/*++ Copyright (c) 1990-2000 Microsoft Corporation Module Name: AllocSup.c Abstract: This module implements the Allocation support routines for Cdfs. The data structure used here is the CD_MCB. There is an entry in the Mcb for each dirent for a file. The entry will map the offset within some file to a starting disk offset and number of bytes. The Mcb also contains the interleave information for an extent. An interleave consists of a number of blocks with data and a (possibly different) number of blocks to skip. Any number of data/skip pairs may exist in an extent but the data and skip sizes are the same throughout the extent. We store the following information into an Mcb entry for an extent. FileOffset Offset in file for start of extent DiskOffset Offset on disk for start of extent ByteCount Number of file bytes in extent, no skip bytes DataBlockByteCount Number of bytes in each data block TotalBlockByteCount Number of bytes is data block and skip block The disk offset in the Mcb has already been biased by the size of the Xar block if present. All of the byte count fields are aligned on logical block boundaries. If this is a directory or path table then the file offset has been biased to round the initial disk offset down to a sector boundary. The biasing is done when loading the values into an Mcb entry. An XA file has a header prepended to the file and each sector is 2352 bytes. The allocation information ignores the header and only deals with 2048 byte sectors. Callers into the allocation package have adjusted the starting offset value to reflect 2048 sectors. On return from this package the caller will have to convert from 2048 sector values into raw XA sector values. --*/ #include "cdprocs.h" // // The Bug check file id for this module // #define BugCheckFileId (CDFS_BUG_CHECK_ALLOCSUP) // // Local support routines // ULONG CdFindMcbEntry ( _In_ PIRP_CONTEXT IrpContext, _In_ PFCB Fcb, _In_ LONGLONG FileOffset ); VOID CdDiskOffsetFromMcbEntry ( _In_ PIRP_CONTEXT IrpContext, _In_ PCD_MCB_ENTRY McbEntry, _In_ LONGLONG FileOffset, _Out_ PLONGLONG DiskOffset, _Out_ PULONG ByteCount ); #ifdef ALLOC_PRAGMA #pragma alloc_text(PAGE, CdAddInitialAllocation) #pragma alloc_text(PAGE, CdAddAllocationFromDirent) #pragma alloc_text(PAGE, CdDiskOffsetFromMcbEntry) #pragma alloc_text(PAGE, CdFindMcbEntry) #pragma alloc_text(PAGE, CdInitializeMcb) #pragma alloc_text(PAGE, CdLookupAllocation) #pragma alloc_text(PAGE, CdTruncateAllocation) #pragma alloc_text(PAGE, CdUninitializeMcb) #endif _Requires_lock_held_(_Global_critical_region_) VOID #ifdef _MSC_VER // PREFast currently has no way to express the Fcb==Fcb->Vcb->VolumeDasdFcb early return #pragma warning(suppress: 6001 6101) #endif CdLookupAllocation ( _In_ PIRP_CONTEXT IrpContext, _In_ PFCB Fcb, _In_ LONGLONG FileOffset, _Out_ PLONGLONG DiskOffset, _Out_ PULONG ByteCount ) /*++ Routine Description: This routine looks through the mapping information for the file to find the logical diskoffset and number of bytes at that offset. We only deal with logical 2048 byte sectors here. If the mapping isn't present we will look it up on disk now. This routine assumes we are looking up a valid range in the file. This routine raises if it can't find mapping for the file offset. The Fcb may not be locked prior to calling this routine. We will always acquire it here. Arguments: Fcb - Fcb representing this stream. FileOffset - Lookup the allocation beginning at this point. DiskOffset - Address to store the logical disk offset. ByteCount - Address to store the number of contiguous bytes beginning at DiskOffset above. Return Value: None. --*/ { BOOLEAN FirstPass = TRUE; ULONG McbEntryOffset; PFCB ParentFcb = NULL; BOOLEAN CleanupParent = FALSE; BOOLEAN UnlockFcb = FALSE; LONGLONG