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_alloc_send_control_pkt(struct sockaddr_vm *src,
 277                                       struct sockaddr_vm *dst,
 278                                       enum vmci_transport_packet_type type,
 279                                       u64 size,
 280                                       u64 mode,
 281                                       struct vmci_transport_waiting_info *wait,
 282                                       u16 proto,
 283                                       struct vmci_handle handle)
 284 {
 285         struct vmci_transport_packet *pkt;
 286         int err;
 287
 288         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 289         if (!pkt)
 290                 return -ENOMEM;
 291
 292         err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
 293                                                 mode, wait, proto, handle,
 294                                                 true);
 295         kfree(pkt);
 296
 297         return err;
 298 }
 299
 300 static int
 301 vmci_transport_send_control_pkt(struct sock *sk,
 302                                 enum vmci_transport_packet_type type,
 303                                 u64 size,
 304                                 u64 mode,
 305                                 struct vmci_transport_waiting_info *wait,
 306                                 u16 proto,
 307                                 struct vmci_handle handle)
 308 {
 309         struct vsock_sock *vsk;
 310
 311         vsk = vsock_sk(sk);
 312
 313         if (!vsock_addr_bound(&vsk->local_addr))
 314                 return -EINVAL;
 315
 316         if (!vsock_addr_bound(&vsk->remote_addr))
 317                 return -EINVAL;
 318
 319         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
 320                                                      &vsk->remote_addr,
 321                                                      type, size, mode,
 322                                                      wait, proto, handle);
 323 }
 324
 325 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 326                                         struct sockaddr_vm *src,
 327                                         struct vmci_transport_packet *pkt)
 328 {
 329         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 330                 return 0;
 331         return vmci_transport_send_control_pkt_bh(
 332                                         dst, src,
 333                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 334                                         0, NULL, VMCI_INVALID_HANDLE);
 335 }
 336
 337 static int vmci_transport_send_reset(struct sock *sk,
 338                                      struct vmci_transport_packet *pkt)
 339 {
 340         struct sockaddr_vm *dst_ptr;
 341         struct sockaddr_vm dst;
 342         struct vsock_sock *vsk;
 343
 344         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 345                 return 0;
 346
 347         vsk = vsock_sk(sk);
 348
 349         if (!vsock_addr_bound(&vsk->local_addr))
 350                 return -EINVAL;
 351
 352         if (vsock_addr_bound(&vsk->remote_addr)) {
 353                 dst_ptr = &vsk->remote_addr;
 354         } else {
 355                 vsock_addr_init(&dst, pkt->dg.src.context,
 356                                 pkt->src_port);
 357                 dst_ptr = &dst;
 358         }
 359         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
 360                                              VMCI_TRANSPORT_PACKET_TYPE_RST,
 361                                              0, 0, NULL, VSOCK_PROTO_INVALID,
 362                                              VMCI_INVALID_HANDLE);
 363 }
 364
 365 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 366 {
 367         return vmci_transport_send_control_pkt(
 368                                         sk,
 369                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 370                                         size, 0, NULL,
 371                                         VSOCK_PROTO_INVALID,
 372                                         VMCI_INVALID_HANDLE);
 373 }
 374
 375 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 376                                           u16 version)
 377 {
 378         return vmci_transport_send_control_pkt(
 379                                         sk,
 380                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 381                                         size, 0, NULL, version,
 382                                         VMCI_INVALID_HANDLE);
 383 }
 384
 385 static int vmci_transport_send_qp_offer(struct sock *sk,
 386                                         struct vmci_handle handle)
 387 {
 388         return vmci_transport_send_control_pkt(
 389                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 390                                         0, NULL,
 391                                         VSOCK_PROTO_INVALID, handle);
 392 }
 393
 394 static int vmci_transport_send_attach(struct sock *sk,
 395                                       struct vmci_handle handle)
 396 {
 397         return vmci_transport_send_control_pkt(
 398                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 399                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 400                                         handle);
 401 }
 402
 403 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 404 {
 405         return vmci_transport_reply_control_pkt_fast(
 406                                                 pkt,
 407                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 408                                                 0, 0, NULL,
 409                                                 VMCI_INVALID_HANDLE);
 410 }
 411
 412 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 413                                           struct sockaddr_vm *src)
 414 {
 415         return vmci_transport_send_control_pkt_bh(
 416                                         dst, src,
 417                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 418                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 419 }
 420
 421 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 422                                  struct sockaddr_vm *src)
 423 {
 424         return vmci_transport_send_control_pkt_bh(
 425                                         dst, src,
 426                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 427                                         0, NULL, VMCI_INVALID_HANDLE);
 428 }
 429
 430 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 431                                 struct sockaddr_vm *src)
 432 {
 433         return vmci_transport_send_control_pkt_bh(
 434                                         dst, src,
 435                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 436                                         0, NULL, VMCI_INVALID_HANDLE);
 437 }
 438
 439 int vmci_transport_send_wrote(struct sock *sk)
 440 {
 441         return vmci_transport_send_control_pkt(
 442                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 443                                         0, NULL, VSOCK_PROTO_INVALID,
 444                                         VMCI_INVALID_HANDLE);
 445 }
 446
 447 int vmci_transport_send_read(struct sock *sk)
 448 {
 449         return vmci_transport_send_control_pkt(
 450                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 451                                         0, NULL, VSOCK_PROTO_INVALID,
 452                                         VMCI_INVALID_HANDLE);
 453 }
 454
 455 int vmci_transport_send_waiting_write(struct sock *sk,
 456                                       struct vmci_transport_waiting_info *wait)
 457 {
 458         return vmci_transport_send_control_pkt(
 459                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 460                                 0, 0, wait, VSOCK_PROTO_INVALID,
 461                                 VMCI_INVALID_HANDLE);
 462 }
 463
 464 int vmci_transport_send_waiting_read(struct sock *sk,
 465                                      struct vmci_transport_waiting_info *wait)
 466 {
 467         return vmci_transport_send_control_pkt(
 468                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 469                                 0, 0, wait, VSOCK_PROTO_INVALID,
 470                                 VMCI_INVALID_HANDLE);
 471 }
 472
 473 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 474 {
 475         return vmci_transport_send_control_pkt(
 476                                         &vsk->sk,
 477                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 478                                         0, mode, NULL,
 479                                         VSOCK_PROTO_INVALID,
 480                                         VMCI_INVALID_HANDLE);
 481 }
 482
 483 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 484 {
 485         return vmci_transport_send_control_pkt(sk,
 486                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 487                                         size, 0, NULL,
 488                                         VSOCK_PROTO_INVALID,
 489                                         VMCI_INVALID_HANDLE);
 490 }
 491
 492 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 493                                              u16 version)
 494 {
 495         return vmci_transport_send_control_pkt(
 496                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 497                                         size, 0, NULL, version,
 498                                         VMCI_INVALID_HANDLE);
 499 }
 500
 501 static struct sock *vmci_transport_get_pending(
 502                                         struct sock *listener,
 503                                         struct vmci_transport_packet *pkt)
 504 {
 505         struct vsock_sock *vlistener;
 506         struct vsock_sock *vpending;
 507         struct sock *pending;
 508         struct sockaddr_vm src;
 509
 510         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 511
 512         vlistener = vsock_sk(listener);
 513
 514         list_for_each_entry(vpending, &vlistener->pending_links,
 515                             pending_links) {
 516                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 517                     pkt->dst_port == vpending->local_addr.svm_port) {
 518                         pending = sk_vsock(vpending);
 519                         sock_hold(pending);
 520                         goto found;
 521                 }
 522         }
 523
 524         pending = NULL;
 525 found:
 526         return pending;
 527
 528 }
 529
 530 static void vmci_transport_release_pending(struct sock *pending)
 531 {
 532         sock_put(pending);
 533 }
 534
 535 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 536  * trusted sockets 2) sockets from applications running as the same user as the
 537  * VM (this is only true for the host side and only when using hosted products)
 538  */
 539
 540 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 541 {
 542         return vsock->trusted ||
 543                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 544 }
 545
 546 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 547  * only if it is trusted as described in vmci_transport_is_trusted.
