kernel/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/miscdevice.h>
  25 #include <linux/module.h>
  26 #include <linux/mutex.h>
  27 #include <linux/net.h>
  28 #include <linux/poll.h>
  29 #include <linux/skbuff.h>
  30 #include <linux/smp.h>
  31 #include <linux/socket.h>
  32 #include <linux/stddef.h>
  33 #include <linux/unistd.h>
  34 #include <linux/wait.h>
  35 #include <linux/workqueue.h>
  36 #include <net/sock.h>
  37 #include <net/af_vsock.h>
  38
  39 #include "vmci_transport_notify.h"
  40
  41 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
  42 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
  43 static void vmci_transport_peer_detach_cb(u32 sub_id,
  44                                           const struct vmci_event_data *ed,
  45                                           void *client_data);
  46 static void vmci_transport_recv_pkt_work(struct work_struct *work);
  47 static void vmci_transport_cleanup(struct work_struct *work);
  48 static int vmci_transport_recv_listen(struct sock *sk,
  49                                       struct vmci_transport_packet *pkt);
  50 static int vmci_transport_recv_connecting_server(
  51                                         struct sock *sk,
  52                                         struct sock *pending,
  53                                         struct vmci_transport_packet *pkt);
  54 static int vmci_transport_recv_connecting_client(
  55                                         struct sock *sk,
  56                                         struct vmci_transport_packet *pkt);
  57 static int vmci_transport_recv_connecting_client_negotiate(
  58                                         struct sock *sk,
  59                                         struct vmci_transport_packet *pkt);
  60 static int vmci_transport_recv_connecting_client_invalid(
  61                                         struct sock *sk,
  62                                         struct vmci_transport_packet *pkt);
  63 static int vmci_transport_recv_connected(struct sock *sk,
  64                                          struct vmci_transport_packet *pkt);
  65 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
  66 static u16 vmci_transport_new_proto_supported_versions(void);
  67 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
  68                                                   bool old_pkt_proto);
  69
  70 struct vmci_transport_recv_pkt_info {
  71         struct work_struct work;
  72         struct sock *sk;
  73         struct vmci_transport_packet pkt;
  74 };
  75
  76 static LIST_HEAD(vmci_transport_cleanup_list);
  77 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
  78 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
  79
  80 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  81                                                            VMCI_INVALID_ID };
  82 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  83
  84 static int PROTOCOL_OVERRIDE = -1;
  85
  86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN   128
  87 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE       262144
  88 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX   262144
  89
  90 /* The default peer timeout indicates how long we will wait for a peer response
  91  * to a control message.
  92  */
  93 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
  94
  95 /* Helper function to convert from a VMCI error code to a VSock error code. */
  96
  97 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  98 {
  99         int err;
 100
 101         switch (vmci_error) {
 102         case VMCI_ERROR_NO_MEM:
 103                 err = ENOMEM;
 104                 break;
 105         case VMCI_ERROR_DUPLICATE_ENTRY:
 106         case VMCI_ERROR_ALREADY_EXISTS:
 107                 err = EADDRINUSE;
 108                 break;
 109         case VMCI_ERROR_NO_ACCESS:
 110                 err = EPERM;
 111                 break;
 112         case VMCI_ERROR_NO_RESOURCES:
 113                 err = ENOBUFS;
 114                 break;
 115         case VMCI_ERROR_INVALID_RESOURCE:
 116                 err = EHOSTUNREACH;
 117                 break;
 118         case VMCI_ERROR_INVALID_ARGS:
 119         default:
 120                 err = EINVAL;
 121         }
 122
 123         return err > 0 ? -err : err;
 124 }
 125
 126 static u32 vmci_transport_peer_rid(u32 peer_cid)
 127 {
 128         if (VMADDR_CID_HYPERVISOR == peer_cid)
 129                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 130
 131         return VMCI_TRANSPORT_PACKET_RID;
 132 }
 133
 134 static inline void
 135 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 136                            struct sockaddr_vm *src,
 137                            struct sockaddr_vm *dst,
 138                            u8 type,
 139                            u64 size,
 140                            u64 mode,
 141                            struct vmci_transport_waiting_info *wait,
 142                            u16 proto,
 143                            struct vmci_handle handle)
 144 {
 145         /* We register the stream control handler as an any cid handle so we
 146          * must always send from a source address of VMADDR_CID_ANY
 147          */
 148         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 149                                        VMCI_TRANSPORT_PACKET_RID);
 150         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 151                                        vmci_transport_peer_rid(dst->svm_cid));
 152         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 153         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 154         pkt->type = type;
 155         pkt->src_port = src->svm_port;
 156         pkt->dst_port = dst->svm_port;
 157         memset(&pkt->proto, 0, sizeof(pkt->proto));
 158         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 159
 160         switch (pkt->type) {
 161         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 162                 pkt->u.size = 0;
 163                 break;
 164
 165         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 166         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 167                 pkt->u.size = size;
 168                 break;
 169
 170         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 171         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 172                 pkt->u.handle = handle;
 173                 break;
 174
 175         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 176         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 177         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 178                 pkt->u.size = 0;
 179                 break;
 180
 181         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 182                 pkt->u.mode = mode;
 183                 break;
 184
 185         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 186         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 187                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 188                 break;
 189
 190         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 191         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 192                 pkt->u.size = size;
 193                 pkt->proto = proto;
 194                 break;
 195         }
 196 }
 197
 198 static inline void
 199 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 200                                     struct sockaddr_vm *local,
 201                                     struct sockaddr_vm *remote)
 202 {
 203         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 204         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 205 }
 206
 207 static int
 208 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 209                                   struct sockaddr_vm *src,
 210                                   struct sockaddr_vm *dst,
 211                                   enum vmci_transport_packet_type type,
 212                                   u64 size,
 213                                   u64 mode,
 214                                   struct vmci_transport_waiting_info *wait,
 215                                   u16 proto,
 216                                   struct vmci_handle handle,
 217                                   bool convert_error)
 218 {
 219         int err;
 220
 221         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 222                                    proto, handle);
 223         err = vmci_datagram_send(&pkt->dg);
 224         if (convert_error && (err < 0))
 225                 return vmci_transport_error_to_vsock_error(err);
 226
 227         return err;
 228 }
 229
 230 static int
 231 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 232                                       enum vmci_transport_packet_type type,
 233                                       u64 size,
 234                                       u64 mode,
 235                                       struct vmci_transport_waiting_info *wait,
 236                                       struct vmci_handle handle)
 237 {
 238         struct vmci_transport_packet reply;
 239         struct sockaddr_vm src, dst;
 240
 241         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 242                 return 0;
 243         } else {
 244                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 245                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 246                                                          type,
 247                                                          size, mode, wait,
 248                                                          VSOCK_PROTO_INVALID,
 249                                                          handle, true);
 250         }
 251 }
 252
 253 static int
 254 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 255                                    struct sockaddr_vm *dst,
 256                                    enum vmci_transport_packet_type type,
 257                                    u64 size,
 258                                    u64 mode,
 259                                    struct vmci_transport_waiting_info *wait,
 260                                    struct vmci_handle handle)
 261 {
 262         /* Note that it is safe to use a single packet across all CPUs since
 263          * two tasklets of the same type are guaranteed to not ever run
 264          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 265          * we can use per-cpu packets.
