xref: /openbmc/linux/kernel/sys.c (revision 6342a20e)
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/mm_inline.h>
11 #include <linux/utsname.h>
12 #include <linux/mman.h>
13 #include <linux/reboot.h>
14 #include <linux/prctl.h>
15 #include <linux/highuid.h>
16 #include <linux/fs.h>
17 #include <linux/kmod.h>
18 #include <linux/ksm.h>
19 #include <linux/perf_event.h>
20 #include <linux/resource.h>
21 #include <linux/kernel.h>
22 #include <linux/workqueue.h>
23 #include <linux/capability.h>
24 #include <linux/device.h>
25 #include <linux/key.h>
26 #include <linux/times.h>
27 #include <linux/posix-timers.h>
28 #include <linux/security.h>
29 #include <linux/random.h>
30 #include <linux/suspend.h>
31 #include <linux/tty.h>
32 #include <linux/signal.h>
33 #include <linux/cn_proc.h>
34 #include <linux/getcpu.h>
35 #include <linux/task_io_accounting_ops.h>
36 #include <linux/seccomp.h>
37 #include <linux/cpu.h>
38 #include <linux/personality.h>
39 #include <linux/ptrace.h>
40 #include <linux/fs_struct.h>
41 #include <linux/file.h>
42 #include <linux/mount.h>
43 #include <linux/gfp.h>
44 #include <linux/syscore_ops.h>
45 #include <linux/version.h>
46 #include <linux/ctype.h>
47 #include <linux/syscall_user_dispatch.h>
48 
49 #include <linux/compat.h>
50 #include <linux/syscalls.h>
51 #include <linux/kprobes.h>
52 #include <linux/user_namespace.h>
53 #include <linux/time_namespace.h>
54 #include <linux/binfmts.h>
55 
56 #include <linux/sched.h>
57 #include <linux/sched/autogroup.h>
58 #include <linux/sched/loadavg.h>
59 #include <linux/sched/stat.h>
60 #include <linux/sched/mm.h>
61 #include <linux/sched/coredump.h>
62 #include <linux/sched/task.h>
63 #include <linux/sched/cputime.h>
64 #include <linux/rcupdate.h>
65 #include <linux/uidgid.h>
66 #include <linux/cred.h>
67 
68 #include <linux/nospec.h>
69 
70 #include <linux/kmsg_dump.h>
71 /* Move somewhere else to avoid recompiling? */
72 #include <generated/utsrelease.h>
73 
74 #include <linux/uaccess.h>
75 #include <asm/io.h>
76 #include <asm/unistd.h>
77 
78 #include "uid16.h"
79 
80 #ifndef SET_UNALIGN_CTL
81 # define SET_UNALIGN_CTL(a, b)	(-EINVAL)
82 #endif
83 #ifndef GET_UNALIGN_CTL
84 # define GET_UNALIGN_CTL(a, b)	(-EINVAL)
85 #endif
86 #ifndef SET_FPEMU_CTL
87 # define SET_FPEMU_CTL(a, b)	(-EINVAL)
88 #endif
89 #ifndef GET_FPEMU_CTL
90 # define GET_FPEMU_CTL(a, b)	(-EINVAL)
91 #endif
92 #ifndef SET_FPEXC_CTL
93 # define SET_FPEXC_CTL(a, b)	(-EINVAL)
94 #endif
95 #ifndef GET_FPEXC_CTL
96 # define GET_FPEXC_CTL(a, b)	(-EINVAL)
97 #endif
98 #ifndef GET_ENDIAN
99 # define GET_ENDIAN(a, b)	(-EINVAL)
100 #endif
101 #ifndef SET_ENDIAN
102 # define SET_ENDIAN(a, b)	(-EINVAL)
103 #endif
104 #ifndef GET_TSC_CTL
105 # define GET_TSC_CTL(a)		(-EINVAL)
106 #endif
107 #ifndef SET_TSC_CTL
108 # define SET_TSC_CTL(a)		(-EINVAL)
109 #endif
110 #ifndef GET_FP_MODE
111 # define GET_FP_MODE(a)		(-EINVAL)
112 #endif
113 #ifndef SET_FP_MODE
114 # define SET_FP_MODE(a,b)	(-EINVAL)
115 #endif
116 #ifndef SVE_SET_VL
117 # define SVE_SET_VL(a)		(-EINVAL)
118 #endif
119 #ifndef SVE_GET_VL
120 # define SVE_GET_VL()		(-EINVAL)
121 #endif
122 #ifndef SME_SET_VL
123 # define SME_SET_VL(a)		(-EINVAL)
124 #endif
125 #ifndef SME_GET_VL
126 # define SME_GET_VL()		(-EINVAL)
127 #endif
128 #ifndef PAC_RESET_KEYS
129 # define PAC_RESET_KEYS(a, b)	(-EINVAL)
130 #endif
131 #ifndef PAC_SET_ENABLED_KEYS
132 # define PAC_SET_ENABLED_KEYS(a, b, c)	(-EINVAL)
133 #endif
134 #ifndef PAC_GET_ENABLED_KEYS
135 # define PAC_GET_ENABLED_KEYS(a)	(-EINVAL)
136 #endif
137 #ifndef SET_TAGGED_ADDR_CTRL
138 # define SET_TAGGED_ADDR_CTRL(a)	(-EINVAL)
139 #endif
140 #ifndef GET_TAGGED_ADDR_CTRL
141 # define GET_TAGGED_ADDR_CTRL()		(-EINVAL)
142 #endif
143 
144 /*
145  * this is where the system-wide overflow UID and GID are defined, for
146  * architectures that now have 32-bit UID/GID but didn't in the past
147  */
148 
149 int overflowuid = DEFAULT_OVERFLOWUID;
150 int overflowgid = DEFAULT_OVERFLOWGID;
151 
152 EXPORT_SYMBOL(overflowuid);
153 EXPORT_SYMBOL(overflowgid);
154 
155 /*
156  * the same as above, but for filesystems which can only store a 16-bit
157  * UID and GID. as such, this is needed on all architectures
158  */
159 
160 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
161 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
162 
163 EXPORT_SYMBOL(fs_overflowuid);
164 EXPORT_SYMBOL(fs_overflowgid);
165 
166 /*
167  * Returns true if current's euid is same as p's uid or euid,
168  * or has CAP_SYS_NICE to p's user_ns.
169  *
170  * Called with rcu_read_lock, creds are safe
171  */
172 static bool set_one_prio_perm(struct task_struct *p)
173 {
174 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
175 
176 	if (uid_eq(pcred->uid,  cred->euid) ||
177 	    uid_eq(pcred->euid, cred->euid))
178 		return true;
179 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
180 		return true;
181 	return false;
182 }
183 
184 /*
185  * set the priority of a task
186  * - the caller must hold the RCU read lock
187  */
188 static int set_one_prio(struct task_struct *p, int niceval, int error)
189 {
190 	int no_nice;
191 
192 	if (!set_one_prio_perm(p)) {
193 		error = -EPERM;
194 		goto out;
195 	}
196 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
197 		error = -EACCES;
198 		goto out;
199 	}
200 	no_nice = security_task_setnice(p, niceval);
201 	if (no_nice) {
202 		error = no_nice;
203 		goto out;
204 	}
205 	if (error == -ESRCH)
206 		error = 0;
207 	set_user_nice(p, niceval);
208 out:
209 	return error;
210 }
211 
212 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
213 {
214 	struct task_struct *g, *p;
215 	struct user_struct *user;
216 	const struct cred *cred = current_cred();
217 	int error = -EINVAL;
218 	struct pid *pgrp;
219 	kuid_t uid;
220 
221 	if (which > PRIO_USER || which < PRIO_PROCESS)
222 		goto out;
223 
224 	/* normalize: avoid signed division (rounding problems) */
225 	error = -ESRCH;
226 	if (niceval < MIN_NICE)
227 		niceval = MIN_NICE;
228 	if (niceval > MAX_NICE)
229 		niceval = MAX_NICE;
230 
231 	rcu_read_lock();
232 	switch (which) {
233 	case PRIO_PROCESS:
234 		if (who)
235 			p = find_task_by_vpid(who);
236 		else
237 			p = current;
238 		if (p)
239 			error = set_one_prio(p, niceval, error);
240 		break;
241 	case PRIO_PGRP:
242 		if (who)
243 			pgrp = find_vpid(who);
244 		else
245 			pgrp = task_pgrp(current);
246 		read_lock(&tasklist_lock);
247 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
248 			error = set_one_prio(p, niceval, error);
249 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
250 		read_unlock(&tasklist_lock);
251 		break;
252 	case PRIO_USER:
253 		uid = make_kuid(cred->user_ns, who);
254 		user = cred->user;
255 		if (!who)
256 			uid = cred->uid;
257 		else if (!uid_eq(uid, cred->uid)) {
258 			user = find_user(uid);
259 			if (!user)
260 				goto out_unlock;	/* No processes for this user */
261 		}
262 		for_each_process_thread(g, p) {
263 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
264 				error = set_one_prio(p, niceval, error);
265 		}
266 		if (!uid_eq(uid, cred->uid))
267 			free_uid(user);		/* For find_user() */
268 		break;
269 	}
270 out_unlock:
271 	rcu_read_unlock();
272 out:
273 	return error;
274 }
275 
276 /*
277  * Ugh. To avoid negative return values, "getpriority()" will
278  * not return the normal nice-value, but a negated value that
279  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
280  * to stay compatible.
281  */
282 SYSCALL_DEFINE2(getpriority, int, which, int, who)
283 {
284 	struct task_struct *g, *p;
285 	struct user_struct *user;
286 	const struct cred *cred = current_cred();
287 	long niceval, retval = -ESRCH;
288 	struct pid *pgrp;
289 	kuid_t uid;
290 
291 	if (which > PRIO_USER || which < PRIO_PROCESS)
292 		return -EINVAL;
293 
294 	rcu_read_lock();
295 	switch (which) {
296 	case PRIO_PROCESS:
297 		if (who)
298 			p = find_task_by_vpid(who);
299 		else
300 			p = current;
301 		if (p) {
302 			niceval = nice_to_rlimit(task_nice(p));
303 			if (niceval > retval)
304 				retval = niceval;
305 		}
306 		break;
307 	case PRIO_PGRP:
308 		if (who)
309 			pgrp = find_vpid(who);
310 		else
311 			pgrp = task_pgrp(current);
312 		read_lock(&tasklist_lock);
313 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
314 			niceval = nice_to_rlimit(task_nice(p));
315 			if (niceval > retval)
316 				retval = niceval;
317 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
318 		read_unlock(&tasklist_lock);
319 		break;
320 	case PRIO_USER:
321 		uid = make_kuid(cred->user_ns, who);
322 		user = cred->user;
323 		if (!who)
324 			uid = cred->uid;
325 		else if (!uid_eq(uid, cred->uid)) {
326 			user = find_user(uid);
327 			if (!user)
328 				goto out_unlock;	/* No processes for this user */
329 		}
330 		for_each_process_thread(g, p) {
331 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
332 				niceval = nice_to_rlimit(task_nice(p));
333 				if (niceval > retval)
334 					retval = niceval;
335 			}
336 		}
337 		if (!uid_eq(uid, cred->uid))
338 			free_uid(user);		/* for find_user() */
339 		break;
340 	}
341 out_unlock:
342 	rcu_read_unlock();
343 
344 	return retval;
345 }
346 
347 /*
348  * Unprivileged users may change the real gid to the effective gid
349  * or vice versa.  (BSD-style)
350  *
351  * If you set the real gid at all, or set the effective gid to a value not
352  * equal to the real gid, then the saved gid is set to the new effective gid.
