xref: /openbmc/linux/kernel/sys.c (revision 6f52b16c5b29b89d92c0e7236f4655dc8491ad70)
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 
115 /*
116  * this is where the system-wide overflow UID and GID are defined, for
117  * architectures that now have 32-bit UID/GID but didn't in the past
118  */
119 
120 int overflowuid = DEFAULT_OVERFLOWUID;
121 int overflowgid = DEFAULT_OVERFLOWGID;
122 
123 EXPORT_SYMBOL(overflowuid);
124 EXPORT_SYMBOL(overflowgid);
125 
126 /*
127  * the same as above, but for filesystems which can only store a 16-bit
128  * UID and GID. as such, this is needed on all architectures
129  */
130 
131 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
132 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
133 
134 EXPORT_SYMBOL(fs_overflowuid);
135 EXPORT_SYMBOL(fs_overflowgid);
136 
137 /*
138  * Returns true if current's euid is same as p's uid or euid,
139  * or has CAP_SYS_NICE to p's user_ns.
140  *
141  * Called with rcu_read_lock, creds are safe
142  */
143 static bool set_one_prio_perm(struct task_struct *p)
144 {
145 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
146 
147 	if (uid_eq(pcred->uid,  cred->euid) ||
148 	    uid_eq(pcred->euid, cred->euid))
149 		return true;
150 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
151 		return true;
152 	return false;
153 }
154 
155 /*
156  * set the priority of a task
157  * - the caller must hold the RCU read lock
158  */
159 static int set_one_prio(struct task_struct *p, int niceval, int error)
160 {
161 	int no_nice;
162 
163 	if (!set_one_prio_perm(p)) {
164 		error = -EPERM;
165 		goto out;
166 	}
167 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
168 		error = -EACCES;
169 		goto out;
170 	}
171 	no_nice = security_task_setnice(p, niceval);
172 	if (no_nice) {
173 		error = no_nice;
174 		goto out;
175 	}
176 	if (error == -ESRCH)
177 		error = 0;
178 	set_user_nice(p, niceval);
179 out:
180 	return error;
181 }
182 
183 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
184 {
185 	struct task_struct *g, *p;
186 	struct user_struct *user;
187 	const struct cred *cred = current_cred();
188 	int error = -EINVAL;
189 	struct pid *pgrp;
190 	kuid_t uid;
191 
192 	if (which > PRIO_USER || which < PRIO_PROCESS)
193 		goto out;
194 
195 	/* normalize: avoid signed division (rounding problems) */
196 	error = -ESRCH;
197 	if (niceval < MIN_NICE)
198 		niceval = MIN_NICE;
199 	if (niceval > MAX_NICE)
200 		niceval = MAX_NICE;
201 
202 	rcu_read_lock();
203 	read_lock(&tasklist_lock);
204 	switch (which) {
205 	case PRIO_PROCESS:
206 		if (who)
207 			p = find_task_by_vpid(who);
208 		else
209 			p = current;
210 		if (p)
211 			error = set_one_prio(p, niceval, error);
212 		break;
213 	case PRIO_PGRP:
214 		if (who)
215 			pgrp = find_vpid(who);
216 		else
217 			pgrp = task_pgrp(current);
218 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
219 			error = set_one_prio(p, niceval, error);
220 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
221 		break;
222 	case PRIO_USER:
223 		uid = make_kuid(cred->user_ns, who);
224 		user = cred->user;
225 		if (!who)
226 			uid = cred->uid;
227 		else if (!uid_eq(uid, cred->uid)) {
228 			user = find_user(uid);
229 			if (!user)
230 				goto out_unlock;	/* No processes for this user */
231 		}
232 		do_each_thread(g, p) {
233 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
234 				error = set_one_prio(p, niceval, error);
235 		} while_each_thread(g, p);
236 		if (!uid_eq(uid, cred->uid))
237 			free_uid(user);		/* For find_user() */
238 		break;
239 	}
240 out_unlock:
241 	read_unlock(&tasklist_lock);
242 	rcu_read_unlock();
243 out:
244 	return error;
245 }
246 
247 /*
248  * Ugh. To avoid negative return values, "getpriority()" will
249  * not return the normal nice-value, but a negated value that
250  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
251  * to stay compatible.
252  */
253 SYSCALL_DEFINE2(getpriority, int, which, int, who)
254 {
255 	struct task_struct *g, *p;
256 	struct user_struct *user;
257 	const struct cred *cred = current_cred();
258 	long niceval, retval = -ESRCH;
259 	struct pid *pgrp;
260 	kuid_t uid;
261 
262 	if (which > PRIO_USER || which < PRIO_PROCESS)
263 		return -EINVAL;
264 
265 	rcu_read_lock();
266 	read_lock(&tasklist_lock);
267 	switch (which) {
268 	case PRIO_PROCESS:
269 		if (who)
270 			p = find_task_by_vpid(who);
271 		else
272 			p = current;
273 		if (p) {
274 			niceval = nice_to_rlimit(task_nice(p));
275 			if (niceval > retval)
276 				retval = niceval;
277 		}
278 		break;
279 	case PRIO_PGRP:
280 		if (who)
281 			pgrp = find_vpid(who);
282 		else
283 			pgrp = task_pgrp(current);
284 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
285 			niceval = nice_to_rlimit(task_nice(p));
286 			if (niceval > retval)
287 				retval = niceval;
288 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
289 		break;
290 	case PRIO_USER:
291 		uid = make_kuid(cred->user_ns, who);
292 		user = cred->user;
293 		if (!who)
294 			uid = cred->uid;
295 		else if (!uid_eq(uid, cred->uid)) {
296 			user = find_user(uid);
297 			if (!user)
298 				goto out_unlock;	/* No processes for this user */
299 		}
300 		do_each_thread(g, p) {
301 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
302 				niceval = nice_to_rlimit(task_nice(p));
303 				if (niceval > retval)
304 					retval = niceval;
305 			}
306 		} while_each_thread(g, p);
307 		if (!uid_eq(uid, cred->uid))
308 			free_uid(user);		/* for find_user() */
309 		break;
310 	}
311 out_unlock:
312 	read_unlock(&tasklist_lock);
313 	rcu_read_unlock();
314 
315 	return retval;
316 }
317 
318 /*
319  * Unprivileged users may change the real gid to the effective gid
320  * or vice versa.  (BSD-style)
321  *
322  * If you set the real gid at all, or set the effective gid to a value not
323  * equal to the real gid, then the saved gid is set to the new effective gid.
324  *
325  * This makes it possible for a setgid program to completely drop its
326  * privileges, which is often a useful assertion to make when you are doing
327  * a security audit over a program.
328  *
329  * The general idea is that a program which uses just setregid() will be
330  * 100% compatible with BSD.  A program which uses just setgid() will be
331  * 100% compatible with POSIX with saved IDs.
332  *
333  * SMP: There are not races, the GIDs are checked only by filesystem
334  *      operations (as far as semantic preservation is concerned).
335  */
336 #ifdef CONFIG_MULTIUSER
337 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
338 {
339 	struct user_namespace *ns = current_user_ns();
340 	const struct cred *old;
341 	struct cred *new;
342 	int retval;
343 	kgid_t krgid, kegid;
344 
345 	krgid = make_kgid(ns, rgid);
346 	kegid = make_kgid(ns, egid);
347 
348 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
349 		return -EINVAL;
350 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
351 		return -EINVAL;
352 
353 	new = prepare_creds();
354 	if (!new)
355 		return -ENOMEM;
356 	old = current_cred();
357 
358 	retval = -EPERM;
359 	if (rgid != (gid_t) -1) {
360 		if (gid_eq(old->gid, krgid) ||
361 		    gid_eq(old->egid, krgid) ||
362 		    ns_capable(old->user_ns, CAP_SETGID))
363 			new->gid = krgid;
364 		else
365 			goto error;
366 	}
367 	if (egid != (gid_t) -1) {
368 		if (gid_eq(old->gid, kegid) ||
369 		    gid_eq(old->egid, kegid) ||
370 		    gid_eq(old->sgid, kegid) ||
371 		    ns_capable(old->user_ns, CAP_SETGID))
372 			new->egid = kegid;
373 		else
374 			goto error;
375 	}
376 
377 	if (rgid != (gid_t) -1 ||
378 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
379 		new->sgid = new->egid;
380 	new->fsgid = new->egid;
381 
382 	return commit_creds(new);
383 
384 error:
385 	abort_creds(new);
386 	return retval;
387 }
388 
389 /*
390  * setgid() is implemented like SysV w/ SAVED_IDS
391  *
392  * SMP: Same implicit races as above.
393  */
394 SYSCALL_DEFINE1(setgid, gid_t, gid)
395 {
396 	struct user_namespace *ns = current_user_ns();
397 	const struct cred *old;
398 	struct cred *new;
399 	int retval;
400 	kgid_t kgid;
401 
402 	kgid = make_kgid(ns, gid);
403 	if (!gid_valid(kgid))
404 		return -EINVAL;
405 
406 	new = prepare_creds();
407 	if (!new)
408 		return -ENOMEM;
409 	old = current_cred();
410 
411 	retval = -EPERM;
412 	if (ns_capable(old->user_ns, CAP_SETGID))
413 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
414 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
415 		new->egid = new->fsgid = kgid;
416 	else
417 		goto error;
418 
419 	return commit_creds(new);
420 
421 error:
422 	abort_creds(new);
423 	return retval;
424 }
425 
426 /*
427  * change the user struct in a credentials set to match the new UID
428  */
429 static int set_user(struct cred *new)
430 {
431 	struct user_struct *new_user;
432 
433 	new_user = alloc_uid(new->uid);
434 	if (!new_user)
435 		return -EAGAIN;
436 
437 	/*
438 	 * We don't fail in case of NPROC limit excess here because too many
439 	 * poorly written programs don't check set*uid() return code, assuming
440 	 * it never fails if called by root.  We may still enforce NPROC limit
441 	 * for programs doing set*uid()+execve() by harmlessly deferring the
442 	 * failure to the execve() stage.
