xref: /linux-6.15/kernel/sys.c (revision 6b9391b5)
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 #ifndef RISCV_V_SET_CONTROL
144 # define RISCV_V_SET_CONTROL(a)		(-EINVAL)
145 #endif
146 #ifndef RISCV_V_GET_CONTROL
147 # define RISCV_V_GET_CONTROL()		(-EINVAL)
148 #endif
149 #ifndef RISCV_SET_ICACHE_FLUSH_CTX
150 # define RISCV_SET_ICACHE_FLUSH_CTX(a, b)	(-EINVAL)
151 #endif
152 
153 /*
154  * this is where the system-wide overflow UID and GID are defined, for
155  * architectures that now have 32-bit UID/GID but didn't in the past
156  */
157 
158 int overflowuid = DEFAULT_OVERFLOWUID;
159 int overflowgid = DEFAULT_OVERFLOWGID;
160 
161 EXPORT_SYMBOL(overflowuid);
162 EXPORT_SYMBOL(overflowgid);
163 
164 /*
165  * the same as above, but for filesystems which can only store a 16-bit
166  * UID and GID. as such, this is needed on all architectures
167  */
168 
169 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
170 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
171 
172 EXPORT_SYMBOL(fs_overflowuid);
173 EXPORT_SYMBOL(fs_overflowgid);
174 
175 /*
176  * Returns true if current's euid is same as p's uid or euid,
177  * or has CAP_SYS_NICE to p's user_ns.
178  *
179  * Called with rcu_read_lock, creds are safe
180  */
181 static bool set_one_prio_perm(struct task_struct *p)
182 {
183 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
184 
185 	if (uid_eq(pcred->uid,  cred->euid) ||
186 	    uid_eq(pcred->euid, cred->euid))
187 		return true;
188 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
189 		return true;
190 	return false;
191 }
192 
193 /*
194  * set the priority of a task
195  * - the caller must hold the RCU read lock
196  */
197 static int set_one_prio(struct task_struct *p, int niceval, int error)
198 {
199 	int no_nice;
200 
201 	if (!set_one_prio_perm(p)) {
202 		error = -EPERM;
203 		goto out;
204 	}
205 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
206 		error = -EACCES;
207 		goto out;
208 	}
209 	no_nice = security_task_setnice(p, niceval);
210 	if (no_nice) {
211 		error = no_nice;
212 		goto out;
213 	}
214 	if (error == -ESRCH)
215 		error = 0;
216 	set_user_nice(p, niceval);
217 out:
218 	return error;
219 }
220 
221 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
222 {
223 	struct task_struct *g, *p;
224 	struct user_struct *user;
225 	const struct cred *cred = current_cred();
226 	int error = -EINVAL;
227 	struct pid *pgrp;
228 	kuid_t uid;
229 
230 	if (which > PRIO_USER || which < PRIO_PROCESS)
231 		goto out;
232 
233 	/* normalize: avoid signed division (rounding problems) */
234 	error = -ESRCH;
235 	if (niceval < MIN_NICE)
236 		niceval = MIN_NICE;
237 	if (niceval > MAX_NICE)
238 		niceval = MAX_NICE;
239 
240 	rcu_read_lock();
241 	switch (which) {
242 	case PRIO_PROCESS:
243 		if (who)
244 			p = find_task_by_vpid(who);
245 		else
246 			p = current;
247 		if (p)
248 			error = set_one_prio(p, niceval, error);
249 		break;
250 	case PRIO_PGRP:
251 		if (who)
252 			pgrp = find_vpid(who);
253 		else
254 			pgrp = task_pgrp(current);
255 		read_lock(&tasklist_lock);
256 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
257 			error = set_one_prio(p, niceval, error);
258 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
259 		read_unlock(&tasklist_lock);
260 		break;
261 	case PRIO_USER:
262 		uid = make_kuid(cred->user_ns, who);
263 		user = cred->user;
264 		if (!who)
265 			uid = cred->uid;
266 		else if (!uid_eq(uid, cred->uid)) {
267 			user = find_user(uid);
268 			if (!user)
269 				goto out_unlock;	/* No processes for this user */
270 		}
271 		for_each_process_thread(g, p) {
272 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
273 				error = set_one_prio(p, niceval, error);
274 		}
275 		if (!uid_eq(uid, cred->uid))
276 			free_uid(user);		/* For find_user() */
277 		break;
278 	}
279 out_unlock:
280 	rcu_read_unlock();
281 out:
282 	return error;
283 }
284 
285 /*
286  * Ugh. To avoid negative return values, "getpriority()" will
287  * not return the normal nice-value, but a negated value that
288  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
289  * to stay compatible.
290  */
291 SYSCALL_DEFINE2(getpriority, int, which, int, who)
292 {
293 	struct task_struct *g, *p;
294 	struct user_struct *user;
295 	const struct cred *cred = current_cred();
296 	long niceval, retval = -ESRCH;
297 	struct pid *pgrp;
298 	kuid_t uid;
299 
300 	if (which > PRIO_USER || which < PRIO_PROCESS)
301 		return -EINVAL;
302 
303 	rcu_read_lock();
304 	switch (which) {
305 	case PRIO_PROCESS:
306 		if (who)
307 			p = find_task_by_vpid(who);
308 		else
309 			p = current;
310 		if (p) {
311 			niceval = nice_to_rlimit(task_nice(p));
312 			if (niceval > retval)
313 				retval = niceval;
314 		}
315 		break;
316 	case PRIO_PGRP:
317 		if (who)
318 			pgrp = find_vpid(who);
319 		else
320 			pgrp = task_pgrp(current);
321 		read_lock(&tasklist_lock);
322 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
323 			niceval = nice_to_rlimit(task_nice(p));
324 			if (niceval > retval)
325 				retval = niceval;
326 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
327 		read_unlock(&tasklist_lock);
328 		break;
329 	case PRIO_USER:
330 		uid = make_kuid(cred->user_ns, who);
331 		user = cred->user;
332 		if (!who)
333 			uid = cred->uid;
334 		else if (!uid_eq(uid, cred->uid)) {
335 			user = find_user(uid);
336 			if (!user)
337 				goto out_unlock;	/* No processes for this user */
338 		}
339 		for_each_process_thread(g, p) {
340 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
341 				niceval = nice_to_rlimit(task_nice(p));
342 				if (niceval > retval)
343 					retval = niceval;
344 			}
345 		}
346 		if (!uid_eq(uid, cred->uid))
347 			free_uid(user);		/* for find_user() */
348 		break;
349 	}
350 out_unlock:
351 	rcu_read_unlock();
352 
353 	return retval;
354 }
355 
356 /*
357  * Unprivileged users may change the real gid to the effective gid
358  * or vice versa.  (BSD-style)
359  *
360  * If you set the real gid at all, or set the effective gid to a value not
361  * equal to the real gid, then the saved gid is set to the new effective gid.
362  *
363  * This makes it possible for a setgid program to completely drop its
364  * privileges, which is often a useful assertion to make when you are doing
365  * a security audit over a program.
366  *
367  * The general idea is that a program which uses just setregid() will be
368  * 100% compatible with BSD.  A program which uses just setgid() will be
369  * 100% compatible with POSIX with saved IDs.
370  *
371  * SMP: There are not races, the GIDs are checked only by filesystem
372  *      operations (as far as semantic preservation is concerned).
373  */
374 #ifdef CONFIG_MULTIUSER
375 long __sys_setregid(gid_t rgid, gid_t egid)
376 {
377 	struct user_namespace *ns = current_user_ns();
378 	const struct cred *old;
379 	struct cred *new;
380 	int retval;
381 	kgid_t krgid, kegid;
382 
383 	krgid = make_kgid(ns, rgid);
384 	kegid = make_kgid(ns, egid);
385 
386 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
387 		return -EINVAL;
388 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
389 		return -EINVAL;
390 
391 	new = prepare_creds();
392 	if (!new)
393 		return -ENOMEM;
394 	old = current_cred();
395 
396 	retval = -EPERM;
397 	if (rgid != (gid_t) -1) {
398 		if (gid_eq(old->gid, krgid) ||
399 		    gid_eq(old->egid, krgid) ||
400 		    ns_capable_setid(old->user_ns, CAP_SETGID))
401 			new->gid = krgid;
402 		else
403 			goto error;
404 	}
405 	if (egid != (gid_t) -1) {
406 		if (gid_eq(old->gid, kegid) ||
407 		    gid_eq(old->egid, kegid) ||
408 		    gid_eq(old->sgid, kegid) ||
409 		    ns_capable_setid(old->user_ns, CAP_SETGID))
410 			new->egid = kegid;
411 		else
412 			goto error;
413 	}
414 
415 	if (rgid != (gid_t) -1 ||
416 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
417 		new->sgid = new->egid;
418 	new->fsgid = new->egid;
419 
420 	retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
421 	if (retval < 0)
422 		goto error;
423 
424 	return commit_creds(new);
425 
426 error:
427 	abort_creds(new);
428 	return retval;
429 }
430 
431 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
432 {
433 	return __sys_setregid(rgid, egid);
434 }
435 
436 /*
437  * setgid() is implemented like SysV w/ SAVED_IDS
438  *
439  * SMP: Same implicit races as above.
440  */
441 long __sys_setgid(gid_t gid)
442 {
443 	struct user_namespace *ns = current_user_ns();
444 	const struct cred *old;
445 	struct cred *new;
446 	int retval;
447 	kgid_t kgid;
448 
449 	kgid = make_kgid(ns, gid);
450 	if (!gid_valid(kgid))
451 		return -EINVAL;
452 
453 	new = prepare_creds();
454 	if (!new)
455 		return -ENOMEM;
456 	old = current_cred();
457 
458 	retval = -EPERM;
459 	if (ns_capable_setid(old->user_ns, CAP_SETGID))
460 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
461 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
462 		new->egid = new->fsgid = kgid;
463 	else
464 		goto error;
465 
466 	retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
467 	if (retval < 0)
468 		goto error;
469 
470 	return commit_creds(new);
471 
472 error:
473 	abort_creds(new);
474 	return retval;
475 }
476 
477 SYSCALL_DEFINE1(setgid, gid_t, gid)
478 {
479 	return __sys_setgid(gid);
480 }
481 
482 /*
483  * change the user struct in a credentials set to match the new UID
484  */
485 static int set_user(struct cred *new)
486 {
487 	struct user_struct *new_user;
488 
489 	new_user = alloc_uid(new->uid);
490 	if (!new_user)
491 		return -EAGAIN;
492 
493 	free_uid(new->user);
494 	new->user = new_user;
495 	return 0;
496 }
497 
498 static void flag_nproc_exceeded(struct cred *new)
499 {
500 	if (new->ucounts == current_ucounts())
501 		return;
502 
503 	/*
504 	 * We don't fail in case of NPROC limit excess here because too many
505 	 * poorly written programs don't check set*uid() return code, assuming
506 	 * it never fails if called by root.  We may still enforce NPROC limit
507 	 * for programs doing set*uid()+execve() by harmlessly deferring the
508 	 * failure to the execve() stage.
