xref: /linux/kernel/sys.c (revision 99d2592023e5d0a31f5f5a83c694df48239a1e6c)
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 #include <linux/futex.h>
56 #include <linux/rseq.h>
57 
58 #include <linux/sched.h>
59 #include <linux/sched/autogroup.h>
60 #include <linux/sched/loadavg.h>
61 #include <linux/sched/stat.h>
62 #include <linux/sched/mm.h>
63 #include <linux/sched/coredump.h>
64 #include <linux/sched/task.h>
65 #include <linux/sched/cputime.h>
66 #include <linux/rcupdate.h>
67 #include <linux/uidgid.h>
68 #include <linux/cred.h>
69 
70 #include <linux/nospec.h>
71 
72 #include <linux/kmsg_dump.h>
73 /* Move somewhere else to avoid recompiling? */
74 #include <generated/utsrelease.h>
75 
76 #include <linux/uaccess.h>
77 #include <asm/io.h>
78 #include <asm/unistd.h>
79 
80 #include <trace/events/task.h>
81 
82 #include "uid16.h"
83 
84 #ifndef SET_UNALIGN_CTL
85 # define SET_UNALIGN_CTL(a, b)	(-EINVAL)
86 #endif
87 #ifndef GET_UNALIGN_CTL
88 # define GET_UNALIGN_CTL(a, b)	(-EINVAL)
89 #endif
90 #ifndef SET_FPEMU_CTL
91 # define SET_FPEMU_CTL(a, b)	(-EINVAL)
92 #endif
93 #ifndef GET_FPEMU_CTL
94 # define GET_FPEMU_CTL(a, b)	(-EINVAL)
95 #endif
96 #ifndef SET_FPEXC_CTL
97 # define SET_FPEXC_CTL(a, b)	(-EINVAL)
98 #endif
99 #ifndef GET_FPEXC_CTL
100 # define GET_FPEXC_CTL(a, b)	(-EINVAL)
101 #endif
102 #ifndef GET_ENDIAN
103 # define GET_ENDIAN(a, b)	(-EINVAL)
104 #endif
105 #ifndef SET_ENDIAN
106 # define SET_ENDIAN(a, b)	(-EINVAL)
107 #endif
108 #ifndef GET_TSC_CTL
109 # define GET_TSC_CTL(a)		(-EINVAL)
110 #endif
111 #ifndef SET_TSC_CTL
112 # define SET_TSC_CTL(a)		(-EINVAL)
113 #endif
114 #ifndef GET_FP_MODE
115 # define GET_FP_MODE(a)		(-EINVAL)
116 #endif
117 #ifndef SET_FP_MODE
118 # define SET_FP_MODE(a,b)	(-EINVAL)
119 #endif
120 #ifndef SVE_SET_VL
121 # define SVE_SET_VL(a)		(-EINVAL)
122 #endif
123 #ifndef SVE_GET_VL
124 # define SVE_GET_VL()		(-EINVAL)
125 #endif
126 #ifndef SME_SET_VL
127 # define SME_SET_VL(a)		(-EINVAL)
128 #endif
129 #ifndef SME_GET_VL
130 # define SME_GET_VL()		(-EINVAL)
131 #endif
132 #ifndef PAC_RESET_KEYS
133 # define PAC_RESET_KEYS(a, b)	(-EINVAL)
134 #endif
135 #ifndef PAC_SET_ENABLED_KEYS
136 # define PAC_SET_ENABLED_KEYS(a, b, c)	(-EINVAL)
137 #endif
138 #ifndef PAC_GET_ENABLED_KEYS
139 # define PAC_GET_ENABLED_KEYS(a)	(-EINVAL)
140 #endif
141 #ifndef SET_TAGGED_ADDR_CTRL
142 # define SET_TAGGED_ADDR_CTRL(a)	(-EINVAL)
143 #endif
144 #ifndef GET_TAGGED_ADDR_CTRL
145 # define GET_TAGGED_ADDR_CTRL()		(-EINVAL)
146 #endif
147 #ifndef RISCV_V_SET_CONTROL
148 # define RISCV_V_SET_CONTROL(a)		(-EINVAL)
149 #endif
150 #ifndef RISCV_V_GET_CONTROL
151 # define RISCV_V_GET_CONTROL()		(-EINVAL)
152 #endif
153 #ifndef RISCV_SET_ICACHE_FLUSH_CTX
154 # define RISCV_SET_ICACHE_FLUSH_CTX(a, b)	(-EINVAL)
155 #endif
156 #ifndef PPC_GET_DEXCR_ASPECT
157 # define PPC_GET_DEXCR_ASPECT(a, b)	(-EINVAL)
158 #endif
159 #ifndef PPC_SET_DEXCR_ASPECT
160 # define PPC_SET_DEXCR_ASPECT(a, b, c)	(-EINVAL)
161 #endif
162 
163 /*
164  * this is where the system-wide overflow UID and GID are defined, for
165  * architectures that now have 32-bit UID/GID but didn't in the past
166  */
167 
168 int overflowuid = DEFAULT_OVERFLOWUID;
169 int overflowgid = DEFAULT_OVERFLOWGID;
170 
171 EXPORT_SYMBOL(overflowuid);
172 EXPORT_SYMBOL(overflowgid);
173 
174 /*
175  * the same as above, but for filesystems which can only store a 16-bit
176  * UID and GID. as such, this is needed on all architectures
177  */
178 
179 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
180 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
181 
182 EXPORT_SYMBOL(fs_overflowuid);
183 EXPORT_SYMBOL(fs_overflowgid);
184 
185 static const struct ctl_table overflow_sysctl_table[] = {
186 	{
187 		.procname	= "overflowuid",
188 		.data		= &overflowuid,
189 		.maxlen		= sizeof(int),
190 		.mode		= 0644,
191 		.proc_handler	= proc_dointvec_minmax,
192 		.extra1		= SYSCTL_ZERO,
193 		.extra2		= SYSCTL_MAXOLDUID,
194 	},
195 	{
196 		.procname	= "overflowgid",
197 		.data		= &overflowgid,
198 		.maxlen		= sizeof(int),
199 		.mode		= 0644,
200 		.proc_handler	= proc_dointvec_minmax,
201 		.extra1		= SYSCTL_ZERO,
202 		.extra2		= SYSCTL_MAXOLDUID,
203 	},
204 };
205 
206 static int __init init_overflow_sysctl(void)
207 {
208 	register_sysctl_init("kernel", overflow_sysctl_table);
209 	return 0;
210 }
211 
212 postcore_initcall(init_overflow_sysctl);
213 
214 /*
215  * Returns true if current's euid is same as p's uid or euid,
216  * or has CAP_SYS_NICE to p's user_ns.
217  *
218  * Called with rcu_read_lock, creds are safe
219  */
220 static bool set_one_prio_perm(struct task_struct *p)
221 {
222 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
223 
224 	if (uid_eq(pcred->uid,  cred->euid) ||
225 	    uid_eq(pcred->euid, cred->euid))
226 		return true;
227 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
228 		return true;
229 	return false;
230 }
231 
232 /*
233  * set the priority of a task
234  * - the caller must hold the RCU read lock
235  */
236 static int set_one_prio(struct task_struct *p, int niceval, int error)
237 {
238 	int no_nice;
239 
240 	if (!set_one_prio_perm(p)) {
241 		error = -EPERM;
242 		goto out;
243 	}
244 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
245 		error = -EACCES;
246 		goto out;
247 	}
248 	no_nice = security_task_setnice(p, niceval);
249 	if (no_nice) {
250 		error = no_nice;
251 		goto out;
252 	}
253 	if (error == -ESRCH)
254 		error = 0;
255 	set_user_nice(p, niceval);
256 out:
257 	return error;
258 }
259 
260 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
261 {
262 	struct task_struct *g, *p;
263 	struct user_struct *user;
264 	const struct cred *cred = current_cred();
265 	int error = -EINVAL;
266 	struct pid *pgrp;
267 	kuid_t uid;
268 
269 	if (which > PRIO_USER || which < PRIO_PROCESS)
270 		goto out;
271 
272 	/* normalize: avoid signed division (rounding problems) */
273 	error = -ESRCH;
274 	if (niceval < MIN_NICE)
275 		niceval = MIN_NICE;
276 	if (niceval > MAX_NICE)
277 		niceval = MAX_NICE;
278 
279 	rcu_read_lock();
280 	switch (which) {
281 	case PRIO_PROCESS:
282 		if (who)
283 			p = find_task_by_vpid(who);
284 		else
285 			p = current;
286 		if (p)
287 			error = set_one_prio(p, niceval, error);
288 		break;
289 	case PRIO_PGRP:
290 		if (who)
291 			pgrp = find_vpid(who);
292 		else
293 			pgrp = task_pgrp(current);
294 		read_lock(&tasklist_lock);
295 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
296 			error = set_one_prio(p, niceval, error);
297 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
298 		read_unlock(&tasklist_lock);
299 		break;
300 	case PRIO_USER:
301 		uid = make_kuid(cred->user_ns, who);
302 		user = cred->user;
303 		if (!who)
304 			uid = cred->uid;
305 		else if (!uid_eq(uid, cred->uid)) {
306 			user = find_user(uid);
307 			if (!user)
308 				goto out_unlock;	/* No processes for this user */
309 		}
310 		for_each_process_thread(g, p) {
311 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
312 				error = set_one_prio(p, niceval, error);
313 		}
314 		if (!uid_eq(uid, cred->uid))
315 			free_uid(user);		/* For find_user() */
316 		break;
317 	}
318 out_unlock:
319 	rcu_read_unlock();
320 out:
321 	return error;
322 }
323 
324 /*
325  * Ugh. To avoid negative return values, "getpriority()" will
326  * not return the normal nice-value, but a negated value that
327  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
328  * to stay compatible.
329  */
330 SYSCALL_DEFINE2(getpriority, int, which, int, who)
331 {
332 	struct task_struct *g, *p;
333 	struct user_struct *user;
334 	const struct cred *cred = current_cred();
335 	long niceval, retval = -ESRCH;
336 	struct pid *pgrp;
337 	kuid_t uid;
338 
339 	if (which > PRIO_USER || which < PRIO_PROCESS)
340 		return -EINVAL;
341 
342 	rcu_read_lock();
343 	switch (which) {
344 	case PRIO_PROCESS:
345 		if (who)
346 			p = find_task_by_vpid(who);
347 		else
348 			p = current;
349 		if (p) {
350 			niceval = nice_to_rlimit(task_nice(p));
351 			if (niceval > retval)
352 				retval = niceval;
353 		}
354 		break;
355 	case PRIO_PGRP:
356 		if (who)
357 			pgrp = find_vpid(who);
358 		else
359 			pgrp = task_pgrp(current);
360 		read_lock(&tasklist_lock);
361 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
362 			niceval = nice_to_rlimit(task_nice(p));
363 			if (niceval > retval)
364 				retval = niceval;
365 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
366 		read_unlock(&tasklist_lock);
367 		break;
368 	case PRIO_USER:
369 		uid = make_kuid(cred->user_ns, who);
370 		user = cred->user;
371 		if (!who)
372 			uid = cred->uid;
373 		else if (!uid_eq(uid, cred->uid)) {
374 			user = find_user(uid);
375 			if (!user)
376 				goto out_unlock;	/* No processes for this user */
377 		}
378 		for_each_process_thread(g, p) {
379 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
380 				niceval = nice_to_rlimit(task_nice(p));
381 				if (niceval > retval)
382 					retval = niceval;
383 			}
384 		}
385 		if (!uid_eq(uid, cred->uid))
386 			free_uid(user);		/* for find_user() */
387 		break;
388 	}
389 out_unlock:
390 	rcu_read_unlock();
391 
392 	return retval;
393 }
394 
395 /*
396  * Unprivileged users may change the real gid to the effective gid
397  * or vice versa.  (BSD-style)
398  *
399  * If you set the real gid at all, or set the effective gid to a value not
400  * equal to the real gid, then the saved gid is set to the new effective gid.
401  *
402  * This makes it possible for a setgid program to completely drop its
403  * privileges, which is often a useful assertion to make when you are doing
404  * a security audit over a program.
405  *
406  * The general idea is that a program which uses just setregid() will be
407  * 100% compatible with BSD.  A program which uses just setgid() will be
408  * 100% compatible with POSIX with saved IDs.
