vmci_host.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * VMware VMCI Driver
   4 *
   5 * Copyright (C) 2012 VMware, Inc. All rights reserved.
   6 */
   7
   8#include <linux/vmw_vmci_defs.h>
   9#include <linux/vmw_vmci_api.h>
  10#include <linux/miscdevice.h>
  11#include <linux/interrupt.h>
  12#include <linux/highmem.h>
  13#include <linux/atomic.h>
  14#include <linux/kernel.h>
  15#include <linux/module.h>
  16#include <linux/mutex.h>
  17#include <linux/sched.h>
  18#include <linux/cred.h>
  19#include <linux/slab.h>
  20#include <linux/file.h>
  21#include <linux/init.h>
  22#include <linux/poll.h>
  23#include <linux/pci.h>
  24#include <linux/smp.h>
  25#include <linux/fs.h>
  26#include <linux/io.h>
  27
  28#include "vmci_handle_array.h"
  29#include "vmci_queue_pair.h"
  30#include "vmci_datagram.h"
  31#include "vmci_doorbell.h"
  32#include "vmci_resource.h"
  33#include "vmci_context.h"
  34#include "vmci_driver.h"
  35#include "vmci_event.h"
  36
  37#define VMCI_UTIL_NUM_RESOURCES 1
  38
  39enum {
  40	VMCI_NOTIFY_RESOURCE_QUEUE_PAIR = 0,
  41	VMCI_NOTIFY_RESOURCE_DOOR_BELL = 1,
  42};
  43
  44enum {
  45	VMCI_NOTIFY_RESOURCE_ACTION_NOTIFY = 0,
  46	VMCI_NOTIFY_RESOURCE_ACTION_CREATE = 1,
  47	VMCI_NOTIFY_RESOURCE_ACTION_DESTROY = 2,
  48};
  49
  50/*
  51 * VMCI driver initialization. This block can also be used to
  52 * pass initial group membership etc.
  53 */
  54struct vmci_init_blk {
  55	u32 cid;
  56	u32 flags;
  57};
  58
  59/* VMCIqueue_pairAllocInfo_VMToVM */
  60struct vmci_qp_alloc_info_vmvm {
  61	struct vmci_handle handle;
  62	u32 peer;
  63	u32 flags;
  64	u64 produce_size;
  65	u64 consume_size;
  66	u64 produce_page_file;	  /* User VA. */
  67	u64 consume_page_file;	  /* User VA. */
  68	u64 produce_page_file_size;  /* Size of the file name array. */
  69	u64 consume_page_file_size;  /* Size of the file name array. */
  70	s32 result;
  71	u32 _pad;
  72};
  73
  74/* VMCISetNotifyInfo: Used to pass notify flag's address to the host driver. */
  75struct vmci_set_notify_info {
  76	u64 notify_uva;
  77	s32 result;
  78	u32 _pad;
  79};
  80
  81/*
  82 * Per-instance host state
  83 */
  84struct vmci_host_dev {
  85	struct vmci_ctx *context;
  86	int user_version;
  87	enum vmci_obj_type ct_type;
  88	struct mutex lock;  /* Mutex lock for vmci context access */
  89};
  90
  91static struct vmci_ctx *host_context;
  92static bool vmci_host_device_initialized;
  93static atomic_t vmci_host_active_users = ATOMIC_INIT(0);
  94
  95/*
  96 * Determines whether the VMCI host personality is
  97 * available. Since the core functionality of the host driver is
  98 * always present, all guests could possibly use the host
  99 * personality. However, to minimize the deviation from the
 100 * pre-unified driver state of affairs, we only consider the host
 101 * device active if there is no active guest device or if there
 102 * are VMX'en with active VMCI contexts using the host device.
 103 */
 104bool vmci_host_code_active(void)
 105{
 106	return vmci_host_device_initialized &&
 107	    (!vmci_guest_code_active() ||
 108	     atomic_read(&vmci_host_active_users) > 0);
 109}
 110
 111int vmci_host_users(void)
 112{
 113	return atomic_read(&vmci_host_active_users);
 114}
 115
 116/*
 117 * Called on open of /dev/vmci.
 118 */
 119static int vmci_host_open(struct inode *inode, struct file *filp)
 120{
 121	struct vmci_host_dev *vmci_host_dev;
 122
 123	vmci_host_dev = kzalloc(sizeof(struct vmci_host_dev), GFP_KERNEL);
 124	if (vmci_host_dev == NULL)
 125		return -ENOMEM;
 126
 127	vmci_host_dev->ct_type = VMCIOBJ_NOT_SET;
 128	mutex_init(&vmci_host_dev->lock);
 129	filp->private_data = vmci_host_dev;
 130
 131	return 0;
 132}
 133
 134/*
 135 * Called on close of /dev/vmci, most often when the process
 136 * exits.
