OILS / core / process.py View on Github | oilshell.org

1957 lines, 948 significant
1# Copyright 2016 Andy Chu. All rights reserved.
2# Licensed under the Apache License, Version 2.0 (the "License");
3# you may not use this file except in compliance with the License.
4# You may obtain a copy of the License at
5#
6# http://www.apache.org/licenses/LICENSE-2.0
7"""
8process.py - Launch processes and manipulate file descriptors.
9"""
10from __future__ import print_function
11
12from errno import EACCES, EBADF, ECHILD, EINTR, ENOENT, ENOEXEC, EEXIST
13import fcntl as fcntl_
14from fcntl import F_DUPFD, F_GETFD, F_SETFD, FD_CLOEXEC
15from signal import (SIG_DFL, SIG_IGN, SIGINT, SIGPIPE, SIGQUIT, SIGTSTP,
16 SIGTTOU, SIGTTIN, SIGWINCH)
17
18from _devbuild.gen.id_kind_asdl import Id
19from _devbuild.gen.runtime_asdl import (job_state_e, job_state_t,
20 job_state_str, wait_status,
21 wait_status_t, RedirValue,
22 redirect_arg, redirect_arg_e, trace,
23 trace_t)
24from _devbuild.gen.syntax_asdl import (
25 loc_t,
26 redir_loc,
27 redir_loc_e,
28 redir_loc_t,
29)
30from _devbuild.gen.value_asdl import (value, value_e)
31from core import dev
32from core import error
33from core.error import e_die
34from core import pyutil
35from core import pyos
36from core import state
37from display import ui
38from core import util
39from data_lang import j8_lite
40from frontend import location
41from frontend import match
42from mycpp import mylib
43from mycpp.mylib import log, print_stderr, probe, tagswitch, iteritems
44
45import posix_ as posix
46from posix_ import (
47 # translated by mycpp and directly called! No wrapper!
48 WIFSIGNALED,
49 WIFEXITED,
50 WIFSTOPPED,
51 WEXITSTATUS,
52 WSTOPSIG,
53 WTERMSIG,
54 WNOHANG,
55 O_APPEND,
56 O_CREAT,
57 O_EXCL,
58 O_NONBLOCK,
59 O_NOCTTY,
60 O_RDONLY,
61 O_RDWR,
62 O_WRONLY,
63 O_TRUNC,
64)
65
66from typing import IO, List, Tuple, Dict, Optional, Any, cast, TYPE_CHECKING
67
68if TYPE_CHECKING:
69 from _devbuild.gen.runtime_asdl import cmd_value
70 from _devbuild.gen.syntax_asdl import command_t
71 from builtin import trap_osh
72 from core import optview
73 from core.util import _DebugFile
74 from osh.cmd_eval import CommandEvaluator
75
76NO_FD = -1
77
78# Minimum file descriptor that the shell can use. Other descriptors can be
79# directly used by user programs, e.g. exec 9>&1
80#
81# Oil uses 100 because users are allowed TWO digits in frontend/lexer_def.py.
82# This is a compromise between bash (unlimited, but requires crazy
83# bookkeeping), and dash/zsh (10) and mksh (24)
84_SHELL_MIN_FD = 100
85
86# Style for 'jobs' builtin
87STYLE_DEFAULT = 0
88STYLE_LONG = 1
89STYLE_PID_ONLY = 2
90
91# To save on allocations in JobList::GetJobWithSpec()
92CURRENT_JOB_SPECS = ['', '%', '%%', '%+']
93
94
95class ctx_FileCloser(object):
96
97 def __init__(self, f):
98 # type: (mylib.LineReader) -> None
99 self.f = f
100
101 def __enter__(self):
102 # type: () -> None
103 pass
104
105 def __exit__(self, type, value, traceback):
106 # type: (Any, Any, Any) -> None
107 self.f.close()
108
109
110def InitInteractiveShell():
111 # type: () -> None
112 """Called when initializing an interactive shell."""
113
114 # The shell itself should ignore Ctrl-\.
115 pyos.Sigaction(SIGQUIT, SIG_IGN)
116
117 # This prevents Ctrl-Z from suspending OSH in interactive mode.
118 pyos.Sigaction(SIGTSTP, SIG_IGN)
119
120 # More signals from
121 # https://www.gnu.org/software/libc/manual/html_node/Initializing-the-Shell.html
122 # (but not SIGCHLD)
123 pyos.Sigaction(SIGTTOU, SIG_IGN)
124 pyos.Sigaction(SIGTTIN, SIG_IGN)
125
126 # Register a callback to receive terminal width changes.
127 # NOTE: In line_input.c, we turned off rl_catch_sigwinch.
128
129 # This is ALWAYS on, which means that it can cause EINTR, and wait() and
130 # read() have to handle it
131 pyos.RegisterSignalInterest(SIGWINCH)
132
133
134def SaveFd(fd):
135 # type: (int) -> int
136 saved = fcntl_.fcntl(fd, F_DUPFD, _SHELL_MIN_FD) # type: int
137 return saved
138
139
140class _RedirFrame(object):
141
142 def __init__(self, saved_fd, orig_fd, forget):
143 # type: (int, int, bool) -> None
144 self.saved_fd = saved_fd
145 self.orig_fd = orig_fd
146 self.forget = forget
147
148
149class _FdFrame(object):
150
151 def __init__(self):
152 # type: () -> None
153 self.saved = [] # type: List[_RedirFrame]
154 self.need_wait = [] # type: List[Process]
155
156 def Forget(self):
157 # type: () -> None
158 """For exec 1>&2."""
159 for rf in reversed(self.saved):
160 if rf.saved_fd != NO_FD and rf.forget:
161 posix.close(rf.saved_fd)
162
163 del self.saved[:] # like list.clear() in Python 3.3
164 del self.need_wait[:]
165
166 def __repr__(self):
167 # type: () -> str
168 return '<_FdFrame %s>' % self.saved
169
170
171class FdState(object):
172 """File descriptor state for the current process.
173
174 For example, you can do 'myfunc > out.txt' without forking. Child
175 processes inherit our state.
176 """
177
178 def __init__(
179 self,
180 errfmt, # type: ui.ErrorFormatter
181 job_control, # type: JobControl
182 job_list, # type: JobList
183 mem, # type: state.Mem
184 tracer, # type: Optional[dev.Tracer]
185 waiter, # type: Optional[Waiter]
186 exec_opts, # type: optview.Exec
187 ):
188 # type: (...) -> None
189 """
190 Args:
191 errfmt: for errors
192 job_list: For keeping track of _HereDocWriterThunk
193 """
194 self.errfmt = errfmt
195 self.job_control = job_control
196 self.job_list = job_list
197 self.cur_frame = _FdFrame() # for the top level
198 self.stack = [self.cur_frame]
199 self.mem = mem
200 self.tracer = tracer
201 self.waiter = waiter
202 self.exec_opts = exec_opts
203
204 def Open(self, path):
205 # type: (str) -> mylib.LineReader
206 """Opens a path for read, but moves it out of the reserved 3-9 fd
207 range.
208
209 Returns:
210 A Python file object. The caller is responsible for Close().
211
212 Raises:
213 IOError or OSError if the path can't be found. (This is Python-induced wart)
214 """
215 fd_mode = O_RDONLY
216 f = self._Open(path, 'r', fd_mode)
217
218 # Hacky downcast
219 return cast('mylib.LineReader', f)
220
221 # used for util.DebugFile
222 def OpenForWrite(self, path):
223 # type: (str) -> mylib.Writer
224 fd_mode = O_CREAT | O_RDWR
225 f = self._Open(path, 'w', fd_mode)
226
227 # Hacky downcast
228 return cast('mylib.Writer', f)
229
230 def _Open(self, path, c_mode, fd_mode):
231 # type: (str, str, int) -> IO[str]
232 fd = posix.open(path, fd_mode, 0o666) # may raise OSError
233
234 # Immediately move it to a new location
235 new_fd = SaveFd(fd)
236 posix.close(fd)
237
238 # Return a Python file handle
239 f = posix.fdopen(new_fd, c_mode) # may raise IOError
240 return f
241
242 def _WriteFdToMem(self, fd_name, fd):
243 # type: (str, int) -> None
244 if self.mem:
245 # setvar, not setref
246 state.OshLanguageSetValue(self.mem, location.LName(fd_name),
247 value.Str(str(fd)))
248
249 def _ReadFdFromMem(self, fd_name):
250 # type: (str) -> int
251 val = self.mem.GetValue(fd_name)
252 if val.tag() == value_e.Str:
253 try:
254 return int(cast(value.Str, val).s)
255 except ValueError:
256 return NO_FD
257 return NO_FD
258
259 def _PushSave(self, fd):
260 # type: (int) -> bool
261 """Save fd to a new location and remember to restore it later."""
