xref: /linux/kernel/sys.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
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 #include <linux/ptrace.h>
37 
38 #include <linux/compat.h>
39 #include <linux/syscalls.h>
40 #include <linux/kprobes.h>
41 #include <linux/user_namespace.h>
42 
43 #include <asm/uaccess.h>
44 #include <asm/io.h>
45 #include <asm/unistd.h>
46 
47 #ifndef SET_UNALIGN_CTL
48 # define SET_UNALIGN_CTL(a,b)	(-EINVAL)
49 #endif
50 #ifndef GET_UNALIGN_CTL
51 # define GET_UNALIGN_CTL(a,b)	(-EINVAL)
52 #endif
53 #ifndef SET_FPEMU_CTL
54 # define SET_FPEMU_CTL(a,b)	(-EINVAL)
55 #endif
56 #ifndef GET_FPEMU_CTL
57 # define GET_FPEMU_CTL(a,b)	(-EINVAL)
58 #endif
59 #ifndef SET_FPEXC_CTL
60 # define SET_FPEXC_CTL(a,b)	(-EINVAL)
61 #endif
62 #ifndef GET_FPEXC_CTL
63 # define GET_FPEXC_CTL(a,b)	(-EINVAL)
64 #endif
65 #ifndef GET_ENDIAN
66 # define GET_ENDIAN(a,b)	(-EINVAL)
67 #endif
68 #ifndef SET_ENDIAN
69 # define SET_ENDIAN(a,b)	(-EINVAL)
70 #endif
71 #ifndef GET_TSC_CTL
72 # define GET_TSC_CTL(a)		(-EINVAL)
73 #endif
74 #ifndef SET_TSC_CTL
75 # define SET_TSC_CTL(a)		(-EINVAL)
76 #endif
77 
78 /*
79  * this is where the system-wide overflow UID and GID are defined, for
80  * architectures that now have 32-bit UID/GID but didn't in the past
81  */
82 
83 int overflowuid = DEFAULT_OVERFLOWUID;
84 int overflowgid = DEFAULT_OVERFLOWGID;
85 
86 #ifdef CONFIG_UID16
87 EXPORT_SYMBOL(overflowuid);
88 EXPORT_SYMBOL(overflowgid);
89 #endif
90 
91 /*
92  * the same as above, but for filesystems which can only store a 16-bit
93  * UID and GID. as such, this is needed on all architectures
94  */
95 
96 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
97 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
98 
99 EXPORT_SYMBOL(fs_overflowuid);
100 EXPORT_SYMBOL(fs_overflowgid);
101 
102 /*
103  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
104  */
105 
106 int C_A_D = 1;
107 struct pid *cad_pid;
108 EXPORT_SYMBOL(cad_pid);
109 
110 /*
111  * If set, this is used for preparing the system to power off.
112  */
113 
114 void (*pm_power_off_prepare)(void);
115 
116 /*
117  * set the priority of a task
118  * - the caller must hold the RCU read lock
119  */
120 static int set_one_prio(struct task_struct *p, int niceval, int error)
121 {
122 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
123 	int no_nice;
124 
125 	if (pcred->uid  != cred->euid &&
126 	    pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
127 		error = -EPERM;
128 		goto out;
129 	}
130 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
131 		error = -EACCES;
132 		goto out;
133 	}
134 	no_nice = security_task_setnice(p, niceval);
135 	if (no_nice) {
136 		error = no_nice;
137 		goto out;
138 	}
139 	if (error == -ESRCH)
140 		error = 0;
141 	set_user_nice(p, niceval);
142 out:
143 	return error;
144 }
145 
146 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
147 {
148 	struct task_struct *g, *p;
149 	struct user_struct *user;
150 	const struct cred *cred = current_cred();
151 	int error = -EINVAL;
152 	struct pid *pgrp;
153 
154 	if (which > PRIO_USER || which < PRIO_PROCESS)
155 		goto out;
156 
157 	/* normalize: avoid signed division (rounding problems) */
158 	error = -ESRCH;
159 	if (niceval < -20)
160 		niceval = -20;
161 	if (niceval > 19)
162 		niceval = 19;
163 
164 	read_lock(&tasklist_lock);
165 	switch (which) {
166 		case PRIO_PROCESS:
167 			if (who)
168 				p = find_task_by_vpid(who);
169 			else
170 				p = current;
171 			if (p)
172 				error = set_one_prio(p, niceval, error);
173 			break;
174 		case PRIO_PGRP:
175 			if (who)
176 				pgrp = find_vpid(who);
177 			else
178 				pgrp = task_pgrp(current);
179 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
180 				error = set_one_prio(p, niceval, error);
181 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
182 			break;
183 		case PRIO_USER:
184 			user = (struct user_struct *) cred->user;
185 			if (!who)
186 				who = cred->uid;
187 			else if ((who != cred->uid) &&
188 				 !(user = find_user(who)))
189 				goto out_unlock;	/* No processes for this user */
190 
191 			do_each_thread(g, p)
192 				if (__task_cred(p)->uid == who)
193 					error = set_one_prio(p, niceval, error);
194 			while_each_thread(g, p);
195 			if (who != cred->uid)
196 				free_uid(user);		/* For find_user() */
197 			break;
198 	}
199 out_unlock:
200 	read_unlock(&tasklist_lock);
201 out:
202 	return error;
203 }
204 
205 /*
206  * Ugh. To avoid negative return values, "getpriority()" will
207  * not return the normal nice-value, but a negated value that
208  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
209  * to stay compatible.
210  */
211 SYSCALL_DEFINE2(getpriority, int, which, int, who)
212 {
213 	struct task_struct *g, *p;
214 	struct user_struct *user;
215 	const struct cred *cred = current_cred();
216 	long niceval, retval = -ESRCH;
217 	struct pid *pgrp;
218 
219 	if (which > PRIO_USER || which < PRIO_PROCESS)
220 		return -EINVAL;
221 
222 	read_lock(&tasklist_lock);
223 	switch (which) {
224 		case PRIO_PROCESS:
225 			if (who)
226 				p = find_task_by_vpid(who);
227 			else
228 				p = current;
229 			if (p) {
230 				niceval = 20 - task_nice(p);
231 				if (niceval > retval)
232 					retval = niceval;
233 			}
234 			break;
235 		case PRIO_PGRP:
236 			if (who)
237 				pgrp = find_vpid(who);
238 			else
239 				pgrp = task_pgrp(current);
240 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
241 				niceval = 20 - task_nice(p);
242 				if (niceval > retval)
243 					retval = niceval;
244 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
245 			break;
246 		case PRIO_USER:
247 			user = (struct user_struct *) cred->user;
248 			if (!who)
249 				who = cred->uid;
250 			else if ((who != cred->uid) &&
251 				 !(user = find_user(who)))
252 				goto out_unlock;	/* No processes for this user */
253 
254 			do_each_thread(g, p)
255 				if (__task_cred(p)->uid == who) {
256 					niceval = 20 - task_nice(p);
257 					if (niceval > retval)
258 						retval = niceval;
259 				}
260 			while_each_thread(g, p);
261 			if (who != cred->uid)
262 				free_uid(user);		/* for find_user() */
263 			break;
264 	}
265 out_unlock:
266 	read_unlock(&tasklist_lock);
267 
268 	return retval;
269 }
270 
271 /**
272  *	emergency_restart - reboot the system
273  *
274  *	Without shutting down any hardware or taking any locks
275  *	reboot the system.  This is called when we know we are in
276  *	trouble so this is our best effort to reboot.  This is
277  *	safe to call in interrupt context.