509 	 */
510 	if (is_rlimit_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
511 			new->user != INIT_USER)
512 		current->flags |= PF_NPROC_EXCEEDED;
513 	else
514 		current->flags &= ~PF_NPROC_EXCEEDED;
515 }
516 
517 /*
518  * Unprivileged users may change the real uid to the effective uid
519  * or vice versa.  (BSD-style)
520  *
521  * If you set the real uid at all, or set the effective uid to a value not
522  * equal to the real uid, then the saved uid is set to the new effective uid.
523  *
524  * This makes it possible for a setuid program to completely drop its
525  * privileges, which is often a useful assertion to make when you are doing
526  * a security audit over a program.
527  *
528  * The general idea is that a program which uses just setreuid() will be
529  * 100% compatible with BSD.  A program which uses just setuid() will be
530  * 100% compatible with POSIX with saved IDs.
531  */
532 long __sys_setreuid(uid_t ruid, uid_t euid)
533 {
534 	struct user_namespace *ns = current_user_ns();
535 	const struct cred *old;
536 	struct cred *new;
537 	int retval;
538 	kuid_t kruid, keuid;
539 
540 	kruid = make_kuid(ns, ruid);
541 	keuid = make_kuid(ns, euid);
542 
543 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
544 		return -EINVAL;
545 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
546 		return -EINVAL;
547 
548 	new = prepare_creds();
549 	if (!new)
550 		return -ENOMEM;
551 	old = current_cred();
552 
553 	retval = -EPERM;
554 	if (ruid != (uid_t) -1) {
555 		new->uid = kruid;
556 		if (!uid_eq(old->uid, kruid) &&
557 		    !uid_eq(old->euid, kruid) &&
558 		    !ns_capable_setid(old->user_ns, CAP_SETUID))
559 			goto error;
560 	}
561 
562 	if (euid != (uid_t) -1) {
563 		new->euid = keuid;
564 		if (!uid_eq(old->uid, keuid) &&
565 		    !uid_eq(old->euid, keuid) &&
566 		    !uid_eq(old->suid, keuid) &&
567 		    !ns_capable_setid(old->user_ns, CAP_SETUID))
568 			goto error;
569 	}
570 
571 	if (!uid_eq(new->uid, old->uid)) {
572 		retval = set_user(new);
573 		if (retval < 0)
574 			goto error;
575 	}
576 	if (ruid != (uid_t) -1 ||
577 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
578 		new->suid = new->euid;
579 	new->fsuid = new->euid;
580 
581 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
582 	if (retval < 0)
583 		goto error;
584 
585 	retval = set_cred_ucounts(new);
586 	if (retval < 0)
587 		goto error;
588 
589 	flag_nproc_exceeded(new);
590 	return commit_creds(new);
591 
592 error:
593 	abort_creds(new);
594 	return retval;
595 }
596 
597 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
598 {
599 	return __sys_setreuid(ruid, euid);
600 }
601 
602 /*
603  * setuid() is implemented like SysV with SAVED_IDS
604  *
605  * Note that SAVED_ID's is deficient in that a setuid root program
606  * like sendmail, for example, cannot set its uid to be a normal
607  * user and then switch back, because if you're root, setuid() sets
608  * the saved uid too.  If you don't like this, blame the bright people
609  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
610  * will allow a root program to temporarily drop privileges and be able to
611  * regain them by swapping the real and effective uid.
612  */
613 long __sys_setuid(uid_t uid)
614 {
615 	struct user_namespace *ns = current_user_ns();
616 	const struct cred *old;
617 	struct cred *new;
618 	int retval;
619 	kuid_t kuid;
620 
621 	kuid = make_kuid(ns, uid);
622 	if (!uid_valid(kuid))
623 		return -EINVAL;
624 
625 	new = prepare_creds();
626 	if (!new)
627 		return -ENOMEM;
628 	old = current_cred();
629 
630 	retval = -EPERM;
631 	if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
632 		new->suid = new->uid = kuid;
633 		if (!uid_eq(kuid, old->uid)) {
634 			retval = set_user(new);
635 			if (retval < 0)
636 				goto error;
637 		}
638 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
639 		goto error;
640 	}
641 
642 	new->fsuid = new->euid = kuid;
643 
644 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
645 	if (retval < 0)
646 		goto error;
647 
648 	retval = set_cred_ucounts(new);
649 	if (retval < 0)
650 		goto error;
651 
652 	flag_nproc_exceeded(new);
653 	return commit_creds(new);
654 
655 error:
656 	abort_creds(new);
657 	return retval;
658 }
659 
660 SYSCALL_DEFINE1(setuid, uid_t, uid)
661 {
662 	return __sys_setuid(uid);
663 }
664 
665 
666 /*
667  * This function implements a generic ability to update ruid, euid,
668  * and suid.  This allows you to implement the 4.4 compatible seteuid().
669  */
670 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
671 {
672 	struct user_namespace *ns = current_user_ns();
673 	const struct cred *old;
674 	struct cred *new;
675 	int retval;
676 	kuid_t kruid, keuid, ksuid;
677 	bool ruid_new, euid_new, suid_new;
678 
679 	kruid = make_kuid(ns, ruid);
680 	keuid = make_kuid(ns, euid);
681 	ksuid = make_kuid(ns, suid);
682 
683 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
684 		return -EINVAL;
685 
686 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
687 		return -EINVAL;
688 
689 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
690 		return -EINVAL;
691 
692 	old = current_cred();
693 
694 	/* check for no-op */
695 	if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
696 	    (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
697 				    uid_eq(keuid, old->fsuid))) &&
698 	    (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
699 		return 0;
700 
701 	ruid_new = ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
702 		   !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
703 	euid_new = euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
704 		   !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
705 	suid_new = suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
706 		   !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
707 	if ((ruid_new || euid_new || suid_new) &&
708 	    !ns_capable_setid(old->user_ns, CAP_SETUID))
709 		return -EPERM;
710 
711 	new = prepare_creds();
712 	if (!new)
713 		return -ENOMEM;
714 
715 	if (ruid != (uid_t) -1) {
716 		new->uid = kruid;
717 		if (!uid_eq(kruid, old->uid)) {
718 			retval = set_user(new);
719 			if (retval < 0)
720 				goto error;
721 		}
722 	}
723 	if (euid != (uid_t) -1)
724 		new->euid = keuid;
725 	if (suid != (uid_t) -1)
726 		new->suid = ksuid;
727 	new->fsuid = new->euid;
728 
729 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
730 	if (retval < 0)
731 		goto error;
732 
733 	retval = set_cred_ucounts(new);
734 	if (retval < 0)
735 		goto error;
736 
737 	flag_nproc_exceeded(new);
738 	return commit_creds(new);
739 
740 error:
741 	abort_creds(new);
742 	return retval;
743 }
744 
745 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
746 {
747 	return __sys_setresuid(ruid, euid, suid);
748 }
749 
750 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
751 {
752 	const struct cred *cred = current_cred();
753 	int retval;
754 	uid_t ruid, euid, suid;
755 
756 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
757 	euid = from_kuid_munged(cred->user_ns, cred->euid);
758 	suid = from_kuid_munged(cred->user_ns, cred->suid);
759 
760 	retval = put_user(ruid, ruidp);
761 	if (!retval) {
762 		retval = put_user(euid, euidp);
763 		if (!retval)
764 			return put_user(suid, suidp);
765 	}
766 	return retval;
767 }
768 
769 /*
770  * Same as above, but for rgid, egid, sgid.
771  */
772 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
773 {
774 	struct user_namespace *ns = current_user_ns();
775 	const struct cred *old;
776 	struct cred *new;
777 	int retval;
778 	kgid_t krgid, kegid, ksgid;
779 	bool rgid_new, egid_new, sgid_new;
780 
781 	krgid = make_kgid(ns, rgid);
782 	kegid = make_kgid(ns, egid);
783 	ksgid = make_kgid(ns, sgid);
784 
785 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
786 		return -EINVAL;
787 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
788 		return -EINVAL;
789 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
790 		return -EINVAL;
791 
792 	old = current_cred();
793 
794 	/* check for no-op */
795 	if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
796 	    (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
797 				    gid_eq(kegid, old->fsgid))) &&
798 	    (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
799 		return 0;
800 
801 	rgid_new = rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
802 		   !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
803 	egid_new = egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
804 		   !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
805 	sgid_new = sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
806 		   !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
807 	if ((rgid_new || egid_new || sgid_new) &&
808 	    !ns_capable_setid(old->user_ns, CAP_SETGID))
809 		return -EPERM;
810 
811 	new = prepare_creds();
812 	if (!new)
813 		return -ENOMEM;
814 
815 	if (rgid != (gid_t) -1)
816 		new->gid = krgid;
817 	if (egid != (gid_t) -1)
818 		new->egid = kegid;
819 	if (sgid != (gid_t) -1)
820 		new->sgid = ksgid;
821 	new->fsgid = new->egid;
822 
823 	retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
824 	if (retval < 0)
825 		goto error;
826 
827 	return commit_creds(new);
828 
829 error:
830 	abort_creds(new);
831 	return retval;
832 }
833 
834 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
835 {
836 	return __sys_setresgid(rgid, egid, sgid);
837 }
838 
839 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
840 {
841 	const struct cred *cred = current_cred();
842 	int retval;
843 	gid_t rgid, egid, sgid;
844 
845 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
846 	egid = from_kgid_munged(cred->user_ns, cred->egid);
847 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
848 
849 	retval = put_user(rgid, rgidp);
850 	if (!retval) {
851 		retval = put_user(egid, egidp);
852 		if (!retval)
853 			retval = put_user(sgid, sgidp);
854 	}
855 
856 	return retval;
857 }
858 
859 
860 /*
861  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
862  * is used for "access()" and for the NFS daemon (letting nfsd stay at
863  * whatever uid it wants to). It normally shadows "euid", except when
864  * explicitly set by setfsuid() or for access..
865  */
866 long __sys_setfsuid(uid_t uid)
867 {
868 	const struct cred *old;
869 	struct cred *new;
870 	uid_t old_fsuid;
871 	kuid_t kuid;
872 
873 	old = current_cred();
874 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
875 
876 	kuid = make_kuid(old->user_ns, uid);
877 	if (!uid_valid(kuid))
878 		return old_fsuid;
879 
880 	new = prepare_creds();
881 	if (!new)
882 		return old_fsuid;
883 
884 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
885 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
886 	    ns_capable_setid(old->user_ns, CAP_SETUID)) {
887 		if (!uid_eq(kuid, old->fsuid)) {
888 			new->fsuid = kuid;
889 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
890 				goto change_okay;
891 		}
892 	}
893 
894 	abort_creds(new);
895 	return old_fsuid;
896 
897 change_okay:
898 	commit_creds(new);
899 	return old_fsuid;
900 }
901 
902 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
903 {
904 	return __sys_setfsuid(uid);
905 }
906 
907 /*
908  * Samma på svenska..
