1/* POSIX spawn interface. Linux version.
2 Copyright (C) 2016-2017 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 <http://www.gnu.org/licenses/>. */
18
19#include <spawn.h>
20#include <fcntl.h>
21#include <paths.h>
22#include <string.h>
23#include <sys/resource.h>
24#include <sys/wait.h>
25#include <sys/param.h>
26#include <sys/mman.h>
27#include <not-cancel.h>
28#include <local-setxid.h>
29#include <shlib-compat.h>
30#include <nptl/pthreadP.h>
31#include <dl-sysdep.h>
32#include <ldsodefs.h>
33#include <libc-internal.h>
34#include "spawn_int.h"
35
36/* The Linux implementation of posix_spawn{p} uses the clone syscall directly
37 with CLONE_VM and CLONE_VFORK flags and an allocated stack. The new stack
38 and start function solves most the vfork limitation (possible parent
39 clobber due stack spilling). The remaining issue are:
40
41 1. That no signal handlers must run in child context, to avoid corrupting
42 parent's state.
43 2. The parent must ensure child's stack freeing.
44 3. Child must synchronize with parent to enforce 2. and to possible
45 return execv issues.
46
47 The first issue is solved by blocking all signals in child, even
48 the NPTL-internal ones (SIGCANCEL and SIGSETXID). The second and
49 third issue is done by a stack allocation in parent, and by using a
50 field in struct spawn_args where the child can write an error
51 code. CLONE_VFORK ensures that the parent does not run until the
52 child has either exec'ed successfully or exited. */
53
54
55/* The Unix standard contains a long explanation of the way to signal
56 an error after the fork() was successful. Since no new wait status
57 was wanted there is no way to signal an error using one of the
58 available methods. The committee chose to signal an error by a
59 normal program exit with the exit code 127. */
60#define SPAWN_ERROR 127
61
62#ifdef __ia64__
63# define CLONE(__fn, __stackbase, __stacksize, __flags, __args) \
64 __clone2 (__fn, __stackbase, __stacksize, __flags, __args, 0, 0, 0)
65#else
66# define CLONE(__fn, __stack, __stacksize, __flags, __args) \
67 __clone (__fn, __stack, __flags, __args)
68#endif
69
70/* Since ia64 wants the stackbase w/clone2, re-use the grows-up macro. */
71#if _STACK_GROWS_UP || defined (__ia64__)
72# define STACK(__stack, __stack_size) (__stack)
73#elif _STACK_GROWS_DOWN
74# define STACK(__stack, __stack_size) (__stack + __stack_size)
75#endif
76
77
78struct posix_spawn_args
79{
80 sigset_t oldmask;
81 const char *file;
82 int (*exec) (const char *, char *const *, char *const *);
83 const posix_spawn_file_actions_t *fa;
84 const posix_spawnattr_t *restrict attr;
85 char *const *argv;
86 ptrdiff_t argc;
87 char *const *envp;
88 int xflags;
89 int err;
90};
91
92/* Older version requires that shell script without shebang definition
93 to be called explicitly using /bin/sh (_PATH_BSHELL). */
94static void
95maybe_script_execute (struct posix_spawn_args *args)
96{
97 if (SHLIB_COMPAT (libc, GLIBC_2_2, GLIBC_2_15)
98 && (args->xflags & SPAWN_XFLAGS_TRY_SHELL) && errno == ENOEXEC)
99 {
100 char *const *argv = args->argv;
101 ptrdiff_t argc = args->argc;
102
103 /* Construct an argument list for the shell. */
104 char *new_argv[argc + 2];
105 new_argv[0] = (char *) _PATH_BSHELL;
106 new_argv[1] = (char *) args->file;
107 if (argc > 1)
108 memcpy (new_argv + 2, argv + 1, argc * sizeof(char *));
109 else
110 new_argv[2] = NULL;
111
112 /* Execute the shell. */
113 args->exec (new_argv[0], new_argv, args->envp);
114 }
115}
116
117/* Function used in the clone call to setup the signals mask, posix_spawn
118 attributes, and file actions. It run on its own stack (provided by the
119 posix_spawn call). */
120static int
121__spawni_child (void *arguments)
122{
123 struct posix_spawn_args *args = arguments;
124 const posix_spawnattr_t *restrict attr = args->attr;
125 const posix_spawn_file_actions_t *file_actions = args->fa;
126 int ret;
127
128 /* The child must ensure that no signal handler are enabled because it shared
129 memory with parent, so the signal disposition must be either SIG_DFL or
130 SIG_IGN. It does by iterating over all signals and although it could
131 possibly be more optimized (by tracking which signal potentially have a
132 signal handler), it might requires system specific solutions (since the
133 sigset_t data type can be very different on different architectures). */
