| 1 |
/* Waiting for a subprocess to finish. |
| 2 |
Copyright (C) 2001-2003, 2005-2024 Free Software Foundation, Inc. |
| 3 |
Written by Bruno Haible <[email protected]>, 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 <https://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 _WIN32 && ! defined __CYGWIN__ |
| 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 |
#ifdef __KLIBC__ |
| 54 |
# include <dlfcn.h> |
| 55 |
|
| 56 |
# undef waitpid |
| 57 |
|
| 58 |
/* Replacement of waitpid() to support spawn2() of LIBCx which is the kLIBC |
| 59 |
extension library. See for details: |
| 60 |
<https://github.com/bitwiseworks/libcx/blob/master/src/spawn/libcx/spawn2.h#L194>. |
| 61 |
*/ |
| 62 |
static pid_t |
| 63 |
klibc_waitpid (pid_t pid, int *statusp, int options) |
| 64 |
{ |
| 65 |
static pid_t (*waitpid_pfn) (pid_t, int *, int) = NULL; |
| 66 |
|
| 67 |
if (waitpid_pfn == NULL) |
| 68 |
{ |
| 69 |
void *libcx_handle; |
| 70 |
|
| 71 |
/* Try to use waitpid() of LIBCx first if available because it can |
| 72 |
process the return value of spawn-family of kLIBC as well as spawn2() |
| 73 |
of LIBCx. */ |
| 74 |
libcx_handle = dlopen ("libcx0", RTLD_LAZY); |
| 75 |
if (libcx_handle != NULL) |
| 76 |
waitpid_pfn = dlsym (libcx_handle, "_waitpid"); |
| 77 |
/* If not available, falls back to waitpid() of kLIBC. */ |
| 78 |
if (waitpid_pfn == NULL) |
| 79 |
waitpid_pfn = waitpid; |
| 80 |
} |
| 81 |
|
| 82 |
return waitpid_pfn (pid, statusp, options); |
| 83 |
} |
| 84 |
|
| 85 |
# define waitpid klibc_waitpid |
| 86 |
#endif |
| 87 |
|
| 88 |
|
| 89 |
/* Type of an entry in the slaves array. |
| 90 |
The 'used' bit determines whether this entry is currently in use. |
| 91 |
(If pid_t was an atomic type like sig_atomic_t, we could just set the |
| 92 |
'child' field to 0 when unregistering a slave process, and wouldn't need |
| 93 |
the 'used' field.) |
| 94 |
The 'used' and 'child' fields are accessed from within the cleanup_slaves() |
| 95 |
action, therefore we mark them as 'volatile'. */ |
| 96 |
typedef struct |
| 97 |
{ |
| 98 |
volatile sig_atomic_t used; |
| 99 |
volatile pid_t child; |
| 100 |
} |
| 101 |
slaves_entry_t; |
| 102 |
|
| 103 |
/* The registered slave subprocesses. */ |
| 104 |
static slaves_entry_t static_slaves[32]; |
| 105 |
static slaves_entry_t * volatile slaves = static_slaves; |
| 106 |
static sig_atomic_t volatile slaves_count = 0; |
| 107 |
static size_t slaves_allocated = SIZEOF (static_slaves); |
| 108 |
|
| 109 |
/* The termination signal for slave subprocesses. |
| 110 |
2003-10-07: Terminator becomes Governator. */ |
| 111 |
#ifdef SIGHUP |
| 112 |
# define TERMINATOR SIGHUP |
| 113 |
#else |
| 114 |
# define TERMINATOR SIGTERM |
| 115 |
#endif |
| 116 |
|
| 117 |
/* The cleanup action. It gets called asynchronously. */ |
| 118 |
static _GL_ASYNC_SAFE void |
| 119 |
cleanup_slaves (void) |
| 120 |
{ |
| 121 |
for (;;) |
| 122 |
{ |
| 123 |
/* Get the last registered slave. */ |
| 124 |
size_t n = slaves_count; |
| 125 |
if (n == 0) |
| 126 |
break; |
| 127 |
n--; |
| 128 |
slaves_count = n; |
| 129 |
/* Skip unused entries in the slaves array. */ |
| 130 |
if (slaves[n].used) |
| 131 |
{ |
| 132 |
pid_t slave = slaves[n].child; |
| 133 |
|
| 134 |
/* Kill the slave. */ |
