projects/linux·blob v3.9-rc8 · refs

2247 lines · 52.3 KiB

1/*
2 * linux/kernel/sys.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/export.h>
8#include <linux/mm.h>
9#include <linux/utsname.h>
10#include <linux/mman.h>
11#include <linux/reboot.h>
12#include <linux/prctl.h>
13#include <linux/highuid.h>
14#include <linux/fs.h>
15#include <linux/kmod.h>
16#include <linux/perf_event.h>
17#include <linux/resource.h>
18#include <linux/kernel.h>
19#include <linux/kexec.h>
20#include <linux/workqueue.h>
21#include <linux/capability.h>
22#include <linux/device.h>
23#include <linux/key.h>
24#include <linux/times.h>
25#include <linux/posix-timers.h>
26#include <linux/security.h>
27#include <linux/dcookies.h>
28#include <linux/suspend.h>
29#include <linux/tty.h>
30#include <linux/signal.h>
31#include <linux/cn_proc.h>
32#include <linux/getcpu.h>
33#include <linux/task_io_accounting_ops.h>
34#include <linux/seccomp.h>
35#include <linux/cpu.h>
36#include <linux/personality.h>
37#include <linux/ptrace.h>
38#include <linux/fs_struct.h>
39#include <linux/file.h>
40#include <linux/mount.h>
41#include <linux/gfp.h>
42#include <linux/syscore_ops.h>
43#include <linux/version.h>
44#include <linux/ctype.h>
45
46#include <linux/compat.h>
47#include <linux/syscalls.h>
48#include <linux/kprobes.h>
49#include <linux/user_namespace.h>
50#include <linux/binfmts.h>
51
52#include <linux/kmsg_dump.h>
53/* Move somewhere else to avoid recompiling? */
54#include <generated/utsrelease.h>
55
56#include <asm/uaccess.h>
57#include <asm/io.h>
58#include <asm/unistd.h>
59
60#ifndef SET_UNALIGN_CTL
61# define SET_UNALIGN_CTL(a,b) (-EINVAL)
62#endif
63#ifndef GET_UNALIGN_CTL
64# define GET_UNALIGN_CTL(a,b) (-EINVAL)
65#endif
66#ifndef SET_FPEMU_CTL
67# define SET_FPEMU_CTL(a,b) (-EINVAL)
68#endif
69#ifndef GET_FPEMU_CTL
70# define GET_FPEMU_CTL(a,b) (-EINVAL)
71#endif
72#ifndef SET_FPEXC_CTL
73# define SET_FPEXC_CTL(a,b) (-EINVAL)
74#endif
75#ifndef GET_FPEXC_CTL
76# define GET_FPEXC_CTL(a,b) (-EINVAL)
77#endif
78#ifndef GET_ENDIAN
79# define GET_ENDIAN(a,b) (-EINVAL)
80#endif
81#ifndef SET_ENDIAN
82# define SET_ENDIAN(a,b) (-EINVAL)
83#endif
84#ifndef GET_TSC_CTL
85# define GET_TSC_CTL(a) (-EINVAL)
86#endif
87#ifndef SET_TSC_CTL
88# define SET_TSC_CTL(a) (-EINVAL)
89#endif
90
91/*
92 * this is where the system-wide overflow UID and GID are defined, for
93 * architectures that now have 32-bit UID/GID but didn't in the past
94 */
95
96int overflowuid = DEFAULT_OVERFLOWUID;
97int overflowgid = DEFAULT_OVERFLOWGID;
98
99EXPORT_SYMBOL(overflowuid);
100EXPORT_SYMBOL(overflowgid);
101
102/*
103 * the same as above, but for filesystems which can only store a 16-bit
104 * UID and GID. as such, this is needed on all architectures
105 */
106
107int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
108int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
109
110EXPORT_SYMBOL(fs_overflowuid);
111EXPORT_SYMBOL(fs_overflowgid);
112
113/*
114 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
115 */
116
117int C_A_D = 1;
118struct pid *cad_pid;
119EXPORT_SYMBOL(cad_pid);
120
121/*
122 * If set, this is used for preparing the system to power off.
123 */
124
125void (*pm_power_off_prepare)(void);
126
127/*
128 * Returns true if current's euid is same as p's uid or euid,
129 * or has CAP_SYS_NICE to p's user_ns.
130 *
131 * Called with rcu_read_lock, creds are safe
132 */
133static bool set_one_prio_perm(struct task_struct *p)
134{
135 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
136
137 if (uid_eq(pcred->uid, cred->euid) ||
138 uid_eq(pcred->euid, cred->euid))
139 return true;
140 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
141 return true;
142 return false;
143}
144
145/*
146 * set the priority of a task
147 * - the caller must hold the RCU read lock
148 */
149static int set_one_prio(struct task_struct *p, int niceval, int error)
150{
151 int no_nice;
152
153 if (!set_one_prio_perm(p)) {
154 error = -EPERM;
155 goto out;
156 }
157 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
158 error = -EACCES;
159 goto out;
160 }
161 no_nice = security_task_setnice(p, niceval);
162 if (no_nice) {
163 error = no_nice;
164 goto out;
165 }
166 if (error == -ESRCH)
167 error = 0;
168 set_user_nice(p, niceval);
169out:
170 return error;
171}
172
173SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
174{
175 struct task_struct *g, *p;
176 struct user_struct *user;
177 const struct cred *cred = current_cred();
178 int error = -EINVAL;
179 struct pid *pgrp;
180 kuid_t uid;
181
182 if (which > PRIO_USER || which < PRIO_PROCESS)
183 goto out;
184
185 /* normalize: avoid signed division (rounding problems) */
186 error = -ESRCH;
187 if (niceval < -20)
188 niceval = -20;
189 if (niceval > 19)
190 niceval = 19;
191
192 rcu_read_lock();
193 read_lock(&tasklist_lock);
194 switch (which) {
195 case PRIO_PROCESS:
196 if (who)
197 p = find_task_by_vpid(who);
198 else
199 p = current;
200 if (p)
201 error = set_one_prio(p, niceval, error);
202 break;
203 case PRIO_PGRP:
204 if (who)
205 pgrp = find_vpid(who);
206 else
207 pgrp = task_pgrp(current);
208 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
209 error = set_one_prio(p, niceval, error);
210 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
211 break;
212 case PRIO_USER:
213 uid = make_kuid(cred->user_ns, who);
214 user = cred->user;
215 if (!who)
216 uid = cred->uid;
217 else if (!uid_eq(uid, cred->uid) &&
218 !(user = find_user(uid)))
219 goto out_unlock; /* No processes for this user */
220
221 do_each_thread(g, p) {
222 if (uid_eq(task_uid(p), uid))
223 error = set_one_prio(p, niceval, error);
224 } while_each_thread(g, p);
225 if (!uid_eq(uid, cred->uid))
226 free_uid(user); /* For find_user() */
227 break;
228 }
229out_unlock:
230 read_unlock(&tasklist_lock);
231 rcu_read_unlock();
232out:
233 return error;
234}
235
236/*
237 * Ugh. To avoid negative return values, "getpriority()" will
238 * not return the normal nice-value, but a negated value that
239 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
240 * to stay compatible.
241 */
242SYSCALL_DEFINE2(getpriority, int, which, int, who)
243{
244 struct task_struct *g, *p;
245 struct user_struct *user;
246 const struct cred *cred = current_cred();
247 long niceval, retval = -ESRCH;
248 struct pid *pgrp;
249 kuid_t uid;
250
251 if (which > PRIO_USER || which < PRIO_PROCESS)
252 return -EINVAL;
253
254 rcu_read_lock();
255 read_lock(&tasklist_lock);
256 switch (which) {
257 case PRIO_PROCESS:
258 if (who)
259 p = find_task_by_vpid(who);
260 else
261 p = current;
262 if (p) {
263 niceval = 20 - task_nice(p);
264 if (niceval > retval)
265 retval = niceval;
266 }
267 break;
268 case PRIO_PGRP:
269 if (who)
270 pgrp = find_vpid(who);
271 else
272 pgrp = task_pgrp(current);
273 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
274 niceval = 20 - task_nice(p);
275 if (niceval > retval)
276 retval = niceval;
277 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
278 break;
279 case PRIO_USER:
280 uid = make_kuid(cred->user_ns, who);
281 user = cred->user;
282 if (!who)
283 uid = cred->uid;
284 else if (!uid_eq(uid, cred->uid) &&
285 !(user = find_user(uid)))
286 goto out_unlock; /* No processes for this user */
287
288 do_each_thread(g, p) {
289 if (uid_eq(task_uid(p), uid)) {
290 niceval = 20 - task_nice(p);
291 if (niceval > retval)
292 retval = niceval;
293 }
294 } while_each_thread(g, p);
295 if (!uid_eq(uid, cred->uid))
296 free_uid(user); /* for find_user() */
297 break;
298 }
299out_unlock:
300 read_unlock(&tasklist_lock);
301 rcu_read_unlock();
302
303 return retval;
304}
305
306/**
307 * emergency_restart - reboot the system
308 *
309 * Without shutting down any hardware or taking any locks
310 * reboot the system. This is called when we know we are in
311 * trouble so this is our best effort to reboot. This is
312 * safe to call in interrupt context.
