projects/linux·blob v2.6.31-rc8 · refs

1606 lines · 36.2 KiB

1/*
2 * linux/kernel/sys.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/module.h>
8#include <linux/mm.h>
9#include <linux/utsname.h>
10#include <linux/mman.h>
11#include <linux/smp_lock.h>
12#include <linux/notifier.h>
13#include <linux/reboot.h>
14#include <linux/prctl.h>
15#include <linux/highuid.h>
16#include <linux/fs.h>
17#include <linux/perf_counter.h>
18#include <linux/resource.h>
19#include <linux/kernel.h>
20#include <linux/kexec.h>
21#include <linux/workqueue.h>
22#include <linux/capability.h>
23#include <linux/device.h>
24#include <linux/key.h>
25#include <linux/times.h>
26#include <linux/posix-timers.h>
27#include <linux/security.h>
28#include <linux/dcookies.h>
29#include <linux/suspend.h>
30#include <linux/tty.h>
31#include <linux/signal.h>
32#include <linux/cn_proc.h>
33#include <linux/getcpu.h>
34#include <linux/task_io_accounting_ops.h>
35#include <linux/seccomp.h>
36#include <linux/cpu.h>
37#include <linux/ptrace.h>
38#include <linux/fs_struct.h>
39
40#include <linux/compat.h>
41#include <linux/syscalls.h>
42#include <linux/kprobes.h>
43#include <linux/user_namespace.h>
44
45#include <asm/uaccess.h>
46#include <asm/io.h>
47#include <asm/unistd.h>
48
49#ifndef SET_UNALIGN_CTL
50# define SET_UNALIGN_CTL(a,b) (-EINVAL)
51#endif
52#ifndef GET_UNALIGN_CTL
53# define GET_UNALIGN_CTL(a,b) (-EINVAL)
54#endif
55#ifndef SET_FPEMU_CTL
56# define SET_FPEMU_CTL(a,b) (-EINVAL)
57#endif
58#ifndef GET_FPEMU_CTL
59# define GET_FPEMU_CTL(a,b) (-EINVAL)
60#endif
61#ifndef SET_FPEXC_CTL
62# define SET_FPEXC_CTL(a,b) (-EINVAL)
63#endif
64#ifndef GET_FPEXC_CTL
65# define GET_FPEXC_CTL(a,b) (-EINVAL)
66#endif
67#ifndef GET_ENDIAN
68# define GET_ENDIAN(a,b) (-EINVAL)
69#endif
70#ifndef SET_ENDIAN
71# define SET_ENDIAN(a,b) (-EINVAL)
72#endif
73#ifndef GET_TSC_CTL
74# define GET_TSC_CTL(a) (-EINVAL)
75#endif
76#ifndef SET_TSC_CTL
77# define SET_TSC_CTL(a) (-EINVAL)
78#endif
79
80/*
81 * this is where the system-wide overflow UID and GID are defined, for
82 * architectures that now have 32-bit UID/GID but didn't in the past
83 */
84
85int overflowuid = DEFAULT_OVERFLOWUID;
86int overflowgid = DEFAULT_OVERFLOWGID;
87
88#ifdef CONFIG_UID16
89EXPORT_SYMBOL(overflowuid);
90EXPORT_SYMBOL(overflowgid);
91#endif
92
93/*
94 * the same as above, but for filesystems which can only store a 16-bit
95 * UID and GID. as such, this is needed on all architectures
96 */
97
98int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
99int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
100
101EXPORT_SYMBOL(fs_overflowuid);
102EXPORT_SYMBOL(fs_overflowgid);
103
104/*
105 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
106 */
107
108int C_A_D = 1;
109struct pid *cad_pid;
110EXPORT_SYMBOL(cad_pid);
111
112/*
113 * If set, this is used for preparing the system to power off.
114 */
115
116void (*pm_power_off_prepare)(void);
117
118/*
119 * set the priority of a task
120 * - the caller must hold the RCU read lock
121 */
122static int set_one_prio(struct task_struct *p, int niceval, int error)
123{
124 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
125 int no_nice;
126
127 if (pcred->uid != cred->euid &&
128 pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
129 error = -EPERM;
130 goto out;
131 }
132 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
133 error = -EACCES;
134 goto out;
135 }
136 no_nice = security_task_setnice(p, niceval);
137 if (no_nice) {
138 error = no_nice;
139 goto out;
140 }
141 if (error == -ESRCH)
142 error = 0;
143 set_user_nice(p, niceval);
144out:
145 return error;
146}
147
148SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
149{
150 struct task_struct *g, *p;
151 struct user_struct *user;
152 const struct cred *cred = current_cred();
153 int error = -EINVAL;
154 struct pid *pgrp;
155
156 if (which > PRIO_USER || which < PRIO_PROCESS)
157 goto out;
158
159 /* normalize: avoid signed division (rounding problems) */
160 error = -ESRCH;
161 if (niceval < -20)
162 niceval = -20;
163 if (niceval > 19)
164 niceval = 19;
165
166 read_lock(&tasklist_lock);
167 switch (which) {
168 case PRIO_PROCESS:
169 if (who)
170 p = find_task_by_vpid(who);
171 else
172 p = current;
173 if (p)
174 error = set_one_prio(p, niceval, error);
175 break;
176 case PRIO_PGRP:
177 if (who)
178 pgrp = find_vpid(who);
179 else
180 pgrp = task_pgrp(current);
181 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
182 error = set_one_prio(p, niceval, error);
183 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
184 break;
185 case PRIO_USER:
186 user = (struct user_struct *) cred->user;
187 if (!who)
188 who = cred->uid;
189 else if ((who != cred->uid) &&
190 !(user = find_user(who)))
191 goto out_unlock; /* No processes for this user */
192
193 do_each_thread(g, p)
194 if (__task_cred(p)->uid == who)
195 error = set_one_prio(p, niceval, error);
196 while_each_thread(g, p);
197 if (who != cred->uid)
198 free_uid(user); /* For find_user() */
199 break;
200 }
201out_unlock:
202 read_unlock(&tasklist_lock);
203out:
204 return error;
205}
206
207/*
208 * Ugh. To avoid negative return values, "getpriority()" will
209 * not return the normal nice-value, but a negated value that
210 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
211 * to stay compatible.
