xref: /linux-6.15/kernel/sys.c (revision 7ec7fb39)
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/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 
37 #include <linux/compat.h>
38 #include <linux/syscalls.h>
39 #include <linux/kprobes.h>
40 #include <linux/user_namespace.h>
41 
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/unistd.h>
45 
46 #ifndef SET_UNALIGN_CTL
47 # define SET_UNALIGN_CTL(a,b)	(-EINVAL)
48 #endif
49 #ifndef GET_UNALIGN_CTL
50 # define GET_UNALIGN_CTL(a,b)	(-EINVAL)
51 #endif
52 #ifndef SET_FPEMU_CTL
53 # define SET_FPEMU_CTL(a,b)	(-EINVAL)
54 #endif
55 #ifndef GET_FPEMU_CTL
56 # define GET_FPEMU_CTL(a,b)	(-EINVAL)
57 #endif
58 #ifndef SET_FPEXC_CTL
59 # define SET_FPEXC_CTL(a,b)	(-EINVAL)
60 #endif
61 #ifndef GET_FPEXC_CTL
62 # define GET_FPEXC_CTL(a,b)	(-EINVAL)
63 #endif
64 #ifndef GET_ENDIAN
65 # define GET_ENDIAN(a,b)	(-EINVAL)
66 #endif
67 #ifndef SET_ENDIAN
68 # define SET_ENDIAN(a,b)	(-EINVAL)
69 #endif
70 #ifndef GET_TSC_CTL
71 # define GET_TSC_CTL(a)		(-EINVAL)
72 #endif
73 #ifndef SET_TSC_CTL
74 # define SET_TSC_CTL(a)		(-EINVAL)
75 #endif
76 
77 /*
78  * this is where the system-wide overflow UID and GID are defined, for
79  * architectures that now have 32-bit UID/GID but didn't in the past
80  */
81 
82 int overflowuid = DEFAULT_OVERFLOWUID;
83 int overflowgid = DEFAULT_OVERFLOWGID;
84 
85 #ifdef CONFIG_UID16
86 EXPORT_SYMBOL(overflowuid);
87 EXPORT_SYMBOL(overflowgid);
88 #endif
89 
90 /*
91  * the same as above, but for filesystems which can only store a 16-bit
92  * UID and GID. as such, this is needed on all architectures
93  */
94 
95 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
96 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
97 
98 EXPORT_SYMBOL(fs_overflowuid);
99 EXPORT_SYMBOL(fs_overflowgid);
100 
101 /*
102  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
103  */
104 
105 int C_A_D = 1;
106 struct pid *cad_pid;
107 EXPORT_SYMBOL(cad_pid);
108 
109 /*
110  * If set, this is used for preparing the system to power off.
111  */
112 
113 void (*pm_power_off_prepare)(void);
114 
115 /*
116  * set the priority of a task
117  * - the caller must hold the RCU read lock
118  */
119 static int set_one_prio(struct task_struct *p, int niceval, int error)
120 {
121 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
122 	int no_nice;
123 
124 	if (pcred->uid  != cred->euid &&
125 	    pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
126 		error = -EPERM;
127 		goto out;
128 	}
129 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
130 		error = -EACCES;
131 		goto out;
132 	}
133 	no_nice = security_task_setnice(p, niceval);
134 	if (no_nice) {
135 		error = no_nice;
136 		goto out;
137 	}
138 	if (error == -ESRCH)
139 		error = 0;
140 	set_user_nice(p, niceval);
141 out:
142 	return error;
143 }
144 
145 asmlinkage long sys_setpriority(int which, int who, int niceval)
146 {
147 	struct task_struct *g, *p;
148 	struct user_struct *user;
149 	const struct cred *cred = current_cred();
150 	int error = -EINVAL;
151 	struct pid *pgrp;
152 
153 	if (which > PRIO_USER || which < PRIO_PROCESS)
154 		goto out;
155 
156 	/* normalize: avoid signed division (rounding problems) */
157 	error = -ESRCH;
158 	if (niceval < -20)
159 		niceval = -20;
160 	if (niceval > 19)
161 		niceval = 19;
162 
163 	read_lock(&tasklist_lock);
164 	switch (which) {
165 		case PRIO_PROCESS:
166 			if (who)
167 				p = find_task_by_vpid(who);
168 			else
169 				p = current;
170 			if (p)
171 				error = set_one_prio(p, niceval, error);
172 			break;
173 		case PRIO_PGRP:
174 			if (who)
175 				pgrp = find_vpid(who);
176 			else
177 				pgrp = task_pgrp(current);
178 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
179 				error = set_one_prio(p, niceval, error);
180 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
181 			break;
182 		case PRIO_USER:
183 			user = (struct user_struct *) cred->user;
184 			if (!who)
185 				who = cred->uid;
186 			else if ((who != cred->uid) &&
187 				 !(user = find_user(who)))
188 				goto out_unlock;	/* No processes for this user */
189 
190 			do_each_thread(g, p)
191 				if (__task_cred(p)->uid == who)
192 					error = set_one_prio(p, niceval, error);
193 			while_each_thread(g, p);
194 			if (who != cred->uid)
195 				free_uid(user);		/* For find_user() */
196 			break;
197 	}
198 out_unlock:
199 	read_unlock(&tasklist_lock);
200 out:
201 	return error;
202 }
203 
204 /*
205  * Ugh. To avoid negative return values, "getpriority()" will
206  * not return the normal nice-value, but a negated value that
207  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
208  * to stay compatible.
209  */
210 asmlinkage long sys_getpriority(int which, int who)
211 {
212 	struct task_struct *g, *p;
213 	struct user_struct *user;
214 	const struct cred *cred = current_cred();
215 	long niceval, retval = -ESRCH;
216 	struct pid *pgrp;
217 
218 	if (which > PRIO_USER || which < PRIO_PROCESS)
219 		return -EINVAL;
220 
221 	read_lock(&tasklist_lock);
222 	switch (which) {
223 		case PRIO_PROCESS:
224 			if (who)
225 				p = find_task_by_vpid(who);
226 			else
227 				p = current;
228 			if (p) {
229 				niceval = 20 - task_nice(p);
230 				if (niceval > retval)
231 					retval = niceval;
232 			}
233 			break;
234 		case PRIO_PGRP:
235 			if (who)
236 				pgrp = find_vpid(who);
237 			else
238 				pgrp = task_pgrp(current);
239 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
240 				niceval = 20 - task_nice(p);
241 				if (niceval > retval)
242 					retval = niceval;
243 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
244 			break;
245 		case PRIO_USER:
246 			user = (struct user_struct *) cred->user;
247 			if (!who)
248 				who = cred->uid;
249 			else if ((who != cred->uid) &&
250 				 !(user = find_user(who)))
251 				goto out_unlock;	/* No processes for this user */
252 
253 			do_each_thread(g, p)
254 				if (__task_cred(p)->uid == who) {
255 					niceval = 20 - task_nice(p);
256 					if (niceval > retval)
257 						retval = niceval;
258 				}
259 			while_each_thread(g, p);
260 			if (who != cred->uid)
261 				free_uid(user);		/* for find_user() */
262 			break;
263 	}
264 out_unlock:
265 	read_unlock(&tasklist_lock);
266 
267 	return retval;
268 }
269 
270 /**
271  *	emergency_restart - reboot the system
272  *
273  *	Without shutting down any hardware or taking any locks
274  *	reboot the system.  This is called when we know we are in
275  *	trouble so this is our best effort to reboot.  This is
276  *	safe to call in interrupt context.