 266          */
 267         static struct vmci_transport_packet pkt;
 268
 269         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 270                                                  size, mode, wait,
 271                                                  VSOCK_PROTO_INVALID, handle,
 272                                                  false);
 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 vmci_transport_packet *pkt;
 285         struct vsock_sock *vsk;
 286         int err;
 287
 288         vsk = vsock_sk(sk);
 289
 290         if (!vsock_addr_bound(&vsk->local_addr))
 291                 return -EINVAL;
 292
 293         if (!vsock_addr_bound(&vsk->remote_addr))
 294                 return -EINVAL;
 295
 296         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 297         if (!pkt)
 298                 return -ENOMEM;
 299
 300         err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
 301                                                 &vsk->remote_addr, type, size,
 302                                                 mode, wait, proto, handle,
 303                                                 true);
 304         kfree(pkt);
 305
 306         return err;
 307 }
 308
 309 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 310                                         struct sockaddr_vm *src,
 311                                         struct vmci_transport_packet *pkt)
 312 {
 313         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 314                 return 0;
 315         return vmci_transport_send_control_pkt_bh(
 316                                         dst, src,
 317                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 318                                         0, NULL, VMCI_INVALID_HANDLE);
 319 }
 320
 321 static int vmci_transport_send_reset(struct sock *sk,
 322                                      struct vmci_transport_packet *pkt)
 323 {
 324         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 325                 return 0;
 326         return vmci_transport_send_control_pkt(sk,
 327                                         VMCI_TRANSPORT_PACKET_TYPE_RST,
 328                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 329                                         VMCI_INVALID_HANDLE);
 330 }
 331
 332 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 333 {
 334         return vmci_transport_send_control_pkt(
 335                                         sk,
 336                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 337                                         size, 0, NULL,
 338                                         VSOCK_PROTO_INVALID,
 339                                         VMCI_INVALID_HANDLE);
 340 }
 341
 342 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 343                                           u16 version)
 344 {
 345         return vmci_transport_send_control_pkt(
 346                                         sk,
 347                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 348                                         size, 0, NULL, version,
 349                                         VMCI_INVALID_HANDLE);
 350 }
 351
 352 static int vmci_transport_send_qp_offer(struct sock *sk,
 353                                         struct vmci_handle handle)
 354 {
 355         return vmci_transport_send_control_pkt(
 356                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 357                                         0, NULL,
 358                                         VSOCK_PROTO_INVALID, handle);
 359 }
 360
 361 static int vmci_transport_send_attach(struct sock *sk,
 362                                       struct vmci_handle handle)
 363 {
 364         return vmci_transport_send_control_pkt(
 365                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 366                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 367                                         handle);
 368 }
 369
 370 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 371 {
 372         return vmci_transport_reply_control_pkt_fast(
 373                                                 pkt,
 374                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 375                                                 0, 0, NULL,
 376                                                 VMCI_INVALID_HANDLE);
 377 }
 378
 379 static int vmci_transport_send_invalid_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_INVALID,
 385                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 386 }
 387
 388 int vmci_transport_send_wrote_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_WROTE, 0,
 394                                         0, NULL, VMCI_INVALID_HANDLE);
 395 }
 396
 397 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 398                                 struct sockaddr_vm *src)
 399 {
 400         return vmci_transport_send_control_pkt_bh(
 401                                         dst, src,
 402                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 403                                         0, NULL, VMCI_INVALID_HANDLE);
 404 }
 405
 406 int vmci_transport_send_wrote(struct sock *sk)
 407 {
 408         return vmci_transport_send_control_pkt(
 409                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 410                                         0, NULL, VSOCK_PROTO_INVALID,
 411                                         VMCI_INVALID_HANDLE);
 412 }
 413
 414 int vmci_transport_send_read(struct sock *sk)
 415 {
 416         return vmci_transport_send_control_pkt(
 417                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 418                                         0, NULL, VSOCK_PROTO_INVALID,
 419                                         VMCI_INVALID_HANDLE);
 420 }
 421
 422 int vmci_transport_send_waiting_write(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_WRITE,
 427                                 0, 0, wait, VSOCK_PROTO_INVALID,
 428                                 VMCI_INVALID_HANDLE);
 429 }
 430
 431 int vmci_transport_send_waiting_read(struct sock *sk,
 432                                      struct vmci_transport_waiting_info *wait)
 433 {
 434         return vmci_transport_send_control_pkt(
 435                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 436                                 0, 0, wait, VSOCK_PROTO_INVALID,
 437                                 VMCI_INVALID_HANDLE);
 438 }
 439
 440 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 441 {
 442         return vmci_transport_send_control_pkt(
 443                                         &vsk->sk,
 444                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 445                                         0, mode, NULL,
 446                                         VSOCK_PROTO_INVALID,
 447                                         VMCI_INVALID_HANDLE);
 448 }
 449
 450 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 451 {
 452         return vmci_transport_send_control_pkt(sk,
 453                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 454                                         size, 0, NULL,
 455                                         VSOCK_PROTO_INVALID,
 456                                         VMCI_INVALID_HANDLE);
 457 }
 458
 459 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 460                                              u16 version)
 461 {
 462         return vmci_transport_send_control_pkt(
 463                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 464                                         size, 0, NULL, version,
 465                                         VMCI_INVALID_HANDLE);
 466 }
 467
 468 static struct sock *vmci_transport_get_pending(
 469                                         struct sock *listener,
 470                                         struct vmci_transport_packet *pkt)
 471 {
 472         struct vsock_sock *vlistener;
 473         struct vsock_sock *vpending;
 474         struct sock *pending;
 475         struct sockaddr_vm src;
 476
 477         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 478
 479         vlistener = vsock_sk(listener);
 480
 481         list_for_each_entry(vpending, &vlistener->pending_links,
 482                             pending_links) {
 483                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 484                     pkt->dst_port == vpending->local_addr.svm_port) {
 485                         pending = sk_vsock(vpending);
 486                         sock_hold(pending);
 487                         goto found;
 488                 }
 489         }
 490
 491         pending = NULL;
 492 found:
 493         return pending;
 494
 495 }
 496
 497 static void vmci_transport_release_pending(struct sock *pending)
 498 {
 499         sock_put(pending);
 500 }
 501
 502 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 503  * trusted sockets 2) sockets from applications running as the same user as the
 504  * VM (this is only true for the host side and only when using hosted products)
 505  */
 506
 507 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 508 {
 509         return vsock->trusted ||
 510                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 511 }
 512
 513 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 514  * only if it is trusted as described in vmci_transport_is_trusted.
 515  */
 516
 517 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 518 {
 519         if (VMADDR_CID_HYPERVISOR == peer_cid)
 520                 return true;
 521
 522         if (vsock->cached_peer != peer_cid) {
 523                 vsock->cached_peer = peer_cid;
 524                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 525                     (vmci_context_get_priv_flags(peer_cid) &
 526                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 527                         vsock->cached_peer_allow_dgram = false;
 528                 } else {
 529                         vsock->cached_peer_allow_dgram = true;
 530                 }
 531         }
 532
 533         return vsock->cached_peer_allow_dgram;
 534 }
 535
 536 static int
 537 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 538                                 struct vmci_handle *handle,
 539                                 u64 produce_size,
 540                                 u64 consume_size,
 541                                 u32 peer, u32 flags, bool trusted)
 542 {
 543         int err = 0;
 544
 545         if (trusted) {
 546                 /* Try to allocate our queue pair as trusted. This will only
 547                  * work if vsock is running in the host.
 548                  */
 549
 550                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 551                                        consume_size,
 552                                        peer, flags,
 553                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 554                 if (err != VMCI_ERROR_NO_ACCESS)
 555                         goto out;
 556
 557         }
 558
 559         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 560                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 561 out:
 562         if (err < 0) {
 563                 pr_err("Could not attach to queue pair with %d\n",
 564                        err);
 565                 err = vmci_transport_error_to_vsock_error(err);
 566         }
 567
 568         return err;
 569 }
 570
 571 static int
 572 vmci_transport_datagram_create_hnd(u32 resource_id,
 573                                    u32 flags,
 574                                    vmci_datagram_recv_cb recv_cb,
 575                                    void *client_data,
 576                                    struct vmci_handle *out_handle)
 577 {
 578         int err = 0;
 579
 580         /* Try to allocate our datagram handler as trusted. This will only work
 581          * if vsock is running in the host.
 582          */
 583
 584         err = vmci_datagram_create_handle_priv(resource_id, flags,
 585                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 586                                                recv_cb,
 587                                                client_data, out_handle);
 588
 589         if (err == VMCI_ERROR_NO_ACCESS)
 590                 err = vmci_datagram_create_handle(resource_id, flags,
 591                                                   recv_cb, client_data,
 592                                                   out_handle);
 593
 594         return err;
 595 }
 596
 597 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 598  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 599  * sleep it should defer that work to a work queue.
 600  */
 601
 602 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 603 {
 604         struct sock *sk;
 605         size_t size;
 606         struct sk_buff *skb;
 607         struct vsock_sock *vsk;
 608
 609         sk = (struct sock *)data;
 610
 611         /* This handler is privileged when this module is running on the host.