909  */
910 long __sys_setfsgid(gid_t gid)
911 {
912 	const struct cred *old;
913 	struct cred *new;
914 	gid_t old_fsgid;
915 	kgid_t kgid;
916 
917 	old = current_cred();
918 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
919 
920 	kgid = make_kgid(old->user_ns, gid);
921 	if (!gid_valid(kgid))
922 		return old_fsgid;
923 
924 	new = prepare_creds();
925 	if (!new)
926 		return old_fsgid;
927 
928 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
929 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
930 	    ns_capable_setid(old->user_ns, CAP_SETGID)) {
931 		if (!gid_eq(kgid, old->fsgid)) {
932 			new->fsgid = kgid;
933 			if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
934 				goto change_okay;
935 		}
936 	}
937 
938 	abort_creds(new);
939 	return old_fsgid;
940 
941 change_okay:
942 	commit_creds(new);
943 	return old_fsgid;
944 }
945 
946 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
947 {
948 	return __sys_setfsgid(gid);
949 }
950 #endif /* CONFIG_MULTIUSER */
951 
952 /**
953  * sys_getpid - return the thread group id of the current process
954  *
955  * Note, despite the name, this returns the tgid not the pid.  The tgid and
956  * the pid are identical unless CLONE_THREAD was specified on clone() in
957  * which case the tgid is the same in all threads of the same group.
958  *
959  * This is SMP safe as current->tgid does not change.
960  */
961 SYSCALL_DEFINE0(getpid)
962 {
963 	return task_tgid_vnr(current);
964 }
965 
966 /* Thread ID - the internal kernel "pid" */
967 SYSCALL_DEFINE0(gettid)
968 {
969 	return task_pid_vnr(current);
970 }
971 
972 /*
973  * Accessing ->real_parent is not SMP-safe, it could
974  * change from under us. However, we can use a stale
975  * value of ->real_parent under rcu_read_lock(), see
976  * release_task()->call_rcu(delayed_put_task_struct).
977  */
978 SYSCALL_DEFINE0(getppid)
979 {
980 	int pid;
981 
982 	rcu_read_lock();
983 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
984 	rcu_read_unlock();
985 
986 	return pid;
987 }
988 
989 SYSCALL_DEFINE0(getuid)
990 {
991 	/* Only we change this so SMP safe */
992 	return from_kuid_munged(current_user_ns(), current_uid());
993 }
994 
995 SYSCALL_DEFINE0(geteuid)
996 {
997 	/* Only we change this so SMP safe */
998 	return from_kuid_munged(current_user_ns(), current_euid());
999 }
1000 
1001 SYSCALL_DEFINE0(getgid)
1002 {
1003 	/* Only we change this so SMP safe */
1004 	return from_kgid_munged(current_user_ns(), current_gid());
1005 }
1006 
1007 SYSCALL_DEFINE0(getegid)
1008 {
1009 	/* Only we change this so SMP safe */
1010 	return from_kgid_munged(current_user_ns(), current_egid());
1011 }
1012 
1013 static void do_sys_times(struct tms *tms)
1014 {
1015 	u64 tgutime, tgstime, cutime, cstime;
1016 
1017 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1018 	cutime = current->signal->cutime;
1019 	cstime = current->signal->cstime;
1020 	tms->tms_utime = nsec_to_clock_t(tgutime);
1021 	tms->tms_stime = nsec_to_clock_t(tgstime);
1022 	tms->tms_cutime = nsec_to_clock_t(cutime);
1023 	tms->tms_cstime = nsec_to_clock_t(cstime);
1024 }
1025 
1026 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1027 {
1028 	if (tbuf) {
1029 		struct tms tmp;
1030 
1031 		do_sys_times(&tmp);
1032 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1033 			return -EFAULT;
1034 	}
1035 	force_successful_syscall_return();
1036 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
1037 }
1038 
1039 #ifdef CONFIG_COMPAT
1040 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1041 {
1042 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1043 }
1044 
1045 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1046 {
1047 	if (tbuf) {
1048 		struct tms tms;
1049 		struct compat_tms tmp;
1050 
1051 		do_sys_times(&tms);
1052 		/* Convert our struct tms to the compat version. */
1053 		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1054 		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1055 		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1056 		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1057 		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1058 			return -EFAULT;
1059 	}
1060 	force_successful_syscall_return();
1061 	return compat_jiffies_to_clock_t(jiffies);
1062 }
1063 #endif
1064 
1065 /*
1066  * This needs some heavy checking ...
1067  * I just haven't the stomach for it. I also don't fully
1068  * understand sessions/pgrp etc. Let somebody who does explain it.
1069  *
1070  * OK, I think I have the protection semantics right.... this is really
1071  * only important on a multi-user system anyway, to make sure one user
1072  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1073  *
1074  * !PF_FORKNOEXEC check to conform completely to POSIX.
1075  */
1076 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1077 {
1078 	struct task_struct *p;
1079 	struct task_struct *group_leader = current->group_leader;
1080 	struct pid *pgrp;
1081 	int err;
1082 
1083 	if (!pid)
1084 		pid = task_pid_vnr(group_leader);
1085 	if (!pgid)
1086 		pgid = pid;
1087 	if (pgid < 0)
1088 		return -EINVAL;
1089 	rcu_read_lock();
1090 
1091 	/* From this point forward we keep holding onto the tasklist lock
1092 	 * so that our parent does not change from under us. -DaveM
1093 	 */
1094 	write_lock_irq(&tasklist_lock);
1095 
1096 	err = -ESRCH;
1097 	p = find_task_by_vpid(pid);
1098 	if (!p)
1099 		goto out;
1100 
1101 	err = -EINVAL;
1102 	if (!thread_group_leader(p))
1103 		goto out;
1104 
1105 	if (same_thread_group(p->real_parent, group_leader)) {
1106 		err = -EPERM;
1107 		if (task_session(p) != task_session(group_leader))
1108 			goto out;
1109 		err = -EACCES;
1110 		if (!(p->flags & PF_FORKNOEXEC))
1111 			goto out;
1112 	} else {
1113 		err = -ESRCH;
1114 		if (p != group_leader)
1115 			goto out;
1116 	}
1117 
1118 	err = -EPERM;
1119 	if (p->signal->leader)
1120 		goto out;
1121 
1122 	pgrp = task_pid(p);
1123 	if (pgid != pid) {
1124 		struct task_struct *g;
1125 
1126 		pgrp = find_vpid(pgid);
1127 		g = pid_task(pgrp, PIDTYPE_PGID);
1128 		if (!g || task_session(g) != task_session(group_leader))
1129 			goto out;
1130 	}
1131 
1132 	err = security_task_setpgid(p, pgid);
1133 	if (err)
1134 		goto out;
1135 
1136 	if (task_pgrp(p) != pgrp)
1137 		change_pid(p, PIDTYPE_PGID, pgrp);
1138 
1139 	err = 0;
1140 out:
1141 	/* All paths lead to here, thus we are safe. -DaveM */
1142 	write_unlock_irq(&tasklist_lock);
1143 	rcu_read_unlock();
1144 	return err;
1145 }
1146 
1147 static int do_getpgid(pid_t pid)
1148 {
1149 	struct task_struct *p;
1150 	struct pid *grp;
1151 	int retval;
1152 
1153 	rcu_read_lock();
1154 	if (!pid)
1155 		grp = task_pgrp(current);
1156 	else {
1157 		retval = -ESRCH;
1158 		p = find_task_by_vpid(pid);
1159 		if (!p)
1160 			goto out;
1161 		grp = task_pgrp(p);
1162 		if (!grp)
1163 			goto out;
1164 
1165 		retval = security_task_getpgid(p);
1166 		if (retval)
1167 			goto out;
1168 	}
1169 	retval = pid_vnr(grp);
1170 out:
1171 	rcu_read_unlock();
1172 	return retval;
1173 }
1174 
1175 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1176 {
1177 	return do_getpgid(pid);
1178 }
1179 
1180 #ifdef __ARCH_WANT_SYS_GETPGRP
1181 
1182 SYSCALL_DEFINE0(getpgrp)
1183 {
1184 	return do_getpgid(0);
1185 }
1186 
1187 #endif
1188 
1189 SYSCALL_DEFINE1(getsid, pid_t, pid)
1190 {
1191 	struct task_struct *p;
1192 	struct pid *sid;
1193 	int retval;
1194 
1195 	rcu_read_lock();
1196 	if (!pid)
1197 		sid = task_session(current);
1198 	else {
1199 		retval = -ESRCH;
1200 		p = find_task_by_vpid(pid);
1201 		if (!p)
1202 			goto out;
1203 		sid = task_session(p);
1204 		if (!sid)
1205 			goto out;
1206 
1207 		retval = security_task_getsid(p);
1208 		if (retval)
1209 			goto out;
1210 	}
1211 	retval = pid_vnr(sid);
1212 out:
1213 	rcu_read_unlock();
1214 	return retval;
1215 }
1216 
1217 static void set_special_pids(struct pid *pid)
1218 {
1219 	struct task_struct *curr = current->group_leader;
1220 
1221 	if (task_session(curr) != pid)
1222 		change_pid(curr, PIDTYPE_SID, pid);
1223 
1224 	if (task_pgrp(curr) != pid)
1225 		change_pid(curr, PIDTYPE_PGID, pid);
1226 }
1227 
1228 int ksys_setsid(void)
1229 {
1230 	struct task_struct *group_leader = current->group_leader;
1231 	struct pid *sid = task_pid(group_leader);
1232 	pid_t session = pid_vnr(sid);
1233 	int err = -EPERM;
1234 
1235 	write_lock_irq(&tasklist_lock);
1236 	/* Fail if I am already a session leader */
1237 	if (group_leader->signal->leader)
1238 		goto out;
1239 
1240 	/* Fail if a process group id already exists that equals the
1241 	 * proposed session id.
1242 	 */
1243 	if (pid_task(sid, PIDTYPE_PGID))
1244 		goto out;
1245 
1246 	group_leader->signal->leader = 1;
1247 	set_special_pids(sid);
1248 
1249 	proc_clear_tty(group_leader);
1250 
1251 	err = session;
1252 out:
1253 	write_unlock_irq(&tasklist_lock);
1254 	if (err > 0) {
1255 		proc_sid_connector(group_leader);
1256 		sched_autogroup_create_attach(group_leader);
1257 	}
1258 	return err;
1259 }
1260 
1261 SYSCALL_DEFINE0(setsid)
1262 {
1263 	return ksys_setsid();
1264 }
1265 
1266 DECLARE_RWSEM(uts_sem);
1267 
1268 #ifdef COMPAT_UTS_MACHINE
1269 #define override_architecture(name) \
1270 	(personality(current->personality) == PER_LINUX32 && \
1271 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1272 		      sizeof(COMPAT_UTS_MACHINE)))
1273 #else
1274 #define override_architecture(name)	0
1275 #endif
1276 
1277 /*
1278  * Work around broken programs that cannot handle "Linux 3.0".
1279  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1280  * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1281  * 2.6.60.
1282  */
1283 static int override_release(char __user *release, size_t len)
1284 {
1285 	int ret = 0;
1286 
1287 	if (current->personality & UNAME26) {
1288 		const char *rest = UTS_RELEASE;
1289 		char buf[65] = { 0 };
1290 		int ndots = 0;
1291 		unsigned v;
1292 		size_t copy;
1293 
1294 		while (*rest) {
1295 			if (*rest == '.' && ++ndots >= 3)
1296 				break;
1297 			if (!isdigit(*rest) && *rest != '.')