409  *
410  * SMP: There are not races, the GIDs are checked only by filesystem
411  *      operations (as far as semantic preservation is concerned).
412  */
413 #ifdef CONFIG_MULTIUSER
414 long __sys_setregid(gid_t rgid, gid_t egid)
415 {
416 	struct user_namespace *ns = current_user_ns();
417 	const struct cred *old;
418 	struct cred *new;
419 	int retval;
420 	kgid_t krgid, kegid;
421 
422 	krgid = make_kgid(ns, rgid);
423 	kegid = make_kgid(ns, egid);
424 
425 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
426 		return -EINVAL;
427 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
428 		return -EINVAL;
429 
430 	new = prepare_creds();
431 	if (!new)
432 		return -ENOMEM;
433 	old = current_cred();
434 
435 	retval = -EPERM;
436 	if (rgid != (gid_t) -1) {
437 		if (gid_eq(old->gid, krgid) ||
438 		    gid_eq(old->egid, krgid) ||
439 		    ns_capable_setid(old->user_ns, CAP_SETGID))
440 			new->gid = krgid;
441 		else
442 			goto error;
443 	}
444 	if (egid != (gid_t) -1) {
445 		if (gid_eq(old->gid, kegid) ||
446 		    gid_eq(old->egid, kegid) ||
447 		    gid_eq(old->sgid, kegid) ||
448 		    ns_capable_setid(old->user_ns, CAP_SETGID))
449 			new->egid = kegid;
450 		else
451 			goto error;
452 	}
453 
454 	if (rgid != (gid_t) -1 ||
455 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
456 		new->sgid = new->egid;
457 	new->fsgid = new->egid;
458 
459 	retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
460 	if (retval < 0)
461 		goto error;
462 
463 	return commit_creds(new);
464 
465 error:
466 	abort_creds(new);
467 	return retval;
468 }
469 
470 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
471 {
472 	return __sys_setregid(rgid, egid);
473 }
474 
475 /*
476  * setgid() is implemented like SysV w/ SAVED_IDS
477  *
478  * SMP: Same implicit races as above.
479  */
480 long __sys_setgid(gid_t gid)
481 {
482 	struct user_namespace *ns = current_user_ns();
483 	const struct cred *old;
484 	struct cred *new;
485 	int retval;
486 	kgid_t kgid;
487 
488 	kgid = make_kgid(ns, gid);
489 	if (!gid_valid(kgid))
490 		return -EINVAL;
491 
492 	new = prepare_creds();
493 	if (!new)
494 		return -ENOMEM;
495 	old = current_cred();
496 
497 	retval = -EPERM;
498 	if (ns_capable_setid(old->user_ns, CAP_SETGID))
499 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
500 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
501 		new->egid = new->fsgid = kgid;
502 	else
503 		goto error;
504 
505 	retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
506 	if (retval < 0)
507 		goto error;
508 
509 	return commit_creds(new);
510 
511 error:
512 	abort_creds(new);
513 	return retval;
514 }
515 
516 SYSCALL_DEFINE1(setgid, gid_t, gid)
517 {
518 	return __sys_setgid(gid);
519 }
520 
521 /*
522  * change the user struct in a credentials set to match the new UID
523  */
524 static int set_user(struct cred *new)
525 {
526 	struct user_struct *new_user;
527 
528 	new_user = alloc_uid(new->uid);
529 	if (!new_user)
530 		return -EAGAIN;
531 
532 	free_uid(new->user);
533 	new->user = new_user;
534 	return 0;
535 }
536 
537 static void flag_nproc_exceeded(struct cred *new)
538 {
539 	if (new->ucounts == current_ucounts())
540 		return;
541 
542 	/*
543 	 * We don't fail in case of NPROC limit excess here because too many
544 	 * poorly written programs don't check set*uid() return code, assuming
545 	 * it never fails if called by root.  We may still enforce NPROC limit
546 	 * for programs doing set*uid()+execve() by harmlessly deferring the
547 	 * failure to the execve() stage.
548 	 */
549 	if (is_rlimit_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
550 			new->user != INIT_USER)
551 		current->flags |= PF_NPROC_EXCEEDED;
552 	else
553 		current->flags &= ~PF_NPROC_EXCEEDED;
554 }
555 
556 /*
557  * Unprivileged users may change the real uid to the effective uid
558  * or vice versa.  (BSD-style)
559  *
560  * If you set the real uid at all, or set the effective uid to a value not
561  * equal to the real uid, then the saved uid is set to the new effective uid.
562  *
563  * This makes it possible for a setuid program to completely drop its
564  * privileges, which is often a useful assertion to make when you are doing
565  * a security audit over a program.
566  *
567  * The general idea is that a program which uses just setreuid() will be
568  * 100% compatible with BSD.  A program which uses just setuid() will be
569  * 100% compatible with POSIX with saved IDs.
570  */
571 long __sys_setreuid(uid_t ruid, uid_t euid)
572 {
573 	struct user_namespace *ns = current_user_ns();
574 	const struct cred *old;
575 	struct cred *new;
576 	int retval;
577 	kuid_t kruid, keuid;
578 
579 	kruid = make_kuid(ns, ruid);
580 	keuid = make_kuid(ns, euid);
581 
582 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
583 		return -EINVAL;
584 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
585 		return -EINVAL;
586 
587 	new = prepare_creds();
588 	if (!new)
589 		return -ENOMEM;
590 	old = current_cred();
591 
592 	retval = -EPERM;
593 	if (ruid != (uid_t) -1) {
594 		new->uid = kruid;
595 		if (!uid_eq(old->uid, kruid) &&
596 		    !uid_eq(old->euid, kruid) &&
597 		    !ns_capable_setid(old->user_ns, CAP_SETUID))
598 			goto error;
599 	}
600 
601 	if (euid != (uid_t) -1) {
602 		new->euid = keuid;
603 		if (!uid_eq(old->uid, keuid) &&
604 		    !uid_eq(old->euid, keuid) &&
605 		    !uid_eq(old->suid, keuid) &&
606 		    !ns_capable_setid(old->user_ns, CAP_SETUID))
607 			goto error;
608 	}
609 
610 	if (!uid_eq(new->uid, old->uid)) {
611 		retval = set_user(new);
612 		if (retval < 0)
613 			goto error;
614 	}
615 	if (ruid != (uid_t) -1 ||
616 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
617 		new->suid = new->euid;
618 	new->fsuid = new->euid;
619 
620 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
621 	if (retval < 0)
622 		goto error;
623 
624 	retval = set_cred_ucounts(new);
625 	if (retval < 0)
626 		goto error;
627 
628 	flag_nproc_exceeded(new);
629 	return commit_creds(new);
630 
631 error:
632 	abort_creds(new);
633 	return retval;
634 }
635 
636 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
637 {
638 	return __sys_setreuid(ruid, euid);
639 }
640 
641 /*
642  * setuid() is implemented like SysV with SAVED_IDS
643  *
644  * Note that SAVED_ID's is deficient in that a setuid root program
645  * like sendmail, for example, cannot set its uid to be a normal
646  * user and then switch back, because if you're root, setuid() sets
647  * the saved uid too.  If you don't like this, blame the bright people
648  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
649  * will allow a root program to temporarily drop privileges and be able to
650  * regain them by swapping the real and effective uid.
651  */
652 long __sys_setuid(uid_t uid)
653 {
654 	struct user_namespace *ns = current_user_ns();
655 	const struct cred *old;
656 	struct cred *new;
657 	int retval;
658 	kuid_t kuid;
659 
660 	kuid = make_kuid(ns, uid);
661 	if (!uid_valid(kuid))
662 		return -EINVAL;
663 
664 	new = prepare_creds();
665 	if (!new)
666 		return -ENOMEM;
667 	old = current_cred();
668 
669 	retval = -EPERM;
670 	if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
671 		new->suid = new->uid = kuid;
672 		if (!uid_eq(kuid, old->uid)) {
673 			retval = set_user(new);
674 			if (retval < 0)
675 				goto error;
676 		}
677 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
678 		goto error;
679 	}
680 
681 	new->fsuid = new->euid = kuid;
682 
683 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
684 	if (retval < 0)
685 		goto error;
686 
687 	retval = set_cred_ucounts(new);
688 	if (retval < 0)
689 		goto error;
690 
691 	flag_nproc_exceeded(new);
692 	return commit_creds(new);
693 
694 error:
695 	abort_creds(new);
696 	return retval;
697 }
698 
699 SYSCALL_DEFINE1(setuid, uid_t, uid)
700 {
701 	return __sys_setuid(uid);
702 }
703 
704 
705 /*
706  * This function implements a generic ability to update ruid, euid,
707  * and suid.  This allows you to implement the 4.4 compatible seteuid().
708  */
709 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
710 {
711 	struct user_namespace *ns = current_user_ns();
712 	const struct cred *old;
713 	struct cred *new;
714 	int retval;
715 	kuid_t kruid, keuid, ksuid;
716 	bool ruid_new, euid_new, suid_new;
717 
718 	kruid = make_kuid(ns, ruid);
719 	keuid = make_kuid(ns, euid);
720 	ksuid = make_kuid(ns, suid);
721 
722 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
723 		return -EINVAL;
724 
725 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
726 		return -EINVAL;
727 
728 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
729 		return -EINVAL;
730 
731 	old = current_cred();
732 
733 	/* check for no-op */
734 	if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
735 	    (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
736 				    uid_eq(keuid, old->fsuid))) &&
737 	    (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
738 		return 0;
739 
740 	ruid_new = ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
741 		   !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
742 	euid_new = euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
743 		   !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
744 	suid_new = suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
745 		   !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
746 	if ((ruid_new || euid_new || suid_new) &&
747 	    !ns_capable_setid(old->user_ns, CAP_SETUID))
748 		return -EPERM;
749 
750 	new = prepare_creds();
751 	if (!new)
752 		return -ENOMEM;
753 
754 	if (ruid != (uid_t) -1) {
755 		new->uid = kruid;
756 		if (!uid_eq(kruid, old->uid)) {
757 			retval = set_user(new);
758 			if (retval < 0)
759 				goto error;
760 		}
761 	}
762 	if (euid != (uid_t) -1)
763 		new->euid = keuid;
764 	if (suid != (uid_t) -1)
765 		new->suid = ksuid;
766 	new->fsuid = new->euid;
767 
768 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
769 	if (retval < 0)
770 		goto error;
771 
772 	retval = set_cred_ucounts(new);
773 	if (retval < 0)
774 		goto error;
775 
776 	flag_nproc_exceeded(new);
777 	return commit_creds(new);
778 
779 error:
780 	abort_creds(new);
781 	return retval;
782 }
783 
784 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
785 {
786 	return __sys_setresuid(ruid, euid, suid);
787 }
788 
789 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
790 {
791 	const struct cred *cred = current_cred();
792 	int retval;
793 	uid_t ruid, euid, suid;
794 
795 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
796 	euid = from_kuid_munged(cred->user_ns, cred->euid);
797 	suid = from_kuid_munged(cred->user_ns, cred->suid);
798 
799 	retval = put_user(ruid, ruidp);
800 	if (!retval) {
801 		retval = put_user(euid, euidp);
802 		if (!retval)
803 			return put_user(suid, suidp);
804 	}
805 	return retval;
806 }
807 
808 /*
809  * Same as above, but for rgid, egid, sgid.
