net/vmw_vsock/vmci_transport.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VMware vSockets Driver
   4  *
   5  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   6  */
   7
   8 #include <linux/types.h>
   9 #include <linux/bitops.h>
  10 #include <linux/cred.h>
  11 #include <linux/init.h>
  12 #include <linux/io.h>
  13 #include <linux/kernel.h>
  14 #include <linux/kmod.h>
  15 #include <linux/list.h>
  16 #include <linux/module.h>
  17 #include <linux/mutex.h>
  18 #include <linux/net.h>
  19 #include <linux/poll.h>
  20 #include <linux/skbuff.h>
  21 #include <linux/smp.h>
  22 #include <linux/socket.h>
  23 #include <linux/stddef.h>
  24 #include <linux/unistd.h>
  25 #include <linux/wait.h>
  26 #include <linux/workqueue.h>
  27 #include <net/sock.h>
  28 #include <net/af_vsock.h>
  29
  30 #include "vmci_transport_notify.h"
  31
  32 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
  33 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
  34 static void vmci_transport_peer_detach_cb(u32 sub_id,
  35                                           const struct vmci_event_data *ed,
  36                                           void *client_data);
  37 static void vmci_transport_recv_pkt_work(struct work_struct *work);
  38 static void vmci_transport_cleanup(struct work_struct *work);
  39 static int vmci_transport_recv_listen(struct sock *sk,
  40                                       struct vmci_transport_packet *pkt);
  41 static int vmci_transport_recv_connecting_server(
  42                                         struct sock *sk,
  43                                         struct sock *pending,
  44                                         struct vmci_transport_packet *pkt);
  45 static int vmci_transport_recv_connecting_client(
  46                                         struct sock *sk,
  47                                         struct vmci_transport_packet *pkt);
  48 static int vmci_transport_recv_connecting_client_negotiate(
  49                                         struct sock *sk,
  50                                         struct vmci_transport_packet *pkt);
  51 static int vmci_transport_recv_connecting_client_invalid(
  52                                         struct sock *sk,
  53                                         struct vmci_transport_packet *pkt);
  54 static int vmci_transport_recv_connected(struct sock *sk,
  55                                          struct vmci_transport_packet *pkt);
  56 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
  57 static u16 vmci_transport_new_proto_supported_versions(void);
  58 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
  59                                                   bool old_pkt_proto);
  60 static bool vmci_check_transport(struct vsock_sock *vsk);
  61
  62 struct vmci_transport_recv_pkt_info {
  63         struct work_struct work;
  64         struct sock *sk;
  65         struct vmci_transport_packet pkt;
  66 };
  67
  68 static LIST_HEAD(vmci_transport_cleanup_list);
  69 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
  70 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
  71
  72 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  73                                                            VMCI_INVALID_ID };
  74 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  75
  76 static int PROTOCOL_OVERRIDE = -1;
  77
  78 /* Helper function to convert from a VMCI error code to a VSock error code. */
  79
  80 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  81 {
  82         switch (vmci_error) {
  83         case VMCI_ERROR_NO_MEM:
  84                 return -ENOMEM;
  85         case VMCI_ERROR_DUPLICATE_ENTRY:
  86         case VMCI_ERROR_ALREADY_EXISTS:
  87                 return -EADDRINUSE;
  88         case VMCI_ERROR_NO_ACCESS:
  89                 return -EPERM;
  90         case VMCI_ERROR_NO_RESOURCES:
  91                 return -ENOBUFS;
  92         case VMCI_ERROR_INVALID_RESOURCE:
  93                 return -EHOSTUNREACH;
  94         case VMCI_ERROR_INVALID_ARGS:
  95         default:
  96                 break;
  97         }
  98         return -EINVAL;
  99 }
 100
 101 static u32 vmci_transport_peer_rid(u32 peer_cid)
 102 {
 103         if (VMADDR_CID_HYPERVISOR == peer_cid)
 104                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 105
 106         return VMCI_TRANSPORT_PACKET_RID;
 107 }
 108
 109 static inline void
 110 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 111                            struct sockaddr_vm *src,
 112                            struct sockaddr_vm *dst,
 113                            u8 type,
 114                            u64 size,
 115                            u64 mode,
 116                            struct vmci_transport_waiting_info *wait,
 117                            u16 proto,
 118                            struct vmci_handle handle)
 119 {
 120         /* We register the stream control handler as an any cid handle so we
 121          * must always send from a source address of VMADDR_CID_ANY
 122          */
 123         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 124                                        VMCI_TRANSPORT_PACKET_RID);
 125         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 126                                        vmci_transport_peer_rid(dst->svm_cid));
 127         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 128         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 129         pkt->type = type;
 130         pkt->src_port = src->svm_port;
 131         pkt->dst_port = dst->svm_port;
 132         memset(&pkt->proto, 0, sizeof(pkt->proto));
 133         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 134
 135         switch (pkt->type) {
 136         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 137                 pkt->u.size = 0;
 138                 break;
 139
 140         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 141         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 142                 pkt->u.size = size;
 143                 break;
 144
 145         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 146         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 147                 pkt->u.handle = handle;
 148                 break;
 149
 150         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 151         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 152         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 153                 pkt->u.size = 0;
 154                 break;
 155
 156         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 157                 pkt->u.mode = mode;
 158                 break;
 159
 160         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 161         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 162                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 163                 break;
 164
 165         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 166         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 167                 pkt->u.size = size;
 168                 pkt->proto = proto;
 169                 break;
 170         }
 171 }
 172
 173 static inline void
 174 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 175                                     struct sockaddr_vm *local,
 176                                     struct sockaddr_vm *remote)
 177 {
 178         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 179         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 180 }
 181
 182 static int
 183 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 184                                   struct sockaddr_vm *src,
 185                                   struct sockaddr_vm *dst,
 186                                   enum vmci_transport_packet_type type,
 187                                   u64 size,
 188                                   u64 mode,
 189                                   struct vmci_transport_waiting_info *wait,
 190                                   u16 proto,
 191                                   struct vmci_handle handle,
 192                                   bool convert_error)
 193 {
 194         int err;
 195
 196         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 197                                    proto, handle);
 198         err = vmci_datagram_send(&pkt->dg);
 199         if (convert_error && (err < 0))
 200                 return vmci_transport_error_to_vsock_error(err);
 201
 202         return err;
 203 }
 204
 205 static int
 206 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 207                                       enum vmci_transport_packet_type type,
 208                                       u64 size,
 209                                       u64 mode,
 210                                       struct vmci_transport_waiting_info *wait,
 211                                       struct vmci_handle handle)
 212 {
 213         struct vmci_transport_packet reply;
 214         struct sockaddr_vm src, dst;
 215
 216         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 217                 return 0;
 218         } else {
 219                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 220                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 221                                                          type,
 222                                                          size, mode, wait,
 223                                                          VSOCK_PROTO_INVALID,
 224                                                          handle, true);
 225         }
 226 }
 227
 228 static int
 229 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 230                                    struct sockaddr_vm *dst,
 231                                    enum vmci_transport_packet_type type,
 232                                    u64 size,
 233                                    u64 mode,
 234                                    struct vmci_transport_waiting_info *wait,
 235                                    struct vmci_handle handle)
 236 {
 237         /* Note that it is safe to use a single packet across all CPUs since
 238          * two tasklets of the same type are guaranteed to not ever run
 239          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 240          * we can use per-cpu packets.