262 #log('---- _PushSave %s', fd)
263 ok = True
264 try:
265 new_fd = SaveFd(fd)
266 except (IOError, OSError) as e:
267 ok = False
268 # Example program that causes this error: exec 4>&1. Descriptor 4 isn't
269 # open.
270 # This seems to be ignored in dash too in savefd()?
271 if e.errno != EBADF:
272 raise
273 if ok:
274 posix.close(fd)
275 fcntl_.fcntl(new_fd, F_SETFD, FD_CLOEXEC)
276 self.cur_frame.saved.append(_RedirFrame(new_fd, fd, True))
277 else:
278 # if we got EBADF, we still need to close the original on Pop()
279 self._PushClose(fd)
280
281 return ok
282
283 def _PushDup(self, fd1, blame_loc):
284 # type: (int, redir_loc_t) -> int
285 """Save fd2 in a higher range, and dup fd1 onto fd2.
286
287 Returns whether F_DUPFD/dup2 succeeded, and the new descriptor.
288 """
289 UP_loc = blame_loc
290 if blame_loc.tag() == redir_loc_e.VarName:
291 fd2_name = cast(redir_loc.VarName, UP_loc).name
292 try:
293 # F_DUPFD: GREATER than range
294 new_fd = fcntl_.fcntl(fd1, F_DUPFD, _SHELL_MIN_FD) # type: int
295 except (IOError, OSError) as e:
296 if e.errno == EBADF:
297 print_stderr('F_DUPFD fd %d: %s' %
298 (fd1, pyutil.strerror(e)))
299 return NO_FD
300 else:
301 raise # this redirect failed
302
303 self._WriteFdToMem(fd2_name, new_fd)
304
305 elif blame_loc.tag() == redir_loc_e.Fd:
306 fd2 = cast(redir_loc.Fd, UP_loc).fd
307
308 if fd1 == fd2:
309 # The user could have asked for it to be open on descriptor 3, but open()
310 # already returned 3, e.g. echo 3>out.txt
311 return NO_FD
312
313 # Check the validity of fd1 before _PushSave(fd2)
314 try:
315 fcntl_.fcntl(fd1, F_GETFD)
316 except (IOError, OSError) as e:
317 print_stderr('F_GETFD fd %d: %s' % (fd1, pyutil.strerror(e)))
318 raise
319
320 need_restore = self._PushSave(fd2)
321
322 #log('==== dup2 %s %s\n' % (fd1, fd2))
323 try:
324 posix.dup2(fd1, fd2)
325 except (IOError, OSError) as e:
326 # bash/dash give this error too, e.g. for 'echo hi 1>&3'
327 print_stderr('dup2(%d, %d): %s' %
328 (fd1, fd2, pyutil.strerror(e)))
329
330 # Restore and return error
331 if need_restore:
332 rf = self.cur_frame.saved.pop()
333 posix.dup2(rf.saved_fd, rf.orig_fd)
334 posix.close(rf.saved_fd)
335
336 raise # this redirect failed
337
338 new_fd = fd2
339
340 else:
341 raise AssertionError()
342
343 return new_fd
344
345 def _PushCloseFd(self, blame_loc):
346 # type: (redir_loc_t) -> bool
347 """For 2>&-"""
348 # exec {fd}>&- means close the named descriptor
349
350 UP_loc = blame_loc
351 if blame_loc.tag() == redir_loc_e.VarName:
352 fd_name = cast(redir_loc.VarName, UP_loc).name
353 fd = self._ReadFdFromMem(fd_name)
354 if fd == NO_FD:
355 return False
356
357 elif blame_loc.tag() == redir_loc_e.Fd:
358 fd = cast(redir_loc.Fd, UP_loc).fd
359
360 else:
361 raise AssertionError()
362
363 self._PushSave(fd)
364
365 return True
366
367 def _PushClose(self, fd):
368 # type: (int) -> None
369 self.cur_frame.saved.append(_RedirFrame(NO_FD, fd, False))
370
371 def _PushWait(self, proc):
372 # type: (Process) -> None
373 self.cur_frame.need_wait.append(proc)
374
375 def _ApplyRedirect(self, r):
376 # type: (RedirValue) -> None
377 arg = r.arg
378 UP_arg = arg
379 with tagswitch(arg) as case:
380
381 if case(redirect_arg_e.Path):
382 arg = cast(redirect_arg.Path, UP_arg)
383 # noclobber flag is OR'd with other flags when allowed
384 noclobber_mode = O_EXCL if self.exec_opts.noclobber() else 0
385 if r.op_id in (Id.Redir_Great, Id.Redir_AndGreat): # > &>
386 # NOTE: This is different than >| because it respects noclobber, but
387 # that option is almost never used. See test/wild.sh.
388 mode = O_CREAT | O_WRONLY | O_TRUNC | noclobber_mode
389 elif r.op_id == Id.Redir_Clobber: # >|
390 mode = O_CREAT | O_WRONLY | O_TRUNC
391 elif r.op_id in (Id.Redir_DGreat,
392 Id.Redir_AndDGreat): # >> &>>
393 mode = O_CREAT | O_WRONLY | O_APPEND | noclobber_mode
394 elif r.op_id == Id.Redir_Less: # <
395 mode = O_RDONLY
396 elif r.op_id == Id.Redir_LessGreat: # <>
397 mode = O_CREAT | O_RDWR
398 else:
399 raise NotImplementedError(r.op_id)
400
401 # NOTE: 0666 is affected by umask, all shells use it.
402 try:
403 open_fd = posix.open(arg.filename, mode, 0o666)
404 except (IOError, OSError) as e:
405 if e.errno == EEXIST and self.exec_opts.noclobber():
406 extra = ' (noclobber)'
407 else:
408 extra = ''
409 self.errfmt.Print_(
410 "Can't open %r: %s%s" %
411 (arg.filename, pyutil.strerror(e), extra),
412 blame_loc=r.op_loc)
413 raise # redirect failed
414
415 new_fd = self._PushDup(open_fd, r.loc)
416 if new_fd != NO_FD:
417 posix.close(open_fd)
418
419 # Now handle &> and &>> and their variants. These pairs are the same:
420 #
421 # stdout_stderr.py &> out-err.txt
422 # stdout_stderr.py > out-err.txt 2>&1
423 #
424 # stdout_stderr.py 3&> out-err.txt
425 # stdout_stderr.py 3> out-err.txt 2>&3
426 #
427 # Ditto for {fd}> and {fd}&>
428
429 if r.op_id in (Id.Redir_AndGreat, Id.Redir_AndDGreat):
430 self._PushDup(new_fd, redir_loc.Fd(2))
431
432 elif case(redirect_arg_e.CopyFd): # e.g. echo hi 1>&2
433 arg = cast(redirect_arg.CopyFd, UP_arg)
434
435 if r.op_id == Id.Redir_GreatAnd: # 1>&2
436 self._PushDup(arg.target_fd, r.loc)
437
438 elif r.op_id == Id.Redir_LessAnd: # 0<&5
439 # The only difference between >& and <& is the default file
440 # descriptor argument.
441 self._PushDup(arg.target_fd, r.loc)
442
443 else:
444 raise NotImplementedError()
445
446 elif case(redirect_arg_e.MoveFd): # e.g. echo hi 5>&6-
447 arg = cast(redirect_arg.MoveFd, UP_arg)
448 new_fd = self._PushDup(arg.target_fd, r.loc)
449 if new_fd != NO_FD:
450 posix.close(arg.target_fd)
451
452 UP_loc = r.loc
453 if r.loc.tag() == redir_loc_e.Fd:
454 fd = cast(redir_loc.Fd, UP_loc).fd
455 else:
456 fd = NO_FD
457
458 self.cur_frame.saved.append(_RedirFrame(new_fd, fd, False))
459
460 elif case(redirect_arg_e.CloseFd): # e.g. echo hi 5>&-
461 self._PushCloseFd(r.loc)
462
463 elif case(redirect_arg_e.HereDoc):
464 arg = cast(redirect_arg.HereDoc, UP_arg)
465
466 # NOTE: Do these descriptors have to be moved out of the range 0-9?
467 read_fd, write_fd = posix.pipe()
468
469 self._PushDup(read_fd, r.loc) # stdin is now the pipe
470
471 # We can't close like we do in the filename case above? The writer can
472 # get a "broken pipe".
473 self._PushClose(read_fd)
474
475 thunk = _HereDocWriterThunk(write_fd, arg.body)
476
477 # Use PIPE_SIZE to save a process in the case of small here
478 # docs, which are the common case. (dash does this.)
479
480 # Note: could instrument this to see how often it happens.
481 # Though strace -ff can also work.
482 start_process = len(arg.body) > 4096
483 #start_process = True
484
485 if start_process:
486 here_proc = Process(thunk, self.job_control, self.job_list,
487 self.tracer)
488
489 # NOTE: we could close the read pipe here, but it doesn't really
490 # matter because we control the code.
