brintos

brintos / linux-shallow public Read only

0
0
Text · 9.4 KiB · f65b515 Raw
203 lines · plain
1===========================================================2LZO stream format as understood by Linux's LZO decompressor3===========================================================4 5Introduction6============7 8  This is not a specification. No specification seems to be publicly available9  for the LZO stream format. This document describes what input format the LZO10  decompressor as implemented in the Linux kernel understands. The file subject11  of this analysis is lib/lzo/lzo1x_decompress_safe.c. No analysis was made on12  the compressor nor on any other implementations though it seems likely that13  the format matches the standard one. The purpose of this document is to14  better understand what the code does in order to propose more efficient fixes15  for future bug reports.16 17Description18===========19 20  The stream is composed of a series of instructions, operands, and data. The21  instructions consist in a few bits representing an opcode, and bits forming22  the operands for the instruction, whose size and position depend on the23  opcode and on the number of literals copied by previous instruction. The24  operands are used to indicate:25 26    - a distance when copying data from the dictionary (past output buffer)27    - a length (number of bytes to copy from dictionary)28    - the number of literals to copy, which is retained in variable "state"29      as a piece of information for next instructions.30 31  Optionally depending on the opcode and operands, extra data may follow. These32  extra data can be a complement for the operand (eg: a length or a distance33  encoded on larger values), or a literal to be copied to the output buffer.34 35  The first byte of the block follows a different encoding from other bytes, it36  seems to be optimized for literal use only, since there is no dictionary yet37  prior to that byte.38 39  Lengths are always encoded on a variable size starting with a small number40  of bits in the operand. If the number of bits isn't enough to represent the41  length, up to 255 may be added in increments by consuming more bytes with a42  rate of at most 255 per extra byte (thus the compression ratio cannot exceed43  around 255:1). The variable length encoding using #bits is always the same::44 45       length = byte & ((1 << #bits) - 1)46       if (!length) {47               length = ((1 << #bits) - 1)48               length += 255*(number of zero bytes)49               length += first-non-zero-byte50       }51       length += constant (generally 2 or 3)52 53  For references to the dictionary, distances are relative to the output54  pointer. Distances are encoded using very few bits belonging to certain55  ranges, resulting in multiple copy instructions using different encodings.56  Certain encodings involve one extra byte, others involve two extra bytes57  forming a little-endian 16-bit quantity (marked LE16 below).58 59  After any instruction except the large literal copy, 0, 1, 2 or 3 literals60  are copied before starting the next instruction. The number of literals that61  were copied may change the meaning and behaviour of the next instruction. In62  practice, only one instruction needs to know whether 0, less than 4, or more63  literals were copied. This is the information stored in the <state> variable64  in this implementation. This number of immediate literals to be copied is65  generally encoded in the last two bits of the instruction but may also be66  taken from the last two bits of an extra operand (eg: distance).67 68  End of stream is declared when a block copy of distance 0 is seen. Only one69  instruction may encode this distance (0001HLLL), it takes one LE16 operand70  for the distance, thus requiring 3 bytes.71 72  .. important::73 74     In the code some length checks are missing because certain instructions75     are called under the assumption that a certain number of bytes follow76     because it has already been guaranteed before parsing the instructions.77     They just have to "refill" this credit if they consume extra bytes. This78     is an implementation design choice independent on the algorithm or79     encoding.80 81Versions82 830: Original version841: LZO-RLE85 86Version 1 of LZO implements an extension to encode runs of zeros using run87length encoding. This improves speed for data with many zeros, which is a88common case for zram. This modifies the bitstream in a backwards compatible way89(v1 can correctly decompress v0 compressed data, but v0 cannot read v1 data).90 91For maximum compatibility, both versions are available under different names92(lzo and lzo-rle). Differences in the encoding are noted in this document with93e.g.: version 1 only.94 95Byte sequences96==============97 98  First byte encoding::99 100      0..16   : follow regular instruction encoding, see below. It is worth101                noting that code 16 will represent a block copy