merge bitcoin#20434: Parse ELF directly for symbol and security checks

Makefile.am

@@ -55,7 +55,8 @@ DIST_SHARE = \
   $(top_srcdir)/share/rpcauth
 
 BIN_CHECKS=$(top_srcdir)/contrib/devtools/symbol-check.py \
-           $(top_srcdir)/contrib/devtools/security-check.py
+           $(top_srcdir)/contrib/devtools/security-check.py \
+           $(top_srcdir)/contrib/devtools/pixie.py
 
 WINDOWS_PACKAGING = $(top_srcdir)/share/pixmaps/dash.ico \
   $(top_srcdir)/share/pixmaps/nsis-header.bmp \

configure.ac

@@ -106,7 +106,6 @@ AC_PATH_PROG([GIT], [git])
 AC_PATH_PROG(CCACHE,ccache)
 AC_PATH_PROG(XGETTEXT,xgettext)
 AC_PATH_PROG(HEXDUMP,hexdump)
-AC_PATH_TOOL(READELF, readelf)
 AC_PATH_TOOL(CPPFILT, c++filt)
 AC_PATH_TOOL(OBJCOPY, objcopy)
 AC_PATH_TOOL(DSYMUTIL, dsymutil)

contrib/devtools/pixie.py

@@ -0,0 +1,323 @@
+#!/usr/bin/env python3
+# Copyright (c) 2020 Wladimir J. van der Laan
+# Distributed under the MIT software license, see the accompanying
+# file COPYING or http://www.opensource.org/licenses/mit-license.php.
+'''
+Compact, self-contained ELF implementation for bitcoin-core security checks.
+'''
+import struct
+import types
+from typing import Dict, List, Optional, Union, Tuple
+
+# you can find all these values in elf.h
+EI_NIDENT = 16
+
+# Byte indices in e_ident
+EI_CLASS = 4 # ELFCLASSxx
+EI_DATA = 5  # ELFDATAxxxx
+
+ELFCLASS32 = 1 # 32-bit
+ELFCLASS64 = 2 # 64-bit
+
+ELFDATA2LSB = 1 # little endian
+ELFDATA2MSB = 2 # big endian
+
+# relevant values for e_machine
+EM_386 = 3
+EM_PPC64 = 21
+EM_ARM = 40
+EM_AARCH64 = 183
+EM_X86_64 = 62
+EM_RISCV = 243
+
+# relevant values for e_type
+ET_DYN = 3
+
+# relevant values for sh_type
+SHT_PROGBITS = 1
+SHT_STRTAB = 3
+SHT_DYNAMIC = 6
+SHT_DYNSYM = 11
+SHT_GNU_verneed = 0x6ffffffe
+SHT_GNU_versym = 0x6fffffff
+
+# relevant values for p_type
+PT_LOAD = 1
+PT_GNU_STACK = 0x6474e551
+PT_GNU_RELRO = 0x6474e552
+
+# relevant values for p_flags
+PF_X = (1 << 0)
+PF_W = (1 << 1)
+PF_R = (1 << 2)
+
+# relevant values for d_tag
+DT_NEEDED = 1
+DT_FLAGS = 30
+
+# relevant values of `d_un.d_val' in the DT_FLAGS entry
+DF_BIND_NOW = 0x00000008
+
+# relevant d_tags with string payload
+STRING_TAGS = {DT_NEEDED}
+
+# rrlevant values for ST_BIND subfield of st_info (symbol binding)
+STB_LOCAL = 0
+
+class ELFRecord(types.SimpleNamespace):
+    '''Unified parsing for ELF records.'''
+    def __init__(self, data: bytes, offset: int, eh: 'ELFHeader', total_size: Optional[int]) -> None:
+        hdr_struct = self.STRUCT[eh.ei_class][0][eh.ei_data]
+        if total_size is not None and hdr_struct.size > total_size:
+            raise ValueError(f'{self.__class__.__name__} header size too small ({total_size} < {hdr_struct.size})')
+        for field, value in zip(self.STRUCT[eh.ei_class][1], hdr_struct.unpack(data[offset:offset + hdr_struct.size])):
+            setattr(self, field, value)
+
+def BiStruct(chars: str) -> Dict[int, struct.Struct]:
+    '''Compile a struct parser for both endians.'''
+    return {
+        ELFDATA2LSB: struct.Struct('<' + chars),
+        ELFDATA2MSB: struct.Struct('>' + chars),
+    }
+
+class ELFHeader(ELFRecord):
+    FIELDS = ['e_type', 'e_machine', 'e_version', 'e_entry', 'e_phoff', 'e_shoff', 'e_flags', 'e_ehsize', 'e_phentsize', 'e_phnum', 'e_shentsize', 'e_shnum', 'e_shstrndx']
+    STRUCT = {
+        ELFCLASS32: (BiStruct('HHIIIIIHHHHHH'), FIELDS),
+        ELFCLASS64: (BiStruct('HHIQQQIHHHHHH'), FIELDS),
+    }
+
+    def __init__(self, data: bytes, offset: int) -> None:
+        self.e_ident = data[offset:offset + EI_NIDENT]
