xref: /openbmc/linux/kernel/sys.c (revision 4a3fad70)
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/kmsg_dump.h>
65 /* Move somewhere else to avoid recompiling? */
66 #include <generated/utsrelease.h>
67 
68 #include <linux/uaccess.h>
69 #include <asm/io.h>
70 #include <asm/unistd.h>
71 
72 #ifndef SET_UNALIGN_CTL
73 # define SET_UNALIGN_CTL(a, b)	(-EINVAL)
74 #endif
75 #ifndef GET_UNALIGN_CTL
76 # define GET_UNALIGN_CTL(a, b)	(-EINVAL)
77 #endif
78 #ifndef SET_FPEMU_CTL
79 # define SET_FPEMU_CTL(a, b)	(-EINVAL)
80 #endif
81 #ifndef GET_FPEMU_CTL
82 # define GET_FPEMU_CTL(a, b)	(-EINVAL)
83 #endif
84 #ifndef SET_FPEXC_CTL
85 # define SET_FPEXC_CTL(a, b)	(-EINVAL)
86 #endif
87 #ifndef GET_FPEXC_CTL
88 # define GET_FPEXC_CTL(a, b)	(-EINVAL)
89 #endif
90 #ifndef GET_ENDIAN
91 # define GET_ENDIAN(a, b)	(-EINVAL)
92 #endif
93 #ifndef SET_ENDIAN
94 # define SET_ENDIAN(a, b)	(-EINVAL)
95 #endif
96 #ifndef GET_TSC_CTL
97 # define GET_TSC_CTL(a)		(-EINVAL)
98 #endif
99 #ifndef SET_TSC_CTL
100 # define SET_TSC_CTL(a)		(-EINVAL)
101 #endif
102 #ifndef MPX_ENABLE_MANAGEMENT
103 # define MPX_ENABLE_MANAGEMENT()	(-EINVAL)
104 #endif
105 #ifndef MPX_DISABLE_MANAGEMENT
106 # define MPX_DISABLE_MANAGEMENT()	(-EINVAL)
107 #endif
108 #ifndef GET_FP_MODE
109 # define GET_FP_MODE(a)		(-EINVAL)
110 #endif
111 #ifndef SET_FP_MODE
112 # define SET_FP_MODE(a,b)	(-EINVAL)
113 #endif
114 #ifndef SVE_SET_VL
115 # define SVE_SET_VL(a)		(-EINVAL)
116 #endif
117 #ifndef SVE_GET_VL
118 # define SVE_GET_VL()		(-EINVAL)
119 #endif
120 
121 /*
122  * this is where the system-wide overflow UID and GID are defined, for
123  * architectures that now have 32-bit UID/GID but didn't in the past
124  */
125 
126 int overflowuid = DEFAULT_OVERFLOWUID;
127 int overflowgid = DEFAULT_OVERFLOWGID;
128 
129 EXPORT_SYMBOL(overflowuid);
130 EXPORT_SYMBOL(overflowgid);
131 
132 /*
133  * the same as above, but for filesystems which can only store a 16-bit
134  * UID and GID. as such, this is needed on all architectures
135  */
136 
137 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
138 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
139 
140 EXPORT_SYMBOL(fs_overflowuid);
141 EXPORT_SYMBOL(fs_overflowgid);
142 
143 /*
144  * Returns true if current's euid is same as p's uid or euid,
145  * or has CAP_SYS_NICE to p's user_ns.
146  *
147  * Called with rcu_read_lock, creds are safe
148  */
149 static bool set_one_prio_perm(struct task_struct *p)
150 {
151 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
152 
153 	if (uid_eq(pcred->uid,  cred->euid) ||
154 	    uid_eq(pcred->euid, cred->euid))
155 		return true;
156 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
157 		return true;
158 	return false;
159 }
160 
161 /*
162  * set the priority of a task
163  * - the caller must hold the RCU read lock
164  */
165 static int set_one_prio(struct task_struct *p, int niceval, int error)
166 {
167 	int no_nice;
168 
169 	if (!set_one_prio_perm(p)) {
170 		error = -EPERM;
171 		goto out;
172 	}
173 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
174 		error = -EACCES;
175 		goto out;
176 	}
177 	no_nice = security_task_setnice(p, niceval);
178 	if (no_nice) {
179 		error = no_nice;
180 		goto out;
181 	}
182 	if (error == -ESRCH)
183 		error = 0;
184 	set_user_nice(p, niceval);
185 out:
186 	return error;
187 }
188 
189 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
190 {
191 	struct task_struct *g, *p;
192 	struct user_struct *user;
193 	const struct cred *cred = current_cred();
194 	int error = -EINVAL;
195 	struct pid *pgrp;
196 	kuid_t uid;
197 
198 	if (which > PRIO_USER || which < PRIO_PROCESS)
199 		goto out;
200 
201 	/* normalize: avoid signed division (rounding problems) */
202 	error = -ESRCH;
203 	if (niceval < MIN_NICE)
204 		niceval = MIN_NICE;
205 	if (niceval > MAX_NICE)
206 		niceval = MAX_NICE;
207 
208 	rcu_read_lock();
209 	read_lock(&tasklist_lock);
210 	switch (which) {
211 	case PRIO_PROCESS:
212 		if (who)
213 			p = find_task_by_vpid(who);
214 		else
215 			p = current;
216 		if (p)
217 			error = set_one_prio(p, niceval, error);
218 		break;
219 	case PRIO_PGRP:
220 		if (who)
221 			pgrp = find_vpid(who);
222 		else
223 			pgrp = task_pgrp(current);
224 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
225 			error = set_one_prio(p, niceval, error);
226 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
227 		break;
228 	case PRIO_USER:
229 		uid = make_kuid(cred->user_ns, who);
230 		user = cred->user;
231 		if (!who)
232 			uid = cred->uid;
233 		else if (!uid_eq(uid, cred->uid)) {
234 			user = find_user(uid);
235 			if (!user)
236 				goto out_unlock;	/* No processes for this user */
237 		}
238 		do_each_thread(g, p) {
239 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
240 				error = set_one_prio(p, niceval, error);
241 		} while_each_thread(g, p);
242 		if (!uid_eq(uid, cred->uid))
243 			free_uid(user);		/* For find_user() */
244 		break;
245 	}
246 out_unlock:
247 	read_unlock(&tasklist_lock);
248 	rcu_read_unlock();
249 out:
250 	return error;
251 }
252 
253 /*
254  * Ugh. To avoid negative return values, "getpriority()" will
255  * not return the normal nice-value, but a negated value that
256  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
257  * to stay compatible.
258  */
259 SYSCALL_DEFINE2(getpriority, int, which, int, who)
260 {
261 	struct task_struct *g, *p;
262 	struct user_struct *user;
263 	const struct cred *cred = current_cred();
264 	long niceval, retval = -ESRCH;
265 	struct pid *pgrp;
266 	kuid_t uid;
267 
268 	if (which > PRIO_USER || which < PRIO_PROCESS)
269 		return -EINVAL;
270 
271 	rcu_read_lock();
272 	read_lock(&tasklist_lock);
273 	switch (which) {
274 	case PRIO_PROCESS:
275 		if (who)
276 			p = find_task_by_vpid(who);
277 		else
278 			p = current;
279 		if (p) {
280 			niceval = nice_to_rlimit(task_nice(p));
281 			if (niceval > retval)
282 				retval = niceval;
283 		}
284 		break;
285 	case PRIO_PGRP:
286 		if (who)
287 			pgrp = find_vpid(who);
288 		else
289 			pgrp = task_pgrp(current);
290 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
291 			niceval = nice_to_rlimit(task_nice(p));
292 			if (niceval > retval)
293 				retval = niceval;
294 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
295 		break;
296 	case PRIO_USER:
297 		uid = make_kuid(cred->user_ns, who);
298 		user = cred->user;
299 		if (!who)
300 			uid = cred->uid;
301 		else if (!uid_eq(uid, cred->uid)) {
302 			user = find_user(uid);
303 			if (!user)
304 				goto out_unlock;	/* No processes for this user */
305 		}
306 		do_each_thread(g, p) {
307 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
308 				niceval = nice_to_rlimit(task_nice(p));
309 				if (niceval > retval)
310 					retval = niceval;
311 			}
312 		} while_each_thread(g, p);
313 		if (!uid_eq(uid, cred->uid))
314 			free_uid(user);		/* for find_user() */
315 		break;
316 	}
317 out_unlock:
318 	read_unlock(&tasklist_lock);
319 	rcu_read_unlock();
320 
321 	return retval;
322 }
323 
324 /*
325  * Unprivileged users may change the real gid to the effective gid
326  * or vice versa.  (BSD-style)
327  *
328  * If you set the real gid at all, or set the effective gid to a value not
329  * equal to the real gid, then the saved gid is set to the new effective gid.
330  *
331  * This makes it possible for a setgid program to completely drop its
332  * privileges, which is often a useful assertion to make when you are doing
333  * a security audit over a program.
334  *
335  * The general idea is that a program which uses just setregid() will be
336  * 100% compatible with BSD.  A program which uses just setgid() will be
337  * 100% compatible with POSIX with saved IDs.
338  *
339  * SMP: There are not races, the GIDs are checked only by filesystem
340  *      operations (as far as semantic preservation is concerned).