134 struct sigaction sa;
135 memset (&sa, '\0', sizeof (sa));
136
137 sigset_t hset;
138 __sigprocmask (SIG_BLOCK, 0, &hset);
139 for (int sig = 1; sig < _NSIG; ++sig)
140 {
141 if ((attr->__flags & POSIX_SPAWN_SETSIGDEF)
142 && sigismember (&attr->__sd, sig))
143 {
144 sa.sa_handler = SIG_DFL;
145 }
146 else if (sigismember (&hset, sig))
147 {
148 if (__nptl_is_internal_signal (sig))
149 sa.sa_handler = SIG_IGN;
150 else
151 {
152 __libc_sigaction (sig, 0, &sa);
153 if (sa.sa_handler == SIG_IGN)
154 continue;
155 sa.sa_handler = SIG_DFL;
156 }
157 }
158 else
159 continue;
160
161 __libc_sigaction (sig, &sa, 0);
162 }
163
164#ifdef _POSIX_PRIORITY_SCHEDULING
165 /* Set the scheduling algorithm and parameters. */
166 if ((attr->__flags & (POSIX_SPAWN_SETSCHEDPARAM | POSIX_SPAWN_SETSCHEDULER))
167 == POSIX_SPAWN_SETSCHEDPARAM)
168 {
169 if ((ret = __sched_setparam (0, &attr->__sp)) == -1)
170 goto fail;
171 }
172 else if ((attr->__flags & POSIX_SPAWN_SETSCHEDULER) != 0)
173 {
174 if ((ret = __sched_setscheduler (0, attr->__policy, &attr->__sp)) == -1)
175 goto fail;
176 }
177#endif
178
179 /* Set the process group ID. */
180 if ((attr->__flags & POSIX_SPAWN_SETPGROUP) != 0
181 && (ret = __setpgid (0, attr->__pgrp)) != 0)
182 goto fail;
183
184 /* Set the effective user and group IDs. */
185 if ((attr->__flags & POSIX_SPAWN_RESETIDS) != 0
186 && ((ret = local_seteuid (__getuid ())) != 0
187 || (ret = local_setegid (__getgid ())) != 0))
188 goto fail;
189
190 /* Execute the file actions. */
191 if (file_actions != 0)
192 {
193 int cnt;
194 struct rlimit64 fdlimit;
195 bool have_fdlimit = false;
196
197 for (cnt = 0; cnt < file_actions->__used; ++cnt)
198 {
199 struct __spawn_action *action = &file_actions->__actions[cnt];
200
201 switch (action->tag)
202 {
203 case spawn_do_close:
204 if ((ret =
205 close_not_cancel (action->action.close_action.fd)) != 0)
206 {
207 if (!have_fdlimit)
208 {
209 __getrlimit64 (RLIMIT_NOFILE, &fdlimit);
210 have_fdlimit = true;
211 }
212
213 /* Signal errors only for file descriptors out of range. */
214 if (action->action.close_action.fd < 0
215 || action->action.close_action.fd >= fdlimit.rlim_cur)
216 goto fail;
217 }
218 break;
219
220 case spawn_do_open:
221 {
222 /* POSIX states that if fildes was already an open file descriptor,
223 it shall be closed before the new file is opened. This avoid
224 pontential issues when posix_spawn plus addopen action is called
225 with the process already at maximum number of file descriptor
226 opened and also for multiple actions on single-open special
227 paths (like /dev/watchdog). */
228 close_not_cancel (action->action.open_action.fd);
229
230 ret = open_not_cancel (action->action.open_action.path,
231 action->action.
232 open_action.oflag | O_LARGEFILE,
233 action->action.open_action.mode);
234
235 if (ret == -1)
236 goto fail;
237
238 int new_fd = ret;
239
240 /* Make sure the desired file descriptor is used. */
241 if (ret != action->action.open_action.fd)
242 {
243 if ((ret = __dup2 (new_fd, action->action.open_action.fd))
244 != action->action.open_action.fd)
245 goto fail;
246
247 if ((ret = close_not_cancel (new_fd)) != 0)
248 goto fail;
249 }
250 }
251 break;
252
253 case spawn_do_dup2:
254 if ((ret = __dup2 (action->action.dup2_action.fd,
255 action->action.dup2_action.newfd))
256 != action->action.dup2_action.newfd)
257 goto fail;
258 break;
259 }
260 }
261 }
262
263 /* Set the initial signal mask of the child if POSIX_SPAWN_SETSIGMASK
264 is set, otherwise restore the previous one. */
265 __sigprocmask (SIG_SETMASK, (attr->__flags & POSIX_SPAWN_SETSIGMASK)
266 ? &attr->__ss : &args->oldmask, 0);
267
268 args->exec (args->file, args->argv, args->envp);
269
270 /* This is compatibility function required to enable posix_spawn run
271 script without shebang definition for older posix_spawn versions
272 (2.15). */
273 maybe_script_execute (args);
274
275fail:
276 /* errno should have an appropriate non-zero value; otherwise,
277 there's a bug in glibc or the kernel. For lack of an error code
278 (EINTERNALBUG) describing that, use ECHILD. Another option would
279 be to set args->err to some negative sentinel and have the parent
280 abort(), but that seems needlessly harsh. */
281 args->err = errno ? : ECHILD;
282 _exit (SPAWN_ERROR);
283}
284
285/* Spawn a new process executing PATH with the attributes describes in *ATTRP.
