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_once("Could not attach to queue pair with %d\n", err);
 572                 err = vmci_transport_error_to_vsock_error(err);
 573         }
 574
 575         return err;
 576 }
 577
 578 static int
 579 vmci_transport_datagram_create_hnd(u32 resource_id,
 580                                    u32 flags,
 581                                    vmci_datagram_recv_cb recv_cb,
 582                                    void *client_data,
 583                                    struct vmci_handle *out_handle)
 584 {
 585         int err = 0;
 586
 587         /* Try to allocate our datagram handler as trusted. This will only work
 588          * if vsock is running in the host.
 589          */
 590
 591         err = vmci_datagram_create_handle_priv(resource_id, flags,
 592                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 593                                                recv_cb,
 594                                                client_data, out_handle);
 595
 596         if (err == VMCI_ERROR_NO_ACCESS)
 597                 err = vmci_datagram_create_handle(resource_id, flags,
 598                                                   recv_cb, client_data,
 599                                                   out_handle);
 600
 601         return err;
 602 }
 603
 604 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 605  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 606  * sleep it should defer that work to a work queue.
 607  */
 608
 609 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 610 {
 611         struct sock *sk;
 612         size_t size;
 613         struct sk_buff *skb;
 614         struct vsock_sock *vsk;
 615
 616         sk = (struct sock *)data;
 617
 618         /* This handler is privileged when this module is running on the host.
 619          * We will get datagrams from all endpoints (even VMs that are in a
 620          * restricted context). If we get one from a restricted context then
 621          * the destination socket must be trusted.
 622          *
 623          * NOTE: We access the socket struct without holding the lock here.
 624          * This is ok because the field we are interested is never modified
 625          * outside of the create and destruct socket functions.
 626          */
 627         vsk = vsock_sk(sk);
 628         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 629                 return VMCI_ERROR_NO_ACCESS;
 630
 631         size = VMCI_DG_SIZE(dg);
 632
 633         /* Attach the packet to the socket's receive queue as an sk_buff. */
 634         skb = alloc_skb(size, GFP_ATOMIC);
 635         if (!skb)
 636                 return VMCI_ERROR_NO_MEM;
 637
 638         /* sk_receive_skb() will do a sock_put(), so hold here. */
 639         sock_hold(sk);
 640         skb_put(skb, size);
 641         memcpy(skb->data, dg, size);
 642         sk_receive_skb(sk, skb, 0);
 643
 644         return VMCI_SUCCESS;
 645 }
 646
 647 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 648 {
 649         static const u32 non_socket_contexts[] = {
 650                 VMADDR_CID_LOCAL,
 651         };
 652         int i;
 653
 654         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 655
 656         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 657                 if (cid == non_socket_contexts[i])
 658                         return false;
 659         }
 660
 661         return true;
 662 }
 663
 664 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 665  * interrupt fires.  This is run in bottom-half context but it defers most of
 666  * its work to the packet handling work queue.
 667  */
 668
 669 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 670 {
 671         struct sock *sk;
 672         struct sockaddr_vm dst;
 673         struct sockaddr_vm src;
 674         struct vmci_transport_packet *pkt;
 675         struct vsock_sock *vsk;
 676         bool bh_process_pkt;
 677         int err;
 678
 679         sk = NULL;
 680         err = VMCI_SUCCESS;
 681         bh_process_pkt = false;
 682
 683         /* Ignore incoming packets from contexts without sockets, or resources
 684          * that aren't vsock implementations.
 685          */
 686
 687         if (!vmci_transport_stream_allow(dg->src.context, -1)
 688             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 689                 return VMCI_ERROR_NO_ACCESS;
 690
 691         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 692                 /* Drop datagrams that do not contain full VSock packets. */
 693                 return VMCI_ERROR_INVALID_ARGS;
 694
 695         pkt = (struct vmci_transport_packet *)dg;
 696
 697         /* Find the socket that should handle this packet.  First we look for a
 698          * connected socket and if there is none we look for a socket bound to
 699          * the destintation address.
 700          */
 701         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 702         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 703
 704         sk = vsock_find_connected_socket(&src, &dst);
 705         if (!sk) {
 706                 sk = vsock_find_bound_socket(&dst);
 707                 if (!sk) {
 708                         /* We could not find a socket for this specified
 709                          * address.  If this packet is a RST, we just drop it.
 710                          * If it is another packet, we send a RST.  Note that
 711                          * we do not send a RST reply to RSTs so that we do not
 712                          * continually send RSTs between two endpoints.
 713                          *
 714                          * Note that since this is a reply, dst is src and src
 715                          * is dst.
 716                          */
 717                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 718                                 pr_err("unable to send reset\n");
 719
 720                         err = VMCI_ERROR_NOT_FOUND;
 721                         goto out;
 722                 }
 723         }
 724
 725         /* If the received packet type is beyond all types known to this
 726          * implementation, reply with an invalid message.  Hopefully this will
 727          * help when implementing backwards compatibility in the future.
 728          */
 729         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 730                 vmci_transport_send_invalid_bh(&dst, &src);
 731                 err = VMCI_ERROR_INVALID_ARGS;
 732                 goto out;
 733         }
 734
 735         /* This handler is privileged when this module is running on the host.
 736          * We will get datagram connect requests from all endpoints (even VMs
 737          * that are in a restricted context). If we get one from a restricted
 738          * context then the destination socket must be trusted.
 739          *
 740          * NOTE: We access the socket struct without holding the lock here.
 741          * This is ok because the field we are interested is never modified
 742          * outside of the create and destruct socket functions.
 743          */
 744         vsk = vsock_sk(sk);
 745         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 746                 err = VMCI_ERROR_NO_ACCESS;
 747                 goto out;
 748         }
 749
 750         /* We do most everything in a work queue, but let's fast path the
 751          * notification of reads and writes to help data transfer performance.
 752          * We can only do this if there is no process context code executing
 753          * for this socket since that may change the state.