313 */
314void emergency_restart(void)
315{
316 kmsg_dump(KMSG_DUMP_EMERG);
317 machine_emergency_restart();
318}
319EXPORT_SYMBOL_GPL(emergency_restart);
320
321void kernel_restart_prepare(char *cmd)
322{
323 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
324 system_state = SYSTEM_RESTART;
325 usermodehelper_disable();
326 device_shutdown();
327}
328
329/**
330 * register_reboot_notifier - Register function to be called at reboot time
331 * @nb: Info about notifier function to be called
332 *
333 * Registers a function with the list of functions
334 * to be called at reboot time.
335 *
336 * Currently always returns zero, as blocking_notifier_chain_register()
337 * always returns zero.
338 */
339int register_reboot_notifier(struct notifier_block *nb)
340{
341 return blocking_notifier_chain_register(&reboot_notifier_list, nb);
342}
343EXPORT_SYMBOL(register_reboot_notifier);
344
345/**
346 * unregister_reboot_notifier - Unregister previously registered reboot notifier
347 * @nb: Hook to be unregistered
348 *
349 * Unregisters a previously registered reboot
350 * notifier function.
351 *
352 * Returns zero on success, or %-ENOENT on failure.
353 */
354int unregister_reboot_notifier(struct notifier_block *nb)
355{
356 return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
357}
358EXPORT_SYMBOL(unregister_reboot_notifier);
359
360/**
361 * kernel_restart - reboot the system
362 * @cmd: pointer to buffer containing command to execute for restart
363 * or %NULL
364 *
365 * Shutdown everything and perform a clean reboot.
366 * This is not safe to call in interrupt context.
367 */
368void kernel_restart(char *cmd)
369{
370 kernel_restart_prepare(cmd);
371 disable_nonboot_cpus();
372 syscore_shutdown();
373 if (!cmd)
374 printk(KERN_EMERG "Restarting system.\n");
375 else
376 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
377 kmsg_dump(KMSG_DUMP_RESTART);
378 machine_restart(cmd);
379}
380EXPORT_SYMBOL_GPL(kernel_restart);
381
382static void kernel_shutdown_prepare(enum system_states state)
383{
384 blocking_notifier_call_chain(&reboot_notifier_list,
385 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
386 system_state = state;
387 usermodehelper_disable();
388 device_shutdown();
389}
390/**
391 * kernel_halt - halt the system
392 *
393 * Shutdown everything and perform a clean system halt.
394 */
395void kernel_halt(void)
396{
397 kernel_shutdown_prepare(SYSTEM_HALT);
398 disable_nonboot_cpus();
399 syscore_shutdown();
400 printk(KERN_EMERG "System halted.\n");
401 kmsg_dump(KMSG_DUMP_HALT);
402 machine_halt();
403}
404
405EXPORT_SYMBOL_GPL(kernel_halt);
406
407/**
408 * kernel_power_off - power_off the system
409 *
410 * Shutdown everything and perform a clean system power_off.
411 */
412void kernel_power_off(void)
413{
414 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
415 if (pm_power_off_prepare)
416 pm_power_off_prepare();
417 disable_nonboot_cpus();
418 syscore_shutdown();
419 printk(KERN_EMERG "Power down.\n");
420 kmsg_dump(KMSG_DUMP_POWEROFF);
421 machine_power_off();
422}
423EXPORT_SYMBOL_GPL(kernel_power_off);
424
425static DEFINE_MUTEX(reboot_mutex);
426
427/*
428 * Reboot system call: for obvious reasons only root may call it,
429 * and even root needs to set up some magic numbers in the registers
430 * so that some mistake won't make this reboot the whole machine.
431 * You can also set the meaning of the ctrl-alt-del-key here.
432 *
433 * reboot doesn't sync: do that yourself before calling this.
434 */
435SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
436 void __user *, arg)
437{
438 struct pid_namespace *pid_ns = task_active_pid_ns(current);
439 char buffer[256];
440 int ret = 0;
441
442 /* We only trust the superuser with rebooting the system. */
443 if (!ns_capable(pid_ns->user_ns, CAP_SYS_BOOT))
444 return -EPERM;
445
446 /* For safety, we require "magic" arguments. */
447 if (magic1 != LINUX_REBOOT_MAGIC1 ||
448 (magic2 != LINUX_REBOOT_MAGIC2 &&
449 magic2 != LINUX_REBOOT_MAGIC2A &&
450 magic2 != LINUX_REBOOT_MAGIC2B &&
451 magic2 != LINUX_REBOOT_MAGIC2C))
452 return -EINVAL;
453
454 /*
455 * If pid namespaces are enabled and the current task is in a child
456 * pid_namespace, the command is handled by reboot_pid_ns() which will
457 * call do_exit().
458 */
459 ret = reboot_pid_ns(pid_ns, cmd);
460 if (ret)
461 return ret;
462
463 /* Instead of trying to make the power_off code look like
464 * halt when pm_power_off is not set do it the easy way.
465 */
466 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
467 cmd = LINUX_REBOOT_CMD_HALT;
468
469 mutex_lock(&reboot_mutex);
470 switch (cmd) {
471 case LINUX_REBOOT_CMD_RESTART:
472 kernel_restart(NULL);
473 break;
474
475 case LINUX_REBOOT_CMD_CAD_ON:
476 C_A_D = 1;
477 break;
478
479 case LINUX_REBOOT_CMD_CAD_OFF:
480 C_A_D = 0;
481 break;
482
483 case LINUX_REBOOT_CMD_HALT:
484 kernel_halt();
485 do_exit(0);
486 panic("cannot halt");
487
488 case LINUX_REBOOT_CMD_POWER_OFF:
489 kernel_power_off();
490 do_exit(0);
491 break;
492
493 case LINUX_REBOOT_CMD_RESTART2:
494 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
495 ret = -EFAULT;
496 break;
497 }
498 buffer[sizeof(buffer) - 1] = '\0';
499
500 kernel_restart(buffer);
501 break;
502
503#ifdef CONFIG_KEXEC
504 case LINUX_REBOOT_CMD_KEXEC:
505 ret = kernel_kexec();
506 break;
507#endif
508
509#ifdef CONFIG_HIBERNATION
510 case LINUX_REBOOT_CMD_SW_SUSPEND:
511 ret = hibernate();
512 break;
513#endif
514
515 default:
516 ret = -EINVAL;
517 break;
518 }
519 mutex_unlock(&reboot_mutex);
520 return ret;
521}
522
523static void deferred_cad(struct work_struct *dummy)
524{
525 kernel_restart(NULL);
526}
527
528/*
529 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
530 * As it's called within an interrupt, it may NOT sync: the only choice
531 * is whether to reboot at once, or just ignore the ctrl-alt-del.
532 */
533void ctrl_alt_del(void)
534{
535 static DECLARE_WORK(cad_work, deferred_cad);
536
537 if (C_A_D)
538 schedule_work(&cad_work);
539 else
540 kill_cad_pid(SIGINT, 1);
541}
542
543/*
544 * Unprivileged users may change the real gid to the effective gid
545 * or vice versa. (BSD-style)
546 *
547 * If you set the real gid at all, or set the effective gid to a value not
548 * equal to the real gid, then the saved gid is set to the new effective gid.
549 *
550 * This makes it possible for a setgid program to completely drop its
551 * privileges, which is often a useful assertion to make when you are doing
552 * a security audit over a program.
