vmw_vsock/virtio_transport_common.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * common code for virtio vsock
   4  *
   5  * Copyright (C) 2013-2015 Red Hat, Inc.
   6  * Author: Asias He <asias@redhat.com>
   7  *         Stefan Hajnoczi <stefanha@redhat.com>
   8  */
   9 #include <linux/spinlock.h>
  10 #include <linux/module.h>
  11 #include <linux/sched/signal.h>
  12 #include <linux/ctype.h>
  13 #include <linux/list.h>
  14 #include <linux/virtio_vsock.h>
  15 #include <uapi/linux/vsockmon.h>
  16
  17 #include <net/sock.h>
  18 #include <net/af_vsock.h>
  19
  20 #define CREATE_TRACE_POINTS
  21 #include <trace/events/vsock_virtio_transport_common.h>
  22
  23 /* How long to wait for graceful shutdown of a connection */
  24 #define VSOCK_CLOSE_TIMEOUT (8 * HZ)
  25
  26 /* Threshold for detecting small packets to copy */
  27 #define GOOD_COPY_LEN  128
  28
  29 static const struct virtio_transport *
  30 virtio_transport_get_ops(struct vsock_sock *vsk)
  31 {
  32         const struct vsock_transport *t = vsock_core_get_transport(vsk);
  33
  34         if (WARN_ON(!t))
  35                 return NULL;
  36
  37         return container_of(t, struct virtio_transport, transport);
  38 }
  39
  40 /* Returns a new packet on success, otherwise returns NULL.
  41  *
  42  * If NULL is returned, errp is set to a negative errno.
  43  */
  44 static struct sk_buff *
  45 virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info,
  46                            size_t len,
  47                            u32 src_cid,
  48                            u32 src_port,
  49                            u32 dst_cid,
  50                            u32 dst_port)
  51 {
  52         const size_t skb_len = VIRTIO_VSOCK_SKB_HEADROOM + len;
  53         struct virtio_vsock_hdr *hdr;
  54         struct sk_buff *skb;
  55         void *payload;
  56         int err;
  57
  58         skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL);
  59         if (!skb)
  60                 return NULL;
  61
  62         hdr = virtio_vsock_hdr(skb);
  63         hdr->type       = cpu_to_le16(info->type);
  64         hdr->op         = cpu_to_le16(info->op);
  65         hdr->src_cid    = cpu_to_le64(src_cid);
  66         hdr->dst_cid    = cpu_to_le64(dst_cid);
  67         hdr->src_port   = cpu_to_le32(src_port);
  68         hdr->dst_port   = cpu_to_le32(dst_port);
  69         hdr->flags      = cpu_to_le32(info->flags);
  70         hdr->len        = cpu_to_le32(len);
  71         hdr->buf_alloc  = cpu_to_le32(0);
  72         hdr->fwd_cnt    = cpu_to_le32(0);
  73
  74         if (info->msg && len > 0) {
  75                 payload = skb_put(skb, len);
  76                 err = memcpy_from_msg(payload, info->msg, len);
  77                 if (err)
  78                         goto out;
  79
  80                 if (msg_data_left(info->msg) == 0 &&
  81                     info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) {
  82                         hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
  83
  84                         if (info->msg->msg_flags & MSG_EOR)
  85                                 hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
  86                 }
  87         }
  88
  89         if (info->reply)
  90                 virtio_vsock_skb_set_reply(skb);
  91
  92         trace_virtio_transport_alloc_pkt(src_cid, src_port,
  93                                          dst_cid, dst_port,
  94                                          len,
  95                                          info->type,
  96                                          info->op,
  97                                          info->flags);
  98
  99         if (info->vsk && !skb_set_owner_sk_safe(skb, sk_vsock(info->vsk))) {
 100                 WARN_ONCE(1, "failed to allocate skb on vsock socket with sk_refcnt == 0\n");
 101                 goto out;
 102         }
 103
 104         return skb;
 105
 106 out:
 107         kfree_skb(skb);
 108         return NULL;
 109 }
 110
 111 /* Packet capture */
 112 static struct sk_buff *virtio_transport_build_skb(void *opaque)
 113 {
 114         struct virtio_vsock_hdr *pkt_hdr;
 115         struct sk_buff *pkt = opaque;
 116         struct af_vsockmon_hdr *hdr;
 117         struct sk_buff *skb;
 118         size_t payload_len;
 119         void *payload_buf;
 120
 121         /* A packet could be split to fit the RX buffer, so we can retrieve
 122          * the payload length from the header and the buffer pointer taking
 123          * care of the offset in the original packet.
