net/vmw_vsock/vmci_transport.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VMware vSockets Driver
   4  *
   5  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   6  */
   7
   8 #include <linux/types.h>
   9 #include <linux/bitops.h>
  10 #include <linux/cred.h>
  11 #include <linux/init.h>
  12 #include <linux/io.h>
  13 #include <linux/kernel.h>
  14 #include <linux/kmod.h>
  15 #include <linux/list.h>
  16 #include <linux/module.h>
  17 #include <linux/mutex.h>
  18 #include <linux/net.h>
  19 #include <linux/poll.h>
  20 #include <linux/skbuff.h>
  21 #include <linux/smp.h>
  22 #include <linux/socket.h>
  23 #include <linux/stddef.h>
  24 #include <linux/unistd.h>
  25 #include <linux/wait.h>
  26 #include <linux/workqueue.h>
  27 #include <net/sock.h>
  28 #include <net/af_vsock.h>
  29
  30 #include "vmci_transport_notify.h"
  31
  32 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
  33 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
  34 static void vmci_transport_peer_detach_cb(u32 sub_id,
  35                                           const struct vmci_event_data *ed,
  36                                           void *client_data);
  37 static void vmci_transport_recv_pkt_work(struct work_struct *work);
  38 static void vmci_transport_cleanup(struct work_struct *work);
  39 static int vmci_transport_recv_listen(struct sock *sk,
  40                                       struct vmci_transport_packet *pkt);
  41 static int vmci_transport_recv_connecting_server(
  42                                         struct sock *sk,
  43                                         struct sock *pending,
  44                                         struct vmci_transport_packet *pkt);
  45 static int vmci_transport_recv_connecting_client(
  46                                         struct sock *sk,
  47                                         struct vmci_transport_packet *pkt);
  48 static int vmci_transport_recv_connecting_client_negotiate(
  49                                         struct sock *sk,
  50                                         struct vmci_transport_packet *pkt);
  51 static int vmci_transport_recv_connecting_client_invalid(
  52                                         struct sock *sk,
  53                                         struct vmci_transport_packet *pkt);
  54 static int vmci_transport_recv_connected(struct sock *sk,
  55                                          struct vmci_transport_packet *pkt);
  56 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
  57 static u16 vmci_transport_new_proto_supported_versions(void);
  58 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
  59                                                   bool old_pkt_proto);
  60 static bool vmci_check_transport(struct vsock_sock *vsk);
  61
  62 struct vmci_transport_recv_pkt_info {
  63         struct work_struct work;
  64         struct sock *sk;
  65         struct vmci_transport_packet pkt;
  66 };
  67
  68 static LIST_HEAD(vmci_transport_cleanup_list);
  69 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
  70 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
  71
  72 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  73                                                            VMCI_INVALID_ID };
  74 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  75
  76 static int PROTOCOL_OVERRIDE = -1;
  77
  78 static struct vsock_transport vmci_transport; /* forward declaration */
  79
  80 /* Helper function to convert from a VMCI error code to a VSock error code. */
  81
  82 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  83 {
  84         switch (vmci_error) {
  85         case VMCI_ERROR_NO_MEM:
  86                 return -ENOMEM;
  87         case VMCI_ERROR_DUPLICATE_ENTRY:
  88         case VMCI_ERROR_ALREADY_EXISTS:
  89                 return -EADDRINUSE;
  90         case VMCI_ERROR_NO_ACCESS:
  91                 return -EPERM;
  92         case VMCI_ERROR_NO_RESOURCES:
  93                 return -ENOBUFS;
  94         case VMCI_ERROR_INVALID_RESOURCE:
  95                 return -EHOSTUNREACH;
  96         case VMCI_ERROR_INVALID_ARGS:
  97         default:
  98                 break;
  99         }
 100         return -EINVAL;
 101 }
 102
 103 static u32 vmci_transport_peer_rid(u32 peer_cid)
 104 {
 105         if (VMADDR_CID_HYPERVISOR == peer_cid)
 106                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 107
 108         return VMCI_TRANSPORT_PACKET_RID;
 109 }
 110
 111 static inline void
 112 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 113                            struct sockaddr_vm *src,
 114                            struct sockaddr_vm *dst,
 115                            u8 type,
 116                            u64 size,
 117                            u64 mode,
 118                            struct vmci_transport_waiting_info *wait,
 119                            u16 proto,
 120                            struct vmci_handle handle)
 121 {
 122         /* We register the stream control handler as an any cid handle so we
 123          * must always send from a source address of VMADDR_CID_ANY
 124          */
 125         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 126                                        VMCI_TRANSPORT_PACKET_RID);
 127         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 128                                        vmci_transport_peer_rid(dst->svm_cid));
 129         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 130         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 131         pkt->type = type;
 132         pkt->src_port = src->svm_port;
 133         pkt->dst_port = dst->svm_port;
 134         memset(&pkt->proto, 0, sizeof(pkt->proto));
 135         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 136
 137         switch (pkt->type) {
 138         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 139                 pkt->u.size = 0;
 140                 break;
 141
 142         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 143         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 144                 pkt->u.size = size;
 145                 break;
 146
 147         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 148         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 149                 pkt->u.handle = handle;
 150                 break;
 151
 152         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 153         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 154         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 155                 pkt->u.size = 0;
 156                 break;
 157
 158         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 159                 pkt->u.mode = mode;
 160                 break;
 161
 162         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 163         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 164                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 165                 break;
 166
 167         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 168         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 169                 pkt->u.size = size;
 170                 pkt->proto = proto;
 171                 break;
 172         }
 173 }
 174
 175 static inline void
 176 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 177                                     struct sockaddr_vm *local,
 178                                     struct sockaddr_vm *remote)
 179 {
 180         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 181         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 182 }
 183
 184 static int
 185 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 186                                   struct sockaddr_vm *src,
 187                                   struct sockaddr_vm *dst,
 188                                   enum vmci_transport_packet_type type,
 189                                   u64 size,
 190                                   u64 mode,
 191                                   struct vmci_transport_waiting_info *wait,
 192                                   u16 proto,
 193                                   struct vmci_handle handle,
 194                                   bool convert_error)
 195 {
 196         int err;
 197
 198         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 199                                    proto, handle);
 200         err = vmci_datagram_send(&pkt->dg);
 201         if (convert_error && (err < 0))
 202                 return vmci_transport_error_to_vsock_error(err);
 203
 204         return err;
 205 }
 206
 207 static int
 208 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 209                                       enum vmci_transport_packet_type type,
 210                                       u64 size,
 211                                       u64 mode,
 212                                       struct vmci_transport_waiting_info *wait,
 213                                       struct vmci_handle handle)
 214 {
 215         struct vmci_transport_packet reply;
 216         struct sockaddr_vm src, dst;
 217
 218         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 219                 return 0;
 220         } else {
 221                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 222                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 223                                                          type,
 224                                                          size, mode, wait,
 225                                                          VSOCK_PROTO_INVALID,
 226                                                          handle, true);
 227         }
 228 }
 229
 230 static int
 231 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 232                                    struct sockaddr_vm *dst,
 233                                    enum vmci_transport_packet_type type,
 234                                    u64 size,
 235                                    u64 mode,
 236                                    struct vmci_transport_waiting_info *wait,
 237                                    struct vmci_handle handle)
 238 {
 239         /* Note that it is safe to use a single packet across all CPUs since
 240          * two tasklets of the same type are guaranteed to not ever run
 241          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 242          * we can use per-cpu packets.
