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 /* The default peer timeout indicates how long we will wait for a peer response
  82  * to a control message.
  83  */
  84 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
  85
  86 /* Helper function to convert from a VMCI error code to a VSock error code. */
  87
  88 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  89 {
  90         switch (vmci_error) {
  91         case VMCI_ERROR_NO_MEM:
  92                 return -ENOMEM;
  93         case VMCI_ERROR_DUPLICATE_ENTRY:
  94         case VMCI_ERROR_ALREADY_EXISTS:
  95                 return -EADDRINUSE;
  96         case VMCI_ERROR_NO_ACCESS:
  97                 return -EPERM;
  98         case VMCI_ERROR_NO_RESOURCES:
  99                 return -ENOBUFS;
 100         case VMCI_ERROR_INVALID_RESOURCE:
 101                 return -EHOSTUNREACH;
 102         case VMCI_ERROR_INVALID_ARGS:
 103         default:
 104                 break;
 105         }
 106         return -EINVAL;
 107 }
 108
 109 static u32 vmci_transport_peer_rid(u32 peer_cid)
 110 {
 111         if (VMADDR_CID_HYPERVISOR == peer_cid)
 112                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 113
 114         return VMCI_TRANSPORT_PACKET_RID;
 115 }
 116
 117 static inline void
 118 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 119                            struct sockaddr_vm *src,
 120                            struct sockaddr_vm *dst,
 121                            u8 type,
 122                            u64 size,
 123                            u64 mode,
 124                            struct vmci_transport_waiting_info *wait,
 125                            u16 proto,
 126                            struct vmci_handle handle)
 127 {
 128         /* We register the stream control handler as an any cid handle so we
 129          * must always send from a source address of VMADDR_CID_ANY
 130          */
 131         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 132                                        VMCI_TRANSPORT_PACKET_RID);
 133         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 134                                        vmci_transport_peer_rid(dst->svm_cid));
 135         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 136         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 137         pkt->type = type;
 138         pkt->src_port = src->svm_port;
 139         pkt->dst_port = dst->svm_port;
 140         memset(&pkt->proto, 0, sizeof(pkt->proto));
 141         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 142
 143         switch (pkt->type) {
 144         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 145                 pkt->u.size = 0;
 146                 break;
 147
 148         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 149         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 150                 pkt->u.size = size;
 151                 break;
 152
 153         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 154         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 155                 pkt->u.handle = handle;
 156                 break;
 157
 158         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 159         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 160         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 161                 pkt->u.size = 0;
 162                 break;
 163
 164         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 165                 pkt->u.mode = mode;
 166                 break;
 167
 168         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 169         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 170                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 171                 break;
 172
 173         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 174         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 175                 pkt->u.size = size;
 176                 pkt->proto = proto;
 177                 break;
 178         }
 179 }
 180
 181 static inline void
 182 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 183                                     struct sockaddr_vm *local,
 184                                     struct sockaddr_vm *remote)
 185 {
 186         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 187         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 188 }
 189
 190 static int
 191 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 192                                   struct sockaddr_vm *src,
 193                                   struct sockaddr_vm *dst,
 194                                   enum vmci_transport_packet_type type,
 195                                   u64 size,
 196                                   u64 mode,
 197                                   struct vmci_transport_waiting_info *wait,
 198                                   u16 proto,
 199                                   struct vmci_handle handle,
 200                                   bool convert_error)
 201 {
 202         int err;
 203
 204         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 205                                    proto, handle);
 206         err = vmci_datagram_send(&pkt->dg);
 207         if (convert_error && (err < 0))
 208                 return vmci_transport_error_to_vsock_error(err);
 209
 210         return err;
 211 }
 212
 213 static int
 214 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 215                                       enum vmci_transport_packet_type type,
 216                                       u64 size,
 217                                       u64 mode,
 218                                       struct vmci_transport_waiting_info *wait,
 219                                       struct vmci_handle handle)
 220 {
 221         struct vmci_transport_packet reply;
 222         struct sockaddr_vm src, dst;
 223
 224         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 225                 return 0;
 226         } else {
 227                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 228                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 229                                                          type,
 230                                                          size, mode, wait,
 231                                                          VSOCK_PROTO_INVALID,
 232                                                          handle, true);
 233         }
 234 }
 235
 236 static int
 237 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 238                                    struct sockaddr_vm *dst,
 239                                    enum vmci_transport_packet_type type,
 240                                    u64 size,
 241                                    u64 mode,
 242                                    struct vmci_transport_waiting_info *wait,
 243                                    struct vmci_handle handle)
 244 {
 245         /* Note that it is safe to use a single packet across all CPUs since
 246          * two tasklets of the same type are guaranteed to not ever run
 247          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 248          * we can use per-cpu packets.
 249          */
 250         static struct vmci_transport_packet pkt;
 251
 252         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 253                                                  size, mode, wait,
 254                                                  VSOCK_PROTO_INVALID, handle,
 255                                                  false);
 256 }
 257
 258 static int
 259 vmci_transport_alloc_send_control_pkt(struct sockaddr_vm *src,
 260                                       struct sockaddr_vm *dst,
 261                                       enum vmci_transport_packet_type type,
 262                                       u64 size,
 263                                       u64 mode,
 264                                       struct vmci_transport_waiting_info *wait,
 265                                       u16 proto,
 266                                       struct vmci_handle handle)
 267 {
 268         struct vmci_transport_packet *pkt;
 269         int err;
 270
 271         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 272         if (!pkt)
 273                 return -ENOMEM;
 274
 275         err = __vmci_transport_send_control_pkt(pkt, src, dst, type, size,
 276                                                 mode, wait, proto, handle,
 277                                                 true);
 278         kfree(pkt);
 279
 280         return err;
 281 }
 282
 283 static int
 284 vmci_transport_send_control_pkt(struct sock *sk,
 285                                 enum vmci_transport_packet_type type,
 286                                 u64 size,
 287                                 u64 mode,
 288                                 struct vmci_transport_waiting_info *wait,
 289                                 u16 proto,
 290                                 struct vmci_handle handle)
 291 {
 292         struct vsock_sock *vsk;
 293
 294         vsk = vsock_sk(sk);
 295
 296         if (!vsock_addr_bound(&vsk->local_addr))
 297                 return -EINVAL;
 298
 299         if (!vsock_addr_bound(&vsk->remote_addr))
 300                 return -EINVAL;
 301
 302         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr,
 303                                                      &vsk->remote_addr,
 304                                                      type, size, mode,
 305                                                      wait, proto, handle);
 306 }
 307
 308 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 309                                         struct sockaddr_vm *src,
 310                                         struct vmci_transport_packet *pkt)
 311 {
 312         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 313                 return 0;
 314         return vmci_transport_send_control_pkt_bh(
 315                                         dst, src,
 316                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 317                                         0, NULL, VMCI_INVALID_HANDLE);
 318 }
 319
 320 static int vmci_transport_send_reset(struct sock *sk,
 321                                      struct vmci_transport_packet *pkt)
 322 {
 323         struct sockaddr_vm *dst_ptr;
 324         struct sockaddr_vm dst;
 325         struct vsock_sock *vsk;
 326
 327         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 328                 return 0;
 329
 330         vsk = vsock_sk(sk);
 331
 332         if (!vsock_addr_bound(&vsk->local_addr))
 333                 return -EINVAL;
 334
 335         if (vsock_addr_bound(&vsk->remote_addr)) {
 336                 dst_ptr = &vsk->remote_addr;
 337         } else {
 338                 vsock_addr_init(&dst, pkt->dg.src.context,
 339                                 pkt->src_port);
 340                 dst_ptr = &dst;
 341         }
 342         return vmci_transport_alloc_send_control_pkt(&vsk->local_addr, dst_ptr,
 343                                              VMCI_TRANSPORT_PACKET_TYPE_RST,
 344                                              0, 0, NULL, VSOCK_PROTO_INVALID,
 345                                              VMCI_INVALID_HANDLE);
 346 }
 347
 348 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 349 {
 350         return vmci_transport_send_control_pkt(
 351                                         sk,
 352                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 353                                         size, 0, NULL,
