Home | History | Annotate | Download | only in lib
      1 /* Waiting for a subprocess to finish.
      2    Copyright (C) 2001-2003, 2005-2012 Free Software Foundation, Inc.
      3    Written by Bruno Haible <haible (at) 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 3 of the License, or
      8    (at your option) 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, see <http://www.gnu.org/licenses/>.  */
     17 
     18 
     19 #include <config.h>
     20 
     21 /* Specification.  */
     22 #include "wait-process.h"
     23 
     24 #include <errno.h>
     25 #include <stdlib.h>
     26 #include <string.h>
     27 #include <signal.h>
     28 
     29 #include <sys/types.h>
     30 #include <sys/wait.h>
     31 
     32 #include "error.h"
     33 #include "fatal-signal.h"
     34 #include "xalloc.h"
     35 #include "gettext.h"
     36 
     37 #define _(str) gettext (str)
     38 
     39 #define SIZEOF(a) (sizeof(a) / sizeof(a[0]))
     40 
     41 
     42 #if defined _MSC_VER || defined __MINGW32__
     43 
     44 #define WIN32_LEAN_AND_MEAN
     45 #include <windows.h>
     46 
     47 /* The return value of spawnvp() is really a process handle as returned
     48    by CreateProcess().  Therefore we can kill it using TerminateProcess.  */
     49 #define kill(pid,sig) TerminateProcess ((HANDLE) (pid), sig)
     50 
     51 #endif
     52 
     53 
     54 /* Type of an entry in the slaves array.
     55    The 'used' bit determines whether this entry is currently in use.
     56    (If pid_t was an atomic type like sig_atomic_t, we could just set the
     57    'child' field to 0 when unregistering a slave process, and wouldn't need
     58    the 'used' field.)
     59    The 'used' and 'child' fields are accessed from within the cleanup_slaves()
     60    action, therefore we mark them as 'volatile'.  */
     61 typedef struct
     62 {
     63   volatile sig_atomic_t used;
     64   volatile pid_t child;
     65 }
     66 slaves_entry_t;
     67 
     68 /* The registered slave subprocesses.  */
     69 static slaves_entry_t static_slaves[32];
     70 static slaves_entry_t * volatile slaves = static_slaves;
     71 static sig_atomic_t volatile slaves_count = 0;
     72 static size_t slaves_allocated = SIZEOF (static_slaves);
     73 
     74 /* The termination signal for slave subprocesses.
     75    2003-10-07:  Terminator becomes Governator.  */
     76 #ifdef SIGHUP
     77 # define TERMINATOR SIGHUP
     78 #else
     79 # define TERMINATOR SIGTERM
     80 #endif
     81 
     82 /* The cleanup action.  It gets called asynchronously.  */
     83 static void
     84 cleanup_slaves (void)
     85 {
     86   for (;;)
     87     {
     88       /* Get the last registered slave.  */
     89       size_t n = slaves_count;
     90       if (n == 0)
     91         break;
     92       n--;
     93       slaves_count = n;
     94       /* Skip unused entries in the slaves array.  */
     95       if (slaves[n].used)
     96         {
     97           pid_t slave = slaves[n].child;
     98 
     99           /* Kill the slave.  */
    100           kill (slave, TERMINATOR);
    101         }
    102     }
    103 }
    104 
    105 /* Register a subprocess as being a slave process.  This means that the
    106    subprocess will be terminated when its creator receives a catchable fatal
    107    signal or exits normally.  Registration ends when wait_subprocess()
    108    notices that the subprocess has exited.  */
    109 void
    110 register_slave_subprocess (pid_t child)
    111 {
    112   static bool cleanup_slaves_registered = false;
    113   if (!cleanup_slaves_registered)
    114     {
    115       atexit (cleanup_slaves);
    116       at_fatal_signal (cleanup_slaves);
    117       cleanup_slaves_registered = true;
    118     }
    119 
    120   /* Try to store the new slave in an unused entry of the slaves array.  */
    121   {
    122     slaves_entry_t *s = slaves;
    123     slaves_entry_t *s_end = s + slaves_count;
    124 
    125     for (; s < s_end; s++)
    126       if (!s->used)
    127         {
    128           /* The two uses of 'volatile' in the slaves_entry_t type above
    129              (and ISO C 99 section 5.1.2.3.(5)) ensure that we mark the
    130              entry as used only after the child pid has been written to the
    131              memory location s->child.  */
    132           s->child = child;
    133           s->used = 1;
    134           return;
    135         }
    136   }
    137 
    138   if (slaves_count == slaves_allocated)
    139     {
    140       /* Extend the slaves array.  Note that we cannot use xrealloc(),
    141          because then the cleanup_slaves() function could access an already
    142          deallocated array.  */
    143       slaves_entry_t *old_slaves = slaves;
    144       size_t new_slaves_allocated = 2 * slaves_allocated;
    145       slaves_entry_t *new_slaves =
    146         (slaves_entry_t *)
    147         malloc (new_slaves_allocated * sizeof (slaves_entry_t));
    148       if (new_slaves == NULL)
    149         {
    150           /* xalloc_die() will call exit() which will invoke cleanup_slaves().