1298 				break;
1299 			rest++;
1300 		}
1301 		v = LINUX_VERSION_PATCHLEVEL + 60;
1302 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1303 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1304 		ret = copy_to_user(release, buf, copy + 1);
1305 	}
1306 	return ret;
1307 }
1308 
1309 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1310 {
1311 	struct new_utsname tmp;
1312 
1313 	down_read(&uts_sem);
1314 	memcpy(&tmp, utsname(), sizeof(tmp));
1315 	up_read(&uts_sem);
1316 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1317 		return -EFAULT;
1318 
1319 	if (override_release(name->release, sizeof(name->release)))
1320 		return -EFAULT;
1321 	if (override_architecture(name))
1322 		return -EFAULT;
1323 	return 0;
1324 }
1325 
1326 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1327 /*
1328  * Old cruft
1329  */
1330 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1331 {
1332 	struct old_utsname tmp;
1333 
1334 	if (!name)
1335 		return -EFAULT;
1336 
1337 	down_read(&uts_sem);
1338 	memcpy(&tmp, utsname(), sizeof(tmp));
1339 	up_read(&uts_sem);
1340 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1341 		return -EFAULT;
1342 
1343 	if (override_release(name->release, sizeof(name->release)))
1344 		return -EFAULT;
1345 	if (override_architecture(name))
1346 		return -EFAULT;
1347 	return 0;
1348 }
1349 
1350 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1351 {
1352 	struct oldold_utsname tmp;
1353 
1354 	if (!name)
1355 		return -EFAULT;
1356 
1357 	memset(&tmp, 0, sizeof(tmp));
1358 
1359 	down_read(&uts_sem);
1360 	memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1361 	memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1362 	memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1363 	memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1364 	memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1365 	up_read(&uts_sem);
1366 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1367 		return -EFAULT;
1368 
1369 	if (override_architecture(name))
1370 		return -EFAULT;
1371 	if (override_release(name->release, sizeof(name->release)))
1372 		return -EFAULT;
1373 	return 0;
1374 }
1375 #endif
1376 
1377 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1378 {
1379 	int errno;
1380 	char tmp[__NEW_UTS_LEN];
1381 
1382 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1383 		return -EPERM;
1384 
1385 	if (len < 0 || len > __NEW_UTS_LEN)
1386 		return -EINVAL;
1387 	errno = -EFAULT;
1388 	if (!copy_from_user(tmp, name, len)) {
1389 		struct new_utsname *u;
1390 
1391 		add_device_randomness(tmp, len);
1392 		down_write(&uts_sem);
1393 		u = utsname();
1394 		memcpy(u->nodename, tmp, len);
1395 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1396 		errno = 0;
1397 		uts_proc_notify(UTS_PROC_HOSTNAME);
1398 		up_write(&uts_sem);
1399 	}
1400 	return errno;
1401 }
1402 
1403 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1404 
1405 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1406 {
1407 	int i;
1408 	struct new_utsname *u;
1409 	char tmp[__NEW_UTS_LEN + 1];
1410 
1411 	if (len < 0)
1412 		return -EINVAL;
1413 	down_read(&uts_sem);
1414 	u = utsname();
1415 	i = 1 + strlen(u->nodename);
1416 	if (i > len)
1417 		i = len;
1418 	memcpy(tmp, u->nodename, i);
1419 	up_read(&uts_sem);
1420 	if (copy_to_user(name, tmp, i))
1421 		return -EFAULT;
1422 	return 0;
1423 }
1424 
1425 #endif
1426 
1427 /*
1428  * Only setdomainname; getdomainname can be implemented by calling
1429  * uname()
1430  */
1431 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1432 {
1433 	int errno;
1434 	char tmp[__NEW_UTS_LEN];
1435 
1436 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1437 		return -EPERM;
1438 	if (len < 0 || len > __NEW_UTS_LEN)
1439 		return -EINVAL;
1440 
1441 	errno = -EFAULT;
1442 	if (!copy_from_user(tmp, name, len)) {
1443 		struct new_utsname *u;
1444 
1445 		add_device_randomness(tmp, len);
1446 		down_write(&uts_sem);
1447 		u = utsname();
1448 		memcpy(u->domainname, tmp, len);
1449 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1450 		errno = 0;
1451 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1452 		up_write(&uts_sem);
1453 	}
1454 	return errno;
1455 }
1456 
1457 /* make sure you are allowed to change @tsk limits before calling this */
1458 static int do_prlimit(struct task_struct *tsk, unsigned int resource,
1459 		      struct rlimit *new_rlim, struct rlimit *old_rlim)
1460 {
1461 	struct rlimit *rlim;
1462 	int retval = 0;
1463 
1464 	if (resource >= RLIM_NLIMITS)
1465 		return -EINVAL;
1466 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1467 
1468 	if (new_rlim) {
1469 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1470 			return -EINVAL;
1471 		if (resource == RLIMIT_NOFILE &&
1472 				new_rlim->rlim_max > sysctl_nr_open)
1473 			return -EPERM;
1474 	}
1475 
1476 	/* Holding a refcount on tsk protects tsk->signal from disappearing. */
1477 	rlim = tsk->signal->rlim + resource;
1478 	task_lock(tsk->group_leader);
1479 	if (new_rlim) {
1480 		/*
1481 		 * Keep the capable check against init_user_ns until cgroups can
1482 		 * contain all limits.
1483 		 */
1484 		if (new_rlim->rlim_max > rlim->rlim_max &&
1485 				!capable(CAP_SYS_RESOURCE))
1486 			retval = -EPERM;
1487 		if (!retval)
1488 			retval = security_task_setrlimit(tsk, resource, new_rlim);
1489 	}
1490 	if (!retval) {
1491 		if (old_rlim)
1492 			*old_rlim = *rlim;
1493 		if (new_rlim)
1494 			*rlim = *new_rlim;
1495 	}
1496 	task_unlock(tsk->group_leader);
1497 
1498 	/*
1499 	 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1500 	 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1501 	 * ignores the rlimit.
1502 	 */
1503 	if (!retval && new_rlim && resource == RLIMIT_CPU &&
1504 	    new_rlim->rlim_cur != RLIM_INFINITY &&
1505 	    IS_ENABLED(CONFIG_POSIX_TIMERS)) {
1506 		/*
1507 		 * update_rlimit_cpu can fail if the task is exiting, but there
1508 		 * may be other tasks in the thread group that are not exiting,
1509 		 * and they need their cpu timers adjusted.
1510 		 *
1511 		 * The group_leader is the last task to be released, so if we
1512 		 * cannot update_rlimit_cpu on it, then the entire process is
1513 		 * exiting and we do not need to update at all.
1514 		 */
1515 		update_rlimit_cpu(tsk->group_leader, new_rlim->rlim_cur);
1516 	}
1517 
1518 	return retval;
1519 }
1520 
1521 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1522 {
1523 	struct rlimit value;
1524 	int ret;
1525 
1526 	ret = do_prlimit(current, resource, NULL, &value);
1527 	if (!ret)
1528 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1529 
1530 	return ret;
1531 }
1532 
1533 #ifdef CONFIG_COMPAT
1534 
1535 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1536 		       struct compat_rlimit __user *, rlim)
1537 {
1538 	struct rlimit r;
1539 	struct compat_rlimit r32;
1540 
1541 	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1542 		return -EFAULT;
1543 
1544 	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1545 		r.rlim_cur = RLIM_INFINITY;
1546 	else
1547 		r.rlim_cur = r32.rlim_cur;
1548 	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1549 		r.rlim_max = RLIM_INFINITY;
1550 	else
1551 		r.rlim_max = r32.rlim_max;
1552 	return do_prlimit(current, resource, &r, NULL);
1553 }
1554 
1555 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1556 		       struct compat_rlimit __user *, rlim)
1557 {
1558 	struct rlimit r;
1559 	int ret;
1560 
1561 	ret = do_prlimit(current, resource, NULL, &r);
1562 	if (!ret) {
1563 		struct compat_rlimit r32;
1564 		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1565 			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1566 		else
1567 			r32.rlim_cur = r.rlim_cur;
1568 		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1569 			r32.rlim_max = COMPAT_RLIM_INFINITY;
1570 		else
1571 			r32.rlim_max = r.rlim_max;
1572 
1573 		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1574 			return -EFAULT;
1575 	}
1576 	return ret;
1577 }
1578 
1579 #endif
1580 
1581 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1582 
1583 /*
1584  *	Back compatibility for getrlimit. Needed for some apps.
1585  */
1586 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1587 		struct rlimit __user *, rlim)
1588 {
1589 	struct rlimit x;
1590 	if (resource >= RLIM_NLIMITS)
1591 		return -EINVAL;
1592 
1593 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1594 	task_lock(current->group_leader);
1595 	x = current->signal->rlim[resource];
1596 	task_unlock(current->group_leader);
1597 	if (x.rlim_cur > 0x7FFFFFFF)
1598 		x.rlim_cur = 0x7FFFFFFF;
1599 	if (x.rlim_max > 0x7FFFFFFF)
1600 		x.rlim_max = 0x7FFFFFFF;
1601 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1602 }
1603 
1604 #ifdef CONFIG_COMPAT
1605 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1606 		       struct compat_rlimit __user *, rlim)
1607 {
1608 	struct rlimit r;
1609 
1610 	if (resource >= RLIM_NLIMITS)
1611 		return -EINVAL;
1612 
1613 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1614 	task_lock(current->group_leader);
1615 	r = current->signal->rlim[resource];
1616 	task_unlock(current->group_leader);
1617 	if (r.rlim_cur > 0x7FFFFFFF)
1618 		r.rlim_cur = 0x7FFFFFFF;
1619 	if (r.rlim_max > 0x7FFFFFFF)
1620 		r.rlim_max = 0x7FFFFFFF;
1621 
1622 	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1623 	    put_user(r.rlim_max, &rlim->rlim_max))
1624 		return -EFAULT;
1625 	return 0;
1626 }
1627 #endif
1628 
1629 #endif
1630 
1631 static inline bool rlim64_is_infinity(__u64 rlim64)
1632 {
1633 #if BITS_PER_LONG < 64
1634 	return rlim64 >= ULONG_MAX;
1635 #else
1636 	return rlim64 == RLIM64_INFINITY;
1637 #endif
1638 }
1639 
1640 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1641 {
1642 	if (rlim->rlim_cur == RLIM_INFINITY)
1643 		rlim64->rlim_cur = RLIM64_INFINITY;
1644 	else
1645 		rlim64->rlim_cur = rlim->rlim_cur;
1646 	if (rlim->rlim_max == RLIM_INFINITY)
1647 		rlim64->rlim_max = RLIM64_INFINITY;
1648 	else
1649 		rlim64->rlim_max = rlim->rlim_max;
1650 }
1651 
1652 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1653 {
1654 	if (rlim64_is_infinity(rlim64->rlim_cur))
1655 		rlim->rlim_cur = RLIM_INFINITY;
1656 	else
1657 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1658 	if (rlim64_is_infinity(rlim64->rlim_max))
1659 		rlim->rlim_max = RLIM_INFINITY;
1660 	else
1661 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1662 }
1663 
1664 /* rcu lock must be held */
1665 static int check_prlimit_permission(struct task_struct *task,
1666 				    unsigned int flags)
1667 {
1668 	const struct cred *cred = current_cred(), *tcred;
1669 	bool id_match;
1670 
1671 	if (current == task)
1672 		return 0;
1673 
1674 	tcred = __task_cred(task);
1675 	id_match = (uid_eq(cred->uid, tcred->euid) &&
1676 		    uid_eq(cred->uid, tcred->suid) &&
1677 		    uid_eq(cred->uid, tcred->uid)  &&
1678 		    gid_eq(cred->gid, tcred->egid) &&
1679 		    gid_eq(cred->gid, tcred->sgid) &&
1680 		    gid_eq(cred->gid, tcred->gid));
1681 	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1682 		return -EPERM;
1683 
1684 	return security_task_prlimit(cred, tcred, flags);
1685 }
1686 
1687 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1688 		const struct rlimit64 __user *, new_rlim,
1689 		struct rlimit64 __user *, old_rlim)
1690 {
1691 	struct rlimit64 old64, new64;
1692 	struct rlimit old, new;
1693 	struct task_struct *tsk;
1694 	unsigned int checkflags = 0;
1695 	int ret;
1696 
1697 	if (old_rlim)
1698 		checkflags |= LSM_PRLIMIT_READ;
1699 
1700 	if (new_rlim) {
1701 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1702 			return -EFAULT;
1703 		rlim64_to_rlim(&new64, &new);
1704 		checkflags |= LSM_PRLIMIT_WRITE;
1705 	}
1706 
1707 	rcu_read_lock();
1708 	tsk = pid ? find_task_by_vpid(pid) : current;
1709 	if (!tsk) {
1710 		rcu_read_unlock();
1711 		return -ESRCH;
1712 	}
1713 	ret = check_prlimit_permission(tsk, checkflags);
1714 	if (ret) {
1715 		rcu_read_unlock();
1716 		return ret;
1717 	}
1718 	get_task_struct(tsk);
1719 	rcu_read_unlock();
1720 
1721 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1722 			old_rlim ? &old : NULL);
1723 
1724 	if (!ret && old_rlim) {
1725 		rlim_to_rlim64(&old, &old64);
1726 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1727 			ret = -EFAULT;
1728 	}
1729 
1730 	put_task_struct(tsk);
1731 	return ret;
1732 }
1733 
1734 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1735 {
1736 	struct rlimit new_rlim;
1737 
1738 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1739 		return -EFAULT;
1740 	return do_prlimit(current, resource, &new_rlim, NULL);
1741 }
1742 
1743 /*
1744  * It would make sense to put struct rusage in the task_struct,
1745  * except that would make the task_struct be *really big*.  After
1746  * task_struct gets moved into malloc'ed memory, it would
1747  * make sense to do this.  It will make moving the rest of the information
1748  * a lot simpler!  (Which we're not doing right now because we're not
1749  * measuring them yet).