+        if self.e_ident[0:4] != b'\x7fELF':
+            raise ValueError('invalid ELF magic')
+        self.ei_class = self.e_ident[EI_CLASS]
+        self.ei_data = self.e_ident[EI_DATA]
+
+        super().__init__(data, offset + EI_NIDENT, self, None)
+
+    def __repr__(self) -> str:
+        return f'Header(e_ident={self.e_ident!r}, e_type={self.e_type}, e_machine={self.e_machine}, e_version={self.e_version}, e_entry={self.e_entry}, e_phoff={self.e_phoff}, e_shoff={self.e_shoff}, e_flags={self.e_flags}, e_ehsize={self.e_ehsize}, e_phentsize={self.e_phentsize}, e_phnum={self.e_phnum}, e_shentsize={self.e_shentsize}, e_shnum={self.e_shnum}, e_shstrndx={self.e_shstrndx})'
+
+class Section(ELFRecord):
+    name: Optional[bytes] = None
+    FIELDS = ['sh_name', 'sh_type', 'sh_flags', 'sh_addr', 'sh_offset', 'sh_size', 'sh_link', 'sh_info', 'sh_addralign', 'sh_entsize']
+    STRUCT = {
+        ELFCLASS32: (BiStruct('IIIIIIIIII'), FIELDS),
+        ELFCLASS64: (BiStruct('IIQQQQIIQQ'), FIELDS),
+    }
+
+    def __init__(self, data: bytes, offset: int, eh: ELFHeader) -> None:
+        super().__init__(data, offset, eh, eh.e_shentsize)
+        self._data = data
+
+    def __repr__(self) -> str:
+        return f'Section(sh_name={self.sh_name}({self.name!r}), sh_type=0x{self.sh_type:x}, sh_flags={self.sh_flags}, sh_addr=0x{self.sh_addr:x}, sh_offset=0x{self.sh_offset:x}, sh_size={self.sh_size}, sh_link={self.sh_link}, sh_info={self.sh_info}, sh_addralign={self.sh_addralign}, sh_entsize={self.sh_entsize})'
+
+    def contents(self) -> bytes:
+        '''Return section contents.'''
+        return self._data[self.sh_offset:self.sh_offset + self.sh_size]
+
+class ProgramHeader(ELFRecord):
+    STRUCT = {
+        # different ELF classes have the same fields, but in a different order to optimize space versus alignment
+        ELFCLASS32: (BiStruct('IIIIIIII'), ['p_type', 'p_offset', 'p_vaddr', 'p_paddr', 'p_filesz', 'p_memsz', 'p_flags', 'p_align']),
+        ELFCLASS64: (BiStruct('IIQQQQQQ'), ['p_type', 'p_flags', 'p_offset', 'p_vaddr', 'p_paddr', 'p_filesz', 'p_memsz', 'p_align']),
+    }
+
+    def __init__(self, data: bytes, offset: int, eh: ELFHeader) -> None:
+        super().__init__(data, offset, eh, eh.e_phentsize)
+
+    def __repr__(self) -> str:
+        return f'ProgramHeader(p_type={self.p_type}, p_offset={self.p_offset}, p_vaddr={self.p_vaddr}, p_paddr={self.p_paddr}, p_filesz={self.p_filesz}, p_memsz={self.p_memsz}, p_flags={self.p_flags}, p_align={self.p_align})'
+
+class Symbol(ELFRecord):
+    STRUCT = {
+        # different ELF classes have the same fields, but in a different order to optimize space versus alignment
+        ELFCLASS32: (BiStruct('IIIBBH'), ['st_name', 'st_value', 'st_size', 'st_info', 'st_other', 'st_shndx']),
+        ELFCLASS64: (BiStruct('IBBHQQ'), ['st_name', 'st_info', 'st_other', 'st_shndx', 'st_value', 'st_size']),
+    }
+
+    def __init__(self, data: bytes, offset: int, eh: ELFHeader, symtab: Section, strings: bytes, version: Optional[bytes]) -> None:
+        super().__init__(data, offset, eh, symtab.sh_entsize)
+        self.name = _lookup_string(strings, self.st_name)
+        self.version = version
+
+    def __repr__(self) -> str:
+        return f'Symbol(st_name={self.st_name}({self.name!r}), st_value={self.st_value}, st_size={self.st_size}, st_info={self.st_info}, st_other={self.st_other}, st_shndx={self.st_shndx}, version={self.version!r})'
+
+    @property
+    def is_import(self) -> bool:
+        '''Returns whether the symbol is an imported symbol.'''
+        return self.st_bind != STB_LOCAL and self.st_shndx == 0