341  */
342 #ifdef CONFIG_MULTIUSER
343 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
344 {
345 	struct user_namespace *ns = current_user_ns();
346 	const struct cred *old;
347 	struct cred *new;
348 	int retval;
349 	kgid_t krgid, kegid;
350 
351 	krgid = make_kgid(ns, rgid);
352 	kegid = make_kgid(ns, egid);
353 
354 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
355 		return -EINVAL;
356 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
357 		return -EINVAL;
358 
359 	new = prepare_creds();
360 	if (!new)
361 		return -ENOMEM;
362 	old = current_cred();
363 
364 	retval = -EPERM;
365 	if (rgid != (gid_t) -1) {
366 		if (gid_eq(old->gid, krgid) ||
367 		    gid_eq(old->egid, krgid) ||
368 		    ns_capable(old->user_ns, CAP_SETGID))
369 			new->gid = krgid;
370 		else
371 			goto error;
372 	}
373 	if (egid != (gid_t) -1) {
374 		if (gid_eq(old->gid, kegid) ||
375 		    gid_eq(old->egid, kegid) ||
376 		    gid_eq(old->sgid, kegid) ||
377 		    ns_capable(old->user_ns, CAP_SETGID))
378 			new->egid = kegid;
379 		else
380 			goto error;
381 	}
382 
383 	if (rgid != (gid_t) -1 ||
384 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
385 		new->sgid = new->egid;
386 	new->fsgid = new->egid;
387 
388 	return commit_creds(new);
389 
390 error:
391 	abort_creds(new);
392 	return retval;
393 }
394 
395 /*
396  * setgid() is implemented like SysV w/ SAVED_IDS
397  *
398  * SMP: Same implicit races as above.
399  */
400 SYSCALL_DEFINE1(setgid, gid_t, gid)
401 {
402 	struct user_namespace *ns = current_user_ns();
403 	const struct cred *old;
404 	struct cred *new;
405 	int retval;
406 	kgid_t kgid;
407 
408 	kgid = make_kgid(ns, gid);
409 	if (!gid_valid(kgid))
410 		return -EINVAL;
411 
412 	new = prepare_creds();
413 	if (!new)
414 		return -ENOMEM;
415 	old = current_cred();
416 
417 	retval = -EPERM;
418 	if (ns_capable(old->user_ns, CAP_SETGID))
419 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
420 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
421 		new->egid = new->fsgid = kgid;
422 	else
423 		goto error;
424 
425 	return commit_creds(new);
426 
427 error:
428 	abort_creds(new);
429 	return retval;
430 }
431 
432 /*
433  * change the user struct in a credentials set to match the new UID
434  */
435 static int set_user(struct cred *new)
436 {
437 	struct user_struct *new_user;
438 
439 	new_user = alloc_uid(new->uid);
440 	if (!new_user)
441 		return -EAGAIN;
442 
443 	/*
444 	 * We don't fail in case of NPROC limit excess here because too many
445 	 * poorly written programs don't check set*uid() return code, assuming
446 	 * it never fails if called by root.  We may still enforce NPROC limit
447 	 * for programs doing set*uid()+execve() by harmlessly deferring the
448 	 * failure to the execve() stage.
449 	 */
450 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
451 			new_user != INIT_USER)
452 		current->flags |= PF_NPROC_EXCEEDED;
453 	else
454 		current->flags &= ~PF_NPROC_EXCEEDED;
455 
456 	free_uid(new->user);
457 	new->user = new_user;
458 	return 0;
459 }
460 
461 /*
462  * Unprivileged users may change the real uid to the effective uid
463  * or vice versa.  (BSD-style)
464  *
465  * If you set the real uid at all, or set the effective uid to a value not
466  * equal to the real uid, then the saved uid is set to the new effective uid.
467  *
468  * This makes it possible for a setuid program to completely drop its
469  * privileges, which is often a useful assertion to make when you are doing
470  * a security audit over a program.
471  *
472  * The general idea is that a program which uses just setreuid() will be
473  * 100% compatible with BSD.  A program which uses just setuid() will be
474  * 100% compatible with POSIX with saved IDs.
475  */
476 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
477 {
478 	struct user_namespace *ns = current_user_ns();
479 	const struct cred *old;
480 	struct cred *new;
481 	int retval;
482 	kuid_t kruid, keuid;
483 
484 	kruid = make_kuid(ns, ruid);
485 	keuid = make_kuid(ns, euid);
486 
487 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
488 		return -EINVAL;
489 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
490 		return -EINVAL;
491 
492 	new = prepare_creds();
493 	if (!new)
494 		return -ENOMEM;
495 	old = current_cred();
496 
497 	retval = -EPERM;
498 	if (ruid != (uid_t) -1) {
499 		new->uid = kruid;
500 		if (!uid_eq(old->uid, kruid) &&
501 		    !uid_eq(old->euid, kruid) &&
502 		    !ns_capable(old->user_ns, CAP_SETUID))
503 			goto error;
504 	}
505 
506 	if (euid != (uid_t) -1) {
507 		new->euid = keuid;
508 		if (!uid_eq(old->uid, keuid) &&
509 		    !uid_eq(old->euid, keuid) &&
510 		    !uid_eq(old->suid, keuid) &&
511 		    !ns_capable(old->user_ns, CAP_SETUID))
512 			goto error;
513 	}
514 
515 	if (!uid_eq(new->uid, old->uid)) {
516 		retval = set_user(new);
517 		if (retval < 0)
518 			goto error;
519 	}
520 	if (ruid != (uid_t) -1 ||
521 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
522 		new->suid = new->euid;
523 	new->fsuid = new->euid;
524 
525 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
526 	if (retval < 0)
527 		goto error;
528 
529 	return commit_creds(new);
530 
531 error:
532 	abort_creds(new);
533 	return retval;
534 }
535 
536 /*
537  * setuid() is implemented like SysV with SAVED_IDS
538  *
539  * Note that SAVED_ID's is deficient in that a setuid root program
540  * like sendmail, for example, cannot set its uid to be a normal
541  * user and then switch back, because if you're root, setuid() sets
542  * the saved uid too.  If you don't like this, blame the bright people
543  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
544  * will allow a root program to temporarily drop privileges and be able to
545  * regain them by swapping the real and effective uid.
546  */
547 SYSCALL_DEFINE1(setuid, uid_t, uid)
548 {
549 	struct user_namespace *ns = current_user_ns();
550 	const struct cred *old;
551 	struct cred *new;
552 	int retval;
553 	kuid_t kuid;
554 
555 	kuid = make_kuid(ns, uid);
556 	if (!uid_valid(kuid))
557 		return -EINVAL;
558 
559 	new = prepare_creds();
560 	if (!new)
561 		return -ENOMEM;
562 	old = current_cred();
563 
564 	retval = -EPERM;
565 	if (ns_capable(old->user_ns, CAP_SETUID)) {
566 		new->suid = new->uid = kuid;
567 		if (!uid_eq(kuid, old->uid)) {
568 			retval = set_user(new);
569 			if (retval < 0)
570 				goto error;
571 		}
572 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
573 		goto error;
574 	}
575 
576 	new->fsuid = new->euid = kuid;
577 
578 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
579 	if (retval < 0)
580 		goto error;
581 
582 	return commit_creds(new);
583 
584 error:
585 	abort_creds(new);
586 	return retval;
587 }
588 
589 
590 /*
591  * This function implements a generic ability to update ruid, euid,
592  * and suid.  This allows you to implement the 4.4 compatible seteuid().
593  */
594 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
595 {
596 	struct user_namespace *ns = current_user_ns();
597 	const struct cred *old;
598 	struct cred *new;
599 	int retval;
600 	kuid_t kruid, keuid, ksuid;
601 
602 	kruid = make_kuid(ns, ruid);
603 	keuid = make_kuid(ns, euid);
604 	ksuid = make_kuid(ns, suid);
605 
606 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
607 		return -EINVAL;
608 
609 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
610 		return -EINVAL;
611 
612 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
613 		return -EINVAL;
614 
615 	new = prepare_creds();
616 	if (!new)
617 		return -ENOMEM;
618 
619 	old = current_cred();
620 
621 	retval = -EPERM;
622 	if (!ns_capable(old->user_ns, CAP_SETUID)) {
623 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
624 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
625 			goto error;
626 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
627 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
628 			goto error;
629 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
630 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
631 			goto error;
632 	}
633 
634 	if (ruid != (uid_t) -1) {
635 		new->uid = kruid;
636 		if (!uid_eq(kruid, old->uid)) {
637 			retval = set_user(new);
638 			if (retval < 0)
639 				goto error;
640 		}
641 	}
642 	if (euid != (uid_t) -1)
643 		new->euid = keuid;
644 	if (suid != (uid_t) -1)
645 		new->suid = ksuid;
646 	new->fsuid = new->euid;
647 
648 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
649 	if (retval < 0)
650 		goto error;
651 
652 	return commit_creds(new);
653 
654 error:
655 	abort_creds(new);
656 	return retval;
657 }
658 
659 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
660 {
661 	const struct cred *cred = current_cred();
662 	int retval;
663 	uid_t ruid, euid, suid;
664 
665 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
666 	euid = from_kuid_munged(cred->user_ns, cred->euid);
667 	suid = from_kuid_munged(cred->user_ns, cred->suid);
668 
669 	retval = put_user(ruid, ruidp);
670 	if (!retval) {
671 		retval = put_user(euid, euidp);
672 		if (!retval)
673 			return put_user(suid, suidp);
674 	}
675 	return retval;
676 }
677 
678 /*
679  * Same as above, but for rgid, egid, sgid.
680  */
681 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
682 {
683 	struct user_namespace *ns = current_user_ns();
684 	const struct cred *old;
685 	struct cred *new;
686 	int retval;
687 	kgid_t krgid, kegid, ksgid;
688 
689 	krgid = make_kgid(ns, rgid);
690 	kegid = make_kgid(ns, egid);
691 	ksgid = make_kgid(ns, sgid);
692 
693 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
694 		return -EINVAL;
695 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
696 		return -EINVAL;
697 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
698 		return -EINVAL;
699 
700 	new = prepare_creds();
701 	if (!new)
702 		return -ENOMEM;
703 	old = current_cred();
704 
705 	retval = -EPERM;
706 	if (!ns_capable(old->user_ns, CAP_SETGID)) {
707 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
708 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
709 			goto error;
710 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
711 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
712 			goto error;
713 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
714 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
715 			goto error;
716 	}
717 
718 	if (rgid != (gid_t) -1)
719 		new->gid = krgid;
720 	if (egid != (gid_t) -1)
721 		new->egid = kegid;
722 	if (sgid != (gid_t) -1)
723 		new->sgid = ksgid;
724 	new->fsgid = new->egid;
725 
726 	return commit_creds(new);
727 
728 error:
729 	abort_creds(new);
730 	return retval;
731 }
732 
733 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
734 {
735 	const struct cred *cred = current_cred();
736 	int retval;
737 	gid_t rgid, egid, sgid;
738 
739 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
740 	egid = from_kgid_munged(cred->user_ns, cred->egid);
741 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
742 
743 	retval = put_user(rgid, rgidp);
744 	if (!retval) {
745 		retval = put_user(egid, egidp);
746 		if (!retval)
747 			retval = put_user(sgid, sgidp);
748 	}
749 
750 	return retval;
751 }
752 
753 
754 /*
755  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
756  * is used for "access()" and for the NFS daemon (letting nfsd stay at
757  * whatever uid it wants to). It normally shadows "euid", except when
758  * explicitly set by setfsuid() or for access..