278  */
279 void emergency_restart(void)
280 {
281 	machine_emergency_restart();
282 }
283 EXPORT_SYMBOL_GPL(emergency_restart);
284 
285 void kernel_restart_prepare(char *cmd)
286 {
287 	blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
288 	system_state = SYSTEM_RESTART;
289 	device_shutdown();
290 	sysdev_shutdown();
291 }
292 
293 /**
294  *	kernel_restart - reboot the system
295  *	@cmd: pointer to buffer containing command to execute for restart
296  *		or %NULL
297  *
298  *	Shutdown everything and perform a clean reboot.
299  *	This is not safe to call in interrupt context.
300  */
301 void kernel_restart(char *cmd)
302 {
303 	kernel_restart_prepare(cmd);
304 	if (!cmd)
305 		printk(KERN_EMERG "Restarting system.\n");
306 	else
307 		printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
308 	machine_restart(cmd);
309 }
310 EXPORT_SYMBOL_GPL(kernel_restart);
311 
312 static void kernel_shutdown_prepare(enum system_states state)
313 {
314 	blocking_notifier_call_chain(&reboot_notifier_list,
315 		(state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
316 	system_state = state;
317 	device_shutdown();
318 }
319 /**
320  *	kernel_halt - halt the system
321  *
322  *	Shutdown everything and perform a clean system halt.
323  */
324 void kernel_halt(void)
325 {
326 	kernel_shutdown_prepare(SYSTEM_HALT);
327 	sysdev_shutdown();
328 	printk(KERN_EMERG "System halted.\n");
329 	machine_halt();
330 }
331 
332 EXPORT_SYMBOL_GPL(kernel_halt);
333 
334 /**
335  *	kernel_power_off - power_off the system
336  *
337  *	Shutdown everything and perform a clean system power_off.
338  */
339 void kernel_power_off(void)
340 {
341 	kernel_shutdown_prepare(SYSTEM_POWER_OFF);
342 	if (pm_power_off_prepare)
343 		pm_power_off_prepare();
344 	disable_nonboot_cpus();
345 	sysdev_shutdown();
346 	printk(KERN_EMERG "Power down.\n");
347 	machine_power_off();
348 }
349 EXPORT_SYMBOL_GPL(kernel_power_off);
350 /*
351  * Reboot system call: for obvious reasons only root may call it,
352  * and even root needs to set up some magic numbers in the registers
353  * so that some mistake won't make this reboot the whole machine.
354  * You can also set the meaning of the ctrl-alt-del-key here.
355  *
356  * reboot doesn't sync: do that yourself before calling this.
357  */
358 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
359 		void __user *, arg)
360 {
361 	char buffer[256];
362 
363 	/* We only trust the superuser with rebooting the system. */
364 	if (!capable(CAP_SYS_BOOT))
365 		return -EPERM;
366 
367 	/* For safety, we require "magic" arguments. */
368 	if (magic1 != LINUX_REBOOT_MAGIC1 ||
369 	    (magic2 != LINUX_REBOOT_MAGIC2 &&
370 	                magic2 != LINUX_REBOOT_MAGIC2A &&
371 			magic2 != LINUX_REBOOT_MAGIC2B &&
372 	                magic2 != LINUX_REBOOT_MAGIC2C))
373 		return -EINVAL;
374 
375 	/* Instead of trying to make the power_off code look like
376 	 * halt when pm_power_off is not set do it the easy way.
377 	 */
378 	if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
379 		cmd = LINUX_REBOOT_CMD_HALT;
380 
381 	lock_kernel();
382 	switch (cmd) {
383 	case LINUX_REBOOT_CMD_RESTART:
384 		kernel_restart(NULL);
385 		break;
386 
387 	case LINUX_REBOOT_CMD_CAD_ON:
388 		C_A_D = 1;
389 		break;
390 
391 	case LINUX_REBOOT_CMD_CAD_OFF:
392 		C_A_D = 0;
393 		break;
394 
395 	case LINUX_REBOOT_CMD_HALT:
396 		kernel_halt();
397 		unlock_kernel();
398 		do_exit(0);
399 		break;
400 
401 	case LINUX_REBOOT_CMD_POWER_OFF:
402 		kernel_power_off();
403 		unlock_kernel();
404 		do_exit(0);
405 		break;
406 
407 	case LINUX_REBOOT_CMD_RESTART2:
408 		if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
409 			unlock_kernel();
410 			return -EFAULT;
411 		}
412 		buffer[sizeof(buffer) - 1] = '\0';
413 
414 		kernel_restart(buffer);
415 		break;
416 
417 #ifdef CONFIG_KEXEC
418 	case LINUX_REBOOT_CMD_KEXEC:
419 		{
420 			int ret;
421 			ret = kernel_kexec();
422 			unlock_kernel();
423 			return ret;
424 		}
425 #endif
426 
427 #ifdef CONFIG_HIBERNATION
428 	case LINUX_REBOOT_CMD_SW_SUSPEND:
429 		{
430 			int ret = hibernate();
431 			unlock_kernel();
432 			return ret;
433 		}
434 #endif
435 
436 	default:
437 		unlock_kernel();
438 		return -EINVAL;
439 	}
440 	unlock_kernel();
441 	return 0;
442 }
443 
444 static void deferred_cad(struct work_struct *dummy)
445 {
446 	kernel_restart(NULL);
447 }
448 
449 /*
450  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
451  * As it's called within an interrupt, it may NOT sync: the only choice
452  * is whether to reboot at once, or just ignore the ctrl-alt-del.
453  */
454 void ctrl_alt_del(void)
455 {
456 	static DECLARE_WORK(cad_work, deferred_cad);
457 
458 	if (C_A_D)
459 		schedule_work(&cad_work);
460 	else
461 		kill_cad_pid(SIGINT, 1);
462 }
463 
464 /*
465  * Unprivileged users may change the real gid to the effective gid
466  * or vice versa.  (BSD-style)
467  *
468  * If you set the real gid at all, or set the effective gid to a value not
469  * equal to the real gid, then the saved gid is set to the new effective gid.
470  *
471  * This makes it possible for a setgid program to completely drop its
472  * privileges, which is often a useful assertion to make when you are doing
473  * a security audit over a program.
474  *
475  * The general idea is that a program which uses just setregid() will be
476  * 100% compatible with BSD.  A program which uses just setgid() will be
477  * 100% compatible with POSIX with saved IDs.
478  *
479  * SMP: There are not races, the GIDs are checked only by filesystem
480  *      operations (as far as semantic preservation is concerned).