+
+    @property
+    def is_export(self) -> bool:
+        '''Returns whether the symbol is an exported symbol.'''
+        return self.st_bind != STB_LOCAL and self.st_shndx != 0
+
+    @property
+    def st_bind(self) -> int:
+        '''Returns STB_*.'''
+        return self.st_info >> 4
+
+class Verneed(ELFRecord):
+    DEF = (BiStruct('HHIII'), ['vn_version', 'vn_cnt', 'vn_file', 'vn_aux', 'vn_next'])
+    STRUCT = { ELFCLASS32: DEF, ELFCLASS64: DEF }
+
+    def __init__(self, data: bytes, offset: int, eh: ELFHeader) -> None:
+        super().__init__(data, offset, eh, None)
+
+    def __repr__(self) -> str:
+        return f'Verneed(vn_version={self.vn_version}, vn_cnt={self.vn_cnt}, vn_file={self.vn_file}, vn_aux={self.vn_aux}, vn_next={self.vn_next})'
+
+class Vernaux(ELFRecord):
+    DEF = (BiStruct('IHHII'), ['vna_hash', 'vna_flags', 'vna_other', 'vna_name', 'vna_next'])
+    STRUCT = { ELFCLASS32: DEF, ELFCLASS64: DEF }
+
+    def __init__(self, data: bytes, offset: int, eh: ELFHeader, strings: bytes) -> None:
+        super().__init__(data, offset, eh, None)
+        self.name = _lookup_string(strings, self.vna_name)
+
+    def __repr__(self) -> str:
+        return f'Veraux(vna_hash={self.vna_hash}, vna_flags={self.vna_flags}, vna_other={self.vna_other}, vna_name={self.vna_name}({self.name!r}), vna_next={self.vna_next})'
+
+class DynTag(ELFRecord):
+    STRUCT = {
+        ELFCLASS32: (BiStruct('II'), ['d_tag', 'd_val']),
+        ELFCLASS64: (BiStruct('QQ'), ['d_tag', 'd_val']),
+    }
+
+    def __init__(self, data: bytes, offset: int, eh: ELFHeader, section: Section) -> None:
+        super().__init__(data, offset, eh, section.sh_entsize)
+
+    def __repr__(self) -> str:
+        return f'DynTag(d_tag={self.d_tag}, d_val={self.d_val})'
+
+def _lookup_string(data: bytes, index: int) -> bytes:
+    '''Look up string by offset in ELF string table.'''
+    endx = data.find(b'\x00', index)
+    assert endx != -1
+    return data[index:endx]
+
+VERSYM_S = BiStruct('H') # .gnu_version section has a single 16-bit integer per symbol in the linked section
+def _parse_symbol_table(section: Section, strings: bytes, eh: ELFHeader, versym: bytes, verneed: Dict[int, bytes]) -> List[Symbol]:
+    '''Parse symbol table, return a list of symbols.'''
+    data = section.contents()
+    symbols = []
+    versym_iter = (verneed.get(v[0]) for v in VERSYM_S[eh.ei_data].iter_unpack(versym))
+    for ofs, version in zip(range(0, len(data), section.sh_entsize), versym_iter):
+        symbols.append(Symbol(data, ofs, eh, section, strings, version))
+    return symbols
+
+def _parse_verneed(section: Section, strings: bytes, eh: ELFHeader) -> Dict[int, bytes]:
+    '''Parse .gnu.version_r section, return a dictionary of {versym: 'GLIBC_...'}.'''
+    data = section.contents()
+    ofs = 0
+    result = {}
+    while True:
+        verneed = Verneed(data, ofs, eh)
+        aofs = verneed.vn_aux
+        while True:
+            vernaux = Vernaux(data, aofs, eh, strings)
+            result[vernaux.vna_other] = vernaux.name
+            if not vernaux.vna_next:
+                break
+            aofs += vernaux.vna_next
+
+        if not verneed.vn_next:
+            break
+        ofs += verneed.vn_next
+
+    return result
+
+def _parse_dyn_tags(section: Section, strings: bytes, eh: ELFHeader) -> List[Tuple[int, Union[bytes, int]]]:
+    '''Parse dynamic tags. Return array of tuples.'''
+    data = section.contents()
+    ofs = 0
+    result = []
+    for ofs in range(0, len(data), section.sh_entsize):
