xref: /freebsd/lib/libc/gen/posix_spawn.c (revision 9119bafbaf52e5d86e5879e54b38c5aba3efa2e9)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2008 Ed Schouten <ed@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "namespace.h"
33 #include <sys/param.h>
34 #include <sys/queue.h>
35 #include <sys/wait.h>
36 
37 #include <errno.h>
38 #include <fcntl.h>
39 #include <sched.h>
40 #include <spawn.h>
41 #include <signal.h>
42 #include <stdlib.h>
43 #include <string.h>
44 #include <unistd.h>
45 #include "un-namespace.h"
46 #include "libc_private.h"
47 
48 extern char **environ;
49 
50 struct __posix_spawnattr {
51 	short			sa_flags;
52 	pid_t			sa_pgroup;
53 	struct sched_param	sa_schedparam;
54 	int			sa_schedpolicy;
55 	sigset_t		sa_sigdefault;
56 	sigset_t		sa_sigmask;
57 };
58 
59 struct __posix_spawn_file_actions {
60 	STAILQ_HEAD(, __posix_spawn_file_actions_entry) fa_list;
61 };
62 
63 typedef struct __posix_spawn_file_actions_entry {
64 	STAILQ_ENTRY(__posix_spawn_file_actions_entry) fae_list;
65 	enum { FAE_OPEN, FAE_DUP2, FAE_CLOSE } fae_action;
66 
67 	int fae_fildes;
68 	union {
69 		struct {
70 			char *path;
71 #define fae_path	fae_data.open.path
72 			int oflag;
73 #define fae_oflag	fae_data.open.oflag
74 			mode_t mode;
75 #define fae_mode	fae_data.open.mode
76 		} open;
77 		struct {
78 			int newfildes;
79 #define fae_newfildes	fae_data.dup2.newfildes
80 		} dup2;
81 	} fae_data;
82 } posix_spawn_file_actions_entry_t;
83 
84 /*
85  * Spawn routines
86  */
87 
88 static int
89 process_spawnattr(const posix_spawnattr_t sa)
90 {
91 	struct sigaction sigact = { .sa_flags = 0, .sa_handler = SIG_DFL };
92 	int i;
93 
94 	/*
95 	 * POSIX doesn't really describe in which order everything
96 	 * should be set. We'll just set them in the order in which they
97 	 * are mentioned.
98 	 */
99 
100 	/* Set process group */
101 	if (sa->sa_flags & POSIX_SPAWN_SETPGROUP) {
102 		if (setpgid(0, sa->sa_pgroup) != 0)
103 			return (errno);
104 	}
105 
106 	/* Set scheduler policy */
107 	if (sa->sa_flags & POSIX_SPAWN_SETSCHEDULER) {
108 		if (sched_setscheduler(0, sa->sa_schedpolicy,
109 		    &sa->sa_schedparam) != 0)
110 			return (errno);
111 	} else if (sa->sa_flags & POSIX_SPAWN_SETSCHEDPARAM) {
112 		if (sched_setparam(0, &sa->sa_schedparam) != 0)
113 			return (errno);
114 	}
115 
116 	/* Reset user ID's */
117 	if (sa->sa_flags & POSIX_SPAWN_RESETIDS) {
118 		if (setegid(getgid()) != 0)
119 			return (errno);
120 		if (seteuid(getuid()) != 0)
121 			return (errno);
122 	}
123 
124 	/*
125 	 * Set signal masks/defaults.
126 	 * Use unwrapped syscall, libthr is in undefined state after vfork().
