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("Could not attach to queue pair with %d\n",
 588                        err);
 589                 err = vmci_transport_error_to_vsock_error(err);
 590         }
 591
 592         return err;
 593 }
 594
 595 static int
 596 vmci_transport_datagram_create_hnd(u32 resource_id,
 597                                    u32 flags,
 598                                    vmci_datagram_recv_cb recv_cb,
 599                                    void *client_data,
 600                                    struct vmci_handle *out_handle)
 601 {
 602         int err = 0;
 603
 604         /* Try to allocate our datagram handler as trusted. This will only work
 605          * if vsock is running in the host.
 606          */
 607
 608         err = vmci_datagram_create_handle_priv(resource_id, flags,
 609                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 610                                                recv_cb,
 611                                                client_data, out_handle);
 612
 613         if (err == VMCI_ERROR_NO_ACCESS)
 614                 err = vmci_datagram_create_handle(resource_id, flags,
 615                                                   recv_cb, client_data,
 616                                                   out_handle);
 617
 618         return err;
 619 }
 620
 621 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 622  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 623  * sleep it should defer that work to a work queue.
 624  */
 625
 626 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 627 {
 628         struct sock *sk;
 629         size_t size;
 630         struct sk_buff *skb;
 631         struct vsock_sock *vsk;
 632
 633         sk = (struct sock *)data;
 634
 635         /* This handler is privileged when this module is running on the host.
 636          * We will get datagrams from all endpoints (even VMs that are in a
 637          * restricted context). If we get one from a restricted context then
 638          * the destination socket must be trusted.
 639          *
 640          * NOTE: We access the socket struct without holding the lock here.
 641          * This is ok because the field we are interested is never modified
 642          * outside of the create and destruct socket functions.
 643          */
 644         vsk = vsock_sk(sk);
 645         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 646                 return VMCI_ERROR_NO_ACCESS;
 647
 648         size = VMCI_DG_SIZE(dg);
 649
 650         /* Attach the packet to the socket's receive queue as an sk_buff. */
 651         skb = alloc_skb(size, GFP_ATOMIC);
 652         if (!skb)
 653                 return VMCI_ERROR_NO_MEM;
 654
 655         /* sk_receive_skb() will do a sock_put(), so hold here. */
 656         sock_hold(sk);
 657         skb_put(skb, size);
 658         memcpy(skb->data, dg, size);
 659         sk_receive_skb(sk, skb, 0);
 660
 661         return VMCI_SUCCESS;
 662 }
 663
 664 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 665 {
 666         static const u32 non_socket_contexts[] = {
 667                 VMADDR_CID_RESERVED,
 668         };
 669         int i;
 670
 671         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 672
 673         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 674                 if (cid == non_socket_contexts[i])
 675                         return false;
 676         }
 677
 678         return true;
 679 }
 680
 681 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 682  * interrupt fires.  This is run in bottom-half context but it defers most of
 683  * its work to the packet handling work queue.
 684  */
 685
 686 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 687 {
 688         struct sock *sk;
 689         struct sockaddr_vm dst;
 690         struct sockaddr_vm src;
 691         struct vmci_transport_packet *pkt;
 692         struct vsock_sock *vsk;
 693         bool bh_process_pkt;
 694         int err;
 695
 696         sk = NULL;
 697         err = VMCI_SUCCESS;
 698         bh_process_pkt = false;
 699
 700         /* Ignore incoming packets from contexts without sockets, or resources
 701          * that aren't vsock implementations.
 702          */
 703
 704         if (!vmci_transport_stream_allow(dg->src.context, -1)
 705             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 706                 return VMCI_ERROR_NO_ACCESS;
 707
 708         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 709                 /* Drop datagrams that do not contain full VSock packets. */
 710                 return VMCI_ERROR_INVALID_ARGS;
 711
 712         pkt = (struct vmci_transport_packet *)dg;
 713
 714         /* Find the socket that should handle this packet.  First we look for a
 715          * connected socket and if there is none we look for a socket bound to
 716          * the destintation address.
 717          */
 718         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 719         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 720
 721         sk = vsock_find_connected_socket(&src, &dst);
 722         if (!sk) {
 723                 sk = vsock_find_bound_socket(&dst);
 724                 if (!sk) {
 725                         /* We could not find a socket for this specified
 726                          * address.  If this packet is a RST, we just drop it.
 727                          * If it is another packet, we send a RST.  Note that
 728                          * we do not send a RST reply to RSTs so that we do not
 729                          * continually send RSTs between two endpoints.
 730                          *
 731                          * Note that since this is a reply, dst is src and src
 732                          * is dst.
 733                          */
 734                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 735                                 pr_err("unable to send reset\n");
 736
 737                         err = VMCI_ERROR_NOT_FOUND;
 738                         goto out;
 739                 }
 740         }
 741
 742         /* If the received packet type is beyond all types known to this
 743          * implementation, reply with an invalid message.  Hopefully this will
 744          * help when implementing backwards compatibility in the future.
 745          */
 746         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 747                 vmci_transport_send_invalid_bh(&dst, &src);
 748                 err = VMCI_ERROR_INVALID_ARGS;
 749                 goto out;
 750         }
 751
 752         /* This handler is privileged when this module is running on the host.
 753          * We will get datagram connect requests from all endpoints (even VMs
 754          * that are in a restricted context). If we get one from a restricted
 755          * context then the destination socket must be trusted.
 756          *
 757          * NOTE: We access the socket struct without holding the lock here.
 758          * This is ok because the field we are interested is never modified
 759          * outside of the create and destruct socket functions.
 760          */
 761         vsk = vsock_sk(sk);
 762         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 763                 err = VMCI_ERROR_NO_ACCESS;
 764                 goto out;
 765         }
 766
 767         /* We do most everything in a work queue, but let's fast path the
 768          * notification of reads and writes to help data transfer performance.
 769          * We can only do this if there is no process context code executing
 770          * for this socket since that may change the state.