353  *
354  * This makes it possible for a setgid program to completely drop its
355  * privileges, which is often a useful assertion to make when you are doing
356  * a security audit over a program.
357  *
358  * The general idea is that a program which uses just setregid() will be
359  * 100% compatible with BSD.  A program which uses just setgid() will be
360  * 100% compatible with POSIX with saved IDs.
361  *
362  * SMP: There are not races, the GIDs are checked only by filesystem
363  *      operations (as far as semantic preservation is concerned).
364  */
365 #ifdef CONFIG_MULTIUSER
366 long __sys_setregid(gid_t rgid, gid_t egid)
367 {
368 	struct user_namespace *ns = current_user_ns();
369 	const struct cred *old;
370 	struct cred *new;
371 	int retval;
372 	kgid_t krgid, kegid;
373 
374 	krgid = make_kgid(ns, rgid);
375 	kegid = make_kgid(ns, egid);
376 
377 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
378 		return -EINVAL;
379 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
380 		return -EINVAL;
381 
382 	new = prepare_creds();
383 	if (!new)
384 		return -ENOMEM;
385 	old = current_cred();
386 
387 	retval = -EPERM;
388 	if (rgid != (gid_t) -1) {
389 		if (gid_eq(old->gid, krgid) ||
390 		    gid_eq(old->egid, krgid) ||
391 		    ns_capable_setid(old->user_ns, CAP_SETGID))
392 			new->gid = krgid;
393 		else
394 			goto error;
395 	}
396 	if (egid != (gid_t) -1) {
397 		if (gid_eq(old->gid, kegid) ||
398 		    gid_eq(old->egid, kegid) ||
399 		    gid_eq(old->sgid, kegid) ||
400 		    ns_capable_setid(old->user_ns, CAP_SETGID))
401 			new->egid = kegid;
402 		else
403 			goto error;
404 	}
405 
406 	if (rgid != (gid_t) -1 ||
407 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
408 		new->sgid = new->egid;
409 	new->fsgid = new->egid;
410 
411 	retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
412 	if (retval < 0)
413 		goto error;
414 
415 	return commit_creds(new);
416 
417 error:
418 	abort_creds(new);
419 	return retval;
420 }
421 
422 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
423 {
424 	return __sys_setregid(rgid, egid);
425 }
426 
427 /*
428  * setgid() is implemented like SysV w/ SAVED_IDS
429  *
430  * SMP: Same implicit races as above.
431  */
432 long __sys_setgid(gid_t gid)
433 {
434 	struct user_namespace *ns = current_user_ns();
435 	const struct cred *old;
436 	struct cred *new;
437 	int retval;
438 	kgid_t kgid;
439 
440 	kgid = make_kgid(ns, gid);
441 	if (!gid_valid(kgid))
442 		return -EINVAL;
443 
444 	new = prepare_creds();
445 	if (!new)
446 		return -ENOMEM;
447 	old = current_cred();
448 
449 	retval = -EPERM;
450 	if (ns_capable_setid(old->user_ns, CAP_SETGID))
451 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
452 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
453 		new->egid = new->fsgid = kgid;
454 	else
455 		goto error;
456 
457 	retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
458 	if (retval < 0)
459 		goto error;
460 
461 	return commit_creds(new);
462 
463 error:
464 	abort_creds(new);
465 	return retval;
466 }
467 
468 SYSCALL_DEFINE1(setgid, gid_t, gid)
469 {
470 	return __sys_setgid(gid);
471 }
472 
473 /*
474  * change the user struct in a credentials set to match the new UID
475  */
476 static int set_user(struct cred *new)
477 {
478 	struct user_struct *new_user;
479 
480 	new_user = alloc_uid(new->uid);
481 	if (!new_user)
482 		return -EAGAIN;
483 
484 	free_uid(new->user);
485 	new->user = new_user;
486 	return 0;
487 }
488 
489 static void flag_nproc_exceeded(struct cred *new)
490 {
491 	if (new->ucounts == current_ucounts())
492 		return;
493 
494 	/*
495 	 * We don't fail in case of NPROC limit excess here because too many
496 	 * poorly written programs don't check set*uid() return code, assuming
497 	 * it never fails if called by root.  We may still enforce NPROC limit
498 	 * for programs doing set*uid()+execve() by harmlessly deferring the
499 	 * failure to the execve() stage.
500 	 */
501 	if (is_rlimit_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
502 			new->user != INIT_USER)
503 		current->flags |= PF_NPROC_EXCEEDED;
504 	else
505 		current->flags &= ~PF_NPROC_EXCEEDED;
506 }
507 
508 /*
509  * Unprivileged users may change the real uid to the effective uid
510  * or vice versa.  (BSD-style)
511  *
512  * If you set the real uid at all, or set the effective uid to a value not
513  * equal to the real uid, then the saved uid is set to the new effective uid.
514  *
515  * This makes it possible for a setuid program to completely drop its
516  * privileges, which is often a useful assertion to make when you are doing
517  * a security audit over a program.
518  *
519  * The general idea is that a program which uses just setreuid() will be
520  * 100% compatible with BSD.  A program which uses just setuid() will be
521  * 100% compatible with POSIX with saved IDs.
522  */
523 long __sys_setreuid(uid_t ruid, uid_t euid)
524 {
525 	struct user_namespace *ns = current_user_ns();
526 	const struct cred *old;
527 	struct cred *new;
528 	int retval;
529 	kuid_t kruid, keuid;
530 
531 	kruid = make_kuid(ns, ruid);
532 	keuid = make_kuid(ns, euid);
533 
534 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
535 		return -EINVAL;
536 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
537 		return -EINVAL;
538 
539 	new = prepare_creds();
540 	if (!new)
541 		return -ENOMEM;
542 	old = current_cred();
543 
544 	retval = -EPERM;
545 	if (ruid != (uid_t) -1) {
546 		new->uid = kruid;
547 		if (!uid_eq(old->uid, kruid) &&
548 		    !uid_eq(old->euid, kruid) &&
549 		    !ns_capable_setid(old->user_ns, CAP_SETUID))
550 			goto error;
551 	}
552 
553 	if (euid != (uid_t) -1) {
554 		new->euid = keuid;
555 		if (!uid_eq(old->uid, keuid) &&
556 		    !uid_eq(old->euid, keuid) &&
557 		    !uid_eq(old->suid, keuid) &&
558 		    !ns_capable_setid(old->user_ns, CAP_SETUID))
559 			goto error;
560 	}
561 
562 	if (!uid_eq(new->uid, old->uid)) {
563 		retval = set_user(new);
564 		if (retval < 0)
565 			goto error;
566 	}
567 	if (ruid != (uid_t) -1 ||
568 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
569 		new->suid = new->euid;
570 	new->fsuid = new->euid;
571 
572 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
573 	if (retval < 0)
574 		goto error;
575 
576 	retval = set_cred_ucounts(new);
577 	if (retval < 0)
578 		goto error;
579 
580 	flag_nproc_exceeded(new);
581 	return commit_creds(new);
582 
583 error:
584 	abort_creds(new);
585 	return retval;
586 }
587 
588 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
589 {
590 	return __sys_setreuid(ruid, euid);
591 }
592 
593 /*
594  * setuid() is implemented like SysV with SAVED_IDS
595  *
596  * Note that SAVED_ID's is deficient in that a setuid root program
597  * like sendmail, for example, cannot set its uid to be a normal
598  * user and then switch back, because if you're root, setuid() sets
599  * the saved uid too.  If you don't like this, blame the bright people
600  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
601  * will allow a root program to temporarily drop privileges and be able to
602  * regain them by swapping the real and effective uid.
603  */
604 long __sys_setuid(uid_t uid)
605 {
606 	struct user_namespace *ns = current_user_ns();
607 	const struct cred *old;
608 	struct cred *new;
609 	int retval;
610 	kuid_t kuid;
611 
612 	kuid = make_kuid(ns, uid);
613 	if (!uid_valid(kuid))
614 		return -EINVAL;
615 
616 	new = prepare_creds();
617 	if (!new)
618 		return -ENOMEM;
619 	old = current_cred();
620 
621 	retval = -EPERM;
622 	if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
623 		new->suid = new->uid = kuid;
624 		if (!uid_eq(kuid, old->uid)) {
625 			retval = set_user(new);
626 			if (retval < 0)
627 				goto error;
628 		}
629 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
630 		goto error;
631 	}
632 
633 	new->fsuid = new->euid = kuid;
634 
635 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
636 	if (retval < 0)
637 		goto error;
638 
639 	retval = set_cred_ucounts(new);
640 	if (retval < 0)
641 		goto error;
642 
643 	flag_nproc_exceeded(new);
644 	return commit_creds(new);
645 
646 error:
647 	abort_creds(new);
648 	return retval;
649 }
650 
651 SYSCALL_DEFINE1(setuid, uid_t, uid)
652 {
653 	return __sys_setuid(uid);
654 }
655 
656 
657 /*
658  * This function implements a generic ability to update ruid, euid,
659  * and suid.  This allows you to implement the 4.4 compatible seteuid().
660  */
661 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
662 {
663 	struct user_namespace *ns = current_user_ns();
664 	const struct cred *old;
665 	struct cred *new;
666 	int retval;
667 	kuid_t kruid, keuid, ksuid;
668 	bool ruid_new, euid_new, suid_new;
669 
670 	kruid = make_kuid(ns, ruid);
671 	keuid = make_kuid(ns, euid);
672 	ksuid = make_kuid(ns, suid);
673 
674 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
675 		return -EINVAL;
676 
677 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
678 		return -EINVAL;
679 
680 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
681 		return -EINVAL;
682 
683 	old = current_cred();
684 
685 	/* check for no-op */
686 	if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
687 	    (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
688 				    uid_eq(keuid, old->fsuid))) &&
689 	    (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
690 		return 0;
691 
692 	ruid_new = ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
693 		   !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
694 	euid_new = euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
695 		   !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
696 	suid_new = suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
697 		   !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
698 	if ((ruid_new || euid_new || suid_new) &&
699 	    !ns_capable_setid(old->user_ns, CAP_SETUID))
700 		return -EPERM;
701 
702 	new = prepare_creds();
703 	if (!new)
704 		return -ENOMEM;
705 
706 	if (ruid != (uid_t) -1) {
707 		new->uid = kruid;
708 		if (!uid_eq(kruid, old->uid)) {
709 			retval = set_user(new);
710 			if (retval < 0)
711 				goto error;
712 		}
713 	}
714 	if (euid != (uid_t) -1)
715 		new->euid = keuid;
716 	if (suid != (uid_t) -1)
717 		new->suid = ksuid;
718 	new->fsuid = new->euid;
719 
720 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
721 	if (retval < 0)
722 		goto error;
723 
724 	retval = set_cred_ucounts(new);
725 	if (retval < 0)
726 		goto error;
727 
728 	flag_nproc_exceeded(new);
729 	return commit_creds(new);
730 
731 error:
732 	abort_creds(new);
733 	return retval;
734 }
735 
736 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
737 {
738 	return __sys_setresuid(ruid, euid, suid);
739 }
740 
741 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
742 {
743 	const struct cred *cred = current_cred();
744 	int retval;
745 	uid_t ruid, euid, suid;
746 
747 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
748 	euid = from_kuid_munged(cred->user_ns, cred->euid);
749 	suid = from_kuid_munged(cred->user_ns, cred->suid);
750 
751 	retval = put_user(ruid, ruidp);
752 	if (!retval) {
753 		retval = put_user(euid, euidp);
754 		if (!retval)
755 			return put_user(suid, suidp);
756 	}
757 	return retval;
758 }
759 
760 /*
761  * Same as above, but for rgid, egid, sgid.
