net/vmw_vsock/vmci_transport.c

   1 /*
   2  * VMware vSockets Driver
   3  *
   4  * Copyright (C) 2007-2013 VMware, Inc. All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation version 2 and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  */
  15
  16 #include <linux/types.h>
  17 #include <linux/bitops.h>
  18 #include <linux/cred.h>
  19 #include <linux/init.h>
  20 #include <linux/io.h>
  21 #include <linux/kernel.h>
  22 #include <linux/kmod.h>
  23 #include <linux/list.h>
  24 #include <linux/miscdevice.h>
  25 #include <linux/module.h>
  26 #include <linux/mutex.h>
  27 #include <linux/net.h>
  28 #include <linux/poll.h>
  29 #include <linux/skbuff.h>
  30 #include <linux/smp.h>
  31 #include <linux/socket.h>
  32 #include <linux/stddef.h>
  33 #include <linux/unistd.h>
  34 #include <linux/wait.h>
  35 #include <linux/workqueue.h>
  36 #include <net/sock.h>
  37 #include <net/af_vsock.h>
  38
  39 #include "vmci_transport_notify.h"
  40
  41 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg);
  42 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg);
  43 static void vmci_transport_peer_detach_cb(u32 sub_id,
  44                                           const struct vmci_event_data *ed,
  45                                           void *client_data);
  46 static void vmci_transport_recv_pkt_work(struct work_struct *work);
  47 static void vmci_transport_cleanup(struct work_struct *work);
  48 static int vmci_transport_recv_listen(struct sock *sk,
  49                                       struct vmci_transport_packet *pkt);
  50 static int vmci_transport_recv_connecting_server(
  51                                         struct sock *sk,
  52                                         struct sock *pending,
  53                                         struct vmci_transport_packet *pkt);
  54 static int vmci_transport_recv_connecting_client(
  55                                         struct sock *sk,
  56                                         struct vmci_transport_packet *pkt);
  57 static int vmci_transport_recv_connecting_client_negotiate(
  58                                         struct sock *sk,
  59                                         struct vmci_transport_packet *pkt);
  60 static int vmci_transport_recv_connecting_client_invalid(
  61                                         struct sock *sk,
  62                                         struct vmci_transport_packet *pkt);
  63 static int vmci_transport_recv_connected(struct sock *sk,
  64                                          struct vmci_transport_packet *pkt);
  65 static bool vmci_transport_old_proto_override(bool *old_pkt_proto);
  66 static u16 vmci_transport_new_proto_supported_versions(void);
  67 static bool vmci_transport_proto_to_notify_struct(struct sock *sk, u16 *proto,
  68                                                   bool old_pkt_proto);
  69
  70 struct vmci_transport_recv_pkt_info {
  71         struct work_struct work;
  72         struct sock *sk;
  73         struct vmci_transport_packet pkt;
  74 };
  75
  76 static LIST_HEAD(vmci_transport_cleanup_list);
  77 static DEFINE_SPINLOCK(vmci_transport_cleanup_lock);
  78 static DECLARE_WORK(vmci_transport_cleanup_work, vmci_transport_cleanup);
  79
  80 static struct vmci_handle vmci_transport_stream_handle = { VMCI_INVALID_ID,
  81                                                            VMCI_INVALID_ID };
  82 static u32 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
  83
  84 static int PROTOCOL_OVERRIDE = -1;
  85
  86 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN   128
  87 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE       262144
  88 #define VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX   262144
  89
  90 /* The default peer timeout indicates how long we will wait for a peer response
  91  * to a control message.
  92  */
  93 #define VSOCK_DEFAULT_CONNECT_TIMEOUT (2 * HZ)
  94
  95 /* Helper function to convert from a VMCI error code to a VSock error code. */
  96
  97 static s32 vmci_transport_error_to_vsock_error(s32 vmci_error)
  98 {
  99         switch (vmci_error) {
 100         case VMCI_ERROR_NO_MEM:
 101                 return -ENOMEM;
 102         case VMCI_ERROR_DUPLICATE_ENTRY:
 103         case VMCI_ERROR_ALREADY_EXISTS:
 104                 return -EADDRINUSE;
 105         case VMCI_ERROR_NO_ACCESS:
 106                 return -EPERM;
 107         case VMCI_ERROR_NO_RESOURCES:
 108                 return -ENOBUFS;
 109         case VMCI_ERROR_INVALID_RESOURCE:
 110                 return -EHOSTUNREACH;
 111         case VMCI_ERROR_INVALID_ARGS:
 112         default:
 113                 break;
 114         }
 115         return -EINVAL;
 116 }
 117
 118 static u32 vmci_transport_peer_rid(u32 peer_cid)
 119 {
 120         if (VMADDR_CID_HYPERVISOR == peer_cid)
 121                 return VMCI_TRANSPORT_HYPERVISOR_PACKET_RID;
 122
 123         return VMCI_TRANSPORT_PACKET_RID;
 124 }
 125
 126 static inline void
 127 vmci_transport_packet_init(struct vmci_transport_packet *pkt,
 128                            struct sockaddr_vm *src,
 129                            struct sockaddr_vm *dst,
 130                            u8 type,
 131                            u64 size,
 132                            u64 mode,
 133                            struct vmci_transport_waiting_info *wait,
 134                            u16 proto,
 135                            struct vmci_handle handle)
 136 {
 137         /* We register the stream control handler as an any cid handle so we
 138          * must always send from a source address of VMADDR_CID_ANY
 139          */
 140         pkt->dg.src = vmci_make_handle(VMADDR_CID_ANY,
 141                                        VMCI_TRANSPORT_PACKET_RID);
 142         pkt->dg.dst = vmci_make_handle(dst->svm_cid,
 143                                        vmci_transport_peer_rid(dst->svm_cid));
 144         pkt->dg.payload_size = sizeof(*pkt) - sizeof(pkt->dg);
 145         pkt->version = VMCI_TRANSPORT_PACKET_VERSION;
 146         pkt->type = type;
 147         pkt->src_port = src->svm_port;
 148         pkt->dst_port = dst->svm_port;
 149         memset(&pkt->proto, 0, sizeof(pkt->proto));
 150         memset(&pkt->_reserved2, 0, sizeof(pkt->_reserved2));
 151
 152         switch (pkt->type) {
 153         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
 154                 pkt->u.size = 0;
 155                 break;
 156
 157         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST:
 158         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
 159                 pkt->u.size = size;
 160                 break;
 161
 162         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
 163         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
 164                 pkt->u.handle = handle;
 165                 break;
 166
 167         case VMCI_TRANSPORT_PACKET_TYPE_WROTE:
 168         case VMCI_TRANSPORT_PACKET_TYPE_READ:
 169         case VMCI_TRANSPORT_PACKET_TYPE_RST:
 170                 pkt->u.size = 0;
 171                 break;
 172
 173         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
 174                 pkt->u.mode = mode;
 175                 break;
 176
 177         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ:
 178         case VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE:
 179                 memcpy(&pkt->u.wait, wait, sizeof(pkt->u.wait));
 180                 break;
 181
 182         case VMCI_TRANSPORT_PACKET_TYPE_REQUEST2:
 183         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
 184                 pkt->u.size = size;
 185                 pkt->proto = proto;
 186                 break;
 187         }
 188 }
 189
 190 static inline void
 191 vmci_transport_packet_get_addresses(struct vmci_transport_packet *pkt,
 192                                     struct sockaddr_vm *local,
 193                                     struct sockaddr_vm *remote)
 194 {
 195         vsock_addr_init(local, pkt->dg.dst.context, pkt->dst_port);
 196         vsock_addr_init(remote, pkt->dg.src.context, pkt->src_port);
 197 }
 198
 199 static int
 200 __vmci_transport_send_control_pkt(struct vmci_transport_packet *pkt,
 201                                   struct sockaddr_vm *src,
 202                                   struct sockaddr_vm *dst,
 203                                   enum vmci_transport_packet_type type,
 204                                   u64 size,
 205                                   u64 mode,
 206                                   struct vmci_transport_waiting_info *wait,
 207                                   u16 proto,
 208                                   struct vmci_handle handle,
 209                                   bool convert_error)
 210 {
 211         int err;
 212
 213         vmci_transport_packet_init(pkt, src, dst, type, size, mode, wait,
 214                                    proto, handle);
 215         err = vmci_datagram_send(&pkt->dg);
 216         if (convert_error && (err < 0))
 217                 return vmci_transport_error_to_vsock_error(err);
 218
 219         return err;
 220 }
 221
 222 static int
 223 vmci_transport_reply_control_pkt_fast(struct vmci_transport_packet *pkt,
 224                                       enum vmci_transport_packet_type type,
 225                                       u64 size,
 226                                       u64 mode,
 227                                       struct vmci_transport_waiting_info *wait,
 228                                       struct vmci_handle handle)
 229 {
 230         struct vmci_transport_packet reply;
 231         struct sockaddr_vm src, dst;
 232
 233         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST) {
 234                 return 0;
 235         } else {
 236                 vmci_transport_packet_get_addresses(pkt, &src, &dst);
 237                 return __vmci_transport_send_control_pkt(&reply, &src, &dst,
 238                                                          type,
 239                                                          size, mode, wait,
 240                                                          VSOCK_PROTO_INVALID,
 241                                                          handle, true);
 242         }
 243 }
 244
 245 static int
 246 vmci_transport_send_control_pkt_bh(struct sockaddr_vm *src,
 247                                    struct sockaddr_vm *dst,
 248                                    enum vmci_transport_packet_type type,
 249                                    u64 size,
 250                                    u64 mode,
 251                                    struct vmci_transport_waiting_info *wait,
 252                                    struct vmci_handle handle)
 253 {
 254         /* Note that it is safe to use a single packet across all CPUs since
 255          * two tasklets of the same type are guaranteed to not ever run
 256          * simultaneously. If that ever changes, or VMCI stops using tasklets,
 257          * we can use per-cpu packets.
