xref: /linux/kernel/sys.c (revision b31abceb0fec717948e4e8dae6705baf6914db96)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/kernel/sys.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/export.h>
9 #include <linux/mm.h>
10 #include <linux/utsname.h>
11 #include <linux/mman.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
15 #include <linux/fs.h>
16 #include <linux/kmod.h>
17 #include <linux/perf_event.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/file.h>
40 #include <linux/mount.h>
41 #include <linux/gfp.h>
42 #include <linux/syscore_ops.h>
43 #include <linux/version.h>
44 #include <linux/ctype.h>
45 
46 #include <linux/compat.h>
47 #include <linux/syscalls.h>
48 #include <linux/kprobes.h>
49 #include <linux/user_namespace.h>
50 #include <linux/binfmts.h>
51 
52 #include <linux/sched.h>
53 #include <linux/sched/autogroup.h>
54 #include <linux/sched/loadavg.h>
55 #include <linux/sched/stat.h>
56 #include <linux/sched/mm.h>
57 #include <linux/sched/coredump.h>
58 #include <linux/sched/task.h>
59 #include <linux/sched/cputime.h>
60 #include <linux/rcupdate.h>
61 #include <linux/uidgid.h>
62 #include <linux/cred.h>
63 
64 #include <linux/nospec.h>
65 
66 #include <linux/kmsg_dump.h>
67 /* Move somewhere else to avoid recompiling? */
68 #include <generated/utsrelease.h>
69 
70 #include <linux/uaccess.h>
71 #include <asm/io.h>
72 #include <asm/unistd.h>
73 
74 #include "uid16.h"
75 
76 #ifndef SET_UNALIGN_CTL
77 # define SET_UNALIGN_CTL(a, b)	(-EINVAL)
78 #endif
79 #ifndef GET_UNALIGN_CTL
80 # define GET_UNALIGN_CTL(a, b)	(-EINVAL)
81 #endif
82 #ifndef SET_FPEMU_CTL
83 # define SET_FPEMU_CTL(a, b)	(-EINVAL)
84 #endif
85 #ifndef GET_FPEMU_CTL
86 # define GET_FPEMU_CTL(a, b)	(-EINVAL)
87 #endif
88 #ifndef SET_FPEXC_CTL
89 # define SET_FPEXC_CTL(a, b)	(-EINVAL)
90 #endif
91 #ifndef GET_FPEXC_CTL
92 # define GET_FPEXC_CTL(a, b)	(-EINVAL)
93 #endif
94 #ifndef GET_ENDIAN
95 # define GET_ENDIAN(a, b)	(-EINVAL)
96 #endif
97 #ifndef SET_ENDIAN
98 # define SET_ENDIAN(a, b)	(-EINVAL)
99 #endif
100 #ifndef GET_TSC_CTL
101 # define GET_TSC_CTL(a)		(-EINVAL)
102 #endif
103 #ifndef SET_TSC_CTL
104 # define SET_TSC_CTL(a)		(-EINVAL)
105 #endif
106 #ifndef MPX_ENABLE_MANAGEMENT
107 # define MPX_ENABLE_MANAGEMENT()	(-EINVAL)
108 #endif
109 #ifndef MPX_DISABLE_MANAGEMENT
110 # define MPX_DISABLE_MANAGEMENT()	(-EINVAL)
111 #endif
112 #ifndef GET_FP_MODE
113 # define GET_FP_MODE(a)		(-EINVAL)
114 #endif
115 #ifndef SET_FP_MODE
116 # define SET_FP_MODE(a,b)	(-EINVAL)
117 #endif
118 #ifndef SVE_SET_VL
119 # define SVE_SET_VL(a)		(-EINVAL)
120 #endif
121 #ifndef SVE_GET_VL
122 # define SVE_GET_VL()		(-EINVAL)
123 #endif
124 
125 /*
126  * this is where the system-wide overflow UID and GID are defined, for
127  * architectures that now have 32-bit UID/GID but didn't in the past
128  */
129 
130 int overflowuid = DEFAULT_OVERFLOWUID;
131 int overflowgid = DEFAULT_OVERFLOWGID;
132 
133 EXPORT_SYMBOL(overflowuid);
134 EXPORT_SYMBOL(overflowgid);
135 
136 /*
137  * the same as above, but for filesystems which can only store a 16-bit
138  * UID and GID. as such, this is needed on all architectures
139  */
140 
141 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
142 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
143 
144 EXPORT_SYMBOL(fs_overflowuid);
145 EXPORT_SYMBOL(fs_overflowgid);
146 
147 /*
148  * Returns true if current's euid is same as p's uid or euid,
149  * or has CAP_SYS_NICE to p's user_ns.
150  *
151  * Called with rcu_read_lock, creds are safe
152  */
153 static bool set_one_prio_perm(struct task_struct *p)
154 {
155 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
156 
157 	if (uid_eq(pcred->uid,  cred->euid) ||
158 	    uid_eq(pcred->euid, cred->euid))
159 		return true;
160 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
161 		return true;
162 	return false;
163 }
164 
165 /*
166  * set the priority of a task
167  * - the caller must hold the RCU read lock
168  */
169 static int set_one_prio(struct task_struct *p, int niceval, int error)
170 {
171 	int no_nice;
172 
173 	if (!set_one_prio_perm(p)) {
174 		error = -EPERM;
175 		goto out;
176 	}
177 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
178 		error = -EACCES;
179 		goto out;
180 	}
181 	no_nice = security_task_setnice(p, niceval);
182 	if (no_nice) {
183 		error = no_nice;
184 		goto out;
185 	}
186 	if (error == -ESRCH)
187 		error = 0;
188 	set_user_nice(p, niceval);
189 out:
190 	return error;
191 }
192 
193 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
194 {
195 	struct task_struct *g, *p;
196 	struct user_struct *user;
197 	const struct cred *cred = current_cred();
198 	int error = -EINVAL;
199 	struct pid *pgrp;
200 	kuid_t uid;
201 
202 	if (which > PRIO_USER || which < PRIO_PROCESS)
203 		goto out;
204 
205 	/* normalize: avoid signed division (rounding problems) */
206 	error = -ESRCH;
207 	if (niceval < MIN_NICE)
208 		niceval = MIN_NICE;
209 	if (niceval > MAX_NICE)
210 		niceval = MAX_NICE;
211 
212 	rcu_read_lock();
213 	read_lock(&tasklist_lock);
214 	switch (which) {
215 	case PRIO_PROCESS:
216 		if (who)
217 			p = find_task_by_vpid(who);
218 		else
219 			p = current;
220 		if (p)
221 			error = set_one_prio(p, niceval, error);
222 		break;
223 	case PRIO_PGRP:
224 		if (who)
225 			pgrp = find_vpid(who);
226 		else
227 			pgrp = task_pgrp(current);
228 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
229 			error = set_one_prio(p, niceval, error);
230 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
231 		break;
232 	case PRIO_USER:
233 		uid = make_kuid(cred->user_ns, who);
234 		user = cred->user;
235 		if (!who)
236 			uid = cred->uid;
237 		else if (!uid_eq(uid, cred->uid)) {
238 			user = find_user(uid);
239 			if (!user)
240 				goto out_unlock;	/* No processes for this user */
241 		}
242 		do_each_thread(g, p) {
243 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
244 				error = set_one_prio(p, niceval, error);
245 		} while_each_thread(g, p);
246 		if (!uid_eq(uid, cred->uid))
247 			free_uid(user);		/* For find_user() */
248 		break;
249 	}
250 out_unlock:
251 	read_unlock(&tasklist_lock);
252 	rcu_read_unlock();
253 out:
254 	return error;
255 }
256 
257 /*
258  * Ugh. To avoid negative return values, "getpriority()" will
259  * not return the normal nice-value, but a negated value that
260  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
261  * to stay compatible.
262  */
263 SYSCALL_DEFINE2(getpriority, int, which, int, who)
264 {
265 	struct task_struct *g, *p;
266 	struct user_struct *user;
267 	const struct cred *cred = current_cred();
268 	long niceval, retval = -ESRCH;
269 	struct pid *pgrp;
270 	kuid_t uid;
271 
272 	if (which > PRIO_USER || which < PRIO_PROCESS)
273 		return -EINVAL;
274 
275 	rcu_read_lock();
276 	read_lock(&tasklist_lock);
277 	switch (which) {
278 	case PRIO_PROCESS:
279 		if (who)
280 			p = find_task_by_vpid(who);
281 		else
282 			p = current;
283 		if (p) {
284 			niceval = nice_to_rlimit(task_nice(p));
285 			if (niceval > retval)
286 				retval = niceval;
287 		}
288 		break;
289 	case PRIO_PGRP:
290 		if (who)
291 			pgrp = find_vpid(who);
292 		else
293 			pgrp = task_pgrp(current);
294 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
295 			niceval = nice_to_rlimit(task_nice(p));
296 			if (niceval > retval)
297 				retval = niceval;
298 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
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 		do_each_thread(g, p) {
311 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
312 				niceval = nice_to_rlimit(task_nice(p));
313 				if (niceval > retval)
314 					retval = niceval;
315 			}
316 		} while_each_thread(g, p);
317 		if (!uid_eq(uid, cred->uid))
318 			free_uid(user);		/* for find_user() */
319 		break;
320 	}
321 out_unlock:
322 	read_unlock(&tasklist_lock);
323 	rcu_read_unlock();
324 
325 	return retval;
326 }
327 
328 /*
329  * Unprivileged users may change the real gid to the effective gid
330  * or vice versa.  (BSD-style)
331  *
332  * If you set the real gid at all, or set the effective gid to a value not
333  * equal to the real gid, then the saved gid is set to the new effective gid.
334  *
335  * This makes it possible for a setgid program to completely drop its
336  * privileges, which is often a useful assertion to make when you are doing
337  * a security audit over a program.
338  *
339  * The general idea is that a program which uses just setregid() will be
340  * 100% compatible with BSD.  A program which uses just setgid() will be
341  * 100% compatible with POSIX with saved IDs.
342  *
343  * SMP: There are not races, the GIDs are checked only by filesystem
344  *      operations (as far as semantic preservation is concerned).
