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("Could not attach to queue pair with %d\n",
 597                        err);
 598                 err = vmci_transport_error_to_vsock_error(err);
 599         }
 600
 601         return err;
 602 }
 603
 604 static int
 605 vmci_transport_datagram_create_hnd(u32 resource_id,
 606                                    u32 flags,
 607                                    vmci_datagram_recv_cb recv_cb,
 608                                    void *client_data,
 609                                    struct vmci_handle *out_handle)
 610 {
 611         int err = 0;
 612
 613         /* Try to allocate our datagram handler as trusted. This will only work
 614          * if vsock is running in the host.
 615          */
 616
 617         err = vmci_datagram_create_handle_priv(resource_id, flags,
 618                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 619                                                recv_cb,
 620                                                client_data, out_handle);
 621
 622         if (err == VMCI_ERROR_NO_ACCESS)
 623                 err = vmci_datagram_create_handle(resource_id, flags,
 624                                                   recv_cb, client_data,
 625                                                   out_handle);
 626
 627         return err;
 628 }
 629
 630 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 631  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 632  * sleep it should defer that work to a work queue.
 633  */
 634
 635 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 636 {
 637         struct sock *sk;
 638         size_t size;
 639         struct sk_buff *skb;
 640         struct vsock_sock *vsk;
 641
 642         sk = (struct sock *)data;
 643
 644         /* This handler is privileged when this module is running on the host.
 645          * We will get datagrams from all endpoints (even VMs that are in a
 646          * restricted context). If we get one from a restricted context then
 647          * the destination socket must be trusted.
 648          *
 649          * NOTE: We access the socket struct without holding the lock here.
 650          * This is ok because the field we are interested is never modified
 651          * outside of the create and destruct socket functions.
 652          */
 653         vsk = vsock_sk(sk);
 654         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 655                 return VMCI_ERROR_NO_ACCESS;
 656
 657         size = VMCI_DG_SIZE(dg);
 658
 659         /* Attach the packet to the socket's receive queue as an sk_buff. */
 660         skb = alloc_skb(size, GFP_ATOMIC);
 661         if (!skb)
 662                 return VMCI_ERROR_NO_MEM;
 663
 664         /* sk_receive_skb() will do a sock_put(), so hold here. */
 665         sock_hold(sk);
 666         skb_put(skb, size);
 667         memcpy(skb->data, dg, size);
 668         sk_receive_skb(sk, skb, 0);
 669
 670         return VMCI_SUCCESS;
 671 }
 672
 673 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 674 {
 675         static const u32 non_socket_contexts[] = {
 676                 VMADDR_CID_RESERVED,
 677         };
 678         int i;
 679
 680         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 681
 682         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 683                 if (cid == non_socket_contexts[i])
 684                         return false;
 685         }
 686
 687         return true;
 688 }
 689
 690 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 691  * interrupt fires.  This is run in bottom-half context but it defers most of
 692  * its work to the packet handling work queue.
 693  */
 694
 695 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 696 {
 697         struct sock *sk;
 698         struct sockaddr_vm dst;
 699         struct sockaddr_vm src;
 700         struct vmci_transport_packet *pkt;
 701         struct vsock_sock *vsk;
 702         bool bh_process_pkt;
 703         int err;
 704
 705         sk = NULL;
 706         err = VMCI_SUCCESS;
 707         bh_process_pkt = false;
 708
 709         /* Ignore incoming packets from contexts without sockets, or resources
 710          * that aren't vsock implementations.
 711          */
 712
 713         if (!vmci_transport_stream_allow(dg->src.context, -1)
 714             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 715                 return VMCI_ERROR_NO_ACCESS;
 716
 717         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 718                 /* Drop datagrams that do not contain full VSock packets. */
 719                 return VMCI_ERROR_INVALID_ARGS;
 720
 721         pkt = (struct vmci_transport_packet *)dg;
 722
 723         /* Find the socket that should handle this packet.  First we look for a
 724          * connected socket and if there is none we look for a socket bound to
 725          * the destintation address.
 726          */
 727         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 728         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 729
 730         sk = vsock_find_connected_socket(&src, &dst);
 731         if (!sk) {
 732                 sk = vsock_find_bound_socket(&dst);
 733                 if (!sk) {
 734                         /* We could not find a socket for this specified
 735                          * address.  If this packet is a RST, we just drop it.
 736                          * If it is another packet, we send a RST.  Note that
 737                          * we do not send a RST reply to RSTs so that we do not
 738                          * continually send RSTs between two endpoints.
 739                          *
 740                          * Note that since this is a reply, dst is src and src
 741                          * is dst.
 742                          */
 743                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 744                                 pr_err("unable to send reset\n");
 745
 746                         err = VMCI_ERROR_NOT_FOUND;
 747                         goto out;
 748                 }
 749         }
 750
 751         /* If the received packet type is beyond all types known to this
 752          * implementation, reply with an invalid message.  Hopefully this will
 753          * help when implementing backwards compatibility in the future.
 754          */
 755         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 756                 vmci_transport_send_invalid_bh(&dst, &src);
 757                 err = VMCI_ERROR_INVALID_ARGS;
 758                 goto out;
 759         }
 760
 761         /* This handler is privileged when this module is running on the host.
 762          * We will get datagram connect requests from all endpoints (even VMs
 763          * that are in a restricted context). If we get one from a restricted
 764          * context then the destination socket must be trusted.
 765          *
 766          * NOTE: We access the socket struct without holding the lock here.
 767          * This is ok because the field we are interested is never modified
 768          * outside of the create and destruct socket functions.
 769          */
 770         vsk = vsock_sk(sk);
 771         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 772                 err = VMCI_ERROR_NO_ACCESS;
 773                 goto out;
 774         }
 775
 776         /* We do most everything in a work queue, but let's fast path the
 777          * notification of reads and writes to help data transfer performance.
 778          * We can only do this if there is no process context code executing
 779          * for this socket since that may change the state.
