xref: /linux/kernel/sys.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 /*
2  *  linux/kernel/sys.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/export.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
14 #include <linux/fs.h>
15 #include <linux/kmod.h>
16 #include <linux/perf_event.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/workqueue.h>
20 #include <linux/capability.h>
21 #include <linux/device.h>
22 #include <linux/key.h>
23 #include <linux/times.h>
24 #include <linux/posix-timers.h>
25 #include <linux/security.h>
26 #include <linux/dcookies.h>
27 #include <linux/suspend.h>
28 #include <linux/tty.h>
29 #include <linux/signal.h>
30 #include <linux/cn_proc.h>
31 #include <linux/getcpu.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/seccomp.h>
34 #include <linux/cpu.h>
35 #include <linux/personality.h>
36 #include <linux/ptrace.h>
37 #include <linux/fs_struct.h>
38 #include <linux/file.h>
39 #include <linux/mount.h>
40 #include <linux/gfp.h>
41 #include <linux/syscore_ops.h>
42 #include <linux/version.h>
43 #include <linux/ctype.h>
44 
45 #include <linux/compat.h>
46 #include <linux/syscalls.h>
47 #include <linux/kprobes.h>
48 #include <linux/user_namespace.h>
49 #include <linux/binfmts.h>
50 
51 #include <linux/sched.h>
52 #include <linux/rcupdate.h>
53 #include <linux/uidgid.h>
54 #include <linux/cred.h>
55 
56 #include <linux/kmsg_dump.h>
57 /* Move somewhere else to avoid recompiling? */
58 #include <generated/utsrelease.h>
59 
60 #include <asm/uaccess.h>
61 #include <asm/io.h>
62 #include <asm/unistd.h>
63 
64 #ifndef SET_UNALIGN_CTL
65 # define SET_UNALIGN_CTL(a,b)	(-EINVAL)
66 #endif
67 #ifndef GET_UNALIGN_CTL
68 # define GET_UNALIGN_CTL(a,b)	(-EINVAL)
69 #endif
70 #ifndef SET_FPEMU_CTL
71 # define SET_FPEMU_CTL(a,b)	(-EINVAL)
72 #endif
73 #ifndef GET_FPEMU_CTL
74 # define GET_FPEMU_CTL(a,b)	(-EINVAL)
75 #endif
76 #ifndef SET_FPEXC_CTL
77 # define SET_FPEXC_CTL(a,b)	(-EINVAL)
78 #endif
79 #ifndef GET_FPEXC_CTL
80 # define GET_FPEXC_CTL(a,b)	(-EINVAL)
81 #endif
82 #ifndef GET_ENDIAN
83 # define GET_ENDIAN(a,b)	(-EINVAL)
84 #endif
85 #ifndef SET_ENDIAN
86 # define SET_ENDIAN(a,b)	(-EINVAL)
87 #endif
88 #ifndef GET_TSC_CTL
89 # define GET_TSC_CTL(a)		(-EINVAL)
90 #endif
91 #ifndef SET_TSC_CTL
92 # define SET_TSC_CTL(a)		(-EINVAL)
93 #endif
94 
95 /*
96  * this is where the system-wide overflow UID and GID are defined, for
97  * architectures that now have 32-bit UID/GID but didn't in the past
98  */
99 
100 int overflowuid = DEFAULT_OVERFLOWUID;
101 int overflowgid = DEFAULT_OVERFLOWGID;
102 
103 EXPORT_SYMBOL(overflowuid);
104 EXPORT_SYMBOL(overflowgid);
105 
106 /*
107  * the same as above, but for filesystems which can only store a 16-bit
108  * UID and GID. as such, this is needed on all architectures
109  */
110 
111 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
112 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
113 
114 EXPORT_SYMBOL(fs_overflowuid);
115 EXPORT_SYMBOL(fs_overflowgid);
116 
117 /*
118  * Returns true if current's euid is same as p's uid or euid,
119  * or has CAP_SYS_NICE to p's user_ns.
120  *
121  * Called with rcu_read_lock, creds are safe
122  */
123 static bool set_one_prio_perm(struct task_struct *p)
124 {
125 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
126 
127 	if (uid_eq(pcred->uid,  cred->euid) ||
128 	    uid_eq(pcred->euid, cred->euid))
129 		return true;
130 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
131 		return true;
132 	return false;
133 }
134 
135 /*
136  * set the priority of a task
137  * - the caller must hold the RCU read lock
138  */
139 static int set_one_prio(struct task_struct *p, int niceval, int error)
140 {
141 	int no_nice;
142 
143 	if (!set_one_prio_perm(p)) {
144 		error = -EPERM;
145 		goto out;
146 	}
147 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
148 		error = -EACCES;
149 		goto out;
150 	}
151 	no_nice = security_task_setnice(p, niceval);
152 	if (no_nice) {
153 		error = no_nice;
154 		goto out;
155 	}
156 	if (error == -ESRCH)
157 		error = 0;
158 	set_user_nice(p, niceval);
159 out:
160 	return error;
161 }
162 
163 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
164 {
165 	struct task_struct *g, *p;
166 	struct user_struct *user;
167 	const struct cred *cred = current_cred();
168 	int error = -EINVAL;
169 	struct pid *pgrp;
170 	kuid_t uid;
171 
172 	if (which > PRIO_USER || which < PRIO_PROCESS)
173 		goto out;
174 
175 	/* normalize: avoid signed division (rounding problems) */
176 	error = -ESRCH;
177 	if (niceval < MIN_NICE)
178 		niceval = MIN_NICE;
179 	if (niceval > MAX_NICE)
180 		niceval = MAX_NICE;
181 
182 	rcu_read_lock();
183 	read_lock(&tasklist_lock);
184 	switch (which) {
185 		case PRIO_PROCESS:
186 			if (who)
187 				p = find_task_by_vpid(who);
188 			else
189 				p = current;
190 			if (p)
191 				error = set_one_prio(p, niceval, error);
192 			break;
193 		case PRIO_PGRP:
194 			if (who)
195 				pgrp = find_vpid(who);
196 			else
197 				pgrp = task_pgrp(current);
198 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
199 				error = set_one_prio(p, niceval, error);
200 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
201 			break;
202 		case PRIO_USER:
203 			uid = make_kuid(cred->user_ns, who);
204 			user = cred->user;
205 			if (!who)
206 				uid = cred->uid;
207 			else if (!uid_eq(uid, cred->uid) &&
208 				 !(user = find_user(uid)))
209 				goto out_unlock;	/* No processes for this user */
210 
211 			do_each_thread(g, p) {
212 				if (uid_eq(task_uid(p), uid))
213 					error = set_one_prio(p, niceval, error);
214 			} while_each_thread(g, p);
215 			if (!uid_eq(uid, cred->uid))
216 				free_uid(user);		/* For find_user() */
217 			break;
218 	}
219 out_unlock:
220 	read_unlock(&tasklist_lock);
221 	rcu_read_unlock();
222 out:
223 	return error;
224 }
225 
226 /*
227  * Ugh. To avoid negative return values, "getpriority()" will
228  * not return the normal nice-value, but a negated value that
229  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
230  * to stay compatible.
231  */
232 SYSCALL_DEFINE2(getpriority, int, which, int, who)
233 {
234 	struct task_struct *g, *p;
235 	struct user_struct *user;
236 	const struct cred *cred = current_cred();
237 	long niceval, retval = -ESRCH;
238 	struct pid *pgrp;
239 	kuid_t uid;
240 
241 	if (which > PRIO_USER || which < PRIO_PROCESS)
242 		return -EINVAL;
243 
244 	rcu_read_lock();
245 	read_lock(&tasklist_lock);
246 	switch (which) {
247 		case PRIO_PROCESS:
248 			if (who)
249 				p = find_task_by_vpid(who);
250 			else
251 				p = current;
252 			if (p) {
253 				niceval = nice_to_rlimit(task_nice(p));
254 				if (niceval > retval)
255 					retval = niceval;
256 			}
257 			break;
258 		case PRIO_PGRP:
259 			if (who)
260 				pgrp = find_vpid(who);
261 			else
262 				pgrp = task_pgrp(current);
263 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
264 				niceval = nice_to_rlimit(task_nice(p));
265 				if (niceval > retval)
266 					retval = niceval;
267 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
268 			break;
269 		case PRIO_USER:
270 			uid = make_kuid(cred->user_ns, who);
271 			user = cred->user;
272 			if (!who)
273 				uid = cred->uid;
274 			else if (!uid_eq(uid, cred->uid) &&
275 				 !(user = find_user(uid)))
276 				goto out_unlock;	/* No processes for this user */
277 
278 			do_each_thread(g, p) {
279 				if (uid_eq(task_uid(p), uid)) {
280 					niceval = nice_to_rlimit(task_nice(p));
281 					if (niceval > retval)
282 						retval = niceval;
283 				}
284 			} while_each_thread(g, p);
285 			if (!uid_eq(uid, cred->uid))
286 				free_uid(user);		/* for find_user() */
287 			break;
288 	}
289 out_unlock:
290 	read_unlock(&tasklist_lock);
291 	rcu_read_unlock();
292 
293 	return retval;
294 }
295 
296 /*
297  * Unprivileged users may change the real gid to the effective gid
298  * or vice versa.  (BSD-style)
299  *
300  * If you set the real gid at all, or set the effective gid to a value not
301  * equal to the real gid, then the saved gid is set to the new effective gid.