 258          */
 259         static struct vmci_transport_packet pkt;
 260
 261         return __vmci_transport_send_control_pkt(&pkt, src, dst, type,
 262                                                  size, mode, wait,
 263                                                  VSOCK_PROTO_INVALID, handle,
 264                                                  false);
 265 }
 266
 267 static int
 268 vmci_transport_send_control_pkt(struct sock *sk,
 269                                 enum vmci_transport_packet_type type,
 270                                 u64 size,
 271                                 u64 mode,
 272                                 struct vmci_transport_waiting_info *wait,
 273                                 u16 proto,
 274                                 struct vmci_handle handle)
 275 {
 276         struct vmci_transport_packet *pkt;
 277         struct vsock_sock *vsk;
 278         int err;
 279
 280         vsk = vsock_sk(sk);
 281
 282         if (!vsock_addr_bound(&vsk->local_addr))
 283                 return -EINVAL;
 284
 285         if (!vsock_addr_bound(&vsk->remote_addr))
 286                 return -EINVAL;
 287
 288         pkt = kmalloc(sizeof(*pkt), GFP_KERNEL);
 289         if (!pkt)
 290                 return -ENOMEM;
 291
 292         err = __vmci_transport_send_control_pkt(pkt, &vsk->local_addr,
 293                                                 &vsk->remote_addr, type, size,
 294                                                 mode, wait, proto, handle,
 295                                                 true);
 296         kfree(pkt);
 297
 298         return err;
 299 }
 300
 301 static int vmci_transport_send_reset_bh(struct sockaddr_vm *dst,
 302                                         struct sockaddr_vm *src,
 303                                         struct vmci_transport_packet *pkt)
 304 {
 305         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 306                 return 0;
 307         return vmci_transport_send_control_pkt_bh(
 308                                         dst, src,
 309                                         VMCI_TRANSPORT_PACKET_TYPE_RST, 0,
 310                                         0, NULL, VMCI_INVALID_HANDLE);
 311 }
 312
 313 static int vmci_transport_send_reset(struct sock *sk,
 314                                      struct vmci_transport_packet *pkt)
 315 {
 316         if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST)
 317                 return 0;
 318         return vmci_transport_send_control_pkt(sk,
 319                                         VMCI_TRANSPORT_PACKET_TYPE_RST,
 320                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 321                                         VMCI_INVALID_HANDLE);
 322 }
 323
 324 static int vmci_transport_send_negotiate(struct sock *sk, size_t size)
 325 {
 326         return vmci_transport_send_control_pkt(
 327                                         sk,
 328                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE,
 329                                         size, 0, NULL,
 330                                         VSOCK_PROTO_INVALID,
 331                                         VMCI_INVALID_HANDLE);
 332 }
 333
 334 static int vmci_transport_send_negotiate2(struct sock *sk, size_t size,
 335                                           u16 version)
 336 {
 337         return vmci_transport_send_control_pkt(
 338                                         sk,
 339                                         VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2,
 340                                         size, 0, NULL, version,
 341                                         VMCI_INVALID_HANDLE);
 342 }
 343
 344 static int vmci_transport_send_qp_offer(struct sock *sk,
 345                                         struct vmci_handle handle)
 346 {
 347         return vmci_transport_send_control_pkt(
 348                                         sk, VMCI_TRANSPORT_PACKET_TYPE_OFFER, 0,
 349                                         0, NULL,
 350                                         VSOCK_PROTO_INVALID, handle);
 351 }
 352
 353 static int vmci_transport_send_attach(struct sock *sk,
 354                                       struct vmci_handle handle)
 355 {
 356         return vmci_transport_send_control_pkt(
 357                                         sk, VMCI_TRANSPORT_PACKET_TYPE_ATTACH,
 358                                         0, 0, NULL, VSOCK_PROTO_INVALID,
 359                                         handle);
 360 }
 361
 362 static int vmci_transport_reply_reset(struct vmci_transport_packet *pkt)
 363 {
 364         return vmci_transport_reply_control_pkt_fast(
 365                                                 pkt,
 366                                                 VMCI_TRANSPORT_PACKET_TYPE_RST,
 367                                                 0, 0, NULL,
 368                                                 VMCI_INVALID_HANDLE);
 369 }
 370
 371 static int vmci_transport_send_invalid_bh(struct sockaddr_vm *dst,
 372                                           struct sockaddr_vm *src)
 373 {
 374         return vmci_transport_send_control_pkt_bh(
 375                                         dst, src,
 376                                         VMCI_TRANSPORT_PACKET_TYPE_INVALID,
 377                                         0, 0, NULL, VMCI_INVALID_HANDLE);
 378 }
 379
 380 int vmci_transport_send_wrote_bh(struct sockaddr_vm *dst,
 381                                  struct sockaddr_vm *src)
 382 {
 383         return vmci_transport_send_control_pkt_bh(
 384                                         dst, src,
 385                                         VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 386                                         0, NULL, VMCI_INVALID_HANDLE);
 387 }
 388
 389 int vmci_transport_send_read_bh(struct sockaddr_vm *dst,
 390                                 struct sockaddr_vm *src)
 391 {
 392         return vmci_transport_send_control_pkt_bh(
 393                                         dst, src,
 394                                         VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 395                                         0, NULL, VMCI_INVALID_HANDLE);
 396 }
 397
 398 int vmci_transport_send_wrote(struct sock *sk)
 399 {
 400         return vmci_transport_send_control_pkt(
 401                                         sk, VMCI_TRANSPORT_PACKET_TYPE_WROTE, 0,
 402                                         0, NULL, VSOCK_PROTO_INVALID,
 403                                         VMCI_INVALID_HANDLE);
 404 }
 405
 406 int vmci_transport_send_read(struct sock *sk)
 407 {
 408         return vmci_transport_send_control_pkt(
 409                                         sk, VMCI_TRANSPORT_PACKET_TYPE_READ, 0,
 410                                         0, NULL, VSOCK_PROTO_INVALID,
 411                                         VMCI_INVALID_HANDLE);
 412 }
 413
 414 int vmci_transport_send_waiting_write(struct sock *sk,
 415                                       struct vmci_transport_waiting_info *wait)
 416 {
 417         return vmci_transport_send_control_pkt(
 418                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_WRITE,
 419                                 0, 0, wait, VSOCK_PROTO_INVALID,
 420                                 VMCI_INVALID_HANDLE);
 421 }
 422
 423 int vmci_transport_send_waiting_read(struct sock *sk,
 424                                      struct vmci_transport_waiting_info *wait)
 425 {
 426         return vmci_transport_send_control_pkt(
 427                                 sk, VMCI_TRANSPORT_PACKET_TYPE_WAITING_READ,
 428                                 0, 0, wait, VSOCK_PROTO_INVALID,
 429                                 VMCI_INVALID_HANDLE);
 430 }
 431
 432 static int vmci_transport_shutdown(struct vsock_sock *vsk, int mode)
 433 {
 434         return vmci_transport_send_control_pkt(
 435                                         &vsk->sk,
 436                                         VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN,
 437                                         0, mode, NULL,
 438                                         VSOCK_PROTO_INVALID,
 439                                         VMCI_INVALID_HANDLE);
 440 }
 441
 442 static int vmci_transport_send_conn_request(struct sock *sk, size_t size)
 443 {
 444         return vmci_transport_send_control_pkt(sk,
 445                                         VMCI_TRANSPORT_PACKET_TYPE_REQUEST,
 446                                         size, 0, NULL,
 447                                         VSOCK_PROTO_INVALID,
 448                                         VMCI_INVALID_HANDLE);
 449 }
 450
 451 static int vmci_transport_send_conn_request2(struct sock *sk, size_t size,
 452                                              u16 version)
 453 {
 454         return vmci_transport_send_control_pkt(
 455                                         sk, VMCI_TRANSPORT_PACKET_TYPE_REQUEST2,
 456                                         size, 0, NULL, version,
 457                                         VMCI_INVALID_HANDLE);
 458 }
 459
 460 static struct sock *vmci_transport_get_pending(
 461                                         struct sock *listener,
 462                                         struct vmci_transport_packet *pkt)
 463 {
 464         struct vsock_sock *vlistener;
 465         struct vsock_sock *vpending;
 466         struct sock *pending;
 467         struct sockaddr_vm src;
 468
 469         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 470
 471         vlistener = vsock_sk(listener);
 472
 473         list_for_each_entry(vpending, &vlistener->pending_links,
 474                             pending_links) {
 475                 if (vsock_addr_equals_addr(&src, &vpending->remote_addr) &&
 476                     pkt->dst_port == vpending->local_addr.svm_port) {
 477                         pending = sk_vsock(vpending);
 478                         sock_hold(pending);
 479                         goto found;
 480                 }
 481         }
 482
 483         pending = NULL;
 484 found:
 485         return pending;
 486
 487 }
 488
 489 static void vmci_transport_release_pending(struct sock *pending)
 490 {
 491         sock_put(pending);
 492 }
 493
 494 /* We allow two kinds of sockets to communicate with a restricted VM: 1)
 495  * trusted sockets 2) sockets from applications running as the same user as the
 496  * VM (this is only true for the host side and only when using hosted products)
 497  */
 498
 499 static bool vmci_transport_is_trusted(struct vsock_sock *vsock, u32 peer_cid)
 500 {
 501         return vsock->trusted ||
 502                vmci_is_context_owner(peer_cid, vsock->owner->uid);
 503 }
 504
 505 /* We allow sending datagrams to and receiving datagrams from a restricted VM
 506  * only if it is trusted as described in vmci_transport_is_trusted.
 507  */
 508
 509 static bool vmci_transport_allow_dgram(struct vsock_sock *vsock, u32 peer_cid)
 510 {
 511         if (VMADDR_CID_HYPERVISOR == peer_cid)
 512                 return true;
 513
 514         if (vsock->cached_peer != peer_cid) {
 515                 vsock->cached_peer = peer_cid;
 516                 if (!vmci_transport_is_trusted(vsock, peer_cid) &&
 517                     (vmci_context_get_priv_flags(peer_cid) &
 518                      VMCI_PRIVILEGE_FLAG_RESTRICTED)) {
 519                         vsock->cached_peer_allow_dgram = false;
 520                 } else {
 521                         vsock->cached_peer_allow_dgram = true;
 522                 }
 523         }
 524
 525         return vsock->cached_peer_allow_dgram;
 526 }
 527
 528 static int
 529 vmci_transport_queue_pair_alloc(struct vmci_qp **qpair,
 530                                 struct vmci_handle *handle,
 531                                 u64 produce_size,
 532                                 u64 consume_size,
 533                                 u32 peer, u32 flags, bool trusted)
 534 {
 535         int err = 0;
 536
 537         if (trusted) {
 538                 /* Try to allocate our queue pair as trusted. This will only
 539                  * work if vsock is running in the host.