| 135 |
kill (slave, TERMINATOR); |
| 136 |
} |
| 137 |
} |
| 138 |
} |
| 139 |
|
| 140 |
/* The cleanup action, taking a signal argument. |
| 141 |
It gets called asynchronously. */ |
| 142 |
static _GL_ASYNC_SAFE void |
| 143 |
cleanup_slaves_action (_GL_UNUSED int sig) |
| 144 |
{ |
| 145 |
cleanup_slaves (); |
| 146 |
} |
| 147 |
|
| 148 |
/* Register a subprocess as being a slave process. This means that the |
| 149 |
subprocess will be terminated when its creator receives a catchable fatal |
| 150 |
signal or exits normally. Registration ends when wait_subprocess() |
| 151 |
notices that the subprocess has exited. */ |
| 152 |
void |
| 153 |
register_slave_subprocess (pid_t child) |
| 154 |
{ |
| 155 |
static bool cleanup_slaves_registered = false; |
| 156 |
if (!cleanup_slaves_registered) |
| 157 |
{ |
| 158 |
atexit (cleanup_slaves); |
| 159 |
if (at_fatal_signal (cleanup_slaves_action) < 0) |
| 160 |
xalloc_die (); |
| 161 |
cleanup_slaves_registered = true; |
| 162 |
} |
| 163 |
|
| 164 |
/* Try to store the new slave in an unused entry of the slaves array. */ |
| 165 |
{ |
| 166 |
slaves_entry_t *s = slaves; |
| 167 |
slaves_entry_t *s_end = s + slaves_count; |
| 168 |
|
| 169 |
for (; s < s_end; s++) |
| 170 |
if (!s->used) |
| 171 |
{ |
| 172 |
/* The two uses of 'volatile' in the slaves_entry_t type above |
| 173 |
(and ISO C 99 section 5.1.2.3.(5)) ensure that we mark the |
| 174 |
entry as used only after the child pid has been written to the |
| 175 |
memory location s->child. */ |
| 176 |
s->child = child; |
| 177 |
s->used = 1; |
| 178 |
return; |
| 179 |
} |
| 180 |
} |
| 181 |
|
| 182 |
if (slaves_count == slaves_allocated) |
| 183 |
{ |
| 184 |
/* Extend the slaves array. Note that we cannot use xrealloc(), |
| 185 |
because then the cleanup_slaves() function could access an already |
| 186 |
deallocated array. */ |
| 187 |
slaves_entry_t *old_slaves = slaves; |
| 188 |
size_t new_slaves_allocated = 2 * slaves_allocated; |
| 189 |
slaves_entry_t *new_slaves = |
| 190 |
(slaves_entry_t *) |
| 191 |
malloc (new_slaves_allocated * sizeof (slaves_entry_t)); |
| 192 |
if (new_slaves == NULL) |
| 193 |
{ |
| 194 |
/* xalloc_die() will call exit() which will invoke cleanup_slaves(). |
| 195 |
Additionally we need to kill child, because it's not yet among |
| 196 |
the slaves list. */ |
| 197 |
kill (child, TERMINATOR); |
| 198 |
xalloc_die (); |
| 199 |
} |
| 200 |
memcpy (new_slaves, old_slaves, |
| 201 |
slaves_allocated * sizeof (slaves_entry_t)); |
| 202 |
slaves = new_slaves; |
| 203 |
slaves_allocated = new_slaves_allocated; |
| 204 |
/* Now we can free the old slaves array. */ |
| 205 |
if (old_slaves != static_slaves) |
| 206 |
free (old_slaves); |
| 207 |
} |
| 208 |
/* The three uses of 'volatile' in the types above (and ISO C 99 section |
| 209 |
5.1.2.3.(5)) ensure that we increment the slaves_count only after the |
| 210 |
new slave and its 'used' bit have been written to the memory locations |
| 211 |
that make up slaves[slaves_count]. */ |
| 212 |
slaves[slaves_count].child = child; |
| 213 |
slaves[slaves_count].used = 1; |
| 214 |
slaves_count++; |
| 215 |
} |
| 216 |
|
| 217 |
/* Unregister a child from the list of slave subprocesses. */ |
| 218 |
static void |
| 219 |