302  *
303  * This makes it possible for a setgid program to completely drop its
304  * privileges, which is often a useful assertion to make when you are doing
305  * a security audit over a program.
306  *
307  * The general idea is that a program which uses just setregid() will be
308  * 100% compatible with BSD.  A program which uses just setgid() will be
309  * 100% compatible with POSIX with saved IDs.
310  *
311  * SMP: There are not races, the GIDs are checked only by filesystem
312  *      operations (as far as semantic preservation is concerned).
313  */
314 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
315 {
316 	struct user_namespace *ns = current_user_ns();
317 	const struct cred *old;
318 	struct cred *new;
319 	int retval;
320 	kgid_t krgid, kegid;
321 
322 	krgid = make_kgid(ns, rgid);
323 	kegid = make_kgid(ns, egid);
324 
325 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
326 		return -EINVAL;
327 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
328 		return -EINVAL;
329 
330 	new = prepare_creds();
331 	if (!new)
332 		return -ENOMEM;
333 	old = current_cred();
334 
335 	retval = -EPERM;
336 	if (rgid != (gid_t) -1) {
337 		if (gid_eq(old->gid, krgid) ||
338 		    gid_eq(old->egid, krgid) ||
339 		    ns_capable(old->user_ns, CAP_SETGID))
340 			new->gid = krgid;
341 		else
342 			goto error;
343 	}
344 	if (egid != (gid_t) -1) {
345 		if (gid_eq(old->gid, kegid) ||
346 		    gid_eq(old->egid, kegid) ||
347 		    gid_eq(old->sgid, kegid) ||
348 		    ns_capable(old->user_ns, CAP_SETGID))
349 			new->egid = kegid;
350 		else
351 			goto error;
352 	}
353 
354 	if (rgid != (gid_t) -1 ||
355 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
356 		new->sgid = new->egid;
357 	new->fsgid = new->egid;
358 
359 	return commit_creds(new);
360 
361 error:
362 	abort_creds(new);
363 	return retval;
364 }
365 
366 /*
367  * setgid() is implemented like SysV w/ SAVED_IDS
368  *
369  * SMP: Same implicit races as above.
370  */
371 SYSCALL_DEFINE1(setgid, gid_t, gid)
372 {
373 	struct user_namespace *ns = current_user_ns();
374 	const struct cred *old;
375 	struct cred *new;
376 	int retval;
377 	kgid_t kgid;
378 
379 	kgid = make_kgid(ns, gid);
380 	if (!gid_valid(kgid))
381 		return -EINVAL;
382 
383 	new = prepare_creds();
384 	if (!new)
385 		return -ENOMEM;
386 	old = current_cred();
387 
388 	retval = -EPERM;
389 	if (ns_capable(old->user_ns, CAP_SETGID))
390 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
391 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
392 		new->egid = new->fsgid = kgid;
393 	else
394 		goto error;
395 
396 	return commit_creds(new);
397 
398 error:
399 	abort_creds(new);
400 	return retval;
401 }
402 
403 /*
404  * change the user struct in a credentials set to match the new UID
405  */
406 static int set_user(struct cred *new)
407 {
408 	struct user_struct *new_user;
409 
410 	new_user = alloc_uid(new->uid);
411 	if (!new_user)
412 		return -EAGAIN;
413 
414 	/*
415 	 * We don't fail in case of NPROC limit excess here because too many
416 	 * poorly written programs don't check set*uid() return code, assuming
417 	 * it never fails if called by root.  We may still enforce NPROC limit
418 	 * for programs doing set*uid()+execve() by harmlessly deferring the
419 	 * failure to the execve() stage.
420 	 */
421 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
422 			new_user != INIT_USER)
423 		current->flags |= PF_NPROC_EXCEEDED;
424 	else
425 		current->flags &= ~PF_NPROC_EXCEEDED;
426 
427 	free_uid(new->user);
428 	new->user = new_user;
429 	return 0;
430 }
431 
432 /*
433  * Unprivileged users may change the real uid to the effective uid
434  * or vice versa.  (BSD-style)
435  *
436  * If you set the real uid at all, or set the effective uid to a value not
437  * equal to the real uid, then the saved uid is set to the new effective uid.
438  *
439  * This makes it possible for a setuid program to completely drop its
440  * privileges, which is often a useful assertion to make when you are doing
441  * a security audit over a program.
442  *
443  * The general idea is that a program which uses just setreuid() will be
444  * 100% compatible with BSD.  A program which uses just setuid() will be
445  * 100% compatible with POSIX with saved IDs.
446  */
447 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
448 {
449 	struct user_namespace *ns = current_user_ns();
450 	const struct cred *old;
451 	struct cred *new;
452 	int retval;
453 	kuid_t kruid, keuid;
454 
455 	kruid = make_kuid(ns, ruid);
456 	keuid = make_kuid(ns, euid);
457 
458 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
459 		return -EINVAL;
460 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
461 		return -EINVAL;
462 
463 	new = prepare_creds();
464 	if (!new)
465 		return -ENOMEM;
466 	old = current_cred();
467 
468 	retval = -EPERM;
469 	if (ruid != (uid_t) -1) {
470 		new->uid = kruid;
471 		if (!uid_eq(old->uid, kruid) &&
472 		    !uid_eq(old->euid, kruid) &&
473 		    !ns_capable(old->user_ns, CAP_SETUID))
474 			goto error;
475 	}
476 
477 	if (euid != (uid_t) -1) {
478 		new->euid = keuid;
479 		if (!uid_eq(old->uid, keuid) &&
480 		    !uid_eq(old->euid, keuid) &&
481 		    !uid_eq(old->suid, keuid) &&
482 		    !ns_capable(old->user_ns, CAP_SETUID))
483 			goto error;
484 	}
485 
486 	if (!uid_eq(new->uid, old->uid)) {
487 		retval = set_user(new);
488 		if (retval < 0)
489 			goto error;
490 	}
491 	if (ruid != (uid_t) -1 ||
492 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
493 		new->suid = new->euid;
494 	new->fsuid = new->euid;
495 
496 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
497 	if (retval < 0)
498 		goto error;
499 
500 	return commit_creds(new);
501 
502 error:
503 	abort_creds(new);
504 	return retval;
505 }
506 
507 /*
508  * setuid() is implemented like SysV with SAVED_IDS
509  *
510  * Note that SAVED_ID's is deficient in that a setuid root program
511  * like sendmail, for example, cannot set its uid to be a normal
512  * user and then switch back, because if you're root, setuid() sets
513  * the saved uid too.  If you don't like this, blame the bright people
514  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
515  * will allow a root program to temporarily drop privileges and be able to
516  * regain them by swapping the real and effective uid.
517  */
518 SYSCALL_DEFINE1(setuid, uid_t, uid)
519 {
520 	struct user_namespace *ns = current_user_ns();
521 	const struct cred *old;
522 	struct cred *new;
523 	int retval;
524 	kuid_t kuid;
525 
526 	kuid = make_kuid(ns, uid);
527 	if (!uid_valid(kuid))
528 		return -EINVAL;
529 
530 	new = prepare_creds();
531 	if (!new)
532 		return -ENOMEM;
533 	old = current_cred();
534 
535 	retval = -EPERM;
536 	if (ns_capable(old->user_ns, CAP_SETUID)) {
537 		new->suid = new->uid = kuid;
538 		if (!uid_eq(kuid, old->uid)) {
539 			retval = set_user(new);
540 			if (retval < 0)
541 				goto error;
542 		}
543 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
544 		goto error;
545 	}
546 
547 	new->fsuid = new->euid = kuid;
548 
549 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
550 	if (retval < 0)
551 		goto error;
552 
553 	return commit_creds(new);
554 
555 error:
556 	abort_creds(new);
557 	return retval;
558 }
559 
560 
561 /*
562  * This function implements a generic ability to update ruid, euid,
563  * and suid.  This allows you to implement the 4.4 compatible seteuid().