481  */
482 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
483 {
484 	const struct cred *old;
485 	struct cred *new;
486 	int retval;
487 
488 	new = prepare_creds();
489 	if (!new)
490 		return -ENOMEM;
491 	old = current_cred();
492 
493 	retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
494 	if (retval)
495 		goto error;
496 
497 	retval = -EPERM;
498 	if (rgid != (gid_t) -1) {
499 		if (old->gid == rgid ||
500 		    old->egid == rgid ||
501 		    capable(CAP_SETGID))
502 			new->gid = rgid;
503 		else
504 			goto error;
505 	}
506 	if (egid != (gid_t) -1) {
507 		if (old->gid == egid ||
508 		    old->egid == egid ||
509 		    old->sgid == egid ||
510 		    capable(CAP_SETGID))
511 			new->egid = egid;
512 		else
513 			goto error;
514 	}
515 
516 	if (rgid != (gid_t) -1 ||
517 	    (egid != (gid_t) -1 && egid != old->gid))
518 		new->sgid = new->egid;
519 	new->fsgid = new->egid;
520 
521 	return commit_creds(new);
522 
523 error:
524 	abort_creds(new);
525 	return retval;
526 }
527 
528 /*
529  * setgid() is implemented like SysV w/ SAVED_IDS
530  *
531  * SMP: Same implicit races as above.
532  */
533 SYSCALL_DEFINE1(setgid, gid_t, gid)
534 {
535 	const struct cred *old;
536 	struct cred *new;
537 	int retval;
538 
539 	new = prepare_creds();
540 	if (!new)
541 		return -ENOMEM;
542 	old = current_cred();
543 
544 	retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
545 	if (retval)
546 		goto error;
547 
548 	retval = -EPERM;
549 	if (capable(CAP_SETGID))
550 		new->gid = new->egid = new->sgid = new->fsgid = gid;
551 	else if (gid == old->gid || gid == old->sgid)
552 		new->egid = new->fsgid = gid;
553 	else
554 		goto error;
555 
556 	return commit_creds(new);
557 
558 error:
559 	abort_creds(new);
560 	return retval;
561 }
562 
563 /*
564  * change the user struct in a credentials set to match the new UID
565  */
566 static int set_user(struct cred *new)
567 {
568 	struct user_struct *new_user;
569 
570 	new_user = alloc_uid(current_user_ns(), new->uid);
571 	if (!new_user)
572 		return -EAGAIN;
573 
574 	if (!task_can_switch_user(new_user, current)) {
575 		free_uid(new_user);
576 		return -EINVAL;
577 	}
578 
579 	if (atomic_read(&new_user->processes) >=
580 				current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
581 			new_user != INIT_USER) {
582 		free_uid(new_user);
583 		return -EAGAIN;
584 	}
585 
586 	free_uid(new->user);
587 	new->user = new_user;
588 	return 0;
589 }
590 
591 /*
592  * Unprivileged users may change the real uid to the effective uid
593  * or vice versa.  (BSD-style)
594  *
595  * If you set the real uid at all, or set the effective uid to a value not
596  * equal to the real uid, then the saved uid is set to the new effective uid.
597  *
598  * This makes it possible for a setuid program to completely drop its
599  * privileges, which is often a useful assertion to make when you are doing
600  * a security audit over a program.
601  *
602  * The general idea is that a program which uses just setreuid() will be
603  * 100% compatible with BSD.  A program which uses just setuid() will be
604  * 100% compatible with POSIX with saved IDs.
605  */
606 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
607 {
608 	const struct cred *old;
609 	struct cred *new;
610 	int retval;
611 
612 	new = prepare_creds();
613 	if (!new)
614 		return -ENOMEM;
615 	old = current_cred();
616 
617 	retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
618 	if (retval)
619 		goto error;
620 
621 	retval = -EPERM;
622 	if (ruid != (uid_t) -1) {
623 		new->uid = ruid;
624 		if (old->uid != ruid &&
625 		    old->euid != ruid &&
626 		    !capable(CAP_SETUID))
627 			goto error;
628 	}
629 
630 	if (euid != (uid_t) -1) {
631 		new->euid = euid;
632 		if (old->uid != euid &&
633 		    old->euid != euid &&
634 		    old->suid != euid &&
635 		    !capable(CAP_SETUID))
636 			goto error;
637 	}
638 
639 	if (new->uid != old->uid) {
640 		retval = set_user(new);
641 		if (retval < 0)
642 			goto error;
643 	}
644 	if (ruid != (uid_t) -1 ||
645 	    (euid != (uid_t) -1 && euid != old->uid))
646 		new->suid = new->euid;
647 	new->fsuid = new->euid;
648 
649 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
650 	if (retval < 0)
651 		goto error;
652 
653 	return commit_creds(new);
654 
655 error:
656 	abort_creds(new);
657 	return retval;
658 }
659 
660 /*
661  * setuid() is implemented like SysV with SAVED_IDS
662  *
663  * Note that SAVED_ID's is deficient in that a setuid root program
664  * like sendmail, for example, cannot set its uid to be a normal
665  * user and then switch back, because if you're root, setuid() sets
666  * the saved uid too.  If you don't like this, blame the bright people
667  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
668  * will allow a root program to temporarily drop privileges and be able to
669  * regain them by swapping the real and effective uid.
670  */
671 SYSCALL_DEFINE1(setuid, uid_t, uid)
672 {
673 	const struct cred *old;
674 	struct cred *new;
675 	int retval;
676 
677 	new = prepare_creds();
678 	if (!new)
679 		return -ENOMEM;
680 	old = current_cred();
681 
682 	retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
683 	if (retval)
684 		goto error;
685 
686 	retval = -EPERM;
687 	if (capable(CAP_SETUID)) {
688 		new->suid = new->uid = uid;
689 		if (uid != old->uid) {
690 			retval = set_user(new);
691 			if (retval < 0)
692 				goto error;
693 		}
694 	} else if (uid != old->uid && uid != new->suid) {
695 		goto error;
696 	}
697 
698 	new->fsuid = new->euid = uid;
699 
700 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
701 	if (retval < 0)
702 		goto error;
703 
704 	return commit_creds(new);
705 
706 error:
707 	abort_creds(new);
708 	return retval;
709 }
710 
711 
712 /*
713  * This function implements a generic ability to update ruid, euid,
714  * and suid.  This allows you to implement the 4.4 compatible seteuid().
