net/vmw_vsock/vmci_transport.c

   1 /*
   2  * VMware vSockets Driver
   3  *
   4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation version 2 and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  */
  15
  16 #include <linux/types.h>
  17 #include <linux/bitops.h>
  18 #include <linux/cred.h>
  19 #include <linux/init.h>
  20 #include <linux/io.h>
  21 #include <linux/kernel.h>
  22 #include <linux/kmod.h>
  23 #include <linux/list.h>
  24 #include <linux/module.h>
  25 #include <linux/mutex.h>
  26 #include <linux/net.h>
  27 #include <linux/poll.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/smp.h>
  30 #include <linux/socket.h>
  31 #include <linux/stddef.h>
  32 #include <linux/unistd.h>
  33 #include <linux/wait.h>
  34 #include <linux/workqueue.h>
  35 #include <net/sock.h>
  36 #include <net/af_vsock.h>
  37
  38 #include "vmci_transport_notify.h"
  39
  40 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
  41 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
  42 static void vmci_transport_peer_detach_cb(u32 sub_id,
  43                                           const struct vmci_event_data *ed,
  44                                           void *client_data);
  45 static void vmci_transport_recv_pkt_work(struct work_struct *work);
  46 static void vmci_transport_cleanup(struct work_struct *work);
  47 static int vmci_transport_recv_listen(struct sock *sk,
  48                                       struct vmci_transport_packet *pkt);
  49 static int vmci_transport_recv_connecting_server(
  50                                         struct sock *sk,
  51                                         struct sock *pending,
  52                                         struct vmci_transport_packet *pkt);
  53 static int vmci_transport_recv_connecting_client(
  54                                         struct sock *sk,
  55                                         struct vmci_transport_packet *pkt);
  56 static int vmci_transport_recv_connecting_client_negotiate(
  57                                         struct sock *sk,
  58                                         struct vmci_transport_packet *pkt);
  59 static int vmci_transport_recv_connecting_client_invalid(
  60                                         struct sock *sk,
  61                                         struct vmci_transport_packet *pkt);
  62 static int vmci_transport_recv_connected(struct sock *sk,
  63                                          struct vmci_transport_packet *pkt);
  64 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
  65 static u16 vmci_transport_new_proto_supported_versions(void);
  66 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
  67                                                   bool old_pkt_proto);
  68
  69 struct vmci_transport_recv_pkt_info {
  70         struct work_struct work;
  71         struct sock *sk;
  72         struct vmci_transport_packet pkt;
  73 };
  74
  75 static LIST_HEAD(vmci_transport_cleanup_list);
  76 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
  77 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
  78
  79 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  80                                                            VMCI_INVALID_ID };
  81 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  82
  83 static int PROTOCOL_OVERRIDE = -1;
  84
  85 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN   128
  86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE       262144
  87 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX   262144
  88
  89 /* The default peer timeout indicates how long we will wait for a peer response
  90  * to a control message.
  91  */
  92 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
  93
  94 /* Helper function to convert from a VMCI error code to a VSock error code. */
  95
  96 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  97 {
  98         switch (vmci_error) {
  99         case VMCI_ERROR_NO_MEM:
 100                 return -ENOMEM;
 101         case VMCI_ERROR_DUPLICATE_ENTRY:
 102         case VMCI_ERROR_ALREADY_EXISTS:
 103                 return -EADDRINUSE;
 104         case VMCI_ERROR_NO_ACCESS:
 105                 return -EPERM;
 106         case VMCI_ERROR_NO_RESOURCES:
 107                 return -ENOBUFS;
 108         case VMCI_ERROR_INVALID_RESOURCE:
 109                 return -EHOSTUNREACH;
 110         case VMCI_ERROR_INVALID_ARGS:
 111         default:
 112                 break;
 113         }
 114         return -EINVAL;
 115 }
 116
 117 static u32 vmci_transport_peer_rid(u32 peer_cid)
 118 {
 119         if (VMADDR_CID_HYPERVISOR == peer_cid)
 120                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 121
 122         return VMCI_TRANSPORT_PACKET_RID;
 123 }
 124
 125 static inline void
 126 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 127                            struct sockaddr_vm *src,
 128                            struct sockaddr_vm *dst,
 129                            u8 type,
 130                            u64 size,
 131                            u64 mode,
 132                            struct vmci_transport_waiting_info *wait,
 133                            u16 proto,
 134                            struct vmci_handle handle)
 135 {
 136         /* We register the stream control handler as an any cid handle so we
 137          * must always send from a source address of VMADDR_CID_ANY
 138          */
 139         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 140                                        VMCI_TRANSPORT_PACKET_RID);
 141         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 142                                        vmci_transport_peer_rid(dst->svm_cid));
 143         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 144         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 145         pkt->type = type;
 146         pkt->src_port = src->svm_port;
 147         pkt->dst_port = dst->svm_port;
 148         memset(&pkt->proto, 0, sizeof(pkt->proto));
 149         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 150
 151         switch (pkt->type) {
 152         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 153                 pkt->u.size = 0;
 154                 break;
 155
 156         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 157         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 158                 pkt->u.size = size;
 159                 break;
 160
 161         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 162         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 163                 pkt->u.handle = handle;
 164                 break;
 165
 166         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 167         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 168         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 169                 pkt->u.size = 0;
 170                 break;
 171
 172         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 173                 pkt->u.mode = mode;
 174                 break;
 175
 176         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 177         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 178                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 179                 break;
 180
 181         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 182         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 183                 pkt->u.size = size;
 184                 pkt->proto = proto;
 185                 break;
 186         }
 187 }
 188
 189 static inline void
 190 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 191                                     struct sockaddr_vm *local,
 192                                     struct sockaddr_vm *remote)
 193 {
 194         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 195         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 196 }
 197
 198 static int
 199 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 200                                   struct sockaddr_vm *src,
 201                                   struct sockaddr_vm *dst,
 202                                   enum vmci_transport_packet_type type,
 203                                   u64 size,
 204                                   u64 mode,
 205                                   struct vmci_transport_waiting_info *wait,
 206                                   u16 proto,
 207                                   struct vmci_handle handle,
 208                                   bool convert_error)
 209 {
 210         int err;
 211
 212         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 213                                    proto, handle);
 214         err = vmci_datagram_send(&pkt->dg);
 215         if (convert_error && (err < 0))
 216                 return vmci_transport_error_to_vsock_error(err);
 217
 218         return err;
 219 }
 220
 221 static int
 222 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 223                                       enum vmci_transport_packet_type type,
 224                                       u64 size,
 225                                       u64 mode,
 226                                       struct vmci_transport_waiting_info *wait,
 227                                       struct vmci_handle handle)
 228 {
 229         struct vmci_transport_packet reply;
 230         struct sockaddr_vm src, dst;
 231
 232         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 233                 return 0;
 234         } else {
 235                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 236                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 237                                                          type,
 238                                                          size, mode, wait,
 239                                                          VSOCK_PROTO_INVALID,
 240                                                          handle, true);
 241         }
 242 }
 243
 244 static int
 245 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 246                                    struct sockaddr_vm *dst,
 247                                    enum vmci_transport_packet_type type,
 248                                    u64 size,
 249                                    u64 mode,
 250                                    struct vmci_transport_waiting_info *wait,
 251                                    struct vmci_handle handle)
 252 {
 253         /* Note that it is safe to use a single packet across all CPUs since
 254          * two tasklets of the same type are guaranteed to not ever run
 255          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 256          * we can use per-cpu packets.