 137 */
 138static int vmci_host_close(struct inode *inode, struct file *filp)
 139{
 140	struct vmci_host_dev *vmci_host_dev = filp->private_data;
 141
 142	if (vmci_host_dev->ct_type == VMCIOBJ_CONTEXT) {
 143		vmci_ctx_destroy(vmci_host_dev->context);
 144		vmci_host_dev->context = NULL;
 145
 146		/*
 147		 * The number of active contexts is used to track whether any
 148		 * VMX'en are using the host personality. It is incremented when
 149		 * a context is created through the IOCTL_VMCI_INIT_CONTEXT
 150		 * ioctl.
 151		 */
 152		atomic_dec(&vmci_host_active_users);
 153	}
 154	vmci_host_dev->ct_type = VMCIOBJ_NOT_SET;
 155
 156	kfree(vmci_host_dev);
 157	filp->private_data = NULL;
 158	return 0;
 159}
 160
 161/*
 162 * This is used to wake up the VMX when a VMCI call arrives, or
 163 * to wake up select() or poll() at the next clock tick.
 164 */
 165static __poll_t vmci_host_poll(struct file *filp, poll_table *wait)
 166{
 167	struct vmci_host_dev *vmci_host_dev = filp->private_data;
 168	struct vmci_ctx *context;
 169	__poll_t mask = 0;
 170
 171	if (vmci_host_dev->ct_type == VMCIOBJ_CONTEXT) {
 172		/*
 173		 * Read context only if ct_type == VMCIOBJ_CONTEXT to make
 174		 * sure that context is initialized
 175		 */
 176		context = vmci_host_dev->context;
 177
 178		/* Check for VMCI calls to this VM context. */
 179		if (wait)
 180			poll_wait(filp, &context->host_context.wait_queue,
 181				  wait);
 182
 183		spin_lock(&context->lock);
 184		if (context->pending_datagrams > 0 ||
 185		    vmci_handle_arr_get_size(
 186				context->pending_doorbell_array) > 0) {
 187			mask = EPOLLIN;
 188		}
 189		spin_unlock(&context->lock);
 190	}
 191	return mask;
 192}
 193
 194/*
 195 * Copies the handles of a handle array into a user buffer, and
 196 * returns the new length in userBufferSize. If the copy to the
 197 * user buffer fails, the functions still returns VMCI_SUCCESS,
 198 * but retval != 0.
 199 */
 200static int drv_cp_harray_to_user(void __user *user_buf_uva,
 201				 u64 *user_buf_size,
 202				 struct vmci_handle_arr *handle_array,
 203				 int *retval)
 204{
 205	u32 array_size = 0;
 206	struct vmci_handle *handles;
 207
 208	if (handle_array)
 209		array_size = vmci_handle_arr_get_size(handle_array);
 210
 211	if (array_size * sizeof(*handles) > *user_buf_size)
 212		return VMCI_ERROR_MORE_DATA;
 213
 214	*user_buf_size = array_size * sizeof(*handles);
 215	if (*user_buf_size)
 216		*retval = copy_to_user(user_buf_uva,
 217				       vmci_handle_arr_get_handles
 218				       (handle_array), *user_buf_size);
 219
 220	return VMCI_SUCCESS;
 221}
 222
 223/*
 224 * Sets up a given context for notify to work. Maps the notify
 225 * boolean in user VA into kernel space.
 226 */
 227static int vmci_host_setup_notify(struct vmci_ctx *context,
 228				  unsigned long uva)
 229{
 230	struct page *page;
 231	int retval;
 232
 233	if (context->notify_page) {
 234		pr_devel("%s: Notify mechanism is already set up\n", __func__);
 235		return VMCI_ERROR_DUPLICATE_ENTRY;
 236	}
 237
 238	/*
 239	 * We are using 'bool' internally, but let's make sure we explicit
 240	 * about the size.
 241	 */
 242	BUILD_BUG_ON(sizeof(bool) != sizeof(u8));
 243
 244	/*
 245	 * Lock physical page backing a given user VA.
 246	 */
 247	retval = get_user_pages_fast(uva, 1, FOLL_WRITE, &page);
 248	if (retval != 1)
 249		return VMCI_ERROR_GENERIC;
 250
 251	context->notify_page = page;
 252
 253	/*
 254	 * Map the locked page and set up notify pointer.
 255	 */
 256	context->notify = kmap(context->notify_page) + (uva & (PAGE_SIZE - 1));
 257	vmci_ctx_check_signal_notify(context);
 258
 259	return VMCI_SUCCESS;
 260}
 261
 262static int vmci_host_get_version(struct vmci_host_dev *vmci_host_dev,
 263				 unsigned int cmd, void __user *uptr)
 264{
 265	if (cmd == IOCTL_VMCI_VERSION2) {
 266		int __user *vptr = uptr;
 267		if (get_user(vmci_host_dev->user_version, vptr))
 268			return -EFAULT;
 269	}
 270
 271	/*
 272	 * The basic logic here is:
 273	 *
 274	 * If the user sends in a version of 0 tell it our version.
 275	 * If the user didn't send in a version, tell it our version.
 276	 * If the user sent in an old version, tell it -its- version.
 277	 * If the user sent in an newer version, tell it our version.
 278	 *
 279	 * The rationale behind telling the caller its version is that
 280	 * Workstation 6.5 required that VMX and VMCI kernel module were
 281	 * version sync'd.  All new VMX users will be programmed to
 282	 * handle the VMCI kernel module version.