553 *
554 * The general idea is that a program which uses just setregid() will be
555 * 100% compatible with BSD. A program which uses just setgid() will be
556 * 100% compatible with POSIX with saved IDs.
557 *
558 * SMP: There are not races, the GIDs are checked only by filesystem
559 * operations (as far as semantic preservation is concerned).
560 */
561SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
562{
563 struct user_namespace *ns = current_user_ns();
564 const struct cred *old;
565 struct cred *new;
566 int retval;
567 kgid_t krgid, kegid;
568
569 krgid = make_kgid(ns, rgid);
570 kegid = make_kgid(ns, egid);
571
572 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
573 return -EINVAL;
574 if ((egid != (gid_t) -1) && !gid_valid(kegid))
575 return -EINVAL;
576
577 new = prepare_creds();
578 if (!new)
579 return -ENOMEM;
580 old = current_cred();
581
582 retval = -EPERM;
583 if (rgid != (gid_t) -1) {
584 if (gid_eq(old->gid, krgid) ||
585 gid_eq(old->egid, krgid) ||
586 nsown_capable(CAP_SETGID))
587 new->gid = krgid;
588 else
589 goto error;
590 }
591 if (egid != (gid_t) -1) {
592 if (gid_eq(old->gid, kegid) ||
593 gid_eq(old->egid, kegid) ||
594 gid_eq(old->sgid, kegid) ||
595 nsown_capable(CAP_SETGID))
596 new->egid = kegid;
597 else
598 goto error;
599 }
600
601 if (rgid != (gid_t) -1 ||
602 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
603 new->sgid = new->egid;
604 new->fsgid = new->egid;
605
606 return commit_creds(new);
607
608error:
609 abort_creds(new);
610 return retval;
611}
612
613/*
614 * setgid() is implemented like SysV w/ SAVED_IDS
615 *
616 * SMP: Same implicit races as above.
617 */
618SYSCALL_DEFINE1(setgid, gid_t, gid)
619{
620 struct user_namespace *ns = current_user_ns();
621 const struct cred *old;
622 struct cred *new;
623 int retval;
624 kgid_t kgid;
625
626 kgid = make_kgid(ns, gid);
627 if (!gid_valid(kgid))
628 return -EINVAL;
629
630 new = prepare_creds();
631 if (!new)
632 return -ENOMEM;
633 old = current_cred();
634
635 retval = -EPERM;
636 if (nsown_capable(CAP_SETGID))
637 new->gid = new->egid = new->sgid = new->fsgid = kgid;
638 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
639 new->egid = new->fsgid = kgid;
640 else
641 goto error;
642
643 return commit_creds(new);
644
645error:
646 abort_creds(new);
647 return retval;
648}
649
650/*
651 * change the user struct in a credentials set to match the new UID
652 */
653static int set_user(struct cred *new)
654{
655 struct user_struct *new_user;
656
657 new_user = alloc_uid(new->uid);
658 if (!new_user)
659 return -EAGAIN;
660
661 /*
662 * We don't fail in case of NPROC limit excess here because too many
663 * poorly written programs don't check set*uid() return code, assuming
664 * it never fails if called by root. We may still enforce NPROC limit
665 * for programs doing set*uid()+execve() by harmlessly deferring the
666 * failure to the execve() stage.
667 */
668 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
669 new_user != INIT_USER)
670 current->flags |= PF_NPROC_EXCEEDED;
671 else
672 current->flags &= ~PF_NPROC_EXCEEDED;
673
674 free_uid(new->user);
675 new->user = new_user;
676 return 0;
677}
678
679/*
680 * Unprivileged users may change the real uid to the effective uid
681 * or vice versa. (BSD-style)
682 *
683 * If you set the real uid at all, or set the effective uid to a value not
684 * equal to the real uid, then the saved uid is set to the new effective uid.
685 *
686 * This makes it possible for a setuid program to completely drop its
687 * privileges, which is often a useful assertion to make when you are doing
688 * a security audit over a program.
689 *
690 * The general idea is that a program which uses just setreuid() will be
691 * 100% compatible with BSD. A program which uses just setuid() will be
692 * 100% compatible with POSIX with saved IDs.
693 */
694SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
695{
696 struct user_namespace *ns = current_user_ns();
697 const struct cred *old;
698 struct cred *new;
699 int retval;
700 kuid_t kruid, keuid;
701
702 kruid = make_kuid(ns, ruid);
703 keuid = make_kuid(ns, euid);
704
705 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
706 return -EINVAL;
707 if ((euid != (uid_t) -1) && !uid_valid(keuid))
708 return -EINVAL;
709
710 new = prepare_creds();
711 if (!new)
712 return -ENOMEM;
713 old = current_cred();
714
715 retval = -EPERM;
716 if (ruid != (uid_t) -1) {
717 new->uid = kruid;
718 if (!uid_eq(old->uid, kruid) &&
719 !uid_eq(old->euid, kruid) &&
720 !nsown_capable(CAP_SETUID))
721 goto error;
722 }
723
724 if (euid != (uid_t) -1) {
725 new->euid = keuid;
726 if (!uid_eq(old->uid, keuid) &&
727 !uid_eq(old->euid, keuid) &&
728 !uid_eq(old->suid, keuid) &&
729 !nsown_capable(CAP_SETUID))
730 goto error;
731 }
732
733 if (!uid_eq(new->uid, old->uid)) {
734 retval = set_user(new);
735 if (retval < 0)
736 goto error;
737 }
738 if (ruid != (uid_t) -1 ||
739 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
740 new->suid = new->euid;
741 new->fsuid = new->euid;
742
743 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
744 if (retval < 0)
745 goto error;
746
747 return commit_creds(new);
748
749error:
750 abort_creds(new);
751 return retval;
752}
753
754/*
755 * setuid() is implemented like SysV with SAVED_IDS
756 *
757 * Note that SAVED_ID's is deficient in that a setuid root program
758 * like sendmail, for example, cannot set its uid to be a normal
759 * user and then switch back, because if you're root, setuid() sets
760 * the saved uid too. If you don't like this, blame the bright people
761 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
762 * will allow a root program to temporarily drop privileges and be able to
763 * regain them by swapping the real and effective uid.
764 */
765SYSCALL_DEFINE1(setuid, uid_t, uid)
766{
767 struct user_namespace *ns = current_user_ns();
768 const struct cred *old;
769 struct cred *new;
770 int retval;
771 kuid_t kuid;
772
773 kuid = make_kuid(ns, uid);
774 if (!uid_valid(kuid))
775 return -EINVAL;
776
777 new = prepare_creds();
778 if (!new)
779 return -ENOMEM;
780 old = current_cred();
781
782 retval = -EPERM;
783 if (nsown_capable(CAP_SETUID)) {
784 new->suid = new->uid = kuid;
785 if (!uid_eq(kuid, old->uid)) {
786 retval = set_user(new);
787 if (retval < 0)
788 goto error;
789 }
790 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
791 goto error;
792 }
793
794 new->fsuid = new->euid = kuid;
795
796 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
797 if (retval < 0)
798 goto error;
799
800 return commit_creds(new);
801
802error:
803 abort_creds(new);
804 return retval;
805}
806
807
808/*
809 * This function implements a generic ability to update ruid, euid,
810 * and suid. This allows you to implement the 4.4 compatible seteuid().
811 */
812SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
813{
814 struct user_namespace *ns = current_user_ns();
815 const struct cred *old;
816 struct cred *new;
817 int retval;
818 kuid_t kruid, keuid, ksuid;
819
820 kruid = make_kuid(ns, ruid);
821 keuid = make_kuid(ns, euid);
822 ksuid = make_kuid(ns, suid);
823
824 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
825 return -EINVAL;
826
827 if ((euid != (uid_t) -1) && !uid_valid(keuid))
828 return -EINVAL;
829
830 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
831 return -EINVAL;
832
833 new = prepare_creds();
834 if (!new)
835 return -ENOMEM;
836
837 old = current_cred();
838
839 retval = -EPERM;
840 if (!nsown_capable(CAP_SETUID)) {
841 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
842 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
843 goto error;
844 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
845 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
846 goto error;
847 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
848 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
849 goto error;
850 }
851
852 if (ruid != (uid_t) -1) {
853 new->uid = kruid;
854 if (!uid_eq(kruid, old->uid)) {
855 retval = set_user(new);
856 if (retval < 0)
857 goto error;
858 }
859 }
860 if (euid != (uid_t) -1)
861 new->euid = keuid;
862 if (suid != (uid_t) -1)
863 new->suid = ksuid;
864 new->fsuid = new->euid;
865
866 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
867 if (retval < 0)
868 goto error;
869
870 return commit_creds(new);
871
872error:
873 abort_creds(new);
874 return retval;
875}
876
877SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
878{
879 const struct cred *cred = current_cred();
880 int retval;
881 uid_t ruid, euid, suid;
882
883 ruid = from_kuid_munged(cred->user_ns, cred->uid);
884 euid = from_kuid_munged(cred->user_ns, cred->euid);
885 suid = from_kuid_munged(cred->user_ns, cred->suid);
886
887 if (!(retval = put_user(ruid, ruidp)) &&
888 !(retval = put_user(euid, euidp)))
889 retval = put_user(suid, suidp);
890
891 return retval;
892}
893
894/*
895 * Same as above, but for rgid, egid, sgid.