CurrentFileOffset; ULONG CurrentMcbOffset; PCD_MCB_ENTRY CurrentMcbEntry; DIRENT_ENUM_CONTEXT DirContext = {0}; DIRENT Dirent = {0}; PAGED_CODE(); ASSERT_IRP_CONTEXT( IrpContext ); ASSERT_FCB( Fcb ); // // For DASD IO we already have clamped the read to the volume limits. // We'll allow reading beyond those limits for extended DASD IO, so // no MCB lookup here. // if (Fcb == Fcb->Vcb->VolumeDasdFcb) { *DiskOffset = FileOffset; return; } // // Use a try finally to facilitate cleanup. // _SEH2_TRY { // // We use a loop to perform the lookup. If we don't find the mapping in the // first pass then we look up all of the allocation and then look again. while (TRUE) { // // // Lookup the entry containing this file offset. // CdLockFcb( IrpContext, Fcb ); UnlockFcb = TRUE; McbEntryOffset = CdFindMcbEntry( IrpContext, Fcb, FileOffset ); // // If within the Mcb then we use the data out of this entry and are // done. // if (McbEntryOffset < Fcb->Mcb.CurrentEntryCount) { CdDiskOffsetFromMcbEntry( IrpContext, Fcb->Mcb.McbArray + McbEntryOffset, FileOffset, DiskOffset, ByteCount ); break; // // If this is not the first pass then the disk is corrupt. // } else if (!FirstPass) { CdRaiseStatus( IrpContext, STATUS_DISK_CORRUPT_ERROR ); } CdUnlockFcb( IrpContext, Fcb ); UnlockFcb = FALSE; // // Initialize the search dirent structures. // CdInitializeDirContext( IrpContext, &DirContext ); CdInitializeDirent( IrpContext, &Dirent ); // // Otherwise we need to walk the dirents for this file until we find // the one containing this entry. The parent Fcb should always be // present. // ParentFcb = Fcb->ParentFcb; CdAcquireFileShared( IrpContext, ParentFcb ); CleanupParent = TRUE; // // Do an unsafe test to see if we need to create a file object. // CdVerifyOrCreateDirStreamFile( IrpContext, ParentFcb); // // Initialize the local variables to indicate the first dirent // and lookup the first dirent. // CurrentFileOffset = 0; CurrentMcbOffset = 0; CdLookupDirent( IrpContext, ParentFcb, CdQueryFidDirentOffset( Fcb->FileId ), &DirContext ); // // If we are adding allocation to the Mcb then add all of it. // while (TRUE ) { // // Update the dirent from the on-disk dirent. // CdUpdateDirentFromRawDirent( IrpContext, ParentFcb, &DirContext, &Dirent ); // // Add this dirent to the Mcb if not already present. // CdLockFcb( IrpContext, Fcb ); UnlockFcb = TRUE; if (CurrentMcbOffset >= Fcb->Mcb.CurrentEntryCount) { CdAddAllocationFromDirent( IrpContext, Fcb, CurrentMcbOffset, CurrentFileOffset, &Dirent ); } CdUnlockFcb( IrpContext, Fcb ); UnlockFcb = FALSE; // // If this is the last dirent for the file then exit. // if (!FlagOn( Dirent.DirentFlags, CD_ATTRIBUTE_MULTI )) { break; } // // If we couldn't find another entry then the directory is corrupt because // the last dirent for a file doesn't exist. // if (!CdLookupNextDirent( IrpContext, ParentFcb, &DirContext, &DirContext )) { CdRaiseStatus( IrpContext, STATUS_DISK_CORRUPT_ERROR ); } // // Update our loop variables. // CurrentMcbEntry = Fcb->Mcb.McbArray + CurrentMcbOffset; CurrentFileOffset += CurrentMcbEntry->ByteCount; CurrentMcbOffset += 1; } // // All of the allocation is loaded. Go back and look up the mapping again. // It better be there this time. // FirstPass = FALSE; } } _SEH2_FINALLY { if (CleanupParent) { // // Release the parent and cleanup the dirent structures. // CdReleaseFile( IrpContext, ParentFcb ); CdCleanupDirContext( IrpContext, &DirContext ); CdCleanupDirent( IrpContext, &Dirent ); } if (UnlockFcb) { CdUnlockFcb( IrpContext, Fcb ); } } _SEH2_END; return; } VOID CdAddAllocationFromDirent ( _In_ PIRP_CONTEXT IrpContext, _Inout_ PFCB Fcb, _In_ ULONG McbEntryOffset, _In_ LONGLONG StartingFileOffset, _In_ PDIRENT Dirent ) /*++ Routine Description: This routine is called to add an