 548  */
 549
 550 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 551 {
 552         if (VMADDR_CID_HYPERVISOR == peer_cid)
 553                 return true;
 554
 555         if (vsock->cached_peer != peer_cid) {
 556                 vsock->cached_peer = peer_cid;
 557                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 558                     (vmci_context_get_priv_flags(peer_cid) &
 559                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 560                         vsock->cached_peer_allow_dgram = false;
 561                 } else {
 562                         vsock->cached_peer_allow_dgram = true;
 563                 }
 564         }
 565
 566         return vsock->cached_peer_allow_dgram;
 567 }
 568
 569 static int
 570 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 571                                 struct vmci_handle *handle,
 572                                 u64 produce_size,
 573                                 u64 consume_size,
 574                                 u32 peer, u32 flags, bool trusted)
 575 {
 576         int err = 0;
 577
 578         if (trusted) {
 579                 /* Try to allocate our queue pair as trusted. This will only
 580                  * work if vsock is running in the host.
 581                  */
 582
 583                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 584                                        consume_size,
 585                                        peer, flags,
 586                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 587                 if (err != VMCI_ERROR_NO_ACCESS)
 588                         goto out;
 589
 590         }
 591
 592         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 593                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 594 out:
 595         if (err < 0) {
 596                 pr_err_once("Could not attach to queue pair with %d\n", err);
 597                 err = vmci_transport_error_to_vsock_error(err);
 598         }
 599
 600         return err;
 601 }
 602
 603 static int
 604 vmci_transport_datagram_create_hnd(u32 resource_id,
 605                                    u32 flags,
 606                                    vmci_datagram_recv_cb recv_cb,
 607                                    void *client_data,
 608                                    struct vmci_handle *out_handle)
 609 {
 610         int err = 0;
 611
 612         /* Try to allocate our datagram handler as trusted. This will only work
 613          * if vsock is running in the host.
 614          */
 615
 616         err = vmci_datagram_create_handle_priv(resource_id, flags,
 617                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 618                                                recv_cb,
 619                                                client_data, out_handle);
 620
 621         if (err == VMCI_ERROR_NO_ACCESS)
 622                 err = vmci_datagram_create_handle(resource_id, flags,
 623                                                   recv_cb, client_data,
 624                                                   out_handle);
 625
 626         return err;
 627 }
 628
 629 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 630  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 631  * sleep it should defer that work to a work queue.
 632  */
 633
 634 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 635 {
 636         struct sock *sk;
 637         size_t size;
 638         struct sk_buff *skb;
 639         struct vsock_sock *vsk;
 640
 641         sk = (struct sock *)data;
 642
 643         /* This handler is privileged when this module is running on the host.
 644          * We will get datagrams from all endpoints (even VMs that are in a
 645          * restricted context). If we get one from a restricted context then
 646          * the destination socket must be trusted.
 647          *
 648          * NOTE: We access the socket struct without holding the lock here.
 649          * This is ok because the field we are interested is never modified
 650          * outside of the create and destruct socket functions.
 651          */
 652         vsk = vsock_sk(sk);
 653         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 654                 return VMCI_ERROR_NO_ACCESS;
 655
 656         size = VMCI_DG_SIZE(dg);
 657
 658         /* Attach the packet to the socket's receive queue as an sk_buff. */
 659         skb = alloc_skb(size, GFP_ATOMIC);
 660         if (!skb)
 661                 return VMCI_ERROR_NO_MEM;
 662
 663         /* sk_receive_skb() will do a sock_put(), so hold here. */
 664         sock_hold(sk);
 665         skb_put(skb, size);
 666         memcpy(skb->data, dg, size);
 667         sk_receive_skb(sk, skb, 0);
 668
 669         return VMCI_SUCCESS;
 670 }
 671
 672 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 673 {
 674         static const u32 non_socket_contexts[] = {
 675                 VMADDR_CID_RESERVED,
 676         };
 677         int i;
 678
 679         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 680
 681         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 682                 if (cid == non_socket_contexts[i])
 683                         return false;
 684         }
 685
 686         return true;
 687 }
 688
 689 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 690  * interrupt fires.  This is run in bottom-half context but it defers most of
 691  * its work to the packet handling work queue.
 692  */
 693
 694 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 695 {
 696         struct sock *sk;
 697         struct sockaddr_vm dst;
 698         struct sockaddr_vm src;
 699         struct vmci_transport_packet *pkt;
 700         struct vsock_sock *vsk;
 701         bool bh_process_pkt;
 702         int err;
 703
 704         sk = NULL;
 705         err = VMCI_SUCCESS;
 706         bh_process_pkt = false;
 707
 708         /* Ignore incoming packets from contexts without sockets, or resources
 709          * that aren't vsock implementations.
 710          */
 711
 712         if (!vmci_transport_stream_allow(dg->src.context, -1)
 713             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 714                 return VMCI_ERROR_NO_ACCESS;
 715
 716         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 717                 /* Drop datagrams that do not contain full VSock packets. */
 718                 return VMCI_ERROR_INVALID_ARGS;
 719
 720         pkt = (struct vmci_transport_packet *)dg;
 721
 722         /* Find the socket that should handle this packet.  First we look for a
 723          * connected socket and if there is none we look for a socket bound to
 724          * the destintation address.
 725          */
 726         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 727         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 728
 729         sk = vsock_find_connected_socket(&src, &dst);
 730         if (!sk) {
 731                 sk = vsock_find_bound_socket(&dst);
 732                 if (!sk) {
 733                         /* We could not find a socket for this specified
 734                          * address.  If this packet is a RST, we just drop it.
 735                          * If it is another packet, we send a RST.  Note that
 736                          * we do not send a RST reply to RSTs so that we do not
 737                          * continually send RSTs between two endpoints.
 738                          *
 739                          * Note that since this is a reply, dst is src and src
 740                          * is dst.
 741                          */
 742                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 743                                 pr_err("unable to send reset\n");
 744
 745                         err = VMCI_ERROR_NOT_FOUND;
 746                         goto out;
 747                 }
 748         }
 749
 750         /* If the received packet type is beyond all types known to this
 751          * implementation, reply with an invalid message.  Hopefully this will
 752          * help when implementing backwards compatibility in the future.
 753          */
 754         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 755                 vmci_transport_send_invalid_bh(&dst, &src);
 756                 err = VMCI_ERROR_INVALID_ARGS;
 757                 goto out;
 758         }
 759
 760         /* This handler is privileged when this module is running on the host.
 761          * We will get datagram connect requests from all endpoints (even VMs
 762          * that are in a restricted context). If we get one from a restricted
 763          * context then the destination socket must be trusted.
 764          *
 765          * NOTE: We access the socket struct without holding the lock here.
 766          * This is ok because the field we are interested is never modified
 767          * outside of the create and destruct socket functions.
 768          */
 769         vsk = vsock_sk(sk);
 770         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 771                 err = VMCI_ERROR_NO_ACCESS;
 772                 goto out;
 773         }
 774
 775         /* We do most everything in a work queue, but let's fast path the
 776          * notification of reads and writes to help data transfer performance.