 283	 */
 284
 285	if (vmci_host_dev->user_version > 0 &&
 286	    vmci_host_dev->user_version < VMCI_VERSION_HOSTQP) {
 287		return vmci_host_dev->user_version;
 288	}
 289
 290	return VMCI_VERSION;
 291}
 292
 293#define vmci_ioctl_err(fmt, ...)	\
 294	pr_devel("%s: " fmt, ioctl_name, ##__VA_ARGS__)
 295
 296static int vmci_host_do_init_context(struct vmci_host_dev *vmci_host_dev,
 297				     const char *ioctl_name,
 298				     void __user *uptr)
 299{
 300	struct vmci_init_blk init_block;
 301	const struct cred *cred;
 302	int retval;
 303
 304	if (copy_from_user(&init_block, uptr, sizeof(init_block))) {
 305		vmci_ioctl_err("error reading init block\n");
 306		return -EFAULT;
 307	}
 308
 309	mutex_lock(&vmci_host_dev->lock);
 310
 311	if (vmci_host_dev->ct_type != VMCIOBJ_NOT_SET) {
 312		vmci_ioctl_err("received VMCI init on initialized handle\n");
 313		retval = -EINVAL;
 314		goto out;
 315	}
 316
 317	if (init_block.flags & ~VMCI_PRIVILEGE_FLAG_RESTRICTED) {
 318		vmci_ioctl_err("unsupported VMCI restriction flag\n");
 319		retval = -EINVAL;
 320		goto out;
 321	}
 322
 323	cred = get_current_cred();
 324	vmci_host_dev->context = vmci_ctx_create(init_block.cid,
 325						 init_block.flags, 0,
 326						 vmci_host_dev->user_version,
 327						 cred);
 328	put_cred(cred);
 329	if (IS_ERR(vmci_host_dev->context)) {
 330		retval = PTR_ERR(vmci_host_dev->context);
 331		vmci_ioctl_err("error initializing context\n");
 332		goto out;
 333	}
 334
 335	/*
 336	 * Copy cid to userlevel, we do this to allow the VMX
 337	 * to enforce its policy on cid generation.
 338	 */
 339	init_block.cid = vmci_ctx_get_id(vmci_host_dev->context);
 340	if (copy_to_user(uptr, &init_block, sizeof(init_block))) {
 341		vmci_ctx_destroy(vmci_host_dev->context);
 342		vmci_host_dev->context = NULL;
 343		vmci_ioctl_err("error writing init block\n");
 344		retval = -EFAULT;
 345		goto out;
 346	}
 347
 348	vmci_host_dev->ct_type = VMCIOBJ_CONTEXT;
 349	atomic_inc(&vmci_host_active_users);
 350
 351	vmci_call_vsock_callback(true);
 352
 353	retval = 0;
 354
 355out:
 356	mutex_unlock(&vmci_host_dev->lock);
 357	return retval;
 358}
 359
 360static int vmci_host_do_send_datagram(struct vmci_host_dev *vmci_host_dev,
 361				      const char *ioctl_name,
 362				      void __user *uptr)
 363{
 364	struct vmci_datagram_snd_rcv_info send_info;
 365	struct vmci_datagram *dg = NULL;
 366	u32 cid;
 367
 368	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 369		vmci_ioctl_err("only valid for contexts\n");
 370		return -EINVAL;
 371	}
 372
 373	if (copy_from_user(&send_info, uptr, sizeof(send_info)))
 374		return -EFAULT;
 375
 376	if (send_info.len > VMCI_MAX_DG_SIZE) {
 377		vmci_ioctl_err("datagram is too big (size=%d)\n",
 378			       send_info.len);
 379		return -EINVAL;
 380	}
 381
 382	if (send_info.len < sizeof(*dg)) {
 383		vmci_ioctl_err("datagram is too small (size=%d)\n",
 384			       send_info.len);
 385		return -EINVAL;
 386	}
 387
 388	dg = memdup_user((void __user *)(uintptr_t)send_info.addr,
 389			 send_info.len);
 390	if (IS_ERR(dg)) {
 391		vmci_ioctl_err(
 392			"cannot allocate memory to dispatch datagram\n");
 393		return PTR_ERR(dg);
 394	}
 395
 396	if (VMCI_DG_SIZE(dg) != send_info.len) {
 397		vmci_ioctl_err("datagram size mismatch\n");
 398		kfree(dg);
 399		return -EINVAL;
 400	}
 401
 402	pr_devel("Datagram dst (handle=0x%x:0x%x) src (handle=0x%x:0x%x), payload (size=%llu bytes)\n",
 403		 dg->dst.context, dg->dst.resource,
 404		 dg->src.context, dg->src.resource,
 405		 (unsigned long long)dg->payload_size);
 406
 407	/* Get source context id. */
 408	cid = vmci_ctx_get_id(vmci_host_dev->context);
 409	send_info.result = vmci_datagram_dispatch(cid, dg, true);
 410	kfree(dg);
 411
 412	return copy_to_user(uptr, &send_info, sizeof(send_info)) ? -EFAULT : 0;
 413}
 414
 415static int vmci_host_do_receive_datagram(struct vmci_host_dev *vmci_host_dev,
