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.
484 lines
16 KiB
484 lines
16 KiB
5 years ago
|
#-------------------------------------------------------------------------------
|
||
|
# elftools: elf/sections.py
|
||
|
#
|
||
|
# ELF sections
|
||
|
#
|
||
|
# Eli Bendersky (eliben@gmail.com)
|
||
|
# This code is in the public domain
|
||
|
#-------------------------------------------------------------------------------
|
||
|
from ..common.exceptions import ELFCompressionError
|
||
|
from ..common.utils import struct_parse, elf_assert, parse_cstring_from_stream
|
||
|
from collections import defaultdict
|
||
|
from .constants import SH_FLAGS
|
||
|
from .notes import iter_notes
|
||
|
|
||
|
import zlib
|
||
|
|
||
|
|
||
|
class Section(object):
|
||
|
""" Base class for ELF sections. Also used for all sections types that have
|
||
|
no special functionality.
|
||
|
|
||
|
Allows dictionary-like access to the section header. For example:
|
||
|
> sec = Section(...)
|
||
|
> sec['sh_type'] # section type
|
||
|
"""
|
||
|
def __init__(self, header, name, elffile):
|
||
|
self.header = header
|
||
|
self.name = name
|
||
|
self.elffile = elffile
|
||
|
self.stream = self.elffile.stream
|
||
|
self.structs = self.elffile.structs
|
||
|
self._compressed = header['sh_flags'] & SH_FLAGS.SHF_COMPRESSED
|
||
|
|
||
|
if self.compressed:
|
||
|
# Read the compression header now to know about the size/alignment
|
||
|
# of the decompressed data.
|
||
|
header = struct_parse(self.structs.Elf_Chdr,
|
||
|
self.stream,
|
||
|
stream_pos=self['sh_offset'])
|
||
|
self._compression_type = header['ch_type']
|
||
|
self._decompressed_size = header['ch_size']
|
||
|
self._decompressed_align = header['ch_addralign']
|
||
|
else:
|
||
|
self._decompressed_size = header['sh_size']
|
||
|
self._decompressed_align = header['sh_addralign']
|
||
|
|
||
|
@property
|
||
|
def compressed(self):
|
||
|
""" Is this section compressed?
|
||
|
"""
|
||
|
return self._compressed
|
||
|
|
||
|
@property
|
||
|
def data_size(self):
|
||
|
""" Return the logical size for this section's data.
|
||
|
|
||
|
This can be different from the .sh_size header field when the section
|
||
|
is compressed.
|
||
|
"""
|
||
|
return self._decompressed_size
|
||
|
|
||
|
@property
|
||
|
def data_alignment(self):
|
||
|
""" Return the logical alignment for this section's data.
|
||
|
|
||
|
This can be different from the .sh_addralign header field when the
|
||
|
section is compressed.
|
||
|
"""
|
||
|
return self._decompressed_align
|
||
|
|
||
|
def data(self):
|
||
|
""" The section data from the file.
|
||
|
|
||
|
Note that data is decompressed if the stored section data is
|
||
|
compressed.
|
||
|
"""
|
||
|
# If this section is compressed, deflate it
|
||
|
if self.compressed:
|
||
|
c_type = self._compression_type
|
||
|
if c_type == 'ELFCOMPRESS_ZLIB':
|
||
|
# Read the data to decompress starting right after the
|
||
|
# compression header until the end of the section.
|
||
|
hdr_size = self.structs.Elf_Chdr.sizeof()
|
||
|
self.stream.seek(self['sh_offset'] + hdr_size)
|
||
|
compressed = self.stream.read(self['sh_size'] - hdr_size)
|
||
|
|
||
|
decomp = zlib.decompressobj()
|
||
|
result = decomp.decompress(compressed, self.data_size)
|
||
|
else:
|
||
|
raise ELFCompressionError(
|
||
|
'Unknown compression type: {:#0x}'.format(c_type)
|
||
|
)
|
||
|
|
||
|
if len(result) != self._decompressed_size:
|
||
|
raise ELFCompressionError(
|
||
|
'Decompressed data is {} bytes long, should be {} bytes'
|
||
|
' long'.format(len(result), self._decompressed_size)
|
||
|
)
|
||
|
else:
|
||
|
self.stream.seek(self['sh_offset'])
|
||
|
result = self.stream.read(self._decompressed_size)
|
||
|
|
||
|
return result
|
||
|
|
||
|
def is_null(self):
|
||
|
""" Is this a null section?
