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 *virtio_transport_get_ops(void)
  33 {
  34         const struct vsock_transport *t = vsock_core_get_transport();
  35
  36         return container_of(t, struct virtio_transport, transport);
  37 }
  38
  39 static struct virtio_vsock_pkt *
  40 virtio_transport_alloc_pkt(struct virtio_vsock_pkt_info *info,
  41                            size_t len,
  42                            u32 src_cid,
  43                            u32 src_port,
  44                            u32 dst_cid,
  45                            u32 dst_port)
  46 {
  47         struct virtio_vsock_pkt *pkt;
  48         int err;
  49
  50         pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
  51         if (!pkt)
  52                 return NULL;
  53
  54         pkt->hdr.type           = cpu_to_le16(info->type);
  55         pkt->hdr.op             = cpu_to_le16(info->op);
  56         pkt->hdr.src_cid        = cpu_to_le64(src_cid);
  57         pkt->hdr.dst_cid        = cpu_to_le64(dst_cid);
  58         pkt->hdr.src_port       = cpu_to_le32(src_port);
  59         pkt->hdr.dst_port       = cpu_to_le32(dst_port);
  60         pkt->hdr.flags          = cpu_to_le32(info->flags);
  61         pkt->len                = len;
  62         pkt->hdr.len            = cpu_to_le32(len);
  63         pkt->reply              = info->reply;
  64         pkt->vsk                = info->vsk;
  65
  66         if (info->msg && len > 0) {
  67                 pkt->buf = kmalloc(len, GFP_KERNEL);
  68                 if (!pkt->buf)
  69                         goto out_pkt;
  70
  71                 pkt->buf_len = len;
  72
  73                 err = memcpy_from_msg(pkt->buf, info->msg, len);
  74                 if (err)
  75                         goto out;
  76         }
  77
  78         trace_virtio_transport_alloc_pkt(src_cid, src_port,
  79                                          dst_cid, dst_port,
  80                                          len,
  81                                          info->type,
  82                                          info->op,
  83                                          info->flags);
  84
  85         return pkt;
  86
  87 out:
  88         kfree(pkt->buf);
  89 out_pkt:
  90         kfree(pkt);
  91         return NULL;
  92 }
  93
  94 /* Packet capture */
  95 static struct sk_buff *virtio_transport_build_skb(void *opaque)
  96 {
  97         struct virtio_vsock_pkt *pkt = opaque;
  98         struct af_vsockmon_hdr *hdr;
  99         struct sk_buff *skb;
 100         size_t payload_len;
 101         void *payload_buf;
 102
 103         /* A packet could be split to fit the RX buffer, so we can retrieve
 104          * the payload length from the header and the buffer pointer taking
 105          * care of the offset in the original packet.
 106          */
 107         payload_len = le32_to_cpu(pkt->hdr.len);
 108         payload_buf = pkt->buf + pkt->off;
 109
 110         skb = alloc_skb(sizeof(*hdr) + sizeof(pkt->hdr) + payload_len,
 111                         GFP_ATOMIC);
 112         if (!skb)
 113                 return NULL;
 114
 115         hdr = skb_put(skb, sizeof(*hdr));
 116
 117         /* pkt->hdr is little-endian so no need to byteswap here */
 118         hdr->src_cid = pkt->hdr.src_cid;
 119         hdr->src_port = pkt->hdr.src_port;
 120         hdr->dst_cid = pkt->hdr.dst_cid;
 121         hdr->dst_port = pkt->hdr.dst_port;
 122
 123         hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
 124         hdr->len = cpu_to_le16(sizeof(pkt->hdr));
 125         memset(hdr->reserved, 0, sizeof(hdr->reserved));
 126
 127         switch (le16_to_cpu(pkt->hdr.op)) {
 128         case VIRTIO_VSOCK_OP_REQUEST:
 129         case VIRTIO_VSOCK_OP_RESPONSE:
 130                 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
 131                 break;
 132         case VIRTIO_VSOCK_OP_RST:
 133         case VIRTIO_VSOCK_OP_SHUTDOWN:
 134                 hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
 135                 break;
 136         case VIRTIO_VSOCK_OP_RW:
 137                 hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
 138                 break;
 139         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
 140         case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
 141                 hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
 142                 break;
 143         default:
 144                 hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
 145                 break;
 146         }
 147
 148         skb_put_data(skb, &pkt->hdr, sizeof(pkt->hdr));
 149
 150         if (payload_len) {
 151                 skb_put_data(skb, payload_buf, payload_len);
 152         }
 153
 154         return skb;
 155 }
 156
 157 void virtio_transport_deliver_tap_pkt(struct virtio_vsock_pkt *pkt)
 158 {
 159         vsock_deliver_tap(virtio_transport_build_skb, pkt);
 160 }
 161 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
 162
 163 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
 164                                           struct virtio_vsock_pkt_info *info)
 165 {
 166         u32 src_cid, src_port, dst_cid, dst_port;
 167         struct virtio_vsock_sock *vvs;
 168         struct virtio_vsock_pkt *pkt;
 169         u32 pkt_len = info->pkt_len;
 170
 171         src_cid = vm_sockets_get_local_cid();
 172         src_port = vsk->local_addr.svm_port;
 173         if (!info->remote_cid) {
 174                 dst_cid = vsk->remote_addr.svm_cid;
 175                 dst_port = vsk->remote_addr.svm_port;
 176         } else {
 177                 dst_cid = info->remote_cid;
 178                 dst_port = info->remote_port;
 179         }