 124          */
 125         pkt_hdr = virtio_vsock_hdr(pkt);
 126         payload_len = pkt->len;
 127         payload_buf = pkt->data;
 128
 129         skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len,
 130                         GFP_ATOMIC);
 131         if (!skb)
 132                 return NULL;
 133
 134         hdr = skb_put(skb, sizeof(*hdr));
 135
 136         /* pkt->hdr is little-endian so no need to byteswap here */
 137         hdr->src_cid = pkt_hdr->src_cid;
 138         hdr->src_port = pkt_hdr->src_port;
 139         hdr->dst_cid = pkt_hdr->dst_cid;
 140         hdr->dst_port = pkt_hdr->dst_port;
 141
 142         hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
 143         hdr->len = cpu_to_le16(sizeof(*pkt_hdr));
 144         memset(hdr->reserved, 0, sizeof(hdr->reserved));
 145
 146         switch (le16_to_cpu(pkt_hdr->op)) {
 147         case VIRTIO_VSOCK_OP_REQUEST:
 148         case VIRTIO_VSOCK_OP_RESPONSE:
 149                 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
 150                 break;
 151         case VIRTIO_VSOCK_OP_RST:
 152         case VIRTIO_VSOCK_OP_SHUTDOWN:
 153                 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
 154                 break;
 155         case VIRTIO_VSOCK_OP_RW:
 156                 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
 157                 break;
 158         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
 159         case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
 160                 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
 161                 break;
 162         default:
 163                 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
 164                 break;
 165         }
 166
 167         skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr));
 168
 169         if (payload_len) {
 170                 skb_put_data(skb, payload_buf, payload_len);
 171         }
 172
 173         return skb;
 174 }
 175
 176 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb)
 177 {
 178         if (virtio_vsock_skb_tap_delivered(skb))
 179                 return;
 180
 181         vsock_deliver_tap(virtio_transport_build_skb, skb);
 182         virtio_vsock_skb_set_tap_delivered(skb);
 183 }
 184 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
 185
 186 static u16 virtio_transport_get_type(struct sock *sk)
 187 {
 188         if (sk->sk_type == SOCK_STREAM)
 189                 return VIRTIO_VSOCK_TYPE_STREAM;
 190         else
 191                 return VIRTIO_VSOCK_TYPE_SEQPACKET;
 192 }
 193
 194 /* This function can only be used on connecting/connected sockets,
 195  * since a socket assigned to a transport is required.
 196  *
 197  * Do not use on listener sockets!
 198  */
 199 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
 200                                           struct virtio_vsock_pkt_info *info)
 201 {
 202         u32 src_cid, src_port, dst_cid, dst_port;
 203         const struct virtio_transport *t_ops;
 204         struct virtio_vsock_sock *vvs;
 205         u32 pkt_len = info->pkt_len;
 206         struct sk_buff *skb;
 207
 208         info->type = virtio_transport_get_type(sk_vsock(vsk));
 209
 210         t_ops = virtio_transport_get_ops(vsk);
 211         if (unlikely(!t_ops))
 212                 return -EFAULT;
 213
 214         src_cid = t_ops->transport.get_local_cid();
 215         src_port = vsk->local_addr.svm_port;
 216         if (!info->remote_cid) {
 217                 dst_cid = vsk->remote_addr.svm_cid;
 218                 dst_port = vsk->remote_addr.svm_port;
 219         } else {
 220                 dst_cid = info->remote_cid;
 221                 dst_port = info->remote_port;
 222         }
 223
 224         vvs = vsk->trans;
 225
 226         /* we can send less than pkt_len bytes */
 227         if (pkt_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
 228                 pkt_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
 229
 230         /* virtio_transport_get_credit might return less than pkt_len credit */
 231         pkt_len = virtio_transport_get_credit(vvs, pkt_len);
 232
 233         /* Do not send zero length OP_RW pkt */
 234         if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
 235                 return pkt_len;
 236
 237         skb = virtio_transport_alloc_skb(info, pkt_len,
 238                                          src_cid, src_port,
 239                                          dst_cid, dst_port);
 240         if (!skb) {
 241                 virtio_transport_put_credit(vvs, pkt_len);
 242                 return -ENOMEM;
 243         }
 244
 245         virtio_transport_inc_tx_pkt(vvs, skb);
 246
 247         return t_ops->send_pkt(skb);
 248 }
 249
 250 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
 251                                         u32 len)
 252 {
 253         if (vvs->rx_bytes + len > vvs->buf_alloc)
 254                 return false;
 255
 256         vvs->rx_bytes += len;
 257         return true;
 258 }
 259
 260 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
 261                                         u32 len)
 262 {
 263         vvs->rx_bytes -= len;
 264         vvs->fwd_cnt += len;
 265 }
 266
 267 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb)
 268 {
 269         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
 270
 271         spin_lock_bh(&vvs->rx_lock);
 272         vvs->last_fwd_cnt = vvs->fwd_cnt;
 273         hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
 274         hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc);
 275         spin_unlock_bh(&vvs->rx_lock);
 276 }
 277 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
 278
 279 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
 280 {
 281         u32 ret;
 282
 283         spin_lock_bh(&vvs->tx_lock);
 284         ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 285         if (ret > credit)
 286                 ret = credit;
 287         vvs->tx_cnt += ret;
 288         spin_unlock_bh(&vvs->tx_lock);
 289
 290         return ret;
 291 }
 292 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
 293
 294 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
 295 {
 296         spin_lock_bh(&vvs->tx_lock);
 297         vvs->tx_cnt -= credit;
 298         spin_unlock_bh(&vvs->tx_lock);
 299 }
 300 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
 301
 302 static int virtio_transport_send_credit_update(struct vsock_sock *vsk)
 303 {
 304         struct virtio_vsock_pkt_info info = {
 305                 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
 306                 .vsk = vsk,
 307         };
 308
 309         return virtio_transport_send_pkt_info(vsk, &info);
 310 }
 311
 312 static ssize_t
 313 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
 314                                 struct msghdr *msg,
 315                                 size_t len)
 316 {
 317         struct virtio_vsock_sock *vvs = vsk->trans;
 318         size_t bytes, total = 0, off;
 319         struct sk_buff *skb, *tmp;
 320         int err = -EFAULT;
 321
 322         spin_lock_bh(&vvs->rx_lock);
 323
 324         skb_queue_walk_safe(&vvs->rx_queue, skb,  tmp) {
 325                 off = 0;
 326
 327                 if (total == len)
 328                         break;
 329
 330                 while (total < len && off < skb->len) {
 331                         bytes = len - total;
 332                         if (bytes > skb->len - off)
 333                                 bytes = skb->len - off;
 334
 335                         /* sk_lock is held by caller so no one else can dequeue.
