Source code for sos.archive

# Copyright (C) 2012 Red Hat, Inc.,
#   Jesse Jaggars <>
#   Bryn M. Reeves <>
# This file is part of the sos project:
# This copyrighted material is made available to anyone wishing to use,
# modify, copy, or redistribute it subject to the terms and conditions of
# version 2 of the GNU General Public License.
# See the LICENSE file in the source distribution for further information.
import os
import tarfile
import shutil
import logging
import codecs
import errno
import stat
from datetime import datetime
from threading import Lock

from importlib.util import find_spec
from sos.utilities import sos_get_command_output

    import selinux
except ImportError:

P_FILE = "file"
P_LINK = "link"
P_NODE = "node"
P_DIR = "dir"

[docs]class Archive(object): """Abstract base class for archives."""
[docs] @classmethod def archive_type(cls): """Returns the archive class's name as a string. """ return cls.__name__
log = logging.getLogger("sos") _name = "unset" _debug = False _path_lock = Lock() def _format_msg(self, msg): return "[archive:%s] %s" % (self.archive_type(), msg)
[docs] def set_debug(self, debug): self._debug = debug
[docs] def log_error(self, msg): self.log.error(self._format_msg(msg))
[docs] def log_warn(self, msg): self.log.warning(self._format_msg(msg))
[docs] def log_info(self, msg):
[docs] def log_debug(self, msg): if not self._debug: return self.log.debug(self._format_msg(msg))
# this is our contract to clients of the Archive class hierarchy. # All sub-classes need to implement these methods (or inherit concrete # implementations from a parent class.
[docs] def add_file(self, src, dest=None): raise NotImplementedError
[docs] def add_string(self, content, dest, mode='w'): raise NotImplementedError
[docs] def add_binary(self, content, dest): raise NotImplementedError
[docs] def add_dir(self, path): raise NotImplementedError
[docs] def add_node(self, path, mode, device): raise NotImplementedError
[docs] def get_tmp_dir(self): """Return a temporary directory that clients of the archive may use to write content to. The content of the path is guaranteed to be included in the generated archive.""" raise NotImplementedError
[docs] def name_max(self): """Return the maximum file name length this archive can support. This is the lesser of the name length limit of the archive format and any temporary file system based cache.""" raise NotImplementedError
[docs] def get_archive_path(self): """Return a string representing the path to the temporary archive. For archive classes that implement in-line handling this will be the archive file itself. Archives that use a directory based cache prior to packaging should return the path to the temporary directory where the report content is located""" pass
[docs] def cleanup(self): """Clean up any temporary resources used by an Archive class.""" pass
[docs] def finalize(self, method): """Finalize an archive object via method. This may involve creating An archive that is subsequently compressed or simply closing an archive that supports in-line handling. If method is automatic then the following methods are tried in order: xz, gzip""" self.close()
[docs]class FileCacheArchive(Archive): """ Abstract superclass for archive types that use a temporary cache directory in the file system. """ _tmp_dir = "" _archive_root = "" _archive_name = "" def __init__(self, name, tmpdir, policy, threads, enc_opts, sysroot, manifest=None): self._name = name self._tmp_dir = tmpdir self._policy = policy self._threads = threads self.enc_opts = enc_opts self.sysroot = sysroot or '/' self.manifest = manifest self._archive_root = os.path.join(tmpdir, name) with self._path_lock: os.makedirs(self._archive_root, 0o700) self.log_info("initialised empty FileCacheArchive at '%s'" % (self._archive_root,))
[docs] def dest_path(self, name): if os.path.isabs(name): name = name.lstrip(os.sep) return (os.path.join(self._archive_root, name))
[docs] def join_sysroot(self, path): if path.startswith(self.sysroot): return path if path[0] == os.sep: path = path[1:] return os.path.join(self.sysroot, path)