564  */
565 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
566 {
567 	struct user_namespace *ns = current_user_ns();
568 	const struct cred *old;
569 	struct cred *new;
570 	int retval;
571 	kuid_t kruid, keuid, ksuid;
572 
573 	kruid = make_kuid(ns, ruid);
574 	keuid = make_kuid(ns, euid);
575 	ksuid = make_kuid(ns, suid);
576 
577 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
578 		return -EINVAL;
579 
580 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
581 		return -EINVAL;
582 
583 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
584 		return -EINVAL;
585 
586 	new = prepare_creds();
587 	if (!new)
588 		return -ENOMEM;
589 
590 	old = current_cred();
591 
592 	retval = -EPERM;
593 	if (!ns_capable(old->user_ns, CAP_SETUID)) {
594 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
595 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
596 			goto error;
597 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
598 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
599 			goto error;
600 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
601 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
602 			goto error;
603 	}
604 
605 	if (ruid != (uid_t) -1) {
606 		new->uid = kruid;
607 		if (!uid_eq(kruid, old->uid)) {
608 			retval = set_user(new);
609 			if (retval < 0)
610 				goto error;
611 		}
612 	}
613 	if (euid != (uid_t) -1)
614 		new->euid = keuid;
615 	if (suid != (uid_t) -1)
616 		new->suid = ksuid;
617 	new->fsuid = new->euid;
618 
619 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
620 	if (retval < 0)
621 		goto error;
622 
623 	return commit_creds(new);
624 
625 error:
626 	abort_creds(new);
627 	return retval;
628 }
629 
630 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
631 {
632 	const struct cred *cred = current_cred();
633 	int retval;
634 	uid_t ruid, euid, suid;
635 
636 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
637 	euid = from_kuid_munged(cred->user_ns, cred->euid);
638 	suid = from_kuid_munged(cred->user_ns, cred->suid);
639 
640 	if (!(retval   = put_user(ruid, ruidp)) &&
641 	    !(retval   = put_user(euid, euidp)))
642 		retval = put_user(suid, suidp);
643 
644 	return retval;
645 }
646 
647 /*
648  * Same as above, but for rgid, egid, sgid.
649  */
650 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
651 {
652 	struct user_namespace *ns = current_user_ns();
653 	const struct cred *old;
654 	struct cred *new;
655 	int retval;
656 	kgid_t krgid, kegid, ksgid;
657 
658 	krgid = make_kgid(ns, rgid);
659 	kegid = make_kgid(ns, egid);
660 	ksgid = make_kgid(ns, sgid);
661 
662 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
663 		return -EINVAL;
664 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
665 		return -EINVAL;
666 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
667 		return -EINVAL;
668 
669 	new = prepare_creds();
670 	if (!new)
671 		return -ENOMEM;
672 	old = current_cred();
673 
674 	retval = -EPERM;
675 	if (!ns_capable(old->user_ns, CAP_SETGID)) {
676 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
677 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
678 			goto error;
679 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
680 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
681 			goto error;
682 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
683 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
684 			goto error;
685 	}
686 
687 	if (rgid != (gid_t) -1)
688 		new->gid = krgid;
689 	if (egid != (gid_t) -1)
690 		new->egid = kegid;
691 	if (sgid != (gid_t) -1)
692 		new->sgid = ksgid;
693 	new->fsgid = new->egid;
694 
695 	return commit_creds(new);
696 
697 error:
698 	abort_creds(new);
699 	return retval;
700 }
701 
702 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
703 {
704 	const struct cred *cred = current_cred();
705 	int retval;
706 	gid_t rgid, egid, sgid;
707 
708 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
709 	egid = from_kgid_munged(cred->user_ns, cred->egid);
710 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
711 
712 	if (!(retval   = put_user(rgid, rgidp)) &&
713 	    !(retval   = put_user(egid, egidp)))
714 		retval = put_user(sgid, sgidp);
715 
716 	return retval;
717 }
718 
719 
720 /*
721  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
722  * is used for "access()" and for the NFS daemon (letting nfsd stay at
723  * whatever uid it wants to). It normally shadows "euid", except when
724  * explicitly set by setfsuid() or for access..
725  */
726 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
727 {
728 	const struct cred *old;
729 	struct cred *new;
730 	uid_t old_fsuid;
731 	kuid_t kuid;
732 
733 	old = current_cred();
734 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
735 
736 	kuid = make_kuid(old->user_ns, uid);
737 	if (!uid_valid(kuid))
738 		return old_fsuid;
739 
740 	new = prepare_creds();
741 	if (!new)
742 		return old_fsuid;
743 
744 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
745 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
746 	    ns_capable(old->user_ns, CAP_SETUID)) {
747 		if (!uid_eq(kuid, old->fsuid)) {
748 			new->fsuid = kuid;
749 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
750 				goto change_okay;
751 		}
752 	}
753 
754 	abort_creds(new);
755 	return old_fsuid;
756 
757 change_okay:
758 	commit_creds(new);
759 	return old_fsuid;
760 }
761 
762 /*
763  * Samma på svenska..
764  */
765 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
766 {
767 	const struct cred *old;
768 	struct cred *new;
769 	gid_t old_fsgid;
770 	kgid_t kgid;
771 
772 	old = current_cred();
773 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
774 
775 	kgid = make_kgid(old->user_ns, gid);
776 	if (!gid_valid(kgid))
777 		return old_fsgid;
778 
779 	new = prepare_creds();
780 	if (!new)
781 		return old_fsgid;
782 
783 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
784 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
785 	    ns_capable(old->user_ns, CAP_SETGID)) {
786 		if (!gid_eq(kgid, old->fsgid)) {
787 			new->fsgid = kgid;
788 			goto change_okay;
789 		}
790 	}
791 
792 	abort_creds(new);
793 	return old_fsgid;
794 
795 change_okay:
796 	commit_creds(new);
797 	return old_fsgid;
798 }
799 
800 /**
801  * sys_getpid - return the thread group id of the current process
802  *
803  * Note, despite the name, this returns the tgid not the pid.  The tgid and
804  * the pid are identical unless CLONE_THREAD was specified on clone() in
805  * which case the tgid is the same in all threads of the same group.
806  *
807  * This is SMP safe as current->tgid does not change.
808  */
809 SYSCALL_DEFINE0(getpid)
810 {
811 	return task_tgid_vnr(current);
812 }
813 
814 /* Thread ID - the internal kernel "pid" */
815 SYSCALL_DEFINE0(gettid)
816 {
817 	return task_pid_vnr(current);
818 }
819 
820 /*
821  * Accessing ->real_parent is not SMP-safe, it could
822  * change from under us. However, we can use a stale
823  * value of ->real_parent under rcu_read_lock(), see
824  * release_task()->call_rcu(delayed_put_task_struct).
825  */
826 SYSCALL_DEFINE0(getppid)
827 {
828 	int pid;
829 
830 	rcu_read_lock();
831 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
832 	rcu_read_unlock();
833 
834 	return pid;
835 }
836 
837 SYSCALL_DEFINE0(getuid)
838 {
839 	/* Only we change this so SMP safe */
840 	return from_kuid_munged(current_user_ns(), current_uid());
841 }
842 
843 SYSCALL_DEFINE0(geteuid)
844 {
845 	/* Only we change this so SMP safe */
846 	return from_kuid_munged(current_user_ns(), current_euid());
847 }
848 
849 SYSCALL_DEFINE0(getgid)
850 {
851 	/* Only we change this so SMP safe */
852 	return from_kgid_munged(current_user_ns(), current_gid());
853 }
854 
855 SYSCALL_DEFINE0(getegid)
856 {
857 	/* Only we change this so SMP safe */
858 	return from_kgid_munged(current_user_ns(), current_egid());
859 }
860 
861 void do_sys_times(struct tms *tms)
862 {
863 	cputime_t tgutime, tgstime, cutime, cstime;
864 
865 	spin_lock_irq(&current->sighand->siglock);
866 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
867 	cutime = current->signal->cutime;
868 	cstime = current->signal->cstime;
869 	spin_unlock_irq(&current->sighand->siglock);
870 	tms->tms_utime = cputime_to_clock_t(tgutime);
871 	tms->tms_stime = cputime_to_clock_t(tgstime);
872 	tms->tms_cutime = cputime_to_clock_t(cutime);
873 	tms->tms_cstime = cputime_to_clock_t(cstime);
874 }
875 
876 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
877 {
878 	if (tbuf) {
879 		struct tms tmp;
880 
881 		do_sys_times(&tmp);
882 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
883 			return -EFAULT;
884 	}
885 	force_successful_syscall_return();
886 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
887 }
888 
889 /*
890  * This needs some heavy checking ...