443 	 */
444 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
445 			new_user != INIT_USER)
446 		current->flags |= PF_NPROC_EXCEEDED;
447 	else
448 		current->flags &= ~PF_NPROC_EXCEEDED;
449 
450 	free_uid(new->user);
451 	new->user = new_user;
452 	return 0;
453 }
454 
455 /*
456  * Unprivileged users may change the real uid to the effective uid
457  * or vice versa.  (BSD-style)
458  *
459  * If you set the real uid at all, or set the effective uid to a value not
460  * equal to the real uid, then the saved uid is set to the new effective uid.
461  *
462  * This makes it possible for a setuid program to completely drop its
463  * privileges, which is often a useful assertion to make when you are doing
464  * a security audit over a program.
465  *
466  * The general idea is that a program which uses just setreuid() will be
467  * 100% compatible with BSD.  A program which uses just setuid() will be
468  * 100% compatible with POSIX with saved IDs.
469  */
470 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
471 {
472 	struct user_namespace *ns = current_user_ns();
473 	const struct cred *old;
474 	struct cred *new;
475 	int retval;
476 	kuid_t kruid, keuid;
477 
478 	kruid = make_kuid(ns, ruid);
479 	keuid = make_kuid(ns, euid);
480 
481 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
482 		return -EINVAL;
483 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
484 		return -EINVAL;
485 
486 	new = prepare_creds();
487 	if (!new)
488 		return -ENOMEM;
489 	old = current_cred();
490 
491 	retval = -EPERM;
492 	if (ruid != (uid_t) -1) {
493 		new->uid = kruid;
494 		if (!uid_eq(old->uid, kruid) &&
495 		    !uid_eq(old->euid, kruid) &&
496 		    !ns_capable(old->user_ns, CAP_SETUID))
497 			goto error;
498 	}
499 
500 	if (euid != (uid_t) -1) {
501 		new->euid = keuid;
502 		if (!uid_eq(old->uid, keuid) &&
503 		    !uid_eq(old->euid, keuid) &&
504 		    !uid_eq(old->suid, keuid) &&
505 		    !ns_capable(old->user_ns, CAP_SETUID))
506 			goto error;
507 	}
508 
509 	if (!uid_eq(new->uid, old->uid)) {
510 		retval = set_user(new);
511 		if (retval < 0)
512 			goto error;
513 	}
514 	if (ruid != (uid_t) -1 ||
515 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
516 		new->suid = new->euid;
517 	new->fsuid = new->euid;
518 
519 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
520 	if (retval < 0)
521 		goto error;
522 
523 	return commit_creds(new);
524 
525 error:
526 	abort_creds(new);
527 	return retval;
528 }
529 
530 /*
531  * setuid() is implemented like SysV with SAVED_IDS
532  *
533  * Note that SAVED_ID's is deficient in that a setuid root program
534  * like sendmail, for example, cannot set its uid to be a normal
535  * user and then switch back, because if you're root, setuid() sets
536  * the saved uid too.  If you don't like this, blame the bright people
537  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
538  * will allow a root program to temporarily drop privileges and be able to
539  * regain them by swapping the real and effective uid.
540  */
541 SYSCALL_DEFINE1(setuid, uid_t, uid)
542 {
543 	struct user_namespace *ns = current_user_ns();
544 	const struct cred *old;
545 	struct cred *new;
546 	int retval;
547 	kuid_t kuid;
548 
549 	kuid = make_kuid(ns, uid);
550 	if (!uid_valid(kuid))
551 		return -EINVAL;
552 
553 	new = prepare_creds();
554 	if (!new)
555 		return -ENOMEM;
556 	old = current_cred();
557 
558 	retval = -EPERM;
559 	if (ns_capable(old->user_ns, CAP_SETUID)) {
560 		new->suid = new->uid = kuid;
561 		if (!uid_eq(kuid, old->uid)) {
562 			retval = set_user(new);
563 			if (retval < 0)
564 				goto error;
565 		}
566 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
567 		goto error;
568 	}
569 
570 	new->fsuid = new->euid = kuid;
571 
572 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
573 	if (retval < 0)
574 		goto error;
575 
576 	return commit_creds(new);
577 
578 error:
579 	abort_creds(new);
580 	return retval;
581 }
582 
583 
584 /*
585  * This function implements a generic ability to update ruid, euid,
586  * and suid.  This allows you to implement the 4.4 compatible seteuid().
587  */
588 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
589 {
590 	struct user_namespace *ns = current_user_ns();
591 	const struct cred *old;
592 	struct cred *new;
593 	int retval;
594 	kuid_t kruid, keuid, ksuid;
595 
596 	kruid = make_kuid(ns, ruid);
597 	keuid = make_kuid(ns, euid);
598 	ksuid = make_kuid(ns, suid);
599 
600 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
601 		return -EINVAL;
602 
603 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
604 		return -EINVAL;
605 
606 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
607 		return -EINVAL;
608 
609 	new = prepare_creds();
610 	if (!new)
611 		return -ENOMEM;
612 
613 	old = current_cred();
614 
615 	retval = -EPERM;
616 	if (!ns_capable(old->user_ns, CAP_SETUID)) {
617 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
618 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
619 			goto error;
620 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
621 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
622 			goto error;
623 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
624 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
625 			goto error;
626 	}
627 
628 	if (ruid != (uid_t) -1) {
629 		new->uid = kruid;
630 		if (!uid_eq(kruid, old->uid)) {
631 			retval = set_user(new);
632 			if (retval < 0)
633 				goto error;
634 		}
635 	}
636 	if (euid != (uid_t) -1)
637 		new->euid = keuid;
638 	if (suid != (uid_t) -1)
639 		new->suid = ksuid;
640 	new->fsuid = new->euid;
641 
642 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
643 	if (retval < 0)
644 		goto error;
645 
646 	return commit_creds(new);
647 
648 error:
649 	abort_creds(new);
650 	return retval;
651 }
652 
653 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
654 {
655 	const struct cred *cred = current_cred();
656 	int retval;
657 	uid_t ruid, euid, suid;
658 
659 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
660 	euid = from_kuid_munged(cred->user_ns, cred->euid);
661 	suid = from_kuid_munged(cred->user_ns, cred->suid);
662 
663 	retval = put_user(ruid, ruidp);
664 	if (!retval) {
665 		retval = put_user(euid, euidp);
666 		if (!retval)
667 			return put_user(suid, suidp);
668 	}
669 	return retval;
670 }
671 
672 /*
673  * Same as above, but for rgid, egid, sgid.
674  */
675 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
676 {
677 	struct user_namespace *ns = current_user_ns();
678 	const struct cred *old;
679 	struct cred *new;
680 	int retval;
681 	kgid_t krgid, kegid, ksgid;
682 
683 	krgid = make_kgid(ns, rgid);
684 	kegid = make_kgid(ns, egid);
685 	ksgid = make_kgid(ns, sgid);
686 
687 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
688 		return -EINVAL;
689 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
690 		return -EINVAL;
691 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
692 		return -EINVAL;
693 
694 	new = prepare_creds();
695 	if (!new)
696 		return -ENOMEM;
697 	old = current_cred();
698 
699 	retval = -EPERM;
700 	if (!ns_capable(old->user_ns, CAP_SETGID)) {
701 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
702 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
703 			goto error;
704 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
705 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
706 			goto error;
707 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
708 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
709 			goto error;
710 	}
711 
712 	if (rgid != (gid_t) -1)
713 		new->gid = krgid;
714 	if (egid != (gid_t) -1)
715 		new->egid = kegid;
716 	if (sgid != (gid_t) -1)
717 		new->sgid = ksgid;
718 	new->fsgid = new->egid;
719 
720 	return commit_creds(new);
721 
722 error:
723 	abort_creds(new);
724 	return retval;
725 }
726 
727 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
728 {
729 	const struct cred *cred = current_cred();
730 	int retval;
731 	gid_t rgid, egid, sgid;
732 
733 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
734 	egid = from_kgid_munged(cred->user_ns, cred->egid);
735 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
736 
737 	retval = put_user(rgid, rgidp);
738 	if (!retval) {
739 		retval = put_user(egid, egidp);
740 		if (!retval)
741 			retval = put_user(sgid, sgidp);
742 	}
743 
744 	return retval;
745 }
746 
747 
748 /*
749  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
750  * is used for "access()" and for the NFS daemon (letting nfsd stay at
751  * whatever uid it wants to). It normally shadows "euid", except when
752  * explicitly set by setfsuid() or for access..
753  */
754 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
755 {
756 	const struct cred *old;
757 	struct cred *new;
758 	uid_t old_fsuid;
759 	kuid_t kuid;
760 
761 	old = current_cred();
762 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
763 
764 	kuid = make_kuid(old->user_ns, uid);
765 	if (!uid_valid(kuid))
766 		return old_fsuid;
767 
768 	new = prepare_creds();
769 	if (!new)
770 		return old_fsuid;
771 
772 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
773 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
774 	    ns_capable(old->user_ns, CAP_SETUID)) {
775 		if (!uid_eq(kuid, old->fsuid)) {
776 			new->fsuid = kuid;
777 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
778 				goto change_okay;
779 		}
780 	}
781 
782 	abort_creds(new);
783 	return old_fsuid;
784 
785 change_okay:
786 	commit_creds(new);
787 	return old_fsuid;
788 }
789 
790 /*
791  * Samma pÃ¥ svenska..