277  */
278 void emergency_restart(void)
279 {
280 	machine_emergency_restart();
281 }
282 EXPORT_SYMBOL_GPL(emergency_restart);
283 
284 void kernel_restart_prepare(char *cmd)
285 {
286 	blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
287 	system_state = SYSTEM_RESTART;
288 	device_shutdown();
289 	sysdev_shutdown();
290 }
291 
292 /**
293  *	kernel_restart - reboot the system
294  *	@cmd: pointer to buffer containing command to execute for restart
295  *		or %NULL
296  *
297  *	Shutdown everything and perform a clean reboot.
298  *	This is not safe to call in interrupt context.
299  */
300 void kernel_restart(char *cmd)
301 {
302 	kernel_restart_prepare(cmd);
303 	if (!cmd)
304 		printk(KERN_EMERG "Restarting system.\n");
305 	else
306 		printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
307 	machine_restart(cmd);
308 }
309 EXPORT_SYMBOL_GPL(kernel_restart);
310 
311 static void kernel_shutdown_prepare(enum system_states state)
312 {
313 	blocking_notifier_call_chain(&reboot_notifier_list,
314 		(state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
315 	system_state = state;
316 	device_shutdown();
317 }
318 /**
319  *	kernel_halt - halt the system
320  *
321  *	Shutdown everything and perform a clean system halt.
322  */
323 void kernel_halt(void)
324 {
325 	kernel_shutdown_prepare(SYSTEM_HALT);
326 	sysdev_shutdown();
327 	printk(KERN_EMERG "System halted.\n");
328 	machine_halt();
329 }
330 
331 EXPORT_SYMBOL_GPL(kernel_halt);
332 
333 /**
334  *	kernel_power_off - power_off the system
335  *
336  *	Shutdown everything and perform a clean system power_off.
337  */
338 void kernel_power_off(void)
339 {
340 	kernel_shutdown_prepare(SYSTEM_POWER_OFF);
341 	if (pm_power_off_prepare)
342 		pm_power_off_prepare();
343 	disable_nonboot_cpus();
344 	sysdev_shutdown();
345 	printk(KERN_EMERG "Power down.\n");
346 	machine_power_off();
347 }
348 EXPORT_SYMBOL_GPL(kernel_power_off);
349 /*
350  * Reboot system call: for obvious reasons only root may call it,
351  * and even root needs to set up some magic numbers in the registers
352  * so that some mistake won't make this reboot the whole machine.
353  * You can also set the meaning of the ctrl-alt-del-key here.
354  *
355  * reboot doesn't sync: do that yourself before calling this.
356  */
357 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
358 {
359 	char buffer[256];
360 
361 	/* We only trust the superuser with rebooting the system. */
362 	if (!capable(CAP_SYS_BOOT))
363 		return -EPERM;
364 
365 	/* For safety, we require "magic" arguments. */
366 	if (magic1 != LINUX_REBOOT_MAGIC1 ||
367 	    (magic2 != LINUX_REBOOT_MAGIC2 &&
368 	                magic2 != LINUX_REBOOT_MAGIC2A &&
369 			magic2 != LINUX_REBOOT_MAGIC2B &&
370 	                magic2 != LINUX_REBOOT_MAGIC2C))
371 		return -EINVAL;
372 
373 	/* Instead of trying to make the power_off code look like
374 	 * halt when pm_power_off is not set do it the easy way.
375 	 */
376 	if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
377 		cmd = LINUX_REBOOT_CMD_HALT;
378 
379 	lock_kernel();
380 	switch (cmd) {
381 	case LINUX_REBOOT_CMD_RESTART:
382 		kernel_restart(NULL);
383 		break;
384 
385 	case LINUX_REBOOT_CMD_CAD_ON:
386 		C_A_D = 1;
387 		break;
388 
389 	case LINUX_REBOOT_CMD_CAD_OFF:
390 		C_A_D = 0;
391 		break;
392 
393 	case LINUX_REBOOT_CMD_HALT:
394 		kernel_halt();
395 		unlock_kernel();
396 		do_exit(0);
397 		break;
398 
399 	case LINUX_REBOOT_CMD_POWER_OFF:
400 		kernel_power_off();
401 		unlock_kernel();
402 		do_exit(0);
403 		break;
404 
405 	case LINUX_REBOOT_CMD_RESTART2:
406 		if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
407 			unlock_kernel();
408 			return -EFAULT;
409 		}
410 		buffer[sizeof(buffer) - 1] = '\0';
411 
412 		kernel_restart(buffer);
413 		break;
414 
415 #ifdef CONFIG_KEXEC
416 	case LINUX_REBOOT_CMD_KEXEC:
417 		{
418 			int ret;
419 			ret = kernel_kexec();
420 			unlock_kernel();
421 			return ret;
422 		}
423 #endif
424 
425 #ifdef CONFIG_HIBERNATION
426 	case LINUX_REBOOT_CMD_SW_SUSPEND:
427 		{
428 			int ret = hibernate();
429 			unlock_kernel();
430 			return ret;
431 		}
432 #endif
433 
434 	default:
435 		unlock_kernel();
436 		return -EINVAL;
437 	}
438 	unlock_kernel();
439 	return 0;
440 }
441 
442 static void deferred_cad(struct work_struct *dummy)
443 {
444 	kernel_restart(NULL);
445 }
446 
447 /*
448  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
449  * As it's called within an interrupt, it may NOT sync: the only choice
450  * is whether to reboot at once, or just ignore the ctrl-alt-del.
451  */
452 void ctrl_alt_del(void)
453 {
454 	static DECLARE_WORK(cad_work, deferred_cad);
455 
456 	if (C_A_D)
457 		schedule_work(&cad_work);
458 	else
459 		kill_cad_pid(SIGINT, 1);
460 }
461 
462 /*
463  * Unprivileged users may change the real gid to the effective gid
464  * or vice versa.  (BSD-style)
465  *
466  * If you set the real gid at all, or set the effective gid to a value not
467  * equal to the real gid, then the saved gid is set to the new effective gid.
468  *
469  * This makes it possible for a setgid program to completely drop its
470  * privileges, which is often a useful assertion to make when you are doing
471  * a security audit over a program.
472  *
473  * The general idea is that a program which uses just setregid() will be
474  * 100% compatible with BSD.  A program which uses just setgid() will be
475  * 100% compatible with POSIX with saved IDs.
476  *
477  * SMP: There are not races, the GIDs are checked only by filesystem
478  *      operations (as far as semantic preservation is concerned).
