Current Path : /usr/lib/python2.7/dist-packages/apport/ |
Current File : //usr/lib/python2.7/dist-packages/apport/report.py |
'''Representation of and data collection for a problem report.''' # Copyright (C) 2006 - 2009 Canonical Ltd. # Author: Martin Pitt <martin.pitt@ubuntu.com> # # This program is free software; you can redistribute it and/or modify it # under the terms of the GNU General Public License as published by the # Free Software Foundation; either version 2 of the License, or (at your # option) any later version. See http://www.gnu.org/copyleft/gpl.html for # the full text of the license. import subprocess, tempfile, os.path, urllib, re, pwd, grp, os import fnmatch, glob, traceback, errno, sys, imp import xml.dom, xml.dom.minidom from xml.parsers.expat import ExpatError import problem_report import apport import apport.fileutils from apport.packaging_impl import impl as packaging _data_dir = os.environ.get('APPORT_DATA_DIR','/usr/share/apport') _hook_dir = '%s/package-hooks/' % (_data_dir) _common_hook_dir = '%s/general-hooks/' % (_data_dir) # path of the ignore file _ignore_file = '~/.apport-ignore.xml' # system-wide blacklist _blacklist_dir = '/etc/apport/blacklist.d' _whitelist_dir = '/etc/apport/whitelist.d' # programs that we consider interpreters interpreters = ['sh', 'bash', 'dash', 'csh', 'tcsh', 'python*', 'ruby*', 'php', 'perl*', 'mono*', 'awk'] # # helper functions # def _transitive_dependencies(package, depends_set): '''Recursively add dependencies of package to depends_set.''' try: packaging.get_version(package) except ValueError: return for d in packaging.get_dependencies(package): if not d in depends_set: depends_set.add(d) _transitive_dependencies(d, depends_set) def _read_file(path): '''Read file content. Return its content, or return a textual error if it failed. ''' try: with open(path) as fd: return fd.read().strip() except (OSError, IOError) as e: return 'Error: ' + str(e) def _read_maps(pid): '''Read /proc/pid/maps. Since /proc/$pid/maps may become unreadable unless we are ptracing the process, detect this, and attempt to attach/detach. ''' maps = 'Error: unable to read /proc maps file' try: with open('/proc/%d/maps' % pid) as fd: maps = fd.read().strip() except (OSError,IOError) as e: return 'Error: ' + str(e) return maps def _command_output(command, input = None, stderr = subprocess.STDOUT): '''Run command and capture its output. Try to execute given command (argv list) and return its stdout, or return a textual error if it failed. ''' sp = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=stderr, close_fds=True) (out, err) = sp.communicate(input) if sp.returncode == 0: return out else: if err: err = err.decode('UTF-8', errors='replace') else: err = '' raise OSError('Error: command %s failed with exit code %i: %s' % ( str(command), sp.returncode, err)) def _check_bug_pattern(report, pattern): '''Check if given report matches the given bug pattern XML DOM node. Return the bug URL on match, otherwise None. ''' if not pattern.attributes.has_key('url'): return None for c in pattern.childNodes: # regular expression condition if c.nodeType == xml.dom.Node.ELEMENT_NODE and c.nodeName == 're' and \ c.attributes.has_key('key'): key = c.attributes['key'].nodeValue if key not in report: return None c.normalize() if c.hasChildNodes() and \ c.childNodes[0].nodeType == xml.dom.Node.TEXT_NODE: regexp = c.childNodes[0].nodeValue.encode('UTF-8') try: v = report[key] if isinstance(v, problem_report.CompressedValue): v = v.get_value() if not re.search(regexp, v): return None except: return None return pattern.attributes['url'].nodeValue.encode('UTF-8') def _check_bug_patterns(report, patterns): try: dom = xml.dom.minidom.parseString(patterns) except ExpatError: return None for pattern in dom.getElementsByTagName('pattern'): url = _check_bug_pattern(report, pattern) if url: return url return None def _dom_remove_space(node): '''Recursively remove whitespace from given XML DOM node.''' for c in node.childNodes: if c.nodeType == xml.dom.Node.TEXT_NODE and c.nodeValue.strip() == '': c.unlink() node.removeChild(c) else: _dom_remove_space(c) # # Report class # class Report(problem_report.ProblemReport): '''A problem report specific to apport (crash or bug). This class wraps a standard ProblemReport and adds methods for collecting standard debugging data.''' def __init__(self, type='Crash', date=None): '''Initialize a fresh problem report. date is the desired date/time string; if None (default), the current local time is used. If the report is attached to a process ID, this should be set in self.pid, so that e. g. hooks can use it to collect additional data. ''' problem_report.ProblemReport.