 416					 const char *ioctl_name,
 417					 void __user *uptr)
 418{
 419	struct vmci_datagram_snd_rcv_info recv_info;
 420	struct vmci_datagram *dg = NULL;
 421	int retval;
 422	size_t size;
 423
 424	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 425		vmci_ioctl_err("only valid for contexts\n");
 426		return -EINVAL;
 427	}
 428
 429	if (copy_from_user(&recv_info, uptr, sizeof(recv_info)))
 430		return -EFAULT;
 431
 432	size = recv_info.len;
 433	recv_info.result = vmci_ctx_dequeue_datagram(vmci_host_dev->context,
 434						     &size, &dg);
 435
 436	if (recv_info.result >= VMCI_SUCCESS) {
 437		void __user *ubuf = (void __user *)(uintptr_t)recv_info.addr;
 438		retval = copy_to_user(ubuf, dg, VMCI_DG_SIZE(dg));
 439		kfree(dg);
 440		if (retval != 0)
 441			return -EFAULT;
 442	}
 443
 444	return copy_to_user(uptr, &recv_info, sizeof(recv_info)) ? -EFAULT : 0;
 445}
 446
 447static int vmci_host_do_alloc_queuepair(struct vmci_host_dev *vmci_host_dev,
 448					const char *ioctl_name,
 449					void __user *uptr)
 450{
 451	struct vmci_handle handle;
 452	int vmci_status;
 453	int __user *retptr;
 454
 455	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 456		vmci_ioctl_err("only valid for contexts\n");
 457		return -EINVAL;
 458	}
 459
 460	if (vmci_host_dev->user_version < VMCI_VERSION_NOVMVM) {
 461		struct vmci_qp_alloc_info_vmvm alloc_info;
 462		struct vmci_qp_alloc_info_vmvm __user *info = uptr;
 463
 464		if (copy_from_user(&alloc_info, uptr, sizeof(alloc_info)))
 465			return -EFAULT;
 466
 467		handle = alloc_info.handle;
 468		retptr = &info->result;
 469
 470		vmci_status = vmci_qp_broker_alloc(alloc_info.handle,
 471						alloc_info.peer,
 472						alloc_info.flags,
 473						VMCI_NO_PRIVILEGE_FLAGS,
 474						alloc_info.produce_size,
 475						alloc_info.consume_size,
 476						NULL,
 477						vmci_host_dev->context);
 478
 479		if (vmci_status == VMCI_SUCCESS)
 480			vmci_status = VMCI_SUCCESS_QUEUEPAIR_CREATE;
 481	} else {
 482		struct vmci_qp_alloc_info alloc_info;
 483		struct vmci_qp_alloc_info __user *info = uptr;
 484		struct vmci_qp_page_store page_store;
 485
 486		if (copy_from_user(&alloc_info, uptr, sizeof(alloc_info)))
 487			return -EFAULT;
 488
 489		handle = alloc_info.handle;
 490		retptr = &info->result;
 491
 492		page_store.pages = alloc_info.ppn_va;
 493		page_store.len = alloc_info.num_ppns;
 494
 495		vmci_status = vmci_qp_broker_alloc(alloc_info.handle,
 496						alloc_info.peer,
 497						alloc_info.flags,
 498						VMCI_NO_PRIVILEGE_FLAGS,
 499						alloc_info.produce_size,
 500						alloc_info.consume_size,
 501						&page_store,
 502						vmci_host_dev->context);
 503	}
 504
 505	if (put_user(vmci_status, retptr)) {
 506		if (vmci_status >= VMCI_SUCCESS) {
 507			vmci_status = vmci_qp_broker_detach(handle,
 508							vmci_host_dev->context);
 509		}
 510		return -EFAULT;
 511	}
 512
 513	return 0;
 514}
 515
 516static int vmci_host_do_queuepair_setva(struct vmci_host_dev *vmci_host_dev,
 517					const char *ioctl_name,
 518					void __user *uptr)
 519{
 520	struct vmci_qp_set_va_info set_va_info;
 521	struct vmci_qp_set_va_info __user *info = uptr;
 522	s32 result;
 523
 524	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 525		vmci_ioctl_err("only valid for contexts\n");
 526		return -EINVAL;
 527	}
 528
 529	if (vmci_host_dev->user_version < VMCI_VERSION_NOVMVM) {
 530		vmci_ioctl_err("is not allowed\n");
 531		return -EINVAL;
 532	}
 533
 534	if (copy_from_user(&set_va_info, uptr, sizeof(set_va_info)))
 535		return -EFAULT;
 536
 537	if (set_va_info.va) {
 538		/*
 539		 * VMX is passing down a new VA for the queue
 540		 * pair mapping.
 541		 */
 542		result = vmci_qp_broker_map(set_va_info.handle,
 543					    vmci_host_dev->context,
 544					    set_va_info.va);
 545	} else {
 546		/*
 547		 * The queue pair is about to be unmapped by
 548		 * the VMX.
