1/* Waiting for a subprocess to finish.
2   Copyright (C) 2001-2003, 2005-2006 Free Software Foundation, Inc.
3   Written by Bruno Haible <haible@clisp.cons.org>, 2001.
4
5   This program is free software; you can redistribute it and/or modify
6   it under the terms of the GNU General Public License as published by
7   the Free Software Foundation; either version 2, or (at your option)
8   any later version.
9
10   This program is distributed in the hope that it will be useful,
11   but WITHOUT ANY WARRANTY; without even the implied warranty of
12   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13   GNU General Public License for more details.
14
15   You should have received a copy of the GNU General Public License
16   along with this program; if not, write to the Free Software Foundation,
17   Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.  */
18
19
20#include <config.h>
21
22/* Specification.  */
23#include "wait-process.h"
24
25#include <errno.h>
26#include <stdlib.h>
27#include <string.h>
28#include <signal.h>
29
30#include <sys/types.h>
31
32#if defined _MSC_VER || defined __MINGW32__
33
34/* Native Woe32 API.  */
35#include <process.h>
36#define waitpid(pid,statusp,options) _cwait (statusp, pid, WAIT_CHILD)
37#define WAIT_T int
38#define WTERMSIG(x) ((x) & 0xff) /* or: SIGABRT ?? */
39#define WCOREDUMP(x) 0
40#define WEXITSTATUS(x) (((x) >> 8) & 0xff) /* or: (x) ?? */
41#define WIFSIGNALED(x) (WTERMSIG (x) != 0) /* or: ((x) == 3) ?? */
42#define WIFEXITED(x) (WTERMSIG (x) == 0) /* or: ((x) != 3) ?? */
43#define WIFSTOPPED(x) 0
44
45#else
46
47/* Unix API.  */
48#include <sys/wait.h>
49/* On Linux, WEXITSTATUS are bits 15..8 and WTERMSIG are bits 7..0, while
50   BeOS uses the contrary.  Therefore we use the abstract macros.  */
51#if HAVE_UNION_WAIT
52# define WAIT_T union wait
53# ifndef WTERMSIG
54#  define WTERMSIG(x) ((x).w_termsig)
55# endif
56# ifndef WCOREDUMP
57#  define WCOREDUMP(x) ((x).w_coredump)
58# endif
59# ifndef WEXITSTATUS
60#  define WEXITSTATUS(x) ((x).w_retcode)
61# endif
62#else
63# define WAIT_T int
64# ifndef WTERMSIG
65#  define WTERMSIG(x) ((x) & 0x7f)
66# endif
67# ifndef WCOREDUMP
68#  define WCOREDUMP(x) ((x) & 0x80)
69# endif
70# ifndef WEXITSTATUS
71#  define WEXITSTATUS(x) (((x) >> 8) & 0xff)
72# endif
73#endif
74/* For valid x, exactly one of WIFSIGNALED(x), WIFEXITED(x), WIFSTOPPED(x)
75   is true.  */
76#ifndef WIFSIGNALED
77# define WIFSIGNALED(x) (WTERMSIG (x) != 0 && WTERMSIG(x) != 0x7f)
78#endif
79#ifndef WIFEXITED
80# define WIFEXITED(x) (WTERMSIG (x) == 0)
81#endif
82#ifndef WIFSTOPPED
83# define WIFSTOPPED(x) (WTERMSIG (x) == 0x7f)
84#endif
85/* Note that portable applications may access
86   WTERMSIG(x) only if WIFSIGNALED(x) is true, and
87   WEXITSTATUS(x) only if WIFEXITED(x) is true.  */
88
89#endif
90
91#include "error.h"
92#include "exit.h"
93#include "fatal-signal.h"
94#include "xalloc.h"
95#include "gettext.h"
96
97#define _(str) gettext (str)
98
99#define SIZEOF(a) (sizeof(a) / sizeof(a[0]))
100
101
102#if defined _MSC_VER || defined __MINGW32__
103
104#define WIN32_LEAN_AND_MEAN
105#include <windows.h>
106
107/* The return value of spawnvp() is really a process handle as returned
108   by CreateProcess().  Therefore we can kill it using TerminateProcess.  */
109#define kill(pid,sig) TerminateProcess ((HANDLE) (pid), sig)
110
111#endif
112
113
114/* Type of an entry in the slaves array.
115   The 'used' bit determines whether this entry is currently in use.
116   (If pid_t was an atomic type like sig_atomic_t, we could just set the
117   'child' field to 0 when unregistering a slave process, and wouldn't need
118   the 'used' field.)
119   The 'used' and 'child' fields are accessed from within the cleanup_slaves()
120   action, therefore we mark them as 'volatile'.  */
121typedef struct
122{
123  volatile sig_atomic_t used;
124  volatile pid_t child;
125}
126slaves_entry_t;
127
128/* The registered slave subprocesses.  */
129static slaves_entry_t static_slaves[32];
130static slaves_entry_t * volatile slaves = static_slaves;
131static sig_atomic_t volatile slaves_count = 0;
132static size_t slaves_allocated = SIZEOF (static_slaves);
133
134/* The termination signal for slave subprocesses.