 241          */
 242         static struct vmci_transport_packet pkt;
 243
 244         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 245                                                  size, mode, wait,
 246                                                  VSOCK_PROTO_INVALID, handle,
 247                                                  false);
 248 }
 249
 250 static int
 251 vmci_transport_alloc_send_control_pkt(struct sockaddr_vm *src,
 252                                       struct sockaddr_vm *dst,
 253                                       enum vmci_transport_packet_type type,
 254                                       u64 size,
 255                                       u64 mode,
 256                                       struct vmci_transport_waiting_info *wait,
 257                                       u16 proto,
 258                                       struct vmci_handle handle)
 259 {
 260         struct vmci_transport_packet *pkt;
 261         int err;
 262
 263         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 264         if (!pkt)
 265                 return -ENOMEM;
 266
 267         err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
 268                                                 mode, wait, proto, handle,
 269                                                 true);
 270         kfree(pkt);
 271
 272         return err;
 273 }
 274
 275 static int
 276 vmci_transport_send_control_pkt(struct sock *sk,
 277                                 enum vmci_transport_packet_type type,
 278                                 u64 size,
 279                                 u64 mode,
 280                                 struct vmci_transport_waiting_info *wait,
 281                                 u16 proto,
 282                                 struct vmci_handle handle)
 283 {
 284         struct vsock_sock *vsk;
 285
 286         vsk = vsock_sk(sk);
 287
 288         if (!vsock_addr_bound(&vsk->local_addr))
 289                 return -EINVAL;
 290
 291         if (!vsock_addr_bound(&vsk->remote_addr))
 292                 return -EINVAL;
 293
 294         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
 295                                                      &vsk->remote_addr,
 296                                                      type, size, mode,
 297                                                      wait, proto, handle);
 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         struct sockaddr_vm *dst_ptr;
 316         struct sockaddr_vm dst;
 317         struct vsock_sock *vsk;
 318
 319         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 320                 return 0;
 321
 322         vsk = vsock_sk(sk);
 323
 324         if (!vsock_addr_bound(&vsk->local_addr))
 325                 return -EINVAL;
 326
 327         if (vsock_addr_bound(&vsk->remote_addr)) {
 328                 dst_ptr = &vsk->remote_addr;
 329         } else {
 330                 vsock_addr_init(&dst, pkt->dg.src.context,
 331                                 pkt->src_port);
 332                 dst_ptr = &dst;
 333         }
 334         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
 335                                              VMCI_TRANSPORT_PACKET_TYPE_RST,
 336                                              0, 0, NULL, VSOCK_PROTO_INVALID,
 337                                              VMCI_INVALID_HANDLE);
 338 }
 339
 340 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 341 {
 342         return vmci_transport_send_control_pkt(
 343                                         sk,
 344                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 345                                         size, 0, NULL,
 346                                         VSOCK_PROTO_INVALID,
 347                                         VMCI_INVALID_HANDLE);
 348 }
 349
 350 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 351                                           u16 version)
 352 {
 353         return vmci_transport_send_control_pkt(
 354                                         sk,
 355                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 356                                         size, 0, NULL, version,
 357                                         VMCI_INVALID_HANDLE);
 358 }
 359
 360 static int vmci_transport_send_qp_offer(struct sock *sk,
 361                                         struct vmci_handle handle)
 362 {
 363         return vmci_transport_send_control_pkt(
 364                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 365                                         0, NULL,
 366                                         VSOCK_PROTO_INVALID, handle);
 367 }
 368
 369 static int vmci_transport_send_attach(struct sock *sk,
 370                                       struct vmci_handle handle)
 371 {
 372         return vmci_transport_send_control_pkt(
 373                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 374                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 375                                         handle);
 376 }
 377
 378 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 379 {
 380         return vmci_transport_reply_control_pkt_fast(
 381                                                 pkt,
 382                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 383                                                 0, 0, NULL,
 384                                                 VMCI_INVALID_HANDLE);
 385 }
 386
 387 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 388                                           struct sockaddr_vm *src)
 389 {
 390         return vmci_transport_send_control_pkt_bh(
 391                                         dst, src,
 392                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 393                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 394 }
 395
 396 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 397                                  struct sockaddr_vm *src)
 398 {
 399         return vmci_transport_send_control_pkt_bh(
 400                                         dst, src,
 401                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 402                                         0, NULL, VMCI_INVALID_HANDLE);
 403 }
 404
 405 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 406                                 struct sockaddr_vm *src)
 407 {
 408         return vmci_transport_send_control_pkt_bh(
 409                                         dst, src,
 410                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 411                                         0, NULL, VMCI_INVALID_HANDLE);
 412 }
 413
 414 int vmci_transport_send_wrote(struct sock *sk)
 415 {
 416         return vmci_transport_send_control_pkt(
 417                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 418                                         0, NULL, VSOCK_PROTO_INVALID,
 419                                         VMCI_INVALID_HANDLE);
 420 }
 421
 422 int vmci_transport_send_read(struct sock *sk)
 423 {
 424         return vmci_transport_send_control_pkt(
 425                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 426                                         0, NULL, VSOCK_PROTO_INVALID,
 427                                         VMCI_INVALID_HANDLE);
 428 }
 429
 430 int vmci_transport_send_waiting_write(struct sock *sk,
 431                                       struct vmci_transport_waiting_info *wait)
 432 {
 433         return vmci_transport_send_control_pkt(
 434                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 435                                 0, 0, wait, VSOCK_PROTO_INVALID,
 436                                 VMCI_INVALID_HANDLE);
 437 }
 438
 439 int vmci_transport_send_waiting_read(struct sock *sk,
 440                                      struct vmci_transport_waiting_info *wait)
 441 {
 442         return vmci_transport_send_control_pkt(
 443                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 444                                 0, 0, wait, VSOCK_PROTO_INVALID,
 445                                 VMCI_INVALID_HANDLE);
 446 }
 447
 448 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 449 {
 450         return vmci_transport_send_control_pkt(
 451                                         &vsk->sk,
 452                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 453                                         0, mode, NULL,
 454                                         VSOCK_PROTO_INVALID,
 455                                         VMCI_INVALID_HANDLE);
 456 }
 457
 458 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 459 {
 460         return vmci_transport_send_control_pkt(sk,
 461                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 462                                         size, 0, NULL,
 463                                         VSOCK_PROTO_INVALID,
 464                                         VMCI_INVALID_HANDLE);
 465 }
 466
 467 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 468                                              u16 version)
 469 {
 470         return vmci_transport_send_control_pkt(
 471                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 472                                         size, 0, NULL, version,
 473                                         VMCI_INVALID_HANDLE);
 474 }
 475
 476 static struct sock *vmci_transport_get_pending(
 477                                         struct sock *listener,
 478                                         struct vmci_transport_packet *pkt)
 479 {
 480         struct vsock_sock *vlistener;
 481         struct vsock_sock *vpending;
 482         struct sock *pending;
 483         struct sockaddr_vm src;
 484
 485         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 486
 487         vlistener = vsock_sk(listener);
 488
 489         list_for_each_entry(vpending, &vlistener->pending_links,
 490                             pending_links) {
 491                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 492                     pkt->dst_port == vpending->local_addr.svm_port) {
 493                         pending = sk_vsock(vpending);
 494                         sock_hold(pending);
 495                         goto found;
 496                 }
 497         }
 498
 499         pending = NULL;
 500 found:
 501         return pending;
 502
 503 }
 504
 505 static void vmci_transport_release_pending(struct sock *pending)
 506 {
 507         sock_put(pending);
 508 }
 509
 510 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 511  * trusted sockets 2) sockets from applications running as the same user as the
 512  * VM (this is only true for the host side and only when using hosted products)
 513  */
 514
 515 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 516 {
 517         return vsock->trusted ||
 518                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 519 }
 520
 521 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 522  * only if it is trusted as described in vmci_transport_is_trusted.
 523  */
 524
 525 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 526 {
 527         if (VMADDR_CID_HYPERVISOR == peer_cid)
 528                 return true;
 529
 530         if (vsock->cached_peer != peer_cid) {
 531                 vsock->cached_peer = peer_cid;
 532                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 533                     (vmci_context_get_priv_flags(peer_cid) &
 534                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 535                         vsock->cached_peer_allow_dgram = false;
 536                 } else {
 537                         vsock->cached_peer_allow_dgram = true;
 538                 }
 539         }
 540
 541         return vsock->cached_peer_allow_dgram;
 542 }
 543
 544 static int
 545 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 546                                 struct vmci_handle *handle,
 547                                 u64 produce_size,
 548                                 u64 consume_size,
 549                                 u32 peer, u32 flags, bool trusted)
 550 {
 551         int err = 0;
 552
 553         if (trusted) {
 554                 /* Try to allocate our queue pair as trusted. This will only
 555                  * work if vsock is running in the host.
 556                  */
 557
 558                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 559                                        consume_size,
 560                                        peer, flags,
 561                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 562                 if (err != VMCI_ERROR_NO_ACCESS)
 563                         goto out;
 564
 565         }
 566
 567         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 568                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 569 out:
 570         if (err < 0) {
 571                 pr_err("Could not attach to queue pair with %d\n",
 572                        err);
 573                 err = vmci_transport_error_to_vsock_error(err);
 574         }
 575
 576         return err;
 577 }
 578
 579 static int
 580 vmci_transport_datagram_create_hnd(u32 resource_id,
 581                                    u32 flags,
 582                                    vmci_datagram_recv_cb recv_cb,
 583                                    void *client_data,
 584                                    struct vmci_handle *out_handle)
 585 {
 586         int err = 0;
 587
 588         /* Try to allocate our datagram handler as trusted. This will only work
 589          * if vsock is running in the host.
