windows-nt/Source/XPSP1/NT/base/fs/ntfs/mcbsup.c
2020-09-26 16:20:57 +08:00

2849 lines
72 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) 1991 Microsoft Corporation
Module Name:
McbSup.c
Abstract:
This module implements the Ntfs Mcb package.
Author:
Gary Kimura [GaryKi] 10-Sep-1994
Tom Miller [TomM]
Revision History:
--*/
#include "NtfsProc.h"
#define FIRST_RANGE ((PVOID)1)
#ifndef NTFS_VERIFY_MCB
#define NtfsVerifyNtfsMcb(M) NOTHING;
#define NtfsVerifyUncompressedNtfsMcb(M,S,E) NOTHING;
#endif
//
// Define a tag for general pool allocations from this module
//
#undef MODULE_POOL_TAG
#define MODULE_POOL_TAG ('MFtN')
//
// Local procedure prototypes
//
ULONG
NtfsMcbLookupArrayIndex (
IN PNTFS_MCB Mcb,
IN VCN Vcn
);
VOID
NtfsInsertNewRange (
IN PNTFS_MCB Mcb,
IN LONGLONG StartingVcn,
IN ULONG ArrayIndex,
IN BOOLEAN MakeNewRangeEmpty
);
VOID
NtfsCollapseRanges (
IN PNTFS_MCB Mcb,
IN ULONG StartingArrayIndex,
IN ULONG EndingArrayIndex
);
VOID
NtfsMcbCleanupLruQueue (
IN PVOID Parameter
);
#ifdef NTFS_VERIFY_MCB
VOID
NtfsVerifyNtfsMcb (
IN PNTFS_MCB Mcb
);
VOID
NtfsVerifyUncompressedNtfsMcb (
IN PNTFS_MCB Mcb,
IN LONGLONG StartingVcn,
IN LONGLONG EndingVcn
);
#endif
BOOLEAN
NtfsLockNtfsMcb (
IN PNTFS_MCB Mcb
);
VOID
NtfsUnlockNtfsMcb (
IN PNTFS_MCB Mcb
);
NtfsGrowMcbArray(
IN PNTFS_MCB Mcb
);
//
// Local macros to ASSERT that caller's resource is exclusive or restart is
// underway.
//
#define ASSERT_STREAM_EXCLUSIVE(M) { \
ASSERT( FlagOn( ((PSCB) (M)->FcbHeader)->Vcb->VcbState, VCB_STATE_RESTART_IN_PROGRESS ) || \
ExIsResourceAcquiredExclusiveLite((M)->FcbHeader->Resource )); \
}
//
// Local macros to enqueue and dequeue elements from the lru queue
//
#define NtfsMcbEnqueueLruEntry(M,E) { \
InsertTailList( &NtfsMcbLruQueue, &(E)->LruLinks ); \
NtfsMcbCurrentLevel += 1; \
}
#define NtfsMcbDequeueLruEntry(M,E) { \
if ((E)->LruLinks.Flink != NULL) { \
RemoveEntryList( &(E)->LruLinks ); \
NtfsMcbCurrentLevel -= 1; \
} \
}
//
// Local macro to unload a single array entry
//
#define UnloadEntry(M,I) { \
PNTFS_MCB_ENTRY _Entry; \
_Entry = (M)->NtfsMcbArray[(I)].NtfsMcbEntry; \
(M)->NtfsMcbArray[(I)].NtfsMcbEntry = NULL; \
if (_Entry != NULL) { \
ExAcquireFastMutex( &NtfsMcbFastMutex ); \
NtfsMcbDequeueLruEntry( Mcb, _Entry ); \
ExReleaseFastMutex( &NtfsMcbFastMutex ); \
FsRtlUninitializeLargeMcb( &_Entry->LargeMcb ); \
if ((M)->NtfsMcbArraySize != MCB_ARRAY_PHASE1_SIZE) { \
NtfsFreePool( _Entry ); \
} \
} \
}
VOID
NtfsInitializeNtfsMcb (
IN PNTFS_MCB Mcb,
IN PNTFS_ADVANCED_FCB_HEADER FcbHeader,
IN PNTFS_MCB_INITIAL_STRUCTS McbStructs,
IN POOL_TYPE PoolType
)
/*++
Routine Description:
This routine initializes a new Ntfs Mcb structure.
Arguments:
Mcb - Supplies the Mcb being initialized
FcbHeader - Supplies a pointer to the Fcb header containing
the resource to grab when accessing the Mcb
McbStructs - Initial allocation typically coresident in another
structure to handle initial structures for small and
medium files. This structure should be initially zeroed.
PoolType - Supplies the type of pool to use when
allocating mapping information storage
Return Value:
None.
--*/
{
PNTFS_MCB_ARRAY Array;
RtlZeroMemory( McbStructs, sizeof(NTFS_MCB_INITIAL_STRUCTS) );
//
// Initialize the fcb header field of the mcb
//
Mcb->FcbHeader = FcbHeader;
//
// Initialize the pool type
//
Mcb->PoolType = PoolType;
//
// Now initialize the initial array element
//
Mcb->NtfsMcbArray = Array = &McbStructs->Phase1.SingleMcbArrayEntry;
Mcb->NtfsMcbArraySize = MCB_ARRAY_PHASE1_SIZE;
Mcb->NtfsMcbArraySizeInUse = 1;
Mcb->FastMutex = FcbHeader->FastMutex;
//
// Initialize the first array entry.
//
Array[0].StartingVcn = 0;
Array[0].EndingVcn = -1;
//
// And return to our caller
//
NtfsVerifyNtfsMcb(Mcb);
return;
}
VOID
NtfsUninitializeNtfsMcb (
IN PNTFS_MCB Mcb
)
/*++
Routine Description:
This routine uninitializes an Ntfs Mcb structure.
Arguments:
Mcb - Supplies the Mcb being decommissioned
Return Value:
None.
--*/
{
ULONG i;
PNTFS_MCB_ENTRY Entry;
NtfsVerifyNtfsMcb(Mcb);
//
// Take out the global mutex
//
ExAcquireFastMutex( &NtfsMcbFastMutex );
//
// Deallocate the mcb array if it exists. For every entry in the array
// if the mcb entry is not null then remove the entry from the lru
// queue, uninitialize the large mcb, and free the pool.
//
if (Mcb->NtfsMcbArray != NULL) {
for (i = 0; i < Mcb->NtfsMcbArraySizeInUse; i += 1) {
if ((Entry = Mcb->NtfsMcbArray[i].NtfsMcbEntry) != NULL) {
//
// Remove the entry from the lru queue
//
NtfsMcbDequeueLruEntry( Mcb, Entry );
//
// Now release the entry
//
FsRtlUninitializeLargeMcb( &Entry->LargeMcb );
//
// We can tell from the array count whether this is
// the initial entry and does not need to be deallocated.
//
if (Mcb->NtfsMcbArraySize > MCB_ARRAY_PHASE1_SIZE) {
NtfsFreePool( Entry );
}
}
}
//
// We can tell from the array count whether this is
// the initial array entry(s) and do not need to be deallocated.
//
if (Mcb->NtfsMcbArraySize > MCB_ARRAY_PHASE2_SIZE) {
NtfsFreePool( Mcb->NtfsMcbArray );
}
Mcb->NtfsMcbArray = NULL;
//
// Clear the fast mutex field.
//
Mcb->FastMutex = NULL;
}
ExReleaseFastMutex( &NtfsMcbFastMutex );
//
// And return to our caller
//
return;
}
ULONG
NtfsNumberOfRangesInNtfsMcb (
IN PNTFS_MCB Mcb
)
/*++
Routine Description:
This routine returns the total number of ranges stored in
the mcb
Arguments:
Mcb - Supplies the Mcb being queried
Return Value:
ULONG - The number of ranges mapped by the input mcb
--*/
{
ASSERT_STREAM_EXCLUSIVE(Mcb);
//
// Our answer is the number of ranges in use in the mcb
//
NtfsVerifyNtfsMcb(Mcb);
return Mcb->NtfsMcbArraySizeInUse;
}
BOOLEAN
NtfsNumberOfRunsInRange (
IN PNTFS_MCB Mcb,
IN PVOID RangePtr,
OUT PULONG NumberOfRuns
)
/*++
Routine Description:
This routine returns the total number of runs stored withing a range
Arguments:
Mcb - Supplies the Mcb being queried
RangePtr - Supplies the range to being queried
NumberOrRuns - Returns the number of run in the specified range
but only if the range is loaded
Return Value:
BOOLEAN - TRUE if the range is loaded and then output variable
is valid and FALSE if the range is not loaded.
--*/
{
VCN TempVcn;
LCN TempLcn;
PNTFS_MCB_ENTRY Entry = (PNTFS_MCB_ENTRY)RangePtr;
//
// Null RangePtr means first range
//
if (Entry == FIRST_RANGE) {
Entry = Mcb->NtfsMcbArray[0].NtfsMcbEntry;
//
// If not loaded, return FALSE
//
if (Entry == NULL) {
return FALSE;
}
}
ASSERT_STREAM_EXCLUSIVE(Mcb);
NtfsVerifyNtfsMcb(Mcb);
ASSERT( Mcb == Entry->NtfsMcb );
*NumberOfRuns = FsRtlNumberOfRunsInLargeMcb( &Entry->LargeMcb );
//
// Check if the current entry ends with a hole and increment the run count
// to reflect this. Detect the case where the range has length 0 for a
// file with no allocation. EndingVcn will be less than the starting Vcn
// in this case.
