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
  61 struct vmci_transport_recv_pkt_info {
  62         struct work_struct work;
  63         struct sock *sk;
  64         struct vmci_transport_packet pkt;
  65 };
  66
  67 static LIST_HEAD(vmci_transport_cleanup_list);
  68 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
  69 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
  70
  71 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  72                                                            VMCI_INVALID_ID };
  73 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  74
  75 static int PROTOCOL_OVERRIDE = -1;
  76
  77 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN   128
  78 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE       262144
  79 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX   262144
  80
  81 /* Helper function to convert from a VMCI error code to a VSock error code. */
  82
  83 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  84 {
  85         switch (vmci_error) {
  86         case VMCI_ERROR_NO_MEM:
  87                 return -ENOMEM;
  88         case VMCI_ERROR_DUPLICATE_ENTRY:
  89         case VMCI_ERROR_ALREADY_EXISTS:
  90                 return -EADDRINUSE;
  91         case VMCI_ERROR_NO_ACCESS:
  92                 return -EPERM;
  93         case VMCI_ERROR_NO_RESOURCES:
  94                 return -ENOBUFS;
  95         case VMCI_ERROR_INVALID_RESOURCE:
  96                 return -EHOSTUNREACH;
  97         case VMCI_ERROR_INVALID_ARGS:
  98         default:
  99                 break;
 100         }
 101         return -EINVAL;
 102 }
 103
 104 static u32 vmci_transport_peer_rid(u32 peer_cid)
 105 {
 106         if (VMADDR_CID_HYPERVISOR == peer_cid)
 107                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 108
 109         return VMCI_TRANSPORT_PACKET_RID;
 110 }
 111
 112 static inline void
 113 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 114                            struct sockaddr_vm *src,
 115                            struct sockaddr_vm *dst,
 116                            u8 type,
 117                            u64 size,
 118                            u64 mode,
 119                            struct vmci_transport_waiting_info *wait,
 120                            u16 proto,
 121                            struct vmci_handle handle)
 122 {
 123         /* We register the stream control handler as an any cid handle so we
 124          * must always send from a source address of VMADDR_CID_ANY
 125          */
 126         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 127                                        VMCI_TRANSPORT_PACKET_RID);
 128         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 129                                        vmci_transport_peer_rid(dst->svm_cid));
 130         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 131         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 132         pkt->type = type;
 133         pkt->src_port = src->svm_port;
 134         pkt->dst_port = dst->svm_port;
 135         memset(&pkt->proto, 0, sizeof(pkt->proto));
 136         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 137
 138         switch (pkt->type) {
 139         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 140                 pkt->u.size = 0;
 141                 break;
 142
 143         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 144         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 145                 pkt->u.size = size;
 146                 break;
 147
 148         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 149         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 150                 pkt->u.handle = handle;
 151                 break;
 152
 153         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 154         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 155         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 156                 pkt->u.size = 0;
 157                 break;
 158
 159         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 160                 pkt->u.mode = mode;
 161                 break;
 162
 163         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 164         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 165                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 166                 break;
 167
 168         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 169         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 170                 pkt->u.size = size;
 171                 pkt->proto = proto;
 172                 break;
 173         }
 174 }
 175
 176 static inline void
 177 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 178                                     struct sockaddr_vm *local,
 179                                     struct sockaddr_vm *remote)
 180 {
 181         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 182         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 183 }
 184
 185 static int
 186 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 187                                   struct sockaddr_vm *src,
 188                                   struct sockaddr_vm *dst,
 189                                   enum vmci_transport_packet_type type,
 190                                   u64 size,
 191                                   u64 mode,
 192                                   struct vmci_transport_waiting_info *wait,
 193                                   u16 proto,
 194                                   struct vmci_handle handle,
 195                                   bool convert_error)
 196 {
 197         int err;
 198
 199         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 200                                    proto, handle);
 201         err = vmci_datagram_send(&pkt->dg);
 202         if (convert_error && (err < 0))
 203                 return vmci_transport_error_to_vsock_error(err);
 204
 205         return err;
 206 }
 207
 208 static int
 209 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 210                                       enum vmci_transport_packet_type type,
 211                                       u64 size,
 212                                       u64 mode,
 213                                       struct vmci_transport_waiting_info *wait,
 214                                       struct vmci_handle handle)
 215 {
 216         struct vmci_transport_packet reply;
 217         struct sockaddr_vm src, dst;
 218
 219         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 220                 return 0;
 221         } else {
 222                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 223                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 224                                                          type,
 225                                                          size, mode, wait,
 226                                                          VSOCK_PROTO_INVALID,
 227                                                          handle, true);
 228         }
 229 }
 230
 231 static int
 232 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 233                                    struct sockaddr_vm *dst,
 234                                    enum vmci_transport_packet_type type,
 235                                    u64 size,
 236                                    u64 mode,
 237                                    struct vmci_transport_waiting_info *wait,
 238                                    struct vmci_handle handle)
 239 {
 240         /* Note that it is safe to use a single packet across all CPUs since
 241          * two tasklets of the same type are guaranteed to not ever run
 242          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 243          * we can use per-cpu packets.
 244          */
 245         static struct vmci_transport_packet pkt;
 246
 247         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 248                                                  size, mode, wait,
 249                                                  VSOCK_PROTO_INVALID, handle,
 250                                                  false);
 251 }
 252
 253 static int
 254 vmci_transport_alloc_send_control_pkt(struct sockaddr_vm *src,
 255                                       struct sockaddr_vm *dst,
 256                                       enum vmci_transport_packet_type type,
 257                                       u64 size,
 258                                       u64 mode,
 259                                       struct vmci_transport_waiting_info *wait,
 260                                       u16 proto,
 261                                       struct vmci_handle handle)
 262 {
 263         struct vmci_transport_packet *pkt;
 264         int err;
 265
 266         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 267         if (!pkt)
 268                 return -ENOMEM;
 269
 270         err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
 271                                                 mode, wait, proto, handle,
 272                                                 true);
 273         kfree(pkt);
 274
 275         return err;
 276 }
 277
 278 static int
 279 vmci_transport_send_control_pkt(struct sock *sk,
 280                                 enum vmci_transport_packet_type type,
 281                                 u64 size,
 282                                 u64 mode,
 283                                 struct vmci_transport_waiting_info *wait,
 284                                 u16 proto,
 285                                 struct vmci_handle handle)
 286 {
 287         struct vsock_sock *vsk;
 288
 289         vsk = vsock_sk(sk);
 290
 291         if (!vsock_addr_bound(&vsk->local_addr))
 292                 return -EINVAL;
 293
 294         if (!vsock_addr_bound(&vsk->remote_addr))
 295                 return -EINVAL;
 296
 297         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
 298                                                      &vsk->remote_addr,
 299                                                      type, size, mode,
 300                                                      wait, proto, handle);
 301 }
 302
 303 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 304                                         struct sockaddr_vm *src,
 305                                         struct vmci_transport_packet *pkt)
 306 {
 307         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 308                 return 0;
 309         return vmci_transport_send_control_pkt_bh(
 310                                         dst, src,
 311                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 312                                         0, NULL, VMCI_INVALID_HANDLE);
 313 }
 314
 315 static int vmci_transport_send_reset(struct sock *sk,
 316                                      struct vmci_transport_packet *pkt)
 317 {
 318         struct sockaddr_vm *dst_ptr;
 319         struct sockaddr_vm dst;
 320         struct vsock_sock *vsk;
 321
 322         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 323                 return 0;
 324
 325         vsk = vsock_sk(sk);
 326
 327         if (!vsock_addr_bound(&vsk->local_addr))
 328                 return -EINVAL;
 329
 330         if (vsock_addr_bound(&vsk->remote_addr)) {
 331                 dst_ptr = &vsk->remote_addr;
 332         } else {
 333                 vsock_addr_init(&dst, pkt->dg.src.context,
 334                                 pkt->src_port);
 335                 dst_ptr = &dst;
 336         }
 337         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
 338                                              VMCI_TRANSPORT_PACKET_TYPE_RST,
 339                                              0, 0, NULL, VSOCK_PROTO_INVALID,
