net/vmw_vsock/vmci_transport.c

   1 /*
   2  * VMware vSockets Driver
   3  *
   4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation version 2 and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  */
  15
  16 #include <linux/types.h>
  17 #include <linux/bitops.h>
  18 #include <linux/cred.h>
  19 #include <linux/init.h>
  20 #include <linux/io.h>
  21 #include <linux/kernel.h>
  22 #include <linux/kmod.h>
  23 #include <linux/list.h>
  24 #include <linux/module.h>
  25 #include <linux/mutex.h>
  26 #include <linux/net.h>
  27 #include <linux/poll.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/smp.h>
  30 #include <linux/socket.h>
  31 #include <linux/stddef.h>
  32 #include <linux/unistd.h>
  33 #include <linux/wait.h>
  34 #include <linux/workqueue.h>
  35 #include <net/sock.h>
  36 #include <net/af_vsock.h>
  37
  38 #include "vmci_transport_notify.h"
  39
  40 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
  41 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
  42 static void vmci_transport_peer_detach_cb(u32 sub_id,
  43                                           const struct vmci_event_data *ed,
  44                                           void *client_data);
  45 static void vmci_transport_recv_pkt_work(struct work_struct *work);
  46 static void vmci_transport_cleanup(struct work_struct *work);
  47 static int vmci_transport_recv_listen(struct sock *sk,
  48                                       struct vmci_transport_packet *pkt);
  49 static int vmci_transport_recv_connecting_server(
  50                                         struct sock *sk,
  51                                         struct sock *pending,
  52                                         struct vmci_transport_packet *pkt);
  53 static int vmci_transport_recv_connecting_client(
  54                                         struct sock *sk,
  55                                         struct vmci_transport_packet *pkt);
  56 static int vmci_transport_recv_connecting_client_negotiate(
  57                                         struct sock *sk,
  58                                         struct vmci_transport_packet *pkt);
  59 static int vmci_transport_recv_connecting_client_invalid(
  60                                         struct sock *sk,
  61                                         struct vmci_transport_packet *pkt);
  62 static int vmci_transport_recv_connected(struct sock *sk,
  63                                          struct vmci_transport_packet *pkt);
  64 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
  65 static u16 vmci_transport_new_proto_supported_versions(void);
  66 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
  67                                                   bool old_pkt_proto);
  68
  69 struct vmci_transport_recv_pkt_info {
  70         struct work_struct work;
  71         struct sock *sk;
  72         struct vmci_transport_packet pkt;
  73 };
  74
  75 static LIST_HEAD(vmci_transport_cleanup_list);
  76 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
  77 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
  78
  79 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  80                                                            VMCI_INVALID_ID };
  81 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  82
  83 static int PROTOCOL_OVERRIDE = -1;
  84
  85 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN   128
  86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE       262144
  87 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX   262144
  88
  89 /* The default peer timeout indicates how long we will wait for a peer response
  90  * to a control message.
  91  */
  92 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
  93
  94 /* Helper function to convert from a VMCI error code to a VSock error code. */
  95
  96 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  97 {
  98         switch (vmci_error) {
  99         case VMCI_ERROR_NO_MEM:
 100                 return -ENOMEM;
 101         case VMCI_ERROR_DUPLICATE_ENTRY:
 102         case VMCI_ERROR_ALREADY_EXISTS:
 103                 return -EADDRINUSE;
 104         case VMCI_ERROR_NO_ACCESS:
 105                 return -EPERM;
 106         case VMCI_ERROR_NO_RESOURCES:
 107                 return -ENOBUFS;
 108         case VMCI_ERROR_INVALID_RESOURCE:
 109                 return -EHOSTUNREACH;
 110         case VMCI_ERROR_INVALID_ARGS:
 111         default:
 112                 break;
 113         }
 114         return -EINVAL;
 115 }
 116
 117 static u32 vmci_transport_peer_rid(u32 peer_cid)
 118 {
 119         if (VMADDR_CID_HYPERVISOR == peer_cid)
 120                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 121
 122         return VMCI_TRANSPORT_PACKET_RID;
 123 }
 124
 125 static inline void
 126 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 127                            struct sockaddr_vm *src,
 128                            struct sockaddr_vm *dst,
 129                            u8 type,
 130                            u64 size,
 131                            u64 mode,
 132                            struct vmci_transport_waiting_info *wait,
 133                            u16 proto,
 134                            struct vmci_handle handle)
 135 {
 136         /* We register the stream control handler as an any cid handle so we
 137          * must always send from a source address of VMADDR_CID_ANY
 138          */
 139         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 140                                        VMCI_TRANSPORT_PACKET_RID);
 141         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 142                                        vmci_transport_peer_rid(dst->svm_cid));
 143         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 144         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 145         pkt->type = type;
 146         pkt->src_port = src->svm_port;
 147         pkt->dst_port = dst->svm_port;
 148         memset(&pkt->proto, 0, sizeof(pkt->proto));
 149         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 150
 151         switch (pkt->type) {
 152         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 153                 pkt->u.size = 0;
 154                 break;
 155
 156         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 157         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 158                 pkt->u.size = size;
 159                 break;
 160
 161         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 162         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 163                 pkt->u.handle = handle;
 164                 break;
 165
 166         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 167         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 168         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 169                 pkt->u.size = 0;
 170                 break;
 171
 172         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 173                 pkt->u.mode = mode;
 174                 break;
 175
 176         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 177         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 178                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 179                 break;
 180
 181         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 182         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 183                 pkt->u.size = size;
 184                 pkt->proto = proto;
 185                 break;
 186         }
 187 }
 188
 189 static inline void
 190 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 191                                     struct sockaddr_vm *local,
 192                                     struct sockaddr_vm *remote)
 193 {
 194         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 195         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 196 }
 197
 198 static int
 199 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 200                                   struct sockaddr_vm *src,
 201                                   struct sockaddr_vm *dst,
 202                                   enum vmci_transport_packet_type type,
 203                                   u64 size,
 204                                   u64 mode,
 205                                   struct vmci_transport_waiting_info *wait,
 206                                   u16 proto,
 207                                   struct vmci_handle handle,
 208                                   bool convert_error)
 209 {
 210         int err;
 211
 212         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 213                                    proto, handle);
 214         err = vmci_datagram_send(&pkt->dg);
 215         if (convert_error && (err < 0))
 216                 return vmci_transport_error_to_vsock_error(err);
 217
 218         return err;
 219 }
 220
 221 static int
 222 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 223                                       enum vmci_transport_packet_type type,
 224                                       u64 size,
 225                                       u64 mode,
 226                                       struct vmci_transport_waiting_info *wait,
 227                                       struct vmci_handle handle)
 228 {
 229         struct vmci_transport_packet reply;
 230         struct sockaddr_vm src, dst;
 231
 232         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 233                 return 0;
 234         } else {
 235                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 236                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 237                                                          type,
 238                                                          size, mode, wait,
 239                                                          VSOCK_PROTO_INVALID,
 240                                                          handle, true);
 241         }
 242 }
 243
 244 static int
 245 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 246                                    struct sockaddr_vm *dst,
 247                                    enum vmci_transport_packet_type type,
 248                                    u64 size,
 249                                    u64 mode,
 250                                    struct vmci_transport_waiting_info *wait,
 251                                    struct vmci_handle handle)
 252 {
 253         /* Note that it is safe to use a single packet across all CPUs since
 254          * two tasklets of the same type are guaranteed to not ever run
 255          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 256          * we can use per-cpu packets.