+        tag = DynTag(data, ofs, eh, section)
+        val = _lookup_string(strings, tag.d_val) if tag.d_tag in STRING_TAGS else tag.d_val
+        result.append((tag.d_tag, val))
+
+    return result
+
+class ELFFile:
+    sections: List[Section]
+    program_headers: List[ProgramHeader]
+    dyn_symbols: List[Symbol]
+    dyn_tags: List[Tuple[int, Union[bytes, int]]]
+
+    def __init__(self, data: bytes) -> None:
+        self.data = data
+        self.hdr = ELFHeader(self.data, 0)
+        self._load_sections()
+        self._load_program_headers()
+        self._load_dyn_symbols()
+        self._load_dyn_tags()
+        self._section_to_segment_mapping()
+
+    def _load_sections(self) -> None:
+        self.sections = []
+        for idx in range(self.hdr.e_shnum):
+            offset = self.hdr.e_shoff + idx * self.hdr.e_shentsize
+            self.sections.append(Section(self.data, offset, self.hdr))
+
+        shstr = self.sections[self.hdr.e_shstrndx].contents()
+        for section in self.sections:
+            section.name = _lookup_string(shstr, section.sh_name)
+
+    def _load_program_headers(self) -> None:
+        self.program_headers = []
+        for idx in range(self.hdr.e_phnum):
+            offset = self.hdr.e_phoff + idx * self.hdr.e_phentsize
+            self.program_headers.append(ProgramHeader(self.data, offset, self.hdr))
+
+    def _load_dyn_symbols(self) -> None:
+        # first, load 'verneed' section
+        verneed = None
+        for section in self.sections:
+            if section.sh_type == SHT_GNU_verneed:
+                strtab = self.sections[section.sh_link].contents() # associated string table
+                assert verneed is None # only one section of this kind please
+                verneed = _parse_verneed(section, strtab, self.hdr)
+        assert verneed is not None
+
+        # then, correlate GNU versym sections with dynamic symbol sections
+        versym = {}
+        for section in self.sections:
+            if section.sh_type == SHT_GNU_versym:
+                versym[section.sh_link] = section
+
+        # finally, load dynsym sections
+        self.dyn_symbols = []
+        for idx, section in enumerate(self.sections):
+            if section.sh_type == SHT_DYNSYM: # find dynamic symbol tables
+                strtab_data = self.sections[section.sh_link].contents() # associated string table
+                versym_data = versym[idx].contents() # associated symbol version table
+                self.dyn_symbols += _parse_symbol_table(section, strtab_data, self.hdr, versym_data, verneed)
+
+    def _load_dyn_tags(self) -> None:
+        self.dyn_tags = []
+        for idx, section in enumerate(self.sections):
+            if section.sh_type == SHT_DYNAMIC: # find dynamic tag tables
+                strtab = self.sections[section.sh_link].contents() # associated string table
+                self.dyn_tags += _parse_dyn_tags(section, strtab, self.hdr)
+
+    def _section_to_segment_mapping(self) -> None:
+        for ph in self.program_headers:
+            ph.sections = []
+            for section in self.sections:
+                if ph.p_vaddr <= section.sh_addr < (ph.p_vaddr + ph.p_memsz):
+                    ph.sections.append(section)
+
+    def query_dyn_tags(self, tag_in: int) -> List[Union[int, bytes]]:
+        '''Return the values of all dyn tags with the specified tag.'''
+        return [val for (tag, val) in self.dyn_tags if tag == tag_in]
+
+
+def load(filename: str) -> ELFFile:
+    with open(filename, 'rb') as f:
+        data = f.read()
+    return ELFFile(data)