 354                                         VSOCK_PROTO_INVALID,
 355                                         VMCI_INVALID_HANDLE);
 356 }
 357
 358 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 359                                           u16 version)
 360 {
 361         return vmci_transport_send_control_pkt(
 362                                         sk,
 363                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 364                                         size, 0, NULL, version,
 365                                         VMCI_INVALID_HANDLE);
 366 }
 367
 368 static int vmci_transport_send_qp_offer(struct sock *sk,
 369                                         struct vmci_handle handle)
 370 {
 371         return vmci_transport_send_control_pkt(
 372                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 373                                         0, NULL,
 374                                         VSOCK_PROTO_INVALID, handle);
 375 }
 376
 377 static int vmci_transport_send_attach(struct sock *sk,
 378                                       struct vmci_handle handle)
 379 {
 380         return vmci_transport_send_control_pkt(
 381                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 382                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 383                                         handle);
 384 }
 385
 386 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 387 {
 388         return vmci_transport_reply_control_pkt_fast(
 389                                                 pkt,
 390                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 391                                                 0, 0, NULL,
 392                                                 VMCI_INVALID_HANDLE);
 393 }
 394
 395 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 396                                           struct sockaddr_vm *src)
 397 {
 398         return vmci_transport_send_control_pkt_bh(
 399                                         dst, src,
 400                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 401                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 402 }
 403
 404 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 405                                  struct sockaddr_vm *src)
 406 {
 407         return vmci_transport_send_control_pkt_bh(
 408                                         dst, src,
 409                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 410                                         0, NULL, VMCI_INVALID_HANDLE);
 411 }
 412
 413 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 414                                 struct sockaddr_vm *src)
 415 {
 416         return vmci_transport_send_control_pkt_bh(
 417                                         dst, src,
 418                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 419                                         0, NULL, VMCI_INVALID_HANDLE);
 420 }
 421
 422 int vmci_transport_send_wrote(struct sock *sk)
 423 {
 424         return vmci_transport_send_control_pkt(
 425                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 426                                         0, NULL, VSOCK_PROTO_INVALID,
 427                                         VMCI_INVALID_HANDLE);
 428 }
 429
 430 int vmci_transport_send_read(struct sock *sk)
 431 {
 432         return vmci_transport_send_control_pkt(
 433                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 434                                         0, NULL, VSOCK_PROTO_INVALID,
 435                                         VMCI_INVALID_HANDLE);
 436 }
 437
 438 int vmci_transport_send_waiting_write(struct sock *sk,
 439                                       struct vmci_transport_waiting_info *wait)
 440 {
 441         return vmci_transport_send_control_pkt(
 442                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 443                                 0, 0, wait, VSOCK_PROTO_INVALID,
 444                                 VMCI_INVALID_HANDLE);
 445 }
 446
 447 int vmci_transport_send_waiting_read(struct sock *sk,
 448                                      struct vmci_transport_waiting_info *wait)
 449 {
 450         return vmci_transport_send_control_pkt(
 451                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 452                                 0, 0, wait, VSOCK_PROTO_INVALID,
 453                                 VMCI_INVALID_HANDLE);
 454 }
 455
 456 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 457 {
 458         return vmci_transport_send_control_pkt(
 459                                         &vsk->sk,
 460                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 461                                         0, mode, NULL,
 462                                         VSOCK_PROTO_INVALID,
 463                                         VMCI_INVALID_HANDLE);
 464 }
 465
 466 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 467 {
 468         return vmci_transport_send_control_pkt(sk,
 469                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 470                                         size, 0, NULL,
 471                                         VSOCK_PROTO_INVALID,
 472                                         VMCI_INVALID_HANDLE);
 473 }
 474
 475 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 476                                              u16 version)
 477 {
 478         return vmci_transport_send_control_pkt(
 479                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 480                                         size, 0, NULL, version,
 481                                         VMCI_INVALID_HANDLE);
 482 }
 483
 484 static struct sock *vmci_transport_get_pending(
 485                                         struct sock *listener,
 486                                         struct vmci_transport_packet *pkt)
 487 {
 488         struct vsock_sock *vlistener;
 489         struct vsock_sock *vpending;
 490         struct sock *pending;
 491         struct sockaddr_vm src;
 492
 493         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 494
 495         vlistener = vsock_sk(listener);
 496
 497         list_for_each_entry(vpending, &vlistener->pending_links,
 498                             pending_links) {
 499                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 500                     pkt->dst_port == vpending->local_addr.svm_port) {
 501                         pending = sk_vsock(vpending);
 502                         sock_hold(pending);
 503                         goto found;
 504                 }
 505         }
 506
 507         pending = NULL;
 508 found:
 509         return pending;
 510
 511 }
 512
 513 static void vmci_transport_release_pending(struct sock *pending)
 514 {
 515         sock_put(pending);
 516 }
 517
 518 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 519  * trusted sockets 2) sockets from applications running as the same user as the
 520  * VM (this is only true for the host side and only when using hosted products)
 521  */
 522
 523 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 524 {
 525         return vsock->trusted ||
 526                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 527 }
 528
 529 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 530  * only if it is trusted as described in vmci_transport_is_trusted.
 531  */
 532
 533 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 534 {
 535         if (VMADDR_CID_HYPERVISOR == peer_cid)
 536                 return true;
 537
 538         if (vsock->cached_peer != peer_cid) {
 539                 vsock->cached_peer = peer_cid;
 540                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 541                     (vmci_context_get_priv_flags(peer_cid) &
 542                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 543                         vsock->cached_peer_allow_dgram = false;
 544                 } else {
 545                         vsock->cached_peer_allow_dgram = true;
 546                 }
 547         }
 548
 549         return vsock->cached_peer_allow_dgram;
 550 }
 551
 552 static int
 553 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 554                                 struct vmci_handle *handle,
 555                                 u64 produce_size,
 556                                 u64 consume_size,
 557                                 u32 peer, u32 flags, bool trusted)
 558 {
 559         int err = 0;
 560
 561         if (trusted) {
 562                 /* Try to allocate our queue pair as trusted. This will only
 563                  * work if vsock is running in the host.
 564                  */
 565
 566                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 567                                        consume_size,
 568                                        peer, flags,
 569                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 570                 if (err != VMCI_ERROR_NO_ACCESS)
 571                         goto out;
 572
 573         }
 574
 575         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 576                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 577 out:
 578         if (err < 0) {
 579                 pr_err_once("Could not attach to queue pair with %d\n", err);
 580                 err = vmci_transport_error_to_vsock_error(err);
 581         }
 582
 583         return err;
 584 }
 585
 586 static int
 587 vmci_transport_datagram_create_hnd(u32 resource_id,
 588                                    u32 flags,
 589                                    vmci_datagram_recv_cb recv_cb,
 590                                    void *client_data,
 591                                    struct vmci_handle *out_handle)
 592 {
 593         int err = 0;
 594
 595         /* Try to allocate our datagram handler as trusted. This will only work
 596          * if vsock is running in the host.
 597          */
 598
 599         err = vmci_datagram_create_handle_priv(resource_id, flags,
 600                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 601                                                recv_cb,
 602                                                client_data, out_handle);
 603
 604         if (err == VMCI_ERROR_NO_ACCESS)
 605                 err = vmci_datagram_create_handle(resource_id, flags,
 606                                                   recv_cb, client_data,
 607                                                   out_handle);
 608
 609         return err;
 610 }
 611
 612 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 613  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 614  * sleep it should defer that work to a work queue.
 615  */
 616
 617 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 618 {
 619         struct sock *sk;
 620         size_t size;
 621         struct sk_buff *skb;
 622         struct vsock_sock *vsk;
 623
 624         sk = (struct sock *)data;
 625
 626         /* This handler is privileged when this module is running on the host.
 627          * We will get datagrams from all endpoints (even VMs that are in a
 628          * restricted context). If we get one from a restricted context then
 629          * the destination socket must be trusted.
 630          *
 631          * NOTE: We access the socket struct without holding the lock here.
 632          * This is ok because the field we are interested is never modified
 633          * outside of the create and destruct socket functions.
