1/* POSIX spawn interface. Linux version.
2 Copyright (C) 2016-2024 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
9
10 The GNU C Library 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 GNU
13 Lesser General Public License for more details.
14
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, see
17 <https://www.gnu.org/licenses/>. */
18
19#include <internal-signals.h>
20#include <ldsodefs.h>
21#include <local-setxid.h>
22#include <not-cancel.h>
23#include <paths.h>
24#include <shlib-compat.h>
25#include <spawn.h>
26#include <spawn_int.h>
27#include <sysdep.h>
28#include <sys/resource.h>
29#include <clone_internal.h>
30
31/* The Linux implementation of posix_spawn{p} uses the clone syscall directly
32 with CLONE_VM and CLONE_VFORK flags and an allocated stack. The new stack
33 and start function solves most the vfork limitation (possible parent
34 clobber due stack spilling). The remaining issue are:
35
36 1. That no signal handlers must run in child context, to avoid corrupting
37 parent's state.
38 2. The parent must ensure child's stack freeing.
39 3. Child must synchronize with parent to enforce 2. and to possible
40 return execv issues.
41
42 The first issue is solved by blocking all signals in child, even
43 the NPTL-internal ones (SIGCANCEL and SIGSETXID). The second and
44 third issue is done by a stack allocation in parent, and by using a
45 field in struct spawn_args where the child can write an error
46 code. CLONE_VFORK ensures that the parent does not run until the
47 child has either exec'ed successfully or exited. */
48
49
50/* The Unix standard contains a long explanation of the way to signal
51 an error after the fork() was successful. Since no new wait status
52 was wanted there is no way to signal an error using one of the
53 available methods. The committee chose to signal an error by a
54 normal program exit with the exit code 127. */
55#define SPAWN_ERROR 127
56
57
58struct posix_spawn_args
59{
60 internal_sigset_t oldmask;
61 const char *file;
62 int (*exec) (const char *, char *const *, char *const *);
63 const posix_spawn_file_actions_t *fa;
64 const posix_spawnattr_t *restrict attr;
65 char *const *argv;
66 ptrdiff_t argc;
67 char *const *envp;
68 int xflags;
69 bool use_clone3;
70 int err;
71 int pidfd;
72};
73
74/* Older version requires that shell script without shebang definition
75 to be called explicitly using /bin/sh (_PATH_BSHELL). */
76static void
77maybe_script_execute (struct posix_spawn_args *args)
78{
79 if (SHLIB_COMPAT (libc, GLIBC_2_2, GLIBC_2_15)
80 && (args->xflags & SPAWN_XFLAGS_TRY_SHELL) && errno == ENOEXEC)
81 {
82 char *const *argv = args->argv;
83 ptrdiff_t argc = args->argc;
84
85 /* Construct an argument list for the shell. */
86 char *new_argv[argc + 2];
87 new_argv[0] = (char *) _PATH_BSHELL;
88 new_argv[1] = (char *) args->file;
89 if (argc > 1)
90 memcpy (new_argv + 2, argv + 1, argc * sizeof (char *));
91 else
92 new_argv[2] = NULL;
93
94 /* Execute the shell. */
95 args->exec (new_argv[0], new_argv, args->envp);
96 }
97}
98
99/* Function used in the clone call to setup the signals mask, posix_spawn
100 attributes, and file actions. It run on its own stack (provided by the
101 posix_spawn call). */
102static int
103__spawni_child (void *arguments)
104{
105 struct posix_spawn_args *args = arguments;
106 const posix_spawnattr_t *restrict attr = args->attr;
107 const posix_spawn_file_actions_t *file_actions = args->fa;
108
109 /* The child must ensure that no signal handler is enabled because it
110 shares memory with parent, so all signal dispositions must be either
111 SIG_DFL or SIG_IGN. If clone3/CLONE_CLEAR_SIGHAND is used, there is
112 only the need to set the defined signals POSIX_SPAWN_SETSIGDEF to
113 SIG_DFL; otherwise, the code iterates over all signals. */
114 struct sigaction sa;
115 memset (&sa, '\0', sizeof (sa));
116
117 sigset_t hset;
118 __sigprocmask (SIG_BLOCK, 0, &hset);
119 for (int sig = 1; sig < _NSIG; ++sig)
120 {
121 if ((attr->__flags & POSIX_SPAWN_SETSIGDEF)
122 && __sigismember (set: &attr->__sd, sig))
123 {
124 sa.sa_handler = SIG_DFL;
125 }
126 else if (!args->use_clone3 && __sigismember (set: &hset, sig))
127 {
128 if (is_internal_signal (sig))
129 sa.sa_handler = SIG_IGN;