 754          */
 755         bh_lock_sock(sk);
 756
 757         if (!sock_owned_by_user(sk)) {
 758                 /* The local context ID may be out of date, update it. */
 759                 vsk->local_addr.svm_cid = dst.svm_cid;
 760
 761                 if (sk->sk_state == TCP_ESTABLISHED)
 762                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 763                                         sk, pkt, true, &dst, &src,
 764                                         &bh_process_pkt);
 765         }
 766
 767         bh_unlock_sock(sk);
 768
 769         if (!bh_process_pkt) {
 770                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 771
 772                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 773                 if (!recv_pkt_info) {
 774                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 775                                 pr_err("unable to send reset\n");
 776
 777                         err = VMCI_ERROR_NO_MEM;
 778                         goto out;
 779                 }
 780
 781                 recv_pkt_info->sk = sk;
 782                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 783                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 784
 785                 schedule_work(&recv_pkt_info->work);
 786                 /* Clear sk so that the reference count incremented by one of
 787                  * the Find functions above is not decremented below.  We need
 788                  * that reference count for the packet handler we've scheduled
 789                  * to run.
 790                  */
 791                 sk = NULL;
 792         }
 793
 794 out:
 795         if (sk)
 796                 sock_put(sk);
 797
 798         return err;
 799 }
 800
 801 static void vmci_transport_handle_detach(struct sock *sk)
 802 {
 803         struct vsock_sock *vsk;
 804
 805         vsk = vsock_sk(sk);
 806         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 807                 sock_set_flag(sk, SOCK_DONE);
 808
 809                 /* On a detach the peer will not be sending or receiving
 810                  * anymore.
 811                  */
 812                 vsk->peer_shutdown = SHUTDOWN_MASK;
 813
 814                 /* We should not be sending anymore since the peer won't be
 815                  * there to receive, but we can still receive if there is data
 816                  * left in our consume queue. If the local endpoint is a host,
 817                  * we can't call vsock_stream_has_data, since that may block,
 818                  * but a host endpoint can't read data once the VM has
 819                  * detached, so there is no available data in that case.
 820                  */
 821                 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
 822                     vsock_stream_has_data(vsk) <= 0) {
 823                         if (sk->sk_state == TCP_SYN_SENT) {
 824                                 /* The peer may detach from a queue pair while
 825                                  * we are still in the connecting state, i.e.,
 826                                  * if the peer VM is killed after attaching to
 827                                  * a queue pair, but before we complete the
 828                                  * handshake. In that case, we treat the detach
 829                                  * event like a reset.
 830                                  */
 831
 832                                 sk->sk_state = TCP_CLOSE;
 833                                 sk->sk_err = ECONNRESET;
 834                                 sk_error_report(sk);
 835                                 return;
 836                         }
 837                         sk->sk_state = TCP_CLOSE;
 838                 }
 839                 sk->sk_state_change(sk);
 840         }
 841 }
 842
 843 static void vmci_transport_peer_detach_cb(u32 sub_id,
 844                                           const struct vmci_event_data *e_data,
 845                                           void *client_data)
 846 {
 847         struct vmci_transport *trans = client_data;
 848         const struct vmci_event_payload_qp *e_payload;
 849
 850         e_payload = vmci_event_data_const_payload(e_data);
 851
 852         /* XXX This is lame, we should provide a way to lookup sockets by
 853          * qp_handle.
 854          */
 855         if (vmci_handle_is_invalid(e_payload->handle) ||
 856             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 857                 return;
 858
 859         /* We don't ask for delayed CBs when we subscribe to this event (we
 860          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 861          * guarantees in that case about what context we might be running in,
 862          * so it could be BH or process, blockable or non-blockable.  So we
 863          * need to account for all possible contexts here.
 864          */
 865         spin_lock_bh(&trans->lock);
 866         if (!trans->sk)
 867                 goto out;
 868
 869         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 870          * where trans->sk isn't locked.
 871          */
 872         bh_lock_sock(trans->sk);
 873
 874         vmci_transport_handle_detach(trans->sk);
 875
 876         bh_unlock_sock(trans->sk);
 877  out:
 878         spin_unlock_bh(&trans->lock);
 879 }
 880
 881 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 882                                          const struct vmci_event_data *e_data,
 883                                          void *client_data)
 884 {
 885         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 886 }
 887
 888 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 889 {
 890         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 891         struct vmci_transport_packet *pkt;
 892         struct sock *sk;
 893
 894         recv_pkt_info =
 895                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 896         sk = recv_pkt_info->sk;
 897         pkt = &recv_pkt_info->pkt;
 898
 899         lock_sock(sk);
 900
 901         /* The local context ID may be out of date. */
 902         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 903
 904         switch (sk->sk_state) {
 905         case TCP_LISTEN:
 906                 vmci_transport_recv_listen(sk, pkt);
 907                 break;
 908         case TCP_SYN_SENT:
 909                 /* Processing of pending connections for servers goes through
 910                  * the listening socket, so see vmci_transport_recv_listen()
 911                  * for that path.
 912                  */
 913                 vmci_transport_recv_connecting_client(sk, pkt);
 914                 break;
 915         case TCP_ESTABLISHED:
 916                 vmci_transport_recv_connected(sk, pkt);
 917                 break;
 918         default:
 919                 /* Because this function does not run in the same context as
 920                  * vmci_transport_recv_stream_cb it is possible that the
 921                  * socket has closed. We need to let the other side know or it
 922                  * could be sitting in a connect and hang forever. Send a
 923                  * reset to prevent that.
 924                  */
 925                 vmci_transport_send_reset(sk, pkt);
 926                 break;
 927         }
 928
 929         release_sock(sk);
 930         kfree(recv_pkt_info);
 931         /* Release reference obtained in the stream callback when we fetched
 932          * this socket out of the bound or connected list.
 933          */
 934         sock_put(sk);
 935 }
 936
 937 static int vmci_transport_recv_listen(struct sock *sk,
 938                                       struct vmci_transport_packet *pkt)
 939 {
 940         struct sock *pending;
 941         struct vsock_sock *vpending;
 942         int err;
 943         u64 qp_size;
 944         bool old_request = false;
 945         bool old_pkt_proto = false;
 946
 947         /* Because we are in the listen state, we could be receiving a packet
 948          * for ourself or any previous connection requests that we received.
 949          * If it's the latter, we try to find a socket in our list of pending
 950          * connections and, if we do, call the appropriate handler for the
 951          * state that that socket is in.  Otherwise we try to service the
 952          * connection request.