479  */
480 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
481 {
482 	const struct cred *old;
483 	struct cred *new;
484 	int retval;
485 
486 	new = prepare_creds();
487 	if (!new)
488 		return -ENOMEM;
489 	old = current_cred();
490 
491 	retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
492 	if (retval)
493 		goto error;
494 
495 	retval = -EPERM;
496 	if (rgid != (gid_t) -1) {
497 		if (old->gid == rgid ||
498 		    old->egid == rgid ||
499 		    capable(CAP_SETGID))
500 			new->gid = rgid;
501 		else
502 			goto error;
503 	}
504 	if (egid != (gid_t) -1) {
505 		if (old->gid == egid ||
506 		    old->egid == egid ||
507 		    old->sgid == egid ||
508 		    capable(CAP_SETGID))
509 			new->egid = egid;
510 		else
511 			goto error;
512 	}
513 
514 	if (rgid != (gid_t) -1 ||
515 	    (egid != (gid_t) -1 && egid != old->gid))
516 		new->sgid = new->egid;
517 	new->fsgid = new->egid;
518 
519 	return commit_creds(new);
520 
521 error:
522 	abort_creds(new);
523 	return retval;
524 }
525 
526 /*
527  * setgid() is implemented like SysV w/ SAVED_IDS
528  *
529  * SMP: Same implicit races as above.
530  */
531 asmlinkage long sys_setgid(gid_t gid)
532 {
533 	const struct cred *old;
534 	struct cred *new;
535 	int retval;
536 
537 	new = prepare_creds();
538 	if (!new)
539 		return -ENOMEM;
540 	old = current_cred();
541 
542 	retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
543 	if (retval)
544 		goto error;
545 
546 	retval = -EPERM;
547 	if (capable(CAP_SETGID))
548 		new->gid = new->egid = new->sgid = new->fsgid = gid;
549 	else if (gid == old->gid || gid == old->sgid)
550 		new->egid = new->fsgid = gid;
551 	else
552 		goto error;
553 
554 	return commit_creds(new);
555 
556 error:
557 	abort_creds(new);
558 	return retval;
559 }
560 
561 /*
562  * change the user struct in a credentials set to match the new UID
563  */
564 static int set_user(struct cred *new)
565 {
566 	struct user_struct *new_user;
567 
568 	new_user = alloc_uid(current_user_ns(), new->uid);
569 	if (!new_user)
570 		return -EAGAIN;
571 
572 	if (atomic_read(&new_user->processes) >=
573 				current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
574 			new_user != INIT_USER) {
575 		free_uid(new_user);
576 		return -EAGAIN;
577 	}
578 
579 	free_uid(new->user);
580 	new->user = new_user;
581 	return 0;
582 }
583 
584 /*
585  * Unprivileged users may change the real uid to the effective uid
586  * or vice versa.  (BSD-style)
587  *
588  * If you set the real uid at all, or set the effective uid to a value not
589  * equal to the real uid, then the saved uid is set to the new effective uid.
590  *
591  * This makes it possible for a setuid program to completely drop its
592  * privileges, which is often a useful assertion to make when you are doing
593  * a security audit over a program.
594  *
595  * The general idea is that a program which uses just setreuid() will be
596  * 100% compatible with BSD.  A program which uses just setuid() will be
597  * 100% compatible with POSIX with saved IDs.
598  */
599 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
600 {
601 	const struct cred *old;
602 	struct cred *new;
603 	int retval;
604 
605 	new = prepare_creds();
606 	if (!new)
607 		return -ENOMEM;
608 	old = current_cred();
609 
610 	retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
611 	if (retval)
612 		goto error;
613 
614 	retval = -EPERM;
615 	if (ruid != (uid_t) -1) {
616 		new->uid = ruid;
617 		if (old->uid != ruid &&
618 		    old->euid != ruid &&
619 		    !capable(CAP_SETUID))
620 			goto error;
621 	}
622 
623 	if (euid != (uid_t) -1) {
624 		new->euid = euid;
625 		if (old->uid != euid &&
626 		    old->euid != euid &&
627 		    old->suid != euid &&
628 		    !capable(CAP_SETUID))
629 			goto error;
630 	}
631 
632 	retval = -EAGAIN;
633 	if (new->uid != old->uid && set_user(new) < 0)
634 		goto error;
635 
636 	if (ruid != (uid_t) -1 ||
637 	    (euid != (uid_t) -1 && euid != old->uid))
638 		new->suid = new->euid;
639 	new->fsuid = new->euid;
640 
641 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
642 	if (retval < 0)
643 		goto error;
644 
645 	return commit_creds(new);
646 
647 error:
648 	abort_creds(new);
649 	return retval;
650 }
651 
652 /*
653  * setuid() is implemented like SysV with SAVED_IDS
654  *
655  * Note that SAVED_ID's is deficient in that a setuid root program
656  * like sendmail, for example, cannot set its uid to be a normal
657  * user and then switch back, because if you're root, setuid() sets
658  * the saved uid too.  If you don't like this, blame the bright people
659  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
660  * will allow a root program to temporarily drop privileges and be able to
661  * regain them by swapping the real and effective uid.
662  */
663 asmlinkage long sys_setuid(uid_t uid)
664 {
665 	const struct cred *old;
666 	struct cred *new;
667 	int retval;
668 
669 	new = prepare_creds();
670 	if (!new)
671 		return -ENOMEM;
672 	old = current_cred();
673 
674 	retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
675 	if (retval)
676 		goto error;
677 
678 	retval = -EPERM;
679 	if (capable(CAP_SETUID)) {
680 		new->suid = new->uid = uid;
681 		if (uid != old->uid && set_user(new) < 0) {
682 			retval = -EAGAIN;
683 			goto error;
684 		}
685 	} else if (uid != old->uid && uid != new->suid) {
686 		goto error;
687 	}
688 
689 	new->fsuid = new->euid = uid;
690 
691 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
692 	if (retval < 0)
693 		goto error;
694 
695 	return commit_creds(new);
696 
697 error:
698 	abort_creds(new);
699 	return retval;
700 }
701 
702 
703 /*
704  * This function implements a generic ability to update ruid, euid,
705  * and suid.  This allows you to implement the 4.4 compatible seteuid().
706  */
707 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
708 {
709 	const struct cred *old;
710 	struct cred *new;
711 	int retval;
712 
713 	new = prepare_creds();
714 	if (!new)
715 		return -ENOMEM;
716 
717 	retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
718 	if (retval)
719 		goto error;
720 	old = current_cred();
721 
722 	retval = -EPERM;
723 	if (!capable(CAP_SETUID)) {
724 		if (ruid != (uid_t) -1 && ruid != old->uid &&
725 		    ruid != old->euid  && ruid != old->suid)
726 			goto error;
727 		if (euid != (uid_t) -1 && euid != old->uid &&
728 		    euid != old->euid  && euid != old->suid)
729 			goto error;
730 		if (suid != (uid_t) -1 && suid != old->uid &&
731 		    suid != old->euid  && suid != old->suid)
732 			goto error;
733 	}
734 
735 	retval = -EAGAIN;
736 	if (ruid != (uid_t) -1) {
737 		new->uid = ruid;
738 		if (ruid != old->uid && set_user(new) < 0)
739 			goto error;
740 	}
741 	if (euid != (uid_t) -1)
742 		new->euid = euid;
743 	if (suid != (uid_t) -1)
744 		new->suid = suid;
745 	new->fsuid = new->euid;
746 
747 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
748 	if (retval < 0)
749 		goto error;
750 
751 	return commit_creds(new);
752 
753 error:
754 	abort_creds(new);
755 	return retval;
756 }
757 
758 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
759 {
760 	const struct cred *cred = current_cred();
761 	int retval;
762 
763 	if (!(retval   = put_user(cred->uid,  ruid)) &&
764 	    !(retval   = put_user(cred->euid, euid)))
765 		retval = put_user(cred->suid, suid);
766 
767 	return retval;
768 }
769 
770 /*
771  * Same as above, but for rgid, egid, sgid.