//
if (!FsRtlLookupLastLargeMcbEntry( &Entry->LargeMcb, &TempVcn, &TempLcn )) {
//
// If this is a non-zero length range then add one for the implied hole.
//
if (Entry->NtfsMcbArray->EndingVcn >= Entry->NtfsMcbArray->StartingVcn) {
*NumberOfRuns += 1;
}
//
// There is an entry then check if it reaches the end boundary of the range.
//
} else if (TempVcn != (Entry->NtfsMcbArray->EndingVcn - Entry->NtfsMcbArray->StartingVcn)) {
*NumberOfRuns += 1;
}
return TRUE;
}
BOOLEAN
NtfsLookupLastNtfsMcbEntry (
IN PNTFS_MCB Mcb,
OUT PLONGLONG Vcn,
OUT PLONGLONG Lcn
)
/*++
Routine Description:
This routine returns the last mapping stored in the mcb
Arguments:
Mcb - Supplies the Mcb being queried
Vcn - Receives the Vcn of the last mapping
Lcn - Receives the Lcn corresponding to the Vcn
Return Value:
BOOLEAN - TRUE if the mapping exist and FALSE if no mapping has been
defined or it is unloaded
--*/
{
PNTFS_MCB_ENTRY Entry;
LONGLONG StartingVcn;
ASSERT_STREAM_EXCLUSIVE(Mcb);
NtfsVerifyNtfsMcb(Mcb);
//
// Get the last entry and compute its starting vcn, and make sure
// the entry is valid
//
if ((Entry = Mcb->NtfsMcbArray[Mcb->NtfsMcbArraySizeInUse - 1].NtfsMcbEntry) == NULL) {
return FALSE;
}
StartingVcn = Mcb->NtfsMcbArray[Mcb->NtfsMcbArraySizeInUse - 1].StartingVcn;
//
// Otherwise lookup the last entry and compute the real vcn
//
if (FsRtlLookupLastLargeMcbEntry( &Entry->LargeMcb, Vcn, Lcn )) {
*Vcn += StartingVcn;
} else {
*Vcn = Mcb->NtfsMcbArray[Mcb->NtfsMcbArraySizeInUse - 1].EndingVcn;
*Lcn = UNUSED_LCN;
}
return TRUE;
}
BOOLEAN
NtfsLookupNtfsMcbEntry (
IN PNTFS_MCB Mcb,
IN LONGLONG Vcn,
OUT PLONGLONG Lcn OPTIONAL,
OUT PLONGLONG CountFromLcn OPTIONAL,
OUT PLONGLONG StartingLcn OPTIONAL,
OUT PLONGLONG CountFromStartingLcn OPTIONAL,
OUT PVOID *RangePtr OPTIONAL,
OUT PULONG RunIndex OPTIONAL
)
/*++
Routine Description:
This routine is used to query mapping information
Arguments:
Mcb - Supplies the Mcb being queried
Vcn - Supplies the Vcn being queried
Lcn - Optionally receives the lcn corresponding to the input vcn
CountFromLcn - Optionally receives the number of clusters following
the lcn in the run
StartingLcn - Optionally receives the start of the run containing the
input vcn
CountFromStartingLcn - Optionally receives the number of clusters in
the entire run
RangePtr - Optionally receives the index for the range that we're returning
RunIndex - Optionally receives the index for the run within the range that
we're returning
Return Value:
BOOLEAN - TRUE if the mapping exists and FALSE if it doesn't exist
or if it is unloaded.
--*/
{
ULONG LocalRangeIndex;
PNTFS_MCB_ENTRY Entry;
NtfsAcquireNtfsMcbMutex( Mcb );
NtfsVerifyNtfsMcb(Mcb);
//
// Do a basic bounds check
//
ASSERT( Mcb->NtfsMcbArraySizeInUse > 0 );
//
// Locate the array entry that has the hit for the input vcn, and
// make sure it is valid. Also set the output range index if present
//
LocalRangeIndex = NtfsMcbLookupArrayIndex(Mcb, Vcn);
//
// Now lookup the large mcb entry. The Vcn we pass in is
// biased by the starting vcn. If we miss then we'll just return false
//
if (((Entry = Mcb->NtfsMcbArray[LocalRangeIndex].NtfsMcbEntry) == NULL) ||
(Vcn > Entry->NtfsMcbArray->EndingVcn) ||
(Vcn < Entry->NtfsMcbArray->StartingVcn)) {
ASSERT( (Entry == NULL) || (Vcn > Entry->NtfsMcbArray->EndingVcn) || (Vcn < 0) );
if (ARGUMENT_PRESENT(RangePtr)) {
*RangePtr = (PVOID)Entry;
//
// If this is the first range, always normalize back to the reserved pointer,
// since this is the only range which can move if we split out of our
// initial static allocation!
//
if (LocalRangeIndex == 0) {
*RangePtr = FIRST_RANGE;
}
}
NtfsReleaseNtfsMcbMutex( Mcb );
return FALSE;
}
if (!FsRtlLookupLargeMcbEntry( &Entry->LargeMcb,
Vcn - Mcb->NtfsMcbArray[LocalRangeIndex].StartingVcn,
Lcn,
CountFromLcn,
StartingLcn,
CountFromStartingLcn,
RunIndex )) {
//
// If we go off the end of the Mcb, but are in the range, then we
// return a hole to the end of the range.
//
if (ARGUMENT_PRESENT(Lcn)) {
*Lcn = UNUSED_LCN;
}
if (ARGUMENT_PRESENT(CountFromLcn)) {
*CountFromLcn = Mcb->NtfsMcbArray[LocalRangeIndex].EndingVcn - Vcn + 1;
}
if (ARGUMENT_PRESENT(StartingLcn)) {
*StartingLcn = UNUSED_LCN;
}
if (ARGUMENT_PRESENT(RunIndex)) {
*RunIndex = FsRtlNumberOfRunsInLargeMcb( &Entry->LargeMcb );
}
if (ARGUMENT_PRESENT( CountFromStartingLcn )) {
//
// If there are no runs in the Mcb then specify
// a hole for the full range.
//
*CountFromStartingLcn = Mcb->NtfsMcbArray[LocalRangeIndex].EndingVcn -
Mcb->NtfsMcbArray[LocalRangeIndex].StartingVcn + 1;
if (*RunIndex != 0) {
VCN LastVcn;
LCN LastLcn;
FsRtlLookupLastLargeMcbEntry( &Entry->LargeMcb,
&LastVcn,
&LastLcn );
ASSERT( LastVcn <= *CountFromStartingLcn );
*CountFromStartingLcn -= (LastVcn + 1);
}
}
}
if (ARGUMENT_PRESENT(RangePtr)) {
*RangePtr = (PVOID)Entry;
//
// If this is the first range, always normalize back to the reserved pointer,
// since this is the only range which can move if we split out of our
// initial static allocation!
//
if (LocalRangeIndex == 0) {
*RangePtr = FIRST_RANGE;
}
}
//
// Now move this entry to the tail of the lru queue.
// We need to take out the global mutex to do this.
// Only do this if he is already in the queue - we can
// deadlock if we take a fault in the paging file path.
//
if (Entry->LruLinks.Flink != NULL) {
if (ExTryToAcquireFastMutex( &NtfsMcbFastMutex )) {
NtfsMcbDequeueLruEntry( Mcb, Entry );
NtfsMcbEnqueueLruEntry( Mcb, Entry );
ExReleaseFastMutex( &NtfsMcbFastMutex );
}
}
NtfsReleaseNtfsMcbMutex( Mcb );
return TRUE;
}
BOOLEAN
NtfsGetNextNtfsMcbEntry (
IN PNTFS_MCB Mcb,
IN PVOID *RangePtr,
IN ULONG RunIndex,
OUT PLONGLONG Vcn,
OUT PLONGLONG Lcn,
OUT PLONGLONG Count
)
/*++
Routine Description:
This routine returns the range denoted by the type index values
Arguments:
Mcb - Supplies the Mcb being queried
RangePtr - Supplies the pointer to the range being queried, or NULL for the first one,
returns next range
RunIndex - Supplies the index within then being queried, or MAXULONG for first in next
Vcn - Receives the starting Vcn of the run being returned
Lcn - Receives the starting Lcn of the run being returned or unused
lbn value of -1
Count - Receives the number of clusters within this run
Return Value:
BOOLEAN - TRUE if the two input indices are valid and FALSE if the
the index are not valid or if the range is not loaded
--*/
{
PNTFS_MCB_ENTRY Entry = (PNTFS_MCB_ENTRY)*RangePtr;
BOOLEAN Result = FALSE;
NtfsAcquireNtfsMcbMutex( Mcb );
NtfsVerifyNtfsMcb(Mcb);
try {
//
// Null RangePtr means first range
//
if (Entry == FIRST_RANGE) {
Entry = Mcb->NtfsMcbArray[0].NtfsMcbEntry;
}
//
// If there is no entry 0, get out.
//
if (Entry == NULL) {
try_return(Result = FALSE);
}
//
// RunIndex of MAXULONG means first of next
//
if (RunIndex == MAXULONG) {
//
// If we are already in the last range, get out.