 243          */
 244         static struct vmci_transport_packet pkt;
 245
 246         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 247                                                  size, mode, wait,
 248                                                  VSOCK_PROTO_INVALID, handle,
 249                                                  false);
 250 }
 251
 252 static int
 253 vmci_transport_alloc_send_control_pkt(struct sockaddr_vm *src,
 254                                       struct sockaddr_vm *dst,
 255                                       enum vmci_transport_packet_type type,
 256                                       u64 size,
 257                                       u64 mode,
 258                                       struct vmci_transport_waiting_info *wait,
 259                                       u16 proto,
 260                                       struct vmci_handle handle)
 261 {
 262         struct vmci_transport_packet *pkt;
 263         int err;
 264
 265         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 266         if (!pkt)
 267                 return -ENOMEM;
 268
 269         err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
 270                                                 mode, wait, proto, handle,
 271                                                 true);
 272         kfree(pkt);
 273
 274         return err;
 275 }
 276
 277 static int
 278 vmci_transport_send_control_pkt(struct sock *sk,
 279                                 enum vmci_transport_packet_type type,
 280                                 u64 size,
 281                                 u64 mode,
 282                                 struct vmci_transport_waiting_info *wait,
 283                                 u16 proto,
 284                                 struct vmci_handle handle)
 285 {
 286         struct vsock_sock *vsk;
 287
 288         vsk = vsock_sk(sk);
 289
 290         if (!vsock_addr_bound(&vsk->local_addr))
 291                 return -EINVAL;
 292
 293         if (!vsock_addr_bound(&vsk->remote_addr))
 294                 return -EINVAL;
 295
 296         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
 297                                                      &vsk->remote_addr,
 298                                                      type, size, mode,
 299                                                      wait, proto, handle);
 300 }
 301
 302 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 303                                         struct sockaddr_vm *src,
 304                                         struct vmci_transport_packet *pkt)
 305 {
 306         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 307                 return 0;
 308         return vmci_transport_send_control_pkt_bh(
 309                                         dst, src,
 310                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 311                                         0, NULL, VMCI_INVALID_HANDLE);
 312 }
 313
 314 static int vmci_transport_send_reset(struct sock *sk,
 315                                      struct vmci_transport_packet *pkt)
 316 {
 317         struct sockaddr_vm *dst_ptr;
 318         struct sockaddr_vm dst;
 319         struct vsock_sock *vsk;
 320
 321         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 322                 return 0;
 323
 324         vsk = vsock_sk(sk);
 325
 326         if (!vsock_addr_bound(&vsk->local_addr))
 327                 return -EINVAL;
 328
 329         if (vsock_addr_bound(&vsk->remote_addr)) {
 330                 dst_ptr = &vsk->remote_addr;
 331         } else {
 332                 vsock_addr_init(&dst, pkt->dg.src.context,
 333                                 pkt->src_port);
 334                 dst_ptr = &dst;
 335         }
 336         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
 337                                              VMCI_TRANSPORT_PACKET_TYPE_RST,
 338                                              0, 0, NULL, VSOCK_PROTO_INVALID,
 339                                              VMCI_INVALID_HANDLE);
 340 }
 341
 342 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 343 {
 344         return vmci_transport_send_control_pkt(
 345                                         sk,
 346                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 347                                         size, 0, NULL,
 348                                         VSOCK_PROTO_INVALID,
 349                                         VMCI_INVALID_HANDLE);
 350 }
 351
 352 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 353                                           u16 version)
 354 {
 355         return vmci_transport_send_control_pkt(
 356                                         sk,
 357                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 358                                         size, 0, NULL, version,
 359                                         VMCI_INVALID_HANDLE);
 360 }
 361
 362 static int vmci_transport_send_qp_offer(struct sock *sk,
 363                                         struct vmci_handle handle)
 364 {
 365         return vmci_transport_send_control_pkt(
 366                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 367                                         0, NULL,
 368                                         VSOCK_PROTO_INVALID, handle);
 369 }
 370
 371 static int vmci_transport_send_attach(struct sock *sk,
 372                                       struct vmci_handle handle)
 373 {
 374         return vmci_transport_send_control_pkt(
 375                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 376                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 377                                         handle);
 378 }
 379
 380 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 381 {
 382         return vmci_transport_reply_control_pkt_fast(
 383                                                 pkt,
 384                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 385                                                 0, 0, NULL,
 386                                                 VMCI_INVALID_HANDLE);
 387 }
 388
 389 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 390                                           struct sockaddr_vm *src)
 391 {
 392         return vmci_transport_send_control_pkt_bh(
 393                                         dst, src,
 394                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 395                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 396 }
 397
 398 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 399                                  struct sockaddr_vm *src)
 400 {
 401         return vmci_transport_send_control_pkt_bh(
 402                                         dst, src,
 403                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 404                                         0, NULL, VMCI_INVALID_HANDLE);
 405 }
 406
 407 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 408                                 struct sockaddr_vm *src)
 409 {
 410         return vmci_transport_send_control_pkt_bh(
 411                                         dst, src,
 412                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 413                                         0, NULL, VMCI_INVALID_HANDLE);
 414 }
 415
 416 int vmci_transport_send_wrote(struct sock *sk)
 417 {
 418         return vmci_transport_send_control_pkt(
 419                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 420                                         0, NULL, VSOCK_PROTO_INVALID,
 421                                         VMCI_INVALID_HANDLE);
 422 }
 423
 424 int vmci_transport_send_read(struct sock *sk)
 425 {
 426         return vmci_transport_send_control_pkt(
 427                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 428                                         0, NULL, VSOCK_PROTO_INVALID,
 429                                         VMCI_INVALID_HANDLE);
 430 }
 431
 432 int vmci_transport_send_waiting_write(struct sock *sk,
 433                                       struct vmci_transport_waiting_info *wait)
 434 {
 435         return vmci_transport_send_control_pkt(
 436                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 437                                 0, 0, wait, VSOCK_PROTO_INVALID,
 438                                 VMCI_INVALID_HANDLE);
 439 }
 440
 441 int vmci_transport_send_waiting_read(struct sock *sk,
 442                                      struct vmci_transport_waiting_info *wait)
 443 {
 444         return vmci_transport_send_control_pkt(
 445                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 446                                 0, 0, wait, VSOCK_PROTO_INVALID,
 447                                 VMCI_INVALID_HANDLE);
 448 }
 449
 450 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 451 {
 452         return vmci_transport_send_control_pkt(
 453                                         &vsk->sk,
 454                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 455                                         0, mode, NULL,
 456                                         VSOCK_PROTO_INVALID,
 457                                         VMCI_INVALID_HANDLE);
 458 }
 459
 460 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 461 {
 462         return vmci_transport_send_control_pkt(sk,
 463                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 464                                         size, 0, NULL,
 465                                         VSOCK_PROTO_INVALID,
 466                                         VMCI_INVALID_HANDLE);
 467 }
 468
 469 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 470                                              u16 version)
 471 {
 472         return vmci_transport_send_control_pkt(
 473                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 474                                         size, 0, NULL, version,
 475                                         VMCI_INVALID_HANDLE);
 476 }
 477
 478 static struct sock *vmci_transport_get_pending(
 479                                         struct sock *listener,
 480                                         struct vmci_transport_packet *pkt)
 481 {
 482         struct vsock_sock *vlistener;
 483         struct vsock_sock *vpending;
 484         struct sock *pending;
 485         struct sockaddr_vm src;
 486
 487         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 488
 489         vlistener = vsock_sk(listener);
 490
 491         list_for_each_entry(vpending, &vlistener->pending_links,
 492                             pending_links) {
 493                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 494                     pkt->dst_port == vpending->local_addr.svm_port) {
 495                         pending = sk_vsock(vpending);
 496                         sock_hold(pending);
 497                         goto found;
 498                 }
 499         }
 500
 501         pending = NULL;
 502 found:
 503         return pending;
 504
 505 }
 506
 507 static void vmci_transport_release_pending(struct sock *pending)
 508 {
 509         sock_put(pending);
 510 }
 511
 512 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 513  * trusted sockets 2) sockets from applications running as the same user as the
 514  * VM (this is only true for the host side and only when using hosted products)
 515  */
 516
 517 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 518 {
 519         return vsock->trusted ||
 520                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 521 }
 522
 523 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 524  * only if it is trusted as described in vmci_transport_is_trusted.
 525  */
 526
 527 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 528 {
 529         if (VMADDR_CID_HYPERVISOR == peer_cid)
 530                 return true;
 531
 532         if (vsock->cached_peer != peer_cid) {
 533                 vsock->cached_peer = peer_cid;
 534                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 535                     (vmci_context_get_priv_flags(peer_cid) &
 536                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 537                         vsock->cached_peer_allow_dgram = false;
 538                 } else {
 539                         vsock->cached_peer_allow_dgram = true;
 540                 }
 541         }
 542
 543         return vsock->cached_peer_allow_dgram;
 544 }
 545
 546 static int
 547 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 548                                 struct vmci_handle *handle,
 549                                 u64 produce_size,
 550                                 u64 consume_size,
 551                                 u32 peer, u32 flags, bool trusted)
 552 {
 553         int err = 0;
 554
 555         if (trusted) {
 556                 /* Try to allocate our queue pair as trusted. This will only
 557                  * work if vsock is running in the host.
