk_lock is held by caller so no one else can dequeue.
 688                          * Unlock rx_lock since skb_copy_datagram_iter() may sleep.
 689                          */
 690                         err = skb_copy_datagram_iter(skb, VIRTIO_VSOCK_SKB_CB(skb)->offset,
 691                                                      &msg->msg_iter, bytes);
 692                         if (err)
 693                                 return err;
 694
 695                         spin_lock_bh(&vvs->rx_lock);
 696                 }
 697
 698                 total += skb->len;
 699                 hdr = virtio_vsock_hdr(skb);
 700
 701                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
 702                         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
 703                                 msg->msg_flags |= MSG_EOR;
 704
 705                         break;
 706                 }
 707         }
 708
 709         spin_unlock_bh(&vvs->rx_lock);
 710
 711         return total;
 712 }
 713
 714 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
 715                                                  struct msghdr *msg,
 716                                                  int flags)
 717 {
 718         struct virtio_vsock_sock *vvs = vsk->trans;
 719         int dequeued_len = 0;
 720         size_t user_buf_len = msg_data_left(msg);
 721         bool msg_ready = false;
 722         struct sk_buff *skb;
 723
 724         spin_lock_bh(&vvs->rx_lock);
 725
 726         if (vvs->msg_count == 0) {
 727                 spin_unlock_bh(&vvs->rx_lock);
 728                 return 0;
 729         }
 730
 731         while (!msg_ready) {
 732                 struct virtio_vsock_hdr *hdr;
 733                 size_t pkt_len;
 734
 735                 skb = __skb_dequeue(&vvs->rx_queue);
 736                 if (!skb)
 737                         break;
 738                 hdr = virtio_vsock_hdr(skb);
 739                 pkt_len = (size_t)le32_to_cpu(hdr->len);
 740
 741                 if (dequeued_len >= 0) {
 742                         size_t bytes_to_copy;
 743
 744                         bytes_to_copy = min(user_buf_len, pkt_len);
 745
 746                         if (bytes_to_copy) {
 747                                 int err;
 748
 749                                 /* sk_lock is held by caller so no one else can dequeue.
 750                                  * Unlock rx_lock since skb_copy_datagram_iter() may sleep.
 751                                  */
 752                                 spin_unlock_bh(&vvs->rx_lock);
 753
 754                                 err = skb_copy_datagram_iter(skb, 0,
 755                                                              &msg->msg_iter,
 756                                                              bytes_to_copy);
 757                                 if (err) {
 758                                         /* Copy of message failed. Rest of
 759                                          * fragments will be freed without copy.
 760                                          */
 761                                         dequeued_len = err;
 762                                 } else {
 763                                         user_buf_len -= bytes_to_copy;
 764                                 }
 765
 766                                 spin_lock_bh(&vvs->rx_lock);
 767                         }
 768
 769                         if (dequeued_len >= 0)
 770                                 dequeued_len += pkt_len;
 771                 }
 772
 773                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
 774                         msg_ready = true;
 775                         vvs->msg_count--;
 776
 777                         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
 778                                 msg->msg_flags |= MSG_EOR;
 779                 }
 780
 781                 virtio_transport_dec_rx_pkt(vvs, pkt_len);
 782                 kfree_skb(skb);
 783         }
 784
 785         spin_unlock_bh(&vvs->rx_lock);
 786
 787         virtio_transport_send_credit_update(vsk);
 788
 789         return dequeued_len;
 790 }
 791
 792 ssize_t
 793 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
 794                                 struct msghdr *msg,
 795                                 size_t len, int flags)
 796 {
 797         if (flags & MSG_PEEK)
 798                 return virtio_transport_stream_do_peek(vsk, msg, len);
 799         else
 800                 return virtio_transport_stream_do_dequeue(vsk, msg, len);
 801 }
 802 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
 803
 804 ssize_t
 805 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk,
 806                                    struct msghdr *msg,
 807                                    int flags)
 808 {
 809         if (flags & MSG_PEEK)
 810                 return virtio_transport_seqpacket_do_peek(vsk, msg);
 811         else
 812                 return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags);
 813 }
 814 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue);
 815
 816 int
 817 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk,
 818                                    struct msghdr *msg,
 819                                    size_t len)
 820 {
