xref: /openbmc/linux/kernel/sys.c (revision aac5987a)
1 /*
2  *  linux/kernel/sys.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/export.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
14 #include <linux/fs.h>
15 #include <linux/kmod.h>
16 #include <linux/perf_event.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/workqueue.h>
20 #include <linux/capability.h>
21 #include <linux/device.h>
22 #include <linux/key.h>
23 #include <linux/times.h>
24 #include <linux/posix-timers.h>
25 #include <linux/security.h>
26 #include <linux/dcookies.h>
27 #include <linux/suspend.h>
28 #include <linux/tty.h>
29 #include <linux/signal.h>
30 #include <linux/cn_proc.h>
31 #include <linux/getcpu.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/seccomp.h>
34 #include <linux/cpu.h>
35 #include <linux/personality.h>
36 #include <linux/ptrace.h>
37 #include <linux/fs_struct.h>
38 #include <linux/file.h>
39 #include <linux/mount.h>
40 #include <linux/gfp.h>
41 #include <linux/syscore_ops.h>
42 #include <linux/version.h>
43 #include <linux/ctype.h>
44 
45 #include <linux/compat.h>
46 #include <linux/syscalls.h>
47 #include <linux/kprobes.h>
48 #include <linux/user_namespace.h>
49 #include <linux/binfmts.h>
50 
51 #include <linux/sched.h>
52 #include <linux/sched/autogroup.h>
53 #include <linux/sched/loadavg.h>
54 #include <linux/sched/stat.h>
55 #include <linux/sched/mm.h>
56 #include <linux/sched/coredump.h>
57 #include <linux/sched/task.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/rcupdate.h>
60 #include <linux/uidgid.h>
61 #include <linux/cred.h>
62 
63 #include <linux/kmsg_dump.h>
64 /* Move somewhere else to avoid recompiling? */
65 #include <generated/utsrelease.h>
66 
67 #include <linux/uaccess.h>
68 #include <asm/io.h>
69 #include <asm/unistd.h>
70 
71 #ifndef SET_UNALIGN_CTL
72 # define SET_UNALIGN_CTL(a, b)	(-EINVAL)
73 #endif
74 #ifndef GET_UNALIGN_CTL
75 # define GET_UNALIGN_CTL(a, b)	(-EINVAL)
76 #endif
77 #ifndef SET_FPEMU_CTL
78 # define SET_FPEMU_CTL(a, b)	(-EINVAL)
79 #endif
80 #ifndef GET_FPEMU_CTL
81 # define GET_FPEMU_CTL(a, b)	(-EINVAL)
82 #endif
83 #ifndef SET_FPEXC_CTL
84 # define SET_FPEXC_CTL(a, b)	(-EINVAL)
85 #endif
86 #ifndef GET_FPEXC_CTL
87 # define GET_FPEXC_CTL(a, b)	(-EINVAL)
88 #endif
89 #ifndef GET_ENDIAN
90 # define GET_ENDIAN(a, b)	(-EINVAL)
91 #endif
92 #ifndef SET_ENDIAN
93 # define SET_ENDIAN(a, b)	(-EINVAL)
94 #endif
95 #ifndef GET_TSC_CTL
96 # define GET_TSC_CTL(a)		(-EINVAL)
97 #endif
98 #ifndef SET_TSC_CTL
99 # define SET_TSC_CTL(a)		(-EINVAL)
100 #endif
101 #ifndef MPX_ENABLE_MANAGEMENT
102 # define MPX_ENABLE_MANAGEMENT()	(-EINVAL)
103 #endif
104 #ifndef MPX_DISABLE_MANAGEMENT
105 # define MPX_DISABLE_MANAGEMENT()	(-EINVAL)
106 #endif
107 #ifndef GET_FP_MODE
108 # define GET_FP_MODE(a)		(-EINVAL)
109 #endif
110 #ifndef SET_FP_MODE
111 # define SET_FP_MODE(a,b)	(-EINVAL)
112 #endif
113 
114 /*
115  * this is where the system-wide overflow UID and GID are defined, for
116  * architectures that now have 32-bit UID/GID but didn't in the past
117  */
118 
119 int overflowuid = DEFAULT_OVERFLOWUID;
120 int overflowgid = DEFAULT_OVERFLOWGID;
121 
122 EXPORT_SYMBOL(overflowuid);
123 EXPORT_SYMBOL(overflowgid);
124 
125 /*
126  * the same as above, but for filesystems which can only store a 16-bit
127  * UID and GID. as such, this is needed on all architectures
128  */
129 
130 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
131 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
132 
133 EXPORT_SYMBOL(fs_overflowuid);
134 EXPORT_SYMBOL(fs_overflowgid);
135 
136 /*
137  * Returns true if current's euid is same as p's uid or euid,
138  * or has CAP_SYS_NICE to p's user_ns.