 257          */
 258         static struct vmci_transport_packet pkt;
 259
 260         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 261                                                  size, mode, wait,
 262                                                  VSOCK_PROTO_INVALID, handle,
 263                                                  false);
 264 }
 265
 266 static int
 267 vmci_transport_send_control_pkt(struct sock *sk,
 268                                 enum vmci_transport_packet_type type,
 269                                 u64 size,
 270                                 u64 mode,
 271                                 struct vmci_transport_waiting_info *wait,
 272                                 u16 proto,
 273                                 struct vmci_handle handle)
 274 {
 275         struct vmci_transport_packet *pkt;
 276         struct vsock_sock *vsk;
 277         int err;
 278
 279         vsk = vsock_sk(sk);
 280
 281         if (!vsock_addr_bound(&vsk->local_addr))
 282                 return -EINVAL;
 283
 284         if (!vsock_addr_bound(&vsk->remote_addr))
 285                 return -EINVAL;
 286
 287         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 288         if (!pkt)
 289                 return -ENOMEM;
 290
 291         err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
 292                                                 &vsk->remote_addr, type, size,
 293                                                 mode, wait, proto, handle,
 294                                                 true);
 295         kfree(pkt);
 296
 297         return err;
 298 }
 299
 300 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 301                                         struct sockaddr_vm *src,
 302                                         struct vmci_transport_packet *pkt)
 303 {
 304         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 305                 return 0;
 306         return vmci_transport_send_control_pkt_bh(
 307                                         dst, src,
 308                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 309                                         0, NULL, VMCI_INVALID_HANDLE);
 310 }
 311
 312 static int vmci_transport_send_reset(struct sock *sk,
 313                                      struct vmci_transport_packet *pkt)
 314 {
 315         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 316                 return 0;
 317         return vmci_transport_send_control_pkt(sk,
 318                                         VMCI_TRANSPORT_PACKET_TYPE_RST,
 319                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 320                                         VMCI_INVALID_HANDLE);
 321 }
 322
 323 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 324 {
 325         return vmci_transport_send_control_pkt(
 326                                         sk,
 327                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 328                                         size, 0, NULL,
 329                                         VSOCK_PROTO_INVALID,
 330                                         VMCI_INVALID_HANDLE);
 331 }
 332
 333 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 334                                           u16 version)
 335 {
 336         return vmci_transport_send_control_pkt(
 337                                         sk,
 338                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 339                                         size, 0, NULL, version,
 340                                         VMCI_INVALID_HANDLE);
 341 }
 342
 343 static int vmci_transport_send_qp_offer(struct sock *sk,
 344                                         struct vmci_handle handle)
 345 {
 346         return vmci_transport_send_control_pkt(
 347                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 348                                         0, NULL,
 349                                         VSOCK_PROTO_INVALID, handle);
 350 }
 351
 352 static int vmci_transport_send_attach(struct sock *sk,
 353                                       struct vmci_handle handle)
 354 {
 355         return vmci_transport_send_control_pkt(
 356                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 357                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 358                                         handle);
 359 }
 360
 361 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 362 {
 363         return vmci_transport_reply_control_pkt_fast(
 364                                                 pkt,
 365                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 366                                                 0, 0, NULL,
 367                                                 VMCI_INVALID_HANDLE);
 368 }
 369
 370 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 371                                           struct sockaddr_vm *src)
 372 {
 373         return vmci_transport_send_control_pkt_bh(
 374                                         dst, src,
 375                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 376                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 377 }
 378
 379 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 380                                  struct sockaddr_vm *src)
 381 {
 382         return vmci_transport_send_control_pkt_bh(
 383                                         dst, src,
 384                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 385                                         0, NULL, VMCI_INVALID_HANDLE);
 386 }
 387
 388 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 389                                 struct sockaddr_vm *src)
 390 {
 391         return vmci_transport_send_control_pkt_bh(
 392                                         dst, src,
 393                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 394                                         0, NULL, VMCI_INVALID_HANDLE);
 395 }
 396
 397 int vmci_transport_send_wrote(struct sock *sk)
 398 {
 399         return vmci_transport_send_control_pkt(
 400                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 401                                         0, NULL, VSOCK_PROTO_INVALID,
 402                                         VMCI_INVALID_HANDLE);
 403 }
 404
 405 int vmci_transport_send_read(struct sock *sk)
 406 {
 407         return vmci_transport_send_control_pkt(
 408                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 409                                         0, NULL, VSOCK_PROTO_INVALID,
 410                                         VMCI_INVALID_HANDLE);
 411 }
 412
 413 int vmci_transport_send_waiting_write(struct sock *sk,
 414                                       struct vmci_transport_waiting_info *wait)
 415 {
 416         return vmci_transport_send_control_pkt(
 417                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 418                                 0, 0, wait, VSOCK_PROTO_INVALID,
 419                                 VMCI_INVALID_HANDLE);
 420 }
 421
 422 int vmci_transport_send_waiting_read(struct sock *sk,
 423                                      struct vmci_transport_waiting_info *wait)
 424 {
 425         return vmci_transport_send_control_pkt(
 426                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 427                                 0, 0, wait, VSOCK_PROTO_INVALID,
 428                                 VMCI_INVALID_HANDLE);
 429 }
 430
 431 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 432 {
 433         return vmci_transport_send_control_pkt(
 434                                         &vsk->sk,
 435                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 436                                         0, mode, NULL,
 437                                         VSOCK_PROTO_INVALID,
 438                                         VMCI_INVALID_HANDLE);
 439 }
 440
 441 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 442 {
 443         return vmci_transport_send_control_pkt(sk,
 444                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 445                                         size, 0, NULL,
 446                                         VSOCK_PROTO_INVALID,
 447                                         VMCI_INVALID_HANDLE);
 448 }
 449
 450 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 451                                              u16 version)
 452 {
 453         return vmci_transport_send_control_pkt(
 454                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 455                                         size, 0, NULL, version,
 456                                         VMCI_INVALID_HANDLE);
 457 }
 458
 459 static struct sock *vmci_transport_get_pending(
 460                                         struct sock *listener,
 461                                         struct vmci_transport_packet *pkt)
 462 {
 463         struct vsock_sock *vlistener;
 464         struct vsock_sock *vpending;
 465         struct sock *pending;
 466         struct sockaddr_vm src;
 467
 468         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 469
 470         vlistener = vsock_sk(listener);
 471
 472         list_for_each_entry(vpending, &vlistener->pending_links,
 473                             pending_links) {
 474                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 475                     pkt->dst_port == vpending->local_addr.svm_port) {
 476                         pending = sk_vsock(vpending);
 477                         sock_hold(pending);
 478                         goto found;
 479                 }
 480         }
 481
 482         pending = NULL;
 483 found:
 484         return pending;
 485
 486 }
 487
 488 static void vmci_transport_release_pending(struct sock *pending)
 489 {
 490         sock_put(pending);
 491 }
 492
 493 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 494  * trusted sockets 2) sockets from applications running as the same user as the
 495  * VM (this is only true for the host side and only when using hosted products)
 496  */
 497
 498 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 499 {
 500         return vsock->trusted ||
 501                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 502 }
 503
 504 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 505  * only if it is trusted as described in vmci_transport_is_trusted.
 506  */
 507
 508 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 509 {
 510         if (VMADDR_CID_HYPERVISOR == peer_cid)
 511                 return true;
 512
 513         if (vsock->cached_peer != peer_cid) {
 514                 vsock->cached_peer = peer_cid;
 515                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 516                     (vmci_context_get_priv_flags(peer_cid) &
 517                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 518                         vsock->cached_peer_allow_dgram = false;
 519                 } else {
 520                         vsock->cached_peer_allow_dgram = true;
 521                 }
 522         }
 523
 524         return vsock->cached_peer_allow_dgram;
 525 }
 526
 527 static int
 528 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 529                                 struct vmci_handle *handle,
 530                                 u64 produce_size,
 531                                 u64 consume_size,
 532                                 u32 peer, u32 flags, bool trusted)
 533 {
 534         int err = 0;
 535
 536         if (trusted) {
 537                 /* Try to allocate our queue pair as trusted. This will only
 538                  * work if vsock is running in the host.
 539                  */
 540
 541                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 542                                        consume_size,
 543                                        peer, flags,
 544                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 545                 if (err != VMCI_ERROR_NO_ACCESS)
 546                         goto out;
 547
 548         }
 549
 550         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 551                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 552 out:
 553         if (err < 0) {
 554                 pr_err("Could not attach to queue pair with %d\n",
 555                        err);
 556                 err = vmci_transport_error_to_vsock_error(err);
 557         }
 558
 559         return err;
 560 }
 561
 562 static int
 563 vmci_transport_datagram_create_hnd(u32 resource_id,
 564                                    u32 flags,
 565                                    vmci_datagram_recv_cb recv_cb,
 566                                    void *client_data,
 567                                    struct vmci_handle *out_handle)
 568 {
 569         int err = 0;
 570
 571         /* Try to allocate our datagram handler as trusted. This will only work
 572          * if vsock is running in the host.