759  */
760 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
761 {
762 	const struct cred *old;
763 	struct cred *new;
764 	uid_t old_fsuid;
765 	kuid_t kuid;
766 
767 	old = current_cred();
768 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
769 
770 	kuid = make_kuid(old->user_ns, uid);
771 	if (!uid_valid(kuid))
772 		return old_fsuid;
773 
774 	new = prepare_creds();
775 	if (!new)
776 		return old_fsuid;
777 
778 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
779 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
780 	    ns_capable(old->user_ns, CAP_SETUID)) {
781 		if (!uid_eq(kuid, old->fsuid)) {
782 			new->fsuid = kuid;
783 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
784 				goto change_okay;
785 		}
786 	}
787 
788 	abort_creds(new);
789 	return old_fsuid;
790 
791 change_okay:
792 	commit_creds(new);
793 	return old_fsuid;
794 }
795 
796 /*
797  * Samma på svenska..
798  */
799 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
800 {
801 	const struct cred *old;
802 	struct cred *new;
803 	gid_t old_fsgid;
804 	kgid_t kgid;
805 
806 	old = current_cred();
807 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
808 
809 	kgid = make_kgid(old->user_ns, gid);
810 	if (!gid_valid(kgid))
811 		return old_fsgid;
812 
813 	new = prepare_creds();
814 	if (!new)
815 		return old_fsgid;
816 
817 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
818 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
819 	    ns_capable(old->user_ns, CAP_SETGID)) {
820 		if (!gid_eq(kgid, old->fsgid)) {
821 			new->fsgid = kgid;
822 			goto change_okay;
823 		}
824 	}
825 
826 	abort_creds(new);
827 	return old_fsgid;
828 
829 change_okay:
830 	commit_creds(new);
831 	return old_fsgid;
832 }
833 #endif /* CONFIG_MULTIUSER */
834 
835 /**
836  * sys_getpid - return the thread group id of the current process
837  *
838  * Note, despite the name, this returns the tgid not the pid.  The tgid and
839  * the pid are identical unless CLONE_THREAD was specified on clone() in
840  * which case the tgid is the same in all threads of the same group.
841  *
842  * This is SMP safe as current->tgid does not change.
843  */
844 SYSCALL_DEFINE0(getpid)
845 {
846 	return task_tgid_vnr(current);
847 }
848 
849 /* Thread ID - the internal kernel "pid" */
850 SYSCALL_DEFINE0(gettid)
851 {
852 	return task_pid_vnr(current);
853 }
854 
855 /*
856  * Accessing ->real_parent is not SMP-safe, it could
857  * change from under us. However, we can use a stale
858  * value of ->real_parent under rcu_read_lock(), see
859  * release_task()->call_rcu(delayed_put_task_struct).
860  */
861 SYSCALL_DEFINE0(getppid)
862 {
863 	int pid;
864 
865 	rcu_read_lock();
866 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
867 	rcu_read_unlock();
868 
869 	return pid;
870 }
871 
872 SYSCALL_DEFINE0(getuid)
873 {
874 	/* Only we change this so SMP safe */
875 	return from_kuid_munged(current_user_ns(), current_uid());
876 }
877 
878 SYSCALL_DEFINE0(geteuid)
879 {
880 	/* Only we change this so SMP safe */
881 	return from_kuid_munged(current_user_ns(), current_euid());
882 }
883 
884 SYSCALL_DEFINE0(getgid)
885 {
886 	/* Only we change this so SMP safe */
887 	return from_kgid_munged(current_user_ns(), current_gid());
888 }
889 
890 SYSCALL_DEFINE0(getegid)
891 {
892 	/* Only we change this so SMP safe */
893 	return from_kgid_munged(current_user_ns(), current_egid());
894 }
895 
896 static void do_sys_times(struct tms *tms)
897 {
898 	u64 tgutime, tgstime, cutime, cstime;
899 
900 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
901 	cutime = current->signal->cutime;
902 	cstime = current->signal->cstime;
903 	tms->tms_utime = nsec_to_clock_t(tgutime);
904 	tms->tms_stime = nsec_to_clock_t(tgstime);
905 	tms->tms_cutime = nsec_to_clock_t(cutime);
906 	tms->tms_cstime = nsec_to_clock_t(cstime);
907 }
908 
909 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
910 {
911 	if (tbuf) {
912 		struct tms tmp;
913 
914 		do_sys_times(&tmp);
915 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
916 			return -EFAULT;
917 	}
918 	force_successful_syscall_return();
919 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
920 }
921 
922 #ifdef CONFIG_COMPAT
923 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
924 {
925 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
926 }
927 
928 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
929 {
930 	if (tbuf) {
931 		struct tms tms;
932 		struct compat_tms tmp;
933 
934 		do_sys_times(&tms);
935 		/* Convert our struct tms to the compat version. */
936 		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
937 		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
938 		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
939 		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
940 		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
941 			return -EFAULT;
942 	}
943 	force_successful_syscall_return();
944 	return compat_jiffies_to_clock_t(jiffies);
945 }
946 #endif
947 
948 /*
949  * This needs some heavy checking ...
950  * I just haven't the stomach for it. I also don't fully
951  * understand sessions/pgrp etc. Let somebody who does explain it.
952  *
953  * OK, I think I have the protection semantics right.... this is really
954  * only important on a multi-user system anyway, to make sure one user
955  * can't send a signal to a process owned by another.  -TYT, 12/12/91
956  *
957  * !PF_FORKNOEXEC check to conform completely to POSIX.
958  */
959 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
960 {
961 	struct task_struct *p;
962 	struct task_struct *group_leader = current->group_leader;
963 	struct pid *pgrp;
964 	int err;
965 
966 	if (!pid)
967 		pid = task_pid_vnr(group_leader);
968 	if (!pgid)
969 		pgid = pid;
970 	if (pgid < 0)
971 		return -EINVAL;
972 	rcu_read_lock();
973 
974 	/* From this point forward we keep holding onto the tasklist lock
975 	 * so that our parent does not change from under us. -DaveM
976 	 */
977 	write_lock_irq(&tasklist_lock);
978 
979 	err = -ESRCH;
980 	p = find_task_by_vpid(pid);
981 	if (!p)
982 		goto out;
983 
984 	err = -EINVAL;
985 	if (!thread_group_leader(p))
986 		goto out;
987 
988 	if (same_thread_group(p->real_parent, group_leader)) {
989 		err = -EPERM;
990 		if (task_session(p) != task_session(group_leader))
991 			goto out;
992 		err = -EACCES;
993 		if (!(p->flags & PF_FORKNOEXEC))
994 			goto out;
995 	} else {
996 		err = -ESRCH;
997 		if (p != group_leader)
998 			goto out;
999 	}
1000 
1001 	err = -EPERM;
1002 	if (p->signal->leader)
1003 		goto out;
1004 
1005 	pgrp = task_pid(p);
1006 	if (pgid != pid) {
1007 		struct task_struct *g;
1008 
1009 		pgrp = find_vpid(pgid);
1010 		g = pid_task(pgrp, PIDTYPE_PGID);
1011 		if (!g || task_session(g) != task_session(group_leader))
1012 			goto out;
1013 	}
1014 
1015 	err = security_task_setpgid(p, pgid);
1016 	if (err)
1017 		goto out;
1018 
1019 	if (task_pgrp(p) != pgrp)
1020 		change_pid(p, PIDTYPE_PGID, pgrp);
1021 
1022 	err = 0;
1023 out:
1024 	/* All paths lead to here, thus we are safe. -DaveM */
1025 	write_unlock_irq(&tasklist_lock);
1026 	rcu_read_unlock();
1027 	return err;
1028 }
1029 
1030 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1031 {
1032 	struct task_struct *p;
1033 	struct pid *grp;
1034 	int retval;
1035 
1036 	rcu_read_lock();
1037 	if (!pid)
1038 		grp = task_pgrp(current);
1039 	else {
1040 		retval = -ESRCH;
1041 		p = find_task_by_vpid(pid);
1042 		if (!p)
1043 			goto out;
1044 		grp = task_pgrp(p);
1045 		if (!grp)
1046 			goto out;
1047 
1048 		retval = security_task_getpgid(p);
1049 		if (retval)
1050 			goto out;
1051 	}
1052 	retval = pid_vnr(grp);
1053 out:
1054 	rcu_read_unlock();
1055 	return retval;
1056 }
1057 
1058 #ifdef __ARCH_WANT_SYS_GETPGRP
1059 
1060 SYSCALL_DEFINE0(getpgrp)
1061 {
1062 	return sys_getpgid(0);
1063 }
1064 
1065 #endif
1066 
1067 SYSCALL_DEFINE1(getsid, pid_t, pid)
1068 {
1069 	struct task_struct *p;
1070 	struct pid *sid;
1071 	int retval;
1072 
1073 	rcu_read_lock();
1074 	if (!pid)
1075 		sid = task_session(current);
1076 	else {
1077 		retval = -ESRCH;
1078 		p = find_task_by_vpid(pid);
1079 		if (!p)
1080 			goto out;
1081 		sid = task_session(p);
1082 		if (!sid)
1083 			goto out;
1084 
1085 		retval = security_task_getsid(p);
1086 		if (retval)
1087 			goto out;
1088 	}
1089 	retval = pid_vnr(sid);
1090 out:
1091 	rcu_read_unlock();
1092 	return retval;
1093 }
1094 
1095 static void set_special_pids(struct pid *pid)
1096 {
1097 	struct task_struct *curr = current->group_leader;
1098 
1099 	if (task_session(curr) != pid)
1100 		change_pid(curr, PIDTYPE_SID, pid);
1101 
1102 	if (task_pgrp(curr) != pid)
1103 		change_pid(curr, PIDTYPE_PGID, pid);
1104 }
1105 
1106 SYSCALL_DEFINE0(setsid)
1107 {
1108 	struct task_struct *group_leader = current->group_leader;
1109 	struct pid *sid = task_pid(group_leader);
1110 	pid_t session = pid_vnr(sid);
1111 	int err = -EPERM;
1112 
1113 	write_lock_irq(&tasklist_lock);
1114 	/* Fail if I am already a session leader */
1115 	if (group_leader->signal->leader)
1116 		goto out;
1117 
1118 	/* Fail if a process group id already exists that equals the
1119 	 * proposed session id.