1750  *
1751  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1752  * races with threads incrementing their own counters.  But since word
1753  * reads are atomic, we either get new values or old values and we don't
1754  * care which for the sums.  We always take the siglock to protect reading
1755  * the c* fields from p->signal from races with exit.c updating those
1756  * fields when reaping, so a sample either gets all the additions of a
1757  * given child after it's reaped, or none so this sample is before reaping.
1758  *
1759  * Locking:
1760  * We need to take the siglock for CHILDEREN, SELF and BOTH
1761  * for  the cases current multithreaded, non-current single threaded
1762  * non-current multithreaded.  Thread traversal is now safe with
1763  * the siglock held.
1764  * Strictly speaking, we donot need to take the siglock if we are current and
1765  * single threaded,  as no one else can take our signal_struct away, no one
1766  * else can  reap the  children to update signal->c* counters, and no one else
1767  * can race with the signal-> fields. If we do not take any lock, the
1768  * signal-> fields could be read out of order while another thread was just
1769  * exiting. So we should  place a read memory barrier when we avoid the lock.
1770  * On the writer side,  write memory barrier is implied in  __exit_signal
1771  * as __exit_signal releases  the siglock spinlock after updating the signal->
1772  * fields. But we don't do this yet to keep things simple.
1773  *
1774  */
1775 
1776 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1777 {
1778 	r->ru_nvcsw += t->nvcsw;
1779 	r->ru_nivcsw += t->nivcsw;
1780 	r->ru_minflt += t->min_flt;
1781 	r->ru_majflt += t->maj_flt;
1782 	r->ru_inblock += task_io_get_inblock(t);
1783 	r->ru_oublock += task_io_get_oublock(t);
1784 }
1785 
1786 void getrusage(struct task_struct *p, int who, struct rusage *r)
1787 {
1788 	struct task_struct *t;
1789 	unsigned long flags;
1790 	u64 tgutime, tgstime, utime, stime;
1791 	unsigned long maxrss = 0;
1792 	struct signal_struct *sig = p->signal;
1793 
1794 	memset((char *)r, 0, sizeof (*r));
1795 	utime = stime = 0;
1796 
1797 	if (who == RUSAGE_THREAD) {
1798 		task_cputime_adjusted(current, &utime, &stime);
1799 		accumulate_thread_rusage(p, r);
1800 		maxrss = sig->maxrss;
1801 		goto out;
1802 	}
1803 
1804 	if (!lock_task_sighand(p, &flags))
1805 		return;
1806 
1807 	switch (who) {
1808 	case RUSAGE_BOTH:
1809 	case RUSAGE_CHILDREN:
1810 		utime = sig->cutime;
1811 		stime = sig->cstime;
1812 		r->ru_nvcsw = sig->cnvcsw;
1813 		r->ru_nivcsw = sig->cnivcsw;
1814 		r->ru_minflt = sig->cmin_flt;
1815 		r->ru_majflt = sig->cmaj_flt;
1816 		r->ru_inblock = sig->cinblock;
1817 		r->ru_oublock = sig->coublock;
1818 		maxrss = sig->cmaxrss;
1819 
1820 		if (who == RUSAGE_CHILDREN)
1821 			break;
1822 		fallthrough;
1823 
1824 	case RUSAGE_SELF:
1825 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1826 		utime += tgutime;
1827 		stime += tgstime;
1828 		r->ru_nvcsw += sig->nvcsw;
1829 		r->ru_nivcsw += sig->nivcsw;
1830 		r->ru_minflt += sig->min_flt;
1831 		r->ru_majflt += sig->maj_flt;
1832 		r->ru_inblock += sig->inblock;
1833 		r->ru_oublock += sig->oublock;
1834 		if (maxrss < sig->maxrss)
1835 			maxrss = sig->maxrss;
1836 		__for_each_thread(sig, t)
1837 			accumulate_thread_rusage(t, r);
1838 		break;
1839 
1840 	default:
1841 		BUG();
1842 	}
1843 	unlock_task_sighand(p, &flags);
1844 
1845 out:
1846 	r->ru_utime = ns_to_kernel_old_timeval(utime);
1847 	r->ru_stime = ns_to_kernel_old_timeval(stime);
1848 
1849 	if (who != RUSAGE_CHILDREN) {
1850 		struct mm_struct *mm = get_task_mm(p);
1851 
1852 		if (mm) {
1853 			setmax_mm_hiwater_rss(&maxrss, mm);
1854 			mmput(mm);
1855 		}
1856 	}
1857 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1858 }
1859 
1860 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1861 {
1862 	struct rusage r;
1863 
1864 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1865 	    who != RUSAGE_THREAD)
1866 		return -EINVAL;
1867 
1868 	getrusage(current, who, &r);
1869 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1870 }
1871 
1872 #ifdef CONFIG_COMPAT
1873 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1874 {
1875 	struct rusage r;
1876 
1877 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1878 	    who != RUSAGE_THREAD)
1879 		return -EINVAL;
1880 
1881 	getrusage(current, who, &r);
1882 	return put_compat_rusage(&r, ru);
1883 }
1884 #endif
1885 
1886 SYSCALL_DEFINE1(umask, int, mask)
1887 {
1888 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1889 	return mask;
1890 }
1891 
1892 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1893 {
1894 	struct fd exe;
1895 	struct inode *inode;
1896 	int err;
1897 
1898 	exe = fdget(fd);
1899 	if (!exe.file)
1900 		return -EBADF;
1901 
1902 	inode = file_inode(exe.file);
1903 
1904 	/*
1905 	 * Because the original mm->exe_file points to executable file, make
1906 	 * sure that this one is executable as well, to avoid breaking an
1907 	 * overall picture.
1908 	 */
1909 	err = -EACCES;
1910 	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1911 		goto exit;
1912 
1913 	err = file_permission(exe.file, MAY_EXEC);
1914 	if (err)
1915 		goto exit;
1916 
1917 	err = replace_mm_exe_file(mm, exe.file);
1918 exit:
1919 	fdput(exe);
1920 	return err;
1921 }
1922 
1923 /*
1924  * Check arithmetic relations of passed addresses.
1925  *
1926  * WARNING: we don't require any capability here so be very careful
1927  * in what is allowed for modification from userspace.
1928  */
1929 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1930 {
1931 	unsigned long mmap_max_addr = TASK_SIZE;
1932 	int error = -EINVAL, i;
1933 
1934 	static const unsigned char offsets[] = {
1935 		offsetof(struct prctl_mm_map, start_code),
1936 		offsetof(struct prctl_mm_map, end_code),
1937 		offsetof(struct prctl_mm_map, start_data),
1938 		offsetof(struct prctl_mm_map, end_data),
1939 		offsetof(struct prctl_mm_map, start_brk),
1940 		offsetof(struct prctl_mm_map, brk),
1941 		offsetof(struct prctl_mm_map, start_stack),
1942 		offsetof(struct prctl_mm_map, arg_start),
1943 		offsetof(struct prctl_mm_map, arg_end),
1944 		offsetof(struct prctl_mm_map, env_start),
1945 		offsetof(struct prctl_mm_map, env_end),
1946 	};
1947 
1948 	/*
1949 	 * Make sure the members are not somewhere outside
1950 	 * of allowed address space.
1951 	 */
1952 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1953 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1954 
1955 		if ((unsigned long)val >= mmap_max_addr ||
1956 		    (unsigned long)val < mmap_min_addr)
1957 			goto out;
1958 	}
1959 
1960 	/*
1961 	 * Make sure the pairs are ordered.
1962 	 */
1963 #define __prctl_check_order(__m1, __op, __m2)				\
1964 	((unsigned long)prctl_map->__m1 __op				\
1965 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1966 	error  = __prctl_check_order(start_code, <, end_code);
1967 	error |= __prctl_check_order(start_data,<=, end_data);
1968 	error |= __prctl_check_order(start_brk, <=, brk);
1969 	error |= __prctl_check_order(arg_start, <=, arg_end);
1970 	error |= __prctl_check_order(env_start, <=, env_end);
1971 	if (error)
1972 		goto out;
1973 #undef __prctl_check_order
1974 
1975 	error = -EINVAL;
1976 
1977 	/*
1978 	 * Neither we should allow to override limits if they set.
