dissect.util 3.24.dev4__cp314-cp314t-musllinux_1_2_aarch64.whl

dissect/util/compression/lzfse.py

@@ -0,0 +1,467 @@
+# References:
+# - https://github.com/lzfse/lzfse
+from __future__ import annotations
+
+import io
+import struct
+from typing import BinaryIO, NamedTuple
+
+from dissect.util.compression import lzvn
+
+LZFSE_ENDOFSTREAM_BLOCK_MAGIC = b"bvx$"  # 0x24787662 (end of stream)
+LZFSE_UNCOMPRESSED_BLOCK_MAGIC = b"bvx-"  # 0x2d787662 (raw data)
+LZFSE_COMPRESSEDV1_BLOCK_MAGIC = b"bvx1"  # 0x31787662 (lzfse compressed, uncompressed tables)
+LZFSE_COMPRESSEDV2_BLOCK_MAGIC = b"bvx2"  # 0x32787662 (lzfse compressed, compressed tables)
+LZFSE_COMPRESSEDLZVN_BLOCK_MAGIC = b"bvxn"  # 0x6e787662 (lzvn compressed)
+
+# Throughout LZFSE we refer to "L", "M" and "D"; these will always appear as
+# a triplet, and represent a "usual" LZ-style literal and match pair.  "L"
+# is the number of literal bytes, "M" is the number of match bytes, and "D"
+# is the match "distance"; the distance in bytes between the current pointer
+# and the start of the match.
+LZFSE_ENCODE_L_SYMBOLS = 20
+LZFSE_ENCODE_M_SYMBOLS = 20
+LZFSE_ENCODE_D_SYMBOLS = 64
+LZFSE_ENCODE_LITERAL_SYMBOLS = 256
+LZFSE_ENCODE_L_STATES = 64
+LZFSE_ENCODE_M_STATES = 64
+LZFSE_ENCODE_D_STATES = 256
+LZFSE_ENCODE_LITERAL_STATES = 1024
+LZFSE_MATCHES_PER_BLOCK = 10000
+LZFSE_LITERALS_PER_BLOCK = 4 * LZFSE_MATCHES_PER_BLOCK
+
+# fmt: off
+_lzfse_freq_nbits_table = (
+    2, 3, 2, 5, 2, 3, 2, 8, 2, 3, 2, 5, 2, 3, 2, 14,
+    2, 3, 2, 5, 2, 3, 2, 8, 2, 3, 2, 5, 2, 3, 2, 14
+)
+_lzfse_freq_value_table = (
+    0, 2, 1, 4, 0, 3, 1, -1, 0, 2, 1, 5, 0, 3, 1, -1,
+    0, 2, 1, 6, 0, 3, 1, -1, 0, 2, 1, 7, 0, 3, 1, -1
+)
+
+# The L, M, D data streams are all encoded as a "base" value, which is
+# FSE-encoded, and an "extra bits" value, which is the difference between
+# value and base, and is simply represented as a raw bit value (because it
+# is the low-order bits of a larger number, not much entropy can be
+# extracted from these bits by more complex encoding schemes). The following
+# tables represent the number of low-order bits to encode separately and the
+# base values for each of L, M, and D.
+_l_extra_bits = (
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2, 3, 5, 8
+)
+_l_base_value = (
+    0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 20, 28, 60
+)
+_m_extra_bits = (
+    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 5, 8, 11
+)
+_m_base_value = (
+    0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 24, 56, 312
+)
+_d_extra_bits = (
+    0,  0,  0,  0,  1,  1,  1,  1,  2,  2,  2,  2,  3,  3,  3,  3,
+    4,  4,  4,  4,  5,  5,  5,  5,  6,  6,  6,  6,  7,  7,  7,  7,
+    8,  8,  8,  8,  9,  9,  9,  9,  10, 10, 10, 10, 11, 11, 11, 11,
+    12, 12, 12, 12, 13, 13, 13, 13, 14, 14, 14, 14, 15, 15, 15, 15
+)
+_d_base_value = (
+    0,      1,      2,      3,     4,     6,     8,     10,    12,    16,
+    20,     24,     28,     36,    44,    52,    60,    76,    92,    108,
+    124,    156,    188,    220,   252,   316,   380,   444,   508,   636,
+    764,    892,    1020,   1276,  1532,  1788,  2044,  2556,  3068,  3580,
+    4092,   5116,   6140,   7164,  8188,  10236, 12284, 14332, 16380, 20476,
+    24572,  28668,  32764,  40956, 49148, 57340, 65532, 81916, 98300, 114684,
+    131068, 163836, 196604, 229372
+)
+# fmt: on
+
+
+_I = struct.Struct("<I")
+
+
+def _clz(n: int) -> int:
+    if n == 0:
+        return 32
+    return 32 - n.bit_length()
+
+
+class LZFSECompressedBlockHeader(NamedTuple):
+    """LZFSE compressed block header."""