 612          * We will get datagrams from all endpoints (even VMs that are in a
 613          * restricted context). If we get one from a restricted context then
 614          * the destination socket must be trusted.
 615          *
 616          * NOTE: We access the socket struct without holding the lock here.
 617          * This is ok because the field we are interested is never modified
 618          * outside of the create and destruct socket functions.
 619          */
 620         vsk = vsock_sk(sk);
 621         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 622                 return VMCI_ERROR_NO_ACCESS;
 623
 624         size = VMCI_DG_SIZE(dg);
 625
 626         /* Attach the packet to the socket's receive queue as an sk_buff. */
 627         skb = alloc_skb(size, GFP_ATOMIC);
 628         if (!skb)
 629                 return VMCI_ERROR_NO_MEM;
 630
 631         /* sk_receive_skb() will do a sock_put(), so hold here. */
 632         sock_hold(sk);
 633         skb_put(skb, size);
 634         memcpy(skb->data, dg, size);
 635         sk_receive_skb(sk, skb, 0);
 636
 637         return VMCI_SUCCESS;
 638 }
 639
 640 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 641 {
 642         static const u32 non_socket_contexts[] = {
 643                 VMADDR_CID_RESERVED,
 644         };
 645         int i;
 646
 647         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 648
 649         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 650                 if (cid == non_socket_contexts[i])
 651                         return false;
 652         }
 653
 654         return true;
 655 }
 656
 657 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 658  * interrupt fires.  This is run in bottom-half context but it defers most of
 659  * its work to the packet handling work queue.
 660  */
 661
 662 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 663 {
 664         struct sock *sk;
 665         struct sockaddr_vm dst;
 666         struct sockaddr_vm src;
 667         struct vmci_transport_packet *pkt;
 668         struct vsock_sock *vsk;
 669         bool bh_process_pkt;
 670         int err;
 671
 672         sk = NULL;
 673         err = VMCI_SUCCESS;
 674         bh_process_pkt = false;
 675
 676         /* Ignore incoming packets from contexts without sockets, or resources
 677          * that aren't vsock implementations.
 678          */
 679
 680         if (!vmci_transport_stream_allow(dg->src.context, -1)
 681             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 682                 return VMCI_ERROR_NO_ACCESS;
 683
 684         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 685                 /* Drop datagrams that do not contain full VSock packets. */
 686                 return VMCI_ERROR_INVALID_ARGS;
 687
 688         pkt = (struct vmci_transport_packet *)dg;
 689
 690         /* Find the socket that should handle this packet.  First we look for a
 691          * connected socket and if there is none we look for a socket bound to
 692          * the destintation address.
 693          */
 694         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 695         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 696
 697         sk = vsock_find_connected_socket(&src, &dst);
 698         if (!sk) {
 699                 sk = vsock_find_bound_socket(&dst);
 700                 if (!sk) {
 701                         /* We could not find a socket for this specified
 702                          * address.  If this packet is a RST, we just drop it.
 703                          * If it is another packet, we send a RST.  Note that
 704                          * we do not send a RST reply to RSTs so that we do not
 705                          * continually send RSTs between two endpoints.
 706                          *
 707                          * Note that since this is a reply, dst is src and src
 708                          * is dst.
 709                          */
 710                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 711                                 pr_err("unable to send reset\n");
 712
 713                         err = VMCI_ERROR_NOT_FOUND;
 714                         goto out;
 715                 }
 716         }
 717
 718         /* If the received packet type is beyond all types known to this
 719          * implementation, reply with an invalid message.  Hopefully this will
 720          * help when implementing backwards compatibility in the future.
 721          */
 722         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 723                 vmci_transport_send_invalid_bh(&dst, &src);
 724                 err = VMCI_ERROR_INVALID_ARGS;
 725                 goto out;
 726         }
 727
 728         /* This handler is privileged when this module is running on the host.
 729          * We will get datagram connect requests from all endpoints (even VMs
 730          * that are in a restricted context). If we get one from a restricted
 731          * context then the destination socket must be trusted.
 732          *
 733          * NOTE: We access the socket struct without holding the lock here.
 734          * This is ok because the field we are interested is never modified
 735          * outside of the create and destruct socket functions.
 736          */
 737         vsk = vsock_sk(sk);
 738         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 739                 err = VMCI_ERROR_NO_ACCESS;
 740                 goto out;
 741         }
 742
 743         /* We do most everything in a work queue, but let's fast path the
 744          * notification of reads and writes to help data transfer performance.
 745          * We can only do this if there is no process context code executing
 746          * for this socket since that may change the state.
 747          */
 748         bh_lock_sock(sk);
 749
 750         if (!sock_owned_by_user(sk)) {
 751                 /* The local context ID may be out of date, update it. */
 752                 vsk->local_addr.svm_cid = dst.svm_cid;
 753
 754                 if (sk->sk_state == SS_CONNECTED)
 755                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 756                                         sk, pkt, true, &dst, &src,
 757                                         &bh_process_pkt);
 758         }
 759
 760         bh_unlock_sock(sk);
 761
 762         if (!bh_process_pkt) {
 763                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 764
 765                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 766                 if (!recv_pkt_info) {
 767                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 768                                 pr_err("unable to send reset\n");
 769
 770                         err = VMCI_ERROR_NO_MEM;
 771                         goto out;
 772                 }
 773
 774                 recv_pkt_info->sk = sk;
 775                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 776                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 777
 778                 schedule_work(&recv_pkt_info->work);
 779                 /* Clear sk so that the reference count incremented by one of
 780                  * the Find functions above is not decremented below.  We need
 781                  * that reference count for the packet handler we've scheduled
 782                  * to run.
 783                  */
 784                 sk = NULL;
 785         }
 786
 787 out:
 788         if (sk)
 789                 sock_put(sk);
 790
 791         return err;
 792 }
 793
 794 static void vmci_transport_handle_detach(struct sock *sk)
 795 {
 796         struct vsock_sock *vsk;
 797
 798         vsk = vsock_sk(sk);
 799         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 800                 sock_set_flag(sk, SOCK_DONE);
 801
 802                 /* On a detach the peer will not be sending or receiving
 803                  * anymore.
 804                  */
 805                 vsk->peer_shutdown = SHUTDOWN_MASK;
 806
 807                 /* We should not be sending anymore since the peer won't be
 808                  * there to receive, but we can still receive if there is data
 809                  * left in our consume queue.
 810                  */
 811                 if (vsock_stream_has_data(vsk) <= 0) {
 812                         if (sk->sk_state == SS_CONNECTING) {
 813                                 /* The peer may detach from a queue pair while
 814                                  * we are still in the connecting state, i.e.,
 815                                  * if the peer VM is killed after attaching to
 816                                  * a queue pair, but before we complete the
 817                                  * handshake. In that case, we treat the detach
 818                                  * event like a reset.
 819                                  */
 820
 821                                 sk->sk_state = SS_UNCONNECTED;
 822                                 sk->sk_err = ECONNRESET;
 823                                 sk->sk_error_report(sk);
 824                                 return;
 825                         }
 826                         sk->sk_state = SS_UNCONNECTED;
 827                 }
 828                 sk->sk_state_change(sk);
 829         }
 830 }
 831
 832 static void vmci_transport_peer_detach_cb(u32 sub_id,
 833                                           const struct vmci_event_data *e_data,
 834                                           void *client_data)
 835 {
 836         struct vmci_transport *trans = client_data;
 837         const struct vmci_event_payload_qp *e_payload;
 838
 839         e_payload = vmci_event_data_const_payload(e_data);
 840
 841         /* XXX This is lame, we should provide a way to lookup sockets by
 842          * qp_handle.
 843          */
 844         if (vmci_handle_is_invalid(e_payload->handle) ||
 845             vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 846                 return;
 847
 848         /* We don't ask for delayed CBs when we subscribe to this event (we
 849          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 850          * guarantees in that case about what context we might be running in,
 851          * so it could be BH or process, blockable or non-blockable.  So we
 852          * need to account for all possible contexts here.
 853          */
 854         spin_lock_bh(&trans->lock);
 855         if (!trans->sk)
 856                 goto out;
 857
 858         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 859          * where trans->sk isn't locked.
