1/* POSIX spawn interface. Linux version.
2 Copyright (C) 2016-2018 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 <libc-pointer-arith.h>
33#include <ldsodefs.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
127 /* The child must ensure that no signal handler are enabled because it shared
128 memory with parent, so the signal disposition must be either SIG_DFL or
129 SIG_IGN. It does by iterating over all signals and although it could
130 possibly be more optimized (by tracking which signal potentially have a
131 signal handler), it might requires system specific solutions (since the
132 sigset_t data type can be very different on different architectures). */
133 struct sigaction sa;
134 memset (&sa, '\0', sizeof (sa));
135
136 sigset_t hset;
137 __sigprocmask (SIG_BLOCK, 0, &hset);
138 for (int sig = 1; sig < _NSIG; ++sig)
139 {
140 if ((attr->__flags & POSIX_SPAWN_SETSIGDEF)
141 && sigismember (&attr->__sd, sig))
142 {
143 sa.sa_handler = SIG_DFL;
144 }
145 else if (sigismember (&hset, sig))
146 {
147 if (__is_internal_signal (sig))
148 sa.sa_handler = SIG_IGN;
149 else
150 {
151 __libc_sigaction (sig, 0, &sa);
152 if (sa.sa_handler == SIG_IGN)
153 continue;
154 sa.sa_handler = SIG_DFL;
155 }
156 }
157 else
158 continue;
159
160 __libc_sigaction (sig, &sa, 0);
161 }
162
163#ifdef _POSIX_PRIORITY_SCHEDULING
164 /* Set the scheduling algorithm and parameters. */
165 if ((attr->__flags & (POSIX_SPAWN_SETSCHEDPARAM | POSIX_SPAWN_SETSCHEDULER))
166 == POSIX_SPAWN_SETSCHEDPARAM)
167 {
168 if (__sched_setparam (0, &attr->__sp) == -1)
169 goto fail;
170 }
171 else if ((attr->__flags & POSIX_SPAWN_SETSCHEDULER) != 0)
172 {
173 if (__sched_setscheduler (0, attr->__policy, &attr->__sp) == -1)
174 goto fail;
175 }
176#endif
177
178 if ((attr->__flags & POSIX_SPAWN_SETSID) != 0
179 && __setsid () < 0)
180 goto fail;
181
182 /* Set the process group ID. */
183 if ((attr->__flags & POSIX_SPAWN_SETPGROUP) != 0
184 && __setpgid (0, attr->__pgrp) != 0)
185 goto fail;
186
187 /* Set the effective user and group IDs. */
188 if ((attr->__flags & POSIX_SPAWN_RESETIDS) != 0
189 && (local_seteuid (__getuid ()) != 0
190 || local_setegid (__getgid ()) != 0))
191 goto fail;
192
193 /* Execute the file actions. */
194 if (file_actions != 0)
195 {
196 int cnt;
197 struct rlimit64 fdlimit;
198 bool have_fdlimit = false;
199
200 for (cnt = 0; cnt < file_actions->__used; ++cnt)
201 {
202 struct __spawn_action *action = &file_actions->__actions[cnt];
203
204 switch (action->tag)
205 {
206 case spawn_do_close:
207 if (__close_nocancel (action->action.close_action.fd) != 0)
208 {
209 if (!have_fdlimit)
210 {
211 __getrlimit64 (RLIMIT_NOFILE, &fdlimit);
212 have_fdlimit = true;
213 }
214
215 /* Signal errors only for file descriptors out of range. */
216 if (action->action.close_action.fd < 0
217 || action->action.close_action.fd >= fdlimit.rlim_cur)
218 goto fail;
219 }
220 break;
221
222 case spawn_do_open:
223 {
224 /* POSIX states that if fildes was already an open file descriptor,
225 it shall be closed before the new file is opened. This avoid
226 pontential issues when posix_spawn plus addopen action is called
227 with the process already at maximum number of file descriptor
228 opened and also for multiple actions on single-open special
229 paths (like /dev/watchdog). */
230 __close_nocancel (action->action.open_action.fd);
231
232 int ret = __open_nocancel (action->action.open_action.path,
233 action->action.
234 open_action.oflag | O_LARGEFILE,
235 action->action.open_action.mode);
236
237 if (ret == -1)
238 goto fail;
239
240 int new_fd = ret;
241
242 /* Make sure the desired file descriptor is used. */
243 if (ret != action->action.open_action.fd)
244 {
245 if (__dup2 (new_fd, action->action.open_action.fd)
246 != action->action.open_action.fd)
247 goto fail;
248
249 if (__close_nocancel (new_fd) != 0)
250 goto fail;
251 }
252 }
253 break;
254
255 case spawn_do_dup2:
256 if (__dup2 (action->action.dup2_action.fd,
257 action->action.dup2_action.newfd)
258 != action->action.dup2_action.newfd)
259 goto fail;
260 break;
261 }
262 }
263 }
264
265 /* Set the initial signal mask of the child if POSIX_SPAWN_SETSIGMASK
266 is set, otherwise restore the previous one. */
267 __sigprocmask (SIG_SETMASK, (attr->__flags & POSIX_SPAWN_SETSIGMASK)
268 ? &attr->__ss : &args->oldmask, 0);
269
270 args->exec (args->file, args->argv, args->envp);
271
272 /* This is compatibility function required to enable posix_spawn run
273 script without shebang definition for older posix_spawn versions
274 (2.15). */
275 maybe_script_execute (args);
276
277fail:
278 /* errno should have an appropriate non-zero value; otherwise,
279 there's a bug in glibc or the kernel. For lack of an error code
280 (EINTERNALBUG) describing that, use ECHILD. Another option would
281 be to set args->err to some negative sentinel and have the parent
282 abort(), but that seems needlessly harsh. */
283 args->err = errno ? : ECHILD;
284 _exit (SPAWN_ERROR);
285}
286
287/* Spawn a new process executing PATH with the attributes describes in *ATTRP.
