Leaked source code of windows server 2003
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

454 lines
13 KiB

title "Compression and Decompression Engines"
;++
;
; Copyright (c) 1989 Microsoft Corporation
;
; Module Name:
;
; lzntx86.asm
;
; Abstract:
;
; This module implements the compression and decompression engines needed
; to support file system compression. Functions are provided to
; compress a buffer and decompress a buffer.
;
; Author:
;
; Mark Zbikowski (markz) 15-Mar-1994
;
; Environment:
;
; Any mode.
;
; Revision History:
;
; 15-Mar-1994 markz
;
; 386 version created
;
;--
.386p
.xlist
include ks386.inc
include callconv.inc ; calling convention macros
.list
IFDEF NTOS_KERNEL_RUNTIME
_PAGE SEGMENT DWORD PUBLIC 'CODE'
ELSE
_TEXT SEGMENT DWORD PUBLIC 'CODE'
ENDIF
ASSUME DS:FLAT, ES:FLAT, SS:NOTHING, FS:NOTHING, GS:NOTHING
page
subttl "Decompress a buffer"
;++
;
; NTSTATUS
; LZNT1DecompressChunk (
; OUT PUCHAR UncompressedBuffer,
; IN PUCHAR EndOfUncompressedBufferPlus1,
; IN PUCHAR CompressedBuffer,
; IN PUCHAR EndOfCompressedBufferPlus1,
; OUT PULONG FinalUncompressedChunkSize
; )
;
; Routine Description:
;
; This function decodes a stream of compression tokens and places the
; resultant output into the destination buffer. The format of the input
; is described ..\lznt1.c. As the input is decoded, checks are made to
; ensure that no data is read past the end of the compressed input buffer
; and that no data is stored past the end of the output buffer. Violations
; indicate corrupt input and are indicated by a status return.
;
; The following code takes advantage of two distinct observations.
; First, literal tokens occur at least twice as often as copy tokens.
; This argues for having a "fall-through" being the case where a literal
; token is found. We structure the main decomposition loop in eight
; pieces where the first piece is a sequence of literal-test fall-throughs
; and the remainder are a copy token followed by 7,6,...,0 literal-test
; fall-throughs. Each test examines a particular bit in the tag byte
; and jumps to the relevant code piece.
;
; The second observation involves performing bounds checking only
; when needed. Bounds checking the compressed buffer need only be done
; when fetching the tag byte. If there is not enough room left in the
; input for a tag byte and 8 (worst case) copy tokens, a branch is made
; to a second loop that handles a byte-by-byte "safe" copy to finish
; up the decompression. Similarly, at the head of the loop a check is
; made to ensure that there is enough room in the output buffer for 8
; literal bytes. If not enough room is left, then the second loop is
; used. Finally, after performing each copy, the output-buffer check
; is made as well since a copy may take the destination pointer
; arbitrarily close to the end of the destination.
;
; The register conventions used in the loops below are:
;
; (al) contains the current tag byte
; (ebx) contains the current width in bits of the length given
; the maximum offset
; that can be utilized in a copy token. We update this
; value only prior to performing a copy. This width is used
; both to index a mask table (for extracting the length) as
; well as shifting (for extracting the copy offset)
; (ecx) is used to contain counts during copies
; (edx) is used as a temp variable during copies
; (esi) is used mainly as the source of the next compressed token.
; It is also used for copies.
; (edi) is used as the destination of literals and copies
; (ebp) is used as a frame pointer
;
; Arguments:
;
; UncompressedBuffer (ebp+8) - pointer to destination of uncompression.
;
; EndOfUncompressedBufferPlus1 (ebp+12) - pointer just beyond the
; output buffer. This is used for consistency checking of the stored
; compressed data.
;
; CompressedBuffer (ebp+16) - pointer to compressed source. This pointer
; has been adjusted by the caller to point past the header word, so
; the pointer points to the first tag byte describing which of the
; following tokens are literals and which are copy groups.
;
; EndOfCompressedBufferPlus1 (ebp+20) - pointer just beyond end of input
; buffer. This is used to terminate the decompression.
;
; FinalUncompressedChunkSize (ebp+24) - pointer to a returned decompressed
; size. This has meaningful data ONLY when LZNT1DecompressChunk returns
; STATUS_SUCCESS
;
; Return Value:
;
; STATUS_SUCCESS is returned only if the decompression consumes thee entire
; input buffer and does not exceed the output buffer.
; STATUS_BAD_COMPRESSION_BUFFER is returned when the output buffer would be
; overflowed.