896 */
897SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
898{
899 struct user_namespace *ns = current_user_ns();
900 const struct cred *old;
901 struct cred *new;
902 int retval;
903 kgid_t krgid, kegid, ksgid;
904
905 krgid = make_kgid(ns, rgid);
906 kegid = make_kgid(ns, egid);
907 ksgid = make_kgid(ns, sgid);
908
909 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
910 return -EINVAL;
911 if ((egid != (gid_t) -1) && !gid_valid(kegid))
912 return -EINVAL;
913 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
914 return -EINVAL;
915
916 new = prepare_creds();
917 if (!new)
918 return -ENOMEM;
919 old = current_cred();
920
921 retval = -EPERM;
922 if (!nsown_capable(CAP_SETGID)) {
923 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
924 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
925 goto error;
926 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
927 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
928 goto error;
929 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
930 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
931 goto error;
932 }
933
934 if (rgid != (gid_t) -1)
935 new->gid = krgid;
936 if (egid != (gid_t) -1)
937 new->egid = kegid;
938 if (sgid != (gid_t) -1)
939 new->sgid = ksgid;
940 new->fsgid = new->egid;
941
942 return commit_creds(new);
943
944error:
945 abort_creds(new);
946 return retval;
947}
948
949SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
950{
951 const struct cred *cred = current_cred();
952 int retval;
953 gid_t rgid, egid, sgid;
954
955 rgid = from_kgid_munged(cred->user_ns, cred->gid);
956 egid = from_kgid_munged(cred->user_ns, cred->egid);
957 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
958
959 if (!(retval = put_user(rgid, rgidp)) &&
960 !(retval = put_user(egid, egidp)))
961 retval = put_user(sgid, sgidp);
962
963 return retval;
964}
965
966
967/*
968 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
969 * is used for "access()" and for the NFS daemon (letting nfsd stay at
970 * whatever uid it wants to). It normally shadows "euid", except when
971 * explicitly set by setfsuid() or for access..
972 */
973SYSCALL_DEFINE1(setfsuid, uid_t, uid)
974{
975 const struct cred *old;
976 struct cred *new;
977 uid_t old_fsuid;
978 kuid_t kuid;
979
980 old = current_cred();
981 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
982
983 kuid = make_kuid(old->user_ns, uid);
984 if (!uid_valid(kuid))
985 return old_fsuid;
986
987 new = prepare_creds();
988 if (!new)
989 return old_fsuid;
990
991 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
992 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
993 nsown_capable(CAP_SETUID)) {
994 if (!uid_eq(kuid, old->fsuid)) {
995 new->fsuid = kuid;
996 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
997 goto change_okay;
998 }
999 }
1000
1001 abort_creds(new);
1002 return old_fsuid;
1003
1004change_okay:
1005 commit_creds(new);
1006 return old_fsuid;
1007}
1008
1009/*
1010 * Samma på svenska..
1011 */
1012SYSCALL_DEFINE1(setfsgid, gid_t, gid)
1013{
1014 const struct cred *old;
1015 struct cred *new;
1016 gid_t old_fsgid;
1017 kgid_t kgid;
1018
1019 old = current_cred();
1020 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
1021
1022 kgid = make_kgid(old->user_ns, gid);
1023 if (!gid_valid(kgid))
1024 return old_fsgid;
1025
1026 new = prepare_creds();
1027 if (!new)
1028 return old_fsgid;
1029
1030 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
1031 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
1032 nsown_capable(CAP_SETGID)) {
1033 if (!gid_eq(kgid, old->fsgid)) {
1034 new->fsgid = kgid;
1035 goto change_okay;
1036 }
1037 }
1038
1039 abort_creds(new);
1040 return old_fsgid;
1041
1042change_okay:
1043 commit_creds(new);
1044 return old_fsgid;
1045}
1046
1047void do_sys_times(struct tms *tms)
1048{
1049 cputime_t tgutime, tgstime, cutime, cstime;
1050
1051 spin_lock_irq(&current->sighand->siglock);
1052 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1053 cutime = current->signal->cutime;
1054 cstime = current->signal->cstime;
1055 spin_unlock_irq(&current->sighand->siglock);
1056 tms->tms_utime = cputime_to_clock_t(tgutime);
1057 tms->tms_stime = cputime_to_clock_t(tgstime);
1058 tms->tms_cutime = cputime_to_clock_t(cutime);
1059 tms->tms_cstime = cputime_to_clock_t(cstime);
1060}
1061
1062SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1063{
1064 if (tbuf) {
1065 struct tms tmp;
1066
1067 do_sys_times(&tmp);
1068 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1069 return -EFAULT;
1070 }
1071 force_successful_syscall_return();
1072 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1073}
1074
1075/*
1076 * This needs some heavy checking ...
1077 * I just haven't the stomach for it. I also don't fully
1078 * understand sessions/pgrp etc. Let somebody who does explain it.
1079 *
1080 * OK, I think I have the protection semantics right.... this is really
1081 * only important on a multi-user system anyway, to make sure one user
1082 * can't send a signal to a process owned by another. -TYT, 12/12/91
1083 *
1084 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1085 * LBT 04.03.94
1086 */
1087SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1088{
1089 struct task_struct *p;
1090 struct task_struct *group_leader = current->group_leader;
1091 struct pid *pgrp;
1092 int err;
1093
1094 if (!pid)
1095 pid = task_pid_vnr(group_leader);
1096 if (!pgid)
1097 pgid = pid;
1098 if (pgid < 0)
1099 return -EINVAL;
1100 rcu_read_lock();
1101
1102 /* From this point forward we keep holding onto the tasklist lock
1103 * so that our parent does not change from under us. -DaveM
1104 */
1105 write_lock_irq(&tasklist_lock);
1106
1107 err = -ESRCH;
1108 p = find_task_by_vpid(pid);
1109 if (!p)
1110 goto out;
1111
1112 err = -EINVAL;
1113 if (!thread_group_leader(p))
1114 goto out;
1115
1116 if (same_thread_group(p->real_parent, group_leader)) {
1117 err = -EPERM;
1118 if (task_session(p) != task_session(group_leader))
1119 goto out;
1120 err = -EACCES;
1121 if (p->did_exec)
1122 goto out;
1123 } else {
1124 err = -ESRCH;
1125 if (p != group_leader)
1126 goto out;
1127 }
1128
1129 err = -EPERM;
1130 if (p->signal->leader)
1131 goto out;
1132
1133 pgrp = task_pid(p);
1134 if (pgid != pid) {
1135 struct task_struct *g;
1136
1137 pgrp = find_vpid(pgid);
1138 g = pid_task(pgrp, PIDTYPE_PGID);
1139 if (!g || task_session(g) != task_session(group_leader))
1140 goto out;
1141 }
1142
1143 err = security_task_setpgid(p, pgid);
1144 if (err)
1145 goto out;
1146
1147 if (task_pgrp(p) != pgrp)
1148 change_pid(p, PIDTYPE_PGID, pgrp);
1149
1150 err = 0;
1151out:
1152 /* All paths lead to here, thus we are safe. -DaveM */
1153 write_unlock_irq(&tasklist_lock);
1154 rcu_read_unlock();
1155 return err;
1156}
1157
1158SYSCALL_DEFINE1(getpgid, pid_t, pid)
1159{
1160 struct task_struct *p;
1161 struct pid *grp;
1162 int retval;
1163
1164 rcu_read_lock();
1165 if (!pid)
1166 grp = task_pgrp(current);
1167 else {
1168 retval = -ESRCH;
1169 p = find_task_by_vpid(pid);
1170 if (!p)
1171 goto out;
1172 grp = task_pgrp(p);
1173 if (!grp)
1174 goto out;
1175
1176 retval = security_task_getpgid(p);
1177 if (retval)
1178 goto out;
1179 }
1180 retval = pid_vnr(grp);
1181out:
1182 rcu_read_unlock();
1183 return retval;
1184}
1185
1186#ifdef __ARCH_WANT_SYS_GETPGRP
1187
1188SYSCALL_DEFINE0(getpgrp)
1189{
1190 return sys_getpgid(0);
1191}
1192
1193#endif
1194
1195SYSCALL_DEFINE1(getsid, pid_t, pid)
1196{
1197 struct task_struct *p;
1198 struct pid *sid;
1199 int retval;
1200
1201 rcu_read_lock();
1202 if (!pid)
1203 sid = task_session(current);
1204 else {
1205 retval = -ESRCH;
1206 p = find_task_by_vpid(pid);
1207 if (!p)
1208 goto out;
1209 sid = task_session(p);
1210 if (!sid)
1211 goto out;
1212
1213 retval = security_task_getsid(p);
1214 if (retval)
1215 goto out;
1216 }
1217 retval = pid_vnr(sid);
1218out:
1219 rcu_read_unlock();
1220 return retval;
1221}
1222
1223SYSCALL_DEFINE0(setsid)
1224{
1225 struct task_struct *group_leader = current->group_leader;
1226 struct pid *sid = task_pid(group_leader);
1227 pid_t session = pid_vnr(sid);
1228 int err = -EPERM;
1229
1230 write_lock_irq(&tasklist_lock);
1231 /* Fail if I am already a session leader */
1232 if (group_leader->signal->leader)
1233 goto out;
1234
1235 /* Fail if a process group id already exists that equals the
1236 * proposed session id.
