windows-nt/Source/XPSP1/NT/base/ntos/io/iomgr/lock.c
2020-09-26 16:20:57 +08:00

854 lines
27 KiB
C
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/*++
Copyright (c) 1989-1993 Microsoft Corporation
Module Name:
lock.c
Abstract:
This module contains the code to implement the NtLockFile and the
NtUnlockFile system services for the NT I/O system.
Author:
Darryl E. Havens (darrylh) 29-Nov-1989
Environment:
Kernel mode only
Revision History:
--*/
#include "iomgr.h"
#ifdef ALLOC_PRAGMA
#pragma alloc_text(PAGE, NtLockFile)
#pragma alloc_text(PAGE, NtUnlockFile)
#endif
NTSTATUS
NtLockFile(
IN HANDLE FileHandle,
IN HANDLE Event OPTIONAL,
IN PIO_APC_ROUTINE ApcRoutine OPTIONAL,
IN PVOID ApcContext OPTIONAL,
OUT PIO_STATUS_BLOCK IoStatusBlock,
IN PLARGE_INTEGER ByteOffset,
IN PLARGE_INTEGER Length,
IN ULONG Key,
IN BOOLEAN FailImmediately,
IN BOOLEAN ExclusiveLock
)
/*++
Routine Description:
This service locks a specified range of bytes on the file specified by
the FileHandle parameter. The lock may either be an exclusive lock or
a shared lock. Furthermore, the caller has the option of specifying
whether or not the service should return immediately if the lock cannot
be acquired without waiting.
Arguments:
FileHandle - Supplies a handle to an open file.
Event - Supplies an optional event to be set to the Signaled state when
the operation is complete.
ApcRoutine - Supplies an optional APC routine to be executed when the
operation is complete.
ApcContext - Supplies a context parameter to be passed to the ApcRoutine,
if an ApcRoutine was specified.
IoStatusBlock - Address of the caller's I/O status block.
ByteOffset - Specifies the starting byte offset of the range to lock.
Length - Specifies the length of the byte range to be locked.
Key - Specifies the key to be associated with the lock.
FailImmediately - Specifies that if the lock cannot immediately be
acquired that the service should return to the caller.
ExclusiveLock - Specifies, if TRUE, that the lock should be an exclusive
lock; otherwise the lock is a shared lock.
Return Value:
The status returned is success if the operation was properly queued to
the I/O system. Once the operation completes, the status can be
determined by examining the Status field of the I/O status block.
--*/
{
PIRP irp;
NTSTATUS status;
PFILE_OBJECT fileObject;
PDEVICE_OBJECT deviceObject;
PFAST_IO_DISPATCH fastIoDispatch;
PKEVENT eventObject = (PKEVENT) NULL;
KPROCESSOR_MODE requestorMode;
PIO_STACK_LOCATION irpSp;
LARGE_INTEGER fileOffset;
LARGE_INTEGER length;
ACCESS_MASK grantedAccess;
OBJECT_HANDLE_INFORMATION handleInformation;
BOOLEAN synchronousIo;
PETHREAD CurrentThread;
PAGED_CODE();
//
// Get the previous mode; i.e., the mode of the caller.
//
CurrentThread = PsGetCurrentThread ();
requestorMode = KeGetPreviousModeByThread(&CurrentThread->Tcb);
//
// Reference the file object so the target device can be found and the
// access rights mask can be used in the following checks for callers
// in user mode. Note that if the handle does not refer to a file
// object, then it will fail.
//
status = ObReferenceObjectByHandle( FileHandle,
0L,
IoFileObjectType,
requestorMode,
(PVOID *) &fileObject,
&handleInformation);
if (!NT_SUCCESS( status )) {
return status;
}
grantedAccess = handleInformation.GrantedAccess;
if (requestorMode != KernelMode) {
//
// The caller's access mode is not kernel so probe each of the arguments
// and capture them as necessary. If any failures occur, the condition
// handler will be invoked to handle them. It will simply cleanup and
// return an access violation status code back to the system service
// dispatcher.
//
//
// Check to ensure that the caller has either READ or WRITE access to
// the file. If not, cleanup and return an error.
//
if (!SeComputeGrantedAccesses( grantedAccess, FILE_READ_DATA | FILE_WRITE_DATA )) {
ObDereferenceObject( fileObject );
return STATUS_ACCESS_DENIED;
}
try {
//
// The IoStatusBlock parameter must be writeable by the caller.
//
ProbeForWriteIoStatusEx( IoStatusBlock , ApcRoutine);
//
// The ByteOffset parameter must be readable by the caller. Probe
// and capture it.