286 Before running the process perform the actions described in FILE-ACTIONS. */
287static int
288__spawnix (pid_t * pid, const char *file,
289 const posix_spawn_file_actions_t * file_actions,
290 const posix_spawnattr_t * attrp, char *const argv[],
291 char *const envp[], int xflags,
292 int (*exec) (const char *, char *const *, char *const *))
293{
294 pid_t new_pid;
295 struct posix_spawn_args args;
296 int ec;
297
298 /* To avoid imposing hard limits on posix_spawn{p} the total number of
299 arguments is first calculated to allocate a mmap to hold all possible
300 values. */
301 ptrdiff_t argc = 0;
302 /* Linux allows at most max (0x7FFFFFFF, 1/4 stack size) arguments
303 to be used in a execve call. We limit to INT_MAX minus one due the
304 compatiblity code that may execute a shell script (maybe_script_execute)
305 where it will construct another argument list with an additional
306 argument. */
307 ptrdiff_t limit = INT_MAX - 1;
308 while (argv[argc++] != NULL)
309 if (argc == limit)
310 {
311 errno = E2BIG;
312 return errno;
313 }
314
315 int prot = (PROT_READ | PROT_WRITE
316 | ((GL (dl_stack_flags) & PF_X) ? PROT_EXEC : 0));
317
318 /* Add a slack area for child's stack. */
319 size_t argv_size = (argc * sizeof (void *)) + 512;
320 /* We need at least a few pages in case the compiler's stack checking is
321 enabled. In some configs, it is known to use at least 24KiB. We use
322 32KiB to be "safe" from anything the compiler might do. Besides, the
323 extra pages won't actually be allocated unless they get used. */
324 argv_size += (32 * 1024);
325 size_t stack_size = ALIGN_UP (argv_size, GLRO(dl_pagesize));
326 void *stack = __mmap (NULL, stack_size, prot,
327 MAP_PRIVATE | MAP_ANONYMOUS | MAP_STACK, -1, 0);
328 if (__glibc_unlikely (stack == MAP_FAILED))
329 return errno;
330
331 /* Disable asynchronous cancellation. */
332 int state;
333 __libc_ptf_call (__pthread_setcancelstate,
334 (PTHREAD_CANCEL_DISABLE, &state), 0);
335
336 /* Child must set args.err to something non-negative - we rely on
337 the parent and child sharing VM. */
338 args.err = 0;
339 args.file = file;
340 args.exec = exec;
341 args.fa = file_actions;
342 args.attr = attrp ? attrp : &(const posix_spawnattr_t) { 0 };
343 args.argv = argv;
344 args.argc = argc;
345 args.envp = envp;
346 args.xflags = xflags;
347
348 __libc_signal_block_all (&args.oldmask);
349
350 /* The clone flags used will create a new child that will run in the same
351 memory space (CLONE_VM) and the execution of calling thread will be
352 suspend until the child calls execve or _exit.
353
354 Also since the calling thread execution will be suspend, there is not
355 need for CLONE_SETTLS. Although parent and child share the same TLS
356 namespace, there will be no concurrent access for TLS variables (errno
357 for instance). */
358 new_pid = CLONE (__spawni_child, STACK (stack, stack_size), stack_size,
359 CLONE_VM | CLONE_VFORK | SIGCHLD, &args);
360
361 /* It needs to collect the case where the auxiliary process was created
362 but failed to execute the file (due either any preparation step or
363 for execve itself). */
364 if (new_pid > 0)
365 {
366 /* Also, it handles the unlikely case where the auxiliary process was
367 terminated before calling execve as if it was successfully. The
368 args.err is set to 0 as default and changed to a positive value
369 only in case of failure, so in case of premature termination
370 due a signal args.err will remain zeroed and it will be up to
371 caller to actually collect it. */
372 ec = args.err;
373 if (ec > 0)
374 /* There still an unlikely case where the child is cancelled after
375 setting args.err, due to a positive error value. Also there is
376 possible pid reuse race (where the kernel allocated the same pid
377 to an unrelated process). Unfortunately due synchronization
378 issues where the kernel might not have the process collected
379 the waitpid below can not use WNOHANG. */
380 __waitpid (new_pid, NULL, 0);
381 }
382 else
383 ec = -new_pid;
384
385 __munmap (stack, stack_size);
386
387 if ((ec == 0) && (pid != NULL))
388 *pid = new_pid;
389
390 __libc_signal_restore_set (&args.oldmask);
391
392 __libc_ptf_call (__pthread_setcancelstate, (state, NULL), 0);
393
394 return ec;
395}
396
397/* Spawn a new process executing PATH with the attributes describes in *ATTRP.
398 Before running the process perform the actions described in FILE-ACTIONS. */
399int
400__spawni (pid_t * pid, const char *file,
401 const posix_spawn_file_actions_t * acts,
402 const posix_spawnattr_t * attrp, char *const argv[],
403 char *const envp[], int xflags)
404{
405 return __spawnix (pid, file, acts, attrp, argv, envp, xflags,
406 xflags & SPAWN_XFLAGS_USE_PATH ? __execvpe : __execve);
407}
408