def _Symbolize(asan_input): asan_libs = _FindASanLibraries() libraries = collections.defaultdict(list) asan_lines = [] for asan_log_line in [a.rstrip() for a in asan_input]: m = _ParseAsanLogLine(asan_log_line) if m: libraries[m['library']].append(m) asan_lines.append({'raw_log': asan_log_line, 'parsed': m}) all_symbols = collections.defaultdict(dict) for library, items in libraries.iteritems(): libname = _TranslateLibPath(library, asan_libs) lib_relative_addrs = set([i['rel_address'] for i in items]) # pylint: disable=no-member info_dict = symbol.SymbolInformationForSet(libname, lib_relative_addrs, True) if info_dict: all_symbols[library]['symbols'] = info_dict for asan_log_line in asan_lines: m = asan_log_line['parsed'] if not m: print asan_log_line['raw_log'] continue if (m['library'] in all_symbols and m['rel_address'] in all_symbols[m['library']]['symbols']): s = all_symbols[m['library']]['symbols'][m['rel_address']][0] print '%s%s %s %s' % (m['prefix'], m['pos'], s[0], s[1]) else: print asan_log_line['raw_log']
def _PrintSymbolized(asan_input, arch): """Print symbolized logcat output for Asan symbols. Args: asan_input: list of input lines. arch: Target CPU architecture. """ asan_libs = _FindASanLibraries() # Maps library -> [ AsanParsedLine... ] libraries = collections.defaultdict(list) asan_log_lines = [] for line in asan_input: line = line.rstrip() parsed = _ParseAsanLogLine(line) if parsed: libraries[parsed.library].append(parsed) asan_log_lines.append(AsanLogLine(raw=line, parsed=parsed)) # Maps library -> { address -> [(symbol, location, obj_sym_with_offset)...] } all_symbols = collections.defaultdict(dict) for library, items in libraries.items(): libname = _TranslateLibPath(library, asan_libs) lib_relative_addrs = set(i.rel_address for i in items) # pylint: disable=no-member symbols_by_library = symbol.SymbolInformationForSet(libname, lib_relative_addrs, True, cpu_arch=arch) if symbols_by_library: all_symbols[library] = symbols_by_library for log_line in asan_log_lines: m = log_line.parsed if (m and m.library in all_symbols and m.rel_address in all_symbols[m.library]): # NOTE: all_symbols[lib][address] is a never-emtpy list of tuples. # NOTE: The documentation for SymbolInformationForSet() indicates # that usually one wants to display the last list item, not the first. # The code below takes the first, is this the best choice here? s = all_symbols[m.library][m.rel_address][0] symbol_name = s[0] symbol_location = s[1] print('%s%s %s %s @ \'%s\'' % (m.prefix, m.pos, hex( m.rel_address), symbol_name, symbol_location)) else: print(log_line.raw)
def ConvertTrace(lines, more_info): """Convert strings containing native crash to a stack.""" process_info_line = re.compile("(pid: [0-9]+, tid: [0-9]+.*)") signal_line = re.compile("(signal [0-9]+ \(.*\).*)") register_line = re.compile("(([ ]*[0-9a-z]{2} [0-9a-f]{8}){4})") thread_line = re.compile("(.*)(\-\-\- ){15}\-\-\-") dalvik_jni_thread_line = re.compile( "(\".*\" prio=[0-9]+ tid=[0-9]+ NATIVE.*)") dalvik_native_thread_line = re.compile( "(\".*\" sysTid=[0-9]+ nice=[0-9]+.*)") width = "{8}" if symbol.ARCH == "arm64" or symbol.ARCH == "x86_64": width = "{16}" # Matches LOG(FATAL) lines, like the following example: # [FATAL:source_file.cc(33)] Check failed: !instances_.empty() log_fatal_line = re.compile("(\[FATAL\:.*\].*)$") # Note that both trace and value line matching allow for variable amounts of # whitespace (e.g. \t). This is because the we want to allow for the stack # tool to operate on AndroidFeedback provided system logs. AndroidFeedback # strips out double spaces that are found in tombsone files and logcat output. # # Examples of matched trace lines include lines from tombstone files like: # #00 pc 001cf42e /data/data/com.my.project/lib/libmyproject.so # #00 pc 001cf42e /data/data/com.my.project/lib/libmyproject.so (symbol) # Or lines from AndroidFeedback crash report system logs like: # 03-25 00:51:05.520 I/DEBUG ( 65): #00 pc 001cf42e /data/data/com.my.project/lib/libmyproject.so # Please note the spacing differences. trace_line = re.compile( "(.