 573          */
 574
 575         err = vmci_datagram_create_handle_priv(resource_id, flags,
 576                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 577                                                recv_cb,
 578                                                client_data, out_handle);
 579
 580         if (err == VMCI_ERROR_NO_ACCESS)
 581                 err = vmci_datagram_create_handle(resource_id, flags,
 582                                                   recv_cb, client_data,
 583                                                   out_handle);
 584
 585         return err;
 586 }
 587
 588 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 589  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 590  * sleep it should defer that work to a work queue.
 591  */
 592
 593 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 594 {
 595         struct sock *sk;
 596         size_t size;
 597         struct sk_buff *skb;
 598         struct vsock_sock *vsk;
 599
 600         sk = (struct sock *)data;
 601
 602         /* This handler is privileged when this module is running on the host.
 603          * We will get datagrams from all endpoints (even VMs that are in a
 604          * restricted context). If we get one from a restricted context then
 605          * the destination socket must be trusted.
 606          *
 607          * NOTE: We access the socket struct without holding the lock here.
 608          * This is ok because the field we are interested is never modified
 609          * outside of the create and destruct socket functions.
 610          */
 611         vsk = vsock_sk(sk);
 612         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 613                 return VMCI_ERROR_NO_ACCESS;
 614
 615         size = VMCI_DG_SIZE(dg);
 616
 617         /* Attach the packet to the socket's receive queue as an sk_buff. */
 618         skb = alloc_skb(size, GFP_ATOMIC);
 619         if (!skb)
 620                 return VMCI_ERROR_NO_MEM;
 621
 622         /* sk_receive_skb() will do a sock_put(), so hold here. */
 623         sock_hold(sk);
 624         skb_put(skb, size);
 625         memcpy(skb->data, dg, size);
 626         sk_receive_skb(sk, skb, 0);
 627
 628         return VMCI_SUCCESS;
 629 }
 630
 631 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 632 {
 633         static const u32 non_socket_contexts[] = {
 634                 VMADDR_CID_RESERVED,
 635         };
 636         int i;
 637
 638         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 639
 640         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 641                 if (cid == non_socket_contexts[i])
 642                         return false;
 643         }
 644
 645         return true;
 646 }
 647
 648 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 649  * interrupt fires.  This is run in bottom-half context but it defers most of
 650  * its work to the packet handling work queue.
 651  */
 652
 653 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 654 {
 655         struct sock *sk;
 656         struct sockaddr_vm dst;
 657         struct sockaddr_vm src;
 658         struct vmci_transport_packet *pkt;
 659         struct vsock_sock *vsk;
 660         bool bh_process_pkt;
 661         int err;
 662
 663         sk = NULL;
 664         err = VMCI_SUCCESS;
 665         bh_process_pkt = false;
 666
 667         /* Ignore incoming packets from contexts without sockets, or resources
 668          * that aren't vsock implementations.
 669          */
 670
 671         if (!vmci_transport_stream_allow(dg->src.context, -1)
 672             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 673                 return VMCI_ERROR_NO_ACCESS;
 674
 675         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 676                 /* Drop datagrams that do not contain full VSock packets. */
 677                 return VMCI_ERROR_INVALID_ARGS;
 678
 679         pkt = (struct vmci_transport_packet *)dg;
 680
 681         /* Find the socket that should handle this packet.  First we look for a
 682          * connected socket and if there is none we look for a socket bound to
 683          * the destintation address.
 684          */
 685         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 686         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 687
 688         sk = vsock_find_connected_socket(&src, &dst);
 689         if (!sk) {
 690                 sk = vsock_find_bound_socket(&dst);
 691                 if (!sk) {
 692                         /* We could not find a socket for this specified
 693                          * address.  If this packet is a RST, we just drop it.
 694                          * If it is another packet, we send a RST.  Note that
 695                          * we do not send a RST reply to RSTs so that we do not
 696                          * continually send RSTs between two endpoints.
 697                          *
 698                          * Note that since this is a reply, dst is src and src
 699                          * is dst.
 700                          */
 701                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 702                                 pr_err("unable to send reset\n");
 703
 704                         err = VMCI_ERROR_NOT_FOUND;
 705                         goto out;
 706                 }
 707         }
 708
 709         /* If the received packet type is beyond all types known to this
 710          * implementation, reply with an invalid message.  Hopefully this will
 711          * help when implementing backwards compatibility in the future.
 712          */
 713         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 714                 vmci_transport_send_invalid_bh(&dst, &src);
 715                 err = VMCI_ERROR_INVALID_ARGS;
 716                 goto out;
 717         }
 718
 719         /* This handler is privileged when this module is running on the host.
 720          * We will get datagram connect requests from all endpoints (even VMs
 721          * that are in a restricted context). If we get one from a restricted
 722          * context then the destination socket must be trusted.
 723          *
 724          * NOTE: We access the socket struct without holding the lock here.
 725          * This is ok because the field we are interested is never modified
 726          * outside of the create and destruct socket functions.
 727          */
 728         vsk = vsock_sk(sk);
 729         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 730                 err = VMCI_ERROR_NO_ACCESS;
 731                 goto out;
 732         }
 733
 734         /* We do most everything in a work queue, but let's fast path the
 735          * notification of reads and writes to help data transfer performance.
 736          * We can only do this if there is no process context code executing
 737          * for this socket since that may change the state.
 738          */
 739         bh_lock_sock(sk);
 740
 741         if (!sock_owned_by_user(sk)) {
 742                 /* The local context ID may be out of date, update it. */
 743                 vsk->local_addr.svm_cid = dst.svm_cid;
 744
 745                 if (sk->sk_state == TCP_ESTABLISHED)
 746                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 747                                         sk, pkt, true, &dst, &src,
 748                                         &bh_process_pkt);
 749         }
 750
 751         bh_unlock_sock(sk);
 752
 753         if (!bh_process_pkt) {
 754                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 755
 756                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 757                 if (!recv_pkt_info) {
 758                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 759                                 pr_err("unable to send reset\n");
 760
 761                         err = VMCI_ERROR_NO_MEM;
 762                         goto out;
 763                 }
 764
 765                 recv_pkt_info->sk = sk;
 766                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 767                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 768
 769                 schedule_work(&recv_pkt_info->work);
 770                 /* Clear sk so that the reference count incremented by one of
 771                  * the Find functions above is not decremented below.  We need
 772                  * that reference count for the packet handler we've scheduled
 773                  * to run.
 774                  */
 775                 sk = NULL;
 776         }
 777
 778 out:
 779         if (sk)
 780                 sock_put(sk);
 781
 782         return err;
 783 }
 784
 785 static void vmci_transport_handle_detach(struct sock *sk)
 786 {
 787         struct vsock_sock *vsk;
 788
 789         vsk = vsock_sk(sk);
 790         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 791                 sock_set_flag(sk, SOCK_DONE);
 792
 793                 /* On a detach the peer will not be sending or receiving
 794                  * anymore.
 795                  */
 796                 vsk->peer_shutdown = SHUTDOWN_MASK;
 797
 798                 /* We should not be sending anymore since the peer won't be
 799                  * there to receive, but we can still receive if there is data
 800                  * left in our consume queue.
 801                  */
 802                 if (vsock_stream_has_data(vsk) <= 0) {
 803                         sk->sk_state = TCP_CLOSE;
 804
 805                         if (sk->sk_state == TCP_SYN_SENT) {
 806                                 /* The peer may detach from a queue pair while
 807                                  * we are still in the connecting state, i.e.,
 808                                  * if the peer VM is killed after attaching to
 809                                  * a queue pair, but before we complete the
 810                                  * handshake. In that case, we treat the detach
 811                                  * event like a reset.
 812                                  */
 813
 814                                 sk->sk_err = ECONNRESET;
 815                                 sk->sk_error_report(sk);
 816                                 return;
 817                         }
 818                 }
 819                 sk->sk_state_change(sk);
 820         }
 821 }
 822
 823 static void vmci_transport_peer_detach_cb(u32 sub_id,
 824                                           const struct vmci_event_data *e_data,
 825                                           void *client_data)
 826 {
 827         struct vmci_transport *trans = client_data;
 828         const struct vmci_event_payload_qp *e_payload;
 829
 830         e_payload = vmci_event_data_const_payload(e_data);
 831
 832         /* XXX This is lame, we should provide a way to lookup sockets by
 833          * qp_handle.
