xref: /openbmc/linux/kernel/sys.c (revision 6774def6)
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 			if (!user)
210 				goto out_unlock;	/* No processes for this user */
211 		}
212 		do_each_thread(g, p) {
213 			if (uid_eq(task_uid(p), uid))
214 				error = set_one_prio(p, niceval, error);
215 		} while_each_thread(g, p);
216 		if (!uid_eq(uid, cred->uid))
217 			free_uid(user);		/* For find_user() */
218 		break;
219 	}
220 out_unlock:
221 	read_unlock(&tasklist_lock);
222 	rcu_read_unlock();
223 out:
224 	return error;
225 }
226 
227 /*
228  * Ugh. To avoid negative return values, "getpriority()" will
229  * not return the normal nice-value, but a negated value that
230  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
231  * to stay compatible.
232  */
233 SYSCALL_DEFINE2(getpriority, int, which, int, who)
234 {
235 	struct task_struct *g, *p;
236 	struct user_struct *user;
237 	const struct cred *cred = current_cred();
238 	long niceval, retval = -ESRCH;
239 	struct pid *pgrp;
240 	kuid_t uid;
241 
242 	if (which > PRIO_USER || which < PRIO_PROCESS)
243 		return -EINVAL;
244 
245 	rcu_read_lock();
246 	read_lock(&tasklist_lock);
247 	switch (which) {
248 	case PRIO_PROCESS:
249 		if (who)
250 			p = find_task_by_vpid(who);
251 		else
252 			p = current;
253 		if (p) {
254 			niceval = nice_to_rlimit(task_nice(p));
255 			if (niceval > retval)
256 				retval = niceval;
257 		}
258 		break;
259 	case PRIO_PGRP:
260 		if (who)
261 			pgrp = find_vpid(who);
262 		else
263 			pgrp = task_pgrp(current);
264 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
265 			niceval = nice_to_rlimit(task_nice(p));
266 			if (niceval > retval)
267 				retval = niceval;
268 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
269 		break;
270 	case PRIO_USER:
271 		uid = make_kuid(cred->user_ns, who);
272 		user = cred->user;
273 		if (!who)
274 			uid = cred->uid;
275 		else if (!uid_eq(uid, cred->uid)) {
276 			user = find_user(uid);
277 			if (!user)
278 				goto out_unlock;	/* No processes for this user */
279 		}
280 		do_each_thread(g, p) {
281 			if (uid_eq(task_uid(p), uid)) {
282 				niceval = nice_to_rlimit(task_nice(p));
283 				if (niceval > retval)
284 					retval = niceval;
285 			}
286 		} while_each_thread(g, p);
287 		if (!uid_eq(uid, cred->uid))
288 			free_uid(user);		/* for find_user() */
289 		break;
290 	}
291 out_unlock:
292 	read_unlock(&tasklist_lock);
293 	rcu_read_unlock();
294 
295 	return retval;
296 }
297 
298 /*
299  * Unprivileged users may change the real gid to the effective gid
300  * or vice versa.  (BSD-style)
301  *
302  * If you set the real gid at all, or set the effective gid to a value not
303  * equal to the real gid, then the saved gid is set to the new effective gid.
304  *
305  * This makes it possible for a setgid program to completely drop its
306  * privileges, which is often a useful assertion to make when you are doing
307  * a security audit over a program.
308  *
309  * The general idea is that a program which uses just setregid() will be
310  * 100% compatible with BSD.  A program which uses just setgid() will be
311  * 100% compatible with POSIX with saved IDs.
312  *
313  * SMP: There are not races, the GIDs are checked only by filesystem
314  *      operations (as far as semantic preservation is concerned).
315  */
316 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
317 {
318 	struct user_namespace *ns = current_user_ns();
319 	const struct cred *old;
320 	struct cred *new;
321 	int retval;
322 	kgid_t krgid, kegid;
323 
324 	krgid = make_kgid(ns, rgid);
325 	kegid = make_kgid(ns, egid);
326 
327 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
328 		return -EINVAL;
329 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
330 		return -EINVAL;
331 
332 	new = prepare_creds();
333 	if (!new)
334 		return -ENOMEM;
335 	old = current_cred();
336 
337 	retval = -EPERM;
338 	if (rgid != (gid_t) -1) {
339 		if (gid_eq(old->gid, krgid) ||
340 		    gid_eq(old->egid, krgid) ||
341 		    ns_capable(old->user_ns, CAP_SETGID))
342 			new->gid = krgid;
343 		else
344 			goto error;
345 	}
346 	if (egid != (gid_t) -1) {
347 		if (gid_eq(old->gid, kegid) ||
348 		    gid_eq(old->egid, kegid) ||
349 		    gid_eq(old->sgid, kegid) ||
350 		    ns_capable(old->user_ns, CAP_SETGID))
351 			new->egid = kegid;
352 		else
353 			goto error;
354 	}
355 
356 	if (rgid != (gid_t) -1 ||
357 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
358 		new->sgid = new->egid;
359 	new->fsgid = new->egid;
360 
361 	return commit_creds(new);
362 
363 error:
364 	abort_creds(new);
365 	return retval;
366 }
367 
368 /*
369  * setgid() is implemented like SysV w/ SAVED_IDS
370  *
371  * SMP: Same implicit races as above.
372  */
373 SYSCALL_DEFINE1(setgid, gid_t, gid)
374 {
375 	struct user_namespace *ns = current_user_ns();
376 	const struct cred *old;
377 	struct cred *new;
378 	int retval;
379 	kgid_t kgid;
380 
381 	kgid = make_kgid(ns, gid);
382 	if (!gid_valid(kgid))
383 		return -EINVAL;
384 
385 	new = prepare_creds();
386 	if (!new)
387 		return -ENOMEM;
388 	old = current_cred();
389 
390 	retval = -EPERM;
391 	if (ns_capable(old->user_ns, CAP_SETGID))
392 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
393 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
394 		new->egid = new->fsgid = kgid;
395 	else
396 		goto error;
397 
398 	return commit_creds(new);
399 
400 error:
401 	abort_creds(new);
402 	return retval;
403 }
404 
405 /*
406  * change the user struct in a credentials set to match the new UID
407  */
408 static int set_user(struct cred *new)
409 {
410 	struct user_struct *new_user;
411 
412 	new_user = alloc_uid(new->uid);
413 	if (!new_user)
414 		return -EAGAIN;
415 
416 	/*
417 	 * We don't fail in case of NPROC limit excess here because too many
418 	 * poorly written programs don't check set*uid() return code, assuming
419 	 * it never fails if called by root.  We may still enforce NPROC limit
420 	 * for programs doing set*uid()+execve() by harmlessly deferring the
421 	 * failure to the execve() stage.
422 	 */
423 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
424 			new_user != INIT_USER)
425 		current->flags |= PF_NPROC_EXCEEDED;
426 	else
427 		current->flags &= ~PF_NPROC_EXCEEDED;
428 
429 	free_uid(new->user);
430 	new->user = new_user;
431 	return 0;
432 }
433 
434 /*
435  * Unprivileged users may change the real uid to the effective uid
436  * or vice versa.  (BSD-style)
437  *
438  * If you set the real uid at all, or set the effective uid to a value not
439  * equal to the real uid, then the saved uid is set to the new effective uid.
440  *
441  * This makes it possible for a setuid program to completely drop its
442  * privileges, which is often a useful assertion to make when you are doing
443  * a security audit over a program.
444  *
445  * The general idea is that a program which uses just setreuid() will be
446  * 100% compatible with BSD.  A program which uses just setuid() will be
447  * 100% compatible with POSIX with saved IDs.