212 */
213SYSCALL_DEFINE2(getpriority, int, which, int, who)
214{
215 struct task_struct *g, *p;
216 struct user_struct *user;
217 const struct cred *cred = current_cred();
218 long niceval, retval = -ESRCH;
219 struct pid *pgrp;
220
221 if (which > PRIO_USER || which < PRIO_PROCESS)
222 return -EINVAL;
223
224 read_lock(&tasklist_lock);
225 switch (which) {
226 case PRIO_PROCESS:
227 if (who)
228 p = find_task_by_vpid(who);
229 else
230 p = current;
231 if (p) {
232 niceval = 20 - task_nice(p);
233 if (niceval > retval)
234 retval = niceval;
235 }
236 break;
237 case PRIO_PGRP:
238 if (who)
239 pgrp = find_vpid(who);
240 else
241 pgrp = task_pgrp(current);
242 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
243 niceval = 20 - task_nice(p);
244 if (niceval > retval)
245 retval = niceval;
246 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
247 break;
248 case PRIO_USER:
249 user = (struct user_struct *) cred->user;
250 if (!who)
251 who = cred->uid;
252 else if ((who != cred->uid) &&
253 !(user = find_user(who)))
254 goto out_unlock; /* No processes for this user */
255
256 do_each_thread(g, p)
257 if (__task_cred(p)->uid == who) {
258 niceval = 20 - task_nice(p);
259 if (niceval > retval)
260 retval = niceval;
261 }
262 while_each_thread(g, p);
263 if (who != cred->uid)
264 free_uid(user); /* for find_user() */
265 break;
266 }
267out_unlock:
268 read_unlock(&tasklist_lock);
269
270 return retval;
271}
272
273/**
274 * emergency_restart - reboot the system
275 *
276 * Without shutting down any hardware or taking any locks
277 * reboot the system. This is called when we know we are in
278 * trouble so this is our best effort to reboot. This is
279 * safe to call in interrupt context.
280 */
281void emergency_restart(void)
282{
283 machine_emergency_restart();
284}
285EXPORT_SYMBOL_GPL(emergency_restart);
286
287void kernel_restart_prepare(char *cmd)
288{
289 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
290 system_state = SYSTEM_RESTART;
291 device_shutdown();
292 sysdev_shutdown();
293}
294
295/**
296 * kernel_restart - reboot the system
297 * @cmd: pointer to buffer containing command to execute for restart
298 * or %NULL
299 *
300 * Shutdown everything and perform a clean reboot.
301 * This is not safe to call in interrupt context.
302 */
303void kernel_restart(char *cmd)
304{
305 kernel_restart_prepare(cmd);
306 if (!cmd)
307 printk(KERN_EMERG "Restarting system.\n");
308 else
309 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
310 machine_restart(cmd);
311}
312EXPORT_SYMBOL_GPL(kernel_restart);
313
314static void kernel_shutdown_prepare(enum system_states state)
315{
316 blocking_notifier_call_chain(&reboot_notifier_list,
317 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
318 system_state = state;
319 device_shutdown();
320}
321/**
322 * kernel_halt - halt the system
323 *
324 * Shutdown everything and perform a clean system halt.
325 */
326void kernel_halt(void)
327{
328 kernel_shutdown_prepare(SYSTEM_HALT);
329 sysdev_shutdown();
330 printk(KERN_EMERG "System halted.\n");
331 machine_halt();
332}
333
334EXPORT_SYMBOL_GPL(kernel_halt);
335
336/**
337 * kernel_power_off - power_off the system
338 *
339 * Shutdown everything and perform a clean system power_off.
340 */
341void kernel_power_off(void)
342{
343 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
344 if (pm_power_off_prepare)
345 pm_power_off_prepare();
346 disable_nonboot_cpus();
347 sysdev_shutdown();
348 printk(KERN_EMERG "Power down.\n");
349 machine_power_off();
350}
351EXPORT_SYMBOL_GPL(kernel_power_off);
352/*
353 * Reboot system call: for obvious reasons only root may call it,
354 * and even root needs to set up some magic numbers in the registers
355 * so that some mistake won't make this reboot the whole machine.
356 * You can also set the meaning of the ctrl-alt-del-key here.
357 *
358 * reboot doesn't sync: do that yourself before calling this.
359 */
360SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
361 void __user *, arg)
362{
363 char buffer[256];
364 int ret = 0;
365
366 /* We only trust the superuser with rebooting the system. */
367 if (!capable(CAP_SYS_BOOT))
368 return -EPERM;
369
370 /* For safety, we require "magic" arguments. */
371 if (magic1 != LINUX_REBOOT_MAGIC1 ||
372 (magic2 != LINUX_REBOOT_MAGIC2 &&
373 magic2 != LINUX_REBOOT_MAGIC2A &&
374 magic2 != LINUX_REBOOT_MAGIC2B &&
375 magic2 != LINUX_REBOOT_MAGIC2C))
376 return -EINVAL;
377
378 /* Instead of trying to make the power_off code look like
379 * halt when pm_power_off is not set do it the easy way.
