net/vmw_vsock/vmci_transport.c

   1 /*
   2  * VMware vSockets Driver
   3  *
   4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation version 2 and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  */
  15
  16 #include <linux/types.h>
  17 #include <linux/bitops.h>
  18 #include <linux/cred.h>
  19 #include <linux/init.h>
  20 #include <linux/io.h>
  21 #include <linux/kernel.h>
  22 #include <linux/kmod.h>
  23 #include <linux/list.h>
  24 #include <linux/module.h>
  25 #include <linux/mutex.h>
  26 #include <linux/net.h>
  27 #include <linux/poll.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/smp.h>
  30 #include <linux/socket.h>
  31 #include <linux/stddef.h>
  32 #include <linux/unistd.h>
  33 #include <linux/wait.h>
  34 #include <linux/workqueue.h>
  35 #include <net/sock.h>
  36 #include <net/af_vsock.h>
  37
  38 #include "vmci_transport_notify.h"
  39
  40 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
  41 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
  42 static void vmci_transport_peer_detach_cb(u32 sub_id,
  43                                           const struct vmci_event_data *ed,
  44                                           void *client_data);
  45 static void vmci_transport_recv_pkt_work(struct work_struct *work);
  46 static void vmci_transport_cleanup(struct work_struct *work);
  47 static int vmci_transport_recv_listen(struct sock *sk,
  48                                       struct vmci_transport_packet *pkt);
  49 static int vmci_transport_recv_connecting_server(
  50                                         struct sock *sk,
  51                                         struct sock *pending,
  52                                         struct vmci_transport_packet *pkt);
  53 static int vmci_transport_recv_connecting_client(
  54                                         struct sock *sk,
  55                                         struct vmci_transport_packet *pkt);
  56 static int vmci_transport_recv_connecting_client_negotiate(
  57                                         struct sock *sk,
  58                                         struct vmci_transport_packet *pkt);
  59 static int vmci_transport_recv_connecting_client_invalid(
  60                                         struct sock *sk,
  61                                         struct vmci_transport_packet *pkt);
  62 static int vmci_transport_recv_connected(struct sock *sk,
  63                                          struct vmci_transport_packet *pkt);
  64 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
  65 static u16 vmci_transport_new_proto_supported_versions(void);
  66 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
  67                                                   bool old_pkt_proto);
  68
  69 struct vmci_transport_recv_pkt_info {
  70         struct work_struct work;
  71         struct sock *sk;
  72         struct vmci_transport_packet pkt;
  73 };
  74
  75 static LIST_HEAD(vmci_transport_cleanup_list);
  76 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
  77 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
  78
  79 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  80                                                            VMCI_INVALID_ID };
  81 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  82
  83 static int PROTOCOL_OVERRIDE = -1;
  84
  85 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN   128
  86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE       262144
  87 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX   262144
  88
  89 /* The default peer timeout indicates how long we will wait for a peer response
  90  * to a control message.
  91  */
  92 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
  93
  94 /* Helper function to convert from a VMCI error code to a VSock error code. */
  95
  96 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  97 {
  98         switch (vmci_error) {
  99         case VMCI_ERROR_NO_MEM:
 100                 return -ENOMEM;
 101         case VMCI_ERROR_DUPLICATE_ENTRY:
 102         case VMCI_ERROR_ALREADY_EXISTS:
 103                 return -EADDRINUSE;
 104         case VMCI_ERROR_NO_ACCESS:
 105                 return -EPERM;
 106         case VMCI_ERROR_NO_RESOURCES:
 107                 return -ENOBUFS;
 108         case VMCI_ERROR_INVALID_RESOURCE:
 109                 return -EHOSTUNREACH;
 110         case VMCI_ERROR_INVALID_ARGS:
 111         default:
 112                 break;
 113         }
 114         return -EINVAL;
 115 }
 116
 117 static u32 vmci_transport_peer_rid(u32 peer_cid)
 118 {
 119         if (VMADDR_CID_HYPERVISOR == peer_cid)
 120                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 121
 122         return VMCI_TRANSPORT_PACKET_RID;
 123 }
 124
 125 static inline void
 126 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 127                            struct sockaddr_vm *src,
 128                            struct sockaddr_vm *dst,
 129                            u8 type,
 130                            u64 size,
 131                            u64 mode,
 132                            struct vmci_transport_waiting_info *wait,
 133                            u16 proto,
 134                            struct vmci_handle handle)
 135 {
 136         /* We register the stream control handler as an any cid handle so we
 137          * must always send from a source address of VMADDR_CID_ANY
 138          */
 139         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 140                                        VMCI_TRANSPORT_PACKET_RID);
 141         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 142                                        vmci_transport_peer_rid(dst->svm_cid));
 143         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 144         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 145         pkt->type = type;
 146         pkt->src_port = src->svm_port;
 147         pkt->dst_port = dst->svm_port;
 148         memset(&pkt->proto, 0, sizeof(pkt->proto));
 149         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 150
 151         switch (pkt->type) {
 152         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 153                 pkt->u.size = 0;
 154                 break;
 155
 156         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 157         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 158                 pkt->u.size = size;
 159                 break;
 160
 161         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 162         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 163                 pkt->u.handle = handle;
 164                 break;
 165
 166         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 167         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 168         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 169                 pkt->u.size = 0;
 170                 break;
 171
 172         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 173                 pkt->u.mode = mode;
 174                 break;
 175
 176         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 177         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 178                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 179                 break;
 180
 181         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 182         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 183                 pkt->u.size = size;
 184                 pkt->proto = proto;
 185                 break;
 186         }
 187 }
 188
 189 static inline void
 190 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 191                                     struct sockaddr_vm *local,
 192                                     struct sockaddr_vm *remote)
 193 {
 194         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 195         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 196 }
 197
 198 static int
 199 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 200                                   struct sockaddr_vm *src,
 201                                   struct sockaddr_vm *dst,
 202                                   enum vmci_transport_packet_type type,
 203                                   u64 size,
 204                                   u64 mode,
 205                                   struct vmci_transport_waiting_info *wait,
 206                                   u16 proto,
 207                                   struct vmci_handle handle,
 208                                   bool convert_error)
 209 {
 210         int err;
 211
 212         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 213                                    proto, handle);
 214         err = vmci_datagram_send(&pkt->dg);
 215         if (convert_error && (err < 0))
 216                 return vmci_transport_error_to_vsock_error(err);
 217
 218         return err;
 219 }
 220
 221 static int
 222 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 223                                       enum vmci_transport_packet_type type,
 224                                       u64 size,
 225                                       u64 mode,
 226                                       struct vmci_transport_waiting_info *wait,
 227                                       struct vmci_handle handle)
 228 {
 229         struct vmci_transport_packet reply;
 230         struct sockaddr_vm src, dst;
 231
 232         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 233                 return 0;
 234         } else {
 235                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 236                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 237                                                          type,
 238                                                          size, mode, wait,
 239                                                          VSOCK_PROTO_INVALID,
 240                                                          handle, true);
 241         }
 242 }
 243
 244 static int
 245 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 246                                    struct sockaddr_vm *dst,
 247                                    enum vmci_transport_packet_type type,
 248                                    u64 size,
 249                                    u64 mode,
 250                                    struct vmci_transport_waiting_info *wait,
 251                                    struct vmci_handle handle)
 252 {
 253         /* Note that it is safe to use a single packet across all CPUs since
 254          * two tasklets of the same type are guaranteed to not ever run
 255          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 256          * we can use per-cpu packets.