491 here_proc.StartProcess(trace.HereDoc)
492 #log('Started %s as %d', here_proc, pid)
493 self._PushWait(here_proc)
494
495 # Now that we've started the child, close it in the parent.
496 posix.close(write_fd)
497
498 else:
499 posix.write(write_fd, arg.body)
500 posix.close(write_fd)
501
502 def Push(self, redirects, err_out):
503 # type: (List[RedirValue], List[error.IOError_OSError]) -> None
504 """Apply a group of redirects and remember to undo them."""
505
506 #log('> fd_state.Push %s', redirects)
507 new_frame = _FdFrame()
508 self.stack.append(new_frame)
509 self.cur_frame = new_frame
510
511 for r in redirects:
512 #log('apply %s', r)
513 with ui.ctx_Location(self.errfmt, r.op_loc):
514 try:
515 self._ApplyRedirect(r)
516 except (IOError, OSError) as e:
517 err_out.append(e)
518 # This can fail too
519 self.Pop(err_out)
520 return # for bad descriptor, etc.
521
522 def PushStdinFromPipe(self, r):
523 # type: (int) -> bool
524 """Save the current stdin and make it come from descriptor 'r'.
525
526 'r' is typically the read-end of a pipe. For 'lastpipe'/ZSH
527 semantics of
528
529 echo foo | read line; echo $line
530 """
531 new_frame = _FdFrame()
532 self.stack.append(new_frame)
533 self.cur_frame = new_frame
534
535 self._PushDup(r, redir_loc.Fd(0))
536 return True
537
538 def Pop(self, err_out):
539 # type: (List[error.IOError_OSError]) -> None
540 frame = self.stack.pop()
541 #log('< Pop %s', frame)
542 for rf in reversed(frame.saved):
543 if rf.saved_fd == NO_FD:
544 #log('Close %d', orig)
545 try:
546 posix.close(rf.orig_fd)
547 except (IOError, OSError) as e:
548 err_out.append(e)
549 log('Error closing descriptor %d: %s', rf.orig_fd,
550 pyutil.strerror(e))
551 return
552 else:
553 try:
554 posix.dup2(rf.saved_fd, rf.orig_fd)
555 except (IOError, OSError) as e:
556 err_out.append(e)
557 log('dup2(%d, %d) error: %s', rf.saved_fd, rf.orig_fd,
558 pyutil.strerror(e))
559 #log('fd state:')
560 #posix.system('ls -l /proc/%s/fd' % posix.getpid())
561 return
562 posix.close(rf.saved_fd)
563 #log('dup2 %s %s', saved, orig)
564
565 # Wait for here doc processes to finish.
566 for proc in frame.need_wait:
567 unused_status = proc.Wait(self.waiter)
568
569 def MakePermanent(self):
570 # type: () -> None
571 self.cur_frame.Forget()
572
573
574class ChildStateChange(object):
575
576 def __init__(self):
577 # type: () -> None
578 """Empty constructor for mycpp."""
579 pass
580
581 def Apply(self):
582 # type: () -> None
583 raise NotImplementedError()
584
585 def ApplyFromParent(self, proc):
586 # type: (Process) -> None
587 """Noop for all state changes other than SetPgid for mycpp."""
588 pass
589
590
591class StdinFromPipe(ChildStateChange):
592
593 def __init__(self, pipe_read_fd, w):
594 # type: (int, int) -> None
595 self.r = pipe_read_fd
596 self.w = w
597
598 def __repr__(self):
599 # type: () -> str
600 return '<StdinFromPipe %d %d>' % (self.r, self.w)
601
602 def Apply(self):
603 # type: () -> None
604 posix.dup2(self.r, 0)
605 posix.close(self.r) # close after dup
606
607 posix.close(self.w) # we're reading from the pipe, not writing
608 #log('child CLOSE w %d pid=%d', self.w, posix.getpid())
609
610
611class StdoutToPipe(ChildStateChange):
612
613 def __init__(self, r, pipe_write_fd):
614 # type: (int, int) -> None
615 self.r = r
616 self.w = pipe_write_fd
617
618 def __repr__(self):
619 # type: () -> str
620 return '<StdoutToPipe %d %d>' % (self.r, self.w)
621
622 def Apply(self):
623 # type: () -> None
624 posix.dup2(self.w, 1)
625 posix.close(self.w) # close after dup
626
627 posix.close(self.r) # we're writing to the pipe, not reading
628 #log('child CLOSE r %d pid=%d', self.r, posix.getpid())
629
630
631INVALID_PGID = -1
632# argument to setpgid() that means the process is its own leader
633OWN_LEADER = 0
634
635
636class SetPgid(ChildStateChange):
637
638 def __init__(self, pgid, tracer):
639 # type: (int, dev.Tracer) -> None
640 self.pgid = pgid
641 self.tracer = tracer
642
643 def Apply(self):
644 # type: () -> None
645 try:
646 posix.setpgid(0, self.pgid)
647 except (IOError, OSError) as e:
648 self.tracer.OtherMessage(
649 'osh: child %d failed to set its process group to %d: %s' %
650 (posix.getpid(), self.pgid, pyutil.strerror(e)))
651
652 def ApplyFromParent(self, proc):
653 # type: (Process) -> None
654 try:
655 posix.setpgid(proc.pid, self.pgid)
656 except (IOError, OSError) as e:
657 self.tracer.OtherMessage(
658 'osh: parent failed to set process group for PID %d to %d: %s'
659 % (proc.pid, self.pgid, pyutil.strerror(e)))
660
661
662class ExternalProgram(object):
663 """The capability to execute an external program like 'ls'."""
664
665 def __init__(
666 self,
667 hijack_shebang, # type: str
668 fd_state, # type: FdState
669 errfmt, # type: ui.ErrorFormatter
670 debug_f, # type: _DebugFile
671 ):
672 # type: (...) -> None
673 """
674 Args:
675 hijack_shebang: The path of an interpreter to run instead of the one
676 specified in the shebang line. May be empty.
677 """
678 self.hijack_shebang = hijack_shebang
679 self.fd_state = fd_state
680 self.errfmt = errfmt
681 self.debug_f = debug_f
682
683 def Exec(self, argv0_path, cmd_val, environ):
684 # type: (str, cmd_value.Argv, Dict[str, str]) -> None
685 """Execute a program and exit this process.
686
687 Called by: ls / exec ls / ( ls / )
688 """
689 probe('process', 'ExternalProgram_Exec', argv0_path)
690 self._Exec(argv0_path, cmd_val.argv, cmd_val.arg_locs[0], environ,
691 True)
692 assert False, "This line should never execute" # NO RETURN
693
694 def _Exec(self, argv0_path, argv, argv0_loc, environ, should_retry):
695 # type: (str, List[str], loc_t, Dict[str, str], bool) -> None
696 if len(self.hijack_shebang):
697 opened = True
698 try:
699 f = self.fd_state.Open(argv0_path)
700 except (IOError, OSError) as e:
701 opened = False
702
703 if opened:
704 with ctx_FileCloser(f):
705 # Test if the shebang looks like a shell. TODO: The file might be
706 # binary with no newlines, so read 80 bytes instead of readline().
707
708 #line = f.read(80) # type: ignore # TODO: fix this
709 line = f.readline()
710
711 if match.ShouldHijack(line):
712 h_argv = [self.hijack_shebang, argv0_path]
713 h_argv.extend(argv[1:])
714 argv = h_argv
715 argv0_path = self.hijack_shebang
716 self.debug_f.writeln('Hijacked: %s' % argv0_path)
717 else:
718 #self.debug_f.log('Not hijacking %s (%r)', argv, line)
719 pass
720
721 try:
722 posix.execve(argv0_path, argv, environ)
723 except (IOError, OSError) as e:
724 # Run with /bin/sh when ENOEXEC error (no shebang). All shells do this.
725 if e.errno == ENOEXEC and should_retry:
726 new_argv = ['/bin/sh', argv0_path]
727 new_argv.extend(argv[1:])
728 self._Exec('/bin/sh', new_argv, argv0_loc, environ, False)
729 # NO RETURN
730
731 # Would be nice: when the path is relative and ENOENT: print PWD and do
732 # spelling correction?
733
734 self.errfmt.Print_(
735 "Can't execute %r: %s" % (argv0_path, pyutil.strerror(e)),
736 argv0_loc)
737
738 # POSIX mentions 126 and 127 for two specific errors. The rest are
739 # unspecified.
740 #
741 # http://pubs.opengroup.org/onlinepubs/9699919799.2016edition/utilities/V3_chap02.html#tag_18_08_02
742 if e.errno == EACCES:
743 status = 126
744 elif e.errno == ENOENT:
745 # TODO: most shells print 'command not found', rather than strerror()
746 # == "No such file or directory". That's better because it's at the
747 # end of the path search, and we're never searching for a directory.
748 status = 127
749 else:
750 # dash uses 2, but we use that for parse errors. This seems to be
751 # consistent with mksh and zsh.