127 	 */
128 	if (sa->sa_flags & POSIX_SPAWN_SETSIGMASK) {
129 		__sys_sigprocmask(SIG_SETMASK, &sa->sa_sigmask, NULL);
130 	}
131 
132 	if (sa->sa_flags & POSIX_SPAWN_SETSIGDEF) {
133 		for (i = 1; i <= _SIG_MAXSIG; i++) {
134 			if (sigismember(&sa->sa_sigdefault, i))
135 				if (__sys_sigaction(i, &sigact, NULL) != 0)
136 					return (errno);
137 		}
138 	}
139 
140 	return (0);
141 }
142 
143 static int
144 process_file_actions_entry(posix_spawn_file_actions_entry_t *fae)
145 {
146 	int fd, saved_errno;
147 
148 	switch (fae->fae_action) {
149 	case FAE_OPEN:
150 		/* Perform an open(), make it use the right fd */
151 		fd = _open(fae->fae_path, fae->fae_oflag, fae->fae_mode);
152 		if (fd < 0)
153 			return (errno);
154 		if (fd != fae->fae_fildes) {
155 			if (_dup2(fd, fae->fae_fildes) == -1) {
156 				saved_errno = errno;
157 				(void)_close(fd);
158 				return (saved_errno);
159 			}
160 			if (_close(fd) != 0) {
161 				if (errno == EBADF)
162 					return (EBADF);
163 			}
164 		}
165 		if (_fcntl(fae->fae_fildes, F_SETFD, 0) == -1)
166 			return (errno);
167 		break;
168 	case FAE_DUP2:
169 		/* Perform a dup2() */
170 		if (_dup2(fae->fae_fildes, fae->fae_newfildes) == -1)
171 			return (errno);
172 		if (_fcntl(fae->fae_newfildes, F_SETFD, 0) == -1)
173 			return (errno);
174 		break;
175 	case FAE_CLOSE:
176 		/* Perform a close(), do not fail if already closed */
177 		(void)_close(fae->fae_fildes);
178 		break;
179 	}
180 	return (0);
181 }
182 
183 static int
184 process_file_actions(const posix_spawn_file_actions_t fa)
185 {
186 	posix_spawn_file_actions_entry_t *fae;
187 	int error;
188 
189 	/* Replay all file descriptor modifications */
190 	STAILQ_FOREACH(fae, &fa->fa_list, fae_list) {
191 		error = process_file_actions_entry(fae);
192 		if (error)
193 			return (error);
194 	}
195 	return (0);
196 }
197 
198 struct posix_spawn_args {
199 	const char *path;
200 	const posix_spawn_file_actions_t *fa;
201 	const posix_spawnattr_t *sa;
202 	char * const * argv;
203 	char * const * envp;
204 	int use_env_path;
205 	volatile int error;
206 };
207 
208 #define	PSPAWN_STACK_ALIGNMENT	16
209 #define	PSPAWN_STACK_ALIGNBYTES	(PSPAWN_STACK_ALIGNMENT - 1)
210 #define	PSPAWN_STACK_ALIGN(sz) \
211 	(((sz) + PSPAWN_STACK_ALIGNBYTES) & ~PSPAWN_STACK_ALIGNBYTES)
212 
213 #if defined(__i386__) || defined(__amd64__)
214 /*
215  * Below we'll assume that _RFORK_THREAD_STACK_SIZE is appropriately aligned for
216  * the posix_spawn() case where we do not end up calling _execvpe and won't ever
217  * try to allocate space on the stack for argv[].