 558                  */
 559
 560                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 561                                        consume_size,
 562                                        peer, flags,
 563                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 564                 if (err != VMCI_ERROR_NO_ACCESS)
 565                         goto out;
 566
 567         }
 568
 569         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 570                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 571 out:
 572         if (err < 0) {
 573                 pr_err_once("Could not attach to queue pair with %d\n", err);
 574                 err = vmci_transport_error_to_vsock_error(err);
 575         }
 576
 577         return err;
 578 }
 579
 580 static int
 581 vmci_transport_datagram_create_hnd(u32 resource_id,
 582                                    u32 flags,
 583                                    vmci_datagram_recv_cb recv_cb,
 584                                    void *client_data,
 585                                    struct vmci_handle *out_handle)
 586 {
 587         int err = 0;
 588
 589         /* Try to allocate our datagram handler as trusted. This will only work
 590          * if vsock is running in the host.
 591          */
 592
 593         err = vmci_datagram_create_handle_priv(resource_id, flags,
 594                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 595                                                recv_cb,
 596                                                client_data, out_handle);
 597
 598         if (err == VMCI_ERROR_NO_ACCESS)
 599                 err = vmci_datagram_create_handle(resource_id, flags,
 600                                                   recv_cb, client_data,
 601                                                   out_handle);
 602
 603         return err;
 604 }
 605
 606 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 607  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 608  * sleep it should defer that work to a work queue.
 609  */
 610
 611 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 612 {
 613         struct sock *sk;
 614         size_t size;
 615         struct sk_buff *skb;
 616         struct vsock_sock *vsk;
 617
 618         sk = (struct sock *)data;
 619
 620         /* This handler is privileged when this module is running on the host.
 621          * We will get datagrams from all endpoints (even VMs that are in a
 622          * restricted context). If we get one from a restricted context then
 623          * the destination socket must be trusted.
 624          *
 625          * NOTE: We access the socket struct without holding the lock here.
 626          * This is ok because the field we are interested is never modified
 627          * outside of the create and destruct socket functions.
 628          */
 629         vsk = vsock_sk(sk);
 630         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 631                 return VMCI_ERROR_NO_ACCESS;
 632
 633         size = VMCI_DG_SIZE(dg);
 634
 635         /* Attach the packet to the socket's receive queue as an sk_buff. */
 636         skb = alloc_skb(size, GFP_ATOMIC);
 637         if (!skb)
 638                 return VMCI_ERROR_NO_MEM;
 639
 640         /* sk_receive_skb() will do a sock_put(), so hold here. */
 641         sock_hold(sk);
 642         skb_put(skb, size);
 643         memcpy(skb->data, dg, size);
 644         sk_receive_skb(sk, skb, 0);
 645
 646         return VMCI_SUCCESS;
 647 }
 648
 649 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 650 {
 651         static const u32 non_socket_contexts[] = {
 652                 VMADDR_CID_LOCAL,
 653         };
 654         int i;
 655
 656         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 657
 658         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 659                 if (cid == non_socket_contexts[i])
 660                         return false;
 661         }
 662
 663         return true;
 664 }
 665
 666 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 667  * interrupt fires.  This is run in bottom-half context but it defers most of
 668  * its work to the packet handling work queue.
 669  */
 670
 671 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 672 {
 673         struct sock *sk;
 674         struct sockaddr_vm dst;
 675         struct sockaddr_vm src;
 676         struct vmci_transport_packet *pkt;
 677         struct vsock_sock *vsk;
 678         bool bh_process_pkt;
 679         int err;
 680
 681         sk = NULL;
 682         err = VMCI_SUCCESS;
 683         bh_process_pkt = false;
 684
 685         /* Ignore incoming packets from contexts without sockets, or resources
 686          * that aren't vsock implementations.
 687          */
 688
 689         if (!vmci_transport_stream_allow(dg->src.context, -1)
 690             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 691                 return VMCI_ERROR_NO_ACCESS;
 692
 693         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 694                 /* Drop datagrams that do not contain full VSock packets. */
 695                 return VMCI_ERROR_INVALID_ARGS;
 696
 697         pkt = (struct vmci_transport_packet *)dg;
 698
 699         /* Find the socket that should handle this packet.  First we look for a
 700          * connected socket and if there is none we look for a socket bound to
 701          * the destintation address.
 702          */
 703         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 704         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 705
 706         sk = vsock_find_connected_socket(&src, &dst);
 707         if (!sk) {
 708                 sk = vsock_find_bound_socket(&dst);
 709                 if (!sk) {
 710                         /* We could not find a socket for this specified
 711                          * address.  If this packet is a RST, we just drop it.
 712                          * If it is another packet, we send a RST.  Note that
 713                          * we do not send a RST reply to RSTs so that we do not
 714                          * continually send RSTs between two endpoints.
 715                          *
 716                          * Note that since this is a reply, dst is src and src
 717                          * is dst.
 718                          */
 719                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 720                                 pr_err("unable to send reset\n");
 721
 722                         err = VMCI_ERROR_NOT_FOUND;
 723                         goto out;
 724                 }
 725         }
 726
 727         /* If the received packet type is beyond all types known to this
 728          * implementation, reply with an invalid message.  Hopefully this will
 729          * help when implementing backwards compatibility in the future.
 730          */
 731         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 732                 vmci_transport_send_invalid_bh(&dst, &src);
 733                 err = VMCI_ERROR_INVALID_ARGS;
 734                 goto out;
 735         }
 736
 737         /* This handler is privileged when this module is running on the host.
 738          * We will get datagram connect requests from all endpoints (even VMs
 739          * that are in a restricted context). If we get one from a restricted
 740          * context then the destination socket must be trusted.
 741          *
 742          * NOTE: We access the socket struct without holding the lock here.
 743          * This is ok because the field we are interested is never modified
 744          * outside of the create and destruct socket functions.
 745          */
 746         vsk = vsock_sk(sk);
 747         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 748                 err = VMCI_ERROR_NO_ACCESS;
 749                 goto out;
 750         }
 751
 752         /* We do most everything in a work queue, but let's fast path the
 753          * notification of reads and writes to help data transfer performance.
 754          * We can only do this if there is no process context code executing
 755          * for this socket since that may change the state.
 756          */
 757         bh_lock_sock(sk);
 758
 759         if (!sock_owned_by_user(sk)) {
 760                 /* The local context ID may be out of date, update it. */
 761                 vsk->local_addr.svm_cid = dst.svm_cid;
 762
 763                 if (sk->sk_state == TCP_ESTABLISHED)
 764                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 765                                         sk, pkt, true, &dst, &src,
 766                                         &bh_process_pkt);
 767         }
 768
 769         bh_unlock_sock(sk);
 770
 771         if (!bh_process_pkt) {
 772                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 773
 774                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 775                 if (!recv_pkt_info) {
 776                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 777                                 pr_err("unable to send reset\n");
 778
 779                         err = VMCI_ERROR_NO_MEM;
 780                         goto out;
 781                 }
 782
 783                 recv_pkt_info->sk = sk;
 784                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 785                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 786
 787                 schedule_work(&recv_pkt_info->work);
 788                 /* Clear sk so that the reference count incremented by one of
 789                  * the Find functions above is not decremented below.  We need
 790                  * that reference count for the packet handler we've scheduled
 791                  * to run.
 792                  */
 793                 sk = NULL;
 794         }
 795
 796 out:
 797         if (sk)
 798                 sock_put(sk);
 799
 800         return err;
 801 }
 802
 803 static void vmci_transport_handle_detach(struct sock *sk)
 804 {
 805         struct vsock_sock *vsk;
 806
 807         vsk = vsock_sk(sk);
 808         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 809                 sock_set_flag(sk, SOCK_DONE);
 810
 811                 /* On a detach the peer will not be sending or receiving
 812                  * anymore.
 813                  */
 814                 vsk->peer_shutdown = SHUTDOWN_MASK;
 815
 816                 /* We should not be sending anymore since the peer won't be
 817                  * there to receive, but we can still receive if there is data
 818                  * left in our consume queue. If the local endpoint is a host,
 819                  * we can't call vsock_stream_has_data, since that may block,
 820                  * but a host endpoint can't read data once the VM has
 821                  * detached, so there is no available data in that case.
 822                  */
 823                 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
 824                     vsock_stream_has_data(vsk) <= 0) {
 825                         if (sk->sk_state == TCP_SYN_SENT) {
 826                                 /* The peer may detach from a queue pair while
 827                                  * we are still in the connecting state, i.e.,
 828                                  * if the peer VM is killed after attaching to
 829                                  * a queue pair, but before we complete the
 830                                  * handshake. In that case, we treat the detach
 831                                  * event like a reset.