448  */
449 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
450 {
451 	struct user_namespace *ns = current_user_ns();
452 	const struct cred *old;
453 	struct cred *new;
454 	int retval;
455 	kuid_t kruid, keuid;
456 
457 	kruid = make_kuid(ns, ruid);
458 	keuid = make_kuid(ns, euid);
459 
460 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
461 		return -EINVAL;
462 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
463 		return -EINVAL;
464 
465 	new = prepare_creds();
466 	if (!new)
467 		return -ENOMEM;
468 	old = current_cred();
469 
470 	retval = -EPERM;
471 	if (ruid != (uid_t) -1) {
472 		new->uid = kruid;
473 		if (!uid_eq(old->uid, kruid) &&
474 		    !uid_eq(old->euid, kruid) &&
475 		    !ns_capable(old->user_ns, CAP_SETUID))
476 			goto error;
477 	}
478 
479 	if (euid != (uid_t) -1) {
480 		new->euid = keuid;
481 		if (!uid_eq(old->uid, keuid) &&
482 		    !uid_eq(old->euid, keuid) &&
483 		    !uid_eq(old->suid, keuid) &&
484 		    !ns_capable(old->user_ns, CAP_SETUID))
485 			goto error;
486 	}
487 
488 	if (!uid_eq(new->uid, old->uid)) {
489 		retval = set_user(new);
490 		if (retval < 0)
491 			goto error;
492 	}
493 	if (ruid != (uid_t) -1 ||
494 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
495 		new->suid = new->euid;
496 	new->fsuid = new->euid;
497 
498 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
499 	if (retval < 0)
500 		goto error;
501 
502 	return commit_creds(new);
503 
504 error:
505 	abort_creds(new);
506 	return retval;
507 }
508 
509 /*
510  * setuid() is implemented like SysV with SAVED_IDS
511  *
512  * Note that SAVED_ID's is deficient in that a setuid root program
513  * like sendmail, for example, cannot set its uid to be a normal
514  * user and then switch back, because if you're root, setuid() sets
515  * the saved uid too.  If you don't like this, blame the bright people
516  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
517  * will allow a root program to temporarily drop privileges and be able to
518  * regain them by swapping the real and effective uid.
519  */
520 SYSCALL_DEFINE1(setuid, uid_t, uid)
521 {
522 	struct user_namespace *ns = current_user_ns();
523 	const struct cred *old;
524 	struct cred *new;
525 	int retval;
526 	kuid_t kuid;
527 
528 	kuid = make_kuid(ns, uid);
529 	if (!uid_valid(kuid))
530 		return -EINVAL;
531 
532 	new = prepare_creds();
533 	if (!new)
534 		return -ENOMEM;
535 	old = current_cred();
536 
537 	retval = -EPERM;
538 	if (ns_capable(old->user_ns, CAP_SETUID)) {
539 		new->suid = new->uid = kuid;
540 		if (!uid_eq(kuid, old->uid)) {
541 			retval = set_user(new);
542 			if (retval < 0)
543 				goto error;
544 		}
545 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
546 		goto error;
547 	}
548 
549 	new->fsuid = new->euid = kuid;
550 
551 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
552 	if (retval < 0)
553 		goto error;
554 
555 	return commit_creds(new);
556 
557 error:
558 	abort_creds(new);
559 	return retval;
560 }
561 
562 
563 /*
564  * This function implements a generic ability to update ruid, euid,
565  * and suid.  This allows you to implement the 4.4 compatible seteuid().
566  */
567 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
568 {
569 	struct user_namespace *ns = current_user_ns();
570 	const struct cred *old;
571 	struct cred *new;
572 	int retval;
573 	kuid_t kruid, keuid, ksuid;
574 
575 	kruid = make_kuid(ns, ruid);
576 	keuid = make_kuid(ns, euid);
577 	ksuid = make_kuid(ns, suid);
578 
579 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
580 		return -EINVAL;
581 
582 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
583 		return -EINVAL;
584 
585 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
586 		return -EINVAL;
587 
588 	new = prepare_creds();
589 	if (!new)
590 		return -ENOMEM;
591 
592 	old = current_cred();
593 
594 	retval = -EPERM;
595 	if (!ns_capable(old->user_ns, CAP_SETUID)) {
596 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
597 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
598 			goto error;
599 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
600 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
601 			goto error;
602 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
603 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
604 			goto error;
605 	}
606 
607 	if (ruid != (uid_t) -1) {
608 		new->uid = kruid;
609 		if (!uid_eq(kruid, old->uid)) {
610 			retval = set_user(new);
611 			if (retval < 0)
612 				goto error;
613 		}
614 	}
615 	if (euid != (uid_t) -1)
616 		new->euid = keuid;
617 	if (suid != (uid_t) -1)
618 		new->suid = ksuid;
619 	new->fsuid = new->euid;
620 
621 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
622 	if (retval < 0)
623 		goto error;
624 
625 	return commit_creds(new);
626 
627 error:
628 	abort_creds(new);
629 	return retval;
630 }
631 
632 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
633 {
634 	const struct cred *cred = current_cred();
635 	int retval;
636 	uid_t ruid, euid, suid;
637 
638 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
639 	euid = from_kuid_munged(cred->user_ns, cred->euid);
640 	suid = from_kuid_munged(cred->user_ns, cred->suid);
641 
642 	retval = put_user(ruid, ruidp);
643 	if (!retval) {
644 		retval = put_user(euid, euidp);
645 		if (!retval)
646 			return put_user(suid, suidp);
647 	}
648 	return retval;
649 }
650 
651 /*
652  * Same as above, but for rgid, egid, sgid.
653  */
654 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
655 {
656 	struct user_namespace *ns = current_user_ns();
657 	const struct cred *old;
658 	struct cred *new;
659 	int retval;
660 	kgid_t krgid, kegid, ksgid;
661 
662 	krgid = make_kgid(ns, rgid);
663 	kegid = make_kgid(ns, egid);
664 	ksgid = make_kgid(ns, sgid);
665 
666 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
667 		return -EINVAL;
668 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
669 		return -EINVAL;
670 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
671 		return -EINVAL;
672 
673 	new = prepare_creds();
674 	if (!new)
675 		return -ENOMEM;
676 	old = current_cred();
677 
678 	retval = -EPERM;
679 	if (!ns_capable(old->user_ns, CAP_SETGID)) {
680 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
681 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
682 			goto error;
683 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
684 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
685 			goto error;
686 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
687 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
688 			goto error;
689 	}
690 
691 	if (rgid != (gid_t) -1)
692 		new->gid = krgid;
693 	if (egid != (gid_t) -1)
694 		new->egid = kegid;
695 	if (sgid != (gid_t) -1)
696 		new->sgid = ksgid;
697 	new->fsgid = new->egid;
698 
699 	return commit_creds(new);
700 
701 error:
702 	abort_creds(new);
703 	return retval;
704 }
705 
706 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
707 {
708 	const struct cred *cred = current_cred();
709 	int retval;
710 	gid_t rgid, egid, sgid;
711 
712 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
713 	egid = from_kgid_munged(cred->user_ns, cred->egid);
714 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
715 
716 	retval = put_user(rgid, rgidp);
717 	if (!retval) {
718 		retval = put_user(egid, egidp);
719 		if (!retval)
720 			retval = put_user(sgid, sgidp);
721 	}
722 
723 	return retval;
724 }
725 
726 
727 /*
728  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
729  * is used for "access()" and for the NFS daemon (letting nfsd stay at
730  * whatever uid it wants to). It normally shadows "euid", except when
731  * explicitly set by setfsuid() or for access..
732  */
733 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
734 {
735 	const struct cred *old;
736 	struct cred *new;
737 	uid_t old_fsuid;
738 	kuid_t kuid;
739 
740 	old = current_cred();
741 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
742 
743 	kuid = make_kuid(old->user_ns, uid);
744 	if (!uid_valid(kuid))
745 		return old_fsuid;
746 
747 	new = prepare_creds();
748 	if (!new)
749 		return old_fsuid;
750 
751 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
752 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
753 	    ns_capable(old->user_ns, CAP_SETUID)) {
754 		if (!uid_eq(kuid, old->fsuid)) {
755 			new->fsuid = kuid;
756 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
757 				goto change_okay;
758 		}
759 	}
760 
761 	abort_creds(new);
762 	return old_fsuid;
763 
764 change_okay:
765 	commit_creds(new);
766 	return old_fsuid;
767 }
768 
769 /*
770  * Samma på svenska..
771  */
772 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
773 {
774 	const struct cred *old;
775 	struct cred *new;
776 	gid_t old_fsgid;
777 	kgid_t kgid;
778 
779 	old = current_cred();
780 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
781 
782 	kgid = make_kgid(old->user_ns, gid);
783 	if (!gid_valid(kgid))
784 		return old_fsgid;
785 
786 	new = prepare_creds();
787 	if (!new)
788 		return old_fsgid;
789 
790 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
791 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
792 	    ns_capable(old->user_ns, CAP_SETGID)) {
793 		if (!gid_eq(kgid, old->fsgid)) {
794 			new->fsgid = kgid;
795 			goto change_okay;
796 		}
797 	}
798 
799 	abort_creds(new);
800 	return old_fsgid;
801 
802 change_okay:
803 	commit_creds(new);
804 	return old_fsgid;
805 }
806 
807 /**
808  * sys_getpid - return the thread group id of the current process
809  *
810  * Note, despite the name, this returns the tgid not the pid.  The tgid and
811  * the pid are identical unless CLONE_THREAD was specified on clone() in
812  * which case the tgid is the same in all threads of the same group.
813  *
814  * This is SMP safe as current->tgid does not change.