 540                  */
 541
 542                 err = vmci_qpair_alloc(qpair, handle, produce_size,
 543                                        consume_size,
 544                                        peer, flags,
 545                                        VMCI_PRIVILEGE_FLAG_TRUSTED);
 546                 if (err != VMCI_ERROR_NO_ACCESS)
 547                         goto out;
 548
 549         }
 550
 551         err = vmci_qpair_alloc(qpair, handle, produce_size, consume_size,
 552                                peer, flags, VMCI_NO_PRIVILEGE_FLAGS);
 553 out:
 554         if (err < 0) {
 555                 pr_err("Could not attach to queue pair with %d\n",
 556                        err);
 557                 err = vmci_transport_error_to_vsock_error(err);
 558         }
 559
 560         return err;
 561 }
 562
 563 static int
 564 vmci_transport_datagram_create_hnd(u32 resource_id,
 565                                    u32 flags,
 566                                    vmci_datagram_recv_cb recv_cb,
 567                                    void *client_data,
 568                                    struct vmci_handle *out_handle)
 569 {
 570         int err = 0;
 571
 572         /* Try to allocate our datagram handler as trusted. This will only work
 573          * if vsock is running in the host.
 574          */
 575
 576         err = vmci_datagram_create_handle_priv(resource_id, flags,
 577                                                VMCI_PRIVILEGE_FLAG_TRUSTED,
 578                                                recv_cb,
 579                                                client_data, out_handle);
 580
 581         if (err == VMCI_ERROR_NO_ACCESS)
 582                 err = vmci_datagram_create_handle(resource_id, flags,
 583                                                   recv_cb, client_data,
 584                                                   out_handle);
 585
 586         return err;
 587 }
 588
 589 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 590  * interrupt fires.  This is run in bottom-half context and if it ever needs to
 591  * sleep it should defer that work to a work queue.
 592  */
 593
 594 static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
 595 {
 596         struct sock *sk;
 597         size_t size;
 598         struct sk_buff *skb;
 599         struct vsock_sock *vsk;
 600
 601         sk = (struct sock *)data;
 602
 603         /* This handler is privileged when this module is running on the host.
 604          * We will get datagrams from all endpoints (even VMs that are in a
 605          * restricted context). If we get one from a restricted context then
 606          * the destination socket must be trusted.
 607          *
 608          * NOTE: We access the socket struct without holding the lock here.
 609          * This is ok because the field we are interested is never modified
 610          * outside of the create and destruct socket functions.
 611          */
 612         vsk = vsock_sk(sk);
 613         if (!vmci_transport_allow_dgram(vsk, dg->src.context))
 614                 return VMCI_ERROR_NO_ACCESS;
 615
 616         size = VMCI_DG_SIZE(dg);
 617
 618         /* Attach the packet to the socket's receive queue as an sk_buff. */
 619         skb = alloc_skb(size, GFP_ATOMIC);
 620         if (!skb)
 621                 return VMCI_ERROR_NO_MEM;
 622
 623         /* sk_receive_skb() will do a sock_put(), so hold here. */
 624         sock_hold(sk);
 625         skb_put(skb, size);
 626         memcpy(skb->data, dg, size);
 627         sk_receive_skb(sk, skb, 0);
 628
 629         return VMCI_SUCCESS;
 630 }
 631
 632 static bool vmci_transport_stream_allow(u32 cid, u32 port)
 633 {
 634         static const u32 non_socket_contexts[] = {
 635                 VMADDR_CID_RESERVED,
 636         };
 637         int i;
 638
 639         BUILD_BUG_ON(sizeof(cid) != sizeof(*non_socket_contexts));
 640
 641         for (i = 0; i < ARRAY_SIZE(non_socket_contexts); i++) {
 642                 if (cid == non_socket_contexts[i])
 643                         return false;
 644         }
 645
 646         return true;
 647 }
 648
 649 /* This is invoked as part of a tasklet that's scheduled when the VMCI
 650  * interrupt fires.  This is run in bottom-half context but it defers most of
 651  * its work to the packet handling work queue.
 652  */
 653
 654 static int vmci_transport_recv_stream_cb(void *data, struct vmci_datagram *dg)
 655 {
 656         struct sock *sk;
 657         struct sockaddr_vm dst;
 658         struct sockaddr_vm src;
 659         struct vmci_transport_packet *pkt;
 660         struct vsock_sock *vsk;
 661         bool bh_process_pkt;
 662         int err;
 663
 664         sk = NULL;
 665         err = VMCI_SUCCESS;
 666         bh_process_pkt = false;
 667
 668         /* Ignore incoming packets from contexts without sockets, or resources
 669          * that aren't vsock implementations.
 670          */
 671
 672         if (!vmci_transport_stream_allow(dg->src.context, -1)
 673             || vmci_transport_peer_rid(dg->src.context) != dg->src.resource)
 674                 return VMCI_ERROR_NO_ACCESS;
 675
 676         if (VMCI_DG_SIZE(dg) < sizeof(*pkt))
 677                 /* Drop datagrams that do not contain full VSock packets. */
 678                 return VMCI_ERROR_INVALID_ARGS;
 679
 680         pkt = (struct vmci_transport_packet *)dg;
 681
 682         /* Find the socket that should handle this packet.  First we look for a
 683          * connected socket and if there is none we look for a socket bound to
 684          * the destintation address.
 685          */
 686         vsock_addr_init(&src, pkt->dg.src.context, pkt->src_port);
 687         vsock_addr_init(&dst, pkt->dg.dst.context, pkt->dst_port);
 688
 689         sk = vsock_find_connected_socket(&src, &dst);
 690         if (!sk) {
 691                 sk = vsock_find_bound_socket(&dst);
 692                 if (!sk) {
 693                         /* We could not find a socket for this specified
 694                          * address.  If this packet is a RST, we just drop it.
 695                          * If it is another packet, we send a RST.  Note that
 696                          * we do not send a RST reply to RSTs so that we do not
 697                          * continually send RSTs between two endpoints.
 698                          *
 699                          * Note that since this is a reply, dst is src and src
 700                          * is dst.
 701                          */
 702                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 703                                 pr_err("unable to send reset\n");
 704
 705                         err = VMCI_ERROR_NOT_FOUND;
 706                         goto out;
 707                 }
 708         }
 709
 710         /* If the received packet type is beyond all types known to this
 711          * implementation, reply with an invalid message.  Hopefully this will
 712          * help when implementing backwards compatibility in the future.
 713          */
 714         if (pkt->type >= VMCI_TRANSPORT_PACKET_TYPE_MAX) {
 715                 vmci_transport_send_invalid_bh(&dst, &src);
 716                 err = VMCI_ERROR_INVALID_ARGS;
 717                 goto out;
 718         }
 719
 720         /* This handler is privileged when this module is running on the host.
 721          * We will get datagram connect requests from all endpoints (even VMs
 722          * that are in a restricted context). If we get one from a restricted
 723          * context then the destination socket must be trusted.
 724          *
 725          * NOTE: We access the socket struct without holding the lock here.
 726          * This is ok because the field we are interested is never modified
 727          * outside of the create and destruct socket functions.
 728          */
 729         vsk = vsock_sk(sk);
 730         if (!vmci_transport_allow_dgram(vsk, pkt->dg.src.context)) {
 731                 err = VMCI_ERROR_NO_ACCESS;
 732                 goto out;
 733         }
 734
 735         /* We do most everything in a work queue, but let's fast path the
 736          * notification of reads and writes to help data transfer performance.
 737          * We can only do this if there is no process context code executing
 738          * for this socket since that may change the state.
 739          */
 740         bh_lock_sock(sk);
 741
 742         if (!sock_owned_by_user(sk)) {
 743                 /* The local context ID may be out of date, update it. */
 744                 vsk->local_addr.svm_cid = dst.svm_cid;
 745
 746                 if (sk->sk_state == SS_CONNECTED)
 747                         vmci_trans(vsk)->notify_ops->handle_notify_pkt(
 748                                         sk, pkt, true, &dst, &src,
 749                                         &bh_process_pkt);
 750         }
 751
 752         bh_unlock_sock(sk);
 753
 754         if (!bh_process_pkt) {
 755                 struct vmci_transport_recv_pkt_info *recv_pkt_info;
 756
 757                 recv_pkt_info = kmalloc(sizeof(*recv_pkt_info), GFP_ATOMIC);
 758                 if (!recv_pkt_info) {
 759                         if (vmci_transport_send_reset_bh(&dst, &src, pkt) < 0)
 760                                 pr_err("unable to send reset\n");
 761
 762                         err = VMCI_ERROR_NO_MEM;
 763                         goto out;
 764                 }
 765
 766                 recv_pkt_info->sk = sk;
 767                 memcpy(&recv_pkt_info->pkt, pkt, sizeof(recv_pkt_info->pkt));
 768                 INIT_WORK(&recv_pkt_info->work, vmci_transport_recv_pkt_work);
 769
 770                 schedule_work(&recv_pkt_info->work);
 771                 /* Clear sk so that the reference count incremented by one of
 772                  * the Find functions above is not decremented below.  We need
 773                  * that reference count for the packet handler we've scheduled
 774                  * to run.
 775                  */
 776                 sk = NULL;
 777         }
 778
 779 out:
 780         if (sk)
 781                 sock_put(sk);
 782
 783         return err;
 784 }
 785
 786 static void vmci_transport_handle_detach(struct sock *sk)
 787 {
 788         struct vsock_sock *vsk;
 789
 790         vsk = vsock_sk(sk);
 791         if (!vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)) {
 792                 sock_set_flag(sk, SOCK_DONE);
 793
 794                 /* On a detach the peer will not be sending or receiving
 795                  * anymore.
 796                  */
 797                 vsk->peer_shutdown = SHUTDOWN_MASK;
 798
 799                 /* We should not be sending anymore since the peer won't be
 800                  * there to receive, but we can still receive if there is data
 801                  * left in our consume queue.
 802                  */
 803                 if (vsock_stream_has_data(vsk) <= 0) {
 804                         if (sk->sk_state == SS_CONNECTING) {
 805                                 /* The peer may detach from a queue pair while
 806                                  * we are still in the connecting state, i.e.,
 807                                  * if the peer VM is killed after attaching to
 808                                  * a queue pair, but before we complete the
 809                                  * handshake. In that case, we treat the detach
 810                                  * event like a reset.
