Pileus Git - ~andy/linux/blob - 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
  38 #include "af_vsock.h"
  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_attach_cb(u32 sub_id,
  44                                           const struct vmci_event_data *ed,
  45                                           void *client_data);
  46 static void vmci_transport_peer_detach_cb(u32 sub_id,
  47                                           const struct vmci_event_data *ed,
  48                                           void *client_data);
  49 static void vmci_transport_recv_pkt_work(struct work_struct *work);
  50 static int vmci_transport_recv_listen(struct sock *sk,
  51                                       struct vmci_transport_packet *pkt);
  52 static int vmci_transport_recv_connecting_server(
  53                                         struct sock *sk,
  54                                         struct sock *pending,
  55                                         struct vmci_transport_packet *pkt);
  56 static int vmci_transport_recv_connecting_client(
  57                                         struct sock *sk,
  58                                         struct vmci_transport_packet *pkt);
  59 static int vmci_transport_recv_connecting_client_negotiate(
  60                                         struct sock *sk,
  61                                         struct vmci_transport_packet *pkt);
  62 static int vmci_transport_recv_connecting_client_invalid(
  63                                         struct sock *sk,
  64                                         struct vmci_transport_packet *pkt);
  65 static int vmci_transport_recv_connected(struct sock *sk,
  66                                          struct vmci_transport_packet *pkt);
  67 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
  68 static u16 vmci_transport_new_proto_supported_versions(void);
  69 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
  70                                                   bool old_pkt_proto);
  71
  72 struct vmci_transport_recv_pkt_info {
  73         struct work_struct work;
  74         struct sock *sk;
  75         struct vmci_transport_packet pkt;
  76 };
  77
  78 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  79                                                            VMCI_INVALID_ID };
  80 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  81
  82 static int PROTOCOL_OVERRIDE = -1;
  83
  84 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN   128
  85 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE       262144
  86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX   262144
  87
  88 /* The default peer timeout indicates how long we will wait for a peer response
  89  * to a control message.
  90  */
  91 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
  92
  93 #define SS_LISTEN 255
  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 inline void
 127 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 128                            struct sockaddr_vm *src,
 129                            struct sockaddr_vm *dst,
 130                            u8 type,
 131                            u64 size,
 132                            u64 mode,
 133                            struct vmci_transport_waiting_info *wait,
 134                            u16 proto,
 135                            struct vmci_handle handle)
 136 {
 137         /* We register the stream control handler as an any cid handle so we
 138          * must always send from a source address of VMADDR_CID_ANY
 139          */
 140         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 141                                        VMCI_TRANSPORT_PACKET_RID);
 142         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 143                                        VMCI_TRANSPORT_PACKET_RID);
 144         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 145         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 146         pkt->type = type;
 147         pkt->src_port = src->svm_port;
 148         pkt->dst_port = dst->svm_port;
 149         memset(&pkt->proto, 0, sizeof(pkt->proto));
 150         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 151
 152         switch (pkt->type) {
 153         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 154                 pkt->u.size = 0;
 155                 break;
 156
 157         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 158         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 159                 pkt->u.size = size;
 160                 break;
 161
 162         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 163         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 164                 pkt->u.handle = handle;
 165                 break;
 166
 167         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 168         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 169         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 170                 pkt->u.size = 0;
 171                 break;
 172
 173         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 174                 pkt->u.mode = mode;
 175                 break;
 176
 177         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 178         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 179                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 180                 break;
 181
 182         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 183         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 184                 pkt->u.size = size;
 185                 pkt->proto = proto;
 186                 break;
 187         }
 188 }
 189
 190 static inline void
 191 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 192                                     struct sockaddr_vm *local,
 193                                     struct sockaddr_vm *remote)
 194 {
 195         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 196         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 197 }
 198
 199 static int
 200 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 201                                   struct sockaddr_vm *src,
 202                                   struct sockaddr_vm *dst,
 203                                   enum vmci_transport_packet_type type,
 204                                   u64 size,
 205                                   u64 mode,
 206                                   struct vmci_transport_waiting_info *wait,
 207                                   u16 proto,
 208                                   struct vmci_handle handle,
 209                                   bool convert_error)
 210 {
 211         int err;
 212
 213         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 214                                    proto, handle);
 215         err = vmci_datagram_send(&pkt->dg);
 216         if (convert_error && (err < 0))
 217                 return vmci_transport_error_to_vsock_error(err);
 218
 219         return err;
 220 }
 221
 222 static int
 223 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 224                                       enum vmci_transport_packet_type type,
 225                                       u64 size,
 226                                       u64 mode,
 227                                       struct vmci_transport_waiting_info *wait,
 228                                       struct vmci_handle handle)
 229 {
 230         struct vmci_transport_packet reply;
 231         struct sockaddr_vm src, dst;
 232
 233         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 234                 return 0;
 235         } else {
 236                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 237                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 238                                                          type,
 239                                                          size, mode, wait,
 240                                                          VSOCK_PROTO_INVALID,
 241                                                          handle, true);
 242         }
 243 }
 244
 245 static int
 246 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 247                                    struct sockaddr_vm *dst,
 248                                    enum vmci_transport_packet_type type,
 249                                    u64 size,
 250                                    u64 mode,
 251                                    struct vmci_transport_waiting_info *wait,
 252                                    struct vmci_handle handle)
 253 {
 254         /* Note that it is safe to use a single packet across all CPUs since
 255          * two tasklets of the same type are guaranteed to not ever run
 256          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 257          * we can use per-cpu packets.
 258          */
 259         static struct vmci_transport_packet pkt;
 260
 261         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 262                                                  size, mode, wait,
 263                                                  VSOCK_PROTO_INVALID, handle,
 264                                                  false);
 265 }
 266
 267 static int
 268 vmci_transport_send_control_pkt(struct sock *sk,
 269                                 enum vmci_transport_packet_type type,
 270                                 u64 size,
 271                                 u64 mode,
 272                                 struct vmci_transport_waiting_info *wait,
 273                                 u16 proto,
 274                                 struct vmci_handle handle)
 275 {
 276         struct vmci_transport_packet *pkt;
 277         struct vsock_sock *vsk;
 278         int err;
 279
 280         vsk = vsock_sk(sk);
 281
 282         if (!vsock_addr_bound(&vsk->local_addr))
 283                 return -EINVAL;
 284
 285         if (!vsock_addr_bound(&vsk->remote_addr))
 286                 return -EINVAL;
 287
 288         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 289         if (!pkt)
 290                 return -ENOMEM;
 291
 292         err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
 293                                                 &vsk->remote_addr, type, size,
 294                                                 mode, wait, proto, handle,
 295                                                 true);
 296         kfree(pkt);
 297
 298         return err;
 299 }
 300
 301 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 302                                         struct sockaddr_vm *src,
 303                                         struct vmci_transport_packet *pkt)
 304 {
 305         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 306                 return 0;
 307         return vmci_transport_send_control_pkt_bh(
 308                                         dst, src,
 309                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 310                                         0, NULL, VMCI_INVALID_HANDLE);
 311 }
 312
 313 static int vmci_transport_send_reset(struct sock *sk,
 314                                      struct vmci_transport_packet *pkt)
 315 {
 316         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 317                 return 0;
 318         return vmci_transport_send_control_pkt(sk,
 319                                         VMCI_TRANSPORT_PACKET_TYPE_RST,
 320                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 321                                         VMCI_INVALID_HANDLE);
 322 }
 323
 324 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 325 {
 326         return vmci_transport_send_control_pkt(
 327                                         sk,
 328                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 329                                         size, 0, NULL,
 330                                         VSOCK_PROTO_INVALID,
 331                                         VMCI_INVALID_HANDLE);
 332 }
 333
 334 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 335                                           u16 version)
 336 {
 337         return vmci_transport_send_control_pkt(
 338                                         sk,
 339                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 340                                         size, 0, NULL, version,
 341                                         VMCI_INVALID_HANDLE);