 832                                  */
 833
 834                                 sk->sk_state = TCP_CLOSE;
 835                                 sk->sk_err = ECONNRESET;
 836                                 sk->sk_error_report(sk);
 837                                 return;
 838                         }
 839                         sk->sk_state = TCP_CLOSE;
 840                 }
 841                 sk->sk_state_change(sk);
 842         }
 843 }
 844
 845 static void vmci_transport_peer_detach_cb(u32 sub_id,
 846                                           const struct vmci_event_data *e_data,
 847                                           void *client_data)
 848 {
 849         struct vmci_transport *trans = client_data;
 850         const struct vmci_event_payload_qp *e_payload;
 851
 852         e_payload = vmci_event_data_const_payload(e_data);
 853
 854         /* XXX This is lame, we should provide a way to lookup sockets by
 855          * qp_handle.
 856          */
 857         if (vmci_handle_is_invalid(e_payload->handle) ||
 858             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 859                 return;
 860
 861         /* We don't ask for delayed CBs when we subscribe to this event (we
 862          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 863          * guarantees in that case about what context we might be running in,
 864          * so it could be BH or process, blockable or non-blockable.  So we
 865          * need to account for all possible contexts here.
 866          */
 867         spin_lock_bh(&trans->lock);
 868         if (!trans->sk)
 869                 goto out;
 870
 871         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 872          * where trans->sk isn't locked.
 873          */
 874         bh_lock_sock(trans->sk);
 875
 876         vmci_transport_handle_detach(trans->sk);
 877
 878         bh_unlock_sock(trans->sk);
 879  out:
 880         spin_unlock_bh(&trans->lock);
 881 }
 882
 883 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 884                                          const struct vmci_event_data *e_data,
 885                                          void *client_data)
 886 {
 887         vsock_for_each_connected_socket(&vmci_transport,
 888                                         vmci_transport_handle_detach);
 889 }
 890
 891 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 892 {
 893         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 894         struct vmci_transport_packet *pkt;
 895         struct sock *sk;
 896
 897         recv_pkt_info =
 898                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 899         sk = recv_pkt_info->sk;
 900         pkt = &recv_pkt_info->pkt;
 901
 902         lock_sock(sk);
 903
 904         /* The local context ID may be out of date. */
 905         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 906
 907         switch (sk->sk_state) {
 908         case TCP_LISTEN:
 909                 vmci_transport_recv_listen(sk, pkt);
 910                 break;
 911         case TCP_SYN_SENT:
 912                 /* Processing of pending connections for servers goes through
 913                  * the listening socket, so see vmci_transport_recv_listen()
 914                  * for that path.
 915                  */
 916                 vmci_transport_recv_connecting_client(sk, pkt);
 917                 break;
 918         case TCP_ESTABLISHED:
 919                 vmci_transport_recv_connected(sk, pkt);
 920                 break;
 921         default:
 922                 /* Because this function does not run in the same context as
 923                  * vmci_transport_recv_stream_cb it is possible that the
 924                  * socket has closed. We need to let the other side know or it
 925                  * could be sitting in a connect and hang forever. Send a
 926                  * reset to prevent that.
 927                  */
 928                 vmci_transport_send_reset(sk, pkt);
 929                 break;
 930         }
 931
 932         release_sock(sk);
 933         kfree(recv_pkt_info);
 934         /* Release reference obtained in the stream callback when we fetched
 935          * this socket out of the bound or connected list.
 936          */
 937         sock_put(sk);
 938 }
 939
 940 static int vmci_transport_recv_listen(struct sock *sk,
 941                                       struct vmci_transport_packet *pkt)
 942 {
 943         struct sock *pending;
 944         struct vsock_sock *vpending;
 945         int err;
 946         u64 qp_size;
 947         bool old_request = false;
 948         bool old_pkt_proto = false;
 949
 950         err = 0;
 951
 952         /* Because we are in the listen state, we could be receiving a packet
 953          * for ourself or any previous connection requests that we received.
 954          * If it's the latter, we try to find a socket in our list of pending
 955          * connections and, if we do, call the appropriate handler for the
 956          * state that that socket is in.  Otherwise we try to service the
 957          * connection request.
 958          */
 959         pending = vmci_transport_get_pending(sk, pkt);
 960         if (pending) {
 961                 lock_sock(pending);
 962
 963                 /* The local context ID may be out of date. */
 964                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 965
 966                 switch (pending->sk_state) {
 967                 case TCP_SYN_SENT:
 968                         err = vmci_transport_recv_connecting_server(sk,
 969                                                                     pending,
 970                                                                     pkt);
 971                         break;
 972                 default:
 973                         vmci_transport_send_reset(pending, pkt);
 974                         err = -EINVAL;
 975                 }
 976
 977                 if (err < 0)
 978                         vsock_remove_pending(sk, pending);
 979
 980                 release_sock(pending);
 981                 vmci_transport_release_pending(pending);
 982
 983                 return err;
 984         }
 985
 986         /* The listen state only accepts connection requests.  Reply with a
 987          * reset unless we received a reset.
 988          */
 989
 990         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 991               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 992                 vmci_transport_reply_reset(pkt);
 993                 return -EINVAL;
 994         }
 995
 996         if (pkt->u.size == 0) {
 997                 vmci_transport_reply_reset(pkt);
 998                 return -EINVAL;
 999         }
1000
1001         /* If this socket can't accommodate this connection request, we send a
1002          * reset.  Otherwise we create and initialize a child socket and reply
1003          * with a connection negotiation.
1004          */
1005         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1006                 vmci_transport_reply_reset(pkt);
1007                 return -ECONNREFUSED;
1008         }
1009
1010         pending = vsock_create_connected(sk);
1011         if (!pending) {
1012                 vmci_transport_send_reset(sk, pkt);
1013                 return -ENOMEM;
1014         }
1015
1016         vpending = vsock_sk(pending);
1017
1018         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1019                         pkt->dst_port);
1020         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1021                         pkt->src_port);
1022
1023         err = vsock_assign_transport(vpending, vsock_sk(sk));
1024         /* Transport assigned (looking at remote_addr) must be the same
1025          * where we received the request.
1026          */
1027         if (err || !vmci_check_transport(vpending)) {
1028                 vmci_transport_send_reset(sk, pkt);
1029                 sock_put(pending);
1030                 return err;
1031         }
1032
1033         /* If the proposed size fits within our min/max, accept it. Otherwise
1034          * propose our own size.
1035          */
1036         if (pkt->u.size >= vpending->buffer_min_size &&
1037             pkt->u.size <= vpending->buffer_max_size) {
1038                 qp_size = pkt->u.size;
1039         } else {
1040                 qp_size = vpending->buffer_size;
1041         }
1042
1043         /* Figure out if we are using old or new requests based on the
1044          * overrides pkt types sent by our peer.
1045          */
1046         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1047                 old_request = old_pkt_proto;
1048         } else {
1049                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1050                         old_request = true;
1051                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1052                         old_request = false;
1053
1054         }
1055
1056         if (old_request) {
1057                 /* Handle a REQUEST (or override) */
1058                 u16 version = VSOCK_PROTO_INVALID;
1059                 if (vmci_transport_proto_to_notify_struct(
1060                         pending, &version, true))
1061                         err = vmci_transport_send_negotiate(pending, qp_size);
1062                 else
1063                         err = -EINVAL;
1064
1065         } else {
1066                 /* Handle a REQUEST2 (or override) */
1067                 int proto_int = pkt->proto;
1068                 int pos;
1069                 u16 active_proto_version = 0;
1070
1071                 /* The list of possible protocols is the intersection of all
1072                  * protocols the client supports ... plus all the protocols we
1073                  * support.
1074                  */
1075                 proto_int &= vmci_transport_new_proto_supported_versions();
1076
1077                 /* We choose the highest possible protocol version and use that
1078                  * one.