772  */
773 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
774 {
775 	const struct cred *old;
776 	struct cred *new;
777 	int retval;
778 
779 	new = prepare_creds();
780 	if (!new)
781 		return -ENOMEM;
782 	old = current_cred();
783 
784 	retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
785 	if (retval)
786 		goto error;
787 
788 	retval = -EPERM;
789 	if (!capable(CAP_SETGID)) {
790 		if (rgid != (gid_t) -1 && rgid != old->gid &&
791 		    rgid != old->egid  && rgid != old->sgid)
792 			goto error;
793 		if (egid != (gid_t) -1 && egid != old->gid &&
794 		    egid != old->egid  && egid != old->sgid)
795 			goto error;
796 		if (sgid != (gid_t) -1 && sgid != old->gid &&
797 		    sgid != old->egid  && sgid != old->sgid)
798 			goto error;
799 	}
800 
801 	if (rgid != (gid_t) -1)
802 		new->gid = rgid;
803 	if (egid != (gid_t) -1)
804 		new->egid = egid;
805 	if (sgid != (gid_t) -1)
806 		new->sgid = sgid;
807 	new->fsgid = new->egid;
808 
809 	return commit_creds(new);
810 
811 error:
812 	abort_creds(new);
813 	return retval;
814 }
815 
816 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
817 {
818 	const struct cred *cred = current_cred();
819 	int retval;
820 
821 	if (!(retval   = put_user(cred->gid,  rgid)) &&
822 	    !(retval   = put_user(cred->egid, egid)))
823 		retval = put_user(cred->sgid, sgid);
824 
825 	return retval;
826 }
827 
828 
829 /*
830  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
831  * is used for "access()" and for the NFS daemon (letting nfsd stay at
832  * whatever uid it wants to). It normally shadows "euid", except when
833  * explicitly set by setfsuid() or for access..
834  */
835 asmlinkage long sys_setfsuid(uid_t uid)
836 {
837 	const struct cred *old;
838 	struct cred *new;
839 	uid_t old_fsuid;
840 
841 	new = prepare_creds();
842 	if (!new)
843 		return current_fsuid();
844 	old = current_cred();
845 	old_fsuid = old->fsuid;
846 
847 	if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
848 		goto error;
849 
850 	if (uid == old->uid  || uid == old->euid  ||
851 	    uid == old->suid || uid == old->fsuid ||
852 	    capable(CAP_SETUID)) {
853 		if (uid != old_fsuid) {
854 			new->fsuid = uid;
855 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
856 				goto change_okay;
857 		}
858 	}
859 
860 error:
861 	abort_creds(new);
862 	return old_fsuid;
863 
864 change_okay:
865 	commit_creds(new);
866 	return old_fsuid;
867 }
868 
869 /*
870  * Samma på svenska..
871  */
872 asmlinkage long sys_setfsgid(gid_t gid)
873 {
874 	const struct cred *old;
875 	struct cred *new;
876 	gid_t old_fsgid;
877 
878 	new = prepare_creds();
879 	if (!new)
880 		return current_fsgid();
881 	old = current_cred();
882 	old_fsgid = old->fsgid;
883 
884 	if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
885 		goto error;
886 
887 	if (gid == old->gid  || gid == old->egid  ||
888 	    gid == old->sgid || gid == old->fsgid ||
889 	    capable(CAP_SETGID)) {
890 		if (gid != old_fsgid) {
891 			new->fsgid = gid;
892 			goto change_okay;
893 		}
894 	}
895 
896 error:
897 	abort_creds(new);
898 	return old_fsgid;
899 
900 change_okay:
901 	commit_creds(new);
902 	return old_fsgid;
903 }
904 
905 void do_sys_times(struct tms *tms)
906 {
907 	struct task_cputime cputime;
908 	cputime_t cutime, cstime;
909 
910 	thread_group_cputime(current, &cputime);
911 	spin_lock_irq(&current->sighand->siglock);
912 	cutime = current->signal->cutime;
913 	cstime = current->signal->cstime;
914 	spin_unlock_irq(&current->sighand->siglock);
915 	tms->tms_utime = cputime_to_clock_t(cputime.utime);
916 	tms->tms_stime = cputime_to_clock_t(cputime.stime);
917 	tms->tms_cutime = cputime_to_clock_t(cutime);
918 	tms->tms_cstime = cputime_to_clock_t(cstime);
919 }
920 
921 asmlinkage long sys_times(struct tms __user * tbuf)
922 {
923 	if (tbuf) {
924 		struct tms tmp;
925 
926 		do_sys_times(&tmp);
927 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
928 			return -EFAULT;
929 	}
930 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
931 }
932 
933 /*
934  * This needs some heavy checking ...
935  * I just haven't the stomach for it. I also don't fully
936  * understand sessions/pgrp etc. Let somebody who does explain it.