 342 }
 343
 344 static int vmci_transport_send_qp_offer(struct sock *sk,
 345                                         struct vmci_handle handle)
 346 {
 347         return vmci_transport_send_control_pkt(
 348                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 349                                         0, NULL,
 350                                         VSOCK_PROTO_INVALID, handle);
 351 }
 352
 353 static int vmci_transport_send_attach(struct sock *sk,
 354                                       struct vmci_handle handle)
 355 {
 356         return vmci_transport_send_control_pkt(
 357                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 358                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 359                                         handle);
 360 }
 361
 362 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 363 {
 364         return vmci_transport_reply_control_pkt_fast(
 365                                                 pkt,
 366                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 367                                                 0, 0, NULL,
 368                                                 VMCI_INVALID_HANDLE);
 369 }
 370
 371 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 372                                           struct sockaddr_vm *src)
 373 {
 374         return vmci_transport_send_control_pkt_bh(
 375                                         dst, src,
 376                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 377                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 378 }
 379
 380 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 381                                  struct sockaddr_vm *src)
 382 {
 383         return vmci_transport_send_control_pkt_bh(
 384                                         dst, src,
 385                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 386                                         0, NULL, VMCI_INVALID_HANDLE);
 387 }
 388
 389 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 390                                 struct sockaddr_vm *src)
 391 {
 392         return vmci_transport_send_control_pkt_bh(
 393                                         dst, src,
 394                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 395                                         0, NULL, VMCI_INVALID_HANDLE);
 396 }
 397
 398 int vmci_transport_send_wrote(struct sock *sk)
 399 {
 400         return vmci_transport_send_control_pkt(
 401                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 402                                         0, NULL, VSOCK_PROTO_INVALID,
 403                                         VMCI_INVALID_HANDLE);
 404 }
 405
 406 int vmci_transport_send_read(struct sock *sk)
 407 {
 408         return vmci_transport_send_control_pkt(
 409                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 410                                         0, NULL, VSOCK_PROTO_INVALID,
 411                                         VMCI_INVALID_HANDLE);
 412 }
 413
 414 int vmci_transport_send_waiting_write(struct sock *sk,
 415                                       struct vmci_transport_waiting_info *wait)
 416 {
 417         return vmci_transport_send_control_pkt(
 418                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 419                                 0, 0, wait, VSOCK_PROTO_INVALID,
 420                                 VMCI_INVALID_HANDLE);
 421 }
 422
 423 int vmci_transport_send_waiting_read(struct sock *sk,
 424                                      struct vmci_transport_waiting_info *wait)
 425 {
 426         return vmci_transport_send_control_pkt(
 427                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 428                                 0, 0, wait, VSOCK_PROTO_INVALID,
 429                                 VMCI_INVALID_HANDLE);
 430 }
 431
 432 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 433 {
 434         return vmci_transport_send_control_pkt(
 435                                         &vsk->sk,
 436                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 437                                         0, mode, NULL,
 438                                         VSOCK_PROTO_INVALID,
 439                                         VMCI_INVALID_HANDLE);
 440 }
 441
 442 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 443 {
 444         return vmci_transport_send_control_pkt(sk,
 445                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 446                                         size, 0, NULL,
 447                                         VSOCK_PROTO_INVALID,
 448                                         VMCI_INVALID_HANDLE);
 449 }
 450
 451 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 452                                              u16 version)
 453 {
 454         return vmci_transport_send_control_pkt(
 455                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 456                                         size, 0, NULL, version,
 457                                         VMCI_INVALID_HANDLE);
 458 }
 459
 460 static struct sock *vmci_transport_get_pending(
 461                                         struct sock *listener,
 462                                         struct vmci_transport_packet *pkt)
 463 {
 464         struct vsock_sock *vlistener;
 465         struct vsock_sock *vpending;
 466         struct sock *pending;
 467
 468         vlistener = vsock_sk(listener);
 469
 470         list_for_each_entry(vpending, &vlistener->pending_links,
 471                             pending_links) {
 472                 struct sockaddr_vm src;
 473                 struct sockaddr_vm dst;
 474
 475                 vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 476                 vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 477
 478                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 479                     vsock_addr_equals_addr(&dst, &vpending->local_addr)) {
 480                         pending = sk_vsock(vpending);
 481                         sock_hold(pending);
 482                         goto found;
 483                 }
 484         }
 485
 486         pending = NULL;
 487 found:
 488         return pending;
 489
 490 }
 491
 492 static void vmci_transport_release_pending(struct sock *pending)
 493 {
 494         sock_put(pending);
 495 }
 496
 497 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 498  * trusted sockets 2) sockets from applications running as the same user as the
 499  * VM (this is only true for the host side and only when using hosted products)
 500  */
 501
 502 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 503 {
 504         return vsock->trusted ||
 505                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 506 }
 507
 508 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 509  * only if it is trusted as described in vmci_transport_is_trusted.
 510  */
 511
 512 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 513 {
 514         if (vsock->cached_peer != peer_cid) {
 515                 vsock->cached_peer = peer_cid;
 516                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 517                     (vmci_context_get_priv_flags(peer_cid) &
 518                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 519                         vsock->cached_peer_allow_dgram = false;
 520                 } else {
 521                         vsock->cached_peer_allow_dgram = true;
 522                 }
 523         }
 524
 525         return vsock->cached_peer_allow_dgram;
 526 }
 527
 528 static int
 529 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 530                                 struct vmci_handle *handle,
 531                                 u64 produce_size,
 532                                 u64 consume_size,
 533                                 u32 peer, u32 flags, bool trusted)
 534 {
 535         int err = 0;
 536
 537         if (trusted) {
 538                 /* Try to allocate our queue pair as trusted. This will only
 539                  * work if vsock is running in the host.
 540                  */
 541
 542                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 543                                        consume_size,
 544                                        peer, flags,
 545                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 546                 if (err != VMCI_ERROR_NO_ACCESS)
 547                         goto out;
 548
 549         }
 550
 551         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 552                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 553 out:
 554         if (err < 0) {
 555                 pr_err("Could not attach to queue pair with %d\n",
 556                        err);
 557                 err = vmci_transport_error_to_vsock_error(err);
 558         }
 559
 560         return err;
 561 }
 562
 563 static int
 564 vmci_transport_datagram_create_hnd(u32 resource_id,
 565                                    u32 flags,
 566                                    vmci_datagram_recv_cb recv_cb,
 567                                    void *client_data,
 568                                    struct vmci_handle *out_handle)
 569 {
 570         int err = 0;
 571
 572         /* Try to allocate our datagram handler as trusted. This will only work
 573          * if vsock is running in the host.
 574          */
 575
 576         err = vmci_datagram_create_handle_priv(resource_id, flags,
 577                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 578                                                recv_cb,
 579                                                client_data, out_handle);
 580
 581         if (err == VMCI_ERROR_NO_ACCESS)
 582                 err = vmci_datagram_create_handle(resource_id, flags,
 583                                                   recv_cb, client_data,
 584                                                   out_handle);
 585
 586         return err;
 587 }
 588
 589 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 590  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 591  * sleep it should defer that work to a work queue.
 592  */
 593
 594 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 595 {
 596         struct sock *sk;
 597         size_t size;
 598         struct sk_buff *skb;
 599         struct vsock_sock *vsk;
 600
 601         sk = (struct sock *)data;
 602
 603         /* This handler is privileged when this module is running on the host.
 604          * We will get datagrams from all endpoints (even VMs that are in a
 605          * restricted context). If we get one from a restricted context then
 606          * the destination socket must be trusted.
 607          *
 608          * NOTE: We access the socket struct without holding the lock here.
 609          * This is ok because the field we are interested is never modified
 610          * outside of the create and destruct socket functions.
 611          */
 612         vsk = vsock_sk(sk);
 613         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 614                 return VMCI_ERROR_NO_ACCESS;
 615
 616         size = VMCI_DG_SIZE(dg);
 617
 618         /* Attach the packet to the socket's receive queue as an sk_buff. */
 619         skb = alloc_skb(size, GFP_ATOMIC);
 620         if (skb) {
 621                 /* sk_receive_skb() will do a sock_put(), so hold here. */
 622                 sock_hold(sk);
 623                 skb_put(skb, size);
 624                 memcpy(skb->data, dg, size);
 625                 sk_receive_skb(sk, skb, 0);
 626         }
 627
 628         return VMCI_SUCCESS;
 629 }
 630
 631 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 632 {
 633         static const u32 non_socket_contexts[] = {
 634                 VMADDR_CID_HYPERVISOR,
 635                 VMADDR_CID_RESERVED,
 636         };
 637         int i;
 638
 639         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 640
 641         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 642                 if (cid == non_socket_contexts[i])
 643                         return false;
 644         }
 645
 646         return true;
 647 }
 648
 649 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 650  * interrupt fires.  This is run in bottom-half context but it defers most of
 651  * its work to the packet handling work queue.