//
ProbeForReadSmallStructure( ByteOffset,
sizeof( LARGE_INTEGER ),
sizeof( ULONG ) );
fileOffset = *ByteOffset;
//
// Likewise, the Length parameter must also be readable by the
// caller. Probe and capture it as well.
//
ProbeForReadSmallStructure( Length,
sizeof( LARGE_INTEGER ),
sizeof( ULONG ) );
length = *Length;
//
// If this file has an I/O completion port associated w/it, then
// ensure that the caller did not supply an APC routine, as the
// two are mutually exclusive methods for I/O completion
// notification.
//
if (fileObject->CompletionContext && IopApcRoutinePresent( ApcRoutine )) {
ObDereferenceObject( fileObject );
return STATUS_INVALID_PARAMETER;
}
} except(EXCEPTION_EXECUTE_HANDLER) {
//
// An exception was incurred attempting to probe the caller's
// parameters. Dereference the file object and return an
// appropriate error status code.
//
ObDereferenceObject( fileObject );
return GetExceptionCode();
}
} else {
//
// The caller's mode was kernel. Get the ByteOffset and Length
// parameter 's to the expected locations.
//
fileOffset = *ByteOffset;
length = *Length;
}
//
// Get the address of the event object and set the event to the Not-
// Signaled state, if an event was specified. Note here, too, that if
// the handle does not refer to an event, or if the event cannot be
// written, then the reference will fail. Since certain legacy
// applications rely on an old bug in Win32's LockFileEx, we must
// tolerate bad event handles.
//
if (ARGUMENT_PRESENT( Event )) {
status = ObReferenceObjectByHandle( Event,
EVENT_MODIFY_STATE,
ExEventObjectType,
requestorMode,
(PVOID *) &eventObject,
NULL );
if (!NT_SUCCESS( status )) {
ASSERT( !eventObject );
} else {
KeClearEvent( eventObject );
}
}
//
// Get the address of the target device object and the fast Io dispatch
// structure.
//
deviceObject = IoGetRelatedDeviceObject( fileObject );
fastIoDispatch = deviceObject->DriverObject->FastIoDispatch;
//
// Turbo lock support. If the fast Io Dispatch specifies a fast lock
// routine then we'll first try and calling it with the specified lock
// parameters.
//
if (fastIoDispatch && fastIoDispatch->FastIoLock) {
IO_STATUS_BLOCK localIoStatus;
if (fastIoDispatch->FastIoLock( fileObject,
&fileOffset,
&length,
PsGetCurrentProcessByThread(CurrentThread),
Key,
FailImmediately,
ExclusiveLock,
&localIoStatus,
deviceObject )) {
//
// Carefully return the I/O status.
//
try {
#if defined(_WIN64)
//
// If this is a32-bit thread, and the IO request is
// asynchronous, then the IOSB is 32-bit. Wow64 always sends
// the 32-bit IOSB when the I/O is asynchronous.
//
if (IopIsIosb32(ApcRoutine)) {
PIO_STATUS_BLOCK32 UserIosb32 = (PIO_STATUS_BLOCK32)IoStatusBlock;
UserIosb32->Information = (ULONG)localIoStatus.Information;
UserIosb32->Status = (NTSTATUS)localIoStatus.Status;
} else {
*IoStatusBlock = localIoStatus;
}
#else
*IoStatusBlock = localIoStatus;
#endif
} except( EXCEPTION_EXECUTE_HANDLER ) {
localIoStatus.Status = GetExceptionCode();
localIoStatus.Information = 0;
}
//
// If a valid event was specified, set it.
//
if (eventObject) {
KeSetEvent( eventObject, 0, FALSE );
ObDereferenceObject( eventObject );
}
//
// Note that the file object event need not be set to the
// Signaled state, as it is already set.
//
//
// If this file object has a completion port associated with it
// and this request has a non-NULL APC context then a completion
// message needs to be queued.
//
if (fileObject->CompletionContext && ARGUMENT_PRESENT( ApcContext )) {
if (!NT_SUCCESS(IoSetIoCompletion( fileObject->CompletionContext->Port,
fileObject->CompletionContext->Key,
ApcContext,
localIoStatus.Status,
localIoStatus.Information,
TRUE ))) {
localIoStatus.Status = STATUS_INSUFFICIENT_RESOURCES;
}
}
//
// Cleanup and return.
//
fileObject->LockOperation = TRUE;
ObDereferenceObject( fileObject );
return localIoStatus.Status;
}
}
//
// Make a special check here to determine whether this is a synchronous
// I/O operation. If it is, then wait here until the file is owned by
// the current thread.