 590          */
 591
 592         err = vmci_datagram_create_handle_priv(resource_id, flags,
 593                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 594                                                recv_cb,
 595                                                client_data, out_handle);
 596
 597         if (err == VMCI_ERROR_NO_ACCESS)
 598                 err = vmci_datagram_create_handle(resource_id, flags,
 599                                                   recv_cb, client_data,
 600                                                   out_handle);
 601
 602         return err;
 603 }
 604
 605 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 606  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 607  * sleep it should defer that work to a work queue.
 608  */
 609
 610 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 611 {
 612         struct sock *sk;
 613         size_t size;
 614         struct sk_buff *skb;
 615         struct vsock_sock *vsk;
 616
 617         sk = (struct sock *)data;
 618
 619         /* This handler is privileged when this module is running on the host.
 620          * We will get datagrams from all endpoints (even VMs that are in a
 621          * restricted context). If we get one from a restricted context then
 622          * the destination socket must be trusted.
 623          *
 624          * NOTE: We access the socket struct without holding the lock here.
 625          * This is ok because the field we are interested is never modified
 626          * outside of the create and destruct socket functions.
 627          */
 628         vsk = vsock_sk(sk);
 629         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 630                 return VMCI_ERROR_NO_ACCESS;
 631
 632         size = VMCI_DG_SIZE(dg);
 633
 634         /* Attach the packet to the socket's receive queue as an sk_buff. */
 635         skb = alloc_skb(size, GFP_ATOMIC);
 636         if (!skb)
 637                 return VMCI_ERROR_NO_MEM;
 638
 639         /* sk_receive_skb() will do a sock_put(), so hold here. */
 640         sock_hold(sk);
 641         skb_put(skb, size);
 642         memcpy(skb->data, dg, size);
 643         sk_receive_skb(sk, skb, 0);
 644
 645         return VMCI_SUCCESS;
 646 }
 647
 648 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 649 {
 650         static const u32 non_socket_contexts[] = {
 651                 VMADDR_CID_LOCAL,
 652         };
 653         int i;
 654
 655         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 656
 657         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 658                 if (cid == non_socket_contexts[i])
 659                         return false;
 660         }
 661
 662         return true;
 663 }
 664
 665 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 666  * interrupt fires.  This is run in bottom-half context but it defers most of
 667  * its work to the packet handling work queue.
 668  */
 669
 670 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 671 {
 672         struct sock *sk;
 673         struct sockaddr_vm dst;
 674         struct sockaddr_vm src;
 675         struct vmci_transport_packet *pkt;
 676         struct vsock_sock *vsk;
 677         bool bh_process_pkt;
 678         int err;
 679
 680         sk = NULL;
 681         err = VMCI_SUCCESS;
 682         bh_process_pkt = false;
 683
 684         /* Ignore incoming packets from contexts without sockets, or resources
 685          * that aren't vsock implementations.
 686          */
 687
 688         if (!vmci_transport_stream_allow(dg->src.context, -1)
 689             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 690                 return VMCI_ERROR_NO_ACCESS;
 691
 692         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 693                 /* Drop datagrams that do not contain full VSock packets. */
 694                 return VMCI_ERROR_INVALID_ARGS;
 695
 696         pkt = (struct vmci_transport_packet *)dg;
 697
 698         /* Find the socket that should handle this packet.  First we look for a
 699          * connected socket and if there is none we look for a socket bound to
 700          * the destintation address.
 701          */
 702         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 703         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 704
 705         sk = vsock_find_connected_socket(&src, &dst);
 706         if (!sk) {
 707                 sk = vsock_find_bound_socket(&dst);
 708                 if (!sk) {
 709                         /* We could not find a socket for this specified
 710                          * address.  If this packet is a RST, we just drop it.
 711                          * If it is another packet, we send a RST.  Note that
 712                          * we do not send a RST reply to RSTs so that we do not
 713                          * continually send RSTs between two endpoints.
 714                          *
 715                          * Note that since this is a reply, dst is src and src
 716                          * is dst.
 717                          */
 718                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 719                                 pr_err("unable to send reset\n");
 720
 721                         err = VMCI_ERROR_NOT_FOUND;
 722                         goto out;
 723                 }
 724         }
 725
 726         /* If the received packet type is beyond all types known to this
 727          * implementation, reply with an invalid message.  Hopefully this will
 728          * help when implementing backwards compatibility in the future.
 729          */
 730         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 731                 vmci_transport_send_invalid_bh(&dst, &src);
 732                 err = VMCI_ERROR_INVALID_ARGS;
 733                 goto out;
 734         }
 735
 736         /* This handler is privileged when this module is running on the host.
 737          * We will get datagram connect requests from all endpoints (even VMs
 738          * that are in a restricted context). If we get one from a restricted
 739          * context then the destination socket must be trusted.
 740          *
 741          * NOTE: We access the socket struct without holding the lock here.
 742          * This is ok because the field we are interested is never modified
 743          * outside of the create and destruct socket functions.
 744          */
 745         vsk = vsock_sk(sk);
 746         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 747                 err = VMCI_ERROR_NO_ACCESS;
 748                 goto out;
 749         }
 750
 751         /* We do most everything in a work queue, but let's fast path the
 752          * notification of reads and writes to help data transfer performance.
 753          * We can only do this if there is no process context code executing
 754          * for this socket since that may change the state.
 755          */
 756         bh_lock_sock(sk);
 757
 758         if (!sock_owned_by_user(sk)) {
 759                 /* The local context ID may be out of date, update it. */
 760                 vsk->local_addr.svm_cid = dst.svm_cid;
 761
 762                 if (sk->sk_state == TCP_ESTABLISHED)
 763                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 764                                         sk, pkt, true, &dst, &src,
 765                                         &bh_process_pkt);
 766         }
 767
 768         bh_unlock_sock(sk);
 769
 770         if (!bh_process_pkt) {
 771                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 772
 773                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 774                 if (!recv_pkt_info) {
 775                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 776                                 pr_err("unable to send reset\n");
 777
 778                         err = VMCI_ERROR_NO_MEM;
 779                         goto out;
 780                 }
 781
 782                 recv_pkt_info->sk = sk;
 783                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 784                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 785
 786                 schedule_work(&recv_pkt_info->work);
 787                 /* Clear sk so that the reference count incremented by one of
 788                  * the Find functions above is not decremented below.  We need
 789                  * that reference count for the packet handler we've scheduled
 790                  * to run.
 791                  */
 792                 sk = NULL;
 793         }
 794
 795 out:
 796         if (sk)
 797                 sock_put(sk);
 798
 799         return err;
 800 }
 801
 802 static void vmci_transport_handle_detach(struct sock *sk)
 803 {
 804         struct vsock_sock *vsk;
 805
 806         vsk = vsock_sk(sk);
 807         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 808                 sock_set_flag(sk, SOCK_DONE);
 809
 810                 /* On a detach the peer will not be sending or receiving
 811                  * anymore.
 812                  */
 813                 vsk->peer_shutdown = SHUTDOWN_MASK;
 814
 815                 /* We should not be sending anymore since the peer won't be
 816                  * there to receive, but we can still receive if there is data
 817                  * left in our consume queue. If the local endpoint is a host,
 818                  * we can't call vsock_stream_has_data, since that may block,
 819                  * but a host endpoint can't read data once the VM has
 820                  * detached, so there is no available data in that case.
 821                  */
 822                 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
 823                     vsock_stream_has_data(vsk) <= 0) {
 824                         if (sk->sk_state == TCP_SYN_SENT) {
 825                                 /* The peer may detach from a queue pair while
 826                                  * we are still in the connecting state, i.e.,
 827                                  * if the peer VM is killed after attaching to
 828                                  * a queue pair, but before we complete the
 829                                  * handshake. In that case, we treat the detach
 830                                  * event like a reset.