 771          */
 772         bh_lock_sock(sk);
 773
 774         if (!sock_owned_by_user(sk)) {
 775                 /* The local context ID may be out of date, update it. */
 776                 vsk->local_addr.svm_cid = dst.svm_cid;
 777
 778                 if (sk->sk_state == TCP_ESTABLISHED)
 779                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 780                                         sk, pkt, true, &dst, &src,
 781                                         &bh_process_pkt);
 782         }
 783
 784         bh_unlock_sock(sk);
 785
 786         if (!bh_process_pkt) {
 787                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 788
 789                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 790                 if (!recv_pkt_info) {
 791                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 792                                 pr_err("unable to send reset\n");
 793
 794                         err = VMCI_ERROR_NO_MEM;
 795                         goto out;
 796                 }
 797
 798                 recv_pkt_info->sk = sk;
 799                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 800                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 801
 802                 schedule_work(&recv_pkt_info->work);
 803                 /* Clear sk so that the reference count incremented by one of
 804                  * the Find functions above is not decremented below.  We need
 805                  * that reference count for the packet handler we've scheduled
 806                  * to run.
 807                  */
 808                 sk = NULL;
 809         }
 810
 811 out:
 812         if (sk)
 813                 sock_put(sk);
 814
 815         return err;
 816 }
 817
 818 static void vmci_transport_handle_detach(struct sock *sk)
 819 {
 820         struct vsock_sock *vsk;
 821
 822         vsk = vsock_sk(sk);
 823         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 824                 sock_set_flag(sk, SOCK_DONE);
 825
 826                 /* On a detach the peer will not be sending or receiving
 827                  * anymore.
 828                  */
 829                 vsk->peer_shutdown = SHUTDOWN_MASK;
 830
 831                 /* We should not be sending anymore since the peer won't be
 832                  * there to receive, but we can still receive if there is data
 833                  * left in our consume queue. If the local endpoint is a host,
 834                  * we can't call vsock_stream_has_data, since that may block,
 835                  * but a host endpoint can't read data once the VM has
 836                  * detached, so there is no available data in that case.
 837                  */
 838                 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
 839                     vsock_stream_has_data(vsk) <= 0) {
 840                         if (sk->sk_state == TCP_SYN_SENT) {
 841                                 /* The peer may detach from a queue pair while
 842                                  * we are still in the connecting state, i.e.,
 843                                  * if the peer VM is killed after attaching to
 844                                  * a queue pair, but before we complete the
 845                                  * handshake. In that case, we treat the detach
 846                                  * event like a reset.
 847                                  */
 848
 849                                 sk->sk_state = TCP_CLOSE;
 850                                 sk->sk_err = ECONNRESET;
 851                                 sk->sk_error_report(sk);
 852                                 return;
 853                         }
 854                         sk->sk_state = TCP_CLOSE;
 855                 }
 856                 sk->sk_state_change(sk);
 857         }
 858 }
 859
 860 static void vmci_transport_peer_detach_cb(u32 sub_id,
 861                                           const struct vmci_event_data *e_data,
 862                                           void *client_data)
 863 {
 864         struct vmci_transport *trans = client_data;
 865         const struct vmci_event_payload_qp *e_payload;
 866
 867         e_payload = vmci_event_data_const_payload(e_data);
 868
 869         /* XXX This is lame, we should provide a way to lookup sockets by
 870          * qp_handle.
 871          */
 872         if (vmci_handle_is_invalid(e_payload->handle) ||
 873             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 874                 return;
 875
 876         /* We don't ask for delayed CBs when we subscribe to this event (we
 877          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 878          * guarantees in that case about what context we might be running in,
 879          * so it could be BH or process, blockable or non-blockable.  So we
 880          * need to account for all possible contexts here.
 881          */
 882         spin_lock_bh(&trans->lock);
 883         if (!trans->sk)
 884                 goto out;
 885
 886         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 887          * where trans->sk isn't locked.
 888          */
 889         bh_lock_sock(trans->sk);
 890
 891         vmci_transport_handle_detach(trans->sk);
 892
 893         bh_unlock_sock(trans->sk);
 894  out:
 895         spin_unlock_bh(&trans->lock);
 896 }
 897
 898 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 899                                          const struct vmci_event_data *e_data,
 900                                          void *client_data)
 901 {
 902         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 903 }
 904
 905 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 906 {
 907         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 908         struct vmci_transport_packet *pkt;
 909         struct sock *sk;
 910
 911         recv_pkt_info =
 912                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 913         sk = recv_pkt_info->sk;
 914         pkt = &recv_pkt_info->pkt;
 915
 916         lock_sock(sk);
 917
 918         /* The local context ID may be out of date. */
 919         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 920
 921         switch (sk->sk_state) {
 922         case TCP_LISTEN:
 923                 vmci_transport_recv_listen(sk, pkt);
 924                 break;
 925         case TCP_SYN_SENT:
 926                 /* Processing of pending connections for servers goes through
 927                  * the listening socket, so see vmci_transport_recv_listen()
 928                  * for that path.
 929                  */
 930                 vmci_transport_recv_connecting_client(sk, pkt);
 931                 break;
 932         case TCP_ESTABLISHED:
 933                 vmci_transport_recv_connected(sk, pkt);
 934                 break;
 935         default:
 936                 /* Because this function does not run in the same context as
 937                  * vmci_transport_recv_stream_cb it is possible that the
 938                  * socket has closed. We need to let the other side know or it
 939                  * could be sitting in a connect and hang forever. Send a
 940                  * reset to prevent that.
 941                  */
 942                 vmci_transport_send_reset(sk, pkt);
 943                 break;
 944         }
 945
 946         release_sock(sk);
 947         kfree(recv_pkt_info);
 948         /* Release reference obtained in the stream callback when we fetched
 949          * this socket out of the bound or connected list.
 950          */
 951         sock_put(sk);
 952 }
 953
 954 static int vmci_transport_recv_listen(struct sock *sk,
 955                                       struct vmci_transport_packet *pkt)
 956 {
 957         struct sock *pending;
 958         struct vsock_sock *vpending;
 959         int err;
 960         u64 qp_size;
 961         bool old_request = false;
 962         bool old_pkt_proto = false;
 963
 964         err = 0;
 965
 966         /* Because we are in the listen state, we could be receiving a packet
 967          * for ourself or any previous connection requests that we received.
 968          * If it's the latter, we try to find a socket in our list of pending
 969          * connections and, if we do, call the appropriate handler for the
 970          * state that that socket is in.  Otherwise we try to service the
 971          * connection request.
