1""" 2The LLVM Compiler Infrastructure 3 4This file is distributed under the University of Illinois Open Source 5License. See LICENSE.TXT for details. 6 7Provides classes used by the test results reporting infrastructure 8within the LLDB test suite. 9 10 11This module provides process-management support for the LLDB test 12running infrasructure. 13""" 14 15# System imports 16import os 17import re 18import signal 19import subprocess 20import sys 21import threading 22 23 24class CommunicatorThread(threading.Thread): 25 """Provides a thread class that communicates with a subprocess.""" 26 def __init__(self, process, event, output_file): 27 super(CommunicatorThread, self).__init__() 28 # Don't let this thread prevent shutdown. 29 self.daemon = True 30 self.process = process 31 self.pid = process.pid 32 self.event = event 33 self.output_file = output_file 34 self.output = None 35 36 def run(self): 37 try: 38 # Communicate with the child process. 39 # This will not complete until the child process terminates. 40 self.output = self.process.communicate() 41 except Exception as exception: # pylint: disable=broad-except 42 if self.output_file: 43 self.output_file.write( 44 "exception while using communicate() for pid: {}\n".format( 45 exception)) 46 finally: 47 # Signal that the thread's run is complete. 48 self.event.set() 49 50 51# Provides a regular expression for matching gtimeout-based durations. 52TIMEOUT_REGEX = re.compile(r"(^\d+)([smhd])?$") 53 54 55def timeout_to_seconds(timeout): 56 """Converts timeout/gtimeout timeout values into seconds. 57 58 @param timeout a timeout in the form of xm representing x minutes. 59 60 @return None if timeout is None, or the number of seconds as a float 61 if a valid timeout format was specified. 62 """ 63 if timeout is None: 64 return None 65 else: 66 match = TIMEOUT_REGEX.match(timeout) 67 if match: 68 value = float(match.group(1)) 69 units = match.group(2) 70 if units is None: 71 # default is seconds. No conversion necessary. 72 return value 73 elif units == 's': 74 # Seconds. No conversion necessary. 75 return value 76 elif units == 'm': 77 # Value is in minutes. 78 return 60.0 * value 79 elif units == 'h': 80 # Value is in hours. 81 return (60.0 * 60.0) * value 82 elif units == 'd': 83 # Value is in days. 84 return 24 * (60.0 * 60.0) * value 85 else: 86 raise Exception("unexpected units value '{}'".format(units)) 87 else: 88 raise Exception("could not parse TIMEOUT spec '{}'".format( 89 timeout)) 90 91 92class ProcessHelper(object): 93 """Provides an interface for accessing process-related functionality. 94 95 This class provides a factory method that gives the caller a 96 platform-specific implementation instance of the class. 97 98 Clients of the class should stick to the methods provided in this 99 base class. 100 101 @see ProcessHelper.process_helper() 102 """ 103 def __init__(self): 104 super(ProcessHelper, self).__init__() 105 106 @classmethod 107 def process_helper(cls): 108 """Returns a platform-specific ProcessHelper instance. 109 @return a ProcessHelper instance that does the right thing for 110 the current platform. 111 """ 112 113 # If you add a new platform, create an instance here and 114 # return it. 115 if os.name == "nt": 116 return WindowsProcessHelper() 117 else: 118 # For all POSIX-like systems. 119 return UnixProcessHelper() 120 121 def create_piped_process(self, command, new_process_group=True): 122 # pylint: disable=no-self-use,unused-argument 123 # As expected. We want derived classes to implement this. 124 """Creates a subprocess.Popen-based class with I/O piped to the parent. 125 126 @param command the command line list as would be passed to 127 subprocess.Popen(). Use the list form rather than the string form. 128 129 @param new_process_group indicates if the caller wants the 130 process to be created in its own process group. Each OS handles 131 this concept differently. It provides a level of isolation and 132 can simplify or enable terminating the process tree properly. 133 134 @return a subprocess.Popen-like object. 