937  *
938  * OK, I think I have the protection semantics right.... this is really
939  * only important on a multi-user system anyway, to make sure one user
940  * can't send a signal to a process owned by another.  -TYT, 12/12/91
941  *
942  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
943  * LBT 04.03.94
944  */
945 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
946 {
947 	struct task_struct *p;
948 	struct task_struct *group_leader = current->group_leader;
949 	struct pid *pgrp;
950 	int err;
951 
952 	if (!pid)
953 		pid = task_pid_vnr(group_leader);
954 	if (!pgid)
955 		pgid = pid;
956 	if (pgid < 0)
957 		return -EINVAL;
958 
959 	/* From this point forward we keep holding onto the tasklist lock
960 	 * so that our parent does not change from under us. -DaveM
961 	 */
962 	write_lock_irq(&tasklist_lock);
963 
964 	err = -ESRCH;
965 	p = find_task_by_vpid(pid);
966 	if (!p)
967 		goto out;
968 
969 	err = -EINVAL;
970 	if (!thread_group_leader(p))
971 		goto out;
972 
973 	if (same_thread_group(p->real_parent, group_leader)) {
974 		err = -EPERM;
975 		if (task_session(p) != task_session(group_leader))
976 			goto out;
977 		err = -EACCES;
978 		if (p->did_exec)
979 			goto out;
980 	} else {
981 		err = -ESRCH;
982 		if (p != group_leader)
983 			goto out;
984 	}
985 
986 	err = -EPERM;
987 	if (p->signal->leader)
988 		goto out;
989 
990 	pgrp = task_pid(p);
991 	if (pgid != pid) {
992 		struct task_struct *g;
993 
994 		pgrp = find_vpid(pgid);
995 		g = pid_task(pgrp, PIDTYPE_PGID);
996 		if (!g || task_session(g) != task_session(group_leader))
997 			goto out;
998 	}
999 
1000 	err = security_task_setpgid(p, pgid);
1001 	if (err)
1002 		goto out;
1003 
1004 	if (task_pgrp(p) != pgrp) {
1005 		change_pid(p, PIDTYPE_PGID, pgrp);
1006 		set_task_pgrp(p, pid_nr(pgrp));
1007 	}
1008 
1009 	err = 0;
1010 out:
1011 	/* All paths lead to here, thus we are safe. -DaveM */
1012 	write_unlock_irq(&tasklist_lock);
1013 	return err;
1014 }
1015 
1016 asmlinkage long sys_getpgid(pid_t pid)
1017 {
1018 	struct task_struct *p;
1019 	struct pid *grp;
1020 	int retval;
1021 
1022 	rcu_read_lock();
1023 	if (!pid)
1024 		grp = task_pgrp(current);
1025 	else {
1026 		retval = -ESRCH;
1027 		p = find_task_by_vpid(pid);
1028 		if (!p)
1029 			goto out;
1030 		grp = task_pgrp(p);
1031 		if (!grp)
1032 			goto out;
1033 
1034 		retval = security_task_getpgid(p);
1035 		if (retval)
1036 			goto out;
1037 	}
1038 	retval = pid_vnr(grp);
1039 out:
1040 	rcu_read_unlock();
1041 	return retval;
1042 }
1043 
1044 #ifdef __ARCH_WANT_SYS_GETPGRP
1045 
1046 asmlinkage long sys_getpgrp(void)
1047 {
1048 	return sys_getpgid(0);
1049 }
1050 
1051 #endif
1052 
1053 asmlinkage long sys_getsid(pid_t pid)
1054 {
1055 	struct task_struct *p;
1056 	struct pid *sid;
1057 	int retval;
1058 
1059 	rcu_read_lock();
1060 	if (!pid)
1061 		sid = task_session(current);
1062 	else {
1063 		retval = -ESRCH;
1064 		p = find_task_by_vpid(pid);
1065 		if (!p)
1066 			goto out;
1067 		sid = task_session(p);
1068 		if (!sid)
1069 			goto out;
1070 
1071 		retval = security_task_getsid(p);
1072 		if (retval)
1073 			goto out;
1074 	}
1075 	retval = pid_vnr(sid);
1076 out:
1077 	rcu_read_unlock();
1078 	return retval;
1079 }
1080 
1081 asmlinkage long sys_setsid(void)
1082 {
1083 	struct task_struct *group_leader = current->group_leader;
1084 	struct pid *sid = task_pid(group_leader);
1085 	pid_t session = pid_vnr(sid);
1086 	int err = -EPERM;
1087 
1088 	write_lock_irq(&tasklist_lock);
1089 	/* Fail if I am already a session leader */
1090 	if (group_leader->signal->leader)
1091 		goto out;
1092 
1093 	/* Fail if a process group id already exists that equals the
1094 	 * proposed session id.
1095 	 */
1096 	if (pid_task(sid, PIDTYPE_PGID))
1097 		goto out;
1098 
1099 	group_leader->signal->leader = 1;
1100 	__set_special_pids(sid);
1101 
1102 	proc_clear_tty(group_leader);
1103 
1104 	err = session;
1105 out:
1106 	write_unlock_irq(&tasklist_lock);
1107 	return err;
1108 }
1109 
1110 /*
1111  * Supplementary group IDs
1112  */
1113 
1114 /* init to 2 - one for init_task, one to ensure it is never freed */
1115 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1116 
1117 struct group_info *groups_alloc(int gidsetsize)
1118 {
1119 	struct group_info *group_info;
1120 	int nblocks;
1121 	int i;
1122 
1123 	nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1124 	/* Make sure we always allocate at least one indirect block pointer */
1125 	nblocks = nblocks ? : 1;
1126 	group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1127 	if (!group_info)
1128 		return NULL;
1129 	group_info->ngroups = gidsetsize;
1130 	group_info->nblocks = nblocks;
1131 	atomic_set(&group_info->usage, 1);
1132 
1133 	if (gidsetsize <= NGROUPS_SMALL)
1134 		group_info->blocks[0] = group_info->small_block;
1135 	else {
1136 		for (i = 0; i < nblocks; i++) {
1137 			gid_t *b;
1138 			b = (void *)__get_free_page(GFP_USER);
1139 			if (!b)
1140 				goto out_undo_partial_alloc;
1141 			group_info->blocks[i] = b;
1142 		}
1143 	}
1144 	return group_info;
1145 
1146 out_undo_partial_alloc:
1147 	while (--i >= 0) {
1148 		free_page((unsigned long)group_info->blocks[i]);
1149 	}
1150 	kfree(group_info);
1151 	return NULL;
1152 }
1153 
1154 EXPORT_SYMBOL(groups_alloc);
1155 
1156 void groups_free(struct group_info *group_info)
1157 {
1158 	if (group_info->blocks[0] != group_info->small_block) {
1159 		int i;
1160 		for (i = 0; i < group_info->nblocks; i++)
1161 			free_page((unsigned long)group_info->blocks[i]);
1162 	}
1163 	kfree(group_info);
1164 }
1165 
1166 EXPORT_SYMBOL(groups_free);
1167 
1168 /* export the group_info to a user-space array */
1169 static int groups_to_user(gid_t __user *grouplist,
1170 			  const struct group_info *group_info)
1171 {
1172 	int i;
1173 	unsigned int count = group_info->ngroups;
1174 
1175 	for (i = 0; i < group_info->nblocks; i++) {
1176 		unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1177 		unsigned int len = cp_count * sizeof(*grouplist);
1178 
1179 		if (copy_to_user(grouplist, group_info->blocks[i], len))
1180 			return -EFAULT;
1181 
1182 		grouplist += NGROUPS_PER_BLOCK;
1183 		count -= cp_count;
1184 	}
1185 	return 0;
1186 }
1187 
1188 /* fill a group_info from a user-space array - it must be allocated already */
1189 static int groups_from_user(struct group_info *group_info,
1190     gid_t __user *grouplist)
1191 {
1192 	int i;
1193 	unsigned int count = group_info->ngroups;
1194 
1195 	for (i = 0; i < group_info->nblocks; i++) {
1196 		unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1197 		unsigned int len = cp_count * sizeof(*grouplist);
1198 
1199 		if (copy_from_user(group_info->blocks[i], grouplist, len))
1200 			return -EFAULT;
1201 
1202 		grouplist += NGROUPS_PER_BLOCK;
1203 		count -= cp_count;
1204 	}
1205 	return 0;
1206 }
1207 
1208 /* a simple Shell sort */
1209 static void groups_sort(struct group_info *group_info)
1210 {
1211 	int base, max, stride;
1212 	int gidsetsize = group_info->ngroups;
1213 
1214 	for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1215 		; /* nothing */
1216 	stride /= 3;
1217 
1218 	while (stride) {
1219 		max = gidsetsize - stride;
1220 		for (base = 0; base < max; base++) {
1221 			int left = base;
1222 			int right = left + stride;
1223 			gid_t tmp = GROUP_AT(group_info, right);
1224 
1225 			while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1226 				GROUP_AT(group_info, right) =
1227 				    GROUP_AT(group_info, left);
1228 				right = left;
1229 				left -= stride;
1230 			}
1231 			GROUP_AT(group_info, right) = tmp;
1232 		}
1233 		stride /= 3;
1234 	}
1235 }
1236 
1237 /* a simple bsearch */
1238 int groups_search(const struct group_info *group_info, gid_t grp)
1239 {
1240 	unsigned int left, right;
1241 
1242 	if (!group_info)
1243 		return 0;
1244 
1245 	left = 0;
1246 	right = group_info->ngroups;
1247 	while (left < right) {
1248 		unsigned int mid = (left+right)/2;
1249 		int cmp = grp - GROUP_AT(group_info, mid);
1250 		if (cmp > 0)
1251 			left = mid + 1;
1252 		else if (cmp < 0)
1253 			right = mid;
1254 		else
1255 			return 1;
1256 	}
1257 	return 0;
1258 }
1259 
1260 /**
1261  * set_groups - Change a group subscription in a set of credentials
1262  * @new: The newly prepared set of credentials to alter
1263  * @group_info: The group list to install
1264  *
1265  * Validate a group subscription and, if valid, insert it into a set
1266  * of credentials.