891  * I just haven't the stomach for it. I also don't fully
892  * understand sessions/pgrp etc. Let somebody who does explain it.
893  *
894  * OK, I think I have the protection semantics right.... this is really
895  * only important on a multi-user system anyway, to make sure one user
896  * can't send a signal to a process owned by another.  -TYT, 12/12/91
897  *
898  * !PF_FORKNOEXEC check to conform completely to POSIX.
899  */
900 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
901 {
902 	struct task_struct *p;
903 	struct task_struct *group_leader = current->group_leader;
904 	struct pid *pgrp;
905 	int err;
906 
907 	if (!pid)
908 		pid = task_pid_vnr(group_leader);
909 	if (!pgid)
910 		pgid = pid;
911 	if (pgid < 0)
912 		return -EINVAL;
913 	rcu_read_lock();
914 
915 	/* From this point forward we keep holding onto the tasklist lock
916 	 * so that our parent does not change from under us. -DaveM
917 	 */
918 	write_lock_irq(&tasklist_lock);
919 
920 	err = -ESRCH;
921 	p = find_task_by_vpid(pid);
922 	if (!p)
923 		goto out;
924 
925 	err = -EINVAL;
926 	if (!thread_group_leader(p))
927 		goto out;
928 
929 	if (same_thread_group(p->real_parent, group_leader)) {
930 		err = -EPERM;
931 		if (task_session(p) != task_session(group_leader))
932 			goto out;
933 		err = -EACCES;
934 		if (!(p->flags & PF_FORKNOEXEC))
935 			goto out;
936 	} else {
937 		err = -ESRCH;
938 		if (p != group_leader)
939 			goto out;
940 	}
941 
942 	err = -EPERM;
943 	if (p->signal->leader)
944 		goto out;
945 
946 	pgrp = task_pid(p);
947 	if (pgid != pid) {
948 		struct task_struct *g;
949 
950 		pgrp = find_vpid(pgid);
951 		g = pid_task(pgrp, PIDTYPE_PGID);
952 		if (!g || task_session(g) != task_session(group_leader))
953 			goto out;
954 	}
955 
956 	err = security_task_setpgid(p, pgid);
957 	if (err)
958 		goto out;
959 
960 	if (task_pgrp(p) != pgrp)
961 		change_pid(p, PIDTYPE_PGID, pgrp);
962 
963 	err = 0;
964 out:
965 	/* All paths lead to here, thus we are safe. -DaveM */
966 	write_unlock_irq(&tasklist_lock);
967 	rcu_read_unlock();
968 	return err;
969 }
970 
971 SYSCALL_DEFINE1(getpgid, pid_t, pid)
972 {
973 	struct task_struct *p;
974 	struct pid *grp;
975 	int retval;
976 
977 	rcu_read_lock();
978 	if (!pid)
979 		grp = task_pgrp(current);
980 	else {
981 		retval = -ESRCH;
982 		p = find_task_by_vpid(pid);
983 		if (!p)
984 			goto out;
985 		grp = task_pgrp(p);
986 		if (!grp)
987 			goto out;
988 
989 		retval = security_task_getpgid(p);
990 		if (retval)
991 			goto out;
992 	}
993 	retval = pid_vnr(grp);
994 out:
995 	rcu_read_unlock();
996 	return retval;
997 }
998 
999 #ifdef __ARCH_WANT_SYS_GETPGRP
1000 
1001 SYSCALL_DEFINE0(getpgrp)
1002 {
1003 	return sys_getpgid(0);
1004 }
1005 
1006 #endif
1007 
1008 SYSCALL_DEFINE1(getsid, pid_t, pid)
1009 {
1010 	struct task_struct *p;
1011 	struct pid *sid;
1012 	int retval;
1013 
1014 	rcu_read_lock();
1015 	if (!pid)
1016 		sid = task_session(current);
1017 	else {
1018 		retval = -ESRCH;
1019 		p = find_task_by_vpid(pid);
1020 		if (!p)
1021 			goto out;
1022 		sid = task_session(p);
1023 		if (!sid)
1024 			goto out;
1025 
1026 		retval = security_task_getsid(p);
1027 		if (retval)
1028 			goto out;
1029 	}
1030 	retval = pid_vnr(sid);
1031 out:
1032 	rcu_read_unlock();
1033 	return retval;
1034 }
1035 
1036 static void set_special_pids(struct pid *pid)
1037 {
1038 	struct task_struct *curr = current->group_leader;
1039 
1040 	if (task_session(curr) != pid)
1041 		change_pid(curr, PIDTYPE_SID, pid);
1042 
1043 	if (task_pgrp(curr) != pid)
1044 		change_pid(curr, PIDTYPE_PGID, pid);
1045 }
1046 
1047 SYSCALL_DEFINE0(setsid)
1048 {
1049 	struct task_struct *group_leader = current->group_leader;
1050 	struct pid *sid = task_pid(group_leader);
1051 	pid_t session = pid_vnr(sid);
1052 	int err = -EPERM;
1053 
1054 	write_lock_irq(&tasklist_lock);
1055 	/* Fail if I am already a session leader */
1056 	if (group_leader->signal->leader)
1057 		goto out;
1058 
1059 	/* Fail if a process group id already exists that equals the
1060 	 * proposed session id.
1061 	 */
1062 	if (pid_task(sid, PIDTYPE_PGID))
1063 		goto out;
1064 
1065 	group_leader->signal->leader = 1;
1066 	set_special_pids(sid);
1067 
1068 	proc_clear_tty(group_leader);
1069 
1070 	err = session;
1071 out:
1072 	write_unlock_irq(&tasklist_lock);
1073 	if (err > 0) {
1074 		proc_sid_connector(group_leader);
1075 		sched_autogroup_create_attach(group_leader);
1076 	}
1077 	return err;
1078 }
1079 
1080 DECLARE_RWSEM(uts_sem);
1081 
1082 #ifdef COMPAT_UTS_MACHINE
1083 #define override_architecture(name) \
1084 	(personality(current->personality) == PER_LINUX32 && \
1085 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1086 		      sizeof(COMPAT_UTS_MACHINE)))
1087 #else
1088 #define override_architecture(name)	0
1089 #endif
1090 
1091 /*
1092  * Work around broken programs that cannot handle "Linux 3.0".
1093  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1094  */
1095 static int override_release(char __user *release, size_t len)
1096 {
1097 	int ret = 0;
1098 
1099 	if (current->personality & UNAME26) {
1100 		const char *rest = UTS_RELEASE;
1101 		char buf[65] = { 0 };
1102 		int ndots = 0;
1103 		unsigned v;
1104 		size_t copy;
1105 
1106 		while (*rest) {
1107 			if (*rest == '.' && ++ndots >= 3)
1108 				break;
1109 			if (!isdigit(*rest) && *rest != '.')