 972          */
 973         pending = vmci_transport_get_pending(sk, pkt);
 974         if (pending) {
 975                 lock_sock(pending);
 976
 977                 /* The local context ID may be out of date. */
 978                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 979
 980                 switch (pending->sk_state) {
 981                 case TCP_SYN_SENT:
 982                         err = vmci_transport_recv_connecting_server(sk,
 983                                                                     pending,
 984                                                                     pkt);
 985                         break;
 986                 default:
 987                         vmci_transport_send_reset(pending, pkt);
 988                         err = -EINVAL;
 989                 }
 990
 991                 if (err < 0)
 992                         vsock_remove_pending(sk, pending);
 993
 994                 release_sock(pending);
 995                 vmci_transport_release_pending(pending);
 996
 997                 return err;
 998         }
 999
1000         /* The listen state only accepts connection requests.  Reply with a
1001          * reset unless we received a reset.
1002          */
1003
1004         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
1005               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
1006                 vmci_transport_reply_reset(pkt);
1007                 return -EINVAL;
1008         }
1009
1010         if (pkt->u.size == 0) {
1011                 vmci_transport_reply_reset(pkt);
1012                 return -EINVAL;
1013         }
1014
1015         /* If this socket can't accommodate this connection request, we send a
1016          * reset.  Otherwise we create and initialize a child socket and reply
1017          * with a connection negotiation.
1018          */
1019         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1020                 vmci_transport_reply_reset(pkt);
1021                 return -ECONNREFUSED;
1022         }
1023
1024         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1025                                  sk->sk_type, 0);
1026         if (!pending) {
1027                 vmci_transport_send_reset(sk, pkt);
1028                 return -ENOMEM;
1029         }
1030
1031         vpending = vsock_sk(pending);
1032
1033         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1034                         pkt->dst_port);
1035         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1036                         pkt->src_port);
1037
1038         /* If the proposed size fits within our min/max, accept it. Otherwise
1039          * propose our own size.
1040          */
1041         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1042             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1043                 qp_size = pkt->u.size;
1044         } else {
1045                 qp_size = vmci_trans(vpending)->queue_pair_size;
1046         }
1047
1048         /* Figure out if we are using old or new requests based on the
1049          * overrides pkt types sent by our peer.
1050          */
1051         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1052                 old_request = old_pkt_proto;
1053         } else {
1054                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1055                         old_request = true;
1056                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1057                         old_request = false;
1058
1059         }
1060
1061         if (old_request) {
1062                 /* Handle a REQUEST (or override) */
1063                 u16 version = VSOCK_PROTO_INVALID;
1064                 if (vmci_transport_proto_to_notify_struct(
1065                         pending, &version, true))
1066                         err = vmci_transport_send_negotiate(pending, qp_size);
1067                 else
1068                         err = -EINVAL;
1069
1070         } else {
1071                 /* Handle a REQUEST2 (or override) */
1072                 int proto_int = pkt->proto;
1073                 int pos;
1074                 u16 active_proto_version = 0;
1075
1076                 /* The list of possible protocols is the intersection of all
1077                  * protocols the client supports ... plus all the protocols we
1078                  * support.
1079                  */
1080                 proto_int &= vmci_transport_new_proto_supported_versions();
1081
1082                 /* We choose the highest possible protocol version and use that
1083                  * one.
1084                  */
1085                 pos = fls(proto_int);
1086                 if (pos) {
1087                         active_proto_version = (1 << (pos - 1));
1088                         if (vmci_transport_proto_to_notify_struct(
1089                                 pending, &active_proto_version, false))
1090                                 err = vmci_transport_send_negotiate2(pending,
1091                                                         qp_size,
1092                                                         active_proto_version);
1093                         else
1094                                 err = -EINVAL;
1095
1096                 } else {
1097                         err = -EINVAL;
1098                 }
1099         }
1100
1101         if (err < 0) {
1102                 vmci_transport_send_reset(sk, pkt);
1103                 sock_put(pending);
1104                 err = vmci_transport_error_to_vsock_error(err);
1105                 goto out;
1106         }
1107
1108         vsock_add_pending(sk, pending);
1109         sk->sk_ack_backlog++;
1110
1111         pending->sk_state = TCP_SYN_SENT;
1112         vmci_trans(vpending)->produce_size =
1113                 vmci_trans(vpending)->consume_size = qp_size;
1114         vmci_trans(vpending)->queue_pair_size = qp_size;
1115
1116         vmci_trans(vpending)->notify_ops->process_request(pending);
1117
1118         /* We might never receive another message for this socket and it's not
1119          * connected to any process, so we have to ensure it gets cleaned up
1120          * ourself.  Our delayed work function will take care of that.  Note
1121          * that we do not ever cancel this function since we have few
1122          * guarantees about its state when calling cancel_delayed_work().
1123          * Instead we hold a reference on the socket for that function and make
1124          * it capable of handling cases where it needs to do nothing but
1125          * release that reference.
1126          */
1127         vpending->listener = sk;
1128         sock_hold(sk);
1129         sock_hold(pending);
1130         schedule_delayed_work(&vpending->pending_work, HZ);
1131
1132 out:
1133         return err;
1134 }
1135
1136 static int
1137 vmci_transport_recv_connecting_server(struct sock *listener,
1138                                       struct sock *pending,
1139                                       struct vmci_transport_packet *pkt)
1140 {
1141         struct vsock_sock *vpending;
1142         struct vmci_handle handle;
1143         struct vmci_qp *qpair;
1144         bool is_local;
1145         u32 flags;
1146         u32 detach_sub_id;
1147         int err;
1148         int skerr;
1149
1150         vpending = vsock_sk(pending);
1151         detach_sub_id = VMCI_INVALID_ID;
1152
1153         switch (pkt->type) {
1154         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1155                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1156                         vmci_transport_send_reset(pending, pkt);
1157                         skerr = EPROTO;
1158                         err = -EINVAL;
1159                         goto destroy;
1160                 }
1161                 break;
1162         default:
1163                 /* Close and cleanup the connection. */
1164                 vmci_transport_send_reset(pending, pkt);
1165                 skerr = EPROTO;
1166                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1167                 goto destroy;
1168         }
1169
1170         /* In order to complete the connection we need to attach to the offered
1171          * queue pair and send an attach notification.  We also subscribe to the
1172          * detach event so we know when our peer goes away, and we do that
1173          * before attaching so we don't miss an event.  If all this succeeds,
1174          * we update our state and wakeup anything waiting in accept() for a
1175          * connection.
1176          */
1177
1178         /* We don't care about attach since we ensure the other side has
1179          * attached by specifying the ATTACH_ONLY flag below.