752 status = 127
753
754 posix._exit(status)
755 # NO RETURN
756
757
758class Thunk(object):
759 """Abstract base class for things runnable in another process."""
760
761 def __init__(self):
762 # type: () -> None
763 """Empty constructor for mycpp."""
764 pass
765
766 def Run(self):
767 # type: () -> None
768 """Returns a status code."""
769 raise NotImplementedError()
770
771 def UserString(self):
772 # type: () -> str
773 """Display for the 'jobs' list."""
774 raise NotImplementedError()
775
776 def __repr__(self):
777 # type: () -> str
778 return self.UserString()
779
780
781class ExternalThunk(Thunk):
782 """An external executable."""
783
784 def __init__(self, ext_prog, argv0_path, cmd_val, environ):
785 # type: (ExternalProgram, str, cmd_value.Argv, Dict[str, str]) -> None
786 self.ext_prog = ext_prog
787 self.argv0_path = argv0_path
788 self.cmd_val = cmd_val
789 self.environ = environ
790
791 def UserString(self):
792 # type: () -> str
793
794 # NOTE: This is the format the Tracer uses.
795 # bash displays sleep $n & (code)
796 # but OSH displays sleep 1 & (argv array)
797 # We could switch the former but I'm not sure it's necessary.
798 tmp = [j8_lite.MaybeShellEncode(a) for a in self.cmd_val.argv]
799 return '[process] %s' % ' '.join(tmp)
800
801 def Run(self):
802 # type: () -> None
803 """An ExternalThunk is run in parent for the exec builtin."""
804 self.ext_prog.Exec(self.argv0_path, self.cmd_val, self.environ)
805
806
807class SubProgramThunk(Thunk):
808 """A subprogram that can be executed in another process."""
809
810 def __init__(self, cmd_ev, node, trap_state, multi_trace, inherit_errexit,
811 inherit_errtrace):
812 # type: (CommandEvaluator, command_t, trap_osh.TrapState, dev.MultiTracer, bool, bool) -> None
813 self.cmd_ev = cmd_ev
814 self.node = node
815 self.trap_state = trap_state
816 self.multi_trace = multi_trace
817 self.inherit_errexit = inherit_errexit # for bash errexit compatibility
818 self.inherit_errtrace = inherit_errtrace # for bash errtrace compatibility
819
820 def UserString(self):
821 # type: () -> str
822
823 # NOTE: These can be pieces of a pipeline, so they're arbitrary nodes.
824 # TODO: Extract SPIDS from node to display source? Note that
825 # CompoundStatus also has locations of each pipeline component; see
826 # Executor.RunPipeline()
827 thunk_str = ui.CommandType(self.node)
828 return '[subprog] %s' % thunk_str
829
830 def Run(self):
831 # type: () -> None
832 #self.errfmt.OneLineErrExit() # don't quote code in child processes
833 probe('process', 'SubProgramThunk_Run')
834
835 # TODO: break circular dep. Bit flags could go in ASDL or headers.
836 from osh import cmd_eval
837
838 # signal handlers aren't inherited
839 self.trap_state.ClearForSubProgram(self.inherit_errtrace)
840
841 # NOTE: may NOT return due to exec().
842 if not self.inherit_errexit:
843 self.cmd_ev.mutable_opts.DisableErrExit()
844 try:
845 # optimize to eliminate redundant subshells like ( echo hi ) | wc -l etc.
846 self.cmd_ev.ExecuteAndCatch(self.node, cmd_flags=cmd_eval.Optimize)
847 status = self.cmd_ev.LastStatus()
848 # NOTE: We ignore the is_fatal return value. The user should set -o
849 # errexit so failures in subprocesses cause failures in the parent.
850 except util.UserExit as e:
851 status = e.status
852
853 # Handle errors in a subshell. These two cases are repeated from main()
854 # and the core/completion.py hook.
855 except KeyboardInterrupt:
856 print('')
857 status = 130 # 128 + 2
858 except (IOError, OSError) as e:
859 print_stderr('oils I/O error (subprogram): %s' %
860 pyutil.strerror(e))
861 status = 2
862
863 # If ProcessInit() doesn't turn off buffering, this is needed before
864 # _exit()
865 pyos.FlushStdout()
866
867 self.multi_trace.WriteDumps()
868
869 # We do NOT want to raise SystemExit here. Otherwise dev.Tracer::Pop()
870 # gets called in BOTH processes.
871 # The crash dump seems to be unaffected.
872 posix._exit(status)
873
874
875class _HereDocWriterThunk(Thunk):
876 """Write a here doc to one end of a pipe.
877
878 May be be executed in either a child process or the main shell
879 process.
880 """
881
882 def __init__(self, w, body_str):
883 # type: (int, str) -> None
884 self.w = w
885 self.body_str = body_str
886
887 def UserString(self):
888 # type: () -> str
889
890 # You can hit Ctrl-Z and the here doc writer will be suspended! Other
891 # shells don't have this problem because they use temp files! That's a bit
892 # unfortunate.
893 return '[here doc writer]'
894
895 def Run(self):
896 # type: () -> None
897 """do_exit: For small pipelines."""
898 probe('process', 'HereDocWriterThunk_Run')
899 #log('Writing %r', self.body_str)
900 posix.write(self.w, self.body_str)
901 #log('Wrote %r', self.body_str)
902 posix.close(self.w)
903 #log('Closed %d', self.w)
904
905 posix._exit(0)
906
907
908class Job(object):
909 """Interface for both Process and Pipeline.
910
911 They both can be put in the background and waited on.
912
913 Confusing thing about pipelines in the background: They have TOO MANY NAMES.
914
915 sleep 1 | sleep 2 &
916
917 - The LAST PID is what's printed at the prompt. This is $!, a PROCESS ID and
918 not a JOB ID.
919 # https://www.gnu.org/software/bash/manual/html_node/Special-Parameters.html#Special-Parameters
920 - The process group leader (setpgid) is the FIRST PID.
921 - It's also %1 or %+. The last job started.
922 """
923
924 def __init__(self):
925 # type: () -> None
926 # Initial state with & or Ctrl-Z is Running.
927 self.state = job_state_e.Running
928 self.job_id = -1
929 self.in_background = False
930
931 def DisplayJob(self, job_id, f, style):
932 # type: (int, mylib.Writer, int) -> None
933 raise NotImplementedError()
934
935 def State(self):
936 # type: () -> job_state_t
937 return self.state
938
939 def ProcessGroupId(self):
940 # type: () -> int
941 """Return the process group ID associated with this job."""
942 raise NotImplementedError()
943
944 def JobWait(self, waiter):
945 # type: (Waiter) -> wait_status_t
946 """Wait for this process/pipeline to be stopped or finished."""
947 raise NotImplementedError()
948
949 def SetBackground(self):
950 # type: () -> None
951 """Record that this job is running in the background."""
952 self.in_background = True
953
954 def SetForeground(self):
955 # type: () -> None
956 """Record that this job is running in the foreground."""
957 self.in_background = False
958
959
960class Process(Job):
961 """A process to run.
962
963 TODO: Should we make it clear that this is a FOREGROUND process? A
964 background process is wrapped in a "job". It is unevaluated.
965
966 It provides an API to manipulate file descriptor state in parent and child.
967 """
968
969 def __init__(self, thunk, job_control, job_list, tracer):
970 # type: (Thunk, JobControl, JobList, dev.Tracer) -> None
971 """
972 Args:
973 thunk: Thunk instance
974 job_list: for process bookkeeping
975 """
976 Job.__init__(self)
977 assert isinstance(thunk, Thunk), thunk
978 self.thunk = thunk
979 self.job_control = job_control
980 self.job_list = job_list
981 self.tracer = tracer
982
983 # For pipelines
984 self.parent_pipeline = None # type: Pipeline
985 self.state_changes = [] # type: List[ChildStateChange]
986 self.close_r = -1
987 self.close_w = -1
988
989 self.pid = -1
990 self.status = -1
991
992 def Init_ParentPipeline(self, pi):
993 # type: (Pipeline) -> None
994 """For updating PIPESTATUS."""
995 self.parent_pipeline = pi
996
997 def __repr__(self):
998 # type: () -> str
999
1000 # note: be wary of infinite mutual recursion
1001 #s = ' %s' % self.parent_pipeline if self.parent_pipeline else ''
1002 #return '<Process %s%s>' % (self.thunk, s)
1003 return '<Process %s %s>' % (_JobStateStr(self.state), self.thunk)
1004
1005 def ProcessGroupId(self):
1006 # type: () -> int
1007 """Returns the group ID of this process."""
1008 # This should only ever be called AFTER the process has started
1009 assert self.pid != -1
1010 if self.parent_pipeline:
1011 # XXX: Maybe we should die here instead? Unclear if this branch
1012 # should even be reachable with the current builtins.
