1 /* POSIX spawn interface. Linux version.
2 Copyright (C) 2016-2023 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
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.
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.
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/>. */
19 #include <internal-signals.h>
21 #include <local-setxid.h>
22 #include <not-cancel.h>
24 #include <shlib-compat.h>
26 #include <spawn_int.h>
28 #include <sys/resource.h>
29 #include <clone_internal.h>
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:
36 1. That no signal handlers must run in child context, to avoid corrupting
38 2. The parent must ensure child's stack freeing.
39 3. Child must synchronize with parent to enforce 2. and to possible
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. */
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
58 struct posix_spawn_args
60 internal_sigset_t oldmask
;
62 int (*exec
) (const char *, char *const *, char *const *);
63 const posix_spawn_file_actions_t
*fa
;
64 const posix_spawnattr_t
*restrict attr
;
72 /* Older version requires that shell script without shebang definition
73 to be called explicitly using /bin/sh (_PATH_BSHELL). */
75 maybe_script_execute (struct posix_spawn_args
*args
)
77 if (SHLIB_COMPAT (libc
, GLIBC_2_2
, GLIBC_2_15
)
78 && (args
->xflags
& SPAWN_XFLAGS_TRY_SHELL
) && errno
== ENOEXEC
)
80 char *const *argv
= args
->argv
;
81 ptrdiff_t argc
= args
->argc
;
83 /* Construct an argument list for the shell. */
84 char *new_argv
[argc
+ 2];
85 new_argv
[0] = (char *) _PATH_BSHELL
;
86 new_argv
[1] = (char *) args
->file
;
88 memcpy (new_argv
+ 2, argv
+ 1, argc
* sizeof (char *));
92 /* Execute the shell. */
93 args
->exec (new_argv
[0], new_argv
, args
->envp
);
97 /* Function used in the clone call to setup the signals mask, posix_spawn
98 attributes, and file actions. It run on its own stack (provided by the
101 __spawni_child (void *arguments
)
103 struct posix_spawn_args
*args
= arguments
;
104 const posix_spawnattr_t
*restrict attr
= args
->attr
;
105 const posix_spawn_file_actions_t
*file_actions
= args
->fa
;
107 /* The child must ensure that no signal handler are enabled because it shared
108 memory with parent, so the signal disposition must be either SIG_DFL or
109 SIG_IGN. It does by iterating over all signals and although it could
110 possibly be more optimized (by tracking which signal potentially have a
111 signal handler), it might requires system specific solutions (since the
112 sigset_t data type can be very different on different architectures). */
114 memset (&sa
, '\0', sizeof (sa
));
117 __sigprocmask (SIG_BLOCK
, 0, &hset
);
118 for (int sig
= 1; sig
< _NSIG
; ++sig
)
120 if ((attr
->__flags
& POSIX_SPAWN_SETSIGDEF
)
121 && __sigismember (&attr
->__sd
, sig
))
123 sa
.sa_handler
= SIG_DFL
;
125 else if (__sigismember (&hset
, sig
))
127 if (is_internal_signal (sig
))
128 sa
.sa_handler
= SIG_IGN
;
131 __libc_sigaction (sig
, 0, &sa
);
132 if (sa
.sa_handler
== SIG_IGN
)
134 sa
.sa_handler
= SIG_DFL
;
140 __libc_sigaction (sig
, &sa
, 0);
143 #ifdef _POSIX_PRIORITY_SCHEDULING
144 /* Set the scheduling algorithm and parameters. */
145 if ((attr
->__flags
& (POSIX_SPAWN_SETSCHEDPARAM
| POSIX_SPAWN_SETSCHEDULER
))
146 == POSIX_SPAWN_SETSCHEDPARAM
)
148 if (__sched_setparam (0, &attr
->__sp
) == -1)
151 else if ((attr
->__flags
& POSIX_SPAWN_SETSCHEDULER
) != 0)
153 if (__sched_setscheduler (0, attr
->__policy
, &attr
->__sp
) == -1)
158 if ((attr
->__flags
& POSIX_SPAWN_SETSID
) != 0
162 /* Set the process group ID. */
163 if ((attr
->__flags
& POSIX_SPAWN_SETPGROUP
) != 0
164 && __setpgid (0, attr
->__pgrp
) != 0)