 634          */
 635         vsk = vsock_sk(sk);
 636         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 637                 return VMCI_ERROR_NO_ACCESS;
 638
 639         size = VMCI_DG_SIZE(dg);
 640
 641         /* Attach the packet to the socket's receive queue as an sk_buff. */
 642         skb = alloc_skb(size, GFP_ATOMIC);
 643         if (!skb)
 644                 return VMCI_ERROR_NO_MEM;
 645
 646         /* sk_receive_skb() will do a sock_put(), so hold here. */
 647         sock_hold(sk);
 648         skb_put(skb, size);
 649         memcpy(skb->data, dg, size);
 650         sk_receive_skb(sk, skb, 0);
 651
 652         return VMCI_SUCCESS;
 653 }
 654
 655 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 656 {
 657         static const u32 non_socket_contexts[] = {
 658                 VMADDR_CID_RESERVED,
 659         };
 660         int i;
 661
 662         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 663
 664         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 665                 if (cid == non_socket_contexts[i])
 666                         return false;
 667         }
 668
 669         return true;
 670 }
 671
 672 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 673  * interrupt fires.  This is run in bottom-half context but it defers most of
 674  * its work to the packet handling work queue.
 675  */
 676
 677 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 678 {
 679         struct sock *sk;
 680         struct sockaddr_vm dst;
 681         struct sockaddr_vm src;
 682         struct vmci_transport_packet *pkt;
 683         struct vsock_sock *vsk;
 684         bool bh_process_pkt;
 685         int err;
 686
 687         sk = NULL;
 688         err = VMCI_SUCCESS;
 689         bh_process_pkt = false;
 690
 691         /* Ignore incoming packets from contexts without sockets, or resources
 692          * that aren't vsock implementations.
 693          */
 694
 695         if (!vmci_transport_stream_allow(dg->src.context, -1)
 696             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 697                 return VMCI_ERROR_NO_ACCESS;
 698
 699         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 700                 /* Drop datagrams that do not contain full VSock packets. */
 701                 return VMCI_ERROR_INVALID_ARGS;
 702
 703         pkt = (struct vmci_transport_packet *)dg;
 704
 705         /* Find the socket that should handle this packet.  First we look for a
 706          * connected socket and if there is none we look for a socket bound to
 707          * the destintation address.
 708          */
 709         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 710         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 711
 712         sk = vsock_find_connected_socket(&src, &dst);
 713         if (!sk) {
 714                 sk = vsock_find_bound_socket(&dst);
 715                 if (!sk) {
 716                         /* We could not find a socket for this specified
 717                          * address.  If this packet is a RST, we just drop it.
 718                          * If it is another packet, we send a RST.  Note that
 719                          * we do not send a RST reply to RSTs so that we do not
 720                          * continually send RSTs between two endpoints.
 721                          *
 722                          * Note that since this is a reply, dst is src and src
 723                          * is dst.
 724                          */
 725                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 726                                 pr_err("unable to send reset\n");
 727
 728                         err = VMCI_ERROR_NOT_FOUND;
 729                         goto out;
 730                 }
 731         }
 732
 733         /* If the received packet type is beyond all types known to this
 734          * implementation, reply with an invalid message.  Hopefully this will
 735          * help when implementing backwards compatibility in the future.
 736          */
 737         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 738                 vmci_transport_send_invalid_bh(&dst, &src);
 739                 err = VMCI_ERROR_INVALID_ARGS;
 740                 goto out;
 741         }
 742
 743         /* This handler is privileged when this module is running on the host.
 744          * We will get datagram connect requests from all endpoints (even VMs
 745          * that are in a restricted context). If we get one from a restricted
 746          * context then the destination socket must be trusted.
 747          *
 748          * NOTE: We access the socket struct without holding the lock here.
 749          * This is ok because the field we are interested is never modified
 750          * outside of the create and destruct socket functions.
 751          */
 752         vsk = vsock_sk(sk);
 753         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 754                 err = VMCI_ERROR_NO_ACCESS;
 755                 goto out;
 756         }
 757
 758         /* We do most everything in a work queue, but let's fast path the
 759          * notification of reads and writes to help data transfer performance.
 760          * We can only do this if there is no process context code executing
 761          * for this socket since that may change the state.
 762          */
 763         bh_lock_sock(sk);
 764
 765         if (!sock_owned_by_user(sk)) {
 766                 /* The local context ID may be out of date, update it. */
 767                 vsk->local_addr.svm_cid = dst.svm_cid;
 768
 769                 if (sk->sk_state == TCP_ESTABLISHED)
 770                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 771                                         sk, pkt, true, &dst, &src,
 772                                         &bh_process_pkt);
 773         }
 774
 775         bh_unlock_sock(sk);
 776
 777         if (!bh_process_pkt) {
 778                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 779
 780                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 781                 if (!recv_pkt_info) {
 782                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 783                                 pr_err("unable to send reset\n");
 784
 785                         err = VMCI_ERROR_NO_MEM;
 786                         goto out;
 787                 }
 788
 789                 recv_pkt_info->sk = sk;
 790                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 791                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 792
 793                 schedule_work(&recv_pkt_info->work);
 794                 /* Clear sk so that the reference count incremented by one of
 795                  * the Find functions above is not decremented below.  We need
 796                  * that reference count for the packet handler we've scheduled
 797                  * to run.
 798                  */
 799                 sk = NULL;
 800         }
 801
 802 out:
 803         if (sk)
 804                 sock_put(sk);
 805
 806         return err;
 807 }
 808
 809 static void vmci_transport_handle_detach(struct sock *sk)
 810 {
 811         struct vsock_sock *vsk;
 812
 813         vsk = vsock_sk(sk);
 814         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 815                 sock_set_flag(sk, SOCK_DONE);
 816
 817                 /* On a detach the peer will not be sending or receiving
 818                  * anymore.
 819                  */
 820                 vsk->peer_shutdown = SHUTDOWN_MASK;
 821
 822                 /* We should not be sending anymore since the peer won't be
 823                  * there to receive, but we can still receive if there is data
 824                  * left in our consume queue. If the local endpoint is a host,
 825                  * we can't call vsock_stream_has_data, since that may block,
 826                  * but a host endpoint can't read data once the VM has
 827                  * detached, so there is no available data in that case.
 828                  */
 829                 if (vsk->local_addr.svm_cid == VMADDR_CID_HOST ||
 830                     vsock_stream_has_data(vsk) <= 0) {
 831                         if (sk->sk_state == TCP_SYN_SENT) {
 832                                 /* The peer may detach from a queue pair while
 833                                  * we are still in the connecting state, i.e.,
 834                                  * if the peer VM is killed after attaching to
 835                                  * a queue pair, but before we complete the
 836                                  * handshake. In that case, we treat the detach
 837                                  * event like a reset.
 838                                  */
 839
 840                                 sk->sk_state = TCP_CLOSE;
 841                                 sk->sk_err = ECONNRESET;
 842                                 sk->sk_error_report(sk);
 843                                 return;
 844                         }
 845                         sk->sk_state = TCP_CLOSE;
 846                 }
 847                 sk->sk_state_change(sk);
 848         }
 849 }
 850
 851 static void vmci_transport_peer_detach_cb(u32 sub_id,
 852                                           const struct vmci_event_data *e_data,
 853                                           void *client_data)
 854 {
 855         struct vmci_transport *trans = client_data;
 856         const struct vmci_event_payload_qp *e_payload;
 857
 858         e_payload = vmci_event_data_const_payload(e_data);
 859
 860         /* XXX This is lame, we should provide a way to lookup sockets by
 861          * qp_handle.
 862          */
 863         if (vmci_handle_is_invalid(e_payload->handle) ||
 864             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 865                 return;
 866
 867         /* We don't ask for delayed CBs when we subscribe to this event (we
 868          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 869          * guarantees in that case about what context we might be running in,
 870          * so it could be BH or process, blockable or non-blockable.  So we
 871          * need to account for all possible contexts here.
 872          */
 873         spin_lock_bh(&trans->lock);
 874         if (!trans->sk)
 875                 goto out;
 876
 877         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 878          * where trans->sk isn't locked.