 860          */
 861         bh_lock_sock(trans->sk);
 862
 863         vmci_transport_handle_detach(trans->sk);
 864
 865         bh_unlock_sock(trans->sk);
 866  out:
 867         spin_unlock_bh(&trans->lock);
 868 }
 869
 870 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 871                                          const struct vmci_event_data *e_data,
 872                                          void *client_data)
 873 {
 874         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 875 }
 876
 877 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 878 {
 879         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 880         struct vmci_transport_packet *pkt;
 881         struct sock *sk;
 882
 883         recv_pkt_info =
 884                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 885         sk = recv_pkt_info->sk;
 886         pkt = &recv_pkt_info->pkt;
 887
 888         lock_sock(sk);
 889
 890         /* The local context ID may be out of date. */
 891         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 892
 893         switch (sk->sk_state) {
 894         case VSOCK_SS_LISTEN:
 895                 vmci_transport_recv_listen(sk, pkt);
 896                 break;
 897         case SS_CONNECTING:
 898                 /* Processing of pending connections for servers goes through
 899                  * the listening socket, so see vmci_transport_recv_listen()
 900                  * for that path.
 901                  */
 902                 vmci_transport_recv_connecting_client(sk, pkt);
 903                 break;
 904         case SS_CONNECTED:
 905                 vmci_transport_recv_connected(sk, pkt);
 906                 break;
 907         default:
 908                 /* Because this function does not run in the same context as
 909                  * vmci_transport_recv_stream_cb it is possible that the
 910                  * socket has closed. We need to let the other side know or it
 911                  * could be sitting in a connect and hang forever. Send a
 912                  * reset to prevent that.
 913                  */
 914                 vmci_transport_send_reset(sk, pkt);
 915                 break;
 916         }
 917
 918         release_sock(sk);
 919         kfree(recv_pkt_info);
 920         /* Release reference obtained in the stream callback when we fetched
 921          * this socket out of the bound or connected list.
 922          */
 923         sock_put(sk);
 924 }
 925
 926 static int vmci_transport_recv_listen(struct sock *sk,
 927                                       struct vmci_transport_packet *pkt)
 928 {
 929         struct sock *pending;
 930         struct vsock_sock *vpending;
 931         int err;
 932         u64 qp_size;
 933         bool old_request = false;
 934         bool old_pkt_proto = false;
 935
 936         err = 0;
 937
 938         /* Because we are in the listen state, we could be receiving a packet
 939          * for ourself or any previous connection requests that we received.
 940          * If it's the latter, we try to find a socket in our list of pending
 941          * connections and, if we do, call the appropriate handler for the
 942          * state that that socket is in.  Otherwise we try to service the
 943          * connection request.
 944          */
 945         pending = vmci_transport_get_pending(sk, pkt);
 946         if (pending) {
 947                 lock_sock(pending);
 948
 949                 /* The local context ID may be out of date. */
 950                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 951
 952                 switch (pending->sk_state) {
 953                 case SS_CONNECTING:
 954                         err = vmci_transport_recv_connecting_server(sk,
 955                                                                     pending,
 956                                                                     pkt);
 957                         break;
 958                 default:
 959                         vmci_transport_send_reset(pending, pkt);
 960                         err = -EINVAL;
 961                 }
 962
 963                 if (err < 0)
 964                         vsock_remove_pending(sk, pending);
 965
 966                 release_sock(pending);
 967                 vmci_transport_release_pending(pending);
 968
 969                 return err;
 970         }
 971
 972         /* The listen state only accepts connection requests.  Reply with a
 973          * reset unless we received a reset.
 974          */
 975
 976         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 977               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 978                 vmci_transport_reply_reset(pkt);
 979                 return -EINVAL;
 980         }
 981
 982         if (pkt->u.size == 0) {
 983                 vmci_transport_reply_reset(pkt);
 984                 return -EINVAL;
 985         }
 986
 987         /* If this socket can't accommodate this connection request, we send a
 988          * reset.  Otherwise we create and initialize a child socket and reply
 989          * with a connection negotiation.
 990          */
 991         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
 992                 vmci_transport_reply_reset(pkt);
 993                 return -ECONNREFUSED;
 994         }
 995
 996         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
 997                                  sk->sk_type, 0);
 998         if (!pending) {
 999                 vmci_transport_send_reset(sk, pkt);
1000                 return -ENOMEM;
1001         }
1002
1003         vpending = vsock_sk(pending);
1004
1005         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1006                         pkt->dst_port);
1007         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1008                         pkt->src_port);
1009
1010         /* If the proposed size fits within our min/max, accept it. Otherwise
1011          * propose our own size.
1012          */
1013         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1014             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1015                 qp_size = pkt->u.size;
1016         } else {
1017                 qp_size = vmci_trans(vpending)->queue_pair_size;
1018         }
1019
1020         /* Figure out if we are using old or new requests based on the
1021          * overrides pkt types sent by our peer.
1022          */
1023         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1024                 old_request = old_pkt_proto;
1025         } else {
1026                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1027                         old_request = true;
1028                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1029                         old_request = false;
1030
1031         }
1032
1033         if (old_request) {
1034                 /* Handle a REQUEST (or override) */
1035                 u16 version = VSOCK_PROTO_INVALID;
1036                 if (vmci_transport_proto_to_notify_struct(
1037                         pending, &version, true))
1038                         err = vmci_transport_send_negotiate(pending, qp_size);
1039                 else
1040                         err = -EINVAL;
1041
1042         } else {
1043                 /* Handle a REQUEST2 (or override) */
1044                 int proto_int = pkt->proto;
1045                 int pos;
1046                 u16 active_proto_version = 0;
1047
1048                 /* The list of possible protocols is the intersection of all
1049                  * protocols the client supports ... plus all the protocols we
1050                  * support.
1051                  */
1052                 proto_int &= vmci_transport_new_proto_supported_versions();
1053
1054                 /* We choose the highest possible protocol version and use that
1055                  * one.
1056                  */
1057                 pos = fls(proto_int);
1058                 if (pos) {
1059                         active_proto_version = (1 << (pos - 1));
1060                         if (vmci_transport_proto_to_notify_struct(
1061                                 pending, &active_proto_version, false))
1062                                 err = vmci_transport_send_negotiate2(pending,
1063                                                         qp_size,
1064                                                         active_proto_version);
1065                         else
1066                                 err = -EINVAL;
1067
1068                 } else {
1069                         err = -EINVAL;
1070                 }
1071         }
1072
1073         if (err < 0) {
1074                 vmci_transport_send_reset(sk, pkt);
1075                 sock_put(pending);
1076                 err = vmci_transport_error_to_vsock_error(err);
1077                 goto out;
1078         }
1079
1080         vsock_add_pending(sk, pending);
1081         sk->sk_ack_backlog++;
1082
1083         pending->sk_state = SS_CONNECTING;
1084         vmci_trans(vpending)->produce_size =
1085                 vmci_trans(vpending)->consume_size = qp_size;
1086         vmci_trans(vpending)->queue_pair_size = qp_size;
1087
1088         vmci_trans(vpending)->notify_ops->process_request(pending);
1089
1090         /* We might never receive another message for this socket and it's not
1091          * connected to any process, so we have to ensure it gets cleaned up
1092          * ourself.  Our delayed work function will take care of that.  Note
1093          * that we do not ever cancel this function since we have few
1094          * guarantees about its state when calling cancel_delayed_work().
1095          * Instead we hold a reference on the socket for that function and make
1096          * it capable of handling cases where it needs to do nothing but
1097          * release that reference.