 780          */
 781         bh_lock_sock(sk);
 782
 783         if (!sock_owned_by_user(sk)) {
 784                 /* The local context ID may be out of date, update it. */
 785                 vsk->local_addr.svm_cid = dst.svm_cid;
 786
 787                 if (sk->sk_state == SS_CONNECTED)
 788                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 789                                         sk, pkt, true, &dst, &src,
 790                                         &bh_process_pkt);
 791         }
 792
 793         bh_unlock_sock(sk);
 794
 795         if (!bh_process_pkt) {
 796                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 797
 798                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 799                 if (!recv_pkt_info) {
 800                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 801                                 pr_err("unable to send reset\n");
 802
 803                         err = VMCI_ERROR_NO_MEM;
 804                         goto out;
 805                 }
 806
 807                 recv_pkt_info->sk = sk;
 808                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 809                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 810
 811                 schedule_work(&recv_pkt_info->work);
 812                 /* Clear sk so that the reference count incremented by one of
 813                  * the Find functions above is not decremented below.  We need
 814                  * that reference count for the packet handler we've scheduled
 815                  * to run.
 816                  */
 817                 sk = NULL;
 818         }
 819
 820 out:
 821         if (sk)
 822                 sock_put(sk);
 823
 824         return err;
 825 }
 826
 827 static void vmci_transport_handle_detach(struct sock *sk)
 828 {
 829         struct vsock_sock *vsk;
 830
 831         vsk = vsock_sk(sk);
 832         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 833                 sock_set_flag(sk, SOCK_DONE);
 834
 835                 /* On a detach the peer will not be sending or receiving
 836                  * anymore.
 837                  */
 838                 vsk->peer_shutdown = SHUTDOWN_MASK;
 839
 840                 /* We should not be sending anymore since the peer won't be
 841                  * there to receive, but we can still receive if there is data
 842                  * left in our consume queue.
 843                  */
 844                 if (vsock_stream_has_data(vsk) <= 0) {
 845                         if (sk->sk_state == SS_CONNECTING) {
 846                                 /* The peer may detach from a queue pair while
 847                                  * we are still in the connecting state, i.e.,
 848                                  * if the peer VM is killed after attaching to
 849                                  * a queue pair, but before we complete the
 850                                  * handshake. In that case, we treat the detach
 851                                  * event like a reset.
 852                                  */
 853
 854                                 sk->sk_state = SS_UNCONNECTED;
 855                                 sk->sk_err = ECONNRESET;
 856                                 sk->sk_error_report(sk);
 857                                 return;
 858                         }
 859                         sk->sk_state = SS_UNCONNECTED;
 860                 }
 861                 sk->sk_state_change(sk);
 862         }
 863 }
 864
 865 static void vmci_transport_peer_detach_cb(u32 sub_id,
 866                                           const struct vmci_event_data *e_data,
 867                                           void *client_data)
 868 {
 869         struct vmci_transport *trans = client_data;
 870         const struct vmci_event_payload_qp *e_payload;
 871
 872         e_payload = vmci_event_data_const_payload(e_data);
 873
 874         /* XXX This is lame, we should provide a way to lookup sockets by
 875          * qp_handle.
 876          */
 877         if (vmci_handle_is_invalid(e_payload->handle) ||
 878             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 879                 return;
 880
 881         /* We don't ask for delayed CBs when we subscribe to this event (we
 882          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 883          * guarantees in that case about what context we might be running in,
 884          * so it could be BH or process, blockable or non-blockable.  So we
 885          * need to account for all possible contexts here.
 886          */
 887         spin_lock_bh(&trans->lock);
 888         if (!trans->sk)
 889                 goto out;
 890
 891         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 892          * where trans->sk isn't locked.
 893          */
 894         bh_lock_sock(trans->sk);
 895
 896         vmci_transport_handle_detach(trans->sk);
 897
 898         bh_unlock_sock(trans->sk);
 899  out:
 900         spin_unlock_bh(&trans->lock);
 901 }
 902
 903 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 904                                          const struct vmci_event_data *e_data,
 905                                          void *client_data)
 906 {
 907         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 908 }
 909
 910 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 911 {
 912         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 913         struct vmci_transport_packet *pkt;
 914         struct sock *sk;
 915
 916         recv_pkt_info =
 917                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 918         sk = recv_pkt_info->sk;
 919         pkt = &recv_pkt_info->pkt;
 920
 921         lock_sock(sk);
 922
 923         /* The local context ID may be out of date. */
 924         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 925
 926         switch (sk->sk_state) {
 927         case VSOCK_SS_LISTEN:
 928                 vmci_transport_recv_listen(sk, pkt);
 929                 break;
 930         case SS_CONNECTING:
 931                 /* Processing of pending connections for servers goes through
 932                  * the listening socket, so see vmci_transport_recv_listen()
 933                  * for that path.
 934                  */
 935                 vmci_transport_recv_connecting_client(sk, pkt);
 936                 break;
 937         case SS_CONNECTED:
 938                 vmci_transport_recv_connected(sk, pkt);
 939                 break;
 940         default:
 941                 /* Because this function does not run in the same context as
 942                  * vmci_transport_recv_stream_cb it is possible that the
 943                  * socket has closed. We need to let the other side know or it
 944                  * could be sitting in a connect and hang forever. Send a
 945                  * reset to prevent that.
 946                  */
 947                 vmci_transport_send_reset(sk, pkt);
 948                 break;
 949         }
 950
 951         release_sock(sk);
 952         kfree(recv_pkt_info);
 953         /* Release reference obtained in the stream callback when we fetched
 954          * this socket out of the bound or connected list.
 955          */
 956         sock_put(sk);
 957 }
 958
 959 static int vmci_transport_recv_listen(struct sock *sk,
 960                                       struct vmci_transport_packet *pkt)
 961 {
 962         struct sock *pending;
 963         struct vsock_sock *vpending;
 964         int err;
 965         u64 qp_size;
 966         bool old_request = false;
 967         bool old_pkt_proto = false;
 968
 969         err = 0;
 970
 971         /* Because we are in the listen state, we could be receiving a packet
 972          * for ourself or any previous connection requests that we received.
 973          * If it's the latter, we try to find a socket in our list of pending
 974          * connections and, if we do, call the appropriate handler for the
 975          * state that that socket is in.  Otherwise we try to service the
 976          * connection request.
 977          */
 978         pending = vmci_transport_get_pending(sk, pkt);
 979         if (pending) {
 980                 lock_sock(pending);
 981
 982                 /* The local context ID may be out of date. */
 983                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 984
 985                 switch (pending->sk_state) {
 986                 case SS_CONNECTING:
 987                         err = vmci_transport_recv_connecting_server(sk,
 988                                                                     pending,
 989                                                                     pkt);
 990                         break;
 991                 default:
 992                         vmci_transport_send_reset(pending, pkt);
 993                         err = -EINVAL;
 994                 }
 995
 996                 if (err < 0)
 997                         vsock_remove_pending(sk, pending);
 998
 999                 release_sock(pending);
1000                 vmci_transport_release_pending(pending);
1001
1002                 return err;
1003         }
1004
1005         /* The listen state only accepts connection requests.  Reply with a
1006          * reset unless we received a reset.