 879          */
 880         bh_lock_sock(trans->sk);
 881
 882         vmci_transport_handle_detach(trans->sk);
 883
 884         bh_unlock_sock(trans->sk);
 885  out:
 886         spin_unlock_bh(&trans->lock);
 887 }
 888
 889 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 890                                          const struct vmci_event_data *e_data,
 891                                          void *client_data)
 892 {
 893         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 894 }
 895
 896 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 897 {
 898         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 899         struct vmci_transport_packet *pkt;
 900         struct sock *sk;
 901
 902         recv_pkt_info =
 903                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 904         sk = recv_pkt_info->sk;
 905         pkt = &recv_pkt_info->pkt;
 906
 907         lock_sock(sk);
 908
 909         /* The local context ID may be out of date. */
 910         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 911
 912         switch (sk->sk_state) {
 913         case TCP_LISTEN:
 914                 vmci_transport_recv_listen(sk, pkt);
 915                 break;
 916         case TCP_SYN_SENT:
 917                 /* Processing of pending connections for servers goes through
 918                  * the listening socket, so see vmci_transport_recv_listen()
 919                  * for that path.
 920                  */
 921                 vmci_transport_recv_connecting_client(sk, pkt);
 922                 break;
 923         case TCP_ESTABLISHED:
 924                 vmci_transport_recv_connected(sk, pkt);
 925                 break;
 926         default:
 927                 /* Because this function does not run in the same context as
 928                  * vmci_transport_recv_stream_cb it is possible that the
 929                  * socket has closed. We need to let the other side know or it
 930                  * could be sitting in a connect and hang forever. Send a
 931                  * reset to prevent that.
 932                  */
 933                 vmci_transport_send_reset(sk, pkt);
 934                 break;
 935         }
 936
 937         release_sock(sk);
 938         kfree(recv_pkt_info);
 939         /* Release reference obtained in the stream callback when we fetched
 940          * this socket out of the bound or connected list.
 941          */
 942         sock_put(sk);
 943 }
 944
 945 static int vmci_transport_recv_listen(struct sock *sk,
 946                                       struct vmci_transport_packet *pkt)
 947 {
 948         struct sock *pending;
 949         struct vsock_sock *vpending;
 950         int err;
 951         u64 qp_size;
 952         bool old_request = false;
 953         bool old_pkt_proto = false;
 954
 955         err = 0;
 956
 957         /* Because we are in the listen state, we could be receiving a packet
 958          * for ourself or any previous connection requests that we received.
 959          * If it's the latter, we try to find a socket in our list of pending
 960          * connections and, if we do, call the appropriate handler for the
 961          * state that that socket is in.  Otherwise we try to service the
 962          * connection request.
 963          */
 964         pending = vmci_transport_get_pending(sk, pkt);
 965         if (pending) {
 966                 lock_sock(pending);
 967
 968                 /* The local context ID may be out of date. */
 969                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 970
 971                 switch (pending->sk_state) {
 972                 case TCP_SYN_SENT:
 973                         err = vmci_transport_recv_connecting_server(sk,
 974                                                                     pending,
 975                                                                     pkt);
 976                         break;
 977                 default:
 978                         vmci_transport_send_reset(pending, pkt);
 979                         err = -EINVAL;
 980                 }
 981
 982                 if (err < 0)
 983                         vsock_remove_pending(sk, pending);
 984
 985                 release_sock(pending);
 986                 vmci_transport_release_pending(pending);
 987
 988                 return err;
 989         }
 990
 991         /* The listen state only accepts connection requests.  Reply with a
 992          * reset unless we received a reset.
 993          */
 994
 995         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 996               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 997                 vmci_transport_reply_reset(pkt);
 998                 return -EINVAL;
 999         }
1000
1001         if (pkt->u.size == 0) {
1002                 vmci_transport_reply_reset(pkt);
1003                 return -EINVAL;
1004         }
1005
1006         /* If this socket can't accommodate this connection request, we send a
1007          * reset.  Otherwise we create and initialize a child socket and reply
1008          * with a connection negotiation.
1009          */
1010         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
1011                 vmci_transport_reply_reset(pkt);
1012                 return -ECONNREFUSED;
1013         }
1014
1015         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
1016                                  sk->sk_type, 0);
1017         if (!pending) {
1018                 vmci_transport_send_reset(sk, pkt);
1019                 return -ENOMEM;
1020         }
1021
1022         vpending = vsock_sk(pending);
1023
1024         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
1025                         pkt->dst_port);
1026         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1027                         pkt->src_port);
1028
1029         /* If the proposed size fits within our min/max, accept it. Otherwise
1030          * propose our own size.
1031          */
1032         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1033             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1034                 qp_size = pkt->u.size;
1035         } else {
1036                 qp_size = vmci_trans(vpending)->queue_pair_size;
1037         }
1038
1039         /* Figure out if we are using old or new requests based on the
1040          * overrides pkt types sent by our peer.
1041          */
1042         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1043                 old_request = old_pkt_proto;
1044         } else {
1045                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1046                         old_request = true;
1047                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1048                         old_request = false;
1049
1050         }
1051
1052         if (old_request) {
1053                 /* Handle a REQUEST (or override) */
1054                 u16 version = VSOCK_PROTO_INVALID;
1055                 if (vmci_transport_proto_to_notify_struct(
1056                         pending, &version, true))
1057                         err = vmci_transport_send_negotiate(pending, qp_size);
1058                 else
1059                         err = -EINVAL;
1060
1061         } else {
1062                 /* Handle a REQUEST2 (or override) */
1063                 int proto_int = pkt->proto;
1064                 int pos;
1065                 u16 active_proto_version = 0;
1066
1067                 /* The list of possible protocols is the intersection of all
1068                  * protocols the client supports ... plus all the protocols we
1069                  * support.
1070                  */
1071                 proto_int &= vmci_transport_new_proto_supported_versions();
1072
1073                 /* We choose the highest possible protocol version and use that
1074                  * one.
1075                  */
1076                 pos = fls(proto_int);
1077                 if (pos) {
1078                         active_proto_version = (1 << (pos - 1));
1079                         if (vmci_transport_proto_to_notify_struct(
1080                                 pending, &active_proto_version, false))
1081                                 err = vmci_transport_send_negotiate2(pending,
1082                                                         qp_size,
1083                                                         active_proto_version);
1084                         else
1085                                 err = -EINVAL;
1086
1087                 } else {
1088                         err = -EINVAL;
1089                 }
1090         }
1091
1092         if (err < 0) {
1093                 vmci_transport_send_reset(sk, pkt);
1094                 sock_put(pending);
1095                 err = vmci_transport_error_to_vsock_error(err);
1096                 goto out;
1097         }
1098
1099         vsock_add_pending(sk, pending);
1100         sk->sk_ack_backlog++;
1101
1102         pending->sk_state = TCP_SYN_SENT;
1103         vmci_trans(vpending)->produce_size =
1104                 vmci_trans(vpending)->consume_size = qp_size;
1105         vmci_trans(vpending)->queue_pair_size = qp_size;
1106
1107         vmci_trans(vpending)->notify_ops->process_request(pending);
1108
1109         /* We might never receive another message for this socket and it's not
1110          * connected to any process, so we have to ensure it gets cleaned up
1111          * ourself.  Our delayed work function will take care of that.  Note
1112          * that we do not ever cancel this function since we have few
1113          * guarantees about its state when calling cancel_delayed_work().
1114          * Instead we hold a reference on the socket for that function and make
1115          * it capable of handling cases where it needs to do nothing but
1116          * release that reference.
1117          */
1118         vpending->listener = sk;
1119         sock_hold(sk);
1120         sock_hold(pending);
1121         schedule_delayed_work(&vpending->pending_work, HZ);
1122
1123 out:
1124         return err;
1125 }
1126
1127 static int
1128 vmci_transport_recv_connecting_server(struct sock *listener,
1129                                       struct sock *pending,
1130                                       struct vmci_transport_packet *pkt)
1131 {
1132         struct vsock_sock *vpending;
1133         struct vmci_handle handle;
1134         struct vmci_qp *qpair;
1135         bool is_local;
1136         u32 flags;
1137         u32 detach_sub_id;
1138         int err;
1139         int skerr;
1140
1141         vpending = vsock_sk(pending);
1142         detach_sub_id = VMCI_INVALID_ID;
1143
1144         switch (pkt->type) {
1145         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1146                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1147                         vmci_transport_send_reset(pending, pkt);
1148                         skerr = EPROTO;
1149                         err = -EINVAL;
1150                         goto destroy;
1151                 }
1152                 break;
1153         default:
1154                 /* Close and cleanup the connection. */
1155                 vmci_transport_send_reset(pending, pkt);
1156                 skerr = EPROTO;
1157                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1158                 goto destroy;
1159         }
1160
1161         /* In order to complete the connection we need to attach to the offered
1162          * queue pair and send an attach notification.  We also subscribe to the
1163          * detach event so we know when our peer goes away, and we do that
1164          * before attaching so we don't miss an event.  If all this succeeds,
1165          * we update our state and wakeup anything waiting in accept() for a
1166          * connection.