1013 return self.parent_pipeline.ProcessGroupId()
1014
1015 return self.pid
1016
1017 def DisplayJob(self, job_id, f, style):
1018 # type: (int, mylib.Writer, int) -> None
1019 if job_id == -1:
1020 job_id_str = ' '
1021 else:
1022 job_id_str = '%%%d' % job_id
1023 if style == STYLE_PID_ONLY:
1024 f.write('%d\n' % self.pid)
1025 else:
1026 f.write('%s %d %7s ' %
1027 (job_id_str, self.pid, _JobStateStr(self.state)))
1028 f.write(self.thunk.UserString())
1029 f.write('\n')
1030
1031 def AddStateChange(self, s):
1032 # type: (ChildStateChange) -> None
1033 self.state_changes.append(s)
1034
1035 def AddPipeToClose(self, r, w):
1036 # type: (int, int) -> None
1037 self.close_r = r
1038 self.close_w = w
1039
1040 def MaybeClosePipe(self):
1041 # type: () -> None
1042 if self.close_r != -1:
1043 posix.close(self.close_r)
1044 posix.close(self.close_w)
1045
1046 def StartProcess(self, why):
1047 # type: (trace_t) -> int
1048 """Start this process with fork(), handling redirects."""
1049 pid = posix.fork()
1050 if pid < 0:
1051 # When does this happen?
1052 e_die('Fatal error in posix.fork()')
1053
1054 elif pid == 0: # child
1055 # Note: this happens in BOTH interactive and non-interactive shells.
1056 # We technically don't need to do most of it in non-interactive, since we
1057 # did not change state in InitInteractiveShell().
1058
1059 for st in self.state_changes:
1060 st.Apply()
1061
1062 # Python sets SIGPIPE handler to SIG_IGN by default. Child processes
1063 # shouldn't have this.
1064 # https://docs.python.org/2/library/signal.html
1065 # See Python/pythonrun.c.
1066 pyos.Sigaction(SIGPIPE, SIG_DFL)
1067
1068 # Respond to Ctrl-\ (core dump)
1069 pyos.Sigaction(SIGQUIT, SIG_DFL)
1070
1071 # Only standalone children should get Ctrl-Z. Pipelines remain in the
1072 # foreground because suspending them is difficult with our 'lastpipe'
1073 # semantics.
1074 pid = posix.getpid()
1075 if posix.getpgid(0) == pid and self.parent_pipeline is None:
1076 pyos.Sigaction(SIGTSTP, SIG_DFL)
1077
1078 # More signals from
1079 # https://www.gnu.org/software/libc/manual/html_node/Launching-Jobs.html
1080 # (but not SIGCHLD)
1081 pyos.Sigaction(SIGTTOU, SIG_DFL)
1082 pyos.Sigaction(SIGTTIN, SIG_DFL)
1083
1084 self.tracer.OnNewProcess(pid)
1085 # clear foreground pipeline for subshells
1086 self.thunk.Run()
1087 # Never returns
1088
1089 #log('STARTED process %s, pid = %d', self, pid)
1090 self.tracer.OnProcessStart(pid, why)
1091
1092 # Class invariant: after the process is started, it stores its PID.
1093 self.pid = pid
1094
1095 # SetPgid needs to be applied from the child and the parent to avoid
1096 # racing in calls to tcsetpgrp() in the parent. See APUE sec. 9.2.
1097 for st in self.state_changes:
1098 st.ApplyFromParent(self)
1099
1100 # Program invariant: We keep track of every child process!
1101 self.job_list.AddChildProcess(pid, self)
1102
1103 return pid
1104
1105 def Wait(self, waiter):
1106 # type: (Waiter) -> int
1107 """Wait for this process to finish."""
1108 while self.state == job_state_e.Running:
1109 # Only return if there's nothing to wait for. Keep waiting if we were
1110 # interrupted with a signal.
1111 if waiter.WaitForOne() == W1_ECHILD:
1112 break
1113
1114 assert self.status >= 0, self.status
1115 return self.status
1116
1117 def JobWait(self, waiter):
1118 # type: (Waiter) -> wait_status_t
1119 # wait builtin can be interrupted
1120 while self.state == job_state_e.Running:
1121 result = waiter.WaitForOne()
1122
1123 if result >= 0: # signal
1124 return wait_status.Cancelled(result)
1125
1126 if result == W1_ECHILD:
1127 break
1128
1129 return wait_status.Proc(self.status)
1130
1131 def WhenStopped(self, stop_sig):
1132 # type: (int) -> None
1133
1134 # 128 is a shell thing
1135 # https://www.gnu.org/software/bash/manual/html_node/Exit-Status.html
1136 self.status = 128 + stop_sig
1137 self.state = job_state_e.Stopped
1138
1139 if self.job_id == -1:
1140 # This process was started in the foreground
1141 self.job_list.AddJob(self)
1142
1143 if not self.in_background:
1144 self.job_control.MaybeTakeTerminal()
1145 self.SetBackground()
1146
1147 def WhenDone(self, pid, status):
1148 # type: (int, int) -> None
1149 """Called by the Waiter when this Process finishes."""
1150
1151 #log('Process WhenDone %d %d', pid, status)
1152 assert pid == self.pid, 'Expected %d, got %d' % (self.pid, pid)
1153 self.status = status
1154 self.state = job_state_e.Done
1155 if self.parent_pipeline:
1156 self.parent_pipeline.WhenDone(pid, status)
1157 else:
1158 if self.job_id != -1:
1159 # Job might have been brought to the foreground after being
1160 # assigned a job ID.
1161 if self.in_background:
1162 print_stderr('[%d] Done PID %d' % (self.job_id, self.pid))
1163
1164 self.job_list.RemoveJob(self.job_id)
1165
1166 self.job_list.RemoveChildProcess(self.pid)
1167
1168 if not self.in_background:
1169 self.job_control.MaybeTakeTerminal()
1170
1171 def RunProcess(self, waiter, why):
1172 # type: (Waiter, trace_t) -> int
1173 """Run this process synchronously."""
1174 self.StartProcess(why)
1175 # ShellExecutor might be calling this for the last part of a pipeline.
1176 if self.parent_pipeline is None:
1177 # QUESTION: Can the PGID of a single process just be the PID? i.e. avoid
1178 # calling getpgid()?
1179 self.job_control.MaybeGiveTerminal(posix.getpgid(self.pid))
1180 return self.Wait(waiter)
1181
1182
1183class ctx_Pipe(object):
1184
1185 def __init__(self, fd_state, fd, err_out):
1186 # type: (FdState, int, List[error.IOError_OSError]) -> None
1187 fd_state.PushStdinFromPipe(fd)
1188 self.fd_state = fd_state
1189 self.err_out = err_out
1190
1191 def __enter__(self):
1192 # type: () -> None
1193 pass
1194
1195 def __exit__(self, type, value, traceback):
1196 # type: (Any, Any, Any) -> None
1197 self.fd_state.Pop(self.err_out)
1198
1199
1200class Pipeline(Job):
1201 """A pipeline of processes to run.
1202
1203 Cases we handle:
1204
1205 foo | bar
1206 $(foo | bar)
1207 foo | bar | read v
1208 """
1209
1210 def __init__(self, sigpipe_status_ok, job_control, job_list, tracer):
1211 # type: (bool, JobControl, JobList, dev.Tracer) -> None
1212 Job.__init__(self)
1213 self.job_control = job_control
1214 self.job_list = job_list
1215 self.tracer = tracer
1216
1217 self.procs = [] # type: List[Process]
1218 self.pids = [] # type: List[int] # pids in order
1219 self.pipe_status = [] # type: List[int] # status in order
1220 self.status = -1 # for 'wait' jobs
1221
1222 self.pgid = INVALID_PGID
1223
1224 # Optional for foreground
1225 self.last_thunk = None # type: Tuple[CommandEvaluator, command_t]
1226 self.last_pipe = None # type: Tuple[int, int]
1227
1228 self.sigpipe_status_ok = sigpipe_status_ok
1229
1230 def ProcessGroupId(self):
1231 # type: () -> int
1232 """Returns the group ID of this pipeline."""
1233 return self.pgid
1234
1235 def DisplayJob(self, job_id, f, style):
1236 # type: (int, mylib.Writer, int) -> None
1237 if style == STYLE_PID_ONLY:
1238 f.write('%d\n' % self.procs[0].pid)
1239 else:
1240 # Note: this is STYLE_LONG.
1241 for i, proc in enumerate(self.procs):
1242 if i == 0: # show job ID for first element in pipeline
1243 job_id_str = '%%%d' % job_id
1244 else:
1245 job_id_str = ' ' # 2 spaces
1246
1247 f.write('%s %d %7s ' %
1248 (job_id_str, proc.pid, _JobStateStr(proc.state)))
1249 f.write(proc.thunk.UserString())
1250 f.write('\n')
1251
1252 def DebugPrint(self):
1253 # type: () -> None
1254 print('Pipeline in state %s' % _JobStateStr(self.state))
1255 if mylib.PYTHON: # %s for Process not allowed in C++
1256 for proc in self.procs:
1257 print(' proc %s' % proc)
1258 _, last_node = self.last_thunk
1259 print(' last %s' % last_node)
1260 print(' pipe_status %s' % self.pipe_status)
1261
1262 def Add(self, p):
1263 # type: (Process) -> None
1264 """Append a process to the pipeline."""