130 else
131 {
132 __libc_sigaction (sig, 0, &sa);
133 if (sa.sa_handler == SIG_IGN || sa.sa_handler == SIG_DFL)
134 continue;
135 sa.sa_handler = SIG_DFL;
136 }
137 }
138 else
139 continue;
140
141 __libc_sigaction (sig, &sa, 0);
142 }
143
144#ifdef _POSIX_PRIORITY_SCHEDULING
145 /* Set the scheduling algorithm and parameters. */
146 if ((attr->__flags & (POSIX_SPAWN_SETSCHEDPARAM | POSIX_SPAWN_SETSCHEDULER))
147 == POSIX_SPAWN_SETSCHEDPARAM)
148 {
149 if (__sched_setparam (0, &attr->__sp) == -1)
150 goto fail;
151 }
152 else if ((attr->__flags & POSIX_SPAWN_SETSCHEDULER) != 0)
153 {
154 if (__sched_setscheduler (0, attr->__policy, &attr->__sp) == -1)
155 goto fail;
156 }
157#endif
158
159 if ((attr->__flags & POSIX_SPAWN_SETSID) != 0
160 && __setsid () < 0)
161 goto fail;
162
163 /* Set the process group ID. */
164 if ((attr->__flags & POSIX_SPAWN_SETPGROUP) != 0
165 && __setpgid (0, attr->__pgrp) != 0)
166 goto fail;
167
168 /* Set the effective user and group IDs. */
169 if ((attr->__flags & POSIX_SPAWN_RESETIDS) != 0
170 && (local_seteuid (__getuid ()) != 0
171 || local_setegid (__getgid ()) != 0))
172 goto fail;
173
174 /* Execute the file actions. */
175 if (file_actions != 0)
176 {
177 int cnt;
178 struct rlimit64 fdlimit;
179 bool have_fdlimit = false;
180
181 for (cnt = 0; cnt < file_actions->__used; ++cnt)
182 {
183 struct __spawn_action *action = &file_actions->__actions[cnt];
184
185 switch (action->tag)
186 {
187 case spawn_do_close:
188 if (__close_nocancel (action->action.close_action.fd) != 0)
189 {
190 if (!have_fdlimit)
191 {
192 __getrlimit64 (RLIMIT_NOFILE, &fdlimit);
193 have_fdlimit = true;
194 }
195
196 /* Signal errors only for file descriptors out of range. */
197 if (action->action.close_action.fd < 0
198 || action->action.close_action.fd >= fdlimit.rlim_cur)
199 goto fail;
200 }
201 break;
202
203 case spawn_do_open:
204 {
205 /* POSIX states that if fildes was already an open file descriptor,
206 it shall be closed before the new file is opened. This avoid
207 potential issues when posix_spawn plus addopen action is called
208 with the process already at maximum number of file descriptor
209 opened and also for multiple actions on single-open special
210 paths (like /dev/watchdog). */
211 __close_nocancel (action->action.open_action.fd);
212
213 int ret = __open_nocancel (action->action.open_action.path,
214 action->action.
215 open_action.oflag | O_LARGEFILE,
216 action->action.open_action.mode);
217
218 if (ret == -1)
219 goto fail;
220
221 int new_fd = ret;
222
223 /* Make sure the desired file descriptor is used. */
224 if (ret != action->action.open_action.fd)
225 {
226 if (__dup2 (new_fd, action->action.open_action.fd)
227 != action->action.open_action.fd)
228 goto fail;
229
230 if (__close_nocancel (new_fd) != 0)
231 goto fail;
232 }
233 }
234 break;
235
236 case spawn_do_dup2:
237 /* Austin Group issue #411 requires adddup2 action with source
238 and destination being equal to remove close-on-exec flag. */
239 if (action->action.dup2_action.fd
240 == action->action.dup2_action.newfd)
241 {
242 int fd = action->action.dup2_action.newfd;
243 int flags = __fcntl (fd, F_GETFD, 0);
244 if (flags == -1)
245 goto fail;
246 if (__fcntl (fd, F_SETFD, flags & ~FD_CLOEXEC) == -1)
247 goto fail;
248 }
249 else if (__dup2 (action->action.dup2_action.fd,
250 action->action.dup2_action.newfd)
251 != action->action.dup2_action.newfd)
252 goto fail;
253 break;
254
255 case spawn_do_chdir:
256 if (__chdir (path: action->action.chdir_action.path) != 0)
257 goto fail;
258 break;
259
260 case spawn_do_fchdir:
261 if (__fchdir (fd: action->action.fchdir_action.fd) != 0)
262 goto fail;
263 break;
264
265 case spawn_do_closefrom:
266 {
267 int lowfd = action->action.closefrom_action.from;
268 int r = INLINE_SYSCALL_CALL (close_range, lowfd, ~0U, 0);
269 if (r != 0 && !__closefrom_fallback (lowfd: lowfd, false))
270 goto fail;
271 } break;
272
273 case spawn_do_tcsetpgrp:
274 {
275 /* Check if it is possible to avoid an extra syscall. */
276 pid_t pgrp = (attr->__flags & POSIX_SPAWN_SETPGROUP) != 0
277 && attr->__pgrp != 0
278 ? attr->__pgrp : __getpgid (0);
279 if (__tcsetpgrp (action->action.setpgrp_action.fd, pgrp) != 0)
280 goto fail;
281 }
282 }
283 }
284 }