contrib/devtools/security-check.py

@@ -6,15 +6,15 @@
 Perform basic security checks on a series of executables.
 Exit status will be 0 if successful, and the program will be silent.
 Otherwise the exit status will be 1 and it will log which executables failed which checks.
-Needs `readelf` (for ELF), `objdump` (for PE) and `otool` (for MACHO).
+Needs `objdump` (for PE) and `otool` (for MACHO).
 '''
 import subprocess
 import sys
 import os
-
 from typing import List, Optional
 
-READELF_CMD = os.getenv('READELF', '/usr/bin/readelf')
+import pixie
+
 OBJDUMP_CMD = os.getenv('OBJDUMP', '/usr/bin/objdump')
 OTOOL_CMD = os.getenv('OTOOL', '/usr/bin/otool')
 
@@ -26,75 +26,20 @@ def check_ELF_PIE(executable) -> bool:
     '''
     Check for position independent executable (PIE), allowing for address space randomization.
     '''
-    stdout = run_command([READELF_CMD, '-h', '-W', executable])
+    elf = pixie.load(executable)
-
+    return elf.hdr.e_type == pixie.ET_DYN
-    ok = False
-    for line in stdout.splitlines():
-        tokens = line.split()
-        if len(line)>=2 and tokens[0] == 'Type:' and tokens[1] == 'DYN':
-            ok = True
-    return ok
-
-def get_ELF_program_headers(executable):
-    '''Return type and flags for ELF program headers'''
-    stdout = run_command([READELF_CMD, '-l', '-W', executable])
-
-    in_headers = False
-    headers = []
-    for line in stdout.splitlines():
-        if line.startswith('Program Headers:'):
-            in_headers = True
-            count = 0
-        if line == '':
-            in_headers = False
-        if in_headers:
-            if count == 1: # header line
-                header = [x.strip() for x in line.split()]
-                ofs_typ = header.index('Type')
-                ofs_flags = header.index('Flg')
-                # assert readelf output is what we expect
-                if ofs_typ == -1 or ofs_flags == -1:
-                    raise ValueError('Cannot parse elfread -lW output')
-            elif count > 1:
-                splitline = [x.strip() for x in line.split()]
-                typ = splitline[ofs_typ]
-                if not typ.startswith('[R'): # skip [Requesting ...]
-                    splitline = [x.strip() for x in line.split()]
-                    flags = splitline[ofs_flags]
-                    # check for 'R', ' E'
-                    if splitline[ofs_flags + 1] is 'E':
-                        flags += ' E'
-                    headers.append((typ, flags, []))
-            count += 1
-
-        if line.startswith(' Section to Segment mapping:'):
-            in_mapping = True
-            count = 0
-        if line == '':
-            in_mapping = False
-        if in_mapping:
-            if count == 1: # header line
-                ofs_segment = line.find('Segment')
-                ofs_sections = line.find('Sections...')
-                if ofs_segment == -1 or ofs_sections == -1:
-                    raise ValueError('Cannot parse elfread -lW output')
-            elif count > 1:
-                segment = int(line[ofs_segment:ofs_sections].strip())
-                sections = line[ofs_sections:].strip().split()
-                headers[segment][2].extend(sections)
-            count += 1
-    return headers
 
 def check_ELF_NX(executable) -> bool:
     '''
     Check that no sections are writable and executable (including the stack)
     '''
+    elf = pixie.load(executable)
     have_wx = False
     have_gnu_stack = False
-    for (typ, flags, _) in get_ELF_program_headers(executable):
+    for ph in elf.program_headers:
-        if typ == 'GNU_STACK':
+        if ph.p_type == pixie.PT_GNU_STACK:
             have_gnu_stack = True
-        if 'W' in flags and 'E' in flags: # section is both writable and executable
+        if (ph.p_flags & pixie.PF_W) != 0 and (ph.p_flags & pixie.PF_X) != 0: # section is both writable and executable
             have_wx = True
     return have_gnu_stack and not have_wx
 