1180          */
1181         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1182                                    vmci_transport_peer_detach_cb,
1183                                    vmci_trans(vpending), &detach_sub_id);
1184         if (err < VMCI_SUCCESS) {
1185                 vmci_transport_send_reset(pending, pkt);
1186                 err = vmci_transport_error_to_vsock_error(err);
1187                 skerr = -err;
1188                 goto destroy;
1189         }
1190
1191         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1192
1193         /* Now attach to the queue pair the client created. */
1194         handle = pkt->u.handle;
1195
1196         /* vpending->local_addr always has a context id so we do not need to
1197          * worry about VMADDR_CID_ANY in this case.
1198          */
1199         is_local =
1200             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1201         flags = VMCI_QPFLAG_ATTACH_ONLY;
1202         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1203
1204         err = vmci_transport_queue_pair_alloc(
1205                                         &qpair,
1206                                         &handle,
1207                                         vmci_trans(vpending)->produce_size,
1208                                         vmci_trans(vpending)->consume_size,
1209                                         pkt->dg.src.context,
1210                                         flags,
1211                                         vmci_transport_is_trusted(
1212                                                 vpending,
1213                                                 vpending->remote_addr.svm_cid));
1214         if (err < 0) {
1215                 vmci_transport_send_reset(pending, pkt);
1216                 skerr = -err;
1217                 goto destroy;
1218         }
1219
1220         vmci_trans(vpending)->qp_handle = handle;
1221         vmci_trans(vpending)->qpair = qpair;
1222
1223         /* When we send the attach message, we must be ready to handle incoming
1224          * control messages on the newly connected socket. So we move the
1225          * pending socket to the connected state before sending the attach
1226          * message. Otherwise, an incoming packet triggered by the attach being
1227          * received by the peer may be processed concurrently with what happens
1228          * below after sending the attach message, and that incoming packet
1229          * will find the listening socket instead of the (currently) pending
1230          * socket. Note that enqueueing the socket increments the reference
1231          * count, so even if a reset comes before the connection is accepted,
1232          * the socket will be valid until it is removed from the queue.
1233          *
1234          * If we fail sending the attach below, we remove the socket from the
1235          * connected list and move the socket to TCP_CLOSE before
1236          * releasing the lock, so a pending slow path processing of an incoming
1237          * packet will not see the socket in the connected state in that case.
1238          */
1239         pending->sk_state = TCP_ESTABLISHED;
1240
1241         vsock_insert_connected(vpending);
1242
1243         /* Notify our peer of our attach. */
1244         err = vmci_transport_send_attach(pending, handle);
1245         if (err < 0) {
1246                 vsock_remove_connected(vpending);
1247                 pr_err("Could not send attach\n");
1248                 vmci_transport_send_reset(pending, pkt);
1249                 err = vmci_transport_error_to_vsock_error(err);
1250                 skerr = -err;
1251                 goto destroy;
1252         }
1253
1254         /* We have a connection. Move the now connected socket from the
1255          * listener's pending list to the accept queue so callers of accept()
1256          * can find it.
1257          */
1258         vsock_remove_pending(listener, pending);
1259         vsock_enqueue_accept(listener, pending);
1260
1261         /* Callers of accept() will be be waiting on the listening socket, not
1262          * the pending socket.
1263          */
1264         listener->sk_data_ready(listener);
1265
1266         return 0;
1267
1268 destroy:
1269         pending->sk_err = skerr;
1270         pending->sk_state = TCP_CLOSE;
1271         /* As long as we drop our reference, all necessary cleanup will handle
1272          * when the cleanup function drops its reference and our destruct
1273          * implementation is called.  Note that since the listen handler will
1274          * remove pending from the pending list upon our failure, the cleanup
1275          * function won't drop the additional reference, which is why we do it
1276          * here.
1277          */
1278         sock_put(pending);
1279
1280         return err;
1281 }
1282
1283 static int
1284 vmci_transport_recv_connecting_client(struct sock *sk,
1285                                       struct vmci_transport_packet *pkt)
1286 {
1287         struct vsock_sock *vsk;
1288         int err;
1289         int skerr;
1290
1291         vsk = vsock_sk(sk);
1292
1293         switch (pkt->type) {
1294         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1295                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1296                     !vmci_handle_is_equal(pkt->u.handle,
1297                                           vmci_trans(vsk)->qp_handle)) {
1298                         skerr = EPROTO;
1299                         err = -EINVAL;
1300                         goto destroy;
1301                 }
1302
1303                 /* Signify the socket is connected and wakeup the waiter in
1304                  * connect(). Also place the socket in the connected table for
1305                  * accounting (it can already be found since it's in the bound
1306                  * table).
1307                  */
1308                 sk->sk_state = TCP_ESTABLISHED;
1309                 sk->sk_socket->state = SS_CONNECTED;
1310                 vsock_insert_connected(vsk);
1311                 sk->sk_state_change(sk);
1312
1313                 break;
1314         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1315         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1316                 if (pkt->u.size == 0
1317                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1318                     || pkt->src_port != vsk->remote_addr.svm_port
1319                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1320                     || vmci_trans(vsk)->qpair
1321                     || vmci_trans(vsk)->produce_size != 0
1322                     || vmci_trans(vsk)->consume_size != 0
1323                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1324                         skerr = EPROTO;
1325                         err = -EINVAL;
1326
1327                         goto destroy;
1328                 }
1329
1330                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1331                 if (err) {
1332                         skerr = -err;
1333                         goto destroy;
1334                 }
1335
1336                 break;
1337         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1338                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1339                 if (err) {
1340                         skerr = -err;
1341                         goto destroy;
1342                 }
1343
1344                 break;
1345         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1346                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1347                  * continue processing here after they sent an INVALID packet.
1348                  * This meant that we got a RST after the INVALID. We ignore a
1349                  * RST after an INVALID. The common code doesn't send the RST
1350                  * ... so we can hang if an old version of the common code
1351                  * fails between getting a REQUEST and sending an OFFER back.
1352                  * Not much we can do about it... except hope that it doesn't
1353                  * happen.