entry into the Cd Mcb. We grow the Mcb as necessary and update the new entry. NOTE - The Fcb has already been locked prior to makeing this call. Arguments: Fcb - Fcb containing the Mcb to update. McbEntryOffset - Offset into the Mcb array to add this data. StartingFileOffset - Offset in bytes from the start of the file. Dirent - Dirent containing the on-disk data for this entry. Return Value: None --*/ { ULONG NewArraySize; PVOID NewMcbArray; PCD_MCB_ENTRY McbEntry; PAGED_CODE(); UNREFERENCED_PARAMETER( IrpContext ); ASSERT_IRP_CONTEXT( IrpContext ); ASSERT_FCB( Fcb ); ASSERT_LOCKED_FCB( Fcb ); // // If we need to grow the Mcb then do it now. // if (McbEntryOffset >= Fcb->Mcb.MaximumEntryCount) { // // Allocate a new buffer and copy the old data over. // NewArraySize = Fcb->Mcb.MaximumEntryCount * 2 * sizeof( CD_MCB_ENTRY ); NewMcbArray = FsRtlAllocatePoolWithTag( CdPagedPool, NewArraySize, TAG_MCB_ARRAY ); RtlZeroMemory( NewMcbArray, NewArraySize ); RtlCopyMemory( NewMcbArray, Fcb->Mcb.McbArray, Fcb->Mcb.MaximumEntryCount * sizeof( CD_MCB_ENTRY )); // // Deallocate the current array unless it is embedded in the Fcb. // if (Fcb->Mcb.MaximumEntryCount != 1) { CdFreePool( &Fcb->Mcb.McbArray ); } // // Now update the Mcb with the new array. // Fcb->Mcb.MaximumEntryCount *= 2; Fcb->Mcb.McbArray = NewMcbArray; } // // Update the new entry with the input data. // McbEntry = Fcb->Mcb.McbArray + McbEntryOffset; // // Start with the location and length on disk. // McbEntry->DiskOffset = LlBytesFromBlocks( Fcb->Vcb, Dirent->StartingOffset ); McbEntry->ByteCount = Dirent->DataLength; // // Round the byte count up to a logical block boundary if this is // the last extent. // if (!FlagOn( Dirent->DirentFlags, CD_ATTRIBUTE_MULTI )) { McbEntry->ByteCount = BlockAlign( Fcb->Vcb, McbEntry->ByteCount ); } // // The file offset is the logical position within this file. // We know this is correct regardless of whether we bias the // file size or disk offset. // McbEntry->FileOffset = StartingFileOffset; // // Convert the interleave information from logical blocks to // bytes. // if (Dirent->FileUnitSize != 0) { McbEntry->DataBlockByteCount = LlBytesFromBlocks( Fcb->Vcb, Dirent->FileUnitSize ); McbEntry->TotalBlockByteCount = McbEntry->DataBlockByteCount + LlBytesFromBlocks( Fcb->Vcb, Dirent->InterleaveGapSize ); // // If the file is not interleaved then the size of the data block // and total block are the same as the byte count. // } else { McbEntry->DataBlockByteCount = McbEntry->TotalBlockByteCount = McbEntry->ByteCount; } // // Update the number of entries in the Mcb. The Mcb is never sparse // so whenever we add an entry it becomes the last entry in the Mcb. // Fcb->Mcb.CurrentEntryCount = McbEntryOffset + 1; return; } VOID CdAddInitialAllocation ( _In_ PIRP_CONTEXT IrpContext, _Inout_ PFCB Fcb, _In_ ULONG StartingBlock, _In_ LONGLONG DataLength ) /*++ Routine Description: This routine is called to set up the initial entry in an Mcb. This routine handles the single initial entry for a directory file. We will round the start block down to a sector boundary. Our caller has already biased the DataLength with any adjustments. This is used for the case where there is a single entry and we want to align the data on a sector boundary. Arguments: Fcb - Fcb containing the Mcb to update. StartingBlock - Starting logical block for this directory. This is the start of the actual data. We will bias this by the sector offset of the data. DataLength - Length of the data. Return Value: None --*/ { PCD_MCB_ENTRY McbEntry; PAGED_CODE(); UNREFERENCED_PARAMETER( IrpContext ); ASSERT_IRP_CONTEXT( IrpContext ); ASSERT_FCB( Fcb ); ASSERT_LOCKED_FCB( Fcb ); NT_ASSERT( 0 == Fcb->Mcb.CurrentEntryCount); NT_ASSERT( CDFS_NTC_FCB_DATA != Fcb->NodeTypeCode); // // Update