285
286 /* Set the initial signal mask of the child if POSIX_SPAWN_SETSIGMASK
287 is set, otherwise restore the previous one. */
288 if (attr->__flags & POSIX_SPAWN_SETSIGMASK)
289 __sigprocmask (SIG_SETMASK, &attr->__ss, NULL);
290 else
291 internal_sigprocmask (SIG_SETMASK, set: &args->oldmask, NULL);
292
293 args->exec (args->file, args->argv, args->envp);
294
295 /* This is compatibility function required to enable posix_spawn run
296 script without shebang definition for older posix_spawn versions
297 (2.15). */
298 maybe_script_execute (args);
299
300fail:
301 /* errno should have an appropriate non-zero value; otherwise,
302 there's a bug in glibc or the kernel. For lack of an error code
303 (EINTERNALBUG) describing that, use ECHILD. Another option would
304 be to set args->err to some negative sentinel and have the parent
305 abort(), but that seems needlessly harsh. */
306 args->err = errno ? : ECHILD;
307 _exit (SPAWN_ERROR);
308}
309
310/* Spawn a new process executing PATH with the attributes describes in *ATTRP.
311 Before running the process perform the actions described in FILE-ACTIONS. */
312static int
313__spawnix (int *pid, const char *file,
314 const posix_spawn_file_actions_t * file_actions,
315 const posix_spawnattr_t * attrp, char *const argv[],
316 char *const envp[], int xflags,
317 int (*exec) (const char *, char *const *, char *const *))
318{
319 pid_t new_pid;
320 struct posix_spawn_args args;
321 int ec;
322
323 bool use_pidfd = xflags & SPAWN_XFLAGS_RET_PIDFD;
324
325 /* For CLONE_PIDFD, older kernels might not fail with unsupported flags or
326 some versions might not support waitid (P_PIDFD). So to avoid the need
327 to handle the error on the helper process, check for full pidfd
328 support.
329 ENOSYS is returned because without proper waitid support, pidfd_spawn
330 can not be used proporly independently of its arguments. */
331 if (use_pidfd && !__clone_pidfd_supported ())
332 return ENOSYS;
333
334 /* To avoid imposing hard limits on posix_spawn{p} the total number of
335 arguments is first calculated to allocate a mmap to hold all possible
336 values. */
337 ptrdiff_t argc = 0;
338 /* Linux allows at most max (0x7FFFFFFF, 1/4 stack size) arguments
339 to be used in a execve call. We limit to INT_MAX minus one due the
340 compatibility code that may execute a shell script (maybe_script_execute)
341 where it will construct another argument list with an additional
342 argument. */
343 ptrdiff_t limit = INT_MAX - 1;
344 while (argv[argc++] != NULL)
345 if (argc == limit)
346 {
347 errno = E2BIG;
348 return errno;
349 }
350
351 int prot = (PROT_READ | PROT_WRITE
352 | ((GL (dl_stack_flags) & PF_X) ? PROT_EXEC : 0));
353
354 /* Add a slack area for child's stack. */
355 size_t argv_size = (argc * sizeof (void *)) + 512;
356 /* We need at least a few pages in case the compiler's stack checking is
357 enabled. In some configs, it is known to use at least 24KiB. We use
358 32KiB to be "safe" from anything the compiler might do. Besides, the
359 extra pages won't actually be allocated unless they get used.
360 It also acts the slack for spawn_closefrom (including MIPS64 getdents64
361 where it might use about 1k extra stack space). */
362 argv_size += (32 * 1024);
363 size_t stack_size = ALIGN_UP (argv_size, GLRO(dl_pagesize));
364 void *stack = __mmap (NULL, stack_size, prot,
365 MAP_PRIVATE | MAP_ANONYMOUS | MAP_STACK, -1, 0);
366 if (__glibc_unlikely (stack == MAP_FAILED))
367 return errno;
368
369 /* Disable asynchronous cancellation. */
370 int state;
371 __pthread_setcancelstate (PTHREAD_CANCEL_DISABLE, &state);
372
373 /* Child must set args.err to something non-negative - we rely on
374 the parent and child sharing VM. */
375 args.err = 0;
376 args.file = file;
377 args.exec = exec;
378 args.fa = file_actions;
379 args.attr = attrp ? attrp : &(const posix_spawnattr_t) { 0 };
380 args.argv = argv;
381 args.argc = argc;
382 args.envp = envp;
383 args.pidfd = 0;
384 args.xflags = xflags;
385
386 internal_signal_block_all (&args.oldmask);
387
388 /* The clone flags used will create a new child that will run in the same
389 memory space (CLONE_VM) and the execution of calling thread will be
390 suspend until the child calls execve or _exit.
