xref: /linux/kernel/sys.c (revision 413d6ed3eac387a2876893c337174f0c5b99d01d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/kernel/sys.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/export.h>
9 #include <linux/mm.h>
10 #include <linux/utsname.h>
11 #include <linux/mman.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
15 #include <linux/fs.h>
16 #include <linux/kmod.h>
17 #include <linux/perf_event.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/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 #include <linux/syscall_user_dispatch.h>
45 
46 #include <linux/compat.h>
47 #include <linux/syscalls.h>
48 #include <linux/kprobes.h>
49 #include <linux/user_namespace.h>
50 #include <linux/time_namespace.h>
51 #include <linux/binfmts.h>
52 
53 #include <linux/sched.h>
54 #include <linux/sched/autogroup.h>
55 #include <linux/sched/loadavg.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/mm.h>
58 #include <linux/sched/coredump.h>
59 #include <linux/sched/task.h>
60 #include <linux/sched/cputime.h>
61 #include <linux/rcupdate.h>
62 #include <linux/uidgid.h>
63 #include <linux/cred.h>
64 
65 #include <linux/nospec.h>
66 
67 #include <linux/kmsg_dump.h>
68 /* Move somewhere else to avoid recompiling? */
69 #include <generated/utsrelease.h>
70 
71 #include <linux/uaccess.h>
72 #include <asm/io.h>
73 #include <asm/unistd.h>
74 
75 #include "uid16.h"
76 
77 #ifndef SET_UNALIGN_CTL
78 # define SET_UNALIGN_CTL(a, b)	(-EINVAL)
79 #endif
80 #ifndef GET_UNALIGN_CTL
81 # define GET_UNALIGN_CTL(a, b)	(-EINVAL)
82 #endif
83 #ifndef SET_FPEMU_CTL
84 # define SET_FPEMU_CTL(a, b)	(-EINVAL)
85 #endif
86 #ifndef GET_FPEMU_CTL
87 # define GET_FPEMU_CTL(a, b)	(-EINVAL)
88 #endif
89 #ifndef SET_FPEXC_CTL
90 # define SET_FPEXC_CTL(a, b)	(-EINVAL)
91 #endif
92 #ifndef GET_FPEXC_CTL
93 # define GET_FPEXC_CTL(a, b)	(-EINVAL)
94 #endif
95 #ifndef GET_ENDIAN
96 # define GET_ENDIAN(a, b)	(-EINVAL)
97 #endif
98 #ifndef SET_ENDIAN
99 # define SET_ENDIAN(a, b)	(-EINVAL)
100 #endif
101 #ifndef GET_TSC_CTL
102 # define GET_TSC_CTL(a)		(-EINVAL)
103 #endif
104 #ifndef SET_TSC_CTL
105 # define SET_TSC_CTL(a)		(-EINVAL)
106 #endif
107 #ifndef GET_FP_MODE
108 # define GET_FP_MODE(a)		(-EINVAL)
109 #endif
110 #ifndef SET_FP_MODE
111 # define SET_FP_MODE(a,b)	(-EINVAL)
112 #endif
113 #ifndef SVE_SET_VL
114 # define SVE_SET_VL(a)		(-EINVAL)
115 #endif
116 #ifndef SVE_GET_VL
117 # define SVE_GET_VL()		(-EINVAL)
118 #endif
119 #ifndef PAC_RESET_KEYS
120 # define PAC_RESET_KEYS(a, b)	(-EINVAL)
121 #endif
122 #ifndef PAC_SET_ENABLED_KEYS
123 # define PAC_SET_ENABLED_KEYS(a, b, c)	(-EINVAL)
124 #endif
125 #ifndef PAC_GET_ENABLED_KEYS
126 # define PAC_GET_ENABLED_KEYS(a)	(-EINVAL)
127 #endif
128 #ifndef SET_TAGGED_ADDR_CTRL
129 # define SET_TAGGED_ADDR_CTRL(a)	(-EINVAL)
130 #endif
131 #ifndef GET_TAGGED_ADDR_CTRL
132 # define GET_TAGGED_ADDR_CTRL()		(-EINVAL)
133 #endif
134 
135 /*
136  * this is where the system-wide overflow UID and GID are defined, for
137  * architectures that now have 32-bit UID/GID but didn't in the past
138  */
139 
140 int overflowuid = DEFAULT_OVERFLOWUID;
141 int overflowgid = DEFAULT_OVERFLOWGID;
142 
143 EXPORT_SYMBOL(overflowuid);
144 EXPORT_SYMBOL(overflowgid);
145 
146 /*
147  * the same as above, but for filesystems which can only store a 16-bit
148  * UID and GID. as such, this is needed on all architectures
149  */
150 
151 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
152 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
153 
154 EXPORT_SYMBOL(fs_overflowuid);
155 EXPORT_SYMBOL(fs_overflowgid);
156 
157 /*
158  * Returns true if current's euid is same as p's uid or euid,
159  * or has CAP_SYS_NICE to p's user_ns.
160  *
161  * Called with rcu_read_lock, creds are safe
162  */
163 static bool set_one_prio_perm(struct task_struct *p)
164 {
165 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
166 
167 	if (uid_eq(pcred->uid,  cred->euid) ||
168 	    uid_eq(pcred->euid, cred->euid))
169 		return true;
170 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
171 		return true;
172 	return false;
173 }
174 
175 /*
176  * set the priority of a task
177  * - the caller must hold the RCU read lock
178  */
179 static int set_one_prio(struct task_struct *p, int niceval, int error)
180 {
181 	int no_nice;
182 
183 	if (!set_one_prio_perm(p)) {
184 		error = -EPERM;
185 		goto out;
186 	}
187 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
188 		error = -EACCES;
189 		goto out;
190 	}
191 	no_nice = security_task_setnice(p, niceval);
192 	if (no_nice) {
193 		error = no_nice;
194 		goto out;
195 	}
196 	if (error == -ESRCH)
197 		error = 0;
198 	set_user_nice(p, niceval);
199 out:
200 	return error;
201 }
202 
203 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
204 {
205 	struct task_struct *g, *p;
206 	struct user_struct *user;
207 	const struct cred *cred = current_cred();
208 	int error = -EINVAL;
209 	struct pid *pgrp;
210 	kuid_t uid;
211 
212 	if (which > PRIO_USER || which < PRIO_PROCESS)
213 		goto out;
214 
215 	/* normalize: avoid signed division (rounding problems) */
216 	error = -ESRCH;
217 	if (niceval < MIN_NICE)
218 		niceval = MIN_NICE;
219 	if (niceval > MAX_NICE)
220 		niceval = MAX_NICE;
221 
222 	rcu_read_lock();
223 	read_lock(&tasklist_lock);
224 	switch (which) {
225 	case PRIO_PROCESS:
226 		if (who)
227 			p = find_task_by_vpid(who);
228 		else
229 			p = current;
230 		if (p)
231 			error = set_one_prio(p, niceval, error);
232 		break;
233 	case PRIO_PGRP:
234 		if (who)
235 			pgrp = find_vpid(who);
236 		else
237 			pgrp = task_pgrp(current);
238 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
239 			error = set_one_prio(p, niceval, error);
240 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
241 		break;
242 	case PRIO_USER:
243 		uid = make_kuid(cred->user_ns, who);
244 		user = cred->user;
245 		if (!who)
246 			uid = cred->uid;
247 		else if (!uid_eq(uid, cred->uid)) {
248 			user = find_user(uid);
249 			if (!user)
250 				goto out_unlock;	/* No processes for this user */
251 		}
252 		do_each_thread(g, p) {
253 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
254 				error = set_one_prio(p, niceval, error);
255 		} while_each_thread(g, p);
256 		if (!uid_eq(uid, cred->uid))
257 			free_uid(user);		/* For find_user() */
258 		break;
259 	}
260 out_unlock:
261 	read_unlock(&tasklist_lock);
262 	rcu_read_unlock();
263 out:
264 	return error;
265 }
266 
267 /*
268  * Ugh. To avoid negative return values, "getpriority()" will
269  * not return the normal nice-value, but a negated value that
270  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
271  * to stay compatible.
272  */
273 SYSCALL_DEFINE2(getpriority, int, which, int, who)
274 {
275 	struct task_struct *g, *p;
276 	struct user_struct *user;
277 	const struct cred *cred = current_cred();
278 	long niceval, retval = -ESRCH;
279 	struct pid *pgrp;
280 	kuid_t uid;
281 
282 	if (which > PRIO_USER || which < PRIO_PROCESS)
283 		return -EINVAL;
284 
285 	rcu_read_lock();
286 	read_lock(&tasklist_lock);
287 	switch (which) {
288 	case PRIO_PROCESS:
289 		if (who)
290 			p = find_task_by_vpid(who);
291 		else
292 			p = current;
293 		if (p) {
294 			niceval = nice_to_rlimit(task_nice(p));
295 			if (niceval > retval)
296 				retval = niceval;
297 		}
298 		break;
299 	case PRIO_PGRP:
300 		if (who)
301 			pgrp = find_vpid(who);
302 		else
303 			pgrp = task_pgrp(current);
304 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
305 			niceval = nice_to_rlimit(task_nice(p));
306 			if (niceval > retval)
307 				retval = niceval;
308 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
309 		break;
310 	case PRIO_USER:
311 		uid = make_kuid(cred->user_ns, who);
312 		user = cred->user;
313 		if (!who)
314 			uid = cred->uid;
315 		else if (!uid_eq(uid, cred->uid)) {
316 			user = find_user(uid);
317 			if (!user)
318 				goto out_unlock;	/* No processes for this user */
319 		}
320 		do_each_thread(g, p) {
321 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
322 				niceval = nice_to_rlimit(task_nice(p));
323 				if (niceval > retval)
324 					retval = niceval;
325 			}
326 		} while_each_thread(g, p);
327 		if (!uid_eq(uid, cred->uid))
328 			free_uid(user);		/* for find_user() */
329 		break;
330 	}
331 out_unlock:
332 	read_unlock(&tasklist_lock);
333 	rcu_read_unlock();
334 
335 	return retval;
336 }
337 
338 /*
339  * Unprivileged users may change the real gid to the effective gid
340  * or vice versa.  (BSD-style)
341  *
342  * If you set the real gid at all, or set the effective gid to a value not
343  * equal to the real gid, then the saved gid is set to the new effective gid.
344  *
345  * This makes it possible for a setgid program to completely drop its
346  * privileges, which is often a useful assertion to make when you are doing
347  * a security audit over a program.
348  *
349  * The general idea is that a program which uses just setregid() will be
350  * 100% compatible with BSD.  A program which uses just setgid() will be
351  * 100% compatible with POSIX with saved IDs.