 257          */
 258         static struct vmci_transport_packet pkt;
 259
 260         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 261                                                  size, mode, wait,
 262                                                  VSOCK_PROTO_INVALID, handle,
 263                                                  false);
 264 }
 265
 266 static int
 267 vmci_transport_send_control_pkt(struct sock *sk,
 268                                 enum vmci_transport_packet_type type,
 269                                 u64 size,
 270                                 u64 mode,
 271                                 struct vmci_transport_waiting_info *wait,
 272                                 u16 proto,
 273                                 struct vmci_handle handle)
 274 {
 275         struct vmci_transport_packet *pkt;
 276         struct vsock_sock *vsk;
 277         int err;
 278
 279         vsk = vsock_sk(sk);
 280
 281         if (!vsock_addr_bound(&vsk->local_addr))
 282                 return -EINVAL;
 283
 284         if (!vsock_addr_bound(&vsk->remote_addr))
 285                 return -EINVAL;
 286
 287         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 288         if (!pkt)
 289                 return -ENOMEM;
 290
 291         err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
 292                                                 &vsk->remote_addr, type, size,
 293                                                 mode, wait, proto, handle,
 294                                                 true);
 295         kfree(pkt);
 296
 297         return err;
 298 }
 299
 300 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 301                                         struct sockaddr_vm *src,
 302                                         struct vmci_transport_packet *pkt)
 303 {
 304         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 305                 return 0;
 306         return vmci_transport_send_control_pkt_bh(
 307                                         dst, src,
 308                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 309                                         0, NULL, VMCI_INVALID_HANDLE);
 310 }
 311
 312 static int vmci_transport_send_reset(struct sock *sk,
 313                                      struct vmci_transport_packet *pkt)
 314 {
 315         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 316                 return 0;
 317         return vmci_transport_send_control_pkt(sk,
 318                                         VMCI_TRANSPORT_PACKET_TYPE_RST,
 319                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 320                                         VMCI_INVALID_HANDLE);
 321 }
 322
 323 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 324 {
 325         return vmci_transport_send_control_pkt(
 326                                         sk,
 327                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 328                                         size, 0, NULL,
 329                                         VSOCK_PROTO_INVALID,
 330                                         VMCI_INVALID_HANDLE);
 331 }
 332
 333 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 334                                           u16 version)
 335 {
 336         return vmci_transport_send_control_pkt(
 337                                         sk,
 338                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 339                                         size, 0, NULL, version,
 340                                         VMCI_INVALID_HANDLE);
 341 }
 342
 343 static int vmci_transport_send_qp_offer(struct sock *sk,
 344                                         struct vmci_handle handle)
 345 {
 346         return vmci_transport_send_control_pkt(
 347                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 348                                         0, NULL,
 349                                         VSOCK_PROTO_INVALID, handle);
 350 }
 351
 352 static int vmci_transport_send_attach(struct sock *sk,
 353                                       struct vmci_handle handle)
 354 {
 355         return vmci_transport_send_control_pkt(
 356                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 357                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 358                                         handle);
 359 }
 360
 361 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 362 {
 363         return vmci_transport_reply_control_pkt_fast(
 364                                                 pkt,
 365                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 366                                                 0, 0, NULL,
 367                                                 VMCI_INVALID_HANDLE);
 368 }
 369
 370 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 371                                           struct sockaddr_vm *src)
 372 {
 373         return vmci_transport_send_control_pkt_bh(
 374                                         dst, src,
 375                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 376                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 377 }
 378
 379 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 380                                  struct sockaddr_vm *src)
 381 {
 382         return vmci_transport_send_control_pkt_bh(
 383                                         dst, src,
 384                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 385                                         0, NULL, VMCI_INVALID_HANDLE);
 386 }
 387
 388 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 389                                 struct sockaddr_vm *src)
 390 {
 391         return vmci_transport_send_control_pkt_bh(
 392                                         dst, src,
 393                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 394                                         0, NULL, VMCI_INVALID_HANDLE);
 395 }
 396
 397 int vmci_transport_send_wrote(struct sock *sk)
 398 {
 399         return vmci_transport_send_control_pkt(
 400                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 401                                         0, NULL, VSOCK_PROTO_INVALID,
 402                                         VMCI_INVALID_HANDLE);
 403 }
 404
 405 int vmci_transport_send_read(struct sock *sk)
 406 {
 407         return vmci_transport_send_control_pkt(
 408                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 409                                         0, NULL, VSOCK_PROTO_INVALID,
 410                                         VMCI_INVALID_HANDLE);
 411 }
 412
 413 int vmci_transport_send_waiting_write(struct sock *sk,
 414                                       struct vmci_transport_waiting_info *wait)
 415 {
 416         return vmci_transport_send_control_pkt(
 417                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 418                                 0, 0, wait, VSOCK_PROTO_INVALID,
 419                                 VMCI_INVALID_HANDLE);
 420 }
 421
 422 int vmci_transport_send_waiting_read(struct sock *sk,
 423                                      struct vmci_transport_waiting_info *wait)
 424 {
 425         return vmci_transport_send_control_pkt(
 426                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 427                                 0, 0, wait, VSOCK_PROTO_INVALID,
 428                                 VMCI_INVALID_HANDLE);
 429 }
 430
 431 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 432 {
 433         return vmci_transport_send_control_pkt(
 434                                         &vsk->sk,
 435                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 436                                         0, mode, NULL,
 437                                         VSOCK_PROTO_INVALID,
 438                                         VMCI_INVALID_HANDLE);
 439 }
 440
 441 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 442 {
 443         return vmci_transport_send_control_pkt(sk,
 444                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 445                                         size, 0, NULL,
 446                                         VSOCK_PROTO_INVALID,
 447                                         VMCI_INVALID_HANDLE);
 448 }
 449
 450 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 451                                              u16 version)
 452 {
 453         return vmci_transport_send_control_pkt(
 454                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 455                                         size, 0, NULL, version,
 456                                         VMCI_INVALID_HANDLE);
 457 }
 458
 459 static struct sock *vmci_transport_get_pending(
 460                                         struct sock *listener,
 461                                         struct vmci_transport_packet *pkt)
 462 {
 463         struct vsock_sock *vlistener;
 464         struct vsock_sock *vpending;
 465         struct sock *pending;
 466         struct sockaddr_vm src;
 467
 468         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 469
 470         vlistener = vsock_sk(listener);
 471
 472         list_for_each_entry(vpending, &vlistener->pending_links,
 473                             pending_links) {
 474                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 475                     pkt->dst_port == vpending->local_addr.svm_port) {
 476                         pending = sk_vsock(vpending);
 477                         sock_hold(pending);
 478                         goto found;
 479                 }
 480         }
 481
 482         pending = NULL;
 483 found:
 484         return pending;
 485
 486 }
 487
 488 static void vmci_transport_release_pending(struct sock *pending)
 489 {
 490         sock_put(pending);
 491 }
 492
 493 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 494  * trusted sockets 2) sockets from applications running as the same user as the
 495  * VM (this is only true for the host side and only when using hosted products)
 496  */
 497
 498 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 499 {
 500         return vsock->trusted ||
 501                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 502 }
 503
 504 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 505  * only if it is trusted as described in vmci_transport_is_trusted.
 506  */
 507
 508 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 509 {
 510         if (VMADDR_CID_HYPERVISOR == peer_cid)
 511                 return true;
 512
 513         if (vsock->cached_peer != peer_cid) {
 514                 vsock->cached_peer = peer_cid;
 515                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 516                     (vmci_context_get_priv_flags(peer_cid) &
 517                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 518                         vsock->cached_peer_allow_dgram = false;
 519                 } else {
 520                         vsock->cached_peer_allow_dgram = true;
 521                 }
 522         }
 523
 524         return vsock->cached_peer_allow_dgram;
 525 }
 526
 527 static int
 528 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 529                                 struct vmci_handle *handle,
 530                                 u64 produce_size,
 531                                 u64 consume_size,
 532                                 u32 peer, u32 flags, bool trusted)
 533 {
 534         int err = 0;
 535
 536         if (trusted) {
 537                 /* Try to allocate our queue pair as trusted. This will only
 538                  * work if vsock is running in the host.
 539                  */
 540
 541                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 542                                        consume_size,
 543                                        peer, flags,
 544                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 545                 if (err != VMCI_ERROR_NO_ACCESS)
 546                         goto out;
 547
 548         }
 549
 550         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 551                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 552 out:
 553         if (err < 0) {
 554                 pr_err("Could not attach to queue pair with %d\n",
 555                        err);
 556                 err = vmci_transport_error_to_vsock_error(err);
 557         }
 558
 559         return err;
 560 }
 561
 562 static int
 563 vmci_transport_datagram_create_hnd(u32 resource_id,
 564                                    u32 flags,
 565                                    vmci_datagram_recv_cb recv_cb,
 566                                    void *client_data,
 567                                    struct vmci_handle *out_handle)
 568 {
 569         int err = 0;
 570
 571         /* Try to allocate our datagram handler as trusted. This will only work
 572          * if vsock is running in the host.