    151              Additionally we need to kill child, because it's not yet among
    152              the slaves list.  */
    153           kill (child, TERMINATOR);
    154           xalloc_die ();
    155         }
    156       memcpy (new_slaves, old_slaves,
    157               slaves_allocated * sizeof (slaves_entry_t));
    158       slaves = new_slaves;
    159       slaves_allocated = new_slaves_allocated;
    160       /* Now we can free the old slaves array.  */
    161       if (old_slaves != static_slaves)
    162         free (old_slaves);
    163     }
    164   /* The three uses of 'volatile' in the types above (and ISO C 99 section
    165      5.1.2.3.(5)) ensure that we increment the slaves_count only after the
    166      new slave and its 'used' bit have been written to the memory locations
    167      that make up slaves[slaves_count].  */
    168   slaves[slaves_count].child = child;
    169   slaves[slaves_count].used = 1;
    170   slaves_count++;
    171 }
    172 
    173 /* Unregister a child from the list of slave subprocesses.  */
    174 static void
    175 unregister_slave_subprocess (pid_t child)
    176 {
    177   /* The easiest way to remove an entry from a list that can be used by
    178      an asynchronous signal handler is just to mark it as unused.  For this,
    179      we rely on sig_atomic_t.  */
    180   slaves_entry_t *s = slaves;
    181   slaves_entry_t *s_end = s + slaves_count;
    182 
    183   for (; s < s_end; s++)
    184     if (s->used && s->child == child)
    185       s->used = 0;
    186 }
    187 
    188 
    189 /* Wait for a subprocess to finish.  Return its exit code.
    190    If it didn't terminate correctly, exit if exit_on_error is true, otherwise
    191    return 127.  */
    192 int
    193 wait_subprocess (pid_t child, const char *progname,
    194                  bool ignore_sigpipe, bool null_stderr,
    195                  bool slave_process, bool exit_on_error,
    196                  int *termsigp)
    197 {
    198 #if HAVE_WAITID && defined WNOWAIT && 0
    199   /* Commented out because waitid() without WEXITED and with WNOWAIT doesn't
    200      work: On Solaris 7 and OSF/1 4.0, it returns -1 and sets errno = ECHILD,
    201      and on HP-UX 10.20 it just hangs.  */
    202   /* Use of waitid() with WNOWAIT avoids a race condition: If slave_process is
    203      true, and this process sleeps a very long time between the return from
    204      waitpid() and the execution of unregister_slave_subprocess(), and
    205      meanwhile another process acquires the same PID as child, and then - still
    206      before unregister_slave_subprocess() - this process gets a fatal signal,
    207      it would kill the other totally unrelated process.  */
    208   siginfo_t info;
    209 
    210   if (termsigp != NULL)
    211     *termsigp = 0;
    212   for (;;)
    213     {
    214       if (waitid (P_PID, child, &info, WEXITED | (slave_process ? WNOWAIT : 0))
    215           < 0)
    216         {
    217 # ifdef EINTR
    218           if (errno == EINTR)
    219             continue;
    220 # endif
    221           if (exit_on_error || !null_stderr)
    222             error (exit_on_error ? EXIT_FAILURE : 0, errno,
    223                    _("%s subprocess"), progname);
    224           return 127;
    225         }
    226 
    227       /* info.si_code is set to one of CLD_EXITED, CLD_KILLED, CLD_DUMPED,
    228          CLD_TRAPPED, CLD_STOPPED, CLD_CONTINUED.  Loop until the program
    229          terminates.  */
    230       if (info.si_code == CLD_EXITED
    231           || info.si_code == CLD_KILLED || info.si_code == CLD_DUMPED)
    232         break;
    233     }
    234 
    235   /* The child process has exited or was signalled.  */
    236 
    237   if (slave_process)
    238     {
    239       /* Unregister the child from the list of slave subprocesses, so that
    240          later, when we exit, we don't kill a totally unrelated process which
    241          may have acquired the same pid.  */
    242       unregister_slave_subprocess (child);
    243 
    244       /* Now remove the zombie from the process list.  */