 953          */
 954         pending = vmci_transport_get_pending(sk, pkt);
 955         if (pending) {
 956                 lock_sock(pending);
 957
 958                 /* The local context ID may be out of date. */
 959                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 960
 961                 switch (pending->sk_state) {
 962                 case TCP_SYN_SENT:
 963                         err = vmci_transport_recv_connecting_server(sk,
 964                                                                     pending,
 965                                                                     pkt);
 966                         break;
 967                 default:
 968                         vmci_transport_send_reset(pending, pkt);
 969                         err = -EINVAL;
 970                 }
 971
 972                 if (err < 0)
 973                         vsock_remove_pending(sk, pending);
 974
 975                 release_sock(pending);
 976                 vmci_transport_release_pending(pending);
 977
 978                 return err;
 979         }
 980
 981         /* The listen state only accepts connection requests.  Reply with a
 982          * reset unless we received a reset.
 983          */
 984
 985         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 986               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 987                 vmci_transport_reply_reset(pkt);
 988                 return -EINVAL;
 989         }
 990
 991         if (pkt->u.size == 0) {
 992                 vmci_transport_reply_reset(pkt);
 993                 return -EINVAL;
 994         }
 995
 996         /* If this socket can't accommodate this connection request, we send a
 997          * reset.  Otherwise we create and initialize a child socket and reply
 998          * with a connection negotiation.
 999          */
1000         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1001                 vmci_transport_reply_reset(pkt);
1002                 return -ECONNREFUSED;
1003         }
1004
1005         pending = vsock_create_connected(sk);
1006         if (!pending) {
1007                 vmci_transport_send_reset(sk, pkt);
1008                 return -ENOMEM;
1009         }
1010
1011         vpending = vsock_sk(pending);
1012
1013         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1014                         pkt->dst_port);
1015         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1016                         pkt->src_port);
1017
1018         err = vsock_assign_transport(vpending, vsock_sk(sk));
1019         /* Transport assigned (looking at remote_addr) must be the same
1020          * where we received the request.
1021          */
1022         if (err || !vmci_check_transport(vpending)) {
1023                 vmci_transport_send_reset(sk, pkt);
1024                 sock_put(pending);
1025                 return err;
1026         }
1027
1028         /* If the proposed size fits within our min/max, accept it. Otherwise
1029          * propose our own size.
1030          */
1031         if (pkt->u.size >= vpending->buffer_min_size &&
1032             pkt->u.size <= vpending->buffer_max_size) {
1033                 qp_size = pkt->u.size;
1034         } else {
1035                 qp_size = vpending->buffer_size;
1036         }
1037
1038         /* Figure out if we are using old or new requests based on the
1039          * overrides pkt types sent by our peer.
1040          */
1041         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1042                 old_request = old_pkt_proto;
1043         } else {
1044                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1045                         old_request = true;
1046                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1047                         old_request = false;
1048
1049         }
1050
1051         if (old_request) {
1052                 /* Handle a REQUEST (or override) */
1053                 u16 version = VSOCK_PROTO_INVALID;
1054                 if (vmci_transport_proto_to_notify_struct(
1055                         pending, &version, true))
1056                         err = vmci_transport_send_negotiate(pending, qp_size);
1057                 else
1058                         err = -EINVAL;
1059
1060         } else {
1061                 /* Handle a REQUEST2 (or override) */
1062                 int proto_int = pkt->proto;
1063                 int pos;
1064                 u16 active_proto_version = 0;
1065
1066                 /* The list of possible protocols is the intersection of all
1067                  * protocols the client supports ... plus all the protocols we
1068                  * support.
1069                  */
1070                 proto_int &= vmci_transport_new_proto_supported_versions();
1071
1072                 /* We choose the highest possible protocol version and use that
1073                  * one.
1074                  */
1075                 pos = fls(proto_int);
1076                 if (pos) {
1077                         active_proto_version = (1 << (pos - 1));
1078                         if (vmci_transport_proto_to_notify_struct(
1079                                 pending, &active_proto_version, false))
1080                                 err = vmci_transport_send_negotiate2(pending,
1081                                                         qp_size,
1082                                                         active_proto_version);
1083                         else
1084                                 err = -EINVAL;
1085
1086                 } else {
1087                         err = -EINVAL;
1088                 }
1089         }
1090
1091         if (err < 0) {
1092                 vmci_transport_send_reset(sk, pkt);
1093                 sock_put(pending);
1094                 err = vmci_transport_error_to_vsock_error(err);
1095                 goto out;
1096         }
1097
1098         vsock_add_pending(sk, pending);
1099         sk_acceptq_added(sk);
1100
1101         pending->sk_state = TCP_SYN_SENT;
1102         vmci_trans(vpending)->produce_size =
1103                 vmci_trans(vpending)->consume_size = qp_size;
1104         vpending->buffer_size = qp_size;
1105
1106         vmci_trans(vpending)->notify_ops->process_request(pending);
1107
1108         /* We might never receive another message for this socket and it's not
1109          * connected to any process, so we have to ensure it gets cleaned up
1110          * ourself.  Our delayed work function will take care of that.  Note
1111          * that we do not ever cancel this function since we have few
1112          * guarantees about its state when calling cancel_delayed_work().
1113          * Instead we hold a reference on the socket for that function and make
1114          * it capable of handling cases where it needs to do nothing but
1115          * release that reference.
1116          */
1117         vpending->listener = sk;
1118         sock_hold(sk);
1119         sock_hold(pending);
1120         schedule_delayed_work(&vpending->pending_work, HZ);
1121
1122 out:
1123         return err;
1124 }
1125
1126 static int
1127 vmci_transport_recv_connecting_server(struct sock *listener,
1128                                       struct sock *pending,
1129                                       struct vmci_transport_packet *pkt)
1130 {
1131         struct vsock_sock *vpending;
1132         struct vmci_handle handle;
1133         struct vmci_qp *qpair;
1134         bool is_local;
1135         u32 flags;
1136         u32 detach_sub_id;
1137         int err;
1138         int skerr;
1139
1140         vpending = vsock_sk(pending);
1141         detach_sub_id = VMCI_INVALID_ID;
1142
1143         switch (pkt->type) {
1144         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1145                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1146                         vmci_transport_send_reset(pending, pkt);
1147                         skerr = EPROTO;
1148                         err = -EINVAL;
1149                         goto destroy;
1150                 }
1151                 break;
1152         default:
1153                 /* Close and cleanup the connection. */
1154                 vmci_transport_send_reset(pending, pkt);
1155                 skerr = EPROTO;
1156                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1157                 goto destroy;
1158         }
1159
1160         /* In order to complete the connection we need to attach to the offered
1161          * queue pair and send an attach notification.  We also subscribe to the
1162          * detach event so we know when our peer goes away, and we do that
1163          * before attaching so we don't miss an event.  If all this succeeds,
1164          * we update our state and wakeup anything waiting in accept() for a
1165          * connection.