352  *
353  * SMP: There are not races, the GIDs are checked only by filesystem
354  *      operations (as far as semantic preservation is concerned).
355  */
356 #ifdef CONFIG_MULTIUSER
357 long __sys_setregid(gid_t rgid, gid_t egid)
358 {
359 	struct user_namespace *ns = current_user_ns();
360 	const struct cred *old;
361 	struct cred *new;
362 	int retval;
363 	kgid_t krgid, kegid;
364 
365 	krgid = make_kgid(ns, rgid);
366 	kegid = make_kgid(ns, egid);
367 
368 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
369 		return -EINVAL;
370 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
371 		return -EINVAL;
372 
373 	new = prepare_creds();
374 	if (!new)
375 		return -ENOMEM;
376 	old = current_cred();
377 
378 	retval = -EPERM;
379 	if (rgid != (gid_t) -1) {
380 		if (gid_eq(old->gid, krgid) ||
381 		    gid_eq(old->egid, krgid) ||
382 		    ns_capable_setid(old->user_ns, CAP_SETGID))
383 			new->gid = krgid;
384 		else
385 			goto error;
386 	}
387 	if (egid != (gid_t) -1) {
388 		if (gid_eq(old->gid, kegid) ||
389 		    gid_eq(old->egid, kegid) ||
390 		    gid_eq(old->sgid, kegid) ||
391 		    ns_capable_setid(old->user_ns, CAP_SETGID))
392 			new->egid = kegid;
393 		else
394 			goto error;
395 	}
396 
397 	if (rgid != (gid_t) -1 ||
398 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
399 		new->sgid = new->egid;
400 	new->fsgid = new->egid;
401 
402 	retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
403 	if (retval < 0)
404 		goto error;
405 
406 	return commit_creds(new);
407 
408 error:
409 	abort_creds(new);
410 	return retval;
411 }
412 
413 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
414 {
415 	return __sys_setregid(rgid, egid);
416 }
417 
418 /*
419  * setgid() is implemented like SysV w/ SAVED_IDS
420  *
421  * SMP: Same implicit races as above.
422  */
423 long __sys_setgid(gid_t gid)
424 {
425 	struct user_namespace *ns = current_user_ns();
426 	const struct cred *old;
427 	struct cred *new;
428 	int retval;
429 	kgid_t kgid;
430 
431 	kgid = make_kgid(ns, gid);
432 	if (!gid_valid(kgid))
433 		return -EINVAL;
434 
435 	new = prepare_creds();
436 	if (!new)
437 		return -ENOMEM;
438 	old = current_cred();
439 
440 	retval = -EPERM;
441 	if (ns_capable_setid(old->user_ns, CAP_SETGID))
442 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
443 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
444 		new->egid = new->fsgid = kgid;
445 	else
446 		goto error;
447 
448 	retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
449 	if (retval < 0)
450 		goto error;
451 
452 	return commit_creds(new);
453 
454 error:
455 	abort_creds(new);
456 	return retval;
457 }
458 
459 SYSCALL_DEFINE1(setgid, gid_t, gid)
460 {
461 	return __sys_setgid(gid);
462 }
463 
464 /*
465  * change the user struct in a credentials set to match the new UID
466  */
467 static int set_user(struct cred *new)
468 {
469 	struct user_struct *new_user;
470 
471 	new_user = alloc_uid(new->uid);
472 	if (!new_user)
473 		return -EAGAIN;
474 
475 	/*
476 	 * We don't fail in case of NPROC limit excess here because too many
477 	 * poorly written programs don't check set*uid() return code, assuming
478 	 * it never fails if called by root.  We may still enforce NPROC limit
479 	 * for programs doing set*uid()+execve() by harmlessly deferring the
480 	 * failure to the execve() stage.
481 	 */
482 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
483 			new_user != INIT_USER)
484 		current->flags |= PF_NPROC_EXCEEDED;
485 	else
486 		current->flags &= ~PF_NPROC_EXCEEDED;
487 
488 	free_uid(new->user);
489 	new->user = new_user;
490 	return 0;
491 }
492 
493 /*
494  * Unprivileged users may change the real uid to the effective uid
495  * or vice versa.  (BSD-style)
496  *
497  * If you set the real uid at all, or set the effective uid to a value not
498  * equal to the real uid, then the saved uid is set to the new effective uid.
499  *
500  * This makes it possible for a setuid program to completely drop its
501  * privileges, which is often a useful assertion to make when you are doing
502  * a security audit over a program.
503  *
504  * The general idea is that a program which uses just setreuid() will be
505  * 100% compatible with BSD.  A program which uses just setuid() will be
506  * 100% compatible with POSIX with saved IDs.
507  */
508 long __sys_setreuid(uid_t ruid, uid_t euid)
509 {
510 	struct user_namespace *ns = current_user_ns();
511 	const struct cred *old;
512 	struct cred *new;
513 	int retval;
514 	kuid_t kruid, keuid;
515 
516 	kruid = make_kuid(ns, ruid);
517 	keuid = make_kuid(ns, euid);
518 
519 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
520 		return -EINVAL;
521 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
522 		return -EINVAL;
523 
524 	new = prepare_creds();
525 	if (!new)
526 		return -ENOMEM;
527 	old = current_cred();
528 
529 	retval = -EPERM;
530 	if (ruid != (uid_t) -1) {
531 		new->uid = kruid;
532 		if (!uid_eq(old->uid, kruid) &&
533 		    !uid_eq(old->euid, kruid) &&
534 		    !ns_capable_setid(old->user_ns, CAP_SETUID))
535 			goto error;
536 	}
537 
538 	if (euid != (uid_t) -1) {
539 		new->euid = keuid;
540 		if (!uid_eq(old->uid, keuid) &&
541 		    !uid_eq(old->euid, keuid) &&
542 		    !uid_eq(old->suid, keuid) &&
543 		    !ns_capable_setid(old->user_ns, CAP_SETUID))
544 			goto error;
545 	}
546 
547 	if (!uid_eq(new->uid, old->uid)) {
548 		retval = set_user(new);
549 		if (retval < 0)
550 			goto error;
551 	}
552 	if (ruid != (uid_t) -1 ||
553 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
554 		new->suid = new->euid;
555 	new->fsuid = new->euid;
556 
557 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
558 	if (retval < 0)
559 		goto error;
560 
561 	return commit_creds(new);
562 
563 error:
564 	abort_creds(new);
565 	return retval;
566 }
567 
568 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
569 {
570 	return __sys_setreuid(ruid, euid);
571 }
572 
573 /*
574  * setuid() is implemented like SysV with SAVED_IDS
575  *
576  * Note that SAVED_ID's is deficient in that a setuid root program
577  * like sendmail, for example, cannot set its uid to be a normal
578  * user and then switch back, because if you're root, setuid() sets
579  * the saved uid too.  If you don't like this, blame the bright people
580  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
581  * will allow a root program to temporarily drop privileges and be able to
582  * regain them by swapping the real and effective uid.
583  */
584 long __sys_setuid(uid_t uid)
585 {
586 	struct user_namespace *ns = current_user_ns();
587 	const struct cred *old;
588 	struct cred *new;
589 	int retval;
590 	kuid_t kuid;
591 
592 	kuid = make_kuid(ns, uid);
593 	if (!uid_valid(kuid))
594 		return -EINVAL;
595 
596 	new = prepare_creds();
597 	if (!new)
598 		return -ENOMEM;
599 	old = current_cred();
600 
601 	retval = -EPERM;
602 	if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
603 		new->suid = new->uid = kuid;
604 		if (!uid_eq(kuid, old->uid)) {
605 			retval = set_user(new);
606 			if (retval < 0)
607 				goto error;
608 		}
609 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
610 		goto error;
611 	}
612 
613 	new->fsuid = new->euid = kuid;
614 
615 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
616 	if (retval < 0)
617 		goto error;
618 
619 	return commit_creds(new);
620 
621 error:
622 	abort_creds(new);
623 	return retval;
624 }
625 
626 SYSCALL_DEFINE1(setuid, uid_t, uid)
627 {
628 	return __sys_setuid(uid);
629 }
630 
631 
632 /*
633  * This function implements a generic ability to update ruid, euid,
634  * and suid.  This allows you to implement the 4.4 compatible seteuid().
635  */
636 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
637 {
638 	struct user_namespace *ns = current_user_ns();
639 	const struct cred *old;
640 	struct cred *new;
641 	int retval;
642 	kuid_t kruid, keuid, ksuid;
643 
644 	kruid = make_kuid(ns, ruid);
645 	keuid = make_kuid(ns, euid);
646 	ksuid = make_kuid(ns, suid);
647 
648 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
649 		return -EINVAL;
650 
651 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
652 		return -EINVAL;
653 
654 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
655 		return -EINVAL;
656 
657 	new = prepare_creds();
658 	if (!new)
659 		return -ENOMEM;
660 
661 	old = current_cred();
662 
663 	retval = -EPERM;
664 	if (!ns_capable_setid(old->user_ns, CAP_SETUID)) {
665 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
666 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
667 			goto error;
668 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
669 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
670 			goto error;
671 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
672 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
673 			goto error;
674 	}
675 
676 	if (ruid != (uid_t) -1) {
677 		new->uid = kruid;
678 		if (!uid_eq(kruid, old->uid)) {
679 			retval = set_user(new);
680 			if (retval < 0)
681 				goto error;
682 		}
683 	}
684 	if (euid != (uid_t) -1)
685 		new->euid = keuid;
686 	if (suid != (uid_t) -1)
687 		new->suid = ksuid;
688 	new->fsuid = new->euid;
689 
690 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
691 	if (retval < 0)
692 		goto error;
693 
694 	return commit_creds(new);
695 
696 error:
697 	abort_creds(new);
698 	return retval;
699 }
700 
701 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
702 {
703 	return __sys_setresuid(ruid, euid, suid);
704 }
705 
706 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
707 {
708 	const struct cred *cred = current_cred();
709 	int retval;
710 	uid_t ruid, euid, suid;
711 
712 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
713 	euid = from_kuid_munged(cred->user_ns, cred->euid);
714 	suid = from_kuid_munged(cred->user_ns, cred->suid);
715 
716 	retval = put_user(ruid, ruidp);
717 	if (!retval) {
718 		retval = put_user(euid, euidp);
719 		if (!retval)
720 			return put_user(suid, suidp);
721 	}
722 	return retval;
723 }
724 
725 /*
726  * Same as above, but for rgid, egid, sgid.