|
||
|
"""
|
||
|
return False
|
||
|
|
||
|
def __getitem__(self, name):
|
||
|
""" Implement dict-like access to header entries
|
||
|
"""
|
||
|
return self.header[name]
|
||
|
|
||
|
def __eq__(self, other):
|
||
|
try:
|
||
|
return self.header == other.header
|
||
|
except AttributeError:
|
||
|
return False
|
||
|
|
||
|
def __hash__(self):
|
||
|
return hash(self.header)
|
||
|
|
||
|
|
||
|
class NullSection(Section):
|
||
|
""" ELF NULL section
|
||
|
"""
|
||
|
def is_null(self):
|
||
|
return True
|
||
|
|
||
|
|
||
|
class StringTableSection(Section):
|
||
|
""" ELF string table section.
|
||
|
"""
|
||
|
def get_string(self, offset):
|
||
|
""" Get the string stored at the given offset in this string table.
|
||
|
"""
|
||
|
table_offset = self['sh_offset']
|
||
|
s = parse_cstring_from_stream(self.stream, table_offset + offset)
|
||
|
return s.decode('utf-8') if s else ''
|
||
|
|
||
|
|
||
|
class SymbolTableSection(Section):
|
||
|
""" ELF symbol table section. Has an associated StringTableSection that's
|
||
|
passed in the constructor.
|
||
|
"""
|
||
|
def __init__(self, header, name, elffile, stringtable):
|
||
|
super(SymbolTableSection, self).__init__(header, name, elffile)
|
||
|
self.stringtable = stringtable
|
||
|
elf_assert(self['sh_entsize'] > 0,
|
||
|
'Expected entry size of section %r to be > 0' % name)
|
||
|
elf_assert(self['sh_size'] % self['sh_entsize'] == 0,
|
||
|
'Expected section size to be a multiple of entry size in section %r' % name)
|
||
|
self._symbol_name_map = None
|
||
|
|
||
|
def num_symbols(self):
|
||
|
""" Number of symbols in the table
|
||
|
"""
|
||
|
return self['sh_size'] // self['sh_entsize']
|
||
|
|
||
|
def get_symbol(self, n):
|
||
|
""" Get the symbol at index #n from the table (Symbol object)
|
||
|
"""
|
||
|
# Grab the symbol's entry from the stream
|
||
|
entry_offset = self['sh_offset'] + n * self['sh_entsize']
|
||
|
entry = struct_parse(
|
||
|
self.structs.Elf_Sym,
|
||
|
self.stream,
|
||
|
stream_pos=entry_offset)
|
||
|
# Find the symbol name in the associated string table
|
||
|
name = self.stringtable.get_string(entry['st_name'])
|
||
|
return Symbol(entry, name)
|
||
|
|
||
|
def get_symbol_by_name(self, name):
|
||
|
""" Get a symbol(s) by name. Return None if no symbol by the given name
|
||
|
exists.
|
||
|
"""
|
||
|
# The first time this method is called, construct a name to number
|
||
|
# mapping
|
||
|
#
|
||
|
if self._symbol_name_map is None:
|
||
|
self._symbol_name_map = defaultdict(list)
|
||
|
for i, sym in enumerate(self.iter_symbols()):
|
||
|
self._symbol_name_map[sym.name].append(i)
|
||
|
symnums = self._symbol_name_map.get(name)
|
||
|
return [self.get_symbol(i) for i in symnums] if symnums else None
|
||
|
|
||
|
def iter_symbols(self):
|
||
|
""" Yield all the symbols in the table
|
||
|
"""
|
||
|
for i in range(self.num_symbols()):
|
||
|
yield self.get_symbol(i)
|
||
|
|
||
|
|
||
|
class Symbol(object):
|
||
|
""" Symbol object - representing a single symbol entry from a symbol table
|
||
|
section.
