reactos/drivers/filesystems/udfs/shutdown.cpp

238 lines
8.6 KiB
C++

////////////////////////////////////////////////////////////////////
// Copyright (C) Alexander Telyatnikov, Ivan Keliukh, Yegor Anchishkin, SKIF Software, 1999-2013. Kiev, Ukraine
// All rights reserved
// This file was released under the GPLv2 on June 2015.
////////////////////////////////////////////////////////////////////
/*************************************************************************
*
* File: Shutdown.cpp
*
* Module: UDF File System Driver (Kernel mode execution only)
*
* Description:
* Contains code to handle the "shutdown notification" dispatch entry point.
*
*************************************************************************/
#include "udffs.h"
// define the file specific bug-check id
#define UDF_BUG_CHECK_ID UDF_FILE_SHUTDOWN
/*************************************************************************
*
* Function: UDFShutdown()
*
* Description:
* All disk-based FSDs can expect to receive this shutdown notification
* request whenever the system is about to be halted gracefully. If you
* design and implement a network redirector, you must register explicitly
* for shutdown notification by invoking the IoRegisterShutdownNotification()
* routine from your driver entry.
*
* Note that drivers that register to receive shutdown notification get
* invoked BEFORE disk-based FSDs are told about the shutdown notification.
*
* Expected Interrupt Level (for execution) :
*
* IRQL_PASSIVE_LEVEL
*
* Return Value: Irrelevant.
*
*************************************************************************/
NTSTATUS
NTAPI
UDFShutdown(
PDEVICE_OBJECT DeviceObject, // the logical volume device object
PIRP Irp // I/O Request Packet
)
{
NTSTATUS RC = STATUS_SUCCESS;
PtrUDFIrpContext PtrIrpContext = NULL;
BOOLEAN AreWeTopLevel = FALSE;
UDFPrint(("UDFShutDown\n"));
// BrutePoint();
FsRtlEnterFileSystem();
ASSERT(DeviceObject);
ASSERT(Irp);
// set the top level context
AreWeTopLevel = UDFIsIrpTopLevel(Irp);
//ASSERT(!UDFIsFSDevObj(DeviceObject));
_SEH2_TRY {
// get an IRP context structure and issue the request
PtrIrpContext = UDFAllocateIrpContext(Irp, DeviceObject);
if(PtrIrpContext) {
RC = UDFCommonShutdown(PtrIrpContext, Irp);
} else {
RC = STATUS_INSUFFICIENT_RESOURCES;
Irp->IoStatus.Status = RC;
Irp->IoStatus.Information = 0;
// complete the IRP
IoCompleteRequest(Irp, IO_DISK_INCREMENT);
}
} _SEH2_EXCEPT(UDFExceptionFilter(PtrIrpContext, _SEH2_GetExceptionInformation())) {
RC = UDFExceptionHandler(PtrIrpContext, Irp);
UDFLogEvent(UDF_ERROR_INTERNAL_ERROR, RC);
} _SEH2_END;
if (AreWeTopLevel) {
IoSetTopLevelIrp(NULL);
}
FsRtlExitFileSystem();
return(RC);
} // end UDFShutdown()
/*************************************************************************
*
* Function: UDFCommonShutdown()
*
* Description:
* The actual work is performed here. Basically, all we do here is
* internally invoke a flush on all mounted logical volumes. This, in
* tuen, will result in all open file streams being flushed to disk.
*
* Expected Interrupt Level (for execution) :
*
* IRQL_PASSIVE_LEVEL
*
* Return Value: Irrelevant
*
*************************************************************************/
NTSTATUS
UDFCommonShutdown(
PtrUDFIrpContext PtrIrpContext,
PIRP Irp
)
{
NTSTATUS RC = STATUS_SUCCESS;
PIO_STACK_LOCATION IrpSp = NULL;
PVCB Vcb;
PLIST_ENTRY Link;
PPREVENT_MEDIA_REMOVAL_USER_IN Buf = NULL;
LARGE_INTEGER delay;
UDFPrint(("UDFCommonShutdown\n"));
_SEH2_TRY {
// First, get a pointer to the current I/O stack location
IrpSp = IoGetCurrentIrpStackLocation(Irp);
ASSERT(IrpSp);
Buf = (PPREVENT_MEDIA_REMOVAL_USER_IN)MyAllocatePool__(NonPagedPool, sizeof(PREVENT_MEDIA_REMOVAL_USER_IN));
if(!Buf)
try_return(RC = STATUS_INSUFFICIENT_RESOURCES);
// (a) Block all new "mount volume" requests by acquiring an appropriate
// global resource/lock.