345  */
346 #ifdef CONFIG_MULTIUSER
347 long __sys_setregid(gid_t rgid, gid_t egid)
348 {
349 	struct user_namespace *ns = current_user_ns();
350 	const struct cred *old;
351 	struct cred *new;
352 	int retval;
353 	kgid_t krgid, kegid;
354 
355 	krgid = make_kgid(ns, rgid);
356 	kegid = make_kgid(ns, egid);
357 
358 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
359 		return -EINVAL;
360 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
361 		return -EINVAL;
362 
363 	new = prepare_creds();
364 	if (!new)
365 		return -ENOMEM;
366 	old = current_cred();
367 
368 	retval = -EPERM;
369 	if (rgid != (gid_t) -1) {
370 		if (gid_eq(old->gid, krgid) ||
371 		    gid_eq(old->egid, krgid) ||
372 		    ns_capable(old->user_ns, CAP_SETGID))
373 			new->gid = krgid;
374 		else
375 			goto error;
376 	}
377 	if (egid != (gid_t) -1) {
378 		if (gid_eq(old->gid, kegid) ||
379 		    gid_eq(old->egid, kegid) ||
380 		    gid_eq(old->sgid, kegid) ||
381 		    ns_capable(old->user_ns, CAP_SETGID))
382 			new->egid = kegid;
383 		else
384 			goto error;
385 	}
386 
387 	if (rgid != (gid_t) -1 ||
388 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
389 		new->sgid = new->egid;
390 	new->fsgid = new->egid;
391 
392 	return commit_creds(new);
393 
394 error:
395 	abort_creds(new);
396 	return retval;
397 }
398 
399 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
400 {
401 	return __sys_setregid(rgid, egid);
402 }
403 
404 /*
405  * setgid() is implemented like SysV w/ SAVED_IDS
406  *
407  * SMP: Same implicit races as above.
408  */
409 long __sys_setgid(gid_t gid)
410 {
411 	struct user_namespace *ns = current_user_ns();
412 	const struct cred *old;
413 	struct cred *new;
414 	int retval;
415 	kgid_t kgid;
416 
417 	kgid = make_kgid(ns, gid);
418 	if (!gid_valid(kgid))
419 		return -EINVAL;
420 
421 	new = prepare_creds();
422 	if (!new)
423 		return -ENOMEM;
424 	old = current_cred();
425 
426 	retval = -EPERM;
427 	if (ns_capable(old->user_ns, CAP_SETGID))
428 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
429 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
430 		new->egid = new->fsgid = kgid;
431 	else
432 		goto error;
433 
434 	return commit_creds(new);
435 
436 error:
437 	abort_creds(new);
438 	return retval;
439 }
440 
441 SYSCALL_DEFINE1(setgid, gid_t, gid)
442 {
443 	return __sys_setgid(gid);
444 }
445 
446 /*
447  * change the user struct in a credentials set to match the new UID
448  */
449 static int set_user(struct cred *new)
450 {
451 	struct user_struct *new_user;
452 
453 	new_user = alloc_uid(new->uid);
454 	if (!new_user)
455 		return -EAGAIN;
456 
457 	/*
458 	 * We don't fail in case of NPROC limit excess here because too many
459 	 * poorly written programs don't check set*uid() return code, assuming
460 	 * it never fails if called by root.  We may still enforce NPROC limit
461 	 * for programs doing set*uid()+execve() by harmlessly deferring the
462 	 * failure to the execve() stage.
463 	 */
464 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
465 			new_user != INIT_USER)
466 		current->flags |= PF_NPROC_EXCEEDED;
467 	else
468 		current->flags &= ~PF_NPROC_EXCEEDED;
469 
470 	free_uid(new->user);
471 	new->user = new_user;
472 	return 0;
473 }
474 
475 /*
476  * Unprivileged users may change the real uid to the effective uid
477  * or vice versa.  (BSD-style)
478  *
479  * If you set the real uid at all, or set the effective uid to a value not
480  * equal to the real uid, then the saved uid is set to the new effective uid.
481  *
482  * This makes it possible for a setuid program to completely drop its
483  * privileges, which is often a useful assertion to make when you are doing
484  * a security audit over a program.
485  *
486  * The general idea is that a program which uses just setreuid() will be
487  * 100% compatible with BSD.  A program which uses just setuid() will be
488  * 100% compatible with POSIX with saved IDs.
489  */
490 long __sys_setreuid(uid_t ruid, uid_t euid)
491 {
492 	struct user_namespace *ns = current_user_ns();
493 	const struct cred *old;
494 	struct cred *new;
495 	int retval;
496 	kuid_t kruid, keuid;
497 
498 	kruid = make_kuid(ns, ruid);
499 	keuid = make_kuid(ns, euid);
500 
501 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
502 		return -EINVAL;
503 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
504 		return -EINVAL;
505 
506 	new = prepare_creds();
507 	if (!new)
508 		return -ENOMEM;
509 	old = current_cred();
510 
511 	retval = -EPERM;
512 	if (ruid != (uid_t) -1) {
513 		new->uid = kruid;
514 		if (!uid_eq(old->uid, kruid) &&
515 		    !uid_eq(old->euid, kruid) &&
516 		    !ns_capable(old->user_ns, CAP_SETUID))
517 			goto error;
518 	}
519 
520 	if (euid != (uid_t) -1) {
521 		new->euid = keuid;
522 		if (!uid_eq(old->uid, keuid) &&
523 		    !uid_eq(old->euid, keuid) &&
524 		    !uid_eq(old->suid, keuid) &&
525 		    !ns_capable(old->user_ns, CAP_SETUID))
526 			goto error;
527 	}
528 
529 	if (!uid_eq(new->uid, old->uid)) {
530 		retval = set_user(new);
531 		if (retval < 0)
532 			goto error;
533 	}
534 	if (ruid != (uid_t) -1 ||
535 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
536 		new->suid = new->euid;
537 	new->fsuid = new->euid;
538 
539 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
540 	if (retval < 0)
541 		goto error;
542 
543 	return commit_creds(new);
544 
545 error:
546 	abort_creds(new);
547 	return retval;
548 }
549 
550 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
551 {
552 	return __sys_setreuid(ruid, euid);
553 }
554 
555 /*
556  * setuid() is implemented like SysV with SAVED_IDS
557  *
558  * Note that SAVED_ID's is deficient in that a setuid root program
559  * like sendmail, for example, cannot set its uid to be a normal
560  * user and then switch back, because if you're root, setuid() sets
561  * the saved uid too.  If you don't like this, blame the bright people
562  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
563  * will allow a root program to temporarily drop privileges and be able to
564  * regain them by swapping the real and effective uid.
565  */
566 long __sys_setuid(uid_t uid)
567 {
568 	struct user_namespace *ns = current_user_ns();
569 	const struct cred *old;
570 	struct cred *new;
571 	int retval;
572 	kuid_t kuid;
573 
574 	kuid = make_kuid(ns, uid);
575 	if (!uid_valid(kuid))
576 		return -EINVAL;
577 
578 	new = prepare_creds();
579 	if (!new)
580 		return -ENOMEM;
581 	old = current_cred();
582 
583 	retval = -EPERM;
584 	if (ns_capable(old->user_ns, CAP_SETUID)) {
585 		new->suid = new->uid = kuid;
586 		if (!uid_eq(kuid, old->uid)) {
587 			retval = set_user(new);
588 			if (retval < 0)
589 				goto error;
590 		}
591 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
592 		goto error;
593 	}
594 
595 	new->fsuid = new->euid = kuid;
596 
597 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
598 	if (retval < 0)
599 		goto error;
600 
601 	return commit_creds(new);
602 
603 error:
604 	abort_creds(new);
605 	return retval;
606 }
607 
608 SYSCALL_DEFINE1(setuid, uid_t, uid)
609 {
610 	return __sys_setuid(uid);
611 }
612 
613 
614 /*
615  * This function implements a generic ability to update ruid, euid,
616  * and suid.  This allows you to implement the 4.4 compatible seteuid().
617  */
618 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
619 {
620 	struct user_namespace *ns = current_user_ns();
621 	const struct cred *old;
622 	struct cred *new;
623 	int retval;
624 	kuid_t kruid, keuid, ksuid;
625 
626 	kruid = make_kuid(ns, ruid);
627 	keuid = make_kuid(ns, euid);
628 	ksuid = make_kuid(ns, suid);
629 
630 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
631 		return -EINVAL;
632 
633 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
634 		return -EINVAL;
635 
636 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
637 		return -EINVAL;
638 
639 	new = prepare_creds();
640 	if (!new)
641 		return -ENOMEM;
642 
643 	old = current_cred();
644 
645 	retval = -EPERM;
646 	if (!ns_capable(old->user_ns, CAP_SETUID)) {
647 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
648 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
649 			goto error;
650 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
651 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
652 			goto error;
653 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
654 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
655 			goto error;
656 	}
657 
658 	if (ruid != (uid_t) -1) {
659 		new->uid = kruid;
660 		if (!uid_eq(kruid, old->uid)) {
661 			retval = set_user(new);
662 			if (retval < 0)
663 				goto error;
664 		}
665 	}
666 	if (euid != (uid_t) -1)
667 		new->euid = keuid;
668 	if (suid != (uid_t) -1)
669 		new->suid = ksuid;
670 	new->fsuid = new->euid;
671 
672 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
673 	if (retval < 0)
674 		goto error;
675 
676 	return commit_creds(new);
677 
678 error:
679 	abort_creds(new);
680 	return retval;
681 }
682 
683 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
684 {
685 	return __sys_setresuid(ruid, euid, suid);
686 }
687 
688 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
689 {
690 	const struct cred *cred = current_cred();
691 	int retval;
692 	uid_t ruid, euid, suid;
693 
694 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
695 	euid = from_kuid_munged(cred->user_ns, cred->euid);
696 	suid = from_kuid_munged(cred->user_ns, cred->suid);
697 
698 	retval = put_user(ruid, ruidp);
699 	if (!retval) {
700 		retval = put_user(euid, euidp);
701 		if (!retval)
702 			return put_user(suid, suidp);
703 	}
704 	return retval;
705 }
706 
707 /*
708  * Same as above, but for rgid, egid, sgid.