1079                  */
1080                 pos = fls(proto_int);
1081                 if (pos) {
1082                         active_proto_version = (1 << (pos - 1));
1083                         if (vmci_transport_proto_to_notify_struct(
1084                                 pending, &active_proto_version, false))
1085                                 err = vmci_transport_send_negotiate2(pending,
1086                                                         qp_size,
1087                                                         active_proto_version);
1088                         else
1089                                 err = -EINVAL;
1090
1091                 } else {
1092                         err = -EINVAL;
1093                 }
1094         }
1095
1096         if (err < 0) {
1097                 vmci_transport_send_reset(sk, pkt);
1098                 sock_put(pending);
1099                 err = vmci_transport_error_to_vsock_error(err);
1100                 goto out;
1101         }
1102
1103         vsock_add_pending(sk, pending);
1104         sk_acceptq_added(sk);
1105
1106         pending->sk_state = TCP_SYN_SENT;
1107         vmci_trans(vpending)->produce_size =
1108                 vmci_trans(vpending)->consume_size = qp_size;
1109         vpending->buffer_size = qp_size;
1110
1111         vmci_trans(vpending)->notify_ops->process_request(pending);
1112
1113         /* We might never receive another message for this socket and it's not
1114          * connected to any process, so we have to ensure it gets cleaned up
1115          * ourself.  Our delayed work function will take care of that.  Note
1116          * that we do not ever cancel this function since we have few
1117          * guarantees about its state when calling cancel_delayed_work().
1118          * Instead we hold a reference on the socket for that function and make
1119          * it capable of handling cases where it needs to do nothing but
1120          * release that reference.
1121          */
1122         vpending->listener = sk;
1123         sock_hold(sk);
1124         sock_hold(pending);
1125         schedule_delayed_work(&vpending->pending_work, HZ);
1126
1127 out:
1128         return err;
1129 }
1130
1131 static int
1132 vmci_transport_recv_connecting_server(struct sock *listener,
1133                                       struct sock *pending,
1134                                       struct vmci_transport_packet *pkt)
1135 {
1136         struct vsock_sock *vpending;
1137         struct vmci_handle handle;
1138         struct vmci_qp *qpair;
1139         bool is_local;
1140         u32 flags;
1141         u32 detach_sub_id;
1142         int err;
1143         int skerr;
1144
1145         vpending = vsock_sk(pending);
1146         detach_sub_id = VMCI_INVALID_ID;
1147
1148         switch (pkt->type) {
1149         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1150                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1151                         vmci_transport_send_reset(pending, pkt);
1152                         skerr = EPROTO;
1153                         err = -EINVAL;
1154                         goto destroy;
1155                 }
1156                 break;
1157         default:
1158                 /* Close and cleanup the connection. */
1159                 vmci_transport_send_reset(pending, pkt);
1160                 skerr = EPROTO;
1161                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1162                 goto destroy;
1163         }
1164
1165         /* In order to complete the connection we need to attach to the offered
1166          * queue pair and send an attach notification.  We also subscribe to the
1167          * detach event so we know when our peer goes away, and we do that
1168          * before attaching so we don't miss an event.  If all this succeeds,
1169          * we update our state and wakeup anything waiting in accept() for a
1170          * connection.
1171          */
1172
1173         /* We don't care about attach since we ensure the other side has
1174          * attached by specifying the ATTACH_ONLY flag below.
1175          */
1176         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1177                                    vmci_transport_peer_detach_cb,
1178                                    vmci_trans(vpending), &detach_sub_id);
1179         if (err < VMCI_SUCCESS) {
1180                 vmci_transport_send_reset(pending, pkt);
1181                 err = vmci_transport_error_to_vsock_error(err);
1182                 skerr = -err;
1183                 goto destroy;
1184         }
1185
1186         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1187
1188         /* Now attach to the queue pair the client created. */
1189         handle = pkt->u.handle;
1190
1191         /* vpending->local_addr always has a context id so we do not need to
1192          * worry about VMADDR_CID_ANY in this case.
1193          */
1194         is_local =
1195             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1196         flags = VMCI_QPFLAG_ATTACH_ONLY;
1197         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1198
1199         err = vmci_transport_queue_pair_alloc(
1200                                         &qpair,
1201                                         &handle,
1202                                         vmci_trans(vpending)->produce_size,
1203                                         vmci_trans(vpending)->consume_size,
1204                                         pkt->dg.src.context,
1205                                         flags,
1206                                         vmci_transport_is_trusted(
1207                                                 vpending,
1208                                                 vpending->remote_addr.svm_cid));
1209         if (err < 0) {
1210                 vmci_transport_send_reset(pending, pkt);
1211                 skerr = -err;
1212                 goto destroy;
1213         }
1214
1215         vmci_trans(vpending)->qp_handle = handle;
1216         vmci_trans(vpending)->qpair = qpair;
1217
1218         /* When we send the attach message, we must be ready to handle incoming
1219          * control messages on the newly connected socket. So we move the
1220          * pending socket to the connected state before sending the attach
1221          * message. Otherwise, an incoming packet triggered by the attach being
1222          * received by the peer may be processed concurrently with what happens
1223          * below after sending the attach message, and that incoming packet
1224          * will find the listening socket instead of the (currently) pending
1225          * socket. Note that enqueueing the socket increments the reference
1226          * count, so even if a reset comes before the connection is accepted,
1227          * the socket will be valid until it is removed from the queue.
1228          *
1229          * If we fail sending the attach below, we remove the socket from the
1230          * connected list and move the socket to TCP_CLOSE before
1231          * releasing the lock, so a pending slow path processing of an incoming
1232          * packet will not see the socket in the connected state in that case.
1233          */
1234         pending->sk_state = TCP_ESTABLISHED;
1235
1236         vsock_insert_connected(vpending);
1237
1238         /* Notify our peer of our attach. */
1239         err = vmci_transport_send_attach(pending, handle);
1240         if (err < 0) {
1241                 vsock_remove_connected(vpending);
1242                 pr_err("Could not send attach\n");
1243                 vmci_transport_send_reset(pending, pkt);
1244                 err = vmci_transport_error_to_vsock_error(err);
1245                 skerr = -err;
1246                 goto destroy;
1247         }
1248
1249         /* We have a connection. Move the now connected socket from the
1250          * listener's pending list to the accept queue so callers of accept()
1251          * can find it.
1252          */
1253         vsock_remove_pending(listener, pending);
1254         vsock_enqueue_accept(listener, pending);
1255
1256         /* Callers of accept() will be be waiting on the listening socket, not
1257          * the pending socket.
1258          */
1259         listener->sk_data_ready(listener);
1260
1261         return 0;
1262
1263 destroy:
1264         pending->sk_err = skerr;
1265         pending->sk_state = TCP_CLOSE;
1266         /* As long as we drop our reference, all necessary cleanup will handle
1267          * when the cleanup function drops its reference and our destruct
1268          * implementation is called.  Note that since the listen handler will
1269          * remove pending from the pending list upon our failure, the cleanup
1270          * function won't drop the additional reference, which is why we do it
1271          * here.
1272          */
1273         sock_put(pending);
1274
1275         return err;
1276 }
1277
1278 static int
1279 vmci_transport_recv_connecting_client(struct sock *sk,
1280                                       struct vmci_transport_packet *pkt)
1281 {
1282         struct vsock_sock *vsk;
1283         int err;
1284         int skerr;
1285
1286         vsk = vsock_sk(sk);
1287
1288         switch (pkt->type) {
1289         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1290                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1291                     !vmci_handle_is_equal(pkt->u.handle,
1292                                           vmci_trans(vsk)->qp_handle)) {
1293                         skerr = EPROTO;
1294                         err = -EINVAL;
1295                         goto destroy;
1296                 }
1297
1298                 /* Signify the socket is connected and wakeup the waiter in
1299                  * connect(). Also place the socket in the connected table for
1300                  * accounting (it can already be found since it's in the bound
1301                  * table).
1302                  */
1303                 sk->sk_state = TCP_ESTABLISHED;
1304                 sk->sk_socket->state = SS_CONNECTED;
1305                 vsock_insert_connected(vsk);
1306                 sk->sk_state_change(sk);
1307
1308                 break;
1309         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1310         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1311                 if (pkt->u.size == 0
1312                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1313                     || pkt->src_port != vsk->remote_addr.svm_port
1314                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1315                     || vmci_trans(vsk)->qpair
1316                     || vmci_trans(vsk)->produce_size != 0
1317                     || vmci_trans(vsk)->consume_size != 0
1318                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1319                         skerr = EPROTO;
1320                         err = -EINVAL;
1321
1322                         goto destroy;
1323                 }
1324
1325                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1326                 if (err) {
1327                         skerr = -err;
1328                         goto destroy;
1329                 }
1330
1331                 break;
1332         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1333                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1334                 if (err) {
1335                         skerr = -err;
1336                         goto destroy;
1337                 }
1338
1339                 break;
1340         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1341                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1342                  * continue processing here after they sent an INVALID packet.