 336                          * Unlock rx_lock since memcpy_to_msg() may sleep.
 337                          */
 338                         spin_unlock_bh(&vvs->rx_lock);
 339
 340                         err = memcpy_to_msg(msg, skb->data + off, bytes);
 341                         if (err)
 342                                 goto out;
 343
 344                         spin_lock_bh(&vvs->rx_lock);
 345
 346                         total += bytes;
 347                         off += bytes;
 348                 }
 349         }
 350
 351         spin_unlock_bh(&vvs->rx_lock);
 352
 353         return total;
 354
 355 out:
 356         if (total)
 357                 err = total;
 358         return err;
 359 }
 360
 361 static ssize_t
 362 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
 363                                    struct msghdr *msg,
 364                                    size_t len)
 365 {
 366         struct virtio_vsock_sock *vvs = vsk->trans;
 367         size_t bytes, total = 0;
 368         struct sk_buff *skb;
 369         u32 fwd_cnt_delta;
 370         bool low_rx_bytes;
 371         int err = -EFAULT;
 372         u32 free_space;
 373
 374         spin_lock_bh(&vvs->rx_lock);
 375         while (total < len && !skb_queue_empty(&vvs->rx_queue)) {
 376                 skb = skb_peek(&vvs->rx_queue);
 377
 378                 bytes = len - total;
 379                 if (bytes > skb->len)
 380                         bytes = skb->len;
 381
 382                 /* sk_lock is held by caller so no one else can dequeue.
 383                  * Unlock rx_lock since memcpy_to_msg() may sleep.
 384                  */
 385                 spin_unlock_bh(&vvs->rx_lock);
 386
 387                 err = memcpy_to_msg(msg, skb->data, bytes);
 388                 if (err)
 389                         goto out;
 390
 391                 spin_lock_bh(&vvs->rx_lock);
 392
 393                 total += bytes;
 394                 skb_pull(skb, bytes);
 395
 396                 if (skb->len == 0) {
 397                         u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
 398
 399                         virtio_transport_dec_rx_pkt(vvs, pkt_len);
 400                         __skb_unlink(skb, &vvs->rx_queue);
 401                         consume_skb(skb);
 402                 }
 403         }
 404
 405         fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt;
 406         free_space = vvs->buf_alloc - fwd_cnt_delta;
 407         low_rx_bytes = (vvs->rx_bytes <
 408                         sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX));
 409
 410         spin_unlock_bh(&vvs->rx_lock);
 411
 412         /* To reduce the number of credit update messages,
 413          * don't update credits as long as lots of space is available.
 414          * Note: the limit chosen here is arbitrary. Setting the limit
 415          * too high causes extra messages. Too low causes transmitter
 416          * stalls. As stalls are in theory more expensive than extra
 417          * messages, we set the limit to a high value. TODO: experiment
 418          * with different values. Also send credit update message when
 419          * number of bytes in rx queue is not enough to wake up reader.
 420          */
 421         if (fwd_cnt_delta &&
 422             (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes))
 423                 virtio_transport_send_credit_update(vsk);
 424
 425         return total;
 426
 427 out:
 428         if (total)
 429                 err = total;
 430         return err;
 431 }
 432
 433 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
 434                                                  struct msghdr *msg,
 435                                                  int flags)
 436 {
 437         struct virtio_vsock_sock *vvs = vsk->trans;
 438         int dequeued_len = 0;
 439         size_t user_buf_len = msg_data_left(msg);
 440         bool msg_ready = false;
 441         struct sk_buff *skb;
 442
 443         spin_lock_bh(&vvs->rx_lock);
 444
 445         if (vvs->msg_count == 0) {
 446                 spin_unlock_bh(&vvs->rx_lock);
 447                 return 0;
 448         }
 449
 450         while (!msg_ready) {
 451                 struct virtio_vsock_hdr *hdr;
 452                 size_t pkt_len;
 453
 454                 skb = __skb_dequeue(&vvs->rx_queue);
 455                 if (!skb)
 456                         break;
 457                 hdr = virtio_vsock_hdr(skb);
 458                 pkt_len = (size_t)le32_to_cpu(hdr->len);
 459
 460                 if (dequeued_len >= 0) {
 461                         size_t bytes_to_copy;
 462
 463                         bytes_to_copy = min(user_buf_len, pkt_len);
 464
 465                         if (bytes_to_copy) {
 466                                 int err;
 467
 468                                 /* sk_lock is held by caller so no one else can dequeue.
 469                                  * Unlock rx_lock since memcpy_to_msg() may sleep.
 470                                  */
 471                                 spin_unlock_bh(&vvs->rx_lock);
 472
 473                                 err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
 474                                 if (err) {
 475                                         /* Copy of message failed. Rest of
 476                                          * fragments will be freed without copy.