__init__(self, type, date) self.pid = None self._proc_maps_cache = None def _customized_package_suffix(self, package): '''Return a string suitable for appending to Package/Dependencies. If package has only unmodified files, return the empty string. If not, return ' [modified: ...]' with a list of modified files. ''' suffix = '' mod = packaging.get_modified_files(package) if mod: suffix += ' [modified: %s]' % ' '.join(mod) try: if not packaging.is_distro_package(package): origin = packaging.get_package_origin(package) if origin: suffix += ' [origin: %s]' % origin else: suffix += ' [origin: unknown]' except ValueError: # no-op for nonexisting packages pass return suffix def add_package_info(self, package = None): '''Add packaging information. If package is not given, the report must have ExecutablePath. This adds: - Package: package name and installed version - SourcePackage: source package name - PackageArchitecture: processor architecture this package was built for - Dependencies: package names and versions of all dependencies and pre-dependencies; this also checks if the files are unmodified and appends a list of all modified files ''' if not package: # the kernel does not have a executable path but a package if (not 'ExecutablePath' in self and self['ProblemType'] == 'KernelCrash'): package = self['Package'] else: package = apport.fileutils.find_file_package(self['ExecutablePath']) if not package: return try: version = packaging.get_version(package) except ValueError: # package not installed version = None self['Package'] = '%s %s%s' % (package, version or '(not installed)', self._customized_package_suffix(package)) if version or 'SourcePackage' not in self: self['SourcePackage'] = packaging.get_source(package) if not version: return self['PackageArchitecture'] = packaging.get_architecture(package) # get set of all transitive dependencies dependencies = set([]) _transitive_dependencies(package, dependencies) # get dependency versions self['Dependencies'] = '' for dep in sorted(dependencies): try: v = packaging.get_version(dep) except ValueError: # can happen with uninstalled alternate dependencies continue if self['Dependencies']: self['Dependencies'] += '\n' self['Dependencies'] += '%s %s%s' % (dep, v, self._customized_package_suffix(dep)) def add_os_info(self): '''Add operating system information. This adds: - DistroRelease: lsb_release -sir output - Architecture: system architecture in distro specific notation - Uname: uname -srm output - NonfreeKernelModules: loaded kernel modules which are not free (if there are none, this field will not be present) ''' p = subprocess.Popen(['lsb_release', '-sir'], stdout=subprocess.PIPE, stderr=subprocess.PIPE, close_fds=True) self['DistroRelease'] = p.communicate()[0].decode().strip().replace('\n', ' ') u = os.uname() self['Uname'] = '%s %s %s' % (u[0], u[2], u[4]) self['Architecture'] = packaging.get_system_architecture() def add_user_info(self): '''Add information about the user. This adds: - UserGroups: system groups the user is in ''' user = pwd.getpwuid(os.getuid()).pw_name groups = [name for name, p, gid, memb in grp.getgrall() if user in memb and gid < 1000] groups.sort() self['UserGroups'] = ' '.join(groups) def _check_interpreted(self): '''Check if process is a script. Use ExecutablePath, ProcStatus and ProcCmdline to determine if process is an interpreted script. If so, set InterpreterPath accordingly. ''' if 'ExecutablePath' not in self: return exebasename = os.path.basename(self['ExecutablePath']) # check if we consider ExecutablePath an interpreter; we have to do # this, otherwise 'gedit /tmp/foo.txt' would be detected as interpreted # script as well if not filter(lambda i: fnmatch.fnmatch(exebasename, i), interpreters): return # first, determine process name name = None for l in self['ProcStatus'].splitlines(): try: (k, v) = l.split('\t', 1) except ValueError: continue if k == 'Name:': name = v break if not name: return cmdargs = self['ProcCmdline'].split('\0') bindirs = ['/bin/', '/sbin/', '/usr/bin/', '/usr/sbin/'] # filter out interpreter options while len(cmdargs) >= 2 and cmdargs[1].startswith('-'): # check for -m if name.startswith('python') and cmdargs[1] == '-m' and len(cmdargs) >= 3: path = self._python_module_path(cmdargs[2]) if path: self['InterpreterPath'] = self['ExecutablePath'] self['ExecutablePath'] = path else: self['UnreportableReason'] = 'Cannot determine path of python module %s' % cmdargs[2] return del cmdargs[1] # catch scripts explicitly called with interpreter if len(cmdargs) >= 2: # ensure that cmdargs[1] is an absolute path if cmdargs[1].startswith('.') and 'ProcCwd' in self: cmdargs[1] = os.path.join(self['ProcCwd'], cmdargs[1]) if os.access(cmdargs[1], os.R_OK): self['InterpreterPath'] = self['ExecutablePath'] self['ExecutablePath'] = os.path.realpath(cmdargs[1]) # catch directly executed scripts if 'InterpreterPath' not in self and name != exebasename: argvexes = filter(lambda p: os.access(p, os.R_OK), [p+cmdargs[0] for p in bindirs]) if argvexes and os.path.basename(os.path.realpath(argvexes[0])) == name: self['InterpreterPath'] = self['ExecutablePath'] self['ExecutablePath'] = argvexes[0] # special case: crashes from twistd are usually the fault of the # launched program if 'InterpreterPath' in self and os.path.basename(self['ExecutablePath']) == 'twistd': self['InterpreterPath'] = self['ExecutablePath'] exe = self._twistd_executable() if exe: self['ExecutablePath'] = exe else: self['UnreportableReason'] = 'Cannot determine twistd client program' def _twistd_executable(self): '''Determine the twistd client program from ProcCmdline.''' args = self['ProcCmdline'].split('\0')[2:] # search for a -f/--file, -y/--python or -s/--source argument while args: arg = args[0].split('=', 1) if arg[0].startswith('--file') or arg[0].startswith('--python') or \ arg[0].startswith('--source'): if len(arg) == 2: return arg[1] else: return args[1] elif len(arg[0]) > 1 and arg[0][0] == '-' and arg[0][1] != '-': opts = arg[0][1:] if 'f' in opts or 'y' in opts or 's' in opts: return args[1] args.pop(0) return None @classmethod def _python_module_path(klass, module): '''Determine path of given Python module''' module = module.replace('/', '.').split('.') pathlist = sys.path path = None while module: name = module.pop(0) try: (fd, path, desc) = imp.find_module(name, pathlist) except ImportError: path = None break if fd: fd.close() pathlist = [path] if not module and desc[2] == imp.PKG_DIRECTORY: module = ['__init__'] if path and path.endswith('.pyc'): path = path[:-1] return path def add_proc_info(self, pid=None, extraenv=[]): '''Add /proc/pid information. If neither pid nor self.pid are given, it defaults to the process' current pid and sets self.pid. This adds the following fields: - ExecutablePath: /proc/pid/exe contents; if the crashed process is interpreted, this contains the script path instead - InterpreterPath: /proc/pid/exe contents if the crashed process is interpreted; otherwise this key does not exist - ExecutableTimestamp: time stamp of ExecutablePath, for comparing at report time - ProcEnviron: A subset of the process' environment (only some standard variables that do not disclose potentially sensitive information, plus the ones mentioned in extraenv) - ProcCmdline: /proc/pid/cmdline contents - ProcStatus: /proc/pid/status contents - ProcMaps: /proc/pid/maps contents - ProcAttrCurrent: /proc/pid/attr/current contents, if not "unconfined" ''' if not pid: pid = self.pid or os.getpid() if not self.pid: self.pid = int(pid) pid = str(pid) try: self['ProcCwd'] = os.readlink('/proc/' + pid + '/cwd') except OSError: pass self.add_proc_environ(pid, extraenv) self['ProcStatus'] = _read_file('/proc/' + pid + '/status') self['ProcCmdline'] = _read_file('/proc/' + pid + '/cmdline').rstrip('\0') self['ProcMaps'] = _read_maps(int(pid)) try: self['ExecutablePath'] = os.readlink('/proc/' + pid + '/exe') except OSError as e: if e.errno == errno.ENOENT: raise ValueError('invalid process') else: raise for p in ('rofs', 'rwfs', 'squashmnt', 'persistmnt'): if self['ExecutablePath'].startswith('/%s/' % p): self['ExecutablePath'] = self['ExecutablePath'][len('/%s' % p):] break assert os.path.exists(self['ExecutablePath']) # check if we have an interpreted program self._check_interpreted() self['ExecutableTimestamp'] = str(int(os.stat(self['ExecutablePath']).st_mtime)) # make ProcCmdline ASCII friendly, do shell escaping self['ProcCmdline'] = self['ProcCmdline'].replace('\\', '\\\\').replace(' ', '\\ ').replace('\0', ' ') # grab AppArmor or SELinux context # If no LSM is loaded, reading will return -EINVAL try: # On Linux 2.6.28+, 'current' is world readable, but read() gives # EPERM; Python 2.5.3+ crashes on that (LP: #314065) if os.getuid() == 0: with open('/proc/' + pid + '/attr/current') as fd: val = fd.read().strip() if val != 'unconfined': self['ProcAttrCurrent'] = val except (IOError, OSError): pass def add_proc_environ(self, pid=None, extraenv=[]): '''Add environment information. If pid is not given, it defaults to the process' current pid. This adds the following fields: - ProcEnviron: A subset of the process' environment (only some standard variables that do not disclose potentially sensitive information, plus the ones mentioned in extraenv) ''' safe_vars = ['SHELL', 'TERM', 'LANGUAGE', 'LANG', 'LC_CTYPE', 'LC_COLLATE', 'LC_TIME', 'LC_NUMERIC', 'LC_MONETARY', 'LC_MESSAGES', 'LC_PAPER', 'LC_NAME', 'LC_ADDRESS', 'LC_TELEPHONE', 'LC_MEASUREMENT', 'LC_IDENTIFICATION', 'LOCPATH'] + extraenv if not pid: pid = os.getpid() pid = str(pid) self['ProcEnviron'] = '' env = _read_file('/proc/'+ pid + '/environ').replace('\n', '\\n') if env.startswith('Error:'): self['ProcEnviron'] = env else: for l in env.split('\0'): if l.split('=', 1)[0] in safe_vars: if self['ProcEnviron']: self['ProcEnviron'] += '\n' self['ProcEnviron'] += l elif l.startswith('PATH='): p = l.split('=', 1)[1] if '/home' in p or '/tmp' in p: if self['ProcEnviron']: self['ProcEnviron'] += '\n' self['ProcEnviron'] += 'PATH=(custom, user)' elif p != '/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games': if self['ProcEnviron']: self['ProcEnviron'] += '\n' self['ProcEnviron'] += 'PATH=(custom, no user)' def add_kernel_crash_info(self, debugdir=None): '''Add information from kernel crash. This needs a VmCore in the Report. ''' if 'VmCore' not in self: return unlink_core = False ret = False try: if hasattr(self['VmCore'], 'find'): (fd, core) = tempfile.mkstemp() os.write(fd, self['VmCore']) os.close(fd) unlink_core = True kver = self['Uname'].split()[1] command = ['crash', '/usr/lib/debug/boot/vmlinux-%s' % kver, core, ] try: p = subprocess.Popen(command, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.STDOUT) except OSError: return False p.stdin.write('bt -a -f\n') p.stdin.write('ps\n') p.stdin.write('runq\n') p.stdin.write('quit\n') # FIXME: split it up nicely etc out = p.stdout.read() ret = (p.wait() == 0) if ret: self['Stacktrace'] = out finally: if unlink_core: os.unlink(core) return ret def add_gdb_info(self, rootdir=None): '''Add information from gdb. This requires that the report has a CoreDump and an ExecutablePath. This adds the following fields: - Registers: Output of gdb's 'info registers' command - Disassembly: Output of gdb's 'x/16i $pc' command - Stacktrace: Output of gdb's 'bt full' command - ThreadStacktrace: Output of gdb's 'thread apply all bt full' command - StacktraceTop: simplified stacktrace (topmost 5 functions) for inline inclusion into bug reports and easier processing - AssertionMessage: Value of __abort_msg, if present The optional rootdir can specify a root directory which has the executable, libraries, and debug symbols. This does not require chroot() or root privileges, it just instructs gdb to search for the files there. Raises a IOError if the core dump is invalid/truncated. ''' if 'CoreDump' not in self or 'ExecutablePath' not in self: return unlink_core = False try: if hasattr(self['CoreDump'], 'find'): (fd, core) = tempfile.mkstemp() unlink_core = True os.write(fd, self['CoreDump']) os.close(fd) elif hasattr(self['CoreDump'], 'gzipvalue'): (fd, core) = tempfile.mkstemp() unlink_core = True os.close(fd) self['CoreDump'].write(open(core, 'w')) else: core = self['CoreDump'][0] gdb_reports = { 'Registers': 'info registers', 'Disassembly': 'x/16i $pc', 'Stacktrace': 'bt full', 'ThreadStacktrace': 'thread apply all bt full', 'AssertionMessage': 'print __abort_msg->msg', } command = ['gdb', '--batch'] executable = self.get('InterpreterPath', self['ExecutablePath']) if rootdir: command += ['--ex', 'set debug-file-directory %s/usr/lib/debug' % rootdir, '--ex', 'set solib-absolute-prefix ' + rootdir] executable = rootdir + '/' + executable command += ['--ex', 'file ' + executable, '--ex', 'core-file ' + core] # limit maximum backtrace depth (to avoid looped stacks) command += ['--ex', 'set backtrace limit 2000'] value_keys = [] # append the actual commands and something that acts as a separator for name, cmd in gdb_reports.items(): value_keys.append(name) command += ['--ex', 'p -99', '--ex', cmd] assert os.path.exists(executable) # call gdb try: out = _command_output(command).decode('UTF-8', errors='replace') except OSError: return # check for truncated stack trace if 'is truncated: expected core file size' in out or \ 'not a core dump: File format not recognized' in out: warnings = '\n'.join([l for l in out.splitlines() if 'Warning:' in l]) reason = 'Invalid core dump: ' + warnings.strip() self['UnreportableReason'] = reason raise IOError(reason) # split the output into the various fields part_re = re.compile('^\$\d+\s*=\s*-99$', re.MULTILINE) parts = part_re.split(out) # drop the gdb startup text prior to first separator parts.pop(0) for part in parts: self[value_keys.pop(0)] = part.replace('\n\n', '\n.\n').strip() finally: if unlink_core: os.unlink(core) # clean up AssertionMessage if 'AssertionMessage' in self: # chop off "$n = 0x...." prefix, drop empty ones m = re.match('^\$\d+\s+=\s+0x[0-9a-fA-F]+\s+"(.*)"\s*$', self['AssertionMessage']) if m: self['AssertionMessage'] = m.group(1) if self['AssertionMessage'].endswith('\\n'): self['AssertionMessage'] = self['AssertionMessage'][0:-2] else: del self['AssertionMessage'] if 'Stacktrace' in self: self._gen_stacktrace_top() addr_signature = self.crash_signature_addresses() if addr_signature: self['StacktraceAddressSignature'] = addr_signature def _gen_stacktrace_top(self): '''Build field StacktraceTop as the top five functions of Stacktrace. Signal handler invocations and related functions are skipped since they are generally not useful for triaging and duplicate detection. ''' unwind_functions = set(['g_logv', 'g_log', 'IA__g_log', 'IA__g_logv', 'g_assert_warning', 'IA__g_assert_warning', '__GI_abort', '_XError']) toptrace = [''] * 5 depth = 0 unwound = False unwinding = False unwinding_xerror = False bt_fn_re = re.compile('^#(\d+)\s+(?:0x(?:\w+)\s+in\s+\*?