 834          */
 835         if (vmci_handle_is_invalid(e_payload->handle) ||
 836             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 837                 return;
 838
 839         /* We don't ask for delayed CBs when we subscribe to this event (we
 840          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 841          * guarantees in that case about what context we might be running in,
 842          * so it could be BH or process, blockable or non-blockable.  So we
 843          * need to account for all possible contexts here.
 844          */
 845         spin_lock_bh(&trans->lock);
 846         if (!trans->sk)
 847                 goto out;
 848
 849         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 850          * where trans->sk isn't locked.
 851          */
 852         bh_lock_sock(trans->sk);
 853
 854         vmci_transport_handle_detach(trans->sk);
 855
 856         bh_unlock_sock(trans->sk);
 857  out:
 858         spin_unlock_bh(&trans->lock);
 859 }
 860
 861 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 862                                          const struct vmci_event_data *e_data,
 863                                          void *client_data)
 864 {
 865         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 866 }
 867
 868 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 869 {
 870         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 871         struct vmci_transport_packet *pkt;
 872         struct sock *sk;
 873
 874         recv_pkt_info =
 875                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 876         sk = recv_pkt_info->sk;
 877         pkt = &recv_pkt_info->pkt;
 878
 879         lock_sock(sk);
 880
 881         /* The local context ID may be out of date. */
 882         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 883
 884         switch (sk->sk_state) {
 885         case TCP_LISTEN:
 886                 vmci_transport_recv_listen(sk, pkt);
 887                 break;
 888         case TCP_SYN_SENT:
 889                 /* Processing of pending connections for servers goes through
 890                  * the listening socket, so see vmci_transport_recv_listen()
 891                  * for that path.
 892                  */
 893                 vmci_transport_recv_connecting_client(sk, pkt);
 894                 break;
 895         case TCP_ESTABLISHED:
 896                 vmci_transport_recv_connected(sk, pkt);
 897                 break;
 898         default:
 899                 /* Because this function does not run in the same context as
 900                  * vmci_transport_recv_stream_cb it is possible that the
 901                  * socket has closed. We need to let the other side know or it
 902                  * could be sitting in a connect and hang forever. Send a
 903                  * reset to prevent that.
 904                  */
 905                 vmci_transport_send_reset(sk, pkt);
 906                 break;
 907         }
 908
 909         release_sock(sk);
 910         kfree(recv_pkt_info);
 911         /* Release reference obtained in the stream callback when we fetched
 912          * this socket out of the bound or connected list.
 913          */
 914         sock_put(sk);
 915 }
 916
 917 static int vmci_transport_recv_listen(struct sock *sk,
 918                                       struct vmci_transport_packet *pkt)
 919 {
 920         struct sock *pending;
 921         struct vsock_sock *vpending;
 922         int err;
 923         u64 qp_size;
 924         bool old_request = false;
 925         bool old_pkt_proto = false;
 926
 927         err = 0;
 928
 929         /* Because we are in the listen state, we could be receiving a packet
 930          * for ourself or any previous connection requests that we received.
 931          * If it's the latter, we try to find a socket in our list of pending
 932          * connections and, if we do, call the appropriate handler for the
 933          * state that that socket is in.  Otherwise we try to service the
 934          * connection request.
 935          */
 936         pending = vmci_transport_get_pending(sk, pkt);
 937         if (pending) {
 938                 lock_sock(pending);
 939
 940                 /* The local context ID may be out of date. */
 941                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 942
 943                 switch (pending->sk_state) {
 944                 case TCP_SYN_SENT:
 945                         err = vmci_transport_recv_connecting_server(sk,
 946                                                                     pending,
 947                                                                     pkt);
 948                         break;
 949                 default:
 950                         vmci_transport_send_reset(pending, pkt);
 951                         err = -EINVAL;
 952                 }
 953
 954                 if (err < 0)
 955                         vsock_remove_pending(sk, pending);
 956
 957                 release_sock(pending);
 958                 vmci_transport_release_pending(pending);
 959
 960                 return err;
 961         }
 962
 963         /* The listen state only accepts connection requests.  Reply with a
 964          * reset unless we received a reset.
 965          */
 966
 967         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 968               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 969                 vmci_transport_reply_reset(pkt);
 970                 return -EINVAL;
 971         }
 972
 973         if (pkt->u.size == 0) {
 974                 vmci_transport_reply_reset(pkt);
 975                 return -EINVAL;
 976         }
 977
 978         /* If this socket can't accommodate this connection request, we send a
 979          * reset.  Otherwise we create and initialize a child socket and reply
 980          * with a connection negotiation.
 981          */
 982         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
 983                 vmci_transport_reply_reset(pkt);
 984                 return -ECONNREFUSED;
 985         }
 986
 987         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
 988                                  sk->sk_type, 0);
 989         if (!pending) {
 990                 vmci_transport_send_reset(sk, pkt);
 991                 return -ENOMEM;
 992         }
 993
 994         vpending = vsock_sk(pending);
 995
 996         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
 997                         pkt->dst_port);
 998         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
 999                         pkt->src_port);
1000
1001         /* If the proposed size fits within our min/max, accept it. Otherwise
1002          * propose our own size.
1003          */
1004         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1005             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1006                 qp_size = pkt->u.size;
1007         } else {
1008                 qp_size = vmci_trans(vpending)->queue_pair_size;
1009         }
1010
1011         /* Figure out if we are using old or new requests based on the
1012          * overrides pkt types sent by our peer.
1013          */
1014         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1015                 old_request = old_pkt_proto;
1016         } else {
1017                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1018                         old_request = true;
1019                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1020                         old_request = false;
1021
1022         }
1023
1024         if (old_request) {
1025                 /* Handle a REQUEST (or override) */
1026                 u16 version = VSOCK_PROTO_INVALID;
1027                 if (vmci_transport_proto_to_notify_struct(
1028                         pending, &version, true))
1029                         err = vmci_transport_send_negotiate(pending, qp_size);
1030                 else
1031                         err = -EINVAL;
1032
1033         } else {
1034                 /* Handle a REQUEST2 (or override) */
1035                 int proto_int = pkt->proto;
1036                 int pos;
1037                 u16 active_proto_version = 0;
1038
1039                 /* The list of possible protocols is the intersection of all
1040                  * protocols the client supports ... plus all the protocols we
1041                  * support.
1042                  */
1043                 proto_int &= vmci_transport_new_proto_supported_versions();
1044
1045                 /* We choose the highest possible protocol version and use that
1046                  * one.
1047                  */
1048                 pos = fls(proto_int);
1049                 if (pos) {
1050                         active_proto_version = (1 << (pos - 1));
1051                         if (vmci_transport_proto_to_notify_struct(
1052                                 pending, &active_proto_version, false))
1053                                 err = vmci_transport_send_negotiate2(pending,
1054                                                         qp_size,
1055                                                         active_proto_version);
1056                         else
1057                                 err = -EINVAL;
1058
1059                 } else {
1060                         err = -EINVAL;
1061                 }
1062         }
1063
1064         if (err < 0) {
1065                 vmci_transport_send_reset(sk, pkt);
1066                 sock_put(pending);
1067                 err = vmci_transport_error_to_vsock_error(err);
1068                 goto out;
1069         }
1070
1071         vsock_add_pending(sk, pending);
1072         sk->sk_ack_backlog++;
1073
1074         pending->sk_state = TCP_SYN_SENT;
1075         vmci_trans(vpending)->produce_size =
1076                 vmci_trans(vpending)->consume_size = qp_size;
1077         vmci_trans(vpending)->queue_pair_size = qp_size;
1078
1079         vmci_trans(vpending)->notify_ops->process_request(pending);
1080
1081         /* We might never receive another message for this socket and it's not
1082          * connected to any process, so we have to ensure it gets cleaned up
1083          * ourself.  Our delayed work function will take care of that.  Note
1084          * that we do not ever cancel this function since we have few
1085          * guarantees about its state when calling cancel_delayed_work().
1086          * Instead we hold a reference on the socket for that function and make
1087          * it capable of handling cases where it needs to do nothing but
1088          * release that reference.