1120 	 */
1121 	if (pid_task(sid, PIDTYPE_PGID))
1122 		goto out;
1123 
1124 	group_leader->signal->leader = 1;
1125 	set_special_pids(sid);
1126 
1127 	proc_clear_tty(group_leader);
1128 
1129 	err = session;
1130 out:
1131 	write_unlock_irq(&tasklist_lock);
1132 	if (err > 0) {
1133 		proc_sid_connector(group_leader);
1134 		sched_autogroup_create_attach(group_leader);
1135 	}
1136 	return err;
1137 }
1138 
1139 DECLARE_RWSEM(uts_sem);
1140 
1141 #ifdef COMPAT_UTS_MACHINE
1142 #define override_architecture(name) \
1143 	(personality(current->personality) == PER_LINUX32 && \
1144 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1145 		      sizeof(COMPAT_UTS_MACHINE)))
1146 #else
1147 #define override_architecture(name)	0
1148 #endif
1149 
1150 /*
1151  * Work around broken programs that cannot handle "Linux 3.0".
1152  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1153  * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1154  */
1155 static int override_release(char __user *release, size_t len)
1156 {
1157 	int ret = 0;
1158 
1159 	if (current->personality & UNAME26) {
1160 		const char *rest = UTS_RELEASE;
1161 		char buf[65] = { 0 };
1162 		int ndots = 0;
1163 		unsigned v;
1164 		size_t copy;
1165 
1166 		while (*rest) {
1167 			if (*rest == '.' && ++ndots >= 3)
1168 				break;
1169 			if (!isdigit(*rest) && *rest != '.')
1170 				break;
1171 			rest++;
1172 		}
1173 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1174 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1175 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1176 		ret = copy_to_user(release, buf, copy + 1);
1177 	}
1178 	return ret;
1179 }
1180 
1181 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1182 {
1183 	int errno = 0;
1184 
1185 	down_read(&uts_sem);
1186 	if (copy_to_user(name, utsname(), sizeof *name))
1187 		errno = -EFAULT;
1188 	up_read(&uts_sem);
1189 
1190 	if (!errno && override_release(name->release, sizeof(name->release)))
1191 		errno = -EFAULT;
1192 	if (!errno && override_architecture(name))
1193 		errno = -EFAULT;
1194 	return errno;
1195 }
1196 
1197 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1198 /*
1199  * Old cruft
1200  */
1201 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1202 {
1203 	int error = 0;
1204 
1205 	if (!name)
1206 		return -EFAULT;
1207 
1208 	down_read(&uts_sem);
1209 	if (copy_to_user(name, utsname(), sizeof(*name)))
1210 		error = -EFAULT;
1211 	up_read(&uts_sem);
1212 
1213 	if (!error && override_release(name->release, sizeof(name->release)))
1214 		error = -EFAULT;
1215 	if (!error && override_architecture(name))
1216 		error = -EFAULT;
1217 	return error;
1218 }
1219 
1220 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1221 {
1222 	int error;
1223 
1224 	if (!name)
1225 		return -EFAULT;
1226 	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1227 		return -EFAULT;
1228 
1229 	down_read(&uts_sem);
1230 	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1231 			       __OLD_UTS_LEN);
1232 	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1233 	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1234 				__OLD_UTS_LEN);
1235 	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1236 	error |= __copy_to_user(&name->release, &utsname()->release,
1237 				__OLD_UTS_LEN);
1238 	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1239 	error |= __copy_to_user(&name->version, &utsname()->version,
1240 				__OLD_UTS_LEN);
1241 	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1242 	error |= __copy_to_user(&name->machine, &utsname()->machine,
1243 				__OLD_UTS_LEN);
1244 	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1245 	up_read(&uts_sem);
1246 
1247 	if (!error && override_architecture(name))
1248 		error = -EFAULT;
1249 	if (!error && override_release(name->release, sizeof(name->release)))
1250 		error = -EFAULT;
1251 	return error ? -EFAULT : 0;
1252 }
1253 #endif
1254 
1255 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1256 {
1257 	int errno;
1258 	char tmp[__NEW_UTS_LEN];
1259 
1260 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1261 		return -EPERM;
1262 
1263 	if (len < 0 || len > __NEW_UTS_LEN)
1264 		return -EINVAL;
1265 	down_write(&uts_sem);
1266 	errno = -EFAULT;
1267 	if (!copy_from_user(tmp, name, len)) {
1268 		struct new_utsname *u = utsname();
1269 
1270 		memcpy(u->nodename, tmp, len);
1271 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1272 		errno = 0;
1273 		uts_proc_notify(UTS_PROC_HOSTNAME);
1274 	}
1275 	up_write(&uts_sem);
1276 	return errno;
1277 }
1278 
1279 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1280 
1281 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1282 {
1283 	int i, errno;
1284 	struct new_utsname *u;
1285 
1286 	if (len < 0)
1287 		return -EINVAL;
1288 	down_read(&uts_sem);
1289 	u = utsname();
1290 	i = 1 + strlen(u->nodename);
1291 	if (i > len)
1292 		i = len;
1293 	errno = 0;
1294 	if (copy_to_user(name, u->nodename, i))
1295 		errno = -EFAULT;
1296 	up_read(&uts_sem);
1297 	return errno;
1298 }
1299 
1300 #endif
1301 
1302 /*
1303  * Only setdomainname; getdomainname can be implemented by calling
1304  * uname()
1305  */
1306 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1307 {
1308 	int errno;
1309 	char tmp[__NEW_UTS_LEN];
1310 
1311 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1312 		return -EPERM;
1313 	if (len < 0 || len > __NEW_UTS_LEN)
1314 		return -EINVAL;
1315 
1316 	down_write(&uts_sem);
1317 	errno = -EFAULT;
1318 	if (!copy_from_user(tmp, name, len)) {
1319 		struct new_utsname *u = utsname();
1320 
1321 		memcpy(u->domainname, tmp, len);
1322 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1323 		errno = 0;
1324 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1325 	}
1326 	up_write(&uts_sem);
1327 	return errno;
1328 }
1329 
1330 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1331 {
1332 	struct rlimit value;
1333 	int ret;
1334 
1335 	ret = do_prlimit(current, resource, NULL, &value);
1336 	if (!ret)
1337 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1338 
1339 	return ret;
1340 }
1341 
1342 #ifdef CONFIG_COMPAT
1343 
1344 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1345 		       struct compat_rlimit __user *, rlim)
1346 {
1347 	struct rlimit r;
1348 	struct compat_rlimit r32;
1349 
1350 	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1351 		return -EFAULT;
1352 
1353 	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1354 		r.rlim_cur = RLIM_INFINITY;
1355 	else
1356 		r.rlim_cur = r32.rlim_cur;
1357 	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1358 		r.rlim_max = RLIM_INFINITY;
1359 	else
1360 		r.rlim_max = r32.rlim_max;
1361 	return do_prlimit(current, resource, &r, NULL);
1362 }
1363 
1364 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1365 		       struct compat_rlimit __user *, rlim)
1366 {
1367 	struct rlimit r;
1368 	int ret;
1369 
1370 	ret = do_prlimit(current, resource, NULL, &r);
1371 	if (!ret) {
1372 		struct compat_rlimit r32;
1373 		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1374 			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1375 		else
1376 			r32.rlim_cur = r.rlim_cur;
1377 		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1378 			r32.rlim_max = COMPAT_RLIM_INFINITY;
1379 		else
1380 			r32.rlim_max = r.rlim_max;
1381 
1382 		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1383 			return -EFAULT;
1384 	}
1385 	return ret;
1386 }
1387 
1388 #endif
1389 
1390 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1391 
1392 /*
1393  *	Back compatibility for getrlimit. Needed for some apps.