810  */
811 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
812 {
813 	struct user_namespace *ns = current_user_ns();
814 	const struct cred *old;
815 	struct cred *new;
816 	int retval;
817 	kgid_t krgid, kegid, ksgid;
818 	bool rgid_new, egid_new, sgid_new;
819 
820 	krgid = make_kgid(ns, rgid);
821 	kegid = make_kgid(ns, egid);
822 	ksgid = make_kgid(ns, sgid);
823 
824 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
825 		return -EINVAL;
826 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
827 		return -EINVAL;
828 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
829 		return -EINVAL;
830 
831 	old = current_cred();
832 
833 	/* check for no-op */
834 	if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
835 	    (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
836 				    gid_eq(kegid, old->fsgid))) &&
837 	    (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
838 		return 0;
839 
840 	rgid_new = rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
841 		   !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
842 	egid_new = egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
843 		   !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
844 	sgid_new = sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
845 		   !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
846 	if ((rgid_new || egid_new || sgid_new) &&
847 	    !ns_capable_setid(old->user_ns, CAP_SETGID))
848 		return -EPERM;
849 
850 	new = prepare_creds();
851 	if (!new)
852 		return -ENOMEM;
853 
854 	if (rgid != (gid_t) -1)
855 		new->gid = krgid;
856 	if (egid != (gid_t) -1)
857 		new->egid = kegid;
858 	if (sgid != (gid_t) -1)
859 		new->sgid = ksgid;
860 	new->fsgid = new->egid;
861 
862 	retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
863 	if (retval < 0)
864 		goto error;
865 
866 	return commit_creds(new);
867 
868 error:
869 	abort_creds(new);
870 	return retval;
871 }
872 
873 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
874 {
875 	return __sys_setresgid(rgid, egid, sgid);
876 }
877 
878 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
879 {
880 	const struct cred *cred = current_cred();
881 	int retval;
882 	gid_t rgid, egid, sgid;
883 
884 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
885 	egid = from_kgid_munged(cred->user_ns, cred->egid);
886 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
887 
888 	retval = put_user(rgid, rgidp);
889 	if (!retval) {
890 		retval = put_user(egid, egidp);
891 		if (!retval)
892 			retval = put_user(sgid, sgidp);
893 	}
894 
895 	return retval;
896 }
897 
898 
899 /*
900  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
901  * is used for "access()" and for the NFS daemon (letting nfsd stay at
902  * whatever uid it wants to). It normally shadows "euid", except when
903  * explicitly set by setfsuid() or for access..
904  */
905 long __sys_setfsuid(uid_t uid)
906 {
907 	const struct cred *old;
908 	struct cred *new;
909 	uid_t old_fsuid;
910 	kuid_t kuid;
911 
912 	old = current_cred();
913 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
914 
915 	kuid = make_kuid(old->user_ns, uid);
916 	if (!uid_valid(kuid))
917 		return old_fsuid;
918 
919 	new = prepare_creds();
920 	if (!new)
921 		return old_fsuid;
922 
923 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
924 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
925 	    ns_capable_setid(old->user_ns, CAP_SETUID)) {
926 		if (!uid_eq(kuid, old->fsuid)) {
927 			new->fsuid = kuid;
928 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
929 				goto change_okay;
930 		}
931 	}
932 
933 	abort_creds(new);
934 	return old_fsuid;
935 
936 change_okay:
937 	commit_creds(new);
938 	return old_fsuid;
939 }
940 
941 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
942 {
943 	return __sys_setfsuid(uid);
944 }
945 
946 /*
947  * Samma på svenska..
948  */
949 long __sys_setfsgid(gid_t gid)
950 {
951 	const struct cred *old;
952 	struct cred *new;
953 	gid_t old_fsgid;
954 	kgid_t kgid;
955 
956 	old = current_cred();
957 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
958 
959 	kgid = make_kgid(old->user_ns, gid);
960 	if (!gid_valid(kgid))
961 		return old_fsgid;
962 
963 	new = prepare_creds();
964 	if (!new)
965 		return old_fsgid;
966 
967 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
968 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
969 	    ns_capable_setid(old->user_ns, CAP_SETGID)) {
970 		if (!gid_eq(kgid, old->fsgid)) {
971 			new->fsgid = kgid;
972 			if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
973 				goto change_okay;
974 		}
975 	}
976 
977 	abort_creds(new);
978 	return old_fsgid;
979 
980 change_okay:
981 	commit_creds(new);
982 	return old_fsgid;
983 }
984 
985 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
986 {
987 	return __sys_setfsgid(gid);
988 }
989 #endif /* CONFIG_MULTIUSER */
990 
991 /**
992  * sys_getpid - return the thread group id of the current process
993  *
994  * Note, despite the name, this returns the tgid not the pid.  The tgid and
995  * the pid are identical unless CLONE_THREAD was specified on clone() in
996  * which case the tgid is the same in all threads of the same group.
997  *
998  * This is SMP safe as current->tgid does not change.
999  */
1000 SYSCALL_DEFINE0(getpid)
1001 {
1002 	return task_tgid_vnr(current);
1003 }
1004 
1005 /* Thread ID - the internal kernel "pid" */
1006 SYSCALL_DEFINE0(gettid)
1007 {
1008 	return task_pid_vnr(current);
1009 }
1010 
1011 /*
1012  * Accessing ->real_parent is not SMP-safe, it could
1013  * change from under us. However, we can use a stale
1014  * value of ->real_parent under rcu_read_lock(), see
1015  * release_task()->call_rcu(delayed_put_task_struct).
1016  */
1017 SYSCALL_DEFINE0(getppid)
1018 {
1019 	int pid;
1020 
1021 	rcu_read_lock();
1022 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
1023 	rcu_read_unlock();
1024 
1025 	return pid;
1026 }
1027 
1028 SYSCALL_DEFINE0(getuid)
1029 {
1030 	/* Only we change this so SMP safe */
1031 	return from_kuid_munged(current_user_ns(), current_uid());
1032 }
1033 
1034 SYSCALL_DEFINE0(geteuid)
1035 {
1036 	/* Only we change this so SMP safe */
1037 	return from_kuid_munged(current_user_ns(), current_euid());
1038 }
1039 
1040 SYSCALL_DEFINE0(getgid)
1041 {
1042 	/* Only we change this so SMP safe */
1043 	return from_kgid_munged(current_user_ns(), current_gid());
1044 }
1045 
1046 SYSCALL_DEFINE0(getegid)
1047 {
1048 	/* Only we change this so SMP safe */
1049 	return from_kgid_munged(current_user_ns(), current_egid());
1050 }
1051 
1052 static void do_sys_times(struct tms *tms)
1053 {
1054 	u64 tgutime, tgstime, cutime, cstime;
1055 
1056 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1057 	cutime = current->signal->cutime;
1058 	cstime = current->signal->cstime;
1059 	tms->tms_utime = nsec_to_clock_t(tgutime);
1060 	tms->tms_stime = nsec_to_clock_t(tgstime);
1061 	tms->tms_cutime = nsec_to_clock_t(cutime);
1062 	tms->tms_cstime = nsec_to_clock_t(cstime);
1063 }
1064 
1065 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1066 {
1067 	if (tbuf) {
1068 		struct tms tmp;
1069 
1070 		do_sys_times(&tmp);
1071 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1072 			return -EFAULT;
1073 	}
1074 	force_successful_syscall_return();
1075 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
1076 }
1077 
1078 #ifdef CONFIG_COMPAT
1079 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1080 {
1081 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1082 }
1083 
1084 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1085 {
1086 	if (tbuf) {
1087 		struct tms tms;
1088 		struct compat_tms tmp;
1089 
1090 		do_sys_times(&tms);
1091 		/* Convert our struct tms to the compat version. */
1092 		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1093 		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1094 		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1095 		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1096 		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1097 			return -EFAULT;
1098 	}
1099 	force_successful_syscall_return();
1100 	return compat_jiffies_to_clock_t(jiffies);
1101 }
1102 #endif
1103 
1104 /*
1105  * This needs some heavy checking ...
1106  * I just haven't the stomach for it. I also don't fully
1107  * understand sessions/pgrp etc. Let somebody who does explain it.
1108  *
1109  * OK, I think I have the protection semantics right.... this is really
1110  * only important on a multi-user system anyway, to make sure one user
1111  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1112  *
1113  * !PF_FORKNOEXEC check to conform completely to POSIX.
1114  */
1115 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1116 {
1117 	struct task_struct *p;
1118 	struct task_struct *group_leader = current->group_leader;
1119 	struct pid *pids[PIDTYPE_MAX] = { 0 };
1120 	struct pid *pgrp;
1121 	int err;
1122 
1123 	if (!pid)
1124 		pid = task_pid_vnr(group_leader);
1125 	if (!pgid)
1126 		pgid = pid;
1127 	if (pgid < 0)
1128 		return -EINVAL;
1129 	rcu_read_lock();
1130 
1131 	/* From this point forward we keep holding onto the tasklist lock
1132 	 * so that our parent does not change from under us. -DaveM
1133 	 */
1134 	write_lock_irq(&tasklist_lock);
1135 
1136 	err = -ESRCH;
1137 	p = find_task_by_vpid(pid);
1138 	if (!p)
1139 		goto out;
1140 
1141 	err = -EINVAL;
1142 	if (!thread_group_leader(p))
1143 		goto out;
1144 
1145 	if (same_thread_group(p->real_parent, group_leader)) {
1146 		err = -EPERM;
1147 		if (task_session(p) != task_session(group_leader))
1148 			goto out;
1149 		err = -EACCES;
1150 		if (!(p->flags & PF_FORKNOEXEC))
1151 			goto out;
1152 	} else {
1153 		err = -ESRCH;
1154 		if (p != group_leader)
1155 			goto out;
1156 	}
1157 
1158 	err = -EPERM;
1159 	if (p->signal->leader)
1160 		goto out;
1161 
1162 	pgrp = task_pid(p);
1163 	if (pgid != pid) {
1164 		struct task_struct *g;
1165 
1166 		pgrp = find_vpid(pgid);
1167 		g = pid_task(pgrp, PIDTYPE_PGID);
1168 		if (!g || task_session(g) != task_session(group_leader))
1169 			goto out;
1170 	}
1171 
1172 	err = security_task_setpgid(p, pgid);
1173 	if (err)
1174 		goto out;
1175 
1176 	if (task_pgrp(p) != pgrp)
1177 		change_pid(pids, p, PIDTYPE_PGID, pgrp);
1178 
1179 	err = 0;
1180 out:
1181 	/* All paths lead to here, thus we are safe. -DaveM */
1182 	write_unlock_irq(&tasklist_lock);
1183 	rcu_read_unlock();
1184 	free_pids(pids);
1185 	return err;
1186 }
1187 
1188 static int do_getpgid(pid_t pid)
1189 {
1190 	struct task_struct *p;
1191 	struct pid *grp;
1192 	int retval;
1193 
1194 	rcu_read_lock();
1195 	if (!pid)
1196 		grp = task_pgrp(current);
1197 	else {
1198 		retval = -ESRCH;
1199 		p = find_task_by_vpid(pid);
1200 		if (!p)
1201 			goto out;
1202 		grp = task_pgrp(p);
1203 		if (!grp)
1204 			goto out;
1205 
1206 		retval = security_task_getpgid(p);
1207 		if (retval)
1208 			goto out;
1209 	}
1210 	retval = pid_vnr(grp);
1211 out:
1212 	rcu_read_unlock();
1213 	return retval;
1214 }
1215 
1216 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1217 {
1218 	return do_getpgid(pid);
1219 }
1220 
1221 #ifdef __ARCH_WANT_SYS_GETPGRP
1222 
1223 SYSCALL_DEFINE0(getpgrp)
1224 {
1225 	return do_getpgid(0);
1226 }
1227 
1228 #endif
1229 
1230 SYSCALL_DEFINE1(getsid, pid_t, pid)
1231 {
1232 	struct task_struct *p;
1233 	struct pid *sid;
1234 	int retval;
1235 
1236 	rcu_read_lock();
1237 	if (!pid)
1238 		sid = task_session(current);
1239 	else {
1240 		retval = -ESRCH;
1241 		p = find_task_by_vpid(pid);
1242 		if (!p)
1243 			goto out;
1244 		sid = task_session(p);
1245 		if (!sid)
1246 			goto out;
1247 
1248 		retval = security_task_getsid(p);
1249 		if (retval)
1250 			goto out;
1251 	}
1252 	retval = pid_vnr(sid);
1253 out:
1254 	rcu_read_unlock();
1255 	return retval;
1256 }
1257 
1258 static void set_special_pids(struct pid **pids, struct pid *pid)
1259 {
1260 	struct task_struct *curr = current->group_leader;
1261 
1262 	if (task_session(curr) != pid)
1263 		change_pid(pids, curr, PIDTYPE_SID, pid);
1264 
1265 	if (task_pgrp(curr) != pid)
1266 		change_pid(pids, curr, PIDTYPE_PGID, pid);
1267 }
1268 
1269 int ksys_setsid(void)
1270 {
1271 	struct task_struct *group_leader = current->group_leader;
1272 	struct pid *sid = task_pid(group_leader);
1273 	struct pid *pids[PIDTYPE_MAX] = { 0 };
1274 	pid_t session = pid_vnr(sid);
1275 	int err = -EPERM;
1276 
1277 	write_lock_irq(&tasklist_lock);
1278 	/* Fail if I am already a session leader */
1279 	if (group_leader->signal->leader)
1280 		goto out;
1281 
1282 	/* Fail if a process group id already exists that equals the
1283 	 * proposed session id.