//
if (fileObject->Flags & FO_SYNCHRONOUS_IO) {
BOOLEAN interrupted;
if (!IopAcquireFastLock( fileObject )) {
status = IopAcquireFileObjectLock( fileObject,
requestorMode,
(BOOLEAN) ((fileObject->Flags & FO_ALERTABLE_IO) != 0),
&interrupted );
if (interrupted) {
if (eventObject) {
ObDereferenceObject( eventObject );
}
ObDereferenceObject( fileObject );
return status;
}
}
synchronousIo = TRUE;
} else {
synchronousIo = FALSE;
}
//
// Set the file object to the Not-Signaled state and mark it as having had
// a lock operation performed on it.
//
KeClearEvent( &fileObject->Event );
fileObject->LockOperation = TRUE;
//
// Allocate and initialize the I/O Request Packet (IRP) for this operation.
// The allocation is performed with an exception handler in case the
// caller does not have enough quota to allocate the packet.
irp = IoAllocateIrp( deviceObject->StackSize, TRUE );
if (!irp) {
//
// An IRP could not be allocated. Cleanup and return an appropriate
// error status code.
//
IopAllocateIrpCleanup( fileObject, eventObject );
return STATUS_INSUFFICIENT_RESOURCES;
}
irp->Tail.Overlay.OriginalFileObject = fileObject;
irp->Tail.Overlay.Thread = CurrentThread;
irp->RequestorMode = requestorMode;
//
// Fill in the service independent parameters in the IRP.
//
irp->UserEvent = eventObject;
irp->UserIosb = IoStatusBlock;
irp->Overlay.AsynchronousParameters.UserApcRoutine = ApcRoutine;
irp->Overlay.AsynchronousParameters.UserApcContext = ApcContext;
//
// Get a pointer to the stack location for the first driver. This will be
// used to pass the original function codes and parameters.
//
irpSp = IoGetNextIrpStackLocation( irp );
irpSp->MajorFunction = IRP_MJ_LOCK_CONTROL;
irpSp->MinorFunction = IRP_MN_LOCK;
irpSp->FileObject = fileObject;
//
// Copy the caller's parameters to the service-specific portion of the
// IRP.
//
irpSp->Flags = 0;
if (FailImmediately) {
irpSp->Flags = SL_FAIL_IMMEDIATELY;
}
if (ExclusiveLock) {
irpSp->Flags |= SL_EXCLUSIVE_LOCK;
}
irpSp->Parameters.LockControl.Key = Key;
irpSp->Parameters.LockControl.ByteOffset = fileOffset;
try {
PLARGE_INTEGER lengthBuffer;
//
// Attempt to allocate an intermediary buffer to hold the length of
// this lock operation. If it fails, either because there is no
// more quota, or because there are no more resources, then the
// exception handler will be invoked to cleanup and exit.
//
lengthBuffer = ExAllocatePoolWithQuota( NonPagedPool,
sizeof( LARGE_INTEGER ) );
*lengthBuffer = length;
irp->Tail.Overlay.AuxiliaryBuffer = (PCHAR) lengthBuffer;
irpSp->Parameters.LockControl.Length = lengthBuffer;
} except(EXCEPTION_EXECUTE_HANDLER) {
//
// An exception was incurred. Simply clean everything up and
// return an appropriate error status code.
//
IopExceptionCleanup( fileObject,
irp,
eventObject,
(PKEVENT) NULL );
return GetExceptionCode();
}
//
// Queue the packet, call the driver, and synchronize appopriately with
// I/O completion.
//
return IopSynchronousServiceTail( deviceObject,
irp,
fileObject,
FALSE,
requestorMode,
synchronousIo,
OtherTransfer );
}
NTSTATUS
NtUnlockFile(
IN HANDLE FileHandle,
OUT PIO_STATUS_BLOCK IoStatusBlock,
IN PLARGE_INTEGER ByteOffset,
IN PLARGE_INTEGER Length,
IN ULONG Key
)
/*++
Routine Description:
This service releases the lock associated with the specified byte range
for the file specified by the FileHandle parameter.
Arguments:
FileHandle - Supplies a handle to an open file.
IoStatusBlock - Address of the caller's I/O status block.
ByteOffset - Specifies the byte offset of the range to unlock.
Length - Specifies the length of the byte range to unlock.
Key - Specifies the key associated with the locked range.
Return Value:
The status returned is the final completion status of the operation.