//
if (Entry->NtfsMcbArray == (Mcb->NtfsMcbArray + Mcb->NtfsMcbArraySizeInUse - 1)) {
try_return(Result = FALSE);
}
*RangePtr = Entry = (Entry->NtfsMcbArray + 1)->NtfsMcbEntry;
RunIndex = 0;
}
//
// If there is no next entry, get out.
//
if (Entry == NULL) {
try_return(Result = FALSE);
}
ASSERT( Mcb == Entry->NtfsMcb );
//
// Lookup the large mcb entry. If we get a miss then the we're
// beyond the end of the ntfs mcb and should return false
//
if (!FsRtlGetNextLargeMcbEntry( &Entry->LargeMcb, RunIndex, Vcn, Lcn, Count )) {
//
// Our caller should only be off by one or two (if there is
// a hole) runs.
//
ASSERT(RunIndex <= (FsRtlNumberOfRunsInLargeMcb(&Entry->LargeMcb) + 1));
//
// Get the first Vcn past the last Vcn in a run. It is -1 if there
// are no runs.
//
if (!FsRtlLookupLastLargeMcbEntry( &Entry->LargeMcb, Vcn, Lcn )) {
*Vcn = -1;
}
*Vcn += Entry->NtfsMcbArray->StartingVcn + 1;
//
// If that one is beyond the ending Vcn, then get out.
// Otherwise there is a hole at the end of the range, and we
// must return that when he is reading one index beyond the
// last run. If we have a run index beyond that, then it is
// time to return FALSE as well.
//
if ((*Vcn > Entry->NtfsMcbArray->EndingVcn) ||
(RunIndex > FsRtlNumberOfRunsInLargeMcb(&Entry->LargeMcb))) {
try_return(Result = FALSE);
}
//
// If we go off the end of the Mcb, but are in the range, then we
// return a hole to the end of the range.
//
*Lcn = UNUSED_LCN;
*Count = Entry->NtfsMcbArray->EndingVcn - *Vcn + 1;
} else {
//
// Otherwise we have a hit on the large mcb and need to bias the returned
// vcn by the starting vcn value for this range.
//
*Vcn = *Vcn + Entry->NtfsMcbArray->StartingVcn;
}
//
// Make certain we aren't returning a VCN that maps over to
// the next range.
//
ASSERT(*Vcn - 1 != Entry->NtfsMcbArray->EndingVcn);
Result = TRUE;
try_exit: NOTHING;
} finally {
NtfsReleaseNtfsMcbMutex( Mcb );
}
return Result;
}
BOOLEAN
NtfsSplitNtfsMcb (
IN PNTFS_MCB Mcb,
IN LONGLONG Vcn,
IN LONGLONG Amount
)
/*++
Routine Description:
This routine splits an mcb
Arguments:
Mcb - Supplies the Mcb being maniuplated
Vcn - Supplies the Vcn to be shifted
Amount - Supplies the amount to shift by
Return Value:
BOOLEAN - TRUE if worked okay and FALSE otherwise
--*/
{
ULONG RangeIndex;
PNTFS_MCB_ENTRY Entry;
ULONG i;
ASSERT_STREAM_EXCLUSIVE(Mcb);
NtfsVerifyNtfsMcb(Mcb);
//
// Locate the array entry that has the hit for the input vcn
//
RangeIndex = NtfsMcbLookupArrayIndex(Mcb, Vcn);
Entry = Mcb->NtfsMcbArray[RangeIndex].NtfsMcbEntry;
//
// Now if the entry is not null then we have to call the large
// mcb package to split the mcb. Bias the vcn by the starting vcn
//
if (Entry != NULL) {
if (!FsRtlSplitLargeMcb( &Entry->LargeMcb,
Vcn - Mcb->NtfsMcbArray[RangeIndex].StartingVcn,
Amount )) {
NtfsVerifyNtfsMcb(Mcb);
return FALSE;
}
}
//
// Even if the entry is null we will march through the rest of our ranges
// updating the ending vcn and starting vcn as we go. We will update the
// ending vcn for the range we split and only update the starting vcn
// for the last entry, because its ending vcn is already max long long
//
for (i = RangeIndex + 1; i < Mcb->NtfsMcbArraySizeInUse; i += 1) {
Mcb->NtfsMcbArray[i - 1].EndingVcn += Amount;
Mcb->NtfsMcbArray[i].StartingVcn += Amount;
}
//
// And grow the last range unless it would wrap.
//
if ((Mcb->NtfsMcbArray[i - 1].EndingVcn + Amount) > Mcb->NtfsMcbArray[i - 1].EndingVcn) {
Mcb->NtfsMcbArray[i - 1].EndingVcn += Amount;
}
//
// Then return to our caller
//
NtfsVerifyNtfsMcb(Mcb);
return TRUE;
}
VOID
NtfsRemoveNtfsMcbEntry (
IN PNTFS_MCB Mcb,
IN LONGLONG StartingVcn,
IN LONGLONG Count
)
/*++
Routine Description:
This routine removes an range of mappings from the Mcb. After
the call the mapping for the range will be a hole. It is an
error to call this routine with the mapping range being removed
also being unloaded.
Arguments:
Mcb - Supplies the Mcb being maniuplated
StartingVcn - Supplies the starting Vcn to remove
Count - Supplies the number of mappings to remove
Return Value:
None.
--*/
{
LONGLONG Vcn;
LONGLONG RunLength;
LONGLONG RemainingCount;
ULONG RangeIndex;
PNTFS_MCB_ENTRY Entry;
VCN EntryStartingVcn;
VCN EntryEndingVcn;
ASSERT_STREAM_EXCLUSIVE(Mcb);
NtfsVerifyNtfsMcb(Mcb);
//
// Loop through the range of vcn's that we need to remove
//
for (Vcn = StartingVcn, RemainingCount = Count;
Vcn < StartingVcn + Count;
Vcn += RunLength, RemainingCount -= RunLength) {
//
// Locate the array entry that has the hit for the vcn
//
RangeIndex = NtfsMcbLookupArrayIndex(Mcb, Vcn);
Entry = Mcb->NtfsMcbArray[RangeIndex].NtfsMcbEntry;
EntryStartingVcn = Mcb->NtfsMcbArray[RangeIndex].StartingVcn;
EntryEndingVcn = Mcb->NtfsMcbArray[RangeIndex].EndingVcn;
//
// Compute how much to delete from the entry. We will delete to
// to end of the entry or as much as count is remaining
//
RunLength = EntryEndingVcn - Vcn + 1;
//
// If the Mcb is set up correctly, the only way we can get
// RunLength == 0 is if the Mcb is completely empty. Assume
// that this is error recovery, and that it is ok.
//
if ((Entry == NULL) || (RunLength == 0)) {
break;
}
//
// If that is too much, then just delete what we need.
//
if ((ULONGLONG)RunLength > (ULONGLONG)RemainingCount) { RunLength = RemainingCount; }
//
// Now remove the mapping from the large mcb, bias the vcn
// by the start of the range
//
FsRtlRemoveLargeMcbEntry( &Entry->LargeMcb, Vcn - EntryStartingVcn, RunLength );
}
NtfsVerifyNtfsMcb(Mcb);
return;
}
BOOLEAN
NtfsAddNtfsMcbEntry (
IN PNTFS_MCB Mcb,
IN LONGLONG Vcn,
IN LONGLONG Lcn,
IN LONGLONG RunCount,
IN BOOLEAN AlreadySynchronized
)
/*++
Routine Description:
This routine add a new entry to a Mcb
Arguments:
Mcb - Supplies the Mcb being modified
Vcn - Supplies the Vcn that we are providing a mapping for
Lcn - Supplies the Lcn corresponding to the input Vcn if run count is non zero
RunCount - Supplies the size of the run following the hole
AlreadySynchronized - Indicates if the caller has already acquired the mcb mutex
Return Value:
BOOLEAN - TRUE if the mapping was added successfully and FALSE otherwise
--*/
{
LONGLONG LocalVcn;
LONGLONG LocalLcn;
LONGLONG RunLength;
LONGLONG RemainingCount;
ULONG RangeIndex;
PNTFS_MCB_ENTRY Entry;
PNTFS_MCB_ENTRY NewEntry = NULL;
LONGLONG EntryStartingVcn;
LONGLONG EntryEndingVcn;
LONGLONG PrevEndingVcn;
BOOLEAN Result = FALSE;
if (!AlreadySynchronized) { NtfsAcquireNtfsMcbMutex( Mcb ); }
NtfsVerifyNtfsMcb(Mcb);
try {
//
// Loop through the range of vcn's that we need to add
//
for (LocalVcn = Vcn, LocalLcn = Lcn, RemainingCount = RunCount;
LocalVcn < Vcn + RunCount;
LocalVcn += RunLength, LocalLcn += RunLength, RemainingCount -= RunLength) {
//
// Locate the array entry that has the hit for the vcn
//
RangeIndex = NtfsMcbLookupArrayIndex(Mcb, LocalVcn);
Entry = Mcb->NtfsMcbArray[RangeIndex].NtfsMcbEntry;
EntryStartingVcn = Mcb->NtfsMcbArray[RangeIndex].StartingVcn;
//
// Now if the entry doesn't exist then we'll need to create one
//
if (Entry == NULL) {
//
// See if we need to get the first entry in the initial structs.