+
+    __struct_v1__ = struct.Struct(
+        "<IIIIIIi4HiHHH"
+        f"{LZFSE_ENCODE_L_SYMBOLS}H"
+        f"{LZFSE_ENCODE_M_SYMBOLS}H"
+        f"{LZFSE_ENCODE_D_SYMBOLS}H"
+        f"{LZFSE_ENCODE_LITERAL_SYMBOLS}H"
+    )
+    __struct_v2__ = struct.Struct("<I3Q")
+
+    n_raw_bytes: int
+    n_payload_bytes: int
+    n_literals: int
+    n_matches: int
+    n_literal_payload_bytes: int
+    n_lmd_payload_bytes: int
+    literal_bits: int
+    literal_state: tuple[int, int, int, int]
+    lmd_bits: int
+    l_state: int
+    m_state: int
+    d_state: int
+    l_freq: tuple[int, ...]
+    m_freq: tuple[int, ...]
+    d_freq: tuple[int, ...]
+    literal_freq: tuple[int, ...]
+
+    @classmethod
+    def parse_v1(cls, fh: BinaryIO) -> LZFSECompressedBlockHeader:
+        """Decode all fields from a v1 header."""
+        values = cls.__struct_v1__.unpack(fh.read(cls.__struct_v1__.size))
+
+        return cls(
+            n_raw_bytes=values[0],
+            n_payload_bytes=values[1],
+            n_literals=values[2],
+            n_matches=values[3],
+            n_literal_payload_bytes=values[4],
+            n_lmd_payload_bytes=values[5],
+            literal_bits=values[6],
+            literal_state=values[7:11],
+            lmd_bits=values[11],
+            l_state=values[12],
+            m_state=values[13],
+            d_state=values[14],
+            l_freq=values[15 : 15 + LZFSE_ENCODE_L_SYMBOLS],
+            m_freq=values[35 : 35 + LZFSE_ENCODE_M_SYMBOLS],
+            d_freq=values[55 : 55 + LZFSE_ENCODE_D_SYMBOLS],
+            literal_freq=values[119 : 119 + LZFSE_ENCODE_LITERAL_SYMBOLS],
+        )
+
+    @classmethod
+    def parse_v2(cls, fh: BinaryIO) -> LZFSECompressedBlockHeader:
+        """Decode all fields from a v2 header."""
+        values = cls.__struct_v2__.unpack(fh.read(cls.__struct_v2__.size))
+        v0, v1, v2 = values[1:4]
+
+        n_literal_payload_bytes = _get_field(v0, 20, 20)
+        n_lmd_payload_bytes = _get_field(v1, 40, 20)
+        n_payload_bytes = n_literal_payload_bytes + n_lmd_payload_bytes
+
+        freq_tables_size = _get_field(v2, 0, 32) - cls.__struct_v2__.size - 4  # exclude magic
+
+        if freq_tables_size == 0:
+            l_freq = (0,) * 20
+            m_freq = (0,) * 20
+            d_freq = (0,) * 64
+            literal_freq = (0,) * 256
+        else:
+            accum = 0
+            accum = int.from_bytes(fh.read(freq_tables_size), "little")
+            result = [0] * 720
+
+            for i in range(
+                LZFSE_ENCODE_L_SYMBOLS + LZFSE_ENCODE_M_SYMBOLS + LZFSE_ENCODE_D_SYMBOLS + LZFSE_ENCODE_LITERAL_SYMBOLS
+            ):
+                # Decode and store value
+                nbits, value = _decode_v1_freq_value(accum)
+                result[i] = value
+
+                # Consume nbits bits
+                accum >>= nbits
+
+            l_freq = tuple(result[0:20])
+            m_freq = tuple(result[20:40])
+            d_freq = tuple(result[40:104])
+            literal_freq = tuple(result[104:360])
+
+        return cls(
+            n_raw_bytes=values[0],
+            n_payload_bytes=n_payload_bytes,
+            n_literals=_get_field(v0, 0, 20),
+            n_matches=_get_field(v0, 40, 20),
+            n_literal_payload_bytes=n_literal_payload_bytes,
+            n_lmd_payload_bytes=n_lmd_payload_bytes,
+            literal_bits=_get_field(v0, 60, 3) - 7,
+            literal_state=(
+                _get_field(v1, 0, 10),
+                _get_field(v1, 10, 10),
+                _get_field(v1, 20, 10),
+                _get_field(v1, 30, 10),
+            ),
+            lmd_bits=_get_field(v1, 60, 3) - 7,
+            l_state=_get_field(v2, 32, 10),
+            m_state=_get_field(v2, 42, 10),
+            d_state=_get_field(v2, 52, 10),
+            l_freq=l_freq,
+            m_freq=m_freq,
+            d_freq=d_freq,
+            literal_freq=literal_freq,
+        )
+
+
+def _get_field(value: int, offset: int, nbits: int) -> int:
+    """Extracts up to 32 bits from a 64-bit field beginning at offset, and zero-extends them to a 32-bit int.