762  */
763 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
764 {
765 	struct user_namespace *ns = current_user_ns();
766 	const struct cred *old;
767 	struct cred *new;
768 	int retval;
769 	kgid_t krgid, kegid, ksgid;
770 	bool rgid_new, egid_new, sgid_new;
771 
772 	krgid = make_kgid(ns, rgid);
773 	kegid = make_kgid(ns, egid);
774 	ksgid = make_kgid(ns, sgid);
775 
776 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
777 		return -EINVAL;
778 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
779 		return -EINVAL;
780 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
781 		return -EINVAL;
782 
783 	old = current_cred();
784 
785 	/* check for no-op */
786 	if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
787 	    (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
788 				    gid_eq(kegid, old->fsgid))) &&
789 	    (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
790 		return 0;
791 
792 	rgid_new = rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
793 		   !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
794 	egid_new = egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
795 		   !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
796 	sgid_new = sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
797 		   !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
798 	if ((rgid_new || egid_new || sgid_new) &&
799 	    !ns_capable_setid(old->user_ns, CAP_SETGID))
800 		return -EPERM;
801 
802 	new = prepare_creds();
803 	if (!new)
804 		return -ENOMEM;
805 
806 	if (rgid != (gid_t) -1)
807 		new->gid = krgid;
808 	if (egid != (gid_t) -1)
809 		new->egid = kegid;
810 	if (sgid != (gid_t) -1)
811 		new->sgid = ksgid;
812 	new->fsgid = new->egid;
813 
814 	retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
815 	if (retval < 0)
816 		goto error;
817 
818 	return commit_creds(new);
819 
820 error:
821 	abort_creds(new);
822 	return retval;
823 }
824 
825 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
826 {
827 	return __sys_setresgid(rgid, egid, sgid);
828 }
829 
830 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
831 {
832 	const struct cred *cred = current_cred();
833 	int retval;
834 	gid_t rgid, egid, sgid;
835 
836 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
837 	egid = from_kgid_munged(cred->user_ns, cred->egid);
838 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
839 
840 	retval = put_user(rgid, rgidp);
841 	if (!retval) {
842 		retval = put_user(egid, egidp);
843 		if (!retval)
844 			retval = put_user(sgid, sgidp);
845 	}
846 
847 	return retval;
848 }
849 
850 
851 /*
852  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
853  * is used for "access()" and for the NFS daemon (letting nfsd stay at
854  * whatever uid it wants to). It normally shadows "euid", except when
855  * explicitly set by setfsuid() or for access..
856  */
857 long __sys_setfsuid(uid_t uid)
858 {
859 	const struct cred *old;
860 	struct cred *new;
861 	uid_t old_fsuid;
862 	kuid_t kuid;
863 
864 	old = current_cred();
865 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
866 
867 	kuid = make_kuid(old->user_ns, uid);
868 	if (!uid_valid(kuid))
869 		return old_fsuid;
870 
871 	new = prepare_creds();
872 	if (!new)
873 		return old_fsuid;
874 
875 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
876 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
877 	    ns_capable_setid(old->user_ns, CAP_SETUID)) {
878 		if (!uid_eq(kuid, old->fsuid)) {
879 			new->fsuid = kuid;
880 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
881 				goto change_okay;
882 		}
883 	}
884 
885 	abort_creds(new);
886 	return old_fsuid;
887 
888 change_okay:
889 	commit_creds(new);
890 	return old_fsuid;
891 }
892 
893 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
894 {
895 	return __sys_setfsuid(uid);
896 }
897 
898 /*
899  * Samma på svenska..
900  */
901 long __sys_setfsgid(gid_t gid)
902 {
903 	const struct cred *old;
904 	struct cred *new;
905 	gid_t old_fsgid;
906 	kgid_t kgid;
907 
908 	old = current_cred();
909 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
910 
911 	kgid = make_kgid(old->user_ns, gid);
912 	if (!gid_valid(kgid))
913 		return old_fsgid;
914 
915 	new = prepare_creds();
916 	if (!new)
917 		return old_fsgid;
918 
919 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
920 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
921 	    ns_capable_setid(old->user_ns, CAP_SETGID)) {
922 		if (!gid_eq(kgid, old->fsgid)) {
923 			new->fsgid = kgid;
924 			if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
925 				goto change_okay;
926 		}
927 	}
928 
929 	abort_creds(new);
930 	return old_fsgid;
931 
932 change_okay:
933 	commit_creds(new);
934 	return old_fsgid;
935 }
936 
937 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
938 {
939 	return __sys_setfsgid(gid);
940 }
941 #endif /* CONFIG_MULTIUSER */
942 
943 /**
944  * sys_getpid - return the thread group id of the current process
945  *
946  * Note, despite the name, this returns the tgid not the pid.  The tgid and
947  * the pid are identical unless CLONE_THREAD was specified on clone() in
948  * which case the tgid is the same in all threads of the same group.
949  *
950  * This is SMP safe as current->tgid does not change.
951  */
952 SYSCALL_DEFINE0(getpid)
953 {
954 	return task_tgid_vnr(current);
955 }
956 
957 /* Thread ID - the internal kernel "pid" */
958 SYSCALL_DEFINE0(gettid)
959 {
960 	return task_pid_vnr(current);
961 }
962 
963 /*
964  * Accessing ->real_parent is not SMP-safe, it could
965  * change from under us. However, we can use a stale
966  * value of ->real_parent under rcu_read_lock(), see
967  * release_task()->call_rcu(delayed_put_task_struct).
968  */
969 SYSCALL_DEFINE0(getppid)
970 {
971 	int pid;
972 
973 	rcu_read_lock();
974 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
975 	rcu_read_unlock();
976 
977 	return pid;
978 }
979 
980 SYSCALL_DEFINE0(getuid)
981 {
982 	/* Only we change this so SMP safe */
983 	return from_kuid_munged(current_user_ns(), current_uid());
984 }
985 
986 SYSCALL_DEFINE0(geteuid)
987 {
988 	/* Only we change this so SMP safe */
989 	return from_kuid_munged(current_user_ns(), current_euid());
990 }
991 
992 SYSCALL_DEFINE0(getgid)
993 {
994 	/* Only we change this so SMP safe */
995 	return from_kgid_munged(current_user_ns(), current_gid());
996 }
997 
998 SYSCALL_DEFINE0(getegid)
999 {
1000 	/* Only we change this so SMP safe */
1001 	return from_kgid_munged(current_user_ns(), current_egid());
1002 }
1003 
1004 static void do_sys_times(struct tms *tms)
1005 {
1006 	u64 tgutime, tgstime, cutime, cstime;
1007 
1008 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1009 	cutime = current->signal->cutime;
1010 	cstime = current->signal->cstime;
1011 	tms->tms_utime = nsec_to_clock_t(tgutime);
1012 	tms->tms_stime = nsec_to_clock_t(tgstime);
1013 	tms->tms_cutime = nsec_to_clock_t(cutime);
1014 	tms->tms_cstime = nsec_to_clock_t(cstime);
1015 }
1016 
1017 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1018 {
1019 	if (tbuf) {
1020 		struct tms tmp;
1021 
1022 		do_sys_times(&tmp);
1023 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1024 			return -EFAULT;
1025 	}
1026 	force_successful_syscall_return();
1027 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
1028 }
1029 
1030 #ifdef CONFIG_COMPAT
1031 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1032 {
1033 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1034 }
1035 
1036 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1037 {
1038 	if (tbuf) {
1039 		struct tms tms;
1040 		struct compat_tms tmp;
1041 
1042 		do_sys_times(&tms);
1043 		/* Convert our struct tms to the compat version. */
1044 		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1045 		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1046 		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1047 		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1048 		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1049 			return -EFAULT;
1050 	}
1051 	force_successful_syscall_return();
1052 	return compat_jiffies_to_clock_t(jiffies);
1053 }
1054 #endif
1055 
1056 /*
1057  * This needs some heavy checking ...
1058  * I just haven't the stomach for it. I also don't fully
1059  * understand sessions/pgrp etc. Let somebody who does explain it.
1060  *
1061  * OK, I think I have the protection semantics right.... this is really
1062  * only important on a multi-user system anyway, to make sure one user
1063  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1064  *
1065  * !PF_FORKNOEXEC check to conform completely to POSIX.
1066  */
1067 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1068 {
1069 	struct task_struct *p;
1070 	struct task_struct *group_leader = current->group_leader;
1071 	struct pid *pgrp;
1072 	int err;
1073 
1074 	if (!pid)
1075 		pid = task_pid_vnr(group_leader);
1076 	if (!pgid)
1077 		pgid = pid;
1078 	if (pgid < 0)
1079 		return -EINVAL;
1080 	rcu_read_lock();
1081 
1082 	/* From this point forward we keep holding onto the tasklist lock
1083 	 * so that our parent does not change from under us. -DaveM
1084 	 */
1085 	write_lock_irq(&tasklist_lock);
1086 
1087 	err = -ESRCH;
1088 	p = find_task_by_vpid(pid);
1089 	if (!p)
1090 		goto out;
1091 
1092 	err = -EINVAL;
1093 	if (!thread_group_leader(p))
1094 		goto out;
1095 
1096 	if (same_thread_group(p->real_parent, group_leader)) {
1097 		err = -EPERM;
1098 		if (task_session(p) != task_session(group_leader))
1099 			goto out;
1100 		err = -EACCES;
1101 		if (!(p->flags & PF_FORKNOEXEC))
1102 			goto out;
1103 	} else {
1104 		err = -ESRCH;
1105 		if (p != group_leader)
1106 			goto out;
1107 	}
1108 
1109 	err = -EPERM;
1110 	if (p->signal->leader)
1111 		goto out;
1112 
1113 	pgrp = task_pid(p);
1114 	if (pgid != pid) {
1115 		struct task_struct *g;
1116 
1117 		pgrp = find_vpid(pgid);
1118 		g = pid_task(pgrp, PIDTYPE_PGID);
1119 		if (!g || task_session(g) != task_session(group_leader))
1120 			goto out;
1121 	}
1122 
1123 	err = security_task_setpgid(p, pgid);
1124 	if (err)
1125 		goto out;
1126 
1127 	if (task_pgrp(p) != pgrp)
1128 		change_pid(p, PIDTYPE_PGID, pgrp);
1129 
1130 	err = 0;
1131 out:
1132 	/* All paths lead to here, thus we are safe. -DaveM */
1133 	write_unlock_irq(&tasklist_lock);
1134 	rcu_read_unlock();
1135 	return err;
1136 }
1137 
1138 static int do_getpgid(pid_t pid)
1139 {
1140 	struct task_struct *p;
1141 	struct pid *grp;
1142 	int retval;
1143 
1144 	rcu_read_lock();
1145 	if (!pid)
1146 		grp = task_pgrp(current);
1147 	else {
1148 		retval = -ESRCH;
1149 		p = find_task_by_vpid(pid);
1150 		if (!p)
1151 			goto out;
1152 		grp = task_pgrp(p);
1153 		if (!grp)
1154 			goto out;
1155 
1156 		retval = security_task_getpgid(p);
1157 		if (retval)
1158 			goto out;
1159 	}
1160 	retval = pid_vnr(grp);
1161 out:
1162 	rcu_read_unlock();
1163 	return retval;
1164 }
1165 
1166 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1167 {
1168 	return do_getpgid(pid);
1169 }
1170 
1171 #ifdef __ARCH_WANT_SYS_GETPGRP
1172 
1173 SYSCALL_DEFINE0(getpgrp)
1174 {
1175 	return do_getpgid(0);
1176 }
1177 
1178 #endif
1179 
1180 SYSCALL_DEFINE1(getsid, pid_t, pid)
1181 {
1182 	struct task_struct *p;
1183 	struct pid *sid;
1184 	int retval;
1185 
1186 	rcu_read_lock();
1187 	if (!pid)
1188 		sid = task_session(current);
1189 	else {
1190 		retval = -ESRCH;
1191 		p = find_task_by_vpid(pid);
1192 		if (!p)
1193 			goto out;
1194 		sid = task_session(p);
1195 		if (!sid)
1196 			goto out;
1197 
1198 		retval = security_task_getsid(p);
1199 		if (retval)
1200 			goto out;
1201 	}
1202 	retval = pid_vnr(sid);
1203 out:
1204 	rcu_read_unlock();
1205 	return retval;
1206 }
1207 
1208 static void set_special_pids(struct pid *pid)
1209 {
1210 	struct task_struct *curr = current->group_leader;
1211 
1212 	if (task_session(curr) != pid)
1213 		change_pid(curr, PIDTYPE_SID, pid);
1214 
1215 	if (task_pgrp(curr) != pid)
1216 		change_pid(curr, PIDTYPE_PGID, pid);
1217 }
1218 
1219 int ksys_setsid(void)
1220 {
1221 	struct task_struct *group_leader = current->group_leader;
1222 	struct pid *sid = task_pid(group_leader);
1223 	pid_t session = pid_vnr(sid);
1224 	int err = -EPERM;
1225 
1226 	write_lock_irq(&tasklist_lock);
1227 	/* Fail if I am already a session leader */
1228 	if (group_leader->signal->leader)
1229 		goto out;
1230 
1231 	/* Fail if a process group id already exists that equals the
1232 	 * proposed session id.