 811                                  */
 812
 813                                 sk->sk_state = SS_UNCONNECTED;
 814                                 sk->sk_err = ECONNRESET;
 815                                 sk->sk_error_report(sk);
 816                                 return;
 817                         }
 818                         sk->sk_state = SS_UNCONNECTED;
 819                 }
 820                 sk->sk_state_change(sk);
 821         }
 822 }
 823
 824 static void vmci_transport_peer_detach_cb(u32 sub_id,
 825                                           const struct vmci_event_data *e_data,
 826                                           void *client_data)
 827 {
 828         struct vmci_transport *trans = client_data;
 829         const struct vmci_event_payload_qp *e_payload;
 830
 831         e_payload = vmci_event_data_const_payload(e_data);
 832
 833         /* XXX This is lame, we should provide a way to lookup sockets by
 834          * qp_handle.
 835          */
 836         if (vmci_handle_is_invalid(e_payload->handle) ||
 837             !vmci_handle_is_equal(trans->qp_handle, e_payload->handle))
 838                 return;
 839
 840         /* We don't ask for delayed CBs when we subscribe to this event (we
 841          * pass 0 as flags to vmci_event_subscribe()).  VMCI makes no
 842          * guarantees in that case about what context we might be running in,
 843          * so it could be BH or process, blockable or non-blockable.  So we
 844          * need to account for all possible contexts here.
 845          */
 846         spin_lock_bh(&trans->lock);
 847         if (!trans->sk)
 848                 goto out;
 849
 850         /* Apart from here, trans->lock is only grabbed as part of sk destruct,
 851          * where trans->sk isn't locked.
 852          */
 853         bh_lock_sock(trans->sk);
 854
 855         vmci_transport_handle_detach(trans->sk);
 856
 857         bh_unlock_sock(trans->sk);
 858  out:
 859         spin_unlock_bh(&trans->lock);
 860 }
 861
 862 static void vmci_transport_qp_resumed_cb(u32 sub_id,
 863                                          const struct vmci_event_data *e_data,
 864                                          void *client_data)
 865 {
 866         vsock_for_each_connected_socket(vmci_transport_handle_detach);
 867 }
 868
 869 static void vmci_transport_recv_pkt_work(struct work_struct *work)
 870 {
 871         struct vmci_transport_recv_pkt_info *recv_pkt_info;
 872         struct vmci_transport_packet *pkt;
 873         struct sock *sk;
 874
 875         recv_pkt_info =
 876                 container_of(work, struct vmci_transport_recv_pkt_info, work);
 877         sk = recv_pkt_info->sk;
 878         pkt = &recv_pkt_info->pkt;
 879
 880         lock_sock(sk);
 881
 882         /* The local context ID may be out of date. */
 883         vsock_sk(sk)->local_addr.svm_cid = pkt->dg.dst.context;
 884
 885         switch (sk->sk_state) {
 886         case VSOCK_SS_LISTEN:
 887                 vmci_transport_recv_listen(sk, pkt);
 888                 break;
 889         case SS_CONNECTING:
 890                 /* Processing of pending connections for servers goes through
 891                  * the listening socket, so see vmci_transport_recv_listen()
 892                  * for that path.
 893                  */
 894                 vmci_transport_recv_connecting_client(sk, pkt);
 895                 break;
 896         case SS_CONNECTED:
 897                 vmci_transport_recv_connected(sk, pkt);
 898                 break;
 899         default:
 900                 /* Because this function does not run in the same context as
 901                  * vmci_transport_recv_stream_cb it is possible that the
 902                  * socket has closed. We need to let the other side know or it
 903                  * could be sitting in a connect and hang forever. Send a
 904                  * reset to prevent that.
 905                  */
 906                 vmci_transport_send_reset(sk, pkt);
 907                 break;
 908         }
 909
 910         release_sock(sk);
 911         kfree(recv_pkt_info);
 912         /* Release reference obtained in the stream callback when we fetched
 913          * this socket out of the bound or connected list.
 914          */
 915         sock_put(sk);
 916 }
 917
 918 static int vmci_transport_recv_listen(struct sock *sk,
 919                                       struct vmci_transport_packet *pkt)
 920 {
 921         struct sock *pending;
 922         struct vsock_sock *vpending;
 923         int err;
 924         u64 qp_size;
 925         bool old_request = false;
 926         bool old_pkt_proto = false;
 927
 928         err = 0;
 929
 930         /* Because we are in the listen state, we could be receiving a packet
 931          * for ourself or any previous connection requests that we received.
 932          * If it's the latter, we try to find a socket in our list of pending
 933          * connections and, if we do, call the appropriate handler for the
 934          * state that that socket is in.  Otherwise we try to service the
 935          * connection request.
 936          */
 937         pending = vmci_transport_get_pending(sk, pkt);
 938         if (pending) {
 939                 lock_sock(pending);
 940
 941                 /* The local context ID may be out of date. */
 942                 vsock_sk(pending)->local_addr.svm_cid = pkt->dg.dst.context;
 943
 944                 switch (pending->sk_state) {
 945                 case SS_CONNECTING:
 946                         err = vmci_transport_recv_connecting_server(sk,
 947                                                                     pending,
 948                                                                     pkt);
 949                         break;
 950                 default:
 951                         vmci_transport_send_reset(pending, pkt);
 952                         err = -EINVAL;
 953                 }
 954
 955                 if (err < 0)
 956                         vsock_remove_pending(sk, pending);
 957
 958                 release_sock(pending);
 959                 vmci_transport_release_pending(pending);
 960
 961                 return err;
 962         }
 963
 964         /* The listen state only accepts connection requests.  Reply with a
 965          * reset unless we received a reset.
 966          */
 967
 968         if (!(pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST ||
 969               pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)) {
 970                 vmci_transport_reply_reset(pkt);
 971                 return -EINVAL;
 972         }
 973
 974         if (pkt->u.size == 0) {
 975                 vmci_transport_reply_reset(pkt);
 976                 return -EINVAL;
 977         }
 978
 979         /* If this socket can't accommodate this connection request, we send a
 980          * reset.  Otherwise we create and initialize a child socket and reply
 981          * with a connection negotiation.
 982          */
 983         if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) {
 984                 vmci_transport_reply_reset(pkt);
 985                 return -ECONNREFUSED;
 986         }
 987
 988         pending = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
 989                                  sk->sk_type, 0);
 990         if (!pending) {
 991                 vmci_transport_send_reset(sk, pkt);
 992                 return -ENOMEM;
 993         }
 994
 995         vpending = vsock_sk(pending);
 996
 997         vsock_addr_init(&vpending->local_addr, pkt->dg.dst.context,
 998                         pkt->dst_port);
 999         vsock_addr_init(&vpending->remote_addr, pkt->dg.src.context,
1000                         pkt->src_port);
1001
1002         /* If the proposed size fits within our min/max, accept it. Otherwise
1003          * propose our own size.
1004          */
1005         if (pkt->u.size >= vmci_trans(vpending)->queue_pair_min_size &&
1006             pkt->u.size <= vmci_trans(vpending)->queue_pair_max_size) {
1007                 qp_size = pkt->u.size;
1008         } else {
1009                 qp_size = vmci_trans(vpending)->queue_pair_size;
1010         }
1011
1012         /* Figure out if we are using old or new requests based on the
1013          * overrides pkt types sent by our peer.
1014          */
1015         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1016                 old_request = old_pkt_proto;
1017         } else {
1018                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST)
1019                         old_request = true;
1020                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_REQUEST2)
1021                         old_request = false;
1022
1023         }
1024
1025         if (old_request) {
1026                 /* Handle a REQUEST (or override) */
1027                 u16 version = VSOCK_PROTO_INVALID;
1028                 if (vmci_transport_proto_to_notify_struct(
1029                         pending, &version, true))
1030                         err = vmci_transport_send_negotiate(pending, qp_size);
1031                 else
1032                         err = -EINVAL;
1033
1034         } else {
1035                 /* Handle a REQUEST2 (or override) */
1036                 int proto_int = pkt->proto;
1037                 int pos;
1038                 u16 active_proto_version = 0;
1039
1040                 /* The list of possible protocols is the intersection of all
1041                  * protocols the client supports ... plus all the protocols we
1042                  * support.
1043                  */
1044                 proto_int &= vmci_transport_new_proto_supported_versions();
1045
1046                 /* We choose the highest possible protocol version and use that
1047                  * one.
1048                  */
1049                 pos = fls(proto_int);
1050                 if (pos) {
1051                         active_proto_version = (1 << (pos - 1));
1052                         if (vmci_transport_proto_to_notify_struct(
1053                                 pending, &active_proto_version, false))
1054                                 err = vmci_transport_send_negotiate2(pending,
1055                                                         qp_size,
1056                                                         active_proto_version);
1057                         else
1058                                 err = -EINVAL;
1059
1060                 } else {
1061                         err = -EINVAL;
1062                 }
1063         }
1064
1065         if (err < 0) {
1066                 vmci_transport_send_reset(sk, pkt);
1067                 sock_put(pending);
1068                 err = vmci_transport_error_to_vsock_error(err);
1069                 goto out;
1070         }
1071
1072         vsock_add_pending(sk, pending);
1073         sk->sk_ack_backlog++;
1074
1075         pending->sk_state = SS_CONNECTING;
1076         vmci_trans(vpending)->produce_size =
1077                 vmci_trans(vpending)->consume_size = qp_size;
1078         vmci_trans(vpending)->queue_pair_size = qp_size;
1079
1080         vmci_trans(vpending)->notify_ops->process_request(pending);
1081
1082         /* We might never receive another message for this socket and it's not
1083          * connected to any process, so we have to ensure it gets cleaned up
1084          * ourself.  Our delayed work function will take care of that.  Note
1085          * that we do not ever cancel this function since we have few
1086          * guarantees about its state when calling cancel_delayed_work().
1087          * Instead we hold a reference on the socket for that function and make
1088          * it capable of handling cases where it needs to do nothing but
1089          * release that reference.