727  */
728 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
729 {
730 	struct user_namespace *ns = current_user_ns();
731 	const struct cred *old;
732 	struct cred *new;
733 	int retval;
734 	kgid_t krgid, kegid, ksgid;
735 
736 	krgid = make_kgid(ns, rgid);
737 	kegid = make_kgid(ns, egid);
738 	ksgid = make_kgid(ns, sgid);
739 
740 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
741 		return -EINVAL;
742 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
743 		return -EINVAL;
744 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
745 		return -EINVAL;
746 
747 	new = prepare_creds();
748 	if (!new)
749 		return -ENOMEM;
750 	old = current_cred();
751 
752 	retval = -EPERM;
753 	if (!ns_capable_setid(old->user_ns, CAP_SETGID)) {
754 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
755 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
756 			goto error;
757 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
758 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
759 			goto error;
760 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
761 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
762 			goto error;
763 	}
764 
765 	if (rgid != (gid_t) -1)
766 		new->gid = krgid;
767 	if (egid != (gid_t) -1)
768 		new->egid = kegid;
769 	if (sgid != (gid_t) -1)
770 		new->sgid = ksgid;
771 	new->fsgid = new->egid;
772 
773 	retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
774 	if (retval < 0)
775 		goto error;
776 
777 	return commit_creds(new);
778 
779 error:
780 	abort_creds(new);
781 	return retval;
782 }
783 
784 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
785 {
786 	return __sys_setresgid(rgid, egid, sgid);
787 }
788 
789 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
790 {
791 	const struct cred *cred = current_cred();
792 	int retval;
793 	gid_t rgid, egid, sgid;
794 
795 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
796 	egid = from_kgid_munged(cred->user_ns, cred->egid);
797 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
798 
799 	retval = put_user(rgid, rgidp);
800 	if (!retval) {
801 		retval = put_user(egid, egidp);
802 		if (!retval)
803 			retval = put_user(sgid, sgidp);
804 	}
805 
806 	return retval;
807 }
808 
809 
810 /*
811  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
812  * is used for "access()" and for the NFS daemon (letting nfsd stay at
813  * whatever uid it wants to). It normally shadows "euid", except when
814  * explicitly set by setfsuid() or for access..
815  */
816 long __sys_setfsuid(uid_t uid)
817 {
818 	const struct cred *old;
819 	struct cred *new;
820 	uid_t old_fsuid;
821 	kuid_t kuid;
822 
823 	old = current_cred();
824 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
825 
826 	kuid = make_kuid(old->user_ns, uid);
827 	if (!uid_valid(kuid))
828 		return old_fsuid;
829 
830 	new = prepare_creds();
831 	if (!new)
832 		return old_fsuid;
833 
834 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
835 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
836 	    ns_capable_setid(old->user_ns, CAP_SETUID)) {
837 		if (!uid_eq(kuid, old->fsuid)) {
838 			new->fsuid = kuid;
839 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
840 				goto change_okay;
841 		}
842 	}
843 
844 	abort_creds(new);
845 	return old_fsuid;
846 
847 change_okay:
848 	commit_creds(new);
849 	return old_fsuid;
850 }
851 
852 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
853 {
854 	return __sys_setfsuid(uid);
855 }
856 
857 /*
858  * Samma på svenska..
859  */
860 long __sys_setfsgid(gid_t gid)
861 {
862 	const struct cred *old;
863 	struct cred *new;
864 	gid_t old_fsgid;
865 	kgid_t kgid;
866 
867 	old = current_cred();
868 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
869 
870 	kgid = make_kgid(old->user_ns, gid);
871 	if (!gid_valid(kgid))
872 		return old_fsgid;
873 
874 	new = prepare_creds();
875 	if (!new)
876 		return old_fsgid;
877 
878 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
879 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
880 	    ns_capable_setid(old->user_ns, CAP_SETGID)) {
881 		if (!gid_eq(kgid, old->fsgid)) {
882 			new->fsgid = kgid;
883 			if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
884 				goto change_okay;
885 		}
886 	}
887 
888 	abort_creds(new);
889 	return old_fsgid;
890 
891 change_okay:
892 	commit_creds(new);
893 	return old_fsgid;
894 }
895 
896 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
897 {
898 	return __sys_setfsgid(gid);
899 }
900 #endif /* CONFIG_MULTIUSER */
901 
902 /**
903  * sys_getpid - return the thread group id of the current process
904  *
905  * Note, despite the name, this returns the tgid not the pid.  The tgid and
906  * the pid are identical unless CLONE_THREAD was specified on clone() in
907  * which case the tgid is the same in all threads of the same group.
908  *
909  * This is SMP safe as current->tgid does not change.
910  */
911 SYSCALL_DEFINE0(getpid)
912 {
913 	return task_tgid_vnr(current);
914 }
915 
916 /* Thread ID - the internal kernel "pid" */
917 SYSCALL_DEFINE0(gettid)
918 {
919 	return task_pid_vnr(current);
920 }
921 
922 /*
923  * Accessing ->real_parent is not SMP-safe, it could
924  * change from under us. However, we can use a stale
925  * value of ->real_parent under rcu_read_lock(), see
926  * release_task()->call_rcu(delayed_put_task_struct).
927  */
928 SYSCALL_DEFINE0(getppid)
929 {
930 	int pid;
931 
932 	rcu_read_lock();
933 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
934 	rcu_read_unlock();
935 
936 	return pid;
937 }
938 
939 SYSCALL_DEFINE0(getuid)
940 {
941 	/* Only we change this so SMP safe */
942 	return from_kuid_munged(current_user_ns(), current_uid());
943 }
944 
945 SYSCALL_DEFINE0(geteuid)
946 {
947 	/* Only we change this so SMP safe */
948 	return from_kuid_munged(current_user_ns(), current_euid());
949 }
950 
951 SYSCALL_DEFINE0(getgid)
952 {
953 	/* Only we change this so SMP safe */
954 	return from_kgid_munged(current_user_ns(), current_gid());
955 }
956 
957 SYSCALL_DEFINE0(getegid)
958 {
959 	/* Only we change this so SMP safe */
960 	return from_kgid_munged(current_user_ns(), current_egid());
961 }
962 
963 static void do_sys_times(struct tms *tms)
964 {
965 	u64 tgutime, tgstime, cutime, cstime;
966 
967 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
968 	cutime = current->signal->cutime;
969 	cstime = current->signal->cstime;
970 	tms->tms_utime = nsec_to_clock_t(tgutime);
971 	tms->tms_stime = nsec_to_clock_t(tgstime);
972 	tms->tms_cutime = nsec_to_clock_t(cutime);
973 	tms->tms_cstime = nsec_to_clock_t(cstime);
974 }
975 
976 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
977 {
978 	if (tbuf) {
979 		struct tms tmp;
980 
981 		do_sys_times(&tmp);
982 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
983 			return -EFAULT;
984 	}
985 	force_successful_syscall_return();
986 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
987 }
988 
989 #ifdef CONFIG_COMPAT
990 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
991 {
992 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
993 }
994 
995 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
996 {
997 	if (tbuf) {
998 		struct tms tms;
999 		struct compat_tms tmp;
1000 
1001 		do_sys_times(&tms);
1002 		/* Convert our struct tms to the compat version. */
1003 		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1004 		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1005 		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1006 		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1007 		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1008 			return -EFAULT;
1009 	}
1010 	force_successful_syscall_return();
1011 	return compat_jiffies_to_clock_t(jiffies);
1012 }
1013 #endif
1014 
1015 /*
1016  * This needs some heavy checking ...
1017  * I just haven't the stomach for it. I also don't fully
1018  * understand sessions/pgrp etc. Let somebody who does explain it.
1019  *
1020  * OK, I think I have the protection semantics right.... this is really
1021  * only important on a multi-user system anyway, to make sure one user
1022  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1023  *
1024  * !PF_FORKNOEXEC check to conform completely to POSIX.
1025  */
1026 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1027 {
1028 	struct task_struct *p;
1029 	struct task_struct *group_leader = current->group_leader;
1030 	struct pid *pgrp;
1031 	int err;
1032 
1033 	if (!pid)
1034 		pid = task_pid_vnr(group_leader);
1035 	if (!pgid)
1036 		pgid = pid;
1037 	if (pgid < 0)
1038 		return -EINVAL;
1039 	rcu_read_lock();
1040 
1041 	/* From this point forward we keep holding onto the tasklist lock
1042 	 * so that our parent does not change from under us. -DaveM
1043 	 */
1044 	write_lock_irq(&tasklist_lock);
1045 
1046 	err = -ESRCH;
1047 	p = find_task_by_vpid(pid);
1048 	if (!p)
1049 		goto out;
1050 
1051 	err = -EINVAL;
1052 	if (!thread_group_leader(p))
1053 		goto out;
1054 
1055 	if (same_thread_group(p->real_parent, group_leader)) {
1056 		err = -EPERM;
1057 		if (task_session(p) != task_session(group_leader))
1058 			goto out;
1059 		err = -EACCES;
1060 		if (!(p->flags & PF_FORKNOEXEC))
1061 			goto out;
1062 	} else {
1063 		err = -ESRCH;
1064 		if (p != group_leader)
1065 			goto out;
1066 	}
1067 
1068 	err = -EPERM;
1069 	if (p->signal->leader)
1070 		goto out;
1071 
1072 	pgrp = task_pid(p);
1073 	if (pgid != pid) {
1074 		struct task_struct *g;
1075 
1076 		pgrp = find_vpid(pgid);
1077 		g = pid_task(pgrp, PIDTYPE_PGID);
1078 		if (!g || task_session(g) != task_session(group_leader))
1079 			goto out;
1080 	}
1081 
1082 	err = security_task_setpgid(p, pgid);
1083 	if (err)
1084 		goto out;
1085 
1086 	if (task_pgrp(p) != pgrp)
1087 		change_pid(p, PIDTYPE_PGID, pgrp);
1088 
1089 	err = 0;
1090 out:
1091 	/* All paths lead to here, thus we are safe. -DaveM */
1092 	write_unlock_irq(&tasklist_lock);
1093 	rcu_read_unlock();
1094 	return err;
1095 }
1096 
1097 static int do_getpgid(pid_t pid)
1098 {
1099 	struct task_struct *p;
1100 	struct pid *grp;
1101 	int retval;
1102 
1103 	rcu_read_lock();
1104 	if (!pid)
1105 		grp = task_pgrp(current);
1106 	else {
1107 		retval = -ESRCH;
1108 		p = find_task_by_vpid(pid);
1109 		if (!p)
1110 			goto out;
1111 		grp = task_pgrp(p);
1112 		if (!grp)
1113 			goto out;
1114 
1115 		retval = security_task_getpgid(p);
1116 		if (retval)
1117 			goto out;
1118 	}
1119 	retval = pid_vnr(grp);
1120 out:
1121 	rcu_read_unlock();
1122 	return retval;
1123 }
1124 
1125 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1126 {
1127 	return do_getpgid(pid);
1128 }
1129 
1130 #ifdef __ARCH_WANT_SYS_GETPGRP
1131 
1132 SYSCALL_DEFINE0(getpgrp)
1133 {
1134 	return do_getpgid(0);
1135 }
1136 
1137 #endif
1138 
1139 SYSCALL_DEFINE1(getsid, pid_t, pid)
1140 {
1141 	struct task_struct *p;
1142 	struct pid *sid;
1143 	int retval;
1144 
1145 	rcu_read_lock();
1146 	if (!pid)
1147 		sid = task_session(current);
1148 	else {
1149 		retval = -ESRCH;
1150 		p = find_task_by_vpid(pid);
1151 		if (!p)
1152 			goto out;
1153 		sid = task_session(p);
1154 		if (!sid)
1155 			goto out;
1156 
1157 		retval = security_task_getsid(p);
1158 		if (retval)
1159 			goto out;
1160 	}
1161 	retval = pid_vnr(sid);
1162 out:
1163 	rcu_read_unlock();
1164 	return retval;
1165 }
1166 
1167 static void set_special_pids(struct pid *pid)
1168 {
1169 	struct task_struct *curr = current->group_leader;
1170 
1171 	if (task_session(curr) != pid)
1172 		change_pid(curr, PIDTYPE_SID, pid);
1173 
1174 	if (task_pgrp(curr) != pid)
1175 		change_pid(curr, PIDTYPE_PGID, pid);
1176 }
1177 
1178 int ksys_setsid(void)
1179 {
1180 	struct task_struct *group_leader = current->group_leader;
1181 	struct pid *sid = task_pid(group_leader);
1182 	pid_t session = pid_vnr(sid);
1183 	int err = -EPERM;
1184 
1185 	write_lock_irq(&tasklist_lock);
1186 	/* Fail if I am already a session leader */
1187 	if (group_leader->signal->leader)
1188 		goto out;
1189 
1190 	/* Fail if a process group id already exists that equals the
1191 	 * proposed session id.