//
if (Mcb->NtfsMcbArraySize == MCB_ARRAY_PHASE1_SIZE) {
Entry = &CONTAINING_RECORD(&Mcb->NtfsMcbArray[0],
NTFS_MCB_INITIAL_STRUCTS,
Phase1.SingleMcbArrayEntry)->Phase1.McbEntry;
//
// Allocate pool and initialize the fields in of the entry
//
} else {
NewEntry =
Entry = NtfsAllocatePoolWithTag( Mcb->PoolType, sizeof(NTFS_MCB_ENTRY), 'MftN' );
}
//
// Initialize the entry but don't put into the Mcb array until
// initialization is complete.
//
Entry->NtfsMcb = Mcb;
Entry->NtfsMcbArray = &Mcb->NtfsMcbArray[RangeIndex];
FsRtlInitializeLargeMcb( &Entry->LargeMcb, Mcb->PoolType );
//
// Now put the entry into the lru queue under the protection of
// the global mutex
//
ExAcquireFastMutex( &NtfsMcbFastMutex );
//
// Only put paged Mcb entries in the queue.
//
if (Mcb->PoolType == PagedPool) {
NtfsMcbEnqueueLruEntry( Mcb, Entry );
}
//
// Now that the initialization is complete we can store
// this entry in the Mcb array. This will now be cleaned
// up with the Scb if there is a future error.
//
Mcb->NtfsMcbArray[RangeIndex].NtfsMcbEntry = Entry;
NewEntry = NULL;
//
// Check if we should fire off the cleanup lru queue work item
//
if ((NtfsMcbCurrentLevel > NtfsMcbHighWaterMark) && !NtfsMcbCleanupInProgress) {
NtfsMcbCleanupInProgress = TRUE;
ExInitializeWorkItem( &NtfsMcbWorkItem, NtfsMcbCleanupLruQueue, NULL );
ExQueueWorkItem( &NtfsMcbWorkItem, CriticalWorkQueue );
}
ExReleaseFastMutex( &NtfsMcbFastMutex );
}
//
// Get out if he is trying to add a hole. At least we created the LargeMcb
//
if (Lcn == UNUSED_LCN) {
try_return( Result = TRUE );
}
//
// If this request goes beyond the end of the range,
// and it is the last range, and we will simply
// grow it.
//
EntryEndingVcn = LocalVcn + RemainingCount - 1;
if ((EntryEndingVcn > Mcb->NtfsMcbArray[RangeIndex].EndingVcn) &&
((RangeIndex + 1) == Mcb->NtfsMcbArraySizeInUse)) {
PrevEndingVcn = Mcb->NtfsMcbArray[RangeIndex].EndingVcn;
Mcb->NtfsMcbArray[RangeIndex].EndingVcn = EntryEndingVcn;
//
// Otherwise, just insert enough of this run to go to the end
// of the range.
//
} else {
EntryEndingVcn = Mcb->NtfsMcbArray[RangeIndex].EndingVcn;
}
//
// At this point the entry exists so now compute how much to add
// We will add to end of the entry or as much as count allows us
//
RunLength = EntryEndingVcn - LocalVcn + 1;
if (((ULONGLONG)RunLength) > ((ULONGLONG)RemainingCount)) { RunLength = RemainingCount; }
//
// We need to deal with the case where a range is larger than (2^32 - 1) clusters.
// If there are no runs in this range then the state is legal. Otherwise we
// need to split up the entry.
//
if (EntryEndingVcn - EntryStartingVcn >= MAX_CLUSTERS_PER_RANGE) {
//
// We should only be adding this entry as part of a transaction and the
// snapshot limits should force this range to be unloaded on error.
//
ASSERT( ExIsResourceAcquiredExclusiveLite( ((PSCB) (Mcb->FcbHeader))->Header.Resource ));
ASSERT( ((PSCB) (Mcb->FcbHeader))->ScbSnapshot != NULL);
if (Mcb->NtfsMcbArray[RangeIndex].StartingVcn < ((PSCB) (Mcb->FcbHeader))->ScbSnapshot->LowestModifiedVcn) {
((PSCB) (Mcb->FcbHeader))->ScbSnapshot->LowestModifiedVcn = Mcb->NtfsMcbArray[RangeIndex].StartingVcn;
}
if (Mcb->NtfsMcbArray[RangeIndex].EndingVcn > ((PSCB) (Mcb->FcbHeader))->ScbSnapshot->HighestModifiedVcn) {
((PSCB) (Mcb->FcbHeader))->ScbSnapshot->HighestModifiedVcn = Mcb->NtfsMcbArray[RangeIndex].EndingVcn;
}
//
// If the count in the this Mcb is non-zero then we must be growing the
// range. We can simply split at the previoius end of the Mcb. It must
// be legal.
//
if (FsRtlNumberOfRunsInLargeMcb( &Entry->LargeMcb ) != 0) {
ASSERT( PrevEndingVcn < EntryEndingVcn );
NtfsInsertNewRange( Mcb, PrevEndingVcn + 1, RangeIndex, FALSE );
//
// There are no runs currently in this range. If we are at the
// start of the range then split at our maximum range value.
// Otherwise split at the Vcn being inserted. We don't need
// to be too smart here. The mapping pair package will decide where
// the final range values are.
//
} else if (LocalVcn == EntryStartingVcn) {
NtfsInsertNewRange( Mcb,
EntryStartingVcn + MAX_CLUSTERS_PER_RANGE,
RangeIndex,
FALSE );
//
// Go ahead and split at the CurrentVcn. On our next pass we will
// trim the length of this new range if necessary.
//
} else {
NtfsInsertNewRange( Mcb,
LocalVcn,
RangeIndex,
FALSE );
}
//
// Set the run length to 0 and go back to the start of the loop.
// We will encounter the inserted range on the next pass.
//
RunLength = 0;
continue;
}
//
// Now add the mapping from the large mcb, bias the vcn
// by the start of the range
//
ASSERT( (LocalVcn - EntryStartingVcn) >= 0 );
if (!FsRtlAddLargeMcbEntry( &Entry->LargeMcb,
LocalVcn - EntryStartingVcn,
LocalLcn,
RunLength )) {
try_return( Result = FALSE );
}
}
Result = TRUE;
try_exit: NOTHING;
} finally {
NtfsVerifyNtfsMcb(Mcb);
if (!AlreadySynchronized) { NtfsReleaseNtfsMcbMutex( Mcb ); }
if (NewEntry != NULL) { NtfsFreePool( NewEntry ); }
}
return Result;
}
VOID
NtfsUnloadNtfsMcbRange (
IN PNTFS_MCB Mcb,
IN LONGLONG StartingVcn,
IN LONGLONG EndingVcn,
IN BOOLEAN TruncateOnly,
IN BOOLEAN AlreadySynchronized
)
/*++
Routine Description:
This routine unloads the mapping stored in the Mcb. After
the call everything from startingVcn and endingvcn is now unmapped and unknown.
Arguments:
Mcb - Supplies the Mcb being manipulated
StartingVcn - Supplies the first Vcn which is no longer being mapped
EndingVcn - Supplies the last vcn to be unloaded
TruncateOnly - Supplies TRUE if last affected range should only be
truncated, or FALSE if it should be unloaded (as during
error recovery)
AlreadySynchronized - Supplies TRUE if our caller already owns the Mcb mutex.
Return Value:
None.
--*/
{
ULONG StartingRangeIndex;
ULONG EndingRangeIndex;
ULONG i;
if (!AlreadySynchronized) { NtfsAcquireNtfsMcbMutex( Mcb ); }
//
// Verify that we've been called to unload a valid range. If we haven't,
// then there's nothing we can unload, so we just return here. Still,
// we'll assert so we can see why we were called with an invalid range.
//
if ((StartingVcn < 0) || (EndingVcn < StartingVcn)) {
//
// The only legal case is if the range is empty.
//
ASSERT( StartingVcn == EndingVcn + 1 );
if (!AlreadySynchronized) { NtfsReleaseNtfsMcbMutex( Mcb ); }
return;
}
NtfsVerifyNtfsMcb(Mcb);
NtfsVerifyUncompressedNtfsMcb(Mcb,StartingVcn,EndingVcn);
//
// Get the starting and ending range indices for this call
//
StartingRangeIndex = NtfsMcbLookupArrayIndex( Mcb, StartingVcn );
EndingRangeIndex = NtfsMcbLookupArrayIndex( Mcb, EndingVcn );
//
// Use try finally to enforce common termination processing.
//
try {
//
// For all paged Mcbs, just unload all ranges touched by the
// unload range, and collapse with any unloaded neighbors.
//
if (Mcb->PoolType == PagedPool) {
//
// Handle truncate case. The first test insures that we only truncate
// the Mcb were were initialized with (we cannot deallocate it).
//
// Also only truncate if ending is MAXLONGLONG and we are not eliminating
// the entire range, because that is the common truncate case, and we
// do not want to unload the last range every time we truncate on close.
//
if (((StartingRangeIndex == 0) && (Mcb->NtfsMcbArraySizeInUse == 1))
||
(TruncateOnly && (StartingVcn != Mcb->NtfsMcbArray[StartingRangeIndex].StartingVcn))) {
//
// If this is not a truncate call, make sure to eliminate the
// entire range.