1007          */
1008
1009         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
1010               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
1011                 vmci_transport_reply_reset(pkt);
1012                 return -EINVAL;
1013         }
1014
1015         if (pkt->u.size == 0) {
1016                 vmci_transport_reply_reset(pkt);
1017                 return -EINVAL;
1018         }
1019
1020         /* If this socket can't accommodate this connection request, we send a
1021          * reset.  Otherwise we create and initialize a child socket and reply
1022          * with a connection negotiation.
1023          */
1024         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1025                 vmci_transport_reply_reset(pkt);
1026                 return -ECONNREFUSED;
1027         }
1028
1029         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1030                                  sk->sk_type, 0);
1031         if (!pending) {
1032                 vmci_transport_send_reset(sk, pkt);
1033                 return -ENOMEM;
1034         }
1035
1036         vpending = vsock_sk(pending);
1037
1038         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1039                         pkt->dst_port);
1040         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1041                         pkt->src_port);
1042
1043         /* If the proposed size fits within our min/max, accept it. Otherwise
1044          * propose our own size.
1045          */
1046         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1047             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1048                 qp_size = pkt->u.size;
1049         } else {
1050                 qp_size = vmci_trans(vpending)->queue_pair_size;
1051         }
1052
1053         /* Figure out if we are using old or new requests based on the
1054          * overrides pkt types sent by our peer.
1055          */
1056         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1057                 old_request = old_pkt_proto;
1058         } else {
1059                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1060                         old_request = true;
1061                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1062                         old_request = false;
1063
1064         }
1065
1066         if (old_request) {
1067                 /* Handle a REQUEST (or override) */
1068                 u16 version = VSOCK_PROTO_INVALID;
1069                 if (vmci_transport_proto_to_notify_struct(
1070                         pending, &version, true))
1071                         err = vmci_transport_send_negotiate(pending, qp_size);
1072                 else
1073                         err = -EINVAL;
1074
1075         } else {
1076                 /* Handle a REQUEST2 (or override) */
1077                 int proto_int = pkt->proto;
1078                 int pos;
1079                 u16 active_proto_version = 0;
1080
1081                 /* The list of possible protocols is the intersection of all
1082                  * protocols the client supports ... plus all the protocols we
1083                  * support.
1084                  */
1085                 proto_int &= vmci_transport_new_proto_supported_versions();
1086
1087                 /* We choose the highest possible protocol version and use that
1088                  * one.
1089                  */
1090                 pos = fls(proto_int);
1091                 if (pos) {
1092                         active_proto_version = (1 << (pos - 1));
1093                         if (vmci_transport_proto_to_notify_struct(
1094                                 pending, &active_proto_version, false))
1095                                 err = vmci_transport_send_negotiate2(pending,
1096                                                         qp_size,
1097                                                         active_proto_version);
1098                         else
1099                                 err = -EINVAL;
1100
1101                 } else {
1102                         err = -EINVAL;
1103                 }
1104         }
1105
1106         if (err < 0) {
1107                 vmci_transport_send_reset(sk, pkt);
1108                 sock_put(pending);
1109                 err = vmci_transport_error_to_vsock_error(err);
1110                 goto out;
1111         }
1112
1113         vsock_add_pending(sk, pending);
1114         sk->sk_ack_backlog++;
1115
1116         pending->sk_state = SS_CONNECTING;
1117         vmci_trans(vpending)->produce_size =
1118                 vmci_trans(vpending)->consume_size = qp_size;
1119         vmci_trans(vpending)->queue_pair_size = qp_size;
1120
1121         vmci_trans(vpending)->notify_ops->process_request(pending);
1122
1123         /* We might never receive another message for this socket and it's not
1124          * connected to any process, so we have to ensure it gets cleaned up
1125          * ourself.  Our delayed work function will take care of that.  Note
1126          * that we do not ever cancel this function since we have few
1127          * guarantees about its state when calling cancel_delayed_work().
1128          * Instead we hold a reference on the socket for that function and make
1129          * it capable of handling cases where it needs to do nothing but
1130          * release that reference.
1131          */
1132         vpending->listener = sk;
1133         sock_hold(sk);
1134         sock_hold(pending);
1135         schedule_delayed_work(&vpending->pending_work, HZ);
1136
1137 out:
1138         return err;
1139 }
1140
1141 static int
1142 vmci_transport_recv_connecting_server(struct sock *listener,
1143                                       struct sock *pending,
1144                                       struct vmci_transport_packet *pkt)
1145 {
1146         struct vsock_sock *vpending;
1147         struct vmci_handle handle;
1148         struct vmci_qp *qpair;
1149         bool is_local;
1150         u32 flags;
1151         u32 detach_sub_id;
1152         int err;
1153         int skerr;
1154
1155         vpending = vsock_sk(pending);
1156         detach_sub_id = VMCI_INVALID_ID;
1157
1158         switch (pkt->type) {
1159         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1160                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1161                         vmci_transport_send_reset(pending, pkt);
1162                         skerr = EPROTO;
1163                         err = -EINVAL;
1164                         goto destroy;
1165                 }
1166                 break;
1167         default:
1168                 /* Close and cleanup the connection. */
1169                 vmci_transport_send_reset(pending, pkt);
1170                 skerr = EPROTO;
1171                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1172                 goto destroy;
1173         }
1174
1175         /* In order to complete the connection we need to attach to the offered
1176          * queue pair and send an attach notification.  We also subscribe to the
1177          * detach event so we know when our peer goes away, and we do that
1178          * before attaching so we don't miss an event.  If all this succeeds,
1179          * we update our state and wakeup anything waiting in accept() for a
1180          * connection.
1181          */
1182
1183         /* We don't care about attach since we ensure the other side has
1184          * attached by specifying the ATTACH_ONLY flag below.