1192 	 */
1193 	if (pid_task(sid, PIDTYPE_PGID))
1194 		goto out;
1195 
1196 	group_leader->signal->leader = 1;
1197 	set_special_pids(sid);
1198 
1199 	proc_clear_tty(group_leader);
1200 
1201 	err = session;
1202 out:
1203 	write_unlock_irq(&tasklist_lock);
1204 	if (err > 0) {
1205 		proc_sid_connector(group_leader);
1206 		sched_autogroup_create_attach(group_leader);
1207 	}
1208 	return err;
1209 }
1210 
1211 SYSCALL_DEFINE0(setsid)
1212 {
1213 	return ksys_setsid();
1214 }
1215 
1216 DECLARE_RWSEM(uts_sem);
1217 
1218 #ifdef COMPAT_UTS_MACHINE
1219 #define override_architecture(name) \
1220 	(personality(current->personality) == PER_LINUX32 && \
1221 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1222 		      sizeof(COMPAT_UTS_MACHINE)))
1223 #else
1224 #define override_architecture(name)	0
1225 #endif
1226 
1227 /*
1228  * Work around broken programs that cannot handle "Linux 3.0".
1229  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1230  * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1231  * 2.6.60.
1232  */
1233 static int override_release(char __user *release, size_t len)
1234 {
1235 	int ret = 0;
1236 
1237 	if (current->personality & UNAME26) {
1238 		const char *rest = UTS_RELEASE;
1239 		char buf[65] = { 0 };
1240 		int ndots = 0;
1241 		unsigned v;
1242 		size_t copy;
1243 
1244 		while (*rest) {
1245 			if (*rest == '.' && ++ndots >= 3)
1246 				break;
1247 			if (!isdigit(*rest) && *rest != '.')
1248 				break;
1249 			rest++;
1250 		}
1251 		v = LINUX_VERSION_PATCHLEVEL + 60;
1252 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1253 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1254 		ret = copy_to_user(release, buf, copy + 1);
1255 	}
1256 	return ret;
1257 }
1258 
1259 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1260 {
1261 	struct new_utsname tmp;
1262 
1263 	down_read(&uts_sem);
1264 	memcpy(&tmp, utsname(), sizeof(tmp));
1265 	up_read(&uts_sem);
1266 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1267 		return -EFAULT;
1268 
1269 	if (override_release(name->release, sizeof(name->release)))
1270 		return -EFAULT;
1271 	if (override_architecture(name))
1272 		return -EFAULT;
1273 	return 0;
1274 }
1275 
1276 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1277 /*
1278  * Old cruft
1279  */
1280 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1281 {
1282 	struct old_utsname tmp;
1283 
1284 	if (!name)
1285 		return -EFAULT;
1286 
1287 	down_read(&uts_sem);
1288 	memcpy(&tmp, utsname(), sizeof(tmp));
1289 	up_read(&uts_sem);
1290 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1291 		return -EFAULT;
1292 
1293 	if (override_release(name->release, sizeof(name->release)))
1294 		return -EFAULT;
1295 	if (override_architecture(name))
1296 		return -EFAULT;
1297 	return 0;
1298 }
1299 
1300 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1301 {
1302 	struct oldold_utsname tmp;
1303 
1304 	if (!name)
1305 		return -EFAULT;
1306 
1307 	memset(&tmp, 0, sizeof(tmp));
1308 
1309 	down_read(&uts_sem);
1310 	memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1311 	memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1312 	memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1313 	memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1314 	memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1315 	up_read(&uts_sem);
1316 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1317 		return -EFAULT;
1318 
1319 	if (override_architecture(name))
1320 		return -EFAULT;
1321 	if (override_release(name->release, sizeof(name->release)))
1322 		return -EFAULT;
1323 	return 0;
1324 }
1325 #endif
1326 
1327 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1328 {
1329 	int errno;
1330 	char tmp[__NEW_UTS_LEN];
1331 
1332 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1333 		return -EPERM;
1334 
1335 	if (len < 0 || len > __NEW_UTS_LEN)
1336 		return -EINVAL;
1337 	errno = -EFAULT;
1338 	if (!copy_from_user(tmp, name, len)) {
1339 		struct new_utsname *u;
1340 
1341 		down_write(&uts_sem);
1342 		u = utsname();
1343 		memcpy(u->nodename, tmp, len);
1344 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1345 		errno = 0;
1346 		uts_proc_notify(UTS_PROC_HOSTNAME);
1347 		up_write(&uts_sem);
1348 	}
1349 	return errno;
1350 }
1351 
1352 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1353 
1354 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1355 {
1356 	int i;
1357 	struct new_utsname *u;
1358 	char tmp[__NEW_UTS_LEN + 1];
1359 
1360 	if (len < 0)
1361 		return -EINVAL;
1362 	down_read(&uts_sem);
1363 	u = utsname();
1364 	i = 1 + strlen(u->nodename);
1365 	if (i > len)
1366 		i = len;
1367 	memcpy(tmp, u->nodename, i);
1368 	up_read(&uts_sem);
1369 	if (copy_to_user(name, tmp, i))
1370 		return -EFAULT;
1371 	return 0;
1372 }
1373 
1374 #endif
1375 
1376 /*
1377  * Only setdomainname; getdomainname can be implemented by calling
1378  * uname()
1379  */
1380 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1381 {
1382 	int errno;
1383 	char tmp[__NEW_UTS_LEN];
1384 
1385 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1386 		return -EPERM;
1387 	if (len < 0 || len > __NEW_UTS_LEN)
1388 		return -EINVAL;
1389 
1390 	errno = -EFAULT;
1391 	if (!copy_from_user(tmp, name, len)) {
1392 		struct new_utsname *u;
1393 
1394 		down_write(&uts_sem);
1395 		u = utsname();
1396 		memcpy(u->domainname, tmp, len);
1397 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1398 		errno = 0;
1399 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1400 		up_write(&uts_sem);
1401 	}
1402 	return errno;
1403 }
1404 
1405 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1406 {
1407 	struct rlimit value;
1408 	int ret;
1409 
1410 	ret = do_prlimit(current, resource, NULL, &value);
1411 	if (!ret)
1412 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1413 
1414 	return ret;
1415 }
1416 
1417 #ifdef CONFIG_COMPAT
1418 
1419 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1420 		       struct compat_rlimit __user *, rlim)
1421 {
1422 	struct rlimit r;
1423 	struct compat_rlimit r32;
1424 
1425 	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1426 		return -EFAULT;
1427 
1428 	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1429 		r.rlim_cur = RLIM_INFINITY;
1430 	else
1431 		r.rlim_cur = r32.rlim_cur;
1432 	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1433 		r.rlim_max = RLIM_INFINITY;
1434 	else
1435 		r.rlim_max = r32.rlim_max;
1436 	return do_prlimit(current, resource, &r, NULL);
1437 }
1438 
1439 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1440 		       struct compat_rlimit __user *, rlim)
1441 {
1442 	struct rlimit r;
1443 	int ret;
1444 
1445 	ret = do_prlimit(current, resource, NULL, &r);
1446 	if (!ret) {
1447 		struct compat_rlimit r32;
1448 		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1449 			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1450 		else
1451 			r32.rlim_cur = r.rlim_cur;
1452 		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1453 			r32.rlim_max = COMPAT_RLIM_INFINITY;
1454 		else
1455 			r32.rlim_max = r.rlim_max;
1456 
1457 		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1458 			return -EFAULT;
1459 	}
1460 	return ret;
1461 }
1462 
1463 #endif
1464 
1465 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1466 
1467 /*
1468  *	Back compatibility for getrlimit. Needed for some apps.