//
if (!TruncateOnly) {
StartingVcn = 0;
}
if (Mcb->NtfsMcbArray[StartingRangeIndex].NtfsMcbEntry != NULL) {
FsRtlTruncateLargeMcb( &Mcb->NtfsMcbArray[StartingRangeIndex].NtfsMcbEntry->LargeMcb,
StartingVcn - Mcb->NtfsMcbArray[StartingRangeIndex].StartingVcn );
}
Mcb->NtfsMcbArray[StartingRangeIndex].EndingVcn = StartingVcn - 1;
StartingRangeIndex += 1;
}
//
// Unload entries that are beyond the starting range index
//
for (i = StartingRangeIndex; i <= EndingRangeIndex; i += 1) {
UnloadEntry( Mcb, i );
}
//
// If there is a preceding unloaded range, we must collapse him too.
//
if ((StartingRangeIndex != 0) &&
(Mcb->NtfsMcbArray[StartingRangeIndex - 1].NtfsMcbEntry == NULL)) {
StartingRangeIndex -= 1;
}
//
// If there is a subsequent unloaded range, we must collapse him too.
//
if ((EndingRangeIndex != (Mcb->NtfsMcbArraySizeInUse - 1)) &&
(Mcb->NtfsMcbArray[EndingRangeIndex + 1].NtfsMcbEntry == NULL)) {
EndingRangeIndex += 1;
}
//
// Now collapse empty ranges.
//
if (StartingRangeIndex < EndingRangeIndex) {
NtfsCollapseRanges( Mcb, StartingRangeIndex, EndingRangeIndex );
}
try_return(NOTHING);
}
//
// For nonpaged Mcbs, there is only one range and we truncate it.
//
ASSERT((StartingRangeIndex | EndingRangeIndex) == 0);
if (Mcb->NtfsMcbArray[0].NtfsMcbEntry != NULL) {
FsRtlTruncateLargeMcb( &Mcb->NtfsMcbArray[0].NtfsMcbEntry->LargeMcb, StartingVcn );
}
Mcb->NtfsMcbArray[0].EndingVcn = StartingVcn - 1;
try_exit: NOTHING;
} finally {
//
// Truncate all unused entries from the end by dropping ArraySizeInUse
// to be the index of the last loaded entry + 1.
//
for (i = Mcb->NtfsMcbArraySizeInUse - 1;
(Mcb->NtfsMcbArray[i].NtfsMcbEntry == NULL);
i--) {
//
// If the first range is unloaded, set it to its initial state
// (empty) and break out.
//
if (i==0) {
Mcb->NtfsMcbArray[0].EndingVcn = -1;
break;
}
}
Mcb->NtfsMcbArraySizeInUse = i + 1;
//
// See if we broke anything.
//
NtfsVerifyNtfsMcb(Mcb);
NtfsVerifyUncompressedNtfsMcb(Mcb,StartingVcn,EndingVcn);
if (!AlreadySynchronized) { NtfsReleaseNtfsMcbMutex( Mcb ); }
}
return;
}
VOID
NtfsDefineNtfsMcbRange (
IN PNTFS_MCB Mcb,
IN LONGLONG StartingVcn,
IN LONGLONG EndingVcn,
IN BOOLEAN AlreadySynchronized
)
/*++
Routine Description:
This routine splits an existing range within the Mcb into two ranges
Arguments:
Mcb - Supplies the Mcb being modified
StartingVcn - Supplies the beginning of the new range being split
EndingVcn - Supplies the ending vcn to include in this new range
AlreadySynchronized - Indicates if the caller has already acquired the mcb mutex
Return Value:
None.
--*/
{
ULONG StartingRangeIndex, EndingRangeIndex;
if (!AlreadySynchronized) { NtfsAcquireNtfsMcbMutex( Mcb ); }
NtfsVerifyNtfsMcb(Mcb);
//
// Make sure we're of the right pool type
//
// If the ending vcn is less than or equal to the starting vcn then we will no op
// this call
//
if ((Mcb->PoolType != PagedPool) || (EndingVcn < StartingVcn)) {
if (!AlreadySynchronized) { NtfsReleaseNtfsMcbMutex( Mcb ); }
return;
}
try {
PNTFS_MCB_ARRAY StartingArray;
PNTFS_MCB_ARRAY EndingArray;
PNTFS_MCB_ENTRY StartingEntry;
PNTFS_MCB_ENTRY EndingEntry;
ULONG i;
//
// Locate the Starting Mcb
//
StartingRangeIndex = NtfsMcbLookupArrayIndex( Mcb, StartingVcn );
//
// Locate the ending Mcb
//
EndingRangeIndex = NtfsMcbLookupArrayIndex( Mcb, EndingVcn );
EndingArray = &Mcb->NtfsMcbArray[EndingRangeIndex];
EndingEntry = EndingArray->NtfsMcbEntry;
//
// Special case: extending last range where StartingVcn matches
//
if (((EndingRangeIndex + 1) == Mcb->NtfsMcbArraySizeInUse) &&
(StartingVcn == EndingArray->StartingVcn) &&
(EndingArray->EndingVcn <= EndingVcn)) {
//
// Since this range already starts with the desired Vcn
// we adjust the end to match the caller's request
//
EndingArray->EndingVcn = EndingVcn;
ASSERT( ((EndingVcn - StartingVcn) < MAX_CLUSTERS_PER_RANGE) ||
(EndingEntry == NULL) ||
(FsRtlNumberOfRunsInLargeMcb( &EndingEntry->LargeMcb ) == 0) );
leave;
}
//
// Special case: handling defining a range after the end of the file
//
if (StartingVcn > EndingArray->EndingVcn) {
LONGLONG OldEndingVcn = EndingArray->EndingVcn;
//
// Has to be the last range.
//
ASSERT( StartingRangeIndex == EndingRangeIndex );
ASSERT( (EndingRangeIndex + 1) == Mcb->NtfsMcbArraySizeInUse );
//
// First extend the last range to include our new range.
//
EndingArray->EndingVcn = EndingVcn;
//
// We will be adding a new range and inserting or growing the
// previous range up to the new range. If the previous range is
// *empty* but has an NtfsMcbEntry then we want to unload the entry.
// Otherwise we will grow that range to the correct value but
// the Mcb won't contain the clusters for the range. We want
// to unload that range and update the OldEndingVcn value so
// as not to create two empty ranges prior to this.
//
if ((OldEndingVcn == -1) &&
(EndingArray->NtfsMcbEntry != NULL)) {
ASSERT( EndingRangeIndex == 0 );
UnloadEntry( Mcb, EndingRangeIndex );
}
//
// Create the range the caller specified.
//
NtfsInsertNewRange( Mcb, StartingVcn, EndingRangeIndex, TRUE );
DebugDoit( StartingArray = EndingArray = NULL );
DebugDoit( StartingEntry = EndingEntry = NULL );
//
// If this range does not abut the previous last range, *and*
// the previous range was not *empty*, then we have to define a
// range to contain the unloaded space in the middle.
//
if (((OldEndingVcn + 1) < StartingVcn) &&
((OldEndingVcn + 1) != 0)) {
NtfsInsertNewRange( Mcb, OldEndingVcn + 1, StartingRangeIndex, TRUE );
DebugDoit( StartingArray = EndingArray = NULL );
DebugDoit( StartingEntry = EndingEntry = NULL );
}
ASSERT( ((EndingVcn - StartingVcn) < MAX_CLUSTERS_PER_RANGE) ||
(Mcb->NtfsMcbArray[NtfsMcbLookupArrayIndex( Mcb, EndingVcn )].NtfsMcbEntry == NULL) ||
(FsRtlNumberOfRunsInLargeMcb( &Mcb->NtfsMcbArray[NtfsMcbLookupArrayIndex( Mcb, EndingVcn )].NtfsMcbEntry->LargeMcb ) == 0) );
leave;
}
//
// Check if we really need to insert a new range at the ending vcn
// we only need to do the work if there is not already one at that vcn
// and this is not the last range
//
if (EndingVcn < EndingArray->EndingVcn) {
NtfsInsertNewRange( Mcb, EndingVcn + 1, EndingRangeIndex, FALSE );
DebugDoit( StartingArray = EndingArray = NULL );
DebugDoit( StartingEntry = EndingEntry = NULL );
//
// Recache pointers since NtfsMcbArray may have moved
//
EndingArray = &Mcb->NtfsMcbArray[EndingRangeIndex];
EndingEntry = EndingArray->NtfsMcbEntry;
ASSERT( EndingArray->EndingVcn == EndingVcn );
}
//
// Determine location for insertion
//
StartingArray = &Mcb->NtfsMcbArray[StartingRangeIndex];
StartingEntry = StartingArray->NtfsMcbEntry;
//
// Check if we really need to insert a new range at the starting vcn
// we only need to do the work if this Mcb doesn't start at the
// requested Vcn
//
if (StartingArray->StartingVcn < StartingVcn) {
NtfsInsertNewRange( Mcb, StartingVcn, StartingRangeIndex, FALSE );
DebugDoit( StartingArray = EndingArray = NULL );
DebugDoit( StartingEntry = EndingEntry = NULL );
StartingRangeIndex++;
StartingArray = &Mcb->NtfsMcbArray[StartingRangeIndex];
StartingEntry = StartingArray->NtfsMcbEntry;
ASSERT( StartingArray->StartingVcn == StartingVcn );
EndingRangeIndex++;
// EndingArray = &Mcb->NtfsMcbArray[EndingRangeIndex];
// EndingEntry = EndingArray->NtfsMcbEntry;
// ASSERT( EndingArray->EndingVcn == EndingVcn );
}
ASSERT( StartingArray->StartingVcn == StartingVcn );
// ASSERT( EndingArray->EndingVcn == EndingVcn );
//
// At this point, we have a Vcn range beginning at StartingVcn stored in
// NtfsMcbArray[StartingRangeIndex] AND ending at EndingVcb which is the
// end of NtfsMcbArray[StartingRangeIndex]. This is a collection (>= 1)
// of NtfsMcbEntry's. Our caller expects to have these reduced to
// a single run. Note that our caller should never break the restriction
// on maximum number clusters per range.