1267  */
1268 int set_groups(struct cred *new, struct group_info *group_info)
1269 {
1270 	int retval;
1271 
1272 	retval = security_task_setgroups(group_info);
1273 	if (retval)
1274 		return retval;
1275 
1276 	put_group_info(new->group_info);
1277 	groups_sort(group_info);
1278 	get_group_info(group_info);
1279 	new->group_info = group_info;
1280 	return 0;
1281 }
1282 
1283 EXPORT_SYMBOL(set_groups);
1284 
1285 /**
1286  * set_current_groups - Change current's group subscription
1287  * @group_info: The group list to impose
1288  *
1289  * Validate a group subscription and, if valid, impose it upon current's task
1290  * security record.
1291  */
1292 int set_current_groups(struct group_info *group_info)
1293 {
1294 	struct cred *new;
1295 	int ret;
1296 
1297 	new = prepare_creds();
1298 	if (!new)
1299 		return -ENOMEM;
1300 
1301 	ret = set_groups(new, group_info);
1302 	if (ret < 0) {
1303 		abort_creds(new);
1304 		return ret;
1305 	}
1306 
1307 	return commit_creds(new);
1308 }
1309 
1310 EXPORT_SYMBOL(set_current_groups);
1311 
1312 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1313 {
1314 	const struct cred *cred = current_cred();
1315 	int i;
1316 
1317 	if (gidsetsize < 0)
1318 		return -EINVAL;
1319 
1320 	/* no need to grab task_lock here; it cannot change */
1321 	i = cred->group_info->ngroups;
1322 	if (gidsetsize) {
1323 		if (i > gidsetsize) {
1324 			i = -EINVAL;
1325 			goto out;
1326 		}
1327 		if (groups_to_user(grouplist, cred->group_info)) {
1328 			i = -EFAULT;
1329 			goto out;
1330 		}
1331 	}
1332 out:
1333 	return i;
1334 }
1335 
1336 /*
1337  *	SMP: Our groups are copy-on-write. We can set them safely
1338  *	without another task interfering.
1339  */
1340 
1341 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1342 {
1343 	struct group_info *group_info;
1344 	int retval;
1345 
1346 	if (!capable(CAP_SETGID))
1347 		return -EPERM;
1348 	if ((unsigned)gidsetsize > NGROUPS_MAX)
1349 		return -EINVAL;
1350 
1351 	group_info = groups_alloc(gidsetsize);
1352 	if (!group_info)
1353 		return -ENOMEM;
1354 	retval = groups_from_user(group_info, grouplist);
1355 	if (retval) {
1356 		put_group_info(group_info);
1357 		return retval;
1358 	}
1359 
1360 	retval = set_current_groups(group_info);
1361 	put_group_info(group_info);
1362 
1363 	return retval;
1364 }
1365 
1366 /*
1367  * Check whether we're fsgid/egid or in the supplemental group..
1368  */
1369 int in_group_p(gid_t grp)
1370 {
1371 	const struct cred *cred = current_cred();
1372 	int retval = 1;
1373 
1374 	if (grp != cred->fsgid)
1375 		retval = groups_search(cred->group_info, grp);
1376 	return retval;
1377 }
1378 
1379 EXPORT_SYMBOL(in_group_p);
1380 
1381 int in_egroup_p(gid_t grp)
1382 {
1383 	const struct cred *cred = current_cred();
1384 	int retval = 1;
1385 
1386 	if (grp != cred->egid)
1387 		retval = groups_search(cred->group_info, grp);
1388 	return retval;
1389 }
1390 
1391 EXPORT_SYMBOL(in_egroup_p);
1392 
1393 DECLARE_RWSEM(uts_sem);
1394 
1395 asmlinkage long sys_newuname(struct new_utsname __user * name)
1396 {
1397 	int errno = 0;
1398 
1399 	down_read(&uts_sem);
1400 	if (copy_to_user(name, utsname(), sizeof *name))
1401 		errno = -EFAULT;
1402 	up_read(&uts_sem);
1403 	return errno;
1404 }
1405 
1406 asmlinkage long sys_sethostname(char __user *name, int len)
1407 {
1408 	int errno;
1409 	char tmp[__NEW_UTS_LEN];
1410 
1411 	if (!capable(CAP_SYS_ADMIN))
1412 		return -EPERM;
1413 	if (len < 0 || len > __NEW_UTS_LEN)
1414 		return -EINVAL;
1415 	down_write(&uts_sem);
1416 	errno = -EFAULT;
1417 	if (!copy_from_user(tmp, name, len)) {
1418 		struct new_utsname *u = utsname();
1419 
1420 		memcpy(u->nodename, tmp, len);
1421 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1422 		errno = 0;
1423 	}
1424 	up_write(&uts_sem);
1425 	return errno;
1426 }
1427 
1428 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1429 
1430 asmlinkage long sys_gethostname(char __user *name, int len)
1431 {
1432 	int i, errno;
1433 	struct new_utsname *u;
1434 
1435 	if (len < 0)
1436 		return -EINVAL;
1437 	down_read(&uts_sem);
1438 	u = utsname();
1439 	i = 1 + strlen(u->nodename);
1440 	if (i > len)
1441 		i = len;
1442 	errno = 0;
1443 	if (copy_to_user(name, u->nodename, i))
1444 		errno = -EFAULT;
1445 	up_read(&uts_sem);
1446 	return errno;
1447 }
1448 
1449 #endif
1450 
1451 /*
1452  * Only setdomainname; getdomainname can be implemented by calling
1453  * uname()
1454  */
1455 asmlinkage long sys_setdomainname(char __user *name, int len)
1456 {
1457 	int errno;
1458 	char tmp[__NEW_UTS_LEN];
1459 
1460 	if (!capable(CAP_SYS_ADMIN))
1461 		return -EPERM;
1462 	if (len < 0 || len > __NEW_UTS_LEN)
1463 		return -EINVAL;
1464 
1465 	down_write(&uts_sem);
1466 	errno = -EFAULT;
1467 	if (!copy_from_user(tmp, name, len)) {
1468 		struct new_utsname *u = utsname();
1469 
1470 		memcpy(u->domainname, tmp, len);
1471 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1472 		errno = 0;
1473 	}
1474 	up_write(&uts_sem);
1475 	return errno;
1476 }
1477 
1478 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1479 {
1480 	if (resource >= RLIM_NLIMITS)
1481 		return -EINVAL;
1482 	else {
1483 		struct rlimit value;
1484 		task_lock(current->group_leader);
1485 		value = current->signal->rlim[resource];
1486 		task_unlock(current->group_leader);
1487 		return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1488 	}
1489 }
1490 
1491 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1492 
1493 /*
1494  *	Back compatibility for getrlimit. Needed for some apps.