792  */
793 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
794 {
795 	const struct cred *old;
796 	struct cred *new;
797 	gid_t old_fsgid;
798 	kgid_t kgid;
799 
800 	old = current_cred();
801 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
802 
803 	kgid = make_kgid(old->user_ns, gid);
804 	if (!gid_valid(kgid))
805 		return old_fsgid;
806 
807 	new = prepare_creds();
808 	if (!new)
809 		return old_fsgid;
810 
811 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
812 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
813 	    ns_capable(old->user_ns, CAP_SETGID)) {
814 		if (!gid_eq(kgid, old->fsgid)) {
815 			new->fsgid = kgid;
816 			goto change_okay;
817 		}
818 	}
819 
820 	abort_creds(new);
821 	return old_fsgid;
822 
823 change_okay:
824 	commit_creds(new);
825 	return old_fsgid;
826 }
827 #endif /* CONFIG_MULTIUSER */
828 
829 /**
830  * sys_getpid - return the thread group id of the current process
831  *
832  * Note, despite the name, this returns the tgid not the pid.  The tgid and
833  * the pid are identical unless CLONE_THREAD was specified on clone() in
834  * which case the tgid is the same in all threads of the same group.
835  *
836  * This is SMP safe as current->tgid does not change.
837  */
838 SYSCALL_DEFINE0(getpid)
839 {
840 	return task_tgid_vnr(current);
841 }
842 
843 /* Thread ID - the internal kernel "pid" */
844 SYSCALL_DEFINE0(gettid)
845 {
846 	return task_pid_vnr(current);
847 }
848 
849 /*
850  * Accessing ->real_parent is not SMP-safe, it could
851  * change from under us. However, we can use a stale
852  * value of ->real_parent under rcu_read_lock(), see
853  * release_task()->call_rcu(delayed_put_task_struct).
854  */
855 SYSCALL_DEFINE0(getppid)
856 {
857 	int pid;
858 
859 	rcu_read_lock();
860 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
861 	rcu_read_unlock();
862 
863 	return pid;
864 }
865 
866 SYSCALL_DEFINE0(getuid)
867 {
868 	/* Only we change this so SMP safe */
869 	return from_kuid_munged(current_user_ns(), current_uid());
870 }
871 
872 SYSCALL_DEFINE0(geteuid)
873 {
874 	/* Only we change this so SMP safe */
875 	return from_kuid_munged(current_user_ns(), current_euid());
876 }
877 
878 SYSCALL_DEFINE0(getgid)
879 {
880 	/* Only we change this so SMP safe */
881 	return from_kgid_munged(current_user_ns(), current_gid());
882 }
883 
884 SYSCALL_DEFINE0(getegid)
885 {
886 	/* Only we change this so SMP safe */
887 	return from_kgid_munged(current_user_ns(), current_egid());
888 }
889 
890 static void do_sys_times(struct tms *tms)
891 {
892 	u64 tgutime, tgstime, cutime, cstime;
893 
894 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
895 	cutime = current->signal->cutime;
896 	cstime = current->signal->cstime;
897 	tms->tms_utime = nsec_to_clock_t(tgutime);
898 	tms->tms_stime = nsec_to_clock_t(tgstime);
899 	tms->tms_cutime = nsec_to_clock_t(cutime);
900 	tms->tms_cstime = nsec_to_clock_t(cstime);
901 }
902 
903 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
904 {
905 	if (tbuf) {
906 		struct tms tmp;
907 
908 		do_sys_times(&tmp);
909 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
910 			return -EFAULT;
911 	}
912 	force_successful_syscall_return();
913 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
914 }
915 
916 #ifdef CONFIG_COMPAT
917 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
918 {
919 	return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
920 }
921 
922 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
923 {
924 	if (tbuf) {
925 		struct tms tms;
926 		struct compat_tms tmp;
927 
928 		do_sys_times(&tms);
929 		/* Convert our struct tms to the compat version. */
930 		tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
931 		tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
932 		tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
933 		tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
934 		if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
935 			return -EFAULT;
936 	}
937 	force_successful_syscall_return();
938 	return compat_jiffies_to_clock_t(jiffies);
939 }
940 #endif
941 
942 /*
943  * This needs some heavy checking ...
944  * I just haven't the stomach for it. I also don't fully
945  * understand sessions/pgrp etc. Let somebody who does explain it.
946  *
947  * OK, I think I have the protection semantics right.... this is really
948  * only important on a multi-user system anyway, to make sure one user
949  * can't send a signal to a process owned by another.  -TYT, 12/12/91
950  *
951  * !PF_FORKNOEXEC check to conform completely to POSIX.
952  */
953 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
954 {
955 	struct task_struct *p;
956 	struct task_struct *group_leader = current->group_leader;
957 	struct pid *pgrp;
958 	int err;
959 
960 	if (!pid)
961 		pid = task_pid_vnr(group_leader);
962 	if (!pgid)
963 		pgid = pid;
964 	if (pgid < 0)
965 		return -EINVAL;
966 	rcu_read_lock();
967 
968 	/* From this point forward we keep holding onto the tasklist lock
969 	 * so that our parent does not change from under us. -DaveM
970 	 */
971 	write_lock_irq(&tasklist_lock);
972 
973 	err = -ESRCH;
974 	p = find_task_by_vpid(pid);
975 	if (!p)
976 		goto out;
977 
978 	err = -EINVAL;
979 	if (!thread_group_leader(p))
980 		goto out;
981 
982 	if (same_thread_group(p->real_parent, group_leader)) {
983 		err = -EPERM;
984 		if (task_session(p) != task_session(group_leader))
985 			goto out;
986 		err = -EACCES;
987 		if (!(p->flags & PF_FORKNOEXEC))
988 			goto out;
989 	} else {
990 		err = -ESRCH;
991 		if (p != group_leader)
992 			goto out;
993 	}
994 
995 	err = -EPERM;
996 	if (p->signal->leader)
997 		goto out;
998 
999 	pgrp = task_pid(p);
1000 	if (pgid != pid) {
1001 		struct task_struct *g;
1002 
1003 		pgrp = find_vpid(pgid);
1004 		g = pid_task(pgrp, PIDTYPE_PGID);
1005 		if (!g || task_session(g) != task_session(group_leader))
1006 			goto out;
1007 	}
1008 
1009 	err = security_task_setpgid(p, pgid);
1010 	if (err)
1011 		goto out;
1012 
1013 	if (task_pgrp(p) != pgrp)
1014 		change_pid(p, PIDTYPE_PGID, pgrp);
1015 
1016 	err = 0;
1017 out:
1018 	/* All paths lead to here, thus we are safe. -DaveM */
1019 	write_unlock_irq(&tasklist_lock);
1020 	rcu_read_unlock();
1021 	return err;
1022 }
1023 
1024 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1025 {
1026 	struct task_struct *p;
1027 	struct pid *grp;
1028 	int retval;
1029 
1030 	rcu_read_lock();
1031 	if (!pid)
1032 		grp = task_pgrp(current);
1033 	else {
1034 		retval = -ESRCH;
1035 		p = find_task_by_vpid(pid);
1036 		if (!p)
1037 			goto out;
1038 		grp = task_pgrp(p);
1039 		if (!grp)
1040 			goto out;
1041 
1042 		retval = security_task_getpgid(p);
1043 		if (retval)
1044 			goto out;
1045 	}
1046 	retval = pid_vnr(grp);
1047 out:
1048 	rcu_read_unlock();
1049 	return retval;
1050 }
1051 
1052 #ifdef __ARCH_WANT_SYS_GETPGRP
1053 
1054 SYSCALL_DEFINE0(getpgrp)
1055 {
1056 	return sys_getpgid(0);
1057 }
1058 
1059 #endif
1060 
1061 SYSCALL_DEFINE1(getsid, pid_t, pid)
1062 {
1063 	struct task_struct *p;
1064 	struct pid *sid;
1065 	int retval;
1066 
1067 	rcu_read_lock();
1068 	if (!pid)
1069 		sid = task_session(current);
1070 	else {
1071 		retval = -ESRCH;
1072 		p = find_task_by_vpid(pid);
1073 		if (!p)
1074 			goto out;
1075 		sid = task_session(p);
1076 		if (!sid)
1077 			goto out;
1078 
1079 		retval = security_task_getsid(p);
1080 		if (retval)
1081 			goto out;
1082 	}
1083 	retval = pid_vnr(sid);
1084 out:
1085 	rcu_read_unlock();
1086 	return retval;
1087 }
1088 
1089 static void set_special_pids(struct pid *pid)
1090 {
1091 	struct task_struct *curr = current->group_leader;
1092 
1093 	if (task_session(curr) != pid)
1094 		change_pid(curr, PIDTYPE_SID, pid);
1095 
1096 	if (task_pgrp(curr) != pid)
1097 		change_pid(curr, PIDTYPE_PGID, pid);
1098 }
1099 
1100 SYSCALL_DEFINE0(setsid)
1101 {
1102 	struct task_struct *group_leader = current->group_leader;
1103 	struct pid *sid = task_pid(group_leader);
1104 	pid_t session = pid_vnr(sid);
1105 	int err = -EPERM;
1106 
1107 	write_lock_irq(&tasklist_lock);
1108 	/* Fail if I am already a session leader */
1109 	if (group_leader->signal->leader)
1110 		goto out;
1111 
1112 	/* Fail if a process group id already exists that equals the
1113 	 * proposed session id.
