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