 652  */
 653
 654 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 655 {
 656         struct sock *sk;
 657         struct sockaddr_vm dst;
 658         struct sockaddr_vm src;
 659         struct vmci_transport_packet *pkt;
 660         struct vsock_sock *vsk;
 661         bool bh_process_pkt;
 662         int err;
 663
 664         sk = NULL;
 665         err = VMCI_SUCCESS;
 666         bh_process_pkt = false;
 667
 668         /* Ignore incoming packets from contexts without sockets, or resources
 669          * that aren't vsock implementations.
 670          */
 671
 672         if (!vmci_transport_stream_allow(dg->src.context, -1)
 673             || VMCI_TRANSPORT_PACKET_RID != dg->src.resource)
 674                 return VMCI_ERROR_NO_ACCESS;
 675
 676         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 677                 /* Drop datagrams that do not contain full VSock packets. */
 678                 return VMCI_ERROR_INVALID_ARGS;
 679
 680         pkt = (struct vmci_transport_packet *)dg;
 681
 682         /* Find the socket that should handle this packet.  First we look for a
 683          * connected socket and if there is none we look for a socket bound to
 684          * the destintation address.
 685          */
 686         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 687         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 688
 689         sk = vsock_find_connected_socket(&src, &dst);
 690         if (!sk) {
 691                 sk = vsock_find_bound_socket(&dst);
 692                 if (!sk) {
 693                         /* We could not find a socket for this specified
 694                          * address.  If this packet is a RST, we just drop it.
 695                          * If it is another packet, we send a RST.  Note that
 696                          * we do not send a RST reply to RSTs so that we do not
 697                          * continually send RSTs between two endpoints.
 698                          *
 699                          * Note that since this is a reply, dst is src and src
 700                          * is dst.
 701                          */
 702                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 703                                 pr_err("unable to send reset\n");
 704
 705                         err = VMCI_ERROR_NOT_FOUND;
 706                         goto out;
 707                 }
 708         }
 709
 710         /* If the received packet type is beyond all types known to this
 711          * implementation, reply with an invalid message.  Hopefully this will
 712          * help when implementing backwards compatibility in the future.
 713          */
 714         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 715                 vmci_transport_send_invalid_bh(&dst, &src);
 716                 err = VMCI_ERROR_INVALID_ARGS;
 717                 goto out;
 718         }
 719
 720         /* This handler is privileged when this module is running on the host.
 721          * We will get datagram connect requests from all endpoints (even VMs
 722          * that are in a restricted context). If we get one from a restricted
 723          * context then the destination socket must be trusted.
 724          *
 725          * NOTE: We access the socket struct without holding the lock here.
 726          * This is ok because the field we are interested is never modified
 727          * outside of the create and destruct socket functions.
 728          */
 729         vsk = vsock_sk(sk);
 730         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 731                 err = VMCI_ERROR_NO_ACCESS;
 732                 goto out;
 733         }
 734
 735         /* We do most everything in a work queue, but let's fast path the
 736          * notification of reads and writes to help data transfer performance.
 737          * We can only do this if there is no process context code executing
 738          * for this socket since that may change the state.
 739          */
 740         bh_lock_sock(sk);
 741
 742         if (!sock_owned_by_user(sk) && sk->sk_state == SS_CONNECTED)
 743                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 744                                 sk, pkt, true, &dst, &src,
 745                                 &bh_process_pkt);
 746
 747         bh_unlock_sock(sk);
 748
 749         if (!bh_process_pkt) {
 750                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 751
 752                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 753                 if (!recv_pkt_info) {
 754                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 755                                 pr_err("unable to send reset\n");
 756
 757                         err = VMCI_ERROR_NO_MEM;
 758                         goto out;
 759                 }
 760
 761                 recv_pkt_info->sk = sk;
 762                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 763                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 764
 765                 schedule_work(&recv_pkt_info->work);
 766                 /* Clear sk so that the reference count incremented by one of
 767                  * the Find functions above is not decremented below.  We need
 768                  * that reference count for the packet handler we've scheduled
 769                  * to run.
 770                  */
 771                 sk = NULL;
 772         }
 773
 774 out:
 775         if (sk)
 776                 sock_put(sk);
 777
 778         return err;
 779 }
 780
 781 static void vmci_transport_peer_attach_cb(u32 sub_id,
 782                                           const struct vmci_event_data *e_data,
 783                                           void *client_data)
 784 {
 785         struct sock *sk = client_data;
 786         const struct vmci_event_payload_qp *e_payload;
 787         struct vsock_sock *vsk;
 788
 789         e_payload = vmci_event_data_const_payload(e_data);
 790
 791         vsk = vsock_sk(sk);
 792
 793         /* We don't ask for delayed CBs when we subscribe to this event (we
 794          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 795          * guarantees in that case about what context we might be running in,
 796          * so it could be BH or process, blockable or non-blockable.  So we
 797          * need to account for all possible contexts here.
 798          */
 799         local_bh_disable();
 800         bh_lock_sock(sk);
 801
 802         /* XXX This is lame, we should provide a way to lookup sockets by
 803          * qp_handle.
 804          */
 805         if (vmci_handle_is_equal(vmci_trans(vsk)->qp_handle,
 806                                  e_payload->handle)) {
 807                 /* XXX This doesn't do anything, but in the future we may want
 808                  * to set a flag here to verify the attach really did occur and
 809                  * we weren't just sent a datagram claiming it was.
 810                  */
 811                 goto out;
 812         }
 813
 814 out:
 815         bh_unlock_sock(sk);
 816         local_bh_enable();
 817 }
 818
 819 static void vmci_transport_handle_detach(struct sock *sk)
 820 {
 821         struct vsock_sock *vsk;
 822
 823         vsk = vsock_sk(sk);
 824         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 825                 sock_set_flag(sk, SOCK_DONE);
 826
 827                 /* On a detach the peer will not be sending or receiving
 828                  * anymore.
 829                  */
 830                 vsk->peer_shutdown = SHUTDOWN_MASK;
 831
 832                 /* We should not be sending anymore since the peer won't be
 833                  * there to receive, but we can still receive if there is data
 834                  * left in our consume queue.
 835                  */
 836                 if (vsock_stream_has_data(vsk) <= 0) {
 837                         if (sk->sk_state == SS_CONNECTING) {
 838                                 /* The peer may detach from a queue pair while
 839                                  * we are still in the connecting state, i.e.,
 840                                  * if the peer VM is killed after attaching to
 841                                  * a queue pair, but before we complete the
 842                                  * handshake. In that case, we treat the detach
 843                                  * event like a reset.
 844                                  */
 845
 846                                 sk->sk_state = SS_UNCONNECTED;
 847                                 sk->sk_err = ECONNRESET;
 848                                 sk->sk_error_report(sk);
 849                                 return;
 850                         }
 851                         sk->sk_state = SS_UNCONNECTED;
 852                 }
 853                 sk->sk_state_change(sk);
 854         }
 855 }
 856
 857 static void vmci_transport_peer_detach_cb(u32 sub_id,
 858                                           const struct vmci_event_data *e_data,
 859                                           void *client_data)
 860 {
 861         struct sock *sk = client_data;
 862         const struct vmci_event_payload_qp *e_payload;
 863         struct vsock_sock *vsk;
 864
 865         e_payload = vmci_event_data_const_payload(e_data);
 866         vsk = vsock_sk(sk);
 867         if (vmci_handle_is_invalid(e_payload->handle))
 868                 return;
 869
 870         /* Same rules for locking as for peer_attach_cb(). */
 871         local_bh_disable();
 872         bh_lock_sock(sk);
 873
 874         /* XXX This is lame, we should provide a way to lookup sockets by
 875          * qp_handle.