    245       for (;;)
    246         {
    247           if (waitid (P_PID, child, &info, WEXITED) < 0)
    248             {
    249 # ifdef EINTR
    250               if (errno == EINTR)
    251                 continue;
    252 # endif
    253               if (exit_on_error || !null_stderr)
    254                 error (exit_on_error ? EXIT_FAILURE : 0, errno,
    255                        _("%s subprocess"), progname);
    256               return 127;
    257             }
    258           break;
    259         }
    260     }
    261 
    262   switch (info.si_code)
    263     {
    264     case CLD_KILLED:
    265     case CLD_DUMPED:
    266       if (termsigp != NULL)
    267         *termsigp = info.si_status; /* TODO: or info.si_signo? */
    268 # ifdef SIGPIPE
    269       if (info.si_status == SIGPIPE && ignore_sigpipe)
    270         return 0;
    271 # endif
    272       if (exit_on_error || (!null_stderr && termsigp == NULL))
    273         error (exit_on_error ? EXIT_FAILURE : 0, 0,
    274                _("%s subprocess got fatal signal %d"),
    275                progname, info.si_status);
    276       return 127;
    277     case CLD_EXITED:
    278       if (info.si_status == 127)
    279         {
    280           if (exit_on_error || !null_stderr)
    281             error (exit_on_error ? EXIT_FAILURE : 0, 0,
    282                    _("%s subprocess failed"), progname);
    283           return 127;
    284         }
    285       return info.si_status;
    286     default:
    287       abort ();
    288     }
    289 #else
    290   /* waitpid() is just as portable as wait() nowadays.  */
    291   int status;
    292 
    293   if (termsigp != NULL)
    294     *termsigp = 0;
    295   status = 0;
    296   for (;;)
    297     {
    298       int result = waitpid (child, &status, 0);
    299 
    300       if (result != child)
    301         {
    302 # ifdef EINTR
    303           if (errno == EINTR)
    304             continue;
    305 # endif
    306 # if 0 /* defined ECHILD */
    307           if (errno == ECHILD)
    308             {
    309               /* Child process nonexistent?! Assume it terminated
    310                  successfully.  */
    311               status = 0;
    312               break;
    313             }
    314 # endif
    315           if (exit_on_error || !null_stderr)
    316             error (exit_on_error ? EXIT_FAILURE : 0, errno,
    317                    _("%s subprocess"), progname);
    318           return 127;
    319         }
    320 
    321       /* One of WIFSIGNALED (status), WIFEXITED (status), WIFSTOPPED (status)
    322          must always be true, since we did not specify WCONTINUED in the
    323          waitpid() call.  Loop until the program terminates.  */
    324       if (!WIFSTOPPED (status))
    325         break;
    326     }
    327 
    328   /* The child process has exited or was signalled.  */
    329 
    330   if (slave_process)
    331     /* Unregister the child from the list of slave subprocesses, so that
    332        later, when we exit, we don't kill a totally unrelated process which
    333        may have acquired the same pid.  */
    334     unregister_slave_subprocess (child);
    335 
    336   if (WIFSIGNALED (status))
    337     {
    338       if (termsigp != NULL)
    339         *termsigp = WTERMSIG (status);
    340 # ifdef SIGPIPE
    341       if (WTERMSIG (status) == SIGPIPE && ignore_sigpipe)
    342         return 0;
    343 # endif
    344       if (exit_on_error || (!null_stderr && termsigp == NULL))
    345         error (exit_on_error ? EXIT_FAILURE : 0, 0,
    346                _("%s subprocess got fatal signal %d"),
    347                progname, (int) WTERMSIG (status));
    348       return 127;
    349     }
    350   if (!WIFEXITED (status))
    351     abort ();
    352   if (WEXITSTATUS (status) == 127)
    353     {
    354       if (exit_on_error || !null_stderr)
    355         error (exit_on_error ? EXIT_FAILURE : 0, 0,
    356                _("%s subprocess failed"), progname);
    357       return 127;
    358     }
    359   return WEXITSTATUS (status);
    360 #endif
    361 }
    362