+
+    If we number the bits of ``value`` from 0 (least significant) to 63 (most significant),
+    the result is bits ``offset`` to ``offset+nbits-1``.
+    """
+    return (value >> offset) & ((1 << nbits) - 1)
+
+
+def _decode_v1_freq_value(bits: int) -> tuple[int, int]:
+    """Decode an entry value from next bits of stream."""
+    b = bits & 31
+    n = _lzfse_freq_nbits_table[b]
+
+    # Special cases for > 5 bits encoding
+    if n == 8:
+        value = 8 + ((bits >> 4) & 0xF)
+    elif n == 14:
+        value = 24 + ((bits >> 4) & 0x3FF)
+    else:
+        value = _lzfse_freq_value_table[b]
+
+    return n, value
+
+
+class DecoderEntry(NamedTuple):
+    """Entry for one state in the decoder table."""
+
+    k: int  # Number of bits to read
+    symbol: int  # Emitted symbol
+    delta: int  # Signed increment used to compute next state (+bias)
+
+
+class ValueDecoderEntry(NamedTuple):
+    """Entry for one state in the value decoder table."""
+
+    total_bits: int  # state bits + extra value bits = shift for next decode
+    value_bits: int  # extra value bits
+    delta: int  # state base (delta)
+    vbase: int  # value base
+
+
+def _init_decoder_table(nstates: int, freq: tuple[int, ...]) -> list[DecoderEntry]:
+    """Initialize decoder table ``T[nstates]``.
+
+    ``nstates = sum freq[i]`` is the number of states (a power of 2).
+    ``freq`` is a normalized histogram of symbol frequencies, with ``freq[i] >= 0``.
+
+    Some symbols may have a 0 frequency. In that case, they should not be present in the data.
+    """
+    table = []
+
+    n_clz = _clz(nstates)
+    sum_of_freq = 0
+
+    for i, f in enumerate(freq):
+        if f == 0:
+            # skip this symbol, no occurrences
+            continue
+
+        sum_of_freq += f
+        if sum_of_freq > nstates:
+            raise ValueError("Invalid frequency table")
+        k = _clz(f) - n_clz  # shift needed to ensure N <= (F<<K) < 2*N
+        j0 = ((2 * nstates) >> k) - f
+
+        # Initialize all states S reached by this symbol: OFFSET <= S < OFFSET + F
+        table.extend(
+            DecoderEntry(
+                k=k if j < j0 else k - 1,
+                symbol=i,
+                delta=((f + j) << k) - nstates if j < j0 else (j - j0) << (k - 1),
+            )
+            for j in range(f)
+        )
+
+    return table
+
+
+def _init_value_decoder_table(
+    nstates: int, freq: tuple[int, ...], symbol_vbits: tuple[int, ...], symbol_vbase: tuple[int, ...]
+) -> list[ValueDecoderEntry]:
+    """Initialize value decoder table ``T[nstates]``.
+
+    ``nstates = sum freq[i]`` is the number of states (a power of 2).
+    ``freq`` is a normalized histogram of symbol frequencies, with ``freq[i] >= 0``.
+    ``symbol_vbits`` and ``symbol_vbase`` are the number of value bits to read and the base value for each symbol.
+
+    Some symbols may have a 0 frequency. In that case, they should not be present in the data.