1110 				break;
1111 			rest++;
1112 		}
1113 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1114 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1115 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1116 		ret = copy_to_user(release, buf, copy + 1);
1117 	}
1118 	return ret;
1119 }
1120 
1121 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1122 {
1123 	int errno = 0;
1124 
1125 	down_read(&uts_sem);
1126 	if (copy_to_user(name, utsname(), sizeof *name))
1127 		errno = -EFAULT;
1128 	up_read(&uts_sem);
1129 
1130 	if (!errno && override_release(name->release, sizeof(name->release)))
1131 		errno = -EFAULT;
1132 	if (!errno && override_architecture(name))
1133 		errno = -EFAULT;
1134 	return errno;
1135 }
1136 
1137 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1138 /*
1139  * Old cruft
1140  */
1141 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1142 {
1143 	int error = 0;
1144 
1145 	if (!name)
1146 		return -EFAULT;
1147 
1148 	down_read(&uts_sem);
1149 	if (copy_to_user(name, utsname(), sizeof(*name)))
1150 		error = -EFAULT;
1151 	up_read(&uts_sem);
1152 
1153 	if (!error && override_release(name->release, sizeof(name->release)))
1154 		error = -EFAULT;
1155 	if (!error && override_architecture(name))
1156 		error = -EFAULT;
1157 	return error;
1158 }
1159 
1160 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1161 {
1162 	int error;
1163 
1164 	if (!name)
1165 		return -EFAULT;
1166 	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1167 		return -EFAULT;
1168 
1169 	down_read(&uts_sem);
1170 	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1171 			       __OLD_UTS_LEN);
1172 	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1173 	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1174 				__OLD_UTS_LEN);
1175 	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1176 	error |= __copy_to_user(&name->release, &utsname()->release,
1177 				__OLD_UTS_LEN);
1178 	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1179 	error |= __copy_to_user(&name->version, &utsname()->version,
1180 				__OLD_UTS_LEN);
1181 	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1182 	error |= __copy_to_user(&name->machine, &utsname()->machine,
1183 				__OLD_UTS_LEN);
1184 	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1185 	up_read(&uts_sem);
1186 
1187 	if (!error && override_architecture(name))
1188 		error = -EFAULT;
1189 	if (!error && override_release(name->release, sizeof(name->release)))
1190 		error = -EFAULT;
1191 	return error ? -EFAULT : 0;
1192 }
1193 #endif
1194 
1195 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1196 {
1197 	int errno;
1198 	char tmp[__NEW_UTS_LEN];
1199 
1200 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1201 		return -EPERM;
1202 
1203 	if (len < 0 || len > __NEW_UTS_LEN)
1204 		return -EINVAL;
1205 	down_write(&uts_sem);
1206 	errno = -EFAULT;
1207 	if (!copy_from_user(tmp, name, len)) {
1208 		struct new_utsname *u = utsname();
1209 
1210 		memcpy(u->nodename, tmp, len);
1211 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1212 		errno = 0;
1213 		uts_proc_notify(UTS_PROC_HOSTNAME);
1214 	}
1215 	up_write(&uts_sem);
1216 	return errno;
1217 }
1218 
1219 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1220 
1221 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1222 {
1223 	int i, errno;
1224 	struct new_utsname *u;
1225 
1226 	if (len < 0)
1227 		return -EINVAL;
1228 	down_read(&uts_sem);
1229 	u = utsname();
1230 	i = 1 + strlen(u->nodename);
1231 	if (i > len)
1232 		i = len;
1233 	errno = 0;
1234 	if (copy_to_user(name, u->nodename, i))
1235 		errno = -EFAULT;
1236 	up_read(&uts_sem);
1237 	return errno;
1238 }
1239 
1240 #endif
1241 
1242 /*
1243  * Only setdomainname; getdomainname can be implemented by calling
1244  * uname()
1245  */
1246 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1247 {
1248 	int errno;
1249 	char tmp[__NEW_UTS_LEN];
1250 
1251 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1252 		return -EPERM;
1253 	if (len < 0 || len > __NEW_UTS_LEN)
1254 		return -EINVAL;
1255 
1256 	down_write(&uts_sem);
1257 	errno = -EFAULT;
1258 	if (!copy_from_user(tmp, name, len)) {
1259 		struct new_utsname *u = utsname();
1260 
1261 		memcpy(u->domainname, tmp, len);
1262 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1263 		errno = 0;
1264 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1265 	}
1266 	up_write(&uts_sem);
1267 	return errno;
1268 }
1269 
1270 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1271 {
1272 	struct rlimit value;
1273 	int ret;
1274 
1275 	ret = do_prlimit(current, resource, NULL, &value);
1276 	if (!ret)
1277 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1278 
1279 	return ret;
1280 }
1281 
1282 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1283 
1284 /*
1285  *	Back compatibility for getrlimit. Needed for some apps.
1286  */
1287 
1288 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1289 		struct rlimit __user *, rlim)
1290 {
1291 	struct rlimit x;
1292 	if (resource >= RLIM_NLIMITS)
1293 		return -EINVAL;
1294 
1295 	task_lock(current->group_leader);
1296 	x = current->signal->rlim[resource];
1297 	task_unlock(current->group_leader);
1298 	if (x.rlim_cur > 0x7FFFFFFF)
1299 		x.rlim_cur = 0x7FFFFFFF;
1300 	if (x.rlim_max > 0x7FFFFFFF)
1301 		x.rlim_max = 0x7FFFFFFF;
1302 	return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1303 }
1304 
1305 #endif
1306 
1307 static inline bool rlim64_is_infinity(__u64 rlim64)
1308 {
1309 #if BITS_PER_LONG < 64
1310 	return rlim64 >= ULONG_MAX;
1311 #else
1312 	return rlim64 == RLIM64_INFINITY;
1313 #endif
1314 }
1315 
1316 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1317 {
1318 	if (rlim->rlim_cur == RLIM_INFINITY)
1319 		rlim64->rlim_cur = RLIM64_INFINITY;
1320 	else
1321 		rlim64->rlim_cur = rlim->rlim_cur;
1322 	if (rlim->rlim_max == RLIM_INFINITY)
1323 		rlim64->rlim_max = RLIM64_INFINITY;
1324 	else
1325 		rlim64->rlim_max = rlim->rlim_max;
1326 }
1327 
1328 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1329 {
1330 	if (rlim64_is_infinity(rlim64->rlim_cur))
1331 		rlim->rlim_cur = RLIM_INFINITY;
1332 	else
1333 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1334 	if (rlim64_is_infinity(rlim64->rlim_max))
1335 		rlim->rlim_max = RLIM_INFINITY;
1336 	else
1337 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1338 }
1339 
1340 /* make sure you are allowed to change @tsk limits before calling this */
1341 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1342 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1343 {
1344 	struct rlimit *rlim;
1345 	int retval = 0;
1346 
1347 	if (resource >= RLIM_NLIMITS)
1348 		return -EINVAL;
1349 	if (new_rlim) {
1350 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1351 			return -EINVAL;
1352 		if (resource == RLIMIT_NOFILE &&
1353 				new_rlim->rlim_max > sysctl_nr_open)
1354 			return -EPERM;
1355 	}
1356 
1357 	/* protect tsk->signal and tsk->sighand from disappearing */
1358 	read_lock(&tasklist_lock);
1359 	if (!tsk->sighand) {
1360 		retval = -ESRCH;
1361 		goto out;
1362 	}
1363 
1364 	rlim = tsk->signal->rlim + resource;
1365 	task_lock(tsk->group_leader);
1366 	if (new_rlim) {
1367 		/* Keep the capable check against init_user_ns until
1368 		   cgroups can contain all limits */
1369 		if (new_rlim->rlim_max > rlim->rlim_max &&
1370 				!capable(CAP_SYS_RESOURCE))
1371 			retval = -EPERM;
1372 		if (!retval)
1373 			retval = security_task_setrlimit(tsk->group_leader,
1374 					resource, new_rlim);
1375 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1376 			/*
1377 			 * The caller is asking for an immediate RLIMIT_CPU
1378 			 * expiry.  But we use the zero value to mean "it was
1379 			 * never set".  So let's cheat and make it one second
1380 			 * instead
1381 			 */
1382 			new_rlim->rlim_cur = 1;
1383 		}
1384 	}
1385 	if (!retval) {
1386 		if (old_rlim)
1387 			*old_rlim = *rlim;
1388 		if (new_rlim)
1389 			*rlim = *new_rlim;
1390 	}
1391 	task_unlock(tsk->group_leader);
1392 
1393 	/*
1394 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1395 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1396 	 * very long-standing error, and fixing it now risks breakage of
1397 	 * applications, so we live with it
1398 	 */
1399 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1400 			 new_rlim->rlim_cur != RLIM_INFINITY)
1401 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1402 out:
1403 	read_unlock(&tasklist_lock);
1404 	return retval;
1405 }
1406 
1407 /* rcu lock must be held */
1408 static int check_prlimit_permission(struct task_struct *task)
1409 {
1410 	const struct cred *cred = current_cred(), *tcred;
1411 
1412 	if (current == task)
1413 		return 0;
1414 
1415 	tcred = __task_cred(task);
1416 	if (uid_eq(cred->uid, tcred->euid) &&
1417 	    uid_eq(cred->uid, tcred->suid) &&
1418 	    uid_eq(cred->uid, tcred->uid)  &&
1419 	    gid_eq(cred->gid, tcred->egid) &&
1420 	    gid_eq(cred->gid, tcred->sgid) &&
1421 	    gid_eq(cred->gid, tcred->gid))
1422 		return 0;
1423 	if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1424 		return 0;
1425 
1426 	return -EPERM;
1427 }
1428 
1429 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1430 		const struct rlimit64 __user *, new_rlim,
1431 		struct rlimit64 __user *, old_rlim)
1432 {
1433 	struct rlimit64 old64, new64;
1434 	struct rlimit old, new;
1435 	struct task_struct *tsk;
1436 	int ret;
1437 
1438 	if (new_rlim) {
1439 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1440 			return -EFAULT;
1441 		rlim64_to_rlim(&new64, &new);
1442 	}
1443 
1444 	rcu_read_lock();
1445 	tsk = pid ? find_task_by_vpid(pid) : current;
1446 	if (!tsk) {
1447 		rcu_read_unlock();
1448 		return -ESRCH;
1449 	}
1450 	ret = check_prlimit_permission(tsk);
1451 	if (ret) {
1452 		rcu_read_unlock();
1453 		return ret;
1454 	}
1455 	get_task_struct(tsk);
1456 	rcu_read_unlock();
1457 
1458 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1459 			old_rlim ? &old : NULL);
1460 
1461 	if (!ret && old_rlim) {
1462 		rlim_to_rlim64(&old, &old64);
1463 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1464 			ret = -EFAULT;
1465 	}
1466 
1467 	put_task_struct(tsk);
1468 	return ret;
1469 }
1470 
1471 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1472 {
1473 	struct rlimit new_rlim;
1474 
1475 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1476 		return -EFAULT;
1477 	return do_prlimit(current, resource, &new_rlim, NULL);
1478 }
1479 
1480 /*
1481  * It would make sense to put struct rusage in the task_struct,
1482  * except that would make the task_struct be *really big*.  After
1483  * task_struct gets moved into malloc'ed memory, it would
1484  * make sense to do this.  It will make moving the rest of the information
1485  * a lot simpler!  (Which we're not doing right now because we're not
1486  * measuring them yet).