135   2003-10-07:  Terminator becomes Governator.  */
136#ifdef SIGHUP
137# define TERMINATOR SIGHUP
138#else
139# define TERMINATOR SIGTERM
140#endif
141
142/* The cleanup action.  It gets called asynchronously.  */
143static void
144cleanup_slaves (void)
145{
146  for (;;)
147    {
148      /* Get the last registered slave.  */
149      size_t n = slaves_count;
150      if (n == 0)
151	break;
152      n--;
153      slaves_count = n;
154      /* Skip unused entries in the slaves array.  */
155      if (slaves[n].used)
156	{
157	  pid_t slave = slaves[n].child;
158
159	  /* Kill the slave.  */
160	  kill (slave, TERMINATOR);
161	}
162    }
163}
164
165/* Register a subprocess as being a slave process.  This means that the
166   subprocess will be terminated when its creator receives a catchable fatal
167   signal or exits normally.  Registration ends when wait_subprocess()
168   notices that the subprocess has exited.  */
169void
170register_slave_subprocess (pid_t child)
171{
172  static bool cleanup_slaves_registered = false;
173  if (!cleanup_slaves_registered)
174    {
175      atexit (cleanup_slaves);
176      at_fatal_signal (cleanup_slaves);
177      cleanup_slaves_registered = true;
178    }
179
180  /* Try to store the new slave in an unused entry of the slaves array.  */
181  {
182    slaves_entry_t *s = slaves;
183    slaves_entry_t *s_end = s + slaves_count;
184
185    for (; s < s_end; s++)
186      if (!s->used)
187	{
188	  /* The two uses of 'volatile' in the slaves_entry_t type above
189	     (and ISO C 99 section 5.1.2.3.(5)) ensure that we mark the
190	     entry as used only after the child pid has been written to the
191	     memory location s->child.  */
192	  s->child = child;
193	  s->used = 1;
194	  return;
195	}
196  }
197
198  if (slaves_count == slaves_allocated)
199    {
200      /* Extend the slaves array.  Note that we cannot use xrealloc(),
201	 because then the cleanup_slaves() function could access an already
202	 deallocated array.  */
203      slaves_entry_t *old_slaves = slaves;
204      size_t new_slaves_allocated = 2 * slaves_allocated;
205      slaves_entry_t *new_slaves =
206	malloc (new_slaves_allocated * sizeof (slaves_entry_t));
207      if (new_slaves == NULL)
208	{
209	  /* xalloc_die() will call exit() which will invoke cleanup_slaves().
210	     Additionally we need to kill child, because it's not yet among
211	     the slaves list.  */
212	  kill (child, TERMINATOR);
213	  xalloc_die ();
214	}
215      memcpy (new_slaves, old_slaves,
216	      slaves_allocated * sizeof (slaves_entry_t));
217      slaves = new_slaves;
218      slaves_allocated = new_slaves_allocated;
219      /* Now we can free the old slaves array.  */
220      if (old_slaves != static_slaves)
221	free (old_slaves);
222    }
223  /* The three uses of 'volatile' in the types above (and ISO C 99 section
224     5.1.2.3.(5)) ensure that we increment the slaves_count only after the
225     new slave and its 'used' bit have been written to the memory locations
226     that make up slaves[slaves_count].  */
227  slaves[slaves_count].child = child;
228  slaves[slaves_count].used = 1;
229  slaves_count++;
230}
231
232/* Unregister a child from the list of slave subprocesses.  */
233static inline void
234unregister_slave_subprocess (pid_t child)
235{
236  /* The easiest way to remove an entry from a list that can be used by
237     an asynchronous signal handler is just to mark it as unused.  For this,
238     we rely on sig_atomic_t.  */
239  slaves_entry_t *s = slaves;
240  slaves_entry_t *s_end = s + slaves_count;
241
242  for (; s < s_end; s++)
243    if (s->used && s->child == child)
244      s->used = 0;
245}
246
247
248/* Wait for a subprocess to finish.  Return its exit code.