1394  */
1395 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1396 		struct rlimit __user *, rlim)
1397 {
1398 	struct rlimit x;
1399 	if (resource >= RLIM_NLIMITS)
1400 		return -EINVAL;
1401 
1402 	task_lock(current->group_leader);
1403 	x = current->signal->rlim[resource];
1404 	task_unlock(current->group_leader);
1405 	if (x.rlim_cur > 0x7FFFFFFF)
1406 		x.rlim_cur = 0x7FFFFFFF;
1407 	if (x.rlim_max > 0x7FFFFFFF)
1408 		x.rlim_max = 0x7FFFFFFF;
1409 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1410 }
1411 
1412 #ifdef CONFIG_COMPAT
1413 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1414 		       struct compat_rlimit __user *, rlim)
1415 {
1416 	struct rlimit r;
1417 
1418 	if (resource >= RLIM_NLIMITS)
1419 		return -EINVAL;
1420 
1421 	task_lock(current->group_leader);
1422 	r = current->signal->rlim[resource];
1423 	task_unlock(current->group_leader);
1424 	if (r.rlim_cur > 0x7FFFFFFF)
1425 		r.rlim_cur = 0x7FFFFFFF;
1426 	if (r.rlim_max > 0x7FFFFFFF)
1427 		r.rlim_max = 0x7FFFFFFF;
1428 
1429 	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1430 	    put_user(r.rlim_max, &rlim->rlim_max))
1431 		return -EFAULT;
1432 	return 0;
1433 }
1434 #endif
1435 
1436 #endif
1437 
1438 static inline bool rlim64_is_infinity(__u64 rlim64)
1439 {
1440 #if BITS_PER_LONG < 64
1441 	return rlim64 >= ULONG_MAX;
1442 #else
1443 	return rlim64 == RLIM64_INFINITY;
1444 #endif
1445 }
1446 
1447 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1448 {
1449 	if (rlim->rlim_cur == RLIM_INFINITY)
1450 		rlim64->rlim_cur = RLIM64_INFINITY;
1451 	else
1452 		rlim64->rlim_cur = rlim->rlim_cur;
1453 	if (rlim->rlim_max == RLIM_INFINITY)
1454 		rlim64->rlim_max = RLIM64_INFINITY;
1455 	else
1456 		rlim64->rlim_max = rlim->rlim_max;
1457 }
1458 
1459 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1460 {
1461 	if (rlim64_is_infinity(rlim64->rlim_cur))
1462 		rlim->rlim_cur = RLIM_INFINITY;
1463 	else
1464 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1465 	if (rlim64_is_infinity(rlim64->rlim_max))
1466 		rlim->rlim_max = RLIM_INFINITY;
1467 	else
1468 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1469 }
1470 
1471 /* make sure you are allowed to change @tsk limits before calling this */
1472 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1473 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1474 {
1475 	struct rlimit *rlim;
1476 	int retval = 0;
1477 
1478 	if (resource >= RLIM_NLIMITS)
1479 		return -EINVAL;
1480 	if (new_rlim) {
1481 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1482 			return -EINVAL;
1483 		if (resource == RLIMIT_NOFILE &&
1484 				new_rlim->rlim_max > sysctl_nr_open)
1485 			return -EPERM;
1486 	}
1487 
1488 	/* protect tsk->signal and tsk->sighand from disappearing */
1489 	read_lock(&tasklist_lock);
1490 	if (!tsk->sighand) {
1491 		retval = -ESRCH;
1492 		goto out;
1493 	}
1494 
1495 	rlim = tsk->signal->rlim + resource;
1496 	task_lock(tsk->group_leader);
1497 	if (new_rlim) {
1498 		/* Keep the capable check against init_user_ns until
1499 		   cgroups can contain all limits */
1500 		if (new_rlim->rlim_max > rlim->rlim_max &&
1501 				!capable(CAP_SYS_RESOURCE))
1502 			retval = -EPERM;
1503 		if (!retval)
1504 			retval = security_task_setrlimit(tsk, resource, new_rlim);
1505 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1506 			/*
1507 			 * The caller is asking for an immediate RLIMIT_CPU
1508 			 * expiry.  But we use the zero value to mean "it was
1509 			 * never set".  So let's cheat and make it one second
1510 			 * instead
1511 			 */
1512 			new_rlim->rlim_cur = 1;
1513 		}
1514 	}
1515 	if (!retval) {
1516 		if (old_rlim)
1517 			*old_rlim = *rlim;
1518 		if (new_rlim)
1519 			*rlim = *new_rlim;
1520 	}
1521 	task_unlock(tsk->group_leader);
1522 
1523 	/*
1524 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1525 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1526 	 * very long-standing error, and fixing it now risks breakage of
1527 	 * applications, so we live with it
1528 	 */
1529 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1530 	     new_rlim->rlim_cur != RLIM_INFINITY &&
1531 	     IS_ENABLED(CONFIG_POSIX_TIMERS))
1532 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1533 out:
1534 	read_unlock(&tasklist_lock);
1535 	return retval;
1536 }
1537 
1538 /* rcu lock must be held */
1539 static int check_prlimit_permission(struct task_struct *task,
1540 				    unsigned int flags)
1541 {
1542 	const struct cred *cred = current_cred(), *tcred;
1543 	bool id_match;
1544 
1545 	if (current == task)
1546 		return 0;
1547 
1548 	tcred = __task_cred(task);
1549 	id_match = (uid_eq(cred->uid, tcred->euid) &&
1550 		    uid_eq(cred->uid, tcred->suid) &&
1551 		    uid_eq(cred->uid, tcred->uid)  &&
1552 		    gid_eq(cred->gid, tcred->egid) &&
1553 		    gid_eq(cred->gid, tcred->sgid) &&
1554 		    gid_eq(cred->gid, tcred->gid));
1555 	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1556 		return -EPERM;
1557 
1558 	return security_task_prlimit(cred, tcred, flags);
1559 }
1560 
1561 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1562 		const struct rlimit64 __user *, new_rlim,
1563 		struct rlimit64 __user *, old_rlim)
1564 {
1565 	struct rlimit64 old64, new64;
1566 	struct rlimit old, new;
1567 	struct task_struct *tsk;
1568 	unsigned int checkflags = 0;
1569 	int ret;
1570 
1571 	if (old_rlim)
1572 		checkflags |= LSM_PRLIMIT_READ;
1573 
1574 	if (new_rlim) {
1575 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1576 			return -EFAULT;
1577 		rlim64_to_rlim(&new64, &new);
1578 		checkflags |= LSM_PRLIMIT_WRITE;
1579 	}
1580 
1581 	rcu_read_lock();
1582 	tsk = pid ? find_task_by_vpid(pid) : current;
1583 	if (!tsk) {
1584 		rcu_read_unlock();
1585 		return -ESRCH;
1586 	}
1587 	ret = check_prlimit_permission(tsk, checkflags);
1588 	if (ret) {
1589 		rcu_read_unlock();
1590 		return ret;
1591 	}
1592 	get_task_struct(tsk);
1593 	rcu_read_unlock();
1594 
1595 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1596 			old_rlim ? &old : NULL);
1597 
1598 	if (!ret && old_rlim) {
1599 		rlim_to_rlim64(&old, &old64);
1600 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1601 			ret = -EFAULT;
1602 	}
1603 
1604 	put_task_struct(tsk);
1605 	return ret;
1606 }
1607 
1608 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1609 {
1610 	struct rlimit new_rlim;
1611 
1612 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1613 		return -EFAULT;
1614 	return do_prlimit(current, resource, &new_rlim, NULL);
1615 }
1616 
1617 /*
1618  * It would make sense to put struct rusage in the task_struct,
1619  * except that would make the task_struct be *really big*.  After
1620  * task_struct gets moved into malloc'ed memory, it would
1621  * make sense to do this.  It will make moving the rest of the information
1622  * a lot simpler!  (Which we're not doing right now because we're not
1623  * measuring them yet).
1624  *
1625  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1626  * races with threads incrementing their own counters.  But since word
1627  * reads are atomic, we either get new values or old values and we don't
1628  * care which for the sums.  We always take the siglock to protect reading
1629  * the c* fields from p->signal from races with exit.c updating those
1630  * fields when reaping, so a sample either gets all the additions of a
1631  * given child after it's reaped, or none so this sample is before reaping.
1632  *
1633  * Locking:
1634  * We need to take the siglock for CHILDEREN, SELF and BOTH
1635  * for  the cases current multithreaded, non-current single threaded
1636  * non-current multithreaded.  Thread traversal is now safe with
1637  * the siglock held.
1638  * Strictly speaking, we donot need to take the siglock if we are current and
1639  * single threaded,  as no one else can take our signal_struct away, no one
1640  * else can  reap the  children to update signal->c* counters, and no one else
1641  * can race with the signal-> fields. If we do not take any lock, the
1642  * signal-> fields could be read out of order while another thread was just
1643  * exiting. So we should  place a read memory barrier when we avoid the lock.
1644  * On the writer side,  write memory barrier is implied in  __exit_signal
1645  * as __exit_signal releases  the siglock spinlock after updating the signal->
1646  * fields. But we don't do this yet to keep things simple.