 831                                  */
 832
 833                                 sk->sk_state = TCP_CLOSE;
 834                                 sk->sk_err = ECONNRESET;
 835                                 sk->sk_error_report(sk);
 836                                 return;
 837                         }
 838                         sk->sk_state = TCP_CLOSE;
 839                 }
 840                 sk->sk_state_change(sk);
 841         }
 842 }
 843
 844 static void vmci_transport_peer_detach_cb(u32 sub_id,
 845                                           const struct vmci_event_data *e_data,
 846                                           void *client_data)
 847 {
 848         struct vmci_transport *trans = client_data;
 849         const struct vmci_event_payload_qp *e_payload;
 850
 851         e_payload = vmci_event_data_const_payload(e_data);
 852
 853         /* XXX This is lame, we should provide a way to lookup sockets by
 854          * qp_handle.
 855          */
 856         if (vmci_handle_is_invalid(e_payload->handle) ||
 857             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 858                 return;
 859
 860         /* We don't ask for delayed CBs when we subscribe to this event (we
 861          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 862          * guarantees in that case about what context we might be running in,
 863          * so it could be BH or process, blockable or non-blockable.  So we
 864          * need to account for all possible contexts here.
 865          */
 866         spin_lock_bh(&trans->lock);
 867         if (!trans->sk)
 868                 goto out;
 869
 870         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 871          * where trans->sk isn't locked.
 872          */
 873         bh_lock_sock(trans->sk);
 874
 875         vmci_transport_handle_detach(trans->sk);
 876
 877         bh_unlock_sock(trans->sk);
 878  out:
 879         spin_unlock_bh(&trans->lock);
 880 }
 881
 882 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 883                                          const struct vmci_event_data *e_data,
 884                                          void *client_data)
 885 {
 886         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 887 }
 888
 889 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 890 {
 891         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 892         struct vmci_transport_packet *pkt;
 893         struct sock *sk;
 894
 895         recv_pkt_info =
 896                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 897         sk = recv_pkt_info->sk;
 898         pkt = &recv_pkt_info->pkt;
 899
 900         lock_sock(sk);
 901
 902         /* The local context ID may be out of date. */
 903         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 904
 905         switch (sk->sk_state) {
 906         case TCP_LISTEN:
 907                 vmci_transport_recv_listen(sk, pkt);
 908                 break;
 909         case TCP_SYN_SENT:
 910                 /* Processing of pending connections for servers goes through
 911                  * the listening socket, so see vmci_transport_recv_listen()
 912                  * for that path.
 913                  */
 914                 vmci_transport_recv_connecting_client(sk, pkt);
 915                 break;
 916         case TCP_ESTABLISHED:
 917                 vmci_transport_recv_connected(sk, pkt);
 918                 break;
 919         default:
 920                 /* Because this function does not run in the same context as
 921                  * vmci_transport_recv_stream_cb it is possible that the
 922                  * socket has closed. We need to let the other side know or it
 923                  * could be sitting in a connect and hang forever. Send a
 924                  * reset to prevent that.
 925                  */
 926                 vmci_transport_send_reset(sk, pkt);
 927                 break;
 928         }
 929
 930         release_sock(sk);
 931         kfree(recv_pkt_info);
 932         /* Release reference obtained in the stream callback when we fetched
 933          * this socket out of the bound or connected list.
 934          */
 935         sock_put(sk);
 936 }
 937
 938 static int vmci_transport_recv_listen(struct sock *sk,
 939                                       struct vmci_transport_packet *pkt)
 940 {
 941         struct sock *pending;
 942         struct vsock_sock *vpending;
 943         int err;
 944         u64 qp_size;
 945         bool old_request = false;
 946         bool old_pkt_proto = false;
 947
 948         err = 0;
 949
 950         /* Because we are in the listen state, we could be receiving a packet
 951          * for ourself or any previous connection requests that we received.
 952          * If it's the latter, we try to find a socket in our list of pending
 953          * connections and, if we do, call the appropriate handler for the
 954          * state that that socket is in.  Otherwise we try to service the
 955          * connection request.
 956          */
 957         pending = vmci_transport_get_pending(sk, pkt);
 958         if (pending) {
 959                 lock_sock(pending);
 960
 961                 /* The local context ID may be out of date. */
 962                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 963
 964                 switch (pending->sk_state) {
 965                 case TCP_SYN_SENT:
 966                         err = vmci_transport_recv_connecting_server(sk,
 967                                                                     pending,
 968                                                                     pkt);
 969                         break;
 970                 default:
 971                         vmci_transport_send_reset(pending, pkt);
 972                         err = -EINVAL;
 973                 }
 974
 975                 if (err < 0)
 976                         vsock_remove_pending(sk, pending);
 977
 978                 release_sock(pending);
 979                 vmci_transport_release_pending(pending);
 980
 981                 return err;
 982         }
 983
 984         /* The listen state only accepts connection requests.  Reply with a
 985          * reset unless we received a reset.
 986          */
 987
 988         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 989               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 990                 vmci_transport_reply_reset(pkt);
 991                 return -EINVAL;
 992         }
 993
 994         if (pkt->u.size == 0) {
 995                 vmci_transport_reply_reset(pkt);
 996                 return -EINVAL;
 997         }
 998
 999         /* If this socket can't accommodate this connection request, we send a
1000          * reset.  Otherwise we create and initialize a child socket and reply
1001          * with a connection negotiation.
1002          */
1003         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1004                 vmci_transport_reply_reset(pkt);
1005                 return -ECONNREFUSED;
1006         }
1007
1008         pending = vsock_create_connected(sk);
1009         if (!pending) {
1010                 vmci_transport_send_reset(sk, pkt);
1011                 return -ENOMEM;
1012         }
1013
1014         vpending = vsock_sk(pending);
1015
1016         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1017                         pkt->dst_port);
1018         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1019                         pkt->src_port);
1020
1021         err = vsock_assign_transport(vpending, vsock_sk(sk));
1022         /* Transport assigned (looking at remote_addr) must be the same
1023          * where we received the request.
1024          */
1025         if (err || !vmci_check_transport(vpending)) {
1026                 vmci_transport_send_reset(sk, pkt);
1027                 sock_put(pending);
1028                 return err;
1029         }
1030
1031         /* If the proposed size fits within our min/max, accept it. Otherwise
1032          * propose our own size.
1033          */
1034         if (pkt->u.size >= vpending->buffer_min_size &&
1035             pkt->u.size <= vpending->buffer_max_size) {
1036                 qp_size = pkt->u.size;
1037         } else {
1038                 qp_size = vpending->buffer_size;
1039         }
1040
1041         /* Figure out if we are using old or new requests based on the
1042          * overrides pkt types sent by our peer.
1043          */
1044         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1045                 old_request = old_pkt_proto;
1046         } else {
1047                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1048                         old_request = true;
1049                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1050                         old_request = false;
1051
1052         }
1053
1054         if (old_request) {
1055                 /* Handle a REQUEST (or override) */
1056                 u16 version = VSOCK_PROTO_INVALID;
1057                 if (vmci_transport_proto_to_notify_struct(
1058                         pending, &version, true))
1059                         err = vmci_transport_send_negotiate(pending, qp_size);
1060                 else
1061                         err = -EINVAL;
1062
1063         } else {
1064                 /* Handle a REQUEST2 (or override) */
1065                 int proto_int = pkt->proto;
1066                 int pos;
1067                 u16 active_proto_version = 0;
1068
1069                 /* The list of possible protocols is the intersection of all
1070                  * protocols the client supports ... plus all the protocols we
1071                  * support.
1072                  */
1073                 proto_int &= vmci_transport_new_proto_supported_versions();
1074
1075                 /* We choose the highest possible protocol version and use that
1076                  * one.
1077                  */
1078                 pos = fls(proto_int);
1079                 if (pos) {
1080                         active_proto_version = (1 << (pos - 1));
1081                         if (vmci_transport_proto_to_notify_struct(
1082                                 pending, &active_proto_version, false))
1083                                 err = vmci_transport_send_negotiate2(pending,
1084                                                         qp_size,
1085                                                         active_proto_version);
1086                         else
1087                                 err = -EINVAL;
1088
1089                 } else {
1090                         err = -EINVAL;
1091                 }
1092         }
1093
1094         if (err < 0) {
1095                 vmci_transport_send_reset(sk, pkt);
1096                 sock_put(pending);
1097                 err = vmci_transport_error_to_vsock_error(err);
1098                 goto out;
1099         }
1100
1101         vsock_add_pending(sk, pending);
1102         sk_acceptq_added(sk);
1103
1104         pending->sk_state = TCP_SYN_SENT;
1105         vmci_trans(vpending)->produce_size =
1106                 vmci_trans(vpending)->consume_size = qp_size;
1107         vpending->buffer_size = qp_size;
1108
1109         vmci_trans(vpending)->notify_ops->process_request(pending);
1110
1111         /* We might never receive another message for this socket and it's not
1112          * connected to any process, so we have to ensure it gets cleaned up
1113          * ourself.  Our delayed work function will take care of that.  Note
1114          * that we do not ever cancel this function since we have few
1115          * guarantees about its state when calling cancel_delayed_work().