1469  */
1470 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1471 		struct rlimit __user *, rlim)
1472 {
1473 	struct rlimit x;
1474 	if (resource >= RLIM_NLIMITS)
1475 		return -EINVAL;
1476 
1477 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1478 	task_lock(current->group_leader);
1479 	x = current->signal->rlim[resource];
1480 	task_unlock(current->group_leader);
1481 	if (x.rlim_cur > 0x7FFFFFFF)
1482 		x.rlim_cur = 0x7FFFFFFF;
1483 	if (x.rlim_max > 0x7FFFFFFF)
1484 		x.rlim_max = 0x7FFFFFFF;
1485 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1486 }
1487 
1488 #ifdef CONFIG_COMPAT
1489 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1490 		       struct compat_rlimit __user *, rlim)
1491 {
1492 	struct rlimit r;
1493 
1494 	if (resource >= RLIM_NLIMITS)
1495 		return -EINVAL;
1496 
1497 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1498 	task_lock(current->group_leader);
1499 	r = current->signal->rlim[resource];
1500 	task_unlock(current->group_leader);
1501 	if (r.rlim_cur > 0x7FFFFFFF)
1502 		r.rlim_cur = 0x7FFFFFFF;
1503 	if (r.rlim_max > 0x7FFFFFFF)
1504 		r.rlim_max = 0x7FFFFFFF;
1505 
1506 	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1507 	    put_user(r.rlim_max, &rlim->rlim_max))
1508 		return -EFAULT;
1509 	return 0;
1510 }
1511 #endif
1512 
1513 #endif
1514 
1515 static inline bool rlim64_is_infinity(__u64 rlim64)
1516 {
1517 #if BITS_PER_LONG < 64
1518 	return rlim64 >= ULONG_MAX;
1519 #else
1520 	return rlim64 == RLIM64_INFINITY;
1521 #endif
1522 }
1523 
1524 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1525 {
1526 	if (rlim->rlim_cur == RLIM_INFINITY)
1527 		rlim64->rlim_cur = RLIM64_INFINITY;
1528 	else
1529 		rlim64->rlim_cur = rlim->rlim_cur;
1530 	if (rlim->rlim_max == RLIM_INFINITY)
1531 		rlim64->rlim_max = RLIM64_INFINITY;
1532 	else
1533 		rlim64->rlim_max = rlim->rlim_max;
1534 }
1535 
1536 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1537 {
1538 	if (rlim64_is_infinity(rlim64->rlim_cur))
1539 		rlim->rlim_cur = RLIM_INFINITY;
1540 	else
1541 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1542 	if (rlim64_is_infinity(rlim64->rlim_max))
1543 		rlim->rlim_max = RLIM_INFINITY;
1544 	else
1545 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1546 }
1547 
1548 /* make sure you are allowed to change @tsk limits before calling this */
1549 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1550 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1551 {
1552 	struct rlimit *rlim;
1553 	int retval = 0;
1554 
1555 	if (resource >= RLIM_NLIMITS)
1556 		return -EINVAL;
1557 	if (new_rlim) {
1558 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1559 			return -EINVAL;
1560 		if (resource == RLIMIT_NOFILE &&
1561 				new_rlim->rlim_max > sysctl_nr_open)
1562 			return -EPERM;
1563 	}
1564 
1565 	/* protect tsk->signal and tsk->sighand from disappearing */
1566 	read_lock(&tasklist_lock);
1567 	if (!tsk->sighand) {
1568 		retval = -ESRCH;
1569 		goto out;
1570 	}
1571 
1572 	rlim = tsk->signal->rlim + resource;
1573 	task_lock(tsk->group_leader);
1574 	if (new_rlim) {
1575 		/* Keep the capable check against init_user_ns until
1576 		   cgroups can contain all limits */
1577 		if (new_rlim->rlim_max > rlim->rlim_max &&
1578 				!capable(CAP_SYS_RESOURCE))
1579 			retval = -EPERM;
1580 		if (!retval)
1581 			retval = security_task_setrlimit(tsk, resource, new_rlim);
1582 	}
1583 	if (!retval) {
1584 		if (old_rlim)
1585 			*old_rlim = *rlim;
1586 		if (new_rlim)
1587 			*rlim = *new_rlim;
1588 	}
1589 	task_unlock(tsk->group_leader);
1590 
1591 	/*
1592 	 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1593 	 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1594 	 * ignores the rlimit.
1595 	 */
1596 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1597 	     new_rlim->rlim_cur != RLIM_INFINITY &&
1598 	     IS_ENABLED(CONFIG_POSIX_TIMERS))
1599 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1600 out:
1601 	read_unlock(&tasklist_lock);
1602 	return retval;
1603 }
1604 
1605 /* rcu lock must be held */
1606 static int check_prlimit_permission(struct task_struct *task,
1607 				    unsigned int flags)
1608 {
1609 	const struct cred *cred = current_cred(), *tcred;
1610 	bool id_match;
1611 
1612 	if (current == task)
1613 		return 0;
1614 
1615 	tcred = __task_cred(task);
1616 	id_match = (uid_eq(cred->uid, tcred->euid) &&
1617 		    uid_eq(cred->uid, tcred->suid) &&
1618 		    uid_eq(cred->uid, tcred->uid)  &&
1619 		    gid_eq(cred->gid, tcred->egid) &&
1620 		    gid_eq(cred->gid, tcred->sgid) &&
1621 		    gid_eq(cred->gid, tcred->gid));
1622 	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1623 		return -EPERM;
1624 
1625 	return security_task_prlimit(cred, tcred, flags);
1626 }
1627 
1628 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1629 		const struct rlimit64 __user *, new_rlim,
1630 		struct rlimit64 __user *, old_rlim)
1631 {
1632 	struct rlimit64 old64, new64;
1633 	struct rlimit old, new;
1634 	struct task_struct *tsk;
1635 	unsigned int checkflags = 0;
1636 	int ret;
1637 
1638 	if (old_rlim)
1639 		checkflags |= LSM_PRLIMIT_READ;
1640 
1641 	if (new_rlim) {
1642 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1643 			return -EFAULT;
1644 		rlim64_to_rlim(&new64, &new);
1645 		checkflags |= LSM_PRLIMIT_WRITE;
1646 	}
1647 
1648 	rcu_read_lock();
1649 	tsk = pid ? find_task_by_vpid(pid) : current;
1650 	if (!tsk) {
1651 		rcu_read_unlock();
1652 		return -ESRCH;
1653 	}
1654 	ret = check_prlimit_permission(tsk, checkflags);
1655 	if (ret) {
1656 		rcu_read_unlock();
1657 		return ret;
1658 	}
1659 	get_task_struct(tsk);
1660 	rcu_read_unlock();
1661 
1662 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1663 			old_rlim ? &old : NULL);
1664 
1665 	if (!ret && old_rlim) {
1666 		rlim_to_rlim64(&old, &old64);
1667 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1668 			ret = -EFAULT;
1669 	}
1670 
1671 	put_task_struct(tsk);
1672 	return ret;
1673 }
1674 
1675 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1676 {
1677 	struct rlimit new_rlim;
1678 
1679 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1680 		return -EFAULT;
1681 	return do_prlimit(current, resource, &new_rlim, NULL);
1682 }
1683 
1684 /*
1685  * It would make sense to put struct rusage in the task_struct,
1686  * except that would make the task_struct be *really big*.  After
1687  * task_struct gets moved into malloc'ed memory, it would
1688  * make sense to do this.  It will make moving the rest of the information
1689  * a lot simpler!  (Which we're not doing right now because we're not
1690  * measuring them yet).
1691  *
1692  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1693  * races with threads incrementing their own counters.  But since word
1694  * reads are atomic, we either get new values or old values and we don't
1695  * care which for the sums.  We always take the siglock to protect reading
1696  * the c* fields from p->signal from races with exit.c updating those
1697  * fields when reaping, so a sample either gets all the additions of a
1698  * given child after it's reaped, or none so this sample is before reaping.
1699  *
1700  * Locking:
1701  * We need to take the siglock for CHILDEREN, SELF and BOTH
1702  * for  the cases current multithreaded, non-current single threaded
1703  * non-current multithreaded.  Thread traversal is now safe with
1704  * the siglock held.
1705  * Strictly speaking, we donot need to take the siglock if we are current and
1706  * single threaded,  as no one else can take our signal_struct away, no one
1707  * else can  reap the  children to update signal->c* counters, and no one else
1708  * can race with the signal-> fields. If we do not take any lock, the
1709  * signal-> fields could be read out of order while another thread was just
1710  * exiting. So we should  place a read memory barrier when we avoid the lock.
1711  * On the writer side,  write memory barrier is implied in  __exit_signal
1712  * as __exit_signal releases  the siglock spinlock after updating the signal->
1713  * fields. But we don't do this yet to keep things simple.
1714  *
1715  */
1716 
1717 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1718 {
1719 	r->ru_nvcsw += t->nvcsw;
1720 	r->ru_nivcsw += t->nivcsw;
1721 	r->ru_minflt += t->min_flt;
1722 	r->ru_majflt += t->maj_flt;
1723 	r->ru_inblock += task_io_get_inblock(t);
1724 	r->ru_oublock += task_io_get_oublock(t);
1725 }
1726 
1727 void getrusage(struct task_struct *p, int who, struct rusage *r)
1728 {
1729 	struct task_struct *t;
1730 	unsigned long flags;
1731 	u64 tgutime, tgstime, utime, stime;
1732 	unsigned long maxrss = 0;
1733 
1734 	memset((char *)r, 0, sizeof (*r));
1735 	utime = stime = 0;
1736 
1737 	if (who == RUSAGE_THREAD) {
1738 		task_cputime_adjusted(current, &utime, &stime);
1739 		accumulate_thread_rusage(p, r);
1740 		maxrss = p->signal->maxrss;
1741 		goto out;
1742 	}
1743 
1744 	if (!lock_task_sighand(p, &flags))
1745 		return;
1746 
1747 	switch (who) {
1748 	case RUSAGE_BOTH:
1749 	case RUSAGE_CHILDREN:
1750 		utime = p->signal->cutime;
1751 		stime = p->signal->cstime;
1752 		r->ru_nvcsw = p->signal->cnvcsw;
1753 		r->ru_nivcsw = p->signal->cnivcsw;
1754 		r->ru_minflt = p->signal->cmin_flt;
1755 		r->ru_majflt = p->signal->cmaj_flt;
1756 		r->ru_inblock = p->signal->cinblock;
1757 		r->ru_oublock = p->signal->coublock;
1758 		maxrss = p->signal->cmaxrss;
1759 
1760 		if (who == RUSAGE_CHILDREN)
1761 			break;
1762 		fallthrough;
1763 
1764 	case RUSAGE_SELF:
1765 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1766 		utime += tgutime;
1767 		stime += tgstime;
1768 		r->ru_nvcsw += p->signal->nvcsw;
1769 		r->ru_nivcsw += p->signal->nivcsw;
1770 		r->ru_minflt += p->signal->min_flt;
1771 		r->ru_majflt += p->signal->maj_flt;
1772 		r->ru_inblock += p->signal->inblock;
1773 		r->ru_oublock += p->signal->oublock;
1774 		if (maxrss < p->signal->maxrss)
1775 			maxrss = p->signal->maxrss;
1776 		t = p;
1777 		do {
1778 			accumulate_thread_rusage(t, r);
1779 		} while_each_thread(p, t);
1780 		break;
1781 
1782 	default:
1783 		BUG();
1784 	}
1785 	unlock_task_sighand(p, &flags);
1786 
1787 out:
1788 	r->ru_utime = ns_to_kernel_old_timeval(utime);
1789 	r->ru_stime = ns_to_kernel_old_timeval(stime);
1790 
1791 	if (who != RUSAGE_CHILDREN) {
1792 		struct mm_struct *mm = get_task_mm(p);
1793 
1794 		if (mm) {
1795 			setmax_mm_hiwater_rss(&maxrss, mm);
1796 			mmput(mm);
1797 		}
1798 	}
1799 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1800 }
1801 
1802 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1803 {
1804 	struct rusage r;
1805 
1806 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1807 	    who != RUSAGE_THREAD)
1808 		return -EINVAL;
1809 
1810 	getrusage(current, who, &r);
1811 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1812 }
1813 
1814 #ifdef CONFIG_COMPAT
1815 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1816 {
1817 	struct rusage r;
1818 
1819 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1820 	    who != RUSAGE_THREAD)
1821 		return -EINVAL;
1822 
1823 	getrusage(current, who, &r);
1824 	return put_compat_rusage(&r, ru);
1825 }
1826 #endif
1827 
1828 SYSCALL_DEFINE1(umask, int, mask)
1829 {
1830 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1831 	return mask;
1832 }
1833 
1834 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1835 {
1836 	struct fd exe;
1837 	struct file *old_exe, *exe_file;
1838 	struct inode *inode;
1839 	int err;
1840 
1841 	exe = fdget(fd);
1842 	if (!exe.file)
1843 		return -EBADF;
1844 
1845 	inode = file_inode(exe.file);
1846 
1847 	/*
1848 	 * Because the original mm->exe_file points to executable file, make
1849 	 * sure that this one is executable as well, to avoid breaking an
1850 	 * overall picture.