139  *
140  * Called with rcu_read_lock, creds are safe
141  */
142 static bool set_one_prio_perm(struct task_struct *p)
143 {
144 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
145 
146 	if (uid_eq(pcred->uid,  cred->euid) ||
147 	    uid_eq(pcred->euid, cred->euid))
148 		return true;
149 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
150 		return true;
151 	return false;
152 }
153 
154 /*
155  * set the priority of a task
156  * - the caller must hold the RCU read lock
157  */
158 static int set_one_prio(struct task_struct *p, int niceval, int error)
159 {
160 	int no_nice;
161 
162 	if (!set_one_prio_perm(p)) {
163 		error = -EPERM;
164 		goto out;
165 	}
166 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
167 		error = -EACCES;
168 		goto out;
169 	}
170 	no_nice = security_task_setnice(p, niceval);
171 	if (no_nice) {
172 		error = no_nice;
173 		goto out;
174 	}
175 	if (error == -ESRCH)
176 		error = 0;
177 	set_user_nice(p, niceval);
178 out:
179 	return error;
180 }
181 
182 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
183 {
184 	struct task_struct *g, *p;
185 	struct user_struct *user;
186 	const struct cred *cred = current_cred();
187 	int error = -EINVAL;
188 	struct pid *pgrp;
189 	kuid_t uid;
190 
191 	if (which > PRIO_USER || which < PRIO_PROCESS)
192 		goto out;
193 
194 	/* normalize: avoid signed division (rounding problems) */
195 	error = -ESRCH;
196 	if (niceval < MIN_NICE)
197 		niceval = MIN_NICE;
198 	if (niceval > MAX_NICE)
199 		niceval = MAX_NICE;
200 
201 	rcu_read_lock();
202 	read_lock(&tasklist_lock);
203 	switch (which) {
204 	case PRIO_PROCESS:
205 		if (who)
206 			p = find_task_by_vpid(who);
207 		else
208 			p = current;
209 		if (p)
210 			error = set_one_prio(p, niceval, error);
211 		break;
212 	case PRIO_PGRP:
213 		if (who)
214 			pgrp = find_vpid(who);
215 		else
216 			pgrp = task_pgrp(current);
217 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
218 			error = set_one_prio(p, niceval, error);
219 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
220 		break;
221 	case PRIO_USER:
222 		uid = make_kuid(cred->user_ns, who);
223 		user = cred->user;
224 		if (!who)
225 			uid = cred->uid;
226 		else if (!uid_eq(uid, cred->uid)) {
227 			user = find_user(uid);
228 			if (!user)
229 				goto out_unlock;	/* No processes for this user */
230 		}
231 		do_each_thread(g, p) {
232 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
233 				error = set_one_prio(p, niceval, error);
234 		} while_each_thread(g, p);
235 		if (!uid_eq(uid, cred->uid))
236 			free_uid(user);		/* For find_user() */
237 		break;
238 	}
239 out_unlock:
240 	read_unlock(&tasklist_lock);
241 	rcu_read_unlock();
242 out:
243 	return error;
244 }
245 
246 /*
247  * Ugh. To avoid negative return values, "getpriority()" will
248  * not return the normal nice-value, but a negated value that
249  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
250  * to stay compatible.