 180
 181         vvs = vsk->trans;
 182
 183         /* we can send less than pkt_len bytes */
 184         if (pkt_len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE)
 185                 pkt_len = VIRTIO_VSOCK_MAX_PKT_BUF_SIZE;
 186
 187         /* virtio_transport_get_credit might return less than pkt_len credit */
 188         pkt_len = virtio_transport_get_credit(vvs, pkt_len);
 189
 190         /* Do not send zero length OP_RW pkt */
 191         if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
 192                 return pkt_len;
 193
 194         pkt = virtio_transport_alloc_pkt(info, pkt_len,
 195                                          src_cid, src_port,
 196                                          dst_cid, dst_port);
 197         if (!pkt) {
 198                 virtio_transport_put_credit(vvs, pkt_len);
 199                 return -ENOMEM;
 200         }
 201
 202         virtio_transport_inc_tx_pkt(vvs, pkt);
 203
 204         return virtio_transport_get_ops()->send_pkt(pkt);
 205 }
 206
 207 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
 208                                         struct virtio_vsock_pkt *pkt)
 209 {
 210         if (vvs->rx_bytes + pkt->len > vvs->buf_alloc)
 211                 return false;
 212
 213         vvs->rx_bytes += pkt->len;
 214         return true;
 215 }
 216
 217 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
 218                                         struct virtio_vsock_pkt *pkt)
 219 {
 220         vvs->rx_bytes -= pkt->len;
 221         vvs->fwd_cnt += pkt->len;
 222 }
 223
 224 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct virtio_vsock_pkt *pkt)
 225 {
 226         spin_lock_bh(&vvs->rx_lock);
 227         vvs->last_fwd_cnt = vvs->fwd_cnt;
 228         pkt->hdr.fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
 229         pkt->hdr.buf_alloc = cpu_to_le32(vvs->buf_alloc);
 230         spin_unlock_bh(&vvs->rx_lock);
 231 }
 232 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
 233
 234 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
 235 {
 236         u32 ret;
 237
 238         spin_lock_bh(&vvs->tx_lock);
 239         ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 240         if (ret > credit)
 241                 ret = credit;
 242         vvs->tx_cnt += ret;
 243         spin_unlock_bh(&vvs->tx_lock);
 244
 245         return ret;
 246 }
 247 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
 248
 249 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
 250 {
 251         spin_lock_bh(&vvs->tx_lock);
 252         vvs->tx_cnt -= credit;
 253         spin_unlock_bh(&vvs->tx_lock);
 254 }
 255 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
 256
 257 static int virtio_transport_send_credit_update(struct vsock_sock *vsk,
 258                                                int type,
 259                                                struct virtio_vsock_hdr *hdr)
 260 {
 261         struct virtio_vsock_pkt_info info = {
 262                 .op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
 263                 .type = type,
 264                 .vsk = vsk,
 265         };
 266
 267         return virtio_transport_send_pkt_info(vsk, &info);
 268 }
 269
 270 static ssize_t
 271 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
 272                                 struct msghdr *msg,
 273                                 size_t len)
 274 {
 275         struct virtio_vsock_sock *vvs = vsk->trans;
 276         struct virtio_vsock_pkt *pkt;
 277         size_t bytes, total = 0, off;
 278         int err = -EFAULT;
 279
 280         spin_lock_bh(&vvs->rx_lock);
 281
 282         list_for_each_entry(pkt, &vvs->rx_queue, list) {
 283                 off = pkt->off;
 284
 285                 if (total == len)
 286                         break;
 287
 288                 while (total < len && off < pkt->len) {
 289                         bytes = len - total;
 290                         if (bytes > pkt->len - off)
 291                                 bytes = pkt->len - off;
 292
 293                         /* sk_lock is held by caller so no one else can dequeue.
 294                          * Unlock rx_lock since memcpy_to_msg() may sleep.
 295                          */
 296                         spin_unlock_bh(&vvs->rx_lock);
 297
 298                         err = memcpy_to_msg(msg, pkt->buf + off, bytes);
 299                         if (err)
 300                                 goto out;
 301
 302                         spin_lock_bh(&vvs->rx_lock);
 303
 304                         total += bytes;
 305                         off += bytes;
 306                 }
 307         }
 308
 309         spin_unlock_bh(&vvs->rx_lock);
 310
 311         return total;
 312
 313 out:
 314         if (total)
 315                 err = total;
 316         return err;
 317 }
 318
 319 static ssize_t
 320 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
 321                                    struct msghdr *msg,
 322                                    size_t len)
 323 {
 324         struct virtio_vsock_sock *vvs = vsk->trans;
 325         struct virtio_vsock_pkt *pkt;
 326         size_t bytes, total = 0;
 327         u32 free_space;
 328         int err = -EFAULT;
 329
 330         spin_lock_bh(&vvs->rx_lock);
 331         while (total < len && !list_empty(&vvs->rx_queue)) {
 332                 pkt = list_first_entry(&vvs->rx_queue,
 333                                        struct virtio_vsock_pkt, list);
 334
 335                 bytes = len - total;
 336                 if (bytes > pkt->len - pkt->off)
 337                         bytes = pkt->len - pkt->off;
 338
 339                 /* sk_lock is held by caller so no one else can dequeue.