1166          */
1167
1168         /* We don't care about attach since we ensure the other side has
1169          * attached by specifying the ATTACH_ONLY flag below.
1170          */
1171         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1172                                    vmci_transport_peer_detach_cb,
1173                                    vmci_trans(vpending), &detach_sub_id);
1174         if (err < VMCI_SUCCESS) {
1175                 vmci_transport_send_reset(pending, pkt);
1176                 err = vmci_transport_error_to_vsock_error(err);
1177                 skerr = -err;
1178                 goto destroy;
1179         }
1180
1181         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1182
1183         /* Now attach to the queue pair the client created. */
1184         handle = pkt->u.handle;
1185
1186         /* vpending->local_addr always has a context id so we do not need to
1187          * worry about VMADDR_CID_ANY in this case.
1188          */
1189         is_local =
1190             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1191         flags = VMCI_QPFLAG_ATTACH_ONLY;
1192         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1193
1194         err = vmci_transport_queue_pair_alloc(
1195                                         &qpair,
1196                                         &handle,
1197                                         vmci_trans(vpending)->produce_size,
1198                                         vmci_trans(vpending)->consume_size,
1199                                         pkt->dg.src.context,
1200                                         flags,
1201                                         vmci_transport_is_trusted(
1202                                                 vpending,
1203                                                 vpending->remote_addr.svm_cid));
1204         if (err < 0) {
1205                 vmci_transport_send_reset(pending, pkt);
1206                 skerr = -err;
1207                 goto destroy;
1208         }
1209
1210         vmci_trans(vpending)->qp_handle = handle;
1211         vmci_trans(vpending)->qpair = qpair;
1212
1213         /* When we send the attach message, we must be ready to handle incoming
1214          * control messages on the newly connected socket. So we move the
1215          * pending socket to the connected state before sending the attach
1216          * message. Otherwise, an incoming packet triggered by the attach being
1217          * received by the peer may be processed concurrently with what happens
1218          * below after sending the attach message, and that incoming packet
1219          * will find the listening socket instead of the (currently) pending
1220          * socket. Note that enqueueing the socket increments the reference
1221          * count, so even if a reset comes before the connection is accepted,
1222          * the socket will be valid until it is removed from the queue.
1223          *
1224          * If we fail sending the attach below, we remove the socket from the
1225          * connected list and move the socket to TCP_CLOSE before
1226          * releasing the lock, so a pending slow path processing of an incoming
1227          * packet will not see the socket in the connected state in that case.
1228          */
1229         pending->sk_state = TCP_ESTABLISHED;
1230
1231         vsock_insert_connected(vpending);
1232
1233         /* Notify our peer of our attach. */
1234         err = vmci_transport_send_attach(pending, handle);
1235         if (err < 0) {
1236                 vsock_remove_connected(vpending);
1237                 pr_err("Could not send attach\n");
1238                 vmci_transport_send_reset(pending, pkt);
1239                 err = vmci_transport_error_to_vsock_error(err);
1240                 skerr = -err;
1241                 goto destroy;
1242         }
1243
1244         /* We have a connection. Move the now connected socket from the
1245          * listener's pending list to the accept queue so callers of accept()
1246          * can find it.
1247          */
1248         vsock_remove_pending(listener, pending);
1249         vsock_enqueue_accept(listener, pending);
1250
1251         /* Callers of accept() will be waiting on the listening socket, not
1252          * the pending socket.
1253          */
1254         listener->sk_data_ready(listener);
1255
1256         return 0;
1257
1258 destroy:
1259         pending->sk_err = skerr;
1260         pending->sk_state = TCP_CLOSE;
1261         /* As long as we drop our reference, all necessary cleanup will handle
1262          * when the cleanup function drops its reference and our destruct
1263          * implementation is called.  Note that since the listen handler will
1264          * remove pending from the pending list upon our failure, the cleanup
1265          * function won't drop the additional reference, which is why we do it
1266          * here.
1267          */
1268         sock_put(pending);
1269
1270         return err;
1271 }
1272
1273 static int
1274 vmci_transport_recv_connecting_client(struct sock *sk,
1275                                       struct vmci_transport_packet *pkt)
1276 {
1277         struct vsock_sock *vsk;
1278         int err;
1279         int skerr;
1280
1281         vsk = vsock_sk(sk);
1282
1283         switch (pkt->type) {
1284         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1285                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1286                     !vmci_handle_is_equal(pkt->u.handle,
1287                                           vmci_trans(vsk)->qp_handle)) {
1288                         skerr = EPROTO;
1289                         err = -EINVAL;
1290                         goto destroy;
1291                 }
1292
1293                 /* Signify the socket is connected and wakeup the waiter in
1294                  * connect(). Also place the socket in the connected table for
1295                  * accounting (it can already be found since it's in the bound
1296                  * table).
1297                  */
1298                 sk->sk_state = TCP_ESTABLISHED;
1299                 sk->sk_socket->state = SS_CONNECTED;
1300                 vsock_insert_connected(vsk);
1301                 sk->sk_state_change(sk);
1302
1303                 break;
1304         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1305         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1306                 if (pkt->u.size == 0
1307                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1308                     || pkt->src_port != vsk->remote_addr.svm_port
1309                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1310                     || vmci_trans(vsk)->qpair
1311                     || vmci_trans(vsk)->produce_size != 0
1312                     || vmci_trans(vsk)->consume_size != 0
1313                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1314                         skerr = EPROTO;
1315                         err = -EINVAL;
1316
1317                         goto destroy;
1318                 }
1319
1320                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1321                 if (err) {
1322                         skerr = -err;
1323                         goto destroy;
1324                 }
1325
1326                 break;
1327         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1328                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1329                 if (err) {
1330                         skerr = -err;
1331                         goto destroy;
1332                 }
1333
1334                 break;
1335         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1336                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1337                  * continue processing here after they sent an INVALID packet.