1284 	 */
1285 	if (pid_task(sid, PIDTYPE_PGID))
1286 		goto out;
1287 
1288 	group_leader->signal->leader = 1;
1289 	set_special_pids(pids, sid);
1290 
1291 	proc_clear_tty(group_leader);
1292 
1293 	err = session;
1294 out:
1295 	write_unlock_irq(&tasklist_lock);
1296 	free_pids(pids);
1297 	if (err > 0) {
1298 		proc_sid_connector(group_leader);
1299 		sched_autogroup_create_attach(group_leader);
1300 	}
1301 	return err;
1302 }
1303 
1304 SYSCALL_DEFINE0(setsid)
1305 {
1306 	return ksys_setsid();
1307 }
1308 
1309 DECLARE_RWSEM(uts_sem);
1310 
1311 #ifdef COMPAT_UTS_MACHINE
1312 #define override_architecture(name) \
1313 	(personality(current->personality) == PER_LINUX32 && \
1314 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1315 		      sizeof(COMPAT_UTS_MACHINE)))
1316 #else
1317 #define override_architecture(name)	0
1318 #endif
1319 
1320 /*
1321  * Work around broken programs that cannot handle "Linux 3.0".
1322  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1323  * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1324  * 2.6.60.
1325  */
1326 static int override_release(char __user *release, size_t len)
1327 {
1328 	int ret = 0;
1329 
1330 	if (current->personality & UNAME26) {
1331 		const char *rest = UTS_RELEASE;
1332 		char buf[65] = { 0 };
1333 		int ndots = 0;
1334 		unsigned v;
1335 		size_t copy;
1336 
1337 		while (*rest) {
1338 			if (*rest == '.' && ++ndots >= 3)
1339 				break;
1340 			if (!isdigit(*rest) && *rest != '.')
1341 				break;
1342 			rest++;
1343 		}
1344 		v = LINUX_VERSION_PATCHLEVEL + 60;
1345 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1346 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1347 		ret = copy_to_user(release, buf, copy + 1);
1348 	}
1349 	return ret;
1350 }
1351 
1352 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1353 {
1354 	struct new_utsname tmp;
1355 
1356 	down_read(&uts_sem);
1357 	memcpy(&tmp, utsname(), sizeof(tmp));
1358 	up_read(&uts_sem);
1359 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1360 		return -EFAULT;
1361 
1362 	if (override_release(name->release, sizeof(name->release)))
1363 		return -EFAULT;
1364 	if (override_architecture(name))
1365 		return -EFAULT;
1366 	return 0;
1367 }
1368 
1369 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1370 /*
1371  * Old cruft
1372  */
1373 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1374 {
1375 	struct old_utsname tmp;
1376 
1377 	if (!name)
1378 		return -EFAULT;
1379 
1380 	down_read(&uts_sem);
1381 	memcpy(&tmp, utsname(), sizeof(tmp));
1382 	up_read(&uts_sem);
1383 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1384 		return -EFAULT;
1385 
1386 	if (override_release(name->release, sizeof(name->release)))
1387 		return -EFAULT;
1388 	if (override_architecture(name))
1389 		return -EFAULT;
1390 	return 0;
1391 }
1392 
1393 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1394 {
1395 	struct oldold_utsname tmp;
1396 
1397 	if (!name)
1398 		return -EFAULT;
1399 
1400 	memset(&tmp, 0, sizeof(tmp));
1401 
1402 	down_read(&uts_sem);
1403 	memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1404 	memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1405 	memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1406 	memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1407 	memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1408 	up_read(&uts_sem);
1409 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1410 		return -EFAULT;
1411 
1412 	if (override_architecture(name))
1413 		return -EFAULT;
1414 	if (override_release(name->release, sizeof(name->release)))
1415 		return -EFAULT;
1416 	return 0;
1417 }
1418 #endif
1419 
1420 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1421 {
1422 	int errno;
1423 	char tmp[__NEW_UTS_LEN];
1424 
1425 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1426 		return -EPERM;
1427 
1428 	if (len < 0 || len > __NEW_UTS_LEN)
1429 		return -EINVAL;
1430 	errno = -EFAULT;
1431 	if (!copy_from_user(tmp, name, len)) {
1432 		struct new_utsname *u;
1433 
1434 		add_device_randomness(tmp, len);
1435 		down_write(&uts_sem);
1436 		u = utsname();
1437 		memcpy(u->nodename, tmp, len);
1438 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1439 		errno = 0;
1440 		uts_proc_notify(UTS_PROC_HOSTNAME);
1441 		up_write(&uts_sem);
1442 	}
1443 	return errno;
1444 }
1445 
1446 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1447 
1448 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1449 {
1450 	int i;
1451 	struct new_utsname *u;
1452 	char tmp[__NEW_UTS_LEN + 1];
1453 
1454 	if (len < 0)
1455 		return -EINVAL;
1456 	down_read(&uts_sem);
1457 	u = utsname();
1458 	i = 1 + strlen(u->nodename);
1459 	if (i > len)
1460 		i = len;
1461 	memcpy(tmp, u->nodename, i);
1462 	up_read(&uts_sem);
1463 	if (copy_to_user(name, tmp, i))
1464 		return -EFAULT;
1465 	return 0;
1466 }
1467 
1468 #endif
1469 
1470 /*
1471  * Only setdomainname; getdomainname can be implemented by calling
1472  * uname()
1473  */
1474 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1475 {
1476 	int errno;
1477 	char tmp[__NEW_UTS_LEN];
1478 
1479 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1480 		return -EPERM;
1481 	if (len < 0 || len > __NEW_UTS_LEN)
1482 		return -EINVAL;
1483 
1484 	errno = -EFAULT;
1485 	if (!copy_from_user(tmp, name, len)) {
1486 		struct new_utsname *u;
1487 
1488 		add_device_randomness(tmp, len);
1489 		down_write(&uts_sem);
1490 		u = utsname();
1491 		memcpy(u->domainname, tmp, len);
1492 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1493 		errno = 0;
1494 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1495 		up_write(&uts_sem);
1496 	}
1497 	return errno;
1498 }
1499 
1500 /* make sure you are allowed to change @tsk limits before calling this */
1501 static int do_prlimit(struct task_struct *tsk, unsigned int resource,
1502 		      struct rlimit *new_rlim, struct rlimit *old_rlim)
1503 {
1504 	struct rlimit *rlim;
1505 	int retval = 0;
1506 
1507 	if (resource >= RLIM_NLIMITS)
1508 		return -EINVAL;
1509 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1510 
1511 	if (new_rlim) {
1512 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1513 			return -EINVAL;
1514 		if (resource == RLIMIT_NOFILE &&
1515 				new_rlim->rlim_max > sysctl_nr_open)
1516 			return -EPERM;
1517 	}
1518 
1519 	/* Holding a refcount on tsk protects tsk->signal from disappearing. */
1520 	rlim = tsk->signal->rlim + resource;
1521 	task_lock(tsk->group_leader);
1522 	if (new_rlim) {
1523 		/*
1524 		 * Keep the capable check against init_user_ns until cgroups can
1525 		 * contain all limits.
1526 		 */
1527 		if (new_rlim->rlim_max > rlim->rlim_max &&
1528 				!capable(CAP_SYS_RESOURCE))
1529 			retval = -EPERM;
1530 		if (!retval)
1531 			retval = security_task_setrlimit(tsk, resource, new_rlim);
1532 	}
1533 	if (!retval) {
1534 		if (old_rlim)
1535 			*old_rlim = *rlim;
1536 		if (new_rlim)
1537 			*rlim = *new_rlim;
1538 	}
1539 	task_unlock(tsk->group_leader);
1540 
1541 	/*
1542 	 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1543 	 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1544 	 * ignores the rlimit.
1545 	 */
1546 	if (!retval && new_rlim && resource == RLIMIT_CPU &&
1547 	    new_rlim->rlim_cur != RLIM_INFINITY &&
1548 	    IS_ENABLED(CONFIG_POSIX_TIMERS)) {
1549 		/*
1550 		 * update_rlimit_cpu can fail if the task is exiting, but there
1551 		 * may be other tasks in the thread group that are not exiting,
1552 		 * and they need their cpu timers adjusted.
1553 		 *
1554 		 * The group_leader is the last task to be released, so if we
1555 		 * cannot update_rlimit_cpu on it, then the entire process is
1556 		 * exiting and we do not need to update at all.
1557 		 */
1558 		update_rlimit_cpu(tsk->group_leader, new_rlim->rlim_cur);
1559 	}
1560 
1561 	return retval;
1562 }
1563 
1564 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1565 {
1566 	struct rlimit value;
1567 	int ret;
1568 
1569 	ret = do_prlimit(current, resource, NULL, &value);
1570 	if (!ret)
1571 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1572 
1573 	return ret;
1574 }
1575 
1576 #ifdef CONFIG_COMPAT
1577 
1578 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1579 		       struct compat_rlimit __user *, rlim)
1580 {
1581 	struct rlimit r;
1582 	struct compat_rlimit r32;
1583 
1584 	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1585 		return -EFAULT;
1586 
1587 	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1588 		r.rlim_cur = RLIM_INFINITY;
1589 	else
1590 		r.rlim_cur = r32.rlim_cur;
1591 	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1592 		r.rlim_max = RLIM_INFINITY;
1593 	else
1594 		r.rlim_max = r32.rlim_max;
1595 	return do_prlimit(current, resource, &r, NULL);
1596 }
1597 
1598 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1599 		       struct compat_rlimit __user *, rlim)
1600 {
1601 	struct rlimit r;
1602 	int ret;
1603 
1604 	ret = do_prlimit(current, resource, NULL, &r);
1605 	if (!ret) {
1606 		struct compat_rlimit r32;
1607 		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1608 			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1609 		else
1610 			r32.rlim_cur = r.rlim_cur;
1611 		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1612 			r32.rlim_max = COMPAT_RLIM_INFINITY;
1613 		else
1614 			r32.rlim_max = r.rlim_max;
1615 
1616 		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1617 			return -EFAULT;
1618 	}
1619 	return ret;
1620 }
1621 
1622 #endif
1623 
1624 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1625 
1626 /*
1627  *	Back compatibility for getrlimit. Needed for some apps.