 340                                              VMCI_INVALID_HANDLE);
 341 }
 342
 343 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 344 {
 345         return vmci_transport_send_control_pkt(
 346                                         sk,
 347                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 348                                         size, 0, NULL,
 349                                         VSOCK_PROTO_INVALID,
 350                                         VMCI_INVALID_HANDLE);
 351 }
 352
 353 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 354                                           u16 version)
 355 {
 356         return vmci_transport_send_control_pkt(
 357                                         sk,
 358                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 359                                         size, 0, NULL, version,
 360                                         VMCI_INVALID_HANDLE);
 361 }
 362
 363 static int vmci_transport_send_qp_offer(struct sock *sk,
 364                                         struct vmci_handle handle)
 365 {
 366         return vmci_transport_send_control_pkt(
 367                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 368                                         0, NULL,
 369                                         VSOCK_PROTO_INVALID, handle);
 370 }
 371
 372 static int vmci_transport_send_attach(struct sock *sk,
 373                                       struct vmci_handle handle)
 374 {
 375         return vmci_transport_send_control_pkt(
 376                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 377                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 378                                         handle);
 379 }
 380
 381 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 382 {
 383         return vmci_transport_reply_control_pkt_fast(
 384                                                 pkt,
 385                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 386                                                 0, 0, NULL,
 387                                                 VMCI_INVALID_HANDLE);
 388 }
 389
 390 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 391                                           struct sockaddr_vm *src)
 392 {
 393         return vmci_transport_send_control_pkt_bh(
 394                                         dst, src,
 395                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 396                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 397 }
 398
 399 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 400                                  struct sockaddr_vm *src)
 401 {
 402         return vmci_transport_send_control_pkt_bh(
 403                                         dst, src,
 404                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 405                                         0, NULL, VMCI_INVALID_HANDLE);
 406 }
 407
 408 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 409                                 struct sockaddr_vm *src)
 410 {
 411         return vmci_transport_send_control_pkt_bh(
 412                                         dst, src,
 413                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 414                                         0, NULL, VMCI_INVALID_HANDLE);
 415 }
 416
 417 int vmci_transport_send_wrote(struct sock *sk)
 418 {
 419         return vmci_transport_send_control_pkt(
 420                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 421                                         0, NULL, VSOCK_PROTO_INVALID,
 422                                         VMCI_INVALID_HANDLE);
 423 }
 424
 425 int vmci_transport_send_read(struct sock *sk)
 426 {
 427         return vmci_transport_send_control_pkt(
 428                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 429                                         0, NULL, VSOCK_PROTO_INVALID,
 430                                         VMCI_INVALID_HANDLE);
 431 }
 432
 433 int vmci_transport_send_waiting_write(struct sock *sk,
 434                                       struct vmci_transport_waiting_info *wait)
 435 {
 436         return vmci_transport_send_control_pkt(
 437                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 438                                 0, 0, wait, VSOCK_PROTO_INVALID,
 439                                 VMCI_INVALID_HANDLE);
 440 }
 441
 442 int vmci_transport_send_waiting_read(struct sock *sk,
 443                                      struct vmci_transport_waiting_info *wait)
 444 {
 445         return vmci_transport_send_control_pkt(
 446                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 447                                 0, 0, wait, VSOCK_PROTO_INVALID,
 448                                 VMCI_INVALID_HANDLE);
 449 }
 450
 451 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 452 {
 453         return vmci_transport_send_control_pkt(
 454                                         &vsk->sk,
 455                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 456                                         0, mode, NULL,
 457                                         VSOCK_PROTO_INVALID,
 458                                         VMCI_INVALID_HANDLE);
 459 }
 460
 461 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 462 {
 463         return vmci_transport_send_control_pkt(sk,
 464                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 465                                         size, 0, NULL,
 466                                         VSOCK_PROTO_INVALID,
 467                                         VMCI_INVALID_HANDLE);
 468 }
 469
 470 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 471                                              u16 version)
 472 {
 473         return vmci_transport_send_control_pkt(
 474                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 475                                         size, 0, NULL, version,
 476                                         VMCI_INVALID_HANDLE);
 477 }
 478
 479 static struct sock *vmci_transport_get_pending(
 480                                         struct sock *listener,
 481                                         struct vmci_transport_packet *pkt)
 482 {
 483         struct vsock_sock *vlistener;
 484         struct vsock_sock *vpending;
 485         struct sock *pending;
 486         struct sockaddr_vm src;
 487
 488         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 489
 490         vlistener = vsock_sk(listener);
 491
 492         list_for_each_entry(vpending, &vlistener->pending_links,
 493                             pending_links) {
 494                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 495                     pkt->dst_port == vpending->local_addr.svm_port) {
 496                         pending = sk_vsock(vpending);
 497                         sock_hold(pending);
 498                         goto found;
 499                 }
 500         }
 501
 502         pending = NULL;
 503 found:
 504         return pending;
 505
 506 }
 507
 508 static void vmci_transport_release_pending(struct sock *pending)
 509 {
 510         sock_put(pending);
 511 }
 512
 513 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 514  * trusted sockets 2) sockets from applications running as the same user as the
 515  * VM (this is only true for the host side and only when using hosted products)
 516  */
 517
 518 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 519 {
 520         return vsock->trusted ||
 521                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 522 }
 523
 524 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 525  * only if it is trusted as described in vmci_transport_is_trusted.
 526  */
 527
 528 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 529 {
 530         if (VMADDR_CID_HYPERVISOR == peer_cid)
 531                 return true;
 532
 533         if (vsock->cached_peer != peer_cid) {
 534                 vsock->cached_peer = peer_cid;
 535                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 536                     (vmci_context_get_priv_flags(peer_cid) &
 537                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 538                         vsock->cached_peer_allow_dgram = false;
 539                 } else {
 540                         vsock->cached_peer_allow_dgram = true;
 541                 }
 542         }
 543
 544         return vsock->cached_peer_allow_dgram;
 545 }
 546
 547 static int
 548 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 549                                 struct vmci_handle *handle,
 550                                 u64 produce_size,
 551                                 u64 consume_size,
 552                                 u32 peer, u32 flags, bool trusted)
 553 {
 554         int err = 0;
 555
 556         if (trusted) {
 557                 /* Try to allocate our queue pair as trusted. This will only
 558                  * work if vsock is running in the host.
 559                  */
 560
 561                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 562                                        consume_size,
 563                                        peer, flags,
 564                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 565                 if (err != VMCI_ERROR_NO_ACCESS)
 566                         goto out;
 567
 568         }
 569
 570         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 571                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 572 out:
 573         if (err < 0) {
 574                 pr_err("Could not attach to queue pair with %d\n",
 575                        err);
 576                 err = vmci_transport_error_to_vsock_error(err);
 577         }
 578
 579         return err;
 580 }
 581
 582 static int
 583 vmci_transport_datagram_create_hnd(u32 resource_id,
 584                                    u32 flags,
 585                                    vmci_datagram_recv_cb recv_cb,
 586                                    void *client_data,
 587                                    struct vmci_handle *out_handle)
 588 {
 589         int err = 0;
 590
 591         /* Try to allocate our datagram handler as trusted. This will only work
 592          * if vsock is running in the host.
 593          */
 594
 595         err = vmci_datagram_create_handle_priv(resource_id, flags,
 596                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 597                                                recv_cb,
 598                                                client_data, out_handle);
 599
 600         if (err == VMCI_ERROR_NO_ACCESS)
 601                 err = vmci_datagram_create_handle(resource_id, flags,
 602                                                   recv_cb, client_data,
 603                                                   out_handle);
 604
 605         return err;
 606 }
 607
 608 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 609  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 610  * sleep it should defer that work to a work queue.
 611  */
 612
 613 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 614 {
 615         struct sock *sk;
 616         size_t size;
 617         struct sk_buff *skb;
 618         struct vsock_sock *vsk;
 619
 620         sk = (struct sock *)data;
 621
 622         /* This handler is privileged when this module is running on the host.
 623          * We will get datagrams from all endpoints (even VMs that are in a
 624          * restricted context). If we get one from a restricted context then
 625          * the destination socket must be trusted.
 626          *
 627          * NOTE: We access the socket struct without holding the lock here.
 628          * This is ok because the field we are interested is never modified
 629          * outside of the create and destruct socket functions.