1089          */
1090         vpending->listener = sk;
1091         sock_hold(sk);
1092         sock_hold(pending);
1093         INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1094         schedule_delayed_work(&vpending->dwork, HZ);
1095
1096 out:
1097         return err;
1098 }
1099
1100 static int
1101 vmci_transport_recv_connecting_server(struct sock *listener,
1102                                       struct sock *pending,
1103                                       struct vmci_transport_packet *pkt)
1104 {
1105         struct vsock_sock *vpending;
1106         struct vmci_handle handle;
1107         struct vmci_qp *qpair;
1108         bool is_local;
1109         u32 flags;
1110         u32 detach_sub_id;
1111         int err;
1112         int skerr;
1113
1114         vpending = vsock_sk(pending);
1115         detach_sub_id = VMCI_INVALID_ID;
1116
1117         switch (pkt->type) {
1118         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1119                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1120                         vmci_transport_send_reset(pending, pkt);
1121                         skerr = EPROTO;
1122                         err = -EINVAL;
1123                         goto destroy;
1124                 }
1125                 break;
1126         default:
1127                 /* Close and cleanup the connection. */
1128                 vmci_transport_send_reset(pending, pkt);
1129                 skerr = EPROTO;
1130                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1131                 goto destroy;
1132         }
1133
1134         /* In order to complete the connection we need to attach to the offered
1135          * queue pair and send an attach notification.  We also subscribe to the
1136          * detach event so we know when our peer goes away, and we do that
1137          * before attaching so we don't miss an event.  If all this succeeds,
1138          * we update our state and wakeup anything waiting in accept() for a
1139          * connection.
1140          */
1141
1142         /* We don't care about attach since we ensure the other side has
1143          * attached by specifying the ATTACH_ONLY flag below.
1144          */
1145         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1146                                    vmci_transport_peer_detach_cb,
1147                                    vmci_trans(vpending), &detach_sub_id);
1148         if (err < VMCI_SUCCESS) {
1149                 vmci_transport_send_reset(pending, pkt);
1150                 err = vmci_transport_error_to_vsock_error(err);
1151                 skerr = -err;
1152                 goto destroy;
1153         }
1154
1155         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1156
1157         /* Now attach to the queue pair the client created. */
1158         handle = pkt->u.handle;
1159
1160         /* vpending->local_addr always has a context id so we do not need to
1161          * worry about VMADDR_CID_ANY in this case.
1162          */
1163         is_local =
1164             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1165         flags = VMCI_QPFLAG_ATTACH_ONLY;
1166         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1167
1168         err = vmci_transport_queue_pair_alloc(
1169                                         &qpair,
1170                                         &handle,
1171                                         vmci_trans(vpending)->produce_size,
1172                                         vmci_trans(vpending)->consume_size,
1173                                         pkt->dg.src.context,
1174                                         flags,
1175                                         vmci_transport_is_trusted(
1176                                                 vpending,
1177                                                 vpending->remote_addr.svm_cid));
1178         if (err < 0) {
1179                 vmci_transport_send_reset(pending, pkt);
1180                 skerr = -err;
1181                 goto destroy;
1182         }
1183
1184         vmci_trans(vpending)->qp_handle = handle;
1185         vmci_trans(vpending)->qpair = qpair;
1186
1187         /* When we send the attach message, we must be ready to handle incoming
1188          * control messages on the newly connected socket. So we move the
1189          * pending socket to the connected state before sending the attach
1190          * message. Otherwise, an incoming packet triggered by the attach being
1191          * received by the peer may be processed concurrently with what happens
1192          * below after sending the attach message, and that incoming packet
1193          * will find the listening socket instead of the (currently) pending
1194          * socket. Note that enqueueing the socket increments the reference
1195          * count, so even if a reset comes before the connection is accepted,
1196          * the socket will be valid until it is removed from the queue.
1197          *
1198          * If we fail sending the attach below, we remove the socket from the
1199          * connected list and move the socket to TCP_CLOSE before
1200          * releasing the lock, so a pending slow path processing of an incoming
1201          * packet will not see the socket in the connected state in that case.
1202          */
1203         pending->sk_state = TCP_ESTABLISHED;
1204
1205         vsock_insert_connected(vpending);
1206
1207         /* Notify our peer of our attach. */
1208         err = vmci_transport_send_attach(pending, handle);
1209         if (err < 0) {
1210                 vsock_remove_connected(vpending);
1211                 pr_err("Could not send attach\n");
1212                 vmci_transport_send_reset(pending, pkt);
1213                 err = vmci_transport_error_to_vsock_error(err);
1214                 skerr = -err;
1215                 goto destroy;
1216         }
1217
1218         /* We have a connection. Move the now connected socket from the
1219          * listener's pending list to the accept queue so callers of accept()
1220          * can find it.
1221          */
1222         vsock_remove_pending(listener, pending);
1223         vsock_enqueue_accept(listener, pending);
1224
1225         /* Callers of accept() will be be waiting on the listening socket, not
1226          * the pending socket.
1227          */
1228         listener->sk_data_ready(listener);
1229
1230         return 0;
1231
1232 destroy:
1233         pending->sk_err = skerr;
1234         pending->sk_state = TCP_CLOSE;
1235         /* As long as we drop our reference, all necessary cleanup will handle
1236          * when the cleanup function drops its reference and our destruct
1237          * implementation is called.  Note that since the listen handler will
1238          * remove pending from the pending list upon our failure, the cleanup
1239          * function won't drop the additional reference, which is why we do it
1240          * here.
1241          */
1242         sock_put(pending);
1243
1244         return err;
1245 }
1246
1247 static int
1248 vmci_transport_recv_connecting_client(struct sock *sk,
1249                                       struct vmci_transport_packet *pkt)
1250 {
1251         struct vsock_sock *vsk;
1252         int err;
1253         int skerr;
1254
1255         vsk = vsock_sk(sk);
1256
1257         switch (pkt->type) {
1258         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1259                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1260                     !vmci_handle_is_equal(pkt->u.handle,
1261                                           vmci_trans(vsk)->qp_handle)) {
1262                         skerr = EPROTO;
1263                         err = -EINVAL;
1264                         goto destroy;
1265                 }
1266
1267                 /* Signify the socket is connected and wakeup the waiter in
1268                  * connect(). Also place the socket in the connected table for
1269                  * accounting (it can already be found since it's in the bound
1270                  * table).
1271                  */
1272                 sk->sk_state = TCP_ESTABLISHED;
1273                 sk->sk_socket->state = SS_CONNECTED;
1274                 vsock_insert_connected(vsk);
1275                 sk->sk_state_change(sk);
1276
1277                 break;
1278         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1279         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1280                 if (pkt->u.size == 0
1281                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1282                     || pkt->src_port != vsk->remote_addr.svm_port
1283                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1284                     || vmci_trans(vsk)->qpair
1285                     || vmci_trans(vsk)->produce_size != 0
1286                     || vmci_trans(vsk)->consume_size != 0
1287                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1288                         skerr = EPROTO;
1289                         err = -EINVAL;
1290
1291                         goto destroy;
1292                 }
1293
1294                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1295                 if (err) {
1296                         skerr = -err;
1297                         goto destroy;
1298                 }
1299
1300                 break;
1301         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1302                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1303                 if (err) {
1304                         skerr = -err;
1305                         goto destroy;
1306                 }
1307
1308                 break;
1309         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1310                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1311                  * continue processing here after they sent an INVALID packet.
1312                  * This meant that we got a RST after the INVALID. We ignore a
1313                  * RST after an INVALID. The common code doesn't send the RST
1314                  * ... so we can hang if an old version of the common code
1315                  * fails between getting a REQUEST and sending an OFFER back.
1316                  * Not much we can do about it... except hope that it doesn't
1317                  * happen.
1318                  */
1319                 if (vsk->ignore_connecting_rst) {
1320                         vsk->ignore_connecting_rst = false;
1321                 } else {
1322                         skerr = ECONNRESET;
1323                         err = 0;
1324                         goto destroy;
1325                 }
1326
1327                 break;
1328         default:
1329                 /* Close and cleanup the connection. */
1330                 skerr = EPROTO;
1331                 err = -EINVAL;
1332                 goto destroy;
1333         }
1334
1335         return 0;
1336
1337 destroy:
1338         vmci_transport_send_reset(sk, pkt);
1339
1340         sk->sk_state = TCP_CLOSE;
1341         sk->sk_err = skerr;
1342         sk->sk_error_report(sk);
1343         return err;
1344 }
1345
1346 static int vmci_transport_recv_connecting_client_negotiate(
1347                                         struct sock *sk,
1348                                         struct vmci_transport_packet *pkt)
1349 {
1350         int err;
1351         struct vsock_sock *vsk;
1352         struct vmci_handle handle;
1353         struct vmci_qp *qpair;
1354         u32 detach_sub_id;
1355         bool is_local;
1356         u32 flags;
1357         bool old_proto = true;
1358         bool old_pkt_proto;
1359         u16 version;
1360
1361         vsk = vsock_sk(sk);
1362         handle = VMCI_INVALID_HANDLE;
1363         detach_sub_id = VMCI_INVALID_ID;
1364
1365         /* If we have gotten here then we should be past the point where old
1366          * linux vsock could have sent the bogus rst.