def _make_leading_paths(self, src, mode=0o700): """Create leading path components The standard python `os.makedirs` is insufficient for our needs: it will only create directories, and ignores the fact that some path components may be symbolic links. :param src: The source path in the host file system for which leading components should be created, or the path to an sos_* virtual directory inside the archive. Host paths must be absolute (initial '/'), and sos_* directory paths must be a path relative to the root of the archive. :param mode: An optional mode to be used when creating path components. :returns: A rewritten destination path in the case that one or more symbolic links in intermediate components of the path have altered the path destination. """ self.log_debug("Making leading paths for %s" % src) root = self._archive_root dest = src def in_archive(path): """Test whether path ``path`` is inside the archive. """ return path.startswith(os.path.join(root, "")) if not src.startswith("/"): # Sos archive path (sos_commands, sos_logs etc.) src_dir = src else: # Host file path src_dir = (src if os.path.isdir(self.join_sysroot(src)) else os.path.split(src)[0]) # Build a list of path components in root-to-leaf order. path = src_dir path_comps = [] while path != '/' and path != '': head, tail = os.path.split(path) path_comps.append(tail) path = head path_comps.reverse() abs_path = root src_path = "/" # Check and create components as needed for comp in path_comps: abs_path = os.path.join(abs_path, comp) # Do not create components that are above the archive root. if not in_archive(abs_path): continue src_path = os.path.join(src_path, comp) if not os.path.exists(abs_path): self.log_debug("Making path %s" % abs_path) if os.path.islink(src_path) and os.path.isdir(src_path): target = os.readlink(src_path) # The directory containing the source in the host fs, # adjusted for the current level of path creation. target_dir = os.path.split(src_path)[0] # The source path of the target in the host fs to be # recursively copied. target_src = os.path.join(target_dir, target) # Recursively create leading components of target dest = self._make_leading_paths(target_src, mode=mode) dest = os.path.normpath(dest) # In case symlink target is an absolute path, make it # relative to the directory with symlink source if os.path.isabs(target): target = os.path.relpath(target, target_dir) self.log_debug("Making symlink '%s' -> '%s'" % (abs_path, target)) os.symlink(target, abs_path) else: self.log_debug("Making directory %s" % abs_path) os.mkdir(abs_path, mode) dest = src_path return dest def _check_path(self, src, path_type, dest=None, force=False): """Check a new destination path in the archive. Since it is possible for multiple plugins to collect the same paths, and since plugins can now run concurrently, it is possible for two threads to race in archive methods: historically the archive class only needed to test for the actual presence of a path, since it was impossible for another `Archive` client to enter the class while another method invocation was being dispatched. Deal with this by implementing a locking scheme for operations that modify the path structure of the archive, and by testing explicitly for conflicts with any existing content at the specified destination path. It is not an error to attempt to create a path that already exists in the archive so long as the type of the object to be added matches the type of object already found at the path. It is an error to attempt to re-create an existing path with a different path type (for example, creating a symbolic link at a path already occupied by a regular file). :param src: the source path to be copied to the archive :param path_type: the type of object to be copied :param dest: an optional destination path :param force: force file creation even if the path exists :returns: An absolute destination path if the path should be copied now or `None` otherwise """ dest = dest or self.dest_path(src) if path_type == P_DIR: dest_dir = dest else: dest_dir = os.path.split(dest)[0] if not dest_dir: return dest # Check containing directory presence and path type if os.path.exists(dest_dir) and not os.path.isdir(dest_dir): raise ValueError("path '%s' exists and is not a directory" % dest_dir) elif not os.path.exists(dest_dir): src_dir = src if path_type == P_DIR else os.path.split(src)[0] self._make_leading_paths(src_dir) def is_special(mode): return any([ stat.S_ISBLK(mode), stat.S_ISCHR(mode), stat.S_ISFIFO(mode), stat.S_ISSOCK(mode) ]) if force: return dest # Check destination path presence and type if os.path.exists(dest): # Use lstat: we care about the current object, not the referent. st = os.lstat(dest) ve_msg = "path '%s' exists and is not a %s" if path_type == P_FILE and not stat.S_ISREG(st.st_mode): raise ValueError(ve_msg % (dest, "regular file")) if path_type == P_LINK and not stat.S_ISLNK(st.st_mode): raise ValueError(ve_msg % (dest, "symbolic link")) if path_type == P_NODE and not is_special(st.st_mode): raise ValueError(ve_msg % (dest, "special file")) if path_type == P_DIR and not stat.S_ISDIR(st.st_mode): raise ValueError(ve_msg % (dest, "directory")) # Path has already been copied: skip return None return dest def _copy_attributes(self, src, dest): # copy file attributes, skip SELinux xattrs for /sys and /proc try: stat = os.stat(src) if src.startswith("/sys/") or src.startswith("/proc/"): shutil.copymode(src, dest) os.utime(dest, ns=(stat.st_atime_ns, stat.st_mtime_ns)) else: shutil.copystat(src, dest) os.chown(dest, stat.st_uid, stat.st_gid) except Exception as e: self.log_debug("caught '%s' setting attributes of '%s'" % (e, dest))