(.*)|(<signal handler called>)\s*)$') bt_fn_noaddr_re = re.compile('^#(\d+)\s+(?:(.*)|(<signal handler called>)\s*)$') # some internal functions like the SSE stubs cause unnecessary jitter ignore_functions_re = re.compile('^(__.*_s?sse\d+(?:_\w+)?|__kernel_vsyscall)$') for line in self['Stacktrace'].splitlines(): m = bt_fn_re.match(line) if not m: m = bt_fn_noaddr_re.match(line) if not m: continue if not unwound or unwinding: if m.group(2): fn = m.group(2).split()[0].split('(')[0] else: fn = None # handle XErrors if unwinding_xerror: if fn.startswith('_X') or fn in ['handle_response', 'handle_error', 'XWindowEvent']: continue else: unwinding_xerror = False if m.group(3) or fn in unwind_functions: unwinding = True depth = 0 toptrace = [''] * 5 if m.group(3): # we stop unwinding when we found a <signal handler>, # but we continue unwinding otherwise, as e. g. a glib # abort is usually sitting on top of an XError unwound = True if fn == '_XError': unwinding_xerror = True continue else: unwinding = False frame = m.group(2) or m.group(3) function = frame.split()[0] if depth < len(toptrace) and not ignore_functions_re.match(function): toptrace[depth] = frame depth += 1 self['StacktraceTop'] = '\n'.join(toptrace).strip() def add_hooks_info(self, ui, package=None, srcpackage=None): '''Run hook script for collecting package specific data. A hook script needs to be in _hook_dir/<Package>.py or in _common_hook_dir/*.py and has to contain a function 'add_info(report, ui)' that takes and modifies a Report, and gets an UserInterface reference for interactivity. return True if the hook requested to stop the report filing process, False otherwise. ''' symb = {} # common hooks for hook in glob.glob(_common_hook_dir + '/*.py'): try: with open(hook) as fd: exec(compile(fd.read(), hook, 'exec'), symb) try: symb['add_info'](self, ui) except TypeError as e: if str(e).startswith('add_info()'): # older versions of apport did not pass UI, and hooks that # do not require it don't need to take it symb['add_info'](self) else: raise except StopIteration: return True except: apport.error('hook %s crashed:', hook) traceback.print_exc() pass # binary package hook if not package: package = self.get('Package') if package: package = package.split()[0] if '/' in package: self['UnreportableReason'] = 'invalid Package: %s' % package return hook = '%s/%s.py' % (_hook_dir, package) if os.path.exists(hook): try: with open(hook) as fd: exec(compile(fd.read(), hook, 'exec'), symb) try: symb['add_info'](self, ui) except TypeError as e: if str(e).startswith('add_info()'): # older versions of apport did not pass UI, and hooks that # do not require it don't need to take it symb['add_info'](self) else: raise except StopIteration: return True except: apport.error('hook %s crashed:', hook) traceback.print_exc() pass # source package hook if not srcpackage: srcpackage = self.get('SourcePackage') if srcpackage: srcpackage = srcpackage.split()[0] if '/' in srcpackage: self['UnreportableReason'] = 'invalid SourcePackage: %s' % package return hook = '%s/source_%s.py' % (_hook_dir, srcpackage) if os.path.exists(hook): try: with open(hook) as fd: exec(compile(fd.read(), hook, 'exec'), symb) try: symb['add_info'](self, ui) except TypeError as e: if str(e).startswith('add_info()'): # older versions of apport did not pass UI, and hooks that # do not require it don't need to take it symb['add_info'](self) else: raise except StopIteration: return True except: apport.error('hook %s crashed:', hook) traceback.print_exc() pass return False def search_bug_patterns(self, url): '''Check bug patterns loaded from the specified url. Return bug URL on match, or None otherwise. The url must refer to a valid XML document with the following syntax: root element := <patterns> patterns := <pattern url="http://bug.url"> * pattern := <re key="report_key">regular expression*</re> + For example: <?xml version="1.0"?> <patterns> <pattern url="http://bugtracker.net/bugs/1"> <re key="Foo">ba.*r</re> </pattern> <pattern url="http://bugtracker.net/bugs/2"> <re key="Package">^\S* 1-2$</re> <!