+    """
+    table = []
+
+    n_clz = _clz(nstates)
+    for i, f in enumerate(freq):
+        if f == 0:
+            # skip this symbol, no occurrences
+            continue
+
+        k = _clz(f) - n_clz  # shift needed to ensure N <= (F<<K) < 2*N
+        j0 = ((2 * nstates) >> k) - f
+
+        # Initialize all states S reached by this symbol: OFFSET <= S < OFFSET + F
+        table.extend(
+            ValueDecoderEntry(
+                total_bits=k + symbol_vbits[i] if j < j0 else (k - 1) + symbol_vbits[i],
+                value_bits=symbol_vbits[i],
+                delta=(((f + j) << k) - nstates) if j < j0 else ((j - j0) << (k - 1)),
+                vbase=symbol_vbase[i],
+            )
+            for j in range(f)
+        )
+
+    return table
+
+
+class _BitStream:
+    def __init__(self, data: bytes, nbits: int):
+        self.accum = int.from_bytes(data, "little")
+        self.nbits = nbits + (len(data) * 8)
+
+    def pull(self, n: int) -> int:
+        self.nbits -= n
+        result = self.accum >> self.nbits
+        self.accum &= (1 << self.nbits) - 1
+        return result
+
+
+def _decode(state: int, decoder_table: list[DecoderEntry], in_stream: _BitStream) -> tuple[int, int]:
+    """Decode and return symbol using the decoder table, and update state."""
+    e = decoder_table[state]
+
+    # Update state from K bits of input + DELTA
+    state = e.delta + in_stream.pull(e.k)
+
+    # Return the symbol for this state
+    return state, e.symbol
+
+
+def _value_decode(state: int, decoder_table: list[ValueDecoderEntry], in_stream: _BitStream) -> tuple[int, int]:
+    """Decode and return value using the decoder table, and update state."""
+    entry = decoder_table[state]
+
+    state_and_value_bits = in_stream.pull(entry.total_bits)
+    state = entry.delta + (state_and_value_bits >> entry.value_bits)
+
+    return state, entry.vbase + (state_and_value_bits & ((1 << entry.value_bits) - 1))
+
+
+def _decode_lmd(
+    header: LZFSECompressedBlockHeader,
+    literals: list[int],
+    l_decoder: list[ValueDecoderEntry],
+    m_decoder: list[ValueDecoderEntry],
+    d_decoder: list[ValueDecoderEntry],
+    in_stream: _BitStream,
+) -> bytes:
+    symbols = header.n_matches
+
+    l_state = header.l_state
+    m_state = header.m_state
+    d_state = header.d_state
+
+    lit = io.BytesIO(bytes(literals))
+
+    # ruff: noqa: N806
+    L = 0
+    M = 0
+    D = None
+
+    dst = bytearray()
+
+    while symbols:
+        # Decode the next L, M, D symbol from the input stream
+        l_state, L = _value_decode(l_state, l_decoder, in_stream)
+        if (lit.tell() + L) >= LZFSE_LITERALS_PER_BLOCK + 64:
+            raise ValueError("Literal overflow")
+
+        m_state, M = _value_decode(m_state, m_decoder, in_stream)
+        d_state, new_d = _value_decode(d_state, d_decoder, in_stream)
+        D = new_d if new_d != 0 else D
+
+        if D is None or len(dst) + L < D:
+            raise ValueError("Invalid match distance")
+
+        dst += lit.read(L)
+
+        remaining = M
+        while remaining > 0:
+            match_size = min(remaining, D)
+            dst += dst[-D : (-D + match_size) or None]
+            remaining -= match_size
+
+        symbols -= 1
+
+    return bytes(dst)
+
+
+def decompress(src: bytes | BinaryIO) -> bytes:
+    """LZFSE decompress from a file-like object or bytes.
+
+    Decompresses until EOF or EOS of the input data.
+
+    Args:
+        src: File-like object or bytes to decompress.
+
+    Returns:
+        The decompressed data.