1116          * Instead we hold a reference on the socket for that function and make
1117          * it capable of handling cases where it needs to do nothing but
1118          * release that reference.
1119          */
1120         vpending->listener = sk;
1121         sock_hold(sk);
1122         sock_hold(pending);
1123         schedule_delayed_work(&vpending->pending_work, HZ);
1124
1125 out:
1126         return err;
1127 }
1128
1129 static int
1130 vmci_transport_recv_connecting_server(struct sock *listener,
1131                                       struct sock *pending,
1132                                       struct vmci_transport_packet *pkt)
1133 {
1134         struct vsock_sock *vpending;
1135         struct vmci_handle handle;
1136         struct vmci_qp *qpair;
1137         bool is_local;
1138         u32 flags;
1139         u32 detach_sub_id;
1140         int err;
1141         int skerr;
1142
1143         vpending = vsock_sk(pending);
1144         detach_sub_id = VMCI_INVALID_ID;
1145
1146         switch (pkt->type) {
1147         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1148                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1149                         vmci_transport_send_reset(pending, pkt);
1150                         skerr = EPROTO;
1151                         err = -EINVAL;
1152                         goto destroy;
1153                 }
1154                 break;
1155         default:
1156                 /* Close and cleanup the connection. */
1157                 vmci_transport_send_reset(pending, pkt);
1158                 skerr = EPROTO;
1159                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1160                 goto destroy;
1161         }
1162
1163         /* In order to complete the connection we need to attach to the offered
1164          * queue pair and send an attach notification.  We also subscribe to the
1165          * detach event so we know when our peer goes away, and we do that
1166          * before attaching so we don't miss an event.  If all this succeeds,
1167          * we update our state and wakeup anything waiting in accept() for a
1168          * connection.
1169          */
1170
1171         /* We don't care about attach since we ensure the other side has
1172          * attached by specifying the ATTACH_ONLY flag below.
1173          */
1174         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1175                                    vmci_transport_peer_detach_cb,
1176                                    vmci_trans(vpending), &detach_sub_id);
1177         if (err < VMCI_SUCCESS) {
1178                 vmci_transport_send_reset(pending, pkt);
1179                 err = vmci_transport_error_to_vsock_error(err);
1180                 skerr = -err;
1181                 goto destroy;
1182         }
1183
1184         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1185
1186         /* Now attach to the queue pair the client created. */
1187         handle = pkt->u.handle;
1188
1189         /* vpending->local_addr always has a context id so we do not need to
1190          * worry about VMADDR_CID_ANY in this case.
1191          */
1192         is_local =
1193             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1194         flags = VMCI_QPFLAG_ATTACH_ONLY;
1195         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1196
1197         err = vmci_transport_queue_pair_alloc(
1198                                         &qpair,
1199                                         &handle,
1200                                         vmci_trans(vpending)->produce_size,
1201                                         vmci_trans(vpending)->consume_size,
1202                                         pkt->dg.src.context,
1203                                         flags,
1204                                         vmci_transport_is_trusted(
1205                                                 vpending,
1206                                                 vpending->remote_addr.svm_cid));
1207         if (err < 0) {
1208                 vmci_transport_send_reset(pending, pkt);
1209                 skerr = -err;
1210                 goto destroy;
1211         }
1212
1213         vmci_trans(vpending)->qp_handle = handle;
1214         vmci_trans(vpending)->qpair = qpair;
1215
1216         /* When we send the attach message, we must be ready to handle incoming
1217          * control messages on the newly connected socket. So we move the
1218          * pending socket to the connected state before sending the attach
1219          * message. Otherwise, an incoming packet triggered by the attach being
1220          * received by the peer may be processed concurrently with what happens
1221          * below after sending the attach message, and that incoming packet
1222          * will find the listening socket instead of the (currently) pending
1223          * socket. Note that enqueueing the socket increments the reference
1224          * count, so even if a reset comes before the connection is accepted,
1225          * the socket will be valid until it is removed from the queue.
1226          *
1227          * If we fail sending the attach below, we remove the socket from the
1228          * connected list and move the socket to TCP_CLOSE before
1229          * releasing the lock, so a pending slow path processing of an incoming
1230          * packet will not see the socket in the connected state in that case.
1231          */
1232         pending->sk_state = TCP_ESTABLISHED;
1233
1234         vsock_insert_connected(vpending);
1235
1236         /* Notify our peer of our attach. */
1237         err = vmci_transport_send_attach(pending, handle);
1238         if (err < 0) {
1239                 vsock_remove_connected(vpending);
1240                 pr_err("Could not send attach\n");
1241                 vmci_transport_send_reset(pending, pkt);
1242                 err = vmci_transport_error_to_vsock_error(err);
1243                 skerr = -err;
1244                 goto destroy;
1245         }
1246
1247         /* We have a connection. Move the now connected socket from the
1248          * listener's pending list to the accept queue so callers of accept()
1249          * can find it.
1250          */
1251         vsock_remove_pending(listener, pending);
1252         vsock_enqueue_accept(listener, pending);
1253
1254         /* Callers of accept() will be be waiting on the listening socket, not
1255          * the pending socket.
1256          */
1257         listener->sk_data_ready(listener);
1258
1259         return 0;
1260
1261 destroy:
1262         pending->sk_err = skerr;
1263         pending->sk_state = TCP_CLOSE;
1264         /* As long as we drop our reference, all necessary cleanup will handle
1265          * when the cleanup function drops its reference and our destruct
1266          * implementation is called.  Note that since the listen handler will
1267          * remove pending from the pending list upon our failure, the cleanup
1268          * function won't drop the additional reference, which is why we do it
1269          * here.
1270          */
1271         sock_put(pending);
1272
1273         return err;
1274 }
1275
1276 static int
1277 vmci_transport_recv_connecting_client(struct sock *sk,
1278                                       struct vmci_transport_packet *pkt)
1279 {
1280         struct vsock_sock *vsk;
1281         int err;
1282         int skerr;
1283
1284         vsk = vsock_sk(sk);
1285
1286         switch (pkt->type) {
1287         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1288                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1289                     !vmci_handle_is_equal(pkt->u.handle,
1290                                           vmci_trans(vsk)->qp_handle)) {
1291                         skerr = EPROTO;
1292                         err = -EINVAL;
1293                         goto destroy;
1294                 }
1295
1296                 /* Signify the socket is connected and wakeup the waiter in
1297                  * connect(). Also place the socket in the connected table for
1298                  * accounting (it can already be found since it's in the bound
1299                  * table).
1300                  */
1301                 sk->sk_state = TCP_ESTABLISHED;
1302                 sk->sk_socket->state = SS_CONNECTED;
1303                 vsock_insert_connected(vsk);
1304                 sk->sk_state_change(sk);
1305
1306                 break;
1307         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1308         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1309                 if (pkt->u.size == 0
1310                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1311                     || pkt->src_port != vsk->remote_addr.svm_port
1312                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1313                     || vmci_trans(vsk)->qpair
1314                     || vmci_trans(vsk)->produce_size != 0
1315                     || vmci_trans(vsk)->consume_size != 0
1316                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1317                         skerr = EPROTO;
1318                         err = -EINVAL;
1319
1320                         goto destroy;
1321                 }
1322
1323                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1324                 if (err) {
1325                         skerr = -err;
1326                         goto destroy;
1327                 }
1328
1329                 break;
1330         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1331                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1332                 if (err) {
1333                         skerr = -err;
1334                         goto destroy;
1335                 }
1336
1337                 break;
1338         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1339                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1340                  * continue processing here after they sent an INVALID packet.