1354                  */
1355                 if (vsk->ignore_connecting_rst) {
1356                         vsk->ignore_connecting_rst = false;
1357                 } else {
1358                         skerr = ECONNRESET;
1359                         err = 0;
1360                         goto destroy;
1361                 }
1362
1363                 break;
1364         default:
1365                 /* Close and cleanup the connection. */
1366                 skerr = EPROTO;
1367                 err = -EINVAL;
1368                 goto destroy;
1369         }
1370
1371         return 0;
1372
1373 destroy:
1374         vmci_transport_send_reset(sk, pkt);
1375
1376         sk->sk_state = TCP_CLOSE;
1377         sk->sk_err = skerr;
1378         sk->sk_error_report(sk);
1379         return err;
1380 }
1381
1382 static int vmci_transport_recv_connecting_client_negotiate(
1383                                         struct sock *sk,
1384                                         struct vmci_transport_packet *pkt)
1385 {
1386         int err;
1387         struct vsock_sock *vsk;
1388         struct vmci_handle handle;
1389         struct vmci_qp *qpair;
1390         u32 detach_sub_id;
1391         bool is_local;
1392         u32 flags;
1393         bool old_proto = true;
1394         bool old_pkt_proto;
1395         u16 version;
1396
1397         vsk = vsock_sk(sk);
1398         handle = VMCI_INVALID_HANDLE;
1399         detach_sub_id = VMCI_INVALID_ID;
1400
1401         /* If we have gotten here then we should be past the point where old
1402          * linux vsock could have sent the bogus rst.
1403          */
1404         vsk->sent_request = false;
1405         vsk->ignore_connecting_rst = false;
1406
1407         /* Verify that we're OK with the proposed queue pair size */
1408         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1409             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1410                 err = -EINVAL;
1411                 goto destroy;
1412         }
1413
1414         /* At this point we know the CID the peer is using to talk to us. */
1415
1416         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1417                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1418
1419         /* Setup the notify ops to be the highest supported version that both
1420          * the server and the client support.
1421          */
1422
1423         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1424                 old_proto = old_pkt_proto;
1425         } else {
1426                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1427                         old_proto = true;
1428                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1429                         old_proto = false;
1430
1431         }
1432
1433         if (old_proto)
1434                 version = VSOCK_PROTO_INVALID;
1435         else
1436                 version = pkt->proto;
1437
1438         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1439                 err = -EINVAL;
1440                 goto destroy;
1441         }
1442
1443         /* Subscribe to detach events first.
1444          *
1445          * XXX We attach once for each queue pair created for now so it is easy
1446          * to find the socket (it's provided), but later we should only
1447          * subscribe once and add a way to lookup sockets by queue pair handle.
1448          */
1449         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1450                                    vmci_transport_peer_detach_cb,
1451                                    vmci_trans(vsk), &detach_sub_id);
1452         if (err < VMCI_SUCCESS) {
1453                 err = vmci_transport_error_to_vsock_error(err);
1454                 goto destroy;
1455         }
1456
1457         /* Make VMCI select the handle for us. */
1458         handle = VMCI_INVALID_HANDLE;
1459         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1460         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1461
1462         err = vmci_transport_queue_pair_alloc(&qpair,
1463                                               &handle,
1464                                               pkt->u.size,
1465                                               pkt->u.size,
1466                                               vsk->remote_addr.svm_cid,
1467                                               flags,
1468                                               vmci_transport_is_trusted(
1469                                                   vsk,
1470                                                   vsk->
1471                                                   remote_addr.svm_cid));
1472         if (err < 0)
1473                 goto destroy;
1474
1475         err = vmci_transport_send_qp_offer(sk, handle);
1476         if (err < 0) {
1477                 err = vmci_transport_error_to_vsock_error(err);
1478                 goto destroy;
1479         }
1480
1481         vmci_trans(vsk)->qp_handle = handle;
1482         vmci_trans(vsk)->qpair = qpair;
1483
1484         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1485                 pkt->u.size;
1486
1487         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1488
1489         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1490
1491         return 0;
1492
1493 destroy:
1494         if (detach_sub_id != VMCI_INVALID_ID)
1495                 vmci_event_unsubscribe(detach_sub_id);
1496
1497         if (!vmci_handle_is_invalid(handle))
1498                 vmci_qpair_detach(&qpair);
1499
1500         return err;
1501 }
1502
1503 static int
1504 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1505                                               struct vmci_transport_packet *pkt)
1506 {
1507         int err = 0;
1508         struct vsock_sock *vsk = vsock_sk(sk);
1509
1510         if (vsk->sent_request) {
1511                 vsk->sent_request = false;
1512                 vsk->ignore_connecting_rst = true;
1513
1514                 err = vmci_transport_send_conn_request(
1515                         sk, vmci_trans(vsk)->queue_pair_size);
1516                 if (err < 0)
1517                         err = vmci_transport_error_to_vsock_error(err);
1518                 else
1519                         err = 0;
1520
1521         }
1522
1523         return err;
1524 }
1525
1526 static int vmci_transport_recv_connected(struct sock *sk,
1527                                          struct vmci_transport_packet *pkt)
1528 {
1529         struct vsock_sock *vsk;
1530         bool pkt_processed = false;
1531
1532         /* In cases where we are closing the connection, it's sufficient to
1533          * mark the state change (and maybe error) and wake up any waiting
1534          * threads. Since this is a connected socket, it's owned by a user
1535          * process and will be cleaned up when the failure is passed back on
1536          * the current or next system call.  Our system call implementations
1537          * must therefore check for error and state changes on entry and when
1538          * being awoken.
1539          */
1540         switch (pkt->type) {
1541         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1542                 if (pkt->u.mode) {
1543                         vsk = vsock_sk(sk);
1544
1545                         vsk->peer_shutdown |= pkt->u.mode;
1546                         sk->sk_state_change(sk);
1547                 }
1548                 break;
1549
1550         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1551                 vsk = vsock_sk(sk);
1552                 /* It is possible that we sent our peer a message (e.g a
1553                  * WAITING_READ) right before we got notified that the peer had
1554                  * detached. If that happens then we can get a RST pkt back
1555                  * from our peer even though there is data available for us to
1556                  * read. In that case, don't shutdown the socket completely but
1557                  * instead allow the local client to finish reading data off
1558                  * the queuepair. Always treat a RST pkt in connected mode like
1559                  * a clean shutdown.