 777          * We can only do this if there is no process context code executing
 778          * for this socket since that may change the state.
 779          */
 780         bh_lock_sock(sk);
 781
 782         if (!sock_owned_by_user(sk)) {
 783                 /* The local context ID may be out of date, update it. */
 784                 vsk->local_addr.svm_cid = dst.svm_cid;
 785
 786                 if (sk->sk_state == SS_CONNECTED)
 787                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 788                                         sk, pkt, true, &dst, &src,
 789                                         &bh_process_pkt);
 790         }
 791
 792         bh_unlock_sock(sk);
 793
 794         if (!bh_process_pkt) {
 795                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 796
 797                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 798                 if (!recv_pkt_info) {
 799                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 800                                 pr_err("unable to send reset\n");
 801
 802                         err = VMCI_ERROR_NO_MEM;
 803                         goto out;
 804                 }
 805
 806                 recv_pkt_info->sk = sk;
 807                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 808                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 809
 810                 schedule_work(&recv_pkt_info->work);
 811                 /* Clear sk so that the reference count incremented by one of
 812                  * the Find functions above is not decremented below.  We need
 813                  * that reference count for the packet handler we've scheduled
 814                  * to run.
 815                  */
 816                 sk = NULL;
 817         }
 818
 819 out:
 820         if (sk)
 821                 sock_put(sk);
 822
 823         return err;
 824 }
 825
 826 static void vmci_transport_handle_detach(struct sock *sk)
 827 {
 828         struct vsock_sock *vsk;
 829
 830         vsk = vsock_sk(sk);
 831         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 832                 sock_set_flag(sk, SOCK_DONE);
 833
 834                 /* On a detach the peer will not be sending or receiving
 835                  * anymore.
 836                  */
 837                 vsk->peer_shutdown = SHUTDOWN_MASK;
 838
 839                 /* We should not be sending anymore since the peer won't be
 840                  * there to receive, but we can still receive if there is data
 841                  * left in our consume queue.
 842                  */
 843                 if (vsock_stream_has_data(vsk) <= 0) {
 844                         if (sk->sk_state == SS_CONNECTING) {
 845                                 /* The peer may detach from a queue pair while
 846                                  * we are still in the connecting state, i.e.,
 847                                  * if the peer VM is killed after attaching to
 848                                  * a queue pair, but before we complete the
 849                                  * handshake. In that case, we treat the detach
 850                                  * event like a reset.
 851                                  */
 852
 853                                 sk->sk_state = SS_UNCONNECTED;
 854                                 sk->sk_err = ECONNRESET;
 855                                 sk->sk_error_report(sk);
 856                                 return;
 857                         }
 858                         sk->sk_state = SS_UNCONNECTED;
 859                 }
 860                 sk->sk_state_change(sk);
 861         }
 862 }
 863
 864 static void vmci_transport_peer_detach_cb(u32 sub_id,
 865                                           const struct vmci_event_data *e_data,
 866                                           void *client_data)
 867 {
 868         struct vmci_transport *trans = client_data;
 869         const struct vmci_event_payload_qp *e_payload;
 870
 871         e_payload = vmci_event_data_const_payload(e_data);
 872
 873         /* XXX This is lame, we should provide a way to lookup sockets by
 874          * qp_handle.
 875          */
 876         if (vmci_handle_is_invalid(e_payload->handle) ||
 877             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 878                 return;
 879
 880         /* We don't ask for delayed CBs when we subscribe to this event (we
 881          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 882          * guarantees in that case about what context we might be running in,
 883          * so it could be BH or process, blockable or non-blockable.  So we
 884          * need to account for all possible contexts here.
 885          */
 886         spin_lock_bh(&trans->lock);
 887         if (!trans->sk)
 888                 goto out;
 889
 890         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 891          * where trans->sk isn't locked.
 892          */
 893         bh_lock_sock(trans->sk);
 894
 895         vmci_transport_handle_detach(trans->sk);
 896
 897         bh_unlock_sock(trans->sk);
 898  out:
 899         spin_unlock_bh(&trans->lock);
 900 }
 901
 902 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 903                                          const struct vmci_event_data *e_data,
 904                                          void *client_data)
 905 {
 906         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 907 }
 908
 909 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 910 {
 911         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 912         struct vmci_transport_packet *pkt;
 913         struct sock *sk;
 914
 915         recv_pkt_info =
 916                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 917         sk = recv_pkt_info->sk;
 918         pkt = &recv_pkt_info->pkt;
 919
 920         lock_sock(sk);
 921
 922         /* The local context ID may be out of date. */
 923         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 924
 925         switch (sk->sk_state) {
 926         case VSOCK_SS_LISTEN:
 927                 vmci_transport_recv_listen(sk, pkt);
 928                 break;
 929         case SS_CONNECTING:
 930                 /* Processing of pending connections for servers goes through
 931                  * the listening socket, so see vmci_transport_recv_listen()
 932                  * for that path.
 933                  */
 934                 vmci_transport_recv_connecting_client(sk, pkt);
 935                 break;
 936         case SS_CONNECTED:
 937                 vmci_transport_recv_connected(sk, pkt);
 938                 break;
 939         default:
 940                 /* Because this function does not run in the same context as
 941                  * vmci_transport_recv_stream_cb it is possible that the
 942                  * socket has closed. We need to let the other side know or it
 943                  * could be sitting in a connect and hang forever. Send a
 944                  * reset to prevent that.
 945                  */
 946                 vmci_transport_send_reset(sk, pkt);
 947                 break;
 948         }
 949
 950         release_sock(sk);
 951         kfree(recv_pkt_info);
 952         /* Release reference obtained in the stream callback when we fetched
 953          * this socket out of the bound or connected list.
 954          */
 955         sock_put(sk);
 956 }
 957
 958 static int vmci_transport_recv_listen(struct sock *sk,
 959                                       struct vmci_transport_packet *pkt)
 960 {
 961         struct sock *pending;
 962         struct vsock_sock *vpending;
 963         int err;
 964         u64 qp_size;
 965         bool old_request = false;
 966         bool old_pkt_proto = false;
 967
 968         err = 0;
 969
 970         /* Because we are in the listen state, we could be receiving a packet
 971          * for ourself or any previous connection requests that we received.
 972          * If it's the latter, we try to find a socket in our list of pending
 973          * connections and, if we do, call the appropriate handler for the
 974          * state that that socket is in.  Otherwise we try to service the
 975          * connection request.
 976          */
 977         pending = vmci_transport_get_pending(sk, pkt);
 978         if (pending) {
 979                 lock_sock(pending);
 980
 981                 /* The local context ID may be out of date. */
 982                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 983
 984                 switch (pending->sk_state) {
 985                 case SS_CONNECTING:
 986                         err = vmci_transport_recv_connecting_server(sk,
 987                                                                     pending,
 988                                                                     pkt);
 989                         break;
 990                 default:
 991                         vmci_transport_send_reset(pending, pkt);
 992                         err = -EINVAL;
 993                 }
 994
 995                 if (err < 0)
 996                         vsock_remove_pending(sk, pending);
 997
 998                 release_sock(pending);
 999                 vmci_transport_release_pending(pending);
1000
1001                 return err;
1002         }
1003
1004         /* The listen state only accepts connection requests.  Reply with a
1005          * reset unless we received a reset.