 573          */
 574
 575         err = vmci_datagram_create_handle_priv(resource_id, flags,
 576                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 577                                                recv_cb,
 578                                                client_data, out_handle);
 579
 580         if (err == VMCI_ERROR_NO_ACCESS)
 581                 err = vmci_datagram_create_handle(resource_id, flags,
 582                                                   recv_cb, client_data,
 583                                                   out_handle);
 584
 585         return err;
 586 }
 587
 588 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 589  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 590  * sleep it should defer that work to a work queue.
 591  */
 592
 593 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 594 {
 595         struct sock *sk;
 596         size_t size;
 597         struct sk_buff *skb;
 598         struct vsock_sock *vsk;
 599
 600         sk = (struct sock *)data;
 601
 602         /* This handler is privileged when this module is running on the host.
 603          * We will get datagrams from all endpoints (even VMs that are in a
 604          * restricted context). If we get one from a restricted context then
 605          * the destination socket must be trusted.
 606          *
 607          * NOTE: We access the socket struct without holding the lock here.
 608          * This is ok because the field we are interested is never modified
 609          * outside of the create and destruct socket functions.
 610          */
 611         vsk = vsock_sk(sk);
 612         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 613                 return VMCI_ERROR_NO_ACCESS;
 614
 615         size = VMCI_DG_SIZE(dg);
 616
 617         /* Attach the packet to the socket's receive queue as an sk_buff. */
 618         skb = alloc_skb(size, GFP_ATOMIC);
 619         if (!skb)
 620                 return VMCI_ERROR_NO_MEM;
 621
 622         /* sk_receive_skb() will do a sock_put(), so hold here. */
 623         sock_hold(sk);
 624         skb_put(skb, size);
 625         memcpy(skb->data, dg, size);
 626         sk_receive_skb(sk, skb, 0);
 627
 628         return VMCI_SUCCESS;
 629 }
 630
 631 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 632 {
 633         static const u32 non_socket_contexts[] = {
 634                 VMADDR_CID_RESERVED,
 635         };
 636         int i;
 637
 638         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 639
 640         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 641                 if (cid == non_socket_contexts[i])
 642                         return false;
 643         }
 644
 645         return true;
 646 }
 647
 648 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 649  * interrupt fires.  This is run in bottom-half context but it defers most of
 650  * its work to the packet handling work queue.
 651  */
 652
 653 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 654 {
 655         struct sock *sk;
 656         struct sockaddr_vm dst;
 657         struct sockaddr_vm src;
 658         struct vmci_transport_packet *pkt;
 659         struct vsock_sock *vsk;
 660         bool bh_process_pkt;
 661         int err;
 662
 663         sk = NULL;
 664         err = VMCI_SUCCESS;
 665         bh_process_pkt = false;
 666
 667         /* Ignore incoming packets from contexts without sockets, or resources
 668          * that aren't vsock implementations.
 669          */
 670
 671         if (!vmci_transport_stream_allow(dg->src.context, -1)
 672             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 673                 return VMCI_ERROR_NO_ACCESS;
 674
 675         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 676                 /* Drop datagrams that do not contain full VSock packets. */
 677                 return VMCI_ERROR_INVALID_ARGS;
 678
 679         pkt = (struct vmci_transport_packet *)dg;
 680
 681         /* Find the socket that should handle this packet.  First we look for a
 682          * connected socket and if there is none we look for a socket bound to
 683          * the destintation address.
 684          */
 685         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 686         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 687
 688         sk = vsock_find_connected_socket(&src, &dst);
 689         if (!sk) {
 690                 sk = vsock_find_bound_socket(&dst);
 691                 if (!sk) {
 692                         /* We could not find a socket for this specified
 693                          * address.  If this packet is a RST, we just drop it.
 694                          * If it is another packet, we send a RST.  Note that
 695                          * we do not send a RST reply to RSTs so that we do not
 696                          * continually send RSTs between two endpoints.
 697                          *
 698                          * Note that since this is a reply, dst is src and src
 699                          * is dst.
 700                          */
 701                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 702                                 pr_err("unable to send reset\n");
 703
 704                         err = VMCI_ERROR_NOT_FOUND;
 705                         goto out;
 706                 }
 707         }
 708
 709         /* If the received packet type is beyond all types known to this
 710          * implementation, reply with an invalid message.  Hopefully this will
 711          * help when implementing backwards compatibility in the future.
 712          */
 713         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 714                 vmci_transport_send_invalid_bh(&dst, &src);
 715                 err = VMCI_ERROR_INVALID_ARGS;
 716                 goto out;
 717         }
 718
 719         /* This handler is privileged when this module is running on the host.
 720          * We will get datagram connect requests from all endpoints (even VMs
 721          * that are in a restricted context). If we get one from a restricted
 722          * context then the destination socket must be trusted.
 723          *
 724          * NOTE: We access the socket struct without holding the lock here.
 725          * This is ok because the field we are interested is never modified
 726          * outside of the create and destruct socket functions.
 727          */
 728         vsk = vsock_sk(sk);
 729         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 730                 err = VMCI_ERROR_NO_ACCESS;
 731                 goto out;
 732         }
 733
 734         /* We do most everything in a work queue, but let's fast path the
 735          * notification of reads and writes to help data transfer performance.
 736          * We can only do this if there is no process context code executing
 737          * for this socket since that may change the state.
 738          */
 739         bh_lock_sock(sk);
 740
 741         if (!sock_owned_by_user(sk)) {
 742                 /* The local context ID may be out of date, update it. */
 743                 vsk->local_addr.svm_cid = dst.svm_cid;
 744
 745                 if (sk->sk_state == TCP_ESTABLISHED)
 746                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 747                                         sk, pkt, true, &dst, &src,
 748                                         &bh_process_pkt);
 749         }
 750
 751         bh_unlock_sock(sk);
 752
 753         if (!bh_process_pkt) {
 754                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 755
 756                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 757                 if (!recv_pkt_info) {
 758                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 759                                 pr_err("unable to send reset\n");
 760
 761                         err = VMCI_ERROR_NO_MEM;
 762                         goto out;
 763                 }
 764
 765                 recv_pkt_info->sk = sk;
 766                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 767                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 768
 769                 schedule_work(&recv_pkt_info->work);
 770                 /* Clear sk so that the reference count incremented by one of
 771                  * the Find functions above is not decremented below.  We need
 772                  * that reference count for the packet handler we've scheduled
 773                  * to run.
 774                  */
 775                 sk = NULL;
 776         }
 777
 778 out:
 779         if (sk)
 780                 sock_put(sk);
 781
 782         return err;
 783 }
 784
 785 static void vmci_transport_handle_detach(struct sock *sk)
 786 {
 787         struct vsock_sock *vsk;
 788
 789         vsk = vsock_sk(sk);
 790         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 791                 sock_set_flag(sk, SOCK_DONE);
 792
 793                 /* On a detach the peer will not be sending or receiving
 794                  * anymore.
 795                  */
 796                 vsk->peer_shutdown = SHUTDOWN_MASK;
 797
 798                 /* We should not be sending anymore since the peer won't be
 799                  * there to receive, but we can still receive if there is data
 800                  * left in our consume queue. If the local endpoint is a host,
 801                  * we can't call vsock_stream_has_data, since that may block,
 802                  * but a host endpoint can't read data once the VM has
 803                  * detached, so there is no available data in that case.
 804                  */
 805                 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
 806                     vsock_stream_has_data(vsk) <= 0) {
 807                         if (sk->sk_state == TCP_SYN_SENT) {
 808                                 /* The peer may detach from a queue pair while
 809                                  * we are still in the connecting state, i.e.,
 810                                  * if the peer VM is killed after attaching to
 811                                  * a queue pair, but before we complete the
 812                                  * handshake. In that case, we treat the detach
 813                                  * event like a reset.
 814                                  */
 815
 816                                 sk->sk_state = TCP_CLOSE;
 817                                 sk->sk_err = ECONNRESET;
 818                                 sk->sk_error_report(sk);
 819                                 return;
 820                         }
 821                         sk->sk_state = TCP_CLOSE;
 822                 }
 823                 sk->sk_state_change(sk);
 824         }
 825 }
 826
 827 static void vmci_transport_peer_detach_cb(u32 sub_id,
 828                                           const struct vmci_event_data *e_data,
 829                                           void *client_data)
 830 {
 831         struct vmci_transport *trans = client_data;
 832         const struct vmci_event_payload_qp *e_payload;
 833
 834         e_payload = vmci_event_data_const_payload(e_data);
 835
 836         /* XXX This is lame, we should provide a way to lookup sockets by
 837          * qp_handle.