1851 	 */
1852 	err = -EACCES;
1853 	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1854 		goto exit;
1855 
1856 	err = file_permission(exe.file, MAY_EXEC);
1857 	if (err)
1858 		goto exit;
1859 
1860 	/*
1861 	 * Forbid mm->exe_file change if old file still mapped.
1862 	 */
1863 	exe_file = get_mm_exe_file(mm);
1864 	err = -EBUSY;
1865 	if (exe_file) {
1866 		struct vm_area_struct *vma;
1867 
1868 		mmap_read_lock(mm);
1869 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
1870 			if (!vma->vm_file)
1871 				continue;
1872 			if (path_equal(&vma->vm_file->f_path,
1873 				       &exe_file->f_path))
1874 				goto exit_err;
1875 		}
1876 
1877 		mmap_read_unlock(mm);
1878 		fput(exe_file);
1879 	}
1880 
1881 	err = 0;
1882 	/* set the new file, lockless */
1883 	get_file(exe.file);
1884 	old_exe = xchg(&mm->exe_file, exe.file);
1885 	if (old_exe)
1886 		fput(old_exe);
1887 exit:
1888 	fdput(exe);
1889 	return err;
1890 exit_err:
1891 	mmap_read_unlock(mm);
1892 	fput(exe_file);
1893 	goto exit;
1894 }
1895 
1896 /*
1897  * Check arithmetic relations of passed addresses.
1898  *
1899  * WARNING: we don't require any capability here so be very careful
1900  * in what is allowed for modification from userspace.
1901  */
1902 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1903 {
1904 	unsigned long mmap_max_addr = TASK_SIZE;
1905 	int error = -EINVAL, i;
1906 
1907 	static const unsigned char offsets[] = {
1908 		offsetof(struct prctl_mm_map, start_code),
1909 		offsetof(struct prctl_mm_map, end_code),
1910 		offsetof(struct prctl_mm_map, start_data),
1911 		offsetof(struct prctl_mm_map, end_data),
1912 		offsetof(struct prctl_mm_map, start_brk),
1913 		offsetof(struct prctl_mm_map, brk),
1914 		offsetof(struct prctl_mm_map, start_stack),
1915 		offsetof(struct prctl_mm_map, arg_start),
1916 		offsetof(struct prctl_mm_map, arg_end),
1917 		offsetof(struct prctl_mm_map, env_start),
1918 		offsetof(struct prctl_mm_map, env_end),
1919 	};
1920 
1921 	/*
1922 	 * Make sure the members are not somewhere outside
1923 	 * of allowed address space.
1924 	 */
1925 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1926 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1927 
1928 		if ((unsigned long)val >= mmap_max_addr ||
1929 		    (unsigned long)val < mmap_min_addr)
1930 			goto out;
1931 	}
1932 
1933 	/*
1934 	 * Make sure the pairs are ordered.
1935 	 */
1936 #define __prctl_check_order(__m1, __op, __m2)				\
1937 	((unsigned long)prctl_map->__m1 __op				\
1938 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1939 	error  = __prctl_check_order(start_code, <, end_code);
1940 	error |= __prctl_check_order(start_data,<=, end_data);
1941 	error |= __prctl_check_order(start_brk, <=, brk);
1942 	error |= __prctl_check_order(arg_start, <=, arg_end);
1943 	error |= __prctl_check_order(env_start, <=, env_end);
1944 	if (error)
1945 		goto out;
1946 #undef __prctl_check_order
1947 
1948 	error = -EINVAL;
1949 
1950 	/*
1951 	 * @brk should be after @end_data in traditional maps.
1952 	 */
1953 	if (prctl_map->start_brk <= prctl_map->end_data ||
1954 	    prctl_map->brk <= prctl_map->end_data)
1955 		goto out;
1956 
1957 	/*
1958 	 * Neither we should allow to override limits if they set.
1959 	 */
1960 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1961 			      prctl_map->start_brk, prctl_map->end_data,
1962 			      prctl_map->start_data))
1963 			goto out;
1964 
1965 	error = 0;
1966 out:
1967 	return error;
1968 }
1969 
1970 #ifdef CONFIG_CHECKPOINT_RESTORE
1971 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1972 {
1973 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1974 	unsigned long user_auxv[AT_VECTOR_SIZE];
1975 	struct mm_struct *mm = current->mm;
1976 	int error;
1977 
1978 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1979 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1980 
1981 	if (opt == PR_SET_MM_MAP_SIZE)
1982 		return put_user((unsigned int)sizeof(prctl_map),
1983 				(unsigned int __user *)addr);
1984 
1985 	if (data_size != sizeof(prctl_map))
1986 		return -EINVAL;
1987 
1988 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1989 		return -EFAULT;
1990 
1991 	error = validate_prctl_map_addr(&prctl_map);
1992 	if (error)
1993 		return error;
1994 
1995 	if (prctl_map.auxv_size) {
1996 		/*
1997 		 * Someone is trying to cheat the auxv vector.
1998 		 */
1999 		if (!prctl_map.auxv ||
2000 				prctl_map.auxv_size > sizeof(mm->saved_auxv))
2001 			return -EINVAL;
2002 
2003 		memset(user_auxv, 0, sizeof(user_auxv));
2004 		if (copy_from_user(user_auxv,
2005 				   (const void __user *)prctl_map.auxv,
2006 				   prctl_map.auxv_size))
2007 			return -EFAULT;
2008 
2009 		/* Last entry must be AT_NULL as specification requires */
2010 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2011 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2012 	}
2013 
2014 	if (prctl_map.exe_fd != (u32)-1) {
2015 		/*
2016 		 * Check if the current user is checkpoint/restore capable.
2017 		 * At the time of this writing, it checks for CAP_SYS_ADMIN
2018 		 * or CAP_CHECKPOINT_RESTORE.
2019 		 * Note that a user with access to ptrace can masquerade an
2020 		 * arbitrary program as any executable, even setuid ones.
2021 		 * This may have implications in the tomoyo subsystem.
2022 		 */
2023 		if (!checkpoint_restore_ns_capable(current_user_ns()))
2024 			return -EPERM;
2025 
2026 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2027 		if (error)
2028 			return error;
2029 	}
2030 
2031 	/*
2032 	 * arg_lock protects concurrent updates but we still need mmap_lock for
2033 	 * read to exclude races with sys_brk.
2034 	 */
2035 	mmap_read_lock(mm);
2036 
2037 	/*
2038 	 * We don't validate if these members are pointing to
2039 	 * real present VMAs because application may have correspond
2040 	 * VMAs already unmapped and kernel uses these members for statistics
2041 	 * output in procfs mostly, except
2042 	 *
2043 	 *  - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2044 	 *    for VMAs when updating these members so anything wrong written
2045 	 *    here cause kernel to swear at userspace program but won't lead
2046 	 *    to any problem in kernel itself
2047 	 */
2048 
2049 	spin_lock(&mm->arg_lock);
2050 	mm->start_code	= prctl_map.start_code;
2051 	mm->end_code	= prctl_map.end_code;
2052 	mm->start_data	= prctl_map.start_data;
2053 	mm->end_data	= prctl_map.end_data;
2054 	mm->start_brk	= prctl_map.start_brk;
2055 	mm->brk		= prctl_map.brk;
2056 	mm->start_stack	= prctl_map.start_stack;
2057 	mm->arg_start	= prctl_map.arg_start;
2058 	mm->arg_end	= prctl_map.arg_end;
2059 	mm->env_start	= prctl_map.env_start;
2060 	mm->env_end	= prctl_map.env_end;
2061 	spin_unlock(&mm->arg_lock);
2062 
2063 	/*
2064 	 * Note this update of @saved_auxv is lockless thus
2065 	 * if someone reads this member in procfs while we're
2066 	 * updating -- it may get partly updated results. It's
2067 	 * known and acceptable trade off: we leave it as is to
2068 	 * not introduce additional locks here making the kernel
2069 	 * more complex.
2070 	 */
2071 	if (prctl_map.auxv_size)
2072 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2073 
2074 	mmap_read_unlock(mm);
2075 	return 0;
2076 }
2077 #endif /* CONFIG_CHECKPOINT_RESTORE */
2078 
2079 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2080 			  unsigned long len)
2081 {
2082 	/*
2083 	 * This doesn't move the auxiliary vector itself since it's pinned to
2084 	 * mm_struct, but it permits filling the vector with new values.  It's
2085 	 * up to the caller to provide sane values here, otherwise userspace
2086 	 * tools which use this vector might be unhappy.