 630          */
 631         vsk = vsock_sk(sk);
 632         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 633                 return VMCI_ERROR_NO_ACCESS;
 634
 635         size = VMCI_DG_SIZE(dg);
 636
 637         /* Attach the packet to the socket's receive queue as an sk_buff. */
 638         skb = alloc_skb(size, GFP_ATOMIC);
 639         if (!skb)
 640                 return VMCI_ERROR_NO_MEM;
 641
 642         /* sk_receive_skb() will do a sock_put(), so hold here. */
 643         sock_hold(sk);
 644         skb_put(skb, size);
 645         memcpy(skb->data, dg, size);
 646         sk_receive_skb(sk, skb, 0);
 647
 648         return VMCI_SUCCESS;
 649 }
 650
 651 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 652 {
 653         static const u32 non_socket_contexts[] = {
 654                 VMADDR_CID_RESERVED,
 655         };
 656         int i;
 657
 658         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 659
 660         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 661                 if (cid == non_socket_contexts[i])
 662                         return false;
 663         }
 664
 665         return true;
 666 }
 667
 668 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 669  * interrupt fires.  This is run in bottom-half context but it defers most of
 670  * its work to the packet handling work queue.
 671  */
 672
 673 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 674 {
 675         struct sock *sk;
 676         struct sockaddr_vm dst;
 677         struct sockaddr_vm src;
 678         struct vmci_transport_packet *pkt;
 679         struct vsock_sock *vsk;
 680         bool bh_process_pkt;
 681         int err;
 682
 683         sk = NULL;
 684         err = VMCI_SUCCESS;
 685         bh_process_pkt = false;
 686
 687         /* Ignore incoming packets from contexts without sockets, or resources
 688          * that aren't vsock implementations.
 689          */
 690
 691         if (!vmci_transport_stream_allow(dg->src.context, -1)
 692             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 693                 return VMCI_ERROR_NO_ACCESS;
 694
 695         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 696                 /* Drop datagrams that do not contain full VSock packets. */
 697                 return VMCI_ERROR_INVALID_ARGS;
 698
 699         pkt = (struct vmci_transport_packet *)dg;
 700
 701         /* Find the socket that should handle this packet.  First we look for a
 702          * connected socket and if there is none we look for a socket bound to
 703          * the destintation address.
 704          */
 705         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 706         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 707
 708         sk = vsock_find_connected_socket(&src, &dst);
 709         if (!sk) {
 710                 sk = vsock_find_bound_socket(&dst);
 711                 if (!sk) {
 712                         /* We could not find a socket for this specified
 713                          * address.  If this packet is a RST, we just drop it.
 714                          * If it is another packet, we send a RST.  Note that
 715                          * we do not send a RST reply to RSTs so that we do not
 716                          * continually send RSTs between two endpoints.
 717                          *
 718                          * Note that since this is a reply, dst is src and src
 719                          * is dst.
 720                          */
 721                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 722                                 pr_err("unable to send reset\n");
 723
 724                         err = VMCI_ERROR_NOT_FOUND;
 725                         goto out;
 726                 }
 727         }
 728
 729         /* If the received packet type is beyond all types known to this
 730          * implementation, reply with an invalid message.  Hopefully this will
 731          * help when implementing backwards compatibility in the future.
 732          */
 733         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 734                 vmci_transport_send_invalid_bh(&dst, &src);
 735                 err = VMCI_ERROR_INVALID_ARGS;
 736                 goto out;
 737         }
 738
 739         /* This handler is privileged when this module is running on the host.
 740          * We will get datagram connect requests from all endpoints (even VMs
 741          * that are in a restricted context). If we get one from a restricted
 742          * context then the destination socket must be trusted.
 743          *
 744          * NOTE: We access the socket struct without holding the lock here.
 745          * This is ok because the field we are interested is never modified
 746          * outside of the create and destruct socket functions.
 747          */
 748         vsk = vsock_sk(sk);
 749         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 750                 err = VMCI_ERROR_NO_ACCESS;
 751                 goto out;
 752         }
 753
 754         /* We do most everything in a work queue, but let's fast path the
 755          * notification of reads and writes to help data transfer performance.
 756          * We can only do this if there is no process context code executing
 757          * for this socket since that may change the state.
 758          */
 759         bh_lock_sock(sk);
 760
 761         if (!sock_owned_by_user(sk)) {
 762                 /* The local context ID may be out of date, update it. */
 763                 vsk->local_addr.svm_cid = dst.svm_cid;
 764
 765                 if (sk->sk_state == TCP_ESTABLISHED)
 766                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 767                                         sk, pkt, true, &dst, &src,
 768                                         &bh_process_pkt);
 769         }
 770
 771         bh_unlock_sock(sk);
 772
 773         if (!bh_process_pkt) {
 774                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 775
 776                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 777                 if (!recv_pkt_info) {
 778                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 779                                 pr_err("unable to send reset\n");
 780
 781                         err = VMCI_ERROR_NO_MEM;
 782                         goto out;
 783                 }
 784
 785                 recv_pkt_info->sk = sk;
 786                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 787                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 788
 789                 schedule_work(&recv_pkt_info->work);
 790                 /* Clear sk so that the reference count incremented by one of
 791                  * the Find functions above is not decremented below.  We need
 792                  * that reference count for the packet handler we've scheduled
 793                  * to run.
 794                  */
 795                 sk = NULL;
 796         }
 797
 798 out:
 799         if (sk)
 800                 sock_put(sk);
 801
 802         return err;
 803 }
 804
 805 static void vmci_transport_handle_detach(struct sock *sk)
 806 {
 807         struct vsock_sock *vsk;
 808
 809         vsk = vsock_sk(sk);
 810         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 811                 sock_set_flag(sk, SOCK_DONE);
 812
 813                 /* On a detach the peer will not be sending or receiving
 814                  * anymore.
 815                  */
 816                 vsk->peer_shutdown = SHUTDOWN_MASK;
 817
 818                 /* We should not be sending anymore since the peer won't be
 819                  * there to receive, but we can still receive if there is data
 820                  * left in our consume queue. If the local endpoint is a host,
 821                  * we can't call vsock_stream_has_data, since that may block,
 822                  * but a host endpoint can't read data once the VM has
 823                  * detached, so there is no available data in that case.
 824                  */
 825                 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
 826                     vsock_stream_has_data(vsk) <= 0) {
 827                         if (sk->sk_state == TCP_SYN_SENT) {
 828                                 /* The peer may detach from a queue pair while
 829                                  * we are still in the connecting state, i.e.,
 830                                  * if the peer VM is killed after attaching to
 831                                  * a queue pair, but before we complete the
 832                                  * handshake. In that case, we treat the detach
 833                                  * event like a reset.
 834                                  */
 835
 836                                 sk->sk_state = TCP_CLOSE;
 837                                 sk->sk_err = ECONNRESET;
 838                                 sk->sk_error_report(sk);
 839                                 return;
 840                         }
 841                         sk->sk_state = TCP_CLOSE;
 842                 }
 843                 sk->sk_state_change(sk);
 844         }
 845 }
 846
 847 static void vmci_transport_peer_detach_cb(u32 sub_id,
 848                                           const struct vmci_event_data *e_data,
 849                                           void *client_data)
 850 {
 851         struct vmci_transport *trans = client_data;
 852         const struct vmci_event_payload_qp *e_payload;
 853
 854         e_payload = vmci_event_data_const_payload(e_data);
 855
 856         /* XXX This is lame, we should provide a way to lookup sockets by
 857          * qp_handle.
 858          */
 859         if (vmci_handle_is_invalid(e_payload->handle) ||
 860             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 861                 return;
 862
 863         /* We don't ask for delayed CBs when we subscribe to this event (we
 864          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 865          * guarantees in that case about what context we might be running in,
 866          * so it could be BH or process, blockable or non-blockable.  So we
 867          * need to account for all possible contexts here.
 868          */
 869         spin_lock_bh(&trans->lock);
 870         if (!trans->sk)
 871                 goto out;
 872
 873         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 874          * where trans->sk isn't locked.