135 """ 136 raise Exception("derived class must implement") 137 138 def supports_soft_terminate(self): 139 # pylint: disable=no-self-use 140 # As expected. We want derived classes to implement this. 141 """Indicates if the platform supports soft termination. 142 143 Soft termination is the concept of a terminate mechanism that 144 allows the target process to shut down nicely, but with the 145 catch that the process might choose to ignore it. 146 147 Platform supporter note: only mark soft terminate as supported 148 if the target process has some way to evade the soft terminate 149 request; otherwise, just support the hard terminate method. 150 151 @return True if the platform supports a soft terminate mechanism. 152 """ 153 # By default, we do not support a soft terminate mechanism. 154 return False 155 156 def soft_terminate(self, popen_process, log_file=None, want_core=True): 157 # pylint: disable=no-self-use,unused-argument 158 # As expected. We want derived classes to implement this. 159 """Attempts to terminate the process in a polite way. 160 161 This terminate method is intended to give the child process a 162 chance to clean up and exit on its own, possibly with a request 163 to drop a core file or equivalent (i.e. [mini-]crashdump, crashlog, 164 etc.) If new_process_group was set in the process creation method 165 and the platform supports it, this terminate call will attempt to 166 kill the whole process tree rooted in this child process. 167 168 @param popen_process the subprocess.Popen-like object returned 169 by one of the process-creation methods of this class. 170 171 @param log_file file-like object used to emit error-related 172 logging info. May be None if no error-related info is desired. 173 174 @param want_core True if the caller would like to get a core 175 dump (or the analogous crash report) from the terminated process. 176 """ 177 popen_process.terminate() 178 179 def hard_terminate(self, popen_process, log_file=None): 180 # pylint: disable=no-self-use,unused-argument 181 # As expected. We want derived classes to implement this. 182 """Attempts to terminate the process immediately. 183 184 This terminate method is intended to kill child process in 185 a manner in which the child process has no ability to block, 186 and also has no ability to clean up properly. If new_process_group 187 was specified when creating the process, and if the platform 188 implementation supports it, this will attempt to kill the 189 whole process tree rooted in the child process. 190 191 @param popen_process the subprocess.Popen-like object returned 192 by one of the process-creation methods of this class. 193 194 @param log_file file-like object used to emit error-related 195 logging info. May be None if no error-related info is desired. 196 """ 197 popen_process.kill() 198 199 def was_soft_terminate(self, returncode, with_core): 200 # pylint: disable=no-self-use,unused-argument 201 # As expected. We want derived classes to implement this. 202 """Returns if Popen-like object returncode matches soft terminate. 203 204 @param returncode the returncode from the Popen-like object that 205 terminated with a given return code. 206 207 @param with_core indicates whether the returncode should match 208 a core-generating return signal. 209 210 @return True when the returncode represents what the system would 211 issue when a soft_terminate() with the given with_core arg occurred; 212 False otherwise. 213 """ 214 if not self.supports_soft_terminate(): 215 # If we don't support soft termination on this platform, 216 # then this should always be False. 217 return False 218 else: 219 # Once a platform claims to support soft terminate, it 220 # needs to be able to identify it by overriding this method. 221 raise Exception("platform needs to implement") 222 223 def was_hard_terminate(self, returncode): 224 # pylint: disable=no-self-use,unused-argument 225 # As expected. We want derived classes to implement this. 226 """Returns if Popen-like object returncode matches that of a hard 227 terminate attempt. 