path: root/lib/wait-process.c

/* Waiting for a subprocess to finish.
   Copyright (C) 2001-2003, 2005-2020 Free Software Foundation, Inc.
   Written by Bruno Haible <haible@clisp.cons.org>, 2001.

   This program is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 3 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program.  If not, see <https://www.gnu.org/licenses/>.  */


#include <config.h>

/* Specification.  */
#include "wait-process.h"

#include <errno.h>
#include <stdlib.h>
#include <string.h>
#include <signal.h>

#include <sys/types.h>
#include <sys/wait.h>

#include "error.h"
#include "fatal-signal.h"
#include "xalloc.h"
#include "gettext.h"

#define _(str) gettext (str)

#define SIZEOF(a) (sizeof(a) / sizeof(a[0]))


#if defined _WIN32 && ! defined __CYGWIN__

# define WIN32_LEAN_AND_MEAN
# include <windows.h>

/* The return value of _spawnvp() is really a process handle as returned
   by CreateProcess().  Therefore we can kill it using TerminateProcess.  */
# define kill(pid,sig) TerminateProcess ((HANDLE) (pid), sig)

#endif


/* Type of an entry in the slaves array.
   The 'used' bit determines whether this entry is currently in use.
   (If pid_t was an atomic type like sig_atomic_t, we could just set the
   'child' field to 0 when unregistering a slave process, and wouldn't need
   the 'used' field.)
   The 'used' and 'child' fields are accessed from within the cleanup_slaves()
   action, therefore we mark them as 'volatile'.  */
typedef struct
{
  volatile sig_atomic_t used;
  volatile pid_t child;
}
slaves_entry_t;

/* The registered slave subprocesses.  */
static slaves_entry_t static_slaves[32];
static slaves_entry_t * volatile slaves = static_slaves;
static sig_atomic_t volatile slaves_count = 0;
static size_t slaves_allocated = SIZEOF (static_slaves);

/* The termination signal for slave subprocesses.
   2003-10-07:  Terminator becomes Governator.  */
#ifdef SIGHUP
# define TERMINATOR SIGHUP
#else
# define TERMINATOR SIGTERM
#endif

/* The cleanup action.  It gets called asynchronously.  */
static _GL_ASYNC_SAFE void
cleanup_slaves (void)
{
  for (;;)
    {
      /* Get the last registered slave.  */
      size_t n = slaves_count;
      if (n == 0)
        break;
      n--;
      slaves_count = n;
      /* Skip unused entries in the slaves array.  */
      if (slaves[n].used)
        {
          pid_t slave = slaves[n].child;

          /* Kill the slave.  */
          kill (slave, TERMINATOR);
        }
    }
}

/* The cleanup action, taking a signal argument.
   It gets called asynchronously.  */
static _GL_ASYNC_SAFE void
cleanup_slaves_action (int sig _GL_UNUSED)
{
  cleanup_slaves ();
}

/* Register a subprocess as being a slave process.  This means that the
   subprocess will be terminated when its creator receives a catchable fatal
   signal or exits normally.  Registration ends when wait_subprocess()
   notices that the subprocess has exited.  */
void
register_slave_subprocess (pid_t child)
{
  static bool cleanup_slaves_registered = false;
  if (!cleanup_slaves_registered)
    {
      atexit (cleanup_slaves);
      at_fatal_signal (cleanup_slaves_action);
      cleanup_slaves_registered = true;
    }

  /* Try to store the new slave in an unused entry of the slaves array.  */
  {
    slaves_entry_t *s = slaves;
    slaves_entry_t *s_end = s + slaves_count;

    for (; s < s_end; s++)
      if (!s->used)
        {
          /* The two uses of 'volatile' in the slaves_entry_t type above
             (and ISO C 99 section 5.1.2.3.(5)) ensure that we mark the
             entry as used only after the child pid has been written to the
             memory location s->child.  */
          s->child = child;
          s->used = 1;
          return;
        }
  }

  if (slaves_count == slaves_allocated)
    {
      /* Extend the slaves array.  Note that we cannot use xrealloc(),
         because then the cleanup_slaves() function could access an already
         deallocated array.  */
      slaves_entry_t *old_slaves = slaves;
      size_t new_slaves_allocated = 2 * slaves_allocated;
      slaves_entry_t *new_slaves =
        (slaves_entry_t *)
        malloc (new_slaves_allocated * sizeof (slaves_entry_t));
      if (new_slaves == NULL)
        {
          /* xalloc_die() will call exit() which will invoke cleanup_slaves().
             Additionally we need to kill child, because it's not yet among
             the slaves list.  */
          kill (child, TERMINATOR);
          xalloc_die ();
        }
      memcpy (new_slaves, old_slaves,
              slaves_allocated * sizeof (slaves_entry_t));
      slaves = new_slaves;
      slaves_allocated = new_slaves_allocated;
      /* Now we can free the old slaves array.  */
      if (old_slaves != static_slaves)
        free (old_slaves);
    }
  /* The three uses of 'volatile' in the types above (and ISO C 99 section
     5.1.2.3.(5)) ensure that we increment the slaves_count only after the
     new slave and its 'used' bit have been written to the memory locations
     that make up slaves[slaves_count].  */
  slaves[slaves_count].child = child;
  slaves[slaves_count].used = 1;
  slaves_count++;
}

/* Unregister a child from the list of slave subprocesses.  */
static void
unregister_slave_subprocess (pid_t child)
{
  /* The easiest way to remove an entry from a list that can be used by
     an asynchronous signal handler is just to mark it as unused.  For this,
     we rely on sig_atomic_t.  */
  slaves_entry_t *s = slaves;
  slaves_entry_t *s_end = s + slaves_count;

  for (; s < s_end; s++)
    if (s->used && s->child == child)
      s->used = 0;
}