1167          */
1168
1169         /* We don't care about attach since we ensure the other side has
1170          * attached by specifying the ATTACH_ONLY flag below.
1171          */
1172         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1173                                    vmci_transport_peer_detach_cb,
1174                                    vmci_trans(vpending), &detach_sub_id);
1175         if (err < VMCI_SUCCESS) {
1176                 vmci_transport_send_reset(pending, pkt);
1177                 err = vmci_transport_error_to_vsock_error(err);
1178                 skerr = -err;
1179                 goto destroy;
1180         }
1181
1182         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1183
1184         /* Now attach to the queue pair the client created. */
1185         handle = pkt->u.handle;
1186
1187         /* vpending->local_addr always has a context id so we do not need to
1188          * worry about VMADDR_CID_ANY in this case.
1189          */
1190         is_local =
1191             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1192         flags = VMCI_QPFLAG_ATTACH_ONLY;
1193         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1194
1195         err = vmci_transport_queue_pair_alloc(
1196                                         &qpair,
1197                                         &handle,
1198                                         vmci_trans(vpending)->produce_size,
1199                                         vmci_trans(vpending)->consume_size,
1200                                         pkt->dg.src.context,
1201                                         flags,
1202                                         vmci_transport_is_trusted(
1203                                                 vpending,
1204                                                 vpending->remote_addr.svm_cid));
1205         if (err < 0) {
1206                 vmci_transport_send_reset(pending, pkt);
1207                 skerr = -err;
1208                 goto destroy;
1209         }
1210
1211         vmci_trans(vpending)->qp_handle = handle;
1212         vmci_trans(vpending)->qpair = qpair;
1213
1214         /* When we send the attach message, we must be ready to handle incoming
1215          * control messages on the newly connected socket. So we move the
1216          * pending socket to the connected state before sending the attach
1217          * message. Otherwise, an incoming packet triggered by the attach being
1218          * received by the peer may be processed concurrently with what happens
1219          * below after sending the attach message, and that incoming packet
1220          * will find the listening socket instead of the (currently) pending
1221          * socket. Note that enqueueing the socket increments the reference
1222          * count, so even if a reset comes before the connection is accepted,
1223          * the socket will be valid until it is removed from the queue.
1224          *
1225          * If we fail sending the attach below, we remove the socket from the
1226          * connected list and move the socket to TCP_CLOSE before
1227          * releasing the lock, so a pending slow path processing of an incoming
1228          * packet will not see the socket in the connected state in that case.
1229          */
1230         pending->sk_state = TCP_ESTABLISHED;
1231
1232         vsock_insert_connected(vpending);
1233
1234         /* Notify our peer of our attach. */
1235         err = vmci_transport_send_attach(pending, handle);
1236         if (err < 0) {
1237                 vsock_remove_connected(vpending);
1238                 pr_err("Could not send attach\n");
1239                 vmci_transport_send_reset(pending, pkt);
1240                 err = vmci_transport_error_to_vsock_error(err);
1241                 skerr = -err;
1242                 goto destroy;
1243         }
1244
1245         /* We have a connection. Move the now connected socket from the
1246          * listener's pending list to the accept queue so callers of accept()
1247          * can find it.
1248          */
1249         vsock_remove_pending(listener, pending);
1250         vsock_enqueue_accept(listener, pending);
1251
1252         /* Callers of accept() will be be waiting on the listening socket, not
1253          * the pending socket.
1254          */
1255         listener->sk_data_ready(listener);
1256
1257         return 0;
1258
1259 destroy:
1260         pending->sk_err = skerr;
1261         pending->sk_state = TCP_CLOSE;
1262         /* As long as we drop our reference, all necessary cleanup will handle
1263          * when the cleanup function drops its reference and our destruct
1264          * implementation is called.  Note that since the listen handler will
1265          * remove pending from the pending list upon our failure, the cleanup
1266          * function won't drop the additional reference, which is why we do it
1267          * here.
1268          */
1269         sock_put(pending);
1270
1271         return err;
1272 }
1273
1274 static int
1275 vmci_transport_recv_connecting_client(struct sock *sk,
1276                                       struct vmci_transport_packet *pkt)
1277 {
1278         struct vsock_sock *vsk;
1279         int err;
1280         int skerr;
1281
1282         vsk = vsock_sk(sk);
1283
1284         switch (pkt->type) {
1285         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1286                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1287                     !vmci_handle_is_equal(pkt->u.handle,
1288                                           vmci_trans(vsk)->qp_handle)) {
1289                         skerr = EPROTO;
1290                         err = -EINVAL;
1291                         goto destroy;
1292                 }
1293
1294                 /* Signify the socket is connected and wakeup the waiter in
1295                  * connect(). Also place the socket in the connected table for
1296                  * accounting (it can already be found since it's in the bound
1297                  * table).
1298                  */
1299                 sk->sk_state = TCP_ESTABLISHED;
1300                 sk->sk_socket->state = SS_CONNECTED;
1301                 vsock_insert_connected(vsk);
1302                 sk->sk_state_change(sk);
1303
1304                 break;
1305         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1306         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1307                 if (pkt->u.size == 0
1308                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1309                     || pkt->src_port != vsk->remote_addr.svm_port
1310                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1311                     || vmci_trans(vsk)->qpair
1312                     || vmci_trans(vsk)->produce_size != 0
1313                     || vmci_trans(vsk)->consume_size != 0
1314                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1315                         skerr = EPROTO;
1316                         err = -EINVAL;
1317
1318                         goto destroy;
1319                 }
1320
1321                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1322                 if (err) {
1323                         skerr = -err;
1324                         goto destroy;
1325                 }
1326
1327                 break;
1328         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1329                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1330                 if (err) {
1331                         skerr = -err;
1332                         goto destroy;
1333                 }
1334
1335                 break;
1336         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1337                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1338                  * continue processing here after they sent an INVALID packet.
1339                  * This meant that we got a RST after the INVALID. We ignore a
1340                  * RST after an INVALID. The common code doesn't send the RST
1341                  * ... so we can hang if an old version of the common code
1342                  * fails between getting a REQUEST and sending an OFFER back.
1343                  * Not much we can do about it... except hope that it doesn't
1344                  * happen.
1345                  */
1346                 if (vsk->ignore_connecting_rst) {
1347                         vsk->ignore_connecting_rst = false;
1348                 } else {
1349                         skerr = ECONNRESET;
1350                         err = 0;
1351                         goto destroy;
1352                 }
1353
1354                 break;
1355         default:
1356                 /* Close and cleanup the connection. */
1357                 skerr = EPROTO;
1358                 err = -EINVAL;
1359                 goto destroy;
1360         }
1361
1362         return 0;
1363
1364 destroy:
1365         vmci_transport_send_reset(sk, pkt);
1366
1367         sk->sk_state = TCP_CLOSE;
1368         sk->sk_err = skerr;
1369         sk->sk_error_report(sk);
1370         return err;
1371 }
1372
1373 static int vmci_transport_recv_connecting_client_negotiate(
1374                                         struct sock *sk,
1375                                         struct vmci_transport_packet *pkt)
1376 {
1377         int err;
1378         struct vsock_sock *vsk;
1379         struct vmci_handle handle;
1380         struct vmci_qp *qpair;
1381         u32 detach_sub_id;
1382         bool is_local;
1383         u32 flags;
1384         bool old_proto = true;
1385         bool old_pkt_proto;
1386         u16 version;
1387
1388         vsk = vsock_sk(sk);
1389         handle = VMCI_INVALID_HANDLE;
1390         detach_sub_id = VMCI_INVALID_ID;
1391
1392         /* If we have gotten here then we should be past the point where old
1393          * linux vsock could have sent the bogus rst.