1487  *
1488  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1489  * races with threads incrementing their own counters.  But since word
1490  * reads are atomic, we either get new values or old values and we don't
1491  * care which for the sums.  We always take the siglock to protect reading
1492  * the c* fields from p->signal from races with exit.c updating those
1493  * fields when reaping, so a sample either gets all the additions of a
1494  * given child after it's reaped, or none so this sample is before reaping.
1495  *
1496  * Locking:
1497  * We need to take the siglock for CHILDEREN, SELF and BOTH
1498  * for  the cases current multithreaded, non-current single threaded
1499  * non-current multithreaded.  Thread traversal is now safe with
1500  * the siglock held.
1501  * Strictly speaking, we donot need to take the siglock if we are current and
1502  * single threaded,  as no one else can take our signal_struct away, no one
1503  * else can  reap the  children to update signal->c* counters, and no one else
1504  * can race with the signal-> fields. If we do not take any lock, the
1505  * signal-> fields could be read out of order while another thread was just
1506  * exiting. So we should  place a read memory barrier when we avoid the lock.
1507  * On the writer side,  write memory barrier is implied in  __exit_signal
1508  * as __exit_signal releases  the siglock spinlock after updating the signal->
1509  * fields. But we don't do this yet to keep things simple.
1510  *
1511  */
1512 
1513 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1514 {
1515 	r->ru_nvcsw += t->nvcsw;
1516 	r->ru_nivcsw += t->nivcsw;
1517 	r->ru_minflt += t->min_flt;
1518 	r->ru_majflt += t->maj_flt;
1519 	r->ru_inblock += task_io_get_inblock(t);
1520 	r->ru_oublock += task_io_get_oublock(t);
1521 }
1522 
1523 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1524 {
1525 	struct task_struct *t;
1526 	unsigned long flags;
1527 	cputime_t tgutime, tgstime, utime, stime;
1528 	unsigned long maxrss = 0;
1529 
1530 	memset((char *) r, 0, sizeof *r);
1531 	utime = stime = 0;
1532 
1533 	if (who == RUSAGE_THREAD) {
1534 		task_cputime_adjusted(current, &utime, &stime);
1535 		accumulate_thread_rusage(p, r);
1536 		maxrss = p->signal->maxrss;
1537 		goto out;
1538 	}
1539 
1540 	if (!lock_task_sighand(p, &flags))
1541 		return;
1542 
1543 	switch (who) {
1544 		case RUSAGE_BOTH:
1545 		case RUSAGE_CHILDREN:
1546 			utime = p->signal->cutime;
1547 			stime = p->signal->cstime;
1548 			r->ru_nvcsw = p->signal->cnvcsw;
1549 			r->ru_nivcsw = p->signal->cnivcsw;
1550 			r->ru_minflt = p->signal->cmin_flt;
1551 			r->ru_majflt = p->signal->cmaj_flt;
1552 			r->ru_inblock = p->signal->cinblock;
1553 			r->ru_oublock = p->signal->coublock;
1554 			maxrss = p->signal->cmaxrss;
1555 
1556 			if (who == RUSAGE_CHILDREN)
1557 				break;
1558 
1559 		case RUSAGE_SELF:
1560 			thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1561 			utime += tgutime;
1562 			stime += tgstime;
1563 			r->ru_nvcsw += p->signal->nvcsw;
1564 			r->ru_nivcsw += p->signal->nivcsw;
1565 			r->ru_minflt += p->signal->min_flt;
1566 			r->ru_majflt += p->signal->maj_flt;
1567 			r->ru_inblock += p->signal->inblock;
1568 			r->ru_oublock += p->signal->oublock;
1569 			if (maxrss < p->signal->maxrss)
1570 				maxrss = p->signal->maxrss;
1571 			t = p;
1572 			do {
1573 				accumulate_thread_rusage(t, r);
1574 			} while_each_thread(p, t);
1575 			break;
1576 
1577 		default:
1578 			BUG();
1579 	}
1580 	unlock_task_sighand(p, &flags);
1581 
1582 out:
1583 	cputime_to_timeval(utime, &r->ru_utime);
1584 	cputime_to_timeval(stime, &r->ru_stime);
1585 
1586 	if (who != RUSAGE_CHILDREN) {
1587 		struct mm_struct *mm = get_task_mm(p);
1588 		if (mm) {
1589 			setmax_mm_hiwater_rss(&maxrss, mm);
1590 			mmput(mm);
1591 		}
1592 	}
1593 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1594 }
1595 
1596 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1597 {
1598 	struct rusage r;
1599 	k_getrusage(p, who, &r);
1600 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1601 }
1602 
1603 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1604 {
1605 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1606 	    who != RUSAGE_THREAD)
1607 		return -EINVAL;
1608 	return getrusage(current, who, ru);
1609 }
1610 
1611 #ifdef CONFIG_COMPAT
1612 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1613 {
1614 	struct rusage r;
1615 
1616 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1617 	    who != RUSAGE_THREAD)
1618 		return -EINVAL;
1619 
1620 	k_getrusage(current, who, &r);
1621 	return put_compat_rusage(&r, ru);
1622 }
1623 #endif
1624 
1625 SYSCALL_DEFINE1(umask, int, mask)
1626 {
1627 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1628 	return mask;
1629 }
1630 
1631 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1632 {
1633 	struct fd exe;
1634 	struct inode *inode;
1635 	int err;
1636 
1637 	exe = fdget(fd);
1638 	if (!exe.file)
1639 		return -EBADF;
1640 
1641 	inode = file_inode(exe.file);
1642 
1643 	/*
1644 	 * Because the original mm->exe_file points to executable file, make
1645 	 * sure that this one is executable as well, to avoid breaking an
1646 	 * overall picture.
1647 	 */
1648 	err = -EACCES;
1649 	if (!S_ISREG(inode->i_mode)	||
1650 	    exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1651 		goto exit;
1652 
1653 	err = inode_permission(inode, MAY_EXEC);
1654 	if (err)
1655 		goto exit;
1656 
1657 	down_write(&mm->mmap_sem);
1658 
1659 	/*
1660 	 * Forbid mm->exe_file change if old file still mapped.
1661 	 */
1662 	err = -EBUSY;
1663 	if (mm->exe_file) {
1664 		struct vm_area_struct *vma;
1665 
1666 		for (vma = mm->mmap; vma; vma = vma->vm_next)
1667 			if (vma->vm_file &&
1668 			    path_equal(&vma->vm_file->f_path,
1669 				       &mm->exe_file->f_path))
1670 				goto exit_unlock;
1671 	}
1672 
1673 	/*
1674 	 * The symlink can be changed only once, just to disallow arbitrary
1675 	 * transitions malicious software might bring in. This means one
1676 	 * could make a snapshot over all processes running and monitor
1677 	 * /proc/pid/exe changes to notice unusual activity if needed.