1647  *
1648  */
1649 
1650 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1651 {
1652 	r->ru_nvcsw += t->nvcsw;
1653 	r->ru_nivcsw += t->nivcsw;
1654 	r->ru_minflt += t->min_flt;
1655 	r->ru_majflt += t->maj_flt;
1656 	r->ru_inblock += task_io_get_inblock(t);
1657 	r->ru_oublock += task_io_get_oublock(t);
1658 }
1659 
1660 void getrusage(struct task_struct *p, int who, struct rusage *r)
1661 {
1662 	struct task_struct *t;
1663 	unsigned long flags;
1664 	u64 tgutime, tgstime, utime, stime;
1665 	unsigned long maxrss = 0;
1666 
1667 	memset((char *)r, 0, sizeof (*r));
1668 	utime = stime = 0;
1669 
1670 	if (who == RUSAGE_THREAD) {
1671 		task_cputime_adjusted(current, &utime, &stime);
1672 		accumulate_thread_rusage(p, r);
1673 		maxrss = p->signal->maxrss;
1674 		goto out;
1675 	}
1676 
1677 	if (!lock_task_sighand(p, &flags))
1678 		return;
1679 
1680 	switch (who) {
1681 	case RUSAGE_BOTH:
1682 	case RUSAGE_CHILDREN:
1683 		utime = p->signal->cutime;
1684 		stime = p->signal->cstime;
1685 		r->ru_nvcsw = p->signal->cnvcsw;
1686 		r->ru_nivcsw = p->signal->cnivcsw;
1687 		r->ru_minflt = p->signal->cmin_flt;
1688 		r->ru_majflt = p->signal->cmaj_flt;
1689 		r->ru_inblock = p->signal->cinblock;
1690 		r->ru_oublock = p->signal->coublock;
1691 		maxrss = p->signal->cmaxrss;
1692 
1693 		if (who == RUSAGE_CHILDREN)
1694 			break;
1695 
1696 	case RUSAGE_SELF:
1697 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1698 		utime += tgutime;
1699 		stime += tgstime;
1700 		r->ru_nvcsw += p->signal->nvcsw;
1701 		r->ru_nivcsw += p->signal->nivcsw;
1702 		r->ru_minflt += p->signal->min_flt;
1703 		r->ru_majflt += p->signal->maj_flt;
1704 		r->ru_inblock += p->signal->inblock;
1705 		r->ru_oublock += p->signal->oublock;
1706 		if (maxrss < p->signal->maxrss)
1707 			maxrss = p->signal->maxrss;
1708 		t = p;
1709 		do {
1710 			accumulate_thread_rusage(t, r);
1711 		} while_each_thread(p, t);
1712 		break;
1713 
1714 	default:
1715 		BUG();
1716 	}
1717 	unlock_task_sighand(p, &flags);
1718 
1719 out:
1720 	r->ru_utime = ns_to_timeval(utime);
1721 	r->ru_stime = ns_to_timeval(stime);
1722 
1723 	if (who != RUSAGE_CHILDREN) {
1724 		struct mm_struct *mm = get_task_mm(p);
1725 
1726 		if (mm) {
1727 			setmax_mm_hiwater_rss(&maxrss, mm);
1728 			mmput(mm);
1729 		}
1730 	}
1731 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1732 }
1733 
1734 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1735 {
1736 	struct rusage r;
1737 
1738 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1739 	    who != RUSAGE_THREAD)
1740 		return -EINVAL;
1741 
1742 	getrusage(current, who, &r);
1743 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1744 }
1745 
1746 #ifdef CONFIG_COMPAT
1747 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1748 {
1749 	struct rusage r;
1750 
1751 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1752 	    who != RUSAGE_THREAD)
1753 		return -EINVAL;
1754 
1755 	getrusage(current, who, &r);
1756 	return put_compat_rusage(&r, ru);
1757 }
1758 #endif
1759 
1760 SYSCALL_DEFINE1(umask, int, mask)
1761 {
1762 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1763 	return mask;
1764 }
1765 
1766 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1767 {
1768 	struct fd exe;
1769 	struct file *old_exe, *exe_file;
1770 	struct inode *inode;
1771 	int err;
1772 
1773 	exe = fdget(fd);
1774 	if (!exe.file)
1775 		return -EBADF;
1776 
1777 	inode = file_inode(exe.file);
1778 
1779 	/*
1780 	 * Because the original mm->exe_file points to executable file, make
1781 	 * sure that this one is executable as well, to avoid breaking an
1782 	 * overall picture.
1783 	 */
1784 	err = -EACCES;
1785 	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1786 		goto exit;
1787 
1788 	err = inode_permission(inode, MAY_EXEC);
1789 	if (err)
1790 		goto exit;
1791 
1792 	/*
1793 	 * Forbid mm->exe_file change if old file still mapped.
1794 	 */
1795 	exe_file = get_mm_exe_file(mm);
1796 	err = -EBUSY;
1797 	if (exe_file) {
1798 		struct vm_area_struct *vma;
1799 
1800 		down_read(&mm->mmap_sem);
1801 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
1802 			if (!vma->vm_file)
1803 				continue;
1804 			if (path_equal(&vma->vm_file->f_path,
1805 				       &exe_file->f_path))
1806 				goto exit_err;
1807 		}
1808 
1809 		up_read(&mm->mmap_sem);
1810 		fput(exe_file);
1811 	}
1812 
1813 	err = 0;
1814 	/* set the new file, lockless */
1815 	get_file(exe.file);
1816 	old_exe = xchg(&mm->exe_file, exe.file);
1817 	if (old_exe)
1818 		fput(old_exe);
1819 exit:
1820 	fdput(exe);
1821 	return err;
1822 exit_err:
1823 	up_read(&mm->mmap_sem);
1824 	fput(exe_file);
1825 	goto exit;
1826 }
1827 
1828 /*
1829  * WARNING: we don't require any capability here so be very careful
1830  * in what is allowed for modification from userspace.
1831  */
1832 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1833 {
1834 	unsigned long mmap_max_addr = TASK_SIZE;
1835 	struct mm_struct *mm = current->mm;
1836 	int error = -EINVAL, i;
1837 
1838 	static const unsigned char offsets[] = {
1839 		offsetof(struct prctl_mm_map, start_code),
1840 		offsetof(struct prctl_mm_map, end_code),
1841 		offsetof(struct prctl_mm_map, start_data),
1842 		offsetof(struct prctl_mm_map, end_data),
1843 		offsetof(struct prctl_mm_map, start_brk),
1844 		offsetof(struct prctl_mm_map, brk),
1845 		offsetof(struct prctl_mm_map, start_stack),
1846 		offsetof(struct prctl_mm_map, arg_start),
1847 		offsetof(struct prctl_mm_map, arg_end),
1848 		offsetof(struct prctl_mm_map, env_start),
1849 		offsetof(struct prctl_mm_map, env_end),
1850 	};
1851 
1852 	/*
1853 	 * Make sure the members are not somewhere outside
1854 	 * of allowed address space.
1855 	 */
1856 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1857 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1858 
1859 		if ((unsigned long)val >= mmap_max_addr ||
1860 		    (unsigned long)val < mmap_min_addr)
1861 			goto out;
1862 	}
1863 
1864 	/*
1865 	 * Make sure the pairs are ordered.
1866 	 */
1867 #define __prctl_check_order(__m1, __op, __m2)				\
1868 	((unsigned long)prctl_map->__m1 __op				\
1869 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1870 	error  = __prctl_check_order(start_code, <, end_code);
1871 	error |= __prctl_check_order(start_data, <, end_data);
1872 	error |= __prctl_check_order(start_brk, <=, brk);
1873 	error |= __prctl_check_order(arg_start, <=, arg_end);
1874 	error |= __prctl_check_order(env_start, <=, env_end);
1875 	if (error)
1876 		goto out;
1877 #undef __prctl_check_order
1878 
1879 	error = -EINVAL;
1880 
1881 	/*
1882 	 * @brk should be after @end_data in traditional maps.
1883 	 */
1884 	if (prctl_map->start_brk <= prctl_map->end_data ||
1885 	    prctl_map->brk <= prctl_map->end_data)
1886 		goto out;
1887 
1888 	/*
1889 	 * Neither we should allow to override limits if they set.
1890 	 */
1891 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1892 			      prctl_map->start_brk, prctl_map->end_data,
1893 			      prctl_map->start_data))
1894 			goto out;
1895 
1896 	/*
1897 	 * Someone is trying to cheat the auxv vector.
1898 	 */
1899 	if (prctl_map->auxv_size) {
1900 		if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1901 			goto out;
1902 	}
1903 
1904 	/*
1905 	 * Finally, make sure the caller has the rights to
1906 	 * change /proc/pid/exe link: only local sys admin should
1907 	 * be allowed to.
1908 	 */
1909 	if (prctl_map->exe_fd != (u32)-1) {
1910 		if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1911 			goto out;
1912 	}
1913 
1914 	error = 0;
1915 out:
1916 	return error;
1917 }
1918 
1919 #ifdef CONFIG_CHECKPOINT_RESTORE
1920 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1921 {
1922 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1923 	unsigned long user_auxv[AT_VECTOR_SIZE];
1924 	struct mm_struct *mm = current->mm;
1925 	int error;
1926 
1927 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1928 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1929 
1930 	if (opt == PR_SET_MM_MAP_SIZE)
1931 		return put_user((unsigned int)sizeof(prctl_map),
1932 				(unsigned int __user *)addr);
1933 
1934 	if (data_size != sizeof(prctl_map))
1935 		return -EINVAL;
1936 
1937 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1938 		return -EFAULT;
1939 
1940 	error = validate_prctl_map(&prctl_map);
1941 	if (error)
1942 		return error;
1943 
1944 	if (prctl_map.auxv_size) {
1945 		memset(user_auxv, 0, sizeof(user_auxv));
1946 		if (copy_from_user(user_auxv,
1947 				   (const void __user *)prctl_map.auxv,
1948 				   prctl_map.auxv_size))
1949 			return -EFAULT;
1950 
1951 		/* Last entry must be AT_NULL as specification requires */
1952 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1953 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1954 	}
1955 
1956 	if (prctl_map.exe_fd != (u32)-1) {
1957 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1958 		if (error)
1959 			return error;
1960 	}
1961 
1962 	down_write(&mm->mmap_sem);
1963 
1964 	/*
1965 	 * We don't validate if these members are pointing to
1966 	 * real present VMAs because application may have correspond
1967 	 * VMAs already unmapped and kernel uses these members for statistics
1968 	 * output in procfs mostly, except
1969 	 *
1970 	 *  - @start_brk/@brk which are used in do_brk but kernel lookups
1971 	 *    for VMAs when updating these memvers so anything wrong written
1972 	 *    here cause kernel to swear at userspace program but won't lead
1973 	 *    to any problem in kernel itself
1974 	 */
1975 
1976 	mm->start_code	= prctl_map.start_code;
1977 	mm->end_code	= prctl_map.end_code;
1978 	mm->start_data	= prctl_map.start_data;
1979 	mm->end_data	= prctl_map.end_data;
1980 	mm->start_brk	= prctl_map.start_brk;
1981 	mm->brk		= prctl_map.brk;
1982 	mm->start_stack	= prctl_map.start_stack;
1983 	mm->arg_start	= prctl_map.arg_start;
1984 	mm->arg_end	= prctl_map.arg_end;
1985 	mm->env_start	= prctl_map.env_start;
1986 	mm->env_end	= prctl_map.env_end;
1987 
1988 	/*
1989 	 * Note this update of @saved_auxv is lockless thus
1990 	 * if someone reads this member in procfs while we're
1991 	 * updating -- it may get partly updated results. It's
1992 	 * known and acceptable trade off: we leave it as is to
1993 	 * not introduce additional locks here making the kernel
1994 	 * more complex.
