Root/Documentation/lzo.txt

1
2LZO stream format as understood by Linux's LZO decompressor
3===========================================================
4
5Introduction
6
7  This is not a specification. No specification seems to be publicly available
8  for the LZO stream format. This document describes what input format the LZO
9  decompressor as implemented in the Linux kernel understands. The file subject
10  of this analysis is lib/lzo/lzo1x_decompress_safe.c. No analysis was made on
11  the compressor nor on any other implementations though it seems likely that
12  the format matches the standard one. The purpose of this document is to
13  better understand what the code does in order to propose more efficient fixes
14  for future bug reports.
15
16Description
17
18  The stream is composed of a series of instructions, operands, and data. The
19  instructions consist in a few bits representing an opcode, and bits forming
20  the operands for the instruction, whose size and position depend on the
21  opcode and on the number of literals copied by previous instruction. The
22  operands are used to indicate :
23
24    - a distance when copying data from the dictionary (past output buffer)
25    - a length (number of bytes to copy from dictionary)
26    - the number of literals to copy, which is retained in variable "state"
27      as a piece of information for next instructions.
28
29  Optionally depending on the opcode and operands, extra data may follow. These
30  extra data can be a complement for the operand (eg: a length or a distance
31  encoded on larger values), or a literal to be copied to the output buffer.
32
33  The first byte of the block follows a different encoding from other bytes, it
34  seems to be optimized for literal use only, since there is no dictionary yet
35  prior to that byte.
36
37  Lengths are always encoded on a variable size starting with a small number
38  of bits in the operand. If the number of bits isn't enough to represent the
39  length, up to 255 may be added in increments by consuming more bytes with a
40  rate of at most 255 per extra byte (thus the compression ratio cannot exceed
41  around 255:1). The variable length encoding using #bits is always the same :
42
43       length = byte & ((1 << #bits) - 1)
44       if (!length) {
45               length = ((1 << #bits) - 1)
46               length += 255*(number of zero bytes)
47               length += first-non-zero-byte
48       }
49       length += constant (generally 2 or 3)
50
51  For references to the dictionary, distances are relative to the output
52  pointer. Distances are encoded using very few bits belonging to certain
53  ranges, resulting in multiple copy instructions using different encodings.
54  Certain encodings involve one extra byte, others involve two extra bytes
55  forming a little-endian 16-bit quantity (marked LE16 below).
56
57  After any instruction except the large literal copy, 0, 1, 2 or 3 literals
58  are copied before starting the next instruction. The number of literals that
59  were copied may change the meaning and behaviour of the next instruction. In
60  practice, only one instruction needs to know whether 0, less than 4, or more
61  literals were copied. This is the information stored in the <state> variable
62  in this implementation. This number of immediate literals to be copied is
63  generally encoded in the last two bits of the instruction but may also be
64  taken from the last two bits of an extra operand (eg: distance).
65
66  End of stream is declared when a block copy of distance 0 is seen. Only one
67  instruction may encode this distance (0001HLLL), it takes one LE16 operand
68  for the distance, thus requiring 3 bytes.
69
70  IMPORTANT NOTE : in the code some length checks are missing because certain
71  instructions are called under the assumption that a certain number of bytes
72  follow because it has already been garanteed before parsing the instructions.
73  They just have to "refill" this credit if they consume extra bytes. This is
74  an implementation design choice independant on the algorithm or encoding.
75
76Byte sequences
77
78  First byte encoding :
79
80      0..17 : follow regular instruction encoding, see below. It is worth
81                noting that codes 16 and 17 will represent a block copy from
82                the dictionary which is empty, and that they will always be
83                invalid at this place.
84
85      18..21 : copy 0..3 literals
86                state = (byte - 17) = 0..3 [ copy <state> literals ]
87                skip byte
88
89      22..255 : copy literal string
90                length = (byte - 17) = 4..238
91                state = 4 [ don't copy extra literals ]
92                skip byte
93
94  Instruction encoding :
95
96      0 0 0 0 X X X X (0..15)
97        Depends on the number of literals copied by the last instruction.
98        If last instruction did not copy any literal (state == 0), this
99        encoding will be a copy of 4 or more literal, and must be interpreted
100        like this :
101
102           0 0 0 0 L L L L (0..15) : copy long literal string
103           length = 3 + (L ?: 15 + (zero_bytes * 255) + non_zero_byte)
104           state = 4 (no extra literals are copied)
105
106        If last instruction used to copy between 1 to 3 literals (encoded in
107        the instruction's opcode or distance), the instruction is a copy of a
108        2-byte block from the dictionary within a 1kB distance. It is worth
109        noting that this instruction provides little savings since it uses 2
110        bytes to encode a copy of 2 other bytes but it encodes the number of
111        following literals for free. It must be interpreted like this :
112
113           0 0 0 0 D D S S (0..15) : copy 2 bytes from <= 1kB distance
114           length = 2
115           state = S (copy S literals after this block)
116         Always followed by exactly one byte : H H H H H H H H
117           distance = (H << 2) + D + 1
118
119        If last instruction used to copy 4 or more literals (as detected by
120        state == 4), the instruction becomes a copy of a 3-byte block from the
121        dictionary from a 2..3kB distance, and must be interpreted like this :
122
123           0 0 0 0 D D S S (0..15) : copy 3 bytes from 2..3 kB distance
124           length = 3
125           state = S (copy S literals after this block)
126         Always followed by exactly one byte : H H H H H H H H
127           distance = (H << 2) + D + 2049
128
129      0 0 0 1 H L L L (16..31)
130           Copy of a block within 16..48kB distance (preferably less than 10B)
131           length = 2 + (L ?: 7 + (zero_bytes * 255) + non_zero_byte)
132        Always followed by exactly one LE16 : D D D D D D D D : D D D D D D S S
133           distance = 16384 + (H << 14) + D
134           state = S (copy S literals after this block)
135           End of stream is reached if distance == 16384
136
137      0 0 1 L L L L L (32..63)
138           Copy of small block within 16kB distance (preferably less than 34B)
139           length = 2 + (L ?: 31 + (zero_bytes * 255) + non_zero_byte)
140        Always followed by exactly one LE16 : D D D D D D D D : D D D D D D S S
141           distance = D + 1
142           state = S (copy S literals after this block)
143
144      0 1 L D D D S S (64..127)
145           Copy 3-4 bytes from block within 2kB distance
146           state = S (copy S literals after this block)
147           length = 3 + L
148         Always followed by exactly one byte : H H H H H H H H
149           distance = (H << 3) + D + 1
150
151      1 L L D D D S S (128..255)
152           Copy 5-8 bytes from block within 2kB distance
153           state = S (copy S literals after this block)
154           length = 5 + L
155         Always followed by exactly one byte : H H H H H H H H
156           distance = (H << 3) + D + 1
157
158Authors
159
160  This document was written by Willy Tarreau <w@1wt.eu> on 2014/07/19 during an
161  analysis of the decompression code available in Linux 3.16-rc5. The code is
162  tricky, it is possible that this document contains mistakes or that a few
163  corner cases were overlooked. In any case, please report any doubt, fix, or
164  proposed updates to the author(s) so that the document can be updated.
165

Archive Download this file



interactive