380 */
381 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
382 cmd = LINUX_REBOOT_CMD_HALT;
383
384 lock_kernel();
385 switch (cmd) {
386 case LINUX_REBOOT_CMD_RESTART:
387 kernel_restart(NULL);
388 break;
389
390 case LINUX_REBOOT_CMD_CAD_ON:
391 C_A_D = 1;
392 break;
393
394 case LINUX_REBOOT_CMD_CAD_OFF:
395 C_A_D = 0;
396 break;
397
398 case LINUX_REBOOT_CMD_HALT:
399 kernel_halt();
400 unlock_kernel();
401 do_exit(0);
402 panic("cannot halt");
403
404 case LINUX_REBOOT_CMD_POWER_OFF:
405 kernel_power_off();
406 unlock_kernel();
407 do_exit(0);
408 break;
409
410 case LINUX_REBOOT_CMD_RESTART2:
411 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
412 unlock_kernel();
413 return -EFAULT;
414 }
415 buffer[sizeof(buffer) - 1] = '\0';
416
417 kernel_restart(buffer);
418 break;
419
420#ifdef CONFIG_KEXEC
421 case LINUX_REBOOT_CMD_KEXEC:
422 ret = kernel_kexec();
423 break;
424#endif
425
426#ifdef CONFIG_HIBERNATION
427 case LINUX_REBOOT_CMD_SW_SUSPEND:
428 ret = hibernate();
429 break;
430#endif
431
432 default:
433 ret = -EINVAL;
434 break;
435 }
436 unlock_kernel();
437 return ret;
438}
439
440static void deferred_cad(struct work_struct *dummy)
441{
442 kernel_restart(NULL);
443}
444
445/*
446 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
447 * As it's called within an interrupt, it may NOT sync: the only choice
448 * is whether to reboot at once, or just ignore the ctrl-alt-del.
449 */
450void ctrl_alt_del(void)
451{
452 static DECLARE_WORK(cad_work, deferred_cad);
453
454 if (C_A_D)
455 schedule_work(&cad_work);
456 else
457 kill_cad_pid(SIGINT, 1);
458}
459
460/*
461 * Unprivileged users may change the real gid to the effective gid
462 * or vice versa. (BSD-style)
463 *
464 * If you set the real gid at all, or set the effective gid to a value not
465 * equal to the real gid, then the saved gid is set to the new effective gid.
466 *
467 * This makes it possible for a setgid program to completely drop its
468 * privileges, which is often a useful assertion to make when you are doing
469 * a security audit over a program.
470 *
471 * The general idea is that a program which uses just setregid() will be
472 * 100% compatible with BSD. A program which uses just setgid() will be
473 * 100% compatible with POSIX with saved IDs.
474 *
475 * SMP: There are not races, the GIDs are checked only by filesystem
476 * operations (as far as semantic preservation is concerned).
477 */
478SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
479{
480 const struct cred *old;
481 struct cred *new;
482 int retval;
483
484 new = prepare_creds();
485 if (!new)
486 return -ENOMEM;
487 old = current_cred();
488
489 retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
490 if (retval)
491 goto error;
492
493 retval = -EPERM;
494 if (rgid != (gid_t) -1) {
495 if (old->gid == rgid ||
496 old->egid == rgid ||
497 capable(CAP_SETGID))
498 new->gid = rgid;
499 else
500 goto error;
501 }
502 if (egid != (gid_t) -1) {
503 if (old->gid == egid ||
504 old->egid == egid ||
505 old->sgid == egid ||
506 capable(CAP_SETGID))
507 new->egid = egid;
508 else
509 goto error;
510 }
511
512 if (rgid != (gid_t) -1 ||
513 (egid != (gid_t) -1 && egid != old->gid))
514 new->sgid = new->egid;
515 new->fsgid = new->egid;
516
517 return commit_creds(new);
518
519error:
520 abort_creds(new);
521 return retval;
522}
523
524/*
525 * setgid() is implemented like SysV w/ SAVED_IDS
526 *
527 * SMP: Same implicit races as above.
528 */
529SYSCALL_DEFINE1(setgid, gid_t, gid)
530{
531 const struct cred *old;
532 struct cred *new;
533 int retval;
534
535 new = prepare_creds();
536 if (!new)
537 return -ENOMEM;
538 old = current_cred();
539
540 retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
541 if (retval)
542 goto error;
543
544 retval = -EPERM;
545 if (capable(CAP_SETGID))
546 new->gid = new->egid = new->sgid = new->fsgid = gid;
547 else if (gid == old->gid || gid == old->sgid)
548 new->egid = new->fsgid = gid;
549 else
550 goto error;
551
552 return commit_creds(new);
553
554error:
555 abort_creds(new);
556 return retval;
557}
558
559/*
560 * change the user struct in a credentials set to match the new UID
561 */
562static int set_user(struct cred *new)
563{
564 struct user_struct *new_user;
565
566 new_user = alloc_uid(current_user_ns(), new->uid);
567 if (!new_user)
568 return -EAGAIN;
569
570 if (!task_can_switch_user(new_user, current)) {
571 free_uid(new_user);
572 return -EINVAL;
573 }
574
575 if (atomic_read(&new_user->processes) >=
576 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
577 new_user != INIT_USER) {
578 free_uid(new_user);
579 return -EAGAIN;
580 }
581
582 free_uid(new->user);
583 new->user = new_user;
584 return 0;
585}
586
587/*
588 * Unprivileged users may change the real uid to the effective uid
589 * or vice versa. (BSD-style)
590 *
591 * If you set the real uid at all, or set the effective uid to a value not
592 * equal to the real uid, then the saved uid is set to the new effective uid.
593 *
594 * This makes it possible for a setuid program to completely drop its
595 * privileges, which is often a useful assertion to make when you are doing
596 * a security audit over a program.
597 *
598 * The general idea is that a program which uses just setreuid() will be
599 * 100% compatible with BSD. A program which uses just setuid() will be
600 * 100% compatible with POSIX with saved IDs.