1114 	 */
1115 	if (pid_task(sid, PIDTYPE_PGID))
1116 		goto out;
1117 
1118 	group_leader->signal->leader = 1;
1119 	set_special_pids(sid);
1120 
1121 	proc_clear_tty(group_leader);
1122 
1123 	err = session;
1124 out:
1125 	write_unlock_irq(&tasklist_lock);
1126 	if (err > 0) {
1127 		proc_sid_connector(group_leader);
1128 		sched_autogroup_create_attach(group_leader);
1129 	}
1130 	return err;
1131 }
1132 
1133 DECLARE_RWSEM(uts_sem);
1134 
1135 #ifdef COMPAT_UTS_MACHINE
1136 #define override_architecture(name) \
1137 	(personality(current->personality) == PER_LINUX32 && \
1138 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1139 		      sizeof(COMPAT_UTS_MACHINE)))
1140 #else
1141 #define override_architecture(name)	0
1142 #endif
1143 
1144 /*
1145  * Work around broken programs that cannot handle "Linux 3.0".
1146  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1147  * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1148  */
1149 static int override_release(char __user *release, size_t len)
1150 {
1151 	int ret = 0;
1152 
1153 	if (current->personality & UNAME26) {
1154 		const char *rest = UTS_RELEASE;
1155 		char buf[65] = { 0 };
1156 		int ndots = 0;
1157 		unsigned v;
1158 		size_t copy;
1159 
1160 		while (*rest) {
1161 			if (*rest == '.' && ++ndots >= 3)
1162 				break;
1163 			if (!isdigit(*rest) && *rest != '.')
1164 				break;
1165 			rest++;
1166 		}
1167 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1168 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1169 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1170 		ret = copy_to_user(release, buf, copy + 1);
1171 	}
1172 	return ret;
1173 }
1174 
1175 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1176 {
1177 	int errno = 0;
1178 
1179 	down_read(&uts_sem);
1180 	if (copy_to_user(name, utsname(), sizeof *name))
1181 		errno = -EFAULT;
1182 	up_read(&uts_sem);
1183 
1184 	if (!errno && override_release(name->release, sizeof(name->release)))
1185 		errno = -EFAULT;
1186 	if (!errno && override_architecture(name))
1187 		errno = -EFAULT;
1188 	return errno;
1189 }
1190 
1191 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1192 /*
1193  * Old cruft
1194  */
1195 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1196 {
1197 	int error = 0;
1198 
1199 	if (!name)
1200 		return -EFAULT;
1201 
1202 	down_read(&uts_sem);
1203 	if (copy_to_user(name, utsname(), sizeof(*name)))
1204 		error = -EFAULT;
1205 	up_read(&uts_sem);
1206 
1207 	if (!error && override_release(name->release, sizeof(name->release)))
1208 		error = -EFAULT;
1209 	if (!error && override_architecture(name))
1210 		error = -EFAULT;
1211 	return error;
1212 }
1213 
1214 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1215 {
1216 	int error;
1217 
1218 	if (!name)
1219 		return -EFAULT;
1220 	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1221 		return -EFAULT;
1222 
1223 	down_read(&uts_sem);
1224 	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1225 			       __OLD_UTS_LEN);
1226 	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1227 	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1228 				__OLD_UTS_LEN);
1229 	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1230 	error |= __copy_to_user(&name->release, &utsname()->release,
1231 				__OLD_UTS_LEN);
1232 	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1233 	error |= __copy_to_user(&name->version, &utsname()->version,
1234 				__OLD_UTS_LEN);
1235 	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1236 	error |= __copy_to_user(&name->machine, &utsname()->machine,
1237 				__OLD_UTS_LEN);
1238 	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1239 	up_read(&uts_sem);
1240 
1241 	if (!error && override_architecture(name))
1242 		error = -EFAULT;
1243 	if (!error && override_release(name->release, sizeof(name->release)))
1244 		error = -EFAULT;
1245 	return error ? -EFAULT : 0;
1246 }
1247 #endif
1248 
1249 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1250 {
1251 	int errno;
1252 	char tmp[__NEW_UTS_LEN];
1253 
1254 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1255 		return -EPERM;
1256 
1257 	if (len < 0 || len > __NEW_UTS_LEN)
1258 		return -EINVAL;
1259 	down_write(&uts_sem);
1260 	errno = -EFAULT;
1261 	if (!copy_from_user(tmp, name, len)) {
1262 		struct new_utsname *u = utsname();
1263 
1264 		memcpy(u->nodename, tmp, len);
1265 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1266 		errno = 0;
1267 		uts_proc_notify(UTS_PROC_HOSTNAME);
1268 	}
1269 	up_write(&uts_sem);
1270 	return errno;
1271 }
1272 
1273 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1274 
1275 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1276 {
1277 	int i, errno;
1278 	struct new_utsname *u;
1279 
1280 	if (len < 0)
1281 		return -EINVAL;
1282 	down_read(&uts_sem);
1283 	u = utsname();
1284 	i = 1 + strlen(u->nodename);
1285 	if (i > len)
1286 		i = len;
1287 	errno = 0;
1288 	if (copy_to_user(name, u->nodename, i))
1289 		errno = -EFAULT;
1290 	up_read(&uts_sem);
1291 	return errno;
1292 }
1293 
1294 #endif
1295 
1296 /*
1297  * Only setdomainname; getdomainname can be implemented by calling
1298  * uname()
1299  */
1300 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1301 {
1302 	int errno;
1303 	char tmp[__NEW_UTS_LEN];
1304 
1305 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1306 		return -EPERM;
1307 	if (len < 0 || len > __NEW_UTS_LEN)
1308 		return -EINVAL;
1309 
1310 	down_write(&uts_sem);
1311 	errno = -EFAULT;
1312 	if (!copy_from_user(tmp, name, len)) {
1313 		struct new_utsname *u = utsname();
1314 
1315 		memcpy(u->domainname, tmp, len);
1316 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1317 		errno = 0;
1318 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1319 	}
1320 	up_write(&uts_sem);
1321 	return errno;
1322 }
1323 
1324 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1325 {
1326 	struct rlimit value;
1327 	int ret;
1328 
1329 	ret = do_prlimit(current, resource, NULL, &value);
1330 	if (!ret)
1331 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1332 
1333 	return ret;
1334 }
1335 
1336 #ifdef CONFIG_COMPAT
1337 
1338 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1339 		       struct compat_rlimit __user *, rlim)
1340 {
1341 	struct rlimit r;
1342 	struct compat_rlimit r32;
1343 
1344 	if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1345 		return -EFAULT;
1346 
1347 	if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1348 		r.rlim_cur = RLIM_INFINITY;
1349 	else
1350 		r.rlim_cur = r32.rlim_cur;
1351 	if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1352 		r.rlim_max = RLIM_INFINITY;
1353 	else
1354 		r.rlim_max = r32.rlim_max;
1355 	return do_prlimit(current, resource, &r, NULL);
1356 }
1357 
1358 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1359 		       struct compat_rlimit __user *, rlim)
1360 {
1361 	struct rlimit r;
1362 	int ret;
1363 
1364 	ret = do_prlimit(current, resource, NULL, &r);
1365 	if (!ret) {
1366 		struct compat_rlimit r32;
1367 		if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1368 			r32.rlim_cur = COMPAT_RLIM_INFINITY;
1369 		else
1370 			r32.rlim_cur = r.rlim_cur;
1371 		if (r.rlim_max > COMPAT_RLIM_INFINITY)
1372 			r32.rlim_max = COMPAT_RLIM_INFINITY;
1373 		else
1374 			r32.rlim_max = r.rlim_max;
1375 
1376 		if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1377 			return -EFAULT;
1378 	}
1379 	return ret;
1380 }
1381 
1382 #endif
1383 
1384 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1385 
1386 /*
1387  *	Back compatibility for getrlimit. Needed for some apps.