 875          */
 876         bh_lock_sock(trans->sk);
 877
 878         vmci_transport_handle_detach(trans->sk);
 879
 880         bh_unlock_sock(trans->sk);
 881  out:
 882         spin_unlock_bh(&trans->lock);
 883 }
 884
 885 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 886                                          const struct vmci_event_data *e_data,
 887                                          void *client_data)
 888 {
 889         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 890 }
 891
 892 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 893 {
 894         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 895         struct vmci_transport_packet *pkt;
 896         struct sock *sk;
 897
 898         recv_pkt_info =
 899                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 900         sk = recv_pkt_info->sk;
 901         pkt = &recv_pkt_info->pkt;
 902
 903         lock_sock(sk);
 904
 905         /* The local context ID may be out of date. */
 906         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 907
 908         switch (sk->sk_state) {
 909         case TCP_LISTEN:
 910                 vmci_transport_recv_listen(sk, pkt);
 911                 break;
 912         case TCP_SYN_SENT:
 913                 /* Processing of pending connections for servers goes through
 914                  * the listening socket, so see vmci_transport_recv_listen()
 915                  * for that path.
 916                  */
 917                 vmci_transport_recv_connecting_client(sk, pkt);
 918                 break;
 919         case TCP_ESTABLISHED:
 920                 vmci_transport_recv_connected(sk, pkt);
 921                 break;
 922         default:
 923                 /* Because this function does not run in the same context as
 924                  * vmci_transport_recv_stream_cb it is possible that the
 925                  * socket has closed. We need to let the other side know or it
 926                  * could be sitting in a connect and hang forever. Send a
 927                  * reset to prevent that.
 928                  */
 929                 vmci_transport_send_reset(sk, pkt);
 930                 break;
 931         }
 932
 933         release_sock(sk);
 934         kfree(recv_pkt_info);
 935         /* Release reference obtained in the stream callback when we fetched
 936          * this socket out of the bound or connected list.
 937          */
 938         sock_put(sk);
 939 }
 940
 941 static int vmci_transport_recv_listen(struct sock *sk,
 942                                       struct vmci_transport_packet *pkt)
 943 {
 944         struct sock *pending;
 945         struct vsock_sock *vpending;
 946         int err;
 947         u64 qp_size;
 948         bool old_request = false;
 949         bool old_pkt_proto = false;
 950
 951         err = 0;
 952
 953         /* Because we are in the listen state, we could be receiving a packet
 954          * for ourself or any previous connection requests that we received.
 955          * If it's the latter, we try to find a socket in our list of pending
 956          * connections and, if we do, call the appropriate handler for the
 957          * state that that socket is in.  Otherwise we try to service the
 958          * connection request.
 959          */
 960         pending = vmci_transport_get_pending(sk, pkt);
 961         if (pending) {
 962                 lock_sock(pending);
 963
 964                 /* The local context ID may be out of date. */
 965                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 966
 967                 switch (pending->sk_state) {
 968                 case TCP_SYN_SENT:
 969                         err = vmci_transport_recv_connecting_server(sk,
 970                                                                     pending,
 971                                                                     pkt);
 972                         break;
 973                 default:
 974                         vmci_transport_send_reset(pending, pkt);
 975                         err = -EINVAL;
 976                 }
 977
 978                 if (err < 0)
 979                         vsock_remove_pending(sk, pending);
 980
 981                 release_sock(pending);
 982                 vmci_transport_release_pending(pending);
 983
 984                 return err;
 985         }
 986
 987         /* The listen state only accepts connection requests.  Reply with a
 988          * reset unless we received a reset.
 989          */
 990
 991         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 992               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 993                 vmci_transport_reply_reset(pkt);
 994                 return -EINVAL;
 995         }
 996
 997         if (pkt->u.size == 0) {
 998                 vmci_transport_reply_reset(pkt);
 999                 return -EINVAL;
1000         }
1001
1002         /* If this socket can't accommodate this connection request, we send a
1003          * reset.  Otherwise we create and initialize a child socket and reply
1004          * with a connection negotiation.
1005          */
1006         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1007                 vmci_transport_reply_reset(pkt);
1008                 return -ECONNREFUSED;
1009         }
1010
1011         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1012                                  sk->sk_type, 0);
1013         if (!pending) {
1014                 vmci_transport_send_reset(sk, pkt);
1015                 return -ENOMEM;
1016         }
1017
1018         vpending = vsock_sk(pending);
1019
1020         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1021                         pkt->dst_port);
1022         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1023                         pkt->src_port);
1024
1025         /* If the proposed size fits within our min/max, accept it. Otherwise
1026          * propose our own size.
1027          */
1028         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1029             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1030                 qp_size = pkt->u.size;
1031         } else {
1032                 qp_size = vmci_trans(vpending)->queue_pair_size;
1033         }
1034
1035         /* Figure out if we are using old or new requests based on the
1036          * overrides pkt types sent by our peer.
1037          */
1038         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1039                 old_request = old_pkt_proto;
1040         } else {
1041                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1042                         old_request = true;
1043                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1044                         old_request = false;
1045
1046         }
1047
1048         if (old_request) {
1049                 /* Handle a REQUEST (or override) */
1050                 u16 version = VSOCK_PROTO_INVALID;
1051                 if (vmci_transport_proto_to_notify_struct(
1052                         pending, &version, true))
1053                         err = vmci_transport_send_negotiate(pending, qp_size);
1054                 else
1055                         err = -EINVAL;
1056
1057         } else {
1058                 /* Handle a REQUEST2 (or override) */
1059                 int proto_int = pkt->proto;
1060                 int pos;
1061                 u16 active_proto_version = 0;
1062
1063                 /* The list of possible protocols is the intersection of all
1064                  * protocols the client supports ... plus all the protocols we
1065                  * support.
1066                  */
1067                 proto_int &= vmci_transport_new_proto_supported_versions();
1068
1069                 /* We choose the highest possible protocol version and use that
1070                  * one.
1071                  */
1072                 pos = fls(proto_int);
1073                 if (pos) {
1074                         active_proto_version = (1 << (pos - 1));
1075                         if (vmci_transport_proto_to_notify_struct(
1076                                 pending, &active_proto_version, false))
1077                                 err = vmci_transport_send_negotiate2(pending,
1078                                                         qp_size,
1079                                                         active_proto_version);
1080                         else
1081                                 err = -EINVAL;
1082
1083                 } else {
1084                         err = -EINVAL;
1085                 }
1086         }
1087
1088         if (err < 0) {
1089                 vmci_transport_send_reset(sk, pkt);
1090                 sock_put(pending);
1091                 err = vmci_transport_error_to_vsock_error(err);
1092                 goto out;
1093         }
1094
1095         vsock_add_pending(sk, pending);
1096         sk_acceptq_added(sk);
1097
1098         pending->sk_state = TCP_SYN_SENT;
1099         vmci_trans(vpending)->produce_size =
1100                 vmci_trans(vpending)->consume_size = qp_size;
1101         vmci_trans(vpending)->queue_pair_size = qp_size;
1102
1103         vmci_trans(vpending)->notify_ops->process_request(pending);
1104
1105         /* We might never receive another message for this socket and it's not
1106          * connected to any process, so we have to ensure it gets cleaned up
1107          * ourself.  Our delayed work function will take care of that.  Note
1108          * that we do not ever cancel this function since we have few
1109          * guarantees about its state when calling cancel_delayed_work().
1110          * Instead we hold a reference on the socket for that function and make
1111          * it capable of handling cases where it needs to do nothing but
1112          * release that reference.
1113          */
1114         vpending->listener = sk;
1115         sock_hold(sk);
1116         sock_hold(pending);
1117         schedule_delayed_work(&vpending->pending_work, HZ);
1118
1119 out:
1120         return err;
1121 }
1122
1123 static int
1124 vmci_transport_recv_connecting_server(struct sock *listener,
1125                                       struct sock *pending,
1126                                       struct vmci_transport_packet *pkt)
1127 {
1128         struct vsock_sock *vpending;
1129         struct vmci_handle handle;
1130         struct vmci_qp *qpair;
1131         bool is_local;
1132         u32 flags;
1133         u32 detach_sub_id;
1134         int err;
1135         int skerr;
1136
1137         vpending = vsock_sk(pending);
1138         detach_sub_id = VMCI_INVALID_ID;
1139
1140         switch (pkt->type) {
1141         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1142                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1143                         vmci_transport_send_reset(pending, pkt);
1144                         skerr = EPROTO;
1145                         err = -EINVAL;
1146                         goto destroy;
1147                 }
1148                 break;
1149         default:
1150                 /* Close and cleanup the connection. */
1151                 vmci_transport_send_reset(pending, pkt);
1152                 skerr = EPROTO;
1153                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1154                 goto destroy;
1155         }
1156
1157         /* In order to complete the connection we need to attach to the offered
1158          * queue pair and send an attach notification.  We also subscribe to the
1159          * detach event so we know when our peer goes away, and we do that
1160          * before attaching so we don't miss an event.  If all this succeeds,
1161          * we update our state and wakeup anything waiting in accept() for a
1162          * connection.