 838          */
 839         if (vmci_handle_is_invalid(e_payload->handle) ||
 840             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 841                 return;
 842
 843         /* We don't ask for delayed CBs when we subscribe to this event (we
 844          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 845          * guarantees in that case about what context we might be running in,
 846          * so it could be BH or process, blockable or non-blockable.  So we
 847          * need to account for all possible contexts here.
 848          */
 849         spin_lock_bh(&trans->lock);
 850         if (!trans->sk)
 851                 goto out;
 852
 853         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 854          * where trans->sk isn't locked.
 855          */
 856         bh_lock_sock(trans->sk);
 857
 858         vmci_transport_handle_detach(trans->sk);
 859
 860         bh_unlock_sock(trans->sk);
 861  out:
 862         spin_unlock_bh(&trans->lock);
 863 }
 864
 865 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 866                                          const struct vmci_event_data *e_data,
 867                                          void *client_data)
 868 {
 869         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 870 }
 871
 872 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 873 {
 874         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 875         struct vmci_transport_packet *pkt;
 876         struct sock *sk;
 877
 878         recv_pkt_info =
 879                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 880         sk = recv_pkt_info->sk;
 881         pkt = &recv_pkt_info->pkt;
 882
 883         lock_sock(sk);
 884
 885         /* The local context ID may be out of date. */
 886         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 887
 888         switch (sk->sk_state) {
 889         case TCP_LISTEN:
 890                 vmci_transport_recv_listen(sk, pkt);
 891                 break;
 892         case TCP_SYN_SENT:
 893                 /* Processing of pending connections for servers goes through
 894                  * the listening socket, so see vmci_transport_recv_listen()
 895                  * for that path.
 896                  */
 897                 vmci_transport_recv_connecting_client(sk, pkt);
 898                 break;
 899         case TCP_ESTABLISHED:
 900                 vmci_transport_recv_connected(sk, pkt);
 901                 break;
 902         default:
 903                 /* Because this function does not run in the same context as
 904                  * vmci_transport_recv_stream_cb it is possible that the
 905                  * socket has closed. We need to let the other side know or it
 906                  * could be sitting in a connect and hang forever. Send a
 907                  * reset to prevent that.
 908                  */
 909                 vmci_transport_send_reset(sk, pkt);
 910                 break;
 911         }
 912
 913         release_sock(sk);
 914         kfree(recv_pkt_info);
 915         /* Release reference obtained in the stream callback when we fetched
 916          * this socket out of the bound or connected list.
 917          */
 918         sock_put(sk);
 919 }
 920
 921 static int vmci_transport_recv_listen(struct sock *sk,
 922                                       struct vmci_transport_packet *pkt)
 923 {
 924         struct sock *pending;
 925         struct vsock_sock *vpending;
 926         int err;
 927         u64 qp_size;
 928         bool old_request = false;
 929         bool old_pkt_proto = false;
 930
 931         err = 0;
 932
 933         /* Because we are in the listen state, we could be receiving a packet
 934          * for ourself or any previous connection requests that we received.
 935          * If it's the latter, we try to find a socket in our list of pending
 936          * connections and, if we do, call the appropriate handler for the
 937          * state that that socket is in.  Otherwise we try to service the
 938          * connection request.
 939          */
 940         pending = vmci_transport_get_pending(sk, pkt);
 941         if (pending) {
 942                 lock_sock(pending);
 943
 944                 /* The local context ID may be out of date. */
 945                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 946
 947                 switch (pending->sk_state) {
 948                 case TCP_SYN_SENT:
 949                         err = vmci_transport_recv_connecting_server(sk,
 950                                                                     pending,
 951                                                                     pkt);
 952                         break;
 953                 default:
 954                         vmci_transport_send_reset(pending, pkt);
 955                         err = -EINVAL;
 956                 }
 957
 958                 if (err < 0)
 959                         vsock_remove_pending(sk, pending);
 960
 961                 release_sock(pending);
 962                 vmci_transport_release_pending(pending);
 963
 964                 return err;
 965         }
 966
 967         /* The listen state only accepts connection requests.  Reply with a
 968          * reset unless we received a reset.
 969          */
 970
 971         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 972               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 973                 vmci_transport_reply_reset(pkt);
 974                 return -EINVAL;
 975         }
 976
 977         if (pkt->u.size == 0) {
 978                 vmci_transport_reply_reset(pkt);
 979                 return -EINVAL;
 980         }
 981
 982         /* If this socket can't accommodate this connection request, we send a
 983          * reset.  Otherwise we create and initialize a child socket and reply
 984          * with a connection negotiation.
 985          */
 986         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
 987                 vmci_transport_reply_reset(pkt);
 988                 return -ECONNREFUSED;
 989         }
 990
 991         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
 992                                  sk->sk_type, 0);
 993         if (!pending) {
 994                 vmci_transport_send_reset(sk, pkt);
 995                 return -ENOMEM;
 996         }
 997
 998         vpending = vsock_sk(pending);
 999
1000         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1001                         pkt->dst_port);
1002         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1003                         pkt->src_port);
1004
1005         /* If the proposed size fits within our min/max, accept it. Otherwise
1006          * propose our own size.
1007          */
1008         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1009             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1010                 qp_size = pkt->u.size;
1011         } else {
1012                 qp_size = vmci_trans(vpending)->queue_pair_size;
1013         }
1014
1015         /* Figure out if we are using old or new requests based on the
1016          * overrides pkt types sent by our peer.
1017          */
1018         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1019                 old_request = old_pkt_proto;
1020         } else {
1021                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1022                         old_request = true;
1023                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1024                         old_request = false;
1025
1026         }
1027
1028         if (old_request) {
1029                 /* Handle a REQUEST (or override) */
1030                 u16 version = VSOCK_PROTO_INVALID;
1031                 if (vmci_transport_proto_to_notify_struct(
1032                         pending, &version, true))
1033                         err = vmci_transport_send_negotiate(pending, qp_size);
1034                 else
1035                         err = -EINVAL;
1036
1037         } else {
1038                 /* Handle a REQUEST2 (or override) */
1039                 int proto_int = pkt->proto;
1040                 int pos;
1041                 u16 active_proto_version = 0;
1042
1043                 /* The list of possible protocols is the intersection of all
1044                  * protocols the client supports ... plus all the protocols we
1045                  * support.
1046                  */
1047                 proto_int &= vmci_transport_new_proto_supported_versions();
1048
1049                 /* We choose the highest possible protocol version and use that
1050                  * one.
1051                  */
1052                 pos = fls(proto_int);
1053                 if (pos) {
1054                         active_proto_version = (1 << (pos - 1));
1055                         if (vmci_transport_proto_to_notify_struct(
1056                                 pending, &active_proto_version, false))
1057                                 err = vmci_transport_send_negotiate2(pending,
1058                                                         qp_size,
1059                                                         active_proto_version);
1060                         else
1061                                 err = -EINVAL;
1062
1063                 } else {
1064                         err = -EINVAL;
1065                 }
1066         }
1067
1068         if (err < 0) {
1069                 vmci_transport_send_reset(sk, pkt);
1070                 sock_put(pending);
1071                 err = vmci_transport_error_to_vsock_error(err);
1072                 goto out;
1073         }
1074
1075         vsock_add_pending(sk, pending);
1076         sk->sk_ack_backlog++;
1077
1078         pending->sk_state = TCP_SYN_SENT;
1079         vmci_trans(vpending)->produce_size =
1080                 vmci_trans(vpending)->consume_size = qp_size;
1081         vmci_trans(vpending)->queue_pair_size = qp_size;
1082
1083         vmci_trans(vpending)->notify_ops->process_request(pending);
1084
1085         /* We might never receive another message for this socket and it's not
1086          * connected to any process, so we have to ensure it gets cleaned up
1087          * ourself.  Our delayed work function will take care of that.  Note
1088          * that we do not ever cancel this function since we have few
1089          * guarantees about its state when calling cancel_delayed_work().
1090          * Instead we hold a reference on the socket for that function and make
1091          * it capable of handling cases where it needs to do nothing but
1092          * release that reference.