1090          */
1091         vpending->listener = sk;
1092         sock_hold(sk);
1093         sock_hold(pending);
1094         INIT_DELAYED_WORK(&vpending->dwork, vsock_pending_work);
1095         schedule_delayed_work(&vpending->dwork, HZ);
1096
1097 out:
1098         return err;
1099 }
1100
1101 static int
1102 vmci_transport_recv_connecting_server(struct sock *listener,
1103                                       struct sock *pending,
1104                                       struct vmci_transport_packet *pkt)
1105 {
1106         struct vsock_sock *vpending;
1107         struct vmci_handle handle;
1108         struct vmci_qp *qpair;
1109         bool is_local;
1110         u32 flags;
1111         u32 detach_sub_id;
1112         int err;
1113         int skerr;
1114
1115         vpending = vsock_sk(pending);
1116         detach_sub_id = VMCI_INVALID_ID;
1117
1118         switch (pkt->type) {
1119         case VMCI_TRANSPORT_PACKET_TYPE_OFFER:
1120                 if (vmci_handle_is_invalid(pkt->u.handle)) {
1121                         vmci_transport_send_reset(pending, pkt);
1122                         skerr = EPROTO;
1123                         err = -EINVAL;
1124                         goto destroy;
1125                 }
1126                 break;
1127         default:
1128                 /* Close and cleanup the connection. */
1129                 vmci_transport_send_reset(pending, pkt);
1130                 skerr = EPROTO;
1131                 err = pkt->type == VMCI_TRANSPORT_PACKET_TYPE_RST ? 0 : -EINVAL;
1132                 goto destroy;
1133         }
1134
1135         /* In order to complete the connection we need to attach to the offered
1136          * queue pair and send an attach notification.  We also subscribe to the
1137          * detach event so we know when our peer goes away, and we do that
1138          * before attaching so we don't miss an event.  If all this succeeds,
1139          * we update our state and wakeup anything waiting in accept() for a
1140          * connection.
1141          */
1142
1143         /* We don't care about attach since we ensure the other side has
1144          * attached by specifying the ATTACH_ONLY flag below.
1145          */
1146         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1147                                    vmci_transport_peer_detach_cb,
1148                                    vmci_trans(vpending), &detach_sub_id);
1149         if (err < VMCI_SUCCESS) {
1150                 vmci_transport_send_reset(pending, pkt);
1151                 err = vmci_transport_error_to_vsock_error(err);
1152                 skerr = -err;
1153                 goto destroy;
1154         }
1155
1156         vmci_trans(vpending)->detach_sub_id = detach_sub_id;
1157
1158         /* Now attach to the queue pair the client created. */
1159         handle = pkt->u.handle;
1160
1161         /* vpending->local_addr always has a context id so we do not need to
1162          * worry about VMADDR_CID_ANY in this case.
1163          */
1164         is_local =
1165             vpending->remote_addr.svm_cid == vpending->local_addr.svm_cid;
1166         flags = VMCI_QPFLAG_ATTACH_ONLY;
1167         flags |= is_local ? VMCI_QPFLAG_LOCAL : 0;
1168
1169         err = vmci_transport_queue_pair_alloc(
1170                                         &qpair,
1171                                         &handle,
1172                                         vmci_trans(vpending)->produce_size,
1173                                         vmci_trans(vpending)->consume_size,
1174                                         pkt->dg.src.context,
1175                                         flags,
1176                                         vmci_transport_is_trusted(
1177                                                 vpending,
1178                                                 vpending->remote_addr.svm_cid));
1179         if (err < 0) {
1180                 vmci_transport_send_reset(pending, pkt);
1181                 skerr = -err;
1182                 goto destroy;
1183         }
1184
1185         vmci_trans(vpending)->qp_handle = handle;
1186         vmci_trans(vpending)->qpair = qpair;
1187
1188         /* When we send the attach message, we must be ready to handle incoming
1189          * control messages on the newly connected socket. So we move the
1190          * pending socket to the connected state before sending the attach
1191          * message. Otherwise, an incoming packet triggered by the attach being
1192          * received by the peer may be processed concurrently with what happens
1193          * below after sending the attach message, and that incoming packet
1194          * will find the listening socket instead of the (currently) pending
1195          * socket. Note that enqueueing the socket increments the reference
1196          * count, so even if a reset comes before the connection is accepted,
1197          * the socket will be valid until it is removed from the queue.
1198          *
1199          * If we fail sending the attach below, we remove the socket from the
1200          * connected list and move the socket to SS_UNCONNECTED before
1201          * releasing the lock, so a pending slow path processing of an incoming
1202          * packet will not see the socket in the connected state in that case.
1203          */
1204         pending->sk_state = SS_CONNECTED;
1205
1206         vsock_insert_connected(vpending);
1207
1208         /* Notify our peer of our attach. */
1209         err = vmci_transport_send_attach(pending, handle);
1210         if (err < 0) {
1211                 vsock_remove_connected(vpending);
1212                 pr_err("Could not send attach\n");
1213                 vmci_transport_send_reset(pending, pkt);
1214                 err = vmci_transport_error_to_vsock_error(err);
1215                 skerr = -err;
1216                 goto destroy;
1217         }
1218
1219         /* We have a connection. Move the now connected socket from the
1220          * listener's pending list to the accept queue so callers of accept()
1221          * can find it.
1222          */
1223         vsock_remove_pending(listener, pending);
1224         vsock_enqueue_accept(listener, pending);
1225
1226         /* Callers of accept() will be be waiting on the listening socket, not
1227          * the pending socket.
1228          */
1229         listener->sk_data_ready(listener);
1230
1231         return 0;
1232
1233 destroy:
1234         pending->sk_err = skerr;
1235         pending->sk_state = SS_UNCONNECTED;
1236         /* As long as we drop our reference, all necessary cleanup will handle
1237          * when the cleanup function drops its reference and our destruct
1238          * implementation is called.  Note that since the listen handler will
1239          * remove pending from the pending list upon our failure, the cleanup
1240          * function won't drop the additional reference, which is why we do it
1241          * here.
1242          */
1243         sock_put(pending);
1244
1245         return err;
1246 }
1247
1248 static int
1249 vmci_transport_recv_connecting_client(struct sock *sk,
1250                                       struct vmci_transport_packet *pkt)
1251 {
1252         struct vsock_sock *vsk;
1253         int err;
1254         int skerr;
1255
1256         vsk = vsock_sk(sk);
1257
1258         switch (pkt->type) {
1259         case VMCI_TRANSPORT_PACKET_TYPE_ATTACH:
1260                 if (vmci_handle_is_invalid(pkt->u.handle) ||
1261                     !vmci_handle_is_equal(pkt->u.handle,
1262                                           vmci_trans(vsk)->qp_handle)) {
1263                         skerr = EPROTO;
1264                         err = -EINVAL;
1265                         goto destroy;
1266                 }
1267
1268                 /* Signify the socket is connected and wakeup the waiter in
1269                  * connect(). Also place the socket in the connected table for
1270                  * accounting (it can already be found since it's in the bound
1271                  * table).
1272                  */
1273                 sk->sk_state = SS_CONNECTED;
1274                 sk->sk_socket->state = SS_CONNECTED;
1275                 vsock_insert_connected(vsk);
1276                 sk->sk_state_change(sk);
1277
1278                 break;
1279         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE:
1280         case VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2:
1281                 if (pkt->u.size == 0
1282                     || pkt->dg.src.context != vsk->remote_addr.svm_cid
1283                     || pkt->src_port != vsk->remote_addr.svm_port
1284                     || !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle)
1285                     || vmci_trans(vsk)->qpair
1286                     || vmci_trans(vsk)->produce_size != 0
1287                     || vmci_trans(vsk)->consume_size != 0
1288                     || vmci_trans(vsk)->detach_sub_id != VMCI_INVALID_ID) {
1289                         skerr = EPROTO;
1290                         err = -EINVAL;
1291
1292                         goto destroy;
1293                 }
1294
1295                 err = vmci_transport_recv_connecting_client_negotiate(sk, pkt);
1296                 if (err) {
1297                         skerr = -err;
1298                         goto destroy;
1299                 }
1300
1301                 break;
1302         case VMCI_TRANSPORT_PACKET_TYPE_INVALID:
1303                 err = vmci_transport_recv_connecting_client_invalid(sk, pkt);
1304                 if (err) {
1305                         skerr = -err;
1306                         goto destroy;
1307                 }
1308
1309                 break;
1310         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1311                 /* Older versions of the linux code (WS 6.5 / ESX 4.0) used to
1312                  * continue processing here after they sent an INVALID packet.
1313                  * This meant that we got a RST after the INVALID. We ignore a
1314                  * RST after an INVALID. The common code doesn't send the RST
1315                  * ... so we can hang if an old version of the common code
1316                  * fails between getting a REQUEST and sending an OFFER back.
1317                  * Not much we can do about it... except hope that it doesn't
1318                  * happen.
1319                  */
1320                 if (vsk->ignore_connecting_rst) {
1321                         vsk->ignore_connecting_rst = false;
1322                 } else {
1323                         skerr = ECONNRESET;
1324                         err = 0;
1325                         goto destroy;
1326                 }
1327
1328                 break;
1329         default:
1330                 /* Close and cleanup the connection. */
1331                 skerr = EPROTO;
1332                 err = -EINVAL;
1333                 goto destroy;
1334         }
1335
1336         return 0;
1337
1338 destroy:
1339         vmci_transport_send_reset(sk, pkt);
1340
1341         sk->sk_state = SS_UNCONNECTED;
1342         sk->sk_err = skerr;
1343         sk->sk_error_report(sk);
1344         return err;
1345 }
1346
1347 static int vmci_transport_recv_connecting_client_negotiate(
1348                                         struct sock *sk,
1349                                         struct vmci_transport_packet *pkt)
1350 {
1351         int err;
1352         struct vsock_sock *vsk;
1353         struct vmci_handle handle;
1354         struct vmci_qp *qpair;
1355         u32 detach_sub_id;
1356         bool is_local;
1357         u32 flags;
1358         bool old_proto = true;
1359         bool old_pkt_proto;
1360         u16 version;
1361
1362         vsk = vsock_sk(sk);
1363         handle = VMCI_INVALID_HANDLE;
1364         detach_sub_id = VMCI_INVALID_ID;
1365
1366         /* If we have gotten here then we should be past the point where old
1367          * linux vsock could have sent the bogus rst.
1368          */
1369         vsk->sent_request = false;
1370         vsk->ignore_connecting_rst = false;
1371
1372         /* Verify that we're OK with the proposed queue pair size */
1373         if (pkt->u.size < vmci_trans(vsk)->queue_pair_min_size ||
1374             pkt->u.size > vmci_trans(vsk)->queue_pair_max_size) {
1375                 err = -EINVAL;
1376                 goto destroy;
1377         }
1378
1379         /* At this point we know the CID the peer is using to talk to us. */
1380
1381         if (vsk->local_addr.svm_cid == VMADDR_CID_ANY)
1382                 vsk->local_addr.svm_cid = pkt->dg.dst.context;
1383
1384         /* Setup the notify ops to be the highest supported version that both
1385          * the server and the client support.