1343                  * This meant that we got a RST after the INVALID. We ignore a
1344                  * RST after an INVALID. The common code doesn't send the RST
1345                  * ... so we can hang if an old version of the common code
1346                  * fails between getting a REQUEST and sending an OFFER back.
1347                  * Not much we can do about it... except hope that it doesn't
1348                  * happen.
1349                  */
1350                 if (vsk->ignore_connecting_rst) {
1351                         vsk->ignore_connecting_rst = false;
1352                 } else {
1353                         skerr = ECONNRESET;
1354                         err = 0;
1355                         goto destroy;
1356                 }
1357
1358                 break;
1359         default:
1360                 /* Close and cleanup the connection. */
1361                 skerr = EPROTO;
1362                 err = -EINVAL;
1363                 goto destroy;
1364         }
1365
1366         return 0;
1367
1368 destroy:
1369         vmci_transport_send_reset(sk, pkt);
1370
1371         sk->sk_state = TCP_CLOSE;
1372         sk->sk_err = skerr;
1373         sk->sk_error_report(sk);
1374         return err;
1375 }
1376
1377 static int vmci_transport_recv_connecting_client_negotiate(
1378                                         struct sock *sk,
1379                                         struct vmci_transport_packet *pkt)
1380 {
1381         int err;
1382         struct vsock_sock *vsk;
1383         struct vmci_handle handle;
1384         struct vmci_qp *qpair;
1385         u32 detach_sub_id;
1386         bool is_local;
1387         u32 flags;
1388         bool old_proto = true;
1389         bool old_pkt_proto;
1390         u16 version;
1391
1392         vsk = vsock_sk(sk);
1393         handle = VMCI_INVALID_HANDLE;
1394         detach_sub_id = VMCI_INVALID_ID;
1395
1396         /* If we have gotten here then we should be past the point where old
1397          * linux vsock could have sent the bogus rst.
1398          */
1399         vsk->sent_request = false;
1400         vsk->ignore_connecting_rst = false;
1401
1402         /* Verify that we're OK with the proposed queue pair size */
1403         if (pkt->u.size < vsk->buffer_min_size ||
1404             pkt->u.size > vsk->buffer_max_size) {
1405                 err = -EINVAL;
1406                 goto destroy;
1407         }
1408
1409         /* At this point we know the CID the peer is using to talk to us. */
1410
1411         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1412                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1413
1414         /* Setup the notify ops to be the highest supported version that both
1415          * the server and the client support.
1416          */
1417
1418         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1419                 old_proto = old_pkt_proto;
1420         } else {
1421                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1422                         old_proto = true;
1423                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1424                         old_proto = false;
1425
1426         }
1427
1428         if (old_proto)
1429                 version = VSOCK_PROTO_INVALID;
1430         else
1431                 version = pkt->proto;
1432
1433         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1434                 err = -EINVAL;
1435                 goto destroy;
1436         }
1437
1438         /* Subscribe to detach events first.
1439          *
1440          * XXX We attach once for each queue pair created for now so it is easy
1441          * to find the socket (it's provided), but later we should only
1442          * subscribe once and add a way to lookup sockets by queue pair handle.
1443          */
1444         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1445                                    vmci_transport_peer_detach_cb,
1446                                    vmci_trans(vsk), &detach_sub_id);
1447         if (err < VMCI_SUCCESS) {
1448                 err = vmci_transport_error_to_vsock_error(err);
1449                 goto destroy;
1450         }
1451
1452         /* Make VMCI select the handle for us. */
1453         handle = VMCI_INVALID_HANDLE;
1454         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1455         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1456
1457         err = vmci_transport_queue_pair_alloc(&qpair,
1458                                               &handle,
1459                                               pkt->u.size,
1460                                               pkt->u.size,
1461                                               vsk->remote_addr.svm_cid,
1462                                               flags,
1463                                               vmci_transport_is_trusted(
1464                                                   vsk,
1465                                                   vsk->
1466                                                   remote_addr.svm_cid));
1467         if (err < 0)
1468                 goto destroy;
1469
1470         err = vmci_transport_send_qp_offer(sk, handle);
1471         if (err < 0) {
1472                 err = vmci_transport_error_to_vsock_error(err);
1473                 goto destroy;
1474         }
1475
1476         vmci_trans(vsk)->qp_handle = handle;
1477         vmci_trans(vsk)->qpair = qpair;
1478
1479         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1480                 pkt->u.size;
1481
1482         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1483
1484         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1485
1486         return 0;
1487
1488 destroy:
1489         if (detach_sub_id != VMCI_INVALID_ID)
1490                 vmci_event_unsubscribe(detach_sub_id);
1491
1492         if (!vmci_handle_is_invalid(handle))
1493                 vmci_qpair_detach(&qpair);
1494
1495         return err;
1496 }
1497
1498 static int
1499 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1500                                               struct vmci_transport_packet *pkt)
1501 {
1502         int err = 0;
1503         struct vsock_sock *vsk = vsock_sk(sk);
1504
1505         if (vsk->sent_request) {
1506                 vsk->sent_request = false;
1507                 vsk->ignore_connecting_rst = true;
1508
1509                 err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1510                 if (err < 0)
1511                         err = vmci_transport_error_to_vsock_error(err);
1512                 else
1513                         err = 0;
1514
1515         }
1516
1517         return err;
1518 }
1519
1520 static int vmci_transport_recv_connected(struct sock *sk,
1521                                          struct vmci_transport_packet *pkt)
1522 {
1523         struct vsock_sock *vsk;
1524         bool pkt_processed = false;
1525
1526         /* In cases where we are closing the connection, it's sufficient to
1527          * mark the state change (and maybe error) and wake up any waiting
1528          * threads. Since this is a connected socket, it's owned by a user
1529          * process and will be cleaned up when the failure is passed back on
1530          * the current or next system call.  Our system call implementations
1531          * must therefore check for error and state changes on entry and when
1532          * being awoken.
1533          */
1534         switch (pkt->type) {
1535         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1536                 if (pkt->u.mode) {
1537                         vsk = vsock_sk(sk);
1538
1539                         vsk->peer_shutdown |= pkt->u.mode;
1540                         sk->sk_state_change(sk);
1541                 }
1542                 break;
1543
1544         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1545                 vsk = vsock_sk(sk);
1546                 /* It is possible that we sent our peer a message (e.g a
1547                  * WAITING_READ) right before we got notified that the peer had
1548                  * detached. If that happens then we can get a RST pkt back
1549                  * from our peer even though there is data available for us to
1550                  * read. In that case, don't shutdown the socket completely but
1551                  * instead allow the local client to finish reading data off
1552                  * the queuepair. Always treat a RST pkt in connected mode like
1553                  * a clean shutdown.
1554                  */
1555                 sock_set_flag(sk, SOCK_DONE);
1556                 vsk->peer_shutdown = SHUTDOWN_MASK;
1557                 if (vsock_stream_has_data(vsk) <= 0)
1558                         sk->sk_state = TCP_CLOSING;
1559
1560                 sk->sk_state_change(sk);
1561                 break;
1562
1563         default:
1564                 vsk = vsock_sk(sk);
1565                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1566                                 sk, pkt, false, NULL, NULL,
1567                                 &pkt_processed);
1568                 if (!pkt_processed)
1569                         return -EINVAL;
1570
1571                 break;
1572         }
1573
1574         return 0;
1575 }
1576
1577 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1578                                       struct vsock_sock *psk)
1579 {
1580         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1581         if (!vsk->trans)
1582                 return -ENOMEM;
1583
1584         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1585         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1586         vmci_trans(vsk)->qpair = NULL;
1587         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1588         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1589         vmci_trans(vsk)->notify_ops = NULL;
1590         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1591         vmci_trans(vsk)->sk = &vsk->sk;
1592         spin_lock_init(&vmci_trans(vsk)->lock);
1593
1594         return 0;
1595 }
1596
1597 static void vmci_transport_free_resources(struct list_head *transport_list)
1598 {
1599         while (!list_empty(transport_list)) {
1600                 struct vmci_transport *transport =
1601                     list_first_entry(transport_list, struct vmci_transport,
1602                                      elem);
1603                 list_del(&transport->elem);
1604
1605                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1606                         vmci_event_unsubscribe(transport->detach_sub_id);
1607                         transport->detach_sub_id = VMCI_INVALID_ID;
1608                 }
1609
1610                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1611                         vmci_qpair_detach(&transport->qpair);
1612                         transport->qp_handle = VMCI_INVALID_HANDLE;
1613                         transport->produce_size = 0;
1614                         transport->consume_size = 0;
1615                 }
1616
1617                 kfree(transport);
1618         }
1619 }
1620
1621 static void vmci_transport_cleanup(struct work_struct *work)
1622 {
1623         LIST_HEAD(pending);
1624
1625         spin_lock_bh(&vmci_transport_cleanup_lock);
1626         list_replace_init(&vmci_transport_cleanup_list, &pending);
1627         spin_unlock_bh(&vmci_transport_cleanup_lock);
1628         vmci_transport_free_resources(&pending);
1629 }
1630
1631 static void vmci_transport_destruct(struct vsock_sock *vsk)
1632 {
1633         /* transport can be NULL if we hit a failure at init() time */
1634         if (!vmci_trans(vsk))
1635                 return;
1636
1637         /* Ensure that the detach callback doesn't use the sk/vsk
1638          * we are about to destruct.