 257          */
 258         static struct vmci_transport_packet pkt;
 259
 260         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 261                                                  size, mode, wait,
 262                                                  VSOCK_PROTO_INVALID, handle,
 263                                                  false);
 264 }
 265
 266 static int
 267 vmci_transport_alloc_send_control_pkt(struct sockaddr_vm *src,
 268                                       struct sockaddr_vm *dst,
 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         int err;
 278
 279         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 280         if (!pkt)
 281                 return -ENOMEM;
 282
 283         err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
 284                                                 mode, wait, proto, handle,
 285                                                 true);
 286         kfree(pkt);
 287
 288         return err;
 289 }
 290
 291 static int
 292 vmci_transport_send_control_pkt(struct sock *sk,
 293                                 enum vmci_transport_packet_type type,
 294                                 u64 size,
 295                                 u64 mode,
 296                                 struct vmci_transport_waiting_info *wait,
 297                                 u16 proto,
 298                                 struct vmci_handle handle)
 299 {
 300         struct vsock_sock *vsk;
 301
 302         vsk = vsock_sk(sk);
 303
 304         if (!vsock_addr_bound(&vsk->local_addr))
 305                 return -EINVAL;
 306
 307         if (!vsock_addr_bound(&vsk->remote_addr))
 308                 return -EINVAL;
 309
 310         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
 311                                                      &vsk->remote_addr,
 312                                                      type, size, mode,
 313                                                      wait, proto, handle);
 314 }
 315
 316 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 317                                         struct sockaddr_vm *src,
 318                                         struct vmci_transport_packet *pkt)
 319 {
 320         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 321                 return 0;
 322         return vmci_transport_send_control_pkt_bh(
 323                                         dst, src,
 324                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 325                                         0, NULL, VMCI_INVALID_HANDLE);
 326 }
 327
 328 static int vmci_transport_send_reset(struct sock *sk,
 329                                      struct vmci_transport_packet *pkt)
 330 {
 331         struct sockaddr_vm *dst_ptr;
 332         struct sockaddr_vm dst;
 333         struct vsock_sock *vsk;
 334
 335         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 336                 return 0;
 337
 338         vsk = vsock_sk(sk);
 339
 340         if (!vsock_addr_bound(&vsk->local_addr))
 341                 return -EINVAL;
 342
 343         if (vsock_addr_bound(&vsk->remote_addr)) {
 344                 dst_ptr = &vsk->remote_addr;
 345         } else {
 346                 vsock_addr_init(&dst, pkt->dg.src.context,
 347                                 pkt->src_port);
 348                 dst_ptr = &dst;
 349         }
 350         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
 351                                              VMCI_TRANSPORT_PACKET_TYPE_RST,
 352                                              0, 0, NULL, VSOCK_PROTO_INVALID,
 353                                              VMCI_INVALID_HANDLE);
 354 }
 355
 356 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 357 {
 358         return vmci_transport_send_control_pkt(
 359                                         sk,
 360                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 361                                         size, 0, NULL,
 362                                         VSOCK_PROTO_INVALID,
 363                                         VMCI_INVALID_HANDLE);
 364 }
 365
 366 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 367                                           u16 version)
 368 {
 369         return vmci_transport_send_control_pkt(
 370                                         sk,
 371                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 372                                         size, 0, NULL, version,
 373                                         VMCI_INVALID_HANDLE);
 374 }
 375
 376 static int vmci_transport_send_qp_offer(struct sock *sk,
 377                                         struct vmci_handle handle)
 378 {
 379         return vmci_transport_send_control_pkt(
 380                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 381                                         0, NULL,
 382                                         VSOCK_PROTO_INVALID, handle);
 383 }
 384
 385 static int vmci_transport_send_attach(struct sock *sk,
 386                                       struct vmci_handle handle)
 387 {
 388         return vmci_transport_send_control_pkt(
 389                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 390                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 391                                         handle);
 392 }
 393
 394 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 395 {
 396         return vmci_transport_reply_control_pkt_fast(
 397                                                 pkt,
 398                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 399                                                 0, 0, NULL,
 400                                                 VMCI_INVALID_HANDLE);
 401 }
 402
 403 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 404                                           struct sockaddr_vm *src)
 405 {
 406         return vmci_transport_send_control_pkt_bh(
 407                                         dst, src,
 408                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 409                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 410 }
 411
 412 int vmci_transport_send_wrote_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_WROTE, 0,
 418                                         0, NULL, VMCI_INVALID_HANDLE);
 419 }
 420
 421 int vmci_transport_send_read_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_READ, 0,
 427                                         0, NULL, VMCI_INVALID_HANDLE);
 428 }
 429
 430 int vmci_transport_send_wrote(struct sock *sk)
 431 {
 432         return vmci_transport_send_control_pkt(
 433                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 434                                         0, NULL, VSOCK_PROTO_INVALID,
 435                                         VMCI_INVALID_HANDLE);
 436 }
 437
 438 int vmci_transport_send_read(struct sock *sk)
 439 {
 440         return vmci_transport_send_control_pkt(
 441                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 442                                         0, NULL, VSOCK_PROTO_INVALID,
 443                                         VMCI_INVALID_HANDLE);
 444 }
 445
 446 int vmci_transport_send_waiting_write(struct sock *sk,
 447                                       struct vmci_transport_waiting_info *wait)
 448 {
 449         return vmci_transport_send_control_pkt(
 450                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 451                                 0, 0, wait, VSOCK_PROTO_INVALID,
 452                                 VMCI_INVALID_HANDLE);
 453 }
 454
 455 int vmci_transport_send_waiting_read(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_READ,
 460                                 0, 0, wait, VSOCK_PROTO_INVALID,
 461                                 VMCI_INVALID_HANDLE);
 462 }
 463
 464 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 465 {
 466         return vmci_transport_send_control_pkt(
 467                                         &vsk->sk,
 468                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 469                                         0, mode, NULL,
 470                                         VSOCK_PROTO_INVALID,
 471                                         VMCI_INVALID_HANDLE);
 472 }
 473
 474 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 475 {
 476         return vmci_transport_send_control_pkt(sk,
 477                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 478                                         size, 0, NULL,
 479                                         VSOCK_PROTO_INVALID,
 480                                         VMCI_INVALID_HANDLE);
 481 }
 482
 483 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 484                                              u16 version)
 485 {
 486         return vmci_transport_send_control_pkt(
 487                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 488                                         size, 0, NULL, version,
 489                                         VMCI_INVALID_HANDLE);
 490 }
 491
 492 static struct sock *vmci_transport_get_pending(
 493                                         struct sock *listener,
 494                                         struct vmci_transport_packet *pkt)
 495 {
 496         struct vsock_sock *vlistener;
 497         struct vsock_sock *vpending;
 498         struct sock *pending;
 499         struct sockaddr_vm src;
 500
 501         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 502
 503         vlistener = vsock_sk(listener);
 504
 505         list_for_each_entry(vpending, &vlistener->pending_links,
 506                             pending_links) {
 507                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 508                     pkt->dst_port == vpending->local_addr.svm_port) {
 509                         pending = sk_vsock(vpending);
 510                         sock_hold(pending);
 511                         goto found;
 512                 }
 513         }
 514
 515         pending = NULL;
 516 found:
 517         return pending;
 518
 519 }
 520
 521 static void vmci_transport_release_pending(struct sock *pending)
 522 {
 523         sock_put(pending);
 524 }
 525
 526 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 527  * trusted sockets 2) sockets from applications running as the same user as the
 528  * VM (this is only true for the host side and only when using hosted products)
 529  */
 530
 531 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 532 {
 533         return vsock->trusted ||
 534                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 535 }
 536
 537 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 538  * only if it is trusted as described in vmci_transport_is_trusted.
 539  */
 540
 541 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 542 {
 543         if (VMADDR_CID_HYPERVISOR == peer_cid)
 544                 return true;
 545
 546         if (vsock->cached_peer != peer_cid) {
 547                 vsock->cached_peer = peer_cid;
 548                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 549                     (vmci_context_get_priv_flags(peer_cid) &
 550                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 551                         vsock->cached_peer_allow_dgram = false;
 552                 } else {
 553                         vsock->cached_peer_allow_dgram = true;
 554                 }
 555         }
 556
 557         return vsock->cached_peer_allow_dgram;
 558 }
 559
 560 static int
 561 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 562                                 struct vmci_handle *handle,
 563                                 u64 produce_size,
 564                                 u64 consume_size,
 565                                 u32 peer, u32 flags, bool trusted)
 566 {
 567         int err = 0;
 568
 569         if (trusted) {
 570                 /* Try to allocate our queue pair as trusted. This will only
 571                  * work if vsock is running in the host.
 572                  */
 573
 574                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 575                                        consume_size,
 576                                        peer, flags,
 577                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 578                 if (err != VMCI_ERROR_NO_ACCESS)
 579                         goto out;
 580
 581         }
 582
 583         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 584                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 585 out:
 586         if (err < 0) {
 587                 pr_err_once("Could not attach to queue pair with %d\n", err);
 588                 err = vmci_transport_error_to_vsock_error(err);
 589         }
 590
 591         return err;
 592 }
 593
 594 static int
 595 vmci_transport_datagram_create_hnd(u32 resource_id,
 596                                    u32 flags,
 597                                    vmci_datagram_recv_cb recv_cb,
 598                                    void *client_data,
 599                                    struct vmci_handle *out_handle)
 600 {
 601         int err = 0;
 602
 603         /* Try to allocate our datagram handler as trusted. This will only work
 604          * if vsock is running in the host.