709  */
710 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
711 {
712 	struct user_namespace *ns = current_user_ns();
713 	const struct cred *old;
714 	struct cred *new;
715 	int retval;
716 	kgid_t krgid, kegid, ksgid;
717 
718 	krgid = make_kgid(ns, rgid);
719 	kegid = make_kgid(ns, egid);
720 	ksgid = make_kgid(ns, sgid);
721 
722 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
723 		return -EINVAL;
724 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
725 		return -EINVAL;
726 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
727 		return -EINVAL;
728 
729 	new = prepare_creds();
730 	if (!new)
731 		return -ENOMEM;
732 	old = current_cred();
733 
734 	retval = -EPERM;
735 	if (!ns_capable(old->user_ns, CAP_SETGID)) {
736 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
737 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
738 			goto error;
739 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
740 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
741 			goto error;
742 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
743 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
744 			goto error;
745 	}
746 
747 	if (rgid != (gid_t) -1)
748 		new->gid = krgid;
749 	if (egid != (gid_t) -1)
750 		new->egid = kegid;
751 	if (sgid != (gid_t) -1)
752 		new->sgid = ksgid;
753 	new->fsgid = new->egid;
754 
755 	return commit_creds(new);
756 
757 error:
758 	abort_creds(new);
759 	return retval;
760 }
761 
762 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
763 {
764 	return __sys_setresgid(rgid, egid, sgid);
765 }
766 
767 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
768 {
769 	const struct cred *cred = current_cred();
770 	int retval;
771 	gid_t rgid, egid, sgid;
772 
773 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
774 	egid = from_kgid_munged(cred->user_ns, cred->egid);
775 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
776 
777 	retval = put_user(rgid, rgidp);
778 	if (!retval) {
779 		retval = put_user(egid, egidp);
780 		if (!retval)
781 			retval = put_user(sgid, sgidp);
782 	}
783 
784 	return retval;
785 }
786 
787 
788 /*
789  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
790  * is used for "access()" and for the NFS daemon (letting nfsd stay at
791  * whatever uid it wants to). It normally shadows "euid", except when
792  * explicitly set by setfsuid() or for access..
793  */
794 long __sys_setfsuid(uid_t uid)
795 {
796 	const struct cred *old;
797 	struct cred *new;
798 	uid_t old_fsuid;
799 	kuid_t kuid;
800 
801 	old = current_cred();
802 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
803 
804 	kuid = make_kuid(old->user_ns, uid);
805 	if (!uid_valid(kuid))
806 		return old_fsuid;
807 
808 	new = prepare_creds();
809 	if (!new)
810 		return old_fsuid;
811 
812 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
813 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
814 	    ns_capable(old->user_ns, CAP_SETUID)) {
815 		if (!uid_eq(kuid, old->fsuid)) {
816 			new->fsuid = kuid;
817 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
818 				goto change_okay;
819 		}
820 	}
821 
822 	abort_creds(new);
823 	return old_fsuid;
824 
825 change_okay:
826 	commit_creds(new);
827 	return old_fsuid;
828 }
829 
830 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
831 {
832 	return __sys_setfsuid(uid);
833 }
834 
835 /*
836  * Samma på svenska..
837  */
838 long __sys_setfsgid(gid_t gid)
839 {
840 	const struct cred *old;
841 	struct cred *new;
842 	gid_t old_fsgid;
843 	kgid_t kgid;
844 
845 	old = current_cred();
846 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
847 
848 	kgid = make_kgid(old->user_ns, gid);
849 	if (!gid_valid(kgid))
850 		return old_fsgid;
851 
852 	new = prepare_creds();
853 	if (!new)
854 		return old_fsgid;
855 
856 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
857 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
858 	    ns_capable(old->user_ns, CAP_SETGID)) {
859 		if (!gid_eq(kgid, old->fsgid)) {
860 			new->fsgid = kgid;
861 			goto change_okay;
862 		}
863 	}
864 
865 	abort_creds(new);
866 	return old_fsgid;
867 
868 change_okay:
869 	commit_creds(new);
870 	return old_fsgid;
871 }
872 
873 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
874 {
875 	return __sys_setfsgid(gid);
876 }
877 #endif /* CONFIG_MULTIUSER */
878 
879 /**
880  * sys_getpid - return the thread group id of the current process
881  *
882  * Note, despite the name, this returns the tgid not the pid.  The tgid and
883  * the pid are identical unless CLONE_THREAD was specified on clone() in
884  * which case the tgid is the same in all threads of the same group.
885  *
886  * This is SMP safe as current->tgid does not change.
887  */
888 SYSCALL_DEFINE0(getpid)
889 {
890 	return task_tgid_vnr(current);
891 }
892 
893 /* Thread ID - the internal kernel "pid" */
894 SYSCALL_DEFINE0(gettid)
895 {
896 	return task_pid_vnr(current);
897 }
898 
899 /*
900  * Accessing ->real_parent is not SMP-safe, it could
901  * change from under us. However, we can use a stale
902  * value of ->real_parent under rcu_read_lock(), see
903  * release_task()->call_rcu(delayed_put_task_struct).
904  */
905 SYSCALL_DEFINE0(getppid)
906 {
907 	int pid;
908 
909 	rcu_read_lock();
910 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
911 	rcu_read_unlock();
912 
913 	return pid;
914 }
915 
916 SYSCALL_DEFINE0(getuid)
917 {
918 	/* Only we change this so SMP safe */
919 	return from_kuid_munged(current_user_ns(), current_uid());
920 }
921 
922 SYSCALL_DEFINE0(geteuid)
923 {
924 	/* Only we change this so SMP safe */
925 	return from_kuid_munged(current_user_ns(), current_euid());
926 }
927 
928 SYSCALL_DEFINE0(getgid)
929 {
930 	/* Only we change this so SMP safe */
931 	return from_kgid_munged(current_user_ns(), current_gid());
932 }
933 
934 SYSCALL_DEFINE0(getegid)
935 {
936 	/* Only we change this so SMP safe */
937 	return from_kgid_munged(current_user_ns(), current_egid());
938 }
939 
940 static void do_sys_times(struct tms *tms)
941 {
942 	u64 tgutime, tgstime, cutime, cstime;
943 
944 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
945 	cutime = current->signal->cutime;
946 	cstime = current->signal->cstime;
947 	tms->tms_utime = nsec_to_clock_t(tgutime);
948 	tms->tms_stime = nsec_to_clock_t(tgstime);
949 	tms->tms_cutime = nsec_to_clock_t(cutime);
950 	tms->tms_cstime = nsec_to_clock_t(cstime);
951 }
952 
953 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
954 {
955 	if (tbuf) {
956 		struct tms tmp;
957 
958 		do_sys_times(&tmp);
959 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
960 			return -EFAULT;
961 	}
962 	force_successful_syscall_return();
963 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
964 }
965 
966 #ifdef CONFIG_COMPAT
967 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
968 {
969 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
970 }
971 
972 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
973 {
974 	if (tbuf) {
975 		struct tms tms;
976 		struct compat_tms tmp;
977 
978 		do_sys_times(&tms);
979 		/* Convert our struct tms to the compat version. */
980 		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
981 		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
982 		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
983 		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
984 		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
985 			return -EFAULT;
986 	}
987 	force_successful_syscall_return();
988 	return compat_jiffies_to_clock_t(jiffies);
989 }
990 #endif
991 
992 /*
993  * This needs some heavy checking ...
994  * I just haven't the stomach for it. I also don't fully
995  * understand sessions/pgrp etc. Let somebody who does explain it.
996  *
997  * OK, I think I have the protection semantics right.... this is really
998  * only important on a multi-user system anyway, to make sure one user
999  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1000  *
1001  * !PF_FORKNOEXEC check to conform completely to POSIX.
1002  */
1003 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1004 {
1005 	struct task_struct *p;
1006 	struct task_struct *group_leader = current->group_leader;
1007 	struct pid *pgrp;
1008 	int err;
1009 
1010 	if (!pid)
1011 		pid = task_pid_vnr(group_leader);
1012 	if (!pgid)
1013 		pgid = pid;
1014 	if (pgid < 0)
1015 		return -EINVAL;
1016 	rcu_read_lock();
1017 
1018 	/* From this point forward we keep holding onto the tasklist lock
1019 	 * so that our parent does not change from under us. -DaveM
1020 	 */
1021 	write_lock_irq(&tasklist_lock);
1022 
1023 	err = -ESRCH;
1024 	p = find_task_by_vpid(pid);
1025 	if (!p)
1026 		goto out;
1027 
1028 	err = -EINVAL;
1029 	if (!thread_group_leader(p))
1030 		goto out;
1031 
1032 	if (same_thread_group(p->real_parent, group_leader)) {
1033 		err = -EPERM;
1034 		if (task_session(p) != task_session(group_leader))
1035 			goto out;
1036 		err = -EACCES;
1037 		if (!(p->flags & PF_FORKNOEXEC))
1038 			goto out;
1039 	} else {
1040 		err = -ESRCH;
1041 		if (p != group_leader)
1042 			goto out;
1043 	}
1044 
1045 	err = -EPERM;
1046 	if (p->signal->leader)
1047 		goto out;
1048 
1049 	pgrp = task_pid(p);
1050 	if (pgid != pid) {
1051 		struct task_struct *g;
1052 
1053 		pgrp = find_vpid(pgid);
1054 		g = pid_task(pgrp, PIDTYPE_PGID);
1055 		if (!g || task_session(g) != task_session(group_leader))
1056 			goto out;
1057 	}
1058 
1059 	err = security_task_setpgid(p, pgid);
1060 	if (err)
1061 		goto out;
1062 
1063 	if (task_pgrp(p) != pgrp)
1064 		change_pid(p, PIDTYPE_PGID, pgrp);
1065 
1066 	err = 0;
1067 out:
1068 	/* All paths lead to here, thus we are safe. -DaveM */
1069 	write_unlock_irq(&tasklist_lock);
1070 	rcu_read_unlock();
1071 	return err;
1072 }
1073 
1074 static int do_getpgid(pid_t pid)
1075 {
1076 	struct task_struct *p;
1077 	struct pid *grp;
1078 	int retval;
1079 
1080 	rcu_read_lock();
1081 	if (!pid)
1082 		grp = task_pgrp(current);
1083 	else {
1084 		retval = -ESRCH;
1085 		p = find_task_by_vpid(pid);
1086 		if (!p)
1087 			goto out;
1088 		grp = task_pgrp(p);
1089 		if (!grp)
1090 			goto out;
1091 
1092 		retval = security_task_getpgid(p);
1093 		if (retval)
1094 			goto out;
1095 	}
1096 	retval = pid_vnr(grp);
1097 out:
1098 	rcu_read_unlock();
1099 	return retval;
1100 }
1101 
1102 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1103 {
1104 	return do_getpgid(pid);
1105 }
1106 
1107 #ifdef __ARCH_WANT_SYS_GETPGRP
1108 
1109 SYSCALL_DEFINE0(getpgrp)
1110 {
1111 	return do_getpgid(0);
1112 }
1113 
1114 #endif
1115 
1116 SYSCALL_DEFINE1(getsid, pid_t, pid)
1117 {
1118 	struct task_struct *p;
1119 	struct pid *sid;
1120 	int retval;
1121 
1122 	rcu_read_lock();
1123 	if (!pid)
1124 		sid = task_session(current);
1125 	else {
1126 		retval = -ESRCH;
1127 		p = find_task_by_vpid(pid);
1128 		if (!p)
1129 			goto out;
1130 		sid = task_session(p);
1131 		if (!sid)
1132 			goto out;
1133 
1134 		retval = security_task_getsid(p);
1135 		if (retval)
1136 			goto out;
1137 	}
1138 	retval = pid_vnr(sid);
1139 out:
1140 	rcu_read_unlock();
1141 	return retval;
1142 }
1143 
1144 static void set_special_pids(struct pid *pid)
1145 {
1146 	struct task_struct *curr = current->group_leader;
1147 
1148 	if (task_session(curr) != pid)
1149 		change_pid(curr, PIDTYPE_SID, pid);
1150 
1151 	if (task_pgrp(curr) != pid)
1152 		change_pid(curr, PIDTYPE_PGID, pid);
1153 }
1154 
1155 int ksys_setsid(void)
1156 {
1157 	struct task_struct *group_leader = current->group_leader;
1158 	struct pid *sid = task_pid(group_leader);
1159 	pid_t session = pid_vnr(sid);
1160 	int err = -EPERM;
1161 
1162 	write_lock_irq(&tasklist_lock);
1163 	/* Fail if I am already a session leader */
1164 	if (group_leader->signal->leader)
1165 		goto out;
1166 
1167 	/* Fail if a process group id already exists that equals the
1168 	 * proposed session id.