715  */
716 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
717 {
718 	const struct cred *old;
719 	struct cred *new;
720 	int retval;
721 
722 	new = prepare_creds();
723 	if (!new)
724 		return -ENOMEM;
725 
726 	retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
727 	if (retval)
728 		goto error;
729 	old = current_cred();
730 
731 	retval = -EPERM;
732 	if (!capable(CAP_SETUID)) {
733 		if (ruid != (uid_t) -1 && ruid != old->uid &&
734 		    ruid != old->euid  && ruid != old->suid)
735 			goto error;
736 		if (euid != (uid_t) -1 && euid != old->uid &&
737 		    euid != old->euid  && euid != old->suid)
738 			goto error;
739 		if (suid != (uid_t) -1 && suid != old->uid &&
740 		    suid != old->euid  && suid != old->suid)
741 			goto error;
742 	}
743 
744 	if (ruid != (uid_t) -1) {
745 		new->uid = ruid;
746 		if (ruid != old->uid) {
747 			retval = set_user(new);
748 			if (retval < 0)
749 				goto error;
750 		}
751 	}
752 	if (euid != (uid_t) -1)
753 		new->euid = euid;
754 	if (suid != (uid_t) -1)
755 		new->suid = suid;
756 	new->fsuid = new->euid;
757 
758 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
759 	if (retval < 0)
760 		goto error;
761 
762 	return commit_creds(new);
763 
764 error:
765 	abort_creds(new);
766 	return retval;
767 }
768 
769 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
770 {
771 	const struct cred *cred = current_cred();
772 	int retval;
773 
774 	if (!(retval   = put_user(cred->uid,  ruid)) &&
775 	    !(retval   = put_user(cred->euid, euid)))
776 		retval = put_user(cred->suid, suid);
777 
778 	return retval;
779 }
780 
781 /*
782  * Same as above, but for rgid, egid, sgid.
783  */
784 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
785 {
786 	const struct cred *old;
787 	struct cred *new;
788 	int retval;
789 
790 	new = prepare_creds();
791 	if (!new)
792 		return -ENOMEM;
793 	old = current_cred();
794 
795 	retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
796 	if (retval)
797 		goto error;
798 
799 	retval = -EPERM;
800 	if (!capable(CAP_SETGID)) {
801 		if (rgid != (gid_t) -1 && rgid != old->gid &&
802 		    rgid != old->egid  && rgid != old->sgid)
803 			goto error;
804 		if (egid != (gid_t) -1 && egid != old->gid &&
805 		    egid != old->egid  && egid != old->sgid)
806 			goto error;
807 		if (sgid != (gid_t) -1 && sgid != old->gid &&
808 		    sgid != old->egid  && sgid != old->sgid)
809 			goto error;
810 	}
811 
812 	if (rgid != (gid_t) -1)
813 		new->gid = rgid;
814 	if (egid != (gid_t) -1)
815 		new->egid = egid;
816 	if (sgid != (gid_t) -1)
817 		new->sgid = sgid;
818 	new->fsgid = new->egid;
819 
820 	return commit_creds(new);
821 
822 error:
823 	abort_creds(new);
824 	return retval;
825 }
826 
827 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
828 {
829 	const struct cred *cred = current_cred();
830 	int retval;
831 
832 	if (!(retval   = put_user(cred->gid,  rgid)) &&
833 	    !(retval   = put_user(cred->egid, egid)))
834 		retval = put_user(cred->sgid, sgid);
835 
836 	return retval;
837 }
838 
839 
840 /*
841  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
842  * is used for "access()" and for the NFS daemon (letting nfsd stay at
843  * whatever uid it wants to). It normally shadows "euid", except when
844  * explicitly set by setfsuid() or for access..
845  */
846 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
847 {
848 	const struct cred *old;
849 	struct cred *new;
850 	uid_t old_fsuid;
851 
852 	new = prepare_creds();
853 	if (!new)
854 		return current_fsuid();
855 	old = current_cred();
856 	old_fsuid = old->fsuid;
857 
858 	if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
859 		goto error;
860 
861 	if (uid == old->uid  || uid == old->euid  ||
862 	    uid == old->suid || uid == old->fsuid ||
863 	    capable(CAP_SETUID)) {
864 		if (uid != old_fsuid) {
865 			new->fsuid = uid;
866 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
867 				goto change_okay;
868 		}
869 	}
870 
871 error:
872 	abort_creds(new);
873 	return old_fsuid;
874 
875 change_okay:
876 	commit_creds(new);
877 	return old_fsuid;
878 }
879 
880 /*
881  * Samma på svenska..
882  */
883 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
884 {
885 	const struct cred *old;
886 	struct cred *new;
887 	gid_t old_fsgid;
888 
889 	new = prepare_creds();
890 	if (!new)
891 		return current_fsgid();
892 	old = current_cred();
893 	old_fsgid = old->fsgid;
894 
895 	if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
896 		goto error;
897 
898 	if (gid == old->gid  || gid == old->egid  ||
899 	    gid == old->sgid || gid == old->fsgid ||
900 	    capable(CAP_SETGID)) {
901 		if (gid != old_fsgid) {
902 			new->fsgid = gid;
903 			goto change_okay;
904 		}
905 	}
906 
907 error:
908 	abort_creds(new);
909 	return old_fsgid;
910 
911 change_okay:
912 	commit_creds(new);
913 	return old_fsgid;
914 }
915 
916 void do_sys_times(struct tms *tms)
917 {
918 	struct task_cputime cputime;
919 	cputime_t cutime, cstime;
920 
921 	thread_group_cputime(current, &cputime);
922 	spin_lock_irq(&current->sighand->siglock);
923 	cutime = current->signal->cutime;
924 	cstime = current->signal->cstime;
925 	spin_unlock_irq(&current->sighand->siglock);
926 	tms->tms_utime = cputime_to_clock_t(cputime.utime);
927 	tms->tms_stime = cputime_to_clock_t(cputime.stime);
928 	tms->tms_cutime = cputime_to_clock_t(cutime);
929 	tms->tms_cstime = cputime_to_clock_t(cstime);
930 }
931 
932 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
933 {
934 	if (tbuf) {
935 		struct tms tmp;
936 
937 		do_sys_times(&tmp);
938 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
939 			return -EFAULT;
940 	}
941 	force_successful_syscall_return();
942 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
943 }
944 
945 /*
946  * This needs some heavy checking ...
947  * I just haven't the stomach for it. I also don't fully
948  * understand sessions/pgrp etc. Let somebody who does explain it.