 549		 */
 550		result = vmci_qp_broker_unmap(set_va_info.handle,
 551					 vmci_host_dev->context, 0);
 552	}
 553
 554	return put_user(result, &info->result) ? -EFAULT : 0;
 555}
 556
 557static int vmci_host_do_queuepair_setpf(struct vmci_host_dev *vmci_host_dev,
 558					const char *ioctl_name,
 559					void __user *uptr)
 560{
 561	struct vmci_qp_page_file_info page_file_info;
 562	struct vmci_qp_page_file_info __user *info = uptr;
 563	s32 result;
 564
 565	if (vmci_host_dev->user_version < VMCI_VERSION_HOSTQP ||
 566	    vmci_host_dev->user_version >= VMCI_VERSION_NOVMVM) {
 567		vmci_ioctl_err("not supported on this VMX (version=%d)\n",
 568			       vmci_host_dev->user_version);
 569		return -EINVAL;
 570	}
 571
 572	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 573		vmci_ioctl_err("only valid for contexts\n");
 574		return -EINVAL;
 575	}
 576
 577	if (copy_from_user(&page_file_info, uptr, sizeof(*info)))
 578		return -EFAULT;
 579
 580	/*
 581	 * Communicate success pre-emptively to the caller.  Note that the
 582	 * basic premise is that it is incumbent upon the caller not to look at
 583	 * the info.result field until after the ioctl() returns.  And then,
 584	 * only if the ioctl() result indicates no error.  We send up the
 585	 * SUCCESS status before calling SetPageStore() store because failing
 586	 * to copy up the result code means unwinding the SetPageStore().
 587	 *
 588	 * It turns out the logic to unwind a SetPageStore() opens a can of
 589	 * worms.  For example, if a host had created the queue_pair and a
 590	 * guest attaches and SetPageStore() is successful but writing success
 591	 * fails, then ... the host has to be stopped from writing (anymore)
 592	 * data into the queue_pair.  That means an additional test in the
 593	 * VMCI_Enqueue() code path.  Ugh.
 594	 */
 595
 596	if (put_user(VMCI_SUCCESS, &info->result)) {
 597		/*
 598		 * In this case, we can't write a result field of the
 599		 * caller's info block.  So, we don't even try to
 600		 * SetPageStore().
 601		 */
 602		return -EFAULT;
 603	}
 604
 605	result = vmci_qp_broker_set_page_store(page_file_info.handle,
 606						page_file_info.produce_va,
 607						page_file_info.consume_va,
 608						vmci_host_dev->context);
 609	if (result < VMCI_SUCCESS) {
 610		if (put_user(result, &info->result)) {
 611			/*
 612			 * Note that in this case the SetPageStore()
 613			 * call failed but we were unable to
 614			 * communicate that to the caller (because the
 615			 * copy_to_user() call failed).  So, if we
 616			 * simply return an error (in this case
 617			 * -EFAULT) then the caller will know that the
 618			 *  SetPageStore failed even though we couldn't
 619			 *  put the result code in the result field and
 620			 *  indicate exactly why it failed.
 621			 *
 622			 * That says nothing about the issue where we
 623			 * were once able to write to the caller's info
 624			 * memory and now can't.  Something more
 625			 * serious is probably going on than the fact
 626			 * that SetPageStore() didn't work.
 627			 */
 628			return -EFAULT;
 629		}
 630	}
 631
 632	return 0;
 633}
 634
 635static int vmci_host_do_qp_detach(struct vmci_host_dev *vmci_host_dev,
 636				  const char *ioctl_name,
 637				  void __user *uptr)
 638{
 639	struct vmci_qp_dtch_info detach_info;
 640	struct vmci_qp_dtch_info __user *info = uptr;
 641	s32 result;
 642
 643	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 644		vmci_ioctl_err("only valid for contexts\n");
 645		return -EINVAL;
 646	}
 647
 648	if (copy_from_user(&detach_info, uptr, sizeof(detach_info)))
 649		return -EFAULT;
 650
 651	result = vmci_qp_broker_detach(detach_info.handle,
 652				       vmci_host_dev->context);
 653	if (result == VMCI_SUCCESS &&
 654	    vmci_host_dev->user_version < VMCI_VERSION_NOVMVM) {
 655		result = VMCI_SUCCESS_LAST_DETACH;
 656	}
 657
 658	return put_user(result, &info->result) ? -EFAULT : 0;
 659}
 660
 661static int vmci_host_do_ctx_add_notify(struct vmci_host_dev *vmci_host_dev,
 662				       const char *ioctl_name,
 663				       void __user *uptr)
 664{
 665	struct vmci_ctx_info ar_info;
 666	struct vmci_ctx_info __user *info = uptr;