 821         struct virtio_vsock_sock *vvs = vsk->trans;
 822
 823         spin_lock_bh(&vvs->tx_lock);
 824
 825         if (len > vvs->peer_buf_alloc) {
 826                 spin_unlock_bh(&vvs->tx_lock);
 827                 return -EMSGSIZE;
 828         }
 829
 830         spin_unlock_bh(&vvs->tx_lock);
 831
 832         return virtio_transport_stream_enqueue(vsk, msg, len);
 833 }
 834 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue);
 835
 836 int
 837 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
 838                                struct msghdr *msg,
 839                                size_t len, int flags)
 840 {
 841         return -EOPNOTSUPP;
 842 }
 843 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
 844
 845 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
 846 {
 847         struct virtio_vsock_sock *vvs = vsk->trans;
 848         s64 bytes;
 849
 850         spin_lock_bh(&vvs->rx_lock);
 851         bytes = vvs->rx_bytes;
 852         spin_unlock_bh(&vvs->rx_lock);
 853
 854         return bytes;
 855 }
 856 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
 857
 858 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk)
 859 {
 860         struct virtio_vsock_sock *vvs = vsk->trans;
 861         u32 msg_count;
 862
 863         spin_lock_bh(&vvs->rx_lock);
 864         msg_count = vvs->msg_count;
 865         spin_unlock_bh(&vvs->rx_lock);
 866
 867         return msg_count;
 868 }
 869 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data);
 870
 871 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
 872 {
 873         struct virtio_vsock_sock *vvs = vsk->trans;
 874         s64 bytes;
 875
 876         bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 877         if (bytes < 0)
 878                 bytes = 0;
 879
 880         return bytes;
 881 }
 882
 883 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
 884 {
 885         struct virtio_vsock_sock *vvs = vsk->trans;
 886         s64 bytes;
 887
 888         spin_lock_bh(&vvs->tx_lock);
 889         bytes = virtio_transport_has_space(vsk);
 890         spin_unlock_bh(&vvs->tx_lock);
 891
 892         return bytes;
 893 }
 894 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
 895
 896 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
 897                                     struct vsock_sock *psk)
 898 {
 899         struct virtio_vsock_sock *vvs;
 900
 901         vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
 902         if (!vvs)
 903                 return -ENOMEM;
 904
 905         vsk->trans = vvs;
 906         vvs->vsk = vsk;
 907         if (psk && psk->trans) {
 908                 struct virtio_vsock_sock *ptrans = psk->trans;
 909
 910                 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
 911         }
 912
 913         if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
 914                 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
 915
 916         vvs->buf_alloc = vsk->buffer_size;
 917
 918         spin_lock_init(&vvs->rx_lock);
 919         spin_lock_init(&vvs->tx_lock);
 920         skb_queue_head_init(&vvs->rx_queue);
 921
 922         return 0;
 923 }
 924 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
 925
 926 /* sk_lock held by the caller */
 927 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
 928 {
 929         struct virtio_vsock_sock *vvs = vsk->trans;
 930
 931         if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
 932                 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
 933
 934         vvs->buf_alloc = *val;
 935
 936         virtio_transport_send_credit_update(vsk);
 937 }
 938 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
 939
 940 int
 941 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
 942                                 size_t target,
 943                                 bool *data_ready_now)
 944 {
 945         *data_ready_now = vsock_stream_has_data(vsk) >= target;
 946
 947         return 0;
 948 }
 949 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
 950
 951 int
 952 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
 953                                  size_t target,
 954                                  bool *space_avail_now)
 955 {
 956         s64 free_space;
 957
 958         free_space = vsock_stream_has_space(vsk);
 959         if (free_space > 0)
 960                 *space_avail_now = true;
 961         else if (free_space == 0)
 962                 *space_avail_now = false;
 963
 964         return 0;
 965 }
 966 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
 967
 968 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
 969         size_t target, struct vsock_transport_recv_notify_data *data)
 970 {
 971         return 0;
 972 }
 973 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
 974
 975 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
 976         size_t target, struct vsock_transport_recv_notify_data *data)
 977 {
 978         return 0;
 979 }
 980 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
 981
 982 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
 983         size_t target, struct vsock_transport_recv_notify_data *data)
 984 {
 985         return 0;
 986 }
 987 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
 988