1388  */
1389 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1390 		struct rlimit __user *, rlim)
1391 {
1392 	struct rlimit x;
1393 	if (resource >= RLIM_NLIMITS)
1394 		return -EINVAL;
1395 
1396 	task_lock(current->group_leader);
1397 	x = current->signal->rlim[resource];
1398 	task_unlock(current->group_leader);
1399 	if (x.rlim_cur > 0x7FFFFFFF)
1400 		x.rlim_cur = 0x7FFFFFFF;
1401 	if (x.rlim_max > 0x7FFFFFFF)
1402 		x.rlim_max = 0x7FFFFFFF;
1403 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1404 }
1405 
1406 #ifdef CONFIG_COMPAT
1407 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1408 		       struct compat_rlimit __user *, rlim)
1409 {
1410 	struct rlimit r;
1411 
1412 	if (resource >= RLIM_NLIMITS)
1413 		return -EINVAL;
1414 
1415 	task_lock(current->group_leader);
1416 	r = current->signal->rlim[resource];
1417 	task_unlock(current->group_leader);
1418 	if (r.rlim_cur > 0x7FFFFFFF)
1419 		r.rlim_cur = 0x7FFFFFFF;
1420 	if (r.rlim_max > 0x7FFFFFFF)
1421 		r.rlim_max = 0x7FFFFFFF;
1422 
1423 	if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1424 	    put_user(r.rlim_max, &rlim->rlim_max))
1425 		return -EFAULT;
1426 	return 0;
1427 }
1428 #endif
1429 
1430 #endif
1431 
1432 static inline bool rlim64_is_infinity(__u64 rlim64)
1433 {
1434 #if BITS_PER_LONG < 64
1435 	return rlim64 >= ULONG_MAX;
1436 #else
1437 	return rlim64 == RLIM64_INFINITY;
1438 #endif
1439 }
1440 
1441 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1442 {
1443 	if (rlim->rlim_cur == RLIM_INFINITY)
1444 		rlim64->rlim_cur = RLIM64_INFINITY;
1445 	else
1446 		rlim64->rlim_cur = rlim->rlim_cur;
1447 	if (rlim->rlim_max == RLIM_INFINITY)
1448 		rlim64->rlim_max = RLIM64_INFINITY;
1449 	else
1450 		rlim64->rlim_max = rlim->rlim_max;
1451 }
1452 
1453 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1454 {
1455 	if (rlim64_is_infinity(rlim64->rlim_cur))
1456 		rlim->rlim_cur = RLIM_INFINITY;
1457 	else
1458 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1459 	if (rlim64_is_infinity(rlim64->rlim_max))
1460 		rlim->rlim_max = RLIM_INFINITY;
1461 	else
1462 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1463 }
1464 
1465 /* make sure you are allowed to change @tsk limits before calling this */
1466 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1467 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1468 {
1469 	struct rlimit *rlim;
1470 	int retval = 0;
1471 
1472 	if (resource >= RLIM_NLIMITS)
1473 		return -EINVAL;
1474 	if (new_rlim) {
1475 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1476 			return -EINVAL;
1477 		if (resource == RLIMIT_NOFILE &&
1478 				new_rlim->rlim_max > sysctl_nr_open)
1479 			return -EPERM;
1480 	}
1481 
1482 	/* protect tsk->signal and tsk->sighand from disappearing */
1483 	read_lock(&tasklist_lock);
1484 	if (!tsk->sighand) {
1485 		retval = -ESRCH;
1486 		goto out;
1487 	}
1488 
1489 	rlim = tsk->signal->rlim + resource;
1490 	task_lock(tsk->group_leader);
1491 	if (new_rlim) {
1492 		/* Keep the capable check against init_user_ns until
1493 		   cgroups can contain all limits */
1494 		if (new_rlim->rlim_max > rlim->rlim_max &&
1495 				!capable(CAP_SYS_RESOURCE))
1496 			retval = -EPERM;
1497 		if (!retval)
1498 			retval = security_task_setrlimit(tsk, resource, new_rlim);
1499 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1500 			/*
1501 			 * The caller is asking for an immediate RLIMIT_CPU
1502 			 * expiry.  But we use the zero value to mean "it was
1503 			 * never set".  So let's cheat and make it one second
1504 			 * instead
1505 			 */
1506 			new_rlim->rlim_cur = 1;
1507 		}
1508 	}
1509 	if (!retval) {
1510 		if (old_rlim)
1511 			*old_rlim = *rlim;
1512 		if (new_rlim)
1513 			*rlim = *new_rlim;
1514 	}
1515 	task_unlock(tsk->group_leader);
1516 
1517 	/*
1518 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1519 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1520 	 * very long-standing error, and fixing it now risks breakage of
1521 	 * applications, so we live with it
1522 	 */
1523 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1524 	     new_rlim->rlim_cur != RLIM_INFINITY &&
1525 	     IS_ENABLED(CONFIG_POSIX_TIMERS))
1526 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1527 out:
1528 	read_unlock(&tasklist_lock);
1529 	return retval;
1530 }
1531 
1532 /* rcu lock must be held */
1533 static int check_prlimit_permission(struct task_struct *task,
1534 				    unsigned int flags)
1535 {
1536 	const struct cred *cred = current_cred(), *tcred;
1537 	bool id_match;
1538 
1539 	if (current == task)
1540 		return 0;
1541 
1542 	tcred = __task_cred(task);
1543 	id_match = (uid_eq(cred->uid, tcred->euid) &&
1544 		    uid_eq(cred->uid, tcred->suid) &&
1545 		    uid_eq(cred->uid, tcred->uid)  &&
1546 		    gid_eq(cred->gid, tcred->egid) &&
1547 		    gid_eq(cred->gid, tcred->sgid) &&
1548 		    gid_eq(cred->gid, tcred->gid));
1549 	if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1550 		return -EPERM;
1551 
1552 	return security_task_prlimit(cred, tcred, flags);
1553 }
1554 
1555 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1556 		const struct rlimit64 __user *, new_rlim,
1557 		struct rlimit64 __user *, old_rlim)
1558 {
1559 	struct rlimit64 old64, new64;
1560 	struct rlimit old, new;
1561 	struct task_struct *tsk;
1562 	unsigned int checkflags = 0;
1563 	int ret;
1564 
1565 	if (old_rlim)
1566 		checkflags |= LSM_PRLIMIT_READ;
1567 
1568 	if (new_rlim) {
1569 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1570 			return -EFAULT;
1571 		rlim64_to_rlim(&new64, &new);
1572 		checkflags |= LSM_PRLIMIT_WRITE;
1573 	}
1574 
1575 	rcu_read_lock();
1576 	tsk = pid ? find_task_by_vpid(pid) : current;
1577 	if (!tsk) {
1578 		rcu_read_unlock();
1579 		return -ESRCH;
1580 	}
1581 	ret = check_prlimit_permission(tsk, checkflags);
1582 	if (ret) {
1583 		rcu_read_unlock();
1584 		return ret;
1585 	}
1586 	get_task_struct(tsk);
1587 	rcu_read_unlock();
1588 
1589 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1590 			old_rlim ? &old : NULL);
1591 
1592 	if (!ret && old_rlim) {
1593 		rlim_to_rlim64(&old, &old64);
1594 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1595 			ret = -EFAULT;
1596 	}
1597 
1598 	put_task_struct(tsk);
1599 	return ret;
1600 }
1601 
1602 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1603 {
1604 	struct rlimit new_rlim;
1605 
1606 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1607 		return -EFAULT;
1608 	return do_prlimit(current, resource, &new_rlim, NULL);
1609 }
1610 
1611 /*
1612  * It would make sense to put struct rusage in the task_struct,
1613  * except that would make the task_struct be *really big*.  After
1614  * task_struct gets moved into malloc'ed memory, it would
1615  * make sense to do this.  It will make moving the rest of the information
1616  * a lot simpler!  (Which we're not doing right now because we're not
1617  * measuring them yet).
1618  *
1619  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1620  * races with threads incrementing their own counters.  But since word
1621  * reads are atomic, we either get new values or old values and we don't
1622  * care which for the sums.  We always take the siglock to protect reading
1623  * the c* fields from p->signal from races with exit.c updating those
1624  * fields when reaping, so a sample either gets all the additions of a
1625  * given child after it's reaped, or none so this sample is before reaping.
1626  *
1627  * Locking:
1628  * We need to take the siglock for CHILDEREN, SELF and BOTH
1629  * for  the cases current multithreaded, non-current single threaded
1630  * non-current multithreaded.  Thread traversal is now safe with
1631  * the siglock held.
1632  * Strictly speaking, we donot need to take the siglock if we are current and
1633  * single threaded,  as no one else can take our signal_struct away, no one
1634  * else can  reap the  children to update signal->c* counters, and no one else
1635  * can race with the signal-> fields. If we do not take any lock, the
1636  * signal-> fields could be read out of order while another thread was just
1637  * exiting. So we should  place a read memory barrier when we avoid the lock.
1638  * On the writer side,  write memory barrier is implied in  __exit_signal
1639  * as __exit_signal releases  the siglock spinlock after updating the signal->
1640  * fields. But we don't do this yet to keep things simple.
1641  *
1642  */
1643 
1644 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1645 {
1646 	r->ru_nvcsw += t->nvcsw;
1647 	r->ru_nivcsw += t->nivcsw;
1648 	r->ru_minflt += t->min_flt;
1649 	r->ru_majflt += t->maj_flt;
1650 	r->ru_inblock += task_io_get_inblock(t);
1651 	r->ru_oublock += task_io_get_oublock(t);
1652 }
1653 
1654 void getrusage(struct task_struct *p, int who, struct rusage *r)
1655 {
1656 	struct task_struct *t;
1657 	unsigned long flags;
1658 	u64 tgutime, tgstime, utime, stime;
1659 	unsigned long maxrss = 0;
1660 
1661 	memset((char *)r, 0, sizeof (*r));
1662 	utime = stime = 0;
1663 
1664 	if (who == RUSAGE_THREAD) {
1665 		task_cputime_adjusted(current, &utime, &stime);
1666 		accumulate_thread_rusage(p, r);
1667 		maxrss = p->signal->maxrss;
1668 		goto out;
1669 	}
1670 
1671 	if (!lock_task_sighand(p, &flags))
1672 		return;
1673 
1674 	switch (who) {
1675 	case RUSAGE_BOTH:
1676 	case RUSAGE_CHILDREN:
1677 		utime = p->signal->cutime;
1678 		stime = p->signal->cstime;
1679 		r->ru_nvcsw = p->signal->cnvcsw;
1680 		r->ru_nivcsw = p->signal->cnivcsw;
1681 		r->ru_minflt = p->signal->cmin_flt;
1682 		r->ru_majflt = p->signal->cmaj_flt;
1683 		r->ru_inblock = p->signal->cinblock;
1684 		r->ru_oublock = p->signal->coublock;
1685 		maxrss = p->signal->cmaxrss;
1686 
1687 		if (who == RUSAGE_CHILDREN)
1688 			break;
1689 
1690 	case RUSAGE_SELF:
1691 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1692 		utime += tgutime;
1693 		stime += tgstime;
1694 		r->ru_nvcsw += p->signal->nvcsw;
1695 		r->ru_nivcsw += p->signal->nivcsw;
1696 		r->ru_minflt += p->signal->min_flt;
1697 		r->ru_majflt += p->signal->maj_flt;
1698 		r->ru_inblock += p->signal->inblock;
1699 		r->ru_oublock += p->signal->oublock;
1700 		if (maxrss < p->signal->maxrss)
1701 			maxrss = p->signal->maxrss;
1702 		t = p;
1703 		do {
1704 			accumulate_thread_rusage(t, r);
1705 		} while_each_thread(p, t);
1706 		break;
1707 
1708 	default:
1709 		BUG();
1710 	}
1711 	unlock_task_sighand(p, &flags);
1712 
1713 out:
1714 	r->ru_utime = ns_to_timeval(utime);
1715 	r->ru_stime = ns_to_timeval(stime);
1716 
1717 	if (who != RUSAGE_CHILDREN) {
1718 		struct mm_struct *mm = get_task_mm(p);
1719 
1720 		if (mm) {
1721 			setmax_mm_hiwater_rss(&maxrss, mm);
1722 			mmput(mm);
1723 		}
1724 	}
1725 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1726 }
1727 
1728 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1729 {
1730 	struct rusage r;
1731 
1732 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1733 	    who != RUSAGE_THREAD)
1734 		return -EINVAL;
1735 
1736 	getrusage(current, who, &r);
1737 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1738 }
1739 
1740 #ifdef CONFIG_COMPAT
1741 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1742 {
1743 	struct rusage r;
1744 
1745 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1746 	    who != RUSAGE_THREAD)
1747 		return -EINVAL;
1748 
1749 	getrusage(current, who, &r);
1750 	return put_compat_rusage(&r, ru);
1751 }
1752 #endif
1753 
1754 SYSCALL_DEFINE1(umask, int, mask)
1755 {
1756 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1757 	return mask;
1758 }
1759 
1760 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1761 {
1762 	struct fd exe;
1763 	struct file *old_exe, *exe_file;
1764 	struct inode *inode;
1765 	int err;
1766 
1767 	exe = fdget(fd);
1768 	if (!exe.file)
1769 		return -EBADF;
1770 
1771 	inode = file_inode(exe.file);
1772 
1773 	/*
1774 	 * Because the original mm->exe_file points to executable file, make
1775 	 * sure that this one is executable as well, to avoid breaking an
1776 	 * overall picture.