--*/
{
PIRP irp;
NTSTATUS status;
PFILE_OBJECT fileObject;
PDEVICE_OBJECT deviceObject;
PFAST_IO_DISPATCH fastIoDispatch;
PKEVENT event;
KPROCESSOR_MODE requestorMode;
PIO_STACK_LOCATION irpSp;
IO_STATUS_BLOCK localIoStatus;
LARGE_INTEGER fileOffset;
LARGE_INTEGER length;
ACCESS_MASK grantedAccess;
OBJECT_HANDLE_INFORMATION handleInformation;
BOOLEAN synchronousIo;
PETHREAD CurrentThread;
PAGED_CODE();
//
// Get the previous mode; i.e., the mode of the caller.
//
CurrentThread = PsGetCurrentThread ();
requestorMode = KeGetPreviousModeByThread(&CurrentThread->Tcb);
//
// Reference the file object so the target device can be found and the
// access rights mask can be used in the following checks for callers
// in user mode. Note that if the handle does not refer to a file
// object, then it will fail.
//
status = ObReferenceObjectByHandle( FileHandle,
0L,
IoFileObjectType,
requestorMode,
(PVOID *) &fileObject,
&handleInformation);
if (!NT_SUCCESS( status )) {
return status;
}
grantedAccess = handleInformation.GrantedAccess;
//
// Check to see if the requestor mode was user. If so, perform a bunch
// of extra checks.
//
if (requestorMode != KernelMode) {
//
// The caller's access mode is not kernel so probe each of the arguments
// and capture them as necessary. If any failures occur, the condition
// handler will be invoked to handle them. It will simply cleanup and
// return an access violation status code back to the system service
// dispatcher.
//
//
// Check to ensure that the caller has either READ or WRITE access
// to the file. If not, cleanup and return an error.
//
if (!SeComputeGrantedAccesses( grantedAccess, FILE_READ_DATA | FILE_WRITE_DATA )) {
ObDereferenceObject( fileObject );
return STATUS_ACCESS_DENIED;
}
try {
//
// The IoStatusBlock parameter must be writeable by the caller.
//
ProbeForWriteIoStatus( IoStatusBlock );
//
// The ByteOffset parameter must be readable by the caller. Probe
// and capture it.
//
ProbeForReadSmallStructure( ByteOffset,
sizeof( LARGE_INTEGER ),
sizeof( ULONG ) );
fileOffset = *ByteOffset;
//
// Likewise, the Length parameter must also be readable by the
// caller. Probe and capture it as well.
//
ProbeForReadSmallStructure( Length,
sizeof( LARGE_INTEGER ),
sizeof( ULONG ) );
length = *Length;
} except(EXCEPTION_EXECUTE_HANDLER) {
//
// An exception was incurred while attempting to probe the
// caller's parameters. Dereference the file object and return
// an appropriate error status code.
//
ObDereferenceObject( fileObject );
return GetExceptionCode();
}
} else {
//
// The caller's mode was kernel. Get the ByteOffset and Length
// parameter 's to the expected locations.
//
fileOffset = *ByteOffset;
length = *Length;
}
//
// Get the address of the target device object. If this file represents
// a device that was opened directly, then simply use the device or its
// attached device(s) directly. Also get the fast I/O dispatch address.
//
if (!(fileObject->Flags & FO_DIRECT_DEVICE_OPEN)) {
deviceObject = IoGetRelatedDeviceObject( fileObject );
} else {
deviceObject = IoGetAttachedDevice( fileObject->DeviceObject );
}
fastIoDispatch = deviceObject->DriverObject->FastIoDispatch;
//
// Turbo lock support. If the fast Io Dispatch specifies a fast lock
// routine then we'll first try and calling it with the specified lock
// parameters.
//
if (fastIoDispatch && fastIoDispatch->FastIoUnlockSingle) {
IO_STATUS_BLOCK localIoStatus;
if (fastIoDispatch->FastIoUnlockSingle( fileObject,
&fileOffset,
&length,
PsGetCurrentProcessByThread(CurrentThread),
Key,
&localIoStatus,
deviceObject )) {
//
// Carefully return the I/O status.
//
try {
*IoStatusBlock = localIoStatus;
} except( EXCEPTION_EXECUTE_HANDLER ) {
localIoStatus.Status = GetExceptionCode();
localIoStatus.Information = 0;
}
//
// Cleanup and return.
//
ObDereferenceObject( fileObject );
return localIoStatus.Status;
}
}
//
// Make a special check here to determine whether this is a synchronous
// I/O operation. If it is, then wait here until the file is owned by
// the current thread. If this is not a (serialized) synchronous I/O
// operation, then allocate and initialize the local event.