1163          */
1164
1165         /* We don't care about attach since we ensure the other side has
1166          * attached by specifying the ATTACH_ONLY flag below.
1167          */
1168         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1169                                    vmci_transport_peer_detach_cb,
1170                                    vmci_trans(vpending), &detach_sub_id);
1171         if (err < VMCI_SUCCESS) {
1172                 vmci_transport_send_reset(pending, pkt);
1173                 err = vmci_transport_error_to_vsock_error(err);
1174                 skerr = -err;
1175                 goto destroy;
1176         }
1177
1178         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1179
1180         /* Now attach to the queue pair the client created. */
1181         handle = pkt->u.handle;
1182
1183         /* vpending->local_addr always has a context id so we do not need to
1184          * worry about VMADDR_CID_ANY in this case.
1185          */
1186         is_local =
1187             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1188         flags = VMCI_QPFLAG_ATTACH_ONLY;
1189         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1190
1191         err = vmci_transport_queue_pair_alloc(
1192                                         &qpair,
1193                                         &handle,
1194                                         vmci_trans(vpending)->produce_size,
1195                                         vmci_trans(vpending)->consume_size,
1196                                         pkt->dg.src.context,
1197                                         flags,
1198                                         vmci_transport_is_trusted(
1199                                                 vpending,
1200                                                 vpending->remote_addr.svm_cid));
1201         if (err < 0) {
1202                 vmci_transport_send_reset(pending, pkt);
1203                 skerr = -err;
1204                 goto destroy;
1205         }
1206
1207         vmci_trans(vpending)->qp_handle = handle;
1208         vmci_trans(vpending)->qpair = qpair;
1209
1210         /* When we send the attach message, we must be ready to handle incoming
1211          * control messages on the newly connected socket. So we move the
1212          * pending socket to the connected state before sending the attach
1213          * message. Otherwise, an incoming packet triggered by the attach being
1214          * received by the peer may be processed concurrently with what happens
1215          * below after sending the attach message, and that incoming packet
1216          * will find the listening socket instead of the (currently) pending
1217          * socket. Note that enqueueing the socket increments the reference
1218          * count, so even if a reset comes before the connection is accepted,
1219          * the socket will be valid until it is removed from the queue.
1220          *
1221          * If we fail sending the attach below, we remove the socket from the
1222          * connected list and move the socket to TCP_CLOSE before
1223          * releasing the lock, so a pending slow path processing of an incoming
1224          * packet will not see the socket in the connected state in that case.
1225          */
1226         pending->sk_state = TCP_ESTABLISHED;
1227
1228         vsock_insert_connected(vpending);
1229
1230         /* Notify our peer of our attach. */
1231         err = vmci_transport_send_attach(pending, handle);
1232         if (err < 0) {
1233                 vsock_remove_connected(vpending);
1234                 pr_err("Could not send attach\n");
1235                 vmci_transport_send_reset(pending, pkt);
1236                 err = vmci_transport_error_to_vsock_error(err);
1237                 skerr = -err;
1238                 goto destroy;
1239         }
1240
1241         /* We have a connection. Move the now connected socket from the
1242          * listener's pending list to the accept queue so callers of accept()
1243          * can find it.
1244          */
1245         vsock_remove_pending(listener, pending);
1246         vsock_enqueue_accept(listener, pending);
1247
1248         /* Callers of accept() will be be waiting on the listening socket, not
1249          * the pending socket.
1250          */
1251         listener->sk_data_ready(listener);
1252
1253         return 0;
1254
1255 destroy:
1256         pending->sk_err = skerr;
1257         pending->sk_state = TCP_CLOSE;
1258         /* As long as we drop our reference, all necessary cleanup will handle
1259          * when the cleanup function drops its reference and our destruct
1260          * implementation is called.  Note that since the listen handler will
1261          * remove pending from the pending list upon our failure, the cleanup
1262          * function won't drop the additional reference, which is why we do it
1263          * here.
1264          */
1265         sock_put(pending);
1266
1267         return err;
1268 }
1269
1270 static int
1271 vmci_transport_recv_connecting_client(struct sock *sk,
1272                                       struct vmci_transport_packet *pkt)
1273 {
1274         struct vsock_sock *vsk;
1275         int err;
1276         int skerr;
1277
1278         vsk = vsock_sk(sk);
1279
1280         switch (pkt->type) {
1281         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1282                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1283                     !vmci_handle_is_equal(pkt->u.handle,
1284                                           vmci_trans(vsk)->qp_handle)) {
1285                         skerr = EPROTO;
1286                         err = -EINVAL;
1287                         goto destroy;
1288                 }
1289
1290                 /* Signify the socket is connected and wakeup the waiter in
1291                  * connect(). Also place the socket in the connected table for
1292                  * accounting (it can already be found since it's in the bound
1293                  * table).
1294                  */
1295                 sk->sk_state = TCP_ESTABLISHED;
1296                 sk->sk_socket->state = SS_CONNECTED;
1297                 vsock_insert_connected(vsk);
1298                 sk->sk_state_change(sk);
1299
1300                 break;
1301         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1302         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1303                 if (pkt->u.size == 0
1304                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1305                     || pkt->src_port != vsk->remote_addr.svm_port
1306                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1307                     || vmci_trans(vsk)->qpair
1308                     || vmci_trans(vsk)->produce_size != 0
1309                     || vmci_trans(vsk)->consume_size != 0
1310                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1311                         skerr = EPROTO;
1312                         err = -EINVAL;
1313
1314                         goto destroy;
1315                 }
1316
1317                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1318                 if (err) {
1319                         skerr = -err;
1320                         goto destroy;
1321                 }
1322
1323                 break;
1324         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1325                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1326                 if (err) {
1327                         skerr = -err;
1328                         goto destroy;
1329                 }
1330
1331                 break;
1332         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1333                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1334                  * continue processing here after they sent an INVALID packet.
1335                  * This meant that we got a RST after the INVALID. We ignore a
1336                  * RST after an INVALID. The common code doesn't send the RST
1337                  * ... so we can hang if an old version of the common code
1338                  * fails between getting a REQUEST and sending an OFFER back.
1339                  * Not much we can do about it... except hope that it doesn't
1340                  * happen.
1341                  */
1342                 if (vsk->ignore_connecting_rst) {
1343                         vsk->ignore_connecting_rst = false;
1344                 } else {
1345                         skerr = ECONNRESET;
1346                         err = 0;
1347                         goto destroy;
1348                 }
1349
1350                 break;
1351         default:
1352                 /* Close and cleanup the connection. */
1353                 skerr = EPROTO;
1354                 err = -EINVAL;
1355                 goto destroy;
1356         }
1357
1358         return 0;
1359
1360 destroy:
1361         vmci_transport_send_reset(sk, pkt);
1362
1363         sk->sk_state = TCP_CLOSE;
1364         sk->sk_err = skerr;
1365         sk->sk_error_report(sk);
1366         return err;
1367 }
1368
1369 static int vmci_transport_recv_connecting_client_negotiate(
1370                                         struct sock *sk,
1371                                         struct vmci_transport_packet *pkt)
1372 {
1373         int err;
1374         struct vsock_sock *vsk;
1375         struct vmci_handle handle;
1376         struct vmci_qp *qpair;
1377         u32 detach_sub_id;
1378         bool is_local;
1379         u32 flags;
1380         bool old_proto = true;
1381         bool old_pkt_proto;
1382         u16 version;
1383
1384         vsk = vsock_sk(sk);
1385         handle = VMCI_INVALID_HANDLE;
1386         detach_sub_id = VMCI_INVALID_ID;
1387
1388         /* If we have gotten here then we should be past the point where old
1389          * linux vsock could have sent the bogus rst.