1098          */
1099         vpending->listener = sk;
1100         sock_hold(sk);
1101         sock_hold(pending);
1102         INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1103         schedule_delayed_work(&vpending->dwork, HZ);
1104
1105 out:
1106         return err;
1107 }
1108
1109 static int
1110 vmci_transport_recv_connecting_server(struct sock *listener,
1111                                       struct sock *pending,
1112                                       struct vmci_transport_packet *pkt)
1113 {
1114         struct vsock_sock *vpending;
1115         struct vmci_handle handle;
1116         struct vmci_qp *qpair;
1117         bool is_local;
1118         u32 flags;
1119         u32 detach_sub_id;
1120         int err;
1121         int skerr;
1122
1123         vpending = vsock_sk(pending);
1124         detach_sub_id = VMCI_INVALID_ID;
1125
1126         switch (pkt->type) {
1127         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1128                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1129                         vmci_transport_send_reset(pending, pkt);
1130                         skerr = EPROTO;
1131                         err = -EINVAL;
1132                         goto destroy;
1133                 }
1134                 break;
1135         default:
1136                 /* Close and cleanup the connection. */
1137                 vmci_transport_send_reset(pending, pkt);
1138                 skerr = EPROTO;
1139                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1140                 goto destroy;
1141         }
1142
1143         /* In order to complete the connection we need to attach to the offered
1144          * queue pair and send an attach notification.  We also subscribe to the
1145          * detach event so we know when our peer goes away, and we do that
1146          * before attaching so we don't miss an event.  If all this succeeds,
1147          * we update our state and wakeup anything waiting in accept() for a
1148          * connection.
1149          */
1150
1151         /* We don't care about attach since we ensure the other side has
1152          * attached by specifying the ATTACH_ONLY flag below.
1153          */
1154         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1155                                    vmci_transport_peer_detach_cb,
1156                                    vmci_trans(vpending), &detach_sub_id);
1157         if (err < VMCI_SUCCESS) {
1158                 vmci_transport_send_reset(pending, pkt);
1159                 err = vmci_transport_error_to_vsock_error(err);
1160                 skerr = -err;
1161                 goto destroy;
1162         }
1163
1164         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1165
1166         /* Now attach to the queue pair the client created. */
1167         handle = pkt->u.handle;
1168
1169         /* vpending->local_addr always has a context id so we do not need to
1170          * worry about VMADDR_CID_ANY in this case.
1171          */
1172         is_local =
1173             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1174         flags = VMCI_QPFLAG_ATTACH_ONLY;
1175         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1176
1177         err = vmci_transport_queue_pair_alloc(
1178                                         &qpair,
1179                                         &handle,
1180                                         vmci_trans(vpending)->produce_size,
1181                                         vmci_trans(vpending)->consume_size,
1182                                         pkt->dg.src.context,
1183                                         flags,
1184                                         vmci_transport_is_trusted(
1185                                                 vpending,
1186                                                 vpending->remote_addr.svm_cid));
1187         if (err < 0) {
1188                 vmci_transport_send_reset(pending, pkt);
1189                 skerr = -err;
1190                 goto destroy;
1191         }
1192
1193         vmci_trans(vpending)->qp_handle = handle;
1194         vmci_trans(vpending)->qpair = qpair;
1195
1196         /* When we send the attach message, we must be ready to handle incoming
1197          * control messages on the newly connected socket. So we move the
1198          * pending socket to the connected state before sending the attach
1199          * message. Otherwise, an incoming packet triggered by the attach being
1200          * received by the peer may be processed concurrently with what happens
1201          * below after sending the attach message, and that incoming packet
1202          * will find the listening socket instead of the (currently) pending
1203          * socket. Note that enqueueing the socket increments the reference
1204          * count, so even if a reset comes before the connection is accepted,
1205          * the socket will be valid until it is removed from the queue.
1206          *
1207          * If we fail sending the attach below, we remove the socket from the
1208          * connected list and move the socket to SS_UNCONNECTED before
1209          * releasing the lock, so a pending slow path processing of an incoming
1210          * packet will not see the socket in the connected state in that case.
1211          */
1212         pending->sk_state = SS_CONNECTED;
1213
1214         vsock_insert_connected(vpending);
1215
1216         /* Notify our peer of our attach. */
1217         err = vmci_transport_send_attach(pending, handle);
1218         if (err < 0) {
1219                 vsock_remove_connected(vpending);
1220                 pr_err("Could not send attach\n");
1221                 vmci_transport_send_reset(pending, pkt);
1222                 err = vmci_transport_error_to_vsock_error(err);
1223                 skerr = -err;
1224                 goto destroy;
1225         }
1226
1227         /* We have a connection. Move the now connected socket from the
1228          * listener's pending list to the accept queue so callers of accept()
1229          * can find it.
1230          */
1231         vsock_remove_pending(listener, pending);
1232         vsock_enqueue_accept(listener, pending);
1233
1234         /* Callers of accept() will be be waiting on the listening socket, not
1235          * the pending socket.
1236          */
1237         listener->sk_data_ready(listener);
1238
1239         return 0;
1240
1241 destroy:
1242         pending->sk_err = skerr;
1243         pending->sk_state = SS_UNCONNECTED;
1244         /* As long as we drop our reference, all necessary cleanup will handle
1245          * when the cleanup function drops its reference and our destruct
1246          * implementation is called.  Note that since the listen handler will
1247          * remove pending from the pending list upon our failure, the cleanup
1248          * function won't drop the additional reference, which is why we do it
1249          * here.
1250          */
1251         sock_put(pending);
1252
1253         return err;
1254 }
1255
1256 static int
1257 vmci_transport_recv_connecting_client(struct sock *sk,
1258                                       struct vmci_transport_packet *pkt)
1259 {
1260         struct vsock_sock *vsk;
1261         int err;
1262         int skerr;
1263
1264         vsk = vsock_sk(sk);
1265
1266         switch (pkt->type) {
1267         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1268                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1269                     !vmci_handle_is_equal(pkt->u.handle,
1270                                           vmci_trans(vsk)->qp_handle)) {
1271                         skerr = EPROTO;
1272                         err = -EINVAL;
1273                         goto destroy;
1274                 }
1275
1276                 /* Signify the socket is connected and wakeup the waiter in
1277                  * connect(). Also place the socket in the connected table for
1278                  * accounting (it can already be found since it's in the bound
1279                  * table).
1280                  */
1281                 sk->sk_state = SS_CONNECTED;
1282                 sk->sk_socket->state = SS_CONNECTED;
1283                 vsock_insert_connected(vsk);
1284                 sk->sk_state_change(sk);
1285
1286                 break;
1287         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1288         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1289                 if (pkt->u.size == 0
1290                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1291                     || pkt->src_port != vsk->remote_addr.svm_port
1292                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1293                     || vmci_trans(vsk)->qpair
1294                     || vmci_trans(vsk)->produce_size != 0
1295                     || vmci_trans(vsk)->consume_size != 0
1296                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1297                         skerr = EPROTO;
1298                         err = -EINVAL;
1299
1300                         goto destroy;
1301                 }
1302
1303                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1304                 if (err) {
1305                         skerr = -err;
1306                         goto destroy;
1307                 }
1308
1309                 break;
1310         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1311                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1312                 if (err) {
1313                         skerr = -err;
1314                         goto destroy;
1315                 }
1316
1317                 break;
1318         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1319                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1320                  * continue processing here after they sent an INVALID packet.
1321                  * This meant that we got a RST after the INVALID. We ignore a
1322                  * RST after an INVALID. The common code doesn't send the RST
1323                  * ... so we can hang if an old version of the common code
1324                  * fails between getting a REQUEST and sending an OFFER back.
1325                  * Not much we can do about it... except hope that it doesn't
1326                  * happen.
1327                  */
1328                 if (vsk->ignore_connecting_rst) {
1329                         vsk->ignore_connecting_rst = false;
1330                 } else {
1331                         skerr = ECONNRESET;
1332                         err = 0;
1333                         goto destroy;
1334                 }
1335
1336                 break;
1337         default:
1338                 /* Close and cleanup the connection. */
1339                 skerr = EPROTO;
1340                 err = -EINVAL;
1341                 goto destroy;
1342         }
1343
1344         return 0;
1345
1346 destroy:
1347         vmci_transport_send_reset(sk, pkt);
1348
1349         sk->sk_state = SS_UNCONNECTED;
1350         sk->sk_err = skerr;
1351         sk->sk_error_report(sk);
1352         return err;
1353 }
1354
1355 static int vmci_transport_recv_connecting_client_negotiate(
1356                                         struct sock *sk,
1357                                         struct vmci_transport_packet *pkt)
1358 {
1359         int err;
1360         struct vsock_sock *vsk;
1361         struct vmci_handle handle;
1362         struct vmci_qp *qpair;
1363         u32 detach_sub_id;
1364         bool is_local;
1365         u32 flags;
1366         bool old_proto = true;
1367         bool old_pkt_proto;
1368         u16 version;
1369
1370         vsk = vsock_sk(sk);
1371         handle = VMCI_INVALID_HANDLE;
1372         detach_sub_id = VMCI_INVALID_ID;
1373
1374         /* If we have gotten here then we should be past the point where old
1375          * linux vsock could have sent the bogus rst.