1394          */
1395         vsk->sent_request = false;
1396         vsk->ignore_connecting_rst = false;
1397
1398         /* Verify that we're OK with the proposed queue pair size */
1399         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1400             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1401                 err = -EINVAL;
1402                 goto destroy;
1403         }
1404
1405         /* At this point we know the CID the peer is using to talk to us. */
1406
1407         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1408                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1409
1410         /* Setup the notify ops to be the highest supported version that both
1411          * the server and the client support.
1412          */
1413
1414         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1415                 old_proto = old_pkt_proto;
1416         } else {
1417                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1418                         old_proto = true;
1419                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1420                         old_proto = false;
1421
1422         }
1423
1424         if (old_proto)
1425                 version = VSOCK_PROTO_INVALID;
1426         else
1427                 version = pkt->proto;
1428
1429         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1430                 err = -EINVAL;
1431                 goto destroy;
1432         }
1433
1434         /* Subscribe to detach events first.
1435          *
1436          * XXX We attach once for each queue pair created for now so it is easy
1437          * to find the socket (it's provided), but later we should only
1438          * subscribe once and add a way to lookup sockets by queue pair handle.
1439          */
1440         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1441                                    vmci_transport_peer_detach_cb,
1442                                    vmci_trans(vsk), &detach_sub_id);
1443         if (err < VMCI_SUCCESS) {
1444                 err = vmci_transport_error_to_vsock_error(err);
1445                 goto destroy;
1446         }
1447
1448         /* Make VMCI select the handle for us. */
1449         handle = VMCI_INVALID_HANDLE;
1450         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1451         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1452
1453         err = vmci_transport_queue_pair_alloc(&qpair,
1454                                               &handle,
1455                                               pkt->u.size,
1456                                               pkt->u.size,
1457                                               vsk->remote_addr.svm_cid,
1458                                               flags,
1459                                               vmci_transport_is_trusted(
1460                                                   vsk,
1461                                                   vsk->
1462                                                   remote_addr.svm_cid));
1463         if (err < 0)
1464                 goto destroy;
1465
1466         err = vmci_transport_send_qp_offer(sk, handle);
1467         if (err < 0) {
1468                 err = vmci_transport_error_to_vsock_error(err);
1469                 goto destroy;
1470         }
1471
1472         vmci_trans(vsk)->qp_handle = handle;
1473         vmci_trans(vsk)->qpair = qpair;
1474
1475         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1476                 pkt->u.size;
1477
1478         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1479
1480         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1481
1482         return 0;
1483
1484 destroy:
1485         if (detach_sub_id != VMCI_INVALID_ID)
1486                 vmci_event_unsubscribe(detach_sub_id);
1487
1488         if (!vmci_handle_is_invalid(handle))
1489                 vmci_qpair_detach(&qpair);
1490
1491         return err;
1492 }
1493
1494 static int
1495 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1496                                               struct vmci_transport_packet *pkt)
1497 {
1498         int err = 0;
1499         struct vsock_sock *vsk = vsock_sk(sk);
1500
1501         if (vsk->sent_request) {
1502                 vsk->sent_request = false;
1503                 vsk->ignore_connecting_rst = true;
1504
1505                 err = vmci_transport_send_conn_request(
1506                         sk, vmci_trans(vsk)->queue_pair_size);
1507                 if (err < 0)
1508                         err = vmci_transport_error_to_vsock_error(err);
1509                 else
1510                         err = 0;
1511
1512         }
1513
1514         return err;
1515 }
1516
1517 static int vmci_transport_recv_connected(struct sock *sk,
1518                                          struct vmci_transport_packet *pkt)
1519 {
1520         struct vsock_sock *vsk;
1521         bool pkt_processed = false;
1522
1523         /* In cases where we are closing the connection, it's sufficient to
1524          * mark the state change (and maybe error) and wake up any waiting
1525          * threads. Since this is a connected socket, it's owned by a user
1526          * process and will be cleaned up when the failure is passed back on
1527          * the current or next system call.  Our system call implementations
1528          * must therefore check for error and state changes on entry and when
1529          * being awoken.
1530          */
1531         switch (pkt->type) {
1532         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1533                 if (pkt->u.mode) {
1534                         vsk = vsock_sk(sk);
1535
1536                         vsk->peer_shutdown |= pkt->u.mode;
1537                         sk->sk_state_change(sk);
1538                 }
1539                 break;
1540
1541         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1542                 vsk = vsock_sk(sk);
1543                 /* It is possible that we sent our peer a message (e.g a
1544                  * WAITING_READ) right before we got notified that the peer had
1545                  * detached. If that happens then we can get a RST pkt back
1546                  * from our peer even though there is data available for us to
1547                  * read. In that case, don't shutdown the socket completely but
1548                  * instead allow the local client to finish reading data off
1549                  * the queuepair. Always treat a RST pkt in connected mode like
1550                  * a clean shutdown.
1551                  */
1552                 sock_set_flag(sk, SOCK_DONE);
1553                 vsk->peer_shutdown = SHUTDOWN_MASK;
1554                 if (vsock_stream_has_data(vsk) <= 0)
1555                         sk->sk_state = TCP_CLOSING;
1556
1557                 sk->sk_state_change(sk);
1558                 break;
1559
1560         default:
1561                 vsk = vsock_sk(sk);
1562                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1563                                 sk, pkt, false, NULL, NULL,
1564                                 &pkt_processed);
1565                 if (!pkt_processed)
1566                         return -EINVAL;
1567
1568                 break;
1569         }
1570
1571         return 0;
1572 }
1573
1574 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1575                                       struct vsock_sock *psk)
1576 {
1577         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1578         if (!vsk->trans)
1579                 return -ENOMEM;
1580
1581         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1582         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1583         vmci_trans(vsk)->qpair = NULL;
1584         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1585         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1586         vmci_trans(vsk)->notify_ops = NULL;
1587         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1588         vmci_trans(vsk)->sk = &vsk->sk;
1589         spin_lock_init(&vmci_trans(vsk)->lock);
1590         if (psk) {
1591                 vmci_trans(vsk)->queue_pair_size =
1592                         vmci_trans(psk)->queue_pair_size;
1593                 vmci_trans(vsk)->queue_pair_min_size =
1594                         vmci_trans(psk)->queue_pair_min_size;
1595                 vmci_trans(vsk)->queue_pair_max_size =
1596                         vmci_trans(psk)->queue_pair_max_size;
1597         } else {
1598                 vmci_trans(vsk)->queue_pair_size =
1599                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1600                 vmci_trans(vsk)->queue_pair_min_size =
1601                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1602                 vmci_trans(vsk)->queue_pair_max_size =
1603                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1604         }
1605
1606         return 0;
1607 }
1608
1609 static void vmci_transport_free_resources(struct list_head *transport_list)
1610 {
1611         while (!list_empty(transport_list)) {
1612                 struct vmci_transport *transport =
1613                     list_first_entry(transport_list, struct vmci_transport,
1614                                      elem);
1615                 list_del(&transport->elem);
1616
1617                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1618                         vmci_event_unsubscribe(transport->detach_sub_id);
1619                         transport->detach_sub_id = VMCI_INVALID_ID;
1620                 }
1621
1622                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1623                         vmci_qpair_detach(&transport->qpair);
1624                         transport->qp_handle = VMCI_INVALID_HANDLE;
1625                         transport->produce_size = 0;
1626                         transport->consume_size = 0;
1627                 }
1628
1629                 kfree(transport);
1630         }
1631 }
1632
1633 static void vmci_transport_cleanup(struct work_struct *work)
1634 {
1635         LIST_HEAD(pending);
1636
1637         spin_lock_bh(&vmci_transport_cleanup_lock);
1638         list_replace_init(&vmci_transport_cleanup_list, &pending);
1639         spin_unlock_bh(&vmci_transport_cleanup_lock);
1640         vmci_transport_free_resources(&pending);
1641 }
1642
1643 static void vmci_transport_destruct(struct vsock_sock *vsk)
1644 {
1645         /* transport can be NULL if we hit a failure at init() time */
1646         if (!vmci_trans(vsk))
1647                 return;
1648
1649         /* Ensure that the detach callback doesn't use the sk/vsk
1650          * we are about to destruct.