 605          */
 606
 607         err = vmci_datagram_create_handle_priv(resource_id, flags,
 608                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 609                                                recv_cb,
 610                                                client_data, out_handle);
 611
 612         if (err == VMCI_ERROR_NO_ACCESS)
 613                 err = vmci_datagram_create_handle(resource_id, flags,
 614                                                   recv_cb, client_data,
 615                                                   out_handle);
 616
 617         return err;
 618 }
 619
 620 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 621  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 622  * sleep it should defer that work to a work queue.
 623  */
 624
 625 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 626 {
 627         struct sock *sk;
 628         size_t size;
 629         struct sk_buff *skb;
 630         struct vsock_sock *vsk;
 631
 632         sk = (struct sock *)data;
 633
 634         /* This handler is privileged when this module is running on the host.
 635          * We will get datagrams from all endpoints (even VMs that are in a
 636          * restricted context). If we get one from a restricted context then
 637          * the destination socket must be trusted.
 638          *
 639          * NOTE: We access the socket struct without holding the lock here.
 640          * This is ok because the field we are interested is never modified
 641          * outside of the create and destruct socket functions.
 642          */
 643         vsk = vsock_sk(sk);
 644         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 645                 return VMCI_ERROR_NO_ACCESS;
 646
 647         size = VMCI_DG_SIZE(dg);
 648
 649         /* Attach the packet to the socket's receive queue as an sk_buff. */
 650         skb = alloc_skb(size, GFP_ATOMIC);
 651         if (!skb)
 652                 return VMCI_ERROR_NO_MEM;
 653
 654         /* sk_receive_skb() will do a sock_put(), so hold here. */
 655         sock_hold(sk);
 656         skb_put(skb, size);
 657         memcpy(skb->data, dg, size);
 658         sk_receive_skb(sk, skb, 0);
 659
 660         return VMCI_SUCCESS;
 661 }
 662
 663 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 664 {
 665         static const u32 non_socket_contexts[] = {
 666                 VMADDR_CID_RESERVED,
 667         };
 668         int i;
 669
 670         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 671
 672         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 673                 if (cid == non_socket_contexts[i])
 674                         return false;
 675         }
 676
 677         return true;
 678 }
 679
 680 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 681  * interrupt fires.  This is run in bottom-half context but it defers most of
 682  * its work to the packet handling work queue.
 683  */
 684
 685 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 686 {
 687         struct sock *sk;
 688         struct sockaddr_vm dst;
 689         struct sockaddr_vm src;
 690         struct vmci_transport_packet *pkt;
 691         struct vsock_sock *vsk;
 692         bool bh_process_pkt;
 693         int err;
 694
 695         sk = NULL;
 696         err = VMCI_SUCCESS;
 697         bh_process_pkt = false;
 698
 699         /* Ignore incoming packets from contexts without sockets, or resources
 700          * that aren't vsock implementations.
 701          */
 702
 703         if (!vmci_transport_stream_allow(dg->src.context, -1)
 704             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 705                 return VMCI_ERROR_NO_ACCESS;
 706
 707         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 708                 /* Drop datagrams that do not contain full VSock packets. */
 709                 return VMCI_ERROR_INVALID_ARGS;
 710
 711         pkt = (struct vmci_transport_packet *)dg;
 712
 713         /* Find the socket that should handle this packet.  First we look for a
 714          * connected socket and if there is none we look for a socket bound to
 715          * the destintation address.
 716          */
 717         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 718         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 719
 720         sk = vsock_find_connected_socket(&src, &dst);
 721         if (!sk) {
 722                 sk = vsock_find_bound_socket(&dst);
 723                 if (!sk) {
 724                         /* We could not find a socket for this specified
 725                          * address.  If this packet is a RST, we just drop it.
 726                          * If it is another packet, we send a RST.  Note that
 727                          * we do not send a RST reply to RSTs so that we do not
 728                          * continually send RSTs between two endpoints.
 729                          *
 730                          * Note that since this is a reply, dst is src and src
 731                          * is dst.
 732                          */
 733                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 734                                 pr_err("unable to send reset\n");
 735
 736                         err = VMCI_ERROR_NOT_FOUND;
 737                         goto out;
 738                 }
 739         }
 740
 741         /* If the received packet type is beyond all types known to this
 742          * implementation, reply with an invalid message.  Hopefully this will
 743          * help when implementing backwards compatibility in the future.
 744          */
 745         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 746                 vmci_transport_send_invalid_bh(&dst, &src);
 747                 err = VMCI_ERROR_INVALID_ARGS;
 748                 goto out;
 749         }
 750
 751         /* This handler is privileged when this module is running on the host.
 752          * We will get datagram connect requests from all endpoints (even VMs
 753          * that are in a restricted context). If we get one from a restricted
 754          * context then the destination socket must be trusted.
 755          *
 756          * NOTE: We access the socket struct without holding the lock here.
 757          * This is ok because the field we are interested is never modified
 758          * outside of the create and destruct socket functions.
 759          */
 760         vsk = vsock_sk(sk);
 761         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 762                 err = VMCI_ERROR_NO_ACCESS;
 763                 goto out;
 764         }
 765
 766         /* We do most everything in a work queue, but let's fast path the
 767          * notification of reads and writes to help data transfer performance.
 768          * We can only do this if there is no process context code executing
 769          * for this socket since that may change the state.
 770          */
 771         bh_lock_sock(sk);
 772
 773         if (!sock_owned_by_user(sk)) {
 774                 /* The local context ID may be out of date, update it. */
 775                 vsk->local_addr.svm_cid = dst.svm_cid;
 776
 777                 if (sk->sk_state == TCP_ESTABLISHED)
 778                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 779                                         sk, pkt, true, &dst, &src,
 780                                         &bh_process_pkt);
 781         }
 782
 783         bh_unlock_sock(sk);
 784
 785         if (!bh_process_pkt) {
 786                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 787
 788                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 789                 if (!recv_pkt_info) {
 790                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 791                                 pr_err("unable to send reset\n");
 792
 793                         err = VMCI_ERROR_NO_MEM;
 794                         goto out;
 795                 }
 796
 797                 recv_pkt_info->sk = sk;
 798                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 799                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 800
 801                 schedule_work(&recv_pkt_info->work);
 802                 /* Clear sk so that the reference count incremented by one of
 803                  * the Find functions above is not decremented below.  We need
 804                  * that reference count for the packet handler we've scheduled
 805                  * to run.
 806                  */
 807                 sk = NULL;
 808         }
 809
 810 out:
 811         if (sk)
 812                 sock_put(sk);
 813
 814         return err;
 815 }
 816
 817 static void vmci_transport_handle_detach(struct sock *sk)
 818 {
 819         struct vsock_sock *vsk;
 820
 821         vsk = vsock_sk(sk);
 822         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 823                 sock_set_flag(sk, SOCK_DONE);
 824
 825                 /* On a detach the peer will not be sending or receiving
 826                  * anymore.
 827                  */
 828                 vsk->peer_shutdown = SHUTDOWN_MASK;
 829
 830                 /* We should not be sending anymore since the peer won't be
 831                  * there to receive, but we can still receive if there is data
 832                  * left in our consume queue. If the local endpoint is a host,
 833                  * we can't call vsock_stream_has_data, since that may block,
 834                  * but a host endpoint can't read data once the VM has
 835                  * detached, so there is no available data in that case.
 836                  */
 837                 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
 838                     vsock_stream_has_data(vsk) <= 0) {
 839                         if (sk->sk_state == TCP_SYN_SENT) {
 840                                 /* The peer may detach from a queue pair while
 841                                  * we are still in the connecting state, i.e.,
 842                                  * if the peer VM is killed after attaching to
 843                                  * a queue pair, but before we complete the
 844                                  * handshake. In that case, we treat the detach
 845                                  * event like a reset.
