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