[docs] def add_file(self, src, dest=None): with self._path_lock: if not dest: dest = src dest = self._check_path(dest, P_FILE) if not dest: return # Handle adding a file from either a string respresenting # a path, or a File object open for reading. if not getattr(src, "read", None): # path case try: shutil.copy(src, dest) except OSError as e: # Filter out IO errors on virtual file systems. if src.startswith("/sys/") or src.startswith("/proc/"): pass else: self.log_info("File %s not collected: '%s'" % (src, e)) self._copy_attributes(src, dest) file_name = "'%s'" % src else: # Open file case: first rewind the file to obtain # everything written to it. with open(dest, "w") as f: for line in src: f.write(line) file_name = "open file" self.log_debug("added %s to FileCacheArchive '%s'" % (file_name, self._archive_root))
[docs] def add_string(self, content, dest, mode='w'): with self._path_lock: src = dest # add_string() is a special case: it must always take precedence # over any exixting content in the archive, since it is used by # the Plugin postprocessing hooks to perform regex substitution # on file content. dest = self._check_path(dest, P_FILE, force=True) f =, mode, encoding='utf-8') if isinstance(content, bytes): content = content.decode('utf8', 'ignore') f.write(content) if os.path.exists(src): self._copy_attributes(src, dest) self.log_debug("added string at '%s' to FileCacheArchive '%s'" % (src, self._archive_root))
[docs] def add_binary(self, content, dest): with self._path_lock: dest = self._check_path(dest, P_FILE) if not dest: return f =, 'wb', encoding=None) f.write(content) self.log_debug("added binary content at '%s' to archive '%s'" % (dest, self._archive_root))
[docs] def add_dir(self, path): """Create a directory in the archive. :param path: the path in the host file system to add """ # Establish path structure with self._path_lock: self._check_path(path, P_DIR)
[docs] def add_node(self, path, mode, device): dest = self._check_path(path, P_NODE) if not dest: return if not os.path.exists(dest): try: os.mknod(dest, mode, device) except OSError as e: if e.errno == errno.EPERM: msg = "Operation not permitted" self.log_info("add_node: %s - mknod '%s'" % (msg, dest)) return raise e self._copy_attributes(path, dest)
[docs] def name_max(self): if 'PC_NAME_MAX' in os.pathconf_names: pc_name_max = os.pathconf_names['PC_NAME_MAX'] return os.pathconf(self._archive_root, pc_name_max) else: return 255
[docs] def get_tmp_dir(self): return self._archive_root
[docs] def get_archive_path(self): return self._archive_root
[docs] def makedirs(self, path, mode=0o700): """Create path, including leading components. Used by sos.sosreport to set up sos_* directories. """ os.makedirs(os.path.join(self._archive_root, path), mode=mode) self.log_debug("created directory at '%s' in FileCacheArchive '%s'" % (path, self._archive_root))
[docs] def open_file(self, path): path = self.dest_path(path) return, "r", encoding='utf-8', errors='ignore')
[docs] def cleanup(self): if os.path.isdir(self._archive_root): shutil.rmtree(self._archive_root)
[docs] def add_final_manifest_data(self, method): """Adds component-agnostic data to the manifest so that individual SoSComponents do not need to redundantly add these manually """ end = start = self.manifest.start_time run_time = end - start self.manifest.add_field('end_time', end) self.manifest.add_field('run_time', run_time) self.manifest.add_field('compression', method) self.add_string(self.manifest.get_json(indent=4), os.path.join('sos_reports', 'manifest.json'))
[docs] def rename_archive_root(self, cleaner): """Rename the archive to an obfuscated version using an initialized SoSCleaner instance """ self._name = cleaner.obfuscate_string(self._name) _new_root = os.path.join(self._tmp_dir, self._name) os.rename(self._archive_root, _new_root) self._archive_root = _new_root self._archive_name = os.path.join(self._tmp_dir,