949  *
950  * OK, I think I have the protection semantics right.... this is really
951  * only important on a multi-user system anyway, to make sure one user
952  * can't send a signal to a process owned by another.  -TYT, 12/12/91
953  *
954  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
955  * LBT 04.03.94
956  */
957 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
958 {
959 	struct task_struct *p;
960 	struct task_struct *group_leader = current->group_leader;
961 	struct pid *pgrp;
962 	int err;
963 
964 	if (!pid)
965 		pid = task_pid_vnr(group_leader);
966 	if (!pgid)
967 		pgid = pid;
968 	if (pgid < 0)
969 		return -EINVAL;
970 
971 	/* From this point forward we keep holding onto the tasklist lock
972 	 * so that our parent does not change from under us. -DaveM
973 	 */
974 	write_lock_irq(&tasklist_lock);
975 
976 	err = -ESRCH;
977 	p = find_task_by_vpid(pid);
978 	if (!p)
979 		goto out;
980 
981 	err = -EINVAL;
982 	if (!thread_group_leader(p))
983 		goto out;
984 
985 	if (same_thread_group(p->real_parent, group_leader)) {
986 		err = -EPERM;
987 		if (task_session(p) != task_session(group_leader))
988 			goto out;
989 		err = -EACCES;
990 		if (p->did_exec)
991 			goto out;
992 	} else {
993 		err = -ESRCH;
994 		if (p != group_leader)
995 			goto out;
996 	}
997 
998 	err = -EPERM;
999 	if (p->signal->leader)
1000 		goto out;
1001 
1002 	pgrp = task_pid(p);
1003 	if (pgid != pid) {
1004 		struct task_struct *g;
1005 
1006 		pgrp = find_vpid(pgid);
1007 		g = pid_task(pgrp, PIDTYPE_PGID);
1008 		if (!g || task_session(g) != task_session(group_leader))
1009 			goto out;
1010 	}
1011 
1012 	err = security_task_setpgid(p, pgid);
1013 	if (err)
1014 		goto out;
1015 
1016 	if (task_pgrp(p) != pgrp) {
1017 		change_pid(p, PIDTYPE_PGID, pgrp);
1018 		set_task_pgrp(p, pid_nr(pgrp));
1019 	}
1020 
1021 	err = 0;
1022 out:
1023 	/* All paths lead to here, thus we are safe. -DaveM */
1024 	write_unlock_irq(&tasklist_lock);
1025 	return err;
1026 }
1027 
1028 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1029 {
1030 	struct task_struct *p;
1031 	struct pid *grp;
1032 	int retval;
1033 
1034 	rcu_read_lock();
1035 	if (!pid)
1036 		grp = task_pgrp(current);
1037 	else {
1038 		retval = -ESRCH;
1039 		p = find_task_by_vpid(pid);
1040 		if (!p)
1041 			goto out;
1042 		grp = task_pgrp(p);
1043 		if (!grp)
1044 			goto out;
1045 
1046 		retval = security_task_getpgid(p);
1047 		if (retval)
1048 			goto out;
1049 	}
1050 	retval = pid_vnr(grp);
1051 out:
1052 	rcu_read_unlock();
1053 	return retval;
1054 }
1055 
1056 #ifdef __ARCH_WANT_SYS_GETPGRP
1057 
1058 SYSCALL_DEFINE0(getpgrp)
1059 {
1060 	return sys_getpgid(0);
1061 }
1062 
1063 #endif
1064 
1065 SYSCALL_DEFINE1(getsid, pid_t, pid)
1066 {
1067 	struct task_struct *p;
1068 	struct pid *sid;
1069 	int retval;
1070 
1071 	rcu_read_lock();
1072 	if (!pid)
1073 		sid = task_session(current);
1074 	else {
1075 		retval = -ESRCH;
1076 		p = find_task_by_vpid(pid);
1077 		if (!p)
1078 			goto out;
1079 		sid = task_session(p);
1080 		if (!sid)
1081 			goto out;
1082 
1083 		retval = security_task_getsid(p);
1084 		if (retval)
1085 			goto out;
1086 	}
1087 	retval = pid_vnr(sid);
1088 out:
1089 	rcu_read_unlock();
1090 	return retval;
1091 }
1092 
1093 SYSCALL_DEFINE0(setsid)
1094 {
1095 	struct task_struct *group_leader = current->group_leader;
1096 	struct pid *sid = task_pid(group_leader);
1097 	pid_t session = pid_vnr(sid);
1098 	int err = -EPERM;
1099 
1100 	write_lock_irq(&tasklist_lock);
1101 	/* Fail if I am already a session leader */
1102 	if (group_leader->signal->leader)
1103 		goto out;
1104 
1105 	/* Fail if a process group id already exists that equals the
1106 	 * proposed session id.
1107 	 */
1108 	if (pid_task(sid, PIDTYPE_PGID))
1109 		goto out;
1110 
1111 	group_leader->signal->leader = 1;
1112 	__set_special_pids(sid);
1113 
1114 	proc_clear_tty(group_leader);
1115 
1116 	err = session;
1117 out:
1118 	write_unlock_irq(&tasklist_lock);
1119 	return err;
1120 }
1121 
1122 /*
1123  * Supplementary group IDs
1124  */
1125 
1126 /* init to 2 - one for init_task, one to ensure it is never freed */
1127 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1128 
1129 struct group_info *groups_alloc(int gidsetsize)
1130 {
1131 	struct group_info *group_info;
1132 	int nblocks;
1133 	int i;
1134 
1135 	nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1136 	/* Make sure we always allocate at least one indirect block pointer */
1137 	nblocks = nblocks ? : 1;
1138 	group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1139 	if (!group_info)
1140 		return NULL;
1141 	group_info->ngroups = gidsetsize;
1142 	group_info->nblocks = nblocks;
1143 	atomic_set(&group_info->usage, 1);
1144 
1145 	if (gidsetsize <= NGROUPS_SMALL)
1146 		group_info->blocks[0] = group_info->small_block;
1147 	else {
1148 		for (i = 0; i < nblocks; i++) {
1149 			gid_t *b;
1150 			b = (void *)__get_free_page(GFP_USER);
1151 			if (!b)
1152 				goto out_undo_partial_alloc;
1153 			group_info->blocks[i] = b;
1154 		}
1155 	}
1156 	return group_info;
1157 
1158 out_undo_partial_alloc:
1159 	while (--i >= 0) {
1160 		free_page((unsigned long)group_info->blocks[i]);
1161 	}
1162 	kfree(group_info);
1163 	return NULL;
1164 }
1165 
1166 EXPORT_SYMBOL(groups_alloc);
1167 
1168 void groups_free(struct group_info *group_info)
1169 {
1170 	if (group_info->blocks[0] != group_info->small_block) {
1171 		int i;
1172 		for (i = 0; i < group_info->nblocks; i++)
1173 			free_page((unsigned long)group_info->blocks[i]);
1174 	}
1175 	kfree(group_info);
1176 }
1177 
1178 EXPORT_SYMBOL(groups_free);
1179 
1180 /* export the group_info to a user-space array */
1181 static int groups_to_user(gid_t __user *grouplist,
1182 			  const struct group_info *group_info)
1183 {
1184 	int i;
1185 	unsigned int count = group_info->ngroups;
1186 
1187 	for (i = 0; i < group_info->nblocks; i++) {
1188 		unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1189 		unsigned int len = cp_count * sizeof(*grouplist);
1190 
1191 		if (copy_to_user(grouplist, group_info->blocks[i], len))
1192 			return -EFAULT;
1193 
1194 		grouplist += NGROUPS_PER_BLOCK;
1195 		count -= cp_count;
1196 	}
1197 	return 0;
1198 }
1199 
1200 /* fill a group_info from a user-space array - it must be allocated already */
1201 static int groups_from_user(struct group_info *group_info,
1202     gid_t __user *grouplist)
1203 {
1204 	int i;
1205 	unsigned int count = group_info->ngroups;
1206 
1207 	for (i = 0; i < group_info->nblocks; i++) {
1208 		unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1209 		unsigned int len = cp_count * sizeof(*grouplist);
1210 
1211 		if (copy_from_user(group_info->blocks[i], grouplist, len))
1212 			return -EFAULT;
1213 
1214 		grouplist += NGROUPS_PER_BLOCK;
1215 		count -= cp_count;
1216 	}
1217 	return 0;
1218 }
1219 
1220 /* a simple Shell sort */
1221 static void groups_sort(struct group_info *group_info)
1222 {
1223 	int base, max, stride;
1224 	int gidsetsize = group_info->ngroups;
1225 
1226 	for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1227 		; /* nothing */
1228 	stride /= 3;
1229 
1230 	while (stride) {
1231 		max = gidsetsize - stride;
1232 		for (base = 0; base < max; base++) {
1233 			int left = base;
1234 			int right = left + stride;
1235 			gid_t tmp = GROUP_AT(group_info, right);
1236 
1237 			while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1238 				GROUP_AT(group_info, right) =
1239 				    GROUP_AT(group_info, left);
1240 				right = left;
1241 				left -= stride;
1242 			}
1243 			GROUP_AT(group_info, right) = tmp;
1244 		}
1245 		stride /= 3;
1246 	}
1247 }
1248 
1249 /* a simple bsearch */
1250 int groups_search(const struct group_info *group_info, gid_t grp)
1251 {
1252 	unsigned int left, right;
1253 
1254 	if (!group_info)
1255 		return 0;
1256 
1257 	left = 0;
1258 	right = group_info->ngroups;
1259 	while (left < right) {
1260 		unsigned int mid = (left+right)/2;
1261 		int cmp = grp - GROUP_AT(group_info, mid);
1262 		if (cmp > 0)
1263 			left = mid + 1;
1264 		else if (cmp < 0)
1265 			right = mid;
1266 		else
1267 			return 1;
1268 	}
1269 	return 0;
1270 }
1271 
1272 /**
1273  * set_groups - Change a group subscription in a set of credentials
1274  * @new: The newly prepared set of credentials to alter
1275  * @group_info: The group list to install
1276  *
1277  * Validate a group subscription and, if valid, insert it into a set
1278  * of credentials.