/* Wait for a subprocess to finish.  Return its exit code.
   If it didn't terminate correctly, exit if exit_on_error is true, otherwise
   return 127.  */
int
wait_subprocess (pid_t child, const char *progname,
                 bool ignore_sigpipe, bool null_stderr,
                 bool slave_process, bool exit_on_error,
                 int *termsigp)
{
#if HAVE_WAITID && defined WNOWAIT && 0
  /* Commented out because waitid() without WEXITED and with WNOWAIT doesn't
     work: On Solaris 7 and OSF/1 4.0, it returns -1 and sets errno = ECHILD,
     and on HP-UX 10.20 it just hangs.  */
  /* Use of waitid() with WNOWAIT avoids a race condition: If slave_process is
     true, and this process sleeps a very long time between the return from
     waitpid() and the execution of unregister_slave_subprocess(), and
     meanwhile another process acquires the same PID as child, and then - still
     before unregister_slave_subprocess() - this process gets a fatal signal,
     it would kill the other totally unrelated process.  */
  siginfo_t info;

  if (termsigp != NULL)
    *termsigp = 0;
  for (;;)
    {
      if (waitid (P_PID, child, &info, WEXITED | (slave_process ? WNOWAIT : 0))
          < 0)
        {
# ifdef EINTR
          if (errno == EINTR)
            continue;
# endif
          if (exit_on_error || !null_stderr)
            error (exit_on_error ? EXIT_FAILURE : 0, errno,
                   _("%s subprocess"), progname);
          return 127;
        }

      /* info.si_code is set to one of CLD_EXITED, CLD_KILLED, CLD_DUMPED,
         CLD_TRAPPED, CLD_STOPPED, CLD_CONTINUED.  Loop until the program
         terminates.  */
      if (info.si_code == CLD_EXITED
          || info.si_code == CLD_KILLED || info.si_code == CLD_DUMPED)
        break;
    }

  /* The child process has exited or was signalled.  */

  if (slave_process)
    {
      /* Unregister the child from the list of slave subprocesses, so that
         later, when we exit, we don't kill a totally unrelated process which
         may have acquired the same pid.  */
      unregister_slave_subprocess (child);

      /* Now remove the zombie from the process list.  */
      for (;;)
        {
          if (waitid (P_PID, child, &info, WEXITED) < 0)
            {
# ifdef EINTR
              if (errno == EINTR)
                continue;
# endif
              if (exit_on_error || !null_stderr)
                error (exit_on_error ? EXIT_FAILURE : 0, errno,
                       _("%s subprocess"), progname);
              return 127;
            }
          break;
        }
    }

  switch (info.si_code)
    {
    case CLD_KILLED:
    case CLD_DUMPED:
      if (termsigp != NULL)
        *termsigp = info.si_status; /* TODO: or info.si_signo? */
# ifdef SIGPIPE
      if (info.si_status == SIGPIPE && ignore_sigpipe)
        return 0;
# endif
      if (exit_on_error || (!null_stderr && termsigp == NULL))
        error (exit_on_error ? EXIT_FAILURE : 0, 0,
               _("%s subprocess got fatal signal %d"),
               progname, info.si_status);
      return 127;
    case CLD_EXITED:
      if (info.si_status == 127)
        {
          if (exit_on_error || !null_stderr)
            error (exit_on_error ? EXIT_FAILURE : 0, 0,
                   _("%s subprocess failed"), progname);
          return 127;
        }
      return info.si_status;
    default:
      abort ();
    }
#else
  /* waitpid() is just as portable as wait() nowadays.  */
  int status;

  if (termsigp != NULL)
    *termsigp = 0;
  status = 0;
  for (;;)
    {
      int result = waitpid (child, &status, 0);

      if (result != child)
        {
# ifdef EINTR
          if (errno == EINTR)
            continue;
# endif
# if 0 /* defined ECHILD */
          if (errno == ECHILD)
            {
              /* Child process nonexistent?! Assume it terminated
                 successfully.  */
              status = 0;
              break;
            }
# endif
          if (exit_on_error || !null_stderr)
            error (exit_on_error ? EXIT_FAILURE : 0, errno,
                   _("%s subprocess"), progname);
          return 127;
        }

      /* One of WIFSIGNALED (status), WIFEXITED (status), WIFSTOPPED (status)
         must always be true, since we did not specify WCONTINUED in the
         waitpid() call.  Loop until the program terminates.  */
      if (!WIFSTOPPED (status))
        break;
    }

  /* The child process has exited or was signalled.  */

  if (slave_process)
    /* Unregister the child from the list of slave subprocesses, so that
       later, when we exit, we don't kill a totally unrelated process which
       may have acquired the same pid.  */
    unregister_slave_subprocess (child);

  if (WIFSIGNALED (status))
    {
      if (termsigp != NULL)
        *termsigp = WTERMSIG (status);
# ifdef SIGPIPE
      if (WTERMSIG (status) == SIGPIPE && ignore_sigpipe)
        return 0;
# endif
      if (exit_on_error || (!null_stderr && termsigp == NULL))
        error (exit_on_error ? EXIT_FAILURE : 0, 0,
               _("%s subprocess got fatal signal %d"),
               progname, (int) WTERMSIG (status));
      return 127;
    }
  if (!WIFEXITED (status))
    abort ();
  if (WEXITSTATUS (status) == 127)
    {
      if (exit_on_error || !null_stderr)
        error (exit_on_error ? EXIT_FAILURE : 0, 0,
               _("%s subprocess failed"), progname);
      return 127;
    }
  return WEXITSTATUS (status);
#endif
}