//
while (StartingRangeIndex != EndingRangeIndex) {
VCN Vcn;
BOOLEAN MoreEntries;
LCN Lcn;
LONGLONG Count;
ULONG Index;
PNTFS_MCB_ARRAY NextArray;
PNTFS_MCB_ENTRY NextEntry;
//
// We merge the contents of NtfsMcbArray[StartingRangeIndex + 1] into
// NtfsMcbArray[StartingRangeIndex]
//
//
// Look up the first Vcn to move in the second Mcb. If this
// Mcb consists of one large hole then there is nothing to
// move.
//
NextArray = &Mcb->NtfsMcbArray[StartingRangeIndex + 1];
NextEntry = NextArray->NtfsMcbEntry;
//
// We should never exceed our limit on the maximum number of clusters.
//
ASSERT( ((NextArray->EndingVcn - StartingArray->StartingVcn + 1) <= MAX_CLUSTERS_PER_RANGE) ||
((FsRtlNumberOfRunsInLargeMcb( &StartingEntry->LargeMcb ) == 0) &&
(FsRtlNumberOfRunsInLargeMcb( &NextEntry->LargeMcb ) == 0)) );
Vcn = 0;
MoreEntries = FsRtlLookupLargeMcbEntry( &NextEntry->LargeMcb,
Vcn,
&Lcn,
&Count,
NULL,
NULL,
&Index );
//
// Loop to move entries over.
//
//
// this is the case described by bug #9054.
// the mcb has somehow? been incorrectly split
// so this will force everything to be unloaded
// instead of half loaded and half unloaded
//
// the assert is here simply for debug purposes.
// if this assert fires then we simply want to step
// thru the code and examine the mcb state to
// be certain that our assumtions about this bug
// are correct. the actual bug scenario could not
// be reproed so this code path is un-tested.
//
ASSERT( StartingEntry != NULL );
if (StartingEntry != NULL) {
while (MoreEntries) {
//
// If this entry is not a hole, move it.
//
if (Lcn != UNUSED_LCN) {
FsRtlAddLargeMcbEntry( &StartingEntry->LargeMcb,
(Vcn + NextArray->StartingVcn) - StartingArray->StartingVcn,
Lcn,
Count );
}
Index += 1;
MoreEntries = FsRtlGetNextLargeMcbEntry( &NextEntry->LargeMcb,
Index,
&Vcn,
&Lcn,
&Count );
}
ASSERT( StartingArray->EndingVcn < NextArray->EndingVcn );
StartingArray->EndingVcn = NextArray->EndingVcn;
}
//
// We've completely emptied the next Mcb. Unload it.
//
UnloadEntry( Mcb, StartingRangeIndex + 1 );
Mcb->NtfsMcbArraySizeInUse -= 1;
//
// Compact the array
//
RtlMoveMemory( StartingArray + 1,
StartingArray + 2,
sizeof( NTFS_MCB_ARRAY ) * (Mcb->NtfsMcbArraySizeInUse - (StartingRangeIndex + 1))
);
//
// Adjust the backpointers
//
for (i = StartingRangeIndex + 1;
i < Mcb->NtfsMcbArraySizeInUse;
i += 1) {
if (Mcb->NtfsMcbArray[i].NtfsMcbEntry != NULL) {
Mcb->NtfsMcbArray[i].NtfsMcbEntry->NtfsMcbArray = &Mcb->NtfsMcbArray[i];
}
}
EndingRangeIndex--;
}
} finally {
NtfsVerifyNtfsMcb(Mcb);
if (!AlreadySynchronized) { NtfsReleaseNtfsMcbMutex( Mcb ); }
}
return;
}
//
// Local support routines
//
ULONG
NtfsMcbLookupArrayIndex (
IN PNTFS_MCB Mcb,
IN VCN Vcn
)
/*++
Routine Description:
This routines searches the mcb array for an entry that contains
the input vcn value
Arguments:
Mcb - Supplies the Mcb being queried
Vcn - Supplies the Vcn to lookup
Return Value:
ULONG - The index of the entry containing the input Vcn value
--*/
{
ULONG Index;
ULONG MinIndex;
ULONG MaxIndex;
NtfsVerifyNtfsMcb(Mcb);
//
// Do a quick binary search for the entry containing the vcn
//
MinIndex = 0;
MaxIndex = Mcb->NtfsMcbArraySizeInUse - 1;
while (TRUE) {
Index = (MaxIndex + MinIndex) / 2;
if ((Mcb->NtfsMcbArray[Index].StartingVcn > Vcn) &&
(Index != 0)) {
MaxIndex = Index - 1;
} else if ((Mcb->NtfsMcbArray[Index].EndingVcn < Vcn) &&
(Index != Mcb->NtfsMcbArraySizeInUse - 1)) {
MinIndex = Index + 1;
} else {
return Index;
}
}
}
//
// Local support routines
//
VOID
NtfsInsertNewRange (
IN PNTFS_MCB Mcb,
IN LONGLONG StartingVcn,
IN ULONG ArrayIndex,
IN BOOLEAN MakeNewRangeEmpty
)
/*++
This routine is used to add a new range at the specified vcn and index location.
Since this routine will resize the NtfsMcbArray, the caller must be sure to
invalidate all cached pointers to NtfsMcbArray entries.
Arguments:
Mcb - Supplies the Mcb being modified
StartingVcn - Supplies the vcn for the new range
ArrayIndex - Supplies the index currently containing the starting vcn
MakeNewRangeEmpty - TRUE if the caller wants the new range unloaded regardless
of the state of the current range
Return Value:
None.
--*/
{
ULONG i;
PNTFS_MCB_ENTRY Entry;
PNTFS_MCB_ENTRY NewEntry;
NtfsVerifyNtfsMcb(Mcb);
//
// Check if we need to grow the array
//
if (Mcb->NtfsMcbArraySizeInUse >= Mcb->NtfsMcbArraySize) {
NtfsGrowMcbArray( Mcb );
}
//
// Now move entries that are beyond the array index over by one to make
// room for the new entry
//
if (ArrayIndex + 2 <= Mcb->NtfsMcbArraySizeInUse) {
RtlMoveMemory( &Mcb->NtfsMcbArray[ArrayIndex + 2],
&Mcb->NtfsMcbArray[ArrayIndex + 1],
sizeof(NTFS_MCB_ARRAY) * (Mcb->NtfsMcbArraySizeInUse - ArrayIndex - 1));
for (i = ArrayIndex + 2; i < Mcb->NtfsMcbArraySizeInUse + 1; i += 1) {
if (Mcb->NtfsMcbArray[i].NtfsMcbEntry != NULL) {
Mcb->NtfsMcbArray[i].NtfsMcbEntry->NtfsMcbArray = &Mcb->NtfsMcbArray[i];
}
}
}
//
// Increment our in use count by one
//
Mcb->NtfsMcbArraySizeInUse += 1;
//
// Now fix the starting and ending Vcn values for the old entry and the
// new entry
//
Mcb->NtfsMcbArray[ArrayIndex + 1].StartingVcn = StartingVcn;
Mcb->NtfsMcbArray[ArrayIndex + 1].EndingVcn = Mcb->NtfsMcbArray[ArrayIndex].EndingVcn;
Mcb->NtfsMcbArray[ArrayIndex + 1].NtfsMcbEntry = NULL;
Mcb->NtfsMcbArray[ArrayIndex].EndingVcn = StartingVcn - 1;
//
// Now if the entry is old entry is not null then we have a bunch of work to do
//
if (!MakeNewRangeEmpty && (Entry = Mcb->NtfsMcbArray[ArrayIndex].NtfsMcbEntry) != NULL) {
LONGLONG Vcn;
LONGLONG Lcn;
LONGLONG RunLength;
ULONG Index;
BOOLEAN FreeNewEntry = FALSE;
//
// Use a try-finally in case the Mcb initialization fails.
//
try {
//
// Allocate the new entry slot
//
NewEntry = NtfsAllocatePoolWithTag( Mcb->PoolType, sizeof(NTFS_MCB_ENTRY), 'MftN' );
FreeNewEntry = TRUE;
NewEntry->NtfsMcb = Mcb;
NewEntry->NtfsMcbArray = &Mcb->NtfsMcbArray[ArrayIndex + 1];
FsRtlInitializeLargeMcb( &NewEntry->LargeMcb, Mcb->PoolType );
ExAcquireFastMutex( &NtfsMcbFastMutex );
NtfsMcbEnqueueLruEntry( Mcb, NewEntry );
ExReleaseFastMutex( &NtfsMcbFastMutex );
//
// Now that the initialization is complete we can store
// this entry in the Mcb array. This will now be cleaned
// up with the Scb if there is a future error.