1233 	 */
1234 	if (pid_task(sid, PIDTYPE_PGID))
1235 		goto out;
1236 
1237 	group_leader->signal->leader = 1;
1238 	set_special_pids(sid);
1239 
1240 	proc_clear_tty(group_leader);
1241 
1242 	err = session;
1243 out:
1244 	write_unlock_irq(&tasklist_lock);
1245 	if (err > 0) {
1246 		proc_sid_connector(group_leader);
1247 		sched_autogroup_create_attach(group_leader);
1248 	}
1249 	return err;
1250 }
1251 
1252 SYSCALL_DEFINE0(setsid)
1253 {
1254 	return ksys_setsid();
1255 }
1256 
1257 DECLARE_RWSEM(uts_sem);
1258 
1259 #ifdef COMPAT_UTS_MACHINE
1260 #define override_architecture(name) \
1261 	(personality(current->personality) == PER_LINUX32 && \
1262 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1263 		      sizeof(COMPAT_UTS_MACHINE)))
1264 #else
1265 #define override_architecture(name)	0
1266 #endif
1267 
1268 /*
1269  * Work around broken programs that cannot handle "Linux 3.0".
1270  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1271  * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1272  * 2.6.60.
1273  */
1274 static int override_release(char __user *release, size_t len)
1275 {
1276 	int ret = 0;
1277 
1278 	if (current->personality & UNAME26) {
1279 		const char *rest = UTS_RELEASE;
1280 		char buf[65] = { 0 };
1281 		int ndots = 0;
1282 		unsigned v;
1283 		size_t copy;
1284 
1285 		while (*rest) {
1286 			if (*rest == '.' && ++ndots >= 3)
1287 				break;
1288 			if (!isdigit(*rest) && *rest != '.')
1289 				break;
1290 			rest++;
1291 		}
1292 		v = LINUX_VERSION_PATCHLEVEL + 60;
1293 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1294 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1295 		ret = copy_to_user(release, buf, copy + 1);
1296 	}
1297 	return ret;
1298 }
1299 
1300 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1301 {
1302 	struct new_utsname tmp;
1303 
1304 	down_read(&uts_sem);
1305 	memcpy(&tmp, utsname(), sizeof(tmp));
1306 	up_read(&uts_sem);
1307 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1308 		return -EFAULT;
1309 
1310 	if (override_release(name->release, sizeof(name->release)))
1311 		return -EFAULT;
1312 	if (override_architecture(name))
1313 		return -EFAULT;
1314 	return 0;
1315 }
1316 
1317 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1318 /*
1319  * Old cruft
1320  */
1321 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1322 {
1323 	struct old_utsname tmp;
1324 
1325 	if (!name)
1326 		return -EFAULT;
1327 
1328 	down_read(&uts_sem);
1329 	memcpy(&tmp, utsname(), sizeof(tmp));
1330 	up_read(&uts_sem);
1331 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1332 		return -EFAULT;
1333 
1334 	if (override_release(name->release, sizeof(name->release)))
1335 		return -EFAULT;
1336 	if (override_architecture(name))
1337 		return -EFAULT;
1338 	return 0;
1339 }
1340 
1341 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1342 {
1343 	struct oldold_utsname tmp;
1344 
1345 	if (!name)
1346 		return -EFAULT;
1347 
1348 	memset(&tmp, 0, sizeof(tmp));
1349 
1350 	down_read(&uts_sem);
1351 	memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1352 	memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1353 	memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1354 	memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1355 	memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1356 	up_read(&uts_sem);
1357 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1358 		return -EFAULT;
1359 
1360 	if (override_architecture(name))
1361 		return -EFAULT;
1362 	if (override_release(name->release, sizeof(name->release)))
1363 		return -EFAULT;
1364 	return 0;
1365 }
1366 #endif
1367 
1368 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1369 {
1370 	int errno;
1371 	char tmp[__NEW_UTS_LEN];
1372 
1373 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1374 		return -EPERM;
1375 
1376 	if (len < 0 || len > __NEW_UTS_LEN)
1377 		return -EINVAL;
1378 	errno = -EFAULT;
1379 	if (!copy_from_user(tmp, name, len)) {
1380 		struct new_utsname *u;
1381 
1382 		add_device_randomness(tmp, len);
1383 		down_write(&uts_sem);
1384 		u = utsname();
1385 		memcpy(u->nodename, tmp, len);
1386 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1387 		errno = 0;
1388 		uts_proc_notify(UTS_PROC_HOSTNAME);
1389 		up_write(&uts_sem);
1390 	}
1391 	return errno;
1392 }
1393 
1394 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1395 
1396 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1397 {
1398 	int i;
1399 	struct new_utsname *u;
1400 	char tmp[__NEW_UTS_LEN + 1];
1401 
1402 	if (len < 0)
1403 		return -EINVAL;
1404 	down_read(&uts_sem);
1405 	u = utsname();
1406 	i = 1 + strlen(u->nodename);
1407 	if (i > len)
1408 		i = len;
1409 	memcpy(tmp, u->nodename, i);
1410 	up_read(&uts_sem);
1411 	if (copy_to_user(name, tmp, i))
1412 		return -EFAULT;
1413 	return 0;
1414 }
1415 
1416 #endif
1417 
1418 /*
1419  * Only setdomainname; getdomainname can be implemented by calling
1420  * uname()
1421  */
1422 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1423 {
1424 	int errno;
1425 	char tmp[__NEW_UTS_LEN];
1426 
1427 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1428 		return -EPERM;
1429 	if (len < 0 || len > __NEW_UTS_LEN)
1430 		return -EINVAL;
1431 
1432 	errno = -EFAULT;
1433 	if (!copy_from_user(tmp, name, len)) {
1434 		struct new_utsname *u;
1435 
1436 		add_device_randomness(tmp, len);
1437 		down_write(&uts_sem);
1438 		u = utsname();
1439 		memcpy(u->domainname, tmp, len);
1440 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1441 		errno = 0;
1442 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1443 		up_write(&uts_sem);
1444 	}
1445 	return errno;
1446 }
1447 
1448 /* make sure you are allowed to change @tsk limits before calling this */
1449 static int do_prlimit(struct task_struct *tsk, unsigned int resource,
1450 		      struct rlimit *new_rlim, struct rlimit *old_rlim)
1451 {
1452 	struct rlimit *rlim;
1453 	int retval = 0;
1454 
1455 	if (resource >= RLIM_NLIMITS)
1456 		return -EINVAL;
1457 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1458 
1459 	if (new_rlim) {
1460 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1461 			return -EINVAL;
1462 		if (resource == RLIMIT_NOFILE &&
1463 				new_rlim->rlim_max > sysctl_nr_open)
1464 			return -EPERM;
1465 	}
1466 
1467 	/* Holding a refcount on tsk protects tsk->signal from disappearing. */
1468 	rlim = tsk->signal->rlim + resource;
1469 	task_lock(tsk->group_leader);
1470 	if (new_rlim) {
1471 		/*
1472 		 * Keep the capable check against init_user_ns until cgroups can
1473 		 * contain all limits.
1474 		 */
1475 		if (new_rlim->rlim_max > rlim->rlim_max &&
1476 				!capable(CAP_SYS_RESOURCE))
1477 			retval = -EPERM;
1478 		if (!retval)
1479 			retval = security_task_setrlimit(tsk, resource, new_rlim);
1480 	}
1481 	if (!retval) {
1482 		if (old_rlim)
1483 			*old_rlim = *rlim;
1484 		if (new_rlim)
1485 			*rlim = *new_rlim;
1486 	}
1487 	task_unlock(tsk->group_leader);
1488 
1489 	/*
1490 	 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1491 	 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1492 	 * ignores the rlimit.
1493 	 */
1494 	if (!retval && new_rlim && resource == RLIMIT_CPU &&
1495 	    new_rlim->rlim_cur != RLIM_INFINITY &&
1496 	    IS_ENABLED(CONFIG_POSIX_TIMERS)) {
1497 		/*
1498 		 * update_rlimit_cpu can fail if the task is exiting, but there
1499 		 * may be other tasks in the thread group that are not exiting,
1500 		 * and they need their cpu timers adjusted.
1501 		 *
1502 		 * The group_leader is the last task to be released, so if we
1503 		 * cannot update_rlimit_cpu on it, then the entire process is
1504 		 * exiting and we do not need to update at all.
1505 		 */
1506 		update_rlimit_cpu(tsk->group_leader, new_rlim->rlim_cur);
1507 	}
1508 
1509 	return retval;
1510 }
1511 
1512 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1513 {
1514 	struct rlimit value;
1515 	int ret;
1516 
1517 	ret = do_prlimit(current, resource, NULL, &value);
1518 	if (!ret)
1519 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1520 
1521 	return ret;
1522 }
1523 
1524 #ifdef CONFIG_COMPAT
1525 
1526 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1527 		       struct compat_rlimit __user *, rlim)
1528 {
1529 	struct rlimit r;
1530 	struct compat_rlimit r32;
1531 
1532 	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1533 		return -EFAULT;
1534 
1535 	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1536 		r.rlim_cur = RLIM_INFINITY;
1537 	else
1538 		r.rlim_cur = r32.rlim_cur;
1539 	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1540 		r.rlim_max = RLIM_INFINITY;
1541 	else
1542 		r.rlim_max = r32.rlim_max;
1543 	return do_prlimit(current, resource, &r, NULL);
1544 }
1545 
1546 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1547 		       struct compat_rlimit __user *, rlim)
1548 {
1549 	struct rlimit r;
1550 	int ret;
1551 
1552 	ret = do_prlimit(current, resource, NULL, &r);
1553 	if (!ret) {
1554 		struct compat_rlimit r32;
1555 		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1556 			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1557 		else
1558 			r32.rlim_cur = r.rlim_cur;
1559 		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1560 			r32.rlim_max = COMPAT_RLIM_INFINITY;
1561 		else
1562 			r32.rlim_max = r.rlim_max;
1563 
1564 		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1565 			return -EFAULT;
1566 	}
1567 	return ret;
1568 }
1569 
1570 #endif
1571 
1572 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1573 
1574 /*
1575  *	Back compatibility for getrlimit. Needed for some apps.