 477                                          */
 478                                         dequeued_len = err;
 479                                 } else {
 480                                         user_buf_len -= bytes_to_copy;
 481                                 }
 482
 483                                 spin_lock_bh(&vvs->rx_lock);
 484                         }
 485
 486                         if (dequeued_len >= 0)
 487                                 dequeued_len += pkt_len;
 488                 }
 489
 490                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
 491                         msg_ready = true;
 492                         vvs->msg_count--;
 493
 494                         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
 495                                 msg->msg_flags |= MSG_EOR;
 496                 }
 497
 498                 virtio_transport_dec_rx_pkt(vvs, pkt_len);
 499                 kfree_skb(skb);
 500         }
 501
 502         spin_unlock_bh(&vvs->rx_lock);
 503
 504         virtio_transport_send_credit_update(vsk);
 505
 506         return dequeued_len;
 507 }
 508
 509 ssize_t
 510 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
 511                                 struct msghdr *msg,
 512                                 size_t len, int flags)
 513 {
 514         if (flags & MSG_PEEK)
 515                 return virtio_transport_stream_do_peek(vsk, msg, len);
 516         else
 517                 return virtio_transport_stream_do_dequeue(vsk, msg, len);
 518 }
 519 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
 520
 521 ssize_t
 522 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk,
 523                                    struct msghdr *msg,
 524                                    int flags)
 525 {
 526         if (flags & MSG_PEEK)
 527                 return -EOPNOTSUPP;
 528
 529         return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags);
 530 }
 531 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue);
 532
 533 int
 534 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk,
 535                                    struct msghdr *msg,
 536                                    size_t len)
 537 {
 538         struct virtio_vsock_sock *vvs = vsk->trans;
 539
 540         spin_lock_bh(&vvs->tx_lock);
 541
 542         if (len > vvs->peer_buf_alloc) {
 543                 spin_unlock_bh(&vvs->tx_lock);
 544                 return -EMSGSIZE;
 545         }
 546
 547         spin_unlock_bh(&vvs->tx_lock);
 548
 549         return virtio_transport_stream_enqueue(vsk, msg, len);
 550 }
 551 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue);
 552
 553 int
 554 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
 555                                struct msghdr *msg,
 556                                size_t len, int flags)
 557 {
 558         return -EOPNOTSUPP;
 559 }
 560 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
 561
 562 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
 563 {
 564         struct virtio_vsock_sock *vvs = vsk->trans;
 565         s64 bytes;
 566
 567         spin_lock_bh(&vvs->rx_lock);
 568         bytes = vvs->rx_bytes;
 569         spin_unlock_bh(&vvs->rx_lock);
 570
 571         return bytes;
 572 }
 573 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
 574
 575 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk)
 576 {
 577         struct virtio_vsock_sock *vvs = vsk->trans;
 578         u32 msg_count;
 579
 580         spin_lock_bh(&vvs->rx_lock);
 581         msg_count = vvs->msg_count;
 582         spin_unlock_bh(&vvs->rx_lock);
 583
 584         return msg_count;
 585 }
 586 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data);
 587
 588 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
 589 {
 590         struct virtio_vsock_sock *vvs = vsk->trans;
 591         s64 bytes;
 592
 593         bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 594         if (bytes < 0)
 595                 bytes = 0;
 596
 597         return bytes;
 598 }
 599
 600 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
 601 {
 602         struct virtio_vsock_sock *vvs = vsk->trans;
 603         s64 bytes;
 604
 605         spin_lock_bh(&vvs->tx_lock);
 606         bytes = virtio_transport_has_space(vsk);
 607         spin_unlock_bh(&vvs->tx_lock);
 608
 609         return bytes;
 610 }
 611 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
 612
 613 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
 614                                     struct vsock_sock *psk)
 615 {
 616         struct virtio_vsock_sock *vvs;
 617
 618         vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
 619         if (!vvs)
 620                 return -ENOMEM;
 621
 622         vsk->trans = vvs;
 623         vvs->vsk = vsk;
 624         if (psk && psk->trans) {
 625                 struct virtio_vsock_sock *ptrans = psk->trans;
 626
 627                 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
 628         }
 629
 630         if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
 631                 vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
 632
 633         vvs->buf_alloc = vsk->buffer_size;
 634
 635         spin_lock_init(&vvs->rx_lock);
 636         spin_lock_init(&vvs->tx_lock);
 637         skb_queue_head_init(&vvs->rx_queue);
 638
 639         return 0;
 640 }
 641 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
 642
 643 /* sk_lock held by the caller */
 644 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
 645 {
 646         struct virtio_vsock_sock *vvs = vsk->trans;
 647
 648         if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
 649                 *val = VIRTIO_VSOCK_MAX_BUF_SIZE;
 650
 651         vvs->buf_alloc = *val;
 652
 653         virtio_transport_send_credit_update(vsk);
 654 }
 655 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
 656
 657 int
 658 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
 659                                 size_t target,
 660                                 bool *data_ready_now)
 661 {
 662         *data_ready_now = vsock_stream_has_data(vsk) >= target;
 663
 664         return 0;
 665 }
 666 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
 667
 668 int
 669 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
 670                                  size_t target,
 671                                  bool *space_avail_now)
 672 {
 673         s64 free_space;
 674
 675         free_space = vsock_stream_has_space(vsk);
 676         if (free_space > 0)
 677                 *space_avail_now = true;
 678         else if (free_space == 0)
 679                 *space_avail_now = false;
 680
 681         return 0;
 682 }
 683 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
 684
 685 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
 686         size_t target, struct vsock_transport_recv_notify_data *data)
 687 {
 688         return 0;
 689 }
 690 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
 691
 692 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
 693         size_t target, struct vsock_transport_recv_notify_data *data)
 694 {
 695         return 0;
 696 }
 697 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
 698
 699 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
 700         size_t target, struct vsock_transport_recv_notify_data *data)
 701 {
 702         return 0;
 703 }
 704 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
 705
 706 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
 707         size_t target, ssize_t copied, bool data_read,
 708         struct vsock_transport_recv_notify_data *data)
 709 {
 710         return 0;
 711 }
 712 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
 713
 714 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
 715         struct vsock_transport_send_notify_data *data)
 716 {
 717         return 0;
 718 }
 719 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
 720
 721 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
 722         struct vsock_transport_send_notify_data *data)
 723 {
 724         return 0;
 725 }
 726 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
 727
 728 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
 729         struct vsock_transport_send_notify_data *data)
 730 {
 731         return 0;
 732 }
 733 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
 734
 735 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
 736         ssize_t written, struct vsock_transport_send_notify_data *data)
 737 {
 738         return 0;
 739 }
 740 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
 741
 742 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
 743 {
 744         return vsk->buffer_size;
 745 }
 746 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
 747
 748 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
 749 {
 750         return true;
 751 }
 752 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
 753
 754 bool virtio_transport_stream_allow(u32 cid, u32 port)
 755 {
 756         return true;
 757 }
 758 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
 759
 760 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
 761                                 struct sockaddr_vm *addr)
 762 {
 763         return -EOPNOTSUPP;
 764 }
 765 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
 766
 767 bool virtio_transport_dgram_allow(u32 cid, u32 port)
 768 {
 769         return false;
 770 }
 771 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
 772
 773 int virtio_transport_connect(struct vsock_sock *vsk)
 774 {
 775         struct virtio_vsock_pkt_info info = {
 776                 .op = VIRTIO_VSOCK_OP_REQUEST,
 777                 .vsk = vsk,
 778         };
 779
 780         return virtio_transport_send_pkt_info(vsk, &info);
 781 }
 782 EXPORT_SYMBOL_GPL(virtio_transport_connect);
 783
 784 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
 785 {
 786         struct virtio_vsock_pkt_info info = {
 787                 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
 788                 .flags = (mode & RCV_SHUTDOWN ?