815  */
816 SYSCALL_DEFINE0(getpid)
817 {
818 	return task_tgid_vnr(current);
819 }
820 
821 /* Thread ID - the internal kernel "pid" */
822 SYSCALL_DEFINE0(gettid)
823 {
824 	return task_pid_vnr(current);
825 }
826 
827 /*
828  * Accessing ->real_parent is not SMP-safe, it could
829  * change from under us. However, we can use a stale
830  * value of ->real_parent under rcu_read_lock(), see
831  * release_task()->call_rcu(delayed_put_task_struct).
832  */
833 SYSCALL_DEFINE0(getppid)
834 {
835 	int pid;
836 
837 	rcu_read_lock();
838 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
839 	rcu_read_unlock();
840 
841 	return pid;
842 }
843 
844 SYSCALL_DEFINE0(getuid)
845 {
846 	/* Only we change this so SMP safe */
847 	return from_kuid_munged(current_user_ns(), current_uid());
848 }
849 
850 SYSCALL_DEFINE0(geteuid)
851 {
852 	/* Only we change this so SMP safe */
853 	return from_kuid_munged(current_user_ns(), current_euid());
854 }
855 
856 SYSCALL_DEFINE0(getgid)
857 {
858 	/* Only we change this so SMP safe */
859 	return from_kgid_munged(current_user_ns(), current_gid());
860 }
861 
862 SYSCALL_DEFINE0(getegid)
863 {
864 	/* Only we change this so SMP safe */
865 	return from_kgid_munged(current_user_ns(), current_egid());
866 }
867 
868 void do_sys_times(struct tms *tms)
869 {
870 	cputime_t tgutime, tgstime, cutime, cstime;
871 
872 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
873 	cutime = current->signal->cutime;
874 	cstime = current->signal->cstime;
875 	tms->tms_utime = cputime_to_clock_t(tgutime);
876 	tms->tms_stime = cputime_to_clock_t(tgstime);
877 	tms->tms_cutime = cputime_to_clock_t(cutime);
878 	tms->tms_cstime = cputime_to_clock_t(cstime);
879 }
880 
881 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
882 {
883 	if (tbuf) {
884 		struct tms tmp;
885 
886 		do_sys_times(&tmp);
887 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
888 			return -EFAULT;
889 	}
890 	force_successful_syscall_return();
891 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
892 }
893 
894 /*
895  * This needs some heavy checking ...
896  * I just haven't the stomach for it. I also don't fully
897  * understand sessions/pgrp etc. Let somebody who does explain it.
898  *
899  * OK, I think I have the protection semantics right.... this is really
900  * only important on a multi-user system anyway, to make sure one user
901  * can't send a signal to a process owned by another.  -TYT, 12/12/91
902  *
903  * !PF_FORKNOEXEC check to conform completely to POSIX.
904  */
905 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
906 {
907 	struct task_struct *p;
908 	struct task_struct *group_leader = current->group_leader;
909 	struct pid *pgrp;
910 	int err;
911 
912 	if (!pid)
913 		pid = task_pid_vnr(group_leader);
914 	if (!pgid)
915 		pgid = pid;
916 	if (pgid < 0)
917 		return -EINVAL;
918 	rcu_read_lock();
919 
920 	/* From this point forward we keep holding onto the tasklist lock
921 	 * so that our parent does not change from under us. -DaveM
922 	 */
923 	write_lock_irq(&tasklist_lock);
924 
925 	err = -ESRCH;
926 	p = find_task_by_vpid(pid);
927 	if (!p)
928 		goto out;
929 
930 	err = -EINVAL;
931 	if (!thread_group_leader(p))
932 		goto out;
933 
934 	if (same_thread_group(p->real_parent, group_leader)) {
935 		err = -EPERM;
936 		if (task_session(p) != task_session(group_leader))
937 			goto out;
938 		err = -EACCES;
939 		if (!(p->flags & PF_FORKNOEXEC))
940 			goto out;
941 	} else {
942 		err = -ESRCH;
943 		if (p != group_leader)
944 			goto out;
945 	}
946 
947 	err = -EPERM;
948 	if (p->signal->leader)
949 		goto out;
950 
951 	pgrp = task_pid(p);
952 	if (pgid != pid) {
953 		struct task_struct *g;
954 
955 		pgrp = find_vpid(pgid);
956 		g = pid_task(pgrp, PIDTYPE_PGID);
957 		if (!g || task_session(g) != task_session(group_leader))
958 			goto out;
959 	}
960 
961 	err = security_task_setpgid(p, pgid);
962 	if (err)
963 		goto out;
964 
965 	if (task_pgrp(p) != pgrp)
966 		change_pid(p, PIDTYPE_PGID, pgrp);
967 
968 	err = 0;
969 out:
970 	/* All paths lead to here, thus we are safe. -DaveM */
971 	write_unlock_irq(&tasklist_lock);
972 	rcu_read_unlock();
973 	return err;
974 }
975 
976 SYSCALL_DEFINE1(getpgid, pid_t, pid)
977 {
978 	struct task_struct *p;
979 	struct pid *grp;
980 	int retval;
981 
982 	rcu_read_lock();
983 	if (!pid)
984 		grp = task_pgrp(current);
985 	else {
986 		retval = -ESRCH;
987 		p = find_task_by_vpid(pid);
988 		if (!p)
989 			goto out;
990 		grp = task_pgrp(p);
991 		if (!grp)
992 			goto out;
993 
994 		retval = security_task_getpgid(p);
995 		if (retval)
996 			goto out;
997 	}
998 	retval = pid_vnr(grp);
999 out:
1000 	rcu_read_unlock();
1001 	return retval;
1002 }
1003 
1004 #ifdef __ARCH_WANT_SYS_GETPGRP
1005 
1006 SYSCALL_DEFINE0(getpgrp)
1007 {
1008 	return sys_getpgid(0);
1009 }
1010 
1011 #endif
1012 
1013 SYSCALL_DEFINE1(getsid, pid_t, pid)
1014 {
1015 	struct task_struct *p;
1016 	struct pid *sid;
1017 	int retval;
1018 
1019 	rcu_read_lock();
1020 	if (!pid)
1021 		sid = task_session(current);
1022 	else {
1023 		retval = -ESRCH;
1024 		p = find_task_by_vpid(pid);
1025 		if (!p)
1026 			goto out;
1027 		sid = task_session(p);
1028 		if (!sid)
1029 			goto out;
1030 
1031 		retval = security_task_getsid(p);
1032 		if (retval)
1033 			goto out;
1034 	}
1035 	retval = pid_vnr(sid);
1036 out:
1037 	rcu_read_unlock();
1038 	return retval;
1039 }
1040 
1041 static void set_special_pids(struct pid *pid)
1042 {
1043 	struct task_struct *curr = current->group_leader;
1044 
1045 	if (task_session(curr) != pid)
1046 		change_pid(curr, PIDTYPE_SID, pid);
1047 
1048 	if (task_pgrp(curr) != pid)
1049 		change_pid(curr, PIDTYPE_PGID, pid);
1050 }
1051 
1052 SYSCALL_DEFINE0(setsid)
1053 {
1054 	struct task_struct *group_leader = current->group_leader;
1055 	struct pid *sid = task_pid(group_leader);
1056 	pid_t session = pid_vnr(sid);
1057 	int err = -EPERM;
1058 
1059 	write_lock_irq(&tasklist_lock);
1060 	/* Fail if I am already a session leader */
1061 	if (group_leader->signal->leader)
1062 		goto out;
1063 
1064 	/* Fail if a process group id already exists that equals the
1065 	 * proposed session id.
1066 	 */
1067 	if (pid_task(sid, PIDTYPE_PGID))
1068 		goto out;
1069 
1070 	group_leader->signal->leader = 1;
1071 	set_special_pids(sid);
1072 
1073 	proc_clear_tty(group_leader);
1074 
1075 	err = session;
1076 out:
1077 	write_unlock_irq(&tasklist_lock);
1078 	if (err > 0) {
1079 		proc_sid_connector(group_leader);
1080 		sched_autogroup_create_attach(group_leader);
1081 	}
1082 	return err;
1083 }
1084 
1085 DECLARE_RWSEM(uts_sem);
1086 
1087 #ifdef COMPAT_UTS_MACHINE
1088 #define override_architecture(name) \
1089 	(personality(current->personality) == PER_LINUX32 && \
1090 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1091 		      sizeof(COMPAT_UTS_MACHINE)))
1092 #else
1093 #define override_architecture(name)	0
1094 #endif
1095 
1096 /*
1097  * Work around broken programs that cannot handle "Linux 3.0".