1560                  */
1561                 sock_set_flag(sk, SOCK_DONE);
1562                 vsk->peer_shutdown = SHUTDOWN_MASK;
1563                 if (vsock_stream_has_data(vsk) <= 0)
1564                         sk->sk_state = TCP_CLOSING;
1565
1566                 sk->sk_state_change(sk);
1567                 break;
1568
1569         default:
1570                 vsk = vsock_sk(sk);
1571                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1572                                 sk, pkt, false, NULL, NULL,
1573                                 &pkt_processed);
1574                 if (!pkt_processed)
1575                         return -EINVAL;
1576
1577                 break;
1578         }
1579
1580         return 0;
1581 }
1582
1583 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1584                                       struct vsock_sock *psk)
1585 {
1586         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1587         if (!vsk->trans)
1588                 return -ENOMEM;
1589
1590         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1591         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1592         vmci_trans(vsk)->qpair = NULL;
1593         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1594         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1595         vmci_trans(vsk)->notify_ops = NULL;
1596         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1597         vmci_trans(vsk)->sk = &vsk->sk;
1598         spin_lock_init(&vmci_trans(vsk)->lock);
1599         if (psk) {
1600                 vmci_trans(vsk)->queue_pair_size =
1601                         vmci_trans(psk)->queue_pair_size;
1602                 vmci_trans(vsk)->queue_pair_min_size =
1603                         vmci_trans(psk)->queue_pair_min_size;
1604                 vmci_trans(vsk)->queue_pair_max_size =
1605                         vmci_trans(psk)->queue_pair_max_size;
1606         } else {
1607                 vmci_trans(vsk)->queue_pair_size =
1608                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1609                 vmci_trans(vsk)->queue_pair_min_size =
1610                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1611                 vmci_trans(vsk)->queue_pair_max_size =
1612                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1613         }
1614
1615         return 0;
1616 }
1617
1618 static void vmci_transport_free_resources(struct list_head *transport_list)
1619 {
1620         while (!list_empty(transport_list)) {
1621                 struct vmci_transport *transport =
1622                     list_first_entry(transport_list, struct vmci_transport,
1623                                      elem);
1624                 list_del(&transport->elem);
1625
1626                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1627                         vmci_event_unsubscribe(transport->detach_sub_id);
1628                         transport->detach_sub_id = VMCI_INVALID_ID;
1629                 }
1630
1631                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1632                         vmci_qpair_detach(&transport->qpair);
1633                         transport->qp_handle = VMCI_INVALID_HANDLE;
1634                         transport->produce_size = 0;
1635                         transport->consume_size = 0;
1636                 }
1637
1638                 kfree(transport);
1639         }
1640 }
1641
1642 static void vmci_transport_cleanup(struct work_struct *work)
1643 {
1644         LIST_HEAD(pending);
1645
1646         spin_lock_bh(&vmci_transport_cleanup_lock);
1647         list_replace_init(&vmci_transport_cleanup_list, &pending);
1648         spin_unlock_bh(&vmci_transport_cleanup_lock);
1649         vmci_transport_free_resources(&pending);
1650 }
1651
1652 static void vmci_transport_destruct(struct vsock_sock *vsk)
1653 {
1654         /* Ensure that the detach callback doesn't use the sk/vsk
1655          * we are about to destruct.
1656          */
1657         spin_lock_bh(&vmci_trans(vsk)->lock);
1658         vmci_trans(vsk)->sk = NULL;
1659         spin_unlock_bh(&vmci_trans(vsk)->lock);
1660
1661         if (vmci_trans(vsk)->notify_ops)
1662                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1663
1664         spin_lock_bh(&vmci_transport_cleanup_lock);
1665         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1666         spin_unlock_bh(&vmci_transport_cleanup_lock);
1667         schedule_work(&vmci_transport_cleanup_work);
1668
1669         vsk->trans = NULL;
1670 }
1671
1672 static void vmci_transport_release(struct vsock_sock *vsk)
1673 {
1674         vsock_remove_sock(vsk);
1675
1676         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1677                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1678                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1679         }
1680 }
1681
1682 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1683                                      struct sockaddr_vm *addr)
1684 {
1685         u32 port;
1686         u32 flags;
1687         int err;
1688
1689         /* VMCI will select a resource ID for us if we provide
1690          * VMCI_INVALID_ID.
1691          */
1692         port = addr->svm_port == VMADDR_PORT_ANY ?
1693                         VMCI_INVALID_ID : addr->svm_port;
1694
1695         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1696                 return -EACCES;
1697
1698         flags = addr->svm_cid == VMADDR_CID_ANY ?
1699                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1700
1701         err = vmci_transport_datagram_create_hnd(port, flags,
1702                                                  vmci_transport_recv_dgram_cb,
1703                                                  &vsk->sk,
1704                                                  &vmci_trans(vsk)->dg_handle);
1705         if (err < VMCI_SUCCESS)
1706                 return vmci_transport_error_to_vsock_error(err);
1707         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1708                         vmci_trans(vsk)->dg_handle.resource);
1709
1710         return 0;
1711 }
1712
1713 static int vmci_transport_dgram_enqueue(
1714         struct vsock_sock *vsk,
1715         struct sockaddr_vm *remote_addr,
1716         struct msghdr *msg,
1717         size_t len)
1718 {
1719         int err;
1720         struct vmci_datagram *dg;
1721
1722         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1723                 return -EMSGSIZE;
1724
1725         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1726                 return -EPERM;
1727
1728         /* Allocate a buffer for the user's message and our packet header. */
1729         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1730         if (!dg)
1731                 return -ENOMEM;
1732
1733         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1734
1735         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1736                                    remote_addr->svm_port);
1737         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1738                                    vsk->local_addr.svm_port);
1739         dg->payload_size = len;
1740
1741         err = vmci_datagram_send(dg);
1742         kfree(dg);
1743         if (err < 0)
1744                 return vmci_transport_error_to_vsock_error(err);
1745
1746         return err - sizeof(*dg);
1747 }
1748
1749 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1750                                         struct msghdr *msg, size_t len,
1751                                         int flags)
1752 {
1753         int err;
1754         int noblock;
1755         struct vmci_datagram *dg;
1756         size_t payload_len;
1757         struct sk_buff *skb;
1758
1759         noblock = flags & MSG_DONTWAIT;
1760
1761         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1762                 return -EOPNOTSUPP;
1763
1764         /* Retrieve the head sk_buff from the socket's receive queue. */
1765         err = 0;
1766         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1767         if (!skb)
1768                 return err;
1769
1770         dg = (struct vmci_datagram *)skb->data;
1771         if (!dg)
1772                 /* err is 0, meaning we read zero bytes. */
1773                 goto out;
1774
1775         payload_len = dg->payload_size;
1776         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1777         if (payload_len != skb->len - sizeof(*dg)) {
1778                 err = -EINVAL;
1779                 goto out;
1780         }
1781
1782         if (payload_len > len) {
1783                 payload_len = len;
1784                 msg->msg_flags |= MSG_TRUNC;
1785         }
1786
1787         /* Place the datagram payload in the user's iovec. */
1788         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1789         if (err)
1790                 goto out;
1791
1792         if (msg->msg_name) {
1793                 /* Provide the address of the sender. */
1794                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1795                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1796                 msg->msg_namelen = sizeof(*vm_addr);
1797         }
1798         err = payload_len;
1799
1800 out:
1801         skb_free_datagram(&vsk->sk, skb);
1802         return err;
1803 }
1804
1805 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1806 {
1807         if (cid == VMADDR_CID_HYPERVISOR) {
1808                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1809                  * state and are allowed.