@@ -104,35 +49,34 @@ def check_ELF_RELRO(executable) -> bool:
     GNU_RELRO program header must exist
     Dynamic section must have BIND_NOW flag
     '''
+    elf = pixie.load(executable)
     have_gnu_relro = False
-    for (typ, flags, _) in get_ELF_program_headers(executable):
+    for ph in elf.program_headers:
-        # Note: not checking flags == 'R': here as linkers set the permission differently
+        # Note: not checking p_flags == PF_R: here as linkers set the permission differently
         # This does not affect security: the permission flags of the GNU_RELRO program
         # header are ignored, the PT_LOAD header determines the effective permissions.
         # However, the dynamic linker need to write to this area so these are RW.
         # Glibc itself takes care of mprotecting this area R after relocations are finished.
         # See also https://marc.info/?l=binutils&m=1498883354122353
-        if typ == 'GNU_RELRO':
+        if ph.p_type == pixie.PT_GNU_RELRO:
             have_gnu_relro = True
 
     have_bindnow = False
-    stdout = run_command([READELF_CMD, '-d', '-W', executable])
+    for flags in elf.query_dyn_tags(pixie.DT_FLAGS):
-
+        assert isinstance(flags, int)
-    for line in stdout.splitlines():
+        if flags & pixie.DF_BIND_NOW:
-        tokens = line.split()
-        if len(tokens)>1 and tokens[1] == '(BIND_NOW)' or (len(tokens)>2 and tokens[1] == '(FLAGS)' and 'BIND_NOW' in tokens[2:]):
             have_bindnow = True
+
     return have_gnu_relro and have_bindnow
 
 def check_ELF_Canary(executable) -> bool:
     '''
     Check for use of stack canary
     '''
-    stdout = run_command([READELF_CMD, '--dyn-syms', '-W', executable])
+    elf = pixie.load(executable)
-
     ok = False
-    for line in stdout.splitlines():
+    for symbol in elf.dyn_symbols:
-        if '__stack_chk_fail' in line:
+        if symbol.name == b'__stack_chk_fail':
             ok = True
     return ok
 
@@ -142,48 +86,55 @@ def check_ELF_separate_code(executable):
     based on their permissions. This checks for missing -Wl,-z,separate-code
     and potentially other problems.
     '''
+    elf = pixie.load(executable)
+    R = pixie.PF_R
+    W = pixie.PF_W
+    E = pixie.PF_X
     EXPECTED_FLAGS = {
         # Read + execute
-        '.init': 'R E',
+        b'.init': R | E,
-        '.plt': 'R E',
+        b'.plt': R | E,
-        '.plt.got': 'R E',
+        b'.plt.got': R | E,
-        '.plt.sec': 'R E',
+        b'.plt.sec': R | E,
-        '.text': 'R E',
+        b'.text': R | E,
-        '.fini': 'R E',
+        b'.fini': R | E,
         # Read-only data
-        '.interp': 'R',
+        b'.interp': R,
-        '.note.gnu.property': 'R',
+        b'.note.gnu.property': R,
-        '.note.gnu.build-id': 'R',
+        b'.note.gnu.build-id': R,
-        '.note.ABI-tag': 'R',
+        b'.note.ABI-tag': R,
-        '.gnu.hash': 'R',
+        b'.gnu.hash': R,
-        '.dynsym': 'R',
+        b'.dynsym': R,
-        '.dynstr': 'R',
+        b'.dynstr': R,
-        '.gnu.version': 'R',
+        b'.gnu.version': R,
-        '.gnu.version_r': 'R',
+        b'.gnu.version_r': R,
-        '.rela.dyn': 'R',
+        b'.rela.dyn': R,
-        '.rela.plt': 'R',
+        b'.rela.plt': R,
-        '.rodata': 'R',
+        b'.rodata': R,
-        '.eh_frame_hdr': 'R',
+        b'.eh_frame_hdr': R,
-        '.eh_frame': 'R',
+        b'.eh_frame': R,
-        '.qtmetadata': 'R',
+        b'.qtmetadata': R,
-        '.gcc_except_table': 'R',
+        b'.gcc_except_table': R,
-        '.stapsdt.base': 'R',
+        b'.stapsdt.base': R,
         # Writable data
-        '.init_array': 'RW',
+        b'.init_array': R | W,
-        '.fini_array': 'RW',
+        b'.fini_array': R | W,
-        '.dynamic': 'RW',
+        b'.dynamic': R | W,
-        '.got': 'RW',
+        b'.got': R | W,
-        '.data': 'RW',
+        b'.data': R | W,
-        '.bss': 'RW',
+        b'.bss': R | W,
     }
+    if elf.hdr.e_machine == pixie.EM_PPC64:
+        # .plt is RW on ppc64 even with separate-code
+        EXPECTED_FLAGS[b'.plt'] = R | W
     # For all LOAD program headers get mapping to the list of sections,
     # and for each section, remember the flags of the associated program header.
     flags_per_section = {}
-    for (typ, flags, sections) in get_ELF_program_headers(executable):
+    for ph in elf.program_headers:
-        if typ == 'LOAD':
+        if ph.p_type == pixie.PT_LOAD:
-            for section in sections:
+            for section in ph.sections:
-                assert(section not in flags_per_section)
+                assert(section.name not in flags_per_section)
-                flags_per_section[section] = flags
+                flags_per_section[section.name] = ph.p_flags
     # Spot-check ELF LOAD program header flags per section
     # If these sections exist, check them against the expected R/W/E flags
     for (section, flags) in flags_per_section.items():
@@ -236,7 +187,7 @@ def check_PE_NX(executable) -> bool:
 def get_MACHO_executable_flags(executable) -> List[str]:
     stdout = run_command([OTOOL_CMD, '-vh', executable])
 