1376          */
1377         vsk->sent_request = false;
1378         vsk->ignore_connecting_rst = false;
1379
1380         /* Verify that we're OK with the proposed queue pair size */
1381         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1382             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1383                 err = -EINVAL;
1384                 goto destroy;
1385         }
1386
1387         /* At this point we know the CID the peer is using to talk to us. */
1388
1389         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1390                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1391
1392         /* Setup the notify ops to be the highest supported version that both
1393          * the server and the client support.
1394          */
1395
1396         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1397                 old_proto = old_pkt_proto;
1398         } else {
1399                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1400                         old_proto = true;
1401                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1402                         old_proto = false;
1403
1404         }
1405
1406         if (old_proto)
1407                 version = VSOCK_PROTO_INVALID;
1408         else
1409                 version = pkt->proto;
1410
1411         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1412                 err = -EINVAL;
1413                 goto destroy;
1414         }
1415
1416         /* Subscribe to detach events first.
1417          *
1418          * XXX We attach once for each queue pair created for now so it is easy
1419          * to find the socket (it's provided), but later we should only
1420          * subscribe once and add a way to lookup sockets by queue pair handle.
1421          */
1422         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1423                                    vmci_transport_peer_detach_cb,
1424                                    vmci_trans(vsk), &detach_sub_id);
1425         if (err < VMCI_SUCCESS) {
1426                 err = vmci_transport_error_to_vsock_error(err);
1427                 goto destroy;
1428         }
1429
1430         /* Make VMCI select the handle for us. */
1431         handle = VMCI_INVALID_HANDLE;
1432         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1433         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1434
1435         err = vmci_transport_queue_pair_alloc(&qpair,
1436                                               &handle,
1437                                               pkt->u.size,
1438                                               pkt->u.size,
1439                                               vsk->remote_addr.svm_cid,
1440                                               flags,
1441                                               vmci_transport_is_trusted(
1442                                                   vsk,
1443                                                   vsk->
1444                                                   remote_addr.svm_cid));
1445         if (err < 0)
1446                 goto destroy;
1447
1448         err = vmci_transport_send_qp_offer(sk, handle);
1449         if (err < 0) {
1450                 err = vmci_transport_error_to_vsock_error(err);
1451                 goto destroy;
1452         }
1453
1454         vmci_trans(vsk)->qp_handle = handle;
1455         vmci_trans(vsk)->qpair = qpair;
1456
1457         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1458                 pkt->u.size;
1459
1460         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1461
1462         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1463
1464         return 0;
1465
1466 destroy:
1467         if (detach_sub_id != VMCI_INVALID_ID)
1468                 vmci_event_unsubscribe(detach_sub_id);
1469
1470         if (!vmci_handle_is_invalid(handle))
1471                 vmci_qpair_detach(&qpair);
1472
1473         return err;
1474 }
1475
1476 static int
1477 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1478                                               struct vmci_transport_packet *pkt)
1479 {
1480         int err = 0;
1481         struct vsock_sock *vsk = vsock_sk(sk);
1482
1483         if (vsk->sent_request) {
1484                 vsk->sent_request = false;
1485                 vsk->ignore_connecting_rst = true;
1486
1487                 err = vmci_transport_send_conn_request(
1488                         sk, vmci_trans(vsk)->queue_pair_size);
1489                 if (err < 0)
1490                         err = vmci_transport_error_to_vsock_error(err);
1491                 else
1492                         err = 0;
1493
1494         }
1495
1496         return err;
1497 }
1498
1499 static int vmci_transport_recv_connected(struct sock *sk,
1500                                          struct vmci_transport_packet *pkt)
1501 {
1502         struct vsock_sock *vsk;
1503         bool pkt_processed = false;
1504
1505         /* In cases where we are closing the connection, it's sufficient to
1506          * mark the state change (and maybe error) and wake up any waiting
1507          * threads. Since this is a connected socket, it's owned by a user
1508          * process and will be cleaned up when the failure is passed back on
1509          * the current or next system call.  Our system call implementations
1510          * must therefore check for error and state changes on entry and when
1511          * being awoken.
1512          */
1513         switch (pkt->type) {
1514         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1515                 if (pkt->u.mode) {
1516                         vsk = vsock_sk(sk);
1517
1518                         vsk->peer_shutdown |= pkt->u.mode;
1519                         sk->sk_state_change(sk);
1520                 }
1521                 break;
1522
1523         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1524                 vsk = vsock_sk(sk);
1525                 /* It is possible that we sent our peer a message (e.g a
1526                  * WAITING_READ) right before we got notified that the peer had
1527                  * detached. If that happens then we can get a RST pkt back
1528                  * from our peer even though there is data available for us to
1529                  * read. In that case, don't shutdown the socket completely but
1530                  * instead allow the local client to finish reading data off
1531                  * the queuepair. Always treat a RST pkt in connected mode like
1532                  * a clean shutdown.
1533                  */
1534                 sock_set_flag(sk, SOCK_DONE);
1535                 vsk->peer_shutdown = SHUTDOWN_MASK;
1536                 if (vsock_stream_has_data(vsk) <= 0)
1537                         sk->sk_state = SS_DISCONNECTING;
1538
1539                 sk->sk_state_change(sk);
1540                 break;
1541
1542         default:
1543                 vsk = vsock_sk(sk);
1544                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1545                                 sk, pkt, false, NULL, NULL,
1546                                 &pkt_processed);
1547                 if (!pkt_processed)
1548                         return -EINVAL;
1549
1550                 break;
1551         }
1552
1553         return 0;
1554 }
1555
1556 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1557                                       struct vsock_sock *psk)
1558 {
1559         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1560         if (!vsk->trans)
1561                 return -ENOMEM;
1562
1563         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1564         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1565         vmci_trans(vsk)->qpair = NULL;
1566         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1567         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1568         vmci_trans(vsk)->notify_ops = NULL;
1569         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1570         vmci_trans(vsk)->sk = &vsk->sk;
1571         spin_lock_init(&vmci_trans(vsk)->lock);
1572         if (psk) {
1573                 vmci_trans(vsk)->queue_pair_size =
1574                         vmci_trans(psk)->queue_pair_size;
1575                 vmci_trans(vsk)->queue_pair_min_size =
1576                         vmci_trans(psk)->queue_pair_min_size;
1577                 vmci_trans(vsk)->queue_pair_max_size =
1578                         vmci_trans(psk)->queue_pair_max_size;
1579         } else {
1580                 vmci_trans(vsk)->queue_pair_size =
1581                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1582                 vmci_trans(vsk)->queue_pair_min_size =
1583                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1584                 vmci_trans(vsk)->queue_pair_max_size =
1585                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1586         }
1587
1588         return 0;
1589 }
1590
1591 static void vmci_transport_free_resources(struct list_head *transport_list)
1592 {
1593         while (!list_empty(transport_list)) {
1594                 struct vmci_transport *transport =
1595                     list_first_entry(transport_list, struct vmci_transport,
1596                                      elem);
1597                 list_del(&transport->elem);
1598
1599                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1600                         vmci_event_unsubscribe(transport->detach_sub_id);
1601                         transport->detach_sub_id = VMCI_INVALID_ID;
1602                 }
1603
1604                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1605                         vmci_qpair_detach(&transport->qpair);
1606                         transport->qp_handle = VMCI_INVALID_HANDLE;
1607                         transport->produce_size = 0;
1608                         transport->consume_size = 0;
1609                 }
1610
1611                 kfree(transport);
1612         }
1613 }
1614
1615 static void vmci_transport_cleanup(struct work_struct *work)
1616 {
1617         LIST_HEAD(pending);
1618
1619         spin_lock_bh(&vmci_transport_cleanup_lock);
1620         list_replace_init(&vmci_transport_cleanup_list, &pending);
1621         spin_unlock_bh(&vmci_transport_cleanup_lock);
1622         vmci_transport_free_resources(&pending);
1623 }
1624
1625 static void vmci_transport_destruct(struct vsock_sock *vsk)
1626 {
1627         /* Ensure that the detach callback doesn't use the sk/vsk
1628          * we are about to destruct.