1265 if len(self.procs) == 0:
1266 self.procs.append(p)
1267 return
1268
1269 r, w = posix.pipe()
1270 #log('pipe for %s: %d %d', p, r, w)
1271 prev = self.procs[-1]
1272
1273 prev.AddStateChange(StdoutToPipe(r, w)) # applied on StartPipeline()
1274 p.AddStateChange(StdinFromPipe(r, w)) # applied on StartPipeline()
1275
1276 p.AddPipeToClose(r, w) # MaybeClosePipe() on StartPipeline()
1277
1278 self.procs.append(p)
1279
1280 def AddLast(self, thunk):
1281 # type: (Tuple[CommandEvaluator, command_t]) -> None
1282 """Append the last noden to the pipeline.
1283
1284 This is run in the CURRENT process. It is OPTIONAL, because
1285 pipelines in the background are run uniformly.
1286 """
1287 self.last_thunk = thunk
1288
1289 assert len(self.procs) != 0
1290
1291 r, w = posix.pipe()
1292 prev = self.procs[-1]
1293 prev.AddStateChange(StdoutToPipe(r, w))
1294
1295 self.last_pipe = (r, w) # So we can connect it to last_thunk
1296
1297 def StartPipeline(self, waiter):
1298 # type: (Waiter) -> None
1299
1300 # If we are creating a pipeline in a subshell or we aren't running with job
1301 # control, our children should remain in our inherited process group.
1302 # the pipelines's group ID.
1303 if self.job_control.Enabled():
1304 self.pgid = OWN_LEADER # first process in pipeline is the leader
1305
1306 for i, proc in enumerate(self.procs):
1307 if self.pgid != INVALID_PGID:
1308 proc.AddStateChange(SetPgid(self.pgid, self.tracer))
1309
1310 # Figure out the pid
1311 pid = proc.StartProcess(trace.PipelinePart)
1312 if i == 0 and self.pgid != INVALID_PGID:
1313 # Mimic bash and use the PID of the FIRST process as the group for the
1314 # whole pipeline.
1315 self.pgid = pid
1316
1317 self.pids.append(pid)
1318 self.pipe_status.append(-1) # uninitialized
1319
1320 # NOTE: This is done in the SHELL PROCESS after every fork() call.
1321 # It can't be done at the end; otherwise processes will have descriptors
1322 # from non-adjacent pipes.
1323 proc.MaybeClosePipe()
1324
1325 if self.last_thunk:
1326 self.pipe_status.append(-1) # for self.last_thunk
1327
1328 def LastPid(self):
1329 # type: () -> int
1330 """For the odd $! variable.
1331
1332 It would be better if job IDs or PGIDs were used consistently.
1333 """
1334 return self.pids[-1]
1335
1336 def Wait(self, waiter):
1337 # type: (Waiter) -> List[int]
1338 """Wait for this pipeline to finish."""
1339
1340 assert self.procs, "no procs for Wait()"
1341 # waitpid(-1) zero or more times
1342 while self.state == job_state_e.Running:
1343 # Keep waiting until there's nothing to wait for.
1344 if waiter.WaitForOne() == W1_ECHILD:
1345 break
1346
1347 return self.pipe_status
1348
1349 def JobWait(self, waiter):
1350 # type: (Waiter) -> wait_status_t
1351 """Called by 'wait' builtin, e.g. 'wait %1'."""
1352 # wait builtin can be interrupted
1353 assert self.procs, "no procs for Wait()"
1354 while self.state == job_state_e.Running:
1355 result = waiter.WaitForOne()
1356
1357 if result >= 0: # signal
1358 return wait_status.Cancelled(result)
1359
1360 if result == W1_ECHILD:
1361 break
1362
1363 return wait_status.Pipeline(self.pipe_status)
1364
1365 def RunLastPart(self, waiter, fd_state):
1366 # type: (Waiter, FdState) -> List[int]
1367 """Run this pipeline synchronously (foreground pipeline).
1368
1369 Returns:
1370 pipe_status (list of integers).
1371 """
1372 assert len(self.pids) == len(self.procs)
1373
1374 # TODO: break circular dep. Bit flags could go in ASDL or headers.
1375 from osh import cmd_eval
1376
1377 # This is tcsetpgrp()
1378 # TODO: fix race condition -- I believe the first process could have
1379 # stopped already, and thus getpgid() will fail
1380 self.job_control.MaybeGiveTerminal(self.pgid)
1381
1382 # Run the last part of the pipeline IN PARALLEL with other processes. It
1383 # may or may not fork:
1384 # echo foo | read line # no fork, the builtin runs in THIS shell process
1385 # ls | wc -l # fork for 'wc'
1386
1387 cmd_ev, last_node = self.last_thunk
1388
1389 assert self.last_pipe is not None
1390 r, w = self.last_pipe # set in AddLast()
1391 posix.close(w) # we will not write here
1392
1393 # Fix lastpipe / job control / DEBUG trap interaction
1394 cmd_flags = cmd_eval.NoDebugTrap if self.job_control.Enabled() else 0
1395
1396 # The ERR trap only runs for the WHOLE pipeline, not the COMPONENTS in
1397 # a pipeline.
1398 cmd_flags |= cmd_eval.NoErrTrap
1399
1400 io_errors = [] # type: List[error.IOError_OSError]
1401 with ctx_Pipe(fd_state, r, io_errors):
1402 cmd_ev.ExecuteAndCatch(last_node, cmd_flags)
1403
1404 if len(io_errors):
1405 e_die('Error setting up last part of pipeline: %s' %
1406 pyutil.strerror(io_errors[0]))
1407
1408 # We won't read anymore. If we don't do this, then 'cat' in 'cat
1409 # /dev/urandom | sleep 1' will never get SIGPIPE.
1410 posix.close(r)
1411
1412 self.pipe_status[-1] = cmd_ev.LastStatus()
1413 if self.AllDone():
1414 self.state = job_state_e.Done
1415
1416 #log('pipestatus before all have finished = %s', self.pipe_status)
1417 return self.Wait(waiter)
1418
1419 def AllDone(self):
1420 # type: () -> bool
1421
1422 # mycpp rewrite: all(status != -1 for status in self.pipe_status)
1423 for status in self.pipe_status:
1424 if status == -1:
1425 return False
1426 return True
1427
1428 def WhenDone(self, pid, status):
1429 # type: (int, int) -> None
1430 """Called by Process.WhenDone."""
1431 #log('Pipeline WhenDone %d %d', pid, status)
1432 i = self.pids.index(pid)
1433 assert i != -1, 'Unexpected PID %d' % pid
1434
1435 if status == 141 and self.sigpipe_status_ok:
1436 status = 0
1437
1438 self.job_list.RemoveChildProcess(pid)
1439 self.pipe_status[i] = status
1440 if self.AllDone():
1441 if self.job_id != -1:
1442 # Job might have been brought to the foreground after being
1443 # assigned a job ID.
1444 if self.in_background:
1445 print_stderr('[%d] Done PGID %d' %
1446 (self.job_id, self.pids[0]))
1447
1448 self.job_list.RemoveJob(self.job_id)
1449
1450 # status of pipeline is status of last process
1451 self.status = self.pipe_status[-1]
1452 self.state = job_state_e.Done
1453 if not self.in_background:
1454 self.job_control.MaybeTakeTerminal()
1455
1456
1457def _JobStateStr(i):
1458 # type: (job_state_t) -> str
1459 return job_state_str(i)[10:] # remove 'job_state.'
1460
1461
1462def _GetTtyFd():
1463 # type: () -> int
1464 """Returns -1 if stdio is not a TTY."""
1465 try:
1466 return posix.open("/dev/tty", O_NONBLOCK | O_NOCTTY | O_RDWR, 0o666)
1467 except (IOError, OSError) as e:
1468 return -1
1469
1470
1471class ctx_TerminalControl(object):
1472
1473 def __init__(self, job_control, errfmt):
1474 # type: (JobControl, ui.ErrorFormatter) -> None
1475 job_control.InitJobControl()
1476 self.job_control = job_control
1477 self.errfmt = errfmt
1478
1479 def __enter__(self):
1480 # type: () -> None
1481 pass
1482
1483 def __exit__(self, type, value, traceback):
1484 # type: (Any, Any, Any) -> None
1485
1486 # Return the TTY to the original owner before exiting.
1487 try:
1488 self.job_control.MaybeReturnTerminal()
1489 except error.FatalRuntime as e:
1490 # Don't abort the shell on error, just print a message.