1006          */
1007
1008         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
1009               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
1010                 vmci_transport_reply_reset(pkt);
1011                 return -EINVAL;
1012         }
1013
1014         if (pkt->u.size == 0) {
1015                 vmci_transport_reply_reset(pkt);
1016                 return -EINVAL;
1017         }
1018
1019         /* If this socket can't accommodate this connection request, we send a
1020          * reset.  Otherwise we create and initialize a child socket and reply
1021          * with a connection negotiation.
1022          */
1023         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1024                 vmci_transport_reply_reset(pkt);
1025                 return -ECONNREFUSED;
1026         }
1027
1028         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1029                                  sk->sk_type, 0);
1030         if (!pending) {
1031                 vmci_transport_send_reset(sk, pkt);
1032                 return -ENOMEM;
1033         }
1034
1035         vpending = vsock_sk(pending);
1036
1037         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1038                         pkt->dst_port);
1039         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1040                         pkt->src_port);
1041
1042         /* If the proposed size fits within our min/max, accept it. Otherwise
1043          * propose our own size.
1044          */
1045         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1046             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1047                 qp_size = pkt->u.size;
1048         } else {
1049                 qp_size = vmci_trans(vpending)->queue_pair_size;
1050         }
1051
1052         /* Figure out if we are using old or new requests based on the
1053          * overrides pkt types sent by our peer.
1054          */
1055         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1056                 old_request = old_pkt_proto;
1057         } else {
1058                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1059                         old_request = true;
1060                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1061                         old_request = false;
1062
1063         }
1064
1065         if (old_request) {
1066                 /* Handle a REQUEST (or override) */
1067                 u16 version = VSOCK_PROTO_INVALID;
1068                 if (vmci_transport_proto_to_notify_struct(
1069                         pending, &version, true))
1070                         err = vmci_transport_send_negotiate(pending, qp_size);
1071                 else
1072                         err = -EINVAL;
1073
1074         } else {
1075                 /* Handle a REQUEST2 (or override) */
1076                 int proto_int = pkt->proto;
1077                 int pos;
1078                 u16 active_proto_version = 0;
1079
1080                 /* The list of possible protocols is the intersection of all
1081                  * protocols the client supports ... plus all the protocols we
1082                  * support.
1083                  */
1084                 proto_int &= vmci_transport_new_proto_supported_versions();
1085
1086                 /* We choose the highest possible protocol version and use that
1087                  * one.
1088                  */
1089                 pos = fls(proto_int);
1090                 if (pos) {
1091                         active_proto_version = (1 << (pos - 1));
1092                         if (vmci_transport_proto_to_notify_struct(
1093                                 pending, &active_proto_version, false))
1094                                 err = vmci_transport_send_negotiate2(pending,
1095                                                         qp_size,
1096                                                         active_proto_version);
1097                         else
1098                                 err = -EINVAL;
1099
1100                 } else {
1101                         err = -EINVAL;
1102                 }
1103         }
1104
1105         if (err < 0) {
1106                 vmci_transport_send_reset(sk, pkt);
1107                 sock_put(pending);
1108                 err = vmci_transport_error_to_vsock_error(err);
1109                 goto out;
1110         }
1111
1112         vsock_add_pending(sk, pending);
1113         sk->sk_ack_backlog++;
1114
1115         pending->sk_state = SS_CONNECTING;
1116         vmci_trans(vpending)->produce_size =
1117                 vmci_trans(vpending)->consume_size = qp_size;
1118         vmci_trans(vpending)->queue_pair_size = qp_size;
1119
1120         vmci_trans(vpending)->notify_ops->process_request(pending);
1121
1122         /* We might never receive another message for this socket and it's not
1123          * connected to any process, so we have to ensure it gets cleaned up
1124          * ourself.  Our delayed work function will take care of that.  Note
1125          * that we do not ever cancel this function since we have few
1126          * guarantees about its state when calling cancel_delayed_work().
1127          * Instead we hold a reference on the socket for that function and make
1128          * it capable of handling cases where it needs to do nothing but
1129          * release that reference.
1130          */
1131         vpending->listener = sk;
1132         sock_hold(sk);
1133         sock_hold(pending);
1134         schedule_delayed_work(&vpending->pending_work, HZ);
1135
1136 out:
1137         return err;
1138 }
1139
1140 static int
1141 vmci_transport_recv_connecting_server(struct sock *listener,
1142                                       struct sock *pending,
1143                                       struct vmci_transport_packet *pkt)
1144 {
1145         struct vsock_sock *vpending;
1146         struct vmci_handle handle;
1147         struct vmci_qp *qpair;
1148         bool is_local;
1149         u32 flags;
1150         u32 detach_sub_id;
1151         int err;
1152         int skerr;
1153
1154         vpending = vsock_sk(pending);
1155         detach_sub_id = VMCI_INVALID_ID;
1156
1157         switch (pkt->type) {
1158         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1159                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1160                         vmci_transport_send_reset(pending, pkt);
1161                         skerr = EPROTO;
1162                         err = -EINVAL;
1163                         goto destroy;
1164                 }
1165                 break;
1166         default:
1167                 /* Close and cleanup the connection. */
1168                 vmci_transport_send_reset(pending, pkt);
1169                 skerr = EPROTO;
1170                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1171                 goto destroy;
1172         }
1173
1174         /* In order to complete the connection we need to attach to the offered
1175          * queue pair and send an attach notification.  We also subscribe to the
1176          * detach event so we know when our peer goes away, and we do that
1177          * before attaching so we don't miss an event.  If all this succeeds,
1178          * we update our state and wakeup anything waiting in accept() for a
1179          * connection.
1180          */
1181
1182         /* We don't care about attach since we ensure the other side has
1183          * attached by specifying the ATTACH_ONLY flag below.
1184          */
1185         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1186                                    vmci_transport_peer_detach_cb,
1187                                    vmci_trans(vpending), &detach_sub_id);
1188         if (err < VMCI_SUCCESS) {
1189                 vmci_transport_send_reset(pending, pkt);
1190                 err = vmci_transport_error_to_vsock_error(err);
1191                 skerr = -err;
1192                 goto destroy;
1193         }
1194
1195         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1196
1197         /* Now attach to the queue pair the client created. */
1198         handle = pkt->u.handle;
1199
1200         /* vpending->local_addr always has a context id so we do not need to
1201          * worry about VMADDR_CID_ANY in this case.
1202          */
1203         is_local =
1204             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1205         flags = VMCI_QPFLAG_ATTACH_ONLY;
1206         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1207
1208         err = vmci_transport_queue_pair_alloc(
1209                                         &qpair,
1210                                         &handle,
1211                                         vmci_trans(vpending)->produce_size,
1212                                         vmci_trans(vpending)->consume_size,
1213                                         pkt->dg.src.context,
1214                                         flags,
1215                                         vmci_transport_is_trusted(
1216                                                 vpending,
1217                                                 vpending->remote_addr.svm_cid));
1218         if (err < 0) {
1219                 vmci_transport_send_reset(pending, pkt);
1220                 skerr = -err;
1221                 goto destroy;
1222         }
1223
1224         vmci_trans(vpending)->qp_handle = handle;
1225         vmci_trans(vpending)->qpair = qpair;
1226
1227         /* When we send the attach message, we must be ready to handle incoming
1228          * control messages on the newly connected socket. So we move the
1229          * pending socket to the connected state before sending the attach
1230          * message. Otherwise, an incoming packet triggered by the attach being
1231          * received by the peer may be processed concurrently with what happens
1232          * below after sending the attach message, and that incoming packet
1233          * will find the listening socket instead of the (currently) pending
1234          * socket. Note that enqueueing the socket increments the reference
1235          * count, so even if a reset comes before the connection is accepted,
1236          * the socket will be valid until it is removed from the queue.