//
Mcb->NtfsMcbArray[ArrayIndex + 1].NtfsMcbEntry = NewEntry;
FreeNewEntry = FALSE;
//
// Lookup the entry containing the starting vcn in the old entry and put it
// in the new entry. But only if the entry exists otherwise we know that
// the large mcb doesn't extend into the new range
//
if (FsRtlLookupLargeMcbEntry( &Entry->LargeMcb,
StartingVcn - Mcb->NtfsMcbArray[ArrayIndex].StartingVcn,
&Lcn,
&RunLength,
NULL,
NULL,
&Index )) {
if (Lcn != UNUSED_LCN) {
FsRtlAddLargeMcbEntry( &NewEntry->LargeMcb,
0,
Lcn,
RunLength );
}
//
// Now for every run in the old entry that is beyond the starting vcn we will
// copy it into the new entry. This will also copy over the dummy run at the end
// of the mcb if it exists
//
for (i = Index + 1; FsRtlGetNextLargeMcbEntry( &Entry->LargeMcb, i, &Vcn, &Lcn, &RunLength ); i += 1) {
if (Lcn != UNUSED_LCN) {
ASSERT( (Vcn - (StartingVcn - Mcb->NtfsMcbArray[ArrayIndex].StartingVcn)) >= 0 );
FsRtlAddLargeMcbEntry( &NewEntry->LargeMcb,
Vcn - (StartingVcn - Mcb->NtfsMcbArray[ArrayIndex].StartingVcn),
Lcn,
RunLength );
}
}
//
// Now modify the old mcb to be smaller and put in the dummy run
//
FsRtlTruncateLargeMcb( &Entry->LargeMcb,
StartingVcn - Mcb->NtfsMcbArray[ArrayIndex].StartingVcn );
}
} finally {
if (FreeNewEntry) { NtfsFreePool( NewEntry ); }
}
}
NtfsVerifyNtfsMcb(Mcb);
return;
}
//
// Local support routines
//
VOID
NtfsCollapseRanges (
IN PNTFS_MCB Mcb,
IN ULONG StartingArrayIndex,
IN ULONG EndingArrayIndex
)
/*++
Routine Description:
This routine will remove the indicated array entries
Arguments:
Mcb - Supplies the Mcb being modified
StartingArrayIndex - Supplies the first index to remove
EndingArrayIndex - Supplies the last index to remove
Return Value:
None.
--*/
{
ULONG i;
NtfsVerifyNtfsMcb(Mcb);
//
// Make sure all the ranges are unloaded.
//
DebugDoit(
for (i = StartingArrayIndex; i <= EndingArrayIndex; i++) {
ASSERT(Mcb->NtfsMcbArray[i].NtfsMcbEntry == NULL);
}
);
//
// We keep the first entry by we need to copy over
// the ending vcn of the last entry
//
Mcb->NtfsMcbArray[StartingArrayIndex].EndingVcn = Mcb->NtfsMcbArray[EndingArrayIndex].EndingVcn;
//
// Check if we need to move the ending entries up the array
// if so then move them forward, and adjust the back pointers.
//
if (EndingArrayIndex < Mcb->NtfsMcbArraySizeInUse - 1) {
RtlMoveMemory( &Mcb->NtfsMcbArray[StartingArrayIndex + 1],
&Mcb->NtfsMcbArray[EndingArrayIndex + 1],
sizeof(NTFS_MCB_ARRAY) * (Mcb->NtfsMcbArraySizeInUse - EndingArrayIndex - 1));
for (i = StartingArrayIndex + 1;
i <= (StartingArrayIndex + Mcb->NtfsMcbArraySizeInUse - EndingArrayIndex - 1);
i += 1) {
if (Mcb->NtfsMcbArray[i].NtfsMcbEntry != NULL) {
Mcb->NtfsMcbArray[i].NtfsMcbEntry->NtfsMcbArray = &Mcb->NtfsMcbArray[i];
}
}
}
//
// Decrement the in use count and return to our caller
//
Mcb->NtfsMcbArraySizeInUse -= (EndingArrayIndex - StartingArrayIndex);
NtfsVerifyNtfsMcb(Mcb);
return;
}
//
// Local support routine
//
VOID
NtfsMcbCleanupLruQueue (
IN PVOID Parameter
)
/*++
Routine Description:
This routine is called as an ex work queue item and its job is
to free up the lru queue until we reach the low water mark
Arguments:
Parameter - ignored
Return Value:
None.
--*/
{
PLIST_ENTRY Links;
PNTFS_MCB Mcb;
PNTFS_MCB_ARRAY Array;
PNTFS_MCB_ENTRY Entry;
UNREFERENCED_PARAMETER( Parameter );
//
// Grab the global lock
//
ExAcquireFastMutex( &NtfsMcbFastMutex );
try {
//
// Scan through the lru queue until we either exhaust the queue
// or we've trimmed enough
//
for (Links = NtfsMcbLruQueue.Flink;
(Links != &NtfsMcbLruQueue) && (NtfsMcbCurrentLevel > NtfsMcbLowWaterMark);
Links = Links->Flink ) {
//
// Get the entry and the mcb it points to
//
Entry = CONTAINING_RECORD( Links, NTFS_MCB_ENTRY, LruLinks );
Mcb = Entry->NtfsMcb;
//
// Skip this entry if it is in the open attribute table.
//
if (((PSCB)(Mcb->FcbHeader))->NonpagedScb->OpenAttributeTableIndex != 0) {
continue;
}
//
// Try and lock the mcb
//
if (NtfsLockNtfsMcb( Mcb )) {
NtfsVerifyNtfsMcb(Mcb);
//
// The previous test was an unsafe test. Check again in case
// this entry has been added.
//
if (((PSCB)(Mcb->FcbHeader))->NonpagedScb->OpenAttributeTableIndex == 0) {
//
// We locked the mcb so we can remove this entry, but
// first backup our links pointer so we can continue with the loop
//
Links = Links->Blink;
//
// Get a point to the array entry and then remove this entry and return
// it to pool
//
Array = Entry->NtfsMcbArray;
Array->NtfsMcbEntry = NULL;
NtfsMcbDequeueLruEntry( Mcb, Entry );
FsRtlUninitializeLargeMcb( &Entry->LargeMcb );
if (Mcb->NtfsMcbArraySize != 1) {
NtfsFreePool( Entry );
}
}
NtfsUnlockNtfsMcb( Mcb );
}
}
} finally {
//
// Say we're done with the cleanup so that another one can be fired off when
// necessary
//
NtfsMcbCleanupInProgress = FALSE;
ExReleaseFastMutex( &NtfsMcbFastMutex );
}
//
// Return to our caller
//
return;
}
VOID
NtfsSwapMcbs (
IN PNTFS_MCB McbTarget,
IN PNTFS_MCB McbSource
)
/*++
Routine Description:
This routine swaps the mapping pairs between two mcbs atomically
Arguments:
McbTarget -
McbSource -
Return Value:
None.
--*/
{
ULONG TempNtfsMcbArraySizeInUse;
ULONG TempNtfsMcbArraySize;
PNTFS_MCB_ARRAY TempNtfsMcbArray;
ULONG Index;
ASSERT( McbTarget->PoolType == McbSource->PoolType );
//
// Grab the mutex in the original and new mcb to block everyone out
//
NtfsAcquireNtfsMcbMutex( McbTarget );
NtfsAcquireNtfsMcbMutex( McbSource );
try {
//
// Check if we need to grow either array so they are in the general form
// In the general form we can swap the two by switching the array of mcb entries
//
if (McbSource->NtfsMcbArraySize == MCB_ARRAY_PHASE1_SIZE) {
NtfsGrowMcbArray( McbSource );
}
if (McbSource->NtfsMcbArraySize == MCB_ARRAY_PHASE2_SIZE) {
NtfsGrowMcbArray( McbSource );
}
if (McbTarget->NtfsMcbArraySize == MCB_ARRAY_PHASE1_SIZE) {
NtfsGrowMcbArray( McbTarget);
}
if (McbTarget->NtfsMcbArraySize == MCB_ARRAY_PHASE2_SIZE) {
NtfsGrowMcbArray( McbTarget );
}
//
// Swap the arrays in the two mcb's
//
TempNtfsMcbArraySizeInUse = McbTarget->NtfsMcbArraySizeInUse;
TempNtfsMcbArraySize = McbTarget->NtfsMcbArraySize;
TempNtfsMcbArray = McbTarget->NtfsMcbArray;
McbTarget->NtfsMcbArray = McbSource->NtfsMcbArray;
McbTarget->NtfsMcbArraySize = McbSource->NtfsMcbArraySize;
McbTarget->NtfsMcbArraySizeInUse = McbSource->NtfsMcbArraySizeInUse;
McbSource->NtfsMcbArray = TempNtfsMcbArray;
McbSource->NtfsMcbArraySize = TempNtfsMcbArraySize;
McbSource->NtfsMcbArraySizeInUse = TempNtfsMcbArraySizeInUse;
//
// Fixup the backptr in the array entries to point the the correct mcb
//
for (Index=0; Index < McbSource->NtfsMcbArraySize; Index++) {
if (McbSource->NtfsMcbArray[Index].NtfsMcbEntry != NULL) {
McbSource->NtfsMcbArray[Index].NtfsMcbEntry->NtfsMcb = McbSource;
}
}
for (Index=0; Index < McbTarget->NtfsMcbArraySize; Index++) {
if (McbTarget->NtfsMcbArray[Index].NtfsMcbEntry != NULL) {
McbTarget->NtfsMcbArray[Index].NtfsMcbEntry->NtfsMcb = McbTarget;
}
}
} finally {
NtfsReleaseNtfsMcbMutex( McbSource );
NtfsReleaseNtfsMcbMutex( McbTarget );
}
}
//
// Local support routine
//
BOOLEAN
NtfsLockNtfsMcb (
IN PNTFS_MCB Mcb
)
/*++
Routine Description:
This routine attempts to get the Fcb resource(s) exclusive so that
ranges may be unloaded.