1491 self.errfmt.PrettyPrintError(e)
1492
1493
1494class JobControl(object):
1495 """Interface to setpgid(), tcsetpgrp(), etc."""
1496
1497 def __init__(self):
1498 # type: () -> None
1499
1500 # The main shell's PID and group ID.
1501 self.shell_pid = -1
1502 self.shell_pgid = -1
1503
1504 # The fd of the controlling tty. Set to -1 when job control is disabled.
1505 self.shell_tty_fd = -1
1506
1507 # For giving the terminal back to our parent before exiting (if not a login
1508 # shell).
1509 self.original_tty_pgid = -1
1510
1511 def InitJobControl(self):
1512 # type: () -> None
1513 self.shell_pid = posix.getpid()
1514 orig_shell_pgid = posix.getpgid(0)
1515 self.shell_pgid = orig_shell_pgid
1516 self.shell_tty_fd = _GetTtyFd()
1517
1518 # If we aren't the leader of our process group, create a group and mark
1519 # ourselves as the leader.
1520 if self.shell_pgid != self.shell_pid:
1521 try:
1522 posix.setpgid(self.shell_pid, self.shell_pid)
1523 self.shell_pgid = self.shell_pid
1524 except (IOError, OSError) as e:
1525 self.shell_tty_fd = -1
1526
1527 if self.shell_tty_fd != -1:
1528 self.original_tty_pgid = posix.tcgetpgrp(self.shell_tty_fd)
1529
1530 # If stdio is a TTY, put the shell's process group in the foreground.
1531 try:
1532 posix.tcsetpgrp(self.shell_tty_fd, self.shell_pgid)
1533 except (IOError, OSError) as e:
1534 # We probably aren't in the session leader's process group. Disable job
1535 # control.
1536 self.shell_tty_fd = -1
1537 self.shell_pgid = orig_shell_pgid
1538 posix.setpgid(self.shell_pid, self.shell_pgid)
1539
1540 def Enabled(self):
1541 # type: () -> bool
1542
1543 # TODO: get rid of this syscall? SubProgramThunk should set a flag I
1544 # think.
1545 curr_pid = posix.getpid()
1546 # Only the main shell should bother with job control functions.
1547 return curr_pid == self.shell_pid and self.shell_tty_fd != -1
1548
1549 # TODO: This isn't a PID. This is a process group ID?
1550 #
1551 # What should the table look like?
1552 #
1553 # Do we need the last PID? I don't know why bash prints that. Probably so
1554 # you can do wait $!
1555 # wait -n waits for any node to go from job_state_e.Running to job_state_e.Done?
1556 #
1557 # And it needs a flag for CURRENT, for the implicit arg to 'fg'.
1558 # job_id is just an integer. This is sort of lame.
1559 #
1560 # [job_id, flag, pgid, job_state, node]
1561
1562 def MaybeGiveTerminal(self, pgid):
1563 # type: (int) -> None
1564 """If stdio is a TTY, move the given process group to the
1565 foreground."""
1566 if not self.Enabled():
1567 # Only call tcsetpgrp when job control is enabled.
1568 return
1569
1570 try:
1571 posix.tcsetpgrp(self.shell_tty_fd, pgid)
1572 except (IOError, OSError) as e:
1573 e_die('osh: Failed to move process group %d to foreground: %s' %
1574 (pgid, pyutil.strerror(e)))
1575
1576 def MaybeTakeTerminal(self):
1577 # type: () -> None
1578 """If stdio is a TTY, return the main shell's process group to the
1579 foreground."""
1580 self.MaybeGiveTerminal(self.shell_pgid)
1581
1582 def MaybeReturnTerminal(self):
1583 # type: () -> None
1584 """Called before the shell exits."""
1585 self.MaybeGiveTerminal(self.original_tty_pgid)
1586
1587
1588class JobList(object):
1589 """Global list of jobs, used by a few builtins."""
1590
1591 def __init__(self):
1592 # type: () -> None
1593
1594 # job_id -> Job instance
1595 self.jobs = {} # type: Dict[int, Job]
1596
1597 # pid -> Process. This is for STOP notification.
1598 self.child_procs = {} # type: Dict[int, Process]
1599 self.debug_pipelines = [] # type: List[Pipeline]
1600
1601 # Counter used to assign IDs to jobs. It is incremented every time a job
1602 # is created. Once all active jobs are done it is reset to 1. I'm not
1603 # sure if this reset behavior is mandated by POSIX, but other shells do
1604 # it, so we mimic for the sake of compatibility.
1605 self.job_id = 1
1606
1607 def AddJob(self, job):
1608 # type: (Job) -> int
1609 """Add a background job to the list.
1610
1611 A job is either a Process or Pipeline. You can resume a job with 'fg',
1612 kill it with 'kill', etc.
1613
1614 Two cases:
1615
1616 1. async jobs: sleep 5 | sleep 4 &
1617 2. stopped jobs: sleep 5; then Ctrl-Z
1618 """
1619 job_id = self.job_id
1620 self.jobs[job_id] = job
1621 job.job_id = job_id
1622 self.job_id += 1
1623 return job_id
1624
1625 def RemoveJob(self, job_id):
1626 # type: (int) -> None
1627 """Process and Pipeline can call this."""
1628 mylib.dict_erase(self.jobs, job_id)
1629
1630 if len(self.jobs) == 0:
1631 self.job_id = 1
1632
1633 def AddChildProcess(self, pid, proc):
1634 # type: (int, Process) -> None
1635 """Every child process should be added here as soon as we know its PID.
1636
1637 When the Waiter gets an EXITED or STOPPED notification, we need
1638 to know about it so 'jobs' can work.
1639 """
1640 self.child_procs[pid] = proc
1641
1642 def RemoveChildProcess(self, pid):
1643 # type: (int) -> None
1644 """Remove the child process with the given PID."""
1645 mylib.dict_erase(self.child_procs, pid)
1646
1647 if mylib.PYTHON:
1648
1649 def AddPipeline(self, pi):
1650 # type: (Pipeline) -> None
1651 """For debugging only."""
1652 self.debug_pipelines.append(pi)
1653
1654 def ProcessFromPid(self, pid):
1655 # type: (int) -> Process
1656 """For wait $PID.
1657
1658 There's no way to wait for a pipeline with a PID. That uses job
1659 syntax, e.g. %1. Not a great interface.
1660 """
1661 return self.child_procs.get(pid)
1662
1663 def GetCurrentAndPreviousJobs(self):
1664 # type: () -> Tuple[Optional[Job], Optional[Job]]
1665 """Return the "current" and "previous" jobs (AKA `%+` and `%-`).
1666
1667 See the POSIX specification for the `jobs` builtin for details:
1668 https://pubs.opengroup.org/onlinepubs/007904875/utilities/jobs.html
1669
1670 IMPORTANT NOTE: This method assumes that the jobs list will not change
1671 during its execution! This assumption holds for now because we only ever
1672 update the jobs list from the main loop after WaitPid() informs us of a
1673 change. If we implement `set -b` and install a signal handler for
1674 SIGCHLD we should be careful to synchronize it with this function. The
1675 unsafety of mutating GC data structures from a signal handler should
1676 make this a non-issue, but if bugs related to this appear this note may
1677 be helpful...
1678 """
1679 # Split all active jobs by state and sort each group by decreasing job
1680 # ID to approximate newness.
1681 stopped_jobs = [] # type: List[Job]
1682 running_jobs = [] # type: List[Job]
1683 for i in xrange(0, self.job_id):
1684 job = self.jobs.get(i, None)
1685 if not job:
1686 continue
1687
1688 if job.state == job_state_e.Stopped:
1689 stopped_jobs.append(job)
1690
1691 elif job.state == job_state_e.Running:
1692 running_jobs.append(job)
1693
1694 current = None # type: Optional[Job]
1695 previous = None # type: Optional[Job]
1696 # POSIX says: If there is any suspended job, then the current job shall
1697 # be a suspended job. If there are at least two suspended jobs, then the
1698 # previous job also shall be a suspended job.
1699 #
1700 # So, we will only return running jobs from here if there are no recent
1701 # stopped jobs.
1702 if len(stopped_jobs) > 0:
1703 current = stopped_jobs.pop()
1704
1705 if len(stopped_jobs) > 0:
1706 previous = stopped_jobs.pop()
1707
1708 if len(running_jobs) > 0 and not current:
1709 current = running_jobs.pop()
1710
1711 if len(running_jobs) > 0 and not previous:
1712 previous = running_jobs.pop()
1713
1714 if not previous:
1715 previous = current
1716
1717 return current, previous
1718
1719 def GetJobWithSpec(self, job_spec):
1720 # type: (str) -> Optional[Job]
1721 """Parse the given job spec and return the matching job. If there is no
1722 matching job, this function returns None.