1995 	 */
1996 	if (prctl_map.auxv_size)
1997 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1998 
1999 	up_write(&mm->mmap_sem);
2000 	return 0;
2001 }
2002 #endif /* CONFIG_CHECKPOINT_RESTORE */
2003 
2004 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2005 			  unsigned long len)
2006 {
2007 	/*
2008 	 * This doesn't move the auxiliary vector itself since it's pinned to
2009 	 * mm_struct, but it permits filling the vector with new values.  It's
2010 	 * up to the caller to provide sane values here, otherwise userspace
2011 	 * tools which use this vector might be unhappy.
2012 	 */
2013 	unsigned long user_auxv[AT_VECTOR_SIZE];
2014 
2015 	if (len > sizeof(user_auxv))
2016 		return -EINVAL;
2017 
2018 	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2019 		return -EFAULT;
2020 
2021 	/* Make sure the last entry is always AT_NULL */
2022 	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2023 	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2024 
2025 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2026 
2027 	task_lock(current);
2028 	memcpy(mm->saved_auxv, user_auxv, len);
2029 	task_unlock(current);
2030 
2031 	return 0;
2032 }
2033 
2034 static int prctl_set_mm(int opt, unsigned long addr,
2035 			unsigned long arg4, unsigned long arg5)
2036 {
2037 	struct mm_struct *mm = current->mm;
2038 	struct prctl_mm_map prctl_map;
2039 	struct vm_area_struct *vma;
2040 	int error;
2041 
2042 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2043 			      opt != PR_SET_MM_MAP &&
2044 			      opt != PR_SET_MM_MAP_SIZE)))
2045 		return -EINVAL;
2046 
2047 #ifdef CONFIG_CHECKPOINT_RESTORE
2048 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2049 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2050 #endif
2051 
2052 	if (!capable(CAP_SYS_RESOURCE))
2053 		return -EPERM;
2054 
2055 	if (opt == PR_SET_MM_EXE_FILE)
2056 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2057 
2058 	if (opt == PR_SET_MM_AUXV)
2059 		return prctl_set_auxv(mm, addr, arg4);
2060 
2061 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2062 		return -EINVAL;
2063 
2064 	error = -EINVAL;
2065 
2066 	down_write(&mm->mmap_sem);
2067 	vma = find_vma(mm, addr);
2068 
2069 	prctl_map.start_code	= mm->start_code;
2070 	prctl_map.end_code	= mm->end_code;
2071 	prctl_map.start_data	= mm->start_data;
2072 	prctl_map.end_data	= mm->end_data;
2073 	prctl_map.start_brk	= mm->start_brk;
2074 	prctl_map.brk		= mm->brk;
2075 	prctl_map.start_stack	= mm->start_stack;
2076 	prctl_map.arg_start	= mm->arg_start;
2077 	prctl_map.arg_end	= mm->arg_end;
2078 	prctl_map.env_start	= mm->env_start;
2079 	prctl_map.env_end	= mm->env_end;
2080 	prctl_map.auxv		= NULL;
2081 	prctl_map.auxv_size	= 0;
2082 	prctl_map.exe_fd	= -1;
2083 
2084 	switch (opt) {
2085 	case PR_SET_MM_START_CODE:
2086 		prctl_map.start_code = addr;
2087 		break;
2088 	case PR_SET_MM_END_CODE:
2089 		prctl_map.end_code = addr;
2090 		break;
2091 	case PR_SET_MM_START_DATA:
2092 		prctl_map.start_data = addr;
2093 		break;
2094 	case PR_SET_MM_END_DATA:
2095 		prctl_map.end_data = addr;
2096 		break;
2097 	case PR_SET_MM_START_STACK:
2098 		prctl_map.start_stack = addr;
2099 		break;
2100 	case PR_SET_MM_START_BRK:
2101 		prctl_map.start_brk = addr;
2102 		break;
2103 	case PR_SET_MM_BRK:
2104 		prctl_map.brk = addr;
2105 		break;
2106 	case PR_SET_MM_ARG_START:
2107 		prctl_map.arg_start = addr;
2108 		break;
2109 	case PR_SET_MM_ARG_END:
2110 		prctl_map.arg_end = addr;
2111 		break;
2112 	case PR_SET_MM_ENV_START:
2113 		prctl_map.env_start = addr;
2114 		break;
2115 	case PR_SET_MM_ENV_END:
2116 		prctl_map.env_end = addr;
2117 		break;
2118 	default:
2119 		goto out;
2120 	}
2121 
2122 	error = validate_prctl_map(&prctl_map);
2123 	if (error)
2124 		goto out;
2125 
2126 	switch (opt) {
2127 	/*
2128 	 * If command line arguments and environment
2129 	 * are placed somewhere else on stack, we can
2130 	 * set them up here, ARG_START/END to setup
2131 	 * command line argumets and ENV_START/END
2132 	 * for environment.
2133 	 */
2134 	case PR_SET_MM_START_STACK:
2135 	case PR_SET_MM_ARG_START:
2136 	case PR_SET_MM_ARG_END:
2137 	case PR_SET_MM_ENV_START:
2138 	case PR_SET_MM_ENV_END:
2139 		if (!vma) {
2140 			error = -EFAULT;
2141 			goto out;
2142 		}
2143 	}
2144 
2145 	mm->start_code	= prctl_map.start_code;
2146 	mm->end_code	= prctl_map.end_code;
2147 	mm->start_data	= prctl_map.start_data;
2148 	mm->end_data	= prctl_map.end_data;
2149 	mm->start_brk	= prctl_map.start_brk;
2150 	mm->brk		= prctl_map.brk;
2151 	mm->start_stack	= prctl_map.start_stack;
2152 	mm->arg_start	= prctl_map.arg_start;
2153 	mm->arg_end	= prctl_map.arg_end;
2154 	mm->env_start	= prctl_map.env_start;
2155 	mm->env_end	= prctl_map.env_end;
2156 
2157 	error = 0;
2158 out:
2159 	up_write(&mm->mmap_sem);
2160 	return error;
2161 }
2162 
2163 #ifdef CONFIG_CHECKPOINT_RESTORE
2164 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2165 {
2166 	return put_user(me->clear_child_tid, tid_addr);
2167 }
2168 #else
2169 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2170 {
2171 	return -EINVAL;
2172 }
2173 #endif
2174 
2175 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2176 {
2177 	/*
2178 	 * If task has has_child_subreaper - all its decendants
2179 	 * already have these flag too and new decendants will
2180 	 * inherit it on fork, skip them.
2181 	 *
2182 	 * If we've found child_reaper - skip descendants in
2183 	 * it's subtree as they will never get out pidns.
2184 	 */
2185 	if (p->signal->has_child_subreaper ||
2186 	    is_child_reaper(task_pid(p)))
2187 		return 0;
2188 
2189 	p->signal->has_child_subreaper = 1;
2190 	return 1;
2191 }
2192 
2193 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2194 		unsigned long, arg4, unsigned long, arg5)
2195 {
2196 	struct task_struct *me = current;
2197 	unsigned char comm[sizeof(me->comm)];
2198 	long error;
2199 
2200 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2201 	if (error != -ENOSYS)
2202 		return error;
2203 
2204 	error = 0;
2205 	switch (option) {
2206 	case PR_SET_PDEATHSIG:
2207 		if (!valid_signal(arg2)) {
2208 			error = -EINVAL;
2209 			break;
2210 		}
2211 		me->pdeath_signal = arg2;
2212 		break;
2213 	case PR_GET_PDEATHSIG:
2214 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2215 		break;
2216 	case PR_GET_DUMPABLE:
2217 		error = get_dumpable(me->mm);
2218 		break;
2219 	case PR_SET_DUMPABLE:
2220 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2221 			error = -EINVAL;
2222 			break;
2223 		}
2224 		set_dumpable(me->mm, arg2);
2225 		break;
2226 
2227 	case PR_SET_UNALIGN:
2228 		error = SET_UNALIGN_CTL(me, arg2);
2229 		break;
2230 	case PR_GET_UNALIGN:
2231 		error = GET_UNALIGN_CTL(me, arg2);
2232 		break;
2233 	case PR_SET_FPEMU:
2234 		error = SET_FPEMU_CTL(me, arg2);
2235 		break;
2236 	case PR_GET_FPEMU:
2237 		error = GET_FPEMU_CTL(me, arg2);
2238 		break;
2239 	case PR_SET_FPEXC:
2240 		error = SET_FPEXC_CTL(me, arg2);
2241 		break;
2242 	case PR_GET_FPEXC:
2243 		error = GET_FPEXC_CTL(me, arg2);
2244 		break;
2245 	case PR_GET_TIMING:
2246 		error = PR_TIMING_STATISTICAL;
2247 		break;
2248 	case PR_SET_TIMING:
2249 		if (arg2 != PR_TIMING_STATISTICAL)
2250 			error = -EINVAL;
2251 		break;
2252 	case PR_SET_NAME:
2253 		comm[sizeof(me->comm) - 1] = 0;
2254 		if (strncpy_from_user(comm, (char __user *)arg2,
2255 				      sizeof(me->comm) - 1) < 0)
2256 			return -EFAULT;
2257 		set_task_comm(me, comm);
2258 		proc_comm_connector(me);
2259 		break;
2260 	case PR_GET_NAME:
2261 		get_task_comm(comm, me);
2262 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2263 			return -EFAULT;
2264 		break;
2265 	case PR_GET_ENDIAN:
2266 		error = GET_ENDIAN(me, arg2);
2267 		break;
2268 	case PR_SET_ENDIAN:
2269 		error = SET_ENDIAN(me, arg2);
2270 		break;
2271 	case PR_GET_SECCOMP:
2272 		error = prctl_get_seccomp();
2273 		break;
2274 	case PR_SET_SECCOMP:
2275 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2276 		break;
2277 	case PR_GET_TSC:
2278 		error = GET_TSC_CTL(arg2);
2279 		break;
2280 	case PR_SET_TSC:
2281 		error = SET_TSC_CTL(arg2);
2282 		break;
2283 	case PR_TASK_PERF_EVENTS_DISABLE:
2284 		error = perf_event_task_disable();
2285 		break;
2286 	case PR_TASK_PERF_EVENTS_ENABLE:
2287 		error = perf_event_task_enable();
2288 		break;
2289 	case PR_GET_TIMERSLACK:
2290 		if (current->timer_slack_ns > ULONG_MAX)
2291 			error = ULONG_MAX;
2292 		else
2293 			error = current->timer_slack_ns;
2294 		break;
2295 	case PR_SET_TIMERSLACK:
2296 		if (arg2 <= 0)
2297 			current->timer_slack_ns =
2298 					current->default_timer_slack_ns;
2299 		else
2300 			current->timer_slack_ns = arg2;
2301 		break;
2302 	case PR_MCE_KILL:
2303 		if (arg4 | arg5)
2304 			return -EINVAL;
2305 		switch (arg2) {
2306 		case PR_MCE_KILL_CLEAR:
2307 			if (arg3 != 0)
2308 				return -EINVAL;
2309 			current->flags &= ~PF_MCE_PROCESS;
2310 			break;
2311 		case PR_MCE_KILL_SET:
2312 			current->flags |= PF_MCE_PROCESS;
2313 			if (arg3 == PR_MCE_KILL_EARLY)
2314 				current->flags |= PF_MCE_EARLY;
2315 			else if (arg3 == PR_MCE_KILL_LATE)
2316 				current->flags &= ~PF_MCE_EARLY;
2317 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2318 				current->flags &=
2319 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2320 			else
2321 				return -EINVAL;
2322 			break;
2323 		default:
2324 			return -EINVAL;
2325 		}
2326 		break;
2327 	case PR_MCE_KILL_GET:
2328 		if (arg2 | arg3 | arg4 | arg5)
2329 			return -EINVAL;
2330 		if (current->flags & PF_MCE_PROCESS)
2331 			error = (current->flags & PF_MCE_EARLY) ?