2087 	 */
2088 	unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2089 
2090 	if (len > sizeof(user_auxv))
2091 		return -EINVAL;
2092 
2093 	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2094 		return -EFAULT;
2095 
2096 	/* Make sure the last entry is always AT_NULL */
2097 	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2098 	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2099 
2100 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2101 
2102 	task_lock(current);
2103 	memcpy(mm->saved_auxv, user_auxv, len);
2104 	task_unlock(current);
2105 
2106 	return 0;
2107 }
2108 
2109 static int prctl_set_mm(int opt, unsigned long addr,
2110 			unsigned long arg4, unsigned long arg5)
2111 {
2112 	struct mm_struct *mm = current->mm;
2113 	struct prctl_mm_map prctl_map = {
2114 		.auxv = NULL,
2115 		.auxv_size = 0,
2116 		.exe_fd = -1,
2117 	};
2118 	struct vm_area_struct *vma;
2119 	int error;
2120 
2121 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2122 			      opt != PR_SET_MM_MAP &&
2123 			      opt != PR_SET_MM_MAP_SIZE)))
2124 		return -EINVAL;
2125 
2126 #ifdef CONFIG_CHECKPOINT_RESTORE
2127 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2128 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2129 #endif
2130 
2131 	if (!capable(CAP_SYS_RESOURCE))
2132 		return -EPERM;
2133 
2134 	if (opt == PR_SET_MM_EXE_FILE)
2135 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2136 
2137 	if (opt == PR_SET_MM_AUXV)
2138 		return prctl_set_auxv(mm, addr, arg4);
2139 
2140 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2141 		return -EINVAL;
2142 
2143 	error = -EINVAL;
2144 
2145 	/*
2146 	 * arg_lock protects concurrent updates of arg boundaries, we need
2147 	 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2148 	 * validation.
2149 	 */
2150 	mmap_read_lock(mm);
2151 	vma = find_vma(mm, addr);
2152 
2153 	spin_lock(&mm->arg_lock);
2154 	prctl_map.start_code	= mm->start_code;
2155 	prctl_map.end_code	= mm->end_code;
2156 	prctl_map.start_data	= mm->start_data;
2157 	prctl_map.end_data	= mm->end_data;
2158 	prctl_map.start_brk	= mm->start_brk;
2159 	prctl_map.brk		= mm->brk;
2160 	prctl_map.start_stack	= mm->start_stack;
2161 	prctl_map.arg_start	= mm->arg_start;
2162 	prctl_map.arg_end	= mm->arg_end;
2163 	prctl_map.env_start	= mm->env_start;
2164 	prctl_map.env_end	= mm->env_end;
2165 
2166 	switch (opt) {
2167 	case PR_SET_MM_START_CODE:
2168 		prctl_map.start_code = addr;
2169 		break;
2170 	case PR_SET_MM_END_CODE:
2171 		prctl_map.end_code = addr;
2172 		break;
2173 	case PR_SET_MM_START_DATA:
2174 		prctl_map.start_data = addr;
2175 		break;
2176 	case PR_SET_MM_END_DATA:
2177 		prctl_map.end_data = addr;
2178 		break;
2179 	case PR_SET_MM_START_STACK:
2180 		prctl_map.start_stack = addr;
2181 		break;
2182 	case PR_SET_MM_START_BRK:
2183 		prctl_map.start_brk = addr;
2184 		break;
2185 	case PR_SET_MM_BRK:
2186 		prctl_map.brk = addr;
2187 		break;
2188 	case PR_SET_MM_ARG_START:
2189 		prctl_map.arg_start = addr;
2190 		break;
2191 	case PR_SET_MM_ARG_END:
2192 		prctl_map.arg_end = addr;
2193 		break;
2194 	case PR_SET_MM_ENV_START:
2195 		prctl_map.env_start = addr;
2196 		break;
2197 	case PR_SET_MM_ENV_END:
2198 		prctl_map.env_end = addr;
2199 		break;
2200 	default:
2201 		goto out;
2202 	}
2203 
2204 	error = validate_prctl_map_addr(&prctl_map);
2205 	if (error)
2206 		goto out;
2207 
2208 	switch (opt) {
2209 	/*
2210 	 * If command line arguments and environment
2211 	 * are placed somewhere else on stack, we can
2212 	 * set them up here, ARG_START/END to setup
2213 	 * command line arguments and ENV_START/END
2214 	 * for environment.
2215 	 */
2216 	case PR_SET_MM_START_STACK:
2217 	case PR_SET_MM_ARG_START:
2218 	case PR_SET_MM_ARG_END:
2219 	case PR_SET_MM_ENV_START:
2220 	case PR_SET_MM_ENV_END:
2221 		if (!vma) {
2222 			error = -EFAULT;
2223 			goto out;
2224 		}
2225 	}
2226 
2227 	mm->start_code	= prctl_map.start_code;
2228 	mm->end_code	= prctl_map.end_code;
2229 	mm->start_data	= prctl_map.start_data;
2230 	mm->end_data	= prctl_map.end_data;
2231 	mm->start_brk	= prctl_map.start_brk;
2232 	mm->brk		= prctl_map.brk;
2233 	mm->start_stack	= prctl_map.start_stack;
2234 	mm->arg_start	= prctl_map.arg_start;
2235 	mm->arg_end	= prctl_map.arg_end;
2236 	mm->env_start	= prctl_map.env_start;
2237 	mm->env_end	= prctl_map.env_end;
2238 
2239 	error = 0;
2240 out:
2241 	spin_unlock(&mm->arg_lock);
2242 	mmap_read_unlock(mm);
2243 	return error;
2244 }
2245 
2246 #ifdef CONFIG_CHECKPOINT_RESTORE
2247 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2248 {
2249 	return put_user(me->clear_child_tid, tid_addr);
2250 }
2251 #else
2252 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2253 {
2254 	return -EINVAL;
2255 }
2256 #endif
2257 
2258 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2259 {
2260 	/*
2261 	 * If task has has_child_subreaper - all its descendants
2262 	 * already have these flag too and new descendants will
2263 	 * inherit it on fork, skip them.
2264 	 *
2265 	 * If we've found child_reaper - skip descendants in
2266 	 * it's subtree as they will never get out pidns.
2267 	 */
2268 	if (p->signal->has_child_subreaper ||
2269 	    is_child_reaper(task_pid(p)))
2270 		return 0;
2271 
2272 	p->signal->has_child_subreaper = 1;
2273 	return 1;
2274 }
2275 
2276 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2277 {
2278 	return -EINVAL;
2279 }
2280 
2281 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2282 				    unsigned long ctrl)
2283 {
2284 	return -EINVAL;
2285 }
2286 
2287 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2288 
2289 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2290 		unsigned long, arg4, unsigned long, arg5)
2291 {
2292 	struct task_struct *me = current;
2293 	unsigned char comm[sizeof(me->comm)];
2294 	long error;
2295 
2296 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2297 	if (error != -ENOSYS)
2298 		return error;
2299 
2300 	error = 0;
2301 	switch (option) {
2302 	case PR_SET_PDEATHSIG:
2303 		if (!valid_signal(arg2)) {
2304 			error = -EINVAL;
2305 			break;
2306 		}
2307 		me->pdeath_signal = arg2;
2308 		break;
2309 	case PR_GET_PDEATHSIG:
2310 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2311 		break;
2312 	case PR_GET_DUMPABLE:
2313 		error = get_dumpable(me->mm);
2314 		break;
2315 	case PR_SET_DUMPABLE:
2316 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2317 			error = -EINVAL;
2318 			break;
2319 		}
2320 		set_dumpable(me->mm, arg2);
2321 		break;
2322 
2323 	case PR_SET_UNALIGN:
2324 		error = SET_UNALIGN_CTL(me, arg2);
2325 		break;
2326 	case PR_GET_UNALIGN:
2327 		error = GET_UNALIGN_CTL(me, arg2);
2328 		break;
2329 	case PR_SET_FPEMU:
2330 		error = SET_FPEMU_CTL(me, arg2);
2331 		break;
2332 	case PR_GET_FPEMU:
2333 		error = GET_FPEMU_CTL(me, arg2);
2334 		break;
2335 	case PR_SET_FPEXC:
2336 		error = SET_FPEXC_CTL(me, arg2);
2337 		break;
2338 	case PR_GET_FPEXC:
2339 		error = GET_FPEXC_CTL(me, arg2);
2340 		break;
2341 	case PR_GET_TIMING:
2342 		error = PR_TIMING_STATISTICAL;
2343 		break;
2344 	case PR_SET_TIMING:
2345 		if (arg2 != PR_TIMING_STATISTICAL)
2346 			error = -EINVAL;
2347 		break;
2348 	case PR_SET_NAME:
2349 		comm[sizeof(me->comm) - 1] = 0;
2350 		if (strncpy_from_user(comm, (char __user *)arg2,
2351 				      sizeof(me->comm) - 1) < 0)
2352 			return -EFAULT;
2353 		set_task_comm(me, comm);
2354 		proc_comm_connector(me);
2355 		break;
2356 	case PR_GET_NAME:
2357 		get_task_comm(comm, me);
2358 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2359 			return -EFAULT;
2360 		break;
2361 	case PR_GET_ENDIAN:
2362 		error = GET_ENDIAN(me, arg2);
2363 		break;
2364 	case PR_SET_ENDIAN:
2365 		error = SET_ENDIAN(me, arg2);
2366 		break;
2367 	case PR_GET_SECCOMP:
2368 		error = prctl_get_seccomp();
2369 		break;
2370 	case PR_SET_SECCOMP:
2371 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2372 		break;
2373 	case PR_GET_TSC:
2374 		error = GET_TSC_CTL(arg2);
2375 		break;
2376 	case PR_SET_TSC:
2377 		error = SET_TSC_CTL(arg2);
2378 		break;
2379 	case PR_TASK_PERF_EVENTS_DISABLE:
2380 		error = perf_event_task_disable();
2381 		break;
2382 	case PR_TASK_PERF_EVENTS_ENABLE:
2383 		error = perf_event_task_enable();
2384 		break;
2385 	case PR_GET_TIMERSLACK:
2386 		if (current->timer_slack_ns > ULONG_MAX)
2387 			error = ULONG_MAX;
2388 		else
2389 			error = current->timer_slack_ns;
2390 		break;
2391 	case PR_SET_TIMERSLACK:
2392 		if (arg2 <= 0)
2393 			current->timer_slack_ns =
2394 					current->default_timer_slack_ns;
2395 		else
2396 			current->timer_slack_ns = arg2;
2397 		break;
2398 	case PR_MCE_KILL:
2399 		if (arg4 | arg5)
2400 			return -EINVAL;
2401 		switch (arg2) {
2402 		case PR_MCE_KILL_CLEAR:
2403 			if (arg3 != 0)
2404 				return -EINVAL;
2405 			current->flags &= ~PF_MCE_PROCESS;
2406 			break;
2407 		case PR_MCE_KILL_SET:
2408 			current->flags |= PF_MCE_PROCESS;
2409 			if (arg3 == PR_MCE_KILL_EARLY)
2410 				current->flags |= PF_MCE_EARLY;
2411 			else if (arg3 == PR_MCE_KILL_LATE)
2412 				current->flags &= ~PF_MCE_EARLY;
2413 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2414 				current->flags &=
2415 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2416 			else
2417 				return -EINVAL;
2418 			break;
2419 		default:
2420 			return -EINVAL;
2421 		}
2422 		break;
2423 	case PR_MCE_KILL_GET:
2424 		if (arg2 | arg3 | arg4 | arg5)
2425 			return -EINVAL;
2426 		if (current->flags & PF_MCE_PROCESS)
2427 			error = (current->flags & PF_MCE_EARLY) ?