601 */
602SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
603{
604 const struct cred *old;
605 struct cred *new;
606 int retval;
607
608 new = prepare_creds();
609 if (!new)
610 return -ENOMEM;
611 old = current_cred();
612
613 retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
614 if (retval)
615 goto error;
616
617 retval = -EPERM;
618 if (ruid != (uid_t) -1) {
619 new->uid = ruid;
620 if (old->uid != ruid &&
621 old->euid != ruid &&
622 !capable(CAP_SETUID))
623 goto error;
624 }
625
626 if (euid != (uid_t) -1) {
627 new->euid = euid;
628 if (old->uid != euid &&
629 old->euid != euid &&
630 old->suid != euid &&
631 !capable(CAP_SETUID))
632 goto error;
633 }
634
635 if (new->uid != old->uid) {
636 retval = set_user(new);
637 if (retval < 0)
638 goto error;
639 }
640 if (ruid != (uid_t) -1 ||
641 (euid != (uid_t) -1 && euid != old->uid))
642 new->suid = new->euid;
643 new->fsuid = new->euid;
644
645 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
646 if (retval < 0)
647 goto error;
648
649 return commit_creds(new);
650
651error:
652 abort_creds(new);
653 return retval;
654}
655
656/*
657 * setuid() is implemented like SysV with SAVED_IDS
658 *
659 * Note that SAVED_ID's is deficient in that a setuid root program
660 * like sendmail, for example, cannot set its uid to be a normal
661 * user and then switch back, because if you're root, setuid() sets
662 * the saved uid too. If you don't like this, blame the bright people
663 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
664 * will allow a root program to temporarily drop privileges and be able to
665 * regain them by swapping the real and effective uid.
666 */
667SYSCALL_DEFINE1(setuid, uid_t, uid)
668{
669 const struct cred *old;
670 struct cred *new;
671 int retval;
672
673 new = prepare_creds();
674 if (!new)
675 return -ENOMEM;
676 old = current_cred();
677
678 retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
679 if (retval)
680 goto error;
681
682 retval = -EPERM;
683 if (capable(CAP_SETUID)) {
684 new->suid = new->uid = uid;
685 if (uid != old->uid) {
686 retval = set_user(new);
687 if (retval < 0)
688 goto error;
689 }
690 } else if (uid != old->uid && uid != new->suid) {
691 goto error;
692 }
693
694 new->fsuid = new->euid = uid;
695
696 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
697 if (retval < 0)
698 goto error;
699
700 return commit_creds(new);
701
702error:
703 abort_creds(new);
704 return retval;
705}
706
707
708/*
709 * This function implements a generic ability to update ruid, euid,
710 * and suid. This allows you to implement the 4.4 compatible seteuid().
711 */
712SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
713{
714 const struct cred *old;
715 struct cred *new;
716 int retval;
717
718 new = prepare_creds();
719 if (!new)
720 return -ENOMEM;
721
722 retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
723 if (retval)
724 goto error;
725 old = current_cred();
726
727 retval = -EPERM;
728 if (!capable(CAP_SETUID)) {
729 if (ruid != (uid_t) -1 && ruid != old->uid &&
730 ruid != old->euid && ruid != old->suid)
731 goto error;
732 if (euid != (uid_t) -1 && euid != old->uid &&
733 euid != old->euid && euid != old->suid)
734 goto error;
735 if (suid != (uid_t) -1 && suid != old->uid &&
736 suid != old->euid && suid != old->suid)
737 goto error;
738 }
739
740 if (ruid != (uid_t) -1) {
741 new->uid = ruid;
742 if (ruid != old->uid) {
743 retval = set_user(new);
744 if (retval < 0)
745 goto error;
746 }
747 }
748 if (euid != (uid_t) -1)
749 new->euid = euid;
750 if (suid != (uid_t) -1)
751 new->suid = suid;
752 new->fsuid = new->euid;
753
754 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
755 if (retval < 0)
756 goto error;
757
758 return commit_creds(new);
759
760error:
761 abort_creds(new);
762 return retval;
763}
764
765SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
766{
767 const struct cred *cred = current_cred();
768 int retval;
769
770 if (!(retval = put_user(cred->uid, ruid)) &&
771 !(retval = put_user(cred->euid, euid)))
772 retval = put_user(cred->suid, suid);
773
774 return retval;
775}
776
777/*
778 * Same as above, but for rgid, egid, sgid.
779 */
780SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
781{
782 const struct cred *old;
783 struct cred *new;
784 int retval;
785
786 new = prepare_creds();
787 if (!new)
788 return -ENOMEM;
789 old = current_cred();
790
791 retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
792 if (retval)
793 goto error;
794
795 retval = -EPERM;
796 if (!capable(CAP_SETGID)) {
797 if (rgid != (gid_t) -1 && rgid != old->gid &&
798 rgid != old->egid && rgid != old->sgid)
799 goto error;
800 if (egid != (gid_t) -1 && egid != old->gid &&
801 egid != old->egid && egid != old->sgid)
802 goto error;
803 if (sgid != (gid_t) -1 && sgid != old->gid &&
804 sgid != old->egid && sgid != old->sgid)
805 goto error;
806 }
807
808 if (rgid != (gid_t) -1)
809 new->gid = rgid;
810 if (egid != (gid_t) -1)
811 new->egid = egid;
812 if (sgid != (gid_t) -1)
813 new->sgid = sgid;
814 new->fsgid = new->egid;
815
816 return commit_creds(new);
817
818error:
819 abort_creds(new);
820 return retval;
821}
822
823SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
824{
825 const struct cred *cred = current_cred();
826 int retval;
827
828 if (!(retval = put_user(cred->gid, rgid)) &&
829 !(retval = put_user(cred->egid, egid)))
830 retval = put_user(cred->sgid, sgid);
831
832 return retval;
833}
834
835
836/*
837 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
838 * is used for "access()" and for the NFS daemon (letting nfsd stay at
839 * whatever uid it wants to). It normally shadows "euid", except when
840 * explicitly set by setfsuid() or for access..