 846                                  */
 847
 848                                 sk->sk_state = TCP_CLOSE;
 849                                 sk->sk_err = ECONNRESET;
 850                                 sk->sk_error_report(sk);
 851                                 return;
 852                         }
 853                         sk->sk_state = TCP_CLOSE;
 854                 }
 855                 sk->sk_state_change(sk);
 856         }
 857 }
 858
 859 static void vmci_transport_peer_detach_cb(u32 sub_id,
 860                                           const struct vmci_event_data *e_data,
 861                                           void *client_data)
 862 {
 863         struct vmci_transport *trans = client_data;
 864         const struct vmci_event_payload_qp *e_payload;
 865
 866         e_payload = vmci_event_data_const_payload(e_data);
 867
 868         /* XXX This is lame, we should provide a way to lookup sockets by
 869          * qp_handle.
 870          */
 871         if (vmci_handle_is_invalid(e_payload->handle) ||
 872             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 873                 return;
 874
 875         /* We don't ask for delayed CBs when we subscribe to this event (we
 876          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 877          * guarantees in that case about what context we might be running in,
 878          * so it could be BH or process, blockable or non-blockable.  So we
 879          * need to account for all possible contexts here.
 880          */
 881         spin_lock_bh(&trans->lock);
 882         if (!trans->sk)
 883                 goto out;
 884
 885         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 886          * where trans->sk isn't locked.
 887          */
 888         bh_lock_sock(trans->sk);
 889
 890         vmci_transport_handle_detach(trans->sk);
 891
 892         bh_unlock_sock(trans->sk);
 893  out:
 894         spin_unlock_bh(&trans->lock);
 895 }
 896
 897 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 898                                          const struct vmci_event_data *e_data,
 899                                          void *client_data)
 900 {
 901         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 902 }
 903
 904 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 905 {
 906         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 907         struct vmci_transport_packet *pkt;
 908         struct sock *sk;
 909
 910         recv_pkt_info =
 911                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 912         sk = recv_pkt_info->sk;
 913         pkt = &recv_pkt_info->pkt;
 914
 915         lock_sock(sk);
 916
 917         /* The local context ID may be out of date. */
 918         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 919
 920         switch (sk->sk_state) {
 921         case TCP_LISTEN:
 922                 vmci_transport_recv_listen(sk, pkt);
 923                 break;
 924         case TCP_SYN_SENT:
 925                 /* Processing of pending connections for servers goes through
 926                  * the listening socket, so see vmci_transport_recv_listen()
 927                  * for that path.
 928                  */
 929                 vmci_transport_recv_connecting_client(sk, pkt);
 930                 break;
 931         case TCP_ESTABLISHED:
 932                 vmci_transport_recv_connected(sk, pkt);
 933                 break;
 934         default:
 935                 /* Because this function does not run in the same context as
 936                  * vmci_transport_recv_stream_cb it is possible that the
 937                  * socket has closed. We need to let the other side know or it
 938                  * could be sitting in a connect and hang forever. Send a
 939                  * reset to prevent that.
 940                  */
 941                 vmci_transport_send_reset(sk, pkt);
 942                 break;
 943         }
 944
 945         release_sock(sk);
 946         kfree(recv_pkt_info);
 947         /* Release reference obtained in the stream callback when we fetched
 948          * this socket out of the bound or connected list.
 949          */
 950         sock_put(sk);
 951 }
 952
 953 static int vmci_transport_recv_listen(struct sock *sk,
 954                                       struct vmci_transport_packet *pkt)
 955 {
 956         struct sock *pending;
 957         struct vsock_sock *vpending;
 958         int err;
 959         u64 qp_size;
 960         bool old_request = false;
 961         bool old_pkt_proto = false;
 962
 963         err = 0;
 964
 965         /* Because we are in the listen state, we could be receiving a packet
 966          * for ourself or any previous connection requests that we received.
 967          * If it's the latter, we try to find a socket in our list of pending
 968          * connections and, if we do, call the appropriate handler for the
 969          * state that that socket is in.  Otherwise we try to service the
 970          * connection request.
 971          */
 972         pending = vmci_transport_get_pending(sk, pkt);
 973         if (pending) {
 974                 lock_sock(pending);
 975
 976                 /* The local context ID may be out of date. */
 977                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 978
 979                 switch (pending->sk_state) {
 980                 case TCP_SYN_SENT:
 981                         err = vmci_transport_recv_connecting_server(sk,
 982                                                                     pending,
 983                                                                     pkt);
 984                         break;
 985                 default:
 986                         vmci_transport_send_reset(pending, pkt);
 987                         err = -EINVAL;
 988                 }
 989
 990                 if (err < 0)
 991                         vsock_remove_pending(sk, pending);
 992
 993                 release_sock(pending);
 994                 vmci_transport_release_pending(pending);
 995
 996                 return err;
 997         }
 998
 999         /* The listen state only accepts connection requests.  Reply with a
1000          * reset unless we received a reset.
1001          */
1002
1003         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
1004               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
1005                 vmci_transport_reply_reset(pkt);
1006                 return -EINVAL;
1007         }
1008
1009         if (pkt->u.size == 0) {
1010                 vmci_transport_reply_reset(pkt);
1011                 return -EINVAL;
1012         }
1013
1014         /* If this socket can't accommodate this connection request, we send a
1015          * reset.  Otherwise we create and initialize a child socket and reply
1016          * with a connection negotiation.
1017          */
1018         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1019                 vmci_transport_reply_reset(pkt);
1020                 return -ECONNREFUSED;
1021         }
1022
1023         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1024                                  sk->sk_type, 0);
1025         if (!pending) {
1026                 vmci_transport_send_reset(sk, pkt);
1027                 return -ENOMEM;
1028         }
1029
1030         vpending = vsock_sk(pending);
1031
1032         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1033                         pkt->dst_port);
1034         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1035                         pkt->src_port);
1036
1037         /* If the proposed size fits within our min/max, accept it. Otherwise
1038          * propose our own size.
1039          */
1040         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1041             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1042                 qp_size = pkt->u.size;
1043         } else {
1044                 qp_size = vmci_trans(vpending)->queue_pair_size;
1045         }
1046
1047         /* Figure out if we are using old or new requests based on the
1048          * overrides pkt types sent by our peer.
1049          */
1050         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1051                 old_request = old_pkt_proto;
1052         } else {
1053                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1054                         old_request = true;
1055                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1056                         old_request = false;
1057
1058         }
1059
1060         if (old_request) {
1061                 /* Handle a REQUEST (or override) */
1062                 u16 version = VSOCK_PROTO_INVALID;
1063                 if (vmci_transport_proto_to_notify_struct(
1064                         pending, &version, true))
1065                         err = vmci_transport_send_negotiate(pending, qp_size);
1066                 else
1067                         err = -EINVAL;
1068
1069         } else {
1070                 /* Handle a REQUEST2 (or override) */
1071                 int proto_int = pkt->proto;
1072                 int pos;
1073                 u16 active_proto_version = 0;
1074
1075                 /* The list of possible protocols is the intersection of all
1076                  * protocols the client supports ... plus all the protocols we
1077                  * support.
1078                  */
1079                 proto_int &= vmci_transport_new_proto_supported_versions();
1080
1081                 /* We choose the highest possible protocol version and use that
1082                  * one.
1083                  */
1084                 pos = fls(proto_int);
1085                 if (pos) {
1086                         active_proto_version = (1 << (pos - 1));
1087                         if (vmci_transport_proto_to_notify_struct(
1088                                 pending, &active_proto_version, false))
1089                                 err = vmci_transport_send_negotiate2(pending,
1090                                                         qp_size,
1091                                                         active_proto_version);
1092                         else
1093                                 err = -EINVAL;
1094
1095                 } else {
1096                         err = -EINVAL;
1097                 }
1098         }
1099
1100         if (err < 0) {
1101                 vmci_transport_send_reset(sk, pkt);
1102                 sock_put(pending);
1103                 err = vmci_transport_error_to_vsock_error(err);
1104                 goto out;
1105         }
1106
1107         vsock_add_pending(sk, pending);
1108         sk->sk_ack_backlog++;
1109
1110         pending->sk_state = TCP_SYN_SENT;
1111         vmci_trans(vpending)->produce_size =
1112                 vmci_trans(vpending)->consume_size = qp_size;
1113         vmci_trans(vpending)->queue_pair_size = qp_size;
1114
1115         vmci_trans(vpending)->notify_ops->process_request(pending);
1116
1117         /* We might never receive another message for this socket and it's not
1118          * connected to any process, so we have to ensure it gets cleaned up
1119          * ourself.  Our delayed work function will take care of that.  Note
1120          * that we do not ever cancel this function since we have few
1121          * guarantees about its state when calling cancel_delayed_work().