228 229 @param returncode the returncode from the Popen-like object that 230 terminated with a given return code. 231 232 @return True when the returncode represents what the system would 233 issue when a hard_terminate() occurred; False 234 otherwise. 235 """ 236 raise Exception("platform needs to implement") 237 238 def soft_terminate_signals(self): 239 # pylint: disable=no-self-use 240 """Retrieve signal numbers that can be sent to soft terminate. 241 @return a list of signal numbers that can be sent to soft terminate 242 a process, or None if not applicable. 243 """ 244 return None 245 246 def is_exceptional_exit(self, popen_status): 247 """Returns whether the program exit status is exceptional. 248 249 Returns whether the return code from a Popen process is exceptional. 250 251 @return True if the given popen_status represents an exceptional 252 program exit; False otherwise. 253 """ 254 return popen_status != 0 255 256 def exceptional_exit_details(self, popen_status): 257 """Returns the normalized exceptional exit code and a description. 258 259 Given an exceptional exit code, returns the integral value of the 260 exception and a description for the result. 261 262 Derived classes can override this if they want to want custom 263 exceptional exit code handling. 264 265 @return (normalized exception code, symbolic exception description) 266 """ 267 return (popen_status, "exit") 268 269 270class UnixProcessHelper(ProcessHelper): 271 """Provides a ProcessHelper for Unix-like operating systems. 272 273 This implementation supports anything that looks Posix-y 274 (e.g. Darwin, Linux, *BSD, etc.) 275 """ 276 def __init__(self): 277 super(UnixProcessHelper, self).__init__() 278 279 @classmethod 280 def _create_new_process_group(cls): 281 """Creates a new process group for the calling process.""" 282 os.setpgid(os.getpid(), os.getpid()) 283 284 def create_piped_process(self, command, new_process_group=True): 285 # Determine what to run after the fork but before the exec. 286 if new_process_group: 287 preexec_func = self._create_new_process_group 288 else: 289 preexec_func = None 290 291 # Create the process. 292 process = subprocess.Popen( 293 command, 294 stdin=subprocess.PIPE, 295 stdout=subprocess.PIPE, 296 stderr=subprocess.PIPE, 297 universal_newlines=True, # Elicits automatic byte -> string decoding in Py3 298 close_fds=True, 299 preexec_fn=preexec_func) 300 301 # Remember whether we're using process groups for this 302 # process. 303 process.using_process_groups = new_process_group 304 return process 305 306 def supports_soft_terminate(self): 307 # POSIX does support a soft terminate via: 308 # * SIGTERM (no core requested) 309 # * SIGQUIT (core requested if enabled, see ulimit -c) 310 return True 311 312 @classmethod 313 def _validate_pre_terminate(cls, popen_process, log_file): 314 # Validate args. 315 if popen_process is None: 316 raise ValueError("popen_process is None") 317 318 # Ensure we have something that looks like a valid process. 319 if popen_process.pid < 1: 320 if log_file: 321 log_file.write("skipping soft_terminate(): no process id") 322 return False 323 324 # We only do the process liveness check if we're not using 325 # process groups. With process groups, checking if the main 326 # inferior process is dead and short circuiting here is no 327 # good - children of it in the process group could still be 328 # alive, and they should be killed during a timeout. 329 if not popen_process.using_process_groups: 330 # Don't kill if it's already dead. 331 popen_process.poll() 332 if popen_process.returncode is not None: 333 # It has a returncode. It has already stopped. 334 if log_file: 335 log_file.write( 336 "requested to terminate pid {} but it has already " 337 "terminated, returncode {}".format( 338 popen_process.pid, popen_process.returncode)) 339 # Move along... 340 return False 341 342 # Good to go. 343 return True 344 345 def _kill_with_signal(self, popen_process, log_file, signum): 346 # Validate we're ready to terminate this. 347 if not self._validate_pre_terminate(popen_process, log_file): 348 return 349 350 # Choose kill mechanism based on whether we're targeting 351 # a process group or just a process. 