841 */
842SYSCALL_DEFINE1(setfsuid, uid_t, uid)
843{
844 const struct cred *old;
845 struct cred *new;
846 uid_t old_fsuid;
847
848 new = prepare_creds();
849 if (!new)
850 return current_fsuid();
851 old = current_cred();
852 old_fsuid = old->fsuid;
853
854 if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
855 goto error;
856
857 if (uid == old->uid || uid == old->euid ||
858 uid == old->suid || uid == old->fsuid ||
859 capable(CAP_SETUID)) {
860 if (uid != old_fsuid) {
861 new->fsuid = uid;
862 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
863 goto change_okay;
864 }
865 }
866
867error:
868 abort_creds(new);
869 return old_fsuid;
870
871change_okay:
872 commit_creds(new);
873 return old_fsuid;
874}
875
876/*
877 * Samma på svenska..
878 */
879SYSCALL_DEFINE1(setfsgid, gid_t, gid)
880{
881 const struct cred *old;
882 struct cred *new;
883 gid_t old_fsgid;
884
885 new = prepare_creds();
886 if (!new)
887 return current_fsgid();
888 old = current_cred();
889 old_fsgid = old->fsgid;
890
891 if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
892 goto error;
893
894 if (gid == old->gid || gid == old->egid ||
895 gid == old->sgid || gid == old->fsgid ||
896 capable(CAP_SETGID)) {
897 if (gid != old_fsgid) {
898 new->fsgid = gid;
899 goto change_okay;
900 }
901 }
902
903error:
904 abort_creds(new);
905 return old_fsgid;
906
907change_okay:
908 commit_creds(new);
909 return old_fsgid;
910}
911
912void do_sys_times(struct tms *tms)
913{
914 struct task_cputime cputime;
915 cputime_t cutime, cstime;
916
917 thread_group_cputime(current, &cputime);
918 spin_lock_irq(&current->sighand->siglock);
919 cutime = current->signal->cutime;
920 cstime = current->signal->cstime;
921 spin_unlock_irq(&current->sighand->siglock);
922 tms->tms_utime = cputime_to_clock_t(cputime.utime);
923 tms->tms_stime = cputime_to_clock_t(cputime.stime);
924 tms->tms_cutime = cputime_to_clock_t(cutime);
925 tms->tms_cstime = cputime_to_clock_t(cstime);
926}
927
928SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
929{
930 if (tbuf) {
931 struct tms tmp;
932
933 do_sys_times(&tmp);
934 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
935 return -EFAULT;
936 }
937 force_successful_syscall_return();
938 return (long) jiffies_64_to_clock_t(get_jiffies_64());
939}
940
941/*
942 * This needs some heavy checking ...
943 * I just haven't the stomach for it. I also don't fully
944 * understand sessions/pgrp etc. Let somebody who does explain it.
945 *
946 * OK, I think I have the protection semantics right.... this is really
947 * only important on a multi-user system anyway, to make sure one user
948 * can't send a signal to a process owned by another. -TYT, 12/12/91
949 *
950 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
951 * LBT 04.03.94
952 */
953SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
954{
955 struct task_struct *p;
956 struct task_struct *group_leader = current->group_leader;
957 struct pid *pgrp;
958 int err;
959
960 if (!pid)
961 pid = task_pid_vnr(group_leader);
962 if (!pgid)
963 pgid = pid;
964 if (pgid < 0)
965 return -EINVAL;
966
967 /* From this point forward we keep holding onto the tasklist lock
968 * so that our parent does not change from under us. -DaveM
969 */
970 write_lock_irq(&tasklist_lock);
971
972 err = -ESRCH;
973 p = find_task_by_vpid(pid);
974 if (!p)
975 goto out;
976
977 err = -EINVAL;
978 if (!thread_group_leader(p))
979 goto out;
980
981 if (same_thread_group(p->real_parent, group_leader)) {
982 err = -EPERM;
983 if (task_session(p) != task_session(group_leader))
984 goto out;
985 err = -EACCES;
986 if (p->did_exec)
987 goto out;
988 } else {
989 err = -ESRCH;
990 if (p != group_leader)
991 goto out;
992 }
993
994 err = -EPERM;
995 if (p->signal->leader)
996 goto out;
997
998 pgrp = task_pid(p);
999 if (pgid != pid) {
1000 struct task_struct *g;
1001
1002 pgrp = find_vpid(pgid);
1003 g = pid_task(pgrp, PIDTYPE_PGID);
1004 if (!g || task_session(g) != task_session(group_leader))
1005 goto out;
1006 }
1007
1008 err = security_task_setpgid(p, pgid);
1009 if (err)
1010 goto out;
1011
1012 if (task_pgrp(p) != pgrp)
1013 change_pid(p, PIDTYPE_PGID, pgrp);
1014
1015 err = 0;
1016out:
1017 /* All paths lead to here, thus we are safe. -DaveM */
1018 write_unlock_irq(&tasklist_lock);
1019 return err;
1020}
1021
1022SYSCALL_DEFINE1(getpgid, pid_t, pid)
1023{
1024 struct task_struct *p;
1025 struct pid *grp;
1026 int retval;
1027
1028 rcu_read_lock();
1029 if (!pid)
1030 grp = task_pgrp(current);
1031 else {
1032 retval = -ESRCH;
1033 p = find_task_by_vpid(pid);
1034 if (!p)
1035 goto out;
1036 grp = task_pgrp(p);
1037 if (!grp)
1038 goto out;
1039
1040 retval = security_task_getpgid(p);
1041 if (retval)
1042 goto out;
1043 }
1044 retval = pid_vnr(grp);
1045out:
1046 rcu_read_unlock();
1047 return retval;
1048}
1049
1050#ifdef __ARCH_WANT_SYS_GETPGRP
1051
1052SYSCALL_DEFINE0(getpgrp)
1053{
1054 return sys_getpgid(0);
1055}
1056
1057#endif
1058
1059SYSCALL_DEFINE1(getsid, pid_t, pid)
1060{
1061 struct task_struct *p;
1062 struct pid *sid;
1063 int retval;
1064
1065 rcu_read_lock();
1066 if (!pid)
1067 sid = task_session(current);
1068 else {
1069 retval = -ESRCH;
1070 p = find_task_by_vpid(pid);
1071 if (!p)
1072 goto out;
1073 sid = task_session(p);
1074 if (!sid)
1075 goto out;
1076
1077 retval = security_task_getsid(p);
1078 if (retval)
1079 goto out;
1080 }
1081 retval = pid_vnr(sid);
1082out:
1083 rcu_read_unlock();
1084 return retval;
1085}
1086
1087SYSCALL_DEFINE0(setsid)
1088{
1089 struct task_struct *group_leader = current->group_leader;
1090 struct pid *sid = task_pid(group_leader);
1091 pid_t session = pid_vnr(sid);
1092 int err = -EPERM;
1093
1094 write_lock_irq(&tasklist_lock);
1095 /* Fail if I am already a session leader */
1096 if (group_leader->signal->leader)
1097 goto out;
1098
1099 /* Fail if a process group id already exists that equals the
1100 * proposed session id.
