lzo.rst 9.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202
  1. ===========================================================
  2. LZO stream format as understood by Linux's LZO decompressor
  3. ===========================================================
  4. Introduction
  5. ============
  6. This is not a specification. No specification seems to be publicly available
  7. for the LZO stream format. This document describes what input format the LZO
  8. decompressor as implemented in the Linux kernel understands. The file subject
  9. of this analysis is lib/lzo/lzo1x_decompress_safe.c. No analysis was made on
  10. the compressor nor on any other implementations though it seems likely that
  11. the format matches the standard one. The purpose of this document is to
  12. better understand what the code does in order to propose more efficient fixes
  13. for future bug reports.
  14. Description
  15. ===========
  16. The stream is composed of a series of instructions, operands, and data. The
  17. instructions consist in a few bits representing an opcode, and bits forming
  18. the operands for the instruction, whose size and position depend on the
  19. opcode and on the number of literals copied by previous instruction. The
  20. operands are used to indicate:
  21. - a distance when copying data from the dictionary (past output buffer)
  22. - a length (number of bytes to copy from dictionary)
  23. - the number of literals to copy, which is retained in variable "state"
  24. as a piece of information for next instructions.
  25. Optionally depending on the opcode and operands, extra data may follow. These
  26. extra data can be a complement for the operand (eg: a length or a distance
  27. encoded on larger values), or a literal to be copied to the output buffer.
  28. The first byte of the block follows a different encoding from other bytes, it
  29. seems to be optimized for literal use only, since there is no dictionary yet
  30. prior to that byte.
  31. Lengths are always encoded on a variable size starting with a small number
  32. of bits in the operand. If the number of bits isn't enough to represent the
  33. length, up to 255 may be added in increments by consuming more bytes with a
  34. rate of at most 255 per extra byte (thus the compression ratio cannot exceed
  35. around 255:1). The variable length encoding using #bits is always the same::
  36. length = byte & ((1 << #bits) - 1)
  37. if (!length) {
  38. length = ((1 << #bits) - 1)
  39. length += 255*(number of zero bytes)
  40. length += first-non-zero-byte
  41. }
  42. length += constant (generally 2 or 3)
  43. For references to the dictionary, distances are relative to the output
  44. pointer. Distances are encoded using very few bits belonging to certain
  45. ranges, resulting in multiple copy instructions using different encodings.
  46. Certain encodings involve one extra byte, others involve two extra bytes
  47. forming a little-endian 16-bit quantity (marked LE16 below).
  48. After any instruction except the large literal copy, 0, 1, 2 or 3 literals
  49. are copied before starting the next instruction. The number of literals that
  50. were copied may change the meaning and behaviour of the next instruction. In
  51. practice, only one instruction needs to know whether 0, less than 4, or more
  52. literals were copied. This is the information stored in the <state> variable
  53. in this implementation. This number of immediate literals to be copied is
  54. generally encoded in the last two bits of the instruction but may also be
  55. taken from the last two bits of an extra operand (eg: distance).
  56. End of stream is declared when a block copy of distance 0 is seen. Only one
  57. instruction may encode this distance (0001HLLL), it takes one LE16 operand
  58. for the distance, thus requiring 3 bytes.
  59. .. important::
  60. In the code some length checks are missing because certain instructions
  61. are called under the assumption that a certain number of bytes follow
  62. because it has already been guaranteed before parsing the instructions.
  63. They just have to "refill" this credit if they consume extra bytes. This
  64. is an implementation design choice independent on the algorithm or
  65. encoding.
  66. Versions
  67. 0: Original version
  68. 1: LZO-RLE
  69. Version 1 of LZO implements an extension to encode runs of zeros using run
  70. length encoding. This improves speed for data with many zeros, which is a
  71. common case for zram. This modifies the bitstream in a backwards compatible way
  72. (v1 can correctly decompress v0 compressed data, but v0 cannot read v1 data).
  73. For maximum compatibility, both versions are available under different names
  74. (lzo and lzo-rle). Differences in the encoding are noted in this document with
  75. e.g.: version 1 only.
  76. Byte sequences
  77. ==============
  78. First byte encoding::
  79. 0..16 : follow regular instruction encoding, see below. It is worth
  80. noting that code 16 will represent a block copy from the
  81. dictionary which is empty, and that it will always be
  82. invalid at this place.