167 /* Set the effective user and group IDs. */
168 if ((attr
->__flags
& POSIX_SPAWN_RESETIDS
) != 0
169 && (local_seteuid (__getuid ()) != 0
170 || local_setegid (__getgid ()) != 0))
173 /* Execute the file actions. */
174 if (file_actions
!= 0)
177 struct rlimit64 fdlimit
;
178 bool have_fdlimit
= false;
180 for (cnt
= 0; cnt
< file_actions
->__used
; ++cnt
)
182 struct __spawn_action
*action
= &file_actions
->__actions
[cnt
];
187 if (__close_nocancel (action
->action
.close_action
.fd
) != 0)
191 __getrlimit64 (RLIMIT_NOFILE
, &fdlimit
);
195 /* Signal errors only for file descriptors out of range. */
196 if (action
->action
.close_action
.fd
< 0
197 || action
->action
.close_action
.fd
>= fdlimit
.rlim_cur
)
204 /* POSIX states that if fildes was already an open file descriptor,
205 it shall be closed before the new file is opened. This avoid
206 pontential issues when posix_spawn plus addopen action is called
207 with the process already at maximum number of file descriptor
208 opened and also for multiple actions on single-open special
209 paths (like /dev/watchdog). */
210 __close_nocancel (action
->action
.open_action
.fd
);
212 int ret
= __open_nocancel (action
->action
.open_action
.path
,
214 open_action
.oflag
| O_LARGEFILE
,
215 action
->action
.open_action
.mode
);
222 /* Make sure the desired file descriptor is used. */
223 if (ret
!= action
->action
.open_action
.fd
)
225 if (__dup2 (new_fd
, action
->action
.open_action
.fd
)
226 != action
->action
.open_action
.fd
)
229 if (__close_nocancel (new_fd
) != 0)
236 /* Austin Group issue #411 requires adddup2 action with source
237 and destination being equal to remove close-on-exec flag. */
238 if (action
->action
.dup2_action
.fd
239 == action
->action
.dup2_action
.newfd
)
241 int fd
= action
->action
.dup2_action
.newfd
;
242 int flags
= __fcntl (fd
, F_GETFD
, 0);
245 if (__fcntl (fd
, F_SETFD
, flags
& ~FD_CLOEXEC
) == -1)
248 else if (__dup2 (action
->action
.dup2_action
.fd
,
249 action
->action
.dup2_action
.newfd
)
250 != action
->action
.dup2_action
.newfd
)
255 if (__chdir (action
->action
.chdir_action
.path
) != 0)
259 case spawn_do_fchdir
:
260 if (__fchdir (action
->action
.fchdir_action
.fd
) != 0)
264 case spawn_do_closefrom
:
266 int lowfd
= action
->action
.closefrom_action
.from
;
267 int r
= INLINE_SYSCALL_CALL (close_range
, lowfd
, ~0U, 0);
268 if (r
!= 0 && !__closefrom_fallback (lowfd
, false))
272 case spawn_do_tcsetpgrp
:
274 /* Check if it is possible to avoid an extra syscall. */
275 pid_t pgrp
= (attr
->__flags
& POSIX_SPAWN_SETPGROUP
) != 0
277 ? attr
->__pgrp
: __getpgid (0);
278 if (__tcsetpgrp (action
->action
.setpgrp_action
.fd
, pgrp
) != 0)
285 /* Set the initial signal mask of the child if POSIX_SPAWN_SETSIGMASK
286 is set, otherwise restore the previous one. */
287 if (attr
->__flags
& POSIX_SPAWN_SETSIGMASK
)
288 __sigprocmask (SIG_SETMASK
, &attr
->__ss
, NULL
);
290 internal_sigprocmask (SIG_SETMASK
, &args
->oldmask
, NULL
);
292 args
->exec (args
->file
, args
->argv
, args
->envp
);
294 /* This is compatibility function required to enable posix_spawn run
295 script without shebang definition for older posix_spawn versions
297 maybe_script_execute (args
);
300 /* errno should have an appropriate non-zero value; otherwise,
301 there's a bug in glibc or the kernel. For lack of an error code
302 (EINTERNALBUG) describing that, use ECHILD. Another option would
303 be to set args->err to some negative sentinel and have the parent
304 abort(), but that seems needlessly harsh. */
305 args
->err
= errno
? : ECHILD
;
309 /* Spawn a new process executing PATH with the attributes describes in *ATTRP.