1093          */
1094         vpending->listener = sk;
1095         sock_hold(sk);
1096         sock_hold(pending);
1097         INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1098         schedule_delayed_work(&vpending->dwork, HZ);
1099
1100 out:
1101         return err;
1102 }
1103
1104 static int
1105 vmci_transport_recv_connecting_server(struct sock *listener,
1106                                       struct sock *pending,
1107                                       struct vmci_transport_packet *pkt)
1108 {
1109         struct vsock_sock *vpending;
1110         struct vmci_handle handle;
1111         struct vmci_qp *qpair;
1112         bool is_local;
1113         u32 flags;
1114         u32 detach_sub_id;
1115         int err;
1116         int skerr;
1117
1118         vpending = vsock_sk(pending);
1119         detach_sub_id = VMCI_INVALID_ID;
1120
1121         switch (pkt->type) {
1122         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1123                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1124                         vmci_transport_send_reset(pending, pkt);
1125                         skerr = EPROTO;
1126                         err = -EINVAL;
1127                         goto destroy;
1128                 }
1129                 break;
1130         default:
1131                 /* Close and cleanup the connection. */
1132                 vmci_transport_send_reset(pending, pkt);
1133                 skerr = EPROTO;
1134                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1135                 goto destroy;
1136         }
1137
1138         /* In order to complete the connection we need to attach to the offered
1139          * queue pair and send an attach notification.  We also subscribe to the
1140          * detach event so we know when our peer goes away, and we do that
1141          * before attaching so we don't miss an event.  If all this succeeds,
1142          * we update our state and wakeup anything waiting in accept() for a
1143          * connection.
1144          */
1145
1146         /* We don't care about attach since we ensure the other side has
1147          * attached by specifying the ATTACH_ONLY flag below.
1148          */
1149         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1150                                    vmci_transport_peer_detach_cb,
1151                                    vmci_trans(vpending), &detach_sub_id);
1152         if (err < VMCI_SUCCESS) {
1153                 vmci_transport_send_reset(pending, pkt);
1154                 err = vmci_transport_error_to_vsock_error(err);
1155                 skerr = -err;
1156                 goto destroy;
1157         }
1158
1159         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1160
1161         /* Now attach to the queue pair the client created. */
1162         handle = pkt->u.handle;
1163
1164         /* vpending->local_addr always has a context id so we do not need to
1165          * worry about VMADDR_CID_ANY in this case.
1166          */
1167         is_local =
1168             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1169         flags = VMCI_QPFLAG_ATTACH_ONLY;
1170         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1171
1172         err = vmci_transport_queue_pair_alloc(
1173                                         &qpair,
1174                                         &handle,
1175                                         vmci_trans(vpending)->produce_size,
1176                                         vmci_trans(vpending)->consume_size,
1177                                         pkt->dg.src.context,
1178                                         flags,
1179                                         vmci_transport_is_trusted(
1180                                                 vpending,
1181                                                 vpending->remote_addr.svm_cid));
1182         if (err < 0) {
1183                 vmci_transport_send_reset(pending, pkt);
1184                 skerr = -err;
1185                 goto destroy;
1186         }
1187
1188         vmci_trans(vpending)->qp_handle = handle;
1189         vmci_trans(vpending)->qpair = qpair;
1190
1191         /* When we send the attach message, we must be ready to handle incoming
1192          * control messages on the newly connected socket. So we move the
1193          * pending socket to the connected state before sending the attach
1194          * message. Otherwise, an incoming packet triggered by the attach being
1195          * received by the peer may be processed concurrently with what happens
1196          * below after sending the attach message, and that incoming packet
1197          * will find the listening socket instead of the (currently) pending
1198          * socket. Note that enqueueing the socket increments the reference
1199          * count, so even if a reset comes before the connection is accepted,
1200          * the socket will be valid until it is removed from the queue.
1201          *
1202          * If we fail sending the attach below, we remove the socket from the
1203          * connected list and move the socket to TCP_CLOSE before
1204          * releasing the lock, so a pending slow path processing of an incoming
1205          * packet will not see the socket in the connected state in that case.
1206          */
1207         pending->sk_state = TCP_ESTABLISHED;
1208
1209         vsock_insert_connected(vpending);
1210
1211         /* Notify our peer of our attach. */
1212         err = vmci_transport_send_attach(pending, handle);
1213         if (err < 0) {
1214                 vsock_remove_connected(vpending);
1215                 pr_err("Could not send attach\n");
1216                 vmci_transport_send_reset(pending, pkt);
1217                 err = vmci_transport_error_to_vsock_error(err);
1218                 skerr = -err;
1219                 goto destroy;
1220         }
1221
1222         /* We have a connection. Move the now connected socket from the
1223          * listener's pending list to the accept queue so callers of accept()
1224          * can find it.
1225          */
1226         vsock_remove_pending(listener, pending);
1227         vsock_enqueue_accept(listener, pending);
1228
1229         /* Callers of accept() will be be waiting on the listening socket, not
1230          * the pending socket.
1231          */
1232         listener->sk_data_ready(listener);
1233
1234         return 0;
1235
1236 destroy:
1237         pending->sk_err = skerr;
1238         pending->sk_state = TCP_CLOSE;
1239         /* As long as we drop our reference, all necessary cleanup will handle
1240          * when the cleanup function drops its reference and our destruct
1241          * implementation is called.  Note that since the listen handler will
1242          * remove pending from the pending list upon our failure, the cleanup
1243          * function won't drop the additional reference, which is why we do it
1244          * here.
1245          */
1246         sock_put(pending);
1247
1248         return err;
1249 }
1250
1251 static int
1252 vmci_transport_recv_connecting_client(struct sock *sk,
1253                                       struct vmci_transport_packet *pkt)
1254 {
1255         struct vsock_sock *vsk;
1256         int err;
1257         int skerr;
1258
1259         vsk = vsock_sk(sk);
1260
1261         switch (pkt->type) {
1262         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1263                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1264                     !vmci_handle_is_equal(pkt->u.handle,
1265                                           vmci_trans(vsk)->qp_handle)) {
1266                         skerr = EPROTO;
1267                         err = -EINVAL;
1268                         goto destroy;
1269                 }
1270
1271                 /* Signify the socket is connected and wakeup the waiter in
1272                  * connect(). Also place the socket in the connected table for
1273                  * accounting (it can already be found since it's in the bound
1274                  * table).
1275                  */
1276                 sk->sk_state = TCP_ESTABLISHED;
1277                 sk->sk_socket->state = SS_CONNECTED;
1278                 vsock_insert_connected(vsk);
1279                 sk->sk_state_change(sk);
1280
1281                 break;
1282         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1283         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1284                 if (pkt->u.size == 0
1285                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1286                     || pkt->src_port != vsk->remote_addr.svm_port
1287                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1288                     || vmci_trans(vsk)->qpair
1289                     || vmci_trans(vsk)->produce_size != 0
1290                     || vmci_trans(vsk)->consume_size != 0
1291                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1292                         skerr = EPROTO;
1293                         err = -EINVAL;
1294
1295                         goto destroy;
1296                 }
1297
1298                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1299                 if (err) {
1300                         skerr = -err;
1301                         goto destroy;
1302                 }
1303
1304                 break;
1305         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1306                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1307                 if (err) {
1308                         skerr = -err;
1309                         goto destroy;
1310                 }
1311
1312                 break;
1313         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1314                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1315                  * continue processing here after they sent an INVALID packet.
1316                  * This meant that we got a RST after the INVALID. We ignore a
1317                  * RST after an INVALID. The common code doesn't send the RST
1318                  * ... so we can hang if an old version of the common code
1319                  * fails between getting a REQUEST and sending an OFFER back.
1320                  * Not much we can do about it... except hope that it doesn't
1321                  * happen.
1322                  */
1323                 if (vsk->ignore_connecting_rst) {
1324                         vsk->ignore_connecting_rst = false;
1325                 } else {
1326                         skerr = ECONNRESET;
1327                         err = 0;
1328                         goto destroy;
1329                 }
1330
1331                 break;
1332         default:
1333                 /* Close and cleanup the connection. */
1334                 skerr = EPROTO;
1335                 err = -EINVAL;
1336                 goto destroy;
1337         }
1338
1339         return 0;
1340
1341 destroy:
1342         vmci_transport_send_reset(sk, pkt);
1343
1344         sk->sk_state = TCP_CLOSE;
1345         sk->sk_err = skerr;
1346         sk->sk_error_report(sk);
1347         return err;
1348 }
1349
1350 static int vmci_transport_recv_connecting_client_negotiate(
1351                                         struct sock *sk,
1352                                         struct vmci_transport_packet *pkt)
1353 {
1354         int err;
1355         struct vsock_sock *vsk;
1356         struct vmci_handle handle;
1357         struct vmci_qp *qpair;
1358         u32 detach_sub_id;
1359         bool is_local;
1360         u32 flags;
1361         bool old_proto = true;
1362         bool old_pkt_proto;
1363         u16 version;
1364
1365         vsk = vsock_sk(sk);
1366         handle = VMCI_INVALID_HANDLE;
1367         detach_sub_id = VMCI_INVALID_ID;
1368
1369         /* If we have gotten here then we should be past the point where old
1370          * linux vsock could have sent the bogus rst.