2332 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2333 		else
2334 			error = PR_MCE_KILL_DEFAULT;
2335 		break;
2336 	case PR_SET_MM:
2337 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2338 		break;
2339 	case PR_GET_TID_ADDRESS:
2340 		error = prctl_get_tid_address(me, (int __user **)arg2);
2341 		break;
2342 	case PR_SET_CHILD_SUBREAPER:
2343 		me->signal->is_child_subreaper = !!arg2;
2344 		if (!arg2)
2345 			break;
2346 
2347 		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2348 		break;
2349 	case PR_GET_CHILD_SUBREAPER:
2350 		error = put_user(me->signal->is_child_subreaper,
2351 				 (int __user *)arg2);
2352 		break;
2353 	case PR_SET_NO_NEW_PRIVS:
2354 		if (arg2 != 1 || arg3 || arg4 || arg5)
2355 			return -EINVAL;
2356 
2357 		task_set_no_new_privs(current);
2358 		break;
2359 	case PR_GET_NO_NEW_PRIVS:
2360 		if (arg2 || arg3 || arg4 || arg5)
2361 			return -EINVAL;
2362 		return task_no_new_privs(current) ? 1 : 0;
2363 	case PR_GET_THP_DISABLE:
2364 		if (arg2 || arg3 || arg4 || arg5)
2365 			return -EINVAL;
2366 		error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2367 		break;
2368 	case PR_SET_THP_DISABLE:
2369 		if (arg3 || arg4 || arg5)
2370 			return -EINVAL;
2371 		if (down_write_killable(&me->mm->mmap_sem))
2372 			return -EINTR;
2373 		if (arg2)
2374 			set_bit(MMF_DISABLE_THP, &me->mm->flags);
2375 		else
2376 			clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2377 		up_write(&me->mm->mmap_sem);
2378 		break;
2379 	case PR_MPX_ENABLE_MANAGEMENT:
2380 		if (arg2 || arg3 || arg4 || arg5)
2381 			return -EINVAL;
2382 		error = MPX_ENABLE_MANAGEMENT();
2383 		break;
2384 	case PR_MPX_DISABLE_MANAGEMENT:
2385 		if (arg2 || arg3 || arg4 || arg5)
2386 			return -EINVAL;
2387 		error = MPX_DISABLE_MANAGEMENT();
2388 		break;
2389 	case PR_SET_FP_MODE:
2390 		error = SET_FP_MODE(me, arg2);
2391 		break;
2392 	case PR_GET_FP_MODE:
2393 		error = GET_FP_MODE(me);
2394 		break;
2395 	case PR_SVE_SET_VL:
2396 		error = SVE_SET_VL(arg2);
2397 		break;
2398 	case PR_SVE_GET_VL:
2399 		error = SVE_GET_VL();
2400 		break;
2401 	default:
2402 		error = -EINVAL;
2403 		break;
2404 	}
2405 	return error;
2406 }
2407 
2408 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2409 		struct getcpu_cache __user *, unused)
2410 {
2411 	int err = 0;
2412 	int cpu = raw_smp_processor_id();
2413 
2414 	if (cpup)
2415 		err |= put_user(cpu, cpup);
2416 	if (nodep)
2417 		err |= put_user(cpu_to_node(cpu), nodep);
2418 	return err ? -EFAULT : 0;
2419 }
2420 
2421 /**
2422  * do_sysinfo - fill in sysinfo struct
2423  * @info: pointer to buffer to fill
2424  */
2425 static int do_sysinfo(struct sysinfo *info)
2426 {
2427 	unsigned long mem_total, sav_total;
2428 	unsigned int mem_unit, bitcount;
2429 	struct timespec tp;
2430 
2431 	memset(info, 0, sizeof(struct sysinfo));
2432 
2433 	get_monotonic_boottime(&tp);
2434 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2435 
2436 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2437 
2438 	info->procs = nr_threads;
2439 
2440 	si_meminfo(info);
2441 	si_swapinfo(info);
2442 
2443 	/*
2444 	 * If the sum of all the available memory (i.e. ram + swap)
2445 	 * is less than can be stored in a 32 bit unsigned long then
2446 	 * we can be binary compatible with 2.2.x kernels.  If not,
2447 	 * well, in that case 2.2.x was broken anyways...
2448 	 *
2449 	 *  -Erik Andersen <andersee@debian.org>
2450 	 */
2451 
2452 	mem_total = info->totalram + info->totalswap;
2453 	if (mem_total < info->totalram || mem_total < info->totalswap)
2454 		goto out;
2455 	bitcount = 0;
2456 	mem_unit = info->mem_unit;
2457 	while (mem_unit > 1) {
2458 		bitcount++;
2459 		mem_unit >>= 1;
2460 		sav_total = mem_total;
2461 		mem_total <<= 1;
2462 		if (mem_total < sav_total)
2463 			goto out;
2464 	}
2465 
2466 	/*
2467 	 * If mem_total did not overflow, multiply all memory values by
2468 	 * info->mem_unit and set it to 1.  This leaves things compatible
2469 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2470 	 * kernels...
2471 	 */
2472 
2473 	info->mem_unit = 1;
2474 	info->totalram <<= bitcount;
2475 	info->freeram <<= bitcount;
2476 	info->sharedram <<= bitcount;
2477 	info->bufferram <<= bitcount;
2478 	info->totalswap <<= bitcount;
2479 	info->freeswap <<= bitcount;
2480 	info->totalhigh <<= bitcount;
2481 	info->freehigh <<= bitcount;
2482 
2483 out:
2484 	return 0;
2485 }
2486 
2487 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2488 {
2489 	struct sysinfo val;
2490 
2491 	do_sysinfo(&val);
2492 
2493 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2494 		return -EFAULT;
2495 
2496 	return 0;
2497 }
2498 
2499 #ifdef CONFIG_COMPAT
2500 struct compat_sysinfo {
2501 	s32 uptime;
2502 	u32 loads[3];
2503 	u32 totalram;
2504 	u32 freeram;
2505 	u32 sharedram;
2506 	u32 bufferram;
2507 	u32 totalswap;
2508 	u32 freeswap;
2509 	u16 procs;
2510 	u16 pad;
2511 	u32 totalhigh;
2512 	u32 freehigh;
2513 	u32 mem_unit;
2514 	char _f[20-2*sizeof(u32)-sizeof(int)];
2515 };
2516 
2517 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2518 {
2519 	struct sysinfo s;
2520 
2521 	do_sysinfo(&s);
2522 
2523 	/* Check to see if any memory value is too large for 32-bit and scale
2524 	 *  down if needed
2525 	 */
2526 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2527 		int bitcount = 0;
2528 
2529 		while (s.mem_unit < PAGE_SIZE) {
2530 			s.mem_unit <<= 1;
2531 			bitcount++;
2532 		}
2533 
2534 		s.totalram >>= bitcount;
2535 		s.freeram >>= bitcount;
2536 		s.sharedram >>= bitcount;
2537 		s.bufferram >>= bitcount;
2538 		s.totalswap >>= bitcount;
2539 		s.freeswap >>= bitcount;
2540 		s.totalhigh >>= bitcount;
2541 		s.freehigh >>= bitcount;
2542 	}
2543 
2544 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2545 	    __put_user(s.uptime, &info->uptime) ||
2546 	    __put_user(s.loads[0], &info->loads[0]) ||
2547 	    __put_user(s.loads[1], &info->loads[1]) ||
2548 	    __put_user(s.loads[2], &info->loads[2]) ||
2549 	    __put_user(s.totalram, &info->totalram) ||
2550 	    __put_user(s.freeram, &info->freeram) ||
2551 	    __put_user(s.sharedram, &info->sharedram) ||
2552 	    __put_user(s.bufferram, &info->bufferram) ||
2553 	    __put_user(s.totalswap, &info->totalswap) ||
2554 	    __put_user(s.freeswap, &info->freeswap) ||
2555 	    __put_user(s.procs, &info->procs) ||
2556 	    __put_user(s.totalhigh, &info->totalhigh) ||
2557 	    __put_user(s.freehigh, &info->freehigh) ||
2558 	    __put_user(s.mem_unit, &info->mem_unit))
2559 		return -EFAULT;
2560 
2561 	return 0;
2562 }
2563 #endif /* CONFIG_COMPAT */
2564