1386          */
1387
1388         if (vmci_transport_old_proto_override(&old_pkt_proto)) {
1389                 old_proto = old_pkt_proto;
1390         } else {
1391                 if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE)
1392                         old_proto = true;
1393                 else if (pkt->type == VMCI_TRANSPORT_PACKET_TYPE_NEGOTIATE2)
1394                         old_proto = false;
1395
1396         }
1397
1398         if (old_proto)
1399                 version = VSOCK_PROTO_INVALID;
1400         else
1401                 version = pkt->proto;
1402
1403         if (!vmci_transport_proto_to_notify_struct(sk, &version, old_proto)) {
1404                 err = -EINVAL;
1405                 goto destroy;
1406         }
1407
1408         /* Subscribe to detach events first.
1409          *
1410          * XXX We attach once for each queue pair created for now so it is easy
1411          * to find the socket (it's provided), but later we should only
1412          * subscribe once and add a way to lookup sockets by queue pair handle.
1413          */
1414         err = vmci_event_subscribe(VMCI_EVENT_QP_PEER_DETACH,
1415                                    vmci_transport_peer_detach_cb,
1416                                    vmci_trans(vsk), &detach_sub_id);
1417         if (err < VMCI_SUCCESS) {
1418                 err = vmci_transport_error_to_vsock_error(err);
1419                 goto destroy;
1420         }
1421
1422         /* Make VMCI select the handle for us. */
1423         handle = VMCI_INVALID_HANDLE;
1424         is_local = vsk->remote_addr.svm_cid == vsk->local_addr.svm_cid;
1425         flags = is_local ? VMCI_QPFLAG_LOCAL : 0;
1426
1427         err = vmci_transport_queue_pair_alloc(&qpair,
1428                                               &handle,
1429                                               pkt->u.size,
1430                                               pkt->u.size,
1431                                               vsk->remote_addr.svm_cid,
1432                                               flags,
1433                                               vmci_transport_is_trusted(
1434                                                   vsk,
1435                                                   vsk->
1436                                                   remote_addr.svm_cid));
1437         if (err < 0)
1438                 goto destroy;
1439
1440         err = vmci_transport_send_qp_offer(sk, handle);
1441         if (err < 0) {
1442                 err = vmci_transport_error_to_vsock_error(err);
1443                 goto destroy;
1444         }
1445
1446         vmci_trans(vsk)->qp_handle = handle;
1447         vmci_trans(vsk)->qpair = qpair;
1448
1449         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size =
1450                 pkt->u.size;
1451
1452         vmci_trans(vsk)->detach_sub_id = detach_sub_id;
1453
1454         vmci_trans(vsk)->notify_ops->process_negotiate(sk);
1455
1456         return 0;
1457
1458 destroy:
1459         if (detach_sub_id != VMCI_INVALID_ID)
1460                 vmci_event_unsubscribe(detach_sub_id);
1461
1462         if (!vmci_handle_is_invalid(handle))
1463                 vmci_qpair_detach(&qpair);
1464
1465         return err;
1466 }
1467
1468 static int
1469 vmci_transport_recv_connecting_client_invalid(struct sock *sk,
1470                                               struct vmci_transport_packet *pkt)
1471 {
1472         int err = 0;
1473         struct vsock_sock *vsk = vsock_sk(sk);
1474
1475         if (vsk->sent_request) {
1476                 vsk->sent_request = false;
1477                 vsk->ignore_connecting_rst = true;
1478
1479                 err = vmci_transport_send_conn_request(
1480                         sk, vmci_trans(vsk)->queue_pair_size);
1481                 if (err < 0)
1482                         err = vmci_transport_error_to_vsock_error(err);
1483                 else
1484                         err = 0;
1485
1486         }
1487
1488         return err;
1489 }
1490
1491 static int vmci_transport_recv_connected(struct sock *sk,
1492                                          struct vmci_transport_packet *pkt)
1493 {
1494         struct vsock_sock *vsk;
1495         bool pkt_processed = false;
1496
1497         /* In cases where we are closing the connection, it's sufficient to
1498          * mark the state change (and maybe error) and wake up any waiting
1499          * threads. Since this is a connected socket, it's owned by a user
1500          * process and will be cleaned up when the failure is passed back on
1501          * the current or next system call.  Our system call implementations
1502          * must therefore check for error and state changes on entry and when
1503          * being awoken.
1504          */
1505         switch (pkt->type) {
1506         case VMCI_TRANSPORT_PACKET_TYPE_SHUTDOWN:
1507                 if (pkt->u.mode) {
1508                         vsk = vsock_sk(sk);
1509
1510                         vsk->peer_shutdown |= pkt->u.mode;
1511                         sk->sk_state_change(sk);
1512                 }
1513                 break;
1514
1515         case VMCI_TRANSPORT_PACKET_TYPE_RST:
1516                 vsk = vsock_sk(sk);
1517                 /* It is possible that we sent our peer a message (e.g a
1518                  * WAITING_READ) right before we got notified that the peer had
1519                  * detached. If that happens then we can get a RST pkt back
1520                  * from our peer even though there is data available for us to
1521                  * read. In that case, don't shutdown the socket completely but
1522                  * instead allow the local client to finish reading data off
1523                  * the queuepair. Always treat a RST pkt in connected mode like
1524                  * a clean shutdown.
1525                  */
1526                 sock_set_flag(sk, SOCK_DONE);
1527                 vsk->peer_shutdown = SHUTDOWN_MASK;
1528                 if (vsock_stream_has_data(vsk) <= 0)
1529                         sk->sk_state = SS_DISCONNECTING;
1530
1531                 sk->sk_state_change(sk);
1532                 break;
1533
1534         default:
1535                 vsk = vsock_sk(sk);
1536                 vmci_trans(vsk)->notify_ops->handle_notify_pkt(
1537                                 sk, pkt, false, NULL, NULL,
1538                                 &pkt_processed);
1539                 if (!pkt_processed)
1540                         return -EINVAL;
1541
1542                 break;
1543         }
1544
1545         return 0;
1546 }
1547
1548 static int vmci_transport_socket_init(struct vsock_sock *vsk,
1549                                       struct vsock_sock *psk)
1550 {
1551         vsk->trans = kmalloc(sizeof(struct vmci_transport), GFP_KERNEL);
1552         if (!vsk->trans)
1553                 return -ENOMEM;
1554
1555         vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1556         vmci_trans(vsk)->qp_handle = VMCI_INVALID_HANDLE;
1557         vmci_trans(vsk)->qpair = NULL;
1558         vmci_trans(vsk)->produce_size = vmci_trans(vsk)->consume_size = 0;
1559         vmci_trans(vsk)->detach_sub_id = VMCI_INVALID_ID;
1560         vmci_trans(vsk)->notify_ops = NULL;
1561         INIT_LIST_HEAD(&vmci_trans(vsk)->elem);
1562         vmci_trans(vsk)->sk = &vsk->sk;
1563         spin_lock_init(&vmci_trans(vsk)->lock);
1564         if (psk) {
1565                 vmci_trans(vsk)->queue_pair_size =
1566                         vmci_trans(psk)->queue_pair_size;
1567                 vmci_trans(vsk)->queue_pair_min_size =
1568                         vmci_trans(psk)->queue_pair_min_size;
1569                 vmci_trans(vsk)->queue_pair_max_size =
1570                         vmci_trans(psk)->queue_pair_max_size;
1571         } else {
1572                 vmci_trans(vsk)->queue_pair_size =
1573                         VMCI_TRANSPORT_DEFAULT_QP_SIZE;
1574                 vmci_trans(vsk)->queue_pair_min_size =
1575                          VMCI_TRANSPORT_DEFAULT_QP_SIZE_MIN;
1576                 vmci_trans(vsk)->queue_pair_max_size =
1577                         VMCI_TRANSPORT_DEFAULT_QP_SIZE_MAX;
1578         }
1579
1580         return 0;
1581 }
1582
1583 static void vmci_transport_free_resources(struct list_head *transport_list)
1584 {
1585         while (!list_empty(transport_list)) {
1586                 struct vmci_transport *transport =
1587                     list_first_entry(transport_list, struct vmci_transport,
1588                                      elem);
1589                 list_del(&transport->elem);
1590
1591                 if (transport->detach_sub_id != VMCI_INVALID_ID) {
1592                         vmci_event_unsubscribe(transport->detach_sub_id);
1593                         transport->detach_sub_id = VMCI_INVALID_ID;
1594                 }
1595
1596                 if (!vmci_handle_is_invalid(transport->qp_handle)) {
1597                         vmci_qpair_detach(&transport->qpair);
1598                         transport->qp_handle = VMCI_INVALID_HANDLE;
1599                         transport->produce_size = 0;
1600                         transport->consume_size = 0;
1601                 }
1602
1603                 kfree(transport);
1604         }
1605 }
1606
1607 static void vmci_transport_cleanup(struct work_struct *work)
1608 {
1609         LIST_HEAD(pending);
1610
1611         spin_lock_bh(&vmci_transport_cleanup_lock);
1612         list_replace_init(&vmci_transport_cleanup_list, &pending);
1613         spin_unlock_bh(&vmci_transport_cleanup_lock);
1614         vmci_transport_free_resources(&pending);
1615 }
1616
1617 static void vmci_transport_destruct(struct vsock_sock *vsk)
1618 {
1619         /* Ensure that the detach callback doesn't use the sk/vsk
1620          * we are about to destruct.