1390          */
1391         vsk->sent_request = false;
1392         vsk->ignore_connecting_rst = false;
1393
1394         /* Verify that we're OK with the proposed queue pair size */
1395         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1396             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1397                 err = -EINVAL;
1398                 goto destroy;
1399         }
1400
1401         /* At this point we know the CID the peer is using to talk to us. */
1402
1403         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1404                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1405
1406         /* Setup the notify ops to be the highest supported version that both
1407          * the server and the client support.
1408          */
1409
1410         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1411                 old_proto = old_pkt_proto;
1412         } else {
1413                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1414                         old_proto = true;
1415                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1416                         old_proto = false;
1417
1418         }
1419
1420         if (old_proto)
1421                 version = VSOCK_PROTO_INVALID;
1422         else
1423                 version = pkt->proto;
1424
1425         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1426                 err = -EINVAL;
1427                 goto destroy;
1428         }
1429
1430         /* Subscribe to detach events first.
1431          *
1432          * XXX We attach once for each queue pair created for now so it is easy
1433          * to find the socket (it's provided), but later we should only
1434          * subscribe once and add a way to lookup sockets by queue pair handle.
1435          */
1436         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1437                                    vmci_transport_peer_detach_cb,
1438                                    vmci_trans(vsk), &detach_sub_id);
1439         if (err < VMCI_SUCCESS) {
1440                 err = vmci_transport_error_to_vsock_error(err);
1441                 goto destroy;
1442         }
1443
1444         /* Make VMCI select the handle for us. */
1445         handle = VMCI_INVALID_HANDLE;
1446         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1447         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1448
1449         err = vmci_transport_queue_pair_alloc(&qpair,
1450                                               &handle,
1451                                               pkt->u.size,
1452                                               pkt->u.size,
1453                                               vsk->remote_addr.svm_cid,
1454                                               flags,
1455                                               vmci_transport_is_trusted(
1456                                                   vsk,
1457                                                   vsk->
1458                                                   remote_addr.svm_cid));
1459         if (err < 0)
1460                 goto destroy;
1461
1462         err = vmci_transport_send_qp_offer(sk, handle);
1463         if (err < 0) {
1464                 err = vmci_transport_error_to_vsock_error(err);
1465                 goto destroy;
1466         }
1467
1468         vmci_trans(vsk)->qp_handle = handle;
1469         vmci_trans(vsk)->qpair = qpair;
1470
1471         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1472                 pkt->u.size;
1473
1474         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1475
1476         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1477
1478         return 0;
1479
1480 destroy:
1481         if (detach_sub_id != VMCI_INVALID_ID)
1482                 vmci_event_unsubscribe(detach_sub_id);
1483
1484         if (!vmci_handle_is_invalid(handle))
1485                 vmci_qpair_detach(&qpair);
1486
1487         return err;
1488 }
1489
1490 static int
1491 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1492                                               struct vmci_transport_packet *pkt)
1493 {
1494         int err = 0;
1495         struct vsock_sock *vsk = vsock_sk(sk);
1496
1497         if (vsk->sent_request) {
1498                 vsk->sent_request = false;
1499                 vsk->ignore_connecting_rst = true;
1500
1501                 err = vmci_transport_send_conn_request(
1502                         sk, vmci_trans(vsk)->queue_pair_size);
1503                 if (err < 0)
1504                         err = vmci_transport_error_to_vsock_error(err);
1505                 else
1506                         err = 0;
1507
1508         }
1509
1510         return err;
1511 }
1512
1513 static int vmci_transport_recv_connected(struct sock *sk,
1514                                          struct vmci_transport_packet *pkt)
1515 {
1516         struct vsock_sock *vsk;
1517         bool pkt_processed = false;
1518
1519         /* In cases where we are closing the connection, it's sufficient to
1520          * mark the state change (and maybe error) and wake up any waiting
1521          * threads. Since this is a connected socket, it's owned by a user
1522          * process and will be cleaned up when the failure is passed back on
1523          * the current or next system call.  Our system call implementations
1524          * must therefore check for error and state changes on entry and when
1525          * being awoken.
1526          */
1527         switch (pkt->type) {
1528         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1529                 if (pkt->u.mode) {
1530                         vsk = vsock_sk(sk);
1531
1532                         vsk->peer_shutdown |= pkt->u.mode;
1533                         sk->sk_state_change(sk);
1534                 }
1535                 break;
1536
1537         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1538                 vsk = vsock_sk(sk);
1539                 /* It is possible that we sent our peer a message (e.g a
1540                  * WAITING_READ) right before we got notified that the peer had
1541                  * detached. If that happens then we can get a RST pkt back
1542                  * from our peer even though there is data available for us to
1543                  * read. In that case, don't shutdown the socket completely but
1544                  * instead allow the local client to finish reading data off
1545                  * the queuepair. Always treat a RST pkt in connected mode like
1546                  * a clean shutdown.
1547                  */
1548                 sock_set_flag(sk, SOCK_DONE);
1549                 vsk->peer_shutdown = SHUTDOWN_MASK;
1550                 if (vsock_stream_has_data(vsk) <= 0)
1551                         sk->sk_state = TCP_CLOSING;
1552
1553                 sk->sk_state_change(sk);
1554                 break;
1555
1556         default:
1557                 vsk = vsock_sk(sk);
1558                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1559                                 sk, pkt, false, NULL, NULL,
1560                                 &pkt_processed);
1561                 if (!pkt_processed)
1562                         return -EINVAL;
1563
1564                 break;
1565         }
1566
1567         return 0;
1568 }
1569
1570 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1571                                       struct vsock_sock *psk)
1572 {
1573         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1574         if (!vsk->trans)
1575                 return -ENOMEM;
1576
1577         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1578         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1579         vmci_trans(vsk)->qpair = NULL;
1580         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1581         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1582         vmci_trans(vsk)->notify_ops = NULL;
1583         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1584         vmci_trans(vsk)->sk = &vsk->sk;
1585         spin_lock_init(&vmci_trans(vsk)->lock);
1586         if (psk) {
1587                 vmci_trans(vsk)->queue_pair_size =
1588                         vmci_trans(psk)->queue_pair_size;
1589                 vmci_trans(vsk)->queue_pair_min_size =
1590                         vmci_trans(psk)->queue_pair_min_size;
1591                 vmci_trans(vsk)->queue_pair_max_size =
1592                         vmci_trans(psk)->queue_pair_max_size;
1593         } else {
1594                 vmci_trans(vsk)->queue_pair_size =
1595                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1596                 vmci_trans(vsk)->queue_pair_min_size =
1597                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1598                 vmci_trans(vsk)->queue_pair_max_size =
1599                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1600         }
1601
1602         return 0;
1603 }
1604
1605 static void vmci_transport_free_resources(struct list_head *transport_list)
1606 {
1607         while (!list_empty(transport_list)) {
1608                 struct vmci_transport *transport =
1609                     list_first_entry(transport_list, struct vmci_transport,
1610                                      elem);
1611                 list_del(&transport->elem);
1612
1613                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1614                         vmci_event_unsubscribe(transport->detach_sub_id);
1615                         transport->detach_sub_id = VMCI_INVALID_ID;
1616                 }
1617
1618                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1619                         vmci_qpair_detach(&transport->qpair);
1620                         transport->qp_handle = VMCI_INVALID_HANDLE;
1621                         transport->produce_size = 0;
1622                         transport->consume_size = 0;
1623                 }
1624
1625                 kfree(transport);
1626         }
1627 }
1628
1629 static void vmci_transport_cleanup(struct work_struct *work)
1630 {
1631         LIST_HEAD(pending);
1632
1633         spin_lock_bh(&vmci_transport_cleanup_lock);
1634         list_replace_init(&vmci_transport_cleanup_list, &pending);
1635         spin_unlock_bh(&vmci_transport_cleanup_lock);
1636         vmci_transport_free_resources(&pending);
1637 }
1638
1639 static void vmci_transport_destruct(struct vsock_sock *vsk)
1640 {
1641         /* transport can be NULL if we hit a failure at init() time */
1642         if (!vmci_trans(vsk))
1643                 return;
1644
1645         /* Ensure that the detach callback doesn't use the sk/vsk
1646          * we are about to destruct.