-    flags = []
+    flags: List[str] = []
     for line in stdout.splitlines():
         tokens = line.split()
         # filter first two header lines

contrib/devtools/symbol-check.py

@@ -12,10 +12,11 @@ Example usage:
     find ../gitian-builder/build -type f -executable | xargs python3 contrib/devtools/symbol-check.py
 '''
 import subprocess
-import re
 import sys
 import os
-from typing import List, Optional, Tuple
+from typing import List, Optional
+
+import pixie
 
 # Debian 8 (Jessie) EOL: 2020. https://wiki.debian.org/DebianReleases#Production_Releases
 #
@@ -52,7 +53,6 @@ IGNORE_EXPORTS = {
 '_edata', '_end', '__end__', '_init', '__bss_start', '__bss_start__', '_bss_end__', '__bss_end__', '_fini', '_IO_stdin_used', 'stdin', 'stdout', 'stderr',
 'environ', '_environ', '__environ',
 }
-READELF_CMD = os.getenv('READELF', '/usr/bin/readelf')
 CPPFILT_CMD = os.getenv('CPPFILT', '/usr/bin/c++filt')
 OTOOL_CMD = os.getenv('OTOOL', '/usr/bin/otool')
 
@@ -69,6 +69,8 @@ ELF_ALLOWED_LIBRARIES = {
 'ld-linux.so.2', # 32-bit dynamic linker
 'ld-linux-aarch64.so.1', # 64-bit ARM dynamic linker
 'ld-linux-armhf.so.3', # 32-bit ARM dynamic linker
+'ld64.so.1', # POWER64 ABIv1 dynamic linker
+'ld64.so.2', # POWER64 ABIv2 dynamic linker
 'ld-linux-riscv64-lp64d.so.1', # 64-bit RISC-V dynamic linker
 # dash-qt only
 'libxcb.so.1', # part of X11
@@ -79,11 +81,12 @@ ELF_ALLOWED_LIBRARIES = {
 'libdl.so.2' # programming interface to dynamic linker
 }
 ARCH_MIN_GLIBC_VER = {
-'80386':  (2,1),
+pixie.EM_386:    (2,1),
-'X86-64': (2,2,5),
+pixie.EM_X86_64: (2,2,5),
-'ARM':    (2,4),
+pixie.EM_ARM:    (2,4),
-'AArch64':(2,17),
+pixie.EM_AARCH64:(2,17),
-'RISC-V': (2,27)
+pixie.EM_PPC64:  (2,17),
+pixie.EM_RISCV:  (2,27)
 }
 
 MACHO_ALLOWED_LIBRARIES = {
@@ -128,29 +131,6 @@ class CPPFilt(object):
         self.proc.stdout.close()
         self.proc.wait()
 
-def read_symbols(executable, imports=True) -> List[Tuple[str, str, str]]:
-    '''
-    Parse an ELF executable and return a list of (symbol,version, arch) tuples
-    for dynamic, imported symbols.
-    '''
-    p = subprocess.Popen([READELF_CMD, '--dyn-syms', '-W', '-h', executable], stdout=subprocess.PIPE, stderr=subprocess.PIPE, stdin=subprocess.PIPE, universal_newlines=True)
-    (stdout, stderr) = p.communicate()
-    if p.returncode:
-        raise IOError('Could not read symbols for {}: {}'.format(executable, stderr.strip()))
-    syms = []
-    for line in stdout.splitlines():
-        line = line.split()
-        if 'Machine:' in line:
-            arch = line[-1]
-        if len(line)>7 and re.match('[0-9]+:$', line[0]):
-            (sym, _, version) = line[7].partition('@')
-            is_import = line[6] == 'UND'
-            if version.startswith('@'):
-                version = version[1:]
-            if is_import == imports:
-                syms.append((sym, version, arch))
-    return syms
-
 def check_version(max_versions, version, arch) -> bool:
     if '_' in version:
         (lib, _, ver) = version.rpartition('_')
@@ -162,36 +142,30 @@ def check_version(max_versions, version, arch) -> bool:
         return False
     return ver <= max_versions[lib] or lib == 'GLIBC' and ver <= ARCH_MIN_GLIBC_VER[arch]
 