352 if popen_process.using_process_groups: 353 # if log_file: 354 # log_file.write( 355 # "sending signum {} to process group {} now\n".format( 356 # signum, popen_process.pid)) 357 os.killpg(popen_process.pid, signum) 358 else: 359 # if log_file: 360 # log_file.write( 361 # "sending signum {} to process {} now\n".format( 362 # signum, popen_process.pid)) 363 os.kill(popen_process.pid, signum) 364 365 def soft_terminate(self, popen_process, log_file=None, want_core=True): 366 # Choose signal based on desire for core file. 367 if want_core: 368 # SIGQUIT will generate core by default. Can be caught. 369 signum = signal.SIGQUIT 370 else: 371 # SIGTERM is the traditional nice way to kill a process. 372 # Can be caught, doesn't generate a core. 373 signum = signal.SIGTERM 374 375 self._kill_with_signal(popen_process, log_file, signum) 376 377 def hard_terminate(self, popen_process, log_file=None): 378 self._kill_with_signal(popen_process, log_file, signal.SIGKILL) 379 380 def was_soft_terminate(self, returncode, with_core): 381 if with_core: 382 return returncode == -signal.SIGQUIT 383 else: 384 return returncode == -signal.SIGTERM 385 386 def was_hard_terminate(self, returncode): 387 return returncode == -signal.SIGKILL 388 389 def soft_terminate_signals(self): 390 return [signal.SIGQUIT, signal.SIGTERM] 391 392 @classmethod 393 def _signal_names_by_number(cls): 394 return dict( 395 (k, v) for v, k in reversed(sorted(signal.__dict__.items())) 396 if v.startswith('SIG') and not v.startswith('SIG_')) 397 398 def exceptional_exit_details(self, popen_status): 399 if popen_status >= 0: 400 return (popen_status, "exit") 401 signo = -popen_status 402 signal_names_by_number = self._signal_names_by_number() 403 signal_name = signal_names_by_number.get(signo, "") 404 return (signo, signal_name) 405 406class WindowsProcessHelper(ProcessHelper): 407 """Provides a Windows implementation of the ProcessHelper class.""" 408 def __init__(self): 409 super(WindowsProcessHelper, self).__init__() 410 411 def create_piped_process(self, command, new_process_group=True): 412 if new_process_group: 413 # We need this flag if we want os.kill() to work on the subprocess. 414 creation_flags = subprocess.CREATE_NEW_PROCESS_GROUP 415 else: 416 creation_flags = 0 417 418 return subprocess.Popen( 419 command, 420 stdin=subprocess.PIPE, 421 stdout=subprocess.PIPE, 422 stderr=subprocess.PIPE, 423 universal_newlines=True, # Elicits automatic byte -> string decoding in Py3 424 creationflags=creation_flags) 425 426 def was_hard_terminate(self, returncode): 427 return returncode != 0 428 429 430class ProcessDriver(object): 431 """Drives a child process, notifies on important events, and can timeout. 432 433 Clients are expected to derive from this class and override the 434 on_process_started and on_process_exited methods if they want to 435 hook either of those. 436 437 This class supports timing out the child process in a platform-agnostic 438 way. The on_process_exited method is informed if the exit was natural 439 or if it was due to a timeout. 440 """ 441 def __init__(self, soft_terminate_timeout=10.0): 442 super(ProcessDriver, self).__init__() 443 self.process_helper = ProcessHelper.process_helper() 444 self.pid = None 445 # Create the synchronization event for notifying when the 446 # inferior dotest process is complete. 447 self.done_event = threading.Event() 448 self.io_thread = None 449 self.process = None 450 # Number of seconds to wait for the soft terminate to 451 # wrap up, before moving to more drastic measures. 452 # Might want this longer if core dumps are generated and 453 # take a long time to write out. 454 self.soft_terminate_timeout = soft_terminate_timeout 455 # Number of seconds to wait for the hard terminate to 456 # wrap up, before giving up on the io thread. This should 457 # be fast. 458 self.hard_terminate_timeout = 5.0 459 self.returncode = None 460 461 # ============================================= 462 # Methods for subclasses to override if desired. 