|
||
|
|
||
|
Similarly to Section objects, allows dictionary-like access to the
|
||
|
symbol entry.
|
||
|
"""
|
||
|
def __init__(self, entry, name):
|
||
|
self.entry = entry
|
||
|
self.name = name
|
||
|
|
||
|
def __getitem__(self, name):
|
||
|
""" Implement dict-like access to entries
|
||
|
"""
|
||
|
return self.entry[name]
|
||
|
|
||
|
|
||
|
class SUNWSyminfoTableSection(Section):
|
||
|
""" ELF .SUNW Syminfo table section.
|
||
|
Has an associated SymbolTableSection that's passed in the constructor.
|
||
|
"""
|
||
|
def __init__(self, header, name, elffile, symboltable):
|
||
|
super(SUNWSyminfoTableSection, self).__init__(header, name, elffile)
|
||
|
self.symboltable = symboltable
|
||
|
|
||
|
def num_symbols(self):
|
||
|
""" Number of symbols in the table
|
||
|
"""
|
||
|
return self['sh_size'] // self['sh_entsize'] - 1
|
||
|
|
||
|
def get_symbol(self, n):
|
||
|
""" Get the symbol at index #n from the table (Symbol object).
|
||
|
It begins at 1 and not 0 since the first entry is used to
|
||
|
store the current version of the syminfo table.
|
||
|
"""
|
||
|
# Grab the symbol's entry from the stream
|
||
|
entry_offset = self['sh_offset'] + n * self['sh_entsize']
|
||
|
entry = struct_parse(
|
||
|
self.structs.Elf_Sunw_Syminfo,
|
||
|
self.stream,
|
||
|
stream_pos=entry_offset)
|
||
|
# Find the symbol name in the associated symbol table
|
||
|
name = self.symboltable.get_symbol(n).name
|
||
|
return Symbol(entry, name)
|
||
|
|
||
|
def iter_symbols(self):
|
||
|
""" Yield all the symbols in the table
|
||
|
"""
|
||
|
for i in range(1, self.num_symbols() + 1):
|
||
|
yield self.get_symbol(i)
|
||
|
|
||
|
|
||
|
class NoteSection(Section):
|
||
|
""" ELF NOTE section. Knows how to parse notes.
|
||
|
"""
|
||
|
def iter_notes(self):
|
||
|
""" Yield all the notes in the section. Each result is a dictionary-
|
||
|
like object with "n_name", "n_type", and "n_desc" fields, amongst
|
||
|
others.
|
||
|
"""
|
||
|
return iter_notes(self.elffile, self['sh_offset'], self['sh_size'])
|
||
|
|
||
|
|
||
|
class StabSection(Section):
|
||
|
""" ELF stab section.
|
||
|
"""
|
||
|
def iter_stabs(self):
|
||
|
""" Yield all stab entries. Result type is ELFStructs.Elf_Stabs.
|
||
|
"""
|
||
|
offset = self['sh_offset']
|
||
|
size = self['sh_size']
|
||
|
end = offset + size
|
||
|
while offset < end:
|
||
|
stabs = struct_parse(
|
||
|
self.structs.Elf_Stabs,
|
||
|
self.elffile.stream,
|
||
|
stream_pos=offset)
|
||
|
stabs['n_offset'] = offset
|
||
|
offset += self.structs.Elf_Stabs.sizeof()
|
||
|
self.stream.seek(offset)
|
||
|
yield stabs
|
||
|
|
||
|
|
||
|
class ARMAttribute(object):
|
||
|
""" ARM attribute object - representing a build attribute of ARM ELF files.