1371          */
1372         vsk->sent_request = false;
1373         vsk->ignore_connecting_rst = false;
1374
1375         /* Verify that we're OK with the proposed queue pair size */
1376         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1377             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1378                 err = -EINVAL;
1379                 goto destroy;
1380         }
1381
1382         /* At this point we know the CID the peer is using to talk to us. */
1383
1384         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1385                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1386
1387         /* Setup the notify ops to be the highest supported version that both
1388          * the server and the client support.
1389          */
1390
1391         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1392                 old_proto = old_pkt_proto;
1393         } else {
1394                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1395                         old_proto = true;
1396                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1397                         old_proto = false;
1398
1399         }
1400
1401         if (old_proto)
1402                 version = VSOCK_PROTO_INVALID;
1403         else
1404                 version = pkt->proto;
1405
1406         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1407                 err = -EINVAL;
1408                 goto destroy;
1409         }
1410
1411         /* Subscribe to detach events first.
1412          *
1413          * XXX We attach once for each queue pair created for now so it is easy
1414          * to find the socket (it's provided), but later we should only
1415          * subscribe once and add a way to lookup sockets by queue pair handle.
1416          */
1417         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1418                                    vmci_transport_peer_detach_cb,
1419                                    vmci_trans(vsk), &detach_sub_id);
1420         if (err < VMCI_SUCCESS) {
1421                 err = vmci_transport_error_to_vsock_error(err);
1422                 goto destroy;
1423         }
1424
1425         /* Make VMCI select the handle for us. */
1426         handle = VMCI_INVALID_HANDLE;
1427         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1428         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1429
1430         err = vmci_transport_queue_pair_alloc(&qpair,
1431                                               &handle,
1432                                               pkt->u.size,
1433                                               pkt->u.size,
1434                                               vsk->remote_addr.svm_cid,
1435                                               flags,
1436                                               vmci_transport_is_trusted(
1437                                                   vsk,
1438                                                   vsk->
1439                                                   remote_addr.svm_cid));
1440         if (err < 0)
1441                 goto destroy;
1442
1443         err = vmci_transport_send_qp_offer(sk, handle);
1444         if (err < 0) {
1445                 err = vmci_transport_error_to_vsock_error(err);
1446                 goto destroy;
1447         }
1448
1449         vmci_trans(vsk)->qp_handle = handle;
1450         vmci_trans(vsk)->qpair = qpair;
1451
1452         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1453                 pkt->u.size;
1454
1455         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1456
1457         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1458
1459         return 0;
1460
1461 destroy:
1462         if (detach_sub_id != VMCI_INVALID_ID)
1463                 vmci_event_unsubscribe(detach_sub_id);
1464
1465         if (!vmci_handle_is_invalid(handle))
1466                 vmci_qpair_detach(&qpair);
1467
1468         return err;
1469 }
1470
1471 static int
1472 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1473                                               struct vmci_transport_packet *pkt)
1474 {
1475         int err = 0;
1476         struct vsock_sock *vsk = vsock_sk(sk);
1477
1478         if (vsk->sent_request) {
1479                 vsk->sent_request = false;
1480                 vsk->ignore_connecting_rst = true;
1481
1482                 err = vmci_transport_send_conn_request(
1483                         sk, vmci_trans(vsk)->queue_pair_size);
1484                 if (err < 0)
1485                         err = vmci_transport_error_to_vsock_error(err);
1486                 else
1487                         err = 0;
1488
1489         }
1490
1491         return err;
1492 }
1493
1494 static int vmci_transport_recv_connected(struct sock *sk,
1495                                          struct vmci_transport_packet *pkt)
1496 {
1497         struct vsock_sock *vsk;
1498         bool pkt_processed = false;
1499
1500         /* In cases where we are closing the connection, it's sufficient to
1501          * mark the state change (and maybe error) and wake up any waiting
1502          * threads. Since this is a connected socket, it's owned by a user
1503          * process and will be cleaned up when the failure is passed back on
1504          * the current or next system call.  Our system call implementations
1505          * must therefore check for error and state changes on entry and when
1506          * being awoken.
1507          */
1508         switch (pkt->type) {
1509         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1510                 if (pkt->u.mode) {
1511                         vsk = vsock_sk(sk);
1512
1513                         vsk->peer_shutdown |= pkt->u.mode;
1514                         sk->sk_state_change(sk);
1515                 }
1516                 break;
1517
1518         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1519                 vsk = vsock_sk(sk);
1520                 /* It is possible that we sent our peer a message (e.g a
1521                  * WAITING_READ) right before we got notified that the peer had
1522                  * detached. If that happens then we can get a RST pkt back
1523                  * from our peer even though there is data available for us to
1524                  * read. In that case, don't shutdown the socket completely but
1525                  * instead allow the local client to finish reading data off
1526                  * the queuepair. Always treat a RST pkt in connected mode like
1527                  * a clean shutdown.
1528                  */
1529                 sock_set_flag(sk, SOCK_DONE);
1530                 vsk->peer_shutdown = SHUTDOWN_MASK;
1531                 if (vsock_stream_has_data(vsk) <= 0)
1532                         sk->sk_state = TCP_CLOSING;
1533
1534                 sk->sk_state_change(sk);
1535                 break;
1536
1537         default:
1538                 vsk = vsock_sk(sk);
1539                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1540                                 sk, pkt, false, NULL, NULL,
1541                                 &pkt_processed);
1542                 if (!pkt_processed)
1543                         return -EINVAL;
1544
1545                 break;
1546         }
1547
1548         return 0;
1549 }
1550
1551 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1552                                       struct vsock_sock *psk)
1553 {
1554         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1555         if (!vsk->trans)
1556                 return -ENOMEM;
1557
1558         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1559         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1560         vmci_trans(vsk)->qpair = NULL;
1561         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1562         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1563         vmci_trans(vsk)->notify_ops = NULL;
1564         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1565         vmci_trans(vsk)->sk = &vsk->sk;
1566         spin_lock_init(&vmci_trans(vsk)->lock);
1567         if (psk) {
1568                 vmci_trans(vsk)->queue_pair_size =
1569                         vmci_trans(psk)->queue_pair_size;
1570                 vmci_trans(vsk)->queue_pair_min_size =
1571                         vmci_trans(psk)->queue_pair_min_size;
1572                 vmci_trans(vsk)->queue_pair_max_size =
1573                         vmci_trans(psk)->queue_pair_max_size;
1574         } else {
1575                 vmci_trans(vsk)->queue_pair_size =
1576                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1577                 vmci_trans(vsk)->queue_pair_min_size =
1578                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1579                 vmci_trans(vsk)->queue_pair_max_size =
1580                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1581         }
1582
1583         return 0;
1584 }
1585
1586 static void vmci_transport_free_resources(struct list_head *transport_list)
1587 {
1588         while (!list_empty(transport_list)) {
1589                 struct vmci_transport *transport =
1590                     list_first_entry(transport_list, struct vmci_transport,
1591                                      elem);
1592                 list_del(&transport->elem);
1593
1594                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1595                         vmci_event_unsubscribe(transport->detach_sub_id);
1596                         transport->detach_sub_id = VMCI_INVALID_ID;
1597                 }
1598
1599                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1600                         vmci_qpair_detach(&transport->qpair);
1601                         transport->qp_handle = VMCI_INVALID_HANDLE;
1602                         transport->produce_size = 0;
1603                         transport->consume_size = 0;
1604                 }
1605
1606                 kfree(transport);
1607         }
1608 }
1609
1610 static void vmci_transport_cleanup(struct work_struct *work)
1611 {
1612         LIST_HEAD(pending);
1613
1614         spin_lock_bh(&vmci_transport_cleanup_lock);
1615         list_replace_init(&vmci_transport_cleanup_list, &pending);
1616         spin_unlock_bh(&vmci_transport_cleanup_lock);
1617         vmci_transport_free_resources(&pending);
1618 }
1619
1620 static void vmci_transport_destruct(struct vsock_sock *vsk)
1621 {
1622         /* Ensure that the detach callback doesn't use the sk/vsk
1623          * we are about to destruct.