1279  */
1280 int set_groups(struct cred *new, struct group_info *group_info)
1281 {
1282 	int retval;
1283 
1284 	retval = security_task_setgroups(group_info);
1285 	if (retval)
1286 		return retval;
1287 
1288 	put_group_info(new->group_info);
1289 	groups_sort(group_info);
1290 	get_group_info(group_info);
1291 	new->group_info = group_info;
1292 	return 0;
1293 }
1294 
1295 EXPORT_SYMBOL(set_groups);
1296 
1297 /**
1298  * set_current_groups - Change current's group subscription
1299  * @group_info: The group list to impose
1300  *
1301  * Validate a group subscription and, if valid, impose it upon current's task
1302  * security record.
1303  */
1304 int set_current_groups(struct group_info *group_info)
1305 {
1306 	struct cred *new;
1307 	int ret;
1308 
1309 	new = prepare_creds();
1310 	if (!new)
1311 		return -ENOMEM;
1312 
1313 	ret = set_groups(new, group_info);
1314 	if (ret < 0) {
1315 		abort_creds(new);
1316 		return ret;
1317 	}
1318 
1319 	return commit_creds(new);
1320 }
1321 
1322 EXPORT_SYMBOL(set_current_groups);
1323 
1324 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1325 {
1326 	const struct cred *cred = current_cred();
1327 	int i;
1328 
1329 	if (gidsetsize < 0)
1330 		return -EINVAL;
1331 
1332 	/* no need to grab task_lock here; it cannot change */
1333 	i = cred->group_info->ngroups;
1334 	if (gidsetsize) {
1335 		if (i > gidsetsize) {
1336 			i = -EINVAL;
1337 			goto out;
1338 		}
1339 		if (groups_to_user(grouplist, cred->group_info)) {
1340 			i = -EFAULT;
1341 			goto out;
1342 		}
1343 	}
1344 out:
1345 	return i;
1346 }
1347 
1348 /*
1349  *	SMP: Our groups are copy-on-write. We can set them safely
1350  *	without another task interfering.
1351  */
1352 
1353 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1354 {
1355 	struct group_info *group_info;
1356 	int retval;
1357 
1358 	if (!capable(CAP_SETGID))
1359 		return -EPERM;
1360 	if ((unsigned)gidsetsize > NGROUPS_MAX)
1361 		return -EINVAL;
1362 
1363 	group_info = groups_alloc(gidsetsize);
1364 	if (!group_info)
1365 		return -ENOMEM;
1366 	retval = groups_from_user(group_info, grouplist);
1367 	if (retval) {
1368 		put_group_info(group_info);
1369 		return retval;
1370 	}
1371 
1372 	retval = set_current_groups(group_info);
1373 	put_group_info(group_info);
1374 
1375 	return retval;
1376 }
1377 
1378 /*
1379  * Check whether we're fsgid/egid or in the supplemental group..
1380  */
1381 int in_group_p(gid_t grp)
1382 {
1383 	const struct cred *cred = current_cred();
1384 	int retval = 1;
1385 
1386 	if (grp != cred->fsgid)
1387 		retval = groups_search(cred->group_info, grp);
1388 	return retval;
1389 }
1390 
1391 EXPORT_SYMBOL(in_group_p);
1392 
1393 int in_egroup_p(gid_t grp)
1394 {
1395 	const struct cred *cred = current_cred();
1396 	int retval = 1;
1397 
1398 	if (grp != cred->egid)
1399 		retval = groups_search(cred->group_info, grp);
1400 	return retval;
1401 }
1402 
1403 EXPORT_SYMBOL(in_egroup_p);
1404 
1405 DECLARE_RWSEM(uts_sem);
1406 
1407 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1408 {
1409 	int errno = 0;
1410 
1411 	down_read(&uts_sem);
1412 	if (copy_to_user(name, utsname(), sizeof *name))
1413 		errno = -EFAULT;
1414 	up_read(&uts_sem);
1415 	return errno;
1416 }
1417 
1418 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1419 {
1420 	int errno;
1421 	char tmp[__NEW_UTS_LEN];
1422 
1423 	if (!capable(CAP_SYS_ADMIN))
1424 		return -EPERM;
1425 	if (len < 0 || len > __NEW_UTS_LEN)
1426 		return -EINVAL;
1427 	down_write(&uts_sem);
1428 	errno = -EFAULT;
1429 	if (!copy_from_user(tmp, name, len)) {
1430 		struct new_utsname *u = utsname();
1431 
1432 		memcpy(u->nodename, tmp, len);
1433 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1434 		errno = 0;
1435 	}
1436 	up_write(&uts_sem);
1437 	return errno;
1438 }
1439 
1440 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1441 
1442 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1443 {
1444 	int i, errno;
1445 	struct new_utsname *u;
1446 
1447 	if (len < 0)
1448 		return -EINVAL;
1449 	down_read(&uts_sem);
1450 	u = utsname();
1451 	i = 1 + strlen(u->nodename);
1452 	if (i > len)
1453 		i = len;
1454 	errno = 0;
1455 	if (copy_to_user(name, u->nodename, i))
1456 		errno = -EFAULT;
1457 	up_read(&uts_sem);
1458 	return errno;
1459 }
1460 
1461 #endif
1462 
1463 /*
1464  * Only setdomainname; getdomainname can be implemented by calling
1465  * uname()
1466  */
1467 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1468 {
1469 	int errno;
1470 	char tmp[__NEW_UTS_LEN];
1471 
1472 	if (!capable(CAP_SYS_ADMIN))
1473 		return -EPERM;
1474 	if (len < 0 || len > __NEW_UTS_LEN)
1475 		return -EINVAL;
1476 
1477 	down_write(&uts_sem);
1478 	errno = -EFAULT;
1479 	if (!copy_from_user(tmp, name, len)) {
1480 		struct new_utsname *u = utsname();
1481 
1482 		memcpy(u->domainname, tmp, len);
1483 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1484 		errno = 0;
1485 	}
1486 	up_write(&uts_sem);
1487 	return errno;
1488 }
1489 
1490 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1491 {
1492 	if (resource >= RLIM_NLIMITS)
1493 		return -EINVAL;
1494 	else {
1495 		struct rlimit value;
1496 		task_lock(current->group_leader);
1497 		value = current->signal->rlim[resource];
1498 		task_unlock(current->group_leader);
1499 		return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1500 	}
1501 }
1502 
1503 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1504 
1505 /*
1506  *	Back compatibility for getrlimit. Needed for some apps.