;
;--
; Decompression macros
;** TestLiteralAt - tests to see if there's a literal at a specific
; bit position. If so, it branches to the appropriate copy code
; (decorated by the bit being used).
;
; This code does no bounds checking
TestLiteralAt macro CopyLabel,bit,IsMain
test al,1 SHL bit ; is there a copy token at this position?
jnz CopyLabel&bit ; yes, go copy it
mov dl,[esi+bit+1] ; (dl) = literal byte from compressed stream
ifidn <IsMain>,<Y>
mov [edi+bit],dl ; store literal byte
else
mov [edi],dl ; store literal byte
inc edi ; point to next literal
endif
endm
; Jump - allow specific jumps with computed labels.
Jump macro lab,tag
jmp lab&tag
endm
;** DoCopy - perform a copy. If a bit position is specified
; then branch to the appropriate point in the "safe" tail when
; the copy takes us too close to the end of the output buffer
;
; This code checks the bounds of the copy token: copying before the
; beginning of the buffer and copying beyond the end of the buffer.
DoCopy macro AdjustLabel,bit,IsMain
ifidn <IsMain>,<Y>
if bit ne 0
add edi,bit
endif
endif
Test&AdjustLabel&bit:
cmp edi,WidthBoundary
ja Adjust&AdjustLabel&bit
xor ecx,ecx
mov cx,word ptr [esi+bit+1] ; (ecx) = encoded length:offset
lea edx,[esi+1] ; (edx) = next token location
mov Temp,edx
mov esi,ecx ; (esi) = encoded length:offset
and ecx,MaskTab[ebx*4] ; (ecx) = length
xchg ebx,ecx ; (ebx) = length/(ecx) = width
shr esi,cl ; (esi) = offset
xchg ebx,ecx ; (ebx) = width, (ecx) = length
neg esi ; (esi) = negative real offset
lea esi,[esi+edi-1] ; (esi) = pointer to previous string
cmp esi,UncompressedBuffer ; off front of buffer?
jb DOA ; yes, error
add ecx,3 ; (ecx) = real length
lea edx,[edi+ecx] ; (edx) = end of copy
ifidn <IsMain>,<Y>
cmp edx,EndOfSpecialDest ; do we exceed buffer?
jae TailAdd&bit ; yes, handle in safe tail
else
cmp edx,EndOfUncompressedBufferPlus1
; do we exceed buffer?
ja DOA ; yes, error
endif
rep movsb ; Copy the bytes
mov esi,Temp ; (esi) = next token location
ifidn <IsMain>,<Y>
sub edi,bit+1
endif
endm
;** AdjustWidth - adjust width of length based upon current position of
; input buffer (max offset)
AdjustWidth macro l,i
Adjust&l&i:
dec ebx ; (ebx) = new width pointer
mov edx,UncompressedBuffer ; (edx) = pointer to dest buffer
add edx,WidthTab[ebx*4] ; (edx) = new width boundary
mov WidthBoundary,edx ; save boundary for comparison
jmp Test&l&i
endm
;** GenerateBlock - generates the unsafe block of copy/literal pieces.
;
; This code does no checking for simple input/output checking. Only
; the data referred to by the copy tokens is checked.
GenerateBlock macro bit
Copy&bit:
DoCopy Body,bit,Y
j = bit + 1
while j lt 8
TestLiteralAt Copy,%(j),Y
j = j + 1
endm
add esi,9
add edi,8
jmp Top
AdjustWidth Body,bit
endm
;** GenerateTailBlock - generates safe tail block for compression. This
; code checks everything before each byte stored so it is expected
; to be executed only at the end of the buffer.
GenerateTailBlock macro bit
TailAdd&bit:
add EndOfCompressedBufferPlus1,1+2*8
; restore buffer length to true length
mov esi,Temp ; (esi) = source of copy token block
dec esi
Tail&bit:
lea ecx,[esi+bit+1] ; (ecx) = source of next token
cmp ecx,EndOfCompressedBufferPlus1 ; are we done?