 876          */
 877         if (vmci_handle_is_equal(vmci_trans(vsk)->qp_handle,
 878                                  e_payload->handle))
 879                 vmci_transport_handle_detach(sk);
 880
 881         bh_unlock_sock(sk);
 882         local_bh_enable();
 883 }
 884
 885 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 886                                          const struct vmci_event_data *e_data,
 887                                          void *client_data)
 888 {
 889         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 890 }
 891
 892 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 893 {
 894         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 895         struct vmci_transport_packet *pkt;
 896         struct sock *sk;
 897
 898         recv_pkt_info =
 899                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 900         sk = recv_pkt_info->sk;
 901         pkt = &recv_pkt_info->pkt;
 902
 903         lock_sock(sk);
 904
 905         switch (sk->sk_state) {
 906         case SS_LISTEN:
 907                 vmci_transport_recv_listen(sk, pkt);
 908                 break;
 909         case SS_CONNECTING:
 910                 /* Processing of pending connections for servers goes through
 911                  * the listening socket, so see vmci_transport_recv_listen()
 912                  * for that path.
 913                  */
 914                 vmci_transport_recv_connecting_client(sk, pkt);
 915                 break;
 916         case SS_CONNECTED:
 917                 vmci_transport_recv_connected(sk, pkt);
 918                 break;
 919         default:
 920                 /* Because this function does not run in the same context as
 921                  * vmci_transport_recv_stream_cb it is possible that the
 922                  * socket has closed. We need to let the other side know or it
 923                  * could be sitting in a connect and hang forever. Send a
 924                  * reset to prevent that.
 925                  */
 926                 vmci_transport_send_reset(sk, pkt);
 927                 goto out;
 928         }
 929
 930 out:
 931         release_sock(sk);
 932         kfree(recv_pkt_info);
 933         /* Release reference obtained in the stream callback when we fetched
 934          * this socket out of the bound or connected list.
 935          */
 936         sock_put(sk);
 937 }
 938
 939 static int vmci_transport_recv_listen(struct sock *sk,
 940                                       struct vmci_transport_packet *pkt)
 941 {
 942         struct sock *pending;
 943         struct vsock_sock *vpending;
 944         int err;
 945         u64 qp_size;
 946         bool old_request = false;
 947         bool old_pkt_proto = false;
 948
 949         err = 0;
 950
 951         /* Because we are in the listen state, we could be receiving a packet
 952          * for ourself or any previous connection requests that we received.
 953          * If it's the latter, we try to find a socket in our list of pending
 954          * connections and, if we do, call the appropriate handler for the
 955          * state that that socket is in.  Otherwise we try to service the
 956          * connection request.
 957          */
 958         pending = vmci_transport_get_pending(sk, pkt);
 959         if (pending) {
 960                 lock_sock(pending);
 961                 switch (pending->sk_state) {
 962                 case SS_CONNECTING:
 963                         err = vmci_transport_recv_connecting_server(sk,
 964                                                                     pending,
 965                                                                     pkt);
 966                         break;
 967                 default:
 968                         vmci_transport_send_reset(pending, pkt);
 969                         err = -EINVAL;
 970                 }
 971
 972                 if (err < 0)
 973                         vsock_remove_pending(sk, pending);
 974
 975                 release_sock(pending);
 976                 vmci_transport_release_pending(pending);
 977
 978                 return err;
 979         }
 980
 981         /* The listen state only accepts connection requests.  Reply with a
 982          * reset unless we received a reset.
 983          */
 984
 985         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 986               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 987                 vmci_transport_reply_reset(pkt);
 988                 return -EINVAL;
 989         }
 990
 991         if (pkt->u.size == 0) {
 992                 vmci_transport_reply_reset(pkt);
 993                 return -EINVAL;
 994         }
 995
 996         /* If this socket can't accommodate this connection request, we send a
 997          * reset.  Otherwise we create and initialize a child socket and reply
 998          * with a connection negotiation.
 999          */
1000         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1001                 vmci_transport_reply_reset(pkt);
1002                 return -ECONNREFUSED;
1003         }
1004
1005         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1006                                  sk->sk_type);
1007         if (!pending) {
1008                 vmci_transport_send_reset(sk, pkt);
1009                 return -ENOMEM;
1010         }
1011
1012         vpending = vsock_sk(pending);
1013
1014         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1015                         pkt->dst_port);
1016         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1017                         pkt->src_port);
1018
1019         /* If the proposed size fits within our min/max, accept it. Otherwise
1020          * propose our own size.
1021          */
1022         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1023             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1024                 qp_size = pkt->u.size;
1025         } else {
1026                 qp_size = vmci_trans(vpending)->queue_pair_size;
1027         }
1028
1029         /* Figure out if we are using old or new requests based on the
1030          * overrides pkt types sent by our peer.
1031          */
1032         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1033                 old_request = old_pkt_proto;
1034         } else {
1035                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1036                         old_request = true;
1037                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1038                         old_request = false;
1039
1040         }
1041
1042         if (old_request) {
1043                 /* Handle a REQUEST (or override) */
1044                 u16 version = VSOCK_PROTO_INVALID;
1045                 if (vmci_transport_proto_to_notify_struct(
1046                         pending, &version, true))
1047                         err = vmci_transport_send_negotiate(pending, qp_size);
1048                 else
1049                         err = -EINVAL;
1050
1051         } else {
1052                 /* Handle a REQUEST2 (or override) */
1053                 int proto_int = pkt->proto;
1054                 int pos;
1055                 u16 active_proto_version = 0;
1056
1057                 /* The list of possible protocols is the intersection of all
1058                  * protocols the client supports ... plus all the protocols we
1059                  * support.
1060                  */
1061                 proto_int &= vmci_transport_new_proto_supported_versions();
1062
1063                 /* We choose the highest possible protocol version and use that
1064                  * one.
1065                  */
1066                 pos = fls(proto_int);
1067                 if (pos) {
1068                         active_proto_version = (1 << (pos - 1));
1069                         if (vmci_transport_proto_to_notify_struct(
1070                                 pending, &active_proto_version, false))
1071                                 err = vmci_transport_send_negotiate2(pending,
1072                                                         qp_size,
1073                                                         active_proto_version);
1074                         else
1075                                 err = -EINVAL;
1076
1077                 } else {
1078                         err = -EINVAL;
1079                 }
1080         }
1081
1082         if (err < 0) {
1083                 vmci_transport_send_reset(sk, pkt);
1084                 sock_put(pending);
1085                 err = vmci_transport_error_to_vsock_error(err);
1086                 goto out;
1087         }
1088
1089         vsock_add_pending(sk, pending);
1090         sk->sk_ack_backlog++;
1091
1092         pending->sk_state = SS_CONNECTING;
1093         vmci_trans(vpending)->produce_size =
1094                 vmci_trans(vpending)->consume_size = qp_size;
1095         vmci_trans(vpending)->queue_pair_size = qp_size;
1096
1097         vmci_trans(vpending)->notify_ops->process_request(pending);
1098
1099         /* We might never receive another message for this socket and it's not
1100          * connected to any process, so we have to ensure it gets cleaned up
1101          * ourself.  Our delayed work function will take care of that.  Note
1102          * that we do not ever cancel this function since we have few
1103          * guarantees about its state when calling cancel_delayed_work().
1104          * Instead we hold a reference on the socket for that function and make
1105          * it capable of handling cases where it needs to do nothing but
1106          * release that reference.
1107          */
1108         vpending->listener = sk;
1109         sock_hold(sk);
1110         sock_hold(pending);
1111         INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1112         schedule_delayed_work(&vpending->dwork, HZ);
1113
1114 out:
1115         return err;
1116 }
1117
1118 static int
1119 vmci_transport_recv_connecting_server(struct sock *listener,
1120                                       struct sock *pending,
1121                                       struct vmci_transport_packet *pkt)
1122 {
1123         struct vsock_sock *vpending;
1124         struct vmci_handle handle;
1125         struct vmci_qp *qpair;
1126         bool is_local;
1127         u32 flags;
1128         u32 detach_sub_id;
1129         int err;
1130         int skerr;
1131
1132         vpending = vsock_sk(pending);
1133         detach_sub_id = VMCI_INVALID_ID;
1134
1135         switch (pkt->type) {
1136         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1137                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1138                         vmci_transport_send_reset(pending, pkt);
1139                         skerr = EPROTO;
1140                         err = -EINVAL;
1141                         goto destroy;
1142                 }
1143                 break;
1144         default:
1145                 /* Close and cleanup the connection. */
1146                 vmci_transport_send_reset(pending, pkt);
1147                 skerr = EPROTO;
1148                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1149                 goto destroy;
1150         }
1151
1152         /* In order to complete the connection we need to attach to the offered
1153          * queue pair and send an attach notification.  We also subscribe to the
1154          * detach event so we know when our peer goes away, and we do that
1155          * before attaching so we don't miss an event.  If all this succeeds,
1156          * we update our state and wakeup anything waiting in accept() for a
1157          * connection.