1979 	 */
1980 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1981 			      prctl_map->start_brk, prctl_map->end_data,
1982 			      prctl_map->start_data))
1983 			goto out;
1984 
1985 	error = 0;
1986 out:
1987 	return error;
1988 }
1989 
1990 #ifdef CONFIG_CHECKPOINT_RESTORE
1991 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1992 {
1993 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1994 	unsigned long user_auxv[AT_VECTOR_SIZE];
1995 	struct mm_struct *mm = current->mm;
1996 	int error;
1997 
1998 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1999 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
2000 
2001 	if (opt == PR_SET_MM_MAP_SIZE)
2002 		return put_user((unsigned int)sizeof(prctl_map),
2003 				(unsigned int __user *)addr);
2004 
2005 	if (data_size != sizeof(prctl_map))
2006 		return -EINVAL;
2007 
2008 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2009 		return -EFAULT;
2010 
2011 	error = validate_prctl_map_addr(&prctl_map);
2012 	if (error)
2013 		return error;
2014 
2015 	if (prctl_map.auxv_size) {
2016 		/*
2017 		 * Someone is trying to cheat the auxv vector.
2018 		 */
2019 		if (!prctl_map.auxv ||
2020 				prctl_map.auxv_size > sizeof(mm->saved_auxv))
2021 			return -EINVAL;
2022 
2023 		memset(user_auxv, 0, sizeof(user_auxv));
2024 		if (copy_from_user(user_auxv,
2025 				   (const void __user *)prctl_map.auxv,
2026 				   prctl_map.auxv_size))
2027 			return -EFAULT;
2028 
2029 		/* Last entry must be AT_NULL as specification requires */
2030 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2031 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2032 	}
2033 
2034 	if (prctl_map.exe_fd != (u32)-1) {
2035 		/*
2036 		 * Check if the current user is checkpoint/restore capable.
2037 		 * At the time of this writing, it checks for CAP_SYS_ADMIN
2038 		 * or CAP_CHECKPOINT_RESTORE.
2039 		 * Note that a user with access to ptrace can masquerade an
2040 		 * arbitrary program as any executable, even setuid ones.
2041 		 * This may have implications in the tomoyo subsystem.
2042 		 */
2043 		if (!checkpoint_restore_ns_capable(current_user_ns()))
2044 			return -EPERM;
2045 
2046 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2047 		if (error)
2048 			return error;
2049 	}
2050 
2051 	/*
2052 	 * arg_lock protects concurrent updates but we still need mmap_lock for
2053 	 * read to exclude races with sys_brk.
2054 	 */
2055 	mmap_read_lock(mm);
2056 
2057 	/*
2058 	 * We don't validate if these members are pointing to
2059 	 * real present VMAs because application may have correspond
2060 	 * VMAs already unmapped and kernel uses these members for statistics
2061 	 * output in procfs mostly, except
2062 	 *
2063 	 *  - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2064 	 *    for VMAs when updating these members so anything wrong written
2065 	 *    here cause kernel to swear at userspace program but won't lead
2066 	 *    to any problem in kernel itself
2067 	 */
2068 
2069 	spin_lock(&mm->arg_lock);
2070 	mm->start_code	= prctl_map.start_code;
2071 	mm->end_code	= prctl_map.end_code;
2072 	mm->start_data	= prctl_map.start_data;
2073 	mm->end_data	= prctl_map.end_data;
2074 	mm->start_brk	= prctl_map.start_brk;
2075 	mm->brk		= prctl_map.brk;
2076 	mm->start_stack	= prctl_map.start_stack;
2077 	mm->arg_start	= prctl_map.arg_start;
2078 	mm->arg_end	= prctl_map.arg_end;
2079 	mm->env_start	= prctl_map.env_start;
2080 	mm->env_end	= prctl_map.env_end;
2081 	spin_unlock(&mm->arg_lock);
2082 
2083 	/*
2084 	 * Note this update of @saved_auxv is lockless thus
2085 	 * if someone reads this member in procfs while we're
2086 	 * updating -- it may get partly updated results. It's
2087 	 * known and acceptable trade off: we leave it as is to
2088 	 * not introduce additional locks here making the kernel
2089 	 * more complex.
2090 	 */
2091 	if (prctl_map.auxv_size)
2092 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2093 
2094 	mmap_read_unlock(mm);
2095 	return 0;
2096 }
2097 #endif /* CONFIG_CHECKPOINT_RESTORE */
2098 
2099 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2100 			  unsigned long len)
2101 {
2102 	/*
2103 	 * This doesn't move the auxiliary vector itself since it's pinned to
2104 	 * mm_struct, but it permits filling the vector with new values.  It's
2105 	 * up to the caller to provide sane values here, otherwise userspace
2106 	 * tools which use this vector might be unhappy.
2107 	 */
2108 	unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2109 
2110 	if (len > sizeof(user_auxv))
2111 		return -EINVAL;
2112 
2113 	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2114 		return -EFAULT;
2115 
2116 	/* Make sure the last entry is always AT_NULL */
2117 	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2118 	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2119 
2120 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2121 
2122 	task_lock(current);
2123 	memcpy(mm->saved_auxv, user_auxv, len);
2124 	task_unlock(current);
2125 
2126 	return 0;
2127 }
2128 
2129 static int prctl_set_mm(int opt, unsigned long addr,
2130 			unsigned long arg4, unsigned long arg5)
2131 {
2132 	struct mm_struct *mm = current->mm;
2133 	struct prctl_mm_map prctl_map = {
2134 		.auxv = NULL,
2135 		.auxv_size = 0,
2136 		.exe_fd = -1,
2137 	};
2138 	struct vm_area_struct *vma;
2139 	int error;
2140 
2141 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2142 			      opt != PR_SET_MM_MAP &&
2143 			      opt != PR_SET_MM_MAP_SIZE)))
2144 		return -EINVAL;
2145 
2146 #ifdef CONFIG_CHECKPOINT_RESTORE
2147 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2148 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2149 #endif
2150 
2151 	if (!capable(CAP_SYS_RESOURCE))
2152 		return -EPERM;
2153 
2154 	if (opt == PR_SET_MM_EXE_FILE)
2155 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2156 
2157 	if (opt == PR_SET_MM_AUXV)
2158 		return prctl_set_auxv(mm, addr, arg4);
2159 
2160 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2161 		return -EINVAL;
2162 
2163 	error = -EINVAL;
2164 
2165 	/*
2166 	 * arg_lock protects concurrent updates of arg boundaries, we need
2167 	 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2168 	 * validation.
2169 	 */
2170 	mmap_read_lock(mm);
2171 	vma = find_vma(mm, addr);
2172 
2173 	spin_lock(&mm->arg_lock);
2174 	prctl_map.start_code	= mm->start_code;
2175 	prctl_map.end_code	= mm->end_code;
2176 	prctl_map.start_data	= mm->start_data;
2177 	prctl_map.end_data	= mm->end_data;
2178 	prctl_map.start_brk	= mm->start_brk;
2179 	prctl_map.brk		= mm->brk;
2180 	prctl_map.start_stack	= mm->start_stack;
2181 	prctl_map.arg_start	= mm->arg_start;
2182 	prctl_map.arg_end	= mm->arg_end;
2183 	prctl_map.env_start	= mm->env_start;
2184 	prctl_map.env_end	= mm->env_end;
2185 
2186 	switch (opt) {
2187 	case PR_SET_MM_START_CODE:
2188 		prctl_map.start_code = addr;
2189 		break;
2190 	case PR_SET_MM_END_CODE:
2191 		prctl_map.end_code = addr;
2192 		break;
2193 	case PR_SET_MM_START_DATA:
2194 		prctl_map.start_data = addr;
2195 		break;
2196 	case PR_SET_MM_END_DATA:
2197 		prctl_map.end_data = addr;
2198 		break;
2199 	case PR_SET_MM_START_STACK:
2200 		prctl_map.start_stack = addr;
2201 		break;
2202 	case PR_SET_MM_START_BRK:
2203 		prctl_map.start_brk = addr;
2204 		break;
2205 	case PR_SET_MM_BRK:
2206 		prctl_map.brk = addr;
2207 		break;
2208 	case PR_SET_MM_ARG_START:
2209 		prctl_map.arg_start = addr;
2210 		break;
2211 	case PR_SET_MM_ARG_END:
2212 		prctl_map.arg_end = addr;
2213 		break;
2214 	case PR_SET_MM_ENV_START:
2215 		prctl_map.env_start = addr;
2216 		break;
2217 	case PR_SET_MM_ENV_END:
2218 		prctl_map.env_end = addr;
2219 		break;
2220 	default:
2221 		goto out;
2222 	}
2223 
2224 	error = validate_prctl_map_addr(&prctl_map);
2225 	if (error)
2226 		goto out;
2227 
2228 	switch (opt) {
2229 	/*
2230 	 * If command line arguments and environment
2231 	 * are placed somewhere else on stack, we can
2232 	 * set them up here, ARG_START/END to setup
2233 	 * command line arguments and ENV_START/END
2234 	 * for environment.
2235 	 */
2236 	case PR_SET_MM_START_STACK:
2237 	case PR_SET_MM_ARG_START:
2238 	case PR_SET_MM_ARG_END:
2239 	case PR_SET_MM_ENV_START:
2240 	case PR_SET_MM_ENV_END:
2241 		if (!vma) {
2242 			error = -EFAULT;
2243 			goto out;
2244 		}
2245 	}
2246 
2247 	mm->start_code	= prctl_map.start_code;
2248 	mm->end_code	= prctl_map.end_code;
2249 	mm->start_data	= prctl_map.start_data;
2250 	mm->end_data	= prctl_map.end_data;
2251 	mm->start_brk	= prctl_map.start_brk;
2252 	mm->brk		= prctl_map.brk;
2253 	mm->start_stack	= prctl_map.start_stack;
2254 	mm->arg_start	= prctl_map.arg_start;
2255 	mm->arg_end	= prctl_map.arg_end;
2256 	mm->env_start	= prctl_map.env_start;
2257 	mm->env_end	= prctl_map.env_end;
2258 
2259 	error = 0;
2260 out:
2261 	spin_unlock(&mm->arg_lock);
2262 	mmap_read_unlock(mm);
2263 	return error;
2264 }
2265 
2266 #ifdef CONFIG_CHECKPOINT_RESTORE
2267 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2268 {
2269 	return put_user(me->clear_child_tid, tid_addr);
2270 }
2271 #else
2272 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2273 {
2274 	return -EINVAL;
2275 }
2276 #endif
2277 
2278 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2279 {
2280 	/*
2281 	 * If task has has_child_subreaper - all its descendants
2282 	 * already have these flag too and new descendants will
2283 	 * inherit it on fork, skip them.
2284 	 *
2285 	 * If we've found child_reaper - skip descendants in
2286 	 * it's subtree as they will never get out pidns.