288 Before running the process perform the actions described in FILE-ACTIONS. */
289static int
290__spawnix (pid_t * pid, const char *file,
291 const posix_spawn_file_actions_t * file_actions,
292 const posix_spawnattr_t * attrp, char *const argv[],
293 char *const envp[], int xflags,
294 int (*exec) (const char *, char *const *, char *const *))
295{
296 pid_t new_pid;
297 struct posix_spawn_args args;
298 int ec;
299
300 /* To avoid imposing hard limits on posix_spawn{p} the total number of
301 arguments is first calculated to allocate a mmap to hold all possible
302 values. */
303 ptrdiff_t argc = 0;
304 /* Linux allows at most max (0x7FFFFFFF, 1/4 stack size) arguments
305 to be used in a execve call. We limit to INT_MAX minus one due the
306 compatiblity code that may execute a shell script (maybe_script_execute)
307 where it will construct another argument list with an additional
308 argument. */
309 ptrdiff_t limit = INT_MAX - 1;
310 while (argv[argc++] != NULL)
311 if (argc == limit)
312 {
313 errno = E2BIG;
314 return errno;
315 }
316
317 int prot = (PROT_READ | PROT_WRITE
318 | ((GL (dl_stack_flags) & PF_X) ? PROT_EXEC : 0));
319
320 /* Add a slack area for child's stack. */
321 size_t argv_size = (argc * sizeof (void *)) + 512;
322 /* We need at least a few pages in case the compiler's stack checking is
323 enabled. In some configs, it is known to use at least 24KiB. We use
324 32KiB to be "safe" from anything the compiler might do. Besides, the
325 extra pages won't actually be allocated unless they get used. */
326 argv_size += (32 * 1024);
327 size_t stack_size = ALIGN_UP (argv_size, GLRO(dl_pagesize));
328 void *stack = __mmap (NULL, stack_size, prot,
329 MAP_PRIVATE | MAP_ANONYMOUS | MAP_STACK, -1, 0);
330 if (__glibc_unlikely (stack == MAP_FAILED))
331 return errno;
332
333 /* Disable asynchronous cancellation. */
334 int state;
335 __libc_ptf_call (__pthread_setcancelstate,
336 (PTHREAD_CANCEL_DISABLE, &state), 0);
337
338 /* Child must set args.err to something non-negative - we rely on
339 the parent and child sharing VM. */
340 args.err = 0;
341 args.file = file;
342 args.exec = exec;
343 args.fa = file_actions;
344 args.attr = attrp ? attrp : &(const posix_spawnattr_t) { 0 };
345 args.argv = argv;
346 args.argc = argc;
347 args.envp = envp;
348 args.xflags = xflags;
349
350 __libc_signal_block_all (&args.oldmask);
351
352 /* The clone flags used will create a new child that will run in the same
353 memory space (CLONE_VM) and the execution of calling thread will be
354 suspend until the child calls execve or _exit.
355
356 Also since the calling thread execution will be suspend, there is not
357 need for CLONE_SETTLS. Although parent and child share the same TLS
358 namespace, there will be no concurrent access for TLS variables (errno
359 for instance). */
360 new_pid = CLONE (__spawni_child, STACK (stack, stack_size), stack_size,
361 CLONE_VM | CLONE_VFORK | SIGCHLD, &args);
362
363 /* It needs to collect the case where the auxiliary process was created
364 but failed to execute the file (due either any preparation step or
365 for execve itself). */
366 if (new_pid > 0)
367 {
368 /* Also, it handles the unlikely case where the auxiliary process was
369 terminated before calling execve as if it was successfully. The
370 args.err is set to 0 as default and changed to a positive value
371 only in case of failure, so in case of premature termination
372 due a signal args.err will remain zeroed and it will be up to
373 caller to actually collect it. */
374 ec = args.err;
375 if (ec > 0)
376 /* There still an unlikely case where the child is cancelled after
377 setting args.err, due to a positive error value. Also there is
378 possible pid reuse race (where the kernel allocated the same pid
379 to an unrelated process). Unfortunately due synchronization
380 issues where the kernel might not have the process collected
381 the waitpid below can not use WNOHANG. */
382 __waitpid (new_pid, NULL, 0);
383 }
384 else
385 ec = -new_pid;
386
387 __munmap (stack, stack_size);
388
389 if ((ec == 0) && (pid != NULL))
390 *pid = new_pid;
391
392 __libc_signal_restore_set (&args.oldmask);
393
394 __libc_ptf_call (__pthread_setcancelstate, (state, NULL), 0);
395
396 return ec;
397}
398
399/* Spawn a new process executing PATH with the attributes describes in *ATTRP.
400 Before running the process perform the actions described in FILE-ACTIONS. */
401int
402__spawni (pid_t * pid, const char *file,
403 const posix_spawn_file_actions_t * acts,
404 const posix_spawnattr_t * attrp, char *const argv[],
405 char *const envp[], int xflags)
406{
407 /* It uses __execvpex to avoid run ENOEXEC in non compatibility mode (it
408 will be handled by maybe_script_execute). */
409 return __spawnix (pid, file, acts, attrp, argv, envp, xflags,
410 xflags & SPAWN_XFLAGS_USE_PATH ? __execvpex :__execve);
411}
412