1185          */
1186         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1187                                    vmci_transport_peer_detach_cb,
1188                                    vmci_trans(vpending), &detach_sub_id);
1189         if (err < VMCI_SUCCESS) {
1190                 vmci_transport_send_reset(pending, pkt);
1191                 err = vmci_transport_error_to_vsock_error(err);
1192                 skerr = -err;
1193                 goto destroy;
1194         }
1195
1196         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1197
1198         /* Now attach to the queue pair the client created. */
1199         handle = pkt->u.handle;
1200
1201         /* vpending->local_addr always has a context id so we do not need to
1202          * worry about VMADDR_CID_ANY in this case.
1203          */
1204         is_local =
1205             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1206         flags = VMCI_QPFLAG_ATTACH_ONLY;
1207         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1208
1209         err = vmci_transport_queue_pair_alloc(
1210                                         &qpair,
1211                                         &handle,
1212                                         vmci_trans(vpending)->produce_size,
1213                                         vmci_trans(vpending)->consume_size,
1214                                         pkt->dg.src.context,
1215                                         flags,
1216                                         vmci_transport_is_trusted(
1217                                                 vpending,
1218                                                 vpending->remote_addr.svm_cid));
1219         if (err < 0) {
1220                 vmci_transport_send_reset(pending, pkt);
1221                 skerr = -err;
1222                 goto destroy;
1223         }
1224
1225         vmci_trans(vpending)->qp_handle = handle;
1226         vmci_trans(vpending)->qpair = qpair;
1227
1228         /* When we send the attach message, we must be ready to handle incoming
1229          * control messages on the newly connected socket. So we move the
1230          * pending socket to the connected state before sending the attach
1231          * message. Otherwise, an incoming packet triggered by the attach being
1232          * received by the peer may be processed concurrently with what happens
1233          * below after sending the attach message, and that incoming packet
1234          * will find the listening socket instead of the (currently) pending
1235          * socket. Note that enqueueing the socket increments the reference
1236          * count, so even if a reset comes before the connection is accepted,
1237          * the socket will be valid until it is removed from the queue.
1238          *
1239          * If we fail sending the attach below, we remove the socket from the
1240          * connected list and move the socket to SS_UNCONNECTED before
1241          * releasing the lock, so a pending slow path processing of an incoming
1242          * packet will not see the socket in the connected state in that case.
1243          */
1244         pending->sk_state = SS_CONNECTED;
1245
1246         vsock_insert_connected(vpending);
1247
1248         /* Notify our peer of our attach. */
1249         err = vmci_transport_send_attach(pending, handle);
1250         if (err < 0) {
1251                 vsock_remove_connected(vpending);
1252                 pr_err("Could not send attach\n");
1253                 vmci_transport_send_reset(pending, pkt);
1254                 err = vmci_transport_error_to_vsock_error(err);
1255                 skerr = -err;
1256                 goto destroy;
1257         }
1258
1259         /* We have a connection. Move the now connected socket from the
1260          * listener's pending list to the accept queue so callers of accept()
1261          * can find it.
1262          */
1263         vsock_remove_pending(listener, pending);
1264         vsock_enqueue_accept(listener, pending);
1265
1266         /* Callers of accept() will be be waiting on the listening socket, not
1267          * the pending socket.
1268          */
1269         listener->sk_data_ready(listener);
1270
1271         return 0;
1272
1273 destroy:
1274         pending->sk_err = skerr;
1275         pending->sk_state = SS_UNCONNECTED;
1276         /* As long as we drop our reference, all necessary cleanup will handle
1277          * when the cleanup function drops its reference and our destruct
1278          * implementation is called.  Note that since the listen handler will
1279          * remove pending from the pending list upon our failure, the cleanup
1280          * function won't drop the additional reference, which is why we do it
1281          * here.
1282          */
1283         sock_put(pending);
1284
1285         return err;
1286 }
1287
1288 static int
1289 vmci_transport_recv_connecting_client(struct sock *sk,
1290                                       struct vmci_transport_packet *pkt)
1291 {
1292         struct vsock_sock *vsk;
1293         int err;
1294         int skerr;
1295
1296         vsk = vsock_sk(sk);
1297
1298         switch (pkt->type) {
1299         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1300                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1301                     !vmci_handle_is_equal(pkt->u.handle,
1302                                           vmci_trans(vsk)->qp_handle)) {
1303                         skerr = EPROTO;
1304                         err = -EINVAL;
1305                         goto destroy;
1306                 }
1307
1308                 /* Signify the socket is connected and wakeup the waiter in
1309                  * connect(). Also place the socket in the connected table for
1310                  * accounting (it can already be found since it's in the bound
1311                  * table).
1312                  */
1313                 sk->sk_state = SS_CONNECTED;
1314                 sk->sk_socket->state = SS_CONNECTED;
1315                 vsock_insert_connected(vsk);
1316                 sk->sk_state_change(sk);
1317
1318                 break;
1319         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1320         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1321                 if (pkt->u.size == 0
1322                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1323                     || pkt->src_port != vsk->remote_addr.svm_port
1324                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1325                     || vmci_trans(vsk)->qpair
1326                     || vmci_trans(vsk)->produce_size != 0
1327                     || vmci_trans(vsk)->consume_size != 0
1328                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1329                         skerr = EPROTO;
1330                         err = -EINVAL;
1331
1332                         goto destroy;
1333                 }
1334
1335                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1336                 if (err) {
1337                         skerr = -err;
1338                         goto destroy;
1339                 }
1340
1341                 break;
1342         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1343                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1344                 if (err) {
1345                         skerr = -err;
1346                         goto destroy;
1347                 }
1348
1349                 break;
1350         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1351                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1352                  * continue processing here after they sent an INVALID packet.
1353                  * This meant that we got a RST after the INVALID. We ignore a
1354                  * RST after an INVALID. The common code doesn't send the RST
1355                  * ... so we can hang if an old version of the common code
1356                  * fails between getting a REQUEST and sending an OFFER back.
1357                  * Not much we can do about it... except hope that it doesn't
1358                  * happen.