|
||
|
"""
|
||
|
def __init__(self, structs, stream):
|
||
|
self._tag = struct_parse(structs.Elf_Attribute_Tag, stream)
|
||
|
self.extra = None
|
||
|
|
||
|
if self.tag in ('TAG_FILE', 'TAG_SECTION', 'TAG_SYMBOL'):
|
||
|
self.value = struct_parse(structs.Elf_word('value'), stream)
|
||
|
|
||
|
if self.tag != 'TAG_FILE':
|
||
|
self.extra = []
|
||
|
s_number = struct_parse(structs.Elf_uleb128('s_number'), stream)
|
||
|
|
||
|
while s_number != 0:
|
||
|
self.extra.append(s_number)
|
||
|
s_number = struct_parse(structs.Elf_uleb128('s_number'),
|
||
|
stream
|
||
|
)
|
||
|
|
||
|
elif self.tag in ('TAG_CPU_RAW_NAME', 'TAG_CPU_NAME', 'TAG_CONFORMANCE'):
|
||
|
self.value = struct_parse(structs.Elf_ntbs('value',
|
||
|
encoding='utf-8'),
|
||
|
stream)
|
||
|
|
||
|
elif self.tag == 'TAG_COMPATIBILITY':
|
||
|
self.value = struct_parse(structs.Elf_uleb128('value'), stream)
|
||
|
self.extra = struct_parse(structs.Elf_ntbs('vendor_name',
|
||
|
encoding='utf-8'),
|
||
|
stream)
|
||
|
|
||
|
elif self.tag == 'TAG_ALSO_COMPATIBLE_WITH':
|
||
|
self.value = ARMAttribute(structs, stream)
|
||
|
|
||
|
if type(self.value.value) is not str:
|
||
|
nul = struct_parse(structs.Elf_byte('nul'), stream)
|
||
|
elf_assert(nul == 0,
|
||
|
"Invalid terminating byte %r, expecting NUL." % nul)
|
||
|
|
||
|
else:
|
||
|
self.value = struct_parse(structs.Elf_uleb128('value'), stream)
|
||
|
|
||
|
@property
|
||
|
def tag(self):
|
||
|
return self._tag['tag']
|
||
|
|
||
|
def __repr__(self):
|
||
|
s = '<ARMAttribute (%s): %r>' % (self.tag, self.value)
|
||
|
s += ' %s' % self.extra if self.extra is not None else ''
|
||
|
return s
|
||
|
|
||
|
|
||
|
class ARMAttributesSubsubsection(object):
|
||
|
""" Subsubsection of an ELF .ARM.attributes section's subsection.
|
||
|
"""
|
||
|
def __init__(self, stream, structs, offset):
|
||
|
self.stream = stream
|
||
|
self.offset = offset
|
||
|
self.structs = structs
|
||
|
|
||
|
self.header = ARMAttribute(self.structs, self.stream)
|
||
|
|
||
|
self.attr_start = self.stream.tell()
|
||
|
|
||
|
def iter_attributes(self, tag=None):
|
||
|
""" Yield all attributes (limit to |tag| if specified).
|
||
|
"""
|
||
|
for attribute in self._make_attributes():
|
||
|
if tag is None or attribute.tag == tag:
|
||
|
yield attribute
|
||
|
|
||
|
@property
|
||
|
def num_attributes(self):
|
||
|
""" Number of attributes in the subsubsection.
|
||
|
"""
|
||
|
return sum(1 for _ in self.iter_attributes()) + 1
|
||
|
|
||
|
@property
|
||
|
def attributes(self):
|
||
|
""" List of all attributes in the subsubsection.
|
||
|
"""
|
||
|
return [self.header] + list(self.iter_attributes())
|
||
|
|
||
|
def _make_attributes(self):
|
||
|
""" Create all attributes for this subsubsection except the first one
|
||
|
which is the header.
|
||
|
"""
|
||
|
end = self.offset + self.header.value
|
||
|
|
||
|
self.stream.seek(self.attr_start)
|
||
|
|
||
|
while self.stream.tell() != end:
|
||
|
yield ARMAttribute(self.structs, self.stream)
|
||
|
|
||
|
def __repr__(self):
|
||
|
s = "<ARMAttributesSubsubsection (%s): %d bytes>"
|
||
|
return s % (self.header.tag[4:], self.header.value)
|
||
|
|
||
|
|
||
|
class ARMAttributesSubsection(object):
|
||
|
""" Subsection of an ELF .ARM.attributes section.