1341                  * This meant that we got a RST after the INVALID. We ignore a
1342                  * RST after an INVALID. The common code doesn't send the RST
1343                  * ... so we can hang if an old version of the common code
1344                  * fails between getting a REQUEST and sending an OFFER back.
1345                  * Not much we can do about it... except hope that it doesn't
1346                  * happen.
1347                  */
1348                 if (vsk->ignore_connecting_rst) {
1349                         vsk->ignore_connecting_rst = false;
1350                 } else {
1351                         skerr = ECONNRESET;
1352                         err = 0;
1353                         goto destroy;
1354                 }
1355
1356                 break;
1357         default:
1358                 /* Close and cleanup the connection. */
1359                 skerr = EPROTO;
1360                 err = -EINVAL;
1361                 goto destroy;
1362         }
1363
1364         return 0;
1365
1366 destroy:
1367         vmci_transport_send_reset(sk, pkt);
1368
1369         sk->sk_state = TCP_CLOSE;
1370         sk->sk_err = skerr;
1371         sk->sk_error_report(sk);
1372         return err;
1373 }
1374
1375 static int vmci_transport_recv_connecting_client_negotiate(
1376                                         struct sock *sk,
1377                                         struct vmci_transport_packet *pkt)
1378 {
1379         int err;
1380         struct vsock_sock *vsk;
1381         struct vmci_handle handle;
1382         struct vmci_qp *qpair;
1383         u32 detach_sub_id;
1384         bool is_local;
1385         u32 flags;
1386         bool old_proto = true;
1387         bool old_pkt_proto;
1388         u16 version;
1389
1390         vsk = vsock_sk(sk);
1391         handle = VMCI_INVALID_HANDLE;
1392         detach_sub_id = VMCI_INVALID_ID;
1393
1394         /* If we have gotten here then we should be past the point where old
1395          * linux vsock could have sent the bogus rst.
1396          */
1397         vsk->sent_request = false;
1398         vsk->ignore_connecting_rst = false;
1399
1400         /* Verify that we're OK with the proposed queue pair size */
1401         if (pkt->u.size < vsk->buffer_min_size ||
1402             pkt->u.size > vsk->buffer_max_size) {
1403                 err = -EINVAL;
1404                 goto destroy;
1405         }
1406
1407         /* At this point we know the CID the peer is using to talk to us. */
1408
1409         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1410                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1411
1412         /* Setup the notify ops to be the highest supported version that both
1413          * the server and the client support.
1414          */
1415
1416         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1417                 old_proto = old_pkt_proto;
1418         } else {
1419                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1420                         old_proto = true;
1421                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1422                         old_proto = false;
1423
1424         }
1425
1426         if (old_proto)
1427                 version = VSOCK_PROTO_INVALID;
1428         else
1429                 version = pkt->proto;
1430
1431         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1432                 err = -EINVAL;
1433                 goto destroy;
1434         }
1435
1436         /* Subscribe to detach events first.
1437          *
1438          * XXX We attach once for each queue pair created for now so it is easy
1439          * to find the socket (it's provided), but later we should only
1440          * subscribe once and add a way to lookup sockets by queue pair handle.
1441          */
1442         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1443                                    vmci_transport_peer_detach_cb,
1444                                    vmci_trans(vsk), &detach_sub_id);
1445         if (err < VMCI_SUCCESS) {
1446                 err = vmci_transport_error_to_vsock_error(err);
1447                 goto destroy;
1448         }
1449
1450         /* Make VMCI select the handle for us. */
1451         handle = VMCI_INVALID_HANDLE;
1452         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1453         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1454
1455         err = vmci_transport_queue_pair_alloc(&qpair,
1456                                               &handle,
1457                                               pkt->u.size,
1458                                               pkt->u.size,
1459                                               vsk->remote_addr.svm_cid,
1460                                               flags,
1461                                               vmci_transport_is_trusted(
1462                                                   vsk,
1463                                                   vsk->
1464                                                   remote_addr.svm_cid));
1465         if (err < 0)
1466                 goto destroy;
1467
1468         err = vmci_transport_send_qp_offer(sk, handle);
1469         if (err < 0) {
1470                 err = vmci_transport_error_to_vsock_error(err);
1471                 goto destroy;
1472         }
1473
1474         vmci_trans(vsk)->qp_handle = handle;
1475         vmci_trans(vsk)->qpair = qpair;
1476
1477         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1478                 pkt->u.size;
1479
1480         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1481
1482         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1483
1484         return 0;
1485
1486 destroy:
1487         if (detach_sub_id != VMCI_INVALID_ID)
1488                 vmci_event_unsubscribe(detach_sub_id);
1489
1490         if (!vmci_handle_is_invalid(handle))
1491                 vmci_qpair_detach(&qpair);
1492
1493         return err;
1494 }
1495
1496 static int
1497 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1498                                               struct vmci_transport_packet *pkt)
1499 {
1500         int err = 0;
1501         struct vsock_sock *vsk = vsock_sk(sk);
1502
1503         if (vsk->sent_request) {
1504                 vsk->sent_request = false;
1505                 vsk->ignore_connecting_rst = true;
1506
1507                 err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1508                 if (err < 0)
1509                         err = vmci_transport_error_to_vsock_error(err);
1510                 else
1511                         err = 0;
1512
1513         }
1514
1515         return err;
1516 }
1517
1518 static int vmci_transport_recv_connected(struct sock *sk,
1519                                          struct vmci_transport_packet *pkt)
1520 {
1521         struct vsock_sock *vsk;
1522         bool pkt_processed = false;
1523
1524         /* In cases where we are closing the connection, it's sufficient to
1525          * mark the state change (and maybe error) and wake up any waiting
1526          * threads. Since this is a connected socket, it's owned by a user
1527          * process and will be cleaned up when the failure is passed back on
1528          * the current or next system call.  Our system call implementations
1529          * must therefore check for error and state changes on entry and when
1530          * being awoken.
1531          */
1532         switch (pkt->type) {
1533         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1534                 if (pkt->u.mode) {
1535                         vsk = vsock_sk(sk);
1536
1537                         vsk->peer_shutdown |= pkt->u.mode;
1538                         sk->sk_state_change(sk);
1539                 }
1540                 break;
1541
1542         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1543                 vsk = vsock_sk(sk);
1544                 /* It is possible that we sent our peer a message (e.g a
1545                  * WAITING_READ) right before we got notified that the peer had
1546                  * detached. If that happens then we can get a RST pkt back
1547                  * from our peer even though there is data available for us to
1548                  * read. In that case, don't shutdown the socket completely but
1549                  * instead allow the local client to finish reading data off
1550                  * the queuepair. Always treat a RST pkt in connected mode like
1551                  * a clean shutdown.
1552                  */
1553                 sock_set_flag(sk, SOCK_DONE);
1554                 vsk->peer_shutdown = SHUTDOWN_MASK;
1555                 if (vsock_stream_has_data(vsk) <= 0)
1556                         sk->sk_state = TCP_CLOSING;
1557
1558                 sk->sk_state_change(sk);
1559                 break;
1560
1561         default:
1562                 vsk = vsock_sk(sk);
1563                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1564                                 sk, pkt, false, NULL, NULL,
1565                                 &pkt_processed);
1566                 if (!pkt_processed)
1567                         return -EINVAL;
1568
1569                 break;
1570         }
1571
1572         return 0;
1573 }
1574
1575 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1576                                       struct vsock_sock *psk)
1577 {
1578         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1579         if (!vsk->trans)
1580                 return -ENOMEM;
1581
1582         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1583         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1584         vmci_trans(vsk)->qpair = NULL;
1585         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1586         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1587         vmci_trans(vsk)->notify_ops = NULL;
1588         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1589         vmci_trans(vsk)->sk = &vsk->sk;
1590         spin_lock_init(&vmci_trans(vsk)->lock);
1591
1592         return 0;
1593 }
1594
1595 static void vmci_transport_free_resources(struct list_head *transport_list)
1596 {
1597         while (!list_empty(transport_list)) {
1598                 struct vmci_transport *transport =
1599                     list_first_entry(transport_list, struct vmci_transport,
1600                                      elem);
1601                 list_del(&transport->elem);
1602
1603                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1604                         vmci_event_unsubscribe(transport->detach_sub_id);
1605                         transport->detach_sub_id = VMCI_INVALID_ID;
1606                 }
1607
1608                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1609                         vmci_qpair_detach(&transport->qpair);
1610                         transport->qp_handle = VMCI_INVALID_HANDLE;
1611                         transport->produce_size = 0;
1612                         transport->consume_size = 0;
1613                 }
1614
1615                 kfree(transport);
1616         }
1617 }
1618
1619 static void vmci_transport_cleanup(struct work_struct *work)
1620 {
1621         LIST_HEAD(pending);
1622
1623         spin_lock_bh(&vmci_transport_cleanup_lock);
1624         list_replace_init(&vmci_transport_cleanup_list, &pending);
1625         spin_unlock_bh(&vmci_transport_cleanup_lock);
1626         vmci_transport_free_resources(&pending);
1627 }
1628
1629 static void vmci_transport_destruct(struct vsock_sock *vsk)
1630 {
1631         /* transport can be NULL if we hit a failure at init() time */
1632         if (!vmci_trans(vsk))
1633                 return;
1634
1635         /* Ensure that the detach callback doesn't use the sk/vsk
1636          * we are about to destruct.