391
392 Also since the calling thread execution will be suspend, there is not
393 need for CLONE_SETTLS. Although parent and child share the same TLS
394 namespace, there will be no concurrent access for TLS variables (errno
395 for instance). */
396 bool set_cgroup = attrp ? (attrp->__flags & POSIX_SPAWN_SETCGROUP) : false;
397 struct clone_args clone_args =
398 {
399 /* Unsupported flags like CLONE_CLEAR_SIGHAND will be cleared up by
400 __clone_internal_fallback. */
401 .flags = (set_cgroup ? CLONE_INTO_CGROUP : 0)
402 | (use_pidfd ? CLONE_PIDFD : 0)
403 | CLONE_CLEAR_SIGHAND
404 | CLONE_VM
405 | CLONE_VFORK,
406 .exit_signal = SIGCHLD,
407 .stack = (uintptr_t) stack,
408 .stack_size = stack_size,
409 .cgroup = (set_cgroup ? attrp->__cgroup : 0),
410 .pidfd = use_pidfd ? (uintptr_t) &args.pidfd : 0,
411 /* This is require for clone fallback, where pidfd is returned
412 on parent_tid. */
413 .parent_tid = use_pidfd ? (uintptr_t) &args.pidfd : 0,
414 };
415#ifdef HAVE_CLONE3_WRAPPER
416 args.use_clone3 = true;
417 new_pid = __clone3 (&clone_args, sizeof (clone_args), __spawni_child,
418 &args);
419 /* clone3 was added in 5.3 and CLONE_CLEAR_SIGHAND in 5.5. */
420 if (new_pid == -1 && (errno == ENOSYS || errno == EINVAL))
421#endif
422 {
423 args.use_clone3 = false;
424 if (!set_cgroup)
425 new_pid = __clone_internal_fallback (&clone_args, __spawni_child,
426 &args);
427 else
428 {
429 /* No fallback for POSIX_SPAWN_SETCGROUP if clone3 is not
430 supported. */
431 new_pid = -1;
432#ifdef HAVE_CLONE3_WRAPPER
433 if (errno == ENOSYS)
434#endif
435 errno = ENOTSUP;
436 }
437 }
438
439 /* It needs to collect the case where the auxiliary process was created
440 but failed to execute the file (due either any preparation step or
441 for execve itself). */
442 if (new_pid > 0)
443 {
444 /* Also, it handles the unlikely case where the auxiliary process was
445 terminated before calling execve as if it was successfully. The
446 args.err is set to 0 as default and changed to a positive value
447 only in case of failure, so in case of premature termination
448 due a signal args.err will remain zeroed and it will be up to
449 caller to actually collect it. */
450 ec = args.err;
451 if (ec > 0)
452 /* There still an unlikely case where the child is cancelled after
453 setting args.err, due to a positive error value. Also there is
454 possible pid reuse race (where the kernel allocated the same pid
455 to an unrelated process). Unfortunately due synchronization
456 issues where the kernel might not have the process collected
457 the waitpid below can not use WNOHANG. */
458 __waitpid (new_pid, NULL, 0);
459 }
460 else
461 ec = errno;
462
463 __munmap (stack, stack_size);
464
465 if ((ec == 0) && (pid != NULL))
466 *pid = use_pidfd ? args.pidfd : new_pid;
467
468 internal_signal_restore_set (&args.oldmask);
469
470 __pthread_setcancelstate (state, NULL);
471
472 return ec;
473}
474
475/* Spawn a new process executing PATH with the attributes describes in *ATTRP.
476 Before running the process perform the actions described in FILE-ACTIONS. */
477int
478__spawni (pid_t * pid, const char *file,
479 const posix_spawn_file_actions_t * acts,
480 const posix_spawnattr_t * attrp, char *const argv[],
481 char *const envp[], int xflags)
482{
483 /* It uses __execvpex to avoid run ENOEXEC in non compatibility mode (it
484 will be handled by maybe_script_execute). */
485 return __spawnix (pid, file, file_actions: acts, attrp, argv, envp, xflags,
486 exec: xflags & SPAWN_XFLAGS_USE_PATH ? __execvpex :__execve);
487}
488

source code of glibc/sysdeps/unix/sysv/linux/spawni.c