jz Done ; yes - we exactly match end of buffer
; ja DOA ; INTERNAL ERROR only
cmp edi,EndOfUncompressedBufferPlus1
jz Done ; go quit, destination is full
; ja DOA ; INTERNAL ERROR only
TestLiteralAt TailCopy,bit,N
Jump Tail,%(bit+1)
; We expect a copy token to be at [esi+bit+1]. This means that
; esi+bit+1+tokensize must be <= EndOfCompressedBufferPlus1
TailCopy&bit:
lea ecx,[esi+bit+3] ; (ecx) = next input position
cmp ecx,EndOfCompressedBufferPlus1 ; do we go too far
ja DOA ; yes, we are beyond the end of buffer
DoCopy Tail,bit,N ; perform copy
Jump Tail,%(bit+1)
AdjustWidth Tail,bit
endm
cPublicProc _LZNT1DecompressChunk ,5
push ebp ; (tos) = saved frame pointer
mov ebp,esp ; (ebp) = frame pointer to arguments
sub esp,12 ; Open up room for locals
Temp equ dword ptr [ebp-12]
WidthBoundary equ dword ptr [ebp-8]
EndOfSpecialDest equ dword ptr [ebp-4]
;SavedEBP equ dword ptr [ebp]
;ReturnAddress equ dword ptr [ebp+4]
UncompressedBuffer equ dword ptr [ebp+8]
EndOfUncompressedBufferPlus1 equ dword ptr [ebp+12]
CompressedBuffer equ dword ptr [ebp+16]
EndOfCompressedBufferPlus1 equ dword ptr [ebp+20]
FinalUncompressedChunkSize equ dword ptr [ebp+24]
push ebx
push esi
push edi
mov edi,UncompressedBuffer ; (edi) = destination of decompress
mov esi,CompressedBuffer ; (esi) = header
sub EndOfCompressedBufferPlus1,1+2*8 ; make room for special source
mov eax,EndOfUncompressedBufferPlus1 ; (eax) = end of destination
sub eax,8 ; (eax) = beginning of special tail
mov EndOfSpecialDest,eax ; store special tail
mov WidthBoundary,edi ; force initial width mismatch
mov ebx,13 ; initial width of output
Top: cmp esi,EndOfCompressedBufferPlus1 ; Will this be the last tag group in source?
jae DoTail ; yes, go handle specially
cmp edi,EndOfSpecialDest ; are we too close to end of buffer?
jae DoTail ; yes, go skip to end
mov al,byte ptr [esi] ; (al) = tag byte, (esi) points to token
irpc i,<01234567>
TestLiteralAt Copy,%(i),Y
endm
add esi,9
add edi,8
jmp Top
; ; Width of offset Width of length
WidthTab dd 0FFFFh ; 16 0
dd 0FFFFh ; 15 1
dd 0FFFFh ; 14 2
dd 0FFFFh ; 13 3
dd 0FFFFh ; 12 4
dd 2048 ; 11 5
dd 1024 ; 10 6
dd 512 ; 9 7
dd 256 ; 8 8
dd 128 ; 7 9
dd 64 ; 6 10
dd 32 ; 5 11
dd 16 ; 4 12
dd 0 ; 3 13
dd 0 ; 2 14
dd 0 ; 1 15
dd 0 ; 0 16
; ;
MaskTab dd 0000000000000000b ; 0
dd 0000000000000001b ; 1
dd 0000000000000011b ; 2
dd 0000000000000111b ; 3
dd 0000000000001111b ; 4
dd 0000000000011111b ; 5
dd 0000000000111111b ; 6
dd 0000000001111111b ; 7
dd 0000000011111111b ; 8
dd 0000000111111111b ; 9
dd 0000001111111111b ; 10
dd 0000011111111111b ; 11
dd 0000111111111111b ; 12
dd 0001111111111111b ; 13
dd 0011111111111111b ; 14
dd 0111111111111111b ; 15
dd 1111111111111111b ; 16
irpc i,<01234567>
GenerateBlock %(i)
endm
; We're handling a tail specially for this. We must check at all
; spots for running out of input as well as overflowing output.
;
; (esi) = pointer to possible next tag
DoTail: add EndOfCompressedBufferPlus1,1+2*8 ; point to end of compressed input
TailLoop:
cmp esi,EndOfCompressedBufferPlus1 ; are we totally done?
jz Done ; yes, go return
mov al,byte ptr [esi] ; (al) = tag byte
jmp Tail0
irpc i,<01234567>
GenerateTailBlock i
endm
Tail8: add esi,9
jmp TailLoop
DOA: mov eax,STATUS_BAD_COMPRESSION_BUFFER
jmp Final
Done: mov eax,edi ; (eax) = pointer to next byte to store
sub eax,UncompressedBuffer ; (eax) = length of uncompressed
mov edi,FinalUncompressedChunkSize ; (edi) = user return value location
mov [edi],eax ; return total transfer size to user
xor eax,eax ; (eax) = STATUS_SUCCESS
Final: pop edi
pop esi
pop ebx
mov esp,ebp
pop ebp
stdRET _LZNT1DecompressChunk
stdENDP _LZNT1DecompressChunk
IFDEF NTOS_KERNEL_RUNTIME
_PAGE ENDS
ELSE
_TEXT ENDS
ENDIF
end