463 # ============================================= 464 465 def on_process_started(self): 466 pass 467 468 def on_process_exited(self, command, output, was_timeout, exit_status): 469 pass 470 471 def write(self, content): 472 # pylint: disable=no-self-use 473 # Intended - we want derived classes to be able to override 474 # this and use any self state they may contain. 475 sys.stdout.write(content) 476 477 # ============================================================== 478 # Operations used to drive processes. Clients will want to call 479 # one of these. 480 # ============================================================== 481 482 def run_command(self, command): 483 # Start up the child process and the thread that does the 484 # communication pump. 485 self._start_process_and_io_thread(command) 486 487 # Wait indefinitely for the child process to finish 488 # communicating. This indicates it has closed stdout/stderr 489 # pipes and is done. 490 self.io_thread.join() 491 self.returncode = self.process.wait() 492 if self.returncode is None: 493 raise Exception( 494 "no exit status available for pid {} after the " 495 " inferior dotest.py should have completed".format( 496 self.process.pid)) 497 498 # Notify of non-timeout exit. 499 self.on_process_exited( 500 command, 501 self.io_thread.output, 502 False, 503 self.returncode) 504 505 def run_command_with_timeout(self, command, timeout, want_core): 506 # Figure out how many seconds our timeout description is requesting. 507 timeout_seconds = timeout_to_seconds(timeout) 508 509 # Start up the child process and the thread that does the 510 # communication pump. 511 self._start_process_and_io_thread(command) 512 513 self._wait_with_timeout(timeout_seconds, command, want_core) 514 515 # ================ 516 # Internal details. 517 # ================ 518 519 def _start_process_and_io_thread(self, command): 520 # Create the process. 521 self.process = self.process_helper.create_piped_process(command) 522 self.pid = self.process.pid 523 self.on_process_started() 524 525 # Ensure the event is cleared that is used for signaling 526 # from the communication() thread when communication is 527 # complete (i.e. the inferior process has finished). 528 self.done_event.clear() 529 530 self.io_thread = CommunicatorThread( 531 self.process, self.done_event, self.write) 532 self.io_thread.start() 533 534 def _attempt_soft_kill(self, want_core): 535 # The inferior dotest timed out. Attempt to clean it 536 # with a non-drastic method (so it can clean up properly 537 # and/or generate a core dump). Often the OS can't guarantee 538 # that the process will really terminate after this. 539 self.process_helper.soft_terminate( 540 self.process, 541 want_core=want_core, 542 log_file=self) 543 544 # Now wait up to a certain timeout period for the io thread 545 # to say that the communication ended. If that wraps up 546 # within our soft terminate timeout, we're all done here. 547 self.io_thread.join(self.soft_terminate_timeout) 548 if not self.io_thread.is_alive(): 549 # stdout/stderr were closed on the child process side. We 550 # should be able to wait and reap the child process here. 551 self.returncode = self.process.wait() 552 # We terminated, and the done_trying result is n/a 553 terminated = True 554 done_trying = None 555 else: 556 self.write("soft kill attempt of process {} timed out " 557 "after {} seconds\n".format( 558 self.process.pid, self.soft_terminate_timeout)) 559 terminated = False 560 done_trying = False 561 return terminated, done_trying 562 563 def _attempt_hard_kill(self): 564 # Instruct the process to terminate and really force it to 565 # happen. Don't give the process a chance to ignore. 566 self.process_helper.hard_terminate( 567 self.process, 568 log_file=self) 569 570 # Reap the child process. This should not hang as the 571 # hard_kill() mechanism is supposed to really kill it. 572 # Improvement option: 573 # If this does ever hang, convert to a self.process.poll() 574 # loop checking on self.process.returncode until it is not 575 # None or the timeout occurs. 576 self.returncode = self.process.wait() 577 578 # Wait a few moments for the io thread to finish... 