 989 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
 990         size_t target, ssize_t copied, bool data_read,
 991         struct vsock_transport_recv_notify_data *data)
 992 {
 993         return 0;
 994 }
 995 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
 996
 997 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
 998         struct vsock_transport_send_notify_data *data)
 999 {
1000         return 0;
1001 }
1002 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
1003
1004 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
1005         struct vsock_transport_send_notify_data *data)
1006 {
1007         return 0;
1008 }
1009 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
1010
1011 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
1012         struct vsock_transport_send_notify_data *data)
1013 {
1014         return 0;
1015 }
1016 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
1017
1018 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
1019         ssize_t written, struct vsock_transport_send_notify_data *data)
1020 {
1021         return 0;
1022 }
1023 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
1024
1025 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1026 {
1027         return vsk->buffer_size;
1028 }
1029 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
1030
1031 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
1032 {
1033         return true;
1034 }
1035 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
1036
1037 bool virtio_transport_stream_allow(u32 cid, u32 port)
1038 {
1039         return true;
1040 }
1041 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
1042
1043 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
1044                                 struct sockaddr_vm *addr)
1045 {
1046         return -EOPNOTSUPP;
1047 }
1048 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
1049
1050 bool virtio_transport_dgram_allow(u32 cid, u32 port)
1051 {
1052         return false;
1053 }
1054 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
1055
1056 int virtio_transport_connect(struct vsock_sock *vsk)
1057 {
1058         struct virtio_vsock_pkt_info info = {
1059                 .op = VIRTIO_VSOCK_OP_REQUEST,
1060                 .vsk = vsk,
1061         };
1062
1063         return virtio_transport_send_pkt_info(vsk, &info);
1064 }
1065 EXPORT_SYMBOL_GPL(virtio_transport_connect);
1066
1067 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
1068 {
1069         struct virtio_vsock_pkt_info info = {
1070                 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
1071                 .flags = (mode & RCV_SHUTDOWN ?
1072                           VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
1073                          (mode & SEND_SHUTDOWN ?
1074                           VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
1075                 .vsk = vsk,
1076         };
1077
1078         return virtio_transport_send_pkt_info(vsk, &info);
1079 }
1080 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
1081
1082 int
1083 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
1084                                struct sockaddr_vm *remote_addr,
1085                                struct msghdr *msg,
1086                                size_t dgram_len)
1087 {
1088         return -EOPNOTSUPP;
1089 }
1090 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
1091
1092 ssize_t
1093 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
1094                                 struct msghdr *msg,
1095                                 size_t len)
1096 {
1097         struct virtio_vsock_pkt_info info = {
1098                 .op = VIRTIO_VSOCK_OP_RW,
1099                 .msg = msg,
1100                 .pkt_len = len,
1101                 .vsk = vsk,
1102         };
1103
1104         return virtio_transport_send_pkt_info(vsk, &info);
1105 }
1106 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
1107
1108 void virtio_transport_destruct(struct vsock_sock *vsk)
1109 {
1110         struct virtio_vsock_sock *vvs = vsk->trans;
1111
1112         kfree(vvs);
1113         vsk->trans = NULL;
1114 }
1115 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
1116
1117 ssize_t virtio_transport_unsent_bytes(struct vsock_sock *vsk)
1118 {
1119         struct virtio_vsock_sock *vvs = vsk->trans;
1120         size_t ret;
1121
1122         spin_lock_bh(&vvs->tx_lock);
1123         ret = vvs->bytes_unsent;
1124         spin_unlock_bh(&vvs->tx_lock);
1125
1126         return ret;
1127 }
1128 EXPORT_SYMBOL_GPL(virtio_transport_unsent_bytes);
1129
1130 static int virtio_transport_reset(struct vsock_sock *vsk,
1131                                   struct sk_buff *skb)
1132 {
1133         struct virtio_vsock_pkt_info info = {
1134                 .op = VIRTIO_VSOCK_OP_RST,
1135                 .reply = !!skb,
1136                 .vsk = vsk,
1137         };
1138
1139         /* Send RST only if the original pkt is not a RST pkt */
1140         if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST)
1141                 return 0;
1142
1143         return virtio_transport_send_pkt_info(vsk, &info);
1144 }
1145
1146 /* Normally packets are associated with a socket.  There may be no socket if an
1147  * attempt was made to connect to a socket that does not exist.