1359                  */
1360                 if (vsk->ignore_connecting_rst) {
1361                         vsk->ignore_connecting_rst = false;
1362                 } else {
1363                         skerr = ECONNRESET;
1364                         err = 0;
1365                         goto destroy;
1366                 }
1367
1368                 break;
1369         default:
1370                 /* Close and cleanup the connection. */
1371                 skerr = EPROTO;
1372                 err = -EINVAL;
1373                 goto destroy;
1374         }
1375
1376         return 0;
1377
1378 destroy:
1379         vmci_transport_send_reset(sk, pkt);
1380
1381         sk->sk_state = SS_UNCONNECTED;
1382         sk->sk_err = skerr;
1383         sk->sk_error_report(sk);
1384         return err;
1385 }
1386
1387 static int vmci_transport_recv_connecting_client_negotiate(
1388                                         struct sock *sk,
1389                                         struct vmci_transport_packet *pkt)
1390 {
1391         int err;
1392         struct vsock_sock *vsk;
1393         struct vmci_handle handle;
1394         struct vmci_qp *qpair;
1395         u32 detach_sub_id;
1396         bool is_local;
1397         u32 flags;
1398         bool old_proto = true;
1399         bool old_pkt_proto;
1400         u16 version;
1401
1402         vsk = vsock_sk(sk);
1403         handle = VMCI_INVALID_HANDLE;
1404         detach_sub_id = VMCI_INVALID_ID;
1405
1406         /* If we have gotten here then we should be past the point where old
1407          * linux vsock could have sent the bogus rst.
1408          */
1409         vsk->sent_request = false;
1410         vsk->ignore_connecting_rst = false;
1411
1412         /* Verify that we're OK with the proposed queue pair size */
1413         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1414             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1415                 err = -EINVAL;
1416                 goto destroy;
1417         }
1418
1419         /* At this point we know the CID the peer is using to talk to us. */
1420
1421         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1422                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1423
1424         /* Setup the notify ops to be the highest supported version that both
1425          * the server and the client support.
1426          */
1427
1428         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1429                 old_proto = old_pkt_proto;
1430         } else {
1431                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1432                         old_proto = true;
1433                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1434                         old_proto = false;
1435
1436         }
1437
1438         if (old_proto)
1439                 version = VSOCK_PROTO_INVALID;
1440         else
1441                 version = pkt->proto;
1442
1443         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1444                 err = -EINVAL;
1445                 goto destroy;
1446         }
1447
1448         /* Subscribe to detach events first.
1449          *
1450          * XXX We attach once for each queue pair created for now so it is easy
1451          * to find the socket (it's provided), but later we should only
1452          * subscribe once and add a way to lookup sockets by queue pair handle.
1453          */
1454         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1455                                    vmci_transport_peer_detach_cb,
1456                                    vmci_trans(vsk), &detach_sub_id);
1457         if (err < VMCI_SUCCESS) {
1458                 err = vmci_transport_error_to_vsock_error(err);
1459                 goto destroy;
1460         }
1461
1462         /* Make VMCI select the handle for us. */
1463         handle = VMCI_INVALID_HANDLE;
1464         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1465         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1466
1467         err = vmci_transport_queue_pair_alloc(&qpair,
1468                                               &handle,
1469                                               pkt->u.size,
1470                                               pkt->u.size,
1471                                               vsk->remote_addr.svm_cid,
1472                                               flags,
1473                                               vmci_transport_is_trusted(
1474                                                   vsk,
1475                                                   vsk->
1476                                                   remote_addr.svm_cid));
1477         if (err < 0)
1478                 goto destroy;
1479
1480         err = vmci_transport_send_qp_offer(sk, handle);
1481         if (err < 0) {
1482                 err = vmci_transport_error_to_vsock_error(err);
1483                 goto destroy;
1484         }
1485
1486         vmci_trans(vsk)->qp_handle = handle;
1487         vmci_trans(vsk)->qpair = qpair;
1488
1489         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1490                 pkt->u.size;
1491
1492         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1493
1494         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1495
1496         return 0;
1497
1498 destroy:
1499         if (detach_sub_id != VMCI_INVALID_ID)
1500                 vmci_event_unsubscribe(detach_sub_id);
1501
1502         if (!vmci_handle_is_invalid(handle))
1503                 vmci_qpair_detach(&qpair);
1504
1505         return err;
1506 }
1507
1508 static int
1509 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1510                                               struct vmci_transport_packet *pkt)
1511 {
1512         int err = 0;
1513         struct vsock_sock *vsk = vsock_sk(sk);
1514
1515         if (vsk->sent_request) {
1516                 vsk->sent_request = false;
1517                 vsk->ignore_connecting_rst = true;
1518
1519                 err = vmci_transport_send_conn_request(
1520                         sk, vmci_trans(vsk)->queue_pair_size);
1521                 if (err < 0)
1522                         err = vmci_transport_error_to_vsock_error(err);
1523                 else
1524                         err = 0;
1525
1526         }
1527
1528         return err;
1529 }
1530
1531 static int vmci_transport_recv_connected(struct sock *sk,
1532                                          struct vmci_transport_packet *pkt)
1533 {
1534         struct vsock_sock *vsk;
1535         bool pkt_processed = false;
1536
1537         /* In cases where we are closing the connection, it's sufficient to
1538          * mark the state change (and maybe error) and wake up any waiting
1539          * threads. Since this is a connected socket, it's owned by a user
1540          * process and will be cleaned up when the failure is passed back on
1541          * the current or next system call.  Our system call implementations
1542          * must therefore check for error and state changes on entry and when
1543          * being awoken.
1544          */
1545         switch (pkt->type) {
1546         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1547                 if (pkt->u.mode) {
1548                         vsk = vsock_sk(sk);
1549
1550                         vsk->peer_shutdown |= pkt->u.mode;
1551                         sk->sk_state_change(sk);
1552                 }
1553                 break;
1554
1555         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1556                 vsk = vsock_sk(sk);
1557                 /* It is possible that we sent our peer a message (e.g a
1558                  * WAITING_READ) right before we got notified that the peer had
1559                  * detached. If that happens then we can get a RST pkt back
1560                  * from our peer even though there is data available for us to
1561                  * read. In that case, don't shutdown the socket completely but
1562                  * instead allow the local client to finish reading data off
1563                  * the queuepair. Always treat a RST pkt in connected mode like
1564                  * a clean shutdown.