1098  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1099  */
1100 static int override_release(char __user *release, size_t len)
1101 {
1102 	int ret = 0;
1103 
1104 	if (current->personality & UNAME26) {
1105 		const char *rest = UTS_RELEASE;
1106 		char buf[65] = { 0 };
1107 		int ndots = 0;
1108 		unsigned v;
1109 		size_t copy;
1110 
1111 		while (*rest) {
1112 			if (*rest == '.' && ++ndots >= 3)
1113 				break;
1114 			if (!isdigit(*rest) && *rest != '.')
1115 				break;
1116 			rest++;
1117 		}
1118 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1119 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1120 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1121 		ret = copy_to_user(release, buf, copy + 1);
1122 	}
1123 	return ret;
1124 }
1125 
1126 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1127 {
1128 	int errno = 0;
1129 
1130 	down_read(&uts_sem);
1131 	if (copy_to_user(name, utsname(), sizeof *name))
1132 		errno = -EFAULT;
1133 	up_read(&uts_sem);
1134 
1135 	if (!errno && override_release(name->release, sizeof(name->release)))
1136 		errno = -EFAULT;
1137 	if (!errno && override_architecture(name))
1138 		errno = -EFAULT;
1139 	return errno;
1140 }
1141 
1142 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1143 /*
1144  * Old cruft
1145  */
1146 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1147 {
1148 	int error = 0;
1149 
1150 	if (!name)
1151 		return -EFAULT;
1152 
1153 	down_read(&uts_sem);
1154 	if (copy_to_user(name, utsname(), sizeof(*name)))
1155 		error = -EFAULT;
1156 	up_read(&uts_sem);
1157 
1158 	if (!error && override_release(name->release, sizeof(name->release)))
1159 		error = -EFAULT;
1160 	if (!error && override_architecture(name))
1161 		error = -EFAULT;
1162 	return error;
1163 }
1164 
1165 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1166 {
1167 	int error;
1168 
1169 	if (!name)
1170 		return -EFAULT;
1171 	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1172 		return -EFAULT;
1173 
1174 	down_read(&uts_sem);
1175 	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1176 			       __OLD_UTS_LEN);
1177 	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1178 	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1179 				__OLD_UTS_LEN);
1180 	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1181 	error |= __copy_to_user(&name->release, &utsname()->release,
1182 				__OLD_UTS_LEN);
1183 	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1184 	error |= __copy_to_user(&name->version, &utsname()->version,
1185 				__OLD_UTS_LEN);
1186 	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1187 	error |= __copy_to_user(&name->machine, &utsname()->machine,
1188 				__OLD_UTS_LEN);
1189 	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1190 	up_read(&uts_sem);
1191 
1192 	if (!error && override_architecture(name))
1193 		error = -EFAULT;
1194 	if (!error && override_release(name->release, sizeof(name->release)))
1195 		error = -EFAULT;
1196 	return error ? -EFAULT : 0;
1197 }
1198 #endif
1199 
1200 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1201 {
1202 	int errno;
1203 	char tmp[__NEW_UTS_LEN];
1204 
1205 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1206 		return -EPERM;
1207 
1208 	if (len < 0 || len > __NEW_UTS_LEN)
1209 		return -EINVAL;
1210 	down_write(&uts_sem);
1211 	errno = -EFAULT;
1212 	if (!copy_from_user(tmp, name, len)) {
1213 		struct new_utsname *u = utsname();
1214 
1215 		memcpy(u->nodename, tmp, len);
1216 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1217 		errno = 0;
1218 		uts_proc_notify(UTS_PROC_HOSTNAME);
1219 	}
1220 	up_write(&uts_sem);
1221 	return errno;
1222 }
1223 
1224 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1225 
1226 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1227 {
1228 	int i, errno;
1229 	struct new_utsname *u;
1230 
1231 	if (len < 0)
1232 		return -EINVAL;
1233 	down_read(&uts_sem);
1234 	u = utsname();
1235 	i = 1 + strlen(u->nodename);
1236 	if (i > len)
1237 		i = len;
1238 	errno = 0;
1239 	if (copy_to_user(name, u->nodename, i))
1240 		errno = -EFAULT;
1241 	up_read(&uts_sem);
1242 	return errno;
1243 }
1244 
1245 #endif
1246 
1247 /*
1248  * Only setdomainname; getdomainname can be implemented by calling
1249  * uname()
1250  */
1251 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1252 {
1253 	int errno;
1254 	char tmp[__NEW_UTS_LEN];
1255 
1256 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1257 		return -EPERM;
1258 	if (len < 0 || len > __NEW_UTS_LEN)
1259 		return -EINVAL;
1260 
1261 	down_write(&uts_sem);
1262 	errno = -EFAULT;
1263 	if (!copy_from_user(tmp, name, len)) {
1264 		struct new_utsname *u = utsname();
1265 
1266 		memcpy(u->domainname, tmp, len);
1267 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1268 		errno = 0;
1269 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1270 	}
1271 	up_write(&uts_sem);
1272 	return errno;
1273 }
1274 
1275 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1276 {
1277 	struct rlimit value;
1278 	int ret;
1279 
1280 	ret = do_prlimit(current, resource, NULL, &value);
1281 	if (!ret)
1282 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1283 
1284 	return ret;
1285 }
1286 
1287 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1288 
1289 /*
1290  *	Back compatibility for getrlimit. Needed for some apps.
1291  */
1292 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1293 		struct rlimit __user *, rlim)
1294 {
1295 	struct rlimit x;
1296 	if (resource >= RLIM_NLIMITS)
1297 		return -EINVAL;
1298 
1299 	task_lock(current->group_leader);
1300 	x = current->signal->rlim[resource];
1301 	task_unlock(current->group_leader);
1302 	if (x.rlim_cur > 0x7FFFFFFF)
1303 		x.rlim_cur = 0x7FFFFFFF;
1304 	if (x.rlim_max > 0x7FFFFFFF)
1305 		x.rlim_max = 0x7FFFFFFF;
1306 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1307 }
1308 
1309 #endif
1310 
1311 static inline bool rlim64_is_infinity(__u64 rlim64)
1312 {
1313 #if BITS_PER_LONG < 64
1314 	return rlim64 >= ULONG_MAX;
1315 #else
1316 	return rlim64 == RLIM64_INFINITY;
1317 #endif
1318 }
1319 
1320 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1321 {
1322 	if (rlim->rlim_cur == RLIM_INFINITY)
1323 		rlim64->rlim_cur = RLIM64_INFINITY;
1324 	else
1325 		rlim64->rlim_cur = rlim->rlim_cur;
1326 	if (rlim->rlim_max == RLIM_INFINITY)
1327 		rlim64->rlim_max = RLIM64_INFINITY;
1328 	else
1329 		rlim64->rlim_max = rlim->rlim_max;
1330 }
1331 
1332 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1333 {
1334 	if (rlim64_is_infinity(rlim64->rlim_cur))
1335 		rlim->rlim_cur = RLIM_INFINITY;
1336 	else
1337 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1338 	if (rlim64_is_infinity(rlim64->rlim_max))
1339 		rlim->rlim_max = RLIM_INFINITY;
1340 	else
1341 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1342 }
1343 
1344 /* make sure you are allowed to change @tsk limits before calling this */
1345 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1346 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1347 {
1348 	struct rlimit *rlim;
1349 	int retval = 0;
1350 
1351 	if (resource >= RLIM_NLIMITS)
1352 		return -EINVAL;
1353 	if (new_rlim) {
1354 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1355 			return -EINVAL;
1356 		if (resource == RLIMIT_NOFILE &&
1357 				new_rlim->rlim_max > sysctl_nr_open)
1358 			return -EPERM;
1359 	}
1360 
1361 	/* protect tsk->signal and tsk->sighand from disappearing */
1362 	read_lock(&tasklist_lock);
1363 	if (!tsk->sighand) {
1364 		retval = -ESRCH;
1365 		goto out;
1366 	}
1367 
1368 	rlim = tsk->signal->rlim + resource;
1369 	task_lock(tsk->group_leader);
1370 	if (new_rlim) {
1371 		/* Keep the capable check against init_user_ns until
1372 		   cgroups can contain all limits */
1373 		if (new_rlim->rlim_max > rlim->rlim_max &&
1374 				!capable(CAP_SYS_RESOURCE))
1375 			retval = -EPERM;
1376 		if (!retval)
1377 			retval = security_task_setrlimit(tsk->group_leader,
1378 					resource, new_rlim);
1379 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1380 			/*
1381 			 * The caller is asking for an immediate RLIMIT_CPU
1382 			 * expiry.  But we use the zero value to mean "it was
1383 			 * never set".  So let's cheat and make it one second
1384 			 * instead
1385 			 */
1386 			new_rlim->rlim_cur = 1;
1387 		}
1388 	}