the new entry with the input data. // McbEntry = Fcb->Mcb.McbArray; // // Start with the location and length on disk. // McbEntry->DiskOffset = LlBytesFromBlocks( Fcb->Vcb, StartingBlock ); McbEntry->DiskOffset -= Fcb->StreamOffset; McbEntry->ByteCount = DataLength; // // The file offset is the logical position within this file. // We know this is correct regardless of whether we bias the // file size or disk offset. // McbEntry->FileOffset = 0; // // If the file is not interleaved then the size of the data block // and total block are the same as the byte count. // McbEntry->DataBlockByteCount = McbEntry->TotalBlockByteCount = McbEntry->ByteCount; // // Update the number of entries in the Mcb. The Mcb is never sparse // so whenever we add an entry it becomes the last entry in the Mcb. // Fcb->Mcb.CurrentEntryCount = 1; return; } VOID CdTruncateAllocation ( _In_ PIRP_CONTEXT IrpContext, _Inout_ PFCB Fcb, _In_ LONGLONG StartingFileOffset ) /*++ Routine Description: This routine truncates the Mcb for a file by eliminating all of the Mcb entries from the entry which contains the given offset. The Fcb should be locked when this routine is called. Arguments: Fcb - Fcb containing the Mcb to truncate. StartingFileOffset - Offset in the file to truncate the Mcb from. Return Value: None --*/ { ULONG McbEntryOffset; PAGED_CODE(); ASSERT_IRP_CONTEXT( IrpContext ); ASSERT_FCB( Fcb ); ASSERT_LOCKED_FCB( Fcb ); // // Find the entry containg this starting offset. // McbEntryOffset = CdFindMcbEntry( IrpContext, Fcb, StartingFileOffset ); // // Now set the current size of the mcb to this point. // Fcb->Mcb.CurrentEntryCount = McbEntryOffset; return; } _At_(Fcb->NodeByteSize, _In_range_(>=, FIELD_OFFSET( FCB, FcbType ))) VOID CdInitializeMcb ( _In_ PIRP_CONTEXT IrpContext, _Inout_updates_bytes_(Fcb->NodeByteSize) PFCB Fcb ) /*++ Routine Description: This routine is called to initialize the Mcb in an Fcb. We initialize this with an entry count of one and point to the entry in the Fcb itself. Fcb should be acquired exclusively when this is called. Arguments: Fcb - Fcb containing the Mcb to initialize. Return Value: None --*/ { PAGED_CODE(); UNREFERENCED_PARAMETER( IrpContext ); ASSERT_IRP_CONTEXT( IrpContext ); ASSERT_FCB( Fcb ); // // Set the entry counts to show there is one entry in the array and // it is unused. // Fcb->Mcb.MaximumEntryCount = 1; Fcb->Mcb.CurrentEntryCount = 0; Fcb->Mcb.McbArray = &Fcb->McbEntry; return; } _At_(Fcb->NodeByteSize, _In_range_(>=, FIELD_OFFSET( FCB, FcbType ))) _When_(Fcb->NodeTypeCode == CDFS_NTC_FCB_PATH_TABLE, _At_(Fcb->NodeByteSize, _In_range_(==, SIZEOF_FCB_INDEX))) _When_(Fcb->NodeTypeCode == CDFS_NTC_FCB_INDEX, _At_(Fcb->NodeByteSize, _In_range_(==, SIZEOF_FCB_INDEX))) _When_(Fcb->NodeTypeCode == CDFS_NTC_FCB_DATA, _At_(Fcb->NodeByteSize, _In_range_(==, SIZEOF_FCB_DATA))) VOID CdUninitializeMcb ( _In_ PIRP_CONTEXT IrpContext, _Inout_updates_bytes_(Fcb->NodeByteSize) PFCB Fcb ) /*++ Routine Description: This routine is called to cleanup an Mcb in an Fcb. We look at the maximum run count in the Fcb and if greater than one we will deallocate the buffer. Fcb should be acquired exclusively when this is called. Arguments: Fcb - Fcb containing the Mcb to uninitialize. Return Value: None --*/ { PAGED_CODE(); UNREFERENCED_PARAMETER( IrpContext ); ASSERT_IRP_CONTEXT( IrpContext ); ASSERT_FCB( Fcb ); // // If the count is greater than one then this is an allocated buffer. // if (Fcb->Mcb.MaximumEntryCount > 1) { CdFreePool( &Fcb->Mcb.McbArray ); } return; } // // Local suupport routine // ULONG CdFindMcbEntry ( _In_ PIRP_CONTEXT IrpContext, _In_ PFCB Fcb, _In_ LONGLONG FileOffset ) /*++ Routine Description: This routine is called to find the Mcb entry which contains the file offset