1338                  * This meant that we got a RST after the INVALID. We ignore a
1339                  * RST after an INVALID. The common code doesn't send the RST
1340                  * ... so we can hang if an old version of the common code
1341                  * fails between getting a REQUEST and sending an OFFER back.
1342                  * Not much we can do about it... except hope that it doesn't
1343                  * happen.
1344                  */
1345                 if (vsk->ignore_connecting_rst) {
1346                         vsk->ignore_connecting_rst = false;
1347                 } else {
1348                         skerr = ECONNRESET;
1349                         err = 0;
1350                         goto destroy;
1351                 }
1352
1353                 break;
1354         default:
1355                 /* Close and cleanup the connection. */
1356                 skerr = EPROTO;
1357                 err = -EINVAL;
1358                 goto destroy;
1359         }
1360
1361         return 0;
1362
1363 destroy:
1364         vmci_transport_send_reset(sk, pkt);
1365
1366         sk->sk_state = TCP_CLOSE;
1367         sk->sk_err = skerr;
1368         sk_error_report(sk);
1369         return err;
1370 }
1371
1372 static int vmci_transport_recv_connecting_client_negotiate(
1373                                         struct sock *sk,
1374                                         struct vmci_transport_packet *pkt)
1375 {
1376         int err;
1377         struct vsock_sock *vsk;
1378         struct vmci_handle handle;
1379         struct vmci_qp *qpair;
1380         u32 detach_sub_id;
1381         bool is_local;
1382         u32 flags;
1383         bool old_proto = true;
1384         bool old_pkt_proto;
1385         u16 version;
1386
1387         vsk = vsock_sk(sk);
1388         handle = VMCI_INVALID_HANDLE;
1389         detach_sub_id = VMCI_INVALID_ID;
1390
1391         /* If we have gotten here then we should be past the point where old
1392          * linux vsock could have sent the bogus rst.
1393          */
1394         vsk->sent_request = false;
1395         vsk->ignore_connecting_rst = false;
1396
1397         /* Verify that we're OK with the proposed queue pair size */
1398         if (pkt->u.size < vsk->buffer_min_size ||
1399             pkt->u.size > vsk->buffer_max_size) {
1400                 err = -EINVAL;
1401                 goto destroy;
1402         }
1403
1404         /* At this point we know the CID the peer is using to talk to us. */
1405
1406         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1407                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1408
1409         /* Setup the notify ops to be the highest supported version that both
1410          * the server and the client support.
1411          */
1412
1413         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1414                 old_proto = old_pkt_proto;
1415         } else {
1416                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1417                         old_proto = true;
1418                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1419                         old_proto = false;
1420
1421         }
1422
1423         if (old_proto)
1424                 version = VSOCK_PROTO_INVALID;
1425         else
1426                 version = pkt->proto;
1427
1428         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1429                 err = -EINVAL;
1430                 goto destroy;
1431         }
1432
1433         /* Subscribe to detach events first.
1434          *
1435          * XXX We attach once for each queue pair created for now so it is easy
1436          * to find the socket (it's provided), but later we should only
1437          * subscribe once and add a way to lookup sockets by queue pair handle.
1438          */
1439         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1440                                    vmci_transport_peer_detach_cb,
1441                                    vmci_trans(vsk), &detach_sub_id);
1442         if (err < VMCI_SUCCESS) {
1443                 err = vmci_transport_error_to_vsock_error(err);
1444                 goto destroy;
1445         }
1446
1447         /* Make VMCI select the handle for us. */
1448         handle = VMCI_INVALID_HANDLE;
1449         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1450         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1451
1452         err = vmci_transport_queue_pair_alloc(&qpair,
1453                                               &handle,
1454                                               pkt->u.size,
1455                                               pkt->u.size,
1456                                               vsk->remote_addr.svm_cid,
1457                                               flags,
1458                                               vmci_transport_is_trusted(
1459                                                   vsk,
1460                                                   vsk->
1461                                                   remote_addr.svm_cid));
1462         if (err < 0)
1463                 goto destroy;
1464
1465         err = vmci_transport_send_qp_offer(sk, handle);
1466         if (err < 0) {
1467                 err = vmci_transport_error_to_vsock_error(err);
1468                 goto destroy;
1469         }
1470
1471         vmci_trans(vsk)->qp_handle = handle;
1472         vmci_trans(vsk)->qpair = qpair;
1473
1474         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1475                 pkt->u.size;
1476
1477         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1478
1479         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1480
1481         return 0;
1482
1483 destroy:
1484         if (detach_sub_id != VMCI_INVALID_ID)
1485                 vmci_event_unsubscribe(detach_sub_id);
1486
1487         if (!vmci_handle_is_invalid(handle))
1488                 vmci_qpair_detach(&qpair);
1489
1490         return err;
1491 }
1492
1493 static int
1494 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1495                                               struct vmci_transport_packet *pkt)
1496 {
1497         int err = 0;
1498         struct vsock_sock *vsk = vsock_sk(sk);
1499
1500         if (vsk->sent_request) {
1501                 vsk->sent_request = false;
1502                 vsk->ignore_connecting_rst = true;
1503
1504                 err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1505                 if (err < 0)
1506                         err = vmci_transport_error_to_vsock_error(err);
1507                 else
1508                         err = 0;
1509
1510         }
1511
1512         return err;
1513 }
1514
1515 static int vmci_transport_recv_connected(struct sock *sk,
1516                                          struct vmci_transport_packet *pkt)
1517 {
1518         struct vsock_sock *vsk;
1519         bool pkt_processed = false;
1520
1521         /* In cases where we are closing the connection, it's sufficient to
1522          * mark the state change (and maybe error) and wake up any waiting
1523          * threads. Since this is a connected socket, it's owned by a user
1524          * process and will be cleaned up when the failure is passed back on
1525          * the current or next system call.  Our system call implementations
1526          * must therefore check for error and state changes on entry and when
1527          * being awoken.