1389 	if (!retval) {
1390 		if (old_rlim)
1391 			*old_rlim = *rlim;
1392 		if (new_rlim)
1393 			*rlim = *new_rlim;
1394 	}
1395 	task_unlock(tsk->group_leader);
1396 
1397 	/*
1398 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1399 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1400 	 * very long-standing error, and fixing it now risks breakage of
1401 	 * applications, so we live with it
1402 	 */
1403 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1404 			 new_rlim->rlim_cur != RLIM_INFINITY)
1405 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1406 out:
1407 	read_unlock(&tasklist_lock);
1408 	return retval;
1409 }
1410 
1411 /* rcu lock must be held */
1412 static int check_prlimit_permission(struct task_struct *task)
1413 {
1414 	const struct cred *cred = current_cred(), *tcred;
1415 
1416 	if (current == task)
1417 		return 0;
1418 
1419 	tcred = __task_cred(task);
1420 	if (uid_eq(cred->uid, tcred->euid) &&
1421 	    uid_eq(cred->uid, tcred->suid) &&
1422 	    uid_eq(cred->uid, tcred->uid)  &&
1423 	    gid_eq(cred->gid, tcred->egid) &&
1424 	    gid_eq(cred->gid, tcred->sgid) &&
1425 	    gid_eq(cred->gid, tcred->gid))
1426 		return 0;
1427 	if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1428 		return 0;
1429 
1430 	return -EPERM;
1431 }
1432 
1433 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1434 		const struct rlimit64 __user *, new_rlim,
1435 		struct rlimit64 __user *, old_rlim)
1436 {
1437 	struct rlimit64 old64, new64;
1438 	struct rlimit old, new;
1439 	struct task_struct *tsk;
1440 	int ret;
1441 
1442 	if (new_rlim) {
1443 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1444 			return -EFAULT;
1445 		rlim64_to_rlim(&new64, &new);
1446 	}
1447 
1448 	rcu_read_lock();
1449 	tsk = pid ? find_task_by_vpid(pid) : current;
1450 	if (!tsk) {
1451 		rcu_read_unlock();
1452 		return -ESRCH;
1453 	}
1454 	ret = check_prlimit_permission(tsk);
1455 	if (ret) {
1456 		rcu_read_unlock();
1457 		return ret;
1458 	}
1459 	get_task_struct(tsk);
1460 	rcu_read_unlock();
1461 
1462 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1463 			old_rlim ? &old : NULL);
1464 
1465 	if (!ret && old_rlim) {
1466 		rlim_to_rlim64(&old, &old64);
1467 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1468 			ret = -EFAULT;
1469 	}
1470 
1471 	put_task_struct(tsk);
1472 	return ret;
1473 }
1474 
1475 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1476 {
1477 	struct rlimit new_rlim;
1478 
1479 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1480 		return -EFAULT;
1481 	return do_prlimit(current, resource, &new_rlim, NULL);
1482 }
1483 
1484 /*
1485  * It would make sense to put struct rusage in the task_struct,
1486  * except that would make the task_struct be *really big*.  After
1487  * task_struct gets moved into malloc'ed memory, it would
1488  * make sense to do this.  It will make moving the rest of the information
1489  * a lot simpler!  (Which we're not doing right now because we're not
1490  * measuring them yet).
1491  *
1492  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1493  * races with threads incrementing their own counters.  But since word
1494  * reads are atomic, we either get new values or old values and we don't
1495  * care which for the sums.  We always take the siglock to protect reading
1496  * the c* fields from p->signal from races with exit.c updating those
1497  * fields when reaping, so a sample either gets all the additions of a
1498  * given child after it's reaped, or none so this sample is before reaping.
1499  *
1500  * Locking:
1501  * We need to take the siglock for CHILDEREN, SELF and BOTH
1502  * for  the cases current multithreaded, non-current single threaded
1503  * non-current multithreaded.  Thread traversal is now safe with
1504  * the siglock held.
1505  * Strictly speaking, we donot need to take the siglock if we are current and
1506  * single threaded,  as no one else can take our signal_struct away, no one
1507  * else can  reap the  children to update signal->c* counters, and no one else
1508  * can race with the signal-> fields. If we do not take any lock, the
1509  * signal-> fields could be read out of order while another thread was just
1510  * exiting. So we should  place a read memory barrier when we avoid the lock.
1511  * On the writer side,  write memory barrier is implied in  __exit_signal
1512  * as __exit_signal releases  the siglock spinlock after updating the signal->
1513  * fields. But we don't do this yet to keep things simple.
1514  *
1515  */
1516 
1517 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1518 {
1519 	r->ru_nvcsw += t->nvcsw;
1520 	r->ru_nivcsw += t->nivcsw;
1521 	r->ru_minflt += t->min_flt;
1522 	r->ru_majflt += t->maj_flt;
1523 	r->ru_inblock += task_io_get_inblock(t);
1524 	r->ru_oublock += task_io_get_oublock(t);
1525 }
1526 
1527 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1528 {
1529 	struct task_struct *t;
1530 	unsigned long flags;
1531 	cputime_t tgutime, tgstime, utime, stime;
1532 	unsigned long maxrss = 0;
1533 
1534 	memset((char *)r, 0, sizeof (*r));
1535 	utime = stime = 0;
1536 
1537 	if (who == RUSAGE_THREAD) {
1538 		task_cputime_adjusted(current, &utime, &stime);
1539 		accumulate_thread_rusage(p, r);
1540 		maxrss = p->signal->maxrss;
1541 		goto out;
1542 	}
1543 
1544 	if (!lock_task_sighand(p, &flags))
1545 		return;
1546 
1547 	switch (who) {
1548 	case RUSAGE_BOTH:
1549 	case RUSAGE_CHILDREN:
1550 		utime = p->signal->cutime;
1551 		stime = p->signal->cstime;
1552 		r->ru_nvcsw = p->signal->cnvcsw;
1553 		r->ru_nivcsw = p->signal->cnivcsw;
1554 		r->ru_minflt = p->signal->cmin_flt;
1555 		r->ru_majflt = p->signal->cmaj_flt;
1556 		r->ru_inblock = p->signal->cinblock;
1557 		r->ru_oublock = p->signal->coublock;
1558 		maxrss = p->signal->cmaxrss;
1559 
1560 		if (who == RUSAGE_CHILDREN)
1561 			break;
1562 
1563 	case RUSAGE_SELF:
1564 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1565 		utime += tgutime;
1566 		stime += tgstime;
1567 		r->ru_nvcsw += p->signal->nvcsw;
1568 		r->ru_nivcsw += p->signal->nivcsw;
1569 		r->ru_minflt += p->signal->min_flt;
1570 		r->ru_majflt += p->signal->maj_flt;
1571 		r->ru_inblock += p->signal->inblock;
1572 		r->ru_oublock += p->signal->oublock;
1573 		if (maxrss < p->signal->maxrss)
1574 			maxrss = p->signal->maxrss;
1575 		t = p;
1576 		do {
1577 			accumulate_thread_rusage(t, r);
1578 		} while_each_thread(p, t);
1579 		break;
1580 
1581 	default:
1582 		BUG();
1583 	}
1584 	unlock_task_sighand(p, &flags);
1585 
1586 out:
1587 	cputime_to_timeval(utime, &r->ru_utime);
1588 	cputime_to_timeval(stime, &r->ru_stime);
1589 
1590 	if (who != RUSAGE_CHILDREN) {
1591 		struct mm_struct *mm = get_task_mm(p);
1592 
1593 		if (mm) {
1594 			setmax_mm_hiwater_rss(&maxrss, mm);
1595 			mmput(mm);
1596 		}
1597 	}
1598 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1599 }
1600 
1601 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1602 {
1603 	struct rusage r;
1604 
1605 	k_getrusage(p, who, &r);
1606 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1607 }
1608 
1609 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1610 {
1611 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1612 	    who != RUSAGE_THREAD)
1613 		return -EINVAL;
1614 	return getrusage(current, who, ru);
1615 }
1616 
1617 #ifdef CONFIG_COMPAT
1618 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1619 {
1620 	struct rusage r;
1621 
1622 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1623 	    who != RUSAGE_THREAD)
1624 		return -EINVAL;
1625 
1626 	k_getrusage(current, who, &r);
1627 	return put_compat_rusage(&r, ru);
1628 }
1629 #endif
1630 
1631 SYSCALL_DEFINE1(umask, int, mask)
1632 {
1633 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1634 	return mask;
1635 }
1636 
1637 static int prctl_set_mm_exe_file_locked(struct mm_struct *mm, unsigned int fd)
1638 {
1639 	struct fd exe;
1640 	struct inode *inode;
1641 	int err;
1642 
1643 	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
1644 
1645 	exe = fdget(fd);
1646 	if (!exe.file)
1647 		return -EBADF;
1648 
1649 	inode = file_inode(exe.file);
1650 
1651 	/*
1652 	 * Because the original mm->exe_file points to executable file, make
1653 	 * sure that this one is executable as well, to avoid breaking an
1654 	 * overall picture.