1367          */
1368         vsk->sent_request = false;
1369         vsk->ignore_connecting_rst = false;
1370
1371         /* Verify that we're OK with the proposed queue pair size */
1372         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1373             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1374                 err = -EINVAL;
1375                 goto destroy;
1376         }
1377
1378         /* At this point we know the CID the peer is using to talk to us. */
1379
1380         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1381                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1382
1383         /* Setup the notify ops to be the highest supported version that both
1384          * the server and the client support.
1385          */
1386
1387         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1388                 old_proto = old_pkt_proto;
1389         } else {
1390                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1391                         old_proto = true;
1392                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1393                         old_proto = false;
1394
1395         }
1396
1397         if (old_proto)
1398                 version = VSOCK_PROTO_INVALID;
1399         else
1400                 version = pkt->proto;
1401
1402         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1403                 err = -EINVAL;
1404                 goto destroy;
1405         }
1406
1407         /* Subscribe to detach events first.
1408          *
1409          * XXX We attach once for each queue pair created for now so it is easy
1410          * to find the socket (it's provided), but later we should only
1411          * subscribe once and add a way to lookup sockets by queue pair handle.
1412          */
1413         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1414                                    vmci_transport_peer_detach_cb,
1415                                    vmci_trans(vsk), &detach_sub_id);
1416         if (err < VMCI_SUCCESS) {
1417                 err = vmci_transport_error_to_vsock_error(err);
1418                 goto destroy;
1419         }
1420
1421         /* Make VMCI select the handle for us. */
1422         handle = VMCI_INVALID_HANDLE;
1423         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1424         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1425
1426         err = vmci_transport_queue_pair_alloc(&qpair,
1427                                               &handle,
1428                                               pkt->u.size,
1429                                               pkt->u.size,
1430                                               vsk->remote_addr.svm_cid,
1431                                               flags,
1432                                               vmci_transport_is_trusted(
1433                                                   vsk,
1434                                                   vsk->
1435                                                   remote_addr.svm_cid));
1436         if (err < 0)
1437                 goto destroy;
1438
1439         err = vmci_transport_send_qp_offer(sk, handle);
1440         if (err < 0) {
1441                 err = vmci_transport_error_to_vsock_error(err);
1442                 goto destroy;
1443         }
1444
1445         vmci_trans(vsk)->qp_handle = handle;
1446         vmci_trans(vsk)->qpair = qpair;
1447
1448         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1449                 pkt->u.size;
1450
1451         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1452
1453         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1454
1455         return 0;
1456
1457 destroy:
1458         if (detach_sub_id != VMCI_INVALID_ID)
1459                 vmci_event_unsubscribe(detach_sub_id);
1460
1461         if (!vmci_handle_is_invalid(handle))
1462                 vmci_qpair_detach(&qpair);
1463
1464         return err;
1465 }
1466
1467 static int
1468 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1469                                               struct vmci_transport_packet *pkt)
1470 {
1471         int err = 0;
1472         struct vsock_sock *vsk = vsock_sk(sk);
1473
1474         if (vsk->sent_request) {
1475                 vsk->sent_request = false;
1476                 vsk->ignore_connecting_rst = true;
1477
1478                 err = vmci_transport_send_conn_request(
1479                         sk, vmci_trans(vsk)->queue_pair_size);
1480                 if (err < 0)
1481                         err = vmci_transport_error_to_vsock_error(err);
1482                 else
1483                         err = 0;
1484
1485         }
1486
1487         return err;
1488 }
1489
1490 static int vmci_transport_recv_connected(struct sock *sk,
1491                                          struct vmci_transport_packet *pkt)
1492 {
1493         struct vsock_sock *vsk;
1494         bool pkt_processed = false;
1495
1496         /* In cases where we are closing the connection, it's sufficient to
1497          * mark the state change (and maybe error) and wake up any waiting
1498          * threads. Since this is a connected socket, it's owned by a user
1499          * process and will be cleaned up when the failure is passed back on
1500          * the current or next system call.  Our system call implementations
1501          * must therefore check for error and state changes on entry and when
1502          * being awoken.
1503          */
1504         switch (pkt->type) {
1505         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1506                 if (pkt->u.mode) {
1507                         vsk = vsock_sk(sk);
1508
1509                         vsk->peer_shutdown |= pkt->u.mode;
1510                         sk->sk_state_change(sk);
1511                 }
1512                 break;
1513
1514         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1515                 vsk = vsock_sk(sk);
1516                 /* It is possible that we sent our peer a message (e.g a
1517                  * WAITING_READ) right before we got notified that the peer had
1518                  * detached. If that happens then we can get a RST pkt back
1519                  * from our peer even though there is data available for us to
1520                  * read. In that case, don't shutdown the socket completely but
1521                  * instead allow the local client to finish reading data off
1522                  * the queuepair. Always treat a RST pkt in connected mode like
1523                  * a clean shutdown.
1524                  */
1525                 sock_set_flag(sk, SOCK_DONE);
1526                 vsk->peer_shutdown = SHUTDOWN_MASK;
1527                 if (vsock_stream_has_data(vsk) <= 0)
1528                         sk->sk_state = TCP_CLOSING;
1529
1530                 sk->sk_state_change(sk);
1531                 break;
1532
1533         default:
1534                 vsk = vsock_sk(sk);
1535                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1536                                 sk, pkt, false, NULL, NULL,
1537                                 &pkt_processed);
1538                 if (!pkt_processed)
1539                         return -EINVAL;
1540
1541                 break;
1542         }
1543
1544         return 0;
1545 }
1546
1547 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1548                                       struct vsock_sock *psk)
1549 {
1550         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1551         if (!vsk->trans)
1552                 return -ENOMEM;
1553
1554         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1555         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1556         vmci_trans(vsk)->qpair = NULL;
1557         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1558         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1559         vmci_trans(vsk)->notify_ops = NULL;
1560         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1561         vmci_trans(vsk)->sk = &vsk->sk;
1562         spin_lock_init(&vmci_trans(vsk)->lock);
1563         if (psk) {
1564                 vmci_trans(vsk)->queue_pair_size =
1565                         vmci_trans(psk)->queue_pair_size;
1566                 vmci_trans(vsk)->queue_pair_min_size =
1567                         vmci_trans(psk)->queue_pair_min_size;
1568                 vmci_trans(vsk)->queue_pair_max_size =
1569                         vmci_trans(psk)->queue_pair_max_size;
1570         } else {
1571                 vmci_trans(vsk)->queue_pair_size =
1572                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1573                 vmci_trans(vsk)->queue_pair_min_size =
1574                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1575                 vmci_trans(vsk)->queue_pair_max_size =
1576                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1577         }
1578
1579         return 0;
1580 }
1581
1582 static void vmci_transport_free_resources(struct list_head *transport_list)
1583 {
1584         while (!list_empty(transport_list)) {
1585                 struct vmci_transport *transport =
1586                     list_first_entry(transport_list, struct vmci_transport,
1587                                      elem);
1588                 list_del(&transport->elem);
1589
1590                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1591                         vmci_event_unsubscribe(transport->detach_sub_id);
1592                         transport->detach_sub_id = VMCI_INVALID_ID;
1593                 }
1594
1595                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1596                         vmci_qpair_detach(&transport->qpair);
1597                         transport->qp_handle = VMCI_INVALID_HANDLE;
1598                         transport->produce_size = 0;
1599                         transport->consume_size = 0;
1600                 }
1601
1602                 kfree(transport);
1603         }
1604 }
1605
1606 static void vmci_transport_cleanup(struct work_struct *work)
1607 {
1608         LIST_HEAD(pending);
1609
1610         spin_lock_bh(&vmci_transport_cleanup_lock);
1611         list_replace_init(&vmci_transport_cleanup_list, &pending);
1612         spin_unlock_bh(&vmci_transport_cleanup_lock);
1613         vmci_transport_free_resources(&pending);
1614 }
1615
1616 static void vmci_transport_destruct(struct vsock_sock *vsk)
1617 {
1618         /* Ensure that the detach callback doesn't use the sk/vsk
1619          * we are about to destruct.