1629          */
1630         spin_lock_bh(&vmci_trans(vsk)->lock);
1631         vmci_trans(vsk)->sk = NULL;
1632         spin_unlock_bh(&vmci_trans(vsk)->lock);
1633
1634         if (vmci_trans(vsk)->notify_ops)
1635                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1636
1637         spin_lock_bh(&vmci_transport_cleanup_lock);
1638         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1639         spin_unlock_bh(&vmci_transport_cleanup_lock);
1640         schedule_work(&vmci_transport_cleanup_work);
1641
1642         vsk->trans = NULL;
1643 }
1644
1645 static void vmci_transport_release(struct vsock_sock *vsk)
1646 {
1647         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1648                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1649                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1650         }
1651 }
1652
1653 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1654                                      struct sockaddr_vm *addr)
1655 {
1656         u32 port;
1657         u32 flags;
1658         int err;
1659
1660         /* VMCI will select a resource ID for us if we provide
1661          * VMCI_INVALID_ID.
1662          */
1663         port = addr->svm_port == VMADDR_PORT_ANY ?
1664                         VMCI_INVALID_ID : addr->svm_port;
1665
1666         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1667                 return -EACCES;
1668
1669         flags = addr->svm_cid == VMADDR_CID_ANY ?
1670                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1671
1672         err = vmci_transport_datagram_create_hnd(port, flags,
1673                                                  vmci_transport_recv_dgram_cb,
1674                                                  &vsk->sk,
1675                                                  &vmci_trans(vsk)->dg_handle);
1676         if (err < VMCI_SUCCESS)
1677                 return vmci_transport_error_to_vsock_error(err);
1678         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1679                         vmci_trans(vsk)->dg_handle.resource);
1680
1681         return 0;
1682 }
1683
1684 static int vmci_transport_dgram_enqueue(
1685         struct vsock_sock *vsk,
1686         struct sockaddr_vm *remote_addr,
1687         struct msghdr *msg,
1688         size_t len)
1689 {
1690         int err;
1691         struct vmci_datagram *dg;
1692
1693         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1694                 return -EMSGSIZE;
1695
1696         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1697                 return -EPERM;
1698
1699         /* Allocate a buffer for the user's message and our packet header. */
1700         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1701         if (!dg)
1702                 return -ENOMEM;
1703
1704         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1705
1706         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1707                                    remote_addr->svm_port);
1708         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1709                                    vsk->local_addr.svm_port);
1710         dg->payload_size = len;
1711
1712         err = vmci_datagram_send(dg);
1713         kfree(dg);
1714         if (err < 0)
1715                 return vmci_transport_error_to_vsock_error(err);
1716
1717         return err - sizeof(*dg);
1718 }
1719
1720 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1721                                         struct msghdr *msg, size_t len,
1722                                         int flags)
1723 {
1724         int err;
1725         int noblock;
1726         struct vmci_datagram *dg;
1727         size_t payload_len;
1728         struct sk_buff *skb;
1729
1730         noblock = flags & MSG_DONTWAIT;
1731
1732         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1733                 return -EOPNOTSUPP;
1734
1735         /* Retrieve the head sk_buff from the socket's receive queue. */
1736         err = 0;
1737         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1738         if (err)
1739                 return err;
1740
1741         if (!skb)
1742                 return -EAGAIN;
1743
1744         dg = (struct vmci_datagram *)skb->data;
1745         if (!dg)
1746                 /* err is 0, meaning we read zero bytes. */
1747                 goto out;
1748
1749         payload_len = dg->payload_size;
1750         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1751         if (payload_len != skb->len - sizeof(*dg)) {
1752                 err = -EINVAL;
1753                 goto out;
1754         }
1755
1756         if (payload_len > len) {
1757                 payload_len = len;
1758                 msg->msg_flags |= MSG_TRUNC;
1759         }
1760
1761         /* Place the datagram payload in the user's iovec. */
1762         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1763         if (err)
1764                 goto out;
1765
1766         if (msg->msg_name) {
1767                 /* Provide the address of the sender. */
1768                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1769                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1770                 msg->msg_namelen = sizeof(*vm_addr);
1771         }
1772         err = payload_len;
1773
1774 out:
1775         skb_free_datagram(&vsk->sk, skb);
1776         return err;
1777 }
1778
1779 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1780 {
1781         if (cid == VMADDR_CID_HYPERVISOR) {
1782                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1783                  * state and are allowed.
1784                  */
1785                 return port == VMCI_UNITY_PBRPC_REGISTER;
1786         }
1787
1788         return true;
1789 }
1790
1791 static int vmci_transport_connect(struct vsock_sock *vsk)
1792 {
1793         int err;
1794         bool old_pkt_proto = false;
1795         struct sock *sk = &vsk->sk;
1796
1797         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1798                 old_pkt_proto) {
1799                 err = vmci_transport_send_conn_request(
1800                         sk, vmci_trans(vsk)->queue_pair_size);
1801                 if (err < 0) {
1802                         sk->sk_state = SS_UNCONNECTED;
1803                         return err;
1804                 }
1805         } else {
1806                 int supported_proto_versions =
1807                         vmci_transport_new_proto_supported_versions();
1808                 err = vmci_transport_send_conn_request2(
1809                                 sk, vmci_trans(vsk)->queue_pair_size,
1810                                 supported_proto_versions);
1811                 if (err < 0) {
1812                         sk->sk_state = SS_UNCONNECTED;
1813                         return err;
1814                 }
1815
1816                 vsk->sent_request = true;
1817         }
1818
1819         return err;
1820 }
1821
1822 static ssize_t vmci_transport_stream_dequeue(
1823         struct vsock_sock *vsk,
1824         struct msghdr *msg,
1825         size_t len,
1826         int flags)
1827 {
1828         if (flags & MSG_PEEK)
1829                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1830         else
1831                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1832 }
1833
1834 static ssize_t vmci_transport_stream_enqueue(
1835         struct vsock_sock *vsk,
1836         struct msghdr *msg,
1837         size_t len)
1838 {
1839         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1840 }
1841
1842 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1843 {
1844         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1845 }
1846
1847 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1848 {
1849         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1850 }
1851
1852 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1853 {
1854         return vmci_trans(vsk)->consume_size;
1855 }
1856
1857 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1858 {
1859         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1860 }
1861
1862 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1863 {
1864         return vmci_trans(vsk)->queue_pair_size;
1865 }
1866
1867 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1868 {
1869         return vmci_trans(vsk)->queue_pair_min_size;
1870 }
1871
1872 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1873 {
1874         return vmci_trans(vsk)->queue_pair_max_size;
1875 }
1876
1877 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1878 {
1879         if (val < vmci_trans(vsk)->queue_pair_min_size)
1880                 vmci_trans(vsk)->queue_pair_min_size = val;
1881         if (val > vmci_trans(vsk)->queue_pair_max_size)
1882                 vmci_trans(vsk)->queue_pair_max_size = val;
1883         vmci_trans(vsk)->queue_pair_size = val;
1884 }
1885
1886 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1887                                                u64 val)
1888 {
1889         if (val > vmci_trans(vsk)->queue_pair_size)
1890                 vmci_trans(vsk)->queue_pair_size = val;
1891         vmci_trans(vsk)->queue_pair_min_size = val;
1892 }
1893
1894 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1895                                                u64 val)
1896 {
1897         if (val < vmci_trans(vsk)->queue_pair_size)
1898                 vmci_trans(vsk)->queue_pair_size = val;
1899         vmci_trans(vsk)->queue_pair_max_size = val;
1900 }
1901
1902 static int vmci_transport_notify_poll_in(
1903         struct vsock_sock *vsk,
1904         size_t target,
1905         bool *data_ready_now)
1906 {
1907         return vmci_trans(vsk)->notify_ops->poll_in(
1908                         &vsk->sk, target, data_ready_now);
1909 }
1910
1911 static int vmci_transport_notify_poll_out(
1912         struct vsock_sock *vsk,
1913         size_t target,
1914         bool *space_available_now)
1915 {