1628  */
1629 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1630 		struct rlimit __user *, rlim)
1631 {
1632 	struct rlimit x;
1633 	if (resource >= RLIM_NLIMITS)
1634 		return -EINVAL;
1635 
1636 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1637 	task_lock(current->group_leader);
1638 	x = current->signal->rlim[resource];
1639 	task_unlock(current->group_leader);
1640 	if (x.rlim_cur > 0x7FFFFFFF)
1641 		x.rlim_cur = 0x7FFFFFFF;
1642 	if (x.rlim_max > 0x7FFFFFFF)
1643 		x.rlim_max = 0x7FFFFFFF;
1644 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1645 }
1646 
1647 #ifdef CONFIG_COMPAT
1648 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1649 		       struct compat_rlimit __user *, rlim)
1650 {
1651 	struct rlimit r;
1652 
1653 	if (resource >= RLIM_NLIMITS)
1654 		return -EINVAL;
1655 
1656 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1657 	task_lock(current->group_leader);
1658 	r = current->signal->rlim[resource];
1659 	task_unlock(current->group_leader);
1660 	if (r.rlim_cur > 0x7FFFFFFF)
1661 		r.rlim_cur = 0x7FFFFFFF;
1662 	if (r.rlim_max > 0x7FFFFFFF)
1663 		r.rlim_max = 0x7FFFFFFF;
1664 
1665 	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1666 	    put_user(r.rlim_max, &rlim->rlim_max))
1667 		return -EFAULT;
1668 	return 0;
1669 }
1670 #endif
1671 
1672 #endif
1673 
1674 static inline bool rlim64_is_infinity(__u64 rlim64)
1675 {
1676 #if BITS_PER_LONG < 64
1677 	return rlim64 >= ULONG_MAX;
1678 #else
1679 	return rlim64 == RLIM64_INFINITY;
1680 #endif
1681 }
1682 
1683 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1684 {
1685 	if (rlim->rlim_cur == RLIM_INFINITY)
1686 		rlim64->rlim_cur = RLIM64_INFINITY;
1687 	else
1688 		rlim64->rlim_cur = rlim->rlim_cur;
1689 	if (rlim->rlim_max == RLIM_INFINITY)
1690 		rlim64->rlim_max = RLIM64_INFINITY;
1691 	else
1692 		rlim64->rlim_max = rlim->rlim_max;
1693 }
1694 
1695 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1696 {
1697 	if (rlim64_is_infinity(rlim64->rlim_cur))
1698 		rlim->rlim_cur = RLIM_INFINITY;
1699 	else
1700 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1701 	if (rlim64_is_infinity(rlim64->rlim_max))
1702 		rlim->rlim_max = RLIM_INFINITY;
1703 	else
1704 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1705 }
1706 
1707 /* rcu lock must be held */
1708 static int check_prlimit_permission(struct task_struct *task,
1709 				    unsigned int flags)
1710 {
1711 	const struct cred *cred = current_cred(), *tcred;
1712 	bool id_match;
1713 
1714 	if (current == task)
1715 		return 0;
1716 
1717 	tcred = __task_cred(task);
1718 	id_match = (uid_eq(cred->uid, tcred->euid) &&
1719 		    uid_eq(cred->uid, tcred->suid) &&
1720 		    uid_eq(cred->uid, tcred->uid)  &&
1721 		    gid_eq(cred->gid, tcred->egid) &&
1722 		    gid_eq(cred->gid, tcred->sgid) &&
1723 		    gid_eq(cred->gid, tcred->gid));
1724 	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1725 		return -EPERM;
1726 
1727 	return security_task_prlimit(cred, tcred, flags);
1728 }
1729 
1730 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1731 		const struct rlimit64 __user *, new_rlim,
1732 		struct rlimit64 __user *, old_rlim)
1733 {
1734 	struct rlimit64 old64, new64;
1735 	struct rlimit old, new;
1736 	struct task_struct *tsk;
1737 	unsigned int checkflags = 0;
1738 	bool need_tasklist;
1739 	int ret;
1740 
1741 	if (old_rlim)
1742 		checkflags |= LSM_PRLIMIT_READ;
1743 
1744 	if (new_rlim) {
1745 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1746 			return -EFAULT;
1747 		rlim64_to_rlim(&new64, &new);
1748 		checkflags |= LSM_PRLIMIT_WRITE;
1749 	}
1750 
1751 	rcu_read_lock();
1752 	tsk = pid ? find_task_by_vpid(pid) : current;
1753 	if (!tsk) {
1754 		rcu_read_unlock();
1755 		return -ESRCH;
1756 	}
1757 	ret = check_prlimit_permission(tsk, checkflags);
1758 	if (ret) {
1759 		rcu_read_unlock();
1760 		return ret;
1761 	}
1762 	get_task_struct(tsk);
1763 	rcu_read_unlock();
1764 
1765 	need_tasklist = !same_thread_group(tsk, current);
1766 	if (need_tasklist) {
1767 		/*
1768 		 * Ensure we can't race with group exit or de_thread(),
1769 		 * so tsk->group_leader can't be freed or changed until
1770 		 * read_unlock(tasklist_lock) below.
1771 		 */
1772 		read_lock(&tasklist_lock);
1773 		if (!pid_alive(tsk))
1774 			ret = -ESRCH;
1775 	}
1776 
1777 	if (!ret) {
1778 		ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1779 				old_rlim ? &old : NULL);
1780 	}
1781 
1782 	if (need_tasklist)
1783 		read_unlock(&tasklist_lock);
1784 
1785 	if (!ret && old_rlim) {
1786 		rlim_to_rlim64(&old, &old64);
1787 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1788 			ret = -EFAULT;
1789 	}
1790 
1791 	put_task_struct(tsk);
1792 	return ret;
1793 }
1794 
1795 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1796 {
1797 	struct rlimit new_rlim;
1798 
1799 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1800 		return -EFAULT;
1801 	return do_prlimit(current, resource, &new_rlim, NULL);
1802 }
1803 
1804 /*
1805  * It would make sense to put struct rusage in the task_struct,
1806  * except that would make the task_struct be *really big*.  After
1807  * task_struct gets moved into malloc'ed memory, it would
1808  * make sense to do this.  It will make moving the rest of the information
1809  * a lot simpler!  (Which we're not doing right now because we're not
1810  * measuring them yet).
1811  *
1812  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1813  * races with threads incrementing their own counters.  But since word
1814  * reads are atomic, we either get new values or old values and we don't
1815  * care which for the sums.  We always take the siglock to protect reading
1816  * the c* fields from p->signal from races with exit.c updating those
1817  * fields when reaping, so a sample either gets all the additions of a
1818  * given child after it's reaped, or none so this sample is before reaping.
1819  *
1820  * Locking:
1821  * We need to take the siglock for CHILDEREN, SELF and BOTH
1822  * for  the cases current multithreaded, non-current single threaded
1823  * non-current multithreaded.  Thread traversal is now safe with
1824  * the siglock held.
1825  * Strictly speaking, we donot need to take the siglock if we are current and
1826  * single threaded,  as no one else can take our signal_struct away, no one
1827  * else can  reap the  children to update signal->c* counters, and no one else
1828  * can race with the signal-> fields. If we do not take any lock, the
1829  * signal-> fields could be read out of order while another thread was just
1830  * exiting. So we should  place a read memory barrier when we avoid the lock.
1831  * On the writer side,  write memory barrier is implied in  __exit_signal
1832  * as __exit_signal releases  the siglock spinlock after updating the signal->
1833  * fields. But we don't do this yet to keep things simple.
1834  *
1835  */
1836 
1837 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1838 {
1839 	r->ru_nvcsw += t->nvcsw;
1840 	r->ru_nivcsw += t->nivcsw;
1841 	r->ru_minflt += t->min_flt;
1842 	r->ru_majflt += t->maj_flt;
1843 	r->ru_inblock += task_io_get_inblock(t);
1844 	r->ru_oublock += task_io_get_oublock(t);
1845 }
1846 
1847 void getrusage(struct task_struct *p, int who, struct rusage *r)
1848 {
1849 	struct task_struct *t;
1850 	unsigned long flags;
1851 	u64 tgutime, tgstime, utime, stime;
1852 	unsigned long maxrss;
1853 	struct mm_struct *mm;
1854 	struct signal_struct *sig = p->signal;
1855 	unsigned int seq = 0;
1856 
1857 retry:
1858 	memset(r, 0, sizeof(*r));
1859 	utime = stime = 0;
1860 	maxrss = 0;
1861 
1862 	if (who == RUSAGE_THREAD) {
1863 		task_cputime_adjusted(current, &utime, &stime);
1864 		accumulate_thread_rusage(p, r);
1865 		maxrss = sig->maxrss;
1866 		goto out_thread;
1867 	}
1868 
1869 	flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
1870 
1871 	switch (who) {
1872 	case RUSAGE_BOTH:
1873 	case RUSAGE_CHILDREN:
1874 		utime = sig->cutime;
1875 		stime = sig->cstime;
1876 		r->ru_nvcsw = sig->cnvcsw;
1877 		r->ru_nivcsw = sig->cnivcsw;
1878 		r->ru_minflt = sig->cmin_flt;
1879 		r->ru_majflt = sig->cmaj_flt;
1880 		r->ru_inblock = sig->cinblock;
1881 		r->ru_oublock = sig->coublock;
1882 		maxrss = sig->cmaxrss;
1883 
1884 		if (who == RUSAGE_CHILDREN)
1885 			break;
1886 		fallthrough;
1887 
1888 	case RUSAGE_SELF:
1889 		r->ru_nvcsw += sig->nvcsw;
1890 		r->ru_nivcsw += sig->nivcsw;
1891 		r->ru_minflt += sig->min_flt;
1892 		r->ru_majflt += sig->maj_flt;
1893 		r->ru_inblock += sig->inblock;
1894 		r->ru_oublock += sig->oublock;
1895 		if (maxrss < sig->maxrss)
1896 			maxrss = sig->maxrss;
1897 
1898 		rcu_read_lock();
1899 		__for_each_thread(sig, t)
1900 			accumulate_thread_rusage(t, r);
1901 		rcu_read_unlock();
1902 
1903 		break;
1904 
1905 	default:
1906 		BUG();
1907 	}
1908 
1909 	if (need_seqretry(&sig->stats_lock, seq)) {
1910 		seq = 1;
1911 		goto retry;
1912 	}
1913 	done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
1914 
1915 	if (who == RUSAGE_CHILDREN)
1916 		goto out_children;
1917 
1918 	thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1919 	utime += tgutime;
1920 	stime += tgstime;
1921 
1922 out_thread:
1923 	mm = get_task_mm(p);
1924 	if (mm) {
1925 		setmax_mm_hiwater_rss(&maxrss, mm);
1926 		mmput(mm);
1927 	}
1928 
1929 out_children:
1930 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1931 	r->ru_utime = ns_to_kernel_old_timeval(utime);
1932 	r->ru_stime = ns_to_kernel_old_timeval(stime);
1933 }
1934 
1935 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1936 {
1937 	struct rusage r;
1938 
1939 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1940 	    who != RUSAGE_THREAD)
1941 		return -EINVAL;
1942 
1943 	getrusage(current, who, &r);
1944 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1945 }
1946 
1947 #ifdef CONFIG_COMPAT
1948 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1949 {
1950 	struct rusage r;
1951 
1952 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1953 	    who != RUSAGE_THREAD)
1954 		return -EINVAL;
1955 
1956 	getrusage(current, who, &r);
1957 	return put_compat_rusage(&r, ru);
1958 }
1959 #endif
1960 
1961 SYSCALL_DEFINE1(umask, int, mask)
1962 {
1963 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1964 	return mask;
1965 }
1966 
1967 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1968 {
1969 	CLASS(fd, exe)(fd);
1970 	struct inode *inode;
1971 	int err;
1972 
1973 	if (fd_empty(exe))
1974 		return -EBADF;
1975 
1976 	inode = file_inode(fd_file(exe));
1977 
1978 	/*
1979 	 * Because the original mm->exe_file points to executable file, make
1980 	 * sure that this one is executable as well, to avoid breaking an
1981 	 * overall picture.
1982 	 */
1983 	if (!S_ISREG(inode->i_mode) || path_noexec(&fd_file(exe)->f_path))
1984 		return -EACCES;
1985 
1986 	err = file_permission(fd_file(exe), MAY_EXEC);
1987 	if (err)
1988 		return err;
1989 
1990 	return replace_mm_exe_file(mm, fd_file(exe));
1991 }
1992 
1993 /*
1994  * Check arithmetic relations of passed addresses.
1995  *
1996  * WARNING: we don't require any capability here so be very careful
1997  * in what is allowed for modification from userspace.
1998  */
1999 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
2000 {
2001 	unsigned long mmap_max_addr = TASK_SIZE;
2002 	int error = -EINVAL, i;
2003 
2004 	static const unsigned char offsets[] = {
2005 		offsetof(struct prctl_mm_map, start_code),
2006 		offsetof(struct prctl_mm_map, end_code),
2007 		offsetof(struct prctl_mm_map, start_data),
2008 		offsetof(struct prctl_mm_map, end_data),
2009 		offsetof(struct prctl_mm_map, start_brk),
2010 		offsetof(struct prctl_mm_map, brk),
2011 		offsetof(struct prctl_mm_map, start_stack),
2012 		offsetof(struct prctl_mm_map, arg_start),
2013 		offsetof(struct prctl_mm_map, arg_end),
2014 		offsetof(struct prctl_mm_map, env_start),
2015 		offsetof(struct prctl_mm_map, env_end),
2016 	};
2017 
2018 	/*
2019 	 * Make sure the members are not somewhere outside
2020 	 * of allowed address space.
2021 	 */
2022 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
2023 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
2024 
2025 		if ((unsigned long)val >= mmap_max_addr ||
2026 		    (unsigned long)val < mmap_min_addr)
2027 			goto out;
2028 	}
2029 
2030 	/*
2031 	 * Make sure the pairs are ordered.
2032 	 */
2033 #define __prctl_check_order(__m1, __op, __m2)				\
2034 	((unsigned long)prctl_map->__m1 __op				\
2035 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
2036 	error  = __prctl_check_order(start_code, <, end_code);
2037 	error |= __prctl_check_order(start_data,<=, end_data);
2038 	error |= __prctl_check_order(start_brk, <=, brk);
2039 	error |= __prctl_check_order(arg_start, <=, arg_end);
2040 	error |= __prctl_check_order(env_start, <=, env_end);
2041 	if (error)
2042 		goto out;
2043 #undef __prctl_check_order
2044 
2045 	error = -EINVAL;
2046 
2047 	/*
2048 	 * Neither we should allow to override limits if they set.