310 Before running the process perform the actions described in FILE-ACTIONS. */
312 __spawnix (pid_t
* pid
, const char *file
,
313 const posix_spawn_file_actions_t
* file_actions
,
314 const posix_spawnattr_t
* attrp
, char *const argv
[],
315 char *const envp
[], int xflags
,
316 int (*exec
) (const char *, char *const *, char *const *))
319 struct posix_spawn_args args
;
322 /* To avoid imposing hard limits on posix_spawn{p} the total number of
323 arguments is first calculated to allocate a mmap to hold all possible
326 /* Linux allows at most max (0x7FFFFFFF, 1/4 stack size) arguments
327 to be used in a execve call. We limit to INT_MAX minus one due the
328 compatiblity code that may execute a shell script (maybe_script_execute)
329 where it will construct another argument list with an additional
331 ptrdiff_t limit
= INT_MAX
- 1;
332 while (argv
[argc
++] != NULL
)
339 int prot
= (PROT_READ
| PROT_WRITE
340 | ((GL (dl_stack_flags
) & PF_X
) ? PROT_EXEC
: 0));
342 /* Add a slack area for child's stack. */
343 size_t argv_size
= (argc
* sizeof (void *)) + 512;
344 /* We need at least a few pages in case the compiler's stack checking is
345 enabled. In some configs, it is known to use at least 24KiB. We use
346 32KiB to be "safe" from anything the compiler might do. Besides, the
347 extra pages won't actually be allocated unless they get used.
348 It also acts the slack for spawn_closefrom (including MIPS64 getdents64
349 where it might use about 1k extra stack space). */
350 argv_size
+= (32 * 1024);
351 size_t stack_size
= ALIGN_UP (argv_size
, GLRO(dl_pagesize
));
352 void *stack
= __mmap (NULL
, stack_size
, prot
,
353 MAP_PRIVATE
| MAP_ANONYMOUS
| MAP_STACK
, -1, 0);
354 if (__glibc_unlikely (stack
== MAP_FAILED
))
357 /* Disable asynchronous cancellation. */
359 __pthread_setcancelstate (PTHREAD_CANCEL_DISABLE
, &state
);
361 /* Child must set args.err to something non-negative - we rely on
362 the parent and child sharing VM. */
366 args
.fa
= file_actions
;
367 args
.attr
= attrp
? attrp
: &(const posix_spawnattr_t
) { 0 };
371 args
.xflags
= xflags
;
373 internal_signal_block_all (&args
.oldmask
);
375 /* The clone flags used will create a new child that will run in the same
376 memory space (CLONE_VM) and the execution of calling thread will be
377 suspend until the child calls execve or _exit.
379 Also since the calling thread execution will be suspend, there is not
380 need for CLONE_SETTLS. Although parent and child share the same TLS
381 namespace, there will be no concurrent access for TLS variables (errno
383 struct clone_args clone_args
=
385 .flags
= CLONE_VM
| CLONE_VFORK
,
386 .exit_signal
= SIGCHLD
,
387 .stack
= (uintptr_t) stack
,
388 .stack_size
= stack_size
,
390 new_pid
= __clone_internal (&clone_args
, __spawni_child
, &args
);
392 /* It needs to collect the case where the auxiliary process was created
393 but failed to execute the file (due either any preparation step or
394 for execve itself). */
397 /* Also, it handles the unlikely case where the auxiliary process was
398 terminated before calling execve as if it was successfully. The
399 args.err is set to 0 as default and changed to a positive value
400 only in case of failure, so in case of premature termination
401 due a signal args.err will remain zeroed and it will be up to
402 caller to actually collect it. */
405 /* There still an unlikely case where the child is cancelled after
406 setting args.err, due to a positive error value. Also there is
407 possible pid reuse race (where the kernel allocated the same pid
408 to an unrelated process). Unfortunately due synchronization
409 issues where the kernel might not have the process collected
410 the waitpid below can not use WNOHANG. */
411 __waitpid (new_pid
, NULL
, 0);
416 __munmap (stack
, stack_size
);
418 if ((ec
== 0) && (pid
!= NULL
))
421 internal_signal_restore_set (&args
.oldmask
);
423 __pthread_setcancelstate (state
, NULL
);
428 /* Spawn a new process executing PATH with the attributes describes in *ATTRP.
429 Before running the process perform the actions described in FILE-ACTIONS. */
431 __spawni (pid_t
* pid
, const char *file
,
432 const posix_spawn_file_actions_t
* acts
,
433 const posix_spawnattr_t
* attrp
, char *const argv
[],
434 char *const envp
[], int xflags
)
436 /* It uses __execvpex to avoid run ENOEXEC in non compatibility mode (it
437 will be handled by maybe_script_execute). */
438 return __spawnix (pid
, file
, acts
, attrp
, argv
, envp
, xflags
,
439 xflags
& SPAWN_XFLAGS_USE_PATH
? __execvpex
:__execve
);