net/vmw_vsock/virtio_transport_common.c

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