 789                           VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
 790                          (mode & SEND_SHUTDOWN ?
 791                           VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
 792                 .vsk = vsk,
 793         };
 794
 795         return virtio_transport_send_pkt_info(vsk, &info);
 796 }
 797 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
 798
 799 int
 800 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
 801                                struct sockaddr_vm *remote_addr,
 802                                struct msghdr *msg,
 803                                size_t dgram_len)
 804 {
 805         return -EOPNOTSUPP;
 806 }
 807 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
 808
 809 ssize_t
 810 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
 811                                 struct msghdr *msg,
 812                                 size_t len)
 813 {
 814         struct virtio_vsock_pkt_info info = {
 815                 .op = VIRTIO_VSOCK_OP_RW,
 816                 .msg = msg,
 817                 .pkt_len = len,
 818                 .vsk = vsk,
 819         };
 820
 821         return virtio_transport_send_pkt_info(vsk, &info);
 822 }
 823 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
 824
 825 void virtio_transport_destruct(struct vsock_sock *vsk)
 826 {
 827         struct virtio_vsock_sock *vvs = vsk->trans;
 828
 829         kfree(vvs);
 830 }
 831 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
 832
 833 static int virtio_transport_reset(struct vsock_sock *vsk,
 834                                   struct sk_buff *skb)
 835 {
 836         struct virtio_vsock_pkt_info info = {
 837                 .op = VIRTIO_VSOCK_OP_RST,
 838                 .reply = !!skb,
 839                 .vsk = vsk,
 840         };
 841
 842         /* Send RST only if the original pkt is not a RST pkt */
 843         if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST)
 844                 return 0;
 845
 846         return virtio_transport_send_pkt_info(vsk, &info);
 847 }
 848
 849 /* Normally packets are associated with a socket.  There may be no socket if an
 850  * attempt was made to connect to a socket that does not exist.
 851  */
 852 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
 853                                           struct sk_buff *skb)
 854 {
 855         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
 856         struct virtio_vsock_pkt_info info = {
 857                 .op = VIRTIO_VSOCK_OP_RST,
 858                 .type = le16_to_cpu(hdr->type),
 859                 .reply = true,
 860         };
 861         struct sk_buff *reply;
 862
 863         /* Send RST only if the original pkt is not a RST pkt */
 864         if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
 865                 return 0;
 866
 867         reply = virtio_transport_alloc_skb(&info, 0,
 868                                            le64_to_cpu(hdr->dst_cid),
 869                                            le32_to_cpu(hdr->dst_port),
 870                                            le64_to_cpu(hdr->src_cid),
 871                                            le32_to_cpu(hdr->src_port));
 872         if (!reply)
 873                 return -ENOMEM;
 874
 875         if (!t) {
 876                 kfree_skb(reply);
 877                 return -ENOTCONN;
 878         }
 879
 880         return t->send_pkt(reply);
 881 }
 882
 883 /* This function should be called with sk_lock held and SOCK_DONE set */
 884 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
 885 {
 886         struct virtio_vsock_sock *vvs = vsk->trans;
 887
 888         /* We don't need to take rx_lock, as the socket is closing and we are
 889          * removing it.