1576  */
1577 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1578 		struct rlimit __user *, rlim)
1579 {
1580 	struct rlimit x;
1581 	if (resource >= RLIM_NLIMITS)
1582 		return -EINVAL;
1583 
1584 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1585 	task_lock(current->group_leader);
1586 	x = current->signal->rlim[resource];
1587 	task_unlock(current->group_leader);
1588 	if (x.rlim_cur > 0x7FFFFFFF)
1589 		x.rlim_cur = 0x7FFFFFFF;
1590 	if (x.rlim_max > 0x7FFFFFFF)
1591 		x.rlim_max = 0x7FFFFFFF;
1592 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1593 }
1594 
1595 #ifdef CONFIG_COMPAT
1596 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1597 		       struct compat_rlimit __user *, rlim)
1598 {
1599 	struct rlimit r;
1600 
1601 	if (resource >= RLIM_NLIMITS)
1602 		return -EINVAL;
1603 
1604 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1605 	task_lock(current->group_leader);
1606 	r = current->signal->rlim[resource];
1607 	task_unlock(current->group_leader);
1608 	if (r.rlim_cur > 0x7FFFFFFF)
1609 		r.rlim_cur = 0x7FFFFFFF;
1610 	if (r.rlim_max > 0x7FFFFFFF)
1611 		r.rlim_max = 0x7FFFFFFF;
1612 
1613 	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1614 	    put_user(r.rlim_max, &rlim->rlim_max))
1615 		return -EFAULT;
1616 	return 0;
1617 }
1618 #endif
1619 
1620 #endif
1621 
1622 static inline bool rlim64_is_infinity(__u64 rlim64)
1623 {
1624 #if BITS_PER_LONG < 64
1625 	return rlim64 >= ULONG_MAX;
1626 #else
1627 	return rlim64 == RLIM64_INFINITY;
1628 #endif
1629 }
1630 
1631 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1632 {
1633 	if (rlim->rlim_cur == RLIM_INFINITY)
1634 		rlim64->rlim_cur = RLIM64_INFINITY;
1635 	else
1636 		rlim64->rlim_cur = rlim->rlim_cur;
1637 	if (rlim->rlim_max == RLIM_INFINITY)
1638 		rlim64->rlim_max = RLIM64_INFINITY;
1639 	else
1640 		rlim64->rlim_max = rlim->rlim_max;
1641 }
1642 
1643 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1644 {
1645 	if (rlim64_is_infinity(rlim64->rlim_cur))
1646 		rlim->rlim_cur = RLIM_INFINITY;
1647 	else
1648 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1649 	if (rlim64_is_infinity(rlim64->rlim_max))
1650 		rlim->rlim_max = RLIM_INFINITY;
1651 	else
1652 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1653 }
1654 
1655 /* rcu lock must be held */
1656 static int check_prlimit_permission(struct task_struct *task,
1657 				    unsigned int flags)
1658 {
1659 	const struct cred *cred = current_cred(), *tcred;
1660 	bool id_match;
1661 
1662 	if (current == task)
1663 		return 0;
1664 
1665 	tcred = __task_cred(task);
1666 	id_match = (uid_eq(cred->uid, tcred->euid) &&
1667 		    uid_eq(cred->uid, tcred->suid) &&
1668 		    uid_eq(cred->uid, tcred->uid)  &&
1669 		    gid_eq(cred->gid, tcred->egid) &&
1670 		    gid_eq(cred->gid, tcred->sgid) &&
1671 		    gid_eq(cred->gid, tcred->gid));
1672 	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1673 		return -EPERM;
1674 
1675 	return security_task_prlimit(cred, tcred, flags);
1676 }
1677 
1678 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1679 		const struct rlimit64 __user *, new_rlim,
1680 		struct rlimit64 __user *, old_rlim)
1681 {
1682 	struct rlimit64 old64, new64;
1683 	struct rlimit old, new;
1684 	struct task_struct *tsk;
1685 	unsigned int checkflags = 0;
1686 	int ret;
1687 
1688 	if (old_rlim)
1689 		checkflags |= LSM_PRLIMIT_READ;
1690 
1691 	if (new_rlim) {
1692 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1693 			return -EFAULT;
1694 		rlim64_to_rlim(&new64, &new);
1695 		checkflags |= LSM_PRLIMIT_WRITE;
1696 	}
1697 
1698 	rcu_read_lock();
1699 	tsk = pid ? find_task_by_vpid(pid) : current;
1700 	if (!tsk) {
1701 		rcu_read_unlock();
1702 		return -ESRCH;
1703 	}
1704 	ret = check_prlimit_permission(tsk, checkflags);
1705 	if (ret) {
1706 		rcu_read_unlock();
1707 		return ret;
1708 	}
1709 	get_task_struct(tsk);
1710 	rcu_read_unlock();
1711 
1712 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1713 			old_rlim ? &old : NULL);
1714 
1715 	if (!ret && old_rlim) {
1716 		rlim_to_rlim64(&old, &old64);
1717 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1718 			ret = -EFAULT;
1719 	}
1720 
1721 	put_task_struct(tsk);
1722 	return ret;
1723 }
1724 
1725 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1726 {
1727 	struct rlimit new_rlim;
1728 
1729 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1730 		return -EFAULT;
1731 	return do_prlimit(current, resource, &new_rlim, NULL);
1732 }
1733 
1734 /*
1735  * It would make sense to put struct rusage in the task_struct,
1736  * except that would make the task_struct be *really big*.  After
1737  * task_struct gets moved into malloc'ed memory, it would
1738  * make sense to do this.  It will make moving the rest of the information
1739  * a lot simpler!  (Which we're not doing right now because we're not
1740  * measuring them yet).
1741  *
1742  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1743  * races with threads incrementing their own counters.  But since word
1744  * reads are atomic, we either get new values or old values and we don't
1745  * care which for the sums.  We always take the siglock to protect reading
1746  * the c* fields from p->signal from races with exit.c updating those
1747  * fields when reaping, so a sample either gets all the additions of a
1748  * given child after it's reaped, or none so this sample is before reaping.
1749  *
1750  * Locking:
1751  * We need to take the siglock for CHILDEREN, SELF and BOTH
1752  * for  the cases current multithreaded, non-current single threaded
1753  * non-current multithreaded.  Thread traversal is now safe with
1754  * the siglock held.
1755  * Strictly speaking, we donot need to take the siglock if we are current and
1756  * single threaded,  as no one else can take our signal_struct away, no one
1757  * else can  reap the  children to update signal->c* counters, and no one else
1758  * can race with the signal-> fields. If we do not take any lock, the
1759  * signal-> fields could be read out of order while another thread was just
1760  * exiting. So we should  place a read memory barrier when we avoid the lock.
1761  * On the writer side,  write memory barrier is implied in  __exit_signal
1762  * as __exit_signal releases  the siglock spinlock after updating the signal->
1763  * fields. But we don't do this yet to keep things simple.
1764  *
1765  */
1766 
1767 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1768 {
1769 	r->ru_nvcsw += t->nvcsw;
1770 	r->ru_nivcsw += t->nivcsw;
1771 	r->ru_minflt += t->min_flt;
1772 	r->ru_majflt += t->maj_flt;
1773 	r->ru_inblock += task_io_get_inblock(t);
1774 	r->ru_oublock += task_io_get_oublock(t);
1775 }
1776 
1777 void getrusage(struct task_struct *p, int who, struct rusage *r)
1778 {
1779 	struct task_struct *t;
1780 	unsigned long flags;
1781 	u64 tgutime, tgstime, utime, stime;
1782 	unsigned long maxrss = 0;
1783 
1784 	memset((char *)r, 0, sizeof (*r));
1785 	utime = stime = 0;
1786 
1787 	if (who == RUSAGE_THREAD) {
1788 		task_cputime_adjusted(current, &utime, &stime);
1789 		accumulate_thread_rusage(p, r);
1790 		maxrss = p->signal->maxrss;
1791 		goto out;
1792 	}
1793 
1794 	if (!lock_task_sighand(p, &flags))
1795 		return;
1796 
1797 	switch (who) {
1798 	case RUSAGE_BOTH:
1799 	case RUSAGE_CHILDREN:
1800 		utime = p->signal->cutime;
1801 		stime = p->signal->cstime;
1802 		r->ru_nvcsw = p->signal->cnvcsw;
1803 		r->ru_nivcsw = p->signal->cnivcsw;
1804 		r->ru_minflt = p->signal->cmin_flt;
1805 		r->ru_majflt = p->signal->cmaj_flt;
1806 		r->ru_inblock = p->signal->cinblock;
1807 		r->ru_oublock = p->signal->coublock;
1808 		maxrss = p->signal->cmaxrss;
1809 
1810 		if (who == RUSAGE_CHILDREN)
1811 			break;
1812 		fallthrough;
1813 
1814 	case RUSAGE_SELF:
1815 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1816 		utime += tgutime;
1817 		stime += tgstime;
1818 		r->ru_nvcsw += p->signal->nvcsw;
1819 		r->ru_nivcsw += p->signal->nivcsw;
1820 		r->ru_minflt += p->signal->min_flt;
1821 		r->ru_majflt += p->signal->maj_flt;
1822 		r->ru_inblock += p->signal->inblock;
1823 		r->ru_oublock += p->signal->oublock;
1824 		if (maxrss < p->signal->maxrss)
1825 			maxrss = p->signal->maxrss;
1826 		t = p;
1827 		do {
1828 			accumulate_thread_rusage(t, r);
1829 		} while_each_thread(p, t);
1830 		break;
1831 
1832 	default:
1833 		BUG();
1834 	}
1835 	unlock_task_sighand(p, &flags);
1836 
1837 out:
1838 	r->ru_utime = ns_to_kernel_old_timeval(utime);
1839 	r->ru_stime = ns_to_kernel_old_timeval(stime);
1840 
1841 	if (who != RUSAGE_CHILDREN) {
1842 		struct mm_struct *mm = get_task_mm(p);
1843 
1844 		if (mm) {
1845 			setmax_mm_hiwater_rss(&maxrss, mm);
1846 			mmput(mm);
1847 		}
1848 	}
1849 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1850 }
1851 
1852 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1853 {
1854 	struct rusage r;
1855 
1856 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1857 	    who != RUSAGE_THREAD)
1858 		return -EINVAL;
1859 
1860 	getrusage(current, who, &r);
1861 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1862 }
1863 
1864 #ifdef CONFIG_COMPAT
1865 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1866 {
1867 	struct rusage r;
1868 
1869 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1870 	    who != RUSAGE_THREAD)
1871 		return -EINVAL;
1872 
1873 	getrusage(current, who, &r);
1874 	return put_compat_rusage(&r, ru);
1875 }
1876 #endif
1877 
1878 SYSCALL_DEFINE1(umask, int, mask)
1879 {
1880 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1881 	return mask;
1882 }
1883 
1884 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1885 {
1886 	struct fd exe;
1887 	struct inode *inode;
1888 	int err;
1889 
1890 	exe = fdget(fd);
1891 	if (!exe.file)
1892 		return -EBADF;
1893 
1894 	inode = file_inode(exe.file);
1895 
1896 	/*
1897 	 * Because the original mm->exe_file points to executable file, make
1898 	 * sure that this one is executable as well, to avoid breaking an
1899 	 * overall picture.