1122          * Instead we hold a reference on the socket for that function and make
1123          * it capable of handling cases where it needs to do nothing but
1124          * release that reference.
1125          */
1126         vpending->listener = sk;
1127         sock_hold(sk);
1128         sock_hold(pending);
1129         schedule_delayed_work(&vpending->pending_work, HZ);
1130
1131 out:
1132         return err;
1133 }
1134
1135 static int
1136 vmci_transport_recv_connecting_server(struct sock *listener,
1137                                       struct sock *pending,
1138                                       struct vmci_transport_packet *pkt)
1139 {
1140         struct vsock_sock *vpending;
1141         struct vmci_handle handle;
1142         struct vmci_qp *qpair;
1143         bool is_local;
1144         u32 flags;
1145         u32 detach_sub_id;
1146         int err;
1147         int skerr;
1148
1149         vpending = vsock_sk(pending);
1150         detach_sub_id = VMCI_INVALID_ID;
1151
1152         switch (pkt->type) {
1153         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1154                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1155                         vmci_transport_send_reset(pending, pkt);
1156                         skerr = EPROTO;
1157                         err = -EINVAL;
1158                         goto destroy;
1159                 }
1160                 break;
1161         default:
1162                 /* Close and cleanup the connection. */
1163                 vmci_transport_send_reset(pending, pkt);
1164                 skerr = EPROTO;
1165                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1166                 goto destroy;
1167         }
1168
1169         /* In order to complete the connection we need to attach to the offered
1170          * queue pair and send an attach notification.  We also subscribe to the
1171          * detach event so we know when our peer goes away, and we do that
1172          * before attaching so we don't miss an event.  If all this succeeds,
1173          * we update our state and wakeup anything waiting in accept() for a
1174          * connection.
1175          */
1176
1177         /* We don't care about attach since we ensure the other side has
1178          * attached by specifying the ATTACH_ONLY flag below.
1179          */
1180         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1181                                    vmci_transport_peer_detach_cb,
1182                                    vmci_trans(vpending), &detach_sub_id);
1183         if (err < VMCI_SUCCESS) {
1184                 vmci_transport_send_reset(pending, pkt);
1185                 err = vmci_transport_error_to_vsock_error(err);
1186                 skerr = -err;
1187                 goto destroy;
1188         }
1189
1190         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1191
1192         /* Now attach to the queue pair the client created. */
1193         handle = pkt->u.handle;
1194
1195         /* vpending->local_addr always has a context id so we do not need to
1196          * worry about VMADDR_CID_ANY in this case.
1197          */
1198         is_local =
1199             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1200         flags = VMCI_QPFLAG_ATTACH_ONLY;
1201         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1202
1203         err = vmci_transport_queue_pair_alloc(
1204                                         &qpair,
1205                                         &handle,
1206                                         vmci_trans(vpending)->produce_size,
1207                                         vmci_trans(vpending)->consume_size,
1208                                         pkt->dg.src.context,
1209                                         flags,
1210                                         vmci_transport_is_trusted(
1211                                                 vpending,
1212                                                 vpending->remote_addr.svm_cid));
1213         if (err < 0) {
1214                 vmci_transport_send_reset(pending, pkt);
1215                 skerr = -err;
1216                 goto destroy;
1217         }
1218
1219         vmci_trans(vpending)->qp_handle = handle;
1220         vmci_trans(vpending)->qpair = qpair;
1221
1222         /* When we send the attach message, we must be ready to handle incoming
1223          * control messages on the newly connected socket. So we move the
1224          * pending socket to the connected state before sending the attach
1225          * message. Otherwise, an incoming packet triggered by the attach being
1226          * received by the peer may be processed concurrently with what happens
1227          * below after sending the attach message, and that incoming packet
1228          * will find the listening socket instead of the (currently) pending
1229          * socket. Note that enqueueing the socket increments the reference
1230          * count, so even if a reset comes before the connection is accepted,
1231          * the socket will be valid until it is removed from the queue.
1232          *
1233          * If we fail sending the attach below, we remove the socket from the
1234          * connected list and move the socket to TCP_CLOSE before
1235          * releasing the lock, so a pending slow path processing of an incoming
1236          * packet will not see the socket in the connected state in that case.
1237          */
1238         pending->sk_state = TCP_ESTABLISHED;
1239
1240         vsock_insert_connected(vpending);
1241
1242         /* Notify our peer of our attach. */
1243         err = vmci_transport_send_attach(pending, handle);
1244         if (err < 0) {
1245                 vsock_remove_connected(vpending);
1246                 pr_err("Could not send attach\n");
1247                 vmci_transport_send_reset(pending, pkt);
1248                 err = vmci_transport_error_to_vsock_error(err);
1249                 skerr = -err;
1250                 goto destroy;
1251         }
1252
1253         /* We have a connection. Move the now connected socket from the
1254          * listener's pending list to the accept queue so callers of accept()
1255          * can find it.
1256          */
1257         vsock_remove_pending(listener, pending);
1258         vsock_enqueue_accept(listener, pending);
1259
1260         /* Callers of accept() will be be waiting on the listening socket, not
1261          * the pending socket.
1262          */
1263         listener->sk_data_ready(listener);
1264
1265         return 0;
1266
1267 destroy:
1268         pending->sk_err = skerr;
1269         pending->sk_state = TCP_CLOSE;
1270         /* As long as we drop our reference, all necessary cleanup will handle
1271          * when the cleanup function drops its reference and our destruct
1272          * implementation is called.  Note that since the listen handler will
1273          * remove pending from the pending list upon our failure, the cleanup
1274          * function won't drop the additional reference, which is why we do it
1275          * here.
1276          */
1277         sock_put(pending);
1278
1279         return err;
1280 }
1281
1282 static int
1283 vmci_transport_recv_connecting_client(struct sock *sk,
1284                                       struct vmci_transport_packet *pkt)
1285 {
1286         struct vsock_sock *vsk;
1287         int err;
1288         int skerr;
1289
1290         vsk = vsock_sk(sk);
1291
1292         switch (pkt->type) {
1293         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1294                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1295                     !vmci_handle_is_equal(pkt->u.handle,
1296                                           vmci_trans(vsk)->qp_handle)) {
1297                         skerr = EPROTO;
1298                         err = -EINVAL;
1299                         goto destroy;
1300                 }
1301
1302                 /* Signify the socket is connected and wakeup the waiter in
1303                  * connect(). Also place the socket in the connected table for
1304                  * accounting (it can already be found since it's in the bound
1305                  * table).
1306                  */
1307                 sk->sk_state = TCP_ESTABLISHED;
1308                 sk->sk_socket->state = SS_CONNECTED;
1309                 vsock_insert_connected(vsk);
1310                 sk->sk_state_change(sk);
1311
1312                 break;
1313         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1314         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1315                 if (pkt->u.size == 0
1316                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1317                     || pkt->src_port != vsk->remote_addr.svm_port
1318                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1319                     || vmci_trans(vsk)->qpair
1320                     || vmci_trans(vsk)->produce_size != 0
1321                     || vmci_trans(vsk)->consume_size != 0
1322                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1323                         skerr = EPROTO;
1324                         err = -EINVAL;
1325
1326                         goto destroy;
1327                 }
1328
1329                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1330                 if (err) {
1331                         skerr = -err;
1332                         goto destroy;
1333                 }
1334
1335                 break;
1336         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1337                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1338                 if (err) {
1339                         skerr = -err;
1340                         goto destroy;
1341                 }
1342
1343                 break;
1344         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1345                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1346                  * continue processing here after they sent an INVALID packet.
1347                  * This meant that we got a RST after the INVALID. We ignore a
1348                  * RST after an INVALID. The common code doesn't send the RST
1349                  * ... so we can hang if an old version of the common code
1350                  * fails between getting a REQUEST and sending an OFFER back.
1351                  * Not much we can do about it... except hope that it doesn't
1352                  * happen.