1158          */
1159
1160         /* We don't care about attach since we ensure the other side has
1161          * attached by specifying the ATTACH_ONLY flag below.
1162          */
1163         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1164                                    vmci_transport_peer_detach_cb,
1165                                    pending, &detach_sub_id);
1166         if (err < VMCI_SUCCESS) {
1167                 vmci_transport_send_reset(pending, pkt);
1168                 err = vmci_transport_error_to_vsock_error(err);
1169                 skerr = -err;
1170                 goto destroy;
1171         }
1172
1173         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1174
1175         /* Now attach to the queue pair the client created. */
1176         handle = pkt->u.handle;
1177
1178         /* vpending->local_addr always has a context id so we do not need to
1179          * worry about VMADDR_CID_ANY in this case.
1180          */
1181         is_local =
1182             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1183         flags = VMCI_QPFLAG_ATTACH_ONLY;
1184         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1185
1186         err = vmci_transport_queue_pair_alloc(
1187                                         &qpair,
1188                                         &handle,
1189                                         vmci_trans(vpending)->produce_size,
1190                                         vmci_trans(vpending)->consume_size,
1191                                         pkt->dg.src.context,
1192                                         flags,
1193                                         vmci_transport_is_trusted(
1194                                                 vpending,
1195                                                 vpending->remote_addr.svm_cid));
1196         if (err < 0) {
1197                 vmci_transport_send_reset(pending, pkt);
1198                 skerr = -err;
1199                 goto destroy;
1200         }
1201
1202         vmci_trans(vpending)->qp_handle = handle;
1203         vmci_trans(vpending)->qpair = qpair;
1204
1205         /* When we send the attach message, we must be ready to handle incoming
1206          * control messages on the newly connected socket. So we move the
1207          * pending socket to the connected state before sending the attach
1208          * message. Otherwise, an incoming packet triggered by the attach being
1209          * received by the peer may be processed concurrently with what happens
1210          * below after sending the attach message, and that incoming packet
1211          * will find the listening socket instead of the (currently) pending
1212          * socket. Note that enqueueing the socket increments the reference
1213          * count, so even if a reset comes before the connection is accepted,
1214          * the socket will be valid until it is removed from the queue.
1215          *
1216          * If we fail sending the attach below, we remove the socket from the
1217          * connected list and move the socket to SS_UNCONNECTED before
1218          * releasing the lock, so a pending slow path processing of an incoming
1219          * packet will not see the socket in the connected state in that case.
1220          */
1221         pending->sk_state = SS_CONNECTED;
1222
1223         vsock_insert_connected(vpending);
1224
1225         /* Notify our peer of our attach. */
1226         err = vmci_transport_send_attach(pending, handle);
1227         if (err < 0) {
1228                 vsock_remove_connected(vpending);
1229                 pr_err("Could not send attach\n");
1230                 vmci_transport_send_reset(pending, pkt);
1231                 err = vmci_transport_error_to_vsock_error(err);
1232                 skerr = -err;
1233                 goto destroy;
1234         }
1235
1236         /* We have a connection. Move the now connected socket from the
1237          * listener's pending list to the accept queue so callers of accept()
1238          * can find it.
1239          */
1240         vsock_remove_pending(listener, pending);
1241         vsock_enqueue_accept(listener, pending);
1242
1243         /* Callers of accept() will be be waiting on the listening socket, not
1244          * the pending socket.
1245          */
1246         listener->sk_state_change(listener);
1247
1248         return 0;
1249
1250 destroy:
1251         pending->sk_err = skerr;
1252         pending->sk_state = SS_UNCONNECTED;
1253         /* As long as we drop our reference, all necessary cleanup will handle
1254          * when the cleanup function drops its reference and our destruct
1255          * implementation is called.  Note that since the listen handler will
1256          * remove pending from the pending list upon our failure, the cleanup
1257          * function won't drop the additional reference, which is why we do it
1258          * here.
1259          */
1260         sock_put(pending);
1261
1262         return err;
1263 }
1264
1265 static int
1266 vmci_transport_recv_connecting_client(struct sock *sk,
1267                                       struct vmci_transport_packet *pkt)
1268 {
1269         struct vsock_sock *vsk;
1270         int err;
1271         int skerr;
1272
1273         vsk = vsock_sk(sk);
1274
1275         switch (pkt->type) {
1276         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1277                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1278                     !vmci_handle_is_equal(pkt->u.handle,
1279                                           vmci_trans(vsk)->qp_handle)) {
1280                         skerr = EPROTO;
1281                         err = -EINVAL;
1282                         goto destroy;
1283                 }
1284
1285                 /* Signify the socket is connected and wakeup the waiter in
1286                  * connect(). Also place the socket in the connected table for
1287                  * accounting (it can already be found since it's in the bound
1288                  * table).
1289                  */
1290                 sk->sk_state = SS_CONNECTED;
1291                 sk->sk_socket->state = SS_CONNECTED;
1292                 vsock_insert_connected(vsk);
1293                 sk->sk_state_change(sk);
1294
1295                 break;
1296         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1297         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1298                 if (pkt->u.size == 0
1299                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1300                     || pkt->src_port != vsk->remote_addr.svm_port
1301                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1302                     || vmci_trans(vsk)->qpair
1303                     || vmci_trans(vsk)->produce_size != 0
1304                     || vmci_trans(vsk)->consume_size != 0
1305                     || vmci_trans(vsk)->attach_sub_id != VMCI_INVALID_ID
1306                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1307                         skerr = EPROTO;
1308                         err = -EINVAL;
1309
1310                         goto destroy;
1311                 }
1312
1313                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1314                 if (err) {
1315                         skerr = -err;
1316                         goto destroy;
1317                 }
1318
1319                 break;
1320         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1321                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1322                 if (err) {
1323                         skerr = -err;
1324                         goto destroy;
1325                 }
1326
1327                 break;
1328         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1329                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1330                  * continue processing here after they sent an INVALID packet.
1331                  * This meant that we got a RST after the INVALID. We ignore a
1332                  * RST after an INVALID. The common code doesn't send the RST
1333                  * ... so we can hang if an old version of the common code
1334                  * fails between getting a REQUEST and sending an OFFER back.
1335                  * Not much we can do about it... except hope that it doesn't
1336                  * happen.
1337                  */
1338                 if (vsk->ignore_connecting_rst) {
1339                         vsk->ignore_connecting_rst = false;
1340                 } else {
1341                         skerr = ECONNRESET;
1342                         err = 0;
1343                         goto destroy;
1344                 }
1345
1346                 break;
1347         default:
1348                 /* Close and cleanup the connection. */
1349                 skerr = EPROTO;
1350                 err = -EINVAL;
1351                 goto destroy;
1352         }
1353
1354         return 0;
1355
1356 destroy:
1357         vmci_transport_send_reset(sk, pkt);
1358
1359         sk->sk_state = SS_UNCONNECTED;
1360         sk->sk_err = skerr;
1361         sk->sk_error_report(sk);
1362         return err;
1363 }
1364
1365 static int vmci_transport_recv_connecting_client_negotiate(
1366                                         struct sock *sk,
1367                                         struct vmci_transport_packet *pkt)
1368 {
1369         int err;
1370         struct vsock_sock *vsk;
1371         struct vmci_handle handle;
1372         struct vmci_qp *qpair;
1373         u32 attach_sub_id;
1374         u32 detach_sub_id;
1375         bool is_local;
1376         u32 flags;
1377         bool old_proto = true;
1378         bool old_pkt_proto;
1379         u16 version;
1380
1381         vsk = vsock_sk(sk);
1382         handle = VMCI_INVALID_HANDLE;
1383         attach_sub_id = VMCI_INVALID_ID;
1384         detach_sub_id = VMCI_INVALID_ID;
1385
1386         /* If we have gotten here then we should be past the point where old
1387          * linux vsock could have sent the bogus rst.