2287 	 */
2288 	if (p->signal->has_child_subreaper ||
2289 	    is_child_reaper(task_pid(p)))
2290 		return 0;
2291 
2292 	p->signal->has_child_subreaper = 1;
2293 	return 1;
2294 }
2295 
2296 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2297 {
2298 	return -EINVAL;
2299 }
2300 
2301 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2302 				    unsigned long ctrl)
2303 {
2304 	return -EINVAL;
2305 }
2306 
2307 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2308 
2309 #ifdef CONFIG_ANON_VMA_NAME
2310 
2311 #define ANON_VMA_NAME_MAX_LEN		80
2312 #define ANON_VMA_NAME_INVALID_CHARS	"\\`$[]"
2313 
2314 static inline bool is_valid_name_char(char ch)
2315 {
2316 	/* printable ascii characters, excluding ANON_VMA_NAME_INVALID_CHARS */
2317 	return ch > 0x1f && ch < 0x7f &&
2318 		!strchr(ANON_VMA_NAME_INVALID_CHARS, ch);
2319 }
2320 
2321 static int prctl_set_vma(unsigned long opt, unsigned long addr,
2322 			 unsigned long size, unsigned long arg)
2323 {
2324 	struct mm_struct *mm = current->mm;
2325 	const char __user *uname;
2326 	struct anon_vma_name *anon_name = NULL;
2327 	int error;
2328 
2329 	switch (opt) {
2330 	case PR_SET_VMA_ANON_NAME:
2331 		uname = (const char __user *)arg;
2332 		if (uname) {
2333 			char *name, *pch;
2334 
2335 			name = strndup_user(uname, ANON_VMA_NAME_MAX_LEN);
2336 			if (IS_ERR(name))
2337 				return PTR_ERR(name);
2338 
2339 			for (pch = name; *pch != '\0'; pch++) {
2340 				if (!is_valid_name_char(*pch)) {
2341 					kfree(name);
2342 					return -EINVAL;
2343 				}
2344 			}
2345 			/* anon_vma has its own copy */
2346 			anon_name = anon_vma_name_alloc(name);
2347 			kfree(name);
2348 			if (!anon_name)
2349 				return -ENOMEM;
2350 
2351 		}
2352 
2353 		mmap_write_lock(mm);
2354 		error = madvise_set_anon_name(mm, addr, size, anon_name);
2355 		mmap_write_unlock(mm);
2356 		anon_vma_name_put(anon_name);
2357 		break;
2358 	default:
2359 		error = -EINVAL;
2360 	}
2361 
2362 	return error;
2363 }
2364 
2365 #else /* CONFIG_ANON_VMA_NAME */
2366 static int prctl_set_vma(unsigned long opt, unsigned long start,
2367 			 unsigned long size, unsigned long arg)
2368 {
2369 	return -EINVAL;
2370 }
2371 #endif /* CONFIG_ANON_VMA_NAME */
2372 
2373 static inline unsigned long get_current_mdwe(void)
2374 {
2375 	unsigned long ret = 0;
2376 
2377 	if (test_bit(MMF_HAS_MDWE, &current->mm->flags))
2378 		ret |= PR_MDWE_REFUSE_EXEC_GAIN;
2379 	if (test_bit(MMF_HAS_MDWE_NO_INHERIT, &current->mm->flags))
2380 		ret |= PR_MDWE_NO_INHERIT;
2381 
2382 	return ret;
2383 }
2384 
2385 static inline int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
2386 				 unsigned long arg4, unsigned long arg5)
2387 {
2388 	unsigned long current_bits;
2389 
2390 	if (arg3 || arg4 || arg5)
2391 		return -EINVAL;
2392 
2393 	if (bits & ~(PR_MDWE_REFUSE_EXEC_GAIN | PR_MDWE_NO_INHERIT))
2394 		return -EINVAL;
2395 
2396 	/* NO_INHERIT only makes sense with REFUSE_EXEC_GAIN */
2397 	if (bits & PR_MDWE_NO_INHERIT && !(bits & PR_MDWE_REFUSE_EXEC_GAIN))
2398 		return -EINVAL;
2399 
2400 	/* PARISC cannot allow mdwe as it needs writable stacks */
2401 	if (IS_ENABLED(CONFIG_PARISC))
2402 		return -EINVAL;
2403 
2404 	current_bits = get_current_mdwe();
2405 	if (current_bits && current_bits != bits)
2406 		return -EPERM; /* Cannot unset the flags */
2407 
2408 	if (bits & PR_MDWE_NO_INHERIT)
2409 		set_bit(MMF_HAS_MDWE_NO_INHERIT, &current->mm->flags);
2410 	if (bits & PR_MDWE_REFUSE_EXEC_GAIN)
2411 		set_bit(MMF_HAS_MDWE, &current->mm->flags);
2412 
2413 	return 0;
2414 }
2415 
2416 static inline int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
2417 				 unsigned long arg4, unsigned long arg5)
2418 {
2419 	if (arg2 || arg3 || arg4 || arg5)
2420 		return -EINVAL;
2421 	return get_current_mdwe();
2422 }
2423 
2424 static int prctl_get_auxv(void __user *addr, unsigned long len)
2425 {
2426 	struct mm_struct *mm = current->mm;
2427 	unsigned long size = min_t(unsigned long, sizeof(mm->saved_auxv), len);
2428 
2429 	if (size && copy_to_user(addr, mm->saved_auxv, size))
2430 		return -EFAULT;
2431 	return sizeof(mm->saved_auxv);
2432 }
2433 
2434 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2435 		unsigned long, arg4, unsigned long, arg5)
2436 {
2437 	struct task_struct *me = current;
2438 	unsigned char comm[sizeof(me->comm)];
2439 	long error;
2440 
2441 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2442 	if (error != -ENOSYS)
2443 		return error;
2444 
2445 	error = 0;
2446 	switch (option) {
2447 	case PR_SET_PDEATHSIG:
2448 		if (!valid_signal(arg2)) {
2449 			error = -EINVAL;
2450 			break;
2451 		}
2452 		me->pdeath_signal = arg2;
2453 		break;
2454 	case PR_GET_PDEATHSIG:
2455 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2456 		break;
2457 	case PR_GET_DUMPABLE:
2458 		error = get_dumpable(me->mm);
2459 		break;
2460 	case PR_SET_DUMPABLE:
2461 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2462 			error = -EINVAL;
2463 			break;
2464 		}
2465 		set_dumpable(me->mm, arg2);
2466 		break;
2467 
2468 	case PR_SET_UNALIGN:
2469 		error = SET_UNALIGN_CTL(me, arg2);
2470 		break;
2471 	case PR_GET_UNALIGN:
2472 		error = GET_UNALIGN_CTL(me, arg2);
2473 		break;
2474 	case PR_SET_FPEMU:
2475 		error = SET_FPEMU_CTL(me, arg2);
2476 		break;
2477 	case PR_GET_FPEMU:
2478 		error = GET_FPEMU_CTL(me, arg2);
2479 		break;
2480 	case PR_SET_FPEXC:
2481 		error = SET_FPEXC_CTL(me, arg2);
2482 		break;
2483 	case PR_GET_FPEXC:
2484 		error = GET_FPEXC_CTL(me, arg2);
2485 		break;
2486 	case PR_GET_TIMING:
2487 		error = PR_TIMING_STATISTICAL;
2488 		break;
2489 	case PR_SET_TIMING:
2490 		if (arg2 != PR_TIMING_STATISTICAL)
2491 			error = -EINVAL;
2492 		break;
2493 	case PR_SET_NAME:
2494 		comm[sizeof(me->comm) - 1] = 0;
2495 		if (strncpy_from_user(comm, (char __user *)arg2,
2496 				      sizeof(me->comm) - 1) < 0)
2497 			return -EFAULT;
2498 		set_task_comm(me, comm);
2499 		proc_comm_connector(me);
2500 		break;
2501 	case PR_GET_NAME:
2502 		get_task_comm(comm, me);
2503 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2504 			return -EFAULT;
2505 		break;
2506 	case PR_GET_ENDIAN:
2507 		error = GET_ENDIAN(me, arg2);
2508 		break;
2509 	case PR_SET_ENDIAN:
2510 		error = SET_ENDIAN(me, arg2);
2511 		break;
2512 	case PR_GET_SECCOMP:
2513 		error = prctl_get_seccomp();
2514 		break;
2515 	case PR_SET_SECCOMP:
2516 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2517 		break;
2518 	case PR_GET_TSC:
2519 		error = GET_TSC_CTL(arg2);
2520 		break;
2521 	case PR_SET_TSC:
2522 		error = SET_TSC_CTL(arg2);
2523 		break;
2524 	case PR_TASK_PERF_EVENTS_DISABLE:
2525 		error = perf_event_task_disable();
2526 		break;
2527 	case PR_TASK_PERF_EVENTS_ENABLE:
2528 		error = perf_event_task_enable();
2529 		break;
2530 	case PR_GET_TIMERSLACK:
2531 		if (current->timer_slack_ns > ULONG_MAX)
2532 			error = ULONG_MAX;
2533 		else
2534 			error = current->timer_slack_ns;
2535 		break;
2536 	case PR_SET_TIMERSLACK:
2537 		if (arg2 <= 0)
2538 			current->timer_slack_ns =
2539 					current->default_timer_slack_ns;
2540 		else
2541 			current->timer_slack_ns = arg2;
2542 		break;
2543 	case PR_MCE_KILL:
2544 		if (arg4 | arg5)
2545 			return -EINVAL;
2546 		switch (arg2) {
2547 		case PR_MCE_KILL_CLEAR:
2548 			if (arg3 != 0)
2549 				return -EINVAL;
2550 			current->flags &= ~PF_MCE_PROCESS;
2551 			break;
2552 		case PR_MCE_KILL_SET:
2553 			current->flags |= PF_MCE_PROCESS;
2554 			if (arg3 == PR_MCE_KILL_EARLY)
2555 				current->flags |= PF_MCE_EARLY;
2556 			else if (arg3 == PR_MCE_KILL_LATE)
2557 				current->flags &= ~PF_MCE_EARLY;
2558 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2559 				current->flags &=
2560 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2561 			else
2562 				return -EINVAL;
2563 			break;
2564 		default:
2565 			return -EINVAL;
2566 		}
2567 		break;
2568 	case PR_MCE_KILL_GET:
2569 		if (arg2 | arg3 | arg4 | arg5)
2570 			return -EINVAL;
2571 		if (current->flags & PF_MCE_PROCESS)
2572 			error = (current->flags & PF_MCE_EARLY) ?