// (b) Go through your linked list of mounted logical volumes and for
// each such volume, do the following:
// (i) acquire the volume resource exclusively
// (ii) invoke UDFFlushLogicalVolume() (internally) to flush the
// open data streams belonging to the volume from the system
// cache
// (iii) Invoke the physical/virtual/logical target device object
// on which the volume is mounted and inform this device
// about the shutdown request (Use IoBuildSynchronouFsdRequest()
// to create an IRP with MajorFunction = IRP_MJ_SHUTDOWN that you
// will then issue to the target device object).
// (iv) Wait for the completion of the shutdown processing by the target
// device object
// (v) Release the VCB resource we will have acquired in (i) above.
// Acquire GlobalDataResource
UDFAcquireResourceExclusive(&(UDFGlobalData.GlobalDataResource), TRUE);
// Walk through all of the Vcb's attached to the global data.
Link = UDFGlobalData.VCBQueue.Flink;
while (Link != &(UDFGlobalData.VCBQueue)) {
// Get 'next' Vcb
Vcb = CONTAINING_RECORD( Link, VCB, NextVCB );
// Move to the next link now since the current Vcb may be deleted.
Link = Link->Flink;
ASSERT(Link != Link->Flink);
if(!(Vcb->VCBFlags & UDF_VCB_FLAGS_SHUTDOWN)) {
#ifdef UDF_DELAYED_CLOSE
UDFAcquireResourceExclusive(&(Vcb->VCBResource), TRUE);
UDFPrint((" UDFCommonShutdown: set UDF_VCB_FLAGS_NO_DELAYED_CLOSE\n"));
Vcb->VCBFlags |= UDF_VCB_FLAGS_NO_DELAYED_CLOSE;
UDFReleaseResource(&(Vcb->VCBResource));
#endif //UDF_DELAYED_CLOSE
// Note: UDFCloseAllDelayed() doesn't acquire DelayedCloseResource if
// GlobalDataResource is already acquired. Thus for now we should
// release GlobalDataResource and re-acquire it later.
UDFReleaseResource( &(UDFGlobalData.GlobalDataResource) );
if(Vcb->RootDirFCB && Vcb->RootDirFCB->FileInfo) {
UDFPrint((" UDFCommonShutdown: UDFCloseAllSystemDelayedInDir\n"));
RC = UDFCloseAllSystemDelayedInDir(Vcb, Vcb->RootDirFCB->FileInfo);
ASSERT(OS_SUCCESS(RC));
}
#ifdef UDF_DELAYED_CLOSE
UDFCloseAllDelayed(Vcb);
// UDFReleaseResource(&(UDFGlobalData.DelayedCloseResource));
#endif //UDF_DELAYED_CLOSE
// re-acquire GlobalDataResource
UDFAcquireResourceExclusive(&(UDFGlobalData.GlobalDataResource), TRUE);
// disable Eject Waiter
UDFStopEjectWaiter(Vcb);
// Acquire Vcb resource
UDFAcquireResourceExclusive(&(Vcb->VCBResource), TRUE);
ASSERT(!Vcb->OverflowQueueCount);
if(!(Vcb->VCBFlags & UDF_VCB_FLAGS_SHUTDOWN)) {
UDFDoDismountSequence(Vcb, Buf, FALSE);
if(Vcb->VCBFlags & UDF_VCB_FLAGS_REMOVABLE_MEDIA) {
// let drive flush all data before reset
delay.QuadPart = -10000000; // 1 sec
KeDelayExecutionThread(KernelMode, FALSE, &delay);
}
Vcb->VCBFlags |= (UDF_VCB_FLAGS_SHUTDOWN |
UDF_VCB_FLAGS_VOLUME_READ_ONLY);
}
UDFReleaseResource(&(Vcb->VCBResource));
}
}
// Once we have processed all the mounted logical volumes, we can release
// all acquired global resources and leave (in peace :-)
UDFReleaseResource( &(UDFGlobalData.GlobalDataResource) );
RC = STATUS_SUCCESS;
try_exit: NOTHING;
} _SEH2_FINALLY {
if(Buf) MyFreePool__(Buf);
if(!_SEH2_AbnormalTermination()) {
Irp->IoStatus.Status = RC;
Irp->IoStatus.Information = 0;
// Free up the Irp Context
UDFReleaseIrpContext(PtrIrpContext);
// complete the IRP
IoCompleteRequest(Irp, IO_DISK_INCREMENT);
}
} _SEH2_END; // end of "__finally" processing
return(RC);
} // end UDFCommonShutdown()