1353                  */
1354                 if (vsk->ignore_connecting_rst) {
1355                         vsk->ignore_connecting_rst = false;
1356                 } else {
1357                         skerr = ECONNRESET;
1358                         err = 0;
1359                         goto destroy;
1360                 }
1361
1362                 break;
1363         default:
1364                 /* Close and cleanup the connection. */
1365                 skerr = EPROTO;
1366                 err = -EINVAL;
1367                 goto destroy;
1368         }
1369
1370         return 0;
1371
1372 destroy:
1373         vmci_transport_send_reset(sk, pkt);
1374
1375         sk->sk_state = TCP_CLOSE;
1376         sk->sk_err = skerr;
1377         sk->sk_error_report(sk);
1378         return err;
1379 }
1380
1381 static int vmci_transport_recv_connecting_client_negotiate(
1382                                         struct sock *sk,
1383                                         struct vmci_transport_packet *pkt)
1384 {
1385         int err;
1386         struct vsock_sock *vsk;
1387         struct vmci_handle handle;
1388         struct vmci_qp *qpair;
1389         u32 detach_sub_id;
1390         bool is_local;
1391         u32 flags;
1392         bool old_proto = true;
1393         bool old_pkt_proto;
1394         u16 version;
1395
1396         vsk = vsock_sk(sk);
1397         handle = VMCI_INVALID_HANDLE;
1398         detach_sub_id = VMCI_INVALID_ID;
1399
1400         /* If we have gotten here then we should be past the point where old
1401          * linux vsock could have sent the bogus rst.
1402          */
1403         vsk->sent_request = false;
1404         vsk->ignore_connecting_rst = false;
1405
1406         /* Verify that we're OK with the proposed queue pair size */
1407         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1408             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1409                 err = -EINVAL;
1410                 goto destroy;
1411         }
1412
1413         /* At this point we know the CID the peer is using to talk to us. */
1414
1415         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1416                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1417
1418         /* Setup the notify ops to be the highest supported version that both
1419          * the server and the client support.
1420          */
1421
1422         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1423                 old_proto = old_pkt_proto;
1424         } else {
1425                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1426                         old_proto = true;
1427                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1428                         old_proto = false;
1429
1430         }
1431
1432         if (old_proto)
1433                 version = VSOCK_PROTO_INVALID;
1434         else
1435                 version = pkt->proto;
1436
1437         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1438                 err = -EINVAL;
1439                 goto destroy;
1440         }
1441
1442         /* Subscribe to detach events first.
1443          *
1444          * XXX We attach once for each queue pair created for now so it is easy
1445          * to find the socket (it's provided), but later we should only
1446          * subscribe once and add a way to lookup sockets by queue pair handle.
1447          */
1448         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1449                                    vmci_transport_peer_detach_cb,
1450                                    vmci_trans(vsk), &detach_sub_id);
1451         if (err < VMCI_SUCCESS) {
1452                 err = vmci_transport_error_to_vsock_error(err);
1453                 goto destroy;
1454         }
1455
1456         /* Make VMCI select the handle for us. */
1457         handle = VMCI_INVALID_HANDLE;
1458         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1459         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1460
1461         err = vmci_transport_queue_pair_alloc(&qpair,
1462                                               &handle,
1463                                               pkt->u.size,
1464                                               pkt->u.size,
1465                                               vsk->remote_addr.svm_cid,
1466                                               flags,
1467                                               vmci_transport_is_trusted(
1468                                                   vsk,
1469                                                   vsk->
1470                                                   remote_addr.svm_cid));
1471         if (err < 0)
1472                 goto destroy;
1473
1474         err = vmci_transport_send_qp_offer(sk, handle);
1475         if (err < 0) {
1476                 err = vmci_transport_error_to_vsock_error(err);
1477                 goto destroy;
1478         }
1479
1480         vmci_trans(vsk)->qp_handle = handle;
1481         vmci_trans(vsk)->qpair = qpair;
1482
1483         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1484                 pkt->u.size;
1485
1486         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1487
1488         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1489
1490         return 0;
1491
1492 destroy:
1493         if (detach_sub_id != VMCI_INVALID_ID)
1494                 vmci_event_unsubscribe(detach_sub_id);
1495
1496         if (!vmci_handle_is_invalid(handle))
1497                 vmci_qpair_detach(&qpair);
1498
1499         return err;
1500 }
1501
1502 static int
1503 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1504                                               struct vmci_transport_packet *pkt)
1505 {
1506         int err = 0;
1507         struct vsock_sock *vsk = vsock_sk(sk);
1508
1509         if (vsk->sent_request) {
1510                 vsk->sent_request = false;
1511                 vsk->ignore_connecting_rst = true;
1512
1513                 err = vmci_transport_send_conn_request(
1514                         sk, vmci_trans(vsk)->queue_pair_size);
1515                 if (err < 0)
1516                         err = vmci_transport_error_to_vsock_error(err);
1517                 else
1518                         err = 0;
1519
1520         }
1521
1522         return err;
1523 }
1524
1525 static int vmci_transport_recv_connected(struct sock *sk,
1526                                          struct vmci_transport_packet *pkt)
1527 {
1528         struct vsock_sock *vsk;
1529         bool pkt_processed = false;
1530
1531         /* In cases where we are closing the connection, it's sufficient to
1532          * mark the state change (and maybe error) and wake up any waiting
1533          * threads. Since this is a connected socket, it's owned by a user
1534          * process and will be cleaned up when the failure is passed back on
1535          * the current or next system call.  Our system call implementations
1536          * must therefore check for error and state changes on entry and when
1537          * being awoken.
1538          */
1539         switch (pkt->type) {
1540         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1541                 if (pkt->u.mode) {
1542                         vsk = vsock_sk(sk);
1543
1544                         vsk->peer_shutdown |= pkt->u.mode;
1545                         sk->sk_state_change(sk);
1546                 }
1547                 break;
1548
1549         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1550                 vsk = vsock_sk(sk);
1551                 /* It is possible that we sent our peer a message (e.g a
1552                  * WAITING_READ) right before we got notified that the peer had
1553                  * detached. If that happens then we can get a RST pkt back
1554                  * from our peer even though there is data available for us to
1555                  * read. In that case, don't shutdown the socket completely but
1556                  * instead allow the local client to finish reading data off
1557                  * the queuepair. Always treat a RST pkt in connected mode like
1558                  * a clean shutdown.
1559                  */
1560                 sock_set_flag(sk, SOCK_DONE);
1561                 vsk->peer_shutdown = SHUTDOWN_MASK;
1562                 if (vsock_stream_has_data(vsk) <= 0)
1563                         sk->sk_state = TCP_CLOSING;
1564
1565                 sk->sk_state_change(sk);
1566                 break;
1567
1568         default:
1569                 vsk = vsock_sk(sk);
1570                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1571                                 sk, pkt, false, NULL, NULL,
1572                                 &pkt_processed);
1573                 if (!pkt_processed)
1574                         return -EINVAL;
1575
1576                 break;
1577         }
1578
1579         return 0;
1580 }
1581
1582 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1583                                       struct vsock_sock *psk)
1584 {
1585         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1586         if (!vsk->trans)
1587                 return -ENOMEM;
1588
1589         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1590         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1591         vmci_trans(vsk)->qpair = NULL;
1592         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1593         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1594         vmci_trans(vsk)->notify_ops = NULL;
1595         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1596         vmci_trans(vsk)->sk = &vsk->sk;
1597         spin_lock_init(&vmci_trans(vsk)->lock);
1598         if (psk) {
1599                 vmci_trans(vsk)->queue_pair_size =
1600                         vmci_trans(psk)->queue_pair_size;
1601                 vmci_trans(vsk)->queue_pair_min_size =
1602                         vmci_trans(psk)->queue_pair_min_size;
1603                 vmci_trans(vsk)->queue_pair_max_size =
1604                         vmci_trans(psk)->queue_pair_max_size;
1605         } else {
1606                 vmci_trans(vsk)->queue_pair_size =
1607                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1608                 vmci_trans(vsk)->queue_pair_min_size =
1609                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1610                 vmci_trans(vsk)->queue_pair_max_size =
1611                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1612         }
1613
1614         return 0;
1615 }
1616
1617 static void vmci_transport_free_resources(struct list_head *transport_list)
1618 {
1619         while (!list_empty(transport_list)) {
1620                 struct vmci_transport *transport =
1621                     list_first_entry(transport_list, struct vmci_transport,
1622                                      elem);
1623                 list_del(&transport->elem);
1624
1625                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1626                         vmci_event_unsubscribe(transport->detach_sub_id);
1627                         transport->detach_sub_id = VMCI_INVALID_ID;
1628                 }
1629
1630                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1631                         vmci_qpair_detach(&transport->qpair);
1632                         transport->qp_handle = VMCI_INVALID_HANDLE;
1633                         transport->produce_size = 0;
1634                         transport->consume_size = 0;
1635                 }
1636
1637                 kfree(transport);
1638         }
1639 }
1640
1641 static void vmci_transport_cleanup(struct work_struct *work)
1642 {
1643         LIST_HEAD(pending);
1644
1645         spin_lock_bh(&vmci_transport_cleanup_lock);
1646         list_replace_init(&vmci_transport_cleanup_list, &pending);
1647         spin_unlock_bh(&vmci_transport_cleanup_lock);
1648         vmci_transport_free_resources(&pending);
1649 }
1650
1651 static void vmci_transport_destruct(struct vsock_sock *vsk)
1652 {
1653         /* transport can be NULL if we hit a failure at init() time */
1654         if (!vmci_trans(vsk))
1655                 return;
1656
1657         /* Ensure that the detach callback doesn't use the sk/vsk
1658          * we are about to destruct.