1148  */
1149 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
1150                                           struct sk_buff *skb)
1151 {
1152         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1153         struct virtio_vsock_pkt_info info = {
1154                 .op = VIRTIO_VSOCK_OP_RST,
1155                 .type = le16_to_cpu(hdr->type),
1156                 .reply = true,
1157         };
1158         struct sk_buff *reply;
1159
1160         /* Send RST only if the original pkt is not a RST pkt */
1161         if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1162                 return 0;
1163
1164         if (!t)
1165                 return -ENOTCONN;
1166
1167         reply = virtio_transport_alloc_skb(&info, 0, false,
1168                                            le64_to_cpu(hdr->dst_cid),
1169                                            le32_to_cpu(hdr->dst_port),
1170                                            le64_to_cpu(hdr->src_cid),
1171                                            le32_to_cpu(hdr->src_port));
1172         if (!reply)
1173                 return -ENOMEM;
1174
1175         return t->send_pkt(reply);
1176 }
1177
1178 /* This function should be called with sk_lock held and SOCK_DONE set */
1179 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
1180 {
1181         struct virtio_vsock_sock *vvs = vsk->trans;
1182
1183         /* We don't need to take rx_lock, as the socket is closing and we are
1184          * removing it.
1185          */
1186         __skb_queue_purge(&vvs->rx_queue);
1187         vsock_remove_sock(vsk);
1188 }
1189
1190 static void virtio_transport_wait_close(struct sock *sk, long timeout)
1191 {
1192         if (timeout) {
1193                 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1194
1195                 add_wait_queue(sk_sleep(sk), &wait);
1196
1197                 do {
1198                         if (sk_wait_event(sk, &timeout,
1199                                           sock_flag(sk, SOCK_DONE), &wait))
1200                                 break;
1201                 } while (!signal_pending(current) && timeout);
1202
1203                 remove_wait_queue(sk_sleep(sk), &wait);
1204         }
1205 }
1206
1207 static void virtio_transport_do_close(struct vsock_sock *vsk,
1208                                       bool cancel_timeout)
1209 {
1210         struct sock *sk = sk_vsock(vsk);
1211
1212         sock_set_flag(sk, SOCK_DONE);
1213         vsk->peer_shutdown = SHUTDOWN_MASK;
1214         if (vsock_stream_has_data(vsk) <= 0)
1215                 sk->sk_state = TCP_CLOSING;
1216         sk->sk_state_change(sk);
1217
1218         if (vsk->close_work_scheduled &&
1219             (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
1220                 vsk->close_work_scheduled = false;
1221
1222                 virtio_transport_remove_sock(vsk);
1223
1224                 /* Release refcnt obtained when we scheduled the timeout */
1225                 sock_put(sk);
1226         }
1227 }
1228
1229 static void virtio_transport_close_timeout(struct work_struct *work)
1230 {
1231         struct vsock_sock *vsk =
1232                 container_of(work, struct vsock_sock, close_work.work);
1233         struct sock *sk = sk_vsock(vsk);
1234
1235         sock_hold(sk);
1236         lock_sock(sk);
1237
1238         if (!sock_flag(sk, SOCK_DONE)) {
1239                 (void)virtio_transport_reset(vsk, NULL);
1240
1241                 virtio_transport_do_close(vsk, false);
1242         }
1243
1244         vsk->close_work_scheduled = false;
1245
1246         release_sock(sk);
1247         sock_put(sk);
1248 }
1249
1250 /* User context, vsk->sk is locked */
1251 static bool virtio_transport_close(struct vsock_sock *vsk)
1252 {
1253         struct sock *sk = &vsk->sk;
1254
1255         if (!(sk->sk_state == TCP_ESTABLISHED ||
1256               sk->sk_state == TCP_CLOSING))
1257                 return true;
1258
1259         /* Already received SHUTDOWN from peer, reply with RST */
1260         if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
1261                 (void)virtio_transport_reset(vsk, NULL);
1262                 return true;
1263         }
1264
1265         if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
1266                 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
1267
1268         if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
1269                 virtio_transport_wait_close(sk, sk->sk_lingertime);
1270
1271         if (sock_flag(sk, SOCK_DONE)) {
1272                 return true;
1273         }
1274
1275         sock_hold(sk);
1276         INIT_DELAYED_WORK(&vsk->close_work,
1277                           virtio_transport_close_timeout);
1278         vsk->close_work_scheduled = true;
1279         schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
1280         return false;
1281 }
1282
1283 void virtio_transport_release(struct vsock_sock *vsk)
1284 {
1285         struct sock *sk = &vsk->sk;
1286         bool remove_sock = true;
1287
1288         if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)
1289                 remove_sock = virtio_transport_close(vsk);
1290
1291         if (remove_sock) {
1292                 sock_set_flag(sk, SOCK_DONE);
1293                 virtio_transport_remove_sock(vsk);
1294         }
1295 }