251  */
252 SYSCALL_DEFINE2(getpriority, int, which, int, who)
253 {
254 	struct task_struct *g, *p;
255 	struct user_struct *user;
256 	const struct cred *cred = current_cred();
257 	long niceval, retval = -ESRCH;
258 	struct pid *pgrp;
259 	kuid_t uid;
260 
261 	if (which > PRIO_USER || which < PRIO_PROCESS)
262 		return -EINVAL;
263 
264 	rcu_read_lock();
265 	read_lock(&tasklist_lock);
266 	switch (which) {
267 	case PRIO_PROCESS:
268 		if (who)
269 			p = find_task_by_vpid(who);
270 		else
271 			p = current;
272 		if (p) {
273 			niceval = nice_to_rlimit(task_nice(p));
274 			if (niceval > retval)
275 				retval = niceval;
276 		}
277 		break;
278 	case PRIO_PGRP:
279 		if (who)
280 			pgrp = find_vpid(who);
281 		else
282 			pgrp = task_pgrp(current);
283 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
284 			niceval = nice_to_rlimit(task_nice(p));
285 			if (niceval > retval)
286 				retval = niceval;
287 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
288 		break;
289 	case PRIO_USER:
290 		uid = make_kuid(cred->user_ns, who);
291 		user = cred->user;
292 		if (!who)
293 			uid = cred->uid;
294 		else if (!uid_eq(uid, cred->uid)) {
295 			user = find_user(uid);
296 			if (!user)
297 				goto out_unlock;	/* No processes for this user */
298 		}
299 		do_each_thread(g, p) {
300 			if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
301 				niceval = nice_to_rlimit(task_nice(p));
302 				if (niceval > retval)
303 					retval = niceval;
304 			}
305 		} while_each_thread(g, p);
306 		if (!uid_eq(uid, cred->uid))
307 			free_uid(user);		/* for find_user() */
308 		break;
309 	}
310 out_unlock:
311 	read_unlock(&tasklist_lock);
312 	rcu_read_unlock();
313 
314 	return retval;
315 }
316 
317 /*
318  * Unprivileged users may change the real gid to the effective gid
319  * or vice versa.  (BSD-style)
320  *
321  * If you set the real gid at all, or set the effective gid to a value not
322  * equal to the real gid, then the saved gid is set to the new effective gid.
323  *
324  * This makes it possible for a setgid program to completely drop its
325  * privileges, which is often a useful assertion to make when you are doing
326  * a security audit over a program.
327  *
328  * The general idea is that a program which uses just setregid() will be
329  * 100% compatible with BSD.  A program which uses just setgid() will be
330  * 100% compatible with POSIX with saved IDs.
331  *
332  * SMP: There are not races, the GIDs are checked only by filesystem
333  *      operations (as far as semantic preservation is concerned).
334  */
335 #ifdef CONFIG_MULTIUSER
336 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
337 {
338 	struct user_namespace *ns = current_user_ns();
339 	const struct cred *old;
340 	struct cred *new;
341 	int retval;
342 	kgid_t krgid, kegid;
343 
344 	krgid = make_kgid(ns, rgid);
345 	kegid = make_kgid(ns, egid);
346 
347 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
348 		return -EINVAL;
349 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
350 		return -EINVAL;
351 
352 	new = prepare_creds();
353 	if (!new)
354 		return -ENOMEM;
355 	old = current_cred();
356 
357 	retval = -EPERM;
358 	if (rgid != (gid_t) -1) {
359 		if (gid_eq(old->gid, krgid) ||
360 		    gid_eq(old->egid, krgid) ||
361 		    ns_capable(old->user_ns, CAP_SETGID))
362 			new->gid = krgid;
363 		else
364 			goto error;
365 	}
366 	if (egid != (gid_t) -1) {
367 		if (gid_eq(old->gid, kegid) ||
368 		    gid_eq(old->egid, kegid) ||
369 		    gid_eq(old->sgid, kegid) ||
370 		    ns_capable(old->user_ns, CAP_SETGID))
371 			new->egid = kegid;
372 		else
373 			goto error;
374 	}
375 
376 	if (rgid != (gid_t) -1 ||
377 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
378 		new->sgid = new->egid;
379 	new->fsgid = new->egid;
380 
381 	return commit_creds(new);
382 
383 error:
384 	abort_creds(new);
385 	return retval;
386 }
387 
388 /*
389  * setgid() is implemented like SysV w/ SAVED_IDS
390  *
391  * SMP: Same implicit races as above.