1678 	 */
1679 	err = -EPERM;
1680 	if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1681 		goto exit_unlock;
1682 
1683 	err = 0;
1684 	set_mm_exe_file(mm, exe.file);	/* this grabs a reference to exe.file */
1685 exit_unlock:
1686 	up_write(&mm->mmap_sem);
1687 
1688 exit:
1689 	fdput(exe);
1690 	return err;
1691 }
1692 
1693 static int prctl_set_mm(int opt, unsigned long addr,
1694 			unsigned long arg4, unsigned long arg5)
1695 {
1696 	unsigned long rlim = rlimit(RLIMIT_DATA);
1697 	struct mm_struct *mm = current->mm;
1698 	struct vm_area_struct *vma;
1699 	int error;
1700 
1701 	if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
1702 		return -EINVAL;
1703 
1704 	if (!capable(CAP_SYS_RESOURCE))
1705 		return -EPERM;
1706 
1707 	if (opt == PR_SET_MM_EXE_FILE)
1708 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1709 
1710 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
1711 		return -EINVAL;
1712 
1713 	error = -EINVAL;
1714 
1715 	down_read(&mm->mmap_sem);
1716 	vma = find_vma(mm, addr);
1717 
1718 	switch (opt) {
1719 	case PR_SET_MM_START_CODE:
1720 		mm->start_code = addr;
1721 		break;
1722 	case PR_SET_MM_END_CODE:
1723 		mm->end_code = addr;
1724 		break;
1725 	case PR_SET_MM_START_DATA:
1726 		mm->start_data = addr;
1727 		break;
1728 	case PR_SET_MM_END_DATA:
1729 		mm->end_data = addr;
1730 		break;
1731 
1732 	case PR_SET_MM_START_BRK:
1733 		if (addr <= mm->end_data)
1734 			goto out;
1735 
1736 		if (rlim < RLIM_INFINITY &&
1737 		    (mm->brk - addr) +
1738 		    (mm->end_data - mm->start_data) > rlim)
1739 			goto out;
1740 
1741 		mm->start_brk = addr;
1742 		break;
1743 
1744 	case PR_SET_MM_BRK:
1745 		if (addr <= mm->end_data)
1746 			goto out;
1747 
1748 		if (rlim < RLIM_INFINITY &&
1749 		    (addr - mm->start_brk) +
1750 		    (mm->end_data - mm->start_data) > rlim)
1751 			goto out;
1752 
1753 		mm->brk = addr;
1754 		break;
1755 
1756 	/*
1757 	 * If command line arguments and environment
1758 	 * are placed somewhere else on stack, we can
1759 	 * set them up here, ARG_START/END to setup
1760 	 * command line argumets and ENV_START/END
1761 	 * for environment.
1762 	 */
1763 	case PR_SET_MM_START_STACK:
1764 	case PR_SET_MM_ARG_START:
1765 	case PR_SET_MM_ARG_END:
1766 	case PR_SET_MM_ENV_START:
1767 	case PR_SET_MM_ENV_END:
1768 		if (!vma) {
1769 			error = -EFAULT;
1770 			goto out;
1771 		}
1772 		if (opt == PR_SET_MM_START_STACK)
1773 			mm->start_stack = addr;
1774 		else if (opt == PR_SET_MM_ARG_START)
1775 			mm->arg_start = addr;
1776 		else if (opt == PR_SET_MM_ARG_END)
1777 			mm->arg_end = addr;
1778 		else if (opt == PR_SET_MM_ENV_START)
1779 			mm->env_start = addr;
1780 		else if (opt == PR_SET_MM_ENV_END)
1781 			mm->env_end = addr;
1782 		break;
1783 
1784 	/*
1785 	 * This doesn't move auxiliary vector itself
1786 	 * since it's pinned to mm_struct, but allow
1787 	 * to fill vector with new values. It's up
1788 	 * to a caller to provide sane values here
1789 	 * otherwise user space tools which use this
1790 	 * vector might be unhappy.
1791 	 */
1792 	case PR_SET_MM_AUXV: {
1793 		unsigned long user_auxv[AT_VECTOR_SIZE];
1794 
1795 		if (arg4 > sizeof(user_auxv))
1796 			goto out;
1797 		up_read(&mm->mmap_sem);
1798 
1799 		if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
1800 			return -EFAULT;
1801 
1802 		/* Make sure the last entry is always AT_NULL */
1803 		user_auxv[AT_VECTOR_SIZE - 2] = 0;
1804 		user_auxv[AT_VECTOR_SIZE - 1] = 0;
1805 
1806 		BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1807 
1808 		task_lock(current);
1809 		memcpy(mm->saved_auxv, user_auxv, arg4);
1810 		task_unlock(current);
1811 
1812 		return 0;
1813 	}
1814 	default:
1815 		goto out;
1816 	}
1817 
1818 	error = 0;
1819 out:
1820 	up_read(&mm->mmap_sem);
1821 	return error;
1822 }
1823 
1824 #ifdef CONFIG_CHECKPOINT_RESTORE
1825 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1826 {
1827 	return put_user(me->clear_child_tid, tid_addr);
1828 }
1829 #else
1830 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1831 {
1832 	return -EINVAL;
1833 }
1834 #endif
1835 
1836 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1837 		unsigned long, arg4, unsigned long, arg5)
1838 {
1839 	struct task_struct *me = current;
1840 	unsigned char comm[sizeof(me->comm)];
1841 	long error;
1842 
1843 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1844 	if (error != -ENOSYS)
1845 		return error;
1846 
1847 	error = 0;
1848 	switch (option) {
1849 	case PR_SET_PDEATHSIG:
1850 		if (!valid_signal(arg2)) {
1851 			error = -EINVAL;
1852 			break;
1853 		}
1854 		me->pdeath_signal = arg2;
1855 		break;
1856 	case PR_GET_PDEATHSIG:
1857 		error = put_user(me->pdeath_signal, (int __user *)arg2);
1858 		break;
1859 	case PR_GET_DUMPABLE:
1860 		error = get_dumpable(me->mm);
1861 		break;
1862 	case PR_SET_DUMPABLE:
1863 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
1864 			error = -EINVAL;
1865 			break;
1866 		}
1867 		set_dumpable(me->mm, arg2);
1868 		break;
1869 
1870 	case PR_SET_UNALIGN:
1871 		error = SET_UNALIGN_CTL(me, arg2);
1872 		break;
1873 	case PR_GET_UNALIGN:
1874 		error = GET_UNALIGN_CTL(me, arg2);
1875 		break;
1876 	case PR_SET_FPEMU:
1877 		error = SET_FPEMU_CTL(me, arg2);
1878 		break;
1879 	case PR_GET_FPEMU:
1880 		error = GET_FPEMU_CTL(me, arg2);
1881 		break;
1882 	case PR_SET_FPEXC:
1883 		error = SET_FPEXC_CTL(me, arg2);
1884 		break;
1885 	case PR_GET_FPEXC:
1886 		error = GET_FPEXC_CTL(me, arg2);
1887 		break;
1888 	case PR_GET_TIMING:
1889 		error = PR_TIMING_STATISTICAL;
1890 		break;
1891 	case PR_SET_TIMING:
1892 		if (arg2 != PR_TIMING_STATISTICAL)
1893 			error = -EINVAL;
1894 		break;
1895 	case PR_SET_NAME:
1896 		comm[sizeof(me->comm) - 1] = 0;
1897 		if (strncpy_from_user(comm, (char __user *)arg2,
1898 				      sizeof(me->comm) - 1) < 0)
1899 			return -EFAULT;
1900 		set_task_comm(me, comm);
1901 		proc_comm_connector(me);
1902 		break;
1903 	case PR_GET_NAME:
1904 		get_task_comm(comm, me);
1905 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
1906 			return -EFAULT;
1907 		break;
1908 	case PR_GET_ENDIAN:
1909 		error = GET_ENDIAN(me, arg2);
1910 		break;
1911 	case PR_SET_ENDIAN:
1912 		error = SET_ENDIAN(me, arg2);
1913 		break;
1914 	case PR_GET_SECCOMP:
1915 		error = prctl_get_seccomp();
1916 		break;
1917 	case PR_SET_SECCOMP:
1918 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
1919 		break;
1920 	case PR_GET_TSC:
1921 		error = GET_TSC_CTL(arg2);
1922 		break;
1923 	case PR_SET_TSC:
1924 		error = SET_TSC_CTL(arg2);
1925 		break;
1926 	case PR_TASK_PERF_EVENTS_DISABLE:
1927 		error = perf_event_task_disable();
1928 		break;
1929 	case PR_TASK_PERF_EVENTS_ENABLE:
1930 		error = perf_event_task_enable();
1931 		break;
1932 	case PR_GET_TIMERSLACK:
1933 		error = current->timer_slack_ns;
1934 		break;
1935 	case PR_SET_TIMERSLACK:
1936 		if (arg2 <= 0)
1937 			current->timer_slack_ns =
1938 					current->default_timer_slack_ns;
1939 		else
1940 			current->timer_slack_ns = arg2;
1941 		break;
1942 	case PR_MCE_KILL:
1943 		if (arg4 | arg5)
1944 			return -EINVAL;
1945 		switch (arg2) {
1946 		case PR_MCE_KILL_CLEAR:
1947 			if (arg3 != 0)
1948 				return -EINVAL;
1949 			current->flags &= ~PF_MCE_PROCESS;
1950 			break;
1951 		case PR_MCE_KILL_SET:
1952 			current->flags |= PF_MCE_PROCESS;
1953 			if (arg3 == PR_MCE_KILL_EARLY)
1954 				current->flags |= PF_MCE_EARLY;
1955 			else if (arg3 == PR_MCE_KILL_LATE)
1956 				current->flags &= ~PF_MCE_EARLY;
1957 			else if (arg3 == PR_MCE_KILL_DEFAULT)
1958 				current->flags &=
1959 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
1960 			else
1961 				return -EINVAL;
1962 			break;
1963 		default:
1964 			return -EINVAL;
1965 		}
1966 		break;
1967 	case PR_MCE_KILL_GET:
1968 		if (arg2 | arg3 | arg4 | arg5)
1969 			return -EINVAL;
1970 		if (current->flags & PF_MCE_PROCESS)
1971 			error = (current->flags & PF_MCE_EARLY) ?