1101 */
1102 if (pid_task(sid, PIDTYPE_PGID))
1103 goto out;
1104
1105 group_leader->signal->leader = 1;
1106 __set_special_pids(sid);
1107
1108 proc_clear_tty(group_leader);
1109
1110 err = session;
1111out:
1112 write_unlock_irq(&tasklist_lock);
1113 return err;
1114}
1115
1116DECLARE_RWSEM(uts_sem);
1117
1118SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1119{
1120 int errno = 0;
1121
1122 down_read(&uts_sem);
1123 if (copy_to_user(name, utsname(), sizeof *name))
1124 errno = -EFAULT;
1125 up_read(&uts_sem);
1126 return errno;
1127}
1128
1129SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1130{
1131 int errno;
1132 char tmp[__NEW_UTS_LEN];
1133
1134 if (!capable(CAP_SYS_ADMIN))
1135 return -EPERM;
1136 if (len < 0 || len > __NEW_UTS_LEN)
1137 return -EINVAL;
1138 down_write(&uts_sem);
1139 errno = -EFAULT;
1140 if (!copy_from_user(tmp, name, len)) {
1141 struct new_utsname *u = utsname();
1142
1143 memcpy(u->nodename, tmp, len);
1144 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1145 errno = 0;
1146 }
1147 up_write(&uts_sem);
1148 return errno;
1149}
1150
1151#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1152
1153SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1154{
1155 int i, errno;
1156 struct new_utsname *u;
1157
1158 if (len < 0)
1159 return -EINVAL;
1160 down_read(&uts_sem);
1161 u = utsname();
1162 i = 1 + strlen(u->nodename);
1163 if (i > len)
1164 i = len;
1165 errno = 0;
1166 if (copy_to_user(name, u->nodename, i))
1167 errno = -EFAULT;
1168 up_read(&uts_sem);
1169 return errno;
1170}
1171
1172#endif
1173
1174/*
1175 * Only setdomainname; getdomainname can be implemented by calling
1176 * uname()
1177 */
1178SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1179{
1180 int errno;
1181 char tmp[__NEW_UTS_LEN];
1182
1183 if (!capable(CAP_SYS_ADMIN))
1184 return -EPERM;
1185 if (len < 0 || len > __NEW_UTS_LEN)
1186 return -EINVAL;
1187
1188 down_write(&uts_sem);
1189 errno = -EFAULT;
1190 if (!copy_from_user(tmp, name, len)) {
1191 struct new_utsname *u = utsname();
1192
1193 memcpy(u->domainname, tmp, len);
1194 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1195 errno = 0;
1196 }
1197 up_write(&uts_sem);
1198 return errno;
1199}
1200
1201SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1202{
1203 if (resource >= RLIM_NLIMITS)
1204 return -EINVAL;
1205 else {
1206 struct rlimit value;
1207 task_lock(current->group_leader);
1208 value = current->signal->rlim[resource];
1209 task_unlock(current->group_leader);
1210 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1211 }
1212}
1213
1214#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1215
1216/*
1217 * Back compatibility for getrlimit. Needed for some apps.