1620          */
1621         spin_lock_bh(&vmci_trans(vsk)->lock);
1622         vmci_trans(vsk)->sk = NULL;
1623         spin_unlock_bh(&vmci_trans(vsk)->lock);
1624
1625         if (vmci_trans(vsk)->notify_ops)
1626                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1627
1628         spin_lock_bh(&vmci_transport_cleanup_lock);
1629         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1630         spin_unlock_bh(&vmci_transport_cleanup_lock);
1631         schedule_work(&vmci_transport_cleanup_work);
1632
1633         vsk->trans = NULL;
1634 }
1635
1636 static void vmci_transport_release(struct vsock_sock *vsk)
1637 {
1638         vsock_remove_sock(vsk);
1639
1640         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1641                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1642                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1643         }
1644 }
1645
1646 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1647                                      struct sockaddr_vm *addr)
1648 {
1649         u32 port;
1650         u32 flags;
1651         int err;
1652
1653         /* VMCI will select a resource ID for us if we provide
1654          * VMCI_INVALID_ID.
1655          */
1656         port = addr->svm_port == VMADDR_PORT_ANY ?
1657                         VMCI_INVALID_ID : addr->svm_port;
1658
1659         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1660                 return -EACCES;
1661
1662         flags = addr->svm_cid == VMADDR_CID_ANY ?
1663                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1664
1665         err = vmci_transport_datagram_create_hnd(port, flags,
1666                                                  vmci_transport_recv_dgram_cb,
1667                                                  &vsk->sk,
1668                                                  &vmci_trans(vsk)->dg_handle);
1669         if (err < VMCI_SUCCESS)
1670                 return vmci_transport_error_to_vsock_error(err);
1671         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1672                         vmci_trans(vsk)->dg_handle.resource);
1673
1674         return 0;
1675 }
1676
1677 static int vmci_transport_dgram_enqueue(
1678         struct vsock_sock *vsk,
1679         struct sockaddr_vm *remote_addr,
1680         struct msghdr *msg,
1681         size_t len)
1682 {
1683         int err;
1684         struct vmci_datagram *dg;
1685
1686         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1687                 return -EMSGSIZE;
1688
1689         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1690                 return -EPERM;
1691
1692         /* Allocate a buffer for the user's message and our packet header. */
1693         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1694         if (!dg)
1695                 return -ENOMEM;
1696
1697         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1698
1699         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1700                                    remote_addr->svm_port);
1701         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1702                                    vsk->local_addr.svm_port);
1703         dg->payload_size = len;
1704
1705         err = vmci_datagram_send(dg);
1706         kfree(dg);
1707         if (err < 0)
1708                 return vmci_transport_error_to_vsock_error(err);
1709
1710         return err - sizeof(*dg);
1711 }
1712
1713 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1714                                         struct msghdr *msg, size_t len,
1715                                         int flags)
1716 {
1717         int err;
1718         int noblock;
1719         struct vmci_datagram *dg;
1720         size_t payload_len;
1721         struct sk_buff *skb;
1722
1723         noblock = flags & MSG_DONTWAIT;
1724
1725         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1726                 return -EOPNOTSUPP;
1727
1728         /* Retrieve the head sk_buff from the socket's receive queue. */
1729         err = 0;
1730         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1731         if (!skb)
1732                 return err;
1733
1734         dg = (struct vmci_datagram *)skb->data;
1735         if (!dg)
1736                 /* err is 0, meaning we read zero bytes. */
1737                 goto out;
1738
1739         payload_len = dg->payload_size;
1740         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1741         if (payload_len != skb->len - sizeof(*dg)) {
1742                 err = -EINVAL;
1743                 goto out;
1744         }
1745
1746         if (payload_len > len) {
1747                 payload_len = len;
1748                 msg->msg_flags |= MSG_TRUNC;
1749         }
1750
1751         /* Place the datagram payload in the user's iovec. */
1752         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1753         if (err)
1754                 goto out;
1755
1756         if (msg->msg_name) {
1757                 /* Provide the address of the sender. */
1758                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1759                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1760                 msg->msg_namelen = sizeof(*vm_addr);
1761         }
1762         err = payload_len;
1763
1764 out:
1765         skb_free_datagram(&vsk->sk, skb);
1766         return err;
1767 }
1768
1769 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1770 {
1771         if (cid == VMADDR_CID_HYPERVISOR) {
1772                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1773                  * state and are allowed.
1774                  */
1775                 return port == VMCI_UNITY_PBRPC_REGISTER;
1776         }
1777
1778         return true;
1779 }
1780
1781 static int vmci_transport_connect(struct vsock_sock *vsk)
1782 {
1783         int err;
1784         bool old_pkt_proto = false;
1785         struct sock *sk = &vsk->sk;
1786
1787         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1788                 old_pkt_proto) {
1789                 err = vmci_transport_send_conn_request(
1790                         sk, vmci_trans(vsk)->queue_pair_size);
1791                 if (err < 0) {
1792                         sk->sk_state = TCP_CLOSE;
1793                         return err;
1794                 }
1795         } else {
1796                 int supported_proto_versions =
1797                         vmci_transport_new_proto_supported_versions();
1798                 err = vmci_transport_send_conn_request2(
1799                                 sk, vmci_trans(vsk)->queue_pair_size,
1800                                 supported_proto_versions);
1801                 if (err < 0) {
1802                         sk->sk_state = TCP_CLOSE;
1803                         return err;
1804                 }
1805
1806                 vsk->sent_request = true;
1807         }
1808
1809         return err;
1810 }
1811
1812 static ssize_t vmci_transport_stream_dequeue(
1813         struct vsock_sock *vsk,
1814         struct msghdr *msg,
1815         size_t len,
1816         int flags)
1817 {
1818         if (flags & MSG_PEEK)
1819                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1820         else
1821                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1822 }
1823
1824 static ssize_t vmci_transport_stream_enqueue(
1825         struct vsock_sock *vsk,
1826         struct msghdr *msg,
1827         size_t len)
1828 {
1829         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1830 }
1831
1832 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1833 {
1834         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1835 }
1836
1837 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1838 {
1839         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1840 }
1841
1842 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1843 {
1844         return vmci_trans(vsk)->consume_size;
1845 }
1846
1847 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1848 {
1849         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1850 }
1851
1852 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1853 {
1854         return vmci_trans(vsk)->queue_pair_size;
1855 }
1856
1857 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1858 {
1859         return vmci_trans(vsk)->queue_pair_min_size;
1860 }
1861
1862 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1863 {
1864         return vmci_trans(vsk)->queue_pair_max_size;
1865 }
1866
1867 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1868 {
1869         if (val < vmci_trans(vsk)->queue_pair_min_size)
1870                 vmci_trans(vsk)->queue_pair_min_size = val;
1871         if (val > vmci_trans(vsk)->queue_pair_max_size)
1872                 vmci_trans(vsk)->queue_pair_max_size = val;
1873         vmci_trans(vsk)->queue_pair_size = val;
1874 }
1875
1876 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1877                                                u64 val)
1878 {
1879         if (val > vmci_trans(vsk)->queue_pair_size)
1880                 vmci_trans(vsk)->queue_pair_size = val;
1881         vmci_trans(vsk)->queue_pair_min_size = val;
1882 }
1883
1884 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1885                                                u64 val)
1886 {
1887         if (val < vmci_trans(vsk)->queue_pair_size)
1888                 vmci_trans(vsk)->queue_pair_size = val;
1889         vmci_trans(vsk)->queue_pair_max_size = val;
1890 }
1891
1892 static int vmci_transport_notify_poll_in(
1893         struct vsock_sock *vsk,
1894         size_t target,
1895         bool *data_ready_now)
1896 {
1897         return vmci_trans(vsk)->notify_ops->poll_in(
1898                         &vsk->sk, target, data_ready_now);
1899 }
1900
1901 static int vmci_transport_notify_poll_out(
1902         struct vsock_sock *vsk,
1903         size_t target,
1904         bool *space_available_now)
1905 {