1647          */
1648         spin_lock_bh(&vmci_trans(vsk)->lock);
1649         vmci_trans(vsk)->sk = NULL;
1650         spin_unlock_bh(&vmci_trans(vsk)->lock);
1651
1652         if (vmci_trans(vsk)->notify_ops)
1653                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1654
1655         spin_lock_bh(&vmci_transport_cleanup_lock);
1656         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1657         spin_unlock_bh(&vmci_transport_cleanup_lock);
1658         schedule_work(&vmci_transport_cleanup_work);
1659
1660         vsk->trans = NULL;
1661 }
1662
1663 static void vmci_transport_release(struct vsock_sock *vsk)
1664 {
1665         vsock_remove_sock(vsk);
1666
1667         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1668                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1669                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1670         }
1671 }
1672
1673 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1674                                      struct sockaddr_vm *addr)
1675 {
1676         u32 port;
1677         u32 flags;
1678         int err;
1679
1680         /* VMCI will select a resource ID for us if we provide
1681          * VMCI_INVALID_ID.
1682          */
1683         port = addr->svm_port == VMADDR_PORT_ANY ?
1684                         VMCI_INVALID_ID : addr->svm_port;
1685
1686         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1687                 return -EACCES;
1688
1689         flags = addr->svm_cid == VMADDR_CID_ANY ?
1690                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1691
1692         err = vmci_transport_datagram_create_hnd(port, flags,
1693                                                  vmci_transport_recv_dgram_cb,
1694                                                  &vsk->sk,
1695                                                  &vmci_trans(vsk)->dg_handle);
1696         if (err < VMCI_SUCCESS)
1697                 return vmci_transport_error_to_vsock_error(err);
1698         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1699                         vmci_trans(vsk)->dg_handle.resource);
1700
1701         return 0;
1702 }
1703
1704 static int vmci_transport_dgram_enqueue(
1705         struct vsock_sock *vsk,
1706         struct sockaddr_vm *remote_addr,
1707         struct msghdr *msg,
1708         size_t len)
1709 {
1710         int err;
1711         struct vmci_datagram *dg;
1712
1713         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1714                 return -EMSGSIZE;
1715
1716         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1717                 return -EPERM;
1718
1719         /* Allocate a buffer for the user's message and our packet header. */
1720         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1721         if (!dg)
1722                 return -ENOMEM;
1723
1724         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1725
1726         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1727                                    remote_addr->svm_port);
1728         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1729                                    vsk->local_addr.svm_port);
1730         dg->payload_size = len;
1731
1732         err = vmci_datagram_send(dg);
1733         kfree(dg);
1734         if (err < 0)
1735                 return vmci_transport_error_to_vsock_error(err);
1736
1737         return err - sizeof(*dg);
1738 }
1739
1740 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1741                                         struct msghdr *msg, size_t len,
1742                                         int flags)
1743 {
1744         int err;
1745         int noblock;
1746         struct vmci_datagram *dg;
1747         size_t payload_len;
1748         struct sk_buff *skb;
1749
1750         noblock = flags & MSG_DONTWAIT;
1751
1752         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1753                 return -EOPNOTSUPP;
1754
1755         /* Retrieve the head sk_buff from the socket's receive queue. */
1756         err = 0;
1757         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1758         if (!skb)
1759                 return err;
1760
1761         dg = (struct vmci_datagram *)skb->data;
1762         if (!dg)
1763                 /* err is 0, meaning we read zero bytes. */
1764                 goto out;
1765
1766         payload_len = dg->payload_size;
1767         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1768         if (payload_len != skb->len - sizeof(*dg)) {
1769                 err = -EINVAL;
1770                 goto out;
1771         }
1772
1773         if (payload_len > len) {
1774                 payload_len = len;
1775                 msg->msg_flags |= MSG_TRUNC;
1776         }
1777
1778         /* Place the datagram payload in the user's iovec. */
1779         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1780         if (err)
1781                 goto out;
1782
1783         if (msg->msg_name) {
1784                 /* Provide the address of the sender. */
1785                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1786                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1787                 msg->msg_namelen = sizeof(*vm_addr);
1788         }
1789         err = payload_len;
1790
1791 out:
1792         skb_free_datagram(&vsk->sk, skb);
1793         return err;
1794 }
1795
1796 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1797 {
1798         if (cid == VMADDR_CID_HYPERVISOR) {
1799                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1800                  * state and are allowed.
1801                  */
1802                 return port == VMCI_UNITY_PBRPC_REGISTER;
1803         }
1804
1805         return true;
1806 }
1807
1808 static int vmci_transport_connect(struct vsock_sock *vsk)
1809 {
1810         int err;
1811         bool old_pkt_proto = false;
1812         struct sock *sk = &vsk->sk;
1813
1814         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1815                 old_pkt_proto) {
1816                 err = vmci_transport_send_conn_request(
1817                         sk, vmci_trans(vsk)->queue_pair_size);
1818                 if (err < 0) {
1819                         sk->sk_state = TCP_CLOSE;
1820                         return err;
1821                 }
1822         } else {
1823                 int supported_proto_versions =
1824                         vmci_transport_new_proto_supported_versions();
1825                 err = vmci_transport_send_conn_request2(
1826                                 sk, vmci_trans(vsk)->queue_pair_size,
1827                                 supported_proto_versions);
1828                 if (err < 0) {
1829                         sk->sk_state = TCP_CLOSE;
1830                         return err;
1831                 }
1832
1833                 vsk->sent_request = true;
1834         }
1835
1836         return err;
1837 }
1838
1839 static ssize_t vmci_transport_stream_dequeue(
1840         struct vsock_sock *vsk,
1841         struct msghdr *msg,
1842         size_t len,
1843         int flags)
1844 {
1845         if (flags & MSG_PEEK)
1846                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1847         else
1848                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1849 }
1850
1851 static ssize_t vmci_transport_stream_enqueue(
1852         struct vsock_sock *vsk,
1853         struct msghdr *msg,
1854         size_t len)
1855 {
1856         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1857 }
1858
1859 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1860 {
1861         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1862 }
1863
1864 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1865 {
1866         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1867 }
1868
1869 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1870 {
1871         return vmci_trans(vsk)->consume_size;
1872 }
1873
1874 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1875 {
1876         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1877 }
1878
1879 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1880 {
1881         return vmci_trans(vsk)->queue_pair_size;
1882 }
1883
1884 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1885 {
1886         return vmci_trans(vsk)->queue_pair_min_size;
1887 }
1888
1889 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1890 {
1891         return vmci_trans(vsk)->queue_pair_max_size;
1892 }
1893
1894 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1895 {
1896         if (val < vmci_trans(vsk)->queue_pair_min_size)
1897                 vmci_trans(vsk)->queue_pair_min_size = val;
1898         if (val > vmci_trans(vsk)->queue_pair_max_size)
1899                 vmci_trans(vsk)->queue_pair_max_size = val;
1900         vmci_trans(vsk)->queue_pair_size = val;
1901 }
1902
1903 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1904                                                u64 val)
1905 {
1906         if (val > vmci_trans(vsk)->queue_pair_size)
1907                 vmci_trans(vsk)->queue_pair_size = val;
1908         vmci_trans(vsk)->queue_pair_min_size = val;
1909 }
1910
1911 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1912                                                u64 val)
1913 {
1914         if (val < vmci_trans(vsk)->queue_pair_size)
1915                 vmci_trans(vsk)->queue_pair_size = val;
1916         vmci_trans(vsk)->queue_pair_max_size = val;
1917 }
1918
1919 static int vmci_transport_notify_poll_in(
1920         struct vsock_sock *vsk,
1921         size_t target,
1922         bool *data_ready_now)
1923 {
1924         return vmci_trans(vsk)->notify_ops->poll_in(
1925                         &vsk->sk, target, data_ready_now);
1926 }
1927
1928 static int vmci_transport_notify_poll_out(
1929         struct vsock_sock *vsk,
1930         size_t target,
1931         bool *space_available_now)
1932 {