1388          */
1389         vsk->sent_request = false;
1390         vsk->ignore_connecting_rst = false;
1391
1392         /* Verify that we're OK with the proposed queue pair size */
1393         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1394             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1395                 err = -EINVAL;
1396                 goto destroy;
1397         }
1398
1399         /* At this point we know the CID the peer is using to talk to us. */
1400
1401         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1402                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1403
1404         /* Setup the notify ops to be the highest supported version that both
1405          * the server and the client support.
1406          */
1407
1408         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1409                 old_proto = old_pkt_proto;
1410         } else {
1411                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1412                         old_proto = true;
1413                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1414                         old_proto = false;
1415
1416         }
1417
1418         if (old_proto)
1419                 version = VSOCK_PROTO_INVALID;
1420         else
1421                 version = pkt->proto;
1422
1423         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1424                 err = -EINVAL;
1425                 goto destroy;
1426         }
1427
1428         /* Subscribe to attach and detach events first.
1429          *
1430          * XXX We attach once for each queue pair created for now so it is easy
1431          * to find the socket (it's provided), but later we should only
1432          * subscribe once and add a way to lookup sockets by queue pair handle.
1433          */
1434         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_ATTACH,
1435                                    vmci_transport_peer_attach_cb,
1436                                    sk, &attach_sub_id);
1437         if (err < VMCI_SUCCESS) {
1438                 err = vmci_transport_error_to_vsock_error(err);
1439                 goto destroy;
1440         }
1441
1442         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1443                                    vmci_transport_peer_detach_cb,
1444                                    sk, &detach_sub_id);
1445         if (err < VMCI_SUCCESS) {
1446                 err = vmci_transport_error_to_vsock_error(err);
1447                 goto destroy;
1448         }
1449
1450         /* Make VMCI select the handle for us. */
1451         handle = VMCI_INVALID_HANDLE;
1452         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1453         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1454
1455         err = vmci_transport_queue_pair_alloc(&qpair,
1456                                               &handle,
1457                                               pkt->u.size,
1458                                               pkt->u.size,
1459                                               vsk->remote_addr.svm_cid,
1460                                               flags,
1461                                               vmci_transport_is_trusted(
1462                                                   vsk,
1463                                                   vsk->
1464                                                   remote_addr.svm_cid));
1465         if (err < 0)
1466                 goto destroy;
1467
1468         err = vmci_transport_send_qp_offer(sk, handle);
1469         if (err < 0) {
1470                 err = vmci_transport_error_to_vsock_error(err);
1471                 goto destroy;
1472         }
1473
1474         vmci_trans(vsk)->qp_handle = handle;
1475         vmci_trans(vsk)->qpair = qpair;
1476
1477         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1478                 pkt->u.size;
1479
1480         vmci_trans(vsk)->attach_sub_id = attach_sub_id;
1481         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1482
1483         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1484
1485         return 0;
1486
1487 destroy:
1488         if (attach_sub_id != VMCI_INVALID_ID)
1489                 vmci_event_unsubscribe(attach_sub_id);
1490
1491         if (detach_sub_id != VMCI_INVALID_ID)
1492                 vmci_event_unsubscribe(detach_sub_id);
1493
1494         if (!vmci_handle_is_invalid(handle))
1495                 vmci_qpair_detach(&qpair);
1496
1497         return err;
1498 }
1499
1500 static int
1501 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1502                                               struct vmci_transport_packet *pkt)
1503 {
1504         int err = 0;
1505         struct vsock_sock *vsk = vsock_sk(sk);
1506
1507         if (vsk->sent_request) {
1508                 vsk->sent_request = false;
1509                 vsk->ignore_connecting_rst = true;
1510
1511                 err = vmci_transport_send_conn_request(
1512                         sk, vmci_trans(vsk)->queue_pair_size);
1513                 if (err < 0)
1514                         err = vmci_transport_error_to_vsock_error(err);
1515                 else
1516                         err = 0;
1517
1518         }
1519
1520         return err;
1521 }
1522
1523 static int vmci_transport_recv_connected(struct sock *sk,
1524                                          struct vmci_transport_packet *pkt)
1525 {
1526         struct vsock_sock *vsk;
1527         bool pkt_processed = false;
1528
1529         /* In cases where we are closing the connection, it's sufficient to
1530          * mark the state change (and maybe error) and wake up any waiting
1531          * threads. Since this is a connected socket, it's owned by a user
1532          * process and will be cleaned up when the failure is passed back on
1533          * the current or next system call.  Our system call implementations
1534          * must therefore check for error and state changes on entry and when
1535          * being awoken.
1536          */
1537         switch (pkt->type) {
1538         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1539                 if (pkt->u.mode) {
1540                         vsk = vsock_sk(sk);
1541
1542                         vsk->peer_shutdown |= pkt->u.mode;
1543                         sk->sk_state_change(sk);
1544                 }
1545                 break;
1546
1547         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1548                 vsk = vsock_sk(sk);
1549                 /* It is possible that we sent our peer a message (e.g a
1550                  * WAITING_READ) right before we got notified that the peer had
1551                  * detached. If that happens then we can get a RST pkt back
1552                  * from our peer even though there is data available for us to
1553                  * read. In that case, don't shutdown the socket completely but
1554                  * instead allow the local client to finish reading data off
1555                  * the queuepair. Always treat a RST pkt in connected mode like
1556                  * a clean shutdown.
1557                  */
1558                 sock_set_flag(sk, SOCK_DONE);
1559                 vsk->peer_shutdown = SHUTDOWN_MASK;
1560                 if (vsock_stream_has_data(vsk) <= 0)
1561                         sk->sk_state = SS_DISCONNECTING;
1562
1563                 sk->sk_state_change(sk);
1564                 break;
1565
1566         default:
1567                 vsk = vsock_sk(sk);
1568                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1569                                 sk, pkt, false, NULL, NULL,
1570                                 &pkt_processed);
1571                 if (!pkt_processed)
1572                         return -EINVAL;
1573
1574                 break;
1575         }
1576
1577         return 0;
1578 }
1579
1580 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1581                                       struct vsock_sock *psk)
1582 {
1583         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1584         if (!vsk->trans)
1585                 return -ENOMEM;
1586
1587         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1588         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1589         vmci_trans(vsk)->qpair = NULL;
1590         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1591         vmci_trans(vsk)->attach_sub_id = vmci_trans(vsk)->detach_sub_id =
1592                 VMCI_INVALID_ID;
1593         vmci_trans(vsk)->notify_ops = NULL;
1594         if (psk) {
1595                 vmci_trans(vsk)->queue_pair_size =
1596                         vmci_trans(psk)->queue_pair_size;
1597                 vmci_trans(vsk)->queue_pair_min_size =
1598                         vmci_trans(psk)->queue_pair_min_size;
1599                 vmci_trans(vsk)->queue_pair_max_size =
1600                         vmci_trans(psk)->queue_pair_max_size;
1601         } else {
1602                 vmci_trans(vsk)->queue_pair_size =
1603                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1604                 vmci_trans(vsk)->queue_pair_min_size =
1605                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1606                 vmci_trans(vsk)->queue_pair_max_size =
1607                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1608         }
1609
1610         return 0;
1611 }
1612
1613 static void vmci_transport_destruct(struct vsock_sock *vsk)
1614 {
1615         if (vmci_trans(vsk)->attach_sub_id != VMCI_INVALID_ID) {
1616                 vmci_event_unsubscribe(vmci_trans(vsk)->attach_sub_id);
1617                 vmci_trans(vsk)->attach_sub_id = VMCI_INVALID_ID;
1618         }
1619
1620         if (vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1621                 vmci_event_unsubscribe(vmci_trans(vsk)->detach_sub_id);
1622                 vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1623         }
1624
1625         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
1626                 vmci_qpair_detach(&vmci_trans(vsk)->qpair);
1627                 vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1628                 vmci_trans(vsk)->produce_size = 0;
1629                 vmci_trans(vsk)->consume_size = 0;
1630         }
1631
1632         if (vmci_trans(vsk)->notify_ops)
1633                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1634
1635         kfree(vsk->trans);
1636         vsk->trans = NULL;
1637 }
1638
1639 static void vmci_transport_release(struct vsock_sock *vsk)
1640 {
1641         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1642                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1643                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1644         }
1645 }
1646
1647 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1648                                      struct sockaddr_vm *addr)
1649 {
1650         u32 port;
1651         u32 flags;
1652         int err;
1653
1654         /* VMCI will select a resource ID for us if we provide
1655          * VMCI_INVALID_ID.