1237          *
1238          * If we fail sending the attach below, we remove the socket from the
1239          * connected list and move the socket to SS_UNCONNECTED before
1240          * releasing the lock, so a pending slow path processing of an incoming
1241          * packet will not see the socket in the connected state in that case.
1242          */
1243         pending->sk_state = SS_CONNECTED;
1244
1245         vsock_insert_connected(vpending);
1246
1247         /* Notify our peer of our attach. */
1248         err = vmci_transport_send_attach(pending, handle);
1249         if (err < 0) {
1250                 vsock_remove_connected(vpending);
1251                 pr_err("Could not send attach\n");
1252                 vmci_transport_send_reset(pending, pkt);
1253                 err = vmci_transport_error_to_vsock_error(err);
1254                 skerr = -err;
1255                 goto destroy;
1256         }
1257
1258         /* We have a connection. Move the now connected socket from the
1259          * listener's pending list to the accept queue so callers of accept()
1260          * can find it.
1261          */
1262         vsock_remove_pending(listener, pending);
1263         vsock_enqueue_accept(listener, pending);
1264
1265         /* Callers of accept() will be be waiting on the listening socket, not
1266          * the pending socket.
1267          */
1268         listener->sk_data_ready(listener);
1269
1270         return 0;
1271
1272 destroy:
1273         pending->sk_err = skerr;
1274         pending->sk_state = SS_UNCONNECTED;
1275         /* As long as we drop our reference, all necessary cleanup will handle
1276          * when the cleanup function drops its reference and our destruct
1277          * implementation is called.  Note that since the listen handler will
1278          * remove pending from the pending list upon our failure, the cleanup
1279          * function won't drop the additional reference, which is why we do it
1280          * here.
1281          */
1282         sock_put(pending);
1283
1284         return err;
1285 }
1286
1287 static int
1288 vmci_transport_recv_connecting_client(struct sock *sk,
1289                                       struct vmci_transport_packet *pkt)
1290 {
1291         struct vsock_sock *vsk;
1292         int err;
1293         int skerr;
1294
1295         vsk = vsock_sk(sk);
1296
1297         switch (pkt->type) {
1298         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1299                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1300                     !vmci_handle_is_equal(pkt->u.handle,
1301                                           vmci_trans(vsk)->qp_handle)) {
1302                         skerr = EPROTO;
1303                         err = -EINVAL;
1304                         goto destroy;
1305                 }
1306
1307                 /* Signify the socket is connected and wakeup the waiter in
1308                  * connect(). Also place the socket in the connected table for
1309                  * accounting (it can already be found since it's in the bound
1310                  * table).
1311                  */
1312                 sk->sk_state = SS_CONNECTED;
1313                 sk->sk_socket->state = SS_CONNECTED;
1314                 vsock_insert_connected(vsk);
1315                 sk->sk_state_change(sk);
1316
1317                 break;
1318         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1319         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1320                 if (pkt->u.size == 0
1321                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1322                     || pkt->src_port != vsk->remote_addr.svm_port
1323                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1324                     || vmci_trans(vsk)->qpair
1325                     || vmci_trans(vsk)->produce_size != 0
1326                     || vmci_trans(vsk)->consume_size != 0
1327                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1328                         skerr = EPROTO;
1329                         err = -EINVAL;
1330
1331                         goto destroy;
1332                 }
1333
1334                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1335                 if (err) {
1336                         skerr = -err;
1337                         goto destroy;
1338                 }
1339
1340                 break;
1341         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1342                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1343                 if (err) {
1344                         skerr = -err;
1345                         goto destroy;
1346                 }
1347
1348                 break;
1349         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1350                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1351                  * continue processing here after they sent an INVALID packet.
1352                  * This meant that we got a RST after the INVALID. We ignore a
1353                  * RST after an INVALID. The common code doesn't send the RST
1354                  * ... so we can hang if an old version of the common code
1355                  * fails between getting a REQUEST and sending an OFFER back.
1356                  * Not much we can do about it... except hope that it doesn't
1357                  * happen.
1358                  */
1359                 if (vsk->ignore_connecting_rst) {
1360                         vsk->ignore_connecting_rst = false;
1361                 } else {
1362                         skerr = ECONNRESET;
1363                         err = 0;
1364                         goto destroy;
1365                 }
1366
1367                 break;
1368         default:
1369                 /* Close and cleanup the connection. */
1370                 skerr = EPROTO;
1371                 err = -EINVAL;
1372                 goto destroy;
1373         }
1374
1375         return 0;
1376
1377 destroy:
1378         vmci_transport_send_reset(sk, pkt);
1379
1380         sk->sk_state = SS_UNCONNECTED;
1381         sk->sk_err = skerr;
1382         sk->sk_error_report(sk);
1383         return err;
1384 }
1385
1386 static int vmci_transport_recv_connecting_client_negotiate(
1387                                         struct sock *sk,
1388                                         struct vmci_transport_packet *pkt)
1389 {
1390         int err;
1391         struct vsock_sock *vsk;
1392         struct vmci_handle handle;
1393         struct vmci_qp *qpair;
1394         u32 detach_sub_id;
1395         bool is_local;
1396         u32 flags;
1397         bool old_proto = true;
1398         bool old_pkt_proto;
1399         u16 version;
1400
1401         vsk = vsock_sk(sk);
1402         handle = VMCI_INVALID_HANDLE;
1403         detach_sub_id = VMCI_INVALID_ID;
1404
1405         /* If we have gotten here then we should be past the point where old
1406          * linux vsock could have sent the bogus rst.
1407          */
1408         vsk->sent_request = false;
1409         vsk->ignore_connecting_rst = false;
1410
1411         /* Verify that we're OK with the proposed queue pair size */
1412         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1413             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1414                 err = -EINVAL;
1415                 goto destroy;
1416         }
1417
1418         /* At this point we know the CID the peer is using to talk to us. */
1419
1420         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1421                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1422
1423         /* Setup the notify ops to be the highest supported version that both
1424          * the server and the client support.
1425          */
1426
1427         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1428                 old_proto = old_pkt_proto;
1429         } else {
1430                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1431                         old_proto = true;
1432                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1433                         old_proto = false;
1434
1435         }
1436
1437         if (old_proto)
1438                 version = VSOCK_PROTO_INVALID;
1439         else
1440                 version = pkt->proto;
1441
1442         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1443                 err = -EINVAL;
1444                 goto destroy;
1445         }
1446
1447         /* Subscribe to detach events first.
1448          *
1449          * XXX We attach once for each queue pair created for now so it is easy
1450          * to find the socket (it's provided), but later we should only
1451          * subscribe once and add a way to lookup sockets by queue pair handle.