1659          */
1660         spin_lock_bh(&vmci_trans(vsk)->lock);
1661         vmci_trans(vsk)->sk = NULL;
1662         spin_unlock_bh(&vmci_trans(vsk)->lock);
1663
1664         if (vmci_trans(vsk)->notify_ops)
1665                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1666
1667         spin_lock_bh(&vmci_transport_cleanup_lock);
1668         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1669         spin_unlock_bh(&vmci_transport_cleanup_lock);
1670         schedule_work(&vmci_transport_cleanup_work);
1671
1672         vsk->trans = NULL;
1673 }
1674
1675 static void vmci_transport_release(struct vsock_sock *vsk)
1676 {
1677         vsock_remove_sock(vsk);
1678
1679         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1680                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1681                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1682         }
1683 }
1684
1685 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1686                                      struct sockaddr_vm *addr)
1687 {
1688         u32 port;
1689         u32 flags;
1690         int err;
1691
1692         /* VMCI will select a resource ID for us if we provide
1693          * VMCI_INVALID_ID.
1694          */
1695         port = addr->svm_port == VMADDR_PORT_ANY ?
1696                         VMCI_INVALID_ID : addr->svm_port;
1697
1698         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1699                 return -EACCES;
1700
1701         flags = addr->svm_cid == VMADDR_CID_ANY ?
1702                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1703
1704         err = vmci_transport_datagram_create_hnd(port, flags,
1705                                                  vmci_transport_recv_dgram_cb,
1706                                                  &vsk->sk,
1707                                                  &vmci_trans(vsk)->dg_handle);
1708         if (err < VMCI_SUCCESS)
1709                 return vmci_transport_error_to_vsock_error(err);
1710         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1711                         vmci_trans(vsk)->dg_handle.resource);
1712
1713         return 0;
1714 }
1715
1716 static int vmci_transport_dgram_enqueue(
1717         struct vsock_sock *vsk,
1718         struct sockaddr_vm *remote_addr,
1719         struct msghdr *msg,
1720         size_t len)
1721 {
1722         int err;
1723         struct vmci_datagram *dg;
1724
1725         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1726                 return -EMSGSIZE;
1727
1728         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1729                 return -EPERM;
1730
1731         /* Allocate a buffer for the user's message and our packet header. */
1732         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1733         if (!dg)
1734                 return -ENOMEM;
1735
1736         err = memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1737         if (err) {
1738                 kfree(dg);
1739                 return err;
1740         }
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 (!skb)
1775                 return err;
1776
1777         dg = (struct vmci_datagram *)skb->data;
1778         if (!dg)
1779                 /* err is 0, meaning we read zero bytes. */
1780                 goto out;
1781
1782         payload_len = dg->payload_size;
1783         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1784         if (payload_len != skb->len - sizeof(*dg)) {
1785                 err = -EINVAL;
1786                 goto out;
1787         }
1788
1789         if (payload_len > len) {
1790                 payload_len = len;
1791                 msg->msg_flags |= MSG_TRUNC;
1792         }
1793
1794         /* Place the datagram payload in the user's iovec. */
1795         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1796         if (err)
1797                 goto out;
1798
1799         if (msg->msg_name) {
1800                 /* Provide the address of the sender. */
1801                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1802                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1803                 msg->msg_namelen = sizeof(*vm_addr);
1804         }
1805         err = payload_len;
1806
1807 out:
1808         skb_free_datagram(&vsk->sk, skb);
1809         return err;
1810 }
1811
1812 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1813 {
1814         if (cid == VMADDR_CID_HYPERVISOR) {
1815                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1816                  * state and are allowed.
1817                  */
1818                 return port == VMCI_UNITY_PBRPC_REGISTER;
1819         }
1820
1821         return true;
1822 }
1823
1824 static int vmci_transport_connect(struct vsock_sock *vsk)
1825 {
1826         int err;
1827         bool old_pkt_proto = false;
1828         struct sock *sk = &vsk->sk;
1829
1830         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1831                 old_pkt_proto) {
1832                 err = vmci_transport_send_conn_request(
1833                         sk, vmci_trans(vsk)->queue_pair_size);
1834                 if (err < 0) {
1835                         sk->sk_state = TCP_CLOSE;
1836                         return err;
1837                 }
1838         } else {
1839                 int supported_proto_versions =
1840                         vmci_transport_new_proto_supported_versions();
1841                 err = vmci_transport_send_conn_request2(
1842                                 sk, vmci_trans(vsk)->queue_pair_size,
1843                                 supported_proto_versions);
1844                 if (err < 0) {
1845                         sk->sk_state = TCP_CLOSE;
1846                         return err;
1847                 }
1848
1849                 vsk->sent_request = true;
1850         }
1851
1852         return err;
1853 }
1854
1855 static ssize_t vmci_transport_stream_dequeue(
1856         struct vsock_sock *vsk,
1857         struct msghdr *msg,
1858         size_t len,
1859         int flags)
1860 {
1861         if (flags & MSG_PEEK)
1862                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1863         else
1864                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1865 }
1866
1867 static ssize_t vmci_transport_stream_enqueue(
1868         struct vsock_sock *vsk,
1869         struct msghdr *msg,
1870         size_t len)
1871 {
1872         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1873 }
1874
1875 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1876 {
1877         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1878 }
1879
1880 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1881 {
1882         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1883 }
1884
1885 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1886 {
1887         return vmci_trans(vsk)->consume_size;
1888 }
1889
1890 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1891 {
1892         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1893 }
1894
1895 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1896 {
1897         return vmci_trans(vsk)->queue_pair_size;
1898 }
1899
1900 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1901 {
1902         return vmci_trans(vsk)->queue_pair_min_size;
1903 }
1904
1905 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1906 {
1907         return vmci_trans(vsk)->queue_pair_max_size;
1908 }
1909
1910 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1911 {
1912         if (val < vmci_trans(vsk)->queue_pair_min_size)
1913                 vmci_trans(vsk)->queue_pair_min_size = val;
1914         if (val > vmci_trans(vsk)->queue_pair_max_size)
1915                 vmci_trans(vsk)->queue_pair_max_size = val;
1916         vmci_trans(vsk)->queue_pair_size = val;
1917 }
1918
1919 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1920                                                u64 val)
1921 {
1922         if (val > vmci_trans(vsk)->queue_pair_size)
1923                 vmci_trans(vsk)->queue_pair_size = val;
1924         vmci_trans(vsk)->queue_pair_min_size = val;
1925 }
1926
1927 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1928                                                u64 val)
1929 {
1930         if (val < vmci_trans(vsk)->queue_pair_size)
1931                 vmci_trans(vsk)->queue_pair_size = val;
1932         vmci_trans(vsk)->queue_pair_max_size = val;
1933 }
1934
1935 static int vmci_transport_notify_poll_in(
1936         struct vsock_sock *vsk,
1937         size_t target,
1938         bool *data_ready_now)
1939 {
1940         return vmci_trans(vsk)->notify_ops->poll_in(
1941                         &vsk->sk, target, data_ready_now);
1942 }
1943
1944 static int vmci_transport_notify_poll_out(
1945         struct vsock_sock *vsk,
1946         size_t target,
1947         bool *space_available_now)
1948 {
1949         return vmci_trans(vsk)->notify_ops->poll_out(
1950                         &vsk->sk, target, space_available_now);
1951 }
1952
1953 static int vmci_transport_notify_recv_init(
1954         struct vsock_sock *vsk,
1955         size_t target,
1956         struct vsock_transport_recv_notify_data *data)
1957 {
1958         return vmci_trans(vsk)->notify_ops->recv_init(
1959                         &vsk->sk, target,
1960                         (struct vmci_transport_recv_notify_data *)data);
1961 }
1962
1963 static int vmci_transport_notify_recv_pre_block(