1651          */
1652         spin_lock_bh(&vmci_trans(vsk)->lock);
1653         vmci_trans(vsk)->sk = NULL;
1654         spin_unlock_bh(&vmci_trans(vsk)->lock);
1655
1656         if (vmci_trans(vsk)->notify_ops)
1657                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1658
1659         spin_lock_bh(&vmci_transport_cleanup_lock);
1660         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1661         spin_unlock_bh(&vmci_transport_cleanup_lock);
1662         schedule_work(&vmci_transport_cleanup_work);
1663
1664         vsk->trans = NULL;
1665 }
1666
1667 static void vmci_transport_release(struct vsock_sock *vsk)
1668 {
1669         vsock_remove_sock(vsk);
1670
1671         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1672                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1673                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1674         }
1675 }
1676
1677 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1678                                      struct sockaddr_vm *addr)
1679 {
1680         u32 port;
1681         u32 flags;
1682         int err;
1683
1684         /* VMCI will select a resource ID for us if we provide
1685          * VMCI_INVALID_ID.
1686          */
1687         port = addr->svm_port == VMADDR_PORT_ANY ?
1688                         VMCI_INVALID_ID : addr->svm_port;
1689
1690         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1691                 return -EACCES;
1692
1693         flags = addr->svm_cid == VMADDR_CID_ANY ?
1694                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1695
1696         err = vmci_transport_datagram_create_hnd(port, flags,
1697                                                  vmci_transport_recv_dgram_cb,
1698                                                  &vsk->sk,
1699                                                  &vmci_trans(vsk)->dg_handle);
1700         if (err < VMCI_SUCCESS)
1701                 return vmci_transport_error_to_vsock_error(err);
1702         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1703                         vmci_trans(vsk)->dg_handle.resource);
1704
1705         return 0;
1706 }
1707
1708 static int vmci_transport_dgram_enqueue(
1709         struct vsock_sock *vsk,
1710         struct sockaddr_vm *remote_addr,
1711         struct msghdr *msg,
1712         size_t len)
1713 {
1714         int err;
1715         struct vmci_datagram *dg;
1716
1717         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1718                 return -EMSGSIZE;
1719
1720         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1721                 return -EPERM;
1722
1723         /* Allocate a buffer for the user's message and our packet header. */
1724         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1725         if (!dg)
1726                 return -ENOMEM;
1727
1728         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1729
1730         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1731                                    remote_addr->svm_port);
1732         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1733                                    vsk->local_addr.svm_port);
1734         dg->payload_size = len;
1735
1736         err = vmci_datagram_send(dg);
1737         kfree(dg);
1738         if (err < 0)
1739                 return vmci_transport_error_to_vsock_error(err);
1740
1741         return err - sizeof(*dg);
1742 }
1743
1744 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1745                                         struct msghdr *msg, size_t len,
1746                                         int flags)
1747 {
1748         int err;
1749         int noblock;
1750         struct vmci_datagram *dg;
1751         size_t payload_len;
1752         struct sk_buff *skb;
1753
1754         noblock = flags & MSG_DONTWAIT;
1755
1756         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1757                 return -EOPNOTSUPP;
1758
1759         /* Retrieve the head sk_buff from the socket's receive queue. */
1760         err = 0;
1761         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1762         if (!skb)
1763                 return err;
1764
1765         dg = (struct vmci_datagram *)skb->data;
1766         if (!dg)
1767                 /* err is 0, meaning we read zero bytes. */
1768                 goto out;
1769
1770         payload_len = dg->payload_size;
1771         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1772         if (payload_len != skb->len - sizeof(*dg)) {
1773                 err = -EINVAL;
1774                 goto out;
1775         }
1776
1777         if (payload_len > len) {
1778                 payload_len = len;
1779                 msg->msg_flags |= MSG_TRUNC;
1780         }
1781
1782         /* Place the datagram payload in the user's iovec. */
1783         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1784         if (err)
1785                 goto out;
1786
1787         if (msg->msg_name) {
1788                 /* Provide the address of the sender. */
1789                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1790                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1791                 msg->msg_namelen = sizeof(*vm_addr);
1792         }
1793         err = payload_len;
1794
1795 out:
1796         skb_free_datagram(&vsk->sk, skb);
1797         return err;
1798 }
1799
1800 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1801 {
1802         if (cid == VMADDR_CID_HYPERVISOR) {
1803                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1804                  * state and are allowed.
1805                  */
1806                 return port == VMCI_UNITY_PBRPC_REGISTER;
1807         }
1808
1809         return true;
1810 }
1811
1812 static int vmci_transport_connect(struct vsock_sock *vsk)
1813 {
1814         int err;
1815         bool old_pkt_proto = false;
1816         struct sock *sk = &vsk->sk;
1817
1818         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1819                 old_pkt_proto) {
1820                 err = vmci_transport_send_conn_request(
1821                         sk, vmci_trans(vsk)->queue_pair_size);
1822                 if (err < 0) {
1823                         sk->sk_state = TCP_CLOSE;
1824                         return err;
1825                 }
1826         } else {
1827                 int supported_proto_versions =
1828                         vmci_transport_new_proto_supported_versions();
1829                 err = vmci_transport_send_conn_request2(
1830                                 sk, vmci_trans(vsk)->queue_pair_size,
1831                                 supported_proto_versions);
1832                 if (err < 0) {
1833                         sk->sk_state = TCP_CLOSE;
1834                         return err;
1835                 }
1836
1837                 vsk->sent_request = true;
1838         }
1839
1840         return err;
1841 }
1842
1843 static ssize_t vmci_transport_stream_dequeue(
1844         struct vsock_sock *vsk,
1845         struct msghdr *msg,
1846         size_t len,
1847         int flags)
1848 {
1849         if (flags & MSG_PEEK)
1850                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1851         else
1852                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1853 }
1854
1855 static ssize_t vmci_transport_stream_enqueue(
1856         struct vsock_sock *vsk,
1857         struct msghdr *msg,
1858         size_t len)
1859 {
1860         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1861 }
1862
1863 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1864 {
1865         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1866 }
1867
1868 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1869 {
1870         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1871 }
1872
1873 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1874 {
1875         return vmci_trans(vsk)->consume_size;
1876 }
1877
1878 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1879 {
1880         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1881 }
1882
1883 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1884 {
1885         return vmci_trans(vsk)->queue_pair_size;
1886 }
1887
1888 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1889 {
1890         return vmci_trans(vsk)->queue_pair_min_size;
1891 }
1892
1893 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1894 {
1895         return vmci_trans(vsk)->queue_pair_max_size;
1896 }
1897
1898 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1899 {
1900         if (val < vmci_trans(vsk)->queue_pair_min_size)
1901                 vmci_trans(vsk)->queue_pair_min_size = val;
1902         if (val > vmci_trans(vsk)->queue_pair_max_size)
1903                 vmci_trans(vsk)->queue_pair_max_size = val;
1904         vmci_trans(vsk)->queue_pair_size = val;
1905 }
1906
1907 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1908                                                u64 val)
1909 {
1910         if (val > vmci_trans(vsk)->queue_pair_size)
1911                 vmci_trans(vsk)->queue_pair_size = val;
1912         vmci_trans(vsk)->queue_pair_min_size = val;
1913 }
1914
1915 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1916                                                u64 val)
1917 {
1918         if (val < vmci_trans(vsk)->queue_pair_size)
1919                 vmci_trans(vsk)->queue_pair_size = val;
1920         vmci_trans(vsk)->queue_pair_max_size = val;
1921 }
1922
1923 static int vmci_transport_notify_poll_in(
1924         struct vsock_sock *vsk,
1925         size_t target,
1926         bool *data_ready_now)
1927 {
1928         return vmci_trans(vsk)->notify_ops->poll_in(
1929                         &vsk->sk, target, data_ready_now);
1930 }
1931
1932 static int vmci_transport_notify_poll_out(
1933         struct vsock_sock *vsk,
1934         size_t target,
1935         bool *space_available_now)
1936 {