1621          */
1622         spin_lock_bh(&vmci_trans(vsk)->lock);
1623         vmci_trans(vsk)->sk = NULL;
1624         spin_unlock_bh(&vmci_trans(vsk)->lock);
1625
1626         if (vmci_trans(vsk)->notify_ops)
1627                 vmci_trans(vsk)->notify_ops->socket_destruct(vsk);
1628
1629         spin_lock_bh(&vmci_transport_cleanup_lock);
1630         list_add(&vmci_trans(vsk)->elem, &vmci_transport_cleanup_list);
1631         spin_unlock_bh(&vmci_transport_cleanup_lock);
1632         schedule_work(&vmci_transport_cleanup_work);
1633
1634         vsk->trans = NULL;
1635 }
1636
1637 static void vmci_transport_release(struct vsock_sock *vsk)
1638 {
1639         vsock_remove_sock(vsk);
1640
1641         if (!vmci_handle_is_invalid(vmci_trans(vsk)->dg_handle)) {
1642                 vmci_datagram_destroy_handle(vmci_trans(vsk)->dg_handle);
1643                 vmci_trans(vsk)->dg_handle = VMCI_INVALID_HANDLE;
1644         }
1645 }
1646
1647 static int vmci_transport_dgram_bind(struct vsock_sock *vsk,
1648                                      struct sockaddr_vm *addr)
1649 {
1650         u32 port;
1651         u32 flags;
1652         int err;
1653
1654         /* VMCI will select a resource ID for us if we provide
1655          * VMCI_INVALID_ID.
1656          */
1657         port = addr->svm_port == VMADDR_PORT_ANY ?
1658                         VMCI_INVALID_ID : addr->svm_port;
1659
1660         if (port <= LAST_RESERVED_PORT && !capable(CAP_NET_BIND_SERVICE))
1661                 return -EACCES;
1662
1663         flags = addr->svm_cid == VMADDR_CID_ANY ?
1664                                 VMCI_FLAG_ANYCID_DG_HND : 0;
1665
1666         err = vmci_transport_datagram_create_hnd(port, flags,
1667                                                  vmci_transport_recv_dgram_cb,
1668                                                  &vsk->sk,
1669                                                  &vmci_trans(vsk)->dg_handle);
1670         if (err < VMCI_SUCCESS)
1671                 return vmci_transport_error_to_vsock_error(err);
1672         vsock_addr_init(&vsk->local_addr, addr->svm_cid,
1673                         vmci_trans(vsk)->dg_handle.resource);
1674
1675         return 0;
1676 }
1677
1678 static int vmci_transport_dgram_enqueue(
1679         struct vsock_sock *vsk,
1680         struct sockaddr_vm *remote_addr,
1681         struct msghdr *msg,
1682         size_t len)
1683 {
1684         int err;
1685         struct vmci_datagram *dg;
1686
1687         if (len > VMCI_MAX_DG_PAYLOAD_SIZE)
1688                 return -EMSGSIZE;
1689
1690         if (!vmci_transport_allow_dgram(vsk, remote_addr->svm_cid))
1691                 return -EPERM;
1692
1693         /* Allocate a buffer for the user's message and our packet header. */
1694         dg = kmalloc(len + sizeof(*dg), GFP_KERNEL);
1695         if (!dg)
1696                 return -ENOMEM;
1697
1698         memcpy_from_msg(VMCI_DG_PAYLOAD(dg), msg, len);
1699
1700         dg->dst = vmci_make_handle(remote_addr->svm_cid,
1701                                    remote_addr->svm_port);
1702         dg->src = vmci_make_handle(vsk->local_addr.svm_cid,
1703                                    vsk->local_addr.svm_port);
1704         dg->payload_size = len;
1705
1706         err = vmci_datagram_send(dg);
1707         kfree(dg);
1708         if (err < 0)
1709                 return vmci_transport_error_to_vsock_error(err);
1710
1711         return err - sizeof(*dg);
1712 }
1713
1714 static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
1715                                         struct msghdr *msg, size_t len,
1716                                         int flags)
1717 {
1718         int err;
1719         int noblock;
1720         struct vmci_datagram *dg;
1721         size_t payload_len;
1722         struct sk_buff *skb;
1723
1724         noblock = flags & MSG_DONTWAIT;
1725
1726         if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
1727                 return -EOPNOTSUPP;
1728
1729         /* Retrieve the head sk_buff from the socket's receive queue. */
1730         err = 0;
1731         skb = skb_recv_datagram(&vsk->sk, flags, noblock, &err);
1732         if (!skb)
1733                 return err;
1734
1735         dg = (struct vmci_datagram *)skb->data;
1736         if (!dg)
1737                 /* err is 0, meaning we read zero bytes. */
1738                 goto out;
1739
1740         payload_len = dg->payload_size;
1741         /* Ensure the sk_buff matches the payload size claimed in the packet. */
1742         if (payload_len != skb->len - sizeof(*dg)) {
1743                 err = -EINVAL;
1744                 goto out;
1745         }
1746
1747         if (payload_len > len) {
1748                 payload_len = len;
1749                 msg->msg_flags |= MSG_TRUNC;
1750         }
1751
1752         /* Place the datagram payload in the user's iovec. */
1753         err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
1754         if (err)
1755                 goto out;
1756
1757         if (msg->msg_name) {
1758                 /* Provide the address of the sender. */
1759                 DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
1760                 vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
1761                 msg->msg_namelen = sizeof(*vm_addr);
1762         }
1763         err = payload_len;
1764
1765 out:
1766         skb_free_datagram(&vsk->sk, skb);
1767         return err;
1768 }
1769
1770 static bool vmci_transport_dgram_allow(u32 cid, u32 port)
1771 {
1772         if (cid == VMADDR_CID_HYPERVISOR) {
1773                 /* Registrations of PBRPC Servers do not modify VMX/Hypervisor
1774                  * state and are allowed.
1775                  */
1776                 return port == VMCI_UNITY_PBRPC_REGISTER;
1777         }
1778
1779         return true;
1780 }
1781
1782 static int vmci_transport_connect(struct vsock_sock *vsk)
1783 {
1784         int err;
1785         bool old_pkt_proto = false;
1786         struct sock *sk = &vsk->sk;
1787
1788         if (vmci_transport_old_proto_override(&old_pkt_proto) &&
1789                 old_pkt_proto) {
1790                 err = vmci_transport_send_conn_request(
1791                         sk, vmci_trans(vsk)->queue_pair_size);
1792                 if (err < 0) {
1793                         sk->sk_state = SS_UNCONNECTED;
1794                         return err;
1795                 }
1796         } else {
1797                 int supported_proto_versions =
1798                         vmci_transport_new_proto_supported_versions();
1799                 err = vmci_transport_send_conn_request2(
1800                                 sk, vmci_trans(vsk)->queue_pair_size,
1801                                 supported_proto_versions);
1802                 if (err < 0) {
1803                         sk->sk_state = SS_UNCONNECTED;
1804                         return err;
1805                 }
1806
1807                 vsk->sent_request = true;
1808         }
1809
1810         return err;
1811 }
1812
1813 static ssize_t vmci_transport_stream_dequeue(
1814         struct vsock_sock *vsk,
1815         struct msghdr *msg,
1816         size_t len,
1817         int flags)
1818 {
1819         if (flags & MSG_PEEK)
1820                 return vmci_qpair_peekv(vmci_trans(vsk)->qpair, msg, len, 0);
1821         else
1822                 return vmci_qpair_dequev(vmci_trans(vsk)->qpair, msg, len, 0);
1823 }
1824
1825 static ssize_t vmci_transport_stream_enqueue(
1826         struct vsock_sock *vsk,
1827         struct msghdr *msg,
1828         size_t len)
1829 {
1830         return vmci_qpair_enquev(vmci_trans(vsk)->qpair, msg, len, 0);
1831 }
1832
1833 static s64 vmci_transport_stream_has_data(struct vsock_sock *vsk)
1834 {
1835         return vmci_qpair_consume_buf_ready(vmci_trans(vsk)->qpair);
1836 }
1837
1838 static s64 vmci_transport_stream_has_space(struct vsock_sock *vsk)
1839 {
1840         return vmci_qpair_produce_free_space(vmci_trans(vsk)->qpair);
1841 }
1842
1843 static u64 vmci_transport_stream_rcvhiwat(struct vsock_sock *vsk)
1844 {
1845         return vmci_trans(vsk)->consume_size;
1846 }
1847
1848 static bool vmci_transport_stream_is_active(struct vsock_sock *vsk)
1849 {
1850         return !vmci_handle_is_invalid(vmci_trans(vsk)->qp_handle);
1851 }
1852
1853 static u64 vmci_transport_get_buffer_size(struct vsock_sock *vsk)
1854 {
1855         return vmci_trans(vsk)->queue_pair_size;
1856 }
1857
1858 static u64 vmci_transport_get_min_buffer_size(struct vsock_sock *vsk)
1859 {
1860         return vmci_trans(vsk)->queue_pair_min_size;
1861 }
1862
1863 static u64 vmci_transport_get_max_buffer_size(struct vsock_sock *vsk)
1864 {
1865         return vmci_trans(vsk)->queue_pair_max_size;
1866 }
1867
1868 static void vmci_transport_set_buffer_size(struct vsock_sock *vsk, u64 val)
1869 {
1870         if (val < vmci_trans(vsk)->queue_pair_min_size)
1871                 vmci_trans(vsk)->queue_pair_min_size = val;
1872         if (val > vmci_trans(vsk)->queue_pair_max_size)
1873                 vmci_trans(vsk)->queue_pair_max_size = val;
1874         vmci_trans(vsk)->queue_pair_size = val;
1875 }
1876
1877 static void vmci_transport_set_min_buffer_size(struct vsock_sock *vsk,
1878                                                u64 val)
1879 {
1880         if (val > vmci_trans(vsk)->queue_pair_size)
1881                 vmci_trans(vsk)->queue_pair_size = val;
1882         vmci_trans(vsk)->queue_pair_min_size = val;
1883 }
1884
1885 static void vmci_transport_set_max_buffer_size(struct vsock_sock *vsk,
1886                                                u64 val)
1887 {
1888         if (val < vmci_trans(vsk)->queue_pair_size)
1889                 vmci_trans(vsk)->queue_pair_size = val;
1890         vmci_trans(vsk)->queue_pair_max_size = val;
1891 }
1892
1893 static int vmci_transport_notify_poll_in(
1894         struct vsock_sock *vsk,
1895         size_t target,
1896         bool *data_ready_now)
1897 {
1898         return vmci_trans(vsk)->notify_ops->poll_in(
1899                         &vsk->sk, target, data_ready_now);
1900 }
1901
1902 static int vmci_transport_notify_poll_out(
1903         struct vsock_sock *vsk,
1904         size_t target,
1905         bool *space_available_now)