1169 	 */
1170 	if (pid_task(sid, PIDTYPE_PGID))
1171 		goto out;
1172 
1173 	group_leader->signal->leader = 1;
1174 	set_special_pids(sid);
1175 
1176 	proc_clear_tty(group_leader);
1177 
1178 	err = session;
1179 out:
1180 	write_unlock_irq(&tasklist_lock);
1181 	if (err > 0) {
1182 		proc_sid_connector(group_leader);
1183 		sched_autogroup_create_attach(group_leader);
1184 	}
1185 	return err;
1186 }
1187 
1188 SYSCALL_DEFINE0(setsid)
1189 {
1190 	return ksys_setsid();
1191 }
1192 
1193 DECLARE_RWSEM(uts_sem);
1194 
1195 #ifdef COMPAT_UTS_MACHINE
1196 #define override_architecture(name) \
1197 	(personality(current->personality) == PER_LINUX32 && \
1198 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1199 		      sizeof(COMPAT_UTS_MACHINE)))
1200 #else
1201 #define override_architecture(name)	0
1202 #endif
1203 
1204 /*
1205  * Work around broken programs that cannot handle "Linux 3.0".
1206  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1207  * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1208  */
1209 static int override_release(char __user *release, size_t len)
1210 {
1211 	int ret = 0;
1212 
1213 	if (current->personality & UNAME26) {
1214 		const char *rest = UTS_RELEASE;
1215 		char buf[65] = { 0 };
1216 		int ndots = 0;
1217 		unsigned v;
1218 		size_t copy;
1219 
1220 		while (*rest) {
1221 			if (*rest == '.' && ++ndots >= 3)
1222 				break;
1223 			if (!isdigit(*rest) && *rest != '.')
1224 				break;
1225 			rest++;
1226 		}
1227 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1228 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1229 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1230 		ret = copy_to_user(release, buf, copy + 1);
1231 	}
1232 	return ret;
1233 }
1234 
1235 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1236 {
1237 	struct new_utsname tmp;
1238 
1239 	down_read(&uts_sem);
1240 	memcpy(&tmp, utsname(), sizeof(tmp));
1241 	up_read(&uts_sem);
1242 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1243 		return -EFAULT;
1244 
1245 	if (override_release(name->release, sizeof(name->release)))
1246 		return -EFAULT;
1247 	if (override_architecture(name))
1248 		return -EFAULT;
1249 	return 0;
1250 }
1251 
1252 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1253 /*
1254  * Old cruft
1255  */
1256 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1257 {
1258 	struct old_utsname tmp;
1259 
1260 	if (!name)
1261 		return -EFAULT;
1262 
1263 	down_read(&uts_sem);
1264 	memcpy(&tmp, utsname(), sizeof(tmp));
1265 	up_read(&uts_sem);
1266 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1267 		return -EFAULT;
1268 
1269 	if (override_release(name->release, sizeof(name->release)))
1270 		return -EFAULT;
1271 	if (override_architecture(name))
1272 		return -EFAULT;
1273 	return 0;
1274 }
1275 
1276 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1277 {
1278 	struct oldold_utsname tmp = {};
1279 
1280 	if (!name)
1281 		return -EFAULT;
1282 
1283 	down_read(&uts_sem);
1284 	memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1285 	memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1286 	memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1287 	memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1288 	memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1289 	up_read(&uts_sem);
1290 	if (copy_to_user(name, &tmp, sizeof(tmp)))
1291 		return -EFAULT;
1292 
1293 	if (override_architecture(name))
1294 		return -EFAULT;
1295 	if (override_release(name->release, sizeof(name->release)))
1296 		return -EFAULT;
1297 	return 0;
1298 }
1299 #endif
1300 
1301 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1302 {
1303 	int errno;
1304 	char tmp[__NEW_UTS_LEN];
1305 
1306 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1307 		return -EPERM;
1308 
1309 	if (len < 0 || len > __NEW_UTS_LEN)
1310 		return -EINVAL;
1311 	errno = -EFAULT;
1312 	if (!copy_from_user(tmp, name, len)) {
1313 		struct new_utsname *u;
1314 
1315 		down_write(&uts_sem);
1316 		u = utsname();
1317 		memcpy(u->nodename, tmp, len);
1318 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1319 		errno = 0;
1320 		uts_proc_notify(UTS_PROC_HOSTNAME);
1321 		up_write(&uts_sem);
1322 	}
1323 	return errno;
1324 }
1325 
1326 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1327 
1328 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1329 {
1330 	int i;
1331 	struct new_utsname *u;
1332 	char tmp[__NEW_UTS_LEN + 1];
1333 
1334 	if (len < 0)
1335 		return -EINVAL;
1336 	down_read(&uts_sem);
1337 	u = utsname();
1338 	i = 1 + strlen(u->nodename);
1339 	if (i > len)
1340 		i = len;
1341 	memcpy(tmp, u->nodename, i);
1342 	up_read(&uts_sem);
1343 	if (copy_to_user(name, tmp, i))
1344 		return -EFAULT;
1345 	return 0;
1346 }
1347 
1348 #endif
1349 
1350 /*
1351  * Only setdomainname; getdomainname can be implemented by calling
1352  * uname()
1353  */
1354 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1355 {
1356 	int errno;
1357 	char tmp[__NEW_UTS_LEN];
1358 
1359 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1360 		return -EPERM;
1361 	if (len < 0 || len > __NEW_UTS_LEN)
1362 		return -EINVAL;
1363 
1364 	errno = -EFAULT;
1365 	if (!copy_from_user(tmp, name, len)) {
1366 		struct new_utsname *u;
1367 
1368 		down_write(&uts_sem);
1369 		u = utsname();
1370 		memcpy(u->domainname, tmp, len);
1371 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1372 		errno = 0;
1373 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1374 		up_write(&uts_sem);
1375 	}
1376 	return errno;
1377 }
1378 
1379 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1380 {
1381 	struct rlimit value;
1382 	int ret;
1383 
1384 	ret = do_prlimit(current, resource, NULL, &value);
1385 	if (!ret)
1386 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1387 
1388 	return ret;
1389 }
1390 
1391 #ifdef CONFIG_COMPAT
1392 
1393 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1394 		       struct compat_rlimit __user *, rlim)
1395 {
1396 	struct rlimit r;
1397 	struct compat_rlimit r32;
1398 
1399 	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1400 		return -EFAULT;
1401 
1402 	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1403 		r.rlim_cur = RLIM_INFINITY;
1404 	else
1405 		r.rlim_cur = r32.rlim_cur;
1406 	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1407 		r.rlim_max = RLIM_INFINITY;
1408 	else
1409 		r.rlim_max = r32.rlim_max;
1410 	return do_prlimit(current, resource, &r, NULL);
1411 }
1412 
1413 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1414 		       struct compat_rlimit __user *, rlim)
1415 {
1416 	struct rlimit r;
1417 	int ret;
1418 
1419 	ret = do_prlimit(current, resource, NULL, &r);
1420 	if (!ret) {
1421 		struct compat_rlimit r32;
1422 		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1423 			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1424 		else
1425 			r32.rlim_cur = r.rlim_cur;
1426 		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1427 			r32.rlim_max = COMPAT_RLIM_INFINITY;
1428 		else
1429 			r32.rlim_max = r.rlim_max;
1430 
1431 		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1432 			return -EFAULT;
1433 	}
1434 	return ret;
1435 }
1436 
1437 #endif
1438 
1439 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1440 
1441 /*
1442  *	Back compatibility for getrlimit. Needed for some apps.