 340                  * Unlock rx_lock since memcpy_to_msg() may sleep.
 341                  */
 342                 spin_unlock_bh(&vvs->rx_lock);
 343
 344                 err = memcpy_to_msg(msg, pkt->buf + pkt->off, bytes);
 345                 if (err)
 346                         goto out;
 347
 348                 spin_lock_bh(&vvs->rx_lock);
 349
 350                 total += bytes;
 351                 pkt->off += bytes;
 352                 if (pkt->off == pkt->len) {
 353                         virtio_transport_dec_rx_pkt(vvs, pkt);
 354                         list_del(&pkt->list);
 355                         virtio_transport_free_pkt(pkt);
 356                 }
 357         }
 358
 359         free_space = vvs->buf_alloc - (vvs->fwd_cnt - vvs->last_fwd_cnt);
 360
 361         spin_unlock_bh(&vvs->rx_lock);
 362
 363         /* To reduce the number of credit update messages,
 364          * don't update credits as long as lots of space is available.
 365          * Note: the limit chosen here is arbitrary. Setting the limit
 366          * too high causes extra messages. Too low causes transmitter
 367          * stalls. As stalls are in theory more expensive than extra
 368          * messages, we set the limit to a high value. TODO: experiment
 369          * with different values.
 370          */
 371         if (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
 372                 virtio_transport_send_credit_update(vsk,
 373                                                     VIRTIO_VSOCK_TYPE_STREAM,
 374                                                     NULL);
 375         }
 376
 377         return total;
 378
 379 out:
 380         if (total)
 381                 err = total;
 382         return err;
 383 }
 384
 385 ssize_t
 386 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
 387                                 struct msghdr *msg,
 388                                 size_t len, int flags)
 389 {
 390         if (flags & MSG_PEEK)
 391                 return virtio_transport_stream_do_peek(vsk, msg, len);
 392         else
 393                 return virtio_transport_stream_do_dequeue(vsk, msg, len);
 394 }
 395 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
 396
 397 int
 398 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
 399                                struct msghdr *msg,
 400                                size_t len, int flags)
 401 {
 402         return -EOPNOTSUPP;
 403 }
 404 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
 405
 406 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
 407 {
 408         struct virtio_vsock_sock *vvs = vsk->trans;
 409         s64 bytes;
 410
 411         spin_lock_bh(&vvs->rx_lock);
 412         bytes = vvs->rx_bytes;
 413         spin_unlock_bh(&vvs->rx_lock);
 414
 415         return bytes;
 416 }
 417 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
 418
 419 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
 420 {
 421         struct virtio_vsock_sock *vvs = vsk->trans;
 422         s64 bytes;
 423
 424         bytes = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
 425         if (bytes < 0)
 426                 bytes = 0;
 427
 428         return bytes;
 429 }
 430
 431 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
 432 {
 433         struct virtio_vsock_sock *vvs = vsk->trans;
 434         s64 bytes;
 435
 436         spin_lock_bh(&vvs->tx_lock);
 437         bytes = virtio_transport_has_space(vsk);
 438         spin_unlock_bh(&vvs->tx_lock);
 439
 440         return bytes;
 441 }
 442 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
 443
 444 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
 445                                     struct vsock_sock *psk)
 446 {
 447         struct virtio_vsock_sock *vvs;
 448
 449         vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
 450         if (!vvs)
 451                 return -ENOMEM;
 452
 453         vsk->trans = vvs;
 454         vvs->vsk = vsk;
 455         if (psk) {
 456                 struct virtio_vsock_sock *ptrans = psk->trans;
 457
 458                 vvs->buf_size   = ptrans->buf_size;
 459                 vvs->buf_size_min = ptrans->buf_size_min;
 460                 vvs->buf_size_max = ptrans->buf_size_max;
 461                 vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
 462         } else {
 463                 vvs->buf_size = VIRTIO_VSOCK_DEFAULT_BUF_SIZE;
 464                 vvs->buf_size_min = VIRTIO_VSOCK_DEFAULT_MIN_BUF_SIZE;
 465                 vvs->buf_size_max = VIRTIO_VSOCK_DEFAULT_MAX_BUF_SIZE;
 466         }
 467
 468         vvs->buf_alloc = vvs->buf_size;
 469
 470         spin_lock_init(&vvs->rx_lock);
 471         spin_lock_init(&vvs->tx_lock);
 472         INIT_LIST_HEAD(&vvs->rx_queue);
 473
 474         return 0;
 475 }
 476 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
 477
 478 u64 virtio_transport_get_buffer_size(struct vsock_sock *vsk)
 479 {
 480         struct virtio_vsock_sock *vvs = vsk->trans;
 481
 482         return vvs->buf_size;
 483 }
 484 EXPORT_SYMBOL_GPL(virtio_transport_get_buffer_size);
 485
 486 u64 virtio_transport_get_min_buffer_size(struct vsock_sock *vsk)
 487 {
 488         struct virtio_vsock_sock *vvs = vsk->trans;
 489
 490         return vvs->buf_size_min;
 491 }
 492 EXPORT_SYMBOL_GPL(virtio_transport_get_min_buffer_size);
 493
 494 u64 virtio_transport_get_max_buffer_size(struct vsock_sock *vsk)
 495 {
 496         struct virtio_vsock_sock *vvs = vsk->trans;
 497
 498         return vvs->buf_size_max;
 499 }
 500 EXPORT_SYMBOL_GPL(virtio_transport_get_max_buffer_size);
 501
 502 void virtio_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
 503 {
 504         struct virtio_vsock_sock *vvs = vsk->trans;
 505
 506         if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
 507                 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