1624          */
1625         spin_lock_bh(&vmci_trans(vsk)->lock);
1626         vmci_trans(vsk)->sk = NULL;
1627         spin_unlock_bh(&vmci_trans(vsk)->lock);
1628
1629         if (vmci_trans(vsk)->notify_ops)
1630                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1631
1632         spin_lock_bh(&vmci_transport_cleanup_lock);
1633         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1634         spin_unlock_bh(&vmci_transport_cleanup_lock);
1635         schedule_work(&vmci_transport_cleanup_work);
1636
1637         vsk->trans = NULL;
1638 }
1639
1640 static void vmci_transport_release(struct vsock_sock *vsk)
1641 {
1642         vsock_remove_sock(vsk);
1643
1644         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1645                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1646                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1647         }
1648 }
1649
1650 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1651                                      struct sockaddr_vm *addr)
1652 {
1653         u32 port;
1654         u32 flags;
1655         int err;
1656
1657         /* VMCI will select a resource ID for us if we provide
1658          * VMCI_INVALID_ID.
1659          */
1660         port = addr->svm_port == VMADDR_PORT_ANY ?
1661                         VMCI_INVALID_ID : addr->svm_port;
1662
1663         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1664                 return -EACCES;
1665
1666         flags = addr->svm_cid == VMADDR_CID_ANY ?
1667                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1668
1669         err = vmci_transport_datagram_create_hnd(port, flags,
1670                                                  vmci_transport_recv_dgram_cb,
1671                                                  &vsk->sk,
1672                                                  &vmci_trans(vsk)->dg_handle);
1673         if (err < VMCI_SUCCESS)
1674                 return vmci_transport_error_to_vsock_error(err);
1675         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1676                         vmci_trans(vsk)->dg_handle.resource);
1677
1678         return 0;
1679 }
1680
1681 static int vmci_transport_dgram_enqueue(
1682         struct vsock_sock *vsk,
1683         struct sockaddr_vm *remote_addr,
1684         struct msghdr *msg,
1685         size_t len)
1686 {
1687         int err;
1688         struct vmci_datagram *dg;
1689
1690         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1691                 return -EMSGSIZE;
1692
1693         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1694                 return -EPERM;
1695
1696         /* Allocate a buffer for the user's message and our packet header. */
1697         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1698         if (!dg)
1699                 return -ENOMEM;
1700
1701         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1702
1703         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1704                                    remote_addr->svm_port);
1705         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1706                                    vsk->local_addr.svm_port);
1707         dg->payload_size = len;
1708
1709         err = vmci_datagram_send(dg);
1710         kfree(dg);
1711         if (err < 0)
1712                 return vmci_transport_error_to_vsock_error(err);
1713
1714         return err - sizeof(*dg);
1715 }
1716
1717 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1718                                         struct msghdr *msg, size_t len,
1719                                         int flags)
1720 {
1721         int err;
1722         int noblock;
1723         struct vmci_datagram *dg;
1724         size_t payload_len;
1725         struct sk_buff *skb;
1726
1727         noblock = flags & MSG_DONTWAIT;
1728
1729         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1730                 return -EOPNOTSUPP;
1731
1732         /* Retrieve the head sk_buff from the socket's receive queue. */
1733         err = 0;
1734         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1735         if (!skb)
1736                 return err;
1737
1738         dg = (struct vmci_datagram *)skb->data;
1739         if (!dg)
1740                 /* err is 0, meaning we read zero bytes. */
1741                 goto out;
1742
1743         payload_len = dg->payload_size;
1744         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1745         if (payload_len != skb->len - sizeof(*dg)) {
1746                 err = -EINVAL;
1747                 goto out;
1748         }
1749
1750         if (payload_len > len) {
1751                 payload_len = len;
1752                 msg->msg_flags |= MSG_TRUNC;
1753         }
1754
1755         /* Place the datagram payload in the user's iovec. */
1756         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1757         if (err)
1758                 goto out;
1759
1760         if (msg->msg_name) {
1761                 /* Provide the address of the sender. */
1762                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1763                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1764                 msg->msg_namelen = sizeof(*vm_addr);
1765         }
1766         err = payload_len;
1767
1768 out:
1769         skb_free_datagram(&vsk->sk, skb);
1770         return err;
1771 }
1772
1773 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1774 {
1775         if (cid == VMADDR_CID_HYPERVISOR) {
1776                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1777                  * state and are allowed.
1778                  */
1779                 return port == VMCI_UNITY_PBRPC_REGISTER;
1780         }
1781
1782         return true;
1783 }
1784
1785 static int vmci_transport_connect(struct vsock_sock *vsk)
1786 {
1787         int err;
1788         bool old_pkt_proto = false;
1789         struct sock *sk = &vsk->sk;
1790
1791         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1792                 old_pkt_proto) {
1793                 err = vmci_transport_send_conn_request(
1794                         sk, vmci_trans(vsk)->queue_pair_size);
1795                 if (err < 0) {
1796                         sk->sk_state = TCP_CLOSE;
1797                         return err;
1798                 }
1799         } else {
1800                 int supported_proto_versions =
1801                         vmci_transport_new_proto_supported_versions();
1802                 err = vmci_transport_send_conn_request2(
1803                                 sk, vmci_trans(vsk)->queue_pair_size,
1804                                 supported_proto_versions);
1805                 if (err < 0) {
1806                         sk->sk_state = TCP_CLOSE;
1807                         return err;
1808                 }
1809
1810                 vsk->sent_request = true;
1811         }
1812
1813         return err;
1814 }
1815
1816 static ssize_t vmci_transport_stream_dequeue(
1817         struct vsock_sock *vsk,
1818         struct msghdr *msg,
1819         size_t len,
1820         int flags)
1821 {
1822         if (flags & MSG_PEEK)
1823                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1824         else
1825                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1826 }
1827
1828 static ssize_t vmci_transport_stream_enqueue(
1829         struct vsock_sock *vsk,
1830         struct msghdr *msg,
1831         size_t len)
1832 {
1833         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1834 }
1835
1836 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1837 {
1838         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1839 }
1840
1841 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1842 {
1843         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1844 }
1845
1846 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1847 {
1848         return vmci_trans(vsk)->consume_size;
1849 }
1850
1851 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1852 {
1853         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1854 }
1855
1856 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1857 {
1858         return vmci_trans(vsk)->queue_pair_size;
1859 }
1860
1861 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1862 {
1863         return vmci_trans(vsk)->queue_pair_min_size;
1864 }
1865
1866 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1867 {
1868         return vmci_trans(vsk)->queue_pair_max_size;
1869 }
1870
1871 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1872 {
1873         if (val < vmci_trans(vsk)->queue_pair_min_size)
1874                 vmci_trans(vsk)->queue_pair_min_size = val;
1875         if (val > vmci_trans(vsk)->queue_pair_max_size)
1876                 vmci_trans(vsk)->queue_pair_max_size = val;
1877         vmci_trans(vsk)->queue_pair_size = val;
1878 }
1879
1880 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1881                                                u64 val)
1882 {
1883         if (val > vmci_trans(vsk)->queue_pair_size)
1884                 vmci_trans(vsk)->queue_pair_size = val;
1885         vmci_trans(vsk)->queue_pair_min_size = val;
1886 }
1887
1888 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1889                                                u64 val)
1890 {
1891         if (val < vmci_trans(vsk)->queue_pair_size)
1892                 vmci_trans(vsk)->queue_pair_size = val;
1893         vmci_trans(vsk)->queue_pair_max_size = val;
1894 }
1895
1896 static int vmci_transport_notify_poll_in(
1897         struct vsock_sock *vsk,
1898         size_t target,
1899         bool *data_ready_now)
1900 {
1901         return vmci_trans(vsk)->notify_ops->poll_in(
1902                         &vsk->sk, target, data_ready_now);
1903 }
1904
1905 static int vmci_transport_notify_poll_out(
1906         struct vsock_sock *vsk,
1907         size_t target,
1908         bool *space_available_now)
1909 {
