net/vmw_vsock/virtio_transport_common.c

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