-- test for a particular version --> <re key="Foo">write_(hello|goodbye)</re> </pattern> </patterns> ''' # some distros might not want to support these if not url: return try: patterns = urllib.urlopen(url).read() except: # doesn't exist or failed to load return if '<title>404 Not Found' in patterns: return url = _check_bug_patterns(self, patterns) if url: return url return None def _get_ignore_dom(self): '''Read ignore list XML file and return a DOM tree. Return an empty DOM tree if file does not exist. Raises ValueError if the file exists but is invalid XML. ''' ifpath = os.path.expanduser(_ignore_file) if not os.access(ifpath, os.R_OK) or os.path.getsize(ifpath) == 0: # create a document from scratch dom = xml.dom.getDOMImplementation().createDocument(None, 'apport', None) else: try: dom = xml.dom.minidom.parse(ifpath) except ExpatError as e: raise ValueError('%s has invalid format: %s' % (_ignore_file, str(e))) # remove whitespace so that writing back the XML does not accumulate # whitespace dom.documentElement.normalize() _dom_remove_space(dom.documentElement) return dom def check_ignored(self): '''Check if current report should not be presented. Reports can be suppressed by per-user blacklisting in ~/.apport-ignore.xml (in the real UID's home) and /etc/apport/blacklist.d/. For environments where you are only interested in crashes of some programs, you can also create a whitelist in /etc/apport/whitelist.d/, everything which does not match gets ignored then. This requires the ExecutablePath attribute. Throws a ValueError if the file has an invalid format. ''' assert 'ExecutablePath' in self # check blacklist try: for f in os.listdir(_blacklist_dir): try: with open(os.path.join(_blacklist_dir, f)) as fd: for line in fd: if line.strip() == self['ExecutablePath']: return True except IOError: continue except OSError: pass # check whitelist try: whitelist = set() for f in os.listdir(_whitelist_dir): try: with open(os.path.join(_whitelist_dir, f)) as fd: for line in fd: whitelist.add(line.strip()) except IOError: continue if whitelist and self['ExecutablePath'] not in whitelist: return True except OSError: pass dom = self._get_ignore_dom() try: cur_mtime = int(os.stat(self['ExecutablePath']).st_mtime) except OSError: # if it does not exist any more, do nothing return False # search for existing entry and update it for ignore in dom.getElementsByTagName('ignore'): if ignore.getAttribute('program') == self['ExecutablePath']: if float(ignore.getAttribute('mtime')) >= cur_mtime: return True return False def mark_ignore(self): '''Ignore future crashes of this executable. Add a ignore list entry for this report to ~/.apport-ignore.xml, so that future reports for this ExecutablePath are not presented to the user any more. Throws a ValueError if the file already exists and has an invalid format. ''' assert 'ExecutablePath' in self dom = self._get_ignore_dom() try: mtime = str(int(os.stat(self['ExecutablePath']).st_mtime)) except OSError as e: # file went away underneath us, ignore if e.errno == errno.ENOENT: return else: raise # search for existing entry and update it for ignore in dom.getElementsByTagName('ignore'): if ignore.getAttribute('program') == self['ExecutablePath']: ignore.setAttribute('mtime', mtime) break else: # none exists yet, create new ignore node if none exists yet e = dom.createElement('ignore') e.setAttribute('program', self['ExecutablePath']) e.setAttribute('mtime', mtime) dom.documentElement.appendChild(e) # write back file with open(os.path.expanduser(_ignore_file), 'w') as fd: dom.writexml(fd, addindent=' ', newl='\n') dom.unlink() def has_useful_stacktrace(self): '''Check whether StackTrace can be considered 'useful'. The current heuristic is to consider it useless if it either is shorter than three lines and has any unknown function, or for longer traces, a minority of known functions. ''' if not self.get('StacktraceTop'): return False unknown_fn = [f.startswith('??') for f in self['StacktraceTop'].splitlines()] if len(unknown_fn) < 3: return unknown_fn.count(True) == 0 return unknown_fn.count(True) <= len(unknown_fn)/2. def stacktrace_top_function(self): '''Return topmost function in StacktraceTop''' for l in self.get('StacktraceTop', '').splitlines(): fname = l.split('(')[0].strip() if fname != '??': return fname return None def standard_title(self): '''Create an appropriate title for a crash database entry. This contains the topmost function name from the stack trace and the signal (for signal crashes) or the Python exception (for unhandled Python exceptions). Return None if the report is not a crash or a default title could not be generated. ''' # assertion failure if self.get('Signal') == '6' and \ 'ExecutablePath' in self and \ 'AssertionMessage' in self: return '%s assert failure: %s' % ( os.path.basename(self['ExecutablePath']), self['AssertionMessage']) # signal crash if 'Signal' in self and \ 'ExecutablePath' in self and \ 'StacktraceTop' in self: signal_names = { '4': 'SIGILL', '6': 'SIGABRT', '8': 'SIGFPE', '11': 'SIGSEGV', '13': 'SIGPIPE' } fn = self.stacktrace_top_function() if fn: fn = ' in %s()' % fn else: fn = '' arch_mismatch = '' if 'Architecture' in self and \ 'PackageArchitecture' in self and \ self['Architecture'] != self['PackageArchitecture'] and \ self['PackageArchitecture'] != 'all': arch_mismatch = ' [non-native %s package]' % self['PackageArchitecture'] return '%s crashed with %s%s%s' % ( os.path.basename(self['ExecutablePath']), signal_names.get(self.get('Signal'), 'signal ' + self.get('Signal')), fn, arch_mismatch ) # Python exception if 'Traceback' in self and \ 'ExecutablePath' in self: trace = self['Traceback'].splitlines() if len(trace) < 1: return None if len(trace) < 3: return '%s crashed with %s' % ( os.path.basename(self['ExecutablePath']), trace[0]) trace_re = re.compile('^\s*File\s*"(\S+)".