2049 	 */
2050 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
2051 			      prctl_map->start_brk, prctl_map->end_data,
2052 			      prctl_map->start_data))
2053 			goto out;
2054 
2055 	error = 0;
2056 out:
2057 	return error;
2058 }
2059 
2060 #ifdef CONFIG_CHECKPOINT_RESTORE
2061 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
2062 {
2063 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
2064 	unsigned long user_auxv[AT_VECTOR_SIZE];
2065 	struct mm_struct *mm = current->mm;
2066 	int error;
2067 
2068 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2069 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
2070 
2071 	if (opt == PR_SET_MM_MAP_SIZE)
2072 		return put_user((unsigned int)sizeof(prctl_map),
2073 				(unsigned int __user *)addr);
2074 
2075 	if (data_size != sizeof(prctl_map))
2076 		return -EINVAL;
2077 
2078 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2079 		return -EFAULT;
2080 
2081 	error = validate_prctl_map_addr(&prctl_map);
2082 	if (error)
2083 		return error;
2084 
2085 	if (prctl_map.auxv_size) {
2086 		/*
2087 		 * Someone is trying to cheat the auxv vector.
2088 		 */
2089 		if (!prctl_map.auxv ||
2090 				prctl_map.auxv_size > sizeof(mm->saved_auxv))
2091 			return -EINVAL;
2092 
2093 		memset(user_auxv, 0, sizeof(user_auxv));
2094 		if (copy_from_user(user_auxv,
2095 				   (const void __user *)prctl_map.auxv,
2096 				   prctl_map.auxv_size))
2097 			return -EFAULT;
2098 
2099 		/* Last entry must be AT_NULL as specification requires */
2100 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2101 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2102 	}
2103 
2104 	if (prctl_map.exe_fd != (u32)-1) {
2105 		/*
2106 		 * Check if the current user is checkpoint/restore capable.
2107 		 * At the time of this writing, it checks for CAP_SYS_ADMIN
2108 		 * or CAP_CHECKPOINT_RESTORE.
2109 		 * Note that a user with access to ptrace can masquerade an
2110 		 * arbitrary program as any executable, even setuid ones.
2111 		 * This may have implications in the tomoyo subsystem.
2112 		 */
2113 		if (!checkpoint_restore_ns_capable(current_user_ns()))
2114 			return -EPERM;
2115 
2116 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2117 		if (error)
2118 			return error;
2119 	}
2120 
2121 	/*
2122 	 * arg_lock protects concurrent updates but we still need mmap_lock for
2123 	 * read to exclude races with sys_brk.
2124 	 */
2125 	mmap_read_lock(mm);
2126 
2127 	/*
2128 	 * We don't validate if these members are pointing to
2129 	 * real present VMAs because application may have correspond
2130 	 * VMAs already unmapped and kernel uses these members for statistics
2131 	 * output in procfs mostly, except
2132 	 *
2133 	 *  - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2134 	 *    for VMAs when updating these members so anything wrong written
2135 	 *    here cause kernel to swear at userspace program but won't lead
2136 	 *    to any problem in kernel itself
2137 	 */
2138 
2139 	spin_lock(&mm->arg_lock);
2140 	mm->start_code	= prctl_map.start_code;
2141 	mm->end_code	= prctl_map.end_code;
2142 	mm->start_data	= prctl_map.start_data;
2143 	mm->end_data	= prctl_map.end_data;
2144 	mm->start_brk	= prctl_map.start_brk;
2145 	mm->brk		= prctl_map.brk;
2146 	mm->start_stack	= prctl_map.start_stack;
2147 	mm->arg_start	= prctl_map.arg_start;
2148 	mm->arg_end	= prctl_map.arg_end;
2149 	mm->env_start	= prctl_map.env_start;
2150 	mm->env_end	= prctl_map.env_end;
2151 	spin_unlock(&mm->arg_lock);
2152 
2153 	/*
2154 	 * Note this update of @saved_auxv is lockless thus
2155 	 * if someone reads this member in procfs while we're
2156 	 * updating -- it may get partly updated results. It's
2157 	 * known and acceptable trade off: we leave it as is to
2158 	 * not introduce additional locks here making the kernel
2159 	 * more complex.
2160 	 */
2161 	if (prctl_map.auxv_size)
2162 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2163 
2164 	mmap_read_unlock(mm);
2165 	return 0;
2166 }
2167 #endif /* CONFIG_CHECKPOINT_RESTORE */
2168 
2169 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2170 			  unsigned long len)
2171 {
2172 	/*
2173 	 * This doesn't move the auxiliary vector itself since it's pinned to
2174 	 * mm_struct, but it permits filling the vector with new values.  It's
2175 	 * up to the caller to provide sane values here, otherwise userspace
2176 	 * tools which use this vector might be unhappy.
2177 	 */
2178 	unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2179 
2180 	if (len > sizeof(user_auxv))
2181 		return -EINVAL;
2182 
2183 	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2184 		return -EFAULT;
2185 
2186 	/* Make sure the last entry is always AT_NULL */
2187 	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2188 	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2189 
2190 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2191 
2192 	task_lock(current);
2193 	memcpy(mm->saved_auxv, user_auxv, len);
2194 	task_unlock(current);
2195 
2196 	return 0;
2197 }
2198 
2199 static int prctl_set_mm(int opt, unsigned long addr,
2200 			unsigned long arg4, unsigned long arg5)
2201 {
2202 	struct mm_struct *mm = current->mm;
2203 	struct prctl_mm_map prctl_map = {
2204 		.auxv = NULL,
2205 		.auxv_size = 0,
2206 		.exe_fd = -1,
2207 	};
2208 	struct vm_area_struct *vma;
2209 	int error;
2210 
2211 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2212 			      opt != PR_SET_MM_MAP &&
2213 			      opt != PR_SET_MM_MAP_SIZE)))
2214 		return -EINVAL;
2215 
2216 #ifdef CONFIG_CHECKPOINT_RESTORE
2217 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2218 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2219 #endif
2220 
2221 	if (!capable(CAP_SYS_RESOURCE))
2222 		return -EPERM;
2223 
2224 	if (opt == PR_SET_MM_EXE_FILE)
2225 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2226 
2227 	if (opt == PR_SET_MM_AUXV)
2228 		return prctl_set_auxv(mm, addr, arg4);
2229 
2230 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2231 		return -EINVAL;
2232 
2233 	error = -EINVAL;
2234 
2235 	/*
2236 	 * arg_lock protects concurrent updates of arg boundaries, we need
2237 	 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2238 	 * validation.
2239 	 */
2240 	mmap_read_lock(mm);
2241 	vma = find_vma(mm, addr);
2242 
2243 	spin_lock(&mm->arg_lock);
2244 	prctl_map.start_code	= mm->start_code;
2245 	prctl_map.end_code	= mm->end_code;
2246 	prctl_map.start_data	= mm->start_data;
2247 	prctl_map.end_data	= mm->end_data;
2248 	prctl_map.start_brk	= mm->start_brk;
2249 	prctl_map.brk		= mm->brk;
2250 	prctl_map.start_stack	= mm->start_stack;
2251 	prctl_map.arg_start	= mm->arg_start;
2252 	prctl_map.arg_end	= mm->arg_end;
2253 	prctl_map.env_start	= mm->env_start;
2254 	prctl_map.env_end	= mm->env_end;
2255 
2256 	switch (opt) {
2257 	case PR_SET_MM_START_CODE:
2258 		prctl_map.start_code = addr;
2259 		break;
2260 	case PR_SET_MM_END_CODE:
2261 		prctl_map.end_code = addr;
2262 		break;
2263 	case PR_SET_MM_START_DATA:
2264 		prctl_map.start_data = addr;
2265 		break;
2266 	case PR_SET_MM_END_DATA:
2267 		prctl_map.end_data = addr;
2268 		break;
2269 	case PR_SET_MM_START_STACK:
2270 		prctl_map.start_stack = addr;
2271 		break;
2272 	case PR_SET_MM_START_BRK:
2273 		prctl_map.start_brk = addr;
2274 		break;
2275 	case PR_SET_MM_BRK:
2276 		prctl_map.brk = addr;
2277 		break;
2278 	case PR_SET_MM_ARG_START:
2279 		prctl_map.arg_start = addr;
2280 		break;
2281 	case PR_SET_MM_ARG_END:
2282 		prctl_map.arg_end = addr;
2283 		break;
2284 	case PR_SET_MM_ENV_START:
2285 		prctl_map.env_start = addr;
2286 		break;
2287 	case PR_SET_MM_ENV_END:
2288 		prctl_map.env_end = addr;
2289 		break;
2290 	default:
2291 		goto out;
2292 	}
2293 
2294 	error = validate_prctl_map_addr(&prctl_map);
2295 	if (error)
2296 		goto out;
2297 
2298 	switch (opt) {
2299 	/*
2300 	 * If command line arguments and environment
2301 	 * are placed somewhere else on stack, we can
2302 	 * set them up here, ARG_START/END to setup
2303 	 * command line arguments and ENV_START/END
2304 	 * for environment.
2305 	 */
2306 	case PR_SET_MM_START_STACK:
2307 	case PR_SET_MM_ARG_START:
2308 	case PR_SET_MM_ARG_END:
2309 	case PR_SET_MM_ENV_START:
2310 	case PR_SET_MM_ENV_END:
2311 		if (!vma) {
2312 			error = -EFAULT;
2313 			goto out;
2314 		}
2315 	}
2316 
2317 	mm->start_code	= prctl_map.start_code;
2318 	mm->end_code	= prctl_map.end_code;
2319 	mm->start_data	= prctl_map.start_data;
2320 	mm->end_data	= prctl_map.end_data;
2321 	mm->start_brk	= prctl_map.start_brk;
2322 	mm->brk		= prctl_map.brk;
2323 	mm->start_stack	= prctl_map.start_stack;
2324 	mm->arg_start	= prctl_map.arg_start;
2325 	mm->arg_end	= prctl_map.arg_end;
2326 	mm->env_start	= prctl_map.env_start;
2327 	mm->env_end	= prctl_map.env_end;
2328 
2329 	error = 0;
2330 out:
2331 	spin_unlock(&mm->arg_lock);
2332 	mmap_read_unlock(mm);
2333 	return error;
2334 }
2335 
2336 #ifdef CONFIG_CHECKPOINT_RESTORE
2337 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2338 {
2339 	return put_user(me->clear_child_tid, tid_addr);
2340 }
2341 #else
2342 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2343 {
2344 	return -EINVAL;
2345 }
2346 #endif
2347 
2348 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2349 {
2350 	/*
2351 	 * If task has has_child_subreaper - all its descendants
2352 	 * already have these flag too and new descendants will
2353 	 * inherit it on fork, skip them.
2354 	 *
2355 	 * If we've found child_reaper - skip descendants in
2356 	 * it's subtree as they will never get out pidns.