 890          */
 891         __skb_queue_purge(&vvs->rx_queue);
 892         vsock_remove_sock(vsk);
 893 }
 894
 895 static void virtio_transport_wait_close(struct sock *sk, long timeout)
 896 {
 897         if (timeout) {
 898                 DEFINE_WAIT_FUNC(wait, woken_wake_function);
 899
 900                 add_wait_queue(sk_sleep(sk), &wait);
 901
 902                 do {
 903                         if (sk_wait_event(sk, &timeout,
 904                                           sock_flag(sk, SOCK_DONE), &wait))
 905                                 break;
 906                 } while (!signal_pending(current) && timeout);
 907
 908                 remove_wait_queue(sk_sleep(sk), &wait);
 909         }
 910 }
 911
 912 static void virtio_transport_do_close(struct vsock_sock *vsk,
 913                                       bool cancel_timeout)
 914 {
 915         struct sock *sk = sk_vsock(vsk);
 916
 917         sock_set_flag(sk, SOCK_DONE);
 918         vsk->peer_shutdown = SHUTDOWN_MASK;
 919         if (vsock_stream_has_data(vsk) <= 0)
 920                 sk->sk_state = TCP_CLOSING;
 921         sk->sk_state_change(sk);
 922
 923         if (vsk->close_work_scheduled &&
 924             (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
 925                 vsk->close_work_scheduled = false;
 926
 927                 virtio_transport_remove_sock(vsk);
 928
 929                 /* Release refcnt obtained when we scheduled the timeout */
 930                 sock_put(sk);
 931         }
 932 }
 933
 934 static void virtio_transport_close_timeout(struct work_struct *work)
 935 {
 936         struct vsock_sock *vsk =
 937                 container_of(work, struct vsock_sock, close_work.work);
 938         struct sock *sk = sk_vsock(vsk);
 939
 940         sock_hold(sk);
 941         lock_sock(sk);
 942
 943         if (!sock_flag(sk, SOCK_DONE)) {
 944                 (void)virtio_transport_reset(vsk, NULL);
 945
 946                 virtio_transport_do_close(vsk, false);
 947         }
 948
 949         vsk->close_work_scheduled = false;
 950
 951         release_sock(sk);
 952         sock_put(sk);
 953 }
 954
 955 /* User context, vsk->sk is locked */
 956 static bool virtio_transport_close(struct vsock_sock *vsk)
 957 {
 958         struct sock *sk = &vsk->sk;
 959
 960         if (!(sk->sk_state == TCP_ESTABLISHED ||
 961               sk->sk_state == TCP_CLOSING))
 962                 return true;
 963
 964         /* Already received SHUTDOWN from peer, reply with RST */
 965         if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
 966                 (void)virtio_transport_reset(vsk, NULL);
 967                 return true;
 968         }
 969
 970         if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
 971                 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
 972
 973         if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
 974                 virtio_transport_wait_close(sk, sk->sk_lingertime);
 975
 976         if (sock_flag(sk, SOCK_DONE)) {
 977                 return true;
 978         }
 979
 980         sock_hold(sk);
 981         INIT_DELAYED_WORK(&vsk->close_work,
 982                           virtio_transport_close_timeout);
 983         vsk->close_work_scheduled = true;
 984         schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
 985         return false;
 986 }
 987
 988 void virtio_transport_release(struct vsock_sock *vsk)
 989 {
 990         struct sock *sk = &vsk->sk;
 991         bool remove_sock = true;
 992
 993         if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)
 994                 remove_sock = virtio_transport_close(vsk);
 995
 996         if (remove_sock) {
 997                 sock_set_flag(sk, SOCK_DONE);
 998                 virtio_transport_remove_sock(vsk);
 999         }
1000 }
1001 EXPORT_SYMBOL_GPL(virtio_transport_release);
1002
1003 static int
1004 virtio_transport_recv_connecting(struct sock *sk,
1005                                  struct sk_buff *skb)
1006 {
1007         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1008         struct vsock_sock *vsk = vsock_sk(sk);
1009         int skerr;
1010         int err;
1011
1012         switch (le16_to_cpu(hdr->op)) {
1013         case VIRTIO_VSOCK_OP_RESPONSE:
1014                 sk->sk_state = TCP_ESTABLISHED;
1015                 sk->sk_socket->state = SS_CONNECTED;
1016                 vsock_insert_connected(vsk);
1017                 sk->sk_state_change(sk);
1018                 break;
1019         case VIRTIO_VSOCK_OP_INVALID:
1020                 break;
1021         case VIRTIO_VSOCK_OP_RST:
1022                 skerr = ECONNRESET;
1023                 err = 0;
1024                 goto destroy;
1025         default:
1026                 skerr = EPROTO;
1027                 err = -EINVAL;
1028                 goto destroy;
1029         }
1030         return 0;
1031
1032 destroy:
1033         virtio_transport_reset(vsk, skb);
1034         sk->sk_state = TCP_CLOSE;
1035         sk->sk_err = skerr;
1036         sk_error_report(sk);
1037         return err;
1038 }
1039
1040 static void
1041 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
1042                               struct sk_buff *skb)
1043 {
1044         struct virtio_vsock_sock *vvs = vsk->trans;
1045         bool can_enqueue, free_pkt = false;
1046         struct virtio_vsock_hdr *hdr;
1047         u32 len;
1048
1049         hdr = virtio_vsock_hdr(skb);
1050         len = le32_to_cpu(hdr->len);
1051
1052         spin_lock_bh(&vvs->rx_lock);
1053
1054         can_enqueue = virtio_transport_inc_rx_pkt(vvs, len);
1055         if (!can_enqueue) {
1056                 free_pkt = true;
1057                 goto out;
1058         }
1059
1060         if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1061                 vvs->msg_count++;
1062
1063         /* Try to copy small packets into the buffer of last packet queued,
1064          * to avoid wasting memory queueing the entire buffer with a small
1065          * payload.
1066          */
1067         if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
1068                 struct virtio_vsock_hdr *last_hdr;
1069                 struct sk_buff *last_skb;
1070
1071                 last_skb = skb_peek_tail(&vvs->rx_queue);
1072                 last_hdr = virtio_vsock_hdr(last_skb);
1073
1074                 /* If there is space in the last packet queued, we copy the
1075                  * new packet in its buffer. We avoid this if the last packet
1076                  * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is
1077                  * delimiter of SEQPACKET message, so 'pkt' is the first packet
1078                  * of a new message.