1810                  */
1811                 return port == VMCI_UNITY_PBRPC_REGISTER;
1812         }
1813
1814         return true;
1815 }
1816
1817 static int vmci_transport_connect(struct vsock_sock *vsk)
1818 {
1819         int err;
1820         bool old_pkt_proto = false;
1821         struct sock *sk = &vsk->sk;
1822
1823         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1824                 old_pkt_proto) {
1825                 err = vmci_transport_send_conn_request(
1826                         sk, vmci_trans(vsk)->queue_pair_size);
1827                 if (err < 0) {
1828                         sk->sk_state = TCP_CLOSE;
1829                         return err;
1830                 }
1831         } else {
1832                 int supported_proto_versions =
1833                         vmci_transport_new_proto_supported_versions();
1834                 err = vmci_transport_send_conn_request2(
1835                                 sk, vmci_trans(vsk)->queue_pair_size,
1836                                 supported_proto_versions);
1837                 if (err < 0) {
1838                         sk->sk_state = TCP_CLOSE;
1839                         return err;
1840                 }
1841
1842                 vsk->sent_request = true;
1843         }
1844
1845         return err;
1846 }
1847
1848 static ssize_t vmci_transport_stream_dequeue(
1849         struct vsock_sock *vsk,
1850         struct msghdr *msg,
1851         size_t len,
1852         int flags)
1853 {
1854         if (flags & MSG_PEEK)
1855                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1856         else
1857                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1858 }
1859
1860 static ssize_t vmci_transport_stream_enqueue(
1861         struct vsock_sock *vsk,
1862         struct msghdr *msg,
1863         size_t len)
1864 {
1865         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1866 }
1867
1868 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1869 {
1870         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1871 }
1872
1873 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1874 {
1875         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1876 }
1877
1878 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1879 {
1880         return vmci_trans(vsk)->consume_size;
1881 }
1882
1883 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1884 {
1885         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1886 }
1887
1888 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1889 {
1890         return vmci_trans(vsk)->queue_pair_size;
1891 }
1892
1893 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1894 {
1895         return vmci_trans(vsk)->queue_pair_min_size;
1896 }
1897
1898 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1899 {
1900         return vmci_trans(vsk)->queue_pair_max_size;
1901 }
1902
1903 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1904 {
1905         if (val < vmci_trans(vsk)->queue_pair_min_size)
1906                 vmci_trans(vsk)->queue_pair_min_size = val;
1907         if (val > vmci_trans(vsk)->queue_pair_max_size)
1908                 vmci_trans(vsk)->queue_pair_max_size = val;
1909         vmci_trans(vsk)->queue_pair_size = val;
1910 }
1911
1912 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1913                                                u64 val)
1914 {
1915         if (val > vmci_trans(vsk)->queue_pair_size)
1916                 vmci_trans(vsk)->queue_pair_size = val;
1917         vmci_trans(vsk)->queue_pair_min_size = val;
1918 }
1919
1920 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1921                                                u64 val)
1922 {
1923         if (val < vmci_trans(vsk)->queue_pair_size)
1924                 vmci_trans(vsk)->queue_pair_size = val;
1925         vmci_trans(vsk)->queue_pair_max_size = val;
1926 }
1927
1928 static int vmci_transport_notify_poll_in(
1929         struct vsock_sock *vsk,
1930         size_t target,
1931         bool *data_ready_now)
1932 {
1933         return vmci_trans(vsk)->notify_ops->poll_in(
1934                         &vsk->sk, target, data_ready_now);
1935 }
1936
1937 static int vmci_transport_notify_poll_out(
1938         struct vsock_sock *vsk,
1939         size_t target,
1940         bool *space_available_now)
1941 {
1942         return vmci_trans(vsk)->notify_ops->poll_out(
1943                         &vsk->sk, target, space_available_now);
1944 }
1945
1946 static int vmci_transport_notify_recv_init(
1947         struct vsock_sock *vsk,
1948         size_t target,
1949         struct vsock_transport_recv_notify_data *data)
1950 {
1951         return vmci_trans(vsk)->notify_ops->recv_init(
1952                         &vsk->sk, target,
1953                         (struct vmci_transport_recv_notify_data *)data);
1954 }
1955
1956 static int vmci_transport_notify_recv_pre_block(
1957         struct vsock_sock *vsk,
1958         size_t target,
1959         struct vsock_transport_recv_notify_data *data)
1960 {
1961         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1962                         &vsk->sk, target,
1963                         (struct vmci_transport_recv_notify_data *)data);
1964 }
1965
1966 static int vmci_transport_notify_recv_pre_dequeue(
1967         struct vsock_sock *vsk,
1968         size_t target,
1969         struct vsock_transport_recv_notify_data *data)
1970 {
1971         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1972                         &vsk->sk, target,
1973                         (struct vmci_transport_recv_notify_data *)data);
1974 }
1975
1976 static int vmci_transport_notify_recv_post_dequeue(
1977         struct vsock_sock *vsk,
1978         size_t target,
1979         ssize_t copied,
1980         bool data_read,
1981         struct vsock_transport_recv_notify_data *data)
1982 {
1983         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1984                         &vsk->sk, target, copied, data_read,
1985                         (struct vmci_transport_recv_notify_data *)data);
1986 }
1987
1988 static int vmci_transport_notify_send_init(
1989         struct vsock_sock *vsk,
1990         struct vsock_transport_send_notify_data *data)
1991 {
1992         return vmci_trans(vsk)->notify_ops->send_init(
1993                         &vsk->sk,
1994                         (struct vmci_transport_send_notify_data *)data);
1995 }
1996
1997 static int vmci_transport_notify_send_pre_block(
1998         struct vsock_sock *vsk,
1999         struct vsock_transport_send_notify_data *data)
2000 {
2001         return vmci_trans(vsk)->notify_ops->send_pre_block(
2002                         &vsk->sk,
2003                         (struct vmci_transport_send_notify_data *)data);
2004 }
2005
2006 static int vmci_transport_notify_send_pre_enqueue(
2007         struct vsock_sock *vsk,
2008         struct vsock_transport_send_notify_data *data)
2009 {
2010         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2011                         &vsk->sk,
2012                         (struct vmci_transport_send_notify_data *)data);
2013 }
2014
2015 static int vmci_transport_notify_send_post_enqueue(