1900 	 */
1901 	err = -EACCES;
1902 	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1903 		goto exit;
1904 
1905 	err = file_permission(exe.file, MAY_EXEC);
1906 	if (err)
1907 		goto exit;
1908 
1909 	err = replace_mm_exe_file(mm, exe.file);
1910 exit:
1911 	fdput(exe);
1912 	return err;
1913 }
1914 
1915 /*
1916  * Check arithmetic relations of passed addresses.
1917  *
1918  * WARNING: we don't require any capability here so be very careful
1919  * in what is allowed for modification from userspace.
1920  */
1921 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1922 {
1923 	unsigned long mmap_max_addr = TASK_SIZE;
1924 	int error = -EINVAL, i;
1925 
1926 	static const unsigned char offsets[] = {
1927 		offsetof(struct prctl_mm_map, start_code),
1928 		offsetof(struct prctl_mm_map, end_code),
1929 		offsetof(struct prctl_mm_map, start_data),
1930 		offsetof(struct prctl_mm_map, end_data),
1931 		offsetof(struct prctl_mm_map, start_brk),
1932 		offsetof(struct prctl_mm_map, brk),
1933 		offsetof(struct prctl_mm_map, start_stack),
1934 		offsetof(struct prctl_mm_map, arg_start),
1935 		offsetof(struct prctl_mm_map, arg_end),
1936 		offsetof(struct prctl_mm_map, env_start),
1937 		offsetof(struct prctl_mm_map, env_end),
1938 	};
1939 
1940 	/*
1941 	 * Make sure the members are not somewhere outside
1942 	 * of allowed address space.
1943 	 */
1944 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1945 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1946 
1947 		if ((unsigned long)val >= mmap_max_addr ||
1948 		    (unsigned long)val < mmap_min_addr)
1949 			goto out;
1950 	}
1951 
1952 	/*
1953 	 * Make sure the pairs are ordered.
1954 	 */
1955 #define __prctl_check_order(__m1, __op, __m2)				\
1956 	((unsigned long)prctl_map->__m1 __op				\
1957 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1958 	error  = __prctl_check_order(start_code, <, end_code);
1959 	error |= __prctl_check_order(start_data,<=, end_data);
1960 	error |= __prctl_check_order(start_brk, <=, brk);
1961 	error |= __prctl_check_order(arg_start, <=, arg_end);
1962 	error |= __prctl_check_order(env_start, <=, env_end);
1963 	if (error)
1964 		goto out;
1965 #undef __prctl_check_order
1966 
1967 	error = -EINVAL;
1968 
1969 	/*
1970 	 * Neither we should allow to override limits if they set.
1971 	 */
1972 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1973 			      prctl_map->start_brk, prctl_map->end_data,
1974 			      prctl_map->start_data))
1975 			goto out;
1976 
1977 	error = 0;
1978 out:
1979 	return error;
1980 }
1981 
1982 #ifdef CONFIG_CHECKPOINT_RESTORE
1983 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1984 {
1985 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1986 	unsigned long user_auxv[AT_VECTOR_SIZE];
1987 	struct mm_struct *mm = current->mm;
1988 	int error;
1989 
1990 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1991 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1992 
1993 	if (opt == PR_SET_MM_MAP_SIZE)
1994 		return put_user((unsigned int)sizeof(prctl_map),
1995 				(unsigned int __user *)addr);
1996 
1997 	if (data_size != sizeof(prctl_map))
1998 		return -EINVAL;
1999 
2000 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2001 		return -EFAULT;
2002 
2003 	error = validate_prctl_map_addr(&prctl_map);
2004 	if (error)
2005 		return error;
2006 
2007 	if (prctl_map.auxv_size) {
2008 		/*
2009 		 * Someone is trying to cheat the auxv vector.
2010 		 */
2011 		if (!prctl_map.auxv ||
2012 				prctl_map.auxv_size > sizeof(mm->saved_auxv))
2013 			return -EINVAL;
2014 
2015 		memset(user_auxv, 0, sizeof(user_auxv));
2016 		if (copy_from_user(user_auxv,
2017 				   (const void __user *)prctl_map.auxv,
2018 				   prctl_map.auxv_size))
2019 			return -EFAULT;
2020 
2021 		/* Last entry must be AT_NULL as specification requires */
2022 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2023 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2024 	}
2025 
2026 	if (prctl_map.exe_fd != (u32)-1) {
2027 		/*
2028 		 * Check if the current user is checkpoint/restore capable.
2029 		 * At the time of this writing, it checks for CAP_SYS_ADMIN
2030 		 * or CAP_CHECKPOINT_RESTORE.
2031 		 * Note that a user with access to ptrace can masquerade an
2032 		 * arbitrary program as any executable, even setuid ones.
2033 		 * This may have implications in the tomoyo subsystem.
2034 		 */
2035 		if (!checkpoint_restore_ns_capable(current_user_ns()))
2036 			return -EPERM;
2037 
2038 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2039 		if (error)
2040 			return error;
2041 	}
2042 
2043 	/*
2044 	 * arg_lock protects concurrent updates but we still need mmap_lock for
2045 	 * read to exclude races with sys_brk.
2046 	 */
2047 	mmap_read_lock(mm);
2048 
2049 	/*
2050 	 * We don't validate if these members are pointing to
2051 	 * real present VMAs because application may have correspond
2052 	 * VMAs already unmapped and kernel uses these members for statistics
2053 	 * output in procfs mostly, except
2054 	 *
2055 	 *  - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2056 	 *    for VMAs when updating these members so anything wrong written
2057 	 *    here cause kernel to swear at userspace program but won't lead
2058 	 *    to any problem in kernel itself
2059 	 */
2060 
2061 	spin_lock(&mm->arg_lock);
2062 	mm->start_code	= prctl_map.start_code;
2063 	mm->end_code	= prctl_map.end_code;
2064 	mm->start_data	= prctl_map.start_data;
2065 	mm->end_data	= prctl_map.end_data;
2066 	mm->start_brk	= prctl_map.start_brk;
2067 	mm->brk		= prctl_map.brk;
2068 	mm->start_stack	= prctl_map.start_stack;
2069 	mm->arg_start	= prctl_map.arg_start;
2070 	mm->arg_end	= prctl_map.arg_end;
2071 	mm->env_start	= prctl_map.env_start;
2072 	mm->env_end	= prctl_map.env_end;
2073 	spin_unlock(&mm->arg_lock);
2074 
2075 	/*
2076 	 * Note this update of @saved_auxv is lockless thus
2077 	 * if someone reads this member in procfs while we're
2078 	 * updating -- it may get partly updated results. It's
2079 	 * known and acceptable trade off: we leave it as is to
2080 	 * not introduce additional locks here making the kernel
2081 	 * more complex.
2082 	 */
2083 	if (prctl_map.auxv_size)
2084 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2085 
2086 	mmap_read_unlock(mm);
2087 	return 0;
2088 }
2089 #endif /* CONFIG_CHECKPOINT_RESTORE */
2090 
2091 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2092 			  unsigned long len)
2093 {
2094 	/*
2095 	 * This doesn't move the auxiliary vector itself since it's pinned to
2096 	 * mm_struct, but it permits filling the vector with new values.  It's
2097 	 * up to the caller to provide sane values here, otherwise userspace
2098 	 * tools which use this vector might be unhappy.
2099 	 */
2100 	unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2101 
2102 	if (len > sizeof(user_auxv))
2103 		return -EINVAL;
2104 
2105 	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2106 		return -EFAULT;
2107 
2108 	/* Make sure the last entry is always AT_NULL */
2109 	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2110 	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2111 
2112 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2113 
2114 	task_lock(current);
2115 	memcpy(mm->saved_auxv, user_auxv, len);
2116 	task_unlock(current);
2117 
2118 	return 0;
2119 }
2120 
2121 static int prctl_set_mm(int opt, unsigned long addr,
2122 			unsigned long arg4, unsigned long arg5)
2123 {
2124 	struct mm_struct *mm = current->mm;
2125 	struct prctl_mm_map prctl_map = {
2126 		.auxv = NULL,
2127 		.auxv_size = 0,
2128 		.exe_fd = -1,
2129 	};
2130 	struct vm_area_struct *vma;
2131 	int error;
2132 
2133 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2134 			      opt != PR_SET_MM_MAP &&
2135 			      opt != PR_SET_MM_MAP_SIZE)))
2136 		return -EINVAL;
2137 
2138 #ifdef CONFIG_CHECKPOINT_RESTORE
2139 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2140 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2141 #endif
2142 
2143 	if (!capable(CAP_SYS_RESOURCE))
2144 		return -EPERM;
2145 
2146 	if (opt == PR_SET_MM_EXE_FILE)
2147 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2148 
2149 	if (opt == PR_SET_MM_AUXV)
2150 		return prctl_set_auxv(mm, addr, arg4);
2151 
2152 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2153 		return -EINVAL;
2154 
2155 	error = -EINVAL;
2156 
2157 	/*
2158 	 * arg_lock protects concurrent updates of arg boundaries, we need
2159 	 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2160 	 * validation.
2161 	 */
2162 	mmap_read_lock(mm);
2163 	vma = find_vma(mm, addr);
2164 
2165 	spin_lock(&mm->arg_lock);
2166 	prctl_map.start_code	= mm->start_code;
2167 	prctl_map.end_code	= mm->end_code;
2168 	prctl_map.start_data	= mm->start_data;
2169 	prctl_map.end_data	= mm->end_data;
2170 	prctl_map.start_brk	= mm->start_brk;
2171 	prctl_map.brk		= mm->brk;
2172 	prctl_map.start_stack	= mm->start_stack;
2173 	prctl_map.arg_start	= mm->arg_start;
2174 	prctl_map.arg_end	= mm->arg_end;
2175 	prctl_map.env_start	= mm->env_start;
2176 	prctl_map.env_end	= mm->env_end;
2177 
2178 	switch (opt) {
2179 	case PR_SET_MM_START_CODE:
2180 		prctl_map.start_code = addr;
2181 		break;
2182 	case PR_SET_MM_END_CODE:
2183 		prctl_map.end_code = addr;
2184 		break;
2185 	case PR_SET_MM_START_DATA:
2186 		prctl_map.start_data = addr;
2187 		break;
2188 	case PR_SET_MM_END_DATA:
2189 		prctl_map.end_data = addr;
2190 		break;
2191 	case PR_SET_MM_START_STACK:
2192 		prctl_map.start_stack = addr;
2193 		break;
2194 	case PR_SET_MM_START_BRK:
2195 		prctl_map.start_brk = addr;
2196 		break;
2197 	case PR_SET_MM_BRK:
2198 		prctl_map.brk = addr;
2199 		break;
2200 	case PR_SET_MM_ARG_START:
2201 		prctl_map.arg_start = addr;
2202 		break;
2203 	case PR_SET_MM_ARG_END:
2204 		prctl_map.arg_end = addr;
2205 		break;
2206 	case PR_SET_MM_ENV_START:
2207 		prctl_map.env_start = addr;
2208 		break;
2209 	case PR_SET_MM_ENV_END:
2210 		prctl_map.env_end = addr;
2211 		break;
2212 	default:
2213 		goto out;
2214 	}
2215 
2216 	error = validate_prctl_map_addr(&prctl_map);
2217 	if (error)
2218 		goto out;
2219 
2220 	switch (opt) {
2221 	/*
2222 	 * If command line arguments and environment
2223 	 * are placed somewhere else on stack, we can
2224 	 * set them up here, ARG_START/END to setup
2225 	 * command line arguments and ENV_START/END
2226 	 * for environment.