1237 */
1238 if (pid_task(sid, PIDTYPE_PGID))
1239 goto out;
1240
1241 group_leader->signal->leader = 1;
1242 __set_special_pids(sid);
1243
1244 proc_clear_tty(group_leader);
1245
1246 err = session;
1247out:
1248 write_unlock_irq(&tasklist_lock);
1249 if (err > 0) {
1250 proc_sid_connector(group_leader);
1251 sched_autogroup_create_attach(group_leader);
1252 }
1253 return err;
1254}
1255
1256DECLARE_RWSEM(uts_sem);
1257
1258#ifdef COMPAT_UTS_MACHINE
1259#define override_architecture(name) \
1260 (personality(current->personality) == PER_LINUX32 && \
1261 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1262 sizeof(COMPAT_UTS_MACHINE)))
1263#else
1264#define override_architecture(name) 0
1265#endif
1266
1267/*
1268 * Work around broken programs that cannot handle "Linux 3.0".
1269 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1270 */
1271static int override_release(char __user *release, size_t len)
1272{
1273 int ret = 0;
1274
1275 if (current->personality & UNAME26) {
1276 const char *rest = UTS_RELEASE;
1277 char buf[65] = { 0 };
1278 int ndots = 0;
1279 unsigned v;
1280 size_t copy;
1281
1282 while (*rest) {
1283 if (*rest == '.' && ++ndots >= 3)
1284 break;
1285 if (!isdigit(*rest) && *rest != '.')
1286 break;
1287 rest++;
1288 }
1289 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1290 copy = clamp_t(size_t, len, 1, sizeof(buf));
1291 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1292 ret = copy_to_user(release, buf, copy + 1);
1293 }
1294 return ret;
1295}
1296
1297SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1298{
1299 int errno = 0;
1300
1301 down_read(&uts_sem);
1302 if (copy_to_user(name, utsname(), sizeof *name))
1303 errno = -EFAULT;
1304 up_read(&uts_sem);
1305
1306 if (!errno && override_release(name->release, sizeof(name->release)))
1307 errno = -EFAULT;
1308 if (!errno && override_architecture(name))
1309 errno = -EFAULT;
1310 return errno;
1311}
1312
1313#ifdef __ARCH_WANT_SYS_OLD_UNAME
1314/*
1315 * Old cruft
1316 */
1317SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1318{
1319 int error = 0;
1320
1321 if (!name)
1322 return -EFAULT;
1323
1324 down_read(&uts_sem);
1325 if (copy_to_user(name, utsname(), sizeof(*name)))
1326 error = -EFAULT;
1327 up_read(&uts_sem);
1328
1329 if (!error && override_release(name->release, sizeof(name->release)))
1330 error = -EFAULT;
1331 if (!error && override_architecture(name))
1332 error = -EFAULT;
1333 return error;
1334}
1335
1336SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1337{
1338 int error;
1339
1340 if (!name)
1341 return -EFAULT;
1342 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1343 return -EFAULT;
1344
1345 down_read(&uts_sem);
1346 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1347 __OLD_UTS_LEN);
1348 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1349 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1350 __OLD_UTS_LEN);
1351 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1352 error |= __copy_to_user(&name->release, &utsname()->release,
1353 __OLD_UTS_LEN);
1354 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1355 error |= __copy_to_user(&name->version, &utsname()->version,
1356 __OLD_UTS_LEN);
1357 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1358 error |= __copy_to_user(&name->machine, &utsname()->machine,
1359 __OLD_UTS_LEN);
1360 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1361 up_read(&uts_sem);
1362
1363 if (!error && override_architecture(name))
1364 error = -EFAULT;
1365 if (!error && override_release(name->release, sizeof(name->release)))
1366 error = -EFAULT;
1367 return error ? -EFAULT : 0;
1368}
1369#endif
1370
1371SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1372{
1373 int errno;
1374 char tmp[__NEW_UTS_LEN];
1375
1376 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1377 return -EPERM;
1378
1379 if (len < 0 || len > __NEW_UTS_LEN)
1380 return -EINVAL;
1381 down_write(&uts_sem);
1382 errno = -EFAULT;
1383 if (!copy_from_user(tmp, name, len)) {
1384 struct new_utsname *u = utsname();
1385
1386 memcpy(u->nodename, tmp, len);
1387 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1388 errno = 0;
1389 uts_proc_notify(UTS_PROC_HOSTNAME);
1390 }
1391 up_write(&uts_sem);
1392 return errno;
1393}
1394
1395#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1396
1397SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1398{
1399 int i, errno;
1400 struct new_utsname *u;
1401
1402 if (len < 0)
1403 return -EINVAL;
1404 down_read(&uts_sem);
1405 u = utsname();
1406 i = 1 + strlen(u->nodename);
1407 if (i > len)
1408 i = len;
1409 errno = 0;
1410 if (copy_to_user(name, u->nodename, i))
1411 errno = -EFAULT;
1412 up_read(&uts_sem);
1413 return errno;
1414}
1415
1416#endif
1417
1418/*
1419 * Only setdomainname; getdomainname can be implemented by calling
1420 * uname()
1421 */
1422SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1423{
1424 int errno;
1425 char tmp[__NEW_UTS_LEN];
1426
1427 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1428 return -EPERM;
1429 if (len < 0 || len > __NEW_UTS_LEN)
1430 return -EINVAL;
1431
1432 down_write(&uts_sem);
1433 errno = -EFAULT;
1434 if (!copy_from_user(tmp, name, len)) {
1435 struct new_utsname *u = utsname();
1436
1437 memcpy(u->domainname, tmp, len);
1438 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1439 errno = 0;
1440 uts_proc_notify(UTS_PROC_DOMAINNAME);
1441 }
1442 up_write(&uts_sem);
1443 return errno;
1444}
1445
1446SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1447{
1448 struct rlimit value;
1449 int ret;
1450
1451 ret = do_prlimit(current, resource, NULL, &value);
1452 if (!ret)
1453 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1454
1455 return ret;
1456}
1457
1458#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1459
1460/*
1461 * Back compatibility for getrlimit. Needed for some apps.