1218 */
1219
1220SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1221 struct rlimit __user *, rlim)
1222{
1223 struct rlimit x;
1224 if (resource >= RLIM_NLIMITS)
1225 return -EINVAL;
1226
1227 task_lock(current->group_leader);
1228 x = current->signal->rlim[resource];
1229 task_unlock(current->group_leader);
1230 if (x.rlim_cur > 0x7FFFFFFF)
1231 x.rlim_cur = 0x7FFFFFFF;
1232 if (x.rlim_max > 0x7FFFFFFF)
1233 x.rlim_max = 0x7FFFFFFF;
1234 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1235}
1236
1237#endif
1238
1239SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1240{
1241 struct rlimit new_rlim, *old_rlim;
1242 int retval;
1243
1244 if (resource >= RLIM_NLIMITS)
1245 return -EINVAL;
1246 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1247 return -EFAULT;
1248 if (new_rlim.rlim_cur > new_rlim.rlim_max)
1249 return -EINVAL;
1250 old_rlim = current->signal->rlim + resource;
1251 if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1252 !capable(CAP_SYS_RESOURCE))
1253 return -EPERM;
1254 if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1255 return -EPERM;
1256
1257 retval = security_task_setrlimit(resource, &new_rlim);
1258 if (retval)
1259 return retval;
1260
1261 if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1262 /*
1263 * The caller is asking for an immediate RLIMIT_CPU
1264 * expiry. But we use the zero value to mean "it was
1265 * never set". So let's cheat and make it one second
1266 * instead
1267 */
1268 new_rlim.rlim_cur = 1;
1269 }
1270
1271 task_lock(current->group_leader);
1272 *old_rlim = new_rlim;
1273 task_unlock(current->group_leader);
1274
1275 if (resource != RLIMIT_CPU)
1276 goto out;
1277
1278 /*
1279 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1280 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1281 * very long-standing error, and fixing it now risks breakage of
1282 * applications, so we live with it
1283 */
1284 if (new_rlim.rlim_cur == RLIM_INFINITY)
1285 goto out;
1286
1287 update_rlimit_cpu(new_rlim.rlim_cur);
1288out:
1289 return 0;
1290}
1291
1292/*
1293 * It would make sense to put struct rusage in the task_struct,
1294 * except that would make the task_struct be *really big*. After
1295 * task_struct gets moved into malloc'ed memory, it would
1296 * make sense to do this. It will make moving the rest of the information
1297 * a lot simpler! (Which we're not doing right now because we're not
1298 * measuring them yet).
1299 *
1300 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1301 * races with threads incrementing their own counters. But since word
1302 * reads are atomic, we either get new values or old values and we don't
1303 * care which for the sums. We always take the siglock to protect reading
1304 * the c* fields from p->signal from races with exit.c updating those
1305 * fields when reaping, so a sample either gets all the additions of a
1306 * given child after it's reaped, or none so this sample is before reaping.
1307 *
1308 * Locking:
1309 * We need to take the siglock for CHILDEREN, SELF and BOTH
1310 * for the cases current multithreaded, non-current single threaded
1311 * non-current multithreaded. Thread traversal is now safe with
1312 * the siglock held.
1313 * Strictly speaking, we donot need to take the siglock if we are current and
1314 * single threaded, as no one else can take our signal_struct away, no one
1315 * else can reap the children to update signal->c* counters, and no one else
1316 * can race with the signal-> fields. If we do not take any lock, the
1317 * signal-> fields could be read out of order while another thread was just
1318 * exiting. So we should place a read memory barrier when we avoid the lock.
1319 * On the writer side, write memory barrier is implied in __exit_signal
1320 * as __exit_signal releases the siglock spinlock after updating the signal->
1321 * fields. But we don't do this yet to keep things simple.
1322 *
1323 */
1324
1325static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1326{
1327 r->ru_nvcsw += t->nvcsw;
1328 r->ru_nivcsw += t->nivcsw;
1329 r->ru_minflt += t->min_flt;
1330 r->ru_majflt += t->maj_flt;
1331 r->ru_inblock += task_io_get_inblock(t);
1332 r->ru_oublock += task_io_get_oublock(t);
1333}
1334
1335static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1336{
1337 struct task_struct *t;
1338 unsigned long flags;
1339 cputime_t utime, stime;
1340 struct task_cputime cputime;
1341
1342 memset((char *) r, 0, sizeof *r);
1343 utime = stime = cputime_zero;
1344
1345 if (who == RUSAGE_THREAD) {
1346 utime = task_utime(current);
1347 stime = task_stime(current);
1348 accumulate_thread_rusage(p, r);
1349 goto out;
1350 }
1351
1352 if (!lock_task_sighand(p, &flags))
1353 return;
1354
1355 switch (who) {
1356 case RUSAGE_BOTH:
1357 case RUSAGE_CHILDREN:
1358 utime = p->signal->cutime;
1359 stime = p->signal->cstime;
1360 r->ru_nvcsw = p->signal->cnvcsw;
1361 r->ru_nivcsw = p->signal->cnivcsw;
1362 r->ru_minflt = p->signal->cmin_flt;
1363 r->ru_majflt = p->signal->cmaj_flt;
1364 r->ru_inblock = p->signal->cinblock;
1365 r->ru_oublock = p->signal->coublock;
1366
1367 if (who == RUSAGE_CHILDREN)
1368 break;
1369
1370 case RUSAGE_SELF:
1371 thread_group_cputime(p, &cputime);
1372 utime = cputime_add(utime, cputime.utime);
1373 stime = cputime_add(stime, cputime.stime);
1374 r->ru_nvcsw += p->signal->nvcsw;
1375 r->ru_nivcsw += p->signal->nivcsw;
1376 r->ru_minflt += p->signal->min_flt;
1377 r->ru_majflt += p->signal->maj_flt;
1378 r->ru_inblock += p->signal->inblock;
1379 r->ru_oublock += p->signal->oublock;
1380 t = p;