1528          */
1529         switch (pkt->type) {
1530         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1531                 if (pkt->u.mode) {
1532                         vsk = vsock_sk(sk);
1533
1534                         vsk->peer_shutdown |= pkt->u.mode;
1535                         sk->sk_state_change(sk);
1536                 }
1537                 break;
1538
1539         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1540                 vsk = vsock_sk(sk);
1541                 /* It is possible that we sent our peer a message (e.g a
1542                  * WAITING_READ) right before we got notified that the peer had
1543                  * detached. If that happens then we can get a RST pkt back
1544                  * from our peer even though there is data available for us to
1545                  * read. In that case, don't shutdown the socket completely but
1546                  * instead allow the local client to finish reading data off
1547                  * the queuepair. Always treat a RST pkt in connected mode like
1548                  * a clean shutdown.
1549                  */
1550                 sock_set_flag(sk, SOCK_DONE);
1551                 vsk->peer_shutdown = SHUTDOWN_MASK;
1552                 if (vsock_stream_has_data(vsk) <= 0)
1553                         sk->sk_state = TCP_CLOSING;
1554
1555                 sk->sk_state_change(sk);
1556                 break;
1557
1558         default:
1559                 vsk = vsock_sk(sk);
1560                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1561                                 sk, pkt, false, NULL, NULL,
1562                                 &pkt_processed);
1563                 if (!pkt_processed)
1564                         return -EINVAL;
1565
1566                 break;
1567         }
1568
1569         return 0;
1570 }
1571
1572 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1573                                       struct vsock_sock *psk)
1574 {
1575         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1576         if (!vsk->trans)
1577                 return -ENOMEM;
1578
1579         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1580         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1581         vmci_trans(vsk)->qpair = NULL;
1582         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1583         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1584         vmci_trans(vsk)->notify_ops = NULL;
1585         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1586         vmci_trans(vsk)->sk = &vsk->sk;
1587         spin_lock_init(&vmci_trans(vsk)->lock);
1588
1589         return 0;
1590 }
1591
1592 static void vmci_transport_free_resources(struct list_head *transport_list)
1593 {
1594         while (!list_empty(transport_list)) {
1595                 struct vmci_transport *transport =
1596                     list_first_entry(transport_list, struct vmci_transport,
1597                                      elem);
1598                 list_del(&transport->elem);
1599
1600                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1601                         vmci_event_unsubscribe(transport->detach_sub_id);
1602                         transport->detach_sub_id = VMCI_INVALID_ID;
1603                 }
1604
1605                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1606                         vmci_qpair_detach(&transport->qpair);
1607                         transport->qp_handle = VMCI_INVALID_HANDLE;
1608                         transport->produce_size = 0;
1609                         transport->consume_size = 0;
1610                 }
1611
1612                 kfree(transport);
1613         }
1614 }
1615
1616 static void vmci_transport_cleanup(struct work_struct *work)
1617 {
1618         LIST_HEAD(pending);
1619
1620         spin_lock_bh(&vmci_transport_cleanup_lock);
1621         list_replace_init(&vmci_transport_cleanup_list, &pending);
1622         spin_unlock_bh(&vmci_transport_cleanup_lock);
1623         vmci_transport_free_resources(&pending);
1624 }
1625
1626 static void vmci_transport_destruct(struct vsock_sock *vsk)
1627 {
1628         /* transport can be NULL if we hit a failure at init() time */
1629         if (!vmci_trans(vsk))
1630                 return;
1631
1632         /* Ensure that the detach callback doesn't use the sk/vsk
1633          * we are about to destruct.
1634          */
1635         spin_lock_bh(&vmci_trans(vsk)->lock);
1636         vmci_trans(vsk)->sk = NULL;
1637         spin_unlock_bh(&vmci_trans(vsk)->lock);
1638
1639         if (vmci_trans(vsk)->notify_ops)
1640                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1641
1642         spin_lock_bh(&vmci_transport_cleanup_lock);
1643         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1644         spin_unlock_bh(&vmci_transport_cleanup_lock);
1645         schedule_work(&vmci_transport_cleanup_work);
1646
1647         vsk->trans = NULL;
1648 }
1649
1650 static void vmci_transport_release(struct vsock_sock *vsk)
1651 {
1652         vsock_remove_sock(vsk);
1653
1654         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1655                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1656                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1657         }
1658 }
1659
1660 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1661                                      struct sockaddr_vm *addr)
1662 {
1663         u32 port;
1664         u32 flags;
1665         int err;
1666
1667         /* VMCI will select a resource ID for us if we provide
1668          * VMCI_INVALID_ID.
1669          */
1670         port = addr->svm_port == VMADDR_PORT_ANY ?
1671                         VMCI_INVALID_ID : addr->svm_port;
1672
1673         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1674                 return -EACCES;
1675
1676         flags = addr->svm_cid == VMADDR_CID_ANY ?
1677                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1678
1679         err = vmci_transport_datagram_create_hnd(port, flags,
1680                                                  vmci_transport_recv_dgram_cb,
1681                                                  &vsk->sk,
1682                                                  &vmci_trans(vsk)->dg_handle);
1683         if (err < VMCI_SUCCESS)
1684                 return vmci_transport_error_to_vsock_error(err);
1685         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1686                         vmci_trans(vsk)->dg_handle.resource);
1687
1688         return 0;
1689 }
1690
1691 static int vmci_transport_dgram_enqueue(
1692         struct vsock_sock *vsk,
1693         struct sockaddr_vm *remote_addr,
1694         struct msghdr *msg,
1695         size_t len)
1696 {
1697         int err;
1698         struct vmci_datagram *dg;
1699
1700         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1701                 return -EMSGSIZE;
1702
1703         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1704                 return -EPERM;
1705
1706         /* Allocate a buffer for the user's message and our packet header. */
1707         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1708         if (!dg)
1709                 return -ENOMEM;
1710
1711         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1712
1713         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1714                                    remote_addr->svm_port);
1715         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1716                                    vsk->local_addr.svm_port);
1717         dg->payload_size = len;
1718
1719         err = vmci_datagram_send(dg);
1720         kfree(dg);
1721         if (err < 0)
1722                 return vmci_transport_error_to_vsock_error(err);
1723
1724         return err - sizeof(*dg);
1725 }
1726
1727 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1728                                         struct msghdr *msg, size_t len,
1729                                         int flags)
1730 {
1731         int err;
1732         int noblock;
1733         struct vmci_datagram *dg;
1734         size_t payload_len;
1735         struct sk_buff *skb;
1736
1737         noblock = flags & MSG_DONTWAIT;
1738
1739         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1740                 return -EOPNOTSUPP;
1741
1742         /* Retrieve the head sk_buff from the socket's receive queue. */
1743         err = 0;
1744         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1745         if (!skb)
1746                 return err;
1747
1748         dg = (struct vmci_datagram *)skb->data;
1749         if (!dg)
1750                 /* err is 0, meaning we read zero bytes. */
1751                 goto out;
1752
1753         payload_len = dg->payload_size;
1754         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1755         if (payload_len != skb->len - sizeof(*dg)) {
1756                 err = -EINVAL;
1757                 goto out;
1758         }
1759
1760         if (payload_len > len) {
1761                 payload_len = len;
1762                 msg->msg_flags |= MSG_TRUNC;
1763         }
1764
1765         /* Place the datagram payload in the user's iovec. */
1766         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1767         if (err)
1768                 goto out;
1769
1770         if (msg->msg_name) {
1771                 /* Provide the address of the sender. */
1772                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1773                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1774                 msg->msg_namelen = sizeof(*vm_addr);
1775         }
1776         err = payload_len;
1777
1778 out:
1779         skb_free_datagram(&vsk->sk, skb);
1780         return err;
1781 }
1782
1783 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1784 {
1785         if (cid == VMADDR_CID_HYPERVISOR) {
1786                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1787                  * state and are allowed.