1777 	 */
1778 	err = -EACCES;
1779 	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1780 		goto exit;
1781 
1782 	err = inode_permission(inode, MAY_EXEC);
1783 	if (err)
1784 		goto exit;
1785 
1786 	/*
1787 	 * Forbid mm->exe_file change if old file still mapped.
1788 	 */
1789 	exe_file = get_mm_exe_file(mm);
1790 	err = -EBUSY;
1791 	if (exe_file) {
1792 		struct vm_area_struct *vma;
1793 
1794 		down_read(&mm->mmap_sem);
1795 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
1796 			if (!vma->vm_file)
1797 				continue;
1798 			if (path_equal(&vma->vm_file->f_path,
1799 				       &exe_file->f_path))
1800 				goto exit_err;
1801 		}
1802 
1803 		up_read(&mm->mmap_sem);
1804 		fput(exe_file);
1805 	}
1806 
1807 	err = 0;
1808 	/* set the new file, lockless */
1809 	get_file(exe.file);
1810 	old_exe = xchg(&mm->exe_file, exe.file);
1811 	if (old_exe)
1812 		fput(old_exe);
1813 exit:
1814 	fdput(exe);
1815 	return err;
1816 exit_err:
1817 	up_read(&mm->mmap_sem);
1818 	fput(exe_file);
1819 	goto exit;
1820 }
1821 
1822 /*
1823  * WARNING: we don't require any capability here so be very careful
1824  * in what is allowed for modification from userspace.
1825  */
1826 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1827 {
1828 	unsigned long mmap_max_addr = TASK_SIZE;
1829 	struct mm_struct *mm = current->mm;
1830 	int error = -EINVAL, i;
1831 
1832 	static const unsigned char offsets[] = {
1833 		offsetof(struct prctl_mm_map, start_code),
1834 		offsetof(struct prctl_mm_map, end_code),
1835 		offsetof(struct prctl_mm_map, start_data),
1836 		offsetof(struct prctl_mm_map, end_data),
1837 		offsetof(struct prctl_mm_map, start_brk),
1838 		offsetof(struct prctl_mm_map, brk),
1839 		offsetof(struct prctl_mm_map, start_stack),
1840 		offsetof(struct prctl_mm_map, arg_start),
1841 		offsetof(struct prctl_mm_map, arg_end),
1842 		offsetof(struct prctl_mm_map, env_start),
1843 		offsetof(struct prctl_mm_map, env_end),
1844 	};
1845 
1846 	/*
1847 	 * Make sure the members are not somewhere outside
1848 	 * of allowed address space.
1849 	 */
1850 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1851 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1852 
1853 		if ((unsigned long)val >= mmap_max_addr ||
1854 		    (unsigned long)val < mmap_min_addr)
1855 			goto out;
1856 	}
1857 
1858 	/*
1859 	 * Make sure the pairs are ordered.
1860 	 */
1861 #define __prctl_check_order(__m1, __op, __m2)				\
1862 	((unsigned long)prctl_map->__m1 __op				\
1863 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1864 	error  = __prctl_check_order(start_code, <, end_code);
1865 	error |= __prctl_check_order(start_data, <, end_data);
1866 	error |= __prctl_check_order(start_brk, <=, brk);
1867 	error |= __prctl_check_order(arg_start, <=, arg_end);
1868 	error |= __prctl_check_order(env_start, <=, env_end);
1869 	if (error)
1870 		goto out;
1871 #undef __prctl_check_order
1872 
1873 	error = -EINVAL;
1874 
1875 	/*
1876 	 * @brk should be after @end_data in traditional maps.
1877 	 */
1878 	if (prctl_map->start_brk <= prctl_map->end_data ||
1879 	    prctl_map->brk <= prctl_map->end_data)
1880 		goto out;
1881 
1882 	/*
1883 	 * Neither we should allow to override limits if they set.
1884 	 */
1885 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1886 			      prctl_map->start_brk, prctl_map->end_data,
1887 			      prctl_map->start_data))
1888 			goto out;
1889 
1890 	/*
1891 	 * Someone is trying to cheat the auxv vector.
1892 	 */
1893 	if (prctl_map->auxv_size) {
1894 		if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1895 			goto out;
1896 	}
1897 
1898 	/*
1899 	 * Finally, make sure the caller has the rights to
1900 	 * change /proc/pid/exe link: only local sys admin should
1901 	 * be allowed to.
1902 	 */
1903 	if (prctl_map->exe_fd != (u32)-1) {
1904 		if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1905 			goto out;
1906 	}
1907 
1908 	error = 0;
1909 out:
1910 	return error;
1911 }
1912 
1913 #ifdef CONFIG_CHECKPOINT_RESTORE
1914 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1915 {
1916 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1917 	unsigned long user_auxv[AT_VECTOR_SIZE];
1918 	struct mm_struct *mm = current->mm;
1919 	int error;
1920 
1921 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1922 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1923 
1924 	if (opt == PR_SET_MM_MAP_SIZE)
1925 		return put_user((unsigned int)sizeof(prctl_map),
1926 				(unsigned int __user *)addr);
1927 
1928 	if (data_size != sizeof(prctl_map))
1929 		return -EINVAL;
1930 
1931 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1932 		return -EFAULT;
1933 
1934 	error = validate_prctl_map(&prctl_map);
1935 	if (error)
1936 		return error;
1937 
1938 	if (prctl_map.auxv_size) {
1939 		memset(user_auxv, 0, sizeof(user_auxv));
1940 		if (copy_from_user(user_auxv,
1941 				   (const void __user *)prctl_map.auxv,
1942 				   prctl_map.auxv_size))
1943 			return -EFAULT;
1944 
1945 		/* Last entry must be AT_NULL as specification requires */
1946 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1947 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1948 	}
1949 
1950 	if (prctl_map.exe_fd != (u32)-1) {
1951 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1952 		if (error)
1953 			return error;
1954 	}
1955 
1956 	down_write(&mm->mmap_sem);
1957 
1958 	/*
1959 	 * We don't validate if these members are pointing to
1960 	 * real present VMAs because application may have correspond
1961 	 * VMAs already unmapped and kernel uses these members for statistics
1962 	 * output in procfs mostly, except
1963 	 *
1964 	 *  - @start_brk/@brk which are used in do_brk but kernel lookups
1965 	 *    for VMAs when updating these memvers so anything wrong written
1966 	 *    here cause kernel to swear at userspace program but won't lead
1967 	 *    to any problem in kernel itself
1968 	 */
1969 
1970 	mm->start_code	= prctl_map.start_code;
1971 	mm->end_code	= prctl_map.end_code;
1972 	mm->start_data	= prctl_map.start_data;
1973 	mm->end_data	= prctl_map.end_data;
1974 	mm->start_brk	= prctl_map.start_brk;
1975 	mm->brk		= prctl_map.brk;
1976 	mm->start_stack	= prctl_map.start_stack;
1977 	mm->arg_start	= prctl_map.arg_start;
1978 	mm->arg_end	= prctl_map.arg_end;
1979 	mm->env_start	= prctl_map.env_start;
1980 	mm->env_end	= prctl_map.env_end;
1981 
1982 	/*
1983 	 * Note this update of @saved_auxv is lockless thus
1984 	 * if someone reads this member in procfs while we're
1985 	 * updating -- it may get partly updated results. It's
1986 	 * known and acceptable trade off: we leave it as is to
1987 	 * not introduce additional locks here making the kernel
1988 	 * more complex.
1989 	 */
1990 	if (prctl_map.auxv_size)
1991 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1992 
1993 	up_write(&mm->mmap_sem);
1994 	return 0;
1995 }
1996 #endif /* CONFIG_CHECKPOINT_RESTORE */
1997 
1998 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
1999 			  unsigned long len)
2000 {
2001 	/*
2002 	 * This doesn't move the auxiliary vector itself since it's pinned to
2003 	 * mm_struct, but it permits filling the vector with new values.  It's
2004 	 * up to the caller to provide sane values here, otherwise userspace
2005 	 * tools which use this vector might be unhappy.