2227 	 */
2228 	case PR_SET_MM_START_STACK:
2229 	case PR_SET_MM_ARG_START:
2230 	case PR_SET_MM_ARG_END:
2231 	case PR_SET_MM_ENV_START:
2232 	case PR_SET_MM_ENV_END:
2233 		if (!vma) {
2234 			error = -EFAULT;
2235 			goto out;
2236 		}
2237 	}
2238 
2239 	mm->start_code	= prctl_map.start_code;
2240 	mm->end_code	= prctl_map.end_code;
2241 	mm->start_data	= prctl_map.start_data;
2242 	mm->end_data	= prctl_map.end_data;
2243 	mm->start_brk	= prctl_map.start_brk;
2244 	mm->brk		= prctl_map.brk;
2245 	mm->start_stack	= prctl_map.start_stack;
2246 	mm->arg_start	= prctl_map.arg_start;
2247 	mm->arg_end	= prctl_map.arg_end;
2248 	mm->env_start	= prctl_map.env_start;
2249 	mm->env_end	= prctl_map.env_end;
2250 
2251 	error = 0;
2252 out:
2253 	spin_unlock(&mm->arg_lock);
2254 	mmap_read_unlock(mm);
2255 	return error;
2256 }
2257 
2258 #ifdef CONFIG_CHECKPOINT_RESTORE
2259 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2260 {
2261 	return put_user(me->clear_child_tid, tid_addr);
2262 }
2263 #else
2264 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2265 {
2266 	return -EINVAL;
2267 }
2268 #endif
2269 
2270 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2271 {
2272 	/*
2273 	 * If task has has_child_subreaper - all its descendants
2274 	 * already have these flag too and new descendants will
2275 	 * inherit it on fork, skip them.
2276 	 *
2277 	 * If we've found child_reaper - skip descendants in
2278 	 * it's subtree as they will never get out pidns.
2279 	 */
2280 	if (p->signal->has_child_subreaper ||
2281 	    is_child_reaper(task_pid(p)))
2282 		return 0;
2283 
2284 	p->signal->has_child_subreaper = 1;
2285 	return 1;
2286 }
2287 
2288 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2289 {
2290 	return -EINVAL;
2291 }
2292 
2293 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2294 				    unsigned long ctrl)
2295 {
2296 	return -EINVAL;
2297 }
2298 
2299 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2300 
2301 #ifdef CONFIG_ANON_VMA_NAME
2302 
2303 #define ANON_VMA_NAME_MAX_LEN		80
2304 #define ANON_VMA_NAME_INVALID_CHARS	"\\`$[]"
2305 
2306 static inline bool is_valid_name_char(char ch)
2307 {
2308 	/* printable ascii characters, excluding ANON_VMA_NAME_INVALID_CHARS */
2309 	return ch > 0x1f && ch < 0x7f &&
2310 		!strchr(ANON_VMA_NAME_INVALID_CHARS, ch);
2311 }
2312 
2313 static int prctl_set_vma(unsigned long opt, unsigned long addr,
2314 			 unsigned long size, unsigned long arg)
2315 {
2316 	struct mm_struct *mm = current->mm;
2317 	const char __user *uname;
2318 	struct anon_vma_name *anon_name = NULL;
2319 	int error;
2320 
2321 	switch (opt) {
2322 	case PR_SET_VMA_ANON_NAME:
2323 		uname = (const char __user *)arg;
2324 		if (uname) {
2325 			char *name, *pch;
2326 
2327 			name = strndup_user(uname, ANON_VMA_NAME_MAX_LEN);
2328 			if (IS_ERR(name))
2329 				return PTR_ERR(name);
2330 
2331 			for (pch = name; *pch != '\0'; pch++) {
2332 				if (!is_valid_name_char(*pch)) {
2333 					kfree(name);
2334 					return -EINVAL;
2335 				}
2336 			}
2337 			/* anon_vma has its own copy */
2338 			anon_name = anon_vma_name_alloc(name);
2339 			kfree(name);
2340 			if (!anon_name)
2341 				return -ENOMEM;
2342 
2343 		}
2344 
2345 		mmap_write_lock(mm);
2346 		error = madvise_set_anon_name(mm, addr, size, anon_name);
2347 		mmap_write_unlock(mm);
2348 		anon_vma_name_put(anon_name);
2349 		break;
2350 	default:
2351 		error = -EINVAL;
2352 	}
2353 
2354 	return error;
2355 }
2356 
2357 #else /* CONFIG_ANON_VMA_NAME */
2358 static int prctl_set_vma(unsigned long opt, unsigned long start,
2359 			 unsigned long size, unsigned long arg)
2360 {
2361 	return -EINVAL;
2362 }
2363 #endif /* CONFIG_ANON_VMA_NAME */
2364 
2365 static inline int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
2366 				 unsigned long arg4, unsigned long arg5)
2367 {
2368 	if (arg3 || arg4 || arg5)
2369 		return -EINVAL;
2370 
2371 	if (bits & ~(PR_MDWE_REFUSE_EXEC_GAIN))
2372 		return -EINVAL;
2373 
2374 	if (bits & PR_MDWE_REFUSE_EXEC_GAIN)
2375 		set_bit(MMF_HAS_MDWE, &current->mm->flags);
2376 	else if (test_bit(MMF_HAS_MDWE, &current->mm->flags))
2377 		return -EPERM; /* Cannot unset the flag */
2378 
2379 	return 0;
2380 }
2381 
2382 static inline int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
2383 				 unsigned long arg4, unsigned long arg5)
2384 {
2385 	if (arg2 || arg3 || arg4 || arg5)
2386 		return -EINVAL;
2387 
2388 	return test_bit(MMF_HAS_MDWE, &current->mm->flags) ?
2389 		PR_MDWE_REFUSE_EXEC_GAIN : 0;
2390 }
2391 
2392 static int prctl_get_auxv(void __user *addr, unsigned long len)
2393 {
2394 	struct mm_struct *mm = current->mm;
2395 	unsigned long size = min_t(unsigned long, sizeof(mm->saved_auxv), len);
2396 
2397 	if (size && copy_to_user(addr, mm->saved_auxv, size))
2398 		return -EFAULT;
2399 	return sizeof(mm->saved_auxv);
2400 }
2401 
2402 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2403 		unsigned long, arg4, unsigned long, arg5)
2404 {
2405 	struct task_struct *me = current;
2406 	unsigned char comm[sizeof(me->comm)];
2407 	long error;
2408 
2409 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2410 	if (error != -ENOSYS)
2411 		return error;
2412 
2413 	error = 0;
2414 	switch (option) {
2415 	case PR_SET_PDEATHSIG:
2416 		if (!valid_signal(arg2)) {
2417 			error = -EINVAL;
2418 			break;
2419 		}
2420 		me->pdeath_signal = arg2;
2421 		break;
2422 	case PR_GET_PDEATHSIG:
2423 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2424 		break;
2425 	case PR_GET_DUMPABLE:
2426 		error = get_dumpable(me->mm);
2427 		break;
2428 	case PR_SET_DUMPABLE:
2429 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2430 			error = -EINVAL;
2431 			break;
2432 		}
2433 		set_dumpable(me->mm, arg2);
2434 		break;
2435 
2436 	case PR_SET_UNALIGN:
2437 		error = SET_UNALIGN_CTL(me, arg2);
2438 		break;
2439 	case PR_GET_UNALIGN:
2440 		error = GET_UNALIGN_CTL(me, arg2);
2441 		break;
2442 	case PR_SET_FPEMU:
2443 		error = SET_FPEMU_CTL(me, arg2);
2444 		break;
2445 	case PR_GET_FPEMU:
2446 		error = GET_FPEMU_CTL(me, arg2);
2447 		break;
2448 	case PR_SET_FPEXC:
2449 		error = SET_FPEXC_CTL(me, arg2);
2450 		break;
2451 	case PR_GET_FPEXC:
2452 		error = GET_FPEXC_CTL(me, arg2);
2453 		break;
2454 	case PR_GET_TIMING:
2455 		error = PR_TIMING_STATISTICAL;
2456 		break;
2457 	case PR_SET_TIMING:
2458 		if (arg2 != PR_TIMING_STATISTICAL)
2459 			error = -EINVAL;
2460 		break;
2461 	case PR_SET_NAME:
2462 		comm[sizeof(me->comm) - 1] = 0;
2463 		if (strncpy_from_user(comm, (char __user *)arg2,
2464 				      sizeof(me->comm) - 1) < 0)
2465 			return -EFAULT;
2466 		set_task_comm(me, comm);
2467 		proc_comm_connector(me);
2468 		break;
2469 	case PR_GET_NAME:
2470 		get_task_comm(comm, me);
2471 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2472 			return -EFAULT;
2473 		break;
2474 	case PR_GET_ENDIAN:
2475 		error = GET_ENDIAN(me, arg2);
2476 		break;
2477 	case PR_SET_ENDIAN:
2478 		error = SET_ENDIAN(me, arg2);
2479 		break;
2480 	case PR_GET_SECCOMP:
2481 		error = prctl_get_seccomp();
2482 		break;
2483 	case PR_SET_SECCOMP:
2484 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2485 		break;
2486 	case PR_GET_TSC:
2487 		error = GET_TSC_CTL(arg2);
2488 		break;
2489 	case PR_SET_TSC:
2490 		error = SET_TSC_CTL(arg2);
2491 		break;
2492 	case PR_TASK_PERF_EVENTS_DISABLE:
2493 		error = perf_event_task_disable();
2494 		break;
2495 	case PR_TASK_PERF_EVENTS_ENABLE:
2496 		error = perf_event_task_enable();
2497 		break;
2498 	case PR_GET_TIMERSLACK:
2499 		if (current->timer_slack_ns > ULONG_MAX)
2500 			error = ULONG_MAX;
2501 		else
2502 			error = current->timer_slack_ns;
2503 		break;
2504 	case PR_SET_TIMERSLACK:
2505 		if (arg2 <= 0)
2506 			current->timer_slack_ns =
2507 					current->default_timer_slack_ns;
2508 		else
2509 			current->timer_slack_ns = arg2;
2510 		break;
2511 	case PR_MCE_KILL:
2512 		if (arg4 | arg5)
2513 			return -EINVAL;
2514 		switch (arg2) {
2515 		case PR_MCE_KILL_CLEAR:
2516 			if (arg3 != 0)
2517 				return -EINVAL;
2518 			current->flags &= ~PF_MCE_PROCESS;
2519 			break;
2520 		case PR_MCE_KILL_SET:
2521 			current->flags |= PF_MCE_PROCESS;
2522 			if (arg3 == PR_MCE_KILL_EARLY)
2523 				current->flags |= PF_MCE_EARLY;
2524 			else if (arg3 == PR_MCE_KILL_LATE)
2525 				current->flags &= ~PF_MCE_EARLY;
2526 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2527 				current->flags &=
2528 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2529 			else
2530 				return -EINVAL;
2531 			break;
2532 	case PR_GET_AUXV:
2533 		if (arg4 || arg5)
2534 			return -EINVAL;
2535 		error = prctl_get_auxv((void __user *)arg2, arg3);
2536 		break;
2537 		default:
2538 			return -EINVAL;
2539 		}
2540 		break;
2541 	case PR_MCE_KILL_GET:
2542 		if (arg2 | arg3 | arg4 | arg5)
2543 			return -EINVAL;
2544 		if (current->flags & PF_MCE_PROCESS)
2545 			error = (current->flags & PF_MCE_EARLY) ?