1443  */
1444 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1445 		struct rlimit __user *, rlim)
1446 {
1447 	struct rlimit x;
1448 	if (resource >= RLIM_NLIMITS)
1449 		return -EINVAL;
1450 
1451 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1452 	task_lock(current->group_leader);
1453 	x = current->signal->rlim[resource];
1454 	task_unlock(current->group_leader);
1455 	if (x.rlim_cur > 0x7FFFFFFF)
1456 		x.rlim_cur = 0x7FFFFFFF;
1457 	if (x.rlim_max > 0x7FFFFFFF)
1458 		x.rlim_max = 0x7FFFFFFF;
1459 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1460 }
1461 
1462 #ifdef CONFIG_COMPAT
1463 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1464 		       struct compat_rlimit __user *, rlim)
1465 {
1466 	struct rlimit r;
1467 
1468 	if (resource >= RLIM_NLIMITS)
1469 		return -EINVAL;
1470 
1471 	resource = array_index_nospec(resource, RLIM_NLIMITS);
1472 	task_lock(current->group_leader);
1473 	r = current->signal->rlim[resource];
1474 	task_unlock(current->group_leader);
1475 	if (r.rlim_cur > 0x7FFFFFFF)
1476 		r.rlim_cur = 0x7FFFFFFF;
1477 	if (r.rlim_max > 0x7FFFFFFF)
1478 		r.rlim_max = 0x7FFFFFFF;
1479 
1480 	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1481 	    put_user(r.rlim_max, &rlim->rlim_max))
1482 		return -EFAULT;
1483 	return 0;
1484 }
1485 #endif
1486 
1487 #endif
1488 
1489 static inline bool rlim64_is_infinity(__u64 rlim64)
1490 {
1491 #if BITS_PER_LONG < 64
1492 	return rlim64 >= ULONG_MAX;
1493 #else
1494 	return rlim64 == RLIM64_INFINITY;
1495 #endif
1496 }
1497 
1498 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1499 {
1500 	if (rlim->rlim_cur == RLIM_INFINITY)
1501 		rlim64->rlim_cur = RLIM64_INFINITY;
1502 	else
1503 		rlim64->rlim_cur = rlim->rlim_cur;
1504 	if (rlim->rlim_max == RLIM_INFINITY)
1505 		rlim64->rlim_max = RLIM64_INFINITY;
1506 	else
1507 		rlim64->rlim_max = rlim->rlim_max;
1508 }
1509 
1510 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1511 {
1512 	if (rlim64_is_infinity(rlim64->rlim_cur))
1513 		rlim->rlim_cur = RLIM_INFINITY;
1514 	else
1515 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1516 	if (rlim64_is_infinity(rlim64->rlim_max))
1517 		rlim->rlim_max = RLIM_INFINITY;
1518 	else
1519 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1520 }
1521 
1522 /* make sure you are allowed to change @tsk limits before calling this */
1523 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1524 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1525 {
1526 	struct rlimit *rlim;
1527 	int retval = 0;
1528 
1529 	if (resource >= RLIM_NLIMITS)
1530 		return -EINVAL;
1531 	if (new_rlim) {
1532 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1533 			return -EINVAL;
1534 		if (resource == RLIMIT_NOFILE &&
1535 				new_rlim->rlim_max > sysctl_nr_open)
1536 			return -EPERM;
1537 	}
1538 
1539 	/* protect tsk->signal and tsk->sighand from disappearing */
1540 	read_lock(&tasklist_lock);
1541 	if (!tsk->sighand) {
1542 		retval = -ESRCH;
1543 		goto out;
1544 	}
1545 
1546 	rlim = tsk->signal->rlim + resource;
1547 	task_lock(tsk->group_leader);
1548 	if (new_rlim) {
1549 		/* Keep the capable check against init_user_ns until
1550 		   cgroups can contain all limits */
1551 		if (new_rlim->rlim_max > rlim->rlim_max &&
1552 				!capable(CAP_SYS_RESOURCE))
1553 			retval = -EPERM;
1554 		if (!retval)
1555 			retval = security_task_setrlimit(tsk, resource, new_rlim);
1556 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1557 			/*
1558 			 * The caller is asking for an immediate RLIMIT_CPU
1559 			 * expiry.  But we use the zero value to mean "it was
1560 			 * never set".  So let's cheat and make it one second
1561 			 * instead
1562 			 */
1563 			new_rlim->rlim_cur = 1;
1564 		}
1565 	}
1566 	if (!retval) {
1567 		if (old_rlim)
1568 			*old_rlim = *rlim;
1569 		if (new_rlim)
1570 			*rlim = *new_rlim;
1571 	}
1572 	task_unlock(tsk->group_leader);
1573 
1574 	/*
1575 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1576 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1577 	 * very long-standing error, and fixing it now risks breakage of
1578 	 * applications, so we live with it
1579 	 */
1580 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1581 	     new_rlim->rlim_cur != RLIM_INFINITY &&
1582 	     IS_ENABLED(CONFIG_POSIX_TIMERS))
1583 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1584 out:
1585 	read_unlock(&tasklist_lock);
1586 	return retval;
1587 }
1588 
1589 /* rcu lock must be held */
1590 static int check_prlimit_permission(struct task_struct *task,
1591 				    unsigned int flags)
1592 {
1593 	const struct cred *cred = current_cred(), *tcred;
1594 	bool id_match;
1595 
1596 	if (current == task)
1597 		return 0;
1598 
1599 	tcred = __task_cred(task);
1600 	id_match = (uid_eq(cred->uid, tcred->euid) &&
1601 		    uid_eq(cred->uid, tcred->suid) &&
1602 		    uid_eq(cred->uid, tcred->uid)  &&
1603 		    gid_eq(cred->gid, tcred->egid) &&
1604 		    gid_eq(cred->gid, tcred->sgid) &&
1605 		    gid_eq(cred->gid, tcred->gid));
1606 	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1607 		return -EPERM;
1608 
1609 	return security_task_prlimit(cred, tcred, flags);
1610 }
1611 
1612 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1613 		const struct rlimit64 __user *, new_rlim,
1614 		struct rlimit64 __user *, old_rlim)
1615 {
1616 	struct rlimit64 old64, new64;
1617 	struct rlimit old, new;
1618 	struct task_struct *tsk;
1619 	unsigned int checkflags = 0;
1620 	int ret;
1621 
1622 	if (old_rlim)
1623 		checkflags |= LSM_PRLIMIT_READ;
1624 
1625 	if (new_rlim) {
1626 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1627 			return -EFAULT;
1628 		rlim64_to_rlim(&new64, &new);
1629 		checkflags |= LSM_PRLIMIT_WRITE;
1630 	}
1631 
1632 	rcu_read_lock();
1633 	tsk = pid ? find_task_by_vpid(pid) : current;
1634 	if (!tsk) {
1635 		rcu_read_unlock();
1636 		return -ESRCH;
1637 	}
1638 	ret = check_prlimit_permission(tsk, checkflags);
1639 	if (ret) {
1640 		rcu_read_unlock();
1641 		return ret;
1642 	}
1643 	get_task_struct(tsk);
1644 	rcu_read_unlock();
1645 
1646 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1647 			old_rlim ? &old : NULL);
1648 
1649 	if (!ret && old_rlim) {
1650 		rlim_to_rlim64(&old, &old64);
1651 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1652 			ret = -EFAULT;
1653 	}
1654 
1655 	put_task_struct(tsk);
1656 	return ret;
1657 }
1658 
1659 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1660 {
1661 	struct rlimit new_rlim;
1662 
1663 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1664 		return -EFAULT;
1665 	return do_prlimit(current, resource, &new_rlim, NULL);
1666 }
1667 
1668 /*
1669  * It would make sense to put struct rusage in the task_struct,
1670  * except that would make the task_struct be *really big*.  After
1671  * task_struct gets moved into malloc'ed memory, it would
1672  * make sense to do this.  It will make moving the rest of the information
1673  * a lot simpler!  (Which we're not doing right now because we're not
1674  * measuring them yet).
1675  *
1676  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1677  * races with threads incrementing their own counters.  But since word
1678  * reads are atomic, we either get new values or old values and we don't
1679  * care which for the sums.  We always take the siglock to protect reading
1680  * the c* fields from p->signal from races with exit.c updating those
1681  * fields when reaping, so a sample either gets all the additions of a
1682  * given child after it's reaped, or none so this sample is before reaping.
1683  *
1684  * Locking:
1685  * We need to take the siglock for CHILDEREN, SELF and BOTH
1686  * for  the cases current multithreaded, non-current single threaded
1687  * non-current multithreaded.  Thread traversal is now safe with
1688  * the siglock held.
1689  * Strictly speaking, we donot need to take the siglock if we are current and
1690  * single threaded,  as no one else can take our signal_struct away, no one
1691  * else can  reap the  children to update signal->c* counters, and no one else
1692  * can race with the signal-> fields. If we do not take any lock, the
1693  * signal-> fields could be read out of order while another thread was just
1694  * exiting. So we should  place a read memory barrier when we avoid the lock.
1695  * On the writer side,  write memory barrier is implied in  __exit_signal
1696  * as __exit_signal releases  the siglock spinlock after updating the signal->
1697  * fields. But we don't do this yet to keep things simple.