unregister_slave_subprocess (pid_t child) |
| 220 |
{ |
| 221 |
/* The easiest way to remove an entry from a list that can be used by |
| 222 |
an asynchronous signal handler is just to mark it as unused. For this, |
| 223 |
we rely on sig_atomic_t. */ |
| 224 |
slaves_entry_t *s = slaves; |
| 225 |
slaves_entry_t *s_end = s + slaves_count; |
| 226 |
|
| 227 |
for (; s < s_end; s++) |
| 228 |
if (s->used && s->child == child) |
| 229 |
s->used = 0; |
| 230 |
} |
| 231 |
|
| 232 |
|
| 233 |
/* Wait for a subprocess to finish. Return its exit code. |
| 234 |
If it didn't terminate correctly, exit if exit_on_error is true, otherwise |
| 235 |
return 127. */ |
| 236 |
int |
| 237 |
wait_subprocess (pid_t child, const char *progname, |
| 238 |
bool ignore_sigpipe, bool null_stderr, |
| 239 |
bool slave_process, bool exit_on_error, |
| 240 |
int *termsigp) |
| 241 |
{ |
| 242 |
#if HAVE_WAITID && defined WNOWAIT && 0 |
| 243 |
/* Commented out because waitid() without WEXITED and with WNOWAIT doesn't |
| 244 |
work: On Solaris 7 and OSF/1 4.0, it returns -1 and sets errno = ECHILD, |
| 245 |
and on HP-UX 10.20 it just hangs. */ |
| 246 |
/* Use of waitid() with WNOWAIT avoids a race condition: If slave_process is |
| 247 |
true, and this process sleeps a very long time between the return from |
| 248 |
waitpid() and the execution of unregister_slave_subprocess(), and |
| 249 |
meanwhile another process acquires the same PID as child, and then - still |
| 250 |
before unregister_slave_subprocess() - this process gets a fatal signal, |
| 251 |
it would kill the other totally unrelated process. */ |
| 252 |
siginfo_t info; |
| 253 |
|
| 254 |
if (termsigp != NULL) |
| 255 |
*termsigp = 0; |
| 256 |
for (;;) |
| 257 |
{ |
| 258 |
if (waitid (P_PID, child, &info, WEXITED | (slave_process ? WNOWAIT : 0)) |
| 259 |
< 0) |
| 260 |
{ |
| 261 |
# ifdef EINTR |
| 262 |
if (errno == EINTR) |
| 263 |
continue; |
| 264 |
# endif |
| 265 |
if (exit_on_error || !null_stderr) |
| 266 |
error (exit_on_error ? EXIT_FAILURE : 0, errno, |
| 267 |
_("%s subprocess"), progname); |
| 268 |
return 127; |
| 269 |
} |
| 270 |
|
| 271 |
/* info.si_code is set to one of CLD_EXITED, CLD_KILLED, CLD_DUMPED, |
| 272 |
CLD_TRAPPED, CLD_STOPPED, CLD_CONTINUED. Loop until the program |
| 273 |
terminates. */ |
| 274 |
if (info.si_code == CLD_EXITED |
| 275 |
|| info.si_code == CLD_KILLED || info.si_code == CLD_DUMPED) |
| 276 |
break; |
| 277 |
} |
| 278 |
|
| 279 |
/* The child process has exited or was signalled. */ |
| 280 |
|
| 281 |
if (slave_process) |
| 282 |
{ |
| 283 |
/* Unregister the child from the list of slave subprocesses, so that |
| 284 |
later, when we exit, we don't kill a totally unrelated process which |
| 285 |
may have acquired the same pid. */ |
| 286 |
unregister_slave_subprocess (child); |
| 287 |
|
| 288 |
/* Now remove the zombie from the process list. */ |
| 289 |
for (;;) |
| 290 |
{ |
| 291 |
if (waitid (P_PID, child, &info, WEXITED) < 0) |
| 292 |
{ |
| 293 |
# ifdef EINTR |
| 294 |
if (errno == EINTR) |
| 295 |
continue; |
| 296 |
# endif |
| 297 |
if (exit_on_error || !null_stderr) |
| 298 |
error (exit_on_error ? EXIT_FAILURE : 0, errno, |
| 299 |
_("%s subprocess"), progname); |
| 300 |
return 127; |
| 301 |
} |
| 302 |
break; |
| 303 |
} |