from the102                dictionary which is empty, and that it will always be103                invalid at this place.104 105      17      : bitstream version. If the first byte is 17, and compressed106                stream length is at least 5 bytes (length of shortest possible107                versioned bitstream), the next byte gives the bitstream version108                (version 1 only).109                Otherwise, the bitstream version is 0.110 111      18..21  : copy 0..3 literals112                state = (byte - 17) = 0..3  [ copy <state> literals ]113                skip byte114 115      22..255 : copy literal string116                length = (byte - 17) = 4..238117                state = 4 [ don't copy extra literals ]118                skip byte119 120  Instruction encoding::121 122      0 0 0 0 X X X X  (0..15)123        Depends on the number of literals copied by the last instruction.124        If last instruction did not copy any literal (state == 0), this125        encoding will be a copy of 4 or more literal, and must be interpreted126        like this :127 128           0 0 0 0 L L L L  (0..15)  : copy long literal string129           length = 3 + (L ?: 15 + (zero_bytes * 255) + non_zero_byte)130           state = 4  (no extra literals are copied)131 132        If last instruction used to copy between 1 to 3 literals (encoded in133        the instruction's opcode or distance), the instruction is a copy of a134        2-byte block from the dictionary within a 1kB distance. It is worth135        noting that this instruction provides little savings since it uses 2136        bytes to encode a copy of 2 other bytes but it encodes the number of137        following literals for free. It must be interpreted like this :138 139           0 0 0 0 D D S S  (0..15)  : copy 2 bytes from <= 1kB distance140           length = 2141           state = S (copy S literals after this block)142         Always followed by exactly one byte : H H H H H H H H143           distance = (H << 2) + D + 1144 145        If last instruction used to copy 4 or more literals (as detected by146        state == 4), the instruction becomes a copy of a 3-byte block from the147        dictionary from a 2..3kB distance, and must be interpreted like this :148 149           0 0 0 0 D D S S  (0..15)  : copy 3 bytes from 2..3 kB distance150           length = 3151           state = S (copy S literals after this block)152         Always followed by exactly one byte : H H H H H H H H153           distance = (H << 2) + D + 2049154 155      0 0 0 1 H L L L  (16..31)156           Copy of a block within 16..48kB distance (preferably less than 10B)157           length = 2 + (L ?: 7 + (zero_bytes * 255) + non_zero_byte)158        Always followed by exactly one LE16 :  D D D D D D D D : D D D D D D S S159           distance = 16384 + (H << 14) + D160           state = S (copy S literals after this block)161           End of stream is reached if distance == 16384162           In version 1 only, to prevent ambiguity with the RLE case when163           ((distance & 0x803f) == 0x803f) && (261 <= length <= 264), the164           compressor must not emit block copies where distance and length165           meet these conditions.166 167        In version 1 only, this instruction is also used to encode a run of168           zeros if distance = 0xbfff, i.e. H = 1 and the D bits are all 1.169           In this case, it is followed by a fourth byte, X.170           run length = ((X << 3) | (0 0 0 0 0 L L L)) + 4171 172      0 0 1 L L L L L  (32..63)173           Copy of small block within 16kB distance (preferably less than 34B)174           length = 2 + (L ?: 31 + (zero_bytes * 255) + non_zero_byte)175        Always followed by exactly one LE16 :  D D D D D D D D : D D D D D D S S176           distance = D + 1177           state = S (copy S literals after this block)178 179      0 1 L D D D S S  (64..127)180           Copy 3-4 bytes from block within 2kB distance181           state = S (copy S literals after this block)182           length = 3 + L183         Always followed by exactly one byte : H H H H H H H H184           distance = (H << 3) + D + 1185 186      1 L L D D D S S  (128..255)187           Copy 5-8 bytes from block within 2kB distance188           state = S (copy S literals after this block)189           length = 5 + L190         Always followed by exactly one byte : H H H H H H H H191           distance = (H << 3) + D + 1192 193Authors194=======195 196  This document was written by Willy Tarreau <w@1wt.eu> on 2014/07/19 during an197  analysis of the decompression code available in Linux 3.16-rc5, and updated198  by Dave Rodgman <dave.rodgman@arm.com> on 2018/10/30 to introduce run-length199  encoding. The code is tricky, it is possible that this document contains200  mistakes or that a few corner cases were overlooked. In any case, please201  report any doubt, fix, or proposed updates to the author(s) so that the202  document can be updated.203