2016         struct vsock_sock *vsk,
2017         ssize_t written,
2018         struct vsock_transport_send_notify_data *data)
2019 {
2020         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2021                         &vsk->sk, written,
2022                         (struct vmci_transport_send_notify_data *)data);
2023 }
2024
2025 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2026 {
2027         if (PROTOCOL_OVERRIDE != -1) {
2028                 if (PROTOCOL_OVERRIDE == 0)
2029                         *old_pkt_proto = true;
2030                 else
2031                         *old_pkt_proto = false;
2032
2033                 pr_info("Proto override in use\n");
2034                 return true;
2035         }
2036
2037         return false;
2038 }
2039
2040 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2041                                                   u16 *proto,
2042                                                   bool old_pkt_proto)
2043 {
2044         struct vsock_sock *vsk = vsock_sk(sk);
2045
2046         if (old_pkt_proto) {
2047                 if (*proto != VSOCK_PROTO_INVALID) {
2048                         pr_err("Can't set both an old and new protocol\n");
2049                         return false;
2050                 }
2051                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2052                 goto exit;
2053         }
2054
2055         switch (*proto) {
2056         case VSOCK_PROTO_PKT_ON_NOTIFY:
2057                 vmci_trans(vsk)->notify_ops =
2058                         &vmci_transport_notify_pkt_q_state_ops;
2059                 break;
2060         default:
2061                 pr_err("Unknown notify protocol version\n");
2062                 return false;
2063         }
2064
2065 exit:
2066         vmci_trans(vsk)->notify_ops->socket_init(sk);
2067         return true;
2068 }
2069
2070 static u16 vmci_transport_new_proto_supported_versions(void)
2071 {
2072         if (PROTOCOL_OVERRIDE != -1)
2073                 return PROTOCOL_OVERRIDE;
2074
2075         return VSOCK_PROTO_ALL_SUPPORTED;
2076 }
2077
2078 static u32 vmci_transport_get_local_cid(void)
2079 {
2080         return vmci_get_context_id();
2081 }
2082
2083 static const struct vsock_transport vmci_transport = {
2084         .init = vmci_transport_socket_init,
2085         .destruct = vmci_transport_destruct,
2086         .release = vmci_transport_release,
2087         .connect = vmci_transport_connect,
2088         .dgram_bind = vmci_transport_dgram_bind,
2089         .dgram_dequeue = vmci_transport_dgram_dequeue,
2090         .dgram_enqueue = vmci_transport_dgram_enqueue,
2091         .dgram_allow = vmci_transport_dgram_allow,
2092         .stream_dequeue = vmci_transport_stream_dequeue,
2093         .stream_enqueue = vmci_transport_stream_enqueue,
2094         .stream_has_data = vmci_transport_stream_has_data,
2095         .stream_has_space = vmci_transport_stream_has_space,
2096         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2097         .stream_is_active = vmci_transport_stream_is_active,
2098         .stream_allow = vmci_transport_stream_allow,
2099         .notify_poll_in = vmci_transport_notify_poll_in,
2100         .notify_poll_out = vmci_transport_notify_poll_out,
2101         .notify_recv_init = vmci_transport_notify_recv_init,
2102         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2103         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2104         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2105         .notify_send_init = vmci_transport_notify_send_init,
2106         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2107         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2108         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2109         .shutdown = vmci_transport_shutdown,
2110         .set_buffer_size = vmci_transport_set_buffer_size,
2111         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2112         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2113         .get_buffer_size = vmci_transport_get_buffer_size,
2114         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2115         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2116         .get_local_cid = vmci_transport_get_local_cid,
2117 };
2118
2119 static int __init vmci_transport_init(void)
2120 {
2121         int err;
2122
2123         /* Create the datagram handle that we will use to send and receive all
2124          * VSocket control messages for this context.
2125          */
2126         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2127                                                  VMCI_FLAG_ANYCID_DG_HND,
2128                                                  vmci_transport_recv_stream_cb,
2129                                                  NULL,
2130                                                  &vmci_transport_stream_handle);
2131         if (err < VMCI_SUCCESS) {
2132                 pr_err("Unable to create datagram handle. (%d)\n", err);
2133                 return vmci_transport_error_to_vsock_error(err);
2134         }
2135
2136         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2137                                    vmci_transport_qp_resumed_cb,
2138                                    NULL, &vmci_transport_qp_resumed_sub_id);
2139         if (err < VMCI_SUCCESS) {
2140                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2141                 err = vmci_transport_error_to_vsock_error(err);
2142                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2143                 goto err_destroy_stream_handle;
2144         }
2145
2146         err = vsock_core_init(&vmci_transport);
2147         if (err < 0)
2148                 goto err_unsubscribe;
2149
2150         return 0;
2151
2152 err_unsubscribe:
2153         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2154 err_destroy_stream_handle:
2155         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2156         return err;
2157 }
2158 module_init(vmci_transport_init);
2159
2160 static void __exit vmci_transport_exit(void)
2161 {
2162         cancel_work_sync(&vmci_transport_cleanup_work);
2163         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2164
2165         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2166                 if (vmci_datagram_destroy_handle(
2167                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2168                         pr_err("Couldn't destroy datagram handle\n");
2169                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2170         }
2171
2172         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2173                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2174                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2175         }
2176
2177         vsock_core_exit();
2178 }
2179 module_exit(vmci_transport_exit);
2180
2181 MODULE_AUTHOR("VMware, Inc.");
2182 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2183 MODULE_VERSION("1.0.5.0-k");
2184 MODULE_LICENSE("GPL v2");
2185 MODULE_ALIAS("vmware_vsock");
2186 MODULE_ALIAS_NETPROTO(PF_VSOCK);