2573 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2574 		else
2575 			error = PR_MCE_KILL_DEFAULT;
2576 		break;
2577 	case PR_SET_MM:
2578 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2579 		break;
2580 	case PR_GET_TID_ADDRESS:
2581 		error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2582 		break;
2583 	case PR_SET_CHILD_SUBREAPER:
2584 		me->signal->is_child_subreaper = !!arg2;
2585 		if (!arg2)
2586 			break;
2587 
2588 		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2589 		break;
2590 	case PR_GET_CHILD_SUBREAPER:
2591 		error = put_user(me->signal->is_child_subreaper,
2592 				 (int __user *)arg2);
2593 		break;
2594 	case PR_SET_NO_NEW_PRIVS:
2595 		if (arg2 != 1 || arg3 || arg4 || arg5)
2596 			return -EINVAL;
2597 
2598 		task_set_no_new_privs(current);
2599 		break;
2600 	case PR_GET_NO_NEW_PRIVS:
2601 		if (arg2 || arg3 || arg4 || arg5)
2602 			return -EINVAL;
2603 		return task_no_new_privs(current) ? 1 : 0;
2604 	case PR_GET_THP_DISABLE:
2605 		if (arg2 || arg3 || arg4 || arg5)
2606 			return -EINVAL;
2607 		error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2608 		break;
2609 	case PR_SET_THP_DISABLE:
2610 		if (arg3 || arg4 || arg5)
2611 			return -EINVAL;
2612 		if (mmap_write_lock_killable(me->mm))
2613 			return -EINTR;
2614 		if (arg2)
2615 			set_bit(MMF_DISABLE_THP, &me->mm->flags);
2616 		else
2617 			clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2618 		mmap_write_unlock(me->mm);
2619 		break;
2620 	case PR_MPX_ENABLE_MANAGEMENT:
2621 	case PR_MPX_DISABLE_MANAGEMENT:
2622 		/* No longer implemented: */
2623 		return -EINVAL;
2624 	case PR_SET_FP_MODE:
2625 		error = SET_FP_MODE(me, arg2);
2626 		break;
2627 	case PR_GET_FP_MODE:
2628 		error = GET_FP_MODE(me);
2629 		break;
2630 	case PR_SVE_SET_VL:
2631 		error = SVE_SET_VL(arg2);
2632 		break;
2633 	case PR_SVE_GET_VL:
2634 		error = SVE_GET_VL();
2635 		break;
2636 	case PR_SME_SET_VL:
2637 		error = SME_SET_VL(arg2);
2638 		break;
2639 	case PR_SME_GET_VL:
2640 		error = SME_GET_VL();
2641 		break;
2642 	case PR_GET_SPECULATION_CTRL:
2643 		if (arg3 || arg4 || arg5)
2644 			return -EINVAL;
2645 		error = arch_prctl_spec_ctrl_get(me, arg2);
2646 		break;
2647 	case PR_SET_SPECULATION_CTRL:
2648 		if (arg4 || arg5)
2649 			return -EINVAL;
2650 		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2651 		break;
2652 	case PR_PAC_RESET_KEYS:
2653 		if (arg3 || arg4 || arg5)
2654 			return -EINVAL;
2655 		error = PAC_RESET_KEYS(me, arg2);
2656 		break;
2657 	case PR_PAC_SET_ENABLED_KEYS:
2658 		if (arg4 || arg5)
2659 			return -EINVAL;
2660 		error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2661 		break;
2662 	case PR_PAC_GET_ENABLED_KEYS:
2663 		if (arg2 || arg3 || arg4 || arg5)
2664 			return -EINVAL;
2665 		error = PAC_GET_ENABLED_KEYS(me);
2666 		break;
2667 	case PR_SET_TAGGED_ADDR_CTRL:
2668 		if (arg3 || arg4 || arg5)
2669 			return -EINVAL;
2670 		error = SET_TAGGED_ADDR_CTRL(arg2);
2671 		break;
2672 	case PR_GET_TAGGED_ADDR_CTRL:
2673 		if (arg2 || arg3 || arg4 || arg5)
2674 			return -EINVAL;
2675 		error = GET_TAGGED_ADDR_CTRL();
2676 		break;
2677 	case PR_SET_IO_FLUSHER:
2678 		if (!capable(CAP_SYS_RESOURCE))
2679 			return -EPERM;
2680 
2681 		if (arg3 || arg4 || arg5)
2682 			return -EINVAL;
2683 
2684 		if (arg2 == 1)
2685 			current->flags |= PR_IO_FLUSHER;
2686 		else if (!arg2)
2687 			current->flags &= ~PR_IO_FLUSHER;
2688 		else
2689 			return -EINVAL;
2690 		break;
2691 	case PR_GET_IO_FLUSHER:
2692 		if (!capable(CAP_SYS_RESOURCE))
2693 			return -EPERM;
2694 
2695 		if (arg2 || arg3 || arg4 || arg5)
2696 			return -EINVAL;
2697 
2698 		error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2699 		break;
2700 	case PR_SET_SYSCALL_USER_DISPATCH:
2701 		error = set_syscall_user_dispatch(arg2, arg3, arg4,
2702 						  (char __user *) arg5);
2703 		break;
2704 #ifdef CONFIG_SCHED_CORE
2705 	case PR_SCHED_CORE:
2706 		error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2707 		break;
2708 #endif
2709 	case PR_SET_MDWE:
2710 		error = prctl_set_mdwe(arg2, arg3, arg4, arg5);
2711 		break;
2712 	case PR_GET_MDWE:
2713 		error = prctl_get_mdwe(arg2, arg3, arg4, arg5);
2714 		break;
2715 	case PR_SET_VMA:
2716 		error = prctl_set_vma(arg2, arg3, arg4, arg5);
2717 		break;
2718 	case PR_GET_AUXV:
2719 		if (arg4 || arg5)
2720 			return -EINVAL;
2721 		error = prctl_get_auxv((void __user *)arg2, arg3);
2722 		break;
2723 #ifdef CONFIG_KSM
2724 	case PR_SET_MEMORY_MERGE:
2725 		if (arg3 || arg4 || arg5)
2726 			return -EINVAL;
2727 		if (mmap_write_lock_killable(me->mm))
2728 			return -EINTR;
2729 
2730 		if (arg2)
2731 			error = ksm_enable_merge_any(me->mm);
2732 		else
2733 			error = ksm_disable_merge_any(me->mm);
2734 		mmap_write_unlock(me->mm);
2735 		break;
2736 	case PR_GET_MEMORY_MERGE:
2737 		if (arg2 || arg3 || arg4 || arg5)
2738 			return -EINVAL;
2739 
2740 		error = !!test_bit(MMF_VM_MERGE_ANY, &me->mm->flags);
2741 		break;
2742 #endif
2743 	case PR_RISCV_V_SET_CONTROL:
2744 		error = RISCV_V_SET_CONTROL(arg2);
2745 		break;
2746 	case PR_RISCV_V_GET_CONTROL:
2747 		error = RISCV_V_GET_CONTROL();
2748 		break;
2749 	case PR_RISCV_SET_ICACHE_FLUSH_CTX:
2750 		error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
2751 		break;
2752 	default:
2753 		error = -EINVAL;
2754 		break;
2755 	}
2756 	return error;
2757 }
2758 
2759 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2760 		struct getcpu_cache __user *, unused)
2761 {
2762 	int err = 0;
2763 	int cpu = raw_smp_processor_id();
2764 
2765 	if (cpup)
2766 		err |= put_user(cpu, cpup);
2767 	if (nodep)
2768 		err |= put_user(cpu_to_node(cpu), nodep);
2769 	return err ? -EFAULT : 0;
2770 }
2771 
2772 /**
2773  * do_sysinfo - fill in sysinfo struct
2774  * @info: pointer to buffer to fill
2775  */
2776 static int do_sysinfo(struct sysinfo *info)
2777 {
2778 	unsigned long mem_total, sav_total;
2779 	unsigned int mem_unit, bitcount;
2780 	struct timespec64 tp;
2781 
2782 	memset(info, 0, sizeof(struct sysinfo));
2783 
2784 	ktime_get_boottime_ts64(&tp);
2785 	timens_add_boottime(&tp);
2786 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2787 
2788 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2789 
2790 	info->procs = nr_threads;
2791 
2792 	si_meminfo(info);
2793 	si_swapinfo(info);
2794 
2795 	/*
2796 	 * If the sum of all the available memory (i.e. ram + swap)
2797 	 * is less than can be stored in a 32 bit unsigned long then
2798 	 * we can be binary compatible with 2.2.x kernels.  If not,
2799 	 * well, in that case 2.2.x was broken anyways...
2800 	 *
2801 	 *  -Erik Andersen <[email protected]>
2802 	 */
2803 
2804 	mem_total = info->totalram + info->totalswap;
2805 	if (mem_total < info->totalram || mem_total < info->totalswap)
2806 		goto out;
2807 	bitcount = 0;
2808 	mem_unit = info->mem_unit;
2809 	while (mem_unit > 1) {
2810 		bitcount++;
2811 		mem_unit >>= 1;
2812 		sav_total = mem_total;
2813 		mem_total <<= 1;
2814 		if (mem_total < sav_total)
2815 			goto out;
2816 	}
2817 
2818 	/*
2819 	 * If mem_total did not overflow, multiply all memory values by
2820 	 * info->mem_unit and set it to 1.  This leaves things compatible
2821 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2822 	 * kernels...
2823 	 */
2824 
2825 	info->mem_unit = 1;
2826 	info->totalram <<= bitcount;
2827 	info->freeram <<= bitcount;
2828 	info->sharedram <<= bitcount;
2829 	info->bufferram <<= bitcount;
2830 	info->totalswap <<= bitcount;
2831 	info->freeswap <<= bitcount;
2832 	info->totalhigh <<= bitcount;
2833 	info->freehigh <<= bitcount;
2834 
2835 out:
2836 	return 0;
2837 }
2838 
2839 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2840 {
2841 	struct sysinfo val;
2842 
2843 	do_sysinfo(&val);
2844 
2845 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2846 		return -EFAULT;
2847 
2848 	return 0;
2849 }
2850 
2851 #ifdef CONFIG_COMPAT
2852 struct compat_sysinfo {
2853 	s32 uptime;
2854 	u32 loads[3];
2855 	u32 totalram;
2856 	u32 freeram;
2857 	u32 sharedram;
2858 	u32 bufferram;
2859 	u32 totalswap;
2860 	u32 freeswap;
2861 	u16 procs;
2862 	u16 pad;
2863 	u32 totalhigh;
2864 	u32 freehigh;
2865 	u32 mem_unit;
2866 	char _f[20-2*sizeof(u32)-sizeof(int)];
2867 };
2868 
2869 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2870 {
2871 	struct sysinfo s;
2872 	struct compat_sysinfo s_32;
2873 
2874 	do_sysinfo(&s);
2875 
2876 	/* Check to see if any memory value is too large for 32-bit and scale
2877 	 *  down if needed
2878 	 */
2879 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2880 		int bitcount = 0;
2881 
2882 		while (s.mem_unit < PAGE_SIZE) {
2883 			s.mem_unit <<= 1;
2884 			bitcount++;
2885 		}
2886 
2887 		s.totalram >>= bitcount;
2888 		s.freeram >>= bitcount;
2889 		s.sharedram >>= bitcount;
2890 		s.bufferram >>= bitcount;
2891 		s.totalswap >>= bitcount;
2892 		s.freeswap >>= bitcount;
2893 		s.totalhigh >>= bitcount;
2894 		s.freehigh >>= bitcount;
2895 	}
2896 
2897 	memset(&s_32, 0, sizeof(s_32));
2898 	s_32.uptime = s.uptime;
2899 	s_32.loads[0] = s.loads[0];
2900 	s_32.loads[1] = s.loads[1];
2901 	s_32.loads[2] = s.loads[2];
2902 	s_32.totalram = s.totalram;
2903 	s_32.freeram = s.freeram;
2904 	s_32.sharedram = s.sharedram;
2905 	s_32.bufferram = s.bufferram;
2906 	s_32.totalswap = s.totalswap;
2907 	s_32.freeswap = s.freeswap;
2908 	s_32.procs = s.procs;
2909 	s_32.totalhigh = s.totalhigh;
2910 	s_32.freehigh = s.freehigh;
2911 	s_32.mem_unit = s.mem_unit;
2912 	if (copy_to_user(info, &s_32, sizeof(s_32)))
2913 		return -EFAULT;
2914 	return 0;
2915 }
2916 #endif /* CONFIG_COMPAT */
2917