1565                  */
1566                 sock_set_flag(sk, SOCK_DONE);
1567                 vsk->peer_shutdown = SHUTDOWN_MASK;
1568                 if (vsock_stream_has_data(vsk) <= 0)
1569                         sk->sk_state = SS_DISCONNECTING;
1570
1571                 sk->sk_state_change(sk);
1572                 break;
1573
1574         default:
1575                 vsk = vsock_sk(sk);
1576                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1577                                 sk, pkt, false, NULL, NULL,
1578                                 &pkt_processed);
1579                 if (!pkt_processed)
1580                         return -EINVAL;
1581
1582                 break;
1583         }
1584
1585         return 0;
1586 }
1587
1588 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1589                                       struct vsock_sock *psk)
1590 {
1591         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1592         if (!vsk->trans)
1593                 return -ENOMEM;
1594
1595         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1596         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1597         vmci_trans(vsk)->qpair = NULL;
1598         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1599         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1600         vmci_trans(vsk)->notify_ops = NULL;
1601         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1602         vmci_trans(vsk)->sk = &vsk->sk;
1603         spin_lock_init(&vmci_trans(vsk)->lock);
1604         if (psk) {
1605                 vmci_trans(vsk)->queue_pair_size =
1606                         vmci_trans(psk)->queue_pair_size;
1607                 vmci_trans(vsk)->queue_pair_min_size =
1608                         vmci_trans(psk)->queue_pair_min_size;
1609                 vmci_trans(vsk)->queue_pair_max_size =
1610                         vmci_trans(psk)->queue_pair_max_size;
1611         } else {
1612                 vmci_trans(vsk)->queue_pair_size =
1613                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1614                 vmci_trans(vsk)->queue_pair_min_size =
1615                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1616                 vmci_trans(vsk)->queue_pair_max_size =
1617                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1618         }
1619
1620         return 0;
1621 }
1622
1623 static void vmci_transport_free_resources(struct list_head *transport_list)
1624 {
1625         while (!list_empty(transport_list)) {
1626                 struct vmci_transport *transport =
1627                     list_first_entry(transport_list, struct vmci_transport,
1628                                      elem);
1629                 list_del(&transport->elem);
1630
1631                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1632                         vmci_event_unsubscribe(transport->detach_sub_id);
1633                         transport->detach_sub_id = VMCI_INVALID_ID;
1634                 }
1635
1636                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1637                         vmci_qpair_detach(&transport->qpair);
1638                         transport->qp_handle = VMCI_INVALID_HANDLE;
1639                         transport->produce_size = 0;
1640                         transport->consume_size = 0;
1641                 }
1642
1643                 kfree(transport);
1644         }
1645 }
1646
1647 static void vmci_transport_cleanup(struct work_struct *work)
1648 {
1649         LIST_HEAD(pending);
1650
1651         spin_lock_bh(&vmci_transport_cleanup_lock);
1652         list_replace_init(&vmci_transport_cleanup_list, &pending);
1653         spin_unlock_bh(&vmci_transport_cleanup_lock);
1654         vmci_transport_free_resources(&pending);
1655 }
1656
1657 static void vmci_transport_destruct(struct vsock_sock *vsk)
1658 {
1659         /* transport can be NULL if we hit a failure at init() time */
1660         if (!vmci_trans(vsk))
1661                 return;
1662
1663         /* Ensure that the detach callback doesn't use the sk/vsk
1664          * we are about to destruct.
1665          */
1666         spin_lock_bh(&vmci_trans(vsk)->lock);
1667         vmci_trans(vsk)->sk = NULL;
1668         spin_unlock_bh(&vmci_trans(vsk)->lock);
1669
1670         if (vmci_trans(vsk)->notify_ops)
1671                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1672
1673         spin_lock_bh(&vmci_transport_cleanup_lock);
1674         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1675         spin_unlock_bh(&vmci_transport_cleanup_lock);
1676         schedule_work(&vmci_transport_cleanup_work);
1677
1678         vsk->trans = NULL;
1679 }
1680
1681 static void vmci_transport_release(struct vsock_sock *vsk)
1682 {
1683         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1684                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1685                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1686         }
1687 }
1688
1689 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1690                                      struct sockaddr_vm *addr)
1691 {
1692         u32 port;
1693         u32 flags;
1694         int err;
1695
1696         /* VMCI will select a resource ID for us if we provide
1697          * VMCI_INVALID_ID.
1698          */
1699         port = addr->svm_port == VMADDR_PORT_ANY ?
1700                         VMCI_INVALID_ID : addr->svm_port;
1701
1702         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1703                 return -EACCES;
1704
1705         flags = addr->svm_cid == VMADDR_CID_ANY ?
1706                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1707
1708         err = vmci_transport_datagram_create_hnd(port, flags,
1709                                                  vmci_transport_recv_dgram_cb,
1710                                                  &vsk->sk,
1711                                                  &vmci_trans(vsk)->dg_handle);
1712         if (err < VMCI_SUCCESS)
1713                 return vmci_transport_error_to_vsock_error(err);
1714         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1715                         vmci_trans(vsk)->dg_handle.resource);
1716
1717         return 0;
1718 }
1719
1720 static int vmci_transport_dgram_enqueue(
1721         struct vsock_sock *vsk,
1722         struct sockaddr_vm *remote_addr,
1723         struct msghdr *msg,
1724         size_t len)
1725 {
1726         int err;
1727         struct vmci_datagram *dg;
1728
1729         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1730                 return -EMSGSIZE;
1731
1732         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1733                 return -EPERM;
1734
1735         /* Allocate a buffer for the user's message and our packet header. */
1736         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1737         if (!dg)
1738                 return -ENOMEM;
1739
1740         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1741
1742         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1743                                    remote_addr->svm_port);
1744         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1745                                    vsk->local_addr.svm_port);
1746         dg->payload_size = len;
1747
1748         err = vmci_datagram_send(dg);
1749         kfree(dg);
1750         if (err < 0)
1751                 return vmci_transport_error_to_vsock_error(err);
1752
1753         return err - sizeof(*dg);
1754 }
1755
1756 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1757                                         struct msghdr *msg, size_t len,
1758                                         int flags)
1759 {
1760         int err;
1761         int noblock;
1762         struct vmci_datagram *dg;
1763         size_t payload_len;
1764         struct sk_buff *skb;
1765
1766         noblock = flags & MSG_DONTWAIT;
1767
1768         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1769                 return -EOPNOTSUPP;
1770
1771         /* Retrieve the head sk_buff from the socket's receive queue. */
1772         err = 0;
1773         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1774         if (err)
1775                 return err;
1776
1777         if (!skb)
1778                 return -EAGAIN;
1779
1780         dg = (struct vmci_datagram *)skb->data;
1781         if (!dg)
1782                 /* err is 0, meaning we read zero bytes. */
1783                 goto out;
1784
1785         payload_len = dg->payload_size;
1786         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1787         if (payload_len != skb->len - sizeof(*dg)) {
1788                 err = -EINVAL;
1789                 goto out;
1790         }
1791
1792         if (payload_len > len) {
1793                 payload_len = len;
1794                 msg->msg_flags |= MSG_TRUNC;
1795         }
1796
1797         /* Place the datagram payload in the user's iovec. */
1798         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1799         if (err)
1800                 goto out;
1801
1802         if (msg->msg_name) {
1803                 /* Provide the address of the sender. */
1804                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1805                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1806                 msg->msg_namelen = sizeof(*vm_addr);
1807         }
1808         err = payload_len;
1809
1810 out:
1811         skb_free_datagram(&vsk->sk, skb);
1812         return err;
1813 }
1814
1815 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1816 {
1817         if (cid == VMADDR_CID_HYPERVISOR) {
1818                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1819                  * state and are allowed.