1698  *
1699  */
1700 
1701 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1702 {
1703 	r->ru_nvcsw += t->nvcsw;
1704 	r->ru_nivcsw += t->nivcsw;
1705 	r->ru_minflt += t->min_flt;
1706 	r->ru_majflt += t->maj_flt;
1707 	r->ru_inblock += task_io_get_inblock(t);
1708 	r->ru_oublock += task_io_get_oublock(t);
1709 }
1710 
1711 void getrusage(struct task_struct *p, int who, struct rusage *r)
1712 {
1713 	struct task_struct *t;
1714 	unsigned long flags;
1715 	u64 tgutime, tgstime, utime, stime;
1716 	unsigned long maxrss = 0;
1717 
1718 	memset((char *)r, 0, sizeof (*r));
1719 	utime = stime = 0;
1720 
1721 	if (who == RUSAGE_THREAD) {
1722 		task_cputime_adjusted(current, &utime, &stime);
1723 		accumulate_thread_rusage(p, r);
1724 		maxrss = p->signal->maxrss;
1725 		goto out;
1726 	}
1727 
1728 	if (!lock_task_sighand(p, &flags))
1729 		return;
1730 
1731 	switch (who) {
1732 	case RUSAGE_BOTH:
1733 	case RUSAGE_CHILDREN:
1734 		utime = p->signal->cutime;
1735 		stime = p->signal->cstime;
1736 		r->ru_nvcsw = p->signal->cnvcsw;
1737 		r->ru_nivcsw = p->signal->cnivcsw;
1738 		r->ru_minflt = p->signal->cmin_flt;
1739 		r->ru_majflt = p->signal->cmaj_flt;
1740 		r->ru_inblock = p->signal->cinblock;
1741 		r->ru_oublock = p->signal->coublock;
1742 		maxrss = p->signal->cmaxrss;
1743 
1744 		if (who == RUSAGE_CHILDREN)
1745 			break;
1746 
1747 	case RUSAGE_SELF:
1748 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1749 		utime += tgutime;
1750 		stime += tgstime;
1751 		r->ru_nvcsw += p->signal->nvcsw;
1752 		r->ru_nivcsw += p->signal->nivcsw;
1753 		r->ru_minflt += p->signal->min_flt;
1754 		r->ru_majflt += p->signal->maj_flt;
1755 		r->ru_inblock += p->signal->inblock;
1756 		r->ru_oublock += p->signal->oublock;
1757 		if (maxrss < p->signal->maxrss)
1758 			maxrss = p->signal->maxrss;
1759 		t = p;
1760 		do {
1761 			accumulate_thread_rusage(t, r);
1762 		} while_each_thread(p, t);
1763 		break;
1764 
1765 	default:
1766 		BUG();
1767 	}
1768 	unlock_task_sighand(p, &flags);
1769 
1770 out:
1771 	r->ru_utime = ns_to_timeval(utime);
1772 	r->ru_stime = ns_to_timeval(stime);
1773 
1774 	if (who != RUSAGE_CHILDREN) {
1775 		struct mm_struct *mm = get_task_mm(p);
1776 
1777 		if (mm) {
1778 			setmax_mm_hiwater_rss(&maxrss, mm);
1779 			mmput(mm);
1780 		}
1781 	}
1782 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1783 }
1784 
1785 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1786 {
1787 	struct rusage r;
1788 
1789 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1790 	    who != RUSAGE_THREAD)
1791 		return -EINVAL;
1792 
1793 	getrusage(current, who, &r);
1794 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1795 }
1796 
1797 #ifdef CONFIG_COMPAT
1798 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1799 {
1800 	struct rusage r;
1801 
1802 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1803 	    who != RUSAGE_THREAD)
1804 		return -EINVAL;
1805 
1806 	getrusage(current, who, &r);
1807 	return put_compat_rusage(&r, ru);
1808 }
1809 #endif
1810 
1811 SYSCALL_DEFINE1(umask, int, mask)
1812 {
1813 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1814 	return mask;
1815 }
1816 
1817 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1818 {
1819 	struct fd exe;
1820 	struct file *old_exe, *exe_file;
1821 	struct inode *inode;
1822 	int err;
1823 
1824 	exe = fdget(fd);
1825 	if (!exe.file)
1826 		return -EBADF;
1827 
1828 	inode = file_inode(exe.file);
1829 
1830 	/*
1831 	 * Because the original mm->exe_file points to executable file, make
1832 	 * sure that this one is executable as well, to avoid breaking an
1833 	 * overall picture.
1834 	 */
1835 	err = -EACCES;
1836 	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1837 		goto exit;
1838 
1839 	err = inode_permission(inode, MAY_EXEC);
1840 	if (err)
1841 		goto exit;
1842 
1843 	/*
1844 	 * Forbid mm->exe_file change if old file still mapped.
1845 	 */
1846 	exe_file = get_mm_exe_file(mm);
1847 	err = -EBUSY;
1848 	if (exe_file) {
1849 		struct vm_area_struct *vma;
1850 
1851 		down_read(&mm->mmap_sem);
1852 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
1853 			if (!vma->vm_file)
1854 				continue;
1855 			if (path_equal(&vma->vm_file->f_path,
1856 				       &exe_file->f_path))
1857 				goto exit_err;
1858 		}
1859 
1860 		up_read(&mm->mmap_sem);
1861 		fput(exe_file);
1862 	}
1863 
1864 	err = 0;
1865 	/* set the new file, lockless */
1866 	get_file(exe.file);
1867 	old_exe = xchg(&mm->exe_file, exe.file);
1868 	if (old_exe)
1869 		fput(old_exe);
1870 exit:
1871 	fdput(exe);
1872 	return err;
1873 exit_err:
1874 	up_read(&mm->mmap_sem);
1875 	fput(exe_file);
1876 	goto exit;
1877 }
1878 
1879 /*
1880  * WARNING: we don't require any capability here so be very careful
1881  * in what is allowed for modification from userspace.
1882  */
1883 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1884 {
1885 	unsigned long mmap_max_addr = TASK_SIZE;
1886 	struct mm_struct *mm = current->mm;
1887 	int error = -EINVAL, i;
1888 
1889 	static const unsigned char offsets[] = {
1890 		offsetof(struct prctl_mm_map, start_code),
1891 		offsetof(struct prctl_mm_map, end_code),
1892 		offsetof(struct prctl_mm_map, start_data),
1893 		offsetof(struct prctl_mm_map, end_data),
1894 		offsetof(struct prctl_mm_map, start_brk),
1895 		offsetof(struct prctl_mm_map, brk),
1896 		offsetof(struct prctl_mm_map, start_stack),
1897 		offsetof(struct prctl_mm_map, arg_start),
1898 		offsetof(struct prctl_mm_map, arg_end),
1899 		offsetof(struct prctl_mm_map, env_start),
1900 		offsetof(struct prctl_mm_map, env_end),
1901 	};
1902 
1903 	/*
1904 	 * Make sure the members are not somewhere outside
1905 	 * of allowed address space.
1906 	 */
1907 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1908 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1909 
1910 		if ((unsigned long)val >= mmap_max_addr ||
1911 		    (unsigned long)val < mmap_min_addr)
1912 			goto out;
1913 	}
1914 
1915 	/*
1916 	 * Make sure the pairs are ordered.
1917 	 */
1918 #define __prctl_check_order(__m1, __op, __m2)				\
1919 	((unsigned long)prctl_map->__m1 __op				\
1920 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1921 	error  = __prctl_check_order(start_code, <, end_code);
1922 	error |= __prctl_check_order(start_data, <, end_data);
1923 	error |= __prctl_check_order(start_brk, <=, brk);
1924 	error |= __prctl_check_order(arg_start, <=, arg_end);
1925 	error |= __prctl_check_order(env_start, <=, env_end);
1926 	if (error)
1927 		goto out;
1928 #undef __prctl_check_order
1929 
1930 	error = -EINVAL;
1931 
1932 	/*
1933 	 * @brk should be after @end_data in traditional maps.
1934 	 */
1935 	if (prctl_map->start_brk <= prctl_map->end_data ||
1936 	    prctl_map->brk <= prctl_map->end_data)
1937 		goto out;
1938 
1939 	/*
1940 	 * Neither we should allow to override limits if they set.
1941 	 */
1942 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1943 			      prctl_map->start_brk, prctl_map->end_data,
1944 			      prctl_map->start_data))
1945 			goto out;
1946 
1947 	/*
1948 	 * Someone is trying to cheat the auxv vector.
1949 	 */
1950 	if (prctl_map->auxv_size) {
1951 		if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1952 			goto out;
1953 	}
1954 
1955 	/*
1956 	 * Finally, make sure the caller has the rights to
1957 	 * change /proc/pid/exe link: only local sys admin should
1958 	 * be allowed to.
1959 	 */
1960 	if (prctl_map->exe_fd != (u32)-1) {
1961 		if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1962 			goto out;
1963 	}
1964 
1965 	error = 0;
1966 out:
1967 	return error;
1968 }
1969 
1970 #ifdef CONFIG_CHECKPOINT_RESTORE
1971 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1972 {
1973 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1974 	unsigned long user_auxv[AT_VECTOR_SIZE];
1975 	struct mm_struct *mm = current->mm;
1976 	int error;
1977 
1978 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1979 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1980 
1981 	if (opt == PR_SET_MM_MAP_SIZE)
1982 		return put_user((unsigned int)sizeof(prctl_map),
1983 				(unsigned int __user *)addr);
1984 
1985 	if (data_size != sizeof(prctl_map))
1986 		return -EINVAL;
1987 
1988 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1989 		return -EFAULT;
1990 
1991 	error = validate_prctl_map(&prctl_map);
1992 	if (error)
1993 		return error;
1994 
1995 	if (prctl_map.auxv_size) {
1996 		memset(user_auxv, 0, sizeof(user_auxv));
1997 		if (copy_from_user(user_auxv,
1998 				   (const void __user *)prctl_map.auxv,
1999 				   prctl_map.auxv_size))
2000 			return -EFAULT;
2001 
2002 		/* Last entry must be AT_NULL as specification requires */
2003 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2004 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2005 	}
2006 
2007 	if (prctl_map.exe_fd != (u32)-1) {
2008 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2009 		if (error)
2010 			return error;
2011 	}
2012 
2013 	/*
2014 	 * arg_lock protects concurent updates but we still need mmap_sem for
2015 	 * read to exclude races with sys_brk.
2016 	 */
2017 	down_read(&mm->mmap_sem);
2018 
2019 	/*
2020 	 * We don't validate if these members are pointing to
2021 	 * real present VMAs because application may have correspond
2022 	 * VMAs already unmapped and kernel uses these members for statistics
2023 	 * output in procfs mostly, except
2024 	 *
2025 	 *  - @start_brk/@brk which are used in do_brk but kernel lookups
2026 	 *    for VMAs when updating these memvers so anything wrong written
2027 	 *    here cause kernel to swear at userspace program but won't lead
2028 	 *    to any problem in kernel itself
2029 	 */
2030 
2031 	spin_lock(&mm->arg_lock);
2032 	mm->start_code	= prctl_map.start_code;
2033 	mm->end_code	= prctl_map.end_code;
2034 	mm->start_data	= prctl_map.start_data;
2035 	mm->end_data	= prctl_map.end_data;
2036 	mm->start_brk	= prctl_map.start_brk;
2037 	mm->brk		= prctl_map.brk;
2038 	mm->start_stack	= prctl_map.start_stack;
2039 	mm->arg_start	= prctl_map.arg_start;
2040 	mm->arg_end	= prctl_map.arg_end;
2041 	mm->env_start	= prctl_map.env_start;
2042 	mm->env_end	= prctl_map.env_end;
2043 	spin_unlock(&mm->arg_lock);
2044 
2045 	/*
2046 	 * Note this update of @saved_auxv is lockless thus
2047 	 * if someone reads this member in procfs while we're
2048 	 * updating -- it may get partly updated results. It's
2049 	 * known and acceptable trade off: we leave it as is to
2050 	 * not introduce additional locks here making the kernel
2051 	 * more complex.