1723
1724 See the POSIX spec for the `jobs` builtin for details about job specs:
1725 https://pubs.opengroup.org/onlinepubs/007904875/utilities/jobs.html
1726 """
1727 if job_spec in CURRENT_JOB_SPECS:
1728 current, _ = self.GetCurrentAndPreviousJobs()
1729 return current
1730
1731 if job_spec == '%-':
1732 _, previous = self.GetCurrentAndPreviousJobs()
1733 return previous
1734
1735 # TODO: Add support for job specs based on prefixes of process argv.
1736 m = util.RegexSearch(r'^%([0-9]+)$', job_spec)
1737 if m is not None:
1738 assert len(m) == 2
1739 job_id = int(m[1])
1740 if job_id in self.jobs:
1741 return self.jobs[job_id]
1742
1743 return None
1744
1745 def DisplayJobs(self, style):
1746 # type: (int) -> None
1747 """Used by the 'jobs' builtin.
1748
1749 https://pubs.opengroup.org/onlinepubs/9699919799/utilities/jobs.html
1750
1751 "By default, the jobs utility shall display the status of all stopped jobs,
1752 running background jobs and all jobs whose status has changed and have not
1753 been reported by the shell."
1754 """
1755 # NOTE: A job is a background process or pipeline.
1756 #
1757 # echo hi | wc -l -- this starts two processes. Wait for TWO
1758 # echo hi | wc -l & -- this starts a process which starts two processes
1759 # Wait for ONE.
1760 #
1761 # 'jobs -l' GROUPS the PIDs by job. It has the job number, + - indicators
1762 # for %% and %-, PID, status, and "command".
1763 #
1764 # Every component of a pipeline is on the same line with 'jobs', but
1765 # they're separated into different lines with 'jobs -l'.
1766 #
1767 # See demo/jobs-builtin.sh
1768
1769 # $ jobs -l
1770 # [1]+ 24414 Stopped sleep 5
1771 # 24415 | sleep 5
1772 # [2] 24502 Running sleep 6
1773 # 24503 | sleep 6
1774 # 24504 | sleep 5 &
1775 # [3]- 24508 Running sleep 6
1776 # 24509 | sleep 6
1777 # 24510 | sleep 5 &
1778
1779 f = mylib.Stdout()
1780 for job_id, job in iteritems(self.jobs):
1781 # Use the %1 syntax
1782 job.DisplayJob(job_id, f, style)
1783
1784 def DebugPrint(self):
1785 # type: () -> None
1786
1787 f = mylib.Stdout()
1788 f.write('\n')
1789 f.write('[process debug info]\n')
1790
1791 for pid, proc in iteritems(self.child_procs):
1792 proc.DisplayJob(-1, f, STYLE_DEFAULT)
1793 #p = ' |' if proc.parent_pipeline else ''
1794 #print('%d %7s %s%s' % (pid, _JobStateStr(proc.state), proc.thunk.UserString(), p))
1795
1796 if len(self.debug_pipelines):
1797 f.write('\n')
1798 f.write('[pipeline debug info]\n')
1799 for pi in self.debug_pipelines:
1800 pi.DebugPrint()
1801
1802 def ListRecent(self):
1803 # type: () -> None
1804 """For jobs -n, which I think is also used in the interactive
1805 prompt."""
1806 pass
1807
1808 def NumRunning(self):
1809 # type: () -> int
1810 """Return the number of running jobs.
1811
1812 Used by 'wait' and 'wait -n'.
1813 """
1814 count = 0
1815 for _, job in iteritems(self.jobs): # mycpp rewrite: from itervalues()
1816 if job.State() == job_state_e.Running:
1817 count += 1
1818 return count
1819
1820
1821# Some WaitForOne() return values
1822W1_OK = -2 # waitpid(-1) returned
1823W1_ECHILD = -3 # no processes to wait for
1824W1_AGAIN = -4 # WNOHANG was passed and there were no state changes
1825
1826
1827class Waiter(object):
1828 """A capability to wait for processes.
1829
1830 This must be a singleton (and is because CommandEvaluator is a singleton).
1831
1832 Invariants:
1833 - Every child process is registered once
1834 - Every child process is waited for
1835
1836 Canonical example of why we need a GLOBAL waiter:
1837
1838 { sleep 3; echo 'done 3'; } &
1839 { sleep 4; echo 'done 4'; } &
1840
1841 # ... do arbitrary stuff ...
1842
1843 { sleep 1; exit 1; } | { sleep 2; exit 2; }
1844
1845 Now when you do wait() after starting the pipeline, you might get a pipeline
1846 process OR a background process! So you have to distinguish between them.
1847 """
1848
1849 def __init__(self, job_list, exec_opts, signal_safe, tracer):
1850 # type: (JobList, optview.Exec, pyos.SignalSafe, dev.Tracer) -> None
1851 self.job_list = job_list
1852 self.exec_opts = exec_opts
1853 self.signal_safe = signal_safe
1854 self.tracer = tracer
1855 self.last_status = 127 # wait -n error code
1856
1857 def WaitForOne(self, waitpid_options=0):
1858 # type: (int) -> int
1859 """Wait until the next process returns (or maybe Ctrl-C).
1860
1861 Returns:
1862 One of these negative numbers:
1863 W1_ECHILD Nothing to wait for
1864 W1_OK Caller should keep waiting
1865 UNTRAPPED_SIGWINCH
1866 Or
1867 result > 0 Signal that waitpid() was interrupted with
1868
1869 In the interactive shell, we return 0 if we get a Ctrl-C, so the caller
1870 will try again.
1871
1872 Callers:
1873 wait -n -- loop until there is one fewer process (TODO)
1874 wait -- loop until there are no processes
1875 wait $! -- loop until job state is Done (process or pipeline)
1876 Process::Wait() -- loop until Process state is done
1877 Pipeline::Wait() -- loop until Pipeline state is done
1878
1879 Comparisons:
1880 bash: jobs.c waitchld() Has a special case macro(!) CHECK_WAIT_INTR for
1881 the wait builtin
1882
1883 dash: jobs.c waitproc() uses sigfillset(), sigprocmask(), etc. Runs in a
1884 loop while (gotsigchld), but that might be a hack for System V!
1885
1886 Should we have a cleaner API like named posix::wait_for_one() ?
1887
1888 wait_result =
1889 ECHILD -- nothing to wait for
1890 | Done(int pid, int status) -- process done
1891 | EINTR(bool sigint) -- may or may not retry
1892 """
1893 pid, status = pyos.WaitPid(waitpid_options)
1894 if pid == 0: # WNOHANG passed, and no state changes
1895 return W1_AGAIN
1896 elif pid < 0: # error case
1897 err_num = status
1898 #log('waitpid() error => %d %s', e.errno, pyutil.strerror(e))
1899 if err_num == ECHILD:
1900 return W1_ECHILD # nothing to wait for caller should stop
1901 elif err_num == EINTR: # Bug #858 fix
1902 #log('WaitForOne() => %d', self.trap_state.GetLastSignal())
1903 return self.signal_safe.LastSignal() # e.g. 1 for SIGHUP
1904 else:
1905 # The signature of waitpid() means this shouldn't happen
1906 raise AssertionError()
1907
1908 # All child processes are supposed to be in this dict. But this may
1909 # legitimately happen if a grandchild outlives the child (its parent).
1910 # Then it is reparented under this process, so we might receive
1911 # notification of its exit, even though we didn't start it. We can't have
1912 # any knowledge of such processes, so print a warning.
1913 if pid not in self.job_list.child_procs:
1914 print_stderr("osh: PID %d stopped, but osh didn't start it" % pid)
1915 return W1_OK
1916
1917 proc = self.job_list.child_procs[pid]
1918 if 0:
1919 self.job_list.DebugPrint()
1920
1921 if WIFSIGNALED(status):
1922 term_sig = WTERMSIG(status)
1923 status = 128 + term_sig
1924
1925 # Print newline after Ctrl-C.
1926 if term_sig == SIGINT:
1927 print('')
1928
1929 proc.WhenDone(pid, status)
1930
1931 elif WIFEXITED(status):
1932 status = WEXITSTATUS(status)
1933 #log('exit status: %s', status)
1934 proc.WhenDone(pid, status)
1935
1936 elif WIFSTOPPED(status):
1937 #status = WEXITSTATUS(status)
1938 stop_sig = WSTOPSIG(status)
1939
1940 print_stderr('')
1941 print_stderr('[PID %d] Stopped with signal %d' % (pid, stop_sig))
1942 proc.WhenStopped(stop_sig)
1943
1944 else:
1945 raise AssertionError(status)
1946
1947 self.last_status = status # for wait -n
1948 self.tracer.OnProcessEnd(pid, status)
1949 return W1_OK
1950
1951 def PollNotifications(self):
1952 # type: () -> None
1953 """
1954 Process all pending state changes.
1955 """
1956 while self.WaitForOne(waitpid_options=WNOHANG) == W1_OK:
1957 continue