1462 */
1463
1464SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1465 struct rlimit __user *, rlim)
1466{
1467 struct rlimit x;
1468 if (resource >= RLIM_NLIMITS)
1469 return -EINVAL;
1470
1471 task_lock(current->group_leader);
1472 x = current->signal->rlim[resource];
1473 task_unlock(current->group_leader);
1474 if (x.rlim_cur > 0x7FFFFFFF)
1475 x.rlim_cur = 0x7FFFFFFF;
1476 if (x.rlim_max > 0x7FFFFFFF)
1477 x.rlim_max = 0x7FFFFFFF;
1478 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1479}
1480
1481#endif
1482
1483static inline bool rlim64_is_infinity(__u64 rlim64)
1484{
1485#if BITS_PER_LONG < 64
1486 return rlim64 >= ULONG_MAX;
1487#else
1488 return rlim64 == RLIM64_INFINITY;
1489#endif
1490}
1491
1492static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1493{
1494 if (rlim->rlim_cur == RLIM_INFINITY)
1495 rlim64->rlim_cur = RLIM64_INFINITY;
1496 else
1497 rlim64->rlim_cur = rlim->rlim_cur;
1498 if (rlim->rlim_max == RLIM_INFINITY)
1499 rlim64->rlim_max = RLIM64_INFINITY;
1500 else
1501 rlim64->rlim_max = rlim->rlim_max;
1502}
1503
1504static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1505{
1506 if (rlim64_is_infinity(rlim64->rlim_cur))
1507 rlim->rlim_cur = RLIM_INFINITY;
1508 else
1509 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1510 if (rlim64_is_infinity(rlim64->rlim_max))
1511 rlim->rlim_max = RLIM_INFINITY;
1512 else
1513 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1514}
1515
1516/* make sure you are allowed to change @tsk limits before calling this */
1517int do_prlimit(struct task_struct *tsk, unsigned int resource,
1518 struct rlimit *new_rlim, struct rlimit *old_rlim)
1519{
1520 struct rlimit *rlim;
1521 int retval = 0;
1522
1523 if (resource >= RLIM_NLIMITS)
1524 return -EINVAL;
1525 if (new_rlim) {
1526 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1527 return -EINVAL;
1528 if (resource == RLIMIT_NOFILE &&
1529 new_rlim->rlim_max > sysctl_nr_open)
1530 return -EPERM;
1531 }
1532
1533 /* protect tsk->signal and tsk->sighand from disappearing */
1534 read_lock(&tasklist_lock);
1535 if (!tsk->sighand) {
1536 retval = -ESRCH;
1537 goto out;
1538 }
1539
1540 rlim = tsk->signal->rlim + resource;
1541 task_lock(tsk->group_leader);
1542 if (new_rlim) {
1543 /* Keep the capable check against init_user_ns until
1544 cgroups can contain all limits */
1545 if (new_rlim->rlim_max > rlim->rlim_max &&
1546 !capable(CAP_SYS_RESOURCE))
1547 retval = -EPERM;
1548 if (!retval)
1549 retval = security_task_setrlimit(tsk->group_leader,
1550 resource, new_rlim);
1551 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1552 /*
1553 * The caller is asking for an immediate RLIMIT_CPU
1554 * expiry. But we use the zero value to mean "it was
1555 * never set". So let's cheat and make it one second
1556 * instead
1557 */
1558 new_rlim->rlim_cur = 1;
1559 }
1560 }
1561 if (!retval) {
1562 if (old_rlim)
1563 *old_rlim = *rlim;
1564 if (new_rlim)
1565 *rlim = *new_rlim;
1566 }
1567 task_unlock(tsk->group_leader);
1568
1569 /*
1570 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1571 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1572 * very long-standing error, and fixing it now risks breakage of
1573 * applications, so we live with it
1574 */
1575 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1576 new_rlim->rlim_cur != RLIM_INFINITY)
1577 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1578out:
1579 read_unlock(&tasklist_lock);
1580 return retval;
1581}
1582
1583/* rcu lock must be held */
1584static int check_prlimit_permission(struct task_struct *task)
1585{
1586 const struct cred *cred = current_cred(), *tcred;
1587
1588 if (current == task)
1589 return 0;
1590
1591 tcred = __task_cred(task);
1592 if (uid_eq(cred->uid, tcred->euid) &&
1593 uid_eq(cred->uid, tcred->suid) &&
1594 uid_eq(cred->uid, tcred->uid) &&
1595 gid_eq(cred->gid, tcred->egid) &&
1596 gid_eq(cred->gid, tcred->sgid) &&
1597 gid_eq(cred->gid, tcred->gid))
1598 return 0;
1599 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1600 return 0;
1601
1602 return -EPERM;
1603}
1604
1605SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1606 const struct rlimit64 __user *, new_rlim,
1607 struct rlimit64 __user *, old_rlim)
1608{
1609 struct rlimit64 old64, new64;
1610 struct rlimit old, new;
1611 struct task_struct *tsk;
1612 int ret;
1613
1614 if (new_rlim) {
1615 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1616 return -EFAULT;
1617 rlim64_to_rlim(&new64, &new);
1618 }
1619
1620 rcu_read_lock();
1621 tsk = pid ? find_task_by_vpid(pid) : current;
1622 if (!tsk) {
1623 rcu_read_unlock();
1624 return -ESRCH;
1625 }
1626 ret = check_prlimit_permission(tsk);
1627 if (ret) {
1628 rcu_read_unlock();
1629 return ret;
1630 }
1631 get_task_struct(tsk);
1632 rcu_read_unlock();
1633
1634 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1635 old_rlim ? &old : NULL);
1636
1637 if (!ret && old_rlim) {
1638 rlim_to_rlim64(&old, &old64);
1639 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1640 ret = -EFAULT;
1641 }
1642
1643 put_task_struct(tsk);
1644 return ret;
1645}
1646
1647SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1648{
1649 struct rlimit new_rlim;
1650
1651 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1652 return -EFAULT;
1653 return do_prlimit(current, resource, &new_rlim, NULL);
1654}
1655
1656/*
1657 * It would make sense to put struct rusage in the task_struct,
1658 * except that would make the task_struct be *really big*. After
1659 * task_struct gets moved into malloc'ed memory, it would
1660 * make sense to do this. It will make moving the rest of the information
1661 * a lot simpler! (Which we're not doing right now because we're not
1662 * measuring them yet).
1663 *
1664 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1665 * races with threads incrementing their own counters. But since word
1666 * reads are atomic, we either get new values or old values and we don't
1667 * care which for the sums. We always take the siglock to protect reading
1668 * the c* fields from p->signal from races with exit.c updating those
1669 * fields when reaping, so a sample either gets all the additions of a
1670 * given child after it's reaped, or none so this sample is before reaping.
1671 *
1672 * Locking:
1673 * We need to take the siglock for CHILDEREN, SELF and BOTH
1674 * for the cases current multithreaded, non-current single threaded
1675 * non-current multithreaded. Thread traversal is now safe with
1676 * the siglock held.
1677 * Strictly speaking, we donot need to take the siglock if we are current and
1678 * single threaded, as no one else can take our signal_struct away, no one
1679 * else can reap the children to update signal->c* counters, and no one else
1680 * can race with the signal-> fields. If we do not take any lock, the
1681 * signal-> fields could be read out of order while another thread was just
1682 * exiting. So we should place a read memory barrier when we avoid the lock.
1683 * On the writer side, write memory barrier is implied in __exit_signal
1684 * as __exit_signal releases the siglock spinlock after updating the signal->
1685 * fields. But we don't do this yet to keep things simple.