1079                  */
1080                 if (skb->len < skb_tailroom(last_skb) &&
1081                     !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) {
1082                         memcpy(skb_put(last_skb, skb->len), skb->data, skb->len);
1083                         free_pkt = true;
1084                         last_hdr->flags |= hdr->flags;
1085                         le32_add_cpu(&last_hdr->len, len);
1086                         goto out;
1087                 }
1088         }
1089
1090         __skb_queue_tail(&vvs->rx_queue, skb);
1091
1092 out:
1093         spin_unlock_bh(&vvs->rx_lock);
1094         if (free_pkt)
1095                 kfree_skb(skb);
1096 }
1097
1098 static int
1099 virtio_transport_recv_connected(struct sock *sk,
1100                                 struct sk_buff *skb)
1101 {
1102         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1103         struct vsock_sock *vsk = vsock_sk(sk);
1104         int err = 0;
1105
1106         switch (le16_to_cpu(hdr->op)) {
1107         case VIRTIO_VSOCK_OP_RW:
1108                 virtio_transport_recv_enqueue(vsk, skb);
1109                 vsock_data_ready(sk);
1110                 return err;
1111         case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
1112                 virtio_transport_send_credit_update(vsk);
1113                 break;
1114         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
1115                 sk->sk_write_space(sk);
1116                 break;
1117         case VIRTIO_VSOCK_OP_SHUTDOWN:
1118                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
1119                         vsk->peer_shutdown |= RCV_SHUTDOWN;
1120                 if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
1121                         vsk->peer_shutdown |= SEND_SHUTDOWN;
1122                 if (vsk->peer_shutdown == SHUTDOWN_MASK) {
1123                         if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) {
1124                                 (void)virtio_transport_reset(vsk, NULL);
1125                                 virtio_transport_do_close(vsk, true);
1126                         }
1127                         /* Remove this socket anyway because the remote peer sent
1128                          * the shutdown. This way a new connection will succeed
1129                          * if the remote peer uses the same source port,
1130                          * even if the old socket is still unreleased, but now disconnected.
1131                          */
1132                         vsock_remove_sock(vsk);
1133                 }
1134                 if (le32_to_cpu(virtio_vsock_hdr(skb)->flags))
1135                         sk->sk_state_change(sk);
1136                 break;
1137         case VIRTIO_VSOCK_OP_RST:
1138                 virtio_transport_do_close(vsk, true);
1139                 break;
1140         default:
1141                 err = -EINVAL;
1142                 break;
1143         }
1144
1145         kfree_skb(skb);
1146         return err;
1147 }
1148
1149 static void
1150 virtio_transport_recv_disconnecting(struct sock *sk,
1151                                     struct sk_buff *skb)
1152 {
1153         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1154         struct vsock_sock *vsk = vsock_sk(sk);
1155
1156         if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1157                 virtio_transport_do_close(vsk, true);
1158 }
1159
1160 static int
1161 virtio_transport_send_response(struct vsock_sock *vsk,
1162                                struct sk_buff *skb)
1163 {
1164         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1165         struct virtio_vsock_pkt_info info = {
1166                 .op = VIRTIO_VSOCK_OP_RESPONSE,
1167                 .remote_cid = le64_to_cpu(hdr->src_cid),
1168                 .remote_port = le32_to_cpu(hdr->src_port),
1169                 .reply = true,
1170                 .vsk = vsk,
1171         };
1172
1173         return virtio_transport_send_pkt_info(vsk, &info);
1174 }
1175
1176 static bool virtio_transport_space_update(struct sock *sk,
1177                                           struct sk_buff *skb)
1178 {
1179         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1180         struct vsock_sock *vsk = vsock_sk(sk);
1181         struct virtio_vsock_sock *vvs = vsk->trans;
1182         bool space_available;
1183
1184         /* Listener sockets are not associated with any transport, so we are
1185          * not able to take the state to see if there is space available in the
1186          * remote peer, but since they are only used to receive requests, we
1187          * can assume that there is always space available in the other peer.
1188          */
1189         if (!vvs)
1190                 return true;
1191
1192         /* buf_alloc and fwd_cnt is always included in the hdr */
1193         spin_lock_bh(&vvs->tx_lock);
1194         vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc);
1195         vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt);
1196         space_available = virtio_transport_has_space(vsk);
1197         spin_unlock_bh(&vvs->tx_lock);
1198         return space_available;
1199 }
1200
1201 /* Handle server socket */
1202 static int
1203 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb,
1204                              struct virtio_transport *t)
1205 {
1206         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1207         struct vsock_sock *vsk = vsock_sk(sk);
1208         struct vsock_sock *vchild;
1209         struct sock *child;
1210         int ret;
1211
1212         if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) {
1213                 virtio_transport_reset_no_sock(t, skb);
1214                 return -EINVAL;
1215         }
1216
1217         if (sk_acceptq_is_full(sk)) {
1218                 virtio_transport_reset_no_sock(t, skb);
1219                 return -ENOMEM;
1220         }
1221
1222         child = vsock_create_connected(sk);
1223         if (!child) {
1224                 virtio_transport_reset_no_sock(t, skb);
1225                 return -ENOMEM;
1226         }
1227
1228         sk_acceptq_added(sk);
1229
1230         lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1231
1232         child->sk_state = TCP_ESTABLISHED;
1233
1234         vchild = vsock_sk(child);
1235         vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid),
1236                         le32_to_cpu(hdr->dst_port));
1237         vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid),
1238                         le32_to_cpu(hdr->src_port));
1239
1240         ret = vsock_assign_transport(vchild, vsk);
1241         /* Transport assigned (looking at remote_addr) must be the same
1242          * where we received the request.