2052 	 */
2053 	if (prctl_map.auxv_size)
2054 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2055 
2056 	up_read(&mm->mmap_sem);
2057 	return 0;
2058 }
2059 #endif /* CONFIG_CHECKPOINT_RESTORE */
2060 
2061 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2062 			  unsigned long len)
2063 {
2064 	/*
2065 	 * This doesn't move the auxiliary vector itself since it's pinned to
2066 	 * mm_struct, but it permits filling the vector with new values.  It's
2067 	 * up to the caller to provide sane values here, otherwise userspace
2068 	 * tools which use this vector might be unhappy.
2069 	 */
2070 	unsigned long user_auxv[AT_VECTOR_SIZE];
2071 
2072 	if (len > sizeof(user_auxv))
2073 		return -EINVAL;
2074 
2075 	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2076 		return -EFAULT;
2077 
2078 	/* Make sure the last entry is always AT_NULL */
2079 	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2080 	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2081 
2082 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2083 
2084 	task_lock(current);
2085 	memcpy(mm->saved_auxv, user_auxv, len);
2086 	task_unlock(current);
2087 
2088 	return 0;
2089 }
2090 
2091 static int prctl_set_mm(int opt, unsigned long addr,
2092 			unsigned long arg4, unsigned long arg5)
2093 {
2094 	struct mm_struct *mm = current->mm;
2095 	struct prctl_mm_map prctl_map;
2096 	struct vm_area_struct *vma;
2097 	int error;
2098 
2099 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2100 			      opt != PR_SET_MM_MAP &&
2101 			      opt != PR_SET_MM_MAP_SIZE)))
2102 		return -EINVAL;
2103 
2104 #ifdef CONFIG_CHECKPOINT_RESTORE
2105 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2106 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2107 #endif
2108 
2109 	if (!capable(CAP_SYS_RESOURCE))
2110 		return -EPERM;
2111 
2112 	if (opt == PR_SET_MM_EXE_FILE)
2113 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2114 
2115 	if (opt == PR_SET_MM_AUXV)
2116 		return prctl_set_auxv(mm, addr, arg4);
2117 
2118 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2119 		return -EINVAL;
2120 
2121 	error = -EINVAL;
2122 
2123 	down_write(&mm->mmap_sem);
2124 	vma = find_vma(mm, addr);
2125 
2126 	prctl_map.start_code	= mm->start_code;
2127 	prctl_map.end_code	= mm->end_code;
2128 	prctl_map.start_data	= mm->start_data;
2129 	prctl_map.end_data	= mm->end_data;
2130 	prctl_map.start_brk	= mm->start_brk;
2131 	prctl_map.brk		= mm->brk;
2132 	prctl_map.start_stack	= mm->start_stack;
2133 	prctl_map.arg_start	= mm->arg_start;
2134 	prctl_map.arg_end	= mm->arg_end;
2135 	prctl_map.env_start	= mm->env_start;
2136 	prctl_map.env_end	= mm->env_end;
2137 	prctl_map.auxv		= NULL;
2138 	prctl_map.auxv_size	= 0;
2139 	prctl_map.exe_fd	= -1;
2140 
2141 	switch (opt) {
2142 	case PR_SET_MM_START_CODE:
2143 		prctl_map.start_code = addr;
2144 		break;
2145 	case PR_SET_MM_END_CODE:
2146 		prctl_map.end_code = addr;
2147 		break;
2148 	case PR_SET_MM_START_DATA:
2149 		prctl_map.start_data = addr;
2150 		break;
2151 	case PR_SET_MM_END_DATA:
2152 		prctl_map.end_data = addr;
2153 		break;
2154 	case PR_SET_MM_START_STACK:
2155 		prctl_map.start_stack = addr;
2156 		break;
2157 	case PR_SET_MM_START_BRK:
2158 		prctl_map.start_brk = addr;
2159 		break;
2160 	case PR_SET_MM_BRK:
2161 		prctl_map.brk = addr;
2162 		break;
2163 	case PR_SET_MM_ARG_START:
2164 		prctl_map.arg_start = addr;
2165 		break;
2166 	case PR_SET_MM_ARG_END:
2167 		prctl_map.arg_end = addr;
2168 		break;
2169 	case PR_SET_MM_ENV_START:
2170 		prctl_map.env_start = addr;
2171 		break;
2172 	case PR_SET_MM_ENV_END:
2173 		prctl_map.env_end = addr;
2174 		break;
2175 	default:
2176 		goto out;
2177 	}
2178 
2179 	error = validate_prctl_map(&prctl_map);
2180 	if (error)
2181 		goto out;
2182 
2183 	switch (opt) {
2184 	/*
2185 	 * If command line arguments and environment
2186 	 * are placed somewhere else on stack, we can
2187 	 * set them up here, ARG_START/END to setup
2188 	 * command line argumets and ENV_START/END
2189 	 * for environment.
2190 	 */
2191 	case PR_SET_MM_START_STACK:
2192 	case PR_SET_MM_ARG_START:
2193 	case PR_SET_MM_ARG_END:
2194 	case PR_SET_MM_ENV_START:
2195 	case PR_SET_MM_ENV_END:
2196 		if (!vma) {
2197 			error = -EFAULT;
2198 			goto out;
2199 		}
2200 	}
2201 
2202 	mm->start_code	= prctl_map.start_code;
2203 	mm->end_code	= prctl_map.end_code;
2204 	mm->start_data	= prctl_map.start_data;
2205 	mm->end_data	= prctl_map.end_data;
2206 	mm->start_brk	= prctl_map.start_brk;
2207 	mm->brk		= prctl_map.brk;
2208 	mm->start_stack	= prctl_map.start_stack;
2209 	mm->arg_start	= prctl_map.arg_start;
2210 	mm->arg_end	= prctl_map.arg_end;
2211 	mm->env_start	= prctl_map.env_start;
2212 	mm->env_end	= prctl_map.env_end;
2213 
2214 	error = 0;
2215 out:
2216 	up_write(&mm->mmap_sem);
2217 	return error;
2218 }
2219 
2220 #ifdef CONFIG_CHECKPOINT_RESTORE
2221 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2222 {
2223 	return put_user(me->clear_child_tid, tid_addr);
2224 }
2225 #else
2226 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2227 {
2228 	return -EINVAL;
2229 }
2230 #endif
2231 
2232 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2233 {
2234 	/*
2235 	 * If task has has_child_subreaper - all its decendants
2236 	 * already have these flag too and new decendants will
2237 	 * inherit it on fork, skip them.
2238 	 *
2239 	 * If we've found child_reaper - skip descendants in
2240 	 * it's subtree as they will never get out pidns.
2241 	 */
2242 	if (p->signal->has_child_subreaper ||
2243 	    is_child_reaper(task_pid(p)))
2244 		return 0;
2245 
2246 	p->signal->has_child_subreaper = 1;
2247 	return 1;
2248 }
2249 
2250 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2251 {
2252 	return -EINVAL;
2253 }
2254 
2255 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2256 				    unsigned long ctrl)
2257 {
2258 	return -EINVAL;
2259 }
2260 
2261 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2262 		unsigned long, arg4, unsigned long, arg5)
2263 {
2264 	struct task_struct *me = current;
2265 	unsigned char comm[sizeof(me->comm)];
2266 	long error;
2267 
2268 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2269 	if (error != -ENOSYS)
2270 		return error;
2271 
2272 	error = 0;
2273 	switch (option) {
2274 	case PR_SET_PDEATHSIG:
2275 		if (!valid_signal(arg2)) {
2276 			error = -EINVAL;
2277 			break;
2278 		}
2279 		me->pdeath_signal = arg2;
2280 		break;
2281 	case PR_GET_PDEATHSIG:
2282 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2283 		break;
2284 	case PR_GET_DUMPABLE:
2285 		error = get_dumpable(me->mm);
2286 		break;
2287 	case PR_SET_DUMPABLE:
2288 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2289 			error = -EINVAL;
2290 			break;
2291 		}
2292 		set_dumpable(me->mm, arg2);
2293 		break;
2294 
2295 	case PR_SET_UNALIGN:
2296 		error = SET_UNALIGN_CTL(me, arg2);
2297 		break;
2298 	case PR_GET_UNALIGN:
2299 		error = GET_UNALIGN_CTL(me, arg2);
2300 		break;
2301 	case PR_SET_FPEMU:
2302 		error = SET_FPEMU_CTL(me, arg2);
2303 		break;
2304 	case PR_GET_FPEMU:
2305 		error = GET_FPEMU_CTL(me, arg2);
2306 		break;
2307 	case PR_SET_FPEXC:
2308 		error = SET_FPEXC_CTL(me, arg2);
2309 		break;
2310 	case PR_GET_FPEXC:
2311 		error = GET_FPEXC_CTL(me, arg2);
2312 		break;
2313 	case PR_GET_TIMING:
2314 		error = PR_TIMING_STATISTICAL;
2315 		break;
2316 	case PR_SET_TIMING:
2317 		if (arg2 != PR_TIMING_STATISTICAL)
2318 			error = -EINVAL;
2319 		break;
2320 	case PR_SET_NAME:
2321 		comm[sizeof(me->comm) - 1] = 0;
2322 		if (strncpy_from_user(comm, (char __user *)arg2,
2323 				      sizeof(me->comm) - 1) < 0)
2324 			return -EFAULT;
2325 		set_task_comm(me, comm);
2326 		proc_comm_connector(me);
2327 		break;
2328 	case PR_GET_NAME:
2329 		get_task_comm(comm, me);
2330 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2331 			return -EFAULT;
2332 		break;
2333 	case PR_GET_ENDIAN:
2334 		error = GET_ENDIAN(me, arg2);
2335 		break;
2336 	case PR_SET_ENDIAN:
2337 		error = SET_ENDIAN(me, arg2);
2338 		break;
2339 	case PR_GET_SECCOMP:
2340 		error = prctl_get_seccomp();
2341 		break;
2342 	case PR_SET_SECCOMP:
2343 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2344 		break;
2345 	case PR_GET_TSC:
2346 		error = GET_TSC_CTL(arg2);
2347 		break;
2348 	case PR_SET_TSC:
2349 		error = SET_TSC_CTL(arg2);
2350 		break;
2351 	case PR_TASK_PERF_EVENTS_DISABLE:
2352 		error = perf_event_task_disable();
2353 		break;
2354 	case PR_TASK_PERF_EVENTS_ENABLE:
2355 		error = perf_event_task_enable();
2356 		break;
2357 	case PR_GET_TIMERSLACK:
2358 		if (current->timer_slack_ns > ULONG_MAX)
2359 			error = ULONG_MAX;
2360 		else
2361 			error = current->timer_slack_ns;
2362 		break;
2363 	case PR_SET_TIMERSLACK:
2364 		if (arg2 <= 0)
2365 			current->timer_slack_ns =
2366 					current->default_timer_slack_ns;
2367 		else
2368 			current->timer_slack_ns = arg2;
2369 		break;
2370 	case PR_MCE_KILL:
2371 		if (arg4 | arg5)
2372 			return -EINVAL;
2373 		switch (arg2) {
2374 		case PR_MCE_KILL_CLEAR:
2375 			if (arg3 != 0)
2376 				return -EINVAL;
2377 			current->flags &= ~PF_MCE_PROCESS;
2378 			break;
2379 		case PR_MCE_KILL_SET:
2380 			current->flags |= PF_MCE_PROCESS;
2381 			if (arg3 == PR_MCE_KILL_EARLY)
2382 				current->flags |= PF_MCE_EARLY;
2383 			else if (arg3 == PR_MCE_KILL_LATE)
2384 				current->flags &= ~PF_MCE_EARLY;
2385 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2386 				current->flags &=
2387 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2388 			else
2389 				return -EINVAL;
2390 			break;
2391 		default:
2392 			return -EINVAL;
2393 		}
2394 		break;
2395 	case PR_MCE_KILL_GET:
2396 		if (arg2 | arg3 | arg4 | arg5)
2397 			return -EINVAL;
2398 		if (current->flags & PF_MCE_PROCESS)
2399 			error = (current->flags & PF_MCE_EARLY) ?