1686 *
1687 */
1688
1689static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1690{
1691 r->ru_nvcsw += t->nvcsw;
1692 r->ru_nivcsw += t->nivcsw;
1693 r->ru_minflt += t->min_flt;
1694 r->ru_majflt += t->maj_flt;
1695 r->ru_inblock += task_io_get_inblock(t);
1696 r->ru_oublock += task_io_get_oublock(t);
1697}
1698
1699static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1700{
1701 struct task_struct *t;
1702 unsigned long flags;
1703 cputime_t tgutime, tgstime, utime, stime;
1704 unsigned long maxrss = 0;
1705
1706 memset((char *) r, 0, sizeof *r);
1707 utime = stime = 0;
1708
1709 if (who == RUSAGE_THREAD) {
1710 task_cputime_adjusted(current, &utime, &stime);
1711 accumulate_thread_rusage(p, r);
1712 maxrss = p->signal->maxrss;
1713 goto out;
1714 }
1715
1716 if (!lock_task_sighand(p, &flags))
1717 return;
1718
1719 switch (who) {
1720 case RUSAGE_BOTH:
1721 case RUSAGE_CHILDREN:
1722 utime = p->signal->cutime;
1723 stime = p->signal->cstime;
1724 r->ru_nvcsw = p->signal->cnvcsw;
1725 r->ru_nivcsw = p->signal->cnivcsw;
1726 r->ru_minflt = p->signal->cmin_flt;
1727 r->ru_majflt = p->signal->cmaj_flt;
1728 r->ru_inblock = p->signal->cinblock;
1729 r->ru_oublock = p->signal->coublock;
1730 maxrss = p->signal->cmaxrss;
1731
1732 if (who == RUSAGE_CHILDREN)
1733 break;
1734
1735 case RUSAGE_SELF:
1736 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1737 utime += tgutime;
1738 stime += tgstime;
1739 r->ru_nvcsw += p->signal->nvcsw;
1740 r->ru_nivcsw += p->signal->nivcsw;
1741 r->ru_minflt += p->signal->min_flt;
1742 r->ru_majflt += p->signal->maj_flt;
1743 r->ru_inblock += p->signal->inblock;
1744 r->ru_oublock += p->signal->oublock;
1745 if (maxrss < p->signal->maxrss)
1746 maxrss = p->signal->maxrss;
1747 t = p;
1748 do {
1749 accumulate_thread_rusage(t, r);
1750 t = next_thread(t);
1751 } while (t != p);
1752 break;
1753
1754 default:
1755 BUG();
1756 }
1757 unlock_task_sighand(p, &flags);
1758
1759out:
1760 cputime_to_timeval(utime, &r->ru_utime);
1761 cputime_to_timeval(stime, &r->ru_stime);
1762
1763 if (who != RUSAGE_CHILDREN) {
1764 struct mm_struct *mm = get_task_mm(p);
1765 if (mm) {
1766 setmax_mm_hiwater_rss(&maxrss, mm);
1767 mmput(mm);
1768 }
1769 }
1770 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1771}
1772
1773int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1774{
1775 struct rusage r;
1776 k_getrusage(p, who, &r);
1777 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1778}
1779
1780SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1781{
1782 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1783 who != RUSAGE_THREAD)
1784 return -EINVAL;
1785 return getrusage(current, who, ru);
1786}
1787
1788SYSCALL_DEFINE1(umask, int, mask)
1789{
1790 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1791 return mask;
1792}
1793
1794#ifdef CONFIG_CHECKPOINT_RESTORE
1795static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1796{
1797 struct fd exe;
1798 struct inode *inode;
1799 int err;
1800
1801 exe = fdget(fd);
1802 if (!exe.file)
1803 return -EBADF;
1804
1805 inode = file_inode(exe.file);
1806
1807 /*
1808 * Because the original mm->exe_file points to executable file, make
1809 * sure that this one is executable as well, to avoid breaking an
1810 * overall picture.
1811 */
1812 err = -EACCES;
1813 if (!S_ISREG(inode->i_mode) ||
1814 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1815 goto exit;
1816
1817 err = inode_permission(inode, MAY_EXEC);
1818 if (err)
1819 goto exit;
1820
1821 down_write(&mm->mmap_sem);
1822
1823 /*
1824 * Forbid mm->exe_file change if old file still mapped.
1825 */
1826 err = -EBUSY;
1827 if (mm->exe_file) {
1828 struct vm_area_struct *vma;
1829
1830 for (vma = mm->mmap; vma; vma = vma->vm_next)
1831 if (vma->vm_file &&
1832 path_equal(&vma->vm_file->f_path,
1833 &mm->exe_file->f_path))
1834 goto exit_unlock;
1835 }
1836
1837 /*
1838 * The symlink can be changed only once, just to disallow arbitrary
1839 * transitions malicious software might bring in. This means one
1840 * could make a snapshot over all processes running and monitor
1841 * /proc/pid/exe changes to notice unusual activity if needed.
1842 */
1843 err = -EPERM;
1844 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1845 goto exit_unlock;
1846
1847 err = 0;
1848 set_mm_exe_file(mm, exe.file); /* this grabs a reference to exe.file */
1849exit_unlock:
1850 up_write(&mm->mmap_sem);
1851
1852exit:
1853 fdput(exe);
1854 return err;
1855}
1856
1857static int prctl_set_mm(int opt, unsigned long addr,
1858 unsigned long arg4, unsigned long arg5)
1859{
1860 unsigned long rlim = rlimit(RLIMIT_DATA);
1861 struct mm_struct *mm = current->mm;
1862 struct vm_area_struct *vma;
1863 int error;
1864
1865 if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
1866 return -EINVAL;
1867
1868 if (!capable(CAP_SYS_RESOURCE))
1869 return -EPERM;
1870
1871 if (opt == PR_SET_MM_EXE_FILE)
1872 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1873
1874 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1875 return -EINVAL;
1876
1877 error = -EINVAL;
1878
1879 down_read(&mm->mmap_sem);
1880 vma = find_vma(mm, addr);
1881
1882 switch (opt) {
1883 case PR_SET_MM_START_CODE:
1884 mm->start_code = addr;
1885 break;
1886 case PR_SET_MM_END_CODE:
1887 mm->end_code = addr;
1888 break;
1889 case PR_SET_MM_START_DATA:
1890 mm->start_data = addr;
1891 break;
1892 case PR_SET_MM_END_DATA:
1893 mm->end_data = addr;
1894 break;
1895
1896 case PR_SET_MM_START_BRK:
1897 if (addr <= mm->end_data)
1898 goto out;
1899
1900 if (rlim < RLIM_INFINITY &&
1901 (mm->brk - addr) +
1902 (mm->end_data - mm->start_data) > rlim)
1903 goto out;
1904
1905 mm->start_brk = addr;
1906 break;
1907
1908 case PR_SET_MM_BRK:
1909 if (addr <= mm->end_data)
1910 goto out;
1911
1912 if (rlim < RLIM_INFINITY &&
1913 (addr - mm->start_brk) +
1914 (mm->end_data - mm->start_data) > rlim)
1915 goto out;
1916
1917 mm->brk = addr;
1918 break;
1919
1920 /*
1921 * If command line arguments and environment
1922 * are placed somewhere else on stack, we can
1923 * set them up here, ARG_START/END to setup
1924 * command line argumets and ENV_START/END
1925 * for environment.
1926 */
1927 case PR_SET_MM_START_STACK:
1928 case PR_SET_MM_ARG_START:
1929 case PR_SET_MM_ARG_END:
1930 case PR_SET_MM_ENV_START:
1931 case PR_SET_MM_ENV_END:
1932 if (!vma) {
1933 error = -EFAULT;
1934 goto out;
1935 }
1936 if (opt == PR_SET_MM_START_STACK)
1937 mm->start_stack = addr;
1938 else if (opt == PR_SET_MM_ARG_START)
1939 mm->arg_start = addr;
1940 else if (opt == PR_SET_MM_ARG_END)
1941 mm->arg_end = addr;
1942 else if (opt == PR_SET_MM_ENV_START)
1943 mm->env_start = addr;
1944 else if (opt == PR_SET_MM_ENV_END)
1945 mm->env_end = addr;
1946 break;
1947
1948 /*
1949 * This doesn't move auxiliary vector itself
1950 * since it's pinned to mm_struct, but allow
1951 * to fill vector with new values. It's up
1952 * to a caller to provide sane values here
1953 * otherwise user space tools which use this
1954 * vector might be unhappy.
