net/vmw_vsock/virtio_transport_common.c

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