-def elf_read_libraries(filename) -> List[str]:
-    p = subprocess.Popen([READELF_CMD, '-d', '-W', filename], stdout=subprocess.PIPE, stderr=subprocess.PIPE, stdin=subprocess.PIPE, universal_newlines=True)
-    (stdout, stderr) = p.communicate()
-    if p.returncode:
-        raise IOError('Error opening file')
-    libraries = []
-    for line in stdout.splitlines():
-        tokens = line.split()
-        if len(tokens)>2 and tokens[1] == '(NEEDED)':
-            match = re.match(r'^Shared library: \[(.*)\]$', ' '.join(tokens[2:]))
-            if match:
-                libraries.append(match.group(1))
-            else:
-                raise ValueError('Unparseable (NEEDED) specification')
-    return libraries
-
 def check_imported_symbols(filename) -> bool:
+    elf = pixie.load(filename)
     cppfilt = CPPFilt()
     ok = True
-    for sym, version, arch in read_symbols(filename, True):
+
-        if version and not check_version(MAX_VERSIONS, version, arch):
+    for symbol in elf.dyn_symbols:
+        if not symbol.is_import:
+            continue
+        sym = symbol.name.decode()
+        version = symbol.version.decode() if symbol.version is not None else None
+        if version and not check_version(MAX_VERSIONS, version, elf.hdr.e_machine):
             print('{}: symbol {} from unsupported version {}'.format(filename, cppfilt(sym), version))
             ok = False
     return ok
 
 def check_exported_symbols(filename) -> bool:
+    elf = pixie.load(filename)
     cppfilt = CPPFilt()
     ok = True
-    for sym,version,arch in read_symbols(filename, False):
+    for symbol in elf.dyn_symbols:
-        if arch == 'RISC-V' or sym in IGNORE_EXPORTS:
+        if not symbol.is_export:
+            continue
+        sym = symbol.name.decode()
+        if elf.hdr.e_machine == pixie.EM_RISCV or sym in IGNORE_EXPORTS:
             continue
         print('{}: export of symbol {} not allowed'.format(filename, cppfilt(sym)))
         ok = False
@@ -199,9 +173,11 @@ def check_exported_symbols(filename) -> bool:
 
 def check_ELF_libraries(filename) -> bool:
     ok = True
-    for library_name in elf_read_libraries(filename):
+    elf = pixie.load(filename)
-        if library_name not in ELF_ALLOWED_LIBRARIES:
+    for library_name in elf.query_dyn_tags(pixie.DT_NEEDED):
-            print('{}: NEEDED library {} is not allowed'.format(filename, library_name))
+        assert(isinstance(library_name, bytes))
+        if library_name.decode() not in ELF_ALLOWED_LIBRARIES:
+            print('{}: NEEDED library {} is not allowed'.format(filename, library_name.decode()))
             ok = False
     return ok
 

src/Makefile.am

@@ -928,13 +928,13 @@ endif
 
 if GLIBC_BACK_COMPAT
 	@echo "Checking glibc back compat..."
-	$(AM_V_at) READELF=$(READELF) CPPFILT=$(CPPFILT) $(PYTHON) $(top_srcdir)/contrib/devtools/symbol-check.py $(bin_PROGRAMS)
+	$(AM_V_at) CPPFILT=$(CPPFILT) $(PYTHON) $(top_srcdir)/contrib/devtools/symbol-check.py $(bin_PROGRAMS)
 endif
 
 check-security: $(bin_PROGRAMS)
 if HARDEN
 	@echo "Checking binary security..."
-	$(AM_V_at) READELF=$(READELF) OBJDUMP=$(OBJDUMP) OTOOL=$(OTOOL) $(PYTHON) $(top_srcdir)/contrib/devtools/security-check.py $(bin_PROGRAMS)
+	$(AM_V_at) OBJDUMP=$(OBJDUMP) OTOOL=$(OTOOL) $(PYTHON) $(top_srcdir)/contrib/devtools/security-check.py $(bin_PROGRAMS)
 endif