392  */
393 SYSCALL_DEFINE1(setgid, gid_t, gid)
394 {
395 	struct user_namespace *ns = current_user_ns();
396 	const struct cred *old;
397 	struct cred *new;
398 	int retval;
399 	kgid_t kgid;
400 
401 	kgid = make_kgid(ns, gid);
402 	if (!gid_valid(kgid))
403 		return -EINVAL;
404 
405 	new = prepare_creds();
406 	if (!new)
407 		return -ENOMEM;
408 	old = current_cred();
409 
410 	retval = -EPERM;
411 	if (ns_capable(old->user_ns, CAP_SETGID))
412 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
413 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
414 		new->egid = new->fsgid = kgid;
415 	else
416 		goto error;
417 
418 	return commit_creds(new);
419 
420 error:
421 	abort_creds(new);
422 	return retval;
423 }
424 
425 /*
426  * change the user struct in a credentials set to match the new UID
427  */
428 static int set_user(struct cred *new)
429 {
430 	struct user_struct *new_user;
431 
432 	new_user = alloc_uid(new->uid);
433 	if (!new_user)
434 		return -EAGAIN;
435 
436 	/*
437 	 * We don't fail in case of NPROC limit excess here because too many
438 	 * poorly written programs don't check set*uid() return code, assuming
439 	 * it never fails if called by root.  We may still enforce NPROC limit
440 	 * for programs doing set*uid()+execve() by harmlessly deferring the
441 	 * failure to the execve() stage.
442 	 */
443 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
444 			new_user != INIT_USER)
445 		current->flags |= PF_NPROC_EXCEEDED;
446 	else
447 		current->flags &= ~PF_NPROC_EXCEEDED;
448 
449 	free_uid(new->user);
450 	new->user = new_user;
451 	return 0;
452 }
453 
454 /*
455  * Unprivileged users may change the real uid to the effective uid
456  * or vice versa.  (BSD-style)
457  *
458  * If you set the real uid at all, or set the effective uid to a value not
459  * equal to the real uid, then the saved uid is set to the new effective uid.
460  *
461  * This makes it possible for a setuid program to completely drop its
462  * privileges, which is often a useful assertion to make when you are doing
463  * a security audit over a program.
464  *
465  * The general idea is that a program which uses just setreuid() will be
466  * 100% compatible with BSD.  A program which uses just setuid() will be
467  * 100% compatible with POSIX with saved IDs.
468  */
469 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
470 {
471 	struct user_namespace *ns = current_user_ns();
472 	const struct cred *old;
473 	struct cred *new;
474 	int retval;
475 	kuid_t kruid, keuid;
476 
477 	kruid = make_kuid(ns, ruid);
478 	keuid = make_kuid(ns, euid);
479 
480 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
481 		return -EINVAL;
482 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
483 		return -EINVAL;
484 
485 	new = prepare_creds();
486 	if (!new)
487 		return -ENOMEM;
488 	old = current_cred();
489 
490 	retval = -EPERM;
491 	if (ruid != (uid_t) -1) {
492 		new->uid = kruid;
493 		if (!uid_eq(old->uid, kruid) &&
494 		    !uid_eq(old->euid, kruid) &&
495 		    !ns_capable(old->user_ns, CAP_SETUID))
496 			goto error;
497 	}
498 
499 	if (euid != (uid_t) -1) {
500 		new->euid = keuid;
501 		if (!uid_eq(old->uid, keuid) &&
502 		    !uid_eq(old->euid, keuid) &&
503 		    !uid_eq(old->suid, keuid) &&
504 		    !ns_capable(old->user_ns, CAP_SETUID))
505 			goto error;
506 	}
507 
508 	if (!uid_eq(new->uid, old->uid)) {
509 		retval = set_user(new);
510 		if (retval < 0)
511 			goto error;
512 	}
513 	if (ruid != (uid_t) -1 ||
514 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
515 		new->suid = new->euid;
516 	new->fsuid = new->euid;
517 
518 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
519 	if (retval < 0)
520 		goto error;
521 
522 	return commit_creds(new);
523 
524 error:
525 	abort_creds(new);
526 	return retval;
527 }
528 
529 /*
530  * setuid() is implemented like SysV with SAVED_IDS
531  *
532  * Note that SAVED_ID's is deficient in that a setuid root program
533  * like sendmail, for example, cannot set its uid to be a normal
534  * user and then switch back, because if you're root, setuid() sets
535  * the saved uid too.  If you don't like this, blame the bright people
536  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
537  * will allow a root program to temporarily drop privileges and be able to
538  * regain them by swapping the real and effective uid.