 508         if (val < vvs->buf_size_min)
 509                 vvs->buf_size_min = val;
 510         if (val > vvs->buf_size_max)
 511                 vvs->buf_size_max = val;
 512         vvs->buf_size = val;
 513         vvs->buf_alloc = val;
 514
 515         virtio_transport_send_credit_update(vsk, VIRTIO_VSOCK_TYPE_STREAM,
 516                                             NULL);
 517 }
 518 EXPORT_SYMBOL_GPL(virtio_transport_set_buffer_size);
 519
 520 void virtio_transport_set_min_buffer_size(struct vsock_sock *vsk, u64 val)
 521 {
 522         struct virtio_vsock_sock *vvs = vsk->trans;
 523
 524         if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
 525                 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
 526         if (val > vvs->buf_size)
 527                 vvs->buf_size = val;
 528         vvs->buf_size_min = val;
 529 }
 530 EXPORT_SYMBOL_GPL(virtio_transport_set_min_buffer_size);
 531
 532 void virtio_transport_set_max_buffer_size(struct vsock_sock *vsk, u64 val)
 533 {
 534         struct virtio_vsock_sock *vvs = vsk->trans;
 535
 536         if (val > VIRTIO_VSOCK_MAX_BUF_SIZE)
 537                 val = VIRTIO_VSOCK_MAX_BUF_SIZE;
 538         if (val < vvs->buf_size)
 539                 vvs->buf_size = val;
 540         vvs->buf_size_max = val;
 541 }
 542 EXPORT_SYMBOL_GPL(virtio_transport_set_max_buffer_size);
 543
 544 int
 545 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
 546                                 size_t target,
 547                                 bool *data_ready_now)
 548 {
 549         if (vsock_stream_has_data(vsk))
 550                 *data_ready_now = true;
 551         else
 552                 *data_ready_now = false;
 553
 554         return 0;
 555 }
 556 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
 557
 558 int
 559 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
 560                                  size_t target,
 561                                  bool *space_avail_now)
 562 {
 563         s64 free_space;
 564
 565         free_space = vsock_stream_has_space(vsk);
 566         if (free_space > 0)
 567                 *space_avail_now = true;
 568         else if (free_space == 0)
 569                 *space_avail_now = false;
 570
 571         return 0;
 572 }
 573 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
 574
 575 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
 576         size_t target, struct vsock_transport_recv_notify_data *data)
 577 {
 578         return 0;
 579 }
 580 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
 581
 582 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
 583         size_t target, struct vsock_transport_recv_notify_data *data)
 584 {
 585         return 0;
 586 }
 587 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
 588
 589 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
 590         size_t target, struct vsock_transport_recv_notify_data *data)
 591 {
 592         return 0;
 593 }
 594 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
 595
 596 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
 597         size_t target, ssize_t copied, bool data_read,
 598         struct vsock_transport_recv_notify_data *data)
 599 {
 600         return 0;
 601 }
 602 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
 603
 604 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
 605         struct vsock_transport_send_notify_data *data)
 606 {
 607         return 0;
 608 }
 609 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
 610
 611 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
 612         struct vsock_transport_send_notify_data *data)
 613 {
 614         return 0;
 615 }
 616 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
 617
 618 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
 619         struct vsock_transport_send_notify_data *data)
 620 {
 621         return 0;
 622 }
 623 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
 624
 625 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
 626         ssize_t written, struct vsock_transport_send_notify_data *data)
 627 {
 628         return 0;
 629 }
 630 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
 631
 632 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
 633 {
 634         struct virtio_vsock_sock *vvs = vsk->trans;
 635
 636         return vvs->buf_size;
 637 }
 638 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
 639
 640 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
 641 {
 642         return true;
 643 }
 644 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
 645
 646 bool virtio_transport_stream_allow(u32 cid, u32 port)
 647 {
 648         return true;
 649 }
 650 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
 651
 652 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
 653                                 struct sockaddr_vm *addr)
 654 {
 655         return -EOPNOTSUPP;
 656 }
 657 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
 658
 659 bool virtio_transport_dgram_allow(u32 cid, u32 port)
 660 {
 661         return false;
 662 }
 663 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
 664
 665 int virtio_transport_connect(struct vsock_sock *vsk)
 666 {
 667         struct virtio_vsock_pkt_info info = {
 668                 .op = VIRTIO_VSOCK_OP_REQUEST,
 669                 .type = VIRTIO_VSOCK_TYPE_STREAM,
 670                 .vsk = vsk,
 671         };
 672
 673         return virtio_transport_send_pkt_info(vsk, &info);
 674 }
 675 EXPORT_SYMBOL_GPL(virtio_transport_connect);
 676
 677 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
 678 {
 679         struct virtio_vsock_pkt_info info = {
 680                 .op = VIRTIO_VSOCK_OP_SHUTDOWN,
 681                 .type = VIRTIO_VSOCK_TYPE_STREAM,
 682                 .flags = (mode & RCV_SHUTDOWN ?