249   If it didn't terminate correctly, exit if exit_on_error is true, otherwise
250   return 127.  */
251int
252wait_subprocess (pid_t child, const char *progname,
253		 bool ignore_sigpipe, bool null_stderr,
254		 bool slave_process, bool exit_on_error)
255{
256#if HAVE_WAITID && defined WNOWAIT && 0
257  /* Commented out because waitid() with WNOWAIT doesn't work: On Solaris 7
258     and OSF/1 4.0, it returns -1 and sets errno = ECHILD, and on HP-UX 10.20
259     it just hangs.  */
260  /* Use of waitid() with WNOWAIT avoids a race condition: If slave_process is
261     true, and this process sleeps a very long time between the return from
262     waitpid() and the execution of unregister_slave_subprocess(), and
263     meanwhile another process acquires the same PID as child, and then - still
264     before unregister_slave_subprocess() - this process gets a fatal signal,
265     it would kill the other totally unrelated process.  */
266  siginfo_t info;
267  for (;;)
268    {
269      if (waitid (P_PID, child, &info, slave_process ? WNOWAIT : 0) < 0)
270	{
271# ifdef EINTR
272	  if (errno == EINTR)
273	    continue;
274# endif
275	  if (exit_on_error || !null_stderr)
276	    error (exit_on_error ? EXIT_FAILURE : 0, errno,
277		   _("%s subprocess"), progname);
278	  return 127;
279	}
280
281      /* info.si_code is set to one of CLD_EXITED, CLD_KILLED, CLD_DUMPED,
282	 CLD_TRAPPED, CLD_STOPPED, CLD_CONTINUED.  Loop until the program
283	 terminates.  */
284      if (info.si_code == CLD_EXITED
285	  || info.si_code == CLD_KILLED || info.si_code == CLD_DUMPED)
286	break;
287    }
288
289  /* The child process has exited or was signalled.  */
290
291  if (slave_process)
292    {
293      /* Unregister the child from the list of slave subprocesses, so that
294	 later, when we exit, we don't kill a totally unrelated process which
295	 may have acquired the same pid.  */
296      unregister_slave_subprocess (child);
297
298      /* Now remove the zombie from the process list.  */
299      for (;;)
300	{
301	  if (waitid (P_PID, child, &info, 0) < 0)
302	    {
303# ifdef EINTR
304	      if (errno == EINTR)
305		continue;
306# endif
307	      if (exit_on_error || !null_stderr)
308		error (exit_on_error ? EXIT_FAILURE : 0, errno,
309		       _("%s subprocess"), progname);
310	      return 127;
311	    }
312	  break;
313	}
314    }
315
316  switch (info.si_code)
317    {
318    case CLD_KILLED:
319    case CLD_DUMPED:
320# ifdef SIGPIPE
321      if (info.si_status == SIGPIPE && ignore_sigpipe)
322	return 0;
323# endif
324      if (exit_on_error || !null_stderr)
325	error (exit_on_error ? EXIT_FAILURE : 0, 0,
326	       _("%s subprocess got fatal signal %d"),
327	       progname, info.si_status);
328      return 127;
329    case CLD_EXITED:
330      if (info.si_status == 127)
331	{
332	  if (exit_on_error || !null_stderr)
333	    error (exit_on_error ? EXIT_FAILURE : 0, 0,
334		   _("%s subprocess failed"), progname);
335	  return 127;
336	}
337      return info.si_status;
338    default:
339      abort ();
340    }
341#else
342  /* waitpid() is just as portable as wait() nowadays.  */
343  WAIT_T status;
344
345  *(int *) &status = 0;
346  for (;;)
347    {
348      int result = waitpid (child, &status, 0);
349
350      if (result != child)
351	{
352# ifdef EINTR
353	  if (errno == EINTR)
354	    continue;
355# endif
356# if 0 /* defined ECHILD */
357	  if (errno == ECHILD)
358	    {
359	      /* Child process nonexistent?! Assume it terminated
360		 successfully.  */
361	      *(int *) &status = 0;
362	      break;
363	    }
364# endif
365	  if (exit_on_error || !null_stderr)
366	    error (exit_on_error ? EXIT_FAILURE : 0, errno,
367		   _("%s subprocess"), progname);
368	  return 127;
369	}
370
371      /* One of WIFSIGNALED (status), WIFEXITED (status), WIFSTOPPED (status)
372	 must always be true.  Loop until the program terminates.  */
373      if (!WIFSTOPPED (status))
374	break;
375    }
376
377  /* The child process has exited or was signalled.  */
378
379  if (slave_process)
380    /* Unregister the child from the list of slave subprocesses, so that
381       later, when we exit, we don't kill a totally unrelated process which
382       may have acquired the same pid.  */
383    unregister_slave_subprocess (child);
384
385  if (WIFSIGNALED (status))
386    {
387# ifdef SIGPIPE
388      if (WTERMSIG (status) == SIGPIPE && ignore_sigpipe)
389	return 0;
390# endif
391      if (exit_on_error || !null_stderr)
392	error (exit_on_error ? EXIT_FAILURE : 0, 0,
393	       _("%s subprocess got fatal signal %d"),
394	       progname, (int) WTERMSIG (status));
395      return 127;
396    }
397  if (WEXITSTATUS (status) == 127)
398    {
399      if (exit_on_error || !null_stderr)
400	error (exit_on_error ? EXIT_FAILURE : 0, 0,
401	       _("%s subprocess failed"), progname);
402      return 127;
403    }
404  return WEXITSTATUS (status);
405#endif
406}
407