2546 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2547 		else
2548 			error = PR_MCE_KILL_DEFAULT;
2549 		break;
2550 	case PR_SET_MM:
2551 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2552 		break;
2553 	case PR_GET_TID_ADDRESS:
2554 		error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2555 		break;
2556 	case PR_SET_CHILD_SUBREAPER:
2557 		me->signal->is_child_subreaper = !!arg2;
2558 		if (!arg2)
2559 			break;
2560 
2561 		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2562 		break;
2563 	case PR_GET_CHILD_SUBREAPER:
2564 		error = put_user(me->signal->is_child_subreaper,
2565 				 (int __user *)arg2);
2566 		break;
2567 	case PR_SET_NO_NEW_PRIVS:
2568 		if (arg2 != 1 || arg3 || arg4 || arg5)
2569 			return -EINVAL;
2570 
2571 		task_set_no_new_privs(current);
2572 		break;
2573 	case PR_GET_NO_NEW_PRIVS:
2574 		if (arg2 || arg3 || arg4 || arg5)
2575 			return -EINVAL;
2576 		return task_no_new_privs(current) ? 1 : 0;
2577 	case PR_GET_THP_DISABLE:
2578 		if (arg2 || arg3 || arg4 || arg5)
2579 			return -EINVAL;
2580 		error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2581 		break;
2582 	case PR_SET_THP_DISABLE:
2583 		if (arg3 || arg4 || arg5)
2584 			return -EINVAL;
2585 		if (mmap_write_lock_killable(me->mm))
2586 			return -EINTR;
2587 		if (arg2)
2588 			set_bit(MMF_DISABLE_THP, &me->mm->flags);
2589 		else
2590 			clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2591 		mmap_write_unlock(me->mm);
2592 		break;
2593 	case PR_MPX_ENABLE_MANAGEMENT:
2594 	case PR_MPX_DISABLE_MANAGEMENT:
2595 		/* No longer implemented: */
2596 		return -EINVAL;
2597 	case PR_SET_FP_MODE:
2598 		error = SET_FP_MODE(me, arg2);
2599 		break;
2600 	case PR_GET_FP_MODE:
2601 		error = GET_FP_MODE(me);
2602 		break;
2603 	case PR_SVE_SET_VL:
2604 		error = SVE_SET_VL(arg2);
2605 		break;
2606 	case PR_SVE_GET_VL:
2607 		error = SVE_GET_VL();
2608 		break;
2609 	case PR_SME_SET_VL:
2610 		error = SME_SET_VL(arg2);
2611 		break;
2612 	case PR_SME_GET_VL:
2613 		error = SME_GET_VL();
2614 		break;
2615 	case PR_GET_SPECULATION_CTRL:
2616 		if (arg3 || arg4 || arg5)
2617 			return -EINVAL;
2618 		error = arch_prctl_spec_ctrl_get(me, arg2);
2619 		break;
2620 	case PR_SET_SPECULATION_CTRL:
2621 		if (arg4 || arg5)
2622 			return -EINVAL;
2623 		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2624 		break;
2625 	case PR_PAC_RESET_KEYS:
2626 		if (arg3 || arg4 || arg5)
2627 			return -EINVAL;
2628 		error = PAC_RESET_KEYS(me, arg2);
2629 		break;
2630 	case PR_PAC_SET_ENABLED_KEYS:
2631 		if (arg4 || arg5)
2632 			return -EINVAL;
2633 		error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2634 		break;
2635 	case PR_PAC_GET_ENABLED_KEYS:
2636 		if (arg2 || arg3 || arg4 || arg5)
2637 			return -EINVAL;
2638 		error = PAC_GET_ENABLED_KEYS(me);
2639 		break;
2640 	case PR_SET_TAGGED_ADDR_CTRL:
2641 		if (arg3 || arg4 || arg5)
2642 			return -EINVAL;
2643 		error = SET_TAGGED_ADDR_CTRL(arg2);
2644 		break;
2645 	case PR_GET_TAGGED_ADDR_CTRL:
2646 		if (arg2 || arg3 || arg4 || arg5)
2647 			return -EINVAL;
2648 		error = GET_TAGGED_ADDR_CTRL();
2649 		break;
2650 	case PR_SET_IO_FLUSHER:
2651 		if (!capable(CAP_SYS_RESOURCE))
2652 			return -EPERM;
2653 
2654 		if (arg3 || arg4 || arg5)
2655 			return -EINVAL;
2656 
2657 		if (arg2 == 1)
2658 			current->flags |= PR_IO_FLUSHER;
2659 		else if (!arg2)
2660 			current->flags &= ~PR_IO_FLUSHER;
2661 		else
2662 			return -EINVAL;
2663 		break;
2664 	case PR_GET_IO_FLUSHER:
2665 		if (!capable(CAP_SYS_RESOURCE))
2666 			return -EPERM;
2667 
2668 		if (arg2 || arg3 || arg4 || arg5)
2669 			return -EINVAL;
2670 
2671 		error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2672 		break;
2673 	case PR_SET_SYSCALL_USER_DISPATCH:
2674 		error = set_syscall_user_dispatch(arg2, arg3, arg4,
2675 						  (char __user *) arg5);
2676 		break;
2677 #ifdef CONFIG_SCHED_CORE
2678 	case PR_SCHED_CORE:
2679 		error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2680 		break;
2681 #endif
2682 	case PR_SET_MDWE:
2683 		error = prctl_set_mdwe(arg2, arg3, arg4, arg5);
2684 		break;
2685 	case PR_GET_MDWE:
2686 		error = prctl_get_mdwe(arg2, arg3, arg4, arg5);
2687 		break;
2688 	case PR_SET_VMA:
2689 		error = prctl_set_vma(arg2, arg3, arg4, arg5);
2690 		break;
2691 #ifdef CONFIG_KSM
2692 	case PR_SET_MEMORY_MERGE:
2693 		if (arg3 || arg4 || arg5)
2694 			return -EINVAL;
2695 		if (mmap_write_lock_killable(me->mm))
2696 			return -EINTR;
2697 
2698 		if (arg2)
2699 			error = ksm_enable_merge_any(me->mm);
2700 		else
2701 			error = ksm_disable_merge_any(me->mm);
2702 		mmap_write_unlock(me->mm);
2703 		break;
2704 	case PR_GET_MEMORY_MERGE:
2705 		if (arg2 || arg3 || arg4 || arg5)
2706 			return -EINVAL;
2707 
2708 		error = !!test_bit(MMF_VM_MERGE_ANY, &me->mm->flags);
2709 		break;
2710 #endif
2711 	default:
2712 		error = -EINVAL;
2713 		break;
2714 	}
2715 	return error;
2716 }
2717 
2718 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2719 		struct getcpu_cache __user *, unused)
2720 {
2721 	int err = 0;
2722 	int cpu = raw_smp_processor_id();
2723 
2724 	if (cpup)
2725 		err |= put_user(cpu, cpup);
2726 	if (nodep)
2727 		err |= put_user(cpu_to_node(cpu), nodep);
2728 	return err ? -EFAULT : 0;
2729 }
2730 
2731 /**
2732  * do_sysinfo - fill in sysinfo struct
2733  * @info: pointer to buffer to fill
2734  */
2735 static int do_sysinfo(struct sysinfo *info)
2736 {
2737 	unsigned long mem_total, sav_total;
2738 	unsigned int mem_unit, bitcount;
2739 	struct timespec64 tp;
2740 
2741 	memset(info, 0, sizeof(struct sysinfo));
2742 
2743 	ktime_get_boottime_ts64(&tp);
2744 	timens_add_boottime(&tp);
2745 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2746 
2747 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2748 
2749 	info->procs = nr_threads;
2750 
2751 	si_meminfo(info);
2752 	si_swapinfo(info);
2753 
2754 	/*
2755 	 * If the sum of all the available memory (i.e. ram + swap)
2756 	 * is less than can be stored in a 32 bit unsigned long then
2757 	 * we can be binary compatible with 2.2.x kernels.  If not,
2758 	 * well, in that case 2.2.x was broken anyways...
2759 	 *
2760 	 *  -Erik Andersen <andersee@debian.org>
2761 	 */
2762 
2763 	mem_total = info->totalram + info->totalswap;
2764 	if (mem_total < info->totalram || mem_total < info->totalswap)
2765 		goto out;
2766 	bitcount = 0;
2767 	mem_unit = info->mem_unit;
2768 	while (mem_unit > 1) {
2769 		bitcount++;
2770 		mem_unit >>= 1;
2771 		sav_total = mem_total;
2772 		mem_total <<= 1;
2773 		if (mem_total < sav_total)
2774 			goto out;
2775 	}
2776 
2777 	/*
2778 	 * If mem_total did not overflow, multiply all memory values by
2779 	 * info->mem_unit and set it to 1.  This leaves things compatible
2780 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2781 	 * kernels...
2782 	 */
2783 
2784 	info->mem_unit = 1;
2785 	info->totalram <<= bitcount;
2786 	info->freeram <<= bitcount;
2787 	info->sharedram <<= bitcount;
2788 	info->bufferram <<= bitcount;
2789 	info->totalswap <<= bitcount;
2790 	info->freeswap <<= bitcount;
2791 	info->totalhigh <<= bitcount;
2792 	info->freehigh <<= bitcount;
2793 
2794 out:
2795 	return 0;
2796 }
2797 
2798 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2799 {
2800 	struct sysinfo val;
2801 
2802 	do_sysinfo(&val);
2803 
2804 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2805 		return -EFAULT;
2806 
2807 	return 0;
2808 }
2809 
2810 #ifdef CONFIG_COMPAT
2811 struct compat_sysinfo {
2812 	s32 uptime;
2813 	u32 loads[3];
2814 	u32 totalram;
2815 	u32 freeram;
2816 	u32 sharedram;
2817 	u32 bufferram;
2818 	u32 totalswap;
2819 	u32 freeswap;
2820 	u16 procs;
2821 	u16 pad;
2822 	u32 totalhigh;
2823 	u32 freehigh;
2824 	u32 mem_unit;
2825 	char _f[20-2*sizeof(u32)-sizeof(int)];
2826 };
2827 
2828 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2829 {
2830 	struct sysinfo s;
2831 	struct compat_sysinfo s_32;
2832 
2833 	do_sysinfo(&s);
2834 
2835 	/* Check to see if any memory value is too large for 32-bit and scale
2836 	 *  down if needed
2837 	 */
2838 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2839 		int bitcount = 0;
2840 
2841 		while (s.mem_unit < PAGE_SIZE) {
2842 			s.mem_unit <<= 1;
2843 			bitcount++;
2844 		}
2845 
2846 		s.totalram >>= bitcount;
2847 		s.freeram >>= bitcount;
2848 		s.sharedram >>= bitcount;
2849 		s.bufferram >>= bitcount;
2850 		s.totalswap >>= bitcount;
2851 		s.freeswap >>= bitcount;
2852 		s.totalhigh >>= bitcount;
2853 		s.freehigh >>= bitcount;
2854 	}
2855 
2856 	memset(&s_32, 0, sizeof(s_32));
2857 	s_32.uptime = s.uptime;
2858 	s_32.loads[0] = s.loads[0];
2859 	s_32.loads[1] = s.loads[1];
2860 	s_32.loads[2] = s.loads[2];
2861 	s_32.totalram = s.totalram;
2862 	s_32.freeram = s.freeram;
2863 	s_32.sharedram = s.sharedram;
2864 	s_32.bufferram = s.bufferram;
2865 	s_32.totalswap = s.totalswap;
2866 	s_32.freeswap = s.freeswap;
2867 	s_32.procs = s.procs;
2868 	s_32.totalhigh = s.totalhigh;
2869 	s_32.freehigh = s.freehigh;
2870 	s_32.mem_unit = s.mem_unit;
2871 	if (copy_to_user(info, &s_32, sizeof(s_32)))
2872 		return -EFAULT;
2873 	return 0;
2874 }
2875 #endif /* CONFIG_COMPAT */
2876