1820                  */
1821                 return port == VMCI_UNITY_PBRPC_REGISTER;
1822         }
1823
1824         return true;
1825 }
1826
1827 static int vmci_transport_connect(struct vsock_sock *vsk)
1828 {
1829         int err;
1830         bool old_pkt_proto = false;
1831         struct sock *sk = &vsk->sk;
1832
1833         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1834                 old_pkt_proto) {
1835                 err = vmci_transport_send_conn_request(
1836                         sk, vmci_trans(vsk)->queue_pair_size);
1837                 if (err < 0) {
1838                         sk->sk_state = SS_UNCONNECTED;
1839                         return err;
1840                 }
1841         } else {
1842                 int supported_proto_versions =
1843                         vmci_transport_new_proto_supported_versions();
1844                 err = vmci_transport_send_conn_request2(
1845                                 sk, vmci_trans(vsk)->queue_pair_size,
1846                                 supported_proto_versions);
1847                 if (err < 0) {
1848                         sk->sk_state = SS_UNCONNECTED;
1849                         return err;
1850                 }
1851
1852                 vsk->sent_request = true;
1853         }
1854
1855         return err;
1856 }
1857
1858 static ssize_t vmci_transport_stream_dequeue(
1859         struct vsock_sock *vsk,
1860         struct msghdr *msg,
1861         size_t len,
1862         int flags)
1863 {
1864         if (flags & MSG_PEEK)
1865                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1866         else
1867                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1868 }
1869
1870 static ssize_t vmci_transport_stream_enqueue(
1871         struct vsock_sock *vsk,
1872         struct msghdr *msg,
1873         size_t len)
1874 {
1875         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1876 }
1877
1878 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1879 {
1880         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1881 }
1882
1883 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1884 {
1885         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1886 }
1887
1888 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1889 {
1890         return vmci_trans(vsk)->consume_size;
1891 }
1892
1893 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1894 {
1895         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1896 }
1897
1898 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1899 {
1900         return vmci_trans(vsk)->queue_pair_size;
1901 }
1902
1903 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1904 {
1905         return vmci_trans(vsk)->queue_pair_min_size;
1906 }
1907
1908 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1909 {
1910         return vmci_trans(vsk)->queue_pair_max_size;
1911 }
1912
1913 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1914 {
1915         if (val < vmci_trans(vsk)->queue_pair_min_size)
1916                 vmci_trans(vsk)->queue_pair_min_size = val;
1917         if (val > vmci_trans(vsk)->queue_pair_max_size)
1918                 vmci_trans(vsk)->queue_pair_max_size = val;
1919         vmci_trans(vsk)->queue_pair_size = val;
1920 }
1921
1922 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1923                                                u64 val)
1924 {
1925         if (val > vmci_trans(vsk)->queue_pair_size)
1926                 vmci_trans(vsk)->queue_pair_size = val;
1927         vmci_trans(vsk)->queue_pair_min_size = val;
1928 }
1929
1930 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1931                                                u64 val)
1932 {
1933         if (val < vmci_trans(vsk)->queue_pair_size)
1934                 vmci_trans(vsk)->queue_pair_size = val;
1935         vmci_trans(vsk)->queue_pair_max_size = val;
1936 }
1937
1938 static int vmci_transport_notify_poll_in(
1939         struct vsock_sock *vsk,
1940         size_t target,
1941         bool *data_ready_now)
1942 {
1943         return vmci_trans(vsk)->notify_ops->poll_in(
1944                         &vsk->sk, target, data_ready_now);
1945 }
1946
1947 static int vmci_transport_notify_poll_out(
1948         struct vsock_sock *vsk,
1949         size_t target,
1950         bool *space_available_now)
1951 {
1952         return vmci_trans(vsk)->notify_ops->poll_out(
1953                         &vsk->sk, target, space_available_now);
1954 }
1955
1956 static int vmci_transport_notify_recv_init(
1957         struct vsock_sock *vsk,
1958         size_t target,
1959         struct vsock_transport_recv_notify_data *data)
1960 {
1961         return vmci_trans(vsk)->notify_ops->recv_init(
1962                         &vsk->sk, target,
1963                         (struct vmci_transport_recv_notify_data *)data);
1964 }
1965
1966 static int vmci_transport_notify_recv_pre_block(
1967         struct vsock_sock *vsk,
1968         size_t target,
1969         struct vsock_transport_recv_notify_data *data)
1970 {
1971         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1972                         &vsk->sk, target,
1973                         (struct vmci_transport_recv_notify_data *)data);
1974 }
1975
1976 static int vmci_transport_notify_recv_pre_dequeue(
1977         struct vsock_sock *vsk,
1978         size_t target,
1979         struct vsock_transport_recv_notify_data *data)
1980 {
1981         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1982                         &vsk->sk, target,
1983                         (struct vmci_transport_recv_notify_data *)data);
1984 }
1985
1986 static int vmci_transport_notify_recv_post_dequeue(
1987         struct vsock_sock *vsk,
1988         size_t target,
1989         ssize_t copied,
1990         bool data_read,
1991         struct vsock_transport_recv_notify_data *data)
1992 {
1993         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1994                         &vsk->sk, target, copied, data_read,
1995                         (struct vmci_transport_recv_notify_data *)data);
1996 }
1997
1998 static int vmci_transport_notify_send_init(
1999         struct vsock_sock *vsk,
2000         struct vsock_transport_send_notify_data *data)
2001 {
2002         return vmci_trans(vsk)->notify_ops->send_init(
2003                         &vsk->sk,
2004                         (struct vmci_transport_send_notify_data *)data);
2005 }
2006
2007 static int vmci_transport_notify_send_pre_block(
2008         struct vsock_sock *vsk,
2009         struct vsock_transport_send_notify_data *data)
2010 {
2011         return vmci_trans(vsk)->notify_ops->send_pre_block(
2012                         &vsk->sk,
2013                         (struct vmci_transport_send_notify_data *)data);
2014 }
2015
2016 static int vmci_transport_notify_send_pre_enqueue(
2017         struct vsock_sock *vsk,
2018         struct vsock_transport_send_notify_data *data)
2019 {
2020         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2021                         &vsk->sk,
2022                         (struct vmci_transport_send_notify_data *)data);
2023 }
2024
2025 static int vmci_transport_notify_send_post_enqueue(