1906 {
1907         return vmci_trans(vsk)->notify_ops->poll_out(
1908                         &vsk->sk, target, space_available_now);
1909 }
1910
1911 static int vmci_transport_notify_recv_init(
1912         struct vsock_sock *vsk,
1913         size_t target,
1914         struct vsock_transport_recv_notify_data *data)
1915 {
1916         return vmci_trans(vsk)->notify_ops->recv_init(
1917                         &vsk->sk, target,
1918                         (struct vmci_transport_recv_notify_data *)data);
1919 }
1920
1921 static int vmci_transport_notify_recv_pre_block(
1922         struct vsock_sock *vsk,
1923         size_t target,
1924         struct vsock_transport_recv_notify_data *data)
1925 {
1926         return vmci_trans(vsk)->notify_ops->recv_pre_block(
1927                         &vsk->sk, target,
1928                         (struct vmci_transport_recv_notify_data *)data);
1929 }
1930
1931 static int vmci_transport_notify_recv_pre_dequeue(
1932         struct vsock_sock *vsk,
1933         size_t target,
1934         struct vsock_transport_recv_notify_data *data)
1935 {
1936         return vmci_trans(vsk)->notify_ops->recv_pre_dequeue(
1937                         &vsk->sk, target,
1938                         (struct vmci_transport_recv_notify_data *)data);
1939 }
1940
1941 static int vmci_transport_notify_recv_post_dequeue(
1942         struct vsock_sock *vsk,
1943         size_t target,
1944         ssize_t copied,
1945         bool data_read,
1946         struct vsock_transport_recv_notify_data *data)
1947 {
1948         return vmci_trans(vsk)->notify_ops->recv_post_dequeue(
1949                         &vsk->sk, target, copied, data_read,
1950                         (struct vmci_transport_recv_notify_data *)data);
1951 }
1952
1953 static int vmci_transport_notify_send_init(
1954         struct vsock_sock *vsk,
1955         struct vsock_transport_send_notify_data *data)
1956 {
1957         return vmci_trans(vsk)->notify_ops->send_init(
1958                         &vsk->sk,
1959                         (struct vmci_transport_send_notify_data *)data);
1960 }
1961
1962 static int vmci_transport_notify_send_pre_block(
1963         struct vsock_sock *vsk,
1964         struct vsock_transport_send_notify_data *data)
1965 {
1966         return vmci_trans(vsk)->notify_ops->send_pre_block(
1967                         &vsk->sk,
1968                         (struct vmci_transport_send_notify_data *)data);
1969 }
1970
1971 static int vmci_transport_notify_send_pre_enqueue(
1972         struct vsock_sock *vsk,
1973         struct vsock_transport_send_notify_data *data)
1974 {
1975         return vmci_trans(vsk)->notify_ops->send_pre_enqueue(
1976                         &vsk->sk,
1977                         (struct vmci_transport_send_notify_data *)data);
1978 }
1979
1980 static int vmci_transport_notify_send_post_enqueue(
1981         struct vsock_sock *vsk,
1982         ssize_t written,
1983         struct vsock_transport_send_notify_data *data)
1984 {
1985         return vmci_trans(vsk)->notify_ops->send_post_enqueue(
1986                         &vsk->sk, written,
1987                         (struct vmci_transport_send_notify_data *)data);
1988 }
1989
1990 static bool vmci_transport_old_proto_override(bool *old_pkt_proto)
1991 {
1992         if (PROTOCOL_OVERRIDE != -1) {
1993                 if (PROTOCOL_OVERRIDE == 0)
1994                         *old_pkt_proto = true;
1995                 else
1996                         *old_pkt_proto = false;
1997
1998                 pr_info("Proto override in use\n");
1999                 return true;
2000         }
2001
2002         return false;
2003 }
2004
2005 static bool vmci_transport_proto_to_notify_struct(struct sock *sk,
2006                                                   u16 *proto,
2007                                                   bool old_pkt_proto)
2008 {
2009         struct vsock_sock *vsk = vsock_sk(sk);
2010
2011         if (old_pkt_proto) {
2012                 if (*proto != VSOCK_PROTO_INVALID) {
2013                         pr_err("Can't set both an old and new protocol\n");
2014                         return false;
2015                 }
2016                 vmci_trans(vsk)->notify_ops = &vmci_transport_notify_pkt_ops;
2017                 goto exit;
2018         }
2019
2020         switch (*proto) {
2021         case VSOCK_PROTO_PKT_ON_NOTIFY:
2022                 vmci_trans(vsk)->notify_ops =
2023                         &vmci_transport_notify_pkt_q_state_ops;
2024                 break;
2025         default:
2026                 pr_err("Unknown notify protocol version\n");
2027                 return false;
2028         }
2029
2030 exit:
2031         vmci_trans(vsk)->notify_ops->socket_init(sk);
2032         return true;
2033 }
2034
2035 static u16 vmci_transport_new_proto_supported_versions(void)
2036 {
2037         if (PROTOCOL_OVERRIDE != -1)
2038                 return PROTOCOL_OVERRIDE;
2039
2040         return VSOCK_PROTO_ALL_SUPPORTED;
2041 }
2042
2043 static u32 vmci_transport_get_local_cid(void)
2044 {
2045         return vmci_get_context_id();
2046 }
2047
2048 static const struct vsock_transport vmci_transport = {
2049         .init = vmci_transport_socket_init,
2050         .destruct = vmci_transport_destruct,
2051         .release = vmci_transport_release,
2052         .connect = vmci_transport_connect,
2053         .dgram_bind = vmci_transport_dgram_bind,
2054         .dgram_dequeue = vmci_transport_dgram_dequeue,
2055         .dgram_enqueue = vmci_transport_dgram_enqueue,
2056         .dgram_allow = vmci_transport_dgram_allow,
2057         .stream_dequeue = vmci_transport_stream_dequeue,
2058         .stream_enqueue = vmci_transport_stream_enqueue,
2059         .stream_has_data = vmci_transport_stream_has_data,
2060         .stream_has_space = vmci_transport_stream_has_space,
2061         .stream_rcvhiwat = vmci_transport_stream_rcvhiwat,
2062         .stream_is_active = vmci_transport_stream_is_active,
2063         .stream_allow = vmci_transport_stream_allow,
2064         .notify_poll_in = vmci_transport_notify_poll_in,
2065         .notify_poll_out = vmci_transport_notify_poll_out,
2066         .notify_recv_init = vmci_transport_notify_recv_init,
2067         .notify_recv_pre_block = vmci_transport_notify_recv_pre_block,
2068         .notify_recv_pre_dequeue = vmci_transport_notify_recv_pre_dequeue,
2069         .notify_recv_post_dequeue = vmci_transport_notify_recv_post_dequeue,
2070         .notify_send_init = vmci_transport_notify_send_init,
2071         .notify_send_pre_block = vmci_transport_notify_send_pre_block,
2072         .notify_send_pre_enqueue = vmci_transport_notify_send_pre_enqueue,
2073         .notify_send_post_enqueue = vmci_transport_notify_send_post_enqueue,
2074         .shutdown = vmci_transport_shutdown,
2075         .set_buffer_size = vmci_transport_set_buffer_size,
2076         .set_min_buffer_size = vmci_transport_set_min_buffer_size,
2077         .set_max_buffer_size = vmci_transport_set_max_buffer_size,
2078         .get_buffer_size = vmci_transport_get_buffer_size,
2079         .get_min_buffer_size = vmci_transport_get_min_buffer_size,
2080         .get_max_buffer_size = vmci_transport_get_max_buffer_size,
2081         .get_local_cid = vmci_transport_get_local_cid,
2082 };
2083
2084 static int __init vmci_transport_init(void)
2085 {
2086         int err;
2087
2088         /* Create the datagram handle that we will use to send and receive all
2089          * VSocket control messages for this context.
2090          */
2091         err = vmci_transport_datagram_create_hnd(VMCI_TRANSPORT_PACKET_RID,
2092                                                  VMCI_FLAG_ANYCID_DG_HND,
2093                                                  vmci_transport_recv_stream_cb,
2094                                                  NULL,
2095                                                  &vmci_transport_stream_handle);
2096         if (err < VMCI_SUCCESS) {
2097                 pr_err("Unable to create datagram handle. (%d)\n", err);
2098                 return vmci_transport_error_to_vsock_error(err);
2099         }
2100
2101         err = vmci_event_subscribe(VMCI_EVENT_QP_RESUMED,
2102                                    vmci_transport_qp_resumed_cb,
2103                                    NULL, &vmci_transport_qp_resumed_sub_id);
2104         if (err < VMCI_SUCCESS) {
2105                 pr_err("Unable to subscribe to resumed event. (%d)\n", err);
2106                 err = vmci_transport_error_to_vsock_error(err);
2107                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2108                 goto err_destroy_stream_handle;
2109         }
2110
2111         err = vsock_core_init(&vmci_transport);
2112         if (err < 0)
2113                 goto err_unsubscribe;
2114
2115         return 0;
2116
2117 err_unsubscribe:
2118         vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2119 err_destroy_stream_handle:
2120         vmci_datagram_destroy_handle(vmci_transport_stream_handle);
2121         return err;
2122 }
2123 module_init(vmci_transport_init);
2124
2125 static void __exit vmci_transport_exit(void)
2126 {
2127         cancel_work_sync(&vmci_transport_cleanup_work);
2128         vmci_transport_free_resources(&vmci_transport_cleanup_list);
2129
2130         if (!vmci_handle_is_invalid(vmci_transport_stream_handle)) {
2131                 if (vmci_datagram_destroy_handle(
2132                         vmci_transport_stream_handle) != VMCI_SUCCESS)
2133                         pr_err("Couldn't destroy datagram handle\n");
2134                 vmci_transport_stream_handle = VMCI_INVALID_HANDLE;
2135         }
2136
2137         if (vmci_transport_qp_resumed_sub_id != VMCI_INVALID_ID) {
2138                 vmci_event_unsubscribe(vmci_transport_qp_resumed_sub_id);
2139                 vmci_transport_qp_resumed_sub_id = VMCI_INVALID_ID;
2140         }
2141
2142         vsock_core_exit();
2143 }
2144 module_exit(vmci_transport_exit);
2145
2146 MODULE_AUTHOR("VMware, Inc.");
2147 MODULE_DESCRIPTION("VMCI transport for Virtual Sockets");
2148 MODULE_VERSION("1.0.4.0-k");
2149 MODULE_LICENSE("GPL v2");
2150 MODULE_ALIAS("vmware_vsock");
2151 MODULE_ALIAS_NETPROTO(PF_VSOCK);