2428 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2429 		else
2430 			error = PR_MCE_KILL_DEFAULT;
2431 		break;
2432 	case PR_SET_MM:
2433 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2434 		break;
2435 	case PR_GET_TID_ADDRESS:
2436 		error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2437 		break;
2438 	case PR_SET_CHILD_SUBREAPER:
2439 		me->signal->is_child_subreaper = !!arg2;
2440 		if (!arg2)
2441 			break;
2442 
2443 		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2444 		break;
2445 	case PR_GET_CHILD_SUBREAPER:
2446 		error = put_user(me->signal->is_child_subreaper,
2447 				 (int __user *)arg2);
2448 		break;
2449 	case PR_SET_NO_NEW_PRIVS:
2450 		if (arg2 != 1 || arg3 || arg4 || arg5)
2451 			return -EINVAL;
2452 
2453 		task_set_no_new_privs(current);
2454 		break;
2455 	case PR_GET_NO_NEW_PRIVS:
2456 		if (arg2 || arg3 || arg4 || arg5)
2457 			return -EINVAL;
2458 		return task_no_new_privs(current) ? 1 : 0;
2459 	case PR_GET_THP_DISABLE:
2460 		if (arg2 || arg3 || arg4 || arg5)
2461 			return -EINVAL;
2462 		error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2463 		break;
2464 	case PR_SET_THP_DISABLE:
2465 		if (arg3 || arg4 || arg5)
2466 			return -EINVAL;
2467 		if (mmap_write_lock_killable(me->mm))
2468 			return -EINTR;
2469 		if (arg2)
2470 			set_bit(MMF_DISABLE_THP, &me->mm->flags);
2471 		else
2472 			clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2473 		mmap_write_unlock(me->mm);
2474 		break;
2475 	case PR_MPX_ENABLE_MANAGEMENT:
2476 	case PR_MPX_DISABLE_MANAGEMENT:
2477 		/* No longer implemented: */
2478 		return -EINVAL;
2479 	case PR_SET_FP_MODE:
2480 		error = SET_FP_MODE(me, arg2);
2481 		break;
2482 	case PR_GET_FP_MODE:
2483 		error = GET_FP_MODE(me);
2484 		break;
2485 	case PR_SVE_SET_VL:
2486 		error = SVE_SET_VL(arg2);
2487 		break;
2488 	case PR_SVE_GET_VL:
2489 		error = SVE_GET_VL();
2490 		break;
2491 	case PR_GET_SPECULATION_CTRL:
2492 		if (arg3 || arg4 || arg5)
2493 			return -EINVAL;
2494 		error = arch_prctl_spec_ctrl_get(me, arg2);
2495 		break;
2496 	case PR_SET_SPECULATION_CTRL:
2497 		if (arg4 || arg5)
2498 			return -EINVAL;
2499 		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2500 		break;
2501 	case PR_PAC_RESET_KEYS:
2502 		if (arg3 || arg4 || arg5)
2503 			return -EINVAL;
2504 		error = PAC_RESET_KEYS(me, arg2);
2505 		break;
2506 	case PR_PAC_SET_ENABLED_KEYS:
2507 		if (arg4 || arg5)
2508 			return -EINVAL;
2509 		error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2510 		break;
2511 	case PR_PAC_GET_ENABLED_KEYS:
2512 		if (arg2 || arg3 || arg4 || arg5)
2513 			return -EINVAL;
2514 		error = PAC_GET_ENABLED_KEYS(me);
2515 		break;
2516 	case PR_SET_TAGGED_ADDR_CTRL:
2517 		if (arg3 || arg4 || arg5)
2518 			return -EINVAL;
2519 		error = SET_TAGGED_ADDR_CTRL(arg2);
2520 		break;
2521 	case PR_GET_TAGGED_ADDR_CTRL:
2522 		if (arg2 || arg3 || arg4 || arg5)
2523 			return -EINVAL;
2524 		error = GET_TAGGED_ADDR_CTRL();
2525 		break;
2526 	case PR_SET_IO_FLUSHER:
2527 		if (!capable(CAP_SYS_RESOURCE))
2528 			return -EPERM;
2529 
2530 		if (arg3 || arg4 || arg5)
2531 			return -EINVAL;
2532 
2533 		if (arg2 == 1)
2534 			current->flags |= PR_IO_FLUSHER;
2535 		else if (!arg2)
2536 			current->flags &= ~PR_IO_FLUSHER;
2537 		else
2538 			return -EINVAL;
2539 		break;
2540 	case PR_GET_IO_FLUSHER:
2541 		if (!capable(CAP_SYS_RESOURCE))
2542 			return -EPERM;
2543 
2544 		if (arg2 || arg3 || arg4 || arg5)
2545 			return -EINVAL;
2546 
2547 		error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2548 		break;
2549 	case PR_SET_SYSCALL_USER_DISPATCH:
2550 		error = set_syscall_user_dispatch(arg2, arg3, arg4,
2551 						  (char __user *) arg5);
2552 		break;
2553 	default:
2554 		error = -EINVAL;
2555 		break;
2556 	}
2557 	return error;
2558 }
2559 
2560 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2561 		struct getcpu_cache __user *, unused)
2562 {
2563 	int err = 0;
2564 	int cpu = raw_smp_processor_id();
2565 
2566 	if (cpup)
2567 		err |= put_user(cpu, cpup);
2568 	if (nodep)
2569 		err |= put_user(cpu_to_node(cpu), nodep);
2570 	return err ? -EFAULT : 0;
2571 }
2572 
2573 /**
2574  * do_sysinfo - fill in sysinfo struct
2575  * @info: pointer to buffer to fill
2576  */
2577 static int do_sysinfo(struct sysinfo *info)
2578 {
2579 	unsigned long mem_total, sav_total;
2580 	unsigned int mem_unit, bitcount;
2581 	struct timespec64 tp;
2582 
2583 	memset(info, 0, sizeof(struct sysinfo));
2584 
2585 	ktime_get_boottime_ts64(&tp);
2586 	timens_add_boottime(&tp);
2587 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2588 
2589 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2590 
2591 	info->procs = nr_threads;
2592 
2593 	si_meminfo(info);
2594 	si_swapinfo(info);
2595 
2596 	/*
2597 	 * If the sum of all the available memory (i.e. ram + swap)
2598 	 * is less than can be stored in a 32 bit unsigned long then
2599 	 * we can be binary compatible with 2.2.x kernels.  If not,
2600 	 * well, in that case 2.2.x was broken anyways...
2601 	 *
2602 	 *  -Erik Andersen <andersee@debian.org>
2603 	 */
2604 
2605 	mem_total = info->totalram + info->totalswap;
2606 	if (mem_total < info->totalram || mem_total < info->totalswap)
2607 		goto out;
2608 	bitcount = 0;
2609 	mem_unit = info->mem_unit;
2610 	while (mem_unit > 1) {
2611 		bitcount++;
2612 		mem_unit >>= 1;
2613 		sav_total = mem_total;
2614 		mem_total <<= 1;
2615 		if (mem_total < sav_total)
2616 			goto out;
2617 	}
2618 
2619 	/*
2620 	 * If mem_total did not overflow, multiply all memory values by
2621 	 * info->mem_unit and set it to 1.  This leaves things compatible
2622 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2623 	 * kernels...
2624 	 */
2625 
2626 	info->mem_unit = 1;
2627 	info->totalram <<= bitcount;
2628 	info->freeram <<= bitcount;
2629 	info->sharedram <<= bitcount;
2630 	info->bufferram <<= bitcount;
2631 	info->totalswap <<= bitcount;
2632 	info->freeswap <<= bitcount;
2633 	info->totalhigh <<= bitcount;
2634 	info->freehigh <<= bitcount;
2635 
2636 out:
2637 	return 0;
2638 }
2639 
2640 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2641 {
2642 	struct sysinfo val;
2643 
2644 	do_sysinfo(&val);
2645 
2646 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2647 		return -EFAULT;
2648 
2649 	return 0;
2650 }
2651 
2652 #ifdef CONFIG_COMPAT
2653 struct compat_sysinfo {
2654 	s32 uptime;
2655 	u32 loads[3];
2656 	u32 totalram;
2657 	u32 freeram;
2658 	u32 sharedram;
2659 	u32 bufferram;
2660 	u32 totalswap;
2661 	u32 freeswap;
2662 	u16 procs;
2663 	u16 pad;
2664 	u32 totalhigh;
2665 	u32 freehigh;
2666 	u32 mem_unit;
2667 	char _f[20-2*sizeof(u32)-sizeof(int)];
2668 };
2669 
2670 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2671 {
2672 	struct sysinfo s;
2673 	struct compat_sysinfo s_32;
2674 
2675 	do_sysinfo(&s);
2676 
2677 	/* Check to see if any memory value is too large for 32-bit and scale
2678 	 *  down if needed
2679 	 */
2680 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2681 		int bitcount = 0;
2682 
2683 		while (s.mem_unit < PAGE_SIZE) {
2684 			s.mem_unit <<= 1;
2685 			bitcount++;
2686 		}
2687 
2688 		s.totalram >>= bitcount;
2689 		s.freeram >>= bitcount;
2690 		s.sharedram >>= bitcount;
2691 		s.bufferram >>= bitcount;
2692 		s.totalswap >>= bitcount;
2693 		s.freeswap >>= bitcount;
2694 		s.totalhigh >>= bitcount;
2695 		s.freehigh >>= bitcount;
2696 	}
2697 
2698 	memset(&s_32, 0, sizeof(s_32));
2699 	s_32.uptime = s.uptime;
2700 	s_32.loads[0] = s.loads[0];
2701 	s_32.loads[1] = s.loads[1];
2702 	s_32.loads[2] = s.loads[2];
2703 	s_32.totalram = s.totalram;
2704 	s_32.freeram = s.freeram;
2705 	s_32.sharedram = s.sharedram;
2706 	s_32.bufferram = s.bufferram;
2707 	s_32.totalswap = s.totalswap;
2708 	s_32.freeswap = s.freeswap;
2709 	s_32.procs = s.procs;
2710 	s_32.totalhigh = s.totalhigh;
2711 	s_32.freehigh = s.freehigh;
2712 	s_32.mem_unit = s.mem_unit;
2713 	if (copy_to_user(info, &s_32, sizeof(s_32)))
2714 		return -EFAULT;
2715 	return 0;
2716 }
2717 #endif /* CONFIG_COMPAT */
2718