1655 	 */
1656 	err = -EACCES;
1657 	if (!S_ISREG(inode->i_mode)	||
1658 	    exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1659 		goto exit;
1660 
1661 	err = inode_permission(inode, MAY_EXEC);
1662 	if (err)
1663 		goto exit;
1664 
1665 	/*
1666 	 * Forbid mm->exe_file change if old file still mapped.
1667 	 */
1668 	err = -EBUSY;
1669 	if (mm->exe_file) {
1670 		struct vm_area_struct *vma;
1671 
1672 		for (vma = mm->mmap; vma; vma = vma->vm_next)
1673 			if (vma->vm_file &&
1674 			    path_equal(&vma->vm_file->f_path,
1675 				       &mm->exe_file->f_path))
1676 				goto exit;
1677 	}
1678 
1679 	/*
1680 	 * The symlink can be changed only once, just to disallow arbitrary
1681 	 * transitions malicious software might bring in. This means one
1682 	 * could make a snapshot over all processes running and monitor
1683 	 * /proc/pid/exe changes to notice unusual activity if needed.
1684 	 */
1685 	err = -EPERM;
1686 	if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1687 		goto exit;
1688 
1689 	err = 0;
1690 	set_mm_exe_file(mm, exe.file);	/* this grabs a reference to exe.file */
1691 exit:
1692 	fdput(exe);
1693 	return err;
1694 }
1695 
1696 #ifdef CONFIG_CHECKPOINT_RESTORE
1697 /*
1698  * WARNING: we don't require any capability here so be very careful
1699  * in what is allowed for modification from userspace.
1700  */
1701 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1702 {
1703 	unsigned long mmap_max_addr = TASK_SIZE;
1704 	struct mm_struct *mm = current->mm;
1705 	int error = -EINVAL, i;
1706 
1707 	static const unsigned char offsets[] = {
1708 		offsetof(struct prctl_mm_map, start_code),
1709 		offsetof(struct prctl_mm_map, end_code),
1710 		offsetof(struct prctl_mm_map, start_data),
1711 		offsetof(struct prctl_mm_map, end_data),
1712 		offsetof(struct prctl_mm_map, start_brk),
1713 		offsetof(struct prctl_mm_map, brk),
1714 		offsetof(struct prctl_mm_map, start_stack),
1715 		offsetof(struct prctl_mm_map, arg_start),
1716 		offsetof(struct prctl_mm_map, arg_end),
1717 		offsetof(struct prctl_mm_map, env_start),
1718 		offsetof(struct prctl_mm_map, env_end),
1719 	};
1720 
1721 	/*
1722 	 * Make sure the members are not somewhere outside
1723 	 * of allowed address space.
1724 	 */
1725 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1726 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1727 
1728 		if ((unsigned long)val >= mmap_max_addr ||
1729 		    (unsigned long)val < mmap_min_addr)
1730 			goto out;
1731 	}
1732 
1733 	/*
1734 	 * Make sure the pairs are ordered.
1735 	 */
1736 #define __prctl_check_order(__m1, __op, __m2)				\
1737 	((unsigned long)prctl_map->__m1 __op				\
1738 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1739 	error  = __prctl_check_order(start_code, <, end_code);
1740 	error |= __prctl_check_order(start_data, <, end_data);
1741 	error |= __prctl_check_order(start_brk, <=, brk);
1742 	error |= __prctl_check_order(arg_start, <=, arg_end);
1743 	error |= __prctl_check_order(env_start, <=, env_end);
1744 	if (error)
1745 		goto out;
1746 #undef __prctl_check_order
1747 
1748 	error = -EINVAL;
1749 
1750 	/*
1751 	 * @brk should be after @end_data in traditional maps.
1752 	 */
1753 	if (prctl_map->start_brk <= prctl_map->end_data ||
1754 	    prctl_map->brk <= prctl_map->end_data)
1755 		goto out;
1756 
1757 	/*
1758 	 * Neither we should allow to override limits if they set.
1759 	 */
1760 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1761 			      prctl_map->start_brk, prctl_map->end_data,
1762 			      prctl_map->start_data))
1763 			goto out;
1764 
1765 	/*
1766 	 * Someone is trying to cheat the auxv vector.
1767 	 */
1768 	if (prctl_map->auxv_size) {
1769 		if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1770 			goto out;
1771 	}
1772 
1773 	/*
1774 	 * Finally, make sure the caller has the rights to
1775 	 * change /proc/pid/exe link: only local root should
1776 	 * be allowed to.
1777 	 */
1778 	if (prctl_map->exe_fd != (u32)-1) {
1779 		struct user_namespace *ns = current_user_ns();
1780 		const struct cred *cred = current_cred();
1781 
1782 		if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1783 		    !gid_eq(cred->gid, make_kgid(ns, 0)))
1784 			goto out;
1785 	}
1786 
1787 	error = 0;
1788 out:
1789 	return error;
1790 }
1791 
1792 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1793 {
1794 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1795 	unsigned long user_auxv[AT_VECTOR_SIZE];
1796 	struct mm_struct *mm = current->mm;
1797 	int error;
1798 
1799 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1800 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1801 
1802 	if (opt == PR_SET_MM_MAP_SIZE)
1803 		return put_user((unsigned int)sizeof(prctl_map),
1804 				(unsigned int __user *)addr);
1805 
1806 	if (data_size != sizeof(prctl_map))
1807 		return -EINVAL;
1808 
1809 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1810 		return -EFAULT;
1811 
1812 	error = validate_prctl_map(&prctl_map);
1813 	if (error)
1814 		return error;
1815 
1816 	if (prctl_map.auxv_size) {
1817 		memset(user_auxv, 0, sizeof(user_auxv));
1818 		if (copy_from_user(user_auxv,
1819 				   (const void __user *)prctl_map.auxv,
1820 				   prctl_map.auxv_size))
1821 			return -EFAULT;
1822 
1823 		/* Last entry must be AT_NULL as specification requires */
1824 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1825 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1826 	}
1827 
1828 	down_write(&mm->mmap_sem);
1829 	if (prctl_map.exe_fd != (u32)-1)
1830 		error = prctl_set_mm_exe_file_locked(mm, prctl_map.exe_fd);
1831 	downgrade_write(&mm->mmap_sem);
1832 	if (error)
1833 		goto out;
1834 
1835 	/*
1836 	 * We don't validate if these members are pointing to
1837 	 * real present VMAs because application may have correspond
1838 	 * VMAs already unmapped and kernel uses these members for statistics
1839 	 * output in procfs mostly, except
1840 	 *
1841 	 *  - @start_brk/@brk which are used in do_brk but kernel lookups
1842 	 *    for VMAs when updating these memvers so anything wrong written
1843 	 *    here cause kernel to swear at userspace program but won't lead
1844 	 *    to any problem in kernel itself
1845 	 */
1846 
1847 	mm->start_code	= prctl_map.start_code;
1848 	mm->end_code	= prctl_map.end_code;
1849 	mm->start_data	= prctl_map.start_data;
1850 	mm->end_data	= prctl_map.end_data;
1851 	mm->start_brk	= prctl_map.start_brk;
1852 	mm->brk		= prctl_map.brk;
1853 	mm->start_stack	= prctl_map.start_stack;
1854 	mm->arg_start	= prctl_map.arg_start;
1855 	mm->arg_end	= prctl_map.arg_end;
1856 	mm->env_start	= prctl_map.env_start;
1857 	mm->env_end	= prctl_map.env_end;
1858 
1859 	/*
1860 	 * Note this update of @saved_auxv is lockless thus
1861 	 * if someone reads this member in procfs while we're
1862 	 * updating -- it may get partly updated results. It's
1863 	 * known and acceptable trade off: we leave it as is to
1864 	 * not introduce additional locks here making the kernel
1865 	 * more complex.