|
||
|
"""
|
||
|
def __init__(self, stream, structs, offset):
|
||
|
self.stream = stream
|
||
|
self.offset = offset
|
||
|
self.structs = structs
|
||
|
|
||
|
self.header = struct_parse(self.structs.Elf_Attr_Subsection_Header,
|
||
|
self.stream,
|
||
|
self.offset
|
||
|
)
|
||
|
|
||
|
self.subsubsec_start = self.stream.tell()
|
||
|
|
||
|
def iter_subsubsections(self, scope=None):
|
||
|
""" Yield all subsubsections (limit to |scope| if specified).
|
||
|
"""
|
||
|
for subsubsec in self._make_subsubsections():
|
||
|
if scope is None or subsubsec.header.tag == scope:
|
||
|
yield subsubsec
|
||
|
|
||
|
@property
|
||
|
def num_subsubsections(self):
|
||
|
""" Number of subsubsections in the subsection.
|
||
|
"""
|
||
|
return sum(1 for _ in self.iter_subsubsections())
|
||
|
|
||
|
@property
|
||
|
def subsubsections(self):
|
||
|
""" List of all subsubsections in the subsection.
|
||
|
"""
|
||
|
return list(self.iter_subsubsections())
|
||
|
|
||
|
def _make_subsubsections(self):
|
||
|
""" Create all subsubsections for this subsection.
|
||
|
"""
|
||
|
end = self.offset + self['length']
|
||
|
|
||
|
self.stream.seek(self.subsubsec_start)
|
||
|
|
||
|
while self.stream.tell() != end:
|
||
|
subsubsec = ARMAttributesSubsubsection(self.stream,
|
||
|
self.structs,
|
||
|
self.stream.tell())
|
||
|
self.stream.seek(self.subsubsec_start + subsubsec.header.value)
|
||
|
yield subsubsec
|
||
|
|
||
|
def __getitem__(self, name):
|
||
|
""" Implement dict-like access to header entries.
|
||
|
"""
|
||
|
return self.header[name]
|
||
|
|
||
|
def __repr__(self):
|
||
|
s = "<ARMAttributesSubsection (%s): %d bytes>"
|
||
|
return s % (self.header['vendor_name'], self.header['length'])
|
||
|
|
||
|
|
||
|
class ARMAttributesSection(Section):
|
||
|
""" ELF .ARM.attributes section.
|
||
|
"""
|
||
|
def __init__(self, header, name, elffile):
|
||
|
super(ARMAttributesSection, self).__init__(header, name, elffile)
|
||
|
|
||
|
fv = struct_parse(self.structs.Elf_byte('format_version'),
|
||
|
self.stream,
|
||
|
self['sh_offset']
|
||
|
)
|
||
|
|
||
|
elf_assert(chr(fv) == 'A',
|
||
|
"Unknown attributes version %s, expecting 'A'." % chr(fv)
|
||
|
)
|
||
|
|
||
|
self.subsec_start = self.stream.tell()
|
||
|
|
||
|
def iter_subsections(self, vendor_name=None):
|
||
|
""" Yield all subsections (limit to |vendor_name| if specified).
|
||
|
"""
|
||
|
for subsec in self._make_subsections():
|
||
|
if vendor_name is None or subsec['vendor_name'] == vendor_name:
|
||
|
yield subsec
|
||
|
|
||
|
@property
|
||
|
def num_subsections(self):
|
||
|
""" Number of subsections in the section.
|
||
|
"""
|
||
|
return sum(1 for _ in self.iter_subsections())
|
||
|
|
||
|
@property
|
||
|
def subsections(self):
|
||
|
""" List of all subsections in the section.
|
||
|
"""
|
||
|
return list(self.iter_subsections())
|
||
|
|
||
|
def _make_subsections(self):
|
||
|
""" Create all subsections for this section.
|
||
|
"""
|
||
|
end = self['sh_offset'] + self.data_size
|
||
|
|
||
|
self.stream.seek(self.subsec_start)
|
||
|
|
||
|
while self.stream.tell() != end:
|
||
|
subsec = ARMAttributesSubsection(self.stream,
|
||
|
self.structs,
|
||
|
self.stream.tell())
|
||
|
self.stream.seek(self.subsec_start + subsec['length'])
|
||
|
yield subsec
|