 683                           VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
 684                          (mode & SEND_SHUTDOWN ?
 685                           VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
 686                 .vsk = vsk,
 687         };
 688
 689         return virtio_transport_send_pkt_info(vsk, &info);
 690 }
 691 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
 692
 693 int
 694 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
 695                                struct sockaddr_vm *remote_addr,
 696                                struct msghdr *msg,
 697                                size_t dgram_len)
 698 {
 699         return -EOPNOTSUPP;
 700 }
 701 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
 702
 703 ssize_t
 704 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
 705                                 struct msghdr *msg,
 706                                 size_t len)
 707 {
 708         struct virtio_vsock_pkt_info info = {
 709                 .op = VIRTIO_VSOCK_OP_RW,
 710                 .type = VIRTIO_VSOCK_TYPE_STREAM,
 711                 .msg = msg,
 712                 .pkt_len = len,
 713                 .vsk = vsk,
 714         };
 715
 716         return virtio_transport_send_pkt_info(vsk, &info);
 717 }
 718 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
 719
 720 void virtio_transport_destruct(struct vsock_sock *vsk)
 721 {
 722         struct virtio_vsock_sock *vvs = vsk->trans;
 723
 724         kfree(vvs);
 725 }
 726 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
 727
 728 static int virtio_transport_reset(struct vsock_sock *vsk,
 729                                   struct virtio_vsock_pkt *pkt)
 730 {
 731         struct virtio_vsock_pkt_info info = {
 732                 .op = VIRTIO_VSOCK_OP_RST,
 733                 .type = VIRTIO_VSOCK_TYPE_STREAM,
 734                 .reply = !!pkt,
 735                 .vsk = vsk,
 736         };
 737
 738         /* Send RST only if the original pkt is not a RST pkt */
 739         if (pkt && le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
 740                 return 0;
 741
 742         return virtio_transport_send_pkt_info(vsk, &info);
 743 }
 744
 745 /* Normally packets are associated with a socket.  There may be no socket if an
 746  * attempt was made to connect to a socket that does not exist.
 747  */
 748 static int virtio_transport_reset_no_sock(struct virtio_vsock_pkt *pkt)
 749 {
 750         const struct virtio_transport *t;
 751         struct virtio_vsock_pkt *reply;
 752         struct virtio_vsock_pkt_info info = {
 753                 .op = VIRTIO_VSOCK_OP_RST,
 754                 .type = le16_to_cpu(pkt->hdr.type),
 755                 .reply = true,
 756         };
 757
 758         /* Send RST only if the original pkt is not a RST pkt */
 759         if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
 760                 return 0;
 761
 762         reply = virtio_transport_alloc_pkt(&info, 0,
 763                                            le64_to_cpu(pkt->hdr.dst_cid),
 764                                            le32_to_cpu(pkt->hdr.dst_port),
 765                                            le64_to_cpu(pkt->hdr.src_cid),
 766                                            le32_to_cpu(pkt->hdr.src_port));
 767         if (!reply)
 768                 return -ENOMEM;
 769
 770         t = virtio_transport_get_ops();
 771         if (!t) {
 772                 virtio_transport_free_pkt(reply);
 773                 return -ENOTCONN;
 774         }
 775
 776         return t->send_pkt(reply);
 777 }
 778
 779 static void virtio_transport_wait_close(struct sock *sk, long timeout)
 780 {
 781         if (timeout) {
 782                 DEFINE_WAIT_FUNC(wait, woken_wake_function);
 783
 784                 add_wait_queue(sk_sleep(sk), &wait);
 785
 786                 do {
 787                         if (sk_wait_event(sk, &timeout,
 788                                           sock_flag(sk, SOCK_DONE), &wait))
 789                                 break;
 790                 } while (!signal_pending(current) && timeout);
 791
 792                 remove_wait_queue(sk_sleep(sk), &wait);
 793         }
 794 }
 795
 796 static void virtio_transport_do_close(struct vsock_sock *vsk,
 797                                       bool cancel_timeout)
 798 {
 799         struct sock *sk = sk_vsock(vsk);
 800
 801         sock_set_flag(sk, SOCK_DONE);
 802         vsk->peer_shutdown = SHUTDOWN_MASK;
 803         if (vsock_stream_has_data(vsk) <= 0)
 804                 sk->sk_state = TCP_CLOSING;
 805         sk->sk_state_change(sk);
 806
 807         if (vsk->close_work_scheduled &&
 808             (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
 809                 vsk->close_work_scheduled = false;