2006 	 */
2007 	unsigned long user_auxv[AT_VECTOR_SIZE];
2008 
2009 	if (len > sizeof(user_auxv))
2010 		return -EINVAL;
2011 
2012 	if (copy_from_user(user_auxv, (const void __user *)addr, len))
2013 		return -EFAULT;
2014 
2015 	/* Make sure the last entry is always AT_NULL */
2016 	user_auxv[AT_VECTOR_SIZE - 2] = 0;
2017 	user_auxv[AT_VECTOR_SIZE - 1] = 0;
2018 
2019 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2020 
2021 	task_lock(current);
2022 	memcpy(mm->saved_auxv, user_auxv, len);
2023 	task_unlock(current);
2024 
2025 	return 0;
2026 }
2027 
2028 static int prctl_set_mm(int opt, unsigned long addr,
2029 			unsigned long arg4, unsigned long arg5)
2030 {
2031 	struct mm_struct *mm = current->mm;
2032 	struct prctl_mm_map prctl_map;
2033 	struct vm_area_struct *vma;
2034 	int error;
2035 
2036 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2037 			      opt != PR_SET_MM_MAP &&
2038 			      opt != PR_SET_MM_MAP_SIZE)))
2039 		return -EINVAL;
2040 
2041 #ifdef CONFIG_CHECKPOINT_RESTORE
2042 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2043 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2044 #endif
2045 
2046 	if (!capable(CAP_SYS_RESOURCE))
2047 		return -EPERM;
2048 
2049 	if (opt == PR_SET_MM_EXE_FILE)
2050 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2051 
2052 	if (opt == PR_SET_MM_AUXV)
2053 		return prctl_set_auxv(mm, addr, arg4);
2054 
2055 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
2056 		return -EINVAL;
2057 
2058 	error = -EINVAL;
2059 
2060 	down_write(&mm->mmap_sem);
2061 	vma = find_vma(mm, addr);
2062 
2063 	prctl_map.start_code	= mm->start_code;
2064 	prctl_map.end_code	= mm->end_code;
2065 	prctl_map.start_data	= mm->start_data;
2066 	prctl_map.end_data	= mm->end_data;
2067 	prctl_map.start_brk	= mm->start_brk;
2068 	prctl_map.brk		= mm->brk;
2069 	prctl_map.start_stack	= mm->start_stack;
2070 	prctl_map.arg_start	= mm->arg_start;
2071 	prctl_map.arg_end	= mm->arg_end;
2072 	prctl_map.env_start	= mm->env_start;
2073 	prctl_map.env_end	= mm->env_end;
2074 	prctl_map.auxv		= NULL;
2075 	prctl_map.auxv_size	= 0;
2076 	prctl_map.exe_fd	= -1;
2077 
2078 	switch (opt) {
2079 	case PR_SET_MM_START_CODE:
2080 		prctl_map.start_code = addr;
2081 		break;
2082 	case PR_SET_MM_END_CODE:
2083 		prctl_map.end_code = addr;
2084 		break;
2085 	case PR_SET_MM_START_DATA:
2086 		prctl_map.start_data = addr;
2087 		break;
2088 	case PR_SET_MM_END_DATA:
2089 		prctl_map.end_data = addr;
2090 		break;
2091 	case PR_SET_MM_START_STACK:
2092 		prctl_map.start_stack = addr;
2093 		break;
2094 	case PR_SET_MM_START_BRK:
2095 		prctl_map.start_brk = addr;
2096 		break;
2097 	case PR_SET_MM_BRK:
2098 		prctl_map.brk = addr;
2099 		break;
2100 	case PR_SET_MM_ARG_START:
2101 		prctl_map.arg_start = addr;
2102 		break;
2103 	case PR_SET_MM_ARG_END:
2104 		prctl_map.arg_end = addr;
2105 		break;
2106 	case PR_SET_MM_ENV_START:
2107 		prctl_map.env_start = addr;
2108 		break;
2109 	case PR_SET_MM_ENV_END:
2110 		prctl_map.env_end = addr;
2111 		break;
2112 	default:
2113 		goto out;
2114 	}
2115 
2116 	error = validate_prctl_map(&prctl_map);
2117 	if (error)
2118 		goto out;
2119 
2120 	switch (opt) {
2121 	/*
2122 	 * If command line arguments and environment
2123 	 * are placed somewhere else on stack, we can
2124 	 * set them up here, ARG_START/END to setup
2125 	 * command line argumets and ENV_START/END
2126 	 * for environment.
2127 	 */
2128 	case PR_SET_MM_START_STACK:
2129 	case PR_SET_MM_ARG_START:
2130 	case PR_SET_MM_ARG_END:
2131 	case PR_SET_MM_ENV_START:
2132 	case PR_SET_MM_ENV_END:
2133 		if (!vma) {
2134 			error = -EFAULT;
2135 			goto out;
2136 		}
2137 	}
2138 
2139 	mm->start_code	= prctl_map.start_code;
2140 	mm->end_code	= prctl_map.end_code;
2141 	mm->start_data	= prctl_map.start_data;
2142 	mm->end_data	= prctl_map.end_data;
2143 	mm->start_brk	= prctl_map.start_brk;
2144 	mm->brk		= prctl_map.brk;
2145 	mm->start_stack	= prctl_map.start_stack;
2146 	mm->arg_start	= prctl_map.arg_start;
2147 	mm->arg_end	= prctl_map.arg_end;
2148 	mm->env_start	= prctl_map.env_start;
2149 	mm->env_end	= prctl_map.env_end;
2150 
2151 	error = 0;
2152 out:
2153 	up_write(&mm->mmap_sem);
2154 	return error;
2155 }
2156 
2157 #ifdef CONFIG_CHECKPOINT_RESTORE
2158 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2159 {
2160 	return put_user(me->clear_child_tid, tid_addr);
2161 }
2162 #else
2163 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2164 {
2165 	return -EINVAL;
2166 }
2167 #endif
2168 
2169 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2170 {
2171 	/*
2172 	 * If task has has_child_subreaper - all its decendants
2173 	 * already have these flag too and new decendants will
2174 	 * inherit it on fork, skip them.
2175 	 *
2176 	 * If we've found child_reaper - skip descendants in
2177 	 * it's subtree as they will never get out pidns.
2178 	 */
2179 	if (p->signal->has_child_subreaper ||
2180 	    is_child_reaper(task_pid(p)))
2181 		return 0;
2182 
2183 	p->signal->has_child_subreaper = 1;
2184 	return 1;
2185 }
2186 
2187 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2188 		unsigned long, arg4, unsigned long, arg5)
2189 {
2190 	struct task_struct *me = current;
2191 	unsigned char comm[sizeof(me->comm)];
2192 	long error;
2193 
2194 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2195 	if (error != -ENOSYS)
2196 		return error;
2197 
2198 	error = 0;
2199 	switch (option) {
2200 	case PR_SET_PDEATHSIG:
2201 		if (!valid_signal(arg2)) {
2202 			error = -EINVAL;
2203 			break;
2204 		}
2205 		me->pdeath_signal = arg2;
2206 		break;
2207 	case PR_GET_PDEATHSIG:
2208 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2209 		break;
2210 	case PR_GET_DUMPABLE:
2211 		error = get_dumpable(me->mm);
2212 		break;
2213 	case PR_SET_DUMPABLE:
2214 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2215 			error = -EINVAL;
2216 			break;
2217 		}
2218 		set_dumpable(me->mm, arg2);
2219 		break;
2220 
2221 	case PR_SET_UNALIGN:
2222 		error = SET_UNALIGN_CTL(me, arg2);
2223 		break;
2224 	case PR_GET_UNALIGN:
2225 		error = GET_UNALIGN_CTL(me, arg2);
2226 		break;
2227 	case PR_SET_FPEMU:
2228 		error = SET_FPEMU_CTL(me, arg2);
2229 		break;
2230 	case PR_GET_FPEMU:
2231 		error = GET_FPEMU_CTL(me, arg2);
2232 		break;
2233 	case PR_SET_FPEXC:
2234 		error = SET_FPEXC_CTL(me, arg2);
2235 		break;
2236 	case PR_GET_FPEXC:
2237 		error = GET_FPEXC_CTL(me, arg2);
2238 		break;
2239 	case PR_GET_TIMING:
2240 		error = PR_TIMING_STATISTICAL;
2241 		break;
2242 	case PR_SET_TIMING:
2243 		if (arg2 != PR_TIMING_STATISTICAL)
2244 			error = -EINVAL;
2245 		break;
2246 	case PR_SET_NAME:
2247 		comm[sizeof(me->comm) - 1] = 0;
2248 		if (strncpy_from_user(comm, (char __user *)arg2,
2249 				      sizeof(me->comm) - 1) < 0)
2250 			return -EFAULT;
2251 		set_task_comm(me, comm);
2252 		proc_comm_connector(me);
2253 		break;
2254 	case PR_GET_NAME:
2255 		get_task_comm(comm, me);
2256 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2257 			return -EFAULT;
2258 		break;
2259 	case PR_GET_ENDIAN:
2260 		error = GET_ENDIAN(me, arg2);
2261 		break;
2262 	case PR_SET_ENDIAN:
2263 		error = SET_ENDIAN(me, arg2);
2264 		break;
2265 	case PR_GET_SECCOMP:
2266 		error = prctl_get_seccomp();
2267 		break;
2268 	case PR_SET_SECCOMP:
2269 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2270 		break;
2271 	case PR_GET_TSC:
2272 		error = GET_TSC_CTL(arg2);
2273 		break;
2274 	case PR_SET_TSC:
2275 		error = SET_TSC_CTL(arg2);
2276 		break;
2277 	case PR_TASK_PERF_EVENTS_DISABLE:
2278 		error = perf_event_task_disable();
2279 		break;
2280 	case PR_TASK_PERF_EVENTS_ENABLE:
2281 		error = perf_event_task_enable();
2282 		break;
2283 	case PR_GET_TIMERSLACK:
2284 		if (current->timer_slack_ns > ULONG_MAX)
2285 			error = ULONG_MAX;
2286 		else
2287 			error = current->timer_slack_ns;
2288 		break;
2289 	case PR_SET_TIMERSLACK:
2290 		if (arg2 <= 0)
2291 			current->timer_slack_ns =
2292 					current->default_timer_slack_ns;
2293 		else
2294 			current->timer_slack_ns = arg2;
2295 		break;
2296 	case PR_MCE_KILL:
2297 		if (arg4 | arg5)
2298 			return -EINVAL;
2299 		switch (arg2) {
2300 		case PR_MCE_KILL_CLEAR:
2301 			if (arg3 != 0)
2302 				return -EINVAL;
2303 			current->flags &= ~PF_MCE_PROCESS;
2304 			break;
2305 		case PR_MCE_KILL_SET:
2306 			current->flags |= PF_MCE_PROCESS;
2307 			if (arg3 == PR_MCE_KILL_EARLY)
2308 				current->flags |= PF_MCE_EARLY;
2309 			else if (arg3 == PR_MCE_KILL_LATE)
2310 				current->flags &= ~PF_MCE_EARLY;
2311 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2312 				current->flags &=
2313 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2314 			else
2315 				return -EINVAL;
2316 			break;
2317 		default:
2318 			return -EINVAL;
2319 		}
2320 		break;
2321 	case PR_MCE_KILL_GET:
2322 		if (arg2 | arg3 | arg4 | arg5)
2323 			return -EINVAL;
2324 		if (current->flags & PF_MCE_PROCESS)
2325 			error = (current->flags & PF_MCE_EARLY) ?