1637          */
1638         spin_lock_bh(&vmci_trans(vsk)->lock);
1639         vmci_trans(vsk)->sk = NULL;
1640         spin_unlock_bh(&vmci_trans(vsk)->lock);
1641
1642         if (vmci_trans(vsk)->notify_ops)
1643                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1644
1645         spin_lock_bh(&vmci_transport_cleanup_lock);
1646         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1647         spin_unlock_bh(&vmci_transport_cleanup_lock);
1648         schedule_work(&vmci_transport_cleanup_work);
1649
1650         vsk->trans = NULL;
1651 }
1652
1653 static void vmci_transport_release(struct vsock_sock *vsk)
1654 {
1655         vsock_remove_sock(vsk);
1656
1657         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1658                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1659                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1660         }
1661 }
1662
1663 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1664                                      struct sockaddr_vm *addr)
1665 {
1666         u32 port;
1667         u32 flags;
1668         int err;
1669
1670         /* VMCI will select a resource ID for us if we provide
1671          * VMCI_INVALID_ID.
1672          */
1673         port = addr->svm_port == VMADDR_PORT_ANY ?
1674                         VMCI_INVALID_ID : addr->svm_port;
1675
1676         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1677                 return -EACCES;
1678
1679         flags = addr->svm_cid == VMADDR_CID_ANY ?
1680                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1681
1682         err = vmci_transport_datagram_create_hnd(port, flags,
1683                                                  vmci_transport_recv_dgram_cb,
1684                                                  &vsk->sk,
1685                                                  &vmci_trans(vsk)->dg_handle);
1686         if (err < VMCI_SUCCESS)
1687                 return vmci_transport_error_to_vsock_error(err);
1688         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1689                         vmci_trans(vsk)->dg_handle.resource);
1690
1691         return 0;
1692 }
1693
1694 static int vmci_transport_dgram_enqueue(
1695         struct vsock_sock *vsk,
1696         struct sockaddr_vm *remote_addr,
1697         struct msghdr *msg,
1698         size_t len)
1699 {
1700         int err;
1701         struct vmci_datagram *dg;
1702
1703         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1704                 return -EMSGSIZE;
1705
1706         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1707                 return -EPERM;
1708
1709         /* Allocate a buffer for the user's message and our packet header. */
1710         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1711         if (!dg)
1712                 return -ENOMEM;
1713
1714         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1715
1716         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1717                                    remote_addr->svm_port);
1718         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1719                                    vsk->local_addr.svm_port);
1720         dg->payload_size = len;
1721
1722         err = vmci_datagram_send(dg);
1723         kfree(dg);
1724         if (err < 0)
1725                 return vmci_transport_error_to_vsock_error(err);
1726
1727         return err - sizeof(*dg);
1728 }
1729
1730 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1731                                         struct msghdr *msg, size_t len,
1732                                         int flags)
1733 {
1734         int err;
1735         int noblock;
1736         struct vmci_datagram *dg;
1737         size_t payload_len;
1738         struct sk_buff *skb;
1739
1740         noblock = flags & MSG_DONTWAIT;
1741
1742         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1743                 return -EOPNOTSUPP;
1744
1745         /* Retrieve the head sk_buff from the socket's receive queue. */
1746         err = 0;
1747         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1748         if (!skb)
1749                 return err;
1750
1751         dg = (struct vmci_datagram *)skb->data;
1752         if (!dg)
1753                 /* err is 0, meaning we read zero bytes. */
1754                 goto out;
1755
1756         payload_len = dg->payload_size;
1757         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1758         if (payload_len != skb->len - sizeof(*dg)) {
1759                 err = -EINVAL;
1760                 goto out;
1761         }
1762
1763         if (payload_len > len) {
1764                 payload_len = len;
1765                 msg->msg_flags |= MSG_TRUNC;
1766         }
1767
1768         /* Place the datagram payload in the user's iovec. */
1769         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1770         if (err)
1771                 goto out;
1772
1773         if (msg->msg_name) {
1774                 /* Provide the address of the sender. */
1775                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1776                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1777                 msg->msg_namelen = sizeof(*vm_addr);
1778         }
1779         err = payload_len;
1780
1781 out:
1782         skb_free_datagram(&vsk->sk, skb);
1783         return err;
1784 }
1785
1786 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1787 {
1788         if (cid == VMADDR_CID_HYPERVISOR) {
1789                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1790                  * state and are allowed.
1791                  */
1792                 return port == VMCI_UNITY_PBRPC_REGISTER;
1793         }
1794
1795         return true;
1796 }
1797
1798 static int vmci_transport_connect(struct vsock_sock *vsk)
1799 {
1800         int err;
1801         bool old_pkt_proto = false;
1802         struct sock *sk = &vsk->sk;
1803
1804         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1805                 old_pkt_proto) {
1806                 err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1807                 if (err < 0) {
1808                         sk->sk_state = TCP_CLOSE;
1809                         return err;
1810                 }
1811         } else {
1812                 int supported_proto_versions =
1813                         vmci_transport_new_proto_supported_versions();
1814                 err = vmci_transport_send_conn_request2(sk, vsk->buffer_size,
1815                                 supported_proto_versions);
1816                 if (err < 0) {
1817                         sk->sk_state = TCP_CLOSE;
1818                         return err;
1819                 }
1820
1821                 vsk->sent_request = true;
1822         }
1823
1824         return err;
1825 }
1826
1827 static ssize_t vmci_transport_stream_dequeue(
1828         struct vsock_sock *vsk,
1829         struct msghdr *msg,
1830         size_t len,
1831         int flags)
1832 {
1833         if (flags & MSG_PEEK)
1834                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1835         else
1836                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1837 }
1838
1839 static ssize_t vmci_transport_stream_enqueue(
1840         struct vsock_sock *vsk,
1841         struct msghdr *msg,
1842         size_t len)
1843 {
1844         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1845 }
1846
1847 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1848 {
1849         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1850 }
1851
1852 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1853 {
1854         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1855 }
1856
1857 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1858 {
1859         return vmci_trans(vsk)->consume_size;
1860 }
1861
1862 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1863 {
1864         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1865 }
1866
1867 static int vmci_transport_notify_poll_in(
1868         struct vsock_sock *vsk,
1869         size_t target,
1870         bool *data_ready_now)
1871 {
1872         return vmci_trans(vsk)->notify_ops->poll_in(
1873                         &vsk->sk, target, data_ready_now);
1874 }
1875
1876 static int vmci_transport_notify_poll_out(
1877         struct vsock_sock *vsk,
1878         size_t target,
1879         bool *space_available_now)
1880 {
1881         return vmci_trans(vsk)->notify_ops->poll_out(
1882                         &vsk->sk, target, space_available_now);
1883 }
1884
1885 static int vmci_transport_notify_recv_init(
1886         struct vsock_sock *vsk,
1887         size_t target,
1888         struct vsock_transport_recv_notify_data *data)
1889 {
1890         return vmci_trans(vsk)->notify_ops->recv_init(
1891                         &vsk->sk, target,
1892                         (struct vmci_transport_recv_notify_data *)data);
1893 }
1894
1895 static int vmci_transport_notify_recv_pre_block(
1896         struct vsock_sock *vsk,
1897         size_t target,
1898         struct vsock_transport_recv_notify_data *data)
1899 {
1900         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1901                         &vsk->sk, target,
1902                         (struct vmci_transport_recv_notify_data *)data);
1903 }
1904
1905 static int vmci_transport_notify_recv_pre_dequeue(
1906         struct vsock_sock *vsk,
1907         size_t target,
1908         struct vsock_transport_recv_notify_data *data)