*)\#(?P<frame>[0-9]+)[ \t]+(..)[ \t]+(0x)?(?P<address>[0-9a-f]{0,16})[ \t]+(?P<lib>[^\r\n \t]*)(?P<symbol_present> \((?P<symbol_name>.*)\))?" ) # pylint: disable-msg=C6310 # Matches lines emitted by src/base/debug/stack_trace_android.cc, like: # #00 0x7324d92d /data/app-lib/org.chromium.native_test-1/libbase.cr.so+0x0006992d # This pattern includes the unused named capture groups <symbol_present> and # <symbol_name> so that it can interoperate with the |trace_line| regex. debug_trace_line = re.compile('(.*)(?P<frame>\#[0-9]+ 0x[0-9a-f]' + width + ') ' '(?P<lib>[^+]+)\+0x(?P<address>[0-9a-f]' + width + ')' '(?P<symbol_present>)(?P<symbol_name>)') # Examples of matched value lines include: # bea4170c 8018e4e9 /data/data/com.my.project/lib/libmyproject.so # bea4170c 8018e4e9 /data/data/com.my.project/lib/libmyproject.so (symbol) # 03-25 00:51:05.530 I/DEBUG ( 65): bea4170c 8018e4e9 /data/data/com.my.project/lib/libmyproject.so # Again, note the spacing differences. value_line = re.compile("(.*)([0-9a-f]" + width + ")[ \t]+([0-9a-f]" + width + ")[ \t]+([^\r\n \t]*)( \((.*)\))?") # Lines from 'code around' sections of the output will be matched before # value lines because otheriwse the 'code around' sections will be confused as # value lines. # # Examples include: # 801cf40c ffffc4cc 00b2f2c5 00b2f1c7 00c1e1a8 # 03-25 00:51:05.530 I/DEBUG ( 65): 801cf40c ffffc4cc 00b2f2c5 00b2f1c7 00c1e1a8 code_line = re.compile("(.*)[ \t]*[a-f0-9]" + width + "[ \t]*[a-f0-9]" + width + "[ \t]*[a-f0-9]" + width + "[ \t]*[a-f0-9]" + width + "[ \t]*[a-f0-9]" + width + "[ \t]*[ \r\n]") # pylint: disable-msg=C6310 trace_lines = [] value_lines = [] last_frame = -1 # It is faster to get symbol information with a single call rather than with # separate calls for each line. Since symbol.SymbolInformation caches results, # we can extract all the addresses that we will want symbol information for # from the log and call symbol.SymbolInformation so that the results are # cached in the following lookups. code_addresses = {} for ln in lines: line = unicode(ln, errors='ignore') lib, address = None, None match = trace_line.match(line) or debug_trace_line.match(line) if match: address, lib = match.group('address', 'lib') match = value_line.match(line) if match and not code_line.match(line): (_0, _1, address, lib, _2, _3) = match.groups() if lib: code_addresses.setdefault(lib, set()).add(address) for lib in code_addresses: symbol.SymbolInformationForSet(symbol.TranslateLibPath(lib), code_addresses[lib], more_info) for ln in lines: # AndroidFeedback adds zero width spaces into its crash reports. These # should be removed or the regular expresssions will fail to match. line = unicode(ln, errors='ignore') process_header = process_info_line.search(line) signal_header = signal_line.search(line) register_header = register_line.search(line) thread_header = thread_line.search(line) dalvik_jni_thread_header = dalvik_jni_thread_line.search(line) dalvik_native_thread_header = dalvik_native_thread_line.search(line) log_fatal_header = log_fatal_line.search(line) if (process_header or signal_header or register_header or thread_header or dalvik_jni_thread_header or dalvik_native_thread_header or log_fatal_header): if trace_lines or value_lines: PrintOutput(trace_lines, value_lines, more_info) PrintDivider() trace_lines = [] value_lines = [] last_frame = -1 if process_header: print process_header.group(1) if