1639          */
1640         spin_lock_bh(&vmci_trans(vsk)->lock);
1641         vmci_trans(vsk)->sk = NULL;
1642         spin_unlock_bh(&vmci_trans(vsk)->lock);
1643
1644         if (vmci_trans(vsk)->notify_ops)
1645                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1646
1647         spin_lock_bh(&vmci_transport_cleanup_lock);
1648         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1649         spin_unlock_bh(&vmci_transport_cleanup_lock);
1650         schedule_work(&vmci_transport_cleanup_work);
1651
1652         vsk->trans = NULL;
1653 }
1654
1655 static void vmci_transport_release(struct vsock_sock *vsk)
1656 {
1657         vsock_remove_sock(vsk);
1658
1659         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1660                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1661                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1662         }
1663 }
1664
1665 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1666                                      struct sockaddr_vm *addr)
1667 {
1668         u32 port;
1669         u32 flags;
1670         int err;
1671
1672         /* VMCI will select a resource ID for us if we provide
1673          * VMCI_INVALID_ID.
1674          */
1675         port = addr->svm_port == VMADDR_PORT_ANY ?
1676                         VMCI_INVALID_ID : addr->svm_port;
1677
1678         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1679                 return -EACCES;
1680
1681         flags = addr->svm_cid == VMADDR_CID_ANY ?
1682                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1683
1684         err = vmci_transport_datagram_create_hnd(port, flags,
1685                                                  vmci_transport_recv_dgram_cb,
1686                                                  &vsk->sk,
1687                                                  &vmci_trans(vsk)->dg_handle);
1688         if (err < VMCI_SUCCESS)
1689                 return vmci_transport_error_to_vsock_error(err);
1690         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1691                         vmci_trans(vsk)->dg_handle.resource);
1692
1693         return 0;
1694 }
1695
1696 static int vmci_transport_dgram_enqueue(
1697         struct vsock_sock *vsk,
1698         struct sockaddr_vm *remote_addr,
1699         struct msghdr *msg,
1700         size_t len)
1701 {
1702         int err;
1703         struct vmci_datagram *dg;
1704
1705         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1706                 return -EMSGSIZE;
1707
1708         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1709                 return -EPERM;
1710
1711         /* Allocate a buffer for the user's message and our packet header. */
1712         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1713         if (!dg)
1714                 return -ENOMEM;
1715
1716         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1717
1718         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1719                                    remote_addr->svm_port);
1720         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1721                                    vsk->local_addr.svm_port);
1722         dg->payload_size = len;
1723
1724         err = vmci_datagram_send(dg);
1725         kfree(dg);
1726         if (err < 0)
1727                 return vmci_transport_error_to_vsock_error(err);
1728
1729         return err - sizeof(*dg);
1730 }
1731
1732 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1733                                         struct msghdr *msg, size_t len,
1734                                         int flags)
1735 {
1736         int err;
1737         int noblock;
1738         struct vmci_datagram *dg;
1739         size_t payload_len;
1740         struct sk_buff *skb;
1741
1742         noblock = flags & MSG_DONTWAIT;
1743
1744         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1745                 return -EOPNOTSUPP;
1746
1747         /* Retrieve the head sk_buff from the socket's receive queue. */
1748         err = 0;
1749         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1750         if (!skb)
1751                 return err;
1752
1753         dg = (struct vmci_datagram *)skb->data;
1754         if (!dg)
1755                 /* err is 0, meaning we read zero bytes. */
1756                 goto out;
1757
1758         payload_len = dg->payload_size;
1759         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1760         if (payload_len != skb->len - sizeof(*dg)) {
1761                 err = -EINVAL;
1762                 goto out;
1763         }
1764
1765         if (payload_len > len) {
1766                 payload_len = len;
1767                 msg->msg_flags |= MSG_TRUNC;
1768         }
1769
1770         /* Place the datagram payload in the user's iovec. */
1771         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1772         if (err)
1773                 goto out;
1774
1775         if (msg->msg_name) {
1776                 /* Provide the address of the sender. */
1777                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1778                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1779                 msg->msg_namelen = sizeof(*vm_addr);
1780         }
1781         err = payload_len;
1782
1783 out:
1784         skb_free_datagram(&vsk->sk, skb);
1785         return err;
1786 }
1787
1788 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1789 {
1790         if (cid == VMADDR_CID_HYPERVISOR) {
1791                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1792                  * state and are allowed.
1793                  */
1794                 return port == VMCI_UNITY_PBRPC_REGISTER;
1795         }
1796
1797         return true;
1798 }
1799
1800 static int vmci_transport_connect(struct vsock_sock *vsk)
1801 {
1802         int err;
1803         bool old_pkt_proto = false;
1804         struct sock *sk = &vsk->sk;
1805
1806         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1807                 old_pkt_proto) {
1808                 err = vmci_transport_send_conn_request(sk, vsk->buffer_size);
1809                 if (err < 0) {
1810                         sk->sk_state = TCP_CLOSE;
1811                         return err;
1812                 }
1813         } else {
1814                 int supported_proto_versions =
1815                         vmci_transport_new_proto_supported_versions();
1816                 err = vmci_transport_send_conn_request2(sk, vsk->buffer_size,
1817                                 supported_proto_versions);
1818                 if (err < 0) {
1819                         sk->sk_state = TCP_CLOSE;
1820                         return err;
1821                 }
1822
1823                 vsk->sent_request = true;
1824         }
1825
1826         return err;
1827 }
1828
1829 static ssize_t vmci_transport_stream_dequeue(
1830         struct vsock_sock *vsk,
1831         struct msghdr *msg,
1832         size_t len,
1833         int flags)
1834 {
1835         if (flags & MSG_PEEK)
1836                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1837         else
1838                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1839 }
1840
1841 static ssize_t vmci_transport_stream_enqueue(
1842         struct vsock_sock *vsk,
1843         struct msghdr *msg,
1844         size_t len)
1845 {
1846         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1847 }
1848
1849 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1850 {
1851         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1852 }
1853
1854 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1855 {
1856         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1857 }
1858
1859 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1860 {
1861         return vmci_trans(vsk)->consume_size;
1862 }
1863
1864 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1865 {
1866         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1867 }
1868
1869 static int vmci_transport_notify_poll_in(
1870         struct vsock_sock *vsk,
1871         size_t target,
1872         bool *data_ready_now)
1873 {
1874         return vmci_trans(vsk)->notify_ops->poll_in(
1875                         &vsk->sk, target, data_ready_now);
1876 }
1877
1878 static int vmci_transport_notify_poll_out(
1879         struct vsock_sock *vsk,
1880         size_t target,
1881         bool *space_available_now)
1882 {
1883         return vmci_trans(vsk)->notify_ops->poll_out(
1884                         &vsk->sk, target, space_available_now);
1885 }
1886
1887 static int vmci_transport_notify_recv_init(
1888         struct vsock_sock *vsk,
1889         size_t target,
1890         struct vsock_transport_recv_notify_data *data)
1891 {
1892         return vmci_trans(vsk)->notify_ops->recv_init(
1893                         &vsk->sk, target,
1894                         (struct vmci_transport_recv_notify_data *)data);
1895 }
1896
1897 static int vmci_transport_notify_recv_pre_block(
1898         struct vsock_sock *vsk,
1899         size_t target,
1900         struct vsock_transport_recv_notify_data *data)
1901 {
1902         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1903                         &vsk->sk, target,
1904                         (struct vmci_transport_recv_notify_data *)data);
1905 }
1906
1907 static int vmci_transport_notify_recv_pre_dequeue(
1908         struct vsock_sock *vsk,
1909         size_t target,
1910         struct vsock_transport_recv_notify_data *data)