1656          */
1657         port = addr->svm_port == VMADDR_PORT_ANY ?
1658                         VMCI_INVALID_ID : addr->svm_port;
1659
1660         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1661                 return -EACCES;
1662
1663         flags = addr->svm_cid == VMADDR_CID_ANY ?
1664                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1665
1666         err = vmci_transport_datagram_create_hnd(port, flags,
1667                                                  vmci_transport_recv_dgram_cb,
1668                                                  &vsk->sk,
1669                                                  &vmci_trans(vsk)->dg_handle);
1670         if (err < VMCI_SUCCESS)
1671                 return vmci_transport_error_to_vsock_error(err);
1672         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1673                         vmci_trans(vsk)->dg_handle.resource);
1674
1675         return 0;
1676 }
1677
1678 static int vmci_transport_dgram_enqueue(
1679         struct vsock_sock *vsk,
1680         struct sockaddr_vm *remote_addr,
1681         struct iovec *iov,
1682         size_t len)
1683 {
1684         int err;
1685         struct vmci_datagram *dg;
1686
1687         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1688                 return -EMSGSIZE;
1689
1690         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1691                 return -EPERM;
1692
1693         /* Allocate a buffer for the user's message and our packet header. */
1694         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1695         if (!dg)
1696                 return -ENOMEM;
1697
1698         memcpy_fromiovec(VMCI_DG_PAYLOAD(dg), iov, len);
1699
1700         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1701                                    remote_addr->svm_port);
1702         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1703                                    vsk->local_addr.svm_port);
1704         dg->payload_size = len;
1705
1706         err = vmci_datagram_send(dg);
1707         kfree(dg);
1708         if (err < 0)
1709                 return vmci_transport_error_to_vsock_error(err);
1710
1711         return err - sizeof(*dg);
1712 }
1713
1714 static int vmci_transport_dgram_dequeue(struct kiocb *kiocb,
1715                                         struct vsock_sock *vsk,
1716                                         struct msghdr *msg, size_t len,
1717                                         int flags)
1718 {
1719         int err;
1720         int noblock;
1721         struct vmci_datagram *dg;
1722         size_t payload_len;
1723         struct sk_buff *skb;
1724
1725         noblock = flags & MSG_DONTWAIT;
1726
1727         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1728                 return -EOPNOTSUPP;
1729
1730         /* Retrieve the head sk_buff from the socket's receive queue. */
1731         err = 0;
1732         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1733         if (err)
1734                 return err;
1735
1736         if (!skb)
1737                 return -EAGAIN;
1738
1739         dg = (struct vmci_datagram *)skb->data;
1740         if (!dg)
1741                 /* err is 0, meaning we read zero bytes. */
1742                 goto out;
1743
1744         payload_len = dg->payload_size;
1745         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1746         if (payload_len != skb->len - sizeof(*dg)) {
1747                 err = -EINVAL;
1748                 goto out;
1749         }
1750
1751         if (payload_len > len) {
1752                 payload_len = len;
1753                 msg->msg_flags |= MSG_TRUNC;
1754         }
1755
1756         /* Place the datagram payload in the user's iovec. */
1757         err = skb_copy_datagram_iovec(skb, sizeof(*dg), msg->msg_iov,
1758                 payload_len);
1759         if (err)
1760                 goto out;
1761
1762         msg->msg_namelen = 0;
1763         if (msg->msg_name) {
1764                 struct sockaddr_vm *vm_addr;
1765
1766                 /* Provide the address of the sender. */
1767                 vm_addr = (struct sockaddr_vm *)msg->msg_name;
1768                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1769                 msg->msg_namelen = sizeof(*vm_addr);
1770         }
1771         err = payload_len;
1772
1773 out:
1774         skb_free_datagram(&vsk->sk, skb);
1775         return err;
1776 }
1777
1778 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1779 {
1780         if (cid == VMADDR_CID_HYPERVISOR) {
1781                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1782                  * state and are allowed.
1783                  */
1784                 return port == VMCI_UNITY_PBRPC_REGISTER;
1785         }
1786
1787         return true;
1788 }
1789
1790 static int vmci_transport_connect(struct vsock_sock *vsk)
1791 {
1792         int err;
1793         bool old_pkt_proto = false;
1794         struct sock *sk = &vsk->sk;
1795
1796         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1797                 old_pkt_proto) {
1798                 err = vmci_transport_send_conn_request(
1799                         sk, vmci_trans(vsk)->queue_pair_size);
1800                 if (err < 0) {
1801                         sk->sk_state = SS_UNCONNECTED;
1802                         return err;
1803                 }
1804         } else {
1805                 int supported_proto_versions =
1806                         vmci_transport_new_proto_supported_versions();
1807                 err = vmci_transport_send_conn_request2(
1808                                 sk, vmci_trans(vsk)->queue_pair_size,
1809                                 supported_proto_versions);
1810                 if (err < 0) {
1811                         sk->sk_state = SS_UNCONNECTED;
1812                         return err;
1813                 }
1814
1815                 vsk->sent_request = true;
1816         }
1817
1818         return err;
1819 }
1820
1821 static ssize_t vmci_transport_stream_dequeue(
1822         struct vsock_sock *vsk,
1823         struct iovec *iov,
1824         size_t len,
1825         int flags)
1826 {
1827         if (flags & MSG_PEEK)
1828                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, iov, len, 0);
1829         else
1830                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, iov, len, 0);
1831 }
1832
1833 static ssize_t vmci_transport_stream_enqueue(
1834         struct vsock_sock *vsk,
1835         struct iovec *iov,
1836         size_t len)
1837 {
1838         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, iov, len, 0);
1839 }
1840
1841 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1842 {
1843         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1844 }
1845
1846 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1847 {
1848         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1849 }
1850
1851 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1852 {
1853         return vmci_trans(vsk)->consume_size;
1854 }
1855
1856 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1857 {
1858         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1859 }
1860
1861 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1862 {
1863         return vmci_trans(vsk)->queue_pair_size;
1864 }
1865
1866 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1867 {
1868         return vmci_trans(vsk)->queue_pair_min_size;
1869 }
1870
1871 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1872 {
1873         return vmci_trans(vsk)->queue_pair_max_size;
1874 }
1875
1876 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1877 {
1878         if (val < vmci_trans(vsk)->queue_pair_min_size)
1879                 vmci_trans(vsk)->queue_pair_min_size = val;
1880         if (val > vmci_trans(vsk)->queue_pair_max_size)
1881                 vmci_trans(vsk)->queue_pair_max_size = val;
1882         vmci_trans(vsk)->queue_pair_size = val;
1883 }
1884
1885 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1886                                                u64 val)
1887 {
1888         if (val > vmci_trans(vsk)->queue_pair_size)
1889                 vmci_trans(vsk)->queue_pair_size = val;
1890         vmci_trans(vsk)->queue_pair_min_size = val;
1891 }
1892
1893 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1894                                                u64 val)
1895 {
1896         if (val < vmci_trans(vsk)->queue_pair_size)
1897                 vmci_trans(vsk)->queue_pair_size = val;
1898         vmci_trans(vsk)->queue_pair_max_size = val;
1899 }
1900
1901 static int vmci_transport_notify_poll_in(
1902         struct vsock_sock *vsk,
1903         size_t target,
1904         bool *data_ready_now)
1905 {
1906         return vmci_trans(vsk)->notify_ops->poll_in(
1907                         &vsk->sk, target, data_ready_now);
1908 }
1909
1910 static int vmci_transport_notify_poll_out(
1911         struct vsock_sock *vsk,
1912         size_t target,
1913         bool *space_available_now)
1914 {
1915         return vmci_trans(vsk)->notify_ops->poll_out(
1916                         &vsk->sk, target, space_available_now);
1917 }
1918
1919 static int vmci_transport_notify_recv_init(
1920         struct vsock_sock *vsk,
1921         size_t target,
1922         struct vsock_transport_recv_notify_data *data)
1923 {
1924         return vmci_trans(vsk)->notify_ops->recv_init(