 810
 811                 vsock_remove_sock(vsk);
 812
 813                 /* Release refcnt obtained when we scheduled the timeout */
 814                 sock_put(sk);
 815         }
 816 }
 817
 818 static void virtio_transport_close_timeout(struct work_struct *work)
 819 {
 820         struct vsock_sock *vsk =
 821                 container_of(work, struct vsock_sock, close_work.work);
 822         struct sock *sk = sk_vsock(vsk);
 823
 824         sock_hold(sk);
 825         lock_sock(sk);
 826
 827         if (!sock_flag(sk, SOCK_DONE)) {
 828                 (void)virtio_transport_reset(vsk, NULL);
 829
 830                 virtio_transport_do_close(vsk, false);
 831         }
 832
 833         vsk->close_work_scheduled = false;
 834
 835         release_sock(sk);
 836         sock_put(sk);
 837 }
 838
 839 /* User context, vsk->sk is locked */
 840 static bool virtio_transport_close(struct vsock_sock *vsk)
 841 {
 842         struct sock *sk = &vsk->sk;
 843
 844         if (!(sk->sk_state == TCP_ESTABLISHED ||
 845               sk->sk_state == TCP_CLOSING))
 846                 return true;
 847
 848         /* Already received SHUTDOWN from peer, reply with RST */
 849         if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
 850                 (void)virtio_transport_reset(vsk, NULL);
 851                 return true;
 852         }
 853
 854         if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
 855                 (void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
 856
 857         if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
 858                 virtio_transport_wait_close(sk, sk->sk_lingertime);
 859
 860         if (sock_flag(sk, SOCK_DONE)) {
 861                 return true;
 862         }
 863
 864         sock_hold(sk);
 865         INIT_DELAYED_WORK(&vsk->close_work,
 866                           virtio_transport_close_timeout);
 867         vsk->close_work_scheduled = true;
 868         schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
 869         return false;
 870 }
 871
 872 void virtio_transport_release(struct vsock_sock *vsk)
 873 {
 874         struct virtio_vsock_sock *vvs = vsk->trans;
 875         struct virtio_vsock_pkt *pkt, *tmp;
 876         struct sock *sk = &vsk->sk;
 877         bool remove_sock = true;
 878
 879         lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 880         if (sk->sk_type == SOCK_STREAM)
 881                 remove_sock = virtio_transport_close(vsk);
 882
 883         list_for_each_entry_safe(pkt, tmp, &vvs->rx_queue, list) {
 884                 list_del(&pkt->list);
 885                 virtio_transport_free_pkt(pkt);
 886         }
 887         release_sock(sk);
 888
 889         if (remove_sock)
 890                 vsock_remove_sock(vsk);
 891 }
 892 EXPORT_SYMBOL_GPL(virtio_transport_release);
 893
 894 static int
 895 virtio_transport_recv_connecting(struct sock *sk,
 896                                  struct virtio_vsock_pkt *pkt)
 897 {
 898         struct vsock_sock *vsk = vsock_sk(sk);
 899         int err;
 900         int skerr;
 901
 902         switch (le16_to_cpu(pkt->hdr.op)) {
 903         case VIRTIO_VSOCK_OP_RESPONSE:
 904                 sk->sk_state = TCP_ESTABLISHED;
 905                 sk->sk_socket->state = SS_CONNECTED;
 906                 vsock_insert_connected(vsk);
 907                 sk->sk_state_change(sk);
 908                 break;
 909         case VIRTIO_VSOCK_OP_INVALID:
 910                 break;
 911         case VIRTIO_VSOCK_OP_RST:
 912                 skerr = ECONNRESET;
 913                 err = 0;
 914                 goto destroy;
 915         default:
 916                 skerr = EPROTO;
 917                 err = -EINVAL;
 918                 goto destroy;
 919         }
 920         return 0;
 921
 922 destroy:
 923         virtio_transport_reset(vsk, pkt);
 924         sk->sk_state = TCP_CLOSE;
 925         sk->sk_err = skerr;
 926         sk->sk_error_report(sk);
 927         return err;
 928 }
 929
 930 static void
 931 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
 932                               struct virtio_vsock_pkt *pkt)
 933 {
 934         struct virtio_vsock_sock *vvs = vsk->trans;
 935         bool can_enqueue, free_pkt = false;
 936
 937         pkt->len = le32_to_cpu(pkt->hdr.len);
 938         pkt->off = 0;
 939
 940         spin_lock_bh(&vvs->rx_lock);
 941
 942         can_enqueue = virtio_transport_inc_rx_pkt(vvs, pkt);
 943         if (!can_enqueue) {
 944                 free_pkt = true;
 945                 goto out;
 946         }
 947
 948         /* Try to copy small packets into the buffer of last packet queued,
 949          * to avoid wasting memory queueing the entire buffer with a small
 950          * payload.