2357 	 */
2358 	if (p->signal->has_child_subreaper ||
2359 	    is_child_reaper(task_pid(p)))
2360 		return 0;
2361 
2362 	p->signal->has_child_subreaper = 1;
2363 	return 1;
2364 }
2365 
2366 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2367 {
2368 	return -EINVAL;
2369 }
2370 
2371 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2372 				    unsigned long ctrl)
2373 {
2374 	return -EINVAL;
2375 }
2376 
2377 int __weak arch_get_shadow_stack_status(struct task_struct *t, unsigned long __user *status)
2378 {
2379 	return -EINVAL;
2380 }
2381 
2382 int __weak arch_set_shadow_stack_status(struct task_struct *t, unsigned long status)
2383 {
2384 	return -EINVAL;
2385 }
2386 
2387 int __weak arch_lock_shadow_stack_status(struct task_struct *t, unsigned long status)
2388 {
2389 	return -EINVAL;
2390 }
2391 
2392 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2393 
2394 static int prctl_set_vma(unsigned long opt, unsigned long addr,
2395 			 unsigned long size, unsigned long arg)
2396 {
2397 	int error;
2398 
2399 	switch (opt) {
2400 	case PR_SET_VMA_ANON_NAME:
2401 		error = set_anon_vma_name(addr, size, (const char __user *)arg);
2402 		break;
2403 	default:
2404 		error = -EINVAL;
2405 	}
2406 
2407 	return error;
2408 }
2409 
2410 static inline unsigned long get_current_mdwe(void)
2411 {
2412 	unsigned long ret = 0;
2413 
2414 	if (mm_flags_test(MMF_HAS_MDWE, current->mm))
2415 		ret |= PR_MDWE_REFUSE_EXEC_GAIN;
2416 	if (mm_flags_test(MMF_HAS_MDWE_NO_INHERIT, current->mm))
2417 		ret |= PR_MDWE_NO_INHERIT;
2418 
2419 	return ret;
2420 }
2421 
2422 static inline int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
2423 				 unsigned long arg4, unsigned long arg5)
2424 {
2425 	unsigned long current_bits;
2426 
2427 	if (arg3 || arg4 || arg5)
2428 		return -EINVAL;
2429 
2430 	if (bits & ~(PR_MDWE_REFUSE_EXEC_GAIN | PR_MDWE_NO_INHERIT))
2431 		return -EINVAL;
2432 
2433 	/* NO_INHERIT only makes sense with REFUSE_EXEC_GAIN */
2434 	if (bits & PR_MDWE_NO_INHERIT && !(bits & PR_MDWE_REFUSE_EXEC_GAIN))
2435 		return -EINVAL;
2436 
2437 	/*
2438 	 * EOPNOTSUPP might be more appropriate here in principle, but
2439 	 * existing userspace depends on EINVAL specifically.
2440 	 */
2441 	if (!arch_memory_deny_write_exec_supported())
2442 		return -EINVAL;
2443 
2444 	current_bits = get_current_mdwe();
2445 	if (current_bits && current_bits != bits)
2446 		return -EPERM; /* Cannot unset the flags */
2447 
2448 	if (bits & PR_MDWE_NO_INHERIT)
2449 		mm_flags_set(MMF_HAS_MDWE_NO_INHERIT, current->mm);
2450 	if (bits & PR_MDWE_REFUSE_EXEC_GAIN)
2451 		mm_flags_set(MMF_HAS_MDWE, current->mm);
2452 
2453 	return 0;
2454 }
2455 
2456 static inline int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
2457 				 unsigned long arg4, unsigned long arg5)
2458 {
2459 	if (arg2 || arg3 || arg4 || arg5)
2460 		return -EINVAL;
2461 	return get_current_mdwe();
2462 }
2463 
2464 static int prctl_get_auxv(void __user *addr, unsigned long len)
2465 {
2466 	struct mm_struct *mm = current->mm;
2467 	unsigned long size = min_t(unsigned long, sizeof(mm->saved_auxv), len);
2468 
2469 	if (size && copy_to_user(addr, mm->saved_auxv, size))
2470 		return -EFAULT;
2471 	return sizeof(mm->saved_auxv);
2472 }
2473 
2474 static int prctl_get_thp_disable(unsigned long arg2, unsigned long arg3,
2475 				 unsigned long arg4, unsigned long arg5)
2476 {
2477 	struct mm_struct *mm = current->mm;
2478 
2479 	if (arg2 || arg3 || arg4 || arg5)
2480 		return -EINVAL;
2481 
2482 	/* If disabled, we return "1 | flags", otherwise 0. */
2483 	if (mm_flags_test(MMF_DISABLE_THP_COMPLETELY, mm))
2484 		return 1;
2485 	else if (mm_flags_test(MMF_DISABLE_THP_EXCEPT_ADVISED, mm))
2486 		return 1 | PR_THP_DISABLE_EXCEPT_ADVISED;
2487 	return 0;
2488 }
2489 
2490 static int prctl_set_thp_disable(bool thp_disable, unsigned long flags,
2491 				 unsigned long arg4, unsigned long arg5)
2492 {
2493 	struct mm_struct *mm = current->mm;
2494 
2495 	if (arg4 || arg5)
2496 		return -EINVAL;
2497 
2498 	/* Flags are only allowed when disabling. */
2499 	if ((!thp_disable && flags) || (flags & ~PR_THP_DISABLE_EXCEPT_ADVISED))
2500 		return -EINVAL;
2501 	if (mmap_write_lock_killable(current->mm))
2502 		return -EINTR;
2503 	if (thp_disable) {
2504 		if (flags & PR_THP_DISABLE_EXCEPT_ADVISED) {
2505 			mm_flags_clear(MMF_DISABLE_THP_COMPLETELY, mm);
2506 			mm_flags_set(MMF_DISABLE_THP_EXCEPT_ADVISED, mm);
2507 		} else {
2508 			mm_flags_set(MMF_DISABLE_THP_COMPLETELY, mm);
2509 			mm_flags_clear(MMF_DISABLE_THP_EXCEPT_ADVISED, mm);
2510 		}
2511 	} else {
2512 		mm_flags_clear(MMF_DISABLE_THP_COMPLETELY, mm);
2513 		mm_flags_clear(MMF_DISABLE_THP_EXCEPT_ADVISED, mm);
2514 	}
2515 	mmap_write_unlock(current->mm);
2516 	return 0;
2517 }
2518 
2519 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2520 		unsigned long, arg4, unsigned long, arg5)
2521 {
2522 	struct task_struct *me = current;
2523 	unsigned char comm[sizeof(me->comm)];
2524 	long error;
2525 
2526 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2527 	if (error != -ENOSYS)
2528 		return error;
2529 
2530 	error = 0;
2531 	switch (option) {
2532 	case PR_SET_PDEATHSIG:
2533 		if (!valid_signal(arg2)) {
2534 			error = -EINVAL;
2535 			break;
2536 		}
2537 		/*
2538 		 * Ensure that either:
2539 		 *
2540 		 * 1. Subsequent getppid() calls reflect the parent process having died.
2541 		 * 2. forget_original_parent() will send the new me->pdeath_signal.
2542 		 *
2543 		 * Also prevent the read of me->pdeath_signal from being a data race.
2544 		 */
2545 		read_lock(&tasklist_lock);
2546 		me->pdeath_signal = arg2;
2547 		read_unlock(&tasklist_lock);
2548 		break;
2549 	case PR_GET_PDEATHSIG:
2550 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2551 		break;
2552 	case PR_GET_DUMPABLE:
2553 		error = get_dumpable(me->mm);
2554 		break;
2555 	case PR_SET_DUMPABLE:
2556 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2557 			error = -EINVAL;
2558 			break;
2559 		}
2560 		set_dumpable(me->mm, arg2);
2561 		break;
2562 
2563 	case PR_SET_UNALIGN:
2564 		error = SET_UNALIGN_CTL(me, arg2);
2565 		break;
2566 	case PR_GET_UNALIGN:
2567 		error = GET_UNALIGN_CTL(me, arg2);
2568 		break;
2569 	case PR_SET_FPEMU:
2570 		error = SET_FPEMU_CTL(me, arg2);
2571 		break;
2572 	case PR_GET_FPEMU:
2573 		error = GET_FPEMU_CTL(me, arg2);
2574 		break;
2575 	case PR_SET_FPEXC:
2576 		error = SET_FPEXC_CTL(me, arg2);
2577 		break;
2578 	case PR_GET_FPEXC:
2579 		error = GET_FPEXC_CTL(me, arg2);
2580 		break;
2581 	case PR_GET_TIMING:
2582 		error = PR_TIMING_STATISTICAL;
2583 		break;
2584 	case PR_SET_TIMING:
2585 		if (arg2 != PR_TIMING_STATISTICAL)
2586 			error = -EINVAL;
2587 		break;
2588 	case PR_SET_NAME:
2589 		comm[sizeof(me->comm) - 1] = 0;
2590 		if (strncpy_from_user(comm, (char __user *)arg2,
2591 				      sizeof(me->comm) - 1) < 0)
2592 			return -EFAULT;
2593 		set_task_comm(me, comm);
2594 		proc_comm_connector(me);
2595 		break;
2596 	case PR_GET_NAME:
2597 		get_task_comm(comm, me);
2598 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2599 			return -EFAULT;
2600 		break;
2601 	case PR_GET_ENDIAN:
2602 		error = GET_ENDIAN(me, arg2);
2603 		break;
2604 	case PR_SET_ENDIAN:
2605 		error = SET_ENDIAN(me, arg2);
2606 		break;
2607 	case PR_GET_SECCOMP:
2608 		error = prctl_get_seccomp();
2609 		break;
2610 	case PR_SET_SECCOMP:
2611 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2612 		break;
2613 	case PR_GET_TSC:
2614 		error = GET_TSC_CTL(arg2);
2615 		break;
2616 	case PR_SET_TSC:
2617 		error = SET_TSC_CTL(arg2);
2618 		break;
2619 	case PR_TASK_PERF_EVENTS_DISABLE:
2620 		error = perf_event_task_disable();
2621 		break;
2622 	case PR_TASK_PERF_EVENTS_ENABLE:
2623 		error = perf_event_task_enable();
2624 		break;
2625 	case PR_GET_TIMERSLACK:
2626 		if (current->timer_slack_ns > ULONG_MAX)
2627 			error = ULONG_MAX;
2628 		else
2629 			error = current->timer_slack_ns;
2630 		break;
2631 	case PR_SET_TIMERSLACK:
2632 		if (rt_or_dl_task_policy(current))
2633 			break;
2634 		if (arg2 <= 0)
2635 			current->timer_slack_ns =
2636 					current->default_timer_slack_ns;
2637 		else
2638 			current->timer_slack_ns = arg2;
2639 		break;
2640 	case PR_MCE_KILL:
2641 		if (arg4 | arg5)
2642 			return -EINVAL;
2643 		switch (arg2) {
2644 		case PR_MCE_KILL_CLEAR:
2645 			if (arg3 != 0)
2646 				return -EINVAL;
2647 			current->flags &= ~PF_MCE_PROCESS;
2648 			break;
2649 		case PR_MCE_KILL_SET:
2650 			current->flags |= PF_MCE_PROCESS;
2651 			if (arg3 == PR_MCE_KILL_EARLY)
2652 				current->flags |= PF_MCE_EARLY;
2653 			else if (arg3 == PR_MCE_KILL_LATE)
2654 				current->flags &= ~PF_MCE_EARLY;
2655 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2656 				current->flags &=
2657 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2658 			else
2659 				return -EINVAL;
2660 			break;
2661 		default:
2662 			return -EINVAL;
2663 		}
2664 		break;
2665 	case PR_MCE_KILL_GET:
2666 		if (arg2 | arg3 | arg4 | arg5)
2667 			return -EINVAL;
2668 		if (current->flags & PF_MCE_PROCESS)
2669 			error = (current->flags & PF_MCE_EARLY) ?