1937         return vmci_trans(vsk)->notify_ops->poll_out(
1938                         &vsk->sk, target, space_available_now);
1939 }
1940
1941 static int vmci_transport_notify_recv_init(
1942         struct vsock_sock *vsk,
1943         size_t target,
1944         struct vsock_transport_recv_notify_data *data)
1945 {
1946         return vmci_trans(vsk)->notify_ops->recv_init(
1947                         &vsk->sk, target,
1948                         (struct vmci_transport_recv_notify_data *)data);
1949 }
1950
1951 static int vmci_transport_notify_recv_pre_block(
1952         struct vsock_sock *vsk,
1953         size_t target,
1954         struct vsock_transport_recv_notify_data *data)
1955 {
1956         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1957                         &vsk->sk, target,
1958                         (struct vmci_transport_recv_notify_data *)data);
1959 }
1960
1961 static int vmci_transport_notify_recv_pre_dequeue(
1962         struct vsock_sock *vsk,
1963         size_t target,
1964         struct vsock_transport_recv_notify_data *data)
1965 {
1966         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1967                         &vsk->sk, target,
1968                         (struct vmci_transport_recv_notify_data *)data);
1969 }
1970
1971 static int vmci_transport_notify_recv_post_dequeue(
1972         struct vsock_sock *vsk,
1973         size_t target,
1974         ssize_t copied,
1975         bool data_read,
1976         struct vsock_transport_recv_notify_data *data)
1977 {
1978         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1979                         &vsk->sk, target, copied, data_read,
1980                         (struct vmci_transport_recv_notify_data *)data);
1981 }
1982
1983 static int vmci_transport_notify_send_init(
1984         struct vsock_sock *vsk,
1985         struct vsock_transport_send_notify_data *data)
1986 {
1987         return vmci_trans(vsk)->notify_ops->send_init(
1988                         &vsk->sk,
1989                         (struct vmci_transport_send_notify_data *)data);
1990 }
1991
1992 static int vmci_transport_notify_send_pre_block(
1993         struct vsock_sock *vsk,
1994         struct vsock_transport_send_notify_data *data)
1995 {
1996         return vmci_trans(vsk)->notify_ops->send_pre_block(
1997                         &vsk->sk,
1998                         (struct vmci_transport_send_notify_data *)data);
1999 }
2000
2001 static int vmci_transport_notify_send_pre_enqueue(
2002         struct vsock_sock *vsk,
2003         struct vsock_transport_send_notify_data *data)
2004 {
2005         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
2006                         &vsk->sk,
2007                         (struct vmci_transport_send_notify_data *)data);
2008 }
2009
2010 static int vmci_transport_notify_send_post_enqueue(
2011         struct vsock_sock *vsk,
2012         ssize_t written,
2013         struct vsock_transport_send_notify_data *data)
2014 {
2015         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
2016                         &vsk->sk, written,
2017                         (struct vmci_transport_send_notify_data *)data);
2018 }
2019
2020 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
2021 {
2022         if (PROTOCOL_OVERRIDE != -1) {
2023                 if (PROTOCOL_OVERRIDE == 0)
2024                         *old_pkt_proto = true;
2025                 else
2026                         *old_pkt_proto = false;
2027
2028                 pr_info("Proto override in use\n");
2029                 return true;
2030         }
2031
2032         return false;
2033 }
2034
2035 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2036                                                   u16 *proto,
2037                                                   bool old_pkt_proto)
2038 {
2039         struct vsock_sock *vsk = vsock_sk(sk);
2040
2041         if (old_pkt_proto) {
2042                 if (*proto != VSOCK_PROTO_INVALID) {
2043                         pr_err("Can't set both an old and new protocol\n");
2044                         return false;
2045                 }
2046                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2047                 goto exit;
2048         }
2049
2050         switch (*proto) {
2051         case VSOCK_PROTO_PKT_ON_NOTIFY:
2052                 vmci_trans(vsk)->notify_ops =
2053                         &vmci_transport_notify_pkt_q_state_ops;
2054                 break;
2055         default:
2056                 pr_err("Unknown notify protocol version\n");
2057                 return false;
2058         }
2059
2060 exit:
2061         vmci_trans(vsk)->notify_ops->socket_init(sk);
2062         return true;
2063 }
2064
2065 static u16 vmci_transport_new_proto_supported_versions(void)
2066 {
2067         if (PROTOCOL_OVERRIDE != -1)
2068                 return PROTOCOL_OVERRIDE;
2069
2070         return VSOCK_PROTO_ALL_SUPPORTED;
2071 }
2072
2073 static u32 vmci_transport_get_local_cid(void)
2074 {
2075         return vmci_get_context_id();
2076 }
2077
2078 static const struct vsock_transport vmci_transport = {
2079         .init = vmci_transport_socket_init,
2080         .destruct = vmci_transport_destruct,
2081         .release = vmci_transport_release,
2082         .connect = vmci_transport_connect,
2083         .dgram_bind = vmci_transport_dgram_bind,
2084         .dgram_dequeue = vmci_transport_dgram_dequeue,
2085         .dgram_enqueue = vmci_transport_dgram_enqueue,
2086         .dgram_allow = vmci_transport_dgram_allow,
2087         .stream_dequeue = vmci_transport_stream_dequeue,
2088         .stream_enqueue = vmci_transport_stream_enqueue,
2089         .stream_has_data = vmci_transport_stream_has_data,
2090         .stream_has_space = vmci_transport_stream_has_space,
2091         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2092         .stream_is_active = vmci_transport_stream_is_active,
2093         .stream_allow = vmci_transport_stream_allow,
2094         .notify_poll_in = vmci_transport_notify_poll_in,
2095         .notify_poll_out = vmci_transport_notify_poll_out,
2096         .notify_recv_init = vmci_transport_notify_recv_init,
2097         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2098         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2099         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2100         .notify_send_init = vmci_transport_notify_send_init,
2101         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2102         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2103         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2104         .shutdown = vmci_transport_shutdown,
2105         .set_buffer_size = vmci_transport_set_buffer_size,
2106         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2107         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2108         .get_buffer_size = vmci_transport_get_buffer_size,
2109         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2110         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2111         .get_local_cid = vmci_transport_get_local_cid,
2112 };
2113
2114 static int __init vmci_transport_init(void)
2115 {
2116         int err;
2117
2118         /* Create the datagram handle that we will use to send and receive all
2119          * VSocket control messages for this context.
2120          */
2121         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2122                                                  VMCI_FLAG_ANYCID_DG_HND,
2123                                                  vmci_transport_recv_stream_cb,
2124                                                  NULL,
2125                                                  &vmci_transport_stream_handle);
2126         if (err < VMCI_SUCCESS) {
2127                 pr_err("Unable to create datagram handle. (%d)\n", err);
2128                 return vmci_transport_error_to_vsock_error(err);
2129         }
2130
2131         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2132                                    vmci_transport_qp_resumed_cb,
2133                                    NULL, &vmci_transport_qp_resumed_sub_id);
2134         if (err < VMCI_SUCCESS) {
2135                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2136                 err = vmci_transport_error_to_vsock_error(err);
2137                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2138                 goto err_destroy_stream_handle;
2139         }
2140
2141         err = vsock_core_init(&vmci_transport);
2142         if (err < 0)
2143                 goto err_unsubscribe;
2144
2145         return 0;
2146
2147 err_unsubscribe:
2148         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2149 err_destroy_stream_handle:
2150         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2151         return err;
2152 }
2153 module_init(vmci_transport_init);
2154
2155 static void __exit vmci_transport_exit(void)
2156 {
2157         cancel_work_sync(&vmci_transport_cleanup_work);
2158         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2159
2160         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2161                 if (vmci_datagram_destroy_handle(
2162                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2163                         pr_err("Couldn't destroy datagram handle\n");
2164                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2165         }
2166
2167         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2168                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2169                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2170         }
2171
2172         vsock_core_exit();
2173 }
2174 module_exit(vmci_transport_exit);
2175
2176 MODULE_AUTHOR("VMware, Inc.");
2177 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2178 MODULE_VERSION("1.0.5.0-k");
2179 MODULE_LICENSE("GPL v2");
2180 MODULE_ALIAS("vmware_vsock");
2181 MODULE_ALIAS_NETPROTO(PF_VSOCK);