1909 {
1910         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1911                         &vsk->sk, target,
1912                         (struct vmci_transport_recv_notify_data *)data);
1913 }
1914
1915 static int vmci_transport_notify_recv_post_dequeue(
1916         struct vsock_sock *vsk,
1917         size_t target,
1918         ssize_t copied,
1919         bool data_read,
1920         struct vsock_transport_recv_notify_data *data)
1921 {
1922         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1923                         &vsk->sk, target, copied, data_read,
1924                         (struct vmci_transport_recv_notify_data *)data);
1925 }
1926
1927 static int vmci_transport_notify_send_init(
1928         struct vsock_sock *vsk,
1929         struct vsock_transport_send_notify_data *data)
1930 {
1931         return vmci_trans(vsk)->notify_ops->send_init(
1932                         &vsk->sk,
1933                         (struct vmci_transport_send_notify_data *)data);
1934 }
1935
1936 static int vmci_transport_notify_send_pre_block(
1937         struct vsock_sock *vsk,
1938         struct vsock_transport_send_notify_data *data)
1939 {
1940         return vmci_trans(vsk)->notify_ops->send_pre_block(
1941                         &vsk->sk,
1942                         (struct vmci_transport_send_notify_data *)data);
1943 }
1944
1945 static int vmci_transport_notify_send_pre_enqueue(
1946         struct vsock_sock *vsk,
1947         struct vsock_transport_send_notify_data *data)
1948 {
1949         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1950                         &vsk->sk,
1951                         (struct vmci_transport_send_notify_data *)data);
1952 }
1953
1954 static int vmci_transport_notify_send_post_enqueue(
1955         struct vsock_sock *vsk,
1956         ssize_t written,
1957         struct vsock_transport_send_notify_data *data)
1958 {
1959         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1960                         &vsk->sk, written,
1961                         (struct vmci_transport_send_notify_data *)data);
1962 }
1963
1964 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1965 {
1966         if (PROTOCOL_OVERRIDE != -1) {
1967                 if (PROTOCOL_OVERRIDE == 0)
1968                         *old_pkt_proto = true;
1969                 else
1970                         *old_pkt_proto = false;
1971
1972                 pr_info("Proto override in use\n");
1973                 return true;
1974         }
1975
1976         return false;
1977 }
1978
1979 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
1980                                                   u16 *proto,
1981                                                   bool old_pkt_proto)
1982 {
1983         struct vsock_sock *vsk = vsock_sk(sk);
1984
1985         if (old_pkt_proto) {
1986                 if (*proto != VSOCK_PROTO_INVALID) {
1987                         pr_err("Can't set both an old and new protocol\n");
1988                         return false;
1989                 }
1990                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
1991                 goto exit;
1992         }
1993
1994         switch (*proto) {
1995         case VSOCK_PROTO_PKT_ON_NOTIFY:
1996                 vmci_trans(vsk)->notify_ops =
1997                         &vmci_transport_notify_pkt_q_state_ops;
1998                 break;
1999         default:
2000                 pr_err("Unknown notify protocol version\n");
2001                 return false;
2002         }
2003
2004 exit:
2005         vmci_trans(vsk)->notify_ops->socket_init(sk);
2006         return true;
2007 }
2008
2009 static u16 vmci_transport_new_proto_supported_versions(void)
2010 {
2011         if (PROTOCOL_OVERRIDE != -1)
2012                 return PROTOCOL_OVERRIDE;
2013
2014         return VSOCK_PROTO_ALL_SUPPORTED;
2015 }
2016
2017 static u32 vmci_transport_get_local_cid(void)
2018 {
2019         return vmci_get_context_id();
2020 }
2021
2022 static struct vsock_transport vmci_transport = {
2023         .module = THIS_MODULE,
2024         .init = vmci_transport_socket_init,
2025         .destruct = vmci_transport_destruct,
2026         .release = vmci_transport_release,
2027         .connect = vmci_transport_connect,
2028         .dgram_bind = vmci_transport_dgram_bind,
2029         .dgram_dequeue = vmci_transport_dgram_dequeue,
2030         .dgram_enqueue = vmci_transport_dgram_enqueue,
2031         .dgram_allow = vmci_transport_dgram_allow,
2032         .stream_dequeue = vmci_transport_stream_dequeue,
2033         .stream_enqueue = vmci_transport_stream_enqueue,
2034         .stream_has_data = vmci_transport_stream_has_data,
2035         .stream_has_space = vmci_transport_stream_has_space,
2036         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2037         .stream_is_active = vmci_transport_stream_is_active,
2038         .stream_allow = vmci_transport_stream_allow,
2039         .notify_poll_in = vmci_transport_notify_poll_in,
2040         .notify_poll_out = vmci_transport_notify_poll_out,
2041         .notify_recv_init = vmci_transport_notify_recv_init,
2042         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2043         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2044         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2045         .notify_send_init = vmci_transport_notify_send_init,
2046         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2047         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2048         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2049         .shutdown = vmci_transport_shutdown,
2050         .get_local_cid = vmci_transport_get_local_cid,
2051 };
2052
2053 static bool vmci_check_transport(struct vsock_sock *vsk)
2054 {
2055         return vsk->transport == &vmci_transport;
2056 }
2057
2058 static void vmci_vsock_transport_cb(bool is_host)
2059 {
2060         int features;
2061
2062         if (is_host)
2063                 features = VSOCK_TRANSPORT_F_H2G;
2064         else
2065                 features = VSOCK_TRANSPORT_F_G2H;
2066
2067         vsock_core_register(&vmci_transport, features);
2068 }
2069
2070 static int __init vmci_transport_init(void)
2071 {
2072         int err;
2073
2074         /* Create the datagram handle that we will use to send and receive all
2075          * VSocket control messages for this context.
2076          */
2077         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2078                                                  VMCI_FLAG_ANYCID_DG_HND,
2079                                                  vmci_transport_recv_stream_cb,
2080                                                  NULL,
2081                                                  &vmci_transport_stream_handle);
2082         if (err < VMCI_SUCCESS) {
2083                 pr_err("Unable to create datagram handle. (%d)\n", err);
2084                 return vmci_transport_error_to_vsock_error(err);
2085         }
2086         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2087                                    vmci_transport_qp_resumed_cb,
2088                                    NULL, &vmci_transport_qp_resumed_sub_id);
2089         if (err < VMCI_SUCCESS) {
2090                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2091                 err = vmci_transport_error_to_vsock_error(err);
2092                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2093                 goto err_destroy_stream_handle;
2094         }
2095
2096         /* Register only with dgram feature, other features (H2G, G2H) will be
2097          * registered when the first host or guest becomes active.
2098          */
2099         err = vsock_core_register(&vmci_transport, VSOCK_TRANSPORT_F_DGRAM);
2100         if (err < 0)
2101                 goto err_unsubscribe;
2102
2103         err = vmci_register_vsock_callback(vmci_vsock_transport_cb);
2104         if (err < 0)
2105                 goto err_unregister;
2106
2107         return 0;
2108
2109 err_unregister:
2110         vsock_core_unregister(&vmci_transport);
2111 err_unsubscribe:
2112         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2113 err_destroy_stream_handle:
2114         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2115         return err;
2116 }
2117 module_init(vmci_transport_init);
2118
2119 static void __exit vmci_transport_exit(void)
2120 {
2121         cancel_work_sync(&vmci_transport_cleanup_work);
2122         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2123
2124         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2125                 if (vmci_datagram_destroy_handle(
2126                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2127                         pr_err("Couldn't destroy datagram handle\n");
2128                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2129         }
2130
2131         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2132                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2133                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2134         }
2135
2136         vmci_register_vsock_callback(NULL);
2137         vsock_core_unregister(&vmci_transport);
2138 }
2139 module_exit(vmci_transport_exit);
2140
2141 MODULE_AUTHOR("VMware, Inc.");
2142 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2143 MODULE_VERSION("1.0.5.0-k");
2144 MODULE_LICENSE("GPL v2");
2145 MODULE_ALIAS("vmware_vsock");
2146 MODULE_ALIAS_NETPROTO(PF_VSOCK);