579 self.io_thread.join(self.hard_terminate_timeout) 580 if self.io_thread.is_alive(): 581 # ... but this is not critical if it doesn't end for some 582 # reason. 583 self.write( 584 "hard kill of process {} timed out after {} seconds waiting " 585 "for the io thread (ignoring)\n".format( 586 self.process.pid, self.hard_terminate_timeout)) 587 588 # Set if it terminated. (Set up for optional improvement above). 589 terminated = self.returncode is not None 590 # Nothing else to try. 591 done_trying = True 592 593 return terminated, done_trying 594 595 def _attempt_termination(self, attempt_count, want_core): 596 if self.process_helper.supports_soft_terminate(): 597 # When soft termination is supported, we first try to stop 598 # the process with a soft terminate. Failing that, we try 599 # the hard terminate option. 600 if attempt_count == 1: 601 return self._attempt_soft_kill(want_core) 602 elif attempt_count == 2: 603 return self._attempt_hard_kill() 604 else: 605 # We don't have anything else to try. 606 terminated = self.returncode is not None 607 done_trying = True 608 return terminated, done_trying 609 else: 610 # We only try the hard terminate option when there 611 # is no soft terminate available. 612 if attempt_count == 1: 613 return self._attempt_hard_kill() 614 else: 615 # We don't have anything else to try. 616 terminated = self.returncode is not None 617 done_trying = True 618 return terminated, done_trying 619 620 def _wait_with_timeout(self, timeout_seconds, command, want_core): 621 # Allow up to timeout seconds for the io thread to wrap up. 622 # If that completes, the child process should be done. 623 completed_normally = self.done_event.wait(timeout_seconds) 624 if completed_normally: 625 # Reap the child process here. 626 self.returncode = self.process.wait() 627 else: 628 # Prepare to stop the process 629 process_terminated = completed_normally 630 terminate_attempt_count = 0 631 632 # Try as many attempts as we support for trying to shut down 633 # the child process if it's not already shut down. 634 while not process_terminated: 635 terminate_attempt_count += 1 636 # Attempt to terminate. 637 process_terminated, done_trying = self._attempt_termination( 638 terminate_attempt_count, want_core) 639 # Check if there's nothing more to try. 640 if done_trying: 641 # Break out of our termination attempt loop. 642 break 643 644 # At this point, we're calling it good. The process 645 # finished gracefully, was shut down after one or more 646 # attempts, or we failed but gave it our best effort. 647 self.on_process_exited( 648 command, 649 self.io_thread.output, 650 not completed_normally, 651 self.returncode) 652 653 654def patched_init(self, *args, **kwargs): 655 self.original_init(*args, **kwargs) 656 # Initialize our condition variable that protects wait()/poll(). 657 self.wait_condition = threading.Condition() 658 659 660def patched_wait(self, *args, **kwargs): 661 self.wait_condition.acquire() 662 try: 663 result = self.original_wait(*args, **kwargs) 664 # The process finished. Signal the condition. 665 self.wait_condition.notify_all() 666 return result 667 finally: 668 self.wait_condition.release() 669 670 671def patched_poll(self, *args, **kwargs): 672 self.wait_condition.acquire() 673 try: 674 result = self.original_poll(*args, **kwargs) 675 if self.returncode is not None: 676 # We did complete, and we have the return value. 677 # Signal the event to indicate we're done. 678 self.wait_condition.notify_all() 679 return result 680 finally: 681 self.wait_condition.release() 682 683 684def patch_up_subprocess_popen(): 685 subprocess.Popen.original_init = subprocess.Popen.__init__ 686 subprocess.Popen.__init__ = patched_init 687 688 subprocess.Popen.original_wait = subprocess.Popen.wait 689 subprocess.Popen.wait = patched_wait 690 691 subprocess.Popen.original_poll = subprocess.Popen.poll 692 subprocess.Popen.poll = patched_poll 693 694# Replace key subprocess.Popen() threading-unprotected methods with 695# threading-protected versions. 696patch_up_subprocess_popen() 697