 667	s32 result;
 668	u32 cid;
 669
 670	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 671		vmci_ioctl_err("only valid for contexts\n");
 672		return -EINVAL;
 673	}
 674
 675	if (copy_from_user(&ar_info, uptr, sizeof(ar_info)))
 676		return -EFAULT;
 677
 678	cid = vmci_ctx_get_id(vmci_host_dev->context);
 679	result = vmci_ctx_add_notification(cid, ar_info.remote_cid);
 680
 681	return put_user(result, &info->result) ? -EFAULT : 0;
 682}
 683
 684static int vmci_host_do_ctx_remove_notify(struct vmci_host_dev *vmci_host_dev,
 685					  const char *ioctl_name,
 686					  void __user *uptr)
 687{
 688	struct vmci_ctx_info ar_info;
 689	struct vmci_ctx_info __user *info = uptr;
 690	u32 cid;
 691	int result;
 692
 693	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 694		vmci_ioctl_err("only valid for contexts\n");
 695		return -EINVAL;
 696	}
 697
 698	if (copy_from_user(&ar_info, uptr, sizeof(ar_info)))
 699		return -EFAULT;
 700
 701	cid = vmci_ctx_get_id(vmci_host_dev->context);
 702	result = vmci_ctx_remove_notification(cid,
 703					      ar_info.remote_cid);
 704
 705	return put_user(result, &info->result) ? -EFAULT : 0;
 706}
 707
 708static int vmci_host_do_ctx_get_cpt_state(struct vmci_host_dev *vmci_host_dev,
 709					  const char *ioctl_name,
 710					  void __user *uptr)
 711{
 712	struct vmci_ctx_chkpt_buf_info get_info;
 713	u32 cid;
 714	void *cpt_buf;
 715	int retval;
 716
 717	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 718		vmci_ioctl_err("only valid for contexts\n");
 719		return -EINVAL;
 720	}
 721
 722	if (copy_from_user(&get_info, uptr, sizeof(get_info)))
 723		return -EFAULT;
 724
 725	cid = vmci_ctx_get_id(vmci_host_dev->context);
 726	get_info.result = vmci_ctx_get_chkpt_state(cid, get_info.cpt_type,
 727						&get_info.buf_size, &cpt_buf);
 728	if (get_info.result == VMCI_SUCCESS && get_info.buf_size) {
 729		void __user *ubuf = (void __user *)(uintptr_t)get_info.cpt_buf;
 730		retval = copy_to_user(ubuf, cpt_buf, get_info.buf_size);
 731		kfree(cpt_buf);
 732
 733		if (retval)
 734			return -EFAULT;
 735	}
 736
 737	return copy_to_user(uptr, &get_info, sizeof(get_info)) ? -EFAULT : 0;
 738}
 739
 740static int vmci_host_do_ctx_set_cpt_state(struct vmci_host_dev *vmci_host_dev,
 741					  const char *ioctl_name,
 742					  void __user *uptr)
 743{
 744	struct vmci_ctx_chkpt_buf_info set_info;
 745	u32 cid;
 746	void *cpt_buf;
 747	int retval;
 748
 749	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 750		vmci_ioctl_err("only valid for contexts\n");
 751		return -EINVAL;
 752	}
 753
 754	if (copy_from_user(&set_info, uptr, sizeof(set_info)))
 755		return -EFAULT;
 756
 757	cpt_buf = memdup_user((void __user *)(uintptr_t)set_info.cpt_buf,
 758				set_info.buf_size);
 759	if (IS_ERR(cpt_buf))
 760		return PTR_ERR(cpt_buf);
 761
 762	cid = vmci_ctx_get_id(vmci_host_dev->context);
 763	set_info.result = vmci_ctx_set_chkpt_state(cid, set_info.cpt_type,
 764						   set_info.buf_size, cpt_buf);
 765
 766	retval = copy_to_user(uptr, &set_info, sizeof(set_info)) ? -EFAULT : 0;
 767
 768	kfree(cpt_buf);
 769	return retval;
 770}
 771
 772static int vmci_host_do_get_context_id(struct vmci_host_dev *vmci_host_dev,
 773				       const char *ioctl_name,
 774				       void __user *uptr)
 775{
 776	u32 __user *u32ptr = uptr;
 777
 778	return put_user(VMCI_HOST_CONTEXT_ID, u32ptr) ? -EFAULT : 0;
 779}
 780
 781static int vmci_host_do_set_notify(struct vmci_host_dev *vmci_host_dev,
 782				   const char *ioctl_name,
 783				   void __user *uptr)
 784{
 785	struct vmci_set_notify_info notify_info;
 786
 787	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 788		vmci_ioctl_err("only valid for contexts\n");
 789		return -EINVAL;
 790	}
 791
 792	if (copy_from_user(&notify_info, uptr, sizeof(notify_info)))
 793		return -EFAULT;
 794
 795	if (notify_info.notify_uva) {
 796		notify_info.result =
 797			vmci_host_setup_notify(vmci_host_dev->context,
 798					       notify_info.notify_uva);
 799	} else {
 800		vmci_ctx_unset_notify(vmci_host_dev->context);
 801		notify_info.result = VMCI_SUCCESS;
 802	}
 803
 804	return copy_to_user(uptr, &notify_info, sizeof(notify_info)) ?