1452          */
1453         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1454                                    vmci_transport_peer_detach_cb,
1455                                    vmci_trans(vsk), &detach_sub_id);
1456         if (err < VMCI_SUCCESS) {
1457                 err = vmci_transport_error_to_vsock_error(err);
1458                 goto destroy;
1459         }
1460
1461         /* Make VMCI select the handle for us. */
1462         handle = VMCI_INVALID_HANDLE;
1463         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1464         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1465
1466         err = vmci_transport_queue_pair_alloc(&qpair,
1467                                               &handle,
1468                                               pkt->u.size,
1469                                               pkt->u.size,
1470                                               vsk->remote_addr.svm_cid,
1471                                               flags,
1472                                               vmci_transport_is_trusted(
1473                                                   vsk,
1474                                                   vsk->
1475                                                   remote_addr.svm_cid));
1476         if (err < 0)
1477                 goto destroy;
1478
1479         err = vmci_transport_send_qp_offer(sk, handle);
1480         if (err < 0) {
1481                 err = vmci_transport_error_to_vsock_error(err);
1482                 goto destroy;
1483         }
1484
1485         vmci_trans(vsk)->qp_handle = handle;
1486         vmci_trans(vsk)->qpair = qpair;
1487
1488         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1489                 pkt->u.size;
1490
1491         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1492
1493         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1494
1495         return 0;
1496
1497 destroy:
1498         if (detach_sub_id != VMCI_INVALID_ID)
1499                 vmci_event_unsubscribe(detach_sub_id);
1500
1501         if (!vmci_handle_is_invalid(handle))
1502                 vmci_qpair_detach(&qpair);
1503
1504         return err;
1505 }
1506
1507 static int
1508 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1509                                               struct vmci_transport_packet *pkt)
1510 {
1511         int err = 0;
1512         struct vsock_sock *vsk = vsock_sk(sk);
1513
1514         if (vsk->sent_request) {
1515                 vsk->sent_request = false;
1516                 vsk->ignore_connecting_rst = true;
1517
1518                 err = vmci_transport_send_conn_request(
1519                         sk, vmci_trans(vsk)->queue_pair_size);
1520                 if (err < 0)
1521                         err = vmci_transport_error_to_vsock_error(err);
1522                 else
1523                         err = 0;
1524
1525         }
1526
1527         return err;
1528 }
1529
1530 static int vmci_transport_recv_connected(struct sock *sk,
1531                                          struct vmci_transport_packet *pkt)
1532 {
1533         struct vsock_sock *vsk;
1534         bool pkt_processed = false;
1535
1536         /* In cases where we are closing the connection, it's sufficient to
1537          * mark the state change (and maybe error) and wake up any waiting
1538          * threads. Since this is a connected socket, it's owned by a user
1539          * process and will be cleaned up when the failure is passed back on
1540          * the current or next system call.  Our system call implementations
1541          * must therefore check for error and state changes on entry and when
1542          * being awoken.
1543          */
1544         switch (pkt->type) {
1545         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1546                 if (pkt->u.mode) {
1547                         vsk = vsock_sk(sk);
1548
1549                         vsk->peer_shutdown |= pkt->u.mode;
1550                         sk->sk_state_change(sk);
1551                 }
1552                 break;
1553
1554         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1555                 vsk = vsock_sk(sk);
1556                 /* It is possible that we sent our peer a message (e.g a
1557                  * WAITING_READ) right before we got notified that the peer had
1558                  * detached. If that happens then we can get a RST pkt back
1559                  * from our peer even though there is data available for us to
1560                  * read. In that case, don't shutdown the socket completely but
1561                  * instead allow the local client to finish reading data off
1562                  * the queuepair. Always treat a RST pkt in connected mode like
1563                  * a clean shutdown.
1564                  */
1565                 sock_set_flag(sk, SOCK_DONE);
1566                 vsk->peer_shutdown = SHUTDOWN_MASK;
1567                 if (vsock_stream_has_data(vsk) <= 0)
1568                         sk->sk_state = SS_DISCONNECTING;
1569
1570                 sk->sk_state_change(sk);
1571                 break;
1572
1573         default:
1574                 vsk = vsock_sk(sk);
1575                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1576                                 sk, pkt, false, NULL, NULL,
1577                                 &pkt_processed);
1578                 if (!pkt_processed)
1579                         return -EINVAL;
1580
1581                 break;
1582         }
1583
1584         return 0;
1585 }
1586
1587 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1588                                       struct vsock_sock *psk)
1589 {
1590         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1591         if (!vsk->trans)
1592                 return -ENOMEM;
1593
1594         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1595         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1596         vmci_trans(vsk)->qpair = NULL;
1597         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1598         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1599         vmci_trans(vsk)->notify_ops = NULL;
1600         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1601         vmci_trans(vsk)->sk = &vsk->sk;
1602         spin_lock_init(&vmci_trans(vsk)->lock);
1603         if (psk) {
1604                 vmci_trans(vsk)->queue_pair_size =
1605                         vmci_trans(psk)->queue_pair_size;
1606                 vmci_trans(vsk)->queue_pair_min_size =
1607                         vmci_trans(psk)->queue_pair_min_size;
1608                 vmci_trans(vsk)->queue_pair_max_size =
1609                         vmci_trans(psk)->queue_pair_max_size;
1610         } else {
1611                 vmci_trans(vsk)->queue_pair_size =
1612                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1613                 vmci_trans(vsk)->queue_pair_min_size =
1614                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1615                 vmci_trans(vsk)->queue_pair_max_size =
1616                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1617         }
1618
1619         return 0;
1620 }
1621
1622 static void vmci_transport_free_resources(struct list_head *transport_list)
1623 {
1624         while (!list_empty(transport_list)) {
1625                 struct vmci_transport *transport =
1626                     list_first_entry(transport_list, struct vmci_transport,
1627                                      elem);
1628                 list_del(&transport->elem);
1629
1630                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1631                         vmci_event_unsubscribe(transport->detach_sub_id);
1632                         transport->detach_sub_id = VMCI_INVALID_ID;
1633                 }
1634
1635                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1636                         vmci_qpair_detach(&transport->qpair);
1637                         transport->qp_handle = VMCI_INVALID_HANDLE;
1638                         transport->produce_size = 0;
1639                         transport->consume_size = 0;
1640                 }
1641
1642                 kfree(transport);
1643         }
1644 }
1645
1646 static void vmci_transport_cleanup(struct work_struct *work)
1647 {
1648         LIST_HEAD(pending);
1649
1650         spin_lock_bh(&vmci_transport_cleanup_lock);
1651         list_replace_init(&vmci_transport_cleanup_list, &pending);
1652         spin_unlock_bh(&vmci_transport_cleanup_lock);
1653         vmci_transport_free_resources(&pending);
1654 }
1655
1656 static void vmci_transport_destruct(struct vsock_sock *vsk)
1657 {
1658         /* transport can be NULL if we hit a failure at init() time */
1659         if (!vmci_trans(vsk))
1660                 return;
1661
1662         /* Ensure that the detach callback doesn't use the sk/vsk
1663          * we are about to destruct.
1664          */
1665         spin_lock_bh(&vmci_trans(vsk)->lock);
1666         vmci_trans(vsk)->sk = NULL;
1667         spin_unlock_bh(&vmci_trans(vsk)->lock);
1668
1669         if (vmci_trans(vsk)->notify_ops)
1670                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1671
1672         spin_lock_bh(&vmci_transport_cleanup_lock);
1673         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1674         spin_unlock_bh(&vmci_transport_cleanup_lock);
1675         schedule_work(&vmci_transport_cleanup_work);
1676
1677         vsk->trans = NULL;
1678 }
1679
1680 static void vmci_transport_release(struct vsock_sock *vsk)
1681 {
1682         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1683                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1684                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1685         }
1686 }
1687
1688 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1689                                      struct sockaddr_vm *addr)
1690 {
1691         u32 port;
1692         u32 flags;
1693         int err;
1694
1695         /* VMCI will select a resource ID for us if we provide
1696          * VMCI_INVALID_ID.
1697          */
1698         port = addr->svm_port == VMADDR_PORT_ANY ?
1699                         VMCI_INVALID_ID : addr->svm_port;
1700
1701         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1702                 return -EACCES;
1703
1704         flags = addr->svm_cid == VMADDR_CID_ANY ?