2026         struct vsock_sock *vsk,
2027         ssize_t written,
2028         struct vsock_transport_send_notify_data *data)
2029 {
2030         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2031                         &vsk->sk, written,
2032                         (struct vmci_transport_send_notify_data *)data);
2033 }
2034
2035 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2036 {
2037         if (PROTOCOL_OVERRIDE != -1) {
2038                 if (PROTOCOL_OVERRIDE == 0)
2039                         *old_pkt_proto = true;
2040                 else
2041                         *old_pkt_proto = false;
2042
2043                 pr_info("Proto override in use\n");
2044                 return true;
2045         }
2046
2047         return false;
2048 }
2049
2050 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2051                                                   u16 *proto,
2052                                                   bool old_pkt_proto)
2053 {
2054         struct vsock_sock *vsk = vsock_sk(sk);
2055
2056         if (old_pkt_proto) {
2057                 if (*proto != VSOCK_PROTO_INVALID) {
2058                         pr_err("Can't set both an old and new protocol\n");
2059                         return false;
2060                 }
2061                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2062                 goto exit;
2063         }
2064
2065         switch (*proto) {
2066         case VSOCK_PROTO_PKT_ON_NOTIFY:
2067                 vmci_trans(vsk)->notify_ops =
2068                         &vmci_transport_notify_pkt_q_state_ops;
2069                 break;
2070         default:
2071                 pr_err("Unknown notify protocol version\n");
2072                 return false;
2073         }
2074
2075 exit:
2076         vmci_trans(vsk)->notify_ops->socket_init(sk);
2077         return true;
2078 }
2079
2080 static u16 vmci_transport_new_proto_supported_versions(void)
2081 {
2082         if (PROTOCOL_OVERRIDE != -1)
2083                 return PROTOCOL_OVERRIDE;
2084
2085         return VSOCK_PROTO_ALL_SUPPORTED;
2086 }
2087
2088 static u32 vmci_transport_get_local_cid(void)
2089 {
2090         return vmci_get_context_id();
2091 }
2092
2093 static struct vsock_transport vmci_transport = {
2094         .init = vmci_transport_socket_init,
2095         .destruct = vmci_transport_destruct,
2096         .release = vmci_transport_release,
2097         .connect = vmci_transport_connect,
2098         .dgram_bind = vmci_transport_dgram_bind,
2099         .dgram_dequeue = vmci_transport_dgram_dequeue,
2100         .dgram_enqueue = vmci_transport_dgram_enqueue,
2101         .dgram_allow = vmci_transport_dgram_allow,
2102         .stream_dequeue = vmci_transport_stream_dequeue,
2103         .stream_enqueue = vmci_transport_stream_enqueue,
2104         .stream_has_data = vmci_transport_stream_has_data,
2105         .stream_has_space = vmci_transport_stream_has_space,
2106         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2107         .stream_is_active = vmci_transport_stream_is_active,
2108         .stream_allow = vmci_transport_stream_allow,
2109         .notify_poll_in = vmci_transport_notify_poll_in,
2110         .notify_poll_out = vmci_transport_notify_poll_out,
2111         .notify_recv_init = vmci_transport_notify_recv_init,
2112         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2113         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2114         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2115         .notify_send_init = vmci_transport_notify_send_init,
2116         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2117         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2118         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2119         .shutdown = vmci_transport_shutdown,
2120         .set_buffer_size = vmci_transport_set_buffer_size,
2121         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2122         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2123         .get_buffer_size = vmci_transport_get_buffer_size,
2124         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2125         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2126         .get_local_cid = vmci_transport_get_local_cid,
2127 };
2128
2129 static int __init vmci_transport_init(void)
2130 {
2131         int err;
2132
2133         /* Create the datagram handle that we will use to send and receive all
2134          * VSocket control messages for this context.
2135          */
2136         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2137                                                  VMCI_FLAG_ANYCID_DG_HND,
2138                                                  vmci_transport_recv_stream_cb,
2139                                                  NULL,
2140                                                  &vmci_transport_stream_handle);
2141         if (err < VMCI_SUCCESS) {
2142                 pr_err("Unable to create datagram handle. (%d)\n", err);
2143                 return vmci_transport_error_to_vsock_error(err);
2144         }
2145
2146         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2147                                    vmci_transport_qp_resumed_cb,
2148                                    NULL, &vmci_transport_qp_resumed_sub_id);
2149         if (err < VMCI_SUCCESS) {
2150                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2151                 err = vmci_transport_error_to_vsock_error(err);
2152                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2153                 goto err_destroy_stream_handle;
2154         }
2155
2156         err = vsock_core_init(&vmci_transport);
2157         if (err < 0)
2158                 goto err_unsubscribe;
2159
2160         return 0;
2161
2162 err_unsubscribe:
2163         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2164 err_destroy_stream_handle:
2165         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2166         return err;
2167 }
2168 module_init(vmci_transport_init);
2169
2170 static void __exit vmci_transport_exit(void)
2171 {
2172         cancel_work_sync(&vmci_transport_cleanup_work);
2173         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2174
2175         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2176                 if (vmci_datagram_destroy_handle(
2177                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2178                         pr_err("Couldn't destroy datagram handle\n");
2179                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2180         }
2181
2182         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2183                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2184                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2185         }
2186
2187         vsock_core_exit();
2188 }
2189 module_exit(vmci_transport_exit);
2190
2191 MODULE_AUTHOR("VMware, Inc.");
2192 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2193 MODULE_VERSION("1.0.3.0-k");
2194 MODULE_LICENSE("GPL v2");
2195 MODULE_ALIAS("vmware_vsock");
2196 MODULE_ALIAS_NETPROTO(PF_VSOCK);