1788                  */
1789                 return port == VMCI_UNITY_PBRPC_REGISTER;
1790         }
1791
1792         return true;
1793 }
1794
1795 static int vmci_transport_connect(struct vsock_sock *vsk)
1796 {
1797         int err;
1798         bool old_pkt_proto = false;
1799         struct sock *sk = &vsk->sk;
1800
1801         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1802                 old_pkt_proto) {
1803                 err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1804                 if (err < 0) {
1805                         sk->sk_state = TCP_CLOSE;
1806                         return err;
1807                 }
1808         } else {
1809                 int supported_proto_versions =
1810                         vmci_transport_new_proto_supported_versions();
1811                 err = vmci_transport_send_conn_request2(sk, vsk->buffer_size,
1812                                 supported_proto_versions);
1813                 if (err < 0) {
1814                         sk->sk_state = TCP_CLOSE;
1815                         return err;
1816                 }
1817
1818                 vsk->sent_request = true;
1819         }
1820
1821         return err;
1822 }
1823
1824 static ssize_t vmci_transport_stream_dequeue(
1825         struct vsock_sock *vsk,
1826         struct msghdr *msg,
1827         size_t len,
1828         int flags)
1829 {
1830         if (flags & MSG_PEEK)
1831                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1832         else
1833                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1834 }
1835
1836 static ssize_t vmci_transport_stream_enqueue(
1837         struct vsock_sock *vsk,
1838         struct msghdr *msg,
1839         size_t len)
1840 {
1841         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1842 }
1843
1844 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1845 {
1846         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1847 }
1848
1849 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1850 {
1851         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1852 }
1853
1854 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1855 {
1856         return vmci_trans(vsk)->consume_size;
1857 }
1858
1859 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1860 {
1861         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1862 }
1863
1864 static int vmci_transport_notify_poll_in(
1865         struct vsock_sock *vsk,
1866         size_t target,
1867         bool *data_ready_now)
1868 {
1869         return vmci_trans(vsk)->notify_ops->poll_in(
1870                         &vsk->sk, target, data_ready_now);
1871 }
1872
1873 static int vmci_transport_notify_poll_out(
1874         struct vsock_sock *vsk,
1875         size_t target,
1876         bool *space_available_now)
1877 {
1878         return vmci_trans(vsk)->notify_ops->poll_out(
1879                         &vsk->sk, target, space_available_now);
1880 }
1881
1882 static int vmci_transport_notify_recv_init(
1883         struct vsock_sock *vsk,
1884         size_t target,
1885         struct vsock_transport_recv_notify_data *data)
1886 {
1887         return vmci_trans(vsk)->notify_ops->recv_init(
1888                         &vsk->sk, target,
1889                         (struct vmci_transport_recv_notify_data *)data);
1890 }
1891
1892 static int vmci_transport_notify_recv_pre_block(
1893         struct vsock_sock *vsk,
1894         size_t target,
1895         struct vsock_transport_recv_notify_data *data)
1896 {
1897         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1898                         &vsk->sk, target,
1899                         (struct vmci_transport_recv_notify_data *)data);
1900 }
1901
1902 static int vmci_transport_notify_recv_pre_dequeue(
1903         struct vsock_sock *vsk,
1904         size_t target,
1905         struct vsock_transport_recv_notify_data *data)
1906 {
1907         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1908                         &vsk->sk, target,
1909                         (struct vmci_transport_recv_notify_data *)data);
1910 }
1911
1912 static int vmci_transport_notify_recv_post_dequeue(
1913         struct vsock_sock *vsk,
1914         size_t target,
1915         ssize_t copied,
1916         bool data_read,
1917         struct vsock_transport_recv_notify_data *data)
1918 {
1919         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1920                         &vsk->sk, target, copied, data_read,
1921                         (struct vmci_transport_recv_notify_data *)data);
1922 }
1923
1924 static int vmci_transport_notify_send_init(
1925         struct vsock_sock *vsk,
1926         struct vsock_transport_send_notify_data *data)
1927 {
1928         return vmci_trans(vsk)->notify_ops->send_init(
1929                         &vsk->sk,
1930                         (struct vmci_transport_send_notify_data *)data);
1931 }
1932
1933 static int vmci_transport_notify_send_pre_block(
1934         struct vsock_sock *vsk,
1935         struct vsock_transport_send_notify_data *data)
1936 {
1937         return vmci_trans(vsk)->notify_ops->send_pre_block(
1938                         &vsk->sk,
1939                         (struct vmci_transport_send_notify_data *)data);
1940 }
1941
1942 static int vmci_transport_notify_send_pre_enqueue(
1943         struct vsock_sock *vsk,
1944         struct vsock_transport_send_notify_data *data)
1945 {
1946         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1947                         &vsk->sk,
1948                         (struct vmci_transport_send_notify_data *)data);
1949 }
1950
1951 static int vmci_transport_notify_send_post_enqueue(
1952         struct vsock_sock *vsk,
1953         ssize_t written,
1954         struct vsock_transport_send_notify_data *data)
1955 {
1956         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1957                         &vsk->sk, written,
1958                         (struct vmci_transport_send_notify_data *)data);
1959 }
1960
1961 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1962 {
1963         if (PROTOCOL_OVERRIDE != -1) {
1964                 if (PROTOCOL_OVERRIDE == 0)
1965                         *old_pkt_proto = true;
1966                 else
1967                         *old_pkt_proto = false;
1968
1969                 pr_info("Proto override in use\n");
1970                 return true;
1971         }
1972
1973         return false;
1974 }
1975
1976 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
1977                                                   u16 *proto,
1978                                                   bool old_pkt_proto)
1979 {
1980         struct vsock_sock *vsk = vsock_sk(sk);