1296 EXPORT_SYMBOL_GPL(virtio_transport_release);
1297
1298 static int
1299 virtio_transport_recv_connecting(struct sock *sk,
1300                                  struct sk_buff *skb)
1301 {
1302         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1303         struct vsock_sock *vsk = vsock_sk(sk);
1304         int skerr;
1305         int err;
1306
1307         switch (le16_to_cpu(hdr->op)) {
1308         case VIRTIO_VSOCK_OP_RESPONSE:
1309                 sk->sk_state = TCP_ESTABLISHED;
1310                 sk->sk_socket->state = SS_CONNECTED;
1311                 vsock_insert_connected(vsk);
1312                 sk->sk_state_change(sk);
1313                 break;
1314         case VIRTIO_VSOCK_OP_INVALID:
1315                 break;
1316         case VIRTIO_VSOCK_OP_RST:
1317                 skerr = ECONNRESET;
1318                 err = 0;
1319                 goto destroy;
1320         default:
1321                 skerr = EPROTO;
1322                 err = -EINVAL;
1323                 goto destroy;
1324         }
1325         return 0;
1326
1327 destroy:
1328         virtio_transport_reset(vsk, skb);
1329         sk->sk_state = TCP_CLOSE;
1330         sk->sk_err = skerr;
1331         sk_error_report(sk);
1332         return err;
1333 }
1334
1335 static void
1336 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
1337                               struct sk_buff *skb)
1338 {
1339         struct virtio_vsock_sock *vvs = vsk->trans;
1340         bool can_enqueue, free_pkt = false;
1341         struct virtio_vsock_hdr *hdr;
1342         u32 len;
1343
1344         hdr = virtio_vsock_hdr(skb);
1345         len = le32_to_cpu(hdr->len);
1346
1347         spin_lock_bh(&vvs->rx_lock);
1348
1349         can_enqueue = virtio_transport_inc_rx_pkt(vvs, len);
1350         if (!can_enqueue) {
1351                 free_pkt = true;
1352                 goto out;
1353         }
1354
1355         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1356                 vvs->msg_count++;
1357
1358         /* Try to copy small packets into the buffer of last packet queued,
1359          * to avoid wasting memory queueing the entire buffer with a small
1360          * payload.
1361          */
1362         if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
1363                 struct virtio_vsock_hdr *last_hdr;
1364                 struct sk_buff *last_skb;
1365
1366                 last_skb = skb_peek_tail(&vvs->rx_queue);
1367                 last_hdr = virtio_vsock_hdr(last_skb);
1368
1369                 /* If there is space in the last packet queued, we copy the
1370                  * new packet in its buffer. We avoid this if the last packet
1371                  * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is
1372                  * delimiter of SEQPACKET message, so 'pkt' is the first packet
1373                  * of a new message.
1374                  */
1375                 if (skb->len < skb_tailroom(last_skb) &&
1376                     !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) {
1377                         memcpy(skb_put(last_skb, skb->len), skb->data, skb->len);
1378                         free_pkt = true;
1379                         last_hdr->flags |= hdr->flags;
1380                         le32_add_cpu(&last_hdr->len, len);
1381                         goto out;
1382                 }
1383         }
1384
1385         __skb_queue_tail(&vvs->rx_queue, skb);
1386
1387 out:
1388         spin_unlock_bh(&vvs->rx_lock);
1389         if (free_pkt)
1390                 kfree_skb(skb);
1391 }
1392
1393 static int
1394 virtio_transport_recv_connected(struct sock *sk,
1395                                 struct sk_buff *skb)
1396 {
1397         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1398         struct vsock_sock *vsk = vsock_sk(sk);
1399         int err = 0;
1400
1401         switch (le16_to_cpu(hdr->op)) {
1402         case VIRTIO_VSOCK_OP_RW:
1403                 virtio_transport_recv_enqueue(vsk, skb);
1404                 vsock_data_ready(sk);
1405                 return err;
1406         case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
1407                 virtio_transport_send_credit_update(vsk);
1408                 break;
1409         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
1410                 sk->sk_write_space(sk);
1411                 break;
1412         case VIRTIO_VSOCK_OP_SHUTDOWN:
1413                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
1414                         vsk->peer_shutdown |= RCV_SHUTDOWN;
1415                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
1416                         vsk->peer_shutdown |= SEND_SHUTDOWN;
1417                 if (vsk->peer_shutdown == SHUTDOWN_MASK) {
1418                         if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) {
1419                                 (void)virtio_transport_reset(vsk, NULL);
1420                                 virtio_transport_do_close(vsk, true);
1421                         }
1422                         /* Remove this socket anyway because the remote peer sent
1423                          * the shutdown. This way a new connection will succeed
1424                          * if the remote peer uses the same source port,
1425                          * even if the old socket is still unreleased, but now disconnected.