1495  */
1496 
1497 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1498 {
1499 	struct rlimit x;
1500 	if (resource >= RLIM_NLIMITS)
1501 		return -EINVAL;
1502 
1503 	task_lock(current->group_leader);
1504 	x = current->signal->rlim[resource];
1505 	task_unlock(current->group_leader);
1506 	if (x.rlim_cur > 0x7FFFFFFF)
1507 		x.rlim_cur = 0x7FFFFFFF;
1508 	if (x.rlim_max > 0x7FFFFFFF)
1509 		x.rlim_max = 0x7FFFFFFF;
1510 	return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1511 }
1512 
1513 #endif
1514 
1515 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1516 {
1517 	struct rlimit new_rlim, *old_rlim;
1518 	int retval;
1519 
1520 	if (resource >= RLIM_NLIMITS)
1521 		return -EINVAL;
1522 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1523 		return -EFAULT;
1524 	old_rlim = current->signal->rlim + resource;
1525 	if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1526 	    !capable(CAP_SYS_RESOURCE))
1527 		return -EPERM;
1528 
1529 	if (resource == RLIMIT_NOFILE) {
1530 		if (new_rlim.rlim_max == RLIM_INFINITY)
1531 			new_rlim.rlim_max = sysctl_nr_open;
1532 		if (new_rlim.rlim_cur == RLIM_INFINITY)
1533 			new_rlim.rlim_cur = sysctl_nr_open;
1534 		if (new_rlim.rlim_max > sysctl_nr_open)
1535 			return -EPERM;
1536 	}
1537 
1538 	if (new_rlim.rlim_cur > new_rlim.rlim_max)
1539 		return -EINVAL;
1540 
1541 	retval = security_task_setrlimit(resource, &new_rlim);
1542 	if (retval)
1543 		return retval;
1544 
1545 	if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1546 		/*
1547 		 * The caller is asking for an immediate RLIMIT_CPU
1548 		 * expiry.  But we use the zero value to mean "it was
1549 		 * never set".  So let's cheat and make it one second
1550 		 * instead
1551 		 */
1552 		new_rlim.rlim_cur = 1;
1553 	}
1554 
1555 	task_lock(current->group_leader);
1556 	*old_rlim = new_rlim;
1557 	task_unlock(current->group_leader);
1558 
1559 	if (resource != RLIMIT_CPU)
1560 		goto out;
1561 
1562 	/*
1563 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1564 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1565 	 * very long-standing error, and fixing it now risks breakage of
1566 	 * applications, so we live with it
1567 	 */
1568 	if (new_rlim.rlim_cur == RLIM_INFINITY)
1569 		goto out;
1570 
1571 	update_rlimit_cpu(new_rlim.rlim_cur);
1572 out:
1573 	return 0;
1574 }
1575 
1576 /*
1577  * It would make sense to put struct rusage in the task_struct,
1578  * except that would make the task_struct be *really big*.  After
1579  * task_struct gets moved into malloc'ed memory, it would
1580  * make sense to do this.  It will make moving the rest of the information
1581  * a lot simpler!  (Which we're not doing right now because we're not
1582  * measuring them yet).
1583  *
1584  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1585  * races with threads incrementing their own counters.  But since word
1586  * reads are atomic, we either get new values or old values and we don't
1587  * care which for the sums.  We always take the siglock to protect reading
1588  * the c* fields from p->signal from races with exit.c updating those
1589  * fields when reaping, so a sample either gets all the additions of a
1590  * given child after it's reaped, or none so this sample is before reaping.
1591  *
1592  * Locking:
1593  * We need to take the siglock for CHILDEREN, SELF and BOTH
1594  * for  the cases current multithreaded, non-current single threaded
1595  * non-current multithreaded.  Thread traversal is now safe with
1596  * the siglock held.
1597  * Strictly speaking, we donot need to take the siglock if we are current and
1598  * single threaded,  as no one else can take our signal_struct away, no one
1599  * else can  reap the  children to update signal->c* counters, and no one else
1600  * can race with the signal-> fields. If we do not take any lock, the
1601  * signal-> fields could be read out of order while another thread was just
1602  * exiting. So we should  place a read memory barrier when we avoid the lock.
1603  * On the writer side,  write memory barrier is implied in  __exit_signal
1604  * as __exit_signal releases  the siglock spinlock after updating the signal->
1605  * fields. But we don't do this yet to keep things simple.
1606  *
1607  */
1608 
1609 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1610 {
1611 	r->ru_nvcsw += t->nvcsw;
1612 	r->ru_nivcsw += t->nivcsw;
1613 	r->ru_minflt += t->min_flt;
1614 	r->ru_majflt += t->maj_flt;
1615 	r->ru_inblock += task_io_get_inblock(t);
1616 	r->ru_oublock += task_io_get_oublock(t);
1617 }
1618 
1619 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1620 {
1621 	struct task_struct *t;
1622 	unsigned long flags;
1623 	cputime_t utime, stime;
1624 	struct task_cputime cputime;
1625 
1626 	memset((char *) r, 0, sizeof *r);
1627 	utime = stime = cputime_zero;
1628 
1629 	if (who == RUSAGE_THREAD) {
1630 		accumulate_thread_rusage(p, r);
1631 		goto out;
1632 	}
1633 
1634 	if (!lock_task_sighand(p, &flags))
1635 		return;
1636 
1637 	switch (who) {
1638 		case RUSAGE_BOTH:
1639 		case RUSAGE_CHILDREN:
1640 			utime = p->signal->cutime;
1641 			stime = p->signal->cstime;
1642 			r->ru_nvcsw = p->signal->cnvcsw;
1643 			r->ru_nivcsw = p->signal->cnivcsw;
1644 			r->ru_minflt = p->signal->cmin_flt;
1645 			r->ru_majflt = p->signal->cmaj_flt;
1646 			r->ru_inblock = p->signal->cinblock;
1647 			r->ru_oublock = p->signal->coublock;
1648 
1649 			if (who == RUSAGE_CHILDREN)
1650 				break;
1651 
1652 		case RUSAGE_SELF:
1653 			thread_group_cputime(p, &cputime);
1654 			utime = cputime_add(utime, cputime.utime);
1655 			stime = cputime_add(stime, cputime.stime);
1656 			r->ru_nvcsw += p->signal->nvcsw;
1657 			r->ru_nivcsw += p->signal->nivcsw;
1658 			r->ru_minflt += p->signal->min_flt;
1659 			r->ru_majflt += p->signal->maj_flt;
1660 			r->ru_inblock += p->signal->inblock;
1661 			r->ru_oublock += p->signal->oublock;