Arguments:
Mcb - Supplies the mcb being queried
Return Value:
--*/
{
//
// Try to acquire paging resource exclusive.
//
if ((Mcb->FcbHeader->PagingIoResource == NULL) ||
ExAcquireResourceExclusiveLite(Mcb->FcbHeader->PagingIoResource, FALSE)) {
//
// Now we can try to acquire the main resource exclusively as well.
//
if (ExAcquireResourceExclusiveLite(Mcb->FcbHeader->Resource, FALSE)) {
return TRUE;
}
//
// We failed to acquire the paging I/O resource, so free the main one
// on the way out.
//
if (Mcb->FcbHeader->PagingIoResource != NULL) {
ExReleaseResourceLite( Mcb->FcbHeader->PagingIoResource );
}
}
//
// Could not get this file exclusive.
//
return FALSE;
}
//
// Local support routine
//
VOID
NtfsUnlockNtfsMcb (
IN PNTFS_MCB Mcb
)
/*++
Routine Description:
This routine verifies that an mcb is properly formed
Arguments:
Mcb - Supplies the mcb being queried
Return Value:
None.
--*/
{
//
// If there is a paging I/O resource, release it first.
//
if (Mcb->FcbHeader->PagingIoResource != NULL) {
ExReleaseResourceLite(Mcb->FcbHeader->PagingIoResource);
}
//
// Now release the main resource.
//
ExReleaseResourceLite(Mcb->FcbHeader->Resource);
}
//
// Local support routine
//
NtfsGrowMcbArray(
IN PNTFS_MCB Mcb
)
/*++
Routine Description:
This routine grows the mcb array. If its phase1 - then it will be promoted to phase2
If its phase2 it will become the general form. If its the general form 8 new entries will be added.
Arguments:
Mcb - Supplies the mcb being grown
Return Value:
None.
--*/
{
PNTFS_MCB_ARRAY NewArray;
ULONG OldArraySize = Mcb->NtfsMcbArraySize;
PNTFS_MCB_ENTRY Entry;
//
// Test for initial case where we only have one array entry.
//
if (Mcb->NtfsMcbArraySize == MCB_ARRAY_PHASE1_SIZE) {
//
// Convince ourselves that we do not have to move the array entry.
//
ASSERT(FIELD_OFFSET(NTFS_MCB_INITIAL_STRUCTS, Phase1.SingleMcbArrayEntry) ==
FIELD_OFFSET(NTFS_MCB_INITIAL_STRUCTS, Phase2.ThreeMcbArrayEntries));
if (Mcb->NtfsMcbArray[0].NtfsMcbEntry != NULL) {
//
// Allocate a new Mcb Entry, copy the current one over and change the pointer.
//
Entry = NtfsAllocatePoolWithTag( Mcb->PoolType, sizeof(NTFS_MCB_ENTRY), 'MftN' );
//
// Once space is allocated, dequeue the old entry.
//
ExAcquireFastMutex( &NtfsMcbFastMutex );
NtfsMcbDequeueLruEntry( Mcb, Mcb->NtfsMcbArray[0].NtfsMcbEntry );
RtlCopyMemory( Entry, Mcb->NtfsMcbArray[0].NtfsMcbEntry, sizeof(NTFS_MCB_ENTRY) );
Mcb->NtfsMcbArray[0].NtfsMcbEntry = Entry;
NtfsMcbEnqueueLruEntry( Mcb, Entry );
ExReleaseFastMutex( &NtfsMcbFastMutex );
}
//
// Now change to using the three array elements
//
Mcb->NtfsMcbArraySize = MCB_ARRAY_PHASE2_SIZE;
} else {
ULONG i;
//
// If we do then allocate an array that can contain 8 more entires
//
NewArray = NtfsAllocatePoolWithTag( Mcb->PoolType, sizeof(NTFS_MCB_ARRAY) * (Mcb->NtfsMcbArraySize + 8), 'mftN' );
Mcb->NtfsMcbArraySize += 8;
//
// Copy over the memory from the old array to the new array and then
// for every loaded entry we need to adjust its back pointer to the
// array
//
RtlCopyMemory( NewArray, Mcb->NtfsMcbArray, sizeof(NTFS_MCB_ARRAY) * OldArraySize );
for (i = 0; i < Mcb->NtfsMcbArraySizeInUse; i += 1) {
if (NewArray[i].NtfsMcbEntry != NULL) {
NewArray[i].NtfsMcbEntry->NtfsMcbArray = &NewArray[i];
}
}
//
// Free the old array if it was not the original array.
//
if (OldArraySize > MCB_ARRAY_PHASE2_SIZE) {
NtfsFreePool( Mcb->NtfsMcbArray );
}
Mcb->NtfsMcbArray = NewArray;
}
//
// Zero the new part of the array.
//
ASSERT(sizeof(NTFS_MCB_ARRAY) == ((PCHAR)&NewArray[1] - (PCHAR)&NewArray[0]));
RtlZeroMemory( &Mcb->NtfsMcbArray[OldArraySize],
(Mcb->NtfsMcbArraySize - OldArraySize) * sizeof(NTFS_MCB_ARRAY) );
}
#ifdef NTFS_VERIFY_MCB
//
// Local support routine
//
VOID
NtfsVerifyNtfsMcb (
IN PNTFS_MCB Mcb
)
/*++
Routine Description:
This routine verifies that an mcb is properly formed
Arguments:
Mcb - Supplies the mcb being queried
Return Value:
--*/
{
ULONG i;
PNTFS_MCB_ARRAY Array;
PNTFS_MCB_ENTRY Entry;
LONGLONG Vbn;
LONGLONG Lbn;
ASSERT(Mcb->FcbHeader != NULL);
ASSERT(Mcb->FcbHeader->NodeTypeCode != 0);
ASSERT((Mcb->PoolType == PagedPool) || (Mcb->PoolType == NonPagedPool));
ASSERT(Mcb->NtfsMcbArraySizeInUse <= Mcb->NtfsMcbArraySize);
for (i = 0; i < Mcb->NtfsMcbArraySizeInUse; i += 1) {
Array = &Mcb->NtfsMcbArray[i];
ASSERT(((i == 0) && (Array->StartingVcn == 0)) ||
((i != 0) && (Array->StartingVcn != 0)));
ASSERT(Array->StartingVcn <= (Array->EndingVcn + 1));
if ((Entry = Array->NtfsMcbEntry) != NULL) {
ASSERT(Entry->NtfsMcb == Mcb);
ASSERT(Entry->NtfsMcbArray == Array);
if (FsRtlLookupLastLargeMcbEntry( &Entry->LargeMcb, &Vbn, &Lbn )) {
ASSERT( Vbn <= (Array->EndingVcn - Array->StartingVcn) );
}
}
}
}
//
// Local support routine
//
VOID
NtfsVerifyUncompressedNtfsMcb (
IN PNTFS_MCB Mcb,
IN LONGLONG StartingVcn,
IN LONGLONG EndingVcn
)
/*++
Routine Description:
This routines checks if an mcb is for an uncompressed scb and then
checks that there are no holes in the mcb. Holes within the range being
removed are legal provided EndingVcn is max long long.
Arguments:
Mcb - Supplies the Mcb being examined
StartingVcn - The starting Vcn being unloaded
EndingVcn - The ending Vcn being unloaded
Return Value:
None
--*/
{
ULONG i;
ULONG j;
PNTFS_MCB_ARRAY Array;
PNTFS_MCB_ENTRY Entry;
LONGLONG Vbn;
LONGLONG Lbn;
LONGLONG Count;
//
// Check if the scb is compressed
//
if (((PSCB)Mcb->FcbHeader)->CompressionUnit != 0) { return; }
//
// For each large mcb in the ntfs mcb we will make sure it doesn't
// have any holes.
//
for (i = 0; i < Mcb->NtfsMcbArraySizeInUse; i += 1) {
Array = &Mcb->NtfsMcbArray[i];
if ((Entry = Array->NtfsMcbEntry) != NULL) {
for (j = 0; FsRtlGetNextLargeMcbEntry(&Entry->LargeMcb,j,&Vbn,&Lbn,&Count); j += 1) {
ASSERT((Lbn != -1) ||
((Vbn + Array->StartingVcn >= StartingVcn) && (EndingVcn == MAXLONGLONG)) ||
FlagOn(((PSCB)Mcb->FcbHeader)->Vcb->VcbState, VCB_STATE_RESTART_IN_PROGRESS));
}
}
}
return;
}
#endif