218  */
219 #define	_RFORK_THREAD_STACK_SIZE	4096
220 _Static_assert((_RFORK_THREAD_STACK_SIZE % PSPAWN_STACK_ALIGNMENT) == 0,
221     "Inappropriate stack size alignment");
222 #endif
223 
224 static int
225 _posix_spawn_thr(void *data)
226 {
227 	struct posix_spawn_args *psa;
228 	char * const *envp;
229 
230 	psa = data;
231 	if (psa->sa != NULL) {
232 		psa->error = process_spawnattr(*psa->sa);
233 		if (psa->error)
234 			_exit(127);
235 	}
236 	if (psa->fa != NULL) {
237 		psa->error = process_file_actions(*psa->fa);
238 		if (psa->error)
239 			_exit(127);
240 	}
241 	envp = psa->envp != NULL ? psa->envp : environ;
242 	if (psa->use_env_path)
243 		_execvpe(psa->path, psa->argv, envp);
244 	else
245 		_execve(psa->path, psa->argv, envp);
246 	psa->error = errno;
247 
248 	/* This is called in such a way that it must not exit. */
249 	_exit(127);
250 }
251 
252 static int
253 do_posix_spawn(pid_t *pid, const char *path,
254     const posix_spawn_file_actions_t *fa,
255     const posix_spawnattr_t *sa,
256     char * const argv[], char * const envp[], int use_env_path)
257 {
258 	struct posix_spawn_args psa;
259 	pid_t p;
260 #ifdef _RFORK_THREAD_STACK_SIZE
261 	char *stack;
262 	size_t cnt, stacksz;
263 
264 	stacksz = _RFORK_THREAD_STACK_SIZE;
265 	if (use_env_path) {
266 		/*
267 		 * We need to make sure we have enough room on the stack for the
268 		 * potential alloca() in execvPe if it gets kicked back an
269 		 * ENOEXEC from execve(2), plus the original buffer we gave
270 		 * ourselves; this protects us in the event that the caller
271 		 * intentionally or inadvertently supplies enough arguments to
272 		 * make us blow past the stack we've allocated from it.
273 		 */
274 		for (cnt = 0; argv[cnt] != NULL; ++cnt)
275 			;
276 		stacksz += MAX(3, cnt + 2) * sizeof(char *);
277 		stacksz = PSPAWN_STACK_ALIGN(stacksz);
278 	}
279 
280 	/*
281 	 * aligned_alloc is not safe to use here, because we can't guarantee
282 	 * that aligned_alloc and free will be provided by the same
283 	 * implementation.  We've actively hit at least one application that
284 	 * will provide its own malloc/free but not aligned_alloc leading to
285 	 * a free by the wrong allocator.
286 	 */
287 	stack = malloc(stacksz);
288 	if (stack == NULL)
289 		return (ENOMEM);
290 	stacksz = (((uintptr_t)stack + stacksz) & ~PSPAWN_STACK_ALIGNBYTES) -
291 	    (uintptr_t)stack;
292 #endif
293 	psa.path = path;
294 	psa.fa = fa;
295 	psa.sa = sa;
296 	psa.argv = argv;
297 	psa.envp = envp;
298 	psa.use_env_path = use_env_path;
299 	psa.error = 0;
300 
301 	/*
302 	 * Passing RFSPAWN to rfork(2) gives us effectively a vfork that drops
303 	 * non-ignored signal handlers.  We'll fall back to the slightly less
304 	 * ideal vfork(2) if we get an EINVAL from rfork -- this should only
305 	 * happen with newer libc on older kernel that doesn't accept
306 	 * RFSPAWN.
307 	 */
308 #ifdef _RFORK_THREAD_STACK_SIZE
309 	/*
310 	 * x86 stores the return address on the stack, so rfork(2) cannot work
311 	 * as-is because the child would clobber the return address om the
312 	 * parent.  Because of this, we must use rfork_thread instead while
313 	 * almost every other arch stores the return address in a register.
314 	 */
315 	p = rfork_thread(RFSPAWN, stack + stacksz, _posix_spawn_thr, &psa);
316 	free(stack);
317 #else
318 	p = rfork(RFSPAWN);
319 	if (p == 0)
320 		/* _posix_spawn_thr does not return */
321 		_posix_spawn_thr(&psa);
322 #endif
323 	/*
324 	 * The above block should leave us in a state where we've either
325 	 * succeeded and we're ready to process the results, or we need to
326 	 * fallback to vfork() if the kernel didn't like RFSPAWN.