2400 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2401 		else
2402 			error = PR_MCE_KILL_DEFAULT;
2403 		break;
2404 	case PR_SET_MM:
2405 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2406 		break;
2407 	case PR_GET_TID_ADDRESS:
2408 		error = prctl_get_tid_address(me, (int __user **)arg2);
2409 		break;
2410 	case PR_SET_CHILD_SUBREAPER:
2411 		me->signal->is_child_subreaper = !!arg2;
2412 		if (!arg2)
2413 			break;
2414 
2415 		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2416 		break;
2417 	case PR_GET_CHILD_SUBREAPER:
2418 		error = put_user(me->signal->is_child_subreaper,
2419 				 (int __user *)arg2);
2420 		break;
2421 	case PR_SET_NO_NEW_PRIVS:
2422 		if (arg2 != 1 || arg3 || arg4 || arg5)
2423 			return -EINVAL;
2424 
2425 		task_set_no_new_privs(current);
2426 		break;
2427 	case PR_GET_NO_NEW_PRIVS:
2428 		if (arg2 || arg3 || arg4 || arg5)
2429 			return -EINVAL;
2430 		return task_no_new_privs(current) ? 1 : 0;
2431 	case PR_GET_THP_DISABLE:
2432 		if (arg2 || arg3 || arg4 || arg5)
2433 			return -EINVAL;
2434 		error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2435 		break;
2436 	case PR_SET_THP_DISABLE:
2437 		if (arg3 || arg4 || arg5)
2438 			return -EINVAL;
2439 		if (down_write_killable(&me->mm->mmap_sem))
2440 			return -EINTR;
2441 		if (arg2)
2442 			set_bit(MMF_DISABLE_THP, &me->mm->flags);
2443 		else
2444 			clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2445 		up_write(&me->mm->mmap_sem);
2446 		break;
2447 	case PR_MPX_ENABLE_MANAGEMENT:
2448 		if (arg2 || arg3 || arg4 || arg5)
2449 			return -EINVAL;
2450 		error = MPX_ENABLE_MANAGEMENT();
2451 		break;
2452 	case PR_MPX_DISABLE_MANAGEMENT:
2453 		if (arg2 || arg3 || arg4 || arg5)
2454 			return -EINVAL;
2455 		error = MPX_DISABLE_MANAGEMENT();
2456 		break;
2457 	case PR_SET_FP_MODE:
2458 		error = SET_FP_MODE(me, arg2);
2459 		break;
2460 	case PR_GET_FP_MODE:
2461 		error = GET_FP_MODE(me);
2462 		break;
2463 	case PR_SVE_SET_VL:
2464 		error = SVE_SET_VL(arg2);
2465 		break;
2466 	case PR_SVE_GET_VL:
2467 		error = SVE_GET_VL();
2468 		break;
2469 	case PR_GET_SPECULATION_CTRL:
2470 		if (arg3 || arg4 || arg5)
2471 			return -EINVAL;
2472 		error = arch_prctl_spec_ctrl_get(me, arg2);
2473 		break;
2474 	case PR_SET_SPECULATION_CTRL:
2475 		if (arg4 || arg5)
2476 			return -EINVAL;
2477 		error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2478 		break;
2479 	default:
2480 		error = -EINVAL;
2481 		break;
2482 	}
2483 	return error;
2484 }
2485 
2486 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2487 		struct getcpu_cache __user *, unused)
2488 {
2489 	int err = 0;
2490 	int cpu = raw_smp_processor_id();
2491 
2492 	if (cpup)
2493 		err |= put_user(cpu, cpup);
2494 	if (nodep)
2495 		err |= put_user(cpu_to_node(cpu), nodep);
2496 	return err ? -EFAULT : 0;
2497 }
2498 
2499 /**
2500  * do_sysinfo - fill in sysinfo struct
2501  * @info: pointer to buffer to fill
2502  */
2503 static int do_sysinfo(struct sysinfo *info)
2504 {
2505 	unsigned long mem_total, sav_total;
2506 	unsigned int mem_unit, bitcount;
2507 	struct timespec64 tp;
2508 
2509 	memset(info, 0, sizeof(struct sysinfo));
2510 
2511 	ktime_get_boottime_ts64(&tp);
2512 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2513 
2514 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2515 
2516 	info->procs = nr_threads;
2517 
2518 	si_meminfo(info);
2519 	si_swapinfo(info);
2520 
2521 	/*
2522 	 * If the sum of all the available memory (i.e. ram + swap)
2523 	 * is less than can be stored in a 32 bit unsigned long then
2524 	 * we can be binary compatible with 2.2.x kernels.  If not,
2525 	 * well, in that case 2.2.x was broken anyways...
2526 	 *
2527 	 *  -Erik Andersen <andersee@debian.org>
2528 	 */
2529 
2530 	mem_total = info->totalram + info->totalswap;
2531 	if (mem_total < info->totalram || mem_total < info->totalswap)
2532 		goto out;
2533 	bitcount = 0;
2534 	mem_unit = info->mem_unit;
2535 	while (mem_unit > 1) {
2536 		bitcount++;
2537 		mem_unit >>= 1;
2538 		sav_total = mem_total;
2539 		mem_total <<= 1;
2540 		if (mem_total < sav_total)
2541 			goto out;
2542 	}
2543 
2544 	/*
2545 	 * If mem_total did not overflow, multiply all memory values by
2546 	 * info->mem_unit and set it to 1.  This leaves things compatible
2547 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2548 	 * kernels...
2549 	 */
2550 
2551 	info->mem_unit = 1;
2552 	info->totalram <<= bitcount;
2553 	info->freeram <<= bitcount;
2554 	info->sharedram <<= bitcount;
2555 	info->bufferram <<= bitcount;
2556 	info->totalswap <<= bitcount;
2557 	info->freeswap <<= bitcount;
2558 	info->totalhigh <<= bitcount;
2559 	info->freehigh <<= bitcount;
2560 
2561 out:
2562 	return 0;
2563 }
2564 
2565 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2566 {
2567 	struct sysinfo val;
2568 
2569 	do_sysinfo(&val);
2570 
2571 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2572 		return -EFAULT;
2573 
2574 	return 0;
2575 }
2576 
2577 #ifdef CONFIG_COMPAT
2578 struct compat_sysinfo {
2579 	s32 uptime;
2580 	u32 loads[3];
2581 	u32 totalram;
2582 	u32 freeram;
2583 	u32 sharedram;
2584 	u32 bufferram;
2585 	u32 totalswap;
2586 	u32 freeswap;
2587 	u16 procs;
2588 	u16 pad;
2589 	u32 totalhigh;
2590 	u32 freehigh;
2591 	u32 mem_unit;
2592 	char _f[20-2*sizeof(u32)-sizeof(int)];
2593 };
2594 
2595 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2596 {
2597 	struct sysinfo s;
2598 
2599 	do_sysinfo(&s);
2600 
2601 	/* Check to see if any memory value is too large for 32-bit and scale
2602 	 *  down if needed
2603 	 */
2604 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2605 		int bitcount = 0;
2606 
2607 		while (s.mem_unit < PAGE_SIZE) {
2608 			s.mem_unit <<= 1;
2609 			bitcount++;
2610 		}
2611 
2612 		s.totalram >>= bitcount;
2613 		s.freeram >>= bitcount;
2614 		s.sharedram >>= bitcount;
2615 		s.bufferram >>= bitcount;
2616 		s.totalswap >>= bitcount;
2617 		s.freeswap >>= bitcount;
2618 		s.totalhigh >>= bitcount;
2619 		s.freehigh >>= bitcount;
2620 	}
2621 
2622 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2623 	    __put_user(s.uptime, &info->uptime) ||
2624 	    __put_user(s.loads[0], &info->loads[0]) ||
2625 	    __put_user(s.loads[1], &info->loads[1]) ||
2626 	    __put_user(s.loads[2], &info->loads[2]) ||
2627 	    __put_user(s.totalram, &info->totalram) ||
2628 	    __put_user(s.freeram, &info->freeram) ||
2629 	    __put_user(s.sharedram, &info->sharedram) ||
2630 	    __put_user(s.bufferram, &info->bufferram) ||
2631 	    __put_user(s.totalswap, &info->totalswap) ||
2632 	    __put_user(s.freeswap, &info->freeswap) ||
2633 	    __put_user(s.procs, &info->procs) ||
2634 	    __put_user(s.totalhigh, &info->totalhigh) ||
2635 	    __put_user(s.freehigh, &info->freehigh) ||
2636 	    __put_user(s.mem_unit, &info->mem_unit))
2637 		return -EFAULT;
2638 
2639 	return 0;
2640 }
2641 #endif /* CONFIG_COMPAT */
2642