1925                         &vsk->sk, target,
1926                         (struct vmci_transport_recv_notify_data *)data);
1927 }
1928
1929 static int vmci_transport_notify_recv_pre_block(
1930         struct vsock_sock *vsk,
1931         size_t target,
1932         struct vsock_transport_recv_notify_data *data)
1933 {
1934         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1935                         &vsk->sk, target,
1936                         (struct vmci_transport_recv_notify_data *)data);
1937 }
1938
1939 static int vmci_transport_notify_recv_pre_dequeue(
1940         struct vsock_sock *vsk,
1941         size_t target,
1942         struct vsock_transport_recv_notify_data *data)
1943 {
1944         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1945                         &vsk->sk, target,
1946                         (struct vmci_transport_recv_notify_data *)data);
1947 }
1948
1949 static int vmci_transport_notify_recv_post_dequeue(
1950         struct vsock_sock *vsk,
1951         size_t target,
1952         ssize_t copied,
1953         bool data_read,
1954         struct vsock_transport_recv_notify_data *data)
1955 {
1956         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1957                         &vsk->sk, target, copied, data_read,
1958                         (struct vmci_transport_recv_notify_data *)data);
1959 }
1960
1961 static int vmci_transport_notify_send_init(
1962         struct vsock_sock *vsk,
1963         struct vsock_transport_send_notify_data *data)
1964 {
1965         return vmci_trans(vsk)->notify_ops->send_init(
1966                         &vsk->sk,
1967                         (struct vmci_transport_send_notify_data *)data);
1968 }
1969
1970 static int vmci_transport_notify_send_pre_block(
1971         struct vsock_sock *vsk,
1972         struct vsock_transport_send_notify_data *data)
1973 {
1974         return vmci_trans(vsk)->notify_ops->send_pre_block(
1975                         &vsk->sk,
1976                         (struct vmci_transport_send_notify_data *)data);
1977 }
1978
1979 static int vmci_transport_notify_send_pre_enqueue(
1980         struct vsock_sock *vsk,
1981         struct vsock_transport_send_notify_data *data)
1982 {
1983         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1984                         &vsk->sk,
1985                         (struct vmci_transport_send_notify_data *)data);
1986 }
1987
1988 static int vmci_transport_notify_send_post_enqueue(
1989         struct vsock_sock *vsk,
1990         ssize_t written,
1991         struct vsock_transport_send_notify_data *data)
1992 {
1993         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1994                         &vsk->sk, written,
1995                         (struct vmci_transport_send_notify_data *)data);
1996 }
1997
1998 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1999 {
2000         if (PROTOCOL_OVERRIDE != -1) {
2001                 if (PROTOCOL_OVERRIDE == 0)
2002                         *old_pkt_proto = true;
2003                 else
2004                         *old_pkt_proto = false;
2005
2006                 pr_info("Proto override in use\n");
2007                 return true;
2008         }
2009
2010         return false;
2011 }
2012
2013 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2014                                                   u16 *proto,
2015                                                   bool old_pkt_proto)
2016 {
2017         struct vsock_sock *vsk = vsock_sk(sk);
2018
2019         if (old_pkt_proto) {
2020                 if (*proto != VSOCK_PROTO_INVALID) {
2021                         pr_err("Can't set both an old and new protocol\n");
2022                         return false;
2023                 }
2024                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2025                 goto exit;
2026         }
2027
2028         switch (*proto) {
2029         case VSOCK_PROTO_PKT_ON_NOTIFY:
2030                 vmci_trans(vsk)->notify_ops =
2031                         &vmci_transport_notify_pkt_q_state_ops;
2032                 break;
2033         default:
2034                 pr_err("Unknown notify protocol version\n");
2035                 return false;
2036         }
2037
2038 exit:
2039         vmci_trans(vsk)->notify_ops->socket_init(sk);
2040         return true;
2041 }
2042
2043 static u16 vmci_transport_new_proto_supported_versions(void)
2044 {
2045         if (PROTOCOL_OVERRIDE != -1)
2046                 return PROTOCOL_OVERRIDE;
2047
2048         return VSOCK_PROTO_ALL_SUPPORTED;
2049 }
2050
2051 static u32 vmci_transport_get_local_cid(void)
2052 {
2053         return vmci_get_context_id();
2054 }
2055
2056 static struct vsock_transport vmci_transport = {
2057         .init = vmci_transport_socket_init,
2058         .destruct = vmci_transport_destruct,
2059         .release = vmci_transport_release,
2060         .connect = vmci_transport_connect,
2061         .dgram_bind = vmci_transport_dgram_bind,
2062         .dgram_dequeue = vmci_transport_dgram_dequeue,
2063         .dgram_enqueue = vmci_transport_dgram_enqueue,
2064         .dgram_allow = vmci_transport_dgram_allow,
2065         .stream_dequeue = vmci_transport_stream_dequeue,
2066         .stream_enqueue = vmci_transport_stream_enqueue,
2067         .stream_has_data = vmci_transport_stream_has_data,
2068         .stream_has_space = vmci_transport_stream_has_space,
2069         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2070         .stream_is_active = vmci_transport_stream_is_active,
2071         .stream_allow = vmci_transport_stream_allow,
2072         .notify_poll_in = vmci_transport_notify_poll_in,
2073         .notify_poll_out = vmci_transport_notify_poll_out,
2074         .notify_recv_init = vmci_transport_notify_recv_init,
2075         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2076         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2077         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2078         .notify_send_init = vmci_transport_notify_send_init,
2079         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2080         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2081         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2082         .shutdown = vmci_transport_shutdown,
2083         .set_buffer_size = vmci_transport_set_buffer_size,
2084         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2085         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2086         .get_buffer_size = vmci_transport_get_buffer_size,
2087         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2088         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2089         .get_local_cid = vmci_transport_get_local_cid,
2090 };
2091
2092 static int __init vmci_transport_init(void)
2093 {
2094         int err;
2095
2096         /* Create the datagram handle that we will use to send and receive all
2097          * VSocket control messages for this context.
2098          */
2099         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2100                                                  VMCI_FLAG_ANYCID_DG_HND,
2101                                                  vmci_transport_recv_stream_cb,
2102                                                  NULL,
2103                                                  &vmci_transport_stream_handle);
2104         if (err < VMCI_SUCCESS) {
2105                 pr_err("Unable to create datagram handle. (%d)\n", err);
2106                 return vmci_transport_error_to_vsock_error(err);
2107         }
2108
2109         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2110                                    vmci_transport_qp_resumed_cb,
2111                                    NULL, &vmci_transport_qp_resumed_sub_id);
2112         if (err < VMCI_SUCCESS) {
2113                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2114                 err = vmci_transport_error_to_vsock_error(err);
2115                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2116                 goto err_destroy_stream_handle;
2117         }
2118
2119         err = vsock_core_init(&vmci_transport);
2120         if (err < 0)
2121                 goto err_unsubscribe;
2122
2123         return 0;
2124
2125 err_unsubscribe:
2126         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2127 err_destroy_stream_handle:
2128         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2129         return err;
2130 }
2131 module_init(vmci_transport_init);
2132
2133 static void __exit vmci_transport_exit(void)
2134 {
2135         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2136                 if (vmci_datagram_destroy_handle(
2137                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2138                         pr_err("Couldn't destroy datagram handle\n");
2139                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2140         }
2141
2142         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2143                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2144                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2145         }
2146
2147         vsock_core_exit();
2148 }
2149 module_exit(vmci_transport_exit);
2150
2151 MODULE_AUTHOR("VMware, Inc.");
2152 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2153 MODULE_LICENSE("GPL v2");
2154 MODULE_ALIAS("vmware_vsock");
2155 MODULE_ALIAS_NETPROTO(PF_VSOCK);