//
if (fileObject->Flags & FO_SYNCHRONOUS_IO) {
BOOLEAN interrupted;
if (!IopAcquireFastLock( fileObject )) {
status = IopAcquireFileObjectLock( fileObject,
requestorMode,
(BOOLEAN) ((fileObject->Flags & FO_ALERTABLE_IO) != 0),
&interrupted );
if (interrupted) {
ObDereferenceObject( fileObject );
return status;
}
}
synchronousIo = TRUE;
} else {
//
// This is a synchronous API being invoked for a file that is opened
// for asynchronous I/O. This means that this system service is
// to synchronize the completion of the operation before returning
// to the caller. A local event is used to do this.
//
event = ExAllocatePool( NonPagedPool, sizeof( KEVENT ) );
if (event == NULL) {
ObDereferenceObject( fileObject );
return STATUS_INSUFFICIENT_RESOURCES;
}
KeInitializeEvent( event, SynchronizationEvent, FALSE );
synchronousIo = FALSE;
}
//
// Set the file object to the Not-Signaled state.
//
KeClearEvent( &fileObject->Event );
//
// Allocate and initialize the I/O Request Packet (IRP) for this operation.
// The allocation is performed with an exception handler in case the
// caller does not have enough quota to allocate the packet.
//
irp = IoAllocateIrp( deviceObject->StackSize, TRUE );
if (!irp) {
//
// An IRP could not be allocated. Cleanup and return an appropriate
// error status code.
//
if (!(fileObject->Flags & FO_SYNCHRONOUS_IO)) {
ExFreePool( event );
}
IopAllocateIrpCleanup( fileObject, (PKEVENT) NULL );
return STATUS_INSUFFICIENT_RESOURCES;
}
irp->Tail.Overlay.OriginalFileObject = fileObject;
irp->Tail.Overlay.Thread = CurrentThread;
irp->RequestorMode = requestorMode;
//
// Fill in the service independent parameters in the IRP.
//
if (synchronousIo) {
irp->UserEvent = (PKEVENT) NULL;
irp->UserIosb = IoStatusBlock;
} else {
irp->UserEvent = event;
irp->UserIosb = &localIoStatus;
irp->Flags = IRP_SYNCHRONOUS_API;
}
irp->Overlay.AsynchronousParameters.UserApcRoutine = (PIO_APC_ROUTINE) NULL;
//
// Get a pointer to the stack location for the first driver. This will
// be used to pass the original function codes and parameters.
//
irpSp = IoGetNextIrpStackLocation( irp );
irpSp->MajorFunction = IRP_MJ_LOCK_CONTROL;
irpSp->MinorFunction = IRP_MN_UNLOCK_SINGLE;
irpSp->FileObject = fileObject;
try {
PLARGE_INTEGER lengthBuffer;
//
// Attempt to allocate an intermediary buffer to hold the length of
// this lock operation. If it fails, either because there is no
// more quota, or because there are no more resources, then the
// exception handler will be invoked to cleanup and exit.
//
lengthBuffer = ExAllocatePoolWithQuota( NonPagedPool,
sizeof( LARGE_INTEGER ) );
*lengthBuffer = length;
irp->Tail.Overlay.AuxiliaryBuffer = (PCHAR) lengthBuffer;
irpSp->Parameters.LockControl.Length = lengthBuffer;
} except(EXCEPTION_EXECUTE_HANDLER) {
//
// An exception was incurred. Simply clean everything up and
// return an appropriate error status code.
//
if (!(fileObject->Flags & FO_SYNCHRONOUS_IO)) {
ExFreePool( event );
}
IopExceptionCleanup( fileObject,
irp,
NULL,
(PKEVENT) NULL );
return GetExceptionCode();
}
//
// Copy the caller's parameters to the service-specific portion of the
// IRP.
//
irpSp->Parameters.LockControl.Key = Key;
irpSp->Parameters.LockControl.ByteOffset = fileOffset;
//
// Queue the packet, call the driver, and synchronize appopriately with
// I/O completion.
//
status = IopSynchronousServiceTail( deviceObject,
irp,
fileObject,
FALSE,
requestorMode,
synchronousIo,
OtherTransfer );
//
// If the file for this operation was not opened for synchronous I/O, then
// synchronization of completion of the I/O operation has not yet occurred
// since the allocated event must be used for synchronous APIs on files
// opened for asynchronous I/O. Synchronize the completion of the I/O
// operation now.
//
if (!synchronousIo) {
status = IopSynchronousApiServiceTail( status,
event,
irp,
requestorMode,
&localIoStatus,
IoStatusBlock );
}
return status;
}