+    """
+    if not hasattr(src, "read"):
+        src = io.BytesIO(src)
+
+    dst = bytearray()
+
+    while True:
+        magic = src.read(4)
+
+        if magic == LZFSE_ENDOFSTREAM_BLOCK_MAGIC:
+            break
+
+        if magic == LZFSE_UNCOMPRESSED_BLOCK_MAGIC:
+            (n_raw_bytes,) = _I.unpack(src.read(4))
+            if n_raw_bytes == 0:
+                continue
+
+            dst += src.read(n_raw_bytes)
+
+        elif magic in (LZFSE_COMPRESSEDV1_BLOCK_MAGIC, LZFSE_COMPRESSEDV2_BLOCK_MAGIC):
+            if magic == LZFSE_COMPRESSEDV1_BLOCK_MAGIC:
+                header = LZFSECompressedBlockHeader.parse_v1(src)
+            else:
+                header = LZFSECompressedBlockHeader.parse_v2(src)
+
+            literal_decoder = _init_decoder_table(LZFSE_ENCODE_LITERAL_STATES, header.literal_freq)
+            l_decoder = _init_value_decoder_table(LZFSE_ENCODE_L_STATES, header.l_freq, _l_extra_bits, _l_base_value)
+            m_decoder = _init_value_decoder_table(LZFSE_ENCODE_M_STATES, header.m_freq, _m_extra_bits, _m_base_value)
+            d_decoder = _init_value_decoder_table(LZFSE_ENCODE_D_STATES, header.d_freq, _d_extra_bits, _d_base_value)
+
+            in_stream = _BitStream(src.read(header.n_literal_payload_bytes), header.literal_bits)
+
+            literals = []
+            state0, state1, state2, state3 = header.literal_state
+            for _ in range(0, header.n_literals, 4):
+                state0, result = _decode(state0, literal_decoder, in_stream)
+                literals.append(result)
+                state1, result = _decode(state1, literal_decoder, in_stream)
+                literals.append(result)
+                state2, result = _decode(state2, literal_decoder, in_stream)
+                literals.append(result)
+                state3, result = _decode(state3, literal_decoder, in_stream)
+                literals.append(result)
+
+            in_stream = _BitStream(src.read(header.n_lmd_payload_bytes), header.lmd_bits)
+            dst += _decode_lmd(header, literals, l_decoder, m_decoder, d_decoder, in_stream)
+
+        elif magic == LZFSE_COMPRESSEDLZVN_BLOCK_MAGIC:
+            (n_raw_bytes,) = _I.unpack(src.read(4))
+            (n_payload_bytes,) = _I.unpack(src.read(4))
+
+            dst += lzvn.decompress(src.read(n_payload_bytes))
+
+        else:
+            raise ValueError(f"Invalid LZFSE block magic: {magic!r}")
+
+    return bytes(dst)

dissect/util/compression/lznt1.py

@@ -0,0 +1,92 @@
+# Reference: https://github.com/google/rekall/blob/master/rekall-core/rekall/plugins/filesystems/lznt1.py
+from __future__ import annotations
+
+import array
+import io
+import struct
+from typing import BinaryIO
+
+
+def _get_displacement(offset: int) -> int:
+    """Calculate the displacement."""
+    result = 0
+    while offset >= 0x10:
+        offset >>= 1
+        result += 1
+
+    return result
+
+
+DISPLACEMENT_TABLE = array.array("B", [_get_displacement(x) for x in range(8192)])
+
+COMPRESSED_MASK = 1 << 15
+SIGNATURE_MASK = 3 << 12
+SIZE_MASK = (1 << 12) - 1
+TAG_MASKS = [(1 << i) for i in range(8)]
+
+
+def decompress(src: bytes | BinaryIO) -> bytes:
+    """LZNT1 decompress from a file-like object or bytes.
+
+    Args:
+        src: File-like object or bytes to decompress.
+
+    Returns:
+        The decompressed data.
+    """
+    if not hasattr(src, "read"):
+        src = io.BytesIO(src)
+
+    offset = src.tell()
+    src.seek(0, io.SEEK_END)
+    size = src.tell() - offset
+    src.seek(offset)
+
+    dst = io.BytesIO()
+
+    while src.tell() - offset < size:
+        block_offset = src.tell()
+        uncompressed_chunk_offset = dst.tell()
+
+        block_header = struct.unpack("<H", src.read(2))[0]
+        if block_header & SIGNATURE_MASK != SIGNATURE_MASK:
+            break
+
+        hsize = block_header & SIZE_MASK
+
+        block_end = block_offset + hsize + 3
+
+        if block_header & COMPRESSED_MASK:
+            while src.tell() < block_end:
+                header = ord(src.read(1))
+                for mask in TAG_MASKS:
+                    if src.tell() >= block_end:
+                        break
+
+                    if header & mask:
+                        pointer = struct.unpack("<H", src.read(2))[0]
+                        displacement = DISPLACEMENT_TABLE[dst.tell() - uncompressed_chunk_offset - 1]
+
+                        symbol_offset = (pointer >> (12 - displacement)) + 1
+                        symbol_length = (pointer & (0xFFF >> displacement)) + 3
+
+                        dst.seek(-symbol_offset, io.SEEK_END)
+                        data = dst.read(symbol_length)
+
+                        # Pad the data to make it fit.