1972 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1973 		else
1974 			error = PR_MCE_KILL_DEFAULT;
1975 		break;
1976 	case PR_SET_MM:
1977 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
1978 		break;
1979 	case PR_GET_TID_ADDRESS:
1980 		error = prctl_get_tid_address(me, (int __user **)arg2);
1981 		break;
1982 	case PR_SET_CHILD_SUBREAPER:
1983 		me->signal->is_child_subreaper = !!arg2;
1984 		break;
1985 	case PR_GET_CHILD_SUBREAPER:
1986 		error = put_user(me->signal->is_child_subreaper,
1987 				 (int __user *)arg2);
1988 		break;
1989 	case PR_SET_NO_NEW_PRIVS:
1990 		if (arg2 != 1 || arg3 || arg4 || arg5)
1991 			return -EINVAL;
1992 
1993 		task_set_no_new_privs(current);
1994 		break;
1995 	case PR_GET_NO_NEW_PRIVS:
1996 		if (arg2 || arg3 || arg4 || arg5)
1997 			return -EINVAL;
1998 		return task_no_new_privs(current) ? 1 : 0;
1999 	case PR_GET_THP_DISABLE:
2000 		if (arg2 || arg3 || arg4 || arg5)
2001 			return -EINVAL;
2002 		error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2003 		break;
2004 	case PR_SET_THP_DISABLE:
2005 		if (arg3 || arg4 || arg5)
2006 			return -EINVAL;
2007 		down_write(&me->mm->mmap_sem);
2008 		if (arg2)
2009 			me->mm->def_flags |= VM_NOHUGEPAGE;
2010 		else
2011 			me->mm->def_flags &= ~VM_NOHUGEPAGE;
2012 		up_write(&me->mm->mmap_sem);
2013 		break;
2014 	default:
2015 		error = -EINVAL;
2016 		break;
2017 	}
2018 	return error;
2019 }
2020 
2021 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2022 		struct getcpu_cache __user *, unused)
2023 {
2024 	int err = 0;
2025 	int cpu = raw_smp_processor_id();
2026 	if (cpup)
2027 		err |= put_user(cpu, cpup);
2028 	if (nodep)
2029 		err |= put_user(cpu_to_node(cpu), nodep);
2030 	return err ? -EFAULT : 0;
2031 }
2032 
2033 /**
2034  * do_sysinfo - fill in sysinfo struct
2035  * @info: pointer to buffer to fill
2036  */
2037 static int do_sysinfo(struct sysinfo *info)
2038 {
2039 	unsigned long mem_total, sav_total;
2040 	unsigned int mem_unit, bitcount;
2041 	struct timespec tp;
2042 
2043 	memset(info, 0, sizeof(struct sysinfo));
2044 
2045 	get_monotonic_boottime(&tp);
2046 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2047 
2048 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2049 
2050 	info->procs = nr_threads;
2051 
2052 	si_meminfo(info);
2053 	si_swapinfo(info);
2054 
2055 	/*
2056 	 * If the sum of all the available memory (i.e. ram + swap)
2057 	 * is less than can be stored in a 32 bit unsigned long then
2058 	 * we can be binary compatible with 2.2.x kernels.  If not,
2059 	 * well, in that case 2.2.x was broken anyways...
2060 	 *
2061 	 *  -Erik Andersen <andersee@debian.org>
2062 	 */
2063 
2064 	mem_total = info->totalram + info->totalswap;
2065 	if (mem_total < info->totalram || mem_total < info->totalswap)
2066 		goto out;
2067 	bitcount = 0;
2068 	mem_unit = info->mem_unit;
2069 	while (mem_unit > 1) {
2070 		bitcount++;
2071 		mem_unit >>= 1;
2072 		sav_total = mem_total;
2073 		mem_total <<= 1;
2074 		if (mem_total < sav_total)
2075 			goto out;
2076 	}
2077 
2078 	/*
2079 	 * If mem_total did not overflow, multiply all memory values by
2080 	 * info->mem_unit and set it to 1.  This leaves things compatible
2081 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2082 	 * kernels...
2083 	 */
2084 
2085 	info->mem_unit = 1;
2086 	info->totalram <<= bitcount;
2087 	info->freeram <<= bitcount;
2088 	info->sharedram <<= bitcount;
2089 	info->bufferram <<= bitcount;
2090 	info->totalswap <<= bitcount;
2091 	info->freeswap <<= bitcount;
2092 	info->totalhigh <<= bitcount;
2093 	info->freehigh <<= bitcount;
2094 
2095 out:
2096 	return 0;
2097 }
2098 
2099 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2100 {
2101 	struct sysinfo val;
2102 
2103 	do_sysinfo(&val);
2104 
2105 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2106 		return -EFAULT;
2107 
2108 	return 0;
2109 }
2110 
2111 #ifdef CONFIG_COMPAT
2112 struct compat_sysinfo {
2113 	s32 uptime;
2114 	u32 loads[3];
2115 	u32 totalram;
2116 	u32 freeram;
2117 	u32 sharedram;
2118 	u32 bufferram;
2119 	u32 totalswap;
2120 	u32 freeswap;
2121 	u16 procs;
2122 	u16 pad;
2123 	u32 totalhigh;
2124 	u32 freehigh;
2125 	u32 mem_unit;
2126 	char _f[20-2*sizeof(u32)-sizeof(int)];
2127 };
2128 
2129 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2130 {
2131 	struct sysinfo s;
2132 
2133 	do_sysinfo(&s);
2134 
2135 	/* Check to see if any memory value is too large for 32-bit and scale
2136 	 *  down if needed
2137 	 */
2138 	if ((s.totalram >> 32) || (s.totalswap >> 32)) {
2139 		int bitcount = 0;
2140 
2141 		while (s.mem_unit < PAGE_SIZE) {
2142 			s.mem_unit <<= 1;
2143 			bitcount++;
2144 		}
2145 
2146 		s.totalram >>= bitcount;
2147 		s.freeram >>= bitcount;
2148 		s.sharedram >>= bitcount;
2149 		s.bufferram >>= bitcount;
2150 		s.totalswap >>= bitcount;
2151 		s.freeswap >>= bitcount;
2152 		s.totalhigh >>= bitcount;
2153 		s.freehigh >>= bitcount;
2154 	}
2155 
2156 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2157 	    __put_user(s.uptime, &info->uptime) ||
2158 	    __put_user(s.loads[0], &info->loads[0]) ||
2159 	    __put_user(s.loads[1], &info->loads[1]) ||
2160 	    __put_user(s.loads[2], &info->loads[2]) ||
2161 	    __put_user(s.totalram, &info->totalram) ||
2162 	    __put_user(s.freeram, &info->freeram) ||
2163 	    __put_user(s.sharedram, &info->sharedram) ||
2164 	    __put_user(s.bufferram, &info->bufferram) ||
2165 	    __put_user(s.totalswap, &info->totalswap) ||
2166 	    __put_user(s.freeswap, &info->freeswap) ||
2167 	    __put_user(s.procs, &info->procs) ||
2168 	    __put_user(s.totalhigh, &info->totalhigh) ||
2169 	    __put_user(s.freehigh, &info->freehigh) ||
2170 	    __put_user(s.mem_unit, &info->mem_unit))
2171 		return -EFAULT;
2172 
2173 	return 0;
2174 }
2175 #endif /* CONFIG_COMPAT */
2176