1866 	 */
1867 	if (prctl_map.auxv_size)
1868 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1869 
1870 	error = 0;
1871 out:
1872 	up_read(&mm->mmap_sem);
1873 	return error;
1874 }
1875 #endif /* CONFIG_CHECKPOINT_RESTORE */
1876 
1877 static int prctl_set_mm(int opt, unsigned long addr,
1878 			unsigned long arg4, unsigned long arg5)
1879 {
1880 	struct mm_struct *mm = current->mm;
1881 	struct vm_area_struct *vma;
1882 	int error;
1883 
1884 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1885 			      opt != PR_SET_MM_MAP &&
1886 			      opt != PR_SET_MM_MAP_SIZE)))
1887 		return -EINVAL;
1888 
1889 #ifdef CONFIG_CHECKPOINT_RESTORE
1890 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1891 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1892 #endif
1893 
1894 	if (!capable(CAP_SYS_RESOURCE))
1895 		return -EPERM;
1896 
1897 	if (opt == PR_SET_MM_EXE_FILE) {
1898 		down_write(&mm->mmap_sem);
1899 		error = prctl_set_mm_exe_file_locked(mm, (unsigned int)addr);
1900 		up_write(&mm->mmap_sem);
1901 		return error;
1902 	}
1903 
1904 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
1905 		return -EINVAL;
1906 
1907 	error = -EINVAL;
1908 
1909 	down_read(&mm->mmap_sem);
1910 	vma = find_vma(mm, addr);
1911 
1912 	switch (opt) {
1913 	case PR_SET_MM_START_CODE:
1914 		mm->start_code = addr;
1915 		break;
1916 	case PR_SET_MM_END_CODE:
1917 		mm->end_code = addr;
1918 		break;
1919 	case PR_SET_MM_START_DATA:
1920 		mm->start_data = addr;
1921 		break;
1922 	case PR_SET_MM_END_DATA:
1923 		mm->end_data = addr;
1924 		break;
1925 
1926 	case PR_SET_MM_START_BRK:
1927 		if (addr <= mm->end_data)
1928 			goto out;
1929 
1930 		if (check_data_rlimit(rlimit(RLIMIT_DATA), mm->brk, addr,
1931 				      mm->end_data, mm->start_data))
1932 			goto out;
1933 
1934 		mm->start_brk = addr;
1935 		break;
1936 
1937 	case PR_SET_MM_BRK:
1938 		if (addr <= mm->end_data)
1939 			goto out;
1940 
1941 		if (check_data_rlimit(rlimit(RLIMIT_DATA), addr, mm->start_brk,
1942 				      mm->end_data, mm->start_data))
1943 			goto out;
1944 
1945 		mm->brk = addr;
1946 		break;
1947 
1948 	/*
1949 	 * If command line arguments and environment
1950 	 * are placed somewhere else on stack, we can
1951 	 * set them up here, ARG_START/END to setup
1952 	 * command line argumets and ENV_START/END
1953 	 * for environment.
1954 	 */
1955 	case PR_SET_MM_START_STACK:
1956 	case PR_SET_MM_ARG_START:
1957 	case PR_SET_MM_ARG_END:
1958 	case PR_SET_MM_ENV_START:
1959 	case PR_SET_MM_ENV_END:
1960 		if (!vma) {
1961 			error = -EFAULT;
1962 			goto out;
1963 		}
1964 		if (opt == PR_SET_MM_START_STACK)
1965 			mm->start_stack = addr;
1966 		else if (opt == PR_SET_MM_ARG_START)
1967 			mm->arg_start = addr;
1968 		else if (opt == PR_SET_MM_ARG_END)
1969 			mm->arg_end = addr;
1970 		else if (opt == PR_SET_MM_ENV_START)
1971 			mm->env_start = addr;
1972 		else if (opt == PR_SET_MM_ENV_END)
1973 			mm->env_end = addr;
1974 		break;
1975 
1976 	/*
1977 	 * This doesn't move auxiliary vector itself
1978 	 * since it's pinned to mm_struct, but allow
1979 	 * to fill vector with new values. It's up
1980 	 * to a caller to provide sane values here
1981 	 * otherwise user space tools which use this
1982 	 * vector might be unhappy.
1983 	 */
1984 	case PR_SET_MM_AUXV: {
1985 		unsigned long user_auxv[AT_VECTOR_SIZE];
1986 
1987 		if (arg4 > sizeof(user_auxv))
1988 			goto out;
1989 		up_read(&mm->mmap_sem);
1990 
1991 		if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
1992 			return -EFAULT;
1993 
1994 		/* Make sure the last entry is always AT_NULL */
1995 		user_auxv[AT_VECTOR_SIZE - 2] = 0;
1996 		user_auxv[AT_VECTOR_SIZE - 1] = 0;
1997 
1998 		BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1999 
2000 		task_lock(current);
2001 		memcpy(mm->saved_auxv, user_auxv, arg4);
2002 		task_unlock(current);
2003 
2004 		return 0;
2005 	}
2006 	default:
2007 		goto out;
2008 	}
2009 
2010 	error = 0;
2011 out:
2012 	up_read(&mm->mmap_sem);
2013 	return error;
2014 }
2015 
2016 #ifdef CONFIG_CHECKPOINT_RESTORE
2017 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2018 {
2019 	return put_user(me->clear_child_tid, tid_addr);
2020 }
2021 #else
2022 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2023 {
2024 	return -EINVAL;
2025 }
2026 #endif
2027 
2028 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2029 		unsigned long, arg4, unsigned long, arg5)
2030 {
2031 	struct task_struct *me = current;
2032 	unsigned char comm[sizeof(me->comm)];
2033 	long error;
2034 
2035 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2036 	if (error != -ENOSYS)
2037 		return error;
2038 
2039 	error = 0;
2040 	switch (option) {
2041 	case PR_SET_PDEATHSIG:
2042 		if (!valid_signal(arg2)) {
2043 			error = -EINVAL;
2044 			break;
2045 		}
2046 		me->pdeath_signal = arg2;
2047 		break;
2048 	case PR_GET_PDEATHSIG:
2049 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2050 		break;
2051 	case PR_GET_DUMPABLE:
2052 		error = get_dumpable(me->mm);
2053 		break;
2054 	case PR_SET_DUMPABLE:
2055 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2056 			error = -EINVAL;
2057 			break;
2058 		}
2059 		set_dumpable(me->mm, arg2);
2060 		break;
2061 
2062 	case PR_SET_UNALIGN:
2063 		error = SET_UNALIGN_CTL(me, arg2);
2064 		break;
2065 	case PR_GET_UNALIGN:
2066 		error = GET_UNALIGN_CTL(me, arg2);
2067 		break;
2068 	case PR_SET_FPEMU:
2069 		error = SET_FPEMU_CTL(me, arg2);
2070 		break;
2071 	case PR_GET_FPEMU:
2072 		error = GET_FPEMU_CTL(me, arg2);
2073 		break;
2074 	case PR_SET_FPEXC:
2075 		error = SET_FPEXC_CTL(me, arg2);
2076 		break;
2077 	case PR_GET_FPEXC:
2078 		error = GET_FPEXC_CTL(me, arg2);
2079 		break;
2080 	case PR_GET_TIMING:
2081 		error = PR_TIMING_STATISTICAL;
2082 		break;
2083 	case PR_SET_TIMING:
2084 		if (arg2 != PR_TIMING_STATISTICAL)
2085 			error = -EINVAL;
2086 		break;
2087 	case PR_SET_NAME:
2088 		comm[sizeof(me->comm) - 1] = 0;
2089 		if (strncpy_from_user(comm, (char __user *)arg2,
2090 				      sizeof(me->comm) - 1) < 0)
2091 			return -EFAULT;
2092 		set_task_comm(me, comm);
2093 		proc_comm_connector(me);
2094 		break;
2095 	case PR_GET_NAME:
2096 		get_task_comm(comm, me);
2097 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2098 			return -EFAULT;
2099 		break;
2100 	case PR_GET_ENDIAN:
2101 		error = GET_ENDIAN(me, arg2);
2102 		break;
2103 	case PR_SET_ENDIAN:
2104 		error = SET_ENDIAN(me, arg2);
2105 		break;
2106 	case PR_GET_SECCOMP:
2107 		error = prctl_get_seccomp();
2108 		break;
2109 	case PR_SET_SECCOMP:
2110 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2111 		break;
2112 	case PR_GET_TSC:
2113 		error = GET_TSC_CTL(arg2);
2114 		break;
2115 	case PR_SET_TSC:
2116 		error = SET_TSC_CTL(arg2);
2117 		break;
2118 	case PR_TASK_PERF_EVENTS_DISABLE:
2119 		error = perf_event_task_disable();
2120 		break;
2121 	case PR_TASK_PERF_EVENTS_ENABLE:
2122 		error = perf_event_task_enable();
2123 		break;
2124 	case PR_GET_TIMERSLACK:
2125 		error = current->timer_slack_ns;
2126 		break;
2127 	case PR_SET_TIMERSLACK:
2128 		if (arg2 <= 0)
2129 			current->timer_slack_ns =
2130 					current->default_timer_slack_ns;
2131 		else
2132 			current->timer_slack_ns = arg2;
2133 		break;
2134 	case PR_MCE_KILL:
2135 		if (arg4 | arg5)
2136 			return -EINVAL;
2137 		switch (arg2) {
2138 		case PR_MCE_KILL_CLEAR:
2139 			if (arg3 != 0)
2140 				return -EINVAL;
2141 			current->flags &= ~PF_MCE_PROCESS;
2142 			break;
2143 		case PR_MCE_KILL_SET:
2144 			current->flags |= PF_MCE_PROCESS;
2145 			if (arg3 == PR_MCE_KILL_EARLY)
2146 				current->flags |= PF_MCE_EARLY;
2147 			else if (arg3 == PR_MCE_KILL_LATE)
2148 				current->flags &= ~PF_MCE_EARLY;
2149 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2150 				current->flags &=
2151 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2152 			else
2153 				return -EINVAL;
2154 			break;
2155 		default:
2156 			return -EINVAL;
2157 		}
2158 		break;
2159 	case PR_MCE_KILL_GET:
2160 		if (arg2 | arg3 | arg4 | arg5)
2161 			return -EINVAL;
2162 		if (current->flags & PF_MCE_PROCESS)
2163 			error = (current->flags & PF_MCE_EARLY) ?