1916         return vmci_trans(vsk)->notify_ops->poll_out(
1917                         &vsk->sk, target, space_available_now);
1918 }
1919
1920 static int vmci_transport_notify_recv_init(
1921         struct vsock_sock *vsk,
1922         size_t target,
1923         struct vsock_transport_recv_notify_data *data)
1924 {
1925         return vmci_trans(vsk)->notify_ops->recv_init(
1926                         &vsk->sk, target,
1927                         (struct vmci_transport_recv_notify_data *)data);
1928 }
1929
1930 static int vmci_transport_notify_recv_pre_block(
1931         struct vsock_sock *vsk,
1932         size_t target,
1933         struct vsock_transport_recv_notify_data *data)
1934 {
1935         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1936                         &vsk->sk, target,
1937                         (struct vmci_transport_recv_notify_data *)data);
1938 }
1939
1940 static int vmci_transport_notify_recv_pre_dequeue(
1941         struct vsock_sock *vsk,
1942         size_t target,
1943         struct vsock_transport_recv_notify_data *data)
1944 {
1945         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1946                         &vsk->sk, target,
1947                         (struct vmci_transport_recv_notify_data *)data);
1948 }
1949
1950 static int vmci_transport_notify_recv_post_dequeue(
1951         struct vsock_sock *vsk,
1952         size_t target,
1953         ssize_t copied,
1954         bool data_read,
1955         struct vsock_transport_recv_notify_data *data)
1956 {
1957         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1958                         &vsk->sk, target, copied, data_read,
1959                         (struct vmci_transport_recv_notify_data *)data);
1960 }
1961
1962 static int vmci_transport_notify_send_init(
1963         struct vsock_sock *vsk,
1964         struct vsock_transport_send_notify_data *data)
1965 {
1966         return vmci_trans(vsk)->notify_ops->send_init(
1967                         &vsk->sk,
1968                         (struct vmci_transport_send_notify_data *)data);
1969 }
1970
1971 static int vmci_transport_notify_send_pre_block(
1972         struct vsock_sock *vsk,
1973         struct vsock_transport_send_notify_data *data)
1974 {
1975         return vmci_trans(vsk)->notify_ops->send_pre_block(
1976                         &vsk->sk,
1977                         (struct vmci_transport_send_notify_data *)data);
1978 }
1979
1980 static int vmci_transport_notify_send_pre_enqueue(
1981         struct vsock_sock *vsk,
1982         struct vsock_transport_send_notify_data *data)
1983 {
1984         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1985                         &vsk->sk,
1986                         (struct vmci_transport_send_notify_data *)data);
1987 }
1988
1989 static int vmci_transport_notify_send_post_enqueue(
1990         struct vsock_sock *vsk,
1991         ssize_t written,
1992         struct vsock_transport_send_notify_data *data)
1993 {
1994         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1995                         &vsk->sk, written,
1996                         (struct vmci_transport_send_notify_data *)data);
1997 }
1998
1999 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2000 {
2001         if (PROTOCOL_OVERRIDE != -1) {
2002                 if (PROTOCOL_OVERRIDE == 0)
2003                         *old_pkt_proto = true;
2004                 else
2005                         *old_pkt_proto = false;
2006
2007                 pr_info("Proto override in use\n");
2008                 return true;
2009         }
2010
2011         return false;
2012 }
2013
2014 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2015                                                   u16 *proto,
2016                                                   bool old_pkt_proto)
2017 {
2018         struct vsock_sock *vsk = vsock_sk(sk);
2019
2020         if (old_pkt_proto) {
2021                 if (*proto != VSOCK_PROTO_INVALID) {
2022                         pr_err("Can't set both an old and new protocol\n");
2023                         return false;
2024                 }
2025                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2026                 goto exit;
2027         }
2028
2029         switch (*proto) {
2030         case VSOCK_PROTO_PKT_ON_NOTIFY:
2031                 vmci_trans(vsk)->notify_ops =
2032                         &vmci_transport_notify_pkt_q_state_ops;
2033                 break;
2034         default:
2035                 pr_err("Unknown notify protocol version\n");
2036                 return false;
2037         }
2038
2039 exit:
2040         vmci_trans(vsk)->notify_ops->socket_init(sk);
2041         return true;
2042 }
2043
2044 static u16 vmci_transport_new_proto_supported_versions(void)
2045 {
2046         if (PROTOCOL_OVERRIDE != -1)
2047                 return PROTOCOL_OVERRIDE;
2048
2049         return VSOCK_PROTO_ALL_SUPPORTED;
2050 }
2051
2052 static u32 vmci_transport_get_local_cid(void)
2053 {
2054         return vmci_get_context_id();
2055 }
2056
2057 static struct vsock_transport vmci_transport = {
2058         .init = vmci_transport_socket_init,
2059         .destruct = vmci_transport_destruct,
2060         .release = vmci_transport_release,
2061         .connect = vmci_transport_connect,
2062         .dgram_bind = vmci_transport_dgram_bind,
2063         .dgram_dequeue = vmci_transport_dgram_dequeue,
2064         .dgram_enqueue = vmci_transport_dgram_enqueue,
2065         .dgram_allow = vmci_transport_dgram_allow,
2066         .stream_dequeue = vmci_transport_stream_dequeue,
2067         .stream_enqueue = vmci_transport_stream_enqueue,
2068         .stream_has_data = vmci_transport_stream_has_data,
2069         .stream_has_space = vmci_transport_stream_has_space,
2070         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2071         .stream_is_active = vmci_transport_stream_is_active,
2072         .stream_allow = vmci_transport_stream_allow,
2073         .notify_poll_in = vmci_transport_notify_poll_in,
2074         .notify_poll_out = vmci_transport_notify_poll_out,
2075         .notify_recv_init = vmci_transport_notify_recv_init,
2076         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2077         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2078         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2079         .notify_send_init = vmci_transport_notify_send_init,
2080         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2081         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2082         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2083         .shutdown = vmci_transport_shutdown,
2084         .set_buffer_size = vmci_transport_set_buffer_size,
2085         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2086         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2087         .get_buffer_size = vmci_transport_get_buffer_size,
2088         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2089         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2090         .get_local_cid = vmci_transport_get_local_cid,
2091 };
2092
2093 static int __init vmci_transport_init(void)
2094 {
2095         int err;
2096
2097         /* Create the datagram handle that we will use to send and receive all
2098          * VSocket control messages for this context.
2099          */
2100         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2101                                                  VMCI_FLAG_ANYCID_DG_HND,
2102                                                  vmci_transport_recv_stream_cb,
2103                                                  NULL,
2104                                                  &vmci_transport_stream_handle);
2105         if (err < VMCI_SUCCESS) {
2106                 pr_err("Unable to create datagram handle. (%d)\n", err);
2107                 return vmci_transport_error_to_vsock_error(err);
2108         }
2109
2110         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2111                                    vmci_transport_qp_resumed_cb,
2112                                    NULL, &vmci_transport_qp_resumed_sub_id);
2113         if (err < VMCI_SUCCESS) {
2114                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2115                 err = vmci_transport_error_to_vsock_error(err);
2116                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2117                 goto err_destroy_stream_handle;
2118         }
2119
2120         err = vsock_core_init(&vmci_transport);
2121         if (err < 0)
2122                 goto err_unsubscribe;
2123
2124         return 0;
2125
2126 err_unsubscribe:
2127         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2128 err_destroy_stream_handle:
2129         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2130         return err;
2131 }
2132 module_init(vmci_transport_init);
2133
2134 static void __exit vmci_transport_exit(void)
2135 {
2136         cancel_work_sync(&vmci_transport_cleanup_work);
2137         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2138
2139         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2140                 if (vmci_datagram_destroy_handle(
2141                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2142                         pr_err("Couldn't destroy datagram handle\n");
2143                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2144         }
2145
2146         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2147                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2148                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2149         }
2150
2151         vsock_core_exit();
2152 }
2153 module_exit(vmci_transport_exit);
2154
2155 MODULE_AUTHOR("VMware, Inc.");
2156 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2157 MODULE_VERSION("1.0.2.0-k");
2158 MODULE_LICENSE("GPL v2");
2159 MODULE_ALIAS("vmware_vsock");
2160 MODULE_ALIAS_NETPROTO(PF_VSOCK);