1910         return vmci_trans(vsk)->notify_ops->poll_out(
1911                         &vsk->sk, target, space_available_now);
1912 }
1913
1914 static int vmci_transport_notify_recv_init(
1915         struct vsock_sock *vsk,
1916         size_t target,
1917         struct vsock_transport_recv_notify_data *data)
1918 {
1919         return vmci_trans(vsk)->notify_ops->recv_init(
1920                         &vsk->sk, target,
1921                         (struct vmci_transport_recv_notify_data *)data);
1922 }
1923
1924 static int vmci_transport_notify_recv_pre_block(
1925         struct vsock_sock *vsk,
1926         size_t target,
1927         struct vsock_transport_recv_notify_data *data)
1928 {
1929         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1930                         &vsk->sk, target,
1931                         (struct vmci_transport_recv_notify_data *)data);
1932 }
1933
1934 static int vmci_transport_notify_recv_pre_dequeue(
1935         struct vsock_sock *vsk,
1936         size_t target,
1937         struct vsock_transport_recv_notify_data *data)
1938 {
1939         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1940                         &vsk->sk, target,
1941                         (struct vmci_transport_recv_notify_data *)data);
1942 }
1943
1944 static int vmci_transport_notify_recv_post_dequeue(
1945         struct vsock_sock *vsk,
1946         size_t target,
1947         ssize_t copied,
1948         bool data_read,
1949         struct vsock_transport_recv_notify_data *data)
1950 {
1951         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1952                         &vsk->sk, target, copied, data_read,
1953                         (struct vmci_transport_recv_notify_data *)data);
1954 }
1955
1956 static int vmci_transport_notify_send_init(
1957         struct vsock_sock *vsk,
1958         struct vsock_transport_send_notify_data *data)
1959 {
1960         return vmci_trans(vsk)->notify_ops->send_init(
1961                         &vsk->sk,
1962                         (struct vmci_transport_send_notify_data *)data);
1963 }
1964
1965 static int vmci_transport_notify_send_pre_block(
1966         struct vsock_sock *vsk,
1967         struct vsock_transport_send_notify_data *data)
1968 {
1969         return vmci_trans(vsk)->notify_ops->send_pre_block(
1970                         &vsk->sk,
1971                         (struct vmci_transport_send_notify_data *)data);
1972 }
1973
1974 static int vmci_transport_notify_send_pre_enqueue(
1975         struct vsock_sock *vsk,
1976         struct vsock_transport_send_notify_data *data)
1977 {
1978         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1979                         &vsk->sk,
1980                         (struct vmci_transport_send_notify_data *)data);
1981 }
1982
1983 static int vmci_transport_notify_send_post_enqueue(
1984         struct vsock_sock *vsk,
1985         ssize_t written,
1986         struct vsock_transport_send_notify_data *data)
1987 {
1988         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1989                         &vsk->sk, written,
1990                         (struct vmci_transport_send_notify_data *)data);
1991 }
1992
1993 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1994 {
1995         if (PROTOCOL_OVERRIDE != -1) {
1996                 if (PROTOCOL_OVERRIDE == 0)
1997                         *old_pkt_proto = true;
1998                 else
1999                         *old_pkt_proto = false;
2000
2001                 pr_info("Proto override in use\n");
2002                 return true;
2003         }
2004
2005         return false;
2006 }
2007
2008 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2009                                                   u16 *proto,
2010                                                   bool old_pkt_proto)
2011 {
2012         struct vsock_sock *vsk = vsock_sk(sk);
2013
2014         if (old_pkt_proto) {
2015                 if (*proto != VSOCK_PROTO_INVALID) {
2016                         pr_err("Can't set both an old and new protocol\n");
2017                         return false;
2018                 }
2019                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2020                 goto exit;
2021         }
2022
2023         switch (*proto) {
2024         case VSOCK_PROTO_PKT_ON_NOTIFY:
2025                 vmci_trans(vsk)->notify_ops =
2026                         &vmci_transport_notify_pkt_q_state_ops;
2027                 break;
2028         default:
2029                 pr_err("Unknown notify protocol version\n");
2030                 return false;
2031         }
2032
2033 exit:
2034         vmci_trans(vsk)->notify_ops->socket_init(sk);
2035         return true;
2036 }
2037
2038 static u16 vmci_transport_new_proto_supported_versions(void)
2039 {
2040         if (PROTOCOL_OVERRIDE != -1)
2041                 return PROTOCOL_OVERRIDE;
2042
2043         return VSOCK_PROTO_ALL_SUPPORTED;
2044 }
2045
2046 static u32 vmci_transport_get_local_cid(void)
2047 {
2048         return vmci_get_context_id();
2049 }
2050
2051 static const struct vsock_transport vmci_transport = {
2052         .init = vmci_transport_socket_init,
2053         .destruct = vmci_transport_destruct,
2054         .release = vmci_transport_release,
2055         .connect = vmci_transport_connect,
2056         .dgram_bind = vmci_transport_dgram_bind,
2057         .dgram_dequeue = vmci_transport_dgram_dequeue,
2058         .dgram_enqueue = vmci_transport_dgram_enqueue,
2059         .dgram_allow = vmci_transport_dgram_allow,
2060         .stream_dequeue = vmci_transport_stream_dequeue,
2061         .stream_enqueue = vmci_transport_stream_enqueue,
2062         .stream_has_data = vmci_transport_stream_has_data,
2063         .stream_has_space = vmci_transport_stream_has_space,
2064         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2065         .stream_is_active = vmci_transport_stream_is_active,
2066         .stream_allow = vmci_transport_stream_allow,
2067         .notify_poll_in = vmci_transport_notify_poll_in,
2068         .notify_poll_out = vmci_transport_notify_poll_out,
2069         .notify_recv_init = vmci_transport_notify_recv_init,
2070         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2071         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2072         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2073         .notify_send_init = vmci_transport_notify_send_init,
2074         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2075         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2076         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2077         .shutdown = vmci_transport_shutdown,
2078         .set_buffer_size = vmci_transport_set_buffer_size,
2079         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2080         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2081         .get_buffer_size = vmci_transport_get_buffer_size,
2082         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2083         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2084         .get_local_cid = vmci_transport_get_local_cid,
2085 };
2086
2087 static int __init vmci_transport_init(void)
2088 {
2089         int err;
2090
2091         /* Create the datagram handle that we will use to send and receive all
2092          * VSocket control messages for this context.
2093          */
2094         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2095                                                  VMCI_FLAG_ANYCID_DG_HND,
2096                                                  vmci_transport_recv_stream_cb,
2097                                                  NULL,
2098                                                  &vmci_transport_stream_handle);
2099         if (err < VMCI_SUCCESS) {
2100                 pr_err("Unable to create datagram handle. (%d)\n", err);
2101                 return vmci_transport_error_to_vsock_error(err);
2102         }
2103
2104         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2105                                    vmci_transport_qp_resumed_cb,
2106                                    NULL, &vmci_transport_qp_resumed_sub_id);
2107         if (err < VMCI_SUCCESS) {
2108                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2109                 err = vmci_transport_error_to_vsock_error(err);
2110                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2111                 goto err_destroy_stream_handle;
2112         }
2113
2114         err = vsock_core_init(&vmci_transport);
2115         if (err < 0)
2116                 goto err_unsubscribe;
2117
2118         return 0;
2119
2120 err_unsubscribe:
2121         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2122 err_destroy_stream_handle:
2123         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2124         return err;
2125 }
2126 module_init(vmci_transport_init);
2127
2128 static void __exit vmci_transport_exit(void)
2129 {
2130         cancel_work_sync(&vmci_transport_cleanup_work);
2131         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2132
2133         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2134                 if (vmci_datagram_destroy_handle(
2135                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2136                         pr_err("Couldn't destroy datagram handle\n");
2137                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2138         }
2139
2140         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2141                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2142                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2143         }
2144
2145         vsock_core_exit();
2146 }
2147 module_exit(vmci_transport_exit);
2148
2149 MODULE_AUTHOR("VMware, Inc.");
2150 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2151 MODULE_VERSION("1.0.5.0-k");
2152 MODULE_LICENSE("GPL v2");
2153 MODULE_ALIAS("vmware_vsock");
2154 MODULE_ALIAS_NETPROTO(PF_VSOCK);