1426                          */
1427                         vsock_remove_sock(vsk);
1428                 }
1429                 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags))
1430                         sk->sk_state_change(sk);
1431                 break;
1432         case VIRTIO_VSOCK_OP_RST:
1433                 virtio_transport_do_close(vsk, true);
1434                 break;
1435         default:
1436                 err = -EINVAL;
1437                 break;
1438         }
1439
1440         kfree_skb(skb);
1441         return err;
1442 }
1443
1444 static void
1445 virtio_transport_recv_disconnecting(struct sock *sk,
1446                                     struct sk_buff *skb)
1447 {
1448         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1449         struct vsock_sock *vsk = vsock_sk(sk);
1450
1451         if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1452                 virtio_transport_do_close(vsk, true);
1453 }
1454
1455 static int
1456 virtio_transport_send_response(struct vsock_sock *vsk,
1457                                struct sk_buff *skb)
1458 {
1459         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1460         struct virtio_vsock_pkt_info info = {
1461                 .op = VIRTIO_VSOCK_OP_RESPONSE,
1462                 .remote_cid = le64_to_cpu(hdr->src_cid),
1463                 .remote_port = le32_to_cpu(hdr->src_port),
1464                 .reply = true,
1465                 .vsk = vsk,
1466         };
1467
1468         return virtio_transport_send_pkt_info(vsk, &info);
1469 }
1470
1471 static bool virtio_transport_space_update(struct sock *sk,
1472                                           struct sk_buff *skb)
1473 {
1474         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1475         struct vsock_sock *vsk = vsock_sk(sk);
1476         struct virtio_vsock_sock *vvs = vsk->trans;
1477         bool space_available;
1478
1479         /* Listener sockets are not associated with any transport, so we are
1480          * not able to take the state to see if there is space available in the
1481          * remote peer, but since they are only used to receive requests, we
1482          * can assume that there is always space available in the other peer.
1483          */
1484         if (!vvs)
1485                 return true;
1486
1487         /* buf_alloc and fwd_cnt is always included in the hdr */
1488         spin_lock_bh(&vvs->tx_lock);
1489         vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc);
1490         vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt);
1491         space_available = virtio_transport_has_space(vsk);
1492         spin_unlock_bh(&vvs->tx_lock);
1493         return space_available;
1494 }
1495
1496 /* Handle server socket */
1497 static int
1498 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb,
1499                              struct virtio_transport *t)
1500 {
1501         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1502         struct vsock_sock *vsk = vsock_sk(sk);
1503         struct vsock_sock *vchild;
1504         struct sock *child;
1505         int ret;
1506
1507         if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) {
1508                 virtio_transport_reset_no_sock(t, skb);
1509                 return -EINVAL;
1510         }
1511
1512         if (sk_acceptq_is_full(sk)) {
1513                 virtio_transport_reset_no_sock(t, skb);
1514                 return -ENOMEM;
1515         }
1516
1517         /* __vsock_release() might have already flushed accept_queue.
1518          * Subsequent enqueues would lead to a memory leak.