  83. 17 : bitstream version. If the first byte is 17, and compressed
  84. stream length is at least 5 bytes (length of shortest possible
  85. versioned bitstream), the next byte gives the bitstream version
  86. (version 1 only).
  87. Otherwise, the bitstream version is 0.
  88. 18..21 : copy 0..3 literals
  89. state = (byte - 17) = 0..3 [ copy <state> literals ]
  90. skip byte
  91. 22..255 : copy literal string
  92. length = (byte - 17) = 4..238
  93. state = 4 [ don't copy extra literals ]
  94. skip byte
  95. Instruction encoding::
  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. 0 0 0 0 L L L L (0..15) : copy long literal string
  102. length = 3 + (L ?: 15 + (zero_bytes * 255) + non_zero_byte)
  103. state = 4 (no extra literals are copied)
  104. If last instruction used to copy between 1 to 3 literals (encoded in
  105. the instruction's opcode or distance), the instruction is a copy of a
  106. 2-byte block from the dictionary within a 1kB distance. It is worth
  107. noting that this instruction provides little savings since it uses 2
  108. bytes to encode a copy of 2 other bytes but it encodes the number of
  109. following literals for free. It must be interpreted like this :
  110. 0 0 0 0 D D S S (0..15) : copy 2 bytes from <= 1kB distance
  111. length = 2
  112. state = S (copy S literals after this block)
  113. Always followed by exactly one byte : H H H H H H H H
  114. distance = (H << 2) + D + 1
  115. If last instruction used to copy 4 or more literals (as detected by
  116. state == 4), the instruction becomes a copy of a 3-byte block from the
  117. dictionary from a 2..3kB distance, and must be interpreted like this :
  118. 0 0 0 0 D D S S (0..15) : copy 3 bytes from 2..3 kB distance
  119. length = 3
  120. state = S (copy S literals after this block)
  121. Always followed by exactly one byte : H H H H H H H H
  122. distance = (H << 2) + D + 2049
  123. 0 0 0 1 H L L L (16..31)
  124. Copy of a block within 16..48kB distance (preferably less than 10B)
  125. length = 2 + (L ?: 7 + (zero_bytes * 255) + non_zero_byte)
  126. Always followed by exactly one LE16 : D D D D D D D D : D D D D D D S S
  127. distance = 16384 + (H << 14) + D
  128. state = S (copy S literals after this block)
  129. End of stream is reached if distance == 16384
  130. In version 1 only, to prevent ambiguity with the RLE case when
  131. ((distance & 0x803f) == 0x803f) && (261 <= length <= 264), the
  132. compressor must not emit block copies where distance and length
  133. meet these conditions.
  134. In version 1 only, this instruction is also used to encode a run of
  135. zeros if distance = 0xbfff, i.e. H = 1 and the D bits are all 1.
  136. In this case, it is followed by a fourth byte, X.
  137. run length = ((X << 3) | (0 0 0 0 0 L L L)) + 4
  138. 0 0 1 L L L L L (32..63)
  139. Copy of small block within 16kB distance (preferably less than 34B)
  140. length = 2 + (L ?: 31 + (zero_bytes * 255) + non_zero_byte)
  141. Always followed by exactly one LE16 : D D D D D D D D : D D D D D D S S
  142. distance = D + 1
  143. state = S (copy S literals after this block)
  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. 1 L L D D D S S (128..255)
  151. Copy 5-8 bytes from block within 2kB distance
  152. state = S (copy S literals after this block)
  153. length = 5 + L
  154. Always followed by exactly one byte : H H H H H H H H
  155. distance = (H << 3) + D + 1
  156. Authors
  157. =======
  158. This document was written by Willy Tarreau <w@1wt.eu> on 2014/07/19 during an
  159. analysis of the decompression code available in Linux 3.16-rc5, and updated
  160. by Dave Rodgman <dave.rodgman@arm.com> on 2018/10/30 to introduce run-length
  161. encoding. The code is tricky, it is possible that this document contains
  162. mistakes or that a few corner cases were overlooked. In any case, please
  163. report any doubt, fix, or proposed updates to the author(s) so that the
  164. document can be updated.