at the given point. If the file offset is not currently in the Mcb then we return the offset of the entry to add. Fcb should be locked when this is called. Arguments: Fcb - Fcb containing the Mcb to uninitialize. FileOffset - Return the Mcb entry which contains this file offset. Return Value: ULONG - Offset in the Mcb of the entry for this offset. --*/ { ULONG CurrentMcbOffset; PCD_MCB_ENTRY CurrentMcbEntry; PAGED_CODE(); UNREFERENCED_PARAMETER( IrpContext ); ASSERT_IRP_CONTEXT( IrpContext ); ASSERT_FCB( Fcb ); ASSERT_LOCKED_FCB( Fcb ); // // We expect a linear search will be sufficient here. // CurrentMcbOffset = 0; CurrentMcbEntry = Fcb->Mcb.McbArray; while (CurrentMcbOffset < Fcb->Mcb.CurrentEntryCount) { // // Check if the offset lies within the current Mcb position. // if (FileOffset < CurrentMcbEntry->FileOffset + CurrentMcbEntry->ByteCount) { break; } // // Move to the next entry. // CurrentMcbOffset += 1; CurrentMcbEntry += 1; } // // This is the offset containing this file offset (or the point // where an entry should be added). // return CurrentMcbOffset; } // // Local support routine // VOID CdDiskOffsetFromMcbEntry ( _In_ PIRP_CONTEXT IrpContext, _In_ PCD_MCB_ENTRY McbEntry, _In_ LONGLONG FileOffset, _Out_ PLONGLONG DiskOffset, _Out_ PULONG ByteCount ) /*++ Routine Description: This routine is called to return the diskoffset and length of the file data which begins at offset 'FileOffset'. We have the Mcb entry which contains the mapping and interleave information. NOTE - This routine deals with data in 2048 byte logical sectors. If this is an XA file then our caller has already converted from 'raw' file bytes to 'cooked' file bytes. Arguments: McbEntry - Entry in the Mcb containing the allocation information. FileOffset - Starting Offset in the file to find the matching disk offsets. DiskOffset - Address to store the starting disk offset for this operation. ByteCount - Address to store number of contiguous bytes starting at this disk offset. Return Value: None --*/ { LONGLONG ExtentOffset; LONGLONG CurrentDiskOffset; LONGLONG CurrentExtentOffset; LONGLONG LocalByteCount; PAGED_CODE(); UNREFERENCED_PARAMETER( IrpContext ); ASSERT_IRP_CONTEXT( IrpContext ); // // Extent offset is the difference between the file offset and the start // of the extent. // ExtentOffset = FileOffset - McbEntry->FileOffset; // // Optimize the non-interleave case. // if (McbEntry->ByteCount == McbEntry->DataBlockByteCount) { *DiskOffset = McbEntry->DiskOffset + ExtentOffset; LocalByteCount = McbEntry->ByteCount - ExtentOffset; } else { // // Walk though any interleave until we reach the current offset in // this extent. // CurrentExtentOffset = McbEntry->DataBlockByteCount; CurrentDiskOffset = McbEntry->DiskOffset; while (CurrentExtentOffset <= ExtentOffset) { CurrentDiskOffset += McbEntry->TotalBlockByteCount; CurrentExtentOffset += McbEntry->DataBlockByteCount; } // // We are now positioned at the data block containing the starting // file offset we were given. The disk offset is the offset of // the start of this block plus the extent offset into this block. // The byte count is the data block byte count minus our offset into // this block. // *DiskOffset = CurrentDiskOffset + (ExtentOffset + McbEntry->DataBlockByteCount - CurrentExtentOffset); // // Make sure we aren't past the end of the data length. This is possible // if we only use part of the last data block on an interleaved file. // if (CurrentExtentOffset > McbEntry->ByteCount) { CurrentExtentOffset = McbEntry->ByteCount; } LocalByteCount = CurrentExtentOffset - ExtentOffset; } // // If the byte count exceeds our limit then cut it to fit in 32 bits. // if (LocalByteCount > MAXULONG) { *ByteCount = MAXULONG; } else { *ByteCount = (ULONG) LocalByteCount; } return; }