1705                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1706
1707         err = vmci_transport_datagram_create_hnd(port, flags,
1708                                                  vmci_transport_recv_dgram_cb,
1709                                                  &vsk->sk,
1710                                                  &vmci_trans(vsk)->dg_handle);
1711         if (err < VMCI_SUCCESS)
1712                 return vmci_transport_error_to_vsock_error(err);
1713         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1714                         vmci_trans(vsk)->dg_handle.resource);
1715
1716         return 0;
1717 }
1718
1719 static int vmci_transport_dgram_enqueue(
1720         struct vsock_sock *vsk,
1721         struct sockaddr_vm *remote_addr,
1722         struct msghdr *msg,
1723         size_t len)
1724 {
1725         int err;
1726         struct vmci_datagram *dg;
1727
1728         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1729                 return -EMSGSIZE;
1730
1731         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1732                 return -EPERM;
1733
1734         /* Allocate a buffer for the user's message and our packet header. */
1735         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1736         if (!dg)
1737                 return -ENOMEM;
1738
1739         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1740
1741         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1742                                    remote_addr->svm_port);
1743         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1744                                    vsk->local_addr.svm_port);
1745         dg->payload_size = len;
1746
1747         err = vmci_datagram_send(dg);
1748         kfree(dg);
1749         if (err < 0)
1750                 return vmci_transport_error_to_vsock_error(err);
1751
1752         return err - sizeof(*dg);
1753 }
1754
1755 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1756                                         struct msghdr *msg, size_t len,
1757                                         int flags)
1758 {
1759         int err;
1760         int noblock;
1761         struct vmci_datagram *dg;
1762         size_t payload_len;
1763         struct sk_buff *skb;
1764
1765         noblock = flags & MSG_DONTWAIT;
1766
1767         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1768                 return -EOPNOTSUPP;
1769
1770         /* Retrieve the head sk_buff from the socket's receive queue. */
1771         err = 0;
1772         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1773         if (err)
1774                 return err;
1775
1776         if (!skb)
1777                 return -EAGAIN;
1778
1779         dg = (struct vmci_datagram *)skb->data;
1780         if (!dg)
1781                 /* err is 0, meaning we read zero bytes. */
1782                 goto out;
1783
1784         payload_len = dg->payload_size;
1785         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1786         if (payload_len != skb->len - sizeof(*dg)) {
1787                 err = -EINVAL;
1788                 goto out;
1789         }
1790
1791         if (payload_len > len) {
1792                 payload_len = len;
1793                 msg->msg_flags |= MSG_TRUNC;
1794         }
1795
1796         /* Place the datagram payload in the user's iovec. */
1797         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1798         if (err)
1799                 goto out;
1800
1801         if (msg->msg_name) {
1802                 /* Provide the address of the sender. */
1803                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1804                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1805                 msg->msg_namelen = sizeof(*vm_addr);
1806         }
1807         err = payload_len;
1808
1809 out:
1810         skb_free_datagram(&vsk->sk, skb);
1811         return err;
1812 }
1813
1814 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1815 {
1816         if (cid == VMADDR_CID_HYPERVISOR) {
1817                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1818                  * state and are allowed.
1819                  */
1820                 return port == VMCI_UNITY_PBRPC_REGISTER;
1821         }
1822
1823         return true;
1824 }
1825
1826 static int vmci_transport_connect(struct vsock_sock *vsk)
1827 {
1828         int err;
1829         bool old_pkt_proto = false;
1830         struct sock *sk = &vsk->sk;
1831
1832         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1833                 old_pkt_proto) {
1834                 err = vmci_transport_send_conn_request(
1835                         sk, vmci_trans(vsk)->queue_pair_size);
1836                 if (err < 0) {
1837                         sk->sk_state = SS_UNCONNECTED;
1838                         return err;
1839                 }
1840         } else {
1841                 int supported_proto_versions =
1842                         vmci_transport_new_proto_supported_versions();
1843                 err = vmci_transport_send_conn_request2(
1844                                 sk, vmci_trans(vsk)->queue_pair_size,
1845                                 supported_proto_versions);
1846                 if (err < 0) {
1847                         sk->sk_state = SS_UNCONNECTED;
1848                         return err;
1849                 }
1850
1851                 vsk->sent_request = true;
1852         }
1853
1854         return err;
1855 }
1856
1857 static ssize_t vmci_transport_stream_dequeue(
1858         struct vsock_sock *vsk,
1859         struct msghdr *msg,
1860         size_t len,
1861         int flags)
1862 {
1863         if (flags & MSG_PEEK)
1864                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1865         else
1866                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1867 }
1868
1869 static ssize_t vmci_transport_stream_enqueue(
1870         struct vsock_sock *vsk,
1871         struct msghdr *msg,
1872         size_t len)
1873 {
1874         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1875 }
1876
1877 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1878 {
1879         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1880 }
1881
1882 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1883 {
1884         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1885 }
1886
1887 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1888 {
1889         return vmci_trans(vsk)->consume_size;
1890 }
1891
1892 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1893 {
1894         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1895 }
1896
1897 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1898 {
1899         return vmci_trans(vsk)->queue_pair_size;
1900 }
1901
1902 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1903 {
1904         return vmci_trans(vsk)->queue_pair_min_size;
1905 }
1906
1907 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1908 {
1909         return vmci_trans(vsk)->queue_pair_max_size;
1910 }
1911
1912 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1913 {
1914         if (val < vmci_trans(vsk)->queue_pair_min_size)
1915                 vmci_trans(vsk)->queue_pair_min_size = val;
1916         if (val > vmci_trans(vsk)->queue_pair_max_size)
1917                 vmci_trans(vsk)->queue_pair_max_size = val;
1918         vmci_trans(vsk)->queue_pair_size = val;
1919 }
1920
1921 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1922                                                u64 val)
1923 {
1924         if (val > vmci_trans(vsk)->queue_pair_size)
1925                 vmci_trans(vsk)->queue_pair_size = val;
1926         vmci_trans(vsk)->queue_pair_min_size = val;
1927 }
1928
1929 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1930                                                u64 val)
1931 {
1932         if (val < vmci_trans(vsk)->queue_pair_size)
1933                 vmci_trans(vsk)->queue_pair_size = val;
1934         vmci_trans(vsk)->queue_pair_max_size = val;
1935 }
1936
1937 static int vmci_transport_notify_poll_in(
1938         struct vsock_sock *vsk,
1939         size_t target,
1940         bool *data_ready_now)
1941 {
1942         return vmci_trans(vsk)->notify_ops->poll_in(
1943                         &vsk->sk, target, data_ready_now);
1944 }
1945
1946 static int vmci_transport_notify_poll_out(
1947         struct vsock_sock *vsk,
1948         size_t target,
1949         bool *space_available_now)
1950 {
1951         return vmci_trans(vsk)->notify_ops->poll_out(
1952                         &vsk->sk, target, space_available_now);
1953 }
1954
1955 static int vmci_transport_notify_recv_init(
1956         struct vsock_sock *vsk,
1957         size_t target,
1958         struct vsock_transport_recv_notify_data *data)
1959 {
1960         return vmci_trans(vsk)->notify_ops->recv_init(
1961                         &vsk->sk, target,
1962                         (struct vmci_transport_recv_notify_data *)data);
1963 }
1964
1965 static int vmci_transport_notify_recv_pre_block(
1966         struct vsock_sock *vsk,
1967         size_t target,
1968         struct vsock_transport_recv_notify_data *data)
1969 {