1906         return vmci_trans(vsk)->notify_ops->poll_out(
1907                         &vsk->sk, target, space_available_now);
1908 }
1909
1910 static int vmci_transport_notify_recv_init(
1911         struct vsock_sock *vsk,
1912         size_t target,
1913         struct vsock_transport_recv_notify_data *data)
1914 {
1915         return vmci_trans(vsk)->notify_ops->recv_init(
1916                         &vsk->sk, target,
1917                         (struct vmci_transport_recv_notify_data *)data);
1918 }
1919
1920 static int vmci_transport_notify_recv_pre_block(
1921         struct vsock_sock *vsk,
1922         size_t target,
1923         struct vsock_transport_recv_notify_data *data)
1924 {
1925         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1926                         &vsk->sk, target,
1927                         (struct vmci_transport_recv_notify_data *)data);
1928 }
1929
1930 static int vmci_transport_notify_recv_pre_dequeue(
1931         struct vsock_sock *vsk,
1932         size_t target,
1933         struct vsock_transport_recv_notify_data *data)
1934 {
1935         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1936                         &vsk->sk, target,
1937                         (struct vmci_transport_recv_notify_data *)data);
1938 }
1939
1940 static int vmci_transport_notify_recv_post_dequeue(
1941         struct vsock_sock *vsk,
1942         size_t target,
1943         ssize_t copied,
1944         bool data_read,
1945         struct vsock_transport_recv_notify_data *data)
1946 {
1947         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1948                         &vsk->sk, target, copied, data_read,
1949                         (struct vmci_transport_recv_notify_data *)data);
1950 }
1951
1952 static int vmci_transport_notify_send_init(
1953         struct vsock_sock *vsk,
1954         struct vsock_transport_send_notify_data *data)
1955 {
1956         return vmci_trans(vsk)->notify_ops->send_init(
1957                         &vsk->sk,
1958                         (struct vmci_transport_send_notify_data *)data);
1959 }
1960
1961 static int vmci_transport_notify_send_pre_block(
1962         struct vsock_sock *vsk,
1963         struct vsock_transport_send_notify_data *data)
1964 {
1965         return vmci_trans(vsk)->notify_ops->send_pre_block(
1966                         &vsk->sk,
1967                         (struct vmci_transport_send_notify_data *)data);
1968 }
1969
1970 static int vmci_transport_notify_send_pre_enqueue(
1971         struct vsock_sock *vsk,
1972         struct vsock_transport_send_notify_data *data)
1973 {
1974         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1975                         &vsk->sk,
1976                         (struct vmci_transport_send_notify_data *)data);
1977 }
1978
1979 static int vmci_transport_notify_send_post_enqueue(
1980         struct vsock_sock *vsk,
1981         ssize_t written,
1982         struct vsock_transport_send_notify_data *data)
1983 {
1984         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1985                         &vsk->sk, written,
1986                         (struct vmci_transport_send_notify_data *)data);
1987 }
1988
1989 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1990 {
1991         if (PROTOCOL_OVERRIDE != -1) {
1992                 if (PROTOCOL_OVERRIDE == 0)
1993                         *old_pkt_proto = true;
1994                 else
1995                         *old_pkt_proto = false;
1996
1997                 pr_info("Proto override in use\n");
1998                 return true;
1999         }
2000
2001         return false;
2002 }
2003
2004 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2005                                                   u16 *proto,
2006                                                   bool old_pkt_proto)
2007 {
2008         struct vsock_sock *vsk = vsock_sk(sk);
2009
2010         if (old_pkt_proto) {
2011                 if (*proto != VSOCK_PROTO_INVALID) {
2012                         pr_err("Can't set both an old and new protocol\n");
2013                         return false;
2014                 }
2015                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2016                 goto exit;
2017         }
2018
2019         switch (*proto) {
2020         case VSOCK_PROTO_PKT_ON_NOTIFY:
2021                 vmci_trans(vsk)->notify_ops =
2022                         &vmci_transport_notify_pkt_q_state_ops;
2023                 break;
2024         default:
2025                 pr_err("Unknown notify protocol version\n");
2026                 return false;
2027         }
2028
2029 exit:
2030         vmci_trans(vsk)->notify_ops->socket_init(sk);
2031         return true;
2032 }
2033
2034 static u16 vmci_transport_new_proto_supported_versions(void)
2035 {
2036         if (PROTOCOL_OVERRIDE != -1)
2037                 return PROTOCOL_OVERRIDE;
2038
2039         return VSOCK_PROTO_ALL_SUPPORTED;
2040 }
2041
2042 static u32 vmci_transport_get_local_cid(void)
2043 {
2044         return vmci_get_context_id();
2045 }
2046
2047 static const struct vsock_transport vmci_transport = {
2048         .init = vmci_transport_socket_init,
2049         .destruct = vmci_transport_destruct,
2050         .release = vmci_transport_release,
2051         .connect = vmci_transport_connect,
2052         .dgram_bind = vmci_transport_dgram_bind,
2053         .dgram_dequeue = vmci_transport_dgram_dequeue,
2054         .dgram_enqueue = vmci_transport_dgram_enqueue,
2055         .dgram_allow = vmci_transport_dgram_allow,
2056         .stream_dequeue = vmci_transport_stream_dequeue,
2057         .stream_enqueue = vmci_transport_stream_enqueue,
2058         .stream_has_data = vmci_transport_stream_has_data,
2059         .stream_has_space = vmci_transport_stream_has_space,
2060         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2061         .stream_is_active = vmci_transport_stream_is_active,
2062         .stream_allow = vmci_transport_stream_allow,
2063         .notify_poll_in = vmci_transport_notify_poll_in,
2064         .notify_poll_out = vmci_transport_notify_poll_out,
2065         .notify_recv_init = vmci_transport_notify_recv_init,
2066         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2067         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2068         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2069         .notify_send_init = vmci_transport_notify_send_init,
2070         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2071         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2072         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2073         .shutdown = vmci_transport_shutdown,
2074         .set_buffer_size = vmci_transport_set_buffer_size,
2075         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2076         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2077         .get_buffer_size = vmci_transport_get_buffer_size,
2078         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2079         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2080         .get_local_cid = vmci_transport_get_local_cid,
2081 };
2082
2083 static int __init vmci_transport_init(void)
2084 {
2085         int err;
2086
2087         /* Create the datagram handle that we will use to send and receive all
2088          * VSocket control messages for this context.
2089          */
2090         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2091                                                  VMCI_FLAG_ANYCID_DG_HND,
2092                                                  vmci_transport_recv_stream_cb,
2093                                                  NULL,
2094                                                  &vmci_transport_stream_handle);
2095         if (err < VMCI_SUCCESS) {
2096                 pr_err("Unable to create datagram handle. (%d)\n", err);
2097                 return vmci_transport_error_to_vsock_error(err);
2098         }
2099
2100         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2101                                    vmci_transport_qp_resumed_cb,
2102                                    NULL, &vmci_transport_qp_resumed_sub_id);
2103         if (err < VMCI_SUCCESS) {
2104                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2105                 err = vmci_transport_error_to_vsock_error(err);
2106                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2107                 goto err_destroy_stream_handle;
2108         }
2109
2110         err = vsock_core_init(&vmci_transport);
2111         if (err < 0)
2112                 goto err_unsubscribe;
2113
2114         return 0;
2115
2116 err_unsubscribe:
2117         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2118 err_destroy_stream_handle:
2119         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2120         return err;
2121 }
2122 module_init(vmci_transport_init);
2123
2124 static void __exit vmci_transport_exit(void)
2125 {
2126         cancel_work_sync(&vmci_transport_cleanup_work);
2127         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2128
2129         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2130                 if (vmci_datagram_destroy_handle(
2131                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2132                         pr_err("Couldn't destroy datagram handle\n");
2133                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2134         }
2135
2136         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2137                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2138                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2139         }
2140
2141         vsock_core_exit();
2142 }
2143 module_exit(vmci_transport_exit);
2144
2145 MODULE_AUTHOR("VMware, Inc.");
2146 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2147 MODULE_VERSION("1.0.4.0-k");
2148 MODULE_LICENSE("GPL v2");
2149 MODULE_ALIAS("vmware_vsock");
2150 MODULE_ALIAS_NETPROTO(PF_VSOCK);