 805		-EFAULT : 0;
 806}
 807
 808static int vmci_host_do_notify_resource(struct vmci_host_dev *vmci_host_dev,
 809					const char *ioctl_name,
 810					void __user *uptr)
 811{
 812	struct vmci_dbell_notify_resource_info info;
 813	u32 cid;
 814
 815	if (vmci_host_dev->user_version < VMCI_VERSION_NOTIFY) {
 816		vmci_ioctl_err("invalid for current VMX versions\n");
 817		return -EINVAL;
 818	}
 819
 820	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 821		vmci_ioctl_err("only valid for contexts\n");
 822		return -EINVAL;
 823	}
 824
 825	if (copy_from_user(&info, uptr, sizeof(info)))
 826		return -EFAULT;
 827
 828	cid = vmci_ctx_get_id(vmci_host_dev->context);
 829
 830	switch (info.action) {
 831	case VMCI_NOTIFY_RESOURCE_ACTION_NOTIFY:
 832		if (info.resource == VMCI_NOTIFY_RESOURCE_DOOR_BELL) {
 833			u32 flags = VMCI_NO_PRIVILEGE_FLAGS;
 834			info.result = vmci_ctx_notify_dbell(cid, info.handle,
 835							    flags);
 836		} else {
 837			info.result = VMCI_ERROR_UNAVAILABLE;
 838		}
 839		break;
 840
 841	case VMCI_NOTIFY_RESOURCE_ACTION_CREATE:
 842		info.result = vmci_ctx_dbell_create(cid, info.handle);
 843		break;
 844
 845	case VMCI_NOTIFY_RESOURCE_ACTION_DESTROY:
 846		info.result = vmci_ctx_dbell_destroy(cid, info.handle);
 847		break;
 848
 849	default:
 850		vmci_ioctl_err("got unknown action (action=%d)\n",
 851			       info.action);
 852		info.result = VMCI_ERROR_INVALID_ARGS;
 853	}
 854
 855	return copy_to_user(uptr, &info, sizeof(info)) ? -EFAULT : 0;
 856}
 857
 858static int vmci_host_do_recv_notifications(struct vmci_host_dev *vmci_host_dev,
 859					   const char *ioctl_name,
 860					   void __user *uptr)
 861{
 862	struct vmci_ctx_notify_recv_info info;
 863	struct vmci_handle_arr *db_handle_array;
 864	struct vmci_handle_arr *qp_handle_array;
 865	void __user *ubuf;
 866	u32 cid;
 867	int retval = 0;
 868
 869	if (vmci_host_dev->ct_type != VMCIOBJ_CONTEXT) {
 870		vmci_ioctl_err("only valid for contexts\n");
 871		return -EINVAL;
 872	}
 873
 874	if (vmci_host_dev->user_version < VMCI_VERSION_NOTIFY) {
 875		vmci_ioctl_err("not supported for the current vmx version\n");
 876		return -EINVAL;
 877	}
 878
 879	if (copy_from_user(&info, uptr, sizeof(info)))
 880		return -EFAULT;
 881
 882	if ((info.db_handle_buf_size && !info.db_handle_buf_uva) ||
 883	    (info.qp_handle_buf_size && !info.qp_handle_buf_uva)) {
 884		return -EINVAL;
 885	}
 886
 887	cid = vmci_ctx_get_id(vmci_host_dev->context);
 888
 889	info.result = vmci_ctx_rcv_notifications_get(cid,
 890				&db_handle_array, &qp_handle_array);
 891	if (info.result != VMCI_SUCCESS)
 892		return copy_to_user(uptr, &info, sizeof(info)) ? -EFAULT : 0;
 893
 894	ubuf = (void __user *)(uintptr_t)info.db_handle_buf_uva;
 895	info.result = drv_cp_harray_to_user(ubuf, &info.db_handle_buf_size,
 896					    db_handle_array, &retval);
 897	if (info.result == VMCI_SUCCESS && !retval) {
 898		ubuf = (void __user *)(uintptr_t)info.qp_handle_buf_uva;
 899		info.result = drv_cp_harray_to_user(ubuf,
 900						    &info.qp_handle_buf_size,
 901						    qp_handle_array, &retval);
 902	}
 903
 904	if (!retval && copy_to_user(uptr, &info, sizeof(info)))
 905		retval = -EFAULT;
 906
 907	vmci_ctx_rcv_notifications_release(cid,
 908				db_handle_array, qp_handle_array,
 909				info.result == VMCI_SUCCESS && !retval);
 910
 911	return retval;
 912}
 913
 914static long vmci_host_unlocked_ioctl(struct file *filp,
 915				     unsigned int iocmd, unsigned long ioarg)
 916{
 917#define VMCI_DO_IOCTL(ioctl_name, ioctl_fn) do {			\
 918		char *name = "IOCTL_VMCI_" # ioctl_name;		\
