1
0

Merge pull request #9749 from SomberNight/202504_harden_memory_linux

add harden_memory_linux: harder for other processes to read our memory
This commit is contained in:
accumulator
2025-04-28 22:02:48 +02:00
committed by GitHub
2 changed files with 98 additions and 0 deletions

View File

@@ -0,0 +1,94 @@
# Copyright (C) 2020 cptpcrd
# Copyright (C) 2025 The Electrum developers
# Distributed under the MIT software license, see the accompanying
# file LICENCE or http://www.opensource.org/licenses/mit-license.php
#
# based on https://github.com/cptpcrd/pyprctl/blob/578ed3e81066a8a61dede912454d5eeaef37eeea/pyprctl/ffi.py#L28
#
# This module tries to restrict the ability of other processes to access the memory of our process.
# Traditionally, on Linux, one process can access the memory of another arbitrary process
# if both are running as the same user (uid). (Root can ofc access the memory of ~any process)
# Programs can opt-out from this by setting prctl(PR_SET_DUMPABLE, 0);
#
# Besides PR_SET_DUMPABLE, there are ways to globally restrict this for all processes:
# 1. The Yama (Linux Security Module) ptrace scope can be used to reduce these permissions
# This runtime kernel parameter can be set to the following options:
# 0 - Default attach security permissions.
# 1 - Restricted attach. Only child processes plus normal permissions.
# 2 - Admin-only attach. Only executables with CAP_SYS_PTRACE.
# 3 - No attach. No process may call ptrace at all. Irrevocable.
# # Note: The default value of kernel.yama.ptrace_scope is distro-specific.
# # See `$ cat /proc/sys/kernel/yama/ptrace_scope`.
# # - ubuntu 22.04 sets it to 1 (see /etc/sysctl.d/10-ptrace.conf),
# # - debian 12 sets it to 0
# # - manjaro sets it to 1
# 2. SELinux: ptrace can be restricted by setting the selinux deny_ptrace boolean.
#
# For a quick test on your system, try:
# $ cat /proc/$$/mem > /dev/null
# cat: /proc/4907/mem: Permission denied
# Getting "Permission denied" means access failed, "Input/output error" means access succeeded.
import ctypes
import ctypes.util
import os
import sys
from typing import Optional
from .logging import get_logger
_logger = get_logger(__name__)
PR_GET_DUMPABLE = 3
PR_SET_DUMPABLE = 4
_libc = None # type: Optional[ctypes.CDLL]
def _load_libc():
global _libc
if _libc is not None:
return
#assert sys.platform == "linux", sys.platform
# note: find_library can raise FileNotFoundError(OSError), see https://github.com/python/cpython/issues/93094
_libc_path = ctypes.util.find_library("c")
_libc = ctypes.CDLL(_libc_path, use_errno=True)
_libc.prctl.argtypes = (ctypes.c_int, ctypes.c_ulong, ctypes.c_ulong, ctypes.c_ulong, ctypes.c_ulong)
_libc.prctl.restype = ctypes.c_int
def set_dumpable(flag: bool) -> None:
"""Set the "dumpable" attribute on the current process.
This controls whether a core dump will be produced if the process receives a signal whose
default behavior is to produce a core dump.
In addition, processes that are not dumpable cannot be attached with ptrace() PTRACE_ATTACH.
In effect, another process running as the same user as us can read our memory if we are dumpable.
"""
_load_libc()
res = _libc.prctl(PR_SET_DUMPABLE, int(bool(flag)), 0, 0, 0)
if res < 0:
eno = ctypes.get_errno()
raise OSError(eno, os.strerror(eno), None, None, None)
def set_dumpable_safe(flag: bool) -> None:
try:
_load_libc()
except Exception as e:
_logger.exception("error loading libc")
return
assert _libc is not None
try:
set_dumpable(flag)
except OSError as e:
_logger.error(f"libc.prctl(PR_SET_DUMPABLE, {flag}) errored: {e}")
def get_dumpable() -> bool:
_load_libc()
res = _libc.prctl(PR_GET_DUMPABLE, 0, 0, 0, 0)
if res < 0:
eno = ctypes.get_errno()
raise OSError(eno, os.strerror(eno), None, None, None)
return res != 0

View File

@@ -419,6 +419,10 @@ def main():
print_stderr('unknown command:', uri)
sys.exit(1)
if sys.platform == "linux" and not is_android:
import electrum.harden_memory_linux
electrum.harden_memory_linux.set_dumpable_safe(False)
if cmdname == 'daemon' and config.get("detach"):
# detect lockfile.
# This is not as good as get_file_descriptor, but that would require the asyncio loop