1662 			t = p;
1663 			do {
1664 				accumulate_thread_rusage(t, r);
1665 				t = next_thread(t);
1666 			} while (t != p);
1667 			break;
1668 
1669 		default:
1670 			BUG();
1671 	}
1672 	unlock_task_sighand(p, &flags);
1673 
1674 out:
1675 	cputime_to_timeval(utime, &r->ru_utime);
1676 	cputime_to_timeval(stime, &r->ru_stime);
1677 }
1678 
1679 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1680 {
1681 	struct rusage r;
1682 	k_getrusage(p, who, &r);
1683 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1684 }
1685 
1686 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1687 {
1688 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1689 	    who != RUSAGE_THREAD)
1690 		return -EINVAL;
1691 	return getrusage(current, who, ru);
1692 }
1693 
1694 asmlinkage long sys_umask(int mask)
1695 {
1696 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1697 	return mask;
1698 }
1699 
1700 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1701 			  unsigned long arg4, unsigned long arg5)
1702 {
1703 	struct task_struct *me = current;
1704 	unsigned char comm[sizeof(me->comm)];
1705 	long error;
1706 
1707 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1708 	if (error != -ENOSYS)
1709 		return error;
1710 
1711 	error = 0;
1712 	switch (option) {
1713 		case PR_SET_PDEATHSIG:
1714 			if (!valid_signal(arg2)) {
1715 				error = -EINVAL;
1716 				break;
1717 			}
1718 			me->pdeath_signal = arg2;
1719 			error = 0;
1720 			break;
1721 		case PR_GET_PDEATHSIG:
1722 			error = put_user(me->pdeath_signal, (int __user *)arg2);
1723 			break;
1724 		case PR_GET_DUMPABLE:
1725 			error = get_dumpable(me->mm);
1726 			break;
1727 		case PR_SET_DUMPABLE:
1728 			if (arg2 < 0 || arg2 > 1) {
1729 				error = -EINVAL;
1730 				break;
1731 			}
1732 			set_dumpable(me->mm, arg2);
1733 			error = 0;
1734 			break;
1735 
1736 		case PR_SET_UNALIGN:
1737 			error = SET_UNALIGN_CTL(me, arg2);
1738 			break;
1739 		case PR_GET_UNALIGN:
1740 			error = GET_UNALIGN_CTL(me, arg2);
1741 			break;
1742 		case PR_SET_FPEMU:
1743 			error = SET_FPEMU_CTL(me, arg2);
1744 			break;
1745 		case PR_GET_FPEMU:
1746 			error = GET_FPEMU_CTL(me, arg2);
1747 			break;
1748 		case PR_SET_FPEXC:
1749 			error = SET_FPEXC_CTL(me, arg2);
1750 			break;
1751 		case PR_GET_FPEXC:
1752 			error = GET_FPEXC_CTL(me, arg2);
1753 			break;
1754 		case PR_GET_TIMING:
1755 			error = PR_TIMING_STATISTICAL;
1756 			break;
1757 		case PR_SET_TIMING:
1758 			if (arg2 != PR_TIMING_STATISTICAL)
1759 				error = -EINVAL;
1760 			else
1761 				error = 0;
1762 			break;
1763 
1764 		case PR_SET_NAME:
1765 			comm[sizeof(me->comm)-1] = 0;
1766 			if (strncpy_from_user(comm, (char __user *)arg2,
1767 					      sizeof(me->comm) - 1) < 0)
1768 				return -EFAULT;
1769 			set_task_comm(me, comm);
1770 			return 0;
1771 		case PR_GET_NAME:
1772 			get_task_comm(comm, me);
1773 			if (copy_to_user((char __user *)arg2, comm,
1774 					 sizeof(comm)))
1775 				return -EFAULT;
1776 			return 0;
1777 		case PR_GET_ENDIAN:
1778 			error = GET_ENDIAN(me, arg2);
1779 			break;
1780 		case PR_SET_ENDIAN:
1781 			error = SET_ENDIAN(me, arg2);
1782 			break;
1783 
1784 		case PR_GET_SECCOMP:
1785 			error = prctl_get_seccomp();
1786 			break;
1787 		case PR_SET_SECCOMP:
1788 			error = prctl_set_seccomp(arg2);
1789 			break;
1790 		case PR_GET_TSC:
1791 			error = GET_TSC_CTL(arg2);
1792 			break;
1793 		case PR_SET_TSC:
1794 			error = SET_TSC_CTL(arg2);
1795 			break;
1796 		case PR_GET_TIMERSLACK:
1797 			error = current->timer_slack_ns;
1798 			break;
1799 		case PR_SET_TIMERSLACK:
1800 			if (arg2 <= 0)
1801 				current->timer_slack_ns =
1802 					current->default_timer_slack_ns;
1803 			else
1804 				current->timer_slack_ns = arg2;
1805 			error = 0;
1806 			break;
1807 		default:
1808 			error = -EINVAL;
1809 			break;
1810 	}
1811 	return error;
1812 }
1813 
1814 asmlinkage long sys_getcpu(unsigned __user *cpup, unsigned __user *nodep,
1815 			   struct getcpu_cache __user *unused)
1816 {
1817 	int err = 0;
1818 	int cpu = raw_smp_processor_id();
1819 	if (cpup)
1820 		err |= put_user(cpu, cpup);
1821 	if (nodep)
1822 		err |= put_user(cpu_to_node(cpu), nodep);
1823 	return err ? -EFAULT : 0;
1824 }
1825 
1826 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1827 
1828 static void argv_cleanup(char **argv, char **envp)
1829 {
1830 	argv_free(argv);
1831 }
1832 
1833 /**
1834  * orderly_poweroff - Trigger an orderly system poweroff
1835  * @force: force poweroff if command execution fails
1836  *
1837  * This may be called from any context to trigger a system shutdown.
1838  * If the orderly shutdown fails, it will force an immediate shutdown.
1839  */
1840 int orderly_poweroff(bool force)
1841 {
1842 	int argc;
1843 	char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1844 	static char *envp[] = {
1845 		"HOME=/",
1846 		"PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1847 		NULL
1848 	};
1849 	int ret = -ENOMEM;
1850 	struct subprocess_info *info;
1851 
1852 	if (argv == NULL) {
1853 		printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1854 		       __func__, poweroff_cmd);
1855 		goto out;
1856 	}
1857 
1858 	info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1859 	if (info == NULL) {
1860 		argv_free(argv);
1861 		goto out;
1862 	}
1863 
1864 	call_usermodehelper_setcleanup(info, argv_cleanup);
1865 
1866 	ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1867 
1868   out:
1869 	if (ret && force) {
1870 		printk(KERN_WARNING "Failed to start orderly shutdown: "
1871 		       "forcing the issue\n");
1872 
1873 		/* I guess this should try to kick off some daemon to
1874 		   sync and poweroff asap.  Or not even bother syncing
1875 		   if we're doing an emergency shutdown? */
1876 		emergency_sync();
1877 		kernel_power_off();
1878 	}
1879 
1880 	return ret;
1881 }
1882 EXPORT_SYMBOL_GPL(orderly_poweroff);
1883