1507  */
1508 
1509 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1510 		struct rlimit __user *, rlim)
1511 {
1512 	struct rlimit x;
1513 	if (resource >= RLIM_NLIMITS)
1514 		return -EINVAL;
1515 
1516 	task_lock(current->group_leader);
1517 	x = current->signal->rlim[resource];
1518 	task_unlock(current->group_leader);
1519 	if (x.rlim_cur > 0x7FFFFFFF)
1520 		x.rlim_cur = 0x7FFFFFFF;
1521 	if (x.rlim_max > 0x7FFFFFFF)
1522 		x.rlim_max = 0x7FFFFFFF;
1523 	return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1524 }
1525 
1526 #endif
1527 
1528 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1529 {
1530 	struct rlimit new_rlim, *old_rlim;
1531 	int retval;
1532 
1533 	if (resource >= RLIM_NLIMITS)
1534 		return -EINVAL;
1535 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1536 		return -EFAULT;
1537 	if (new_rlim.rlim_cur > new_rlim.rlim_max)
1538 		return -EINVAL;
1539 	old_rlim = current->signal->rlim + resource;
1540 	if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1541 	    !capable(CAP_SYS_RESOURCE))
1542 		return -EPERM;
1543 	if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1544 		return -EPERM;
1545 
1546 	retval = security_task_setrlimit(resource, &new_rlim);
1547 	if (retval)
1548 		return retval;
1549 
1550 	if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1551 		/*
1552 		 * The caller is asking for an immediate RLIMIT_CPU
1553 		 * expiry.  But we use the zero value to mean "it was
1554 		 * never set".  So let's cheat and make it one second
1555 		 * instead
1556 		 */
1557 		new_rlim.rlim_cur = 1;
1558 	}
1559 
1560 	task_lock(current->group_leader);
1561 	*old_rlim = new_rlim;
1562 	task_unlock(current->group_leader);
1563 
1564 	if (resource != RLIMIT_CPU)
1565 		goto out;
1566 
1567 	/*
1568 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1569 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1570 	 * very long-standing error, and fixing it now risks breakage of
1571 	 * applications, so we live with it
1572 	 */
1573 	if (new_rlim.rlim_cur == RLIM_INFINITY)
1574 		goto out;
1575 
1576 	update_rlimit_cpu(new_rlim.rlim_cur);
1577 out:
1578 	return 0;
1579 }
1580 
1581 /*
1582  * It would make sense to put struct rusage in the task_struct,
1583  * except that would make the task_struct be *really big*.  After
1584  * task_struct gets moved into malloc'ed memory, it would
1585  * make sense to do this.  It will make moving the rest of the information
1586  * a lot simpler!  (Which we're not doing right now because we're not
1587  * measuring them yet).
1588  *
1589  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1590  * races with threads incrementing their own counters.  But since word
1591  * reads are atomic, we either get new values or old values and we don't
1592  * care which for the sums.  We always take the siglock to protect reading
1593  * the c* fields from p->signal from races with exit.c updating those
1594  * fields when reaping, so a sample either gets all the additions of a
1595  * given child after it's reaped, or none so this sample is before reaping.
1596  *
1597  * Locking:
1598  * We need to take the siglock for CHILDEREN, SELF and BOTH
1599  * for  the cases current multithreaded, non-current single threaded
1600  * non-current multithreaded.  Thread traversal is now safe with
1601  * the siglock held.
1602  * Strictly speaking, we donot need to take the siglock if we are current and
1603  * single threaded,  as no one else can take our signal_struct away, no one
1604  * else can  reap the  children to update signal->c* counters, and no one else
1605  * can race with the signal-> fields. If we do not take any lock, the
1606  * signal-> fields could be read out of order while another thread was just
1607  * exiting. So we should  place a read memory barrier when we avoid the lock.
1608  * On the writer side,  write memory barrier is implied in  __exit_signal
1609  * as __exit_signal releases  the siglock spinlock after updating the signal->
1610  * fields. But we don't do this yet to keep things simple.
1611  *
1612  */
1613 
1614 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1615 {
1616 	r->ru_nvcsw += t->nvcsw;
1617 	r->ru_nivcsw += t->nivcsw;
1618 	r->ru_minflt += t->min_flt;
1619 	r->ru_majflt += t->maj_flt;
1620 	r->ru_inblock += task_io_get_inblock(t);
1621 	r->ru_oublock += task_io_get_oublock(t);
1622 }
1623 
1624 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1625 {
1626 	struct task_struct *t;
1627 	unsigned long flags;
1628 	cputime_t utime, stime;
1629 	struct task_cputime cputime;
1630 
1631 	memset((char *) r, 0, sizeof *r);
1632 	utime = stime = cputime_zero;
1633 
1634 	if (who == RUSAGE_THREAD) {
1635 		utime = task_utime(current);
1636 		stime = task_stime(current);
1637 		accumulate_thread_rusage(p, r);
1638 		goto out;
1639 	}
1640 
1641 	if (!lock_task_sighand(p, &flags))
1642 		return;
1643 
1644 	switch (who) {
1645 		case RUSAGE_BOTH:
1646 		case RUSAGE_CHILDREN:
1647 			utime = p->signal->cutime;
1648 			stime = p->signal->cstime;
1649 			r->ru_nvcsw = p->signal->cnvcsw;
1650 			r->ru_nivcsw = p->signal->cnivcsw;
1651 			r->ru_minflt = p->signal->cmin_flt;
1652 			r->ru_majflt = p->signal->cmaj_flt;
1653 			r->ru_inblock = p->signal->cinblock;
1654 			r->ru_oublock = p->signal->coublock;
1655 
1656 			if (who == RUSAGE_CHILDREN)
1657 				break;
1658 
1659 		case RUSAGE_SELF:
1660 			thread_group_cputime(p, &cputime);
1661 			utime = cputime_add(utime, cputime.utime);
1662 			stime = cputime_add(stime, cputime.stime);
1663 			r->ru_nvcsw += p->signal->nvcsw;
1664 			r->ru_nivcsw += p->signal->nivcsw;
1665 			r->ru_minflt += p->signal->min_flt;
1666 			r->ru_majflt += p->signal->maj_flt;
1667 			r->ru_inblock += p->signal->inblock;
1668 			r->ru_oublock += p->signal->oublock;
1669 			t = p;
1670 			do {