327 	 */
328 
329 	if (p == -1 && errno == EINVAL) {
330 		p = vfork();
331 		if (p == 0)
332 			/* _posix_spawn_thr does not return */
333 			_posix_spawn_thr(&psa);
334 	}
335 	if (p == -1)
336 		return (errno);
337 	if (psa.error != 0)
338 		/* Failed; ready to reap */
339 		_waitpid(p, NULL, WNOHANG);
340 	else if (pid != NULL)
341 		/* exec succeeded */
342 		*pid = p;
343 	return (psa.error);
344 }
345 
346 int
347 posix_spawn(pid_t *pid, const char *path,
348     const posix_spawn_file_actions_t *fa,
349     const posix_spawnattr_t *sa,
350     char * const argv[], char * const envp[])
351 {
352 	return do_posix_spawn(pid, path, fa, sa, argv, envp, 0);
353 }
354 
355 int
356 posix_spawnp(pid_t *pid, const char *path,
357     const posix_spawn_file_actions_t *fa,
358     const posix_spawnattr_t *sa,
359     char * const argv[], char * const envp[])
360 {
361 	return do_posix_spawn(pid, path, fa, sa, argv, envp, 1);
362 }
363 
364 /*
365  * File descriptor actions
366  */
367 
368 int
369 posix_spawn_file_actions_init(posix_spawn_file_actions_t *ret)
370 {
371 	posix_spawn_file_actions_t fa;
372 
373 	fa = malloc(sizeof(struct __posix_spawn_file_actions));
374 	if (fa == NULL)
375 		return (-1);
376 
377 	STAILQ_INIT(&fa->fa_list);
378 	*ret = fa;
379 	return (0);
380 }
381 
382 int
383 posix_spawn_file_actions_destroy(posix_spawn_file_actions_t *fa)
384 {
385 	posix_spawn_file_actions_entry_t *fae;
386 
387 	while ((fae = STAILQ_FIRST(&(*fa)->fa_list)) != NULL) {
388 		/* Remove file action entry from the queue */
389 		STAILQ_REMOVE_HEAD(&(*fa)->fa_list, fae_list);
390 
391 		/* Deallocate file action entry */
392 		if (fae->fae_action == FAE_OPEN)
393 			free(fae->fae_path);
394 		free(fae);
395 	}
396 
397 	free(*fa);
398 	return (0);
399 }
400 
401 int
402 posix_spawn_file_actions_addopen(posix_spawn_file_actions_t * __restrict fa,
403     int fildes, const char * __restrict path, int oflag, mode_t mode)
404 {
405 	posix_spawn_file_actions_entry_t *fae;
406 	int error;
407 
408 	if (fildes < 0)
409 		return (EBADF);
410 
411 	/* Allocate object */
412 	fae = malloc(sizeof(posix_spawn_file_actions_entry_t));
413 	if (fae == NULL)
414 		return (errno);
415 
416 	/* Set values and store in queue */
417 	fae->fae_action = FAE_OPEN;
418 	fae->fae_path = strdup(path);
419 	if (fae->fae_path == NULL) {
420 		error = errno;
421 		free(fae);
422 		return (error);
423 	}
424 	fae->fae_fildes = fildes;
425 	fae->fae_oflag = oflag;
426 	fae->fae_mode = mode;
427 
428 	STAILQ_INSERT_TAIL(&(*fa)->fa_list, fae, fae_list);
429 	return (0);
430 }
431 
432 int
433 posix_spawn_file_actions_adddup2(posix_spawn_file_actions_t *fa,
434     int fildes, int newfildes)
435 {
436 	posix_spawn_file_actions_entry_t *fae;
437 
438 	if (fildes < 0 || newfildes < 0)
439 		return (EBADF);
440 
441 	/* Allocate object */
442 	fae = malloc(sizeof(posix_spawn_file_actions_entry_t));
443 	if (fae == NULL)
444 		return (errno);
445 
446 	/* Set values and store in queue */
447 	fae->fae_action = FAE_DUP2;
448 	fae->fae_fildes = fildes;
449 	fae->fae_newfildes = newfildes;
450 
451 	STAILQ_INSERT_TAIL(&(*fa)->fa_list, fae, fae_list);
452 	return (0);
453 }