1933         return vmci_trans(vsk)->notify_ops->poll_out(
1934                         &vsk->sk, target, space_available_now);
1935 }
1936
1937 static int vmci_transport_notify_recv_init(
1938         struct vsock_sock *vsk,
1939         size_t target,
1940         struct vsock_transport_recv_notify_data *data)
1941 {
1942         return vmci_trans(vsk)->notify_ops->recv_init(
1943                         &vsk->sk, target,
1944                         (struct vmci_transport_recv_notify_data *)data);
1945 }
1946
1947 static int vmci_transport_notify_recv_pre_block(
1948         struct vsock_sock *vsk,
1949         size_t target,
1950         struct vsock_transport_recv_notify_data *data)
1951 {
1952         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1953                         &vsk->sk, target,
1954                         (struct vmci_transport_recv_notify_data *)data);
1955 }
1956
1957 static int vmci_transport_notify_recv_pre_dequeue(
1958         struct vsock_sock *vsk,
1959         size_t target,
1960         struct vsock_transport_recv_notify_data *data)
1961 {
1962         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1963                         &vsk->sk, target,
1964                         (struct vmci_transport_recv_notify_data *)data);
1965 }
1966
1967 static int vmci_transport_notify_recv_post_dequeue(
1968         struct vsock_sock *vsk,
1969         size_t target,
1970         ssize_t copied,
1971         bool data_read,
1972         struct vsock_transport_recv_notify_data *data)
1973 {
1974         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1975                         &vsk->sk, target, copied, data_read,
1976                         (struct vmci_transport_recv_notify_data *)data);
1977 }
1978
1979 static int vmci_transport_notify_send_init(
1980         struct vsock_sock *vsk,
1981         struct vsock_transport_send_notify_data *data)
1982 {
1983         return vmci_trans(vsk)->notify_ops->send_init(
1984                         &vsk->sk,
1985                         (struct vmci_transport_send_notify_data *)data);
1986 }
1987
1988 static int vmci_transport_notify_send_pre_block(
1989         struct vsock_sock *vsk,
1990         struct vsock_transport_send_notify_data *data)
1991 {
1992         return vmci_trans(vsk)->notify_ops->send_pre_block(
1993                         &vsk->sk,
1994                         (struct vmci_transport_send_notify_data *)data);
1995 }
1996
1997 static int vmci_transport_notify_send_pre_enqueue(
1998         struct vsock_sock *vsk,
1999         struct vsock_transport_send_notify_data *data)
2000 {
2001         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2002                         &vsk->sk,
2003                         (struct vmci_transport_send_notify_data *)data);
2004 }
2005
2006 static int vmci_transport_notify_send_post_enqueue(
2007         struct vsock_sock *vsk,
2008         ssize_t written,
2009         struct vsock_transport_send_notify_data *data)
2010 {
2011         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2012                         &vsk->sk, written,
2013                         (struct vmci_transport_send_notify_data *)data);
2014 }
2015
2016 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2017 {
2018         if (PROTOCOL_OVERRIDE != -1) {
2019                 if (PROTOCOL_OVERRIDE == 0)
2020                         *old_pkt_proto = true;
2021                 else
2022                         *old_pkt_proto = false;
2023
2024                 pr_info("Proto override in use\n");
2025                 return true;
2026         }
2027
2028         return false;
2029 }
2030
2031 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2032                                                   u16 *proto,
2033                                                   bool old_pkt_proto)
2034 {
2035         struct vsock_sock *vsk = vsock_sk(sk);
2036
2037         if (old_pkt_proto) {
2038                 if (*proto != VSOCK_PROTO_INVALID) {
2039                         pr_err("Can't set both an old and new protocol\n");
2040                         return false;
2041                 }
2042                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2043                 goto exit;
2044         }
2045
2046         switch (*proto) {
2047         case VSOCK_PROTO_PKT_ON_NOTIFY:
2048                 vmci_trans(vsk)->notify_ops =
2049                         &vmci_transport_notify_pkt_q_state_ops;
2050                 break;
2051         default:
2052                 pr_err("Unknown notify protocol version\n");
2053                 return false;
2054         }
2055
2056 exit:
2057         vmci_trans(vsk)->notify_ops->socket_init(sk);
2058         return true;
2059 }
2060
2061 static u16 vmci_transport_new_proto_supported_versions(void)
2062 {
2063         if (PROTOCOL_OVERRIDE != -1)
2064                 return PROTOCOL_OVERRIDE;
2065
2066         return VSOCK_PROTO_ALL_SUPPORTED;
2067 }
2068
2069 static u32 vmci_transport_get_local_cid(void)
2070 {
2071         return vmci_get_context_id();
2072 }
2073
2074 static const struct vsock_transport vmci_transport = {
2075         .init = vmci_transport_socket_init,
2076         .destruct = vmci_transport_destruct,
2077         .release = vmci_transport_release,
2078         .connect = vmci_transport_connect,
2079         .dgram_bind = vmci_transport_dgram_bind,
2080         .dgram_dequeue = vmci_transport_dgram_dequeue,
2081         .dgram_enqueue = vmci_transport_dgram_enqueue,
2082         .dgram_allow = vmci_transport_dgram_allow,
2083         .stream_dequeue = vmci_transport_stream_dequeue,
2084         .stream_enqueue = vmci_transport_stream_enqueue,
2085         .stream_has_data = vmci_transport_stream_has_data,
2086         .stream_has_space = vmci_transport_stream_has_space,
2087         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2088         .stream_is_active = vmci_transport_stream_is_active,
2089         .stream_allow = vmci_transport_stream_allow,
2090         .notify_poll_in = vmci_transport_notify_poll_in,
2091         .notify_poll_out = vmci_transport_notify_poll_out,
2092         .notify_recv_init = vmci_transport_notify_recv_init,
2093         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2094         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2095         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2096         .notify_send_init = vmci_transport_notify_send_init,
2097         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2098         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2099         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2100         .shutdown = vmci_transport_shutdown,
2101         .set_buffer_size = vmci_transport_set_buffer_size,
2102         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2103         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2104         .get_buffer_size = vmci_transport_get_buffer_size,
2105         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2106         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2107         .get_local_cid = vmci_transport_get_local_cid,
2108 };
2109
2110 static int __init vmci_transport_init(void)
2111 {
2112         int err;
2113
2114         /* Create the datagram handle that we will use to send and receive all
2115          * VSocket control messages for this context.
2116          */
2117         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2118                                                  VMCI_FLAG_ANYCID_DG_HND,
2119                                                  vmci_transport_recv_stream_cb,
2120                                                  NULL,
2121                                                  &vmci_transport_stream_handle);
2122         if (err < VMCI_SUCCESS) {
2123                 pr_err("Unable to create datagram handle. (%d)\n", err);
2124                 return vmci_transport_error_to_vsock_error(err);
2125         }
2126
2127         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2128                                    vmci_transport_qp_resumed_cb,
2129                                    NULL, &vmci_transport_qp_resumed_sub_id);
2130         if (err < VMCI_SUCCESS) {
2131                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2132                 err = vmci_transport_error_to_vsock_error(err);
2133                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2134                 goto err_destroy_stream_handle;
2135         }
2136
2137         err = vsock_core_init(&vmci_transport);
2138         if (err < 0)
2139                 goto err_unsubscribe;
2140
2141         return 0;
2142
2143 err_unsubscribe:
2144         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2145 err_destroy_stream_handle:
2146         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2147         return err;
2148 }
2149 module_init(vmci_transport_init);
2150
2151 static void __exit vmci_transport_exit(void)
2152 {
2153         cancel_work_sync(&vmci_transport_cleanup_work);
2154         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2155
2156         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2157                 if (vmci_datagram_destroy_handle(
2158                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2159                         pr_err("Couldn't destroy datagram handle\n");
2160                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2161         }
2162
2163         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2164                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2165                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2166         }
2167
2168         vsock_core_exit();
2169 }
2170 module_exit(vmci_transport_exit);
2171
2172 MODULE_AUTHOR("VMware, Inc.");
2173 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2174 MODULE_VERSION("1.0.5.0-k");
2175 MODULE_LICENSE("GPL v2");
2176 MODULE_ALIAS("vmware_vsock");
2177 MODULE_ALIAS_NETPROTO(PF_VSOCK);