1970         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1971                         &vsk->sk, target,
1972                         (struct vmci_transport_recv_notify_data *)data);
1973 }
1974
1975 static int vmci_transport_notify_recv_pre_dequeue(
1976         struct vsock_sock *vsk,
1977         size_t target,
1978         struct vsock_transport_recv_notify_data *data)
1979 {
1980         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1981                         &vsk->sk, target,
1982                         (struct vmci_transport_recv_notify_data *)data);
1983 }
1984
1985 static int vmci_transport_notify_recv_post_dequeue(
1986         struct vsock_sock *vsk,
1987         size_t target,
1988         ssize_t copied,
1989         bool data_read,
1990         struct vsock_transport_recv_notify_data *data)
1991 {
1992         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1993                         &vsk->sk, target, copied, data_read,
1994                         (struct vmci_transport_recv_notify_data *)data);
1995 }
1996
1997 static int vmci_transport_notify_send_init(
1998         struct vsock_sock *vsk,
1999         struct vsock_transport_send_notify_data *data)
2000 {
2001         return vmci_trans(vsk)->notify_ops->send_init(
2002                         &vsk->sk,
2003                         (struct vmci_transport_send_notify_data *)data);
2004 }
2005
2006 static int vmci_transport_notify_send_pre_block(
2007         struct vsock_sock *vsk,
2008         struct vsock_transport_send_notify_data *data)
2009 {
2010         return vmci_trans(vsk)->notify_ops->send_pre_block(
2011                         &vsk->sk,
2012                         (struct vmci_transport_send_notify_data *)data);
2013 }
2014
2015 static int vmci_transport_notify_send_pre_enqueue(
2016         struct vsock_sock *vsk,
2017         struct vsock_transport_send_notify_data *data)
2018 {
2019         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2020                         &vsk->sk,
2021                         (struct vmci_transport_send_notify_data *)data);
2022 }
2023
2024 static int vmci_transport_notify_send_post_enqueue(
2025         struct vsock_sock *vsk,
2026         ssize_t written,
2027         struct vsock_transport_send_notify_data *data)
2028 {
2029         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2030                         &vsk->sk, written,
2031                         (struct vmci_transport_send_notify_data *)data);
2032 }
2033
2034 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2035 {
2036         if (PROTOCOL_OVERRIDE != -1) {
2037                 if (PROTOCOL_OVERRIDE == 0)
2038                         *old_pkt_proto = true;
2039                 else
2040                         *old_pkt_proto = false;
2041
2042                 pr_info("Proto override in use\n");
2043                 return true;
2044         }
2045
2046         return false;
2047 }
2048
2049 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2050                                                   u16 *proto,
2051                                                   bool old_pkt_proto)
2052 {
2053         struct vsock_sock *vsk = vsock_sk(sk);
2054
2055         if (old_pkt_proto) {
2056                 if (*proto != VSOCK_PROTO_INVALID) {
2057                         pr_err("Can't set both an old and new protocol\n");
2058                         return false;
2059                 }
2060                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2061                 goto exit;
2062         }
2063
2064         switch (*proto) {
2065         case VSOCK_PROTO_PKT_ON_NOTIFY:
2066                 vmci_trans(vsk)->notify_ops =
2067                         &vmci_transport_notify_pkt_q_state_ops;
2068                 break;
2069         default:
2070                 pr_err("Unknown notify protocol version\n");
2071                 return false;
2072         }
2073
2074 exit:
2075         vmci_trans(vsk)->notify_ops->socket_init(sk);
2076         return true;
2077 }
2078
2079 static u16 vmci_transport_new_proto_supported_versions(void)
2080 {
2081         if (PROTOCOL_OVERRIDE != -1)
2082                 return PROTOCOL_OVERRIDE;
2083
2084         return VSOCK_PROTO_ALL_SUPPORTED;
2085 }
2086
2087 static u32 vmci_transport_get_local_cid(void)
2088 {
2089         return vmci_get_context_id();
2090 }
2091
2092 static struct vsock_transport vmci_transport = {
2093         .init = vmci_transport_socket_init,
2094         .destruct = vmci_transport_destruct,
2095         .release = vmci_transport_release,
2096         .connect = vmci_transport_connect,
2097         .dgram_bind = vmci_transport_dgram_bind,
2098         .dgram_dequeue = vmci_transport_dgram_dequeue,
2099         .dgram_enqueue = vmci_transport_dgram_enqueue,
2100         .dgram_allow = vmci_transport_dgram_allow,
2101         .stream_dequeue = vmci_transport_stream_dequeue,
2102         .stream_enqueue = vmci_transport_stream_enqueue,
2103         .stream_has_data = vmci_transport_stream_has_data,
2104         .stream_has_space = vmci_transport_stream_has_space,
2105         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2106         .stream_is_active = vmci_transport_stream_is_active,
2107         .stream_allow = vmci_transport_stream_allow,
2108         .notify_poll_in = vmci_transport_notify_poll_in,
2109         .notify_poll_out = vmci_transport_notify_poll_out,
2110         .notify_recv_init = vmci_transport_notify_recv_init,
2111         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2112         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2113         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2114         .notify_send_init = vmci_transport_notify_send_init,
2115         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2116         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2117         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2118         .shutdown = vmci_transport_shutdown,
2119         .set_buffer_size = vmci_transport_set_buffer_size,
2120         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2121         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2122         .get_buffer_size = vmci_transport_get_buffer_size,
2123         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2124         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2125         .get_local_cid = vmci_transport_get_local_cid,
2126 };
2127
2128 static int __init vmci_transport_init(void)
2129 {
2130         int err;
2131
2132         /* Create the datagram handle that we will use to send and receive all
2133          * VSocket control messages for this context.
2134          */
2135         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2136                                                  VMCI_FLAG_ANYCID_DG_HND,
2137                                                  vmci_transport_recv_stream_cb,
2138                                                  NULL,
2139                                                  &vmci_transport_stream_handle);
2140         if (err < VMCI_SUCCESS) {
2141                 pr_err("Unable to create datagram handle. (%d)\n", err);
2142                 return vmci_transport_error_to_vsock_error(err);
2143         }
2144
2145         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2146                                    vmci_transport_qp_resumed_cb,
2147                                    NULL, &vmci_transport_qp_resumed_sub_id);
2148         if (err < VMCI_SUCCESS) {
2149                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2150                 err = vmci_transport_error_to_vsock_error(err);
2151                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2152                 goto err_destroy_stream_handle;
2153         }
2154
2155         err = vsock_core_init(&vmci_transport);
2156         if (err < 0)
2157                 goto err_unsubscribe;
2158
2159         return 0;
2160
2161 err_unsubscribe:
2162         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2163 err_destroy_stream_handle:
2164         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2165         return err;
2166 }
2167 module_init(vmci_transport_init);
2168
2169 static void __exit vmci_transport_exit(void)
2170 {
2171         cancel_work_sync(&vmci_transport_cleanup_work);
2172         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2173
2174         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2175                 if (vmci_datagram_destroy_handle(
2176                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2177                         pr_err("Couldn't destroy datagram handle\n");
2178                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2179         }
2180
2181         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2182                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2183                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2184         }
2185
2186         vsock_core_exit();
2187 }
2188 module_exit(vmci_transport_exit);
2189
2190 MODULE_AUTHOR("VMware, Inc.");
2191 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2192 MODULE_VERSION("1.0.3.0-k");
2193 MODULE_LICENSE("GPL v2");
2194 MODULE_ALIAS("vmware_vsock");
2195 MODULE_ALIAS_NETPROTO(PF_VSOCK);