1955 */
1956 case PR_SET_MM_AUXV: {
1957 unsigned long user_auxv[AT_VECTOR_SIZE];
1958
1959 if (arg4 > sizeof(user_auxv))
1960 goto out;
1961 up_read(&mm->mmap_sem);
1962
1963 if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
1964 return -EFAULT;
1965
1966 /* Make sure the last entry is always AT_NULL */
1967 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1968 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1969
1970 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1971
1972 task_lock(current);
1973 memcpy(mm->saved_auxv, user_auxv, arg4);
1974 task_unlock(current);
1975
1976 return 0;
1977 }
1978 default:
1979 goto out;
1980 }
1981
1982 error = 0;
1983out:
1984 up_read(&mm->mmap_sem);
1985 return error;
1986}
1987
1988static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1989{
1990 return put_user(me->clear_child_tid, tid_addr);
1991}
1992
1993#else /* CONFIG_CHECKPOINT_RESTORE */
1994static int prctl_set_mm(int opt, unsigned long addr,
1995 unsigned long arg4, unsigned long arg5)
1996{
1997 return -EINVAL;
1998}
1999static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2000{
2001 return -EINVAL;
2002}
2003#endif
2004
2005SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2006 unsigned long, arg4, unsigned long, arg5)
2007{
2008 struct task_struct *me = current;
2009 unsigned char comm[sizeof(me->comm)];
2010 long error;
2011
2012 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2013 if (error != -ENOSYS)
2014 return error;
2015
2016 error = 0;
2017 switch (option) {
2018 case PR_SET_PDEATHSIG:
2019 if (!valid_signal(arg2)) {
2020 error = -EINVAL;
2021 break;
2022 }
2023 me->pdeath_signal = arg2;
2024 break;
2025 case PR_GET_PDEATHSIG:
2026 error = put_user(me->pdeath_signal, (int __user *)arg2);
2027 break;
2028 case PR_GET_DUMPABLE:
2029 error = get_dumpable(me->mm);
2030 break;
2031 case PR_SET_DUMPABLE:
2032 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2033 error = -EINVAL;
2034 break;
2035 }
2036 set_dumpable(me->mm, arg2);
2037 break;
2038
2039 case PR_SET_UNALIGN:
2040 error = SET_UNALIGN_CTL(me, arg2);
2041 break;
2042 case PR_GET_UNALIGN:
2043 error = GET_UNALIGN_CTL(me, arg2);
2044 break;
2045 case PR_SET_FPEMU:
2046 error = SET_FPEMU_CTL(me, arg2);
2047 break;
2048 case PR_GET_FPEMU:
2049 error = GET_FPEMU_CTL(me, arg2);
2050 break;
2051 case PR_SET_FPEXC:
2052 error = SET_FPEXC_CTL(me, arg2);
2053 break;
2054 case PR_GET_FPEXC:
2055 error = GET_FPEXC_CTL(me, arg2);
2056 break;
2057 case PR_GET_TIMING:
2058 error = PR_TIMING_STATISTICAL;
2059 break;
2060 case PR_SET_TIMING:
2061 if (arg2 != PR_TIMING_STATISTICAL)
2062 error = -EINVAL;
2063 break;
2064 case PR_SET_NAME:
2065 comm[sizeof(me->comm) - 1] = 0;
2066 if (strncpy_from_user(comm, (char __user *)arg2,
2067 sizeof(me->comm) - 1) < 0)
2068 return -EFAULT;
2069 set_task_comm(me, comm);
2070 proc_comm_connector(me);
2071 break;
2072 case PR_GET_NAME:
2073 get_task_comm(comm, me);
2074 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2075 return -EFAULT;
2076 break;
2077 case PR_GET_ENDIAN:
2078 error = GET_ENDIAN(me, arg2);
2079 break;
2080 case PR_SET_ENDIAN:
2081 error = SET_ENDIAN(me, arg2);
2082 break;
2083 case PR_GET_SECCOMP:
2084 error = prctl_get_seccomp();
2085 break;
2086 case PR_SET_SECCOMP:
2087 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2088 break;
2089 case PR_GET_TSC:
2090 error = GET_TSC_CTL(arg2);
2091 break;
2092 case PR_SET_TSC:
2093 error = SET_TSC_CTL(arg2);
2094 break;
2095 case PR_TASK_PERF_EVENTS_DISABLE:
2096 error = perf_event_task_disable();
2097 break;
2098 case PR_TASK_PERF_EVENTS_ENABLE:
2099 error = perf_event_task_enable();
2100 break;
2101 case PR_GET_TIMERSLACK:
2102 error = current->timer_slack_ns;
2103 break;
2104 case PR_SET_TIMERSLACK:
2105 if (arg2 <= 0)
2106 current->timer_slack_ns =
2107 current->default_timer_slack_ns;
2108 else
2109 current->timer_slack_ns = arg2;
2110 break;
2111 case PR_MCE_KILL:
2112 if (arg4 | arg5)
2113 return -EINVAL;
2114 switch (arg2) {
2115 case PR_MCE_KILL_CLEAR:
2116 if (arg3 != 0)
2117 return -EINVAL;
2118 current->flags &= ~PF_MCE_PROCESS;
2119 break;
2120 case PR_MCE_KILL_SET:
2121 current->flags |= PF_MCE_PROCESS;
2122 if (arg3 == PR_MCE_KILL_EARLY)
2123 current->flags |= PF_MCE_EARLY;
2124 else if (arg3 == PR_MCE_KILL_LATE)
2125 current->flags &= ~PF_MCE_EARLY;
2126 else if (arg3 == PR_MCE_KILL_DEFAULT)
2127 current->flags &=
2128 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2129 else
2130 return -EINVAL;
2131 break;
2132 default:
2133 return -EINVAL;
2134 }
2135 break;
2136 case PR_MCE_KILL_GET:
2137 if (arg2 | arg3 | arg4 | arg5)
2138 return -EINVAL;
2139 if (current->flags & PF_MCE_PROCESS)
2140 error = (current->flags & PF_MCE_EARLY) ?
2141 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2142 else
2143 error = PR_MCE_KILL_DEFAULT;
2144 break;
2145 case PR_SET_MM:
2146 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2147 break;
2148 case PR_GET_TID_ADDRESS:
2149 error = prctl_get_tid_address(me, (int __user **)arg2);
2150 break;
2151 case PR_SET_CHILD_SUBREAPER:
2152 me->signal->is_child_subreaper = !!arg2;
2153 break;
2154 case PR_GET_CHILD_SUBREAPER:
2155 error = put_user(me->signal->is_child_subreaper,
2156 (int __user *)arg2);
2157 break;
2158 case PR_SET_NO_NEW_PRIVS:
2159 if (arg2 != 1 || arg3 || arg4 || arg5)
2160 return -EINVAL;
2161
2162 current->no_new_privs = 1;
2163 break;
2164 case PR_GET_NO_NEW_PRIVS:
2165 if (arg2 || arg3 || arg4 || arg5)
2166 return -EINVAL;
2167 return current->no_new_privs ? 1 : 0;
2168 default:
2169 error = -EINVAL;
2170 break;
2171 }
2172 return error;
2173}
2174
2175SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2176 struct getcpu_cache __user *, unused)
2177{
2178 int err = 0;
2179 int cpu = raw_smp_processor_id();
2180 if (cpup)
2181 err |= put_user(cpu, cpup);
2182 if (nodep)
2183 err |= put_user(cpu_to_node(cpu), nodep);
2184 return err ? -EFAULT : 0;
2185}
2186
2187char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
2188
2189static int __orderly_poweroff(bool force)
2190{
2191 char **argv;
2192 static char *envp[] = {
2193 "HOME=/",
2194 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
2195 NULL
2196 };
2197 int ret;
2198
2199 argv = argv_split(GFP_KERNEL, poweroff_cmd, NULL);
2200 if (argv) {
2201 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
2202 argv_free(argv);
2203 } else {
2204 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
2205 __func__, poweroff_cmd);
2206 ret = -ENOMEM;
2207 }
2208
2209 if (ret && force) {
2210 printk(KERN_WARNING "Failed to start orderly shutdown: "
2211 "forcing the issue\n");
2212 /*
2213 * I guess this should try to kick off some daemon to sync and
2214 * poweroff asap. Or not even bother syncing if we're doing an
2215 * emergency shutdown?
2216 */
2217 emergency_sync();
2218 kernel_power_off();
2219 }
2220
2221 return ret;
2222}
2223
2224static bool poweroff_force;
2225
2226static void poweroff_work_func(struct work_struct *work)
2227{
2228 __orderly_poweroff(poweroff_force);
2229}
2230
2231static DECLARE_WORK(poweroff_work, poweroff_work_func);
2232
2233/**
2234 * orderly_poweroff - Trigger an orderly system poweroff
2235 * @force: force poweroff if command execution fails
2236 *
2237 * This may be called from any context to trigger a system shutdown.
2238 * If the orderly shutdown fails, it will force an immediate shutdown.
2239 */
2240int orderly_poweroff(bool force)
2241{
2242 if (force) /* do not override the pending "true" */
2243 poweroff_force = true;
2244 schedule_work(&poweroff_work);
2245 return 0;
2246}
2247EXPORT_SYMBOL_GPL(orderly_poweroff);