1519          */
1520         if (sk->sk_shutdown == SHUTDOWN_MASK) {
1521                 virtio_transport_reset_no_sock(t, skb);
1522                 return -ESHUTDOWN;
1523         }
1524
1525         child = vsock_create_connected(sk);
1526         if (!child) {
1527                 virtio_transport_reset_no_sock(t, skb);
1528                 return -ENOMEM;
1529         }
1530
1531         sk_acceptq_added(sk);
1532
1533         lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1534
1535         child->sk_state = TCP_ESTABLISHED;
1536
1537         vchild = vsock_sk(child);
1538         vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid),
1539                         le32_to_cpu(hdr->dst_port));
1540         vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid),
1541                         le32_to_cpu(hdr->src_port));
1542
1543         ret = vsock_assign_transport(vchild, vsk);
1544         /* Transport assigned (looking at remote_addr) must be the same
1545          * where we received the request.
1546          */
1547         if (ret || vchild->transport != &t->transport) {
1548                 release_sock(child);
1549                 virtio_transport_reset_no_sock(t, skb);
1550                 sock_put(child);
1551                 return ret;
1552         }
1553
1554         if (virtio_transport_space_update(child, skb))
1555                 child->sk_write_space(child);
1556
1557         vsock_insert_connected(vchild);
1558         vsock_enqueue_accept(sk, child);
1559         virtio_transport_send_response(vchild, skb);
1560
1561         release_sock(child);
1562
1563         sk->sk_data_ready(sk);
1564         return 0;
1565 }
1566
1567 static bool virtio_transport_valid_type(u16 type)
1568 {
1569         return (type == VIRTIO_VSOCK_TYPE_STREAM) ||
1570                (type == VIRTIO_VSOCK_TYPE_SEQPACKET);
1571 }
1572
1573 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1574  * lock.
1575  */
1576 void virtio_transport_recv_pkt(struct virtio_transport *t,
1577                                struct sk_buff *skb)
1578 {
1579         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1580         struct sockaddr_vm src, dst;
1581         struct vsock_sock *vsk;
1582         struct sock *sk;
1583         bool space_available;
1584
1585         vsock_addr_init(&src, le64_to_cpu(hdr->src_cid),
1586                         le32_to_cpu(hdr->src_port));
1587         vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid),
1588                         le32_to_cpu(hdr->dst_port));
1589
1590         trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1591                                         dst.svm_cid, dst.svm_port,
1592                                         le32_to_cpu(hdr->len),
1593                                         le16_to_cpu(hdr->type),
1594                                         le16_to_cpu(hdr->op),
1595                                         le32_to_cpu(hdr->flags),
1596                                         le32_to_cpu(hdr->buf_alloc),
1597                                         le32_to_cpu(hdr->fwd_cnt));
1598
1599         if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) {
1600                 (void)virtio_transport_reset_no_sock(t, skb);
1601                 goto free_pkt;
1602         }
1603
1604         /* The socket must be in connected or bound table
1605          * otherwise send reset back
1606          */
1607         sk = vsock_find_connected_socket(&src, &dst);
1608         if (!sk) {
1609                 sk = vsock_find_bound_socket(&dst);
1610                 if (!sk) {
1611                         (void)virtio_transport_reset_no_sock(t, skb);
1612                         goto free_pkt;
1613                 }
1614         }
1615
1616         if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) {
1617                 (void)virtio_transport_reset_no_sock(t, skb);
1618                 sock_put(sk);
1619                 goto free_pkt;
1620         }
1621
1622         if (!skb_set_owner_sk_safe(skb, sk)) {
1623                 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
1624                 goto free_pkt;
1625         }
1626
1627         vsk = vsock_sk(sk);
1628
1629         lock_sock(sk);
1630
1631         /* Check if sk has been closed or assigned to another transport before
1632          * lock_sock (note: listener sockets are not assigned to any transport)
1633          */
1634         if (sock_flag(sk, SOCK_DONE) ||
1635             (sk->sk_state != TCP_LISTEN && vsk->transport != &t->transport)) {
1636                 (void)virtio_transport_reset_no_sock(t, skb);
1637                 release_sock(sk);
1638                 sock_put(sk);
1639                 goto free_pkt;
1640         }
1641
1642         space_available = virtio_transport_space_update(sk, skb);
1643
1644         /* Update CID in case it has changed after a transport reset event */
1645         if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1646                 vsk->local_addr.svm_cid = dst.svm_cid;
1647
1648         if (space_available)
1649                 sk->sk_write_space(sk);
1650
1651         switch (sk->sk_state) {
1652         case TCP_LISTEN:
1653                 virtio_transport_recv_listen(sk, skb, t);
1654                 kfree_skb(skb);
1655                 break;
1656         case TCP_SYN_SENT:
1657                 virtio_transport_recv_connecting(sk, skb);
1658                 kfree_skb(skb);
1659                 break;
1660         case TCP_ESTABLISHED:
1661                 virtio_transport_recv_connected(sk, skb);
1662                 break;
1663         case TCP_CLOSING:
1664                 virtio_transport_recv_disconnecting(sk, skb);
1665                 kfree_skb(skb);
1666                 break;
1667         default:
1668                 (void)virtio_transport_reset_no_sock(t, skb);
1669                 kfree_skb(skb);
1670                 break;
1671         }
1672
1673         release_sock(sk);
1674
1675         /* Release refcnt obtained when we fetched this socket out of the
1676          * bound or connected list.