1243          */
1244         if (ret || vchild->transport != &t->transport) {
1245                 release_sock(child);
1246                 virtio_transport_reset_no_sock(t, skb);
1247                 sock_put(child);
1248                 return ret;
1249         }
1250
1251         if (virtio_transport_space_update(child, skb))
1252                 child->sk_write_space(child);
1253
1254         vsock_insert_connected(vchild);
1255         vsock_enqueue_accept(sk, child);
1256         virtio_transport_send_response(vchild, skb);
1257
1258         release_sock(child);
1259
1260         sk->sk_data_ready(sk);
1261         return 0;
1262 }
1263
1264 static bool virtio_transport_valid_type(u16 type)
1265 {
1266         return (type == VIRTIO_VSOCK_TYPE_STREAM) ||
1267                (type == VIRTIO_VSOCK_TYPE_SEQPACKET);
1268 }
1269
1270 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1271  * lock.
1272  */
1273 void virtio_transport_recv_pkt(struct virtio_transport *t,
1274                                struct sk_buff *skb)
1275 {
1276         struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1277         struct sockaddr_vm src, dst;
1278         struct vsock_sock *vsk;
1279         struct sock *sk;
1280         bool space_available;
1281
1282         vsock_addr_init(&src, le64_to_cpu(hdr->src_cid),
1283                         le32_to_cpu(hdr->src_port));
1284         vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid),
1285                         le32_to_cpu(hdr->dst_port));
1286
1287         trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1288                                         dst.svm_cid, dst.svm_port,
1289                                         le32_to_cpu(hdr->len),
1290                                         le16_to_cpu(hdr->type),
1291                                         le16_to_cpu(hdr->op),
1292                                         le32_to_cpu(hdr->flags),
1293                                         le32_to_cpu(hdr->buf_alloc),
1294                                         le32_to_cpu(hdr->fwd_cnt));
1295
1296         if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) {
1297                 (void)virtio_transport_reset_no_sock(t, skb);
1298                 goto free_pkt;
1299         }
1300
1301         /* The socket must be in connected or bound table
1302          * otherwise send reset back
1303          */
1304         sk = vsock_find_connected_socket(&src, &dst);
1305         if (!sk) {
1306                 sk = vsock_find_bound_socket(&dst);
1307                 if (!sk) {
1308                         (void)virtio_transport_reset_no_sock(t, skb);
1309                         goto free_pkt;
1310                 }
1311         }
1312
1313         if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) {
1314                 (void)virtio_transport_reset_no_sock(t, skb);
1315                 sock_put(sk);
1316                 goto free_pkt;
1317         }
1318
1319         if (!skb_set_owner_sk_safe(skb, sk)) {
1320                 WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
1321                 goto free_pkt;
1322         }
1323
1324         vsk = vsock_sk(sk);
1325
1326         lock_sock(sk);
1327
1328         /* Check if sk has been closed before lock_sock */
1329         if (sock_flag(sk, SOCK_DONE)) {
1330                 (void)virtio_transport_reset_no_sock(t, skb);
1331                 release_sock(sk);
1332                 sock_put(sk);
1333                 goto free_pkt;
1334         }
1335
1336         space_available = virtio_transport_space_update(sk, skb);
1337
1338         /* Update CID in case it has changed after a transport reset event */
1339         if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1340                 vsk->local_addr.svm_cid = dst.svm_cid;
1341
1342         if (space_available)
1343                 sk->sk_write_space(sk);
1344
1345         switch (sk->sk_state) {
1346         case TCP_LISTEN:
1347                 virtio_transport_recv_listen(sk, skb, t);
1348                 kfree_skb(skb);
1349                 break;
1350         case TCP_SYN_SENT:
1351                 virtio_transport_recv_connecting(sk, skb);
1352                 kfree_skb(skb);
1353                 break;
1354         case TCP_ESTABLISHED:
1355                 virtio_transport_recv_connected(sk, skb);
1356                 break;
1357         case TCP_CLOSING:
1358                 virtio_transport_recv_disconnecting(sk, skb);
1359                 kfree_skb(skb);
1360                 break;
1361         default:
1362                 (void)virtio_transport_reset_no_sock(t, skb);
1363                 kfree_skb(skb);
1364                 break;
1365         }
1366
1367         release_sock(sk);
1368
1369         /* Release refcnt obtained when we fetched this socket out of the
1370          * bound or connected list.
1371          */
1372         sock_put(sk);
1373         return;
1374
1375 free_pkt:
1376         kfree_skb(skb);
1377 }
1378 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1379
1380 /* Remove skbs found in a queue that have a vsk that matches.
1381  *
1382  * Each skb is freed.
1383  *
1384  * Returns the count of skbs that were reply packets.
1385  */
1386 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue)
1387 {
1388         struct sk_buff_head freeme;
1389         struct sk_buff *skb, *tmp;
1390         int cnt = 0;
1391
1392         skb_queue_head_init(&freeme);
1393
1394         spin_lock_bh(&queue->lock);
1395         skb_queue_walk_safe(queue, skb, tmp) {
1396                 if (vsock_sk(skb->sk) != vsk)
1397                         continue;
1398
1399                 __skb_unlink(skb, queue);
1400                 __skb_queue_tail(&freeme, skb);
1401
1402                 if (virtio_vsock_skb_reply(skb))
1403                         cnt++;
1404         }
1405         spin_unlock_bh(&queue->lock);
1406
1407         __skb_queue_purge(&freeme);
1408
1409         return cnt;
1410 }
1411 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs);
1412
1413 MODULE_LICENSE("GPL v2");
1414 MODULE_AUTHOR("Asias He");
1415 MODULE_DESCRIPTION("common code for virtio vsock");