1911 {
1912         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1913                         &vsk->sk, target,
1914                         (struct vmci_transport_recv_notify_data *)data);
1915 }
1916
1917 static int vmci_transport_notify_recv_post_dequeue(
1918         struct vsock_sock *vsk,
1919         size_t target,
1920         ssize_t copied,
1921         bool data_read,
1922         struct vsock_transport_recv_notify_data *data)
1923 {
1924         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1925                         &vsk->sk, target, copied, data_read,
1926                         (struct vmci_transport_recv_notify_data *)data);
1927 }
1928
1929 static int vmci_transport_notify_send_init(
1930         struct vsock_sock *vsk,
1931         struct vsock_transport_send_notify_data *data)
1932 {
1933         return vmci_trans(vsk)->notify_ops->send_init(
1934                         &vsk->sk,
1935                         (struct vmci_transport_send_notify_data *)data);
1936 }
1937
1938 static int vmci_transport_notify_send_pre_block(
1939         struct vsock_sock *vsk,
1940         struct vsock_transport_send_notify_data *data)
1941 {
1942         return vmci_trans(vsk)->notify_ops->send_pre_block(
1943                         &vsk->sk,
1944                         (struct vmci_transport_send_notify_data *)data);
1945 }
1946
1947 static int vmci_transport_notify_send_pre_enqueue(
1948         struct vsock_sock *vsk,
1949         struct vsock_transport_send_notify_data *data)
1950 {
1951         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1952                         &vsk->sk,
1953                         (struct vmci_transport_send_notify_data *)data);
1954 }
1955
1956 static int vmci_transport_notify_send_post_enqueue(
1957         struct vsock_sock *vsk,
1958         ssize_t written,
1959         struct vsock_transport_send_notify_data *data)
1960 {
1961         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1962                         &vsk->sk, written,
1963                         (struct vmci_transport_send_notify_data *)data);
1964 }
1965
1966 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1967 {
1968         if (PROTOCOL_OVERRIDE != -1) {
1969                 if (PROTOCOL_OVERRIDE == 0)
1970                         *old_pkt_proto = true;
1971                 else
1972                         *old_pkt_proto = false;
1973
1974                 pr_info("Proto override in use\n");
1975                 return true;
1976         }
1977
1978         return false;
1979 }
1980
1981 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
1982                                                   u16 *proto,
1983                                                   bool old_pkt_proto)
1984 {
1985         struct vsock_sock *vsk = vsock_sk(sk);
1986
1987         if (old_pkt_proto) {
1988                 if (*proto != VSOCK_PROTO_INVALID) {
1989                         pr_err("Can't set both an old and new protocol\n");
1990                         return false;
1991                 }
1992                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
1993                 goto exit;
1994         }
1995
1996         switch (*proto) {
1997         case VSOCK_PROTO_PKT_ON_NOTIFY:
1998                 vmci_trans(vsk)->notify_ops =
1999                         &vmci_transport_notify_pkt_q_state_ops;
2000                 break;
2001         default:
2002                 pr_err("Unknown notify protocol version\n");
2003                 return false;
2004         }
2005
2006 exit:
2007         vmci_trans(vsk)->notify_ops->socket_init(sk);
2008         return true;
2009 }
2010
2011 static u16 vmci_transport_new_proto_supported_versions(void)
2012 {
2013         if (PROTOCOL_OVERRIDE != -1)
2014                 return PROTOCOL_OVERRIDE;
2015
2016         return VSOCK_PROTO_ALL_SUPPORTED;
2017 }
2018
2019 static u32 vmci_transport_get_local_cid(void)
2020 {
2021         return vmci_get_context_id();
2022 }
2023
2024 static struct vsock_transport vmci_transport = {
2025         .module = THIS_MODULE,
2026         .init = vmci_transport_socket_init,
2027         .destruct = vmci_transport_destruct,
2028         .release = vmci_transport_release,
2029         .connect = vmci_transport_connect,
2030         .dgram_bind = vmci_transport_dgram_bind,
2031         .dgram_dequeue = vmci_transport_dgram_dequeue,
2032         .dgram_enqueue = vmci_transport_dgram_enqueue,
2033         .dgram_allow = vmci_transport_dgram_allow,
2034         .stream_dequeue = vmci_transport_stream_dequeue,
2035         .stream_enqueue = vmci_transport_stream_enqueue,
2036         .stream_has_data = vmci_transport_stream_has_data,
2037         .stream_has_space = vmci_transport_stream_has_space,
2038         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2039         .stream_is_active = vmci_transport_stream_is_active,
2040         .stream_allow = vmci_transport_stream_allow,
2041         .notify_poll_in = vmci_transport_notify_poll_in,
2042         .notify_poll_out = vmci_transport_notify_poll_out,
2043         .notify_recv_init = vmci_transport_notify_recv_init,
2044         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2045         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2046         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2047         .notify_send_init = vmci_transport_notify_send_init,
2048         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2049         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2050         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2051         .shutdown = vmci_transport_shutdown,
2052         .get_local_cid = vmci_transport_get_local_cid,
2053 };
2054
2055 static bool vmci_check_transport(struct vsock_sock *vsk)
2056 {
2057         return vsk->transport == &vmci_transport;
2058 }
2059
2060 static void vmci_vsock_transport_cb(bool is_host)
2061 {
2062         int features;
2063
2064         if (is_host)
2065                 features = VSOCK_TRANSPORT_F_H2G;
2066         else
2067                 features = VSOCK_TRANSPORT_F_G2H;
2068
2069         vsock_core_register(&vmci_transport, features);
2070 }
2071
2072 static int __init vmci_transport_init(void)
2073 {
2074         int err;
2075
2076         /* Create the datagram handle that we will use to send and receive all
2077          * VSocket control messages for this context.
2078          */
2079         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2080                                                  VMCI_FLAG_ANYCID_DG_HND,
2081                                                  vmci_transport_recv_stream_cb,
2082                                                  NULL,
2083                                                  &vmci_transport_stream_handle);
2084         if (err < VMCI_SUCCESS) {
2085                 pr_err("Unable to create datagram handle. (%d)\n", err);
2086                 return vmci_transport_error_to_vsock_error(err);
2087         }
2088         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2089                                    vmci_transport_qp_resumed_cb,
2090                                    NULL, &vmci_transport_qp_resumed_sub_id);
2091         if (err < VMCI_SUCCESS) {
2092                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2093                 err = vmci_transport_error_to_vsock_error(err);
2094                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2095                 goto err_destroy_stream_handle;
2096         }
2097
2098         /* Register only with dgram feature, other features (H2G, G2H) will be
2099          * registered when the first host or guest becomes active.
2100          */
2101         err = vsock_core_register(&vmci_transport, VSOCK_TRANSPORT_F_DGRAM);
2102         if (err < 0)
2103                 goto err_unsubscribe;
2104
2105         err = vmci_register_vsock_callback(vmci_vsock_transport_cb);
2106         if (err < 0)
2107                 goto err_unregister;
2108
2109         return 0;
2110
2111 err_unregister:
2112         vsock_core_unregister(&vmci_transport);
2113 err_unsubscribe:
2114         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2115 err_destroy_stream_handle:
2116         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2117         return err;
2118 }
2119 module_init(vmci_transport_init);
2120
2121 static void __exit vmci_transport_exit(void)
2122 {
2123         cancel_work_sync(&vmci_transport_cleanup_work);
2124         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2125
2126         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2127                 if (vmci_datagram_destroy_handle(
2128                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2129                         pr_err("Couldn't destroy datagram handle\n");
2130                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2131         }
2132
2133         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2134                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2135                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2136         }
2137
2138         vmci_register_vsock_callback(NULL);
2139         vsock_core_unregister(&vmci_transport);
2140 }
2141 module_exit(vmci_transport_exit);
2142
2143 MODULE_AUTHOR("VMware, Inc.");
2144 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2145 MODULE_VERSION("1.0.5.0-k");
2146 MODULE_LICENSE("GPL v2");
2147 MODULE_ALIAS("vmware_vsock");
2148 MODULE_ALIAS_NETPROTO(PF_VSOCK);