1964         struct vsock_sock *vsk,
1965         size_t target,
1966         struct vsock_transport_recv_notify_data *data)
1967 {
1968         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1969                         &vsk->sk, target,
1970                         (struct vmci_transport_recv_notify_data *)data);
1971 }
1972
1973 static int vmci_transport_notify_recv_pre_dequeue(
1974         struct vsock_sock *vsk,
1975         size_t target,
1976         struct vsock_transport_recv_notify_data *data)
1977 {
1978         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1979                         &vsk->sk, target,
1980                         (struct vmci_transport_recv_notify_data *)data);
1981 }
1982
1983 static int vmci_transport_notify_recv_post_dequeue(
1984         struct vsock_sock *vsk,
1985         size_t target,
1986         ssize_t copied,
1987         bool data_read,
1988         struct vsock_transport_recv_notify_data *data)
1989 {
1990         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1991                         &vsk->sk, target, copied, data_read,
1992                         (struct vmci_transport_recv_notify_data *)data);
1993 }
1994
1995 static int vmci_transport_notify_send_init(
1996         struct vsock_sock *vsk,
1997         struct vsock_transport_send_notify_data *data)
1998 {
1999         return vmci_trans(vsk)->notify_ops->send_init(
2000                         &vsk->sk,
2001                         (struct vmci_transport_send_notify_data *)data);
2002 }
2003
2004 static int vmci_transport_notify_send_pre_block(
2005         struct vsock_sock *vsk,
2006         struct vsock_transport_send_notify_data *data)
2007 {
2008         return vmci_trans(vsk)->notify_ops->send_pre_block(
2009                         &vsk->sk,
2010                         (struct vmci_transport_send_notify_data *)data);
2011 }
2012
2013 static int vmci_transport_notify_send_pre_enqueue(
2014         struct vsock_sock *vsk,
2015         struct vsock_transport_send_notify_data *data)
2016 {
2017         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2018                         &vsk->sk,
2019                         (struct vmci_transport_send_notify_data *)data);
2020 }
2021
2022 static int vmci_transport_notify_send_post_enqueue(
2023         struct vsock_sock *vsk,
2024         ssize_t written,
2025         struct vsock_transport_send_notify_data *data)
2026 {
2027         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2028                         &vsk->sk, written,
2029                         (struct vmci_transport_send_notify_data *)data);
2030 }
2031
2032 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2033 {
2034         if (PROTOCOL_OVERRIDE != -1) {
2035                 if (PROTOCOL_OVERRIDE == 0)
2036                         *old_pkt_proto = true;
2037                 else
2038                         *old_pkt_proto = false;
2039
2040                 pr_info("Proto override in use\n");
2041                 return true;
2042         }
2043
2044         return false;
2045 }
2046
2047 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2048                                                   u16 *proto,
2049                                                   bool old_pkt_proto)
2050 {
2051         struct vsock_sock *vsk = vsock_sk(sk);
2052
2053         if (old_pkt_proto) {
2054                 if (*proto != VSOCK_PROTO_INVALID) {
2055                         pr_err("Can't set both an old and new protocol\n");
2056                         return false;
2057                 }
2058                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2059                 goto exit;
2060         }
2061
2062         switch (*proto) {
2063         case VSOCK_PROTO_PKT_ON_NOTIFY:
2064                 vmci_trans(vsk)->notify_ops =
2065                         &vmci_transport_notify_pkt_q_state_ops;
2066                 break;
2067         default:
2068                 pr_err("Unknown notify protocol version\n");
2069                 return false;
2070         }
2071
2072 exit:
2073         vmci_trans(vsk)->notify_ops->socket_init(sk);
2074         return true;
2075 }
2076
2077 static u16 vmci_transport_new_proto_supported_versions(void)
2078 {
2079         if (PROTOCOL_OVERRIDE != -1)
2080                 return PROTOCOL_OVERRIDE;
2081
2082         return VSOCK_PROTO_ALL_SUPPORTED;
2083 }
2084
2085 static u32 vmci_transport_get_local_cid(void)
2086 {
2087         return vmci_get_context_id();
2088 }
2089
2090 static const struct vsock_transport vmci_transport = {
2091         .init = vmci_transport_socket_init,
2092         .destruct = vmci_transport_destruct,
2093         .release = vmci_transport_release,
2094         .connect = vmci_transport_connect,
2095         .dgram_bind = vmci_transport_dgram_bind,
2096         .dgram_dequeue = vmci_transport_dgram_dequeue,
2097         .dgram_enqueue = vmci_transport_dgram_enqueue,
2098         .dgram_allow = vmci_transport_dgram_allow,
2099         .stream_dequeue = vmci_transport_stream_dequeue,
2100         .stream_enqueue = vmci_transport_stream_enqueue,
2101         .stream_has_data = vmci_transport_stream_has_data,
2102         .stream_has_space = vmci_transport_stream_has_space,
2103         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2104         .stream_is_active = vmci_transport_stream_is_active,
2105         .stream_allow = vmci_transport_stream_allow,
2106         .notify_poll_in = vmci_transport_notify_poll_in,
2107         .notify_poll_out = vmci_transport_notify_poll_out,
2108         .notify_recv_init = vmci_transport_notify_recv_init,
2109         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2110         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2111         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2112         .notify_send_init = vmci_transport_notify_send_init,
2113         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2114         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2115         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2116         .shutdown = vmci_transport_shutdown,
2117         .set_buffer_size = vmci_transport_set_buffer_size,
2118         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2119         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2120         .get_buffer_size = vmci_transport_get_buffer_size,
2121         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2122         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2123         .get_local_cid = vmci_transport_get_local_cid,
2124 };
2125
2126 static int __init vmci_transport_init(void)
2127 {
2128         int err;
2129
2130         /* Create the datagram handle that we will use to send and receive all
2131          * VSocket control messages for this context.
2132          */
2133         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2134                                                  VMCI_FLAG_ANYCID_DG_HND,
2135                                                  vmci_transport_recv_stream_cb,
2136                                                  NULL,
2137                                                  &vmci_transport_stream_handle);
2138         if (err < VMCI_SUCCESS) {
2139                 pr_err("Unable to create datagram handle. (%d)\n", err);
2140                 return vmci_transport_error_to_vsock_error(err);
2141         }
2142
2143         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2144                                    vmci_transport_qp_resumed_cb,
2145                                    NULL, &vmci_transport_qp_resumed_sub_id);
2146         if (err < VMCI_SUCCESS) {
2147                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2148                 err = vmci_transport_error_to_vsock_error(err);
2149                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2150                 goto err_destroy_stream_handle;
2151         }
2152
2153         err = vsock_core_init(&vmci_transport);
2154         if (err < 0)
2155                 goto err_unsubscribe;
2156
2157         return 0;
2158
2159 err_unsubscribe:
2160         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2161 err_destroy_stream_handle:
2162         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2163         return err;
2164 }
2165 module_init(vmci_transport_init);
2166
2167 static void __exit vmci_transport_exit(void)
2168 {
2169         cancel_work_sync(&vmci_transport_cleanup_work);
2170         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2171
2172         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2173                 if (vmci_datagram_destroy_handle(
2174                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2175                         pr_err("Couldn't destroy datagram handle\n");
2176                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2177         }
2178
2179         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2180                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2181                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2182         }
2183
2184         vsock_core_exit();
2185 }
2186 module_exit(vmci_transport_exit);
2187
2188 MODULE_AUTHOR("VMware, Inc.");
2189 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2190 MODULE_VERSION("1.0.5.0-k");
2191 MODULE_LICENSE("GPL v2");
2192 MODULE_ALIAS("vmware_vsock");
2193 MODULE_ALIAS_NETPROTO(PF_VSOCK);