 951          */
 952         if (pkt->len <= GOOD_COPY_LEN && !list_empty(&vvs->rx_queue)) {
 953                 struct virtio_vsock_pkt *last_pkt;
 954
 955                 last_pkt = list_last_entry(&vvs->rx_queue,
 956                                            struct virtio_vsock_pkt, list);
 957
 958                 /* If there is space in the last packet queued, we copy the
 959                  * new packet in its buffer.
 960                  */
 961                 if (pkt->len <= last_pkt->buf_len - last_pkt->len) {
 962                         memcpy(last_pkt->buf + last_pkt->len, pkt->buf,
 963                                pkt->len);
 964                         last_pkt->len += pkt->len;
 965                         free_pkt = true;
 966                         goto out;
 967                 }
 968         }
 969
 970         list_add_tail(&pkt->list, &vvs->rx_queue);
 971
 972 out:
 973         spin_unlock_bh(&vvs->rx_lock);
 974         if (free_pkt)
 975                 virtio_transport_free_pkt(pkt);
 976 }
 977
 978 static int
 979 virtio_transport_recv_connected(struct sock *sk,
 980                                 struct virtio_vsock_pkt *pkt)
 981 {
 982         struct vsock_sock *vsk = vsock_sk(sk);
 983         int err = 0;
 984
 985         switch (le16_to_cpu(pkt->hdr.op)) {
 986         case VIRTIO_VSOCK_OP_RW:
 987                 virtio_transport_recv_enqueue(vsk, pkt);
 988                 sk->sk_data_ready(sk);
 989                 return err;
 990         case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
 991                 sk->sk_write_space(sk);
 992                 break;
 993         case VIRTIO_VSOCK_OP_SHUTDOWN:
 994                 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
 995                         vsk->peer_shutdown |= RCV_SHUTDOWN;
 996                 if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
 997                         vsk->peer_shutdown |= SEND_SHUTDOWN;
 998                 if (vsk->peer_shutdown == SHUTDOWN_MASK &&
 999                     vsock_stream_has_data(vsk) <= 0) {
1000                         sock_set_flag(sk, SOCK_DONE);
1001                         sk->sk_state = TCP_CLOSING;
1002                 }
1003                 if (le32_to_cpu(pkt->hdr.flags))
1004                         sk->sk_state_change(sk);
1005                 break;
1006         case VIRTIO_VSOCK_OP_RST:
1007                 virtio_transport_do_close(vsk, true);
1008                 break;
1009         default:
1010                 err = -EINVAL;
1011                 break;
1012         }
1013
1014         virtio_transport_free_pkt(pkt);
1015         return err;
1016 }
1017
1018 static void
1019 virtio_transport_recv_disconnecting(struct sock *sk,
1020                                     struct virtio_vsock_pkt *pkt)
1021 {
1022         struct vsock_sock *vsk = vsock_sk(sk);
1023
1024         if (le16_to_cpu(pkt->hdr.op) == VIRTIO_VSOCK_OP_RST)
1025                 virtio_transport_do_close(vsk, true);
1026 }
1027
1028 static int
1029 virtio_transport_send_response(struct vsock_sock *vsk,
1030                                struct virtio_vsock_pkt *pkt)
1031 {
1032         struct virtio_vsock_pkt_info info = {
1033                 .op = VIRTIO_VSOCK_OP_RESPONSE,
1034                 .type = VIRTIO_VSOCK_TYPE_STREAM,
1035                 .remote_cid = le64_to_cpu(pkt->hdr.src_cid),
1036                 .remote_port = le32_to_cpu(pkt->hdr.src_port),
1037                 .reply = true,
1038                 .vsk = vsk,
1039         };
1040
1041         return virtio_transport_send_pkt_info(vsk, &info);
1042 }
1043
1044 /* Handle server socket */
1045 static int
1046 virtio_transport_recv_listen(struct sock *sk, struct virtio_vsock_pkt *pkt)
1047 {
1048         struct vsock_sock *vsk = vsock_sk(sk);
1049         struct vsock_sock *vchild;
1050         struct sock *child;
1051
1052         if (le16_to_cpu(pkt->hdr.op) != VIRTIO_VSOCK_OP_REQUEST) {
1053                 virtio_transport_reset(vsk, pkt);
1054                 return -EINVAL;
1055         }
1056
1057         if (sk_acceptq_is_full(sk)) {
1058                 virtio_transport_reset(vsk, pkt);
1059                 return -ENOMEM;
1060         }
1061
1062         child = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1063                                sk->sk_type, 0);
1064         if (!child) {
1065                 virtio_transport_reset(vsk, pkt);
1066                 return -ENOMEM;
1067         }
1068
1069         sk->sk_ack_backlog++;
1070
1071         lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1072
1073         child->sk_state = TCP_ESTABLISHED;
1074
1075         vchild = vsock_sk(child);
1076         vsock_addr_init(&vchild->local_addr, le64_to_cpu(pkt->hdr.dst_cid),
1077                         le32_to_cpu(pkt->hdr.dst_port));
1078         vsock_addr_init(&vchild->remote_addr, le64_to_cpu(pkt->hdr.src_cid),
1079                         le32_to_cpu(pkt->hdr.src_port));
1080
1081         vsock_insert_connected(vchild);
1082         vsock_enqueue_accept(sk, child);
1083         virtio_transport_send_response(vchild, pkt);
1084
1085         release_sock(child);
1086
1087         sk->sk_data_ready(sk);
1088         return 0;
1089 }
1090
1091 static bool virtio_transport_space_update(struct sock *sk,
1092                                           struct virtio_vsock_pkt *pkt)
1093 {
1094         struct vsock_sock *vsk = vsock_sk(sk);
1095         struct virtio_vsock_sock *vvs = vsk->trans;
1096         bool space_available;
1097
1098         /* buf_alloc and fwd_cnt is always included in the hdr */
1099         spin_lock_bh(&vvs->tx_lock);
1100         vvs->peer_buf_alloc = le32_to_cpu(pkt->hdr.buf_alloc);
1101         vvs->peer_fwd_cnt = le32_to_cpu(pkt->hdr.fwd_cnt);
1102         space_available = virtio_transport_has_space(vsk);
1103         spin_unlock_bh(&vvs->tx_lock);
1104         return space_available;
1105 }
1106
1107 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1108  * lock.