* in (.+)$') i = len(trace)-1 function = 'unknown' while i >= 0: m = trace_re.match(trace[i]) if m: module_path = m.group(1) function = m.group(2) break i -= 1 path = os.path.basename(self['ExecutablePath']) last_line = trace[-1] exception = last_line.split(':')[0] m = re.match('^%s: (.+)$' % re.escape(exception), last_line) if m: message = m.group(1) else: message = None if function == '<module>': if module_path == self['ExecutablePath']: context = '__main__' else: # Maybe use os.path.basename? context = module_path else: context = '%s()' % function title = '%s crashed with %s in %s' % ( path, exception, context ) if message: title += ': %s' % message return title # package problem if self.get('ProblemType') == 'Package' and \ 'Package' in self: title = 'package %s failed to install/upgrade' % \ self['Package'] if self.get('ErrorMessage'): title += ': ' + self['ErrorMessage'].splitlines()[-1] return title if self.get('ProblemType') == 'KernelOops' and \ 'OopsText' in self: oops = self['OopsText'] if oops.startswith('------------[ cut here ]------------'): title = oops.split('\n', 2)[1] else: title = oops.split('\n', 1)[0] return title if self.get('ProblemType') == 'KernelOops' and \ 'Failure' in self: # Title the report with suspend or hibernate as appropriate, # and mention any non-free modules loaded up front. title = '' if 'MachineType' in self: title += '[' + self['MachineType'] + '] ' title += self['Failure'] + ' failure' if 'NonfreeKernelModules' in self: title += ' [non-free: ' + self['NonfreeKernelModules'] + ']' title += '\n' return title return None def obsolete_packages(self): '''Return list of obsolete packages in Package and Dependencies.''' obsolete = [] for l in (self.get('Package', '') + '\n' + self.get('Dependencies', '')).splitlines(): if not l: continue pkg, ver = l.split()[:2] avail = packaging.get_available_version(pkg) if ver != None and ver != 'None' and avail != None and \ packaging.compare_versions(ver, avail) < 0: obsolete.append(pkg) return obsolete def crash_signature(self): '''Get a signature string for a crash. This is suitable for identifying duplicates. For signal crashes this the concatenation of ExecutablePath, Signal number, and StacktraceTop function names, separated by a colon. If StacktraceTop has unknown functions or the report lacks any of those fields, return None. In this case, you can use crash_signature_addresses() to get a less precise duplicate signature based on addresses instead of symbol names. For assertion failures, it is the concatenation of ExecutablePath and assertion message, separated by colons. For Python crashes, this concatenates the ExecutablePath, exception name, and Traceback function names, again separated by a colon. ''' if ('ExecutablePath' not in self and not self['ProblemType'] == 'KernelCrash'): return None # kernel crash if 'Stacktrace' in self and self['ProblemType'] == 'KernelCrash': sig = 'kernel' regex = re.compile ('^\s*\#\d+\s\[\w+\]\s(\w+)') for line in self['Stacktrace'].splitlines(): m = regex.match(line) if m: sig += ':' + (m.group(1)) return sig # assertion failures if self.get('Signal') == '6' and 'AssertionMessage' in self: sig = self['ExecutablePath'] + ':' + self['AssertionMessage'] # filter out addresses, to help match duplicates more sanely return re.sub(r'0x[0-9a-f]{6,}','ADDR', sig) # signal crashes if 'StacktraceTop' in self and 'Signal' in self: sig = '%s:%s' % (self['ExecutablePath'], self['Signal']) bt_fn_re = re.compile('^(?:([\w:~]+).*|(<signal handler called>)\s*)$') lines = self['StacktraceTop'].splitlines() if len(lines) < 2: return None for line in lines: m = bt_fn_re.match(line) if m: sig += ':' + (m.group(1) or m.group(2)) else: # this will also catch ?? return None return sig # Python crashes if 'Traceback' in self: trace = self['Traceback'].splitlines() sig = '' if len(trace) == 1: # sometimes, Python exceptions do not have file references m = re.match('(\w+): ', trace[0]) if m: return self['ExecutablePath'] + ':' + m.group(1) else: return None elif len(trace) < 3: return None loc_re = re.compile ('^\s+File "([^"]+).*line (\d+).*\sin (.