signal_header: print signal_header.group(1) if register_header: print register_header.group(1) if thread_header: print thread_header.group(1) if dalvik_jni_thread_header: print dalvik_jni_thread_header.group(1) if dalvik_native_thread_header: print dalvik_native_thread_header.group(1) if log_fatal_header: print log_fatal_header.group(1) continue match = trace_line.match(line) or debug_trace_line.match(line) if match: frame, code_addr, area, symbol_present, symbol_name = match.group( 'frame', 'address', 'lib', 'symbol_present', 'symbol_name') if frame <= last_frame and (trace_lines or value_lines): PrintOutput(trace_lines, value_lines, more_info) PrintDivider() trace_lines = [] value_lines = [] last_frame = frame if area == UNKNOWN or area == HEAP or area == STACK: trace_lines.append((code_addr, "", area)) else: # If a calls b which further calls c and c is inlined to b, we want to # display "a -> b -> c" in the stack trace instead of just "a -> c" info = symbol.SymbolInformation(area, code_addr, more_info) nest_count = len(info) - 1 for (source_symbol, source_location, object_symbol_with_offset) in info: if not source_symbol: if symbol_present: source_symbol = symbol.CallCppFilt(symbol_name) else: source_symbol = UNKNOWN if not source_location: source_location = area if nest_count > 0: nest_count = nest_count - 1 trace_lines.append( ("v------>", source_symbol, source_location)) else: if not object_symbol_with_offset: object_symbol_with_offset = source_symbol trace_lines.append( (code_addr, object_symbol_with_offset, source_location)) if code_line.match(line): # Code lines should be ignored. If this were exluded the 'code around' # sections would trigger value_line matches. continue match = value_line.match(line) if match: (unused_, addr, value, area, symbol_present, symbol_name) = match.groups() if area == UNKNOWN or area == HEAP or area == STACK or not area: value_lines.append((addr, value, "", area)) else: info = symbol.SymbolInformation(area, value, more_info) (source_symbol, source_location, object_symbol_with_offset) = info.pop() if not source_symbol: if symbol_present: source_symbol = symbol.CallCppFilt(symbol_name) else: source_symbol = UNKNOWN if not source_location: source_location = area if not object_symbol_with_offset: object_symbol_with_offset = source_symbol value_lines.append( (addr, value, object_symbol_with_offset, source_location)) PrintOutput(trace_lines, value_lines, more_info)
def ResolveCrashSymbol(lines, more_info): """Convert unicode strings which contains native crash to a stack """ trace_lines = [] value_lines = [] last_frame = -1 pid = -1 # It is faster to get symbol information with a single call rather than with # separate calls for each line. Since symbol.SymbolInformation caches results, # we can extract all the addresses that we will want symbol information for # from the log and call symbol.SymbolInformation so that the results are # cached in the following lookups. code_addresses = {} # Collects all java exception lines, keyed by pid for later output during # native crash handling. java_stderr_by_pid = {} for line in lines: lib, address = None, None match = _TRACE_LINE.match(line) or _DEBUG_TRACE_LINE.match(line) if match: address, lib = match.group('address', 'lib') match = _VALUE_LINE.match(line) if match and not code_line.match(line): (_0, _1, address, lib, _2, _3) = match.groups() if lib: code_addresses.setdefault(lib, set()).add(address) java_stderr_match = _JAVA_STDERR_LINE.search(line) if java_stderr_match: pid, msg = java_stderr_match.groups() java_stderr_by_pid.setdefault(pid, []).append(msg) for lib in code_addresses: symbol.SymbolInformationForSet(symbol.TranslateLibPath(lib), code_addresses[lib], more_info) for line in lines: # AndroidFeedback adds zero width spaces into its crash reports. These # should be removed or the regular expresssions will fail to match. process_header = _PROCESS_INFO_LINE.search(line) signal_header = _SIGNAL_LINE.search(line) register_header = _REGISTER_LINE.search(line) thread_header = _THREAD_LINE.search(line) dalvik_jni_thread_header = _DALVIK_JNI_THREAD_LINE.search(line) dalvik_native_thread_header = _DALVIK_NATIVE_THREAD_LINE.search(line) log_fatal_header = _LOG_FATAL_LINE.search(line) if (process_header or signal_header or register_header or thread_header or dalvik_jni_thread_header or dalvik_native_thread_header or log_fatal_header): if trace_lines or value_lines: java_lines = [] if pid != -1 and pid in java_stderr_by_pid: java_lines = java_stderr_by_pid[pid] PrintOutput(trace_lines, value_lines, java_lines, more_info) PrintDivider() trace_lines = [] value_lines = [] last_frame = -1 pid = -1 if process_header: # Track the last reported pid to find java exceptions. pid = _PROCESS_INFO_PID.search( process_header.group(1)).group(1) print process_header.group(1) if signal_header: print signal_header.group(1) if register_header: print register_header.group(1) if thread_header: print thread_header.group(1) if dalvik_jni_thread_header: print dalvik_jni_thread_header.group(1) if dalvik_native_thread_header: print dalvik_native_thread_header.group(1) if log_fatal_header: print log_fatal_header.group(1) continue match = _TRACE_LINE.match(line) or _DEBUG_TRACE_LINE.match(line) if match: frame, code_addr, area, symbol_present, symbol_name = match.group( 'frame', 'address', 'lib', 'symbol_present', 'symbol_name') logging.debug('Found trace line: %s' % line.strip()) if frame <= last_frame and (trace_lines or value_lines): java_lines = [] if pid != -1 and pid in java_stderr_by_pid: java_lines = java_stderr_by_pid[pid] PrintOutput(trace_lines, value_lines, java_lines, more_info) PrintDivider() trace_lines = [] value_lines = [] pid = -1 last_frame = frame if area == UNKNOWN or area == HEAP or area == STACK: trace_lines.append((code_addr, '', area)) else: logging.debug('Identified lib: %s' % area) # If a calls b which further calls c and c is inlined to b, we want to # display "a -> b -> c" in the stack trace instead of just "a -> c" info = symbol.SymbolInformation(area, code_addr, more_info) logging.debug('symbol information: %s' % info) nest_count = len(info) - 1 for (source_symbol, source_location, object_symbol_with_offset) in info: if not source_symbol: if symbol_present: source_symbol = symbol.CallCppFilt(symbol_name) else: source_symbol = UNKNOWN if not source_location: source_location = area if nest_count > 0: nest_count = nest_count - 1 trace_lines.append( ('v------>', source_symbol, source_location)) else: if not object_symbol_with_offset: object_symbol_with_offset = source_symbol trace_lines.append( (code_addr, object_symbol_with_offset, source_location)) match = _VALUE_LINE.match(line) if match: (unused_, addr, value, area, symbol_present, symbol_name) = match.groups() if area == UNKNOWN or area == HEAP or area == STACK or not area: value_lines.append((addr, value, '', area)) else: info = symbol.SymbolInformation(area, value, more_info) (source_symbol, source_location, object_symbol_with_offset) = info.pop() if not source_symbol: if symbol_present: source_symbol = symbol.CallCppFilt(symbol_name) else: source_symbol = UNKNOWN if not source_location: source_location = area if not object_symbol_with_offset: object_symbol_with_offset = source_symbol value_lines.append( (addr, value, object_symbol_with_offset, source_location)) java_lines = [] if pid != -1 and pid in java_stderr_by_pid: java_lines = java_stderr_by_pid[pid] PrintOutput(trace_lines, value_lines, java_lines, more_info)