2326 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2327 		else
2328 			error = PR_MCE_KILL_DEFAULT;
2329 		break;
2330 	case PR_SET_MM:
2331 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2332 		break;
2333 	case PR_GET_TID_ADDRESS:
2334 		error = prctl_get_tid_address(me, (int __user **)arg2);
2335 		break;
2336 	case PR_SET_CHILD_SUBREAPER:
2337 		me->signal->is_child_subreaper = !!arg2;
2338 		if (!arg2)
2339 			break;
2340 
2341 		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2342 		break;
2343 	case PR_GET_CHILD_SUBREAPER:
2344 		error = put_user(me->signal->is_child_subreaper,
2345 				 (int __user *)arg2);
2346 		break;
2347 	case PR_SET_NO_NEW_PRIVS:
2348 		if (arg2 != 1 || arg3 || arg4 || arg5)
2349 			return -EINVAL;
2350 
2351 		task_set_no_new_privs(current);
2352 		break;
2353 	case PR_GET_NO_NEW_PRIVS:
2354 		if (arg2 || arg3 || arg4 || arg5)
2355 			return -EINVAL;
2356 		return task_no_new_privs(current) ? 1 : 0;
2357 	case PR_GET_THP_DISABLE:
2358 		if (arg2 || arg3 || arg4 || arg5)
2359 			return -EINVAL;
2360 		error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2361 		break;
2362 	case PR_SET_THP_DISABLE:
2363 		if (arg3 || arg4 || arg5)
2364 			return -EINVAL;
2365 		if (down_write_killable(&me->mm->mmap_sem))
2366 			return -EINTR;
2367 		if (arg2)
2368 			set_bit(MMF_DISABLE_THP, &me->mm->flags);
2369 		else
2370 			clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2371 		up_write(&me->mm->mmap_sem);
2372 		break;
2373 	case PR_MPX_ENABLE_MANAGEMENT:
2374 		if (arg2 || arg3 || arg4 || arg5)
2375 			return -EINVAL;
2376 		error = MPX_ENABLE_MANAGEMENT();
2377 		break;
2378 	case PR_MPX_DISABLE_MANAGEMENT:
2379 		if (arg2 || arg3 || arg4 || arg5)
2380 			return -EINVAL;
2381 		error = MPX_DISABLE_MANAGEMENT();
2382 		break;
2383 	case PR_SET_FP_MODE:
2384 		error = SET_FP_MODE(me, arg2);
2385 		break;
2386 	case PR_GET_FP_MODE:
2387 		error = GET_FP_MODE(me);
2388 		break;
2389 	default:
2390 		error = -EINVAL;
2391 		break;
2392 	}
2393 	return error;
2394 }
2395 
2396 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2397 		struct getcpu_cache __user *, unused)
2398 {
2399 	int err = 0;
2400 	int cpu = raw_smp_processor_id();
2401 
2402 	if (cpup)
2403 		err |= put_user(cpu, cpup);
2404 	if (nodep)
2405 		err |= put_user(cpu_to_node(cpu), nodep);
2406 	return err ? -EFAULT : 0;
2407 }
2408 
2409 /**
2410  * do_sysinfo - fill in sysinfo struct
2411  * @info: pointer to buffer to fill
2412  */
2413 static int do_sysinfo(struct sysinfo *info)
2414 {
2415 	unsigned long mem_total, sav_total;
2416 	unsigned int mem_unit, bitcount;
2417 	struct timespec tp;
2418 
2419 	memset(info, 0, sizeof(struct sysinfo));
2420 
2421 	get_monotonic_boottime(&tp);
2422 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2423 
2424 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2425 
2426 	info->procs = nr_threads;
2427 
2428 	si_meminfo(info);
2429 	si_swapinfo(info);
2430 
2431 	/*
2432 	 * If the sum of all the available memory (i.e. ram + swap)
2433 	 * is less than can be stored in a 32 bit unsigned long then
2434 	 * we can be binary compatible with 2.2.x kernels.  If not,
2435 	 * well, in that case 2.2.x was broken anyways...
2436 	 *
2437 	 *  -Erik Andersen <andersee@debian.org>
2438 	 */
2439 
2440 	mem_total = info->totalram + info->totalswap;
2441 	if (mem_total < info->totalram || mem_total < info->totalswap)
2442 		goto out;
2443 	bitcount = 0;
2444 	mem_unit = info->mem_unit;
2445 	while (mem_unit > 1) {
2446 		bitcount++;
2447 		mem_unit >>= 1;
2448 		sav_total = mem_total;
2449 		mem_total <<= 1;
2450 		if (mem_total < sav_total)
2451 			goto out;
2452 	}
2453 
2454 	/*
2455 	 * If mem_total did not overflow, multiply all memory values by
2456 	 * info->mem_unit and set it to 1.  This leaves things compatible
2457 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2458 	 * kernels...
2459 	 */
2460 
2461 	info->mem_unit = 1;
2462 	info->totalram <<= bitcount;
2463 	info->freeram <<= bitcount;
2464 	info->sharedram <<= bitcount;
2465 	info->bufferram <<= bitcount;
2466 	info->totalswap <<= bitcount;
2467 	info->freeswap <<= bitcount;
2468 	info->totalhigh <<= bitcount;
2469 	info->freehigh <<= bitcount;
2470 
2471 out:
2472 	return 0;
2473 }
2474 
2475 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2476 {
2477 	struct sysinfo val;
2478 
2479 	do_sysinfo(&val);
2480 
2481 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2482 		return -EFAULT;
2483 
2484 	return 0;
2485 }
2486 
2487 #ifdef CONFIG_COMPAT
2488 struct compat_sysinfo {
2489 	s32 uptime;
2490 	u32 loads[3];
2491 	u32 totalram;
2492 	u32 freeram;
2493 	u32 sharedram;
2494 	u32 bufferram;
2495 	u32 totalswap;
2496 	u32 freeswap;
2497 	u16 procs;
2498 	u16 pad;
2499 	u32 totalhigh;
2500 	u32 freehigh;
2501 	u32 mem_unit;
2502 	char _f[20-2*sizeof(u32)-sizeof(int)];
2503 };
2504 
2505 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2506 {
2507 	struct sysinfo s;
2508 
2509 	do_sysinfo(&s);
2510 
2511 	/* Check to see if any memory value is too large for 32-bit and scale
2512 	 *  down if needed
2513 	 */
2514 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2515 		int bitcount = 0;
2516 
2517 		while (s.mem_unit < PAGE_SIZE) {
2518 			s.mem_unit <<= 1;
2519 			bitcount++;
2520 		}
2521 
2522 		s.totalram >>= bitcount;
2523 		s.freeram >>= bitcount;
2524 		s.sharedram >>= bitcount;
2525 		s.bufferram >>= bitcount;
2526 		s.totalswap >>= bitcount;
2527 		s.freeswap >>= bitcount;
2528 		s.totalhigh >>= bitcount;
2529 		s.freehigh >>= bitcount;
2530 	}
2531 
2532 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2533 	    __put_user(s.uptime, &info->uptime) ||
2534 	    __put_user(s.loads[0], &info->loads[0]) ||
2535 	    __put_user(s.loads[1], &info->loads[1]) ||
2536 	    __put_user(s.loads[2], &info->loads[2]) ||
2537 	    __put_user(s.totalram, &info->totalram) ||
2538 	    __put_user(s.freeram, &info->freeram) ||
2539 	    __put_user(s.sharedram, &info->sharedram) ||
2540 	    __put_user(s.bufferram, &info->bufferram) ||
2541 	    __put_user(s.totalswap, &info->totalswap) ||
2542 	    __put_user(s.freeswap, &info->freeswap) ||
2543 	    __put_user(s.procs, &info->procs) ||
2544 	    __put_user(s.totalhigh, &info->totalhigh) ||
2545 	    __put_user(s.freehigh, &info->freehigh) ||
2546 	    __put_user(s.mem_unit, &info->mem_unit))
2547 		return -EFAULT;
2548 
2549 	return 0;
2550 }
2551 #endif /* CONFIG_COMPAT */
2552