1671 				accumulate_thread_rusage(t, r);
1672 				t = next_thread(t);
1673 			} while (t != p);
1674 			break;
1675 
1676 		default:
1677 			BUG();
1678 	}
1679 	unlock_task_sighand(p, &flags);
1680 
1681 out:
1682 	cputime_to_timeval(utime, &r->ru_utime);
1683 	cputime_to_timeval(stime, &r->ru_stime);
1684 }
1685 
1686 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1687 {
1688 	struct rusage r;
1689 	k_getrusage(p, who, &r);
1690 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1691 }
1692 
1693 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1694 {
1695 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1696 	    who != RUSAGE_THREAD)
1697 		return -EINVAL;
1698 	return getrusage(current, who, ru);
1699 }
1700 
1701 SYSCALL_DEFINE1(umask, int, mask)
1702 {
1703 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1704 	return mask;
1705 }
1706 
1707 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1708 		unsigned long, arg4, unsigned long, arg5)
1709 {
1710 	struct task_struct *me = current;
1711 	unsigned char comm[sizeof(me->comm)];
1712 	long error;
1713 
1714 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1715 	if (error != -ENOSYS)
1716 		return error;
1717 
1718 	error = 0;
1719 	switch (option) {
1720 		case PR_SET_PDEATHSIG:
1721 			if (!valid_signal(arg2)) {
1722 				error = -EINVAL;
1723 				break;
1724 			}
1725 			me->pdeath_signal = arg2;
1726 			error = 0;
1727 			break;
1728 		case PR_GET_PDEATHSIG:
1729 			error = put_user(me->pdeath_signal, (int __user *)arg2);
1730 			break;
1731 		case PR_GET_DUMPABLE:
1732 			error = get_dumpable(me->mm);
1733 			break;
1734 		case PR_SET_DUMPABLE:
1735 			if (arg2 < 0 || arg2 > 1) {
1736 				error = -EINVAL;
1737 				break;
1738 			}
1739 			set_dumpable(me->mm, arg2);
1740 			error = 0;
1741 			break;
1742 
1743 		case PR_SET_UNALIGN:
1744 			error = SET_UNALIGN_CTL(me, arg2);
1745 			break;
1746 		case PR_GET_UNALIGN:
1747 			error = GET_UNALIGN_CTL(me, arg2);
1748 			break;
1749 		case PR_SET_FPEMU:
1750 			error = SET_FPEMU_CTL(me, arg2);
1751 			break;
1752 		case PR_GET_FPEMU:
1753 			error = GET_FPEMU_CTL(me, arg2);
1754 			break;
1755 		case PR_SET_FPEXC:
1756 			error = SET_FPEXC_CTL(me, arg2);
1757 			break;
1758 		case PR_GET_FPEXC:
1759 			error = GET_FPEXC_CTL(me, arg2);
1760 			break;
1761 		case PR_GET_TIMING:
1762 			error = PR_TIMING_STATISTICAL;
1763 			break;
1764 		case PR_SET_TIMING:
1765 			if (arg2 != PR_TIMING_STATISTICAL)
1766 				error = -EINVAL;
1767 			else
1768 				error = 0;
1769 			break;
1770 
1771 		case PR_SET_NAME:
1772 			comm[sizeof(me->comm)-1] = 0;
1773 			if (strncpy_from_user(comm, (char __user *)arg2,
1774 					      sizeof(me->comm) - 1) < 0)
1775 				return -EFAULT;
1776 			set_task_comm(me, comm);
1777 			return 0;
1778 		case PR_GET_NAME:
1779 			get_task_comm(comm, me);
1780 			if (copy_to_user((char __user *)arg2, comm,
1781 					 sizeof(comm)))
1782 				return -EFAULT;
1783 			return 0;
1784 		case PR_GET_ENDIAN:
1785 			error = GET_ENDIAN(me, arg2);
1786 			break;
1787 		case PR_SET_ENDIAN:
1788 			error = SET_ENDIAN(me, arg2);
1789 			break;
1790 
1791 		case PR_GET_SECCOMP:
1792 			error = prctl_get_seccomp();
1793 			break;
1794 		case PR_SET_SECCOMP:
1795 			error = prctl_set_seccomp(arg2);
1796 			break;
1797 		case PR_GET_TSC:
1798 			error = GET_TSC_CTL(arg2);
1799 			break;
1800 		case PR_SET_TSC:
1801 			error = SET_TSC_CTL(arg2);
1802 			break;
1803 		case PR_GET_TIMERSLACK:
1804 			error = current->timer_slack_ns;
1805 			break;
1806 		case PR_SET_TIMERSLACK:
1807 			if (arg2 <= 0)
1808 				current->timer_slack_ns =
1809 					current->default_timer_slack_ns;
1810 			else
1811 				current->timer_slack_ns = arg2;
1812 			error = 0;
1813 			break;
1814 		default:
1815 			error = -EINVAL;
1816 			break;
1817 	}
1818 	return error;
1819 }
1820 
1821 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1822 		struct getcpu_cache __user *, unused)
1823 {
1824 	int err = 0;
1825 	int cpu = raw_smp_processor_id();
1826 	if (cpup)
1827 		err |= put_user(cpu, cpup);
1828 	if (nodep)
1829 		err |= put_user(cpu_to_node(cpu), nodep);
1830 	return err ? -EFAULT : 0;
1831 }
1832 
1833 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1834 
1835 static void argv_cleanup(char **argv, char **envp)
1836 {
1837 	argv_free(argv);
1838 }
1839 
1840 /**
1841  * orderly_poweroff - Trigger an orderly system poweroff
1842  * @force: force poweroff if command execution fails
1843  *
1844  * This may be called from any context to trigger a system shutdown.
1845  * If the orderly shutdown fails, it will force an immediate shutdown.
1846  */
1847 int orderly_poweroff(bool force)
1848 {
1849 	int argc;
1850 	char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1851 	static char *envp[] = {
1852 		"HOME=/",
1853 		"PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1854 		NULL
1855 	};
1856 	int ret = -ENOMEM;
1857 	struct subprocess_info *info;
1858 
1859 	if (argv == NULL) {
1860 		printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1861 		       __func__, poweroff_cmd);
1862 		goto out;
1863 	}
1864 
1865 	info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1866 	if (info == NULL) {
1867 		argv_free(argv);
1868 		goto out;
1869 	}
1870 
1871 	call_usermodehelper_setcleanup(info, argv_cleanup);
1872 
1873 	ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1874 
1875   out:
1876 	if (ret && force) {
1877 		printk(KERN_WARNING "Failed to start orderly shutdown: "
1878 		       "forcing the issue\n");
1879 
1880 		/* I guess this should try to kick off some daemon to
1881 		   sync and poweroff asap.  Or not even bother syncing
1882 		   if we're doing an emergency shutdown? */
1883 		emergency_sync();
1884 		kernel_power_off();
1885 	}
1886 
1887 	return ret;
1888 }
1889 EXPORT_SYMBOL_GPL(orderly_poweroff);
1890