454 
455 int
456 posix_spawn_file_actions_addclose(posix_spawn_file_actions_t *fa,
457     int fildes)
458 {
459 	posix_spawn_file_actions_entry_t *fae;
460 
461 	if (fildes < 0)
462 		return (EBADF);
463 
464 	/* Allocate object */
465 	fae = malloc(sizeof(posix_spawn_file_actions_entry_t));
466 	if (fae == NULL)
467 		return (errno);
468 
469 	/* Set values and store in queue */
470 	fae->fae_action = FAE_CLOSE;
471 	fae->fae_fildes = fildes;
472 
473 	STAILQ_INSERT_TAIL(&(*fa)->fa_list, fae, fae_list);
474 	return (0);
475 }
476 
477 /*
478  * Spawn attributes
479  */
480 
481 int
482 posix_spawnattr_init(posix_spawnattr_t *ret)
483 {
484 	posix_spawnattr_t sa;
485 
486 	sa = calloc(1, sizeof(struct __posix_spawnattr));
487 	if (sa == NULL)
488 		return (errno);
489 
490 	/* Set defaults as specified by POSIX, cleared above */
491 	*ret = sa;
492 	return (0);
493 }
494 
495 int
496 posix_spawnattr_destroy(posix_spawnattr_t *sa)
497 {
498 	free(*sa);
499 	return (0);
500 }
501 
502 int
503 posix_spawnattr_getflags(const posix_spawnattr_t * __restrict sa,
504     short * __restrict flags)
505 {
506 	*flags = (*sa)->sa_flags;
507 	return (0);
508 }
509 
510 int
511 posix_spawnattr_getpgroup(const posix_spawnattr_t * __restrict sa,
512     pid_t * __restrict pgroup)
513 {
514 	*pgroup = (*sa)->sa_pgroup;
515 	return (0);
516 }
517 
518 int
519 posix_spawnattr_getschedparam(const posix_spawnattr_t * __restrict sa,
520     struct sched_param * __restrict schedparam)
521 {
522 	*schedparam = (*sa)->sa_schedparam;
523 	return (0);
524 }
525 
526 int
527 posix_spawnattr_getschedpolicy(const posix_spawnattr_t * __restrict sa,
528     int * __restrict schedpolicy)
529 {
530 	*schedpolicy = (*sa)->sa_schedpolicy;
531 	return (0);
532 }
533 
534 int
535 posix_spawnattr_getsigdefault(const posix_spawnattr_t * __restrict sa,
536     sigset_t * __restrict sigdefault)
537 {
538 	*sigdefault = (*sa)->sa_sigdefault;
539 	return (0);
540 }
541 
542 int
543 posix_spawnattr_getsigmask(const posix_spawnattr_t * __restrict sa,
544     sigset_t * __restrict sigmask)
545 {
546 	*sigmask = (*sa)->sa_sigmask;
547 	return (0);
548 }
549 
550 int
551 posix_spawnattr_setflags(posix_spawnattr_t *sa, short flags)
552 {
553 	(*sa)->sa_flags = flags;
554 	return (0);
555 }
556 
557 int
558 posix_spawnattr_setpgroup(posix_spawnattr_t *sa, pid_t pgroup)
559 {
560 	(*sa)->sa_pgroup = pgroup;
561 	return (0);
562 }
563 
564 int
565 posix_spawnattr_setschedparam(posix_spawnattr_t * __restrict sa,
566     const struct sched_param * __restrict schedparam)
567 {
568 	(*sa)->sa_schedparam = *schedparam;
569 	return (0);
570 }
571 
572 int
573 posix_spawnattr_setschedpolicy(posix_spawnattr_t *sa, int schedpolicy)
574 {
575 	(*sa)->sa_schedpolicy = schedpolicy;
576 	return (0);
577 }
578 
579 int
580 posix_spawnattr_setsigdefault(posix_spawnattr_t * __restrict sa,
581     const sigset_t * __restrict sigdefault)
582 {
583 	(*sa)->sa_sigdefault = *sigdefault;
584 	return (0);
585 }
586 
587 int
588 posix_spawnattr_setsigmask(posix_spawnattr_t * __restrict sa,
589     const sigset_t * __restrict sigmask)
590 {
591 	(*sa)->sa_sigmask = *sigmask;
592 	return (0);
593 }
594