1381 do {
1382 accumulate_thread_rusage(t, r);
1383 t = next_thread(t);
1384 } while (t != p);
1385 break;
1386
1387 default:
1388 BUG();
1389 }
1390 unlock_task_sighand(p, &flags);
1391
1392out:
1393 cputime_to_timeval(utime, &r->ru_utime);
1394 cputime_to_timeval(stime, &r->ru_stime);
1395}
1396
1397int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1398{
1399 struct rusage r;
1400 k_getrusage(p, who, &r);
1401 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1402}
1403
1404SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1405{
1406 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1407 who != RUSAGE_THREAD)
1408 return -EINVAL;
1409 return getrusage(current, who, ru);
1410}
1411
1412SYSCALL_DEFINE1(umask, int, mask)
1413{
1414 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1415 return mask;
1416}
1417
1418SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1419 unsigned long, arg4, unsigned long, arg5)
1420{
1421 struct task_struct *me = current;
1422 unsigned char comm[sizeof(me->comm)];
1423 long error;
1424
1425 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1426 if (error != -ENOSYS)
1427 return error;
1428
1429 error = 0;
1430 switch (option) {
1431 case PR_SET_PDEATHSIG:
1432 if (!valid_signal(arg2)) {
1433 error = -EINVAL;
1434 break;
1435 }
1436 me->pdeath_signal = arg2;
1437 error = 0;
1438 break;
1439 case PR_GET_PDEATHSIG:
1440 error = put_user(me->pdeath_signal, (int __user *)arg2);
1441 break;
1442 case PR_GET_DUMPABLE:
1443 error = get_dumpable(me->mm);
1444 break;
1445 case PR_SET_DUMPABLE:
1446 if (arg2 < 0 || arg2 > 1) {
1447 error = -EINVAL;
1448 break;
1449 }
1450 set_dumpable(me->mm, arg2);
1451 error = 0;
1452 break;
1453
1454 case PR_SET_UNALIGN:
1455 error = SET_UNALIGN_CTL(me, arg2);
1456 break;
1457 case PR_GET_UNALIGN:
1458 error = GET_UNALIGN_CTL(me, arg2);
1459 break;
1460 case PR_SET_FPEMU:
1461 error = SET_FPEMU_CTL(me, arg2);
1462 break;
1463 case PR_GET_FPEMU:
1464 error = GET_FPEMU_CTL(me, arg2);
1465 break;
1466 case PR_SET_FPEXC:
1467 error = SET_FPEXC_CTL(me, arg2);
1468 break;
1469 case PR_GET_FPEXC:
1470 error = GET_FPEXC_CTL(me, arg2);
1471 break;
1472 case PR_GET_TIMING:
1473 error = PR_TIMING_STATISTICAL;
1474 break;
1475 case PR_SET_TIMING:
1476 if (arg2 != PR_TIMING_STATISTICAL)
1477 error = -EINVAL;
1478 else
1479 error = 0;
1480 break;
1481
1482 case PR_SET_NAME:
1483 comm[sizeof(me->comm)-1] = 0;
1484 if (strncpy_from_user(comm, (char __user *)arg2,
1485 sizeof(me->comm) - 1) < 0)
1486 return -EFAULT;
1487 set_task_comm(me, comm);
1488 return 0;
1489 case PR_GET_NAME:
1490 get_task_comm(comm, me);
1491 if (copy_to_user((char __user *)arg2, comm,
1492 sizeof(comm)))
1493 return -EFAULT;
1494 return 0;
1495 case PR_GET_ENDIAN:
1496 error = GET_ENDIAN(me, arg2);
1497 break;
1498 case PR_SET_ENDIAN:
1499 error = SET_ENDIAN(me, arg2);
1500 break;
1501
1502 case PR_GET_SECCOMP:
1503 error = prctl_get_seccomp();
1504 break;
1505 case PR_SET_SECCOMP:
1506 error = prctl_set_seccomp(arg2);
1507 break;
1508 case PR_GET_TSC:
1509 error = GET_TSC_CTL(arg2);
1510 break;
1511 case PR_SET_TSC:
1512 error = SET_TSC_CTL(arg2);
1513 break;
1514 case PR_TASK_PERF_COUNTERS_DISABLE:
1515 error = perf_counter_task_disable();
1516 break;
1517 case PR_TASK_PERF_COUNTERS_ENABLE:
1518 error = perf_counter_task_enable();
1519 break;
1520 case PR_GET_TIMERSLACK:
1521 error = current->timer_slack_ns;
1522 break;
1523 case PR_SET_TIMERSLACK:
1524 if (arg2 <= 0)
1525 current->timer_slack_ns =
1526 current->default_timer_slack_ns;
1527 else
1528 current->timer_slack_ns = arg2;
1529 error = 0;
1530 break;
1531 default:
1532 error = -EINVAL;
1533 break;
1534 }
1535 return error;
1536}
1537
1538SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1539 struct getcpu_cache __user *, unused)
1540{
1541 int err = 0;
1542 int cpu = raw_smp_processor_id();
1543 if (cpup)
1544 err |= put_user(cpu, cpup);
1545 if (nodep)
1546 err |= put_user(cpu_to_node(cpu), nodep);
1547 return err ? -EFAULT : 0;
1548}
1549
1550char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1551
1552static void argv_cleanup(char **argv, char **envp)
1553{
1554 argv_free(argv);
1555}
1556
1557/**
1558 * orderly_poweroff - Trigger an orderly system poweroff
1559 * @force: force poweroff if command execution fails
1560 *
1561 * This may be called from any context to trigger a system shutdown.
1562 * If the orderly shutdown fails, it will force an immediate shutdown.
1563 */
1564int orderly_poweroff(bool force)
1565{
1566 int argc;
1567 char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1568 static char *envp[] = {
1569 "HOME=/",
1570 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1571 NULL
1572 };
1573 int ret = -ENOMEM;
1574 struct subprocess_info *info;
1575
1576 if (argv == NULL) {
1577 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1578 __func__, poweroff_cmd);
1579 goto out;
1580 }
1581
1582 info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1583 if (info == NULL) {
1584 argv_free(argv);
1585 goto out;
1586 }
1587
1588 call_usermodehelper_setcleanup(info, argv_cleanup);
1589
1590 ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1591
1592 out:
1593 if (ret && force) {
1594 printk(KERN_WARNING "Failed to start orderly shutdown: "
1595 "forcing the issue\n");
1596
1597 /* I guess this should try to kick off some daemon to
1598 sync and poweroff asap. Or not even bother syncing
1599 if we're doing an emergency shutdown? */
1600 emergency_sync();
1601 kernel_power_off();
1602 }
1603
1604 return ret;
1605}
1606EXPORT_SYMBOL_GPL(orderly_poweroff);