[docs] def finalize(self, method): self.log_info("finalizing archive '%s' using method '%s'" % (self._archive_root, method)) try: res = self._build_archive(method) except Exception as err: self.log_error("An error occurred compressing the archive: %s" % err) return self.cleanup() self.log_info("built archive at '%s' (size=%d)" % (self._archive_name, os.stat(self._archive_name).st_size)) if self.enc_opts['encrypt']: try: return self._encrypt(res) except Exception as e: exp_msg = "An error occurred encrypting the archive:" self.log_error("%s %s" % (exp_msg, e)) return res else: return res
def _encrypt(self, archive): """Encrypts the compressed archive using GPG. If encryption fails for any reason, it should be logged by sos but not cause execution to stop. The assumption is that the unencrypted archive would still be of use to the user, and/or that the end user has another means of securing the archive. Returns the name of the encrypted archive, or raises an exception to signal that encryption failed and the unencrypted archive name should be used. """ arc_name = archive.replace("sosreport-", "secured-sosreport-") arc_name += ".gpg" enc_cmd = "gpg --batch -o %s " % arc_name env = None if self.enc_opts["key"]: # need to assume a trusted key here to be able to encrypt the # archive non-interactively enc_cmd += "--trust-model always -e -r %s " % self.enc_opts["key"] enc_cmd += archive if self.enc_opts["password"]: # prevent change of gpg options using a long password, but also # prevent the addition of quote characters to the passphrase passwd = "%s" % self.enc_opts["password"].replace('\'"', '') env = {"sos_gpg": passwd} enc_cmd += "-c --passphrase-fd 0 " enc_cmd = "/bin/bash -c \"echo $sos_gpg | %s\"" % enc_cmd enc_cmd += archive r = sos_get_command_output(enc_cmd, timeout=0, env=env) if r["status"] == 0: return arc_name elif r["status"] == 2: if self.enc_opts["key"]: msg = "Specified key not in keyring" else: msg = "Could not read passphrase" else: # TODO: report the actual error from gpg. Currently, we cannot as # sos_get_command_output() does not capture stderr msg = "gpg exited with code %s" % r["status"] raise Exception(msg)
[docs]class TarFileArchive(FileCacheArchive): """ archive class using python TarFile to create tar archives""" method = None _with_selinux_context = False def __init__(self, name, tmpdir, policy, threads, enc_opts, sysroot, manifest=None): super(TarFileArchive, self).__init__(name, tmpdir, policy, threads, enc_opts, sysroot, manifest) self._suffix = "tar" self._archive_name = os.path.join( tmpdir, # lgtm [py/init-calls-subclass] )
[docs] def set_tarinfo_from_stat(self, tar_info, fstat, mode=None): tar_info.mtime = fstat.st_mtime tar_info.pax_headers['atime'] = "%.9f" % fstat.st_atime tar_info.pax_headers['ctime'] = "%.9f" % fstat.st_ctime if mode: tar_info.mode = mode else: tar_info.mode = fstat.st_mode tar_info.uid = fstat.st_uid tar_info.gid = fstat.st_gid
# this can be used to set permissions if using the # tarfile.add() interface to add directory trees.
[docs] def copy_permissions_filter(self, tarinfo): orig_path =[len(os.path.split(self._name)[-1]):] if not orig_path: orig_path = self._archive_root try: fstat = os.stat(orig_path) except OSError: return tarinfo if self._with_selinux_context: context = self.get_selinux_context(orig_path) if(context): tarinfo.pax_headers[''] = context self.set_tarinfo_from_stat(tarinfo, fstat) return tarinfo
[docs] def get_selinux_context(self, path): try: (rc, c) = selinux.getfilecon(path) return c except Exception: return None
[docs] def name(self): return "%s.%s" % (self._name, self._suffix)
[docs] def name_max(self): # GNU Tar format supports unlimited file name length. Just return # the limit of the underlying FileCacheArchive. return super(TarFileArchive, self).name_max()
def _build_archive(self, method): if method == 'auto': method = 'xz' if find_spec('lzma') is not None else 'gzip' _comp_mode = method.strip('ip') self._archive_name = self._archive_name + ".%s" % _comp_mode # tarfile does not currently have a consistent way to define comnpress # level for both xz and gzip ('preset' for xz, 'compresslevel' for gz) if method == 'gzip': kwargs = {'compresslevel': 6} else: kwargs = {'preset': 3} tar =, mode="w:%s" % _comp_mode, **kwargs) # we need to pass the absolute path to the archive root but we # want the names used in the archive to be relative. tar.add(self._archive_root, arcname=os.path.split(self._name)[1], filter=self.copy_permissions_filter) tar.close() self._suffix += ".%s" % _comp_mode return
# vim: set et ts=4 sw=4 :