1109  */
1110 void virtio_transport_recv_pkt(struct virtio_vsock_pkt *pkt)
1111 {
1112         struct sockaddr_vm src, dst;
1113         struct vsock_sock *vsk;
1114         struct sock *sk;
1115         bool space_available;
1116
1117         vsock_addr_init(&src, le64_to_cpu(pkt->hdr.src_cid),
1118                         le32_to_cpu(pkt->hdr.src_port));
1119         vsock_addr_init(&dst, le64_to_cpu(pkt->hdr.dst_cid),
1120                         le32_to_cpu(pkt->hdr.dst_port));
1121
1122         trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1123                                         dst.svm_cid, dst.svm_port,
1124                                         le32_to_cpu(pkt->hdr.len),
1125                                         le16_to_cpu(pkt->hdr.type),
1126                                         le16_to_cpu(pkt->hdr.op),
1127                                         le32_to_cpu(pkt->hdr.flags),
1128                                         le32_to_cpu(pkt->hdr.buf_alloc),
1129                                         le32_to_cpu(pkt->hdr.fwd_cnt));
1130
1131         if (le16_to_cpu(pkt->hdr.type) != VIRTIO_VSOCK_TYPE_STREAM) {
1132                 (void)virtio_transport_reset_no_sock(pkt);
1133                 goto free_pkt;
1134         }
1135
1136         /* The socket must be in connected or bound table
1137          * otherwise send reset back
1138          */
1139         sk = vsock_find_connected_socket(&src, &dst);
1140         if (!sk) {
1141                 sk = vsock_find_bound_socket(&dst);
1142                 if (!sk) {
1143                         (void)virtio_transport_reset_no_sock(pkt);
1144                         goto free_pkt;
1145                 }
1146         }
1147
1148         vsk = vsock_sk(sk);
1149
1150         space_available = virtio_transport_space_update(sk, pkt);
1151
1152         lock_sock(sk);
1153
1154         /* Update CID in case it has changed after a transport reset event */
1155         vsk->local_addr.svm_cid = dst.svm_cid;
1156
1157         if (space_available)
1158                 sk->sk_write_space(sk);
1159
1160         switch (sk->sk_state) {
1161         case TCP_LISTEN:
1162                 virtio_transport_recv_listen(sk, pkt);
1163                 virtio_transport_free_pkt(pkt);
1164                 break;
1165         case TCP_SYN_SENT:
1166                 virtio_transport_recv_connecting(sk, pkt);
1167                 virtio_transport_free_pkt(pkt);
1168                 break;
1169         case TCP_ESTABLISHED:
1170                 virtio_transport_recv_connected(sk, pkt);
1171                 break;
1172         case TCP_CLOSING:
1173                 virtio_transport_recv_disconnecting(sk, pkt);
1174                 virtio_transport_free_pkt(pkt);
1175                 break;
1176         default:
1177                 virtio_transport_free_pkt(pkt);
1178                 break;
1179         }
1180         release_sock(sk);
1181
1182         /* Release refcnt obtained when we fetched this socket out of the
1183          * bound or connected list.
1184          */
1185         sock_put(sk);
1186         return;
1187
1188 free_pkt:
1189         virtio_transport_free_pkt(pkt);
1190 }
1191 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1192
1193 void virtio_transport_free_pkt(struct virtio_vsock_pkt *pkt)
1194 {
1195         kfree(pkt->buf);
1196         kfree(pkt);
1197 }
1198 EXPORT_SYMBOL_GPL(virtio_transport_free_pkt);
1199
1200 MODULE_LICENSE("GPL v2");
1201 MODULE_AUTHOR("Asias He");
1202 MODULE_DESCRIPTION("common code for virtio vsock");