2670 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2671 		else
2672 			error = PR_MCE_KILL_DEFAULT;
2673 		break;
2674 	case PR_SET_MM:
2675 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2676 		break;
2677 	case PR_GET_TID_ADDRESS:
2678 		error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2679 		break;
2680 	case PR_SET_CHILD_SUBREAPER:
2681 		me->signal->is_child_subreaper = !!arg2;
2682 		if (!arg2)
2683 			break;
2684 
2685 		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2686 		break;
2687 	case PR_GET_CHILD_SUBREAPER:
2688 		error = put_user(me->signal->is_child_subreaper,
2689 				 (int __user *)arg2);
2690 		break;
2691 	case PR_SET_NO_NEW_PRIVS:
2692 		if (arg2 != 1 || arg3 || arg4 || arg5)
2693 			return -EINVAL;
2694 
2695 		task_set_no_new_privs(current);
2696 		break;
2697 	case PR_GET_NO_NEW_PRIVS:
2698 		if (arg2 || arg3 || arg4 || arg5)
2699 			return -EINVAL;
2700 		return task_no_new_privs(current) ? 1 : 0;
2701 	case PR_GET_THP_DISABLE:
2702 		error = prctl_get_thp_disable(arg2, arg3, arg4, arg5);
2703 		break;
2704 	case PR_SET_THP_DISABLE:
2705 		error = prctl_set_thp_disable(arg2, arg3, arg4, arg5);
2706 		break;
2707 	case PR_MPX_ENABLE_MANAGEMENT:
2708 	case PR_MPX_DISABLE_MANAGEMENT:
2709 		/* No longer implemented: */
2710 		return -EINVAL;
2711 	case PR_SET_FP_MODE:
2712 		error = SET_FP_MODE(me, arg2);
2713 		break;
2714 	case PR_GET_FP_MODE:
2715 		error = GET_FP_MODE(me);
2716 		break;
2717 	case PR_SVE_SET_VL:
2718 		error = SVE_SET_VL(arg2);
2719 		break;
2720 	case PR_SVE_GET_VL:
2721 		error = SVE_GET_VL();
2722 		break;
2723 	case PR_SME_SET_VL:
2724 		error = SME_SET_VL(arg2);
2725 		break;
2726 	case PR_SME_GET_VL:
2727 		error = SME_GET_VL();
2728 		break;
2729 	case PR_GET_SPECULATION_CTRL:
2730 		if (arg3 || arg4 || arg5)
2731 			return -EINVAL;
2732 		error = arch_prctl_spec_ctrl_get(me, arg2);
2733 		break;
2734 	case PR_SET_SPECULATION_CTRL:
2735 		if (arg4 || arg5)
2736 			return -EINVAL;
2737 		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2738 		break;
2739 	case PR_PAC_RESET_KEYS:
2740 		if (arg3 || arg4 || arg5)
2741 			return -EINVAL;
2742 		error = PAC_RESET_KEYS(me, arg2);
2743 		break;
2744 	case PR_PAC_SET_ENABLED_KEYS:
2745 		if (arg4 || arg5)
2746 			return -EINVAL;
2747 		error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2748 		break;
2749 	case PR_PAC_GET_ENABLED_KEYS:
2750 		if (arg2 || arg3 || arg4 || arg5)
2751 			return -EINVAL;
2752 		error = PAC_GET_ENABLED_KEYS(me);
2753 		break;
2754 	case PR_SET_TAGGED_ADDR_CTRL:
2755 		if (arg3 || arg4 || arg5)
2756 			return -EINVAL;
2757 		error = SET_TAGGED_ADDR_CTRL(arg2);
2758 		break;
2759 	case PR_GET_TAGGED_ADDR_CTRL:
2760 		if (arg2 || arg3 || arg4 || arg5)
2761 			return -EINVAL;
2762 		error = GET_TAGGED_ADDR_CTRL();
2763 		break;
2764 	case PR_SET_IO_FLUSHER:
2765 		if (!capable(CAP_SYS_RESOURCE))
2766 			return -EPERM;
2767 
2768 		if (arg3 || arg4 || arg5)
2769 			return -EINVAL;
2770 
2771 		if (arg2 == 1)
2772 			current->flags |= PR_IO_FLUSHER;
2773 		else if (!arg2)
2774 			current->flags &= ~PR_IO_FLUSHER;
2775 		else
2776 			return -EINVAL;
2777 		break;
2778 	case PR_GET_IO_FLUSHER:
2779 		if (!capable(CAP_SYS_RESOURCE))
2780 			return -EPERM;
2781 
2782 		if (arg2 || arg3 || arg4 || arg5)
2783 			return -EINVAL;
2784 
2785 		error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2786 		break;
2787 	case PR_SET_SYSCALL_USER_DISPATCH:
2788 		error = set_syscall_user_dispatch(arg2, arg3, arg4,
2789 						  (char __user *) arg5);
2790 		break;
2791 #ifdef CONFIG_SCHED_CORE
2792 	case PR_SCHED_CORE:
2793 		error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2794 		break;
2795 #endif
2796 	case PR_SET_MDWE:
2797 		error = prctl_set_mdwe(arg2, arg3, arg4, arg5);
2798 		break;
2799 	case PR_GET_MDWE:
2800 		error = prctl_get_mdwe(arg2, arg3, arg4, arg5);
2801 		break;
2802 	case PR_PPC_GET_DEXCR:
2803 		if (arg3 || arg4 || arg5)
2804 			return -EINVAL;
2805 		error = PPC_GET_DEXCR_ASPECT(me, arg2);
2806 		break;
2807 	case PR_PPC_SET_DEXCR:
2808 		if (arg4 || arg5)
2809 			return -EINVAL;
2810 		error = PPC_SET_DEXCR_ASPECT(me, arg2, arg3);
2811 		break;
2812 	case PR_SET_VMA:
2813 		error = prctl_set_vma(arg2, arg3, arg4, arg5);
2814 		break;
2815 	case PR_GET_AUXV:
2816 		if (arg4 || arg5)
2817 			return -EINVAL;
2818 		error = prctl_get_auxv((void __user *)arg2, arg3);
2819 		break;
2820 #ifdef CONFIG_KSM
2821 	case PR_SET_MEMORY_MERGE:
2822 		if (arg3 || arg4 || arg5)
2823 			return -EINVAL;
2824 		if (mmap_write_lock_killable(me->mm))
2825 			return -EINTR;
2826 
2827 		if (arg2)
2828 			error = ksm_enable_merge_any(me->mm);
2829 		else
2830 			error = ksm_disable_merge_any(me->mm);
2831 		mmap_write_unlock(me->mm);
2832 		break;
2833 	case PR_GET_MEMORY_MERGE:
2834 		if (arg2 || arg3 || arg4 || arg5)
2835 			return -EINVAL;
2836 
2837 		error = !!mm_flags_test(MMF_VM_MERGE_ANY, me->mm);
2838 		break;
2839 #endif
2840 	case PR_RISCV_V_SET_CONTROL:
2841 		error = RISCV_V_SET_CONTROL(arg2);
2842 		break;
2843 	case PR_RISCV_V_GET_CONTROL:
2844 		error = RISCV_V_GET_CONTROL();
2845 		break;
2846 	case PR_RISCV_SET_ICACHE_FLUSH_CTX:
2847 		error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
2848 		break;
2849 	case PR_GET_SHADOW_STACK_STATUS:
2850 		if (arg3 || arg4 || arg5)
2851 			return -EINVAL;
2852 		error = arch_get_shadow_stack_status(me, (unsigned long __user *) arg2);
2853 		break;
2854 	case PR_SET_SHADOW_STACK_STATUS:
2855 		if (arg3 || arg4 || arg5)
2856 			return -EINVAL;
2857 		error = arch_set_shadow_stack_status(me, arg2);
2858 		break;
2859 	case PR_LOCK_SHADOW_STACK_STATUS:
2860 		if (arg3 || arg4 || arg5)
2861 			return -EINVAL;
2862 		error = arch_lock_shadow_stack_status(me, arg2);
2863 		break;
2864 	case PR_TIMER_CREATE_RESTORE_IDS:
2865 		if (arg3 || arg4 || arg5)
2866 			return -EINVAL;
2867 		error = posixtimer_create_prctl(arg2);
2868 		break;
2869 	case PR_FUTEX_HASH:
2870 		error = futex_hash_prctl(arg2, arg3, arg4);
2871 		break;
2872 	case PR_RSEQ_SLICE_EXTENSION:
2873 		if (arg4 || arg5)
2874 			return -EINVAL;
2875 		error = rseq_slice_extension_prctl(arg2, arg3);
2876 		break;
2877 	default:
2878 		trace_task_prctl_unknown(option, arg2, arg3, arg4, arg5);
2879 		error = -EINVAL;
2880 		break;
2881 	}
2882 	return error;
2883 }
2884 
2885 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2886 		struct getcpu_cache __user *, unused)
2887 {
2888 	int err = 0;
2889 	int cpu = raw_smp_processor_id();
2890 
2891 	if (cpup)
2892 		err |= put_user(cpu, cpup);
2893 	if (nodep)
2894 		err |= put_user(cpu_to_node(cpu), nodep);
2895 	return err ? -EFAULT : 0;
2896 }
2897 
2898 /**
2899  * do_sysinfo - fill in sysinfo struct
2900  * @info: pointer to buffer to fill
2901  */
2902 static int do_sysinfo(struct sysinfo *info)
2903 {
2904 	unsigned long mem_total, sav_total;
2905 	unsigned int mem_unit, bitcount;
2906 	struct timespec64 tp;
2907 
2908 	memset(info, 0, sizeof(struct sysinfo));
2909 
2910 	ktime_get_boottime_ts64(&tp);
2911 	timens_add_boottime(&tp);
2912 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2913 
2914 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2915 
2916 	info->procs = nr_threads;
2917 
2918 	si_meminfo(info);
2919 	si_swapinfo(info);
2920 
2921 	/*
2922 	 * If the sum of all the available memory (i.e. ram + swap)
2923 	 * is less than can be stored in a 32 bit unsigned long then
2924 	 * we can be binary compatible with 2.2.x kernels.  If not,
2925 	 * well, in that case 2.2.x was broken anyways...
2926 	 *
2927 	 *  -Erik Andersen <andersee@debian.org>
2928 	 */
2929 
2930 	mem_total = info->totalram + info->totalswap;
2931 	if (mem_total < info->totalram || mem_total < info->totalswap)
2932 		goto out;
2933 	bitcount = 0;
2934 	mem_unit = info->mem_unit;
2935 	while (mem_unit > 1) {
2936 		bitcount++;
2937 		mem_unit >>= 1;
2938 		sav_total = mem_total;
2939 		mem_total <<= 1;
2940 		if (mem_total < sav_total)
2941 			goto out;
2942 	}
2943 
2944 	/*
2945 	 * If mem_total did not overflow, multiply all memory values by
2946 	 * info->mem_unit and set it to 1.  This leaves things compatible
2947 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2948 	 * kernels...
2949 	 */
2950 
2951 	info->mem_unit = 1;
2952 	info->totalram <<= bitcount;
2953 	info->freeram <<= bitcount;
2954 	info->sharedram <<= bitcount;
2955 	info->bufferram <<= bitcount;
2956 	info->totalswap <<= bitcount;
2957 	info->freeswap <<= bitcount;
2958 	info->totalhigh <<= bitcount;
2959 	info->freehigh <<= bitcount;
2960 
2961 out:
2962 	return 0;
2963 }
2964 
2965 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2966 {
2967 	struct sysinfo val;
2968 
2969 	do_sysinfo(&val);
2970 
2971 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2972 		return -EFAULT;
2973 
2974 	return 0;
2975 }
2976 
2977 #ifdef CONFIG_COMPAT
2978 struct compat_sysinfo {
2979 	s32 uptime;
2980 	u32 loads[3];
2981 	u32 totalram;
2982 	u32 freeram;
2983 	u32 sharedram;
2984 	u32 bufferram;
2985 	u32 totalswap;
2986 	u32 freeswap;
2987 	u16 procs;
2988 	u16 pad;
2989 	u32 totalhigh;
2990 	u32 freehigh;
2991 	u32 mem_unit;
2992 	char _f[20-2*sizeof(u32)-sizeof(int)];
2993 };
2994 
2995 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2996 {
2997 	struct sysinfo s;
2998 	struct compat_sysinfo s_32;
2999 
3000 	do_sysinfo(&s);
3001 
3002 	/* Check to see if any memory value is too large for 32-bit and scale
3003 	 *  down if needed
3004 	 */
3005 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
3006 		int bitcount = 0;
3007 
3008 		while (s.mem_unit < PAGE_SIZE) {
3009 			s.mem_unit <<= 1;
3010 			bitcount++;
3011 		}
3012 
3013 		s.totalram >>= bitcount;
3014 		s.freeram >>= bitcount;
3015 		s.sharedram >>= bitcount;
3016 		s.bufferram >>= bitcount;
3017 		s.totalswap >>= bitcount;
3018 		s.freeswap >>= bitcount;
3019 		s.totalhigh >>= bitcount;
3020 		s.freehigh >>= bitcount;
3021 	}
3022 
3023 	memset(&s_32, 0, sizeof(s_32));
3024 	s_32.uptime = s.uptime;
3025 	s_32.loads[0] = s.loads[0];
3026 	s_32.loads[1] = s.loads[1];
3027 	s_32.loads[2] = s.loads[2];
3028 	s_32.totalram = s.totalram;
3029 	s_32.freeram = s.freeram;
3030 	s_32.sharedram = s.sharedram;
3031 	s_32.bufferram = s.bufferram;
3032 	s_32.totalswap = s.totalswap;
3033 	s_32.freeswap = s.freeswap;
3034 	s_32.procs = s.procs;
3035 	s_32.totalhigh = s.totalhigh;
3036 	s_32.freehigh = s.freehigh;
3037 	s_32.mem_unit = s.mem_unit;
3038 	if (copy_to_user(info, &s_32, sizeof(s_32)))
3039 		return -EFAULT;
3040 	return 0;
3041 }
3042 #endif /* CONFIG_COMPAT */
3043