*)$') for l in trace: m = loc_re.match(l) if m: # if we have a function name, use this; for a a crash # outside of a function/method, fall back to the source # file location if m.group(3) != '<module>': sig += ':' + m.group(3) else: sig += ':%s@%s' % (m.group(1), m.group(2)) return self['ExecutablePath'] + ':' + trace[-1].split(':')[0] + sig return None def crash_signature_addresses(self): '''Compute heuristic duplicate signature for a signal crash. This should be used if crash_signature() fails, i. e. Stacktrace does not have enough symbols. This approach only uses addresses in the stack trace and does not rely on symbol resolution. As we can't unwind these stack traces, we cannot limit them to the top five frames and thus will end up with several or many different signatures for a particular crash. But these can be computed and synchronously checked with a crash database at the client side, which avoids having to upload and process the full report. So on the server-side crash database we will only have to deal with all the equivalence classes (i. e. same crash producing a number of possible signatures) instead of every single report. Return None when signature cannot be determined. ''' if not 'ProcMaps' in self or not 'Stacktrace' in self or not 'Signal' in self: return None stack = [] failed = 0 for line in self['Stacktrace'].splitlines(): if line.startswith('#'): addr = line.split()[1] if not addr.startswith('0x'): continue addr = int(addr, 16) # we do want to know about ValueErrors here, so don't catch # ignore impossibly low addresses; these are usually artifacts # from gdb when not having debug symbols if addr < 0x1000: continue offset = self._address_to_offset(addr) if offset: # avoid ':' in ELF paths, we use that as separator stack.append(offset.replace(':', '..')) else: failed += 1 # stack unwinding chops off ~ 5 functions, and we need some more # accuracy because we do not have symbols; but beyond a depth of 15 # we get too much noise, so we can abort there if len(stack) >= 15: break # we only accept a small minority (< 20%) of failed resolutions, otherwise we # discard if failed > 0 and len(stack)/failed < 4: return None # we also discard if the trace is too short if (failed == 0 and len(stack) < 3) or (failed > 0 and len(stack) < 6): return None return '%s:%s:%s:%s' % ( self['ExecutablePath'], self['Signal'], os.uname()[4], ':'.join(stack)) def anonymize(self): '''Remove user identifying strings from the report. This particularly removes the user name, host name, and IPs from attributes which contain data read from the environment, and removes the ProcCwd attribute completely. ''' replacements = [] if (os.getuid() > 0): # do not replace "root" p = pwd.getpwuid(os.getuid()) if len(p[0]) >= 2: replacements.append((re.compile('\\b%s\\b' % p[0]), 'username')) replacements.append((re.compile('\\b%s\\b' % p[5]), '/home/username')) for s in p[4].split(','): s = s.strip() if len(s) > 2: replacements.append((re.compile('\\b%s\\b' % s), 'User Name')) hostname = os.uname()[1] if len(hostname) >= 2: replacements.append((re.compile('\\b%s\\b' % hostname), 'hostname')) try: del self['ProcCwd'] except KeyError: pass for k in self: if (k.startswith('Proc') and \ not k in ['ProcCpuinfo','ProcMaps','ProcStatus', \ 'ProcInterrupts','ProcModules']) or \ 'Stacktrace' in k or \ k in ['Traceback', 'PythonArgs', 'Title']: if not hasattr(self[k], 'isspace'): continue for (pattern, repl) in replacements: if type(self[k]) == type(b''): self[k] = pattern.sub(repl, self[k]) else: self[k] = pattern.sub(repl, self[k].encode('UTF-8')).decode('UTF-8') def _address_to_offset(self, addr): '''Resolve a memory address to an ELF name and offset. This can be used for building duplicate signatures from non-symbolic stack traces. These often do not have enough symbols available to resolve function names, but taking the raw addresses also is not suitable due to ASLR. But the offsets within a library should be constant between crashes (assuming the same version of all libraries). This needs and uses the "ProcMaps" field to resolve addresses. Return 'path+offset' when found, or None if address is not in any mapped range. ''' self._build_proc_maps_cache() for (start, end, elf) in self._proc_maps_cache: if start <= addr and end >= addr: return '%s+%x' % (elf, addr-start) return None def _build_proc_maps_cache(self): '''Generate self._proc_maps_cache from ProcMaps field. This only gets done once. ''' if self._proc_maps_cache: return assert 'ProcMaps' in self self._proc_maps_cache = [] # library paths might have spaces, so we need to make some assumptions # about the intermediate fields. But we know that in between the pre-last # data field and the path there are many spaces, while between the # other data fields there is only one. So we take 4 or more spaces as # the separator of the last data field and the path. fmt = re.compile('^([0-9a-fA-F]+)-([0-9a-fA-F]+).*\s{4,}(\S.*$)') fmt_unknown = re.compile('^([0-9a-fA-F]+)-([0-9a-fA-F]+)\s') for line in self['ProcMaps'].splitlines(): if not line.strip(): continue m = fmt.match(line) if not m: # ignore lines with unknown ELF if fmt_unknown.match(line): continue # but complain otherwise, as this means we encounter an # architecture or new kernel version where the format changed assert m, 'cannot parse ProcMaps line: ' + line self._proc_maps_cache.append((int(m.group(1), 16), int(m.group(2), 16), m.group(3)))