2164 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2165 		else
2166 			error = PR_MCE_KILL_DEFAULT;
2167 		break;
2168 	case PR_SET_MM:
2169 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2170 		break;
2171 	case PR_GET_TID_ADDRESS:
2172 		error = prctl_get_tid_address(me, (int __user **)arg2);
2173 		break;
2174 	case PR_SET_CHILD_SUBREAPER:
2175 		me->signal->is_child_subreaper = !!arg2;
2176 		break;
2177 	case PR_GET_CHILD_SUBREAPER:
2178 		error = put_user(me->signal->is_child_subreaper,
2179 				 (int __user *)arg2);
2180 		break;
2181 	case PR_SET_NO_NEW_PRIVS:
2182 		if (arg2 != 1 || arg3 || arg4 || arg5)
2183 			return -EINVAL;
2184 
2185 		task_set_no_new_privs(current);
2186 		break;
2187 	case PR_GET_NO_NEW_PRIVS:
2188 		if (arg2 || arg3 || arg4 || arg5)
2189 			return -EINVAL;
2190 		return task_no_new_privs(current) ? 1 : 0;
2191 	case PR_GET_THP_DISABLE:
2192 		if (arg2 || arg3 || arg4 || arg5)
2193 			return -EINVAL;
2194 		error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2195 		break;
2196 	case PR_SET_THP_DISABLE:
2197 		if (arg3 || arg4 || arg5)
2198 			return -EINVAL;
2199 		down_write(&me->mm->mmap_sem);
2200 		if (arg2)
2201 			me->mm->def_flags |= VM_NOHUGEPAGE;
2202 		else
2203 			me->mm->def_flags &= ~VM_NOHUGEPAGE;
2204 		up_write(&me->mm->mmap_sem);
2205 		break;
2206 	default:
2207 		error = -EINVAL;
2208 		break;
2209 	}
2210 	return error;
2211 }
2212 
2213 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2214 		struct getcpu_cache __user *, unused)
2215 {
2216 	int err = 0;
2217 	int cpu = raw_smp_processor_id();
2218 
2219 	if (cpup)
2220 		err |= put_user(cpu, cpup);
2221 	if (nodep)
2222 		err |= put_user(cpu_to_node(cpu), nodep);
2223 	return err ? -EFAULT : 0;
2224 }
2225 
2226 /**
2227  * do_sysinfo - fill in sysinfo struct
2228  * @info: pointer to buffer to fill
2229  */
2230 static int do_sysinfo(struct sysinfo *info)
2231 {
2232 	unsigned long mem_total, sav_total;
2233 	unsigned int mem_unit, bitcount;
2234 	struct timespec tp;
2235 
2236 	memset(info, 0, sizeof(struct sysinfo));
2237 
2238 	get_monotonic_boottime(&tp);
2239 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2240 
2241 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2242 
2243 	info->procs = nr_threads;
2244 
2245 	si_meminfo(info);
2246 	si_swapinfo(info);
2247 
2248 	/*
2249 	 * If the sum of all the available memory (i.e. ram + swap)
2250 	 * is less than can be stored in a 32 bit unsigned long then
2251 	 * we can be binary compatible with 2.2.x kernels.  If not,
2252 	 * well, in that case 2.2.x was broken anyways...
2253 	 *
2254 	 *  -Erik Andersen <andersee@debian.org>
2255 	 */
2256 
2257 	mem_total = info->totalram + info->totalswap;
2258 	if (mem_total < info->totalram || mem_total < info->totalswap)
2259 		goto out;
2260 	bitcount = 0;
2261 	mem_unit = info->mem_unit;
2262 	while (mem_unit > 1) {
2263 		bitcount++;
2264 		mem_unit >>= 1;
2265 		sav_total = mem_total;
2266 		mem_total <<= 1;
2267 		if (mem_total < sav_total)
2268 			goto out;
2269 	}
2270 
2271 	/*
2272 	 * If mem_total did not overflow, multiply all memory values by
2273 	 * info->mem_unit and set it to 1.  This leaves things compatible
2274 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2275 	 * kernels...
2276 	 */
2277 
2278 	info->mem_unit = 1;
2279 	info->totalram <<= bitcount;
2280 	info->freeram <<= bitcount;
2281 	info->sharedram <<= bitcount;
2282 	info->bufferram <<= bitcount;
2283 	info->totalswap <<= bitcount;
2284 	info->freeswap <<= bitcount;
2285 	info->totalhigh <<= bitcount;
2286 	info->freehigh <<= bitcount;
2287 
2288 out:
2289 	return 0;
2290 }
2291 
2292 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2293 {
2294 	struct sysinfo val;
2295 
2296 	do_sysinfo(&val);
2297 
2298 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2299 		return -EFAULT;
2300 
2301 	return 0;
2302 }
2303 
2304 #ifdef CONFIG_COMPAT
2305 struct compat_sysinfo {
2306 	s32 uptime;
2307 	u32 loads[3];
2308 	u32 totalram;
2309 	u32 freeram;
2310 	u32 sharedram;
2311 	u32 bufferram;
2312 	u32 totalswap;
2313 	u32 freeswap;
2314 	u16 procs;
2315 	u16 pad;
2316 	u32 totalhigh;
2317 	u32 freehigh;
2318 	u32 mem_unit;
2319 	char _f[20-2*sizeof(u32)-sizeof(int)];
2320 };
2321 
2322 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2323 {
2324 	struct sysinfo s;
2325 
2326 	do_sysinfo(&s);
2327 
2328 	/* Check to see if any memory value is too large for 32-bit and scale
2329 	 *  down if needed
2330 	 */
2331 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2332 		int bitcount = 0;
2333 
2334 		while (s.mem_unit < PAGE_SIZE) {
2335 			s.mem_unit <<= 1;
2336 			bitcount++;
2337 		}
2338 
2339 		s.totalram >>= bitcount;
2340 		s.freeram >>= bitcount;
2341 		s.sharedram >>= bitcount;
2342 		s.bufferram >>= bitcount;
2343 		s.totalswap >>= bitcount;
2344 		s.freeswap >>= bitcount;
2345 		s.totalhigh >>= bitcount;
2346 		s.freehigh >>= bitcount;
2347 	}
2348 
2349 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2350 	    __put_user(s.uptime, &info->uptime) ||
2351 	    __put_user(s.loads[0], &info->loads[0]) ||
2352 	    __put_user(s.loads[1], &info->loads[1]) ||
2353 	    __put_user(s.loads[2], &info->loads[2]) ||
2354 	    __put_user(s.totalram, &info->totalram) ||
2355 	    __put_user(s.freeram, &info->freeram) ||
2356 	    __put_user(s.sharedram, &info->sharedram) ||
2357 	    __put_user(s.bufferram, &info->bufferram) ||
2358 	    __put_user(s.totalswap, &info->totalswap) ||
2359 	    __put_user(s.freeswap, &info->freeswap) ||
2360 	    __put_user(s.procs, &info->procs) ||
2361 	    __put_user(s.totalhigh, &info->totalhigh) ||
2362 	    __put_user(s.freehigh, &info->freehigh) ||
2363 	    __put_user(s.mem_unit, &info->mem_unit))
2364 		return -EFAULT;
2365 
2366 	return 0;
2367 }
2368 #endif /* CONFIG_COMPAT */
2369