| 304 |
} |
| 305 |
|
| 306 |
switch (info.si_code) |
| 307 |
{ |
| 308 |
case CLD_KILLED: |
| 309 |
case CLD_DUMPED: |
| 310 |
if (termsigp != NULL) |
| 311 |
*termsigp = info.si_status; /* TODO: or info.si_signo? */ |
| 312 |
# ifdef SIGPIPE |
| 313 |
if (info.si_status == SIGPIPE && ignore_sigpipe) |
| 314 |
return 0; |
| 315 |
# endif |
| 316 |
if (exit_on_error || (!null_stderr && termsigp == NULL)) |
| 317 |
error (exit_on_error ? EXIT_FAILURE : 0, 0, |
| 318 |
_("%s subprocess got fatal signal %d"), |
| 319 |
progname, info.si_status); |
| 320 |
return 127; |
| 321 |
case CLD_EXITED: |
| 322 |
if (info.si_status == 127) |
| 323 |
{ |
| 324 |
if (exit_on_error || !null_stderr) |
| 325 |
error (exit_on_error ? EXIT_FAILURE : 0, 0, |
| 326 |
_("%s subprocess failed"), progname); |
| 327 |
return 127; |
| 328 |
} |
| 329 |
return info.si_status; |
| 330 |
default: |
| 331 |
abort (); |
| 332 |
} |
| 333 |
#else |
| 334 |
/* waitpid() is just as portable as wait() nowadays. */ |
| 335 |
int status; |
| 336 |
|
| 337 |
if (termsigp != NULL) |
| 338 |
*termsigp = 0; |
| 339 |
status = 0; |
| 340 |
for (;;) |
| 341 |
{ |
| 342 |
int result = waitpid (child, &status, 0); |
| 343 |
|
| 344 |
if (result != child) |
| 345 |
{ |
| 346 |
# ifdef EINTR |
| 347 |
if (errno == EINTR) |
| 348 |
continue; |
| 349 |
# endif |
| 350 |
# if 0 /* defined ECHILD */ |
| 351 |
if (errno == ECHILD) |
| 352 |
{ |
| 353 |
/* Child process nonexistent?! Assume it terminated |
| 354 |
successfully. */ |
| 355 |
status = 0; |
| 356 |
break; |
| 357 |
} |
| 358 |
# endif |
| 359 |
if (exit_on_error || !null_stderr) |
| 360 |
error (exit_on_error ? EXIT_FAILURE : 0, errno, |
| 361 |
_("%s subprocess"), progname); |
| 362 |
return 127; |
| 363 |
} |
| 364 |
|
| 365 |
/* One of WIFSIGNALED (status), WIFEXITED (status), WIFSTOPPED (status) |
| 366 |
must always be true, since we did not specify WCONTINUED in the |
| 367 |
waitpid() call. Loop until the program terminates. */ |
| 368 |
if (!WIFSTOPPED (status)) |
| 369 |
break; |
| 370 |
} |
| 371 |
|
| 372 |
/* The child process has exited or was signalled. */ |
| 373 |
|
| 374 |
if (slave_process) |
| 375 |
/* Unregister the child from the list of slave subprocesses, so that |
| 376 |
later, when we exit, we don't kill a totally unrelated process which |
| 377 |
may have acquired the same pid. */ |
| 378 |
unregister_slave_subprocess (child); |
| 379 |
|
| 380 |
if (WIFSIGNALED (status)) |
| 381 |
{ |
| 382 |
if (termsigp != NULL) |
| 383 |
*termsigp = WTERMSIG (status); |
| 384 |
# ifdef SIGPIPE |
| 385 |
if (WTERMSIG (status) == SIGPIPE && ignore_sigpipe) |
| 386 |
return 0; |
| 387 |
# endif |
| 388 |
if (exit_on_error || (!null_stderr && termsigp == NULL)) |
| 389 |
error (exit_on_error ? EXIT_FAILURE : 0, 0, |
| 390 |
_("%s subprocess got fatal signal %d"), |
| 391 |
progname, (int) WTERMSIG (status)); |
| 392 |
return 127; |
| 393 |
} |
| 394 |
if (!WIFEXITED (status)) |
| 395 |
abort (); |
| 396 |
if (WEXITSTATUS (status) == 127) |
| 397 |
{ |
| 398 |
if (exit_on_error || !null_stderr) |
| 399 |
error (exit_on_error ? EXIT_FAILURE : 0, 0, |
| 400 |
_("%s subprocess failed"), progname); |
| 401 |
return 127; |
| 402 |
} |
| 403 |
return WEXITSTATUS (status); |
| 404 |
#endif |
| 405 |
} |