+                        if 0 < len(data) < symbol_length:
+                            data = data * (symbol_length // len(data) + 1)
+                            data = data[:symbol_length]
+
+                        dst.seek(0, io.SEEK_END)
+                        dst.write(data)
+                    else:
+                        data = src.read(1)
+                        dst.write(data)
+
+        else:
+            # Block is not compressed
+            data = src.read(hsize + 1)
+            dst.write(data)
+
+    return dst.getvalue()

dissect/util/compression/lzo.py

@@ -0,0 +1,118 @@
+# References:
+# - https://github.com/FFmpeg/FFmpeg/blob/master/libavutil/lzo.c
+# - https://docs.kernel.org/staging/lzo.html
+# - https://github.com/torvalds/linux/blob/master/lib/lzo/lzo1x_decompress_safe.c
+from __future__ import annotations
+
+import io
+import struct
+from typing import BinaryIO
+
+MAX_READ_LENGTH = (1 << 32) - 1000
+
+
+def _read_length(src: BinaryIO, val: int, mask: int) -> int:
+    if (length := val & mask) != 0:
+        return length
+
+    while (val := src.read(1)[0]) == 0:
+        if length >= MAX_READ_LENGTH:
+            raise ValueError("Invalid encoded length")
+        length += 255
+
+    return length + mask + val
+
+
+def decompress(src: bytes | BinaryIO, header: bool = True, buflen: int = -1) -> bytes:
+    """LZO decompress from a file-like object or bytes. Assumes no header.
+
+    Arguments are largely compatible with python-lzo API.
+
+    Args:
+        src: File-like object or bytes to decompress.
+        header: Whether the metadata header is included in the input.
+        buflen: If ``header`` is ``False``, a buffer length in bytes must be given that will fit the output.
+
+    Returns:
+        The decompressed data.
+    """
+    if not hasattr(src, "read"):
+        src = io.BytesIO(src)
+
+    dst = bytearray()
+
+    if header:
+        byte = src.read(1)[0]
+        if byte not in [0xF0, 0xF1]:
+            raise ValueError("Invalid header value")
+        out_len = struct.unpack("<I", src.read(4))[0]
+    else:
+        out_len = buflen
+
+    val = src.read(1)[0]
+    offset = src.tell()
+    if src.seek(5) == 5 and val == 17:
+        src.seek(offset)
+        _ = src.read(1)  # This would be the bitstream version but we don't currently use it
+        val = src.read(1)[0]
+    else:
+        src.seek(offset)
+
+    if val > 17:
+        dst += src.read(val - 17)
+        val = src.read(1)[0]
+
+        if val < 16:
+            raise ValueError("Invalid LZO stream")
+
+    state = 0
+    while True:
+        if val > 15:
+            if val > 63:
+                # Copy 3-8 bytes from block within 2kb distance
+                length = (val >> 5) - 1
+                dist = (src.read(1)[0] << 3) + ((val >> 2) & 7) + 1
+            elif val > 31:
+                # Copy of small block within 16kb distance
+                length = _read_length(src, val, 31)
+                val = src.read(1)[0]
+                dist = (src.read(1)[0] << 6) + (val >> 2) + 1
+            else:
+                # Copy of a block within 16...48kB distance
+                length = _read_length(src, val, 7)
+                dist = (1 << 14) + ((val & 8) << 11)
+                val = src.read(1)[0]
+                dist += (src.read(1)[0] << 6) + (val >> 2)
+                if dist == (1 << 14):
+                    if length != 1:
+                        raise ValueError("Invalid LZO stream")
+                    break
+        elif not state:
+            # Copy 4 or more literals, depending on the last 4 bits
+            length = _read_length(src, val, 15)
+            dst += src.read(length + 3)
+            val = src.read(1)[0]
+            if val > 15:
+                continue
+            length = 1
+            dist = (1 << 11) + (src.read(1)[0] << 2) + (val >> 2) + 1
+        else:
+            length = 0
+            dist = (src.read(1)[0] << 2) + (val >> 2) + 1
+
+        remaining = length + 2
+        while remaining > 0:
+            match_size = min(remaining, dist)
+            dst += dst[-dist : (-dist + match_size) or None]
+            remaining -= match_size
+
+        # State is often encoded in the last 2 bits of the value, and used in subsequent iterations
+        state = length = val & 3
+        dst += src.read(length)
+
+        if len(dst) == out_len:
+            break
+
+        val = src.read(1)[0]
+
+    return bytes(dst)