 919		return vmci_host_do_ ## ioctl_fn(			\
 920			vmci_host_dev, name, uptr);			\
 921	} while (0)
 922
 923	struct vmci_host_dev *vmci_host_dev = filp->private_data;
 924	void __user *uptr = (void __user *)ioarg;
 925
 926	switch (iocmd) {
 927	case IOCTL_VMCI_INIT_CONTEXT:
 928		VMCI_DO_IOCTL(INIT_CONTEXT, init_context);
 929	case IOCTL_VMCI_DATAGRAM_SEND:
 930		VMCI_DO_IOCTL(DATAGRAM_SEND, send_datagram);
 931	case IOCTL_VMCI_DATAGRAM_RECEIVE:
 932		VMCI_DO_IOCTL(DATAGRAM_RECEIVE, receive_datagram);
 933	case IOCTL_VMCI_QUEUEPAIR_ALLOC:
 934		VMCI_DO_IOCTL(QUEUEPAIR_ALLOC, alloc_queuepair);
 935	case IOCTL_VMCI_QUEUEPAIR_SETVA:
 936		VMCI_DO_IOCTL(QUEUEPAIR_SETVA, queuepair_setva);
 937	case IOCTL_VMCI_QUEUEPAIR_SETPAGEFILE:
 938		VMCI_DO_IOCTL(QUEUEPAIR_SETPAGEFILE, queuepair_setpf);
 939	case IOCTL_VMCI_QUEUEPAIR_DETACH:
 940		VMCI_DO_IOCTL(QUEUEPAIR_DETACH, qp_detach);
 941	case IOCTL_VMCI_CTX_ADD_NOTIFICATION:
 942		VMCI_DO_IOCTL(CTX_ADD_NOTIFICATION, ctx_add_notify);
 943	case IOCTL_VMCI_CTX_REMOVE_NOTIFICATION:
 944		VMCI_DO_IOCTL(CTX_REMOVE_NOTIFICATION, ctx_remove_notify);
 945	case IOCTL_VMCI_CTX_GET_CPT_STATE:
 946		VMCI_DO_IOCTL(CTX_GET_CPT_STATE, ctx_get_cpt_state);
 947	case IOCTL_VMCI_CTX_SET_CPT_STATE:
 948		VMCI_DO_IOCTL(CTX_SET_CPT_STATE, ctx_set_cpt_state);
 949	case IOCTL_VMCI_GET_CONTEXT_ID:
 950		VMCI_DO_IOCTL(GET_CONTEXT_ID, get_context_id);
 951	case IOCTL_VMCI_SET_NOTIFY:
 952		VMCI_DO_IOCTL(SET_NOTIFY, set_notify);
 953	case IOCTL_VMCI_NOTIFY_RESOURCE:
 954		VMCI_DO_IOCTL(NOTIFY_RESOURCE, notify_resource);
 955	case IOCTL_VMCI_NOTIFICATIONS_RECEIVE:
 956		VMCI_DO_IOCTL(NOTIFICATIONS_RECEIVE, recv_notifications);
 957
 958	case IOCTL_VMCI_VERSION:
 959	case IOCTL_VMCI_VERSION2:
 960		return vmci_host_get_version(vmci_host_dev, iocmd, uptr);
 961
 962	default:
 963		pr_devel("%s: Unknown ioctl (iocmd=%d)\n", __func__, iocmd);
 964		return -EINVAL;
 965	}
 966
 967#undef VMCI_DO_IOCTL
 968}
 969
 970static const struct file_operations vmuser_fops = {
 971	.owner		= THIS_MODULE,
 972	.open		= vmci_host_open,
 973	.release	= vmci_host_close,
 974	.poll		= vmci_host_poll,
 975	.unlocked_ioctl	= vmci_host_unlocked_ioctl,
 976	.compat_ioctl	= compat_ptr_ioctl,
 977};
 978
 979static struct miscdevice vmci_host_miscdev = {
 980	 .name = "vmci",
 981	 .minor = MISC_DYNAMIC_MINOR,
 982	 .fops = &vmuser_fops,
 983};
 984
 985int __init vmci_host_init(void)
 986{
 987	int error;
 988
 989	host_context = vmci_ctx_create(VMCI_HOST_CONTEXT_ID,
 990					VMCI_DEFAULT_PROC_PRIVILEGE_FLAGS,
 991					-1, VMCI_VERSION, NULL);
 992	if (IS_ERR(host_context)) {
 993		error = PTR_ERR(host_context);
 994		pr_warn("Failed to initialize VMCIContext (error%d)\n",
 995			error);
 996		return error;
 997	}
 998
 999	error = misc_register(&vmci_host_miscdev);
1000	if (error) {
1001		pr_warn("Module registration error (name=%s, major=%d, minor=%d, err=%d)\n",
1002			vmci_host_miscdev.name,
1003			MISC_MAJOR, vmci_host_miscdev.minor,
1004			error);
1005		pr_warn("Unable to initialize host personality\n");
1006		vmci_ctx_destroy(host_context);
1007		return error;
1008	}
1009
1010	pr_info("VMCI host device registered (name=%s, major=%d, minor=%d)\n",
1011		vmci_host_miscdev.name, MISC_MAJOR, vmci_host_miscdev.minor);
1012
1013	vmci_host_device_initialized = true;
1014	return 0;
1015}
1016
1017void __exit vmci_host_exit(void)
1018{
1019	vmci_host_device_initialized = false;
1020
1021	misc_deregister(&vmci_host_miscdev);
1022	vmci_ctx_destroy(host_context);
1023	vmci_qp_broker_exit();
1024
1025	pr_debug("VMCI host driver module unloaded\n");
1026}