1677          */
1678         sock_put(sk);
1679         return;
1680
1681 free_pkt:
1682         kfree_skb(skb);
1683 }
1684 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1685
1686 /* Remove skbs found in a queue that have a vsk that matches.
1687  *
1688  * Each skb is freed.
1689  *
1690  * Returns the count of skbs that were reply packets.
1691  */
1692 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue)
1693 {
1694         struct sk_buff_head freeme;
1695         struct sk_buff *skb, *tmp;
1696         int cnt = 0;
1697
1698         skb_queue_head_init(&freeme);
1699
1700         spin_lock_bh(&queue->lock);
1701         skb_queue_walk_safe(queue, skb, tmp) {
1702                 if (vsock_sk(skb->sk) != vsk)
1703                         continue;
1704
1705                 __skb_unlink(skb, queue);
1706                 __skb_queue_tail(&freeme, skb);
1707
1708                 if (virtio_vsock_skb_reply(skb))
1709                         cnt++;
1710         }
1711         spin_unlock_bh(&queue->lock);
1712
1713         __skb_queue_purge(&freeme);
1714
1715         return cnt;
1716 }
1717 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs);
1718
1719 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor)
1720 {
1721         struct virtio_vsock_sock *vvs = vsk->trans;
1722         struct sock *sk = sk_vsock(vsk);
1723         struct virtio_vsock_hdr *hdr;
1724         struct sk_buff *skb;
1725         int off = 0;
1726         int err;
1727
1728         spin_lock_bh(&vvs->rx_lock);
1729         /* Use __skb_recv_datagram() for race-free handling of the receive. It
1730          * works for types other than dgrams.
1731          */
1732         skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err);
1733         if (!skb) {
1734                 spin_unlock_bh(&vvs->rx_lock);
1735                 return err;
1736         }
1737
1738         hdr = virtio_vsock_hdr(skb);
1739         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1740                 vvs->msg_count--;
1741
1742         virtio_transport_dec_rx_pkt(vvs, le32_to_cpu(hdr->len));
1743         spin_unlock_bh(&vvs->rx_lock);
1744
1745         virtio_transport_send_credit_update(vsk);
1746
1747         return recv_actor(sk, skb);
1748 }
1749 EXPORT_SYMBOL_GPL(virtio_transport_read_skb);
1750
1751 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val)
1752 {
1753         struct virtio_vsock_sock *vvs = vsk->trans;
1754         bool send_update;
1755
1756         spin_lock_bh(&vvs->rx_lock);
1757
1758         /* If number of available bytes is less than new SO_RCVLOWAT value,
1759          * kick sender to send more data, because sender may sleep in its
1760          * 'send()' syscall waiting for enough space at our side. Also
1761          * don't send credit update when peer already knows actual value -
1762          * such transmission will be useless.
1763          */
1764         send_update = (vvs->rx_bytes < val) &&
1765                       (vvs->fwd_cnt != vvs->last_fwd_cnt);
1766
1767         spin_unlock_bh(&vvs->rx_lock);
1768
1769         if (send_update) {
1770                 int err;
1771
1772                 err = virtio_transport_send_credit_update(vsk);
1773                 if (err < 0)
1774                         return err;
1775         }
1776
1777         return 0;
1778 }
1779 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat);
1780
1781 MODULE_LICENSE("GPL v2");
1782 MODULE_AUTHOR("Asias He");
1783 MODULE_DESCRIPTION("common code for virtio vsock");