1981
1982         if (old_pkt_proto) {
1983                 if (*proto != VSOCK_PROTO_INVALID) {
1984                         pr_err("Can't set both an old and new protocol\n");
1985                         return false;
1986                 }
1987                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
1988                 goto exit;
1989         }
1990
1991         switch (*proto) {
1992         case VSOCK_PROTO_PKT_ON_NOTIFY:
1993                 vmci_trans(vsk)->notify_ops =
1994                         &vmci_transport_notify_pkt_q_state_ops;
1995                 break;
1996         default:
1997                 pr_err("Unknown notify protocol version\n");
1998                 return false;
1999         }
2000
2001 exit:
2002         vmci_trans(vsk)->notify_ops->socket_init(sk);
2003         return true;
2004 }
2005
2006 static u16 vmci_transport_new_proto_supported_versions(void)
2007 {
2008         if (PROTOCOL_OVERRIDE != -1)
2009                 return PROTOCOL_OVERRIDE;
2010
2011         return VSOCK_PROTO_ALL_SUPPORTED;
2012 }
2013
2014 static u32 vmci_transport_get_local_cid(void)
2015 {
2016         return vmci_get_context_id();
2017 }
2018
2019 static struct vsock_transport vmci_transport = {
2020         .module = THIS_MODULE,
2021         .init = vmci_transport_socket_init,
2022         .destruct = vmci_transport_destruct,
2023         .release = vmci_transport_release,
2024         .connect = vmci_transport_connect,
2025         .dgram_bind = vmci_transport_dgram_bind,
2026         .dgram_dequeue = vmci_transport_dgram_dequeue,
2027         .dgram_enqueue = vmci_transport_dgram_enqueue,
2028         .dgram_allow = vmci_transport_dgram_allow,
2029         .stream_dequeue = vmci_transport_stream_dequeue,
2030         .stream_enqueue = vmci_transport_stream_enqueue,
2031         .stream_has_data = vmci_transport_stream_has_data,
2032         .stream_has_space = vmci_transport_stream_has_space,
2033         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2034         .stream_is_active = vmci_transport_stream_is_active,
2035         .stream_allow = vmci_transport_stream_allow,
2036         .notify_poll_in = vmci_transport_notify_poll_in,
2037         .notify_poll_out = vmci_transport_notify_poll_out,
2038         .notify_recv_init = vmci_transport_notify_recv_init,
2039         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2040         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2041         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2042         .notify_send_init = vmci_transport_notify_send_init,
2043         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2044         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2045         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2046         .shutdown = vmci_transport_shutdown,
2047         .get_local_cid = vmci_transport_get_local_cid,
2048 };
2049
2050 static bool vmci_check_transport(struct vsock_sock *vsk)
2051 {
2052         return vsk->transport == &vmci_transport;
2053 }
2054
2055 static void vmci_vsock_transport_cb(bool is_host)
2056 {
2057         int features;
2058
2059         if (is_host)
2060                 features = VSOCK_TRANSPORT_F_H2G;
2061         else
2062                 features = VSOCK_TRANSPORT_F_G2H;
2063
2064         vsock_core_register(&vmci_transport, features);
2065 }
2066
2067 static int __init vmci_transport_init(void)
2068 {
2069         int err;
2070
2071         /* Create the datagram handle that we will use to send and receive all
2072          * VSocket control messages for this context.
2073          */
2074         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2075                                                  VMCI_FLAG_ANYCID_DG_HND,
2076                                                  vmci_transport_recv_stream_cb,
2077                                                  NULL,
2078                                                  &vmci_transport_stream_handle);
2079         if (err < VMCI_SUCCESS) {
2080                 pr_err("Unable to create datagram handle. (%d)\n", err);
2081                 return vmci_transport_error_to_vsock_error(err);
2082         }
2083         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2084                                    vmci_transport_qp_resumed_cb,
2085                                    NULL, &vmci_transport_qp_resumed_sub_id);
2086         if (err < VMCI_SUCCESS) {
2087                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2088                 err = vmci_transport_error_to_vsock_error(err);
2089                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2090                 goto err_destroy_stream_handle;
2091         }
2092
2093         /* Register only with dgram feature, other features (H2G, G2H) will be
2094          * registered when the first host or guest becomes active.
2095          */
2096         err = vsock_core_register(&vmci_transport, VSOCK_TRANSPORT_F_DGRAM);
2097         if (err < 0)
2098                 goto err_unsubscribe;
2099
2100         err = vmci_register_vsock_callback(vmci_vsock_transport_cb);
2101         if (err < 0)
2102                 goto err_unregister;
2103
2104         return 0;
2105
2106 err_unregister:
2107         vsock_core_unregister(&vmci_transport);
2108 err_unsubscribe:
2109         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2110 err_destroy_stream_handle:
2111         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2112         return err;
2113 }
2114 module_init(vmci_transport_init);
2115
2116 static void __exit vmci_transport_exit(void)
2117 {
2118         cancel_work_sync(&vmci_transport_cleanup_work);
2119         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2120
2121         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2122                 if (vmci_datagram_destroy_handle(
2123                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2124                         pr_err("Couldn't destroy datagram handle\n");
2125                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2126         }
2127
2128         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2129                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2130                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2131         }
2132
2133         vmci_register_vsock_callback(NULL);
2134         vsock_core_unregister(&vmci_transport);
2135 }
2136 module_exit(vmci_transport_exit);
2137
2138 MODULE_AUTHOR("VMware, Inc.");
2139 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2140 MODULE_VERSION("1.0.5.0-k");
2141 MODULE_LICENSE("GPL v2");
2142 MODULE_ALIAS("vmware_vsock");
2143 MODULE_ALIAS_NETPROTO(PF_VSOCK);