539  */
540 SYSCALL_DEFINE1(setuid, uid_t, uid)
541 {
542 	struct user_namespace *ns = current_user_ns();
543 	const struct cred *old;
544 	struct cred *new;
545 	int retval;
546 	kuid_t kuid;
547 
548 	kuid = make_kuid(ns, uid);
549 	if (!uid_valid(kuid))
550 		return -EINVAL;
551 
552 	new = prepare_creds();
553 	if (!new)
554 		return -ENOMEM;
555 	old = current_cred();
556 
557 	retval = -EPERM;
558 	if (ns_capable(old->user_ns, CAP_SETUID)) {
559 		new->suid = new->uid = kuid;
560 		if (!uid_eq(kuid, old->uid)) {
561 			retval = set_user(new);
562 			if (retval < 0)
563 				goto error;
564 		}
565 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
566 		goto error;
567 	}
568 
569 	new->fsuid = new->euid = kuid;
570 
571 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
572 	if (retval < 0)
573 		goto error;
574 
575 	return commit_creds(new);
576 
577 error:
578 	abort_creds(new);
579 	return retval;
580 }
581 
582 
583 /*
584  * This function implements a generic ability to update ruid, euid,
585  * and suid.  This allows you to implement the 4.4 compatible seteuid().
586  */
587 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
588 {
589 	struct user_namespace *ns = current_user_ns();
590 	const struct cred *old;
591 	struct cred *new;
592 	int retval;
593 	kuid_t kruid, keuid, ksuid;
594 
595 	kruid = make_kuid(ns, ruid);
596 	keuid = make_kuid(ns, euid);
597 	ksuid = make_kuid(ns, suid);
598 
599 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
600 		return -EINVAL;
601 
602 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
603 		return -EINVAL;
604 
605 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
606 		return -EINVAL;
607 
608 	new = prepare_creds();
609 	if (!new)
610 		return -ENOMEM;
611 
612 	old = current_cred();
613 
614 	retval = -EPERM;
615 	if (!ns_capable(old->user_ns, CAP_SETUID)) {
616 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
617 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
618 			goto error;
619 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
620 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
621 			goto error;
622 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
623 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
624 			goto error;
625 	}
626 
627 	if (ruid != (uid_t) -1) {
628 		new->uid = kruid;
629 		if (!uid_eq(kruid, old->uid)) {
630 			retval = set_user(new);
631 			if (retval < 0)
632 				goto error;
633 		}
634 	}
635 	if (euid != (uid_t) -1)
636 		new->euid = keuid;
637 	if (suid != (uid_t) -1)
638 		new->suid = ksuid;
639 	new->fsuid = new->euid;
640 
641 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
642 	if (retval < 0)
643 		goto error;
644 
645 	return commit_creds(new);
646 
647 error:
648 	abort_creds(new);
649 	return retval;
650 }
651 
652 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
653 {
654 	const struct cred *cred = current_cred();
655 	int retval;
656 	uid_t ruid, euid, suid;
657 
658 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
659 	euid = from_kuid_munged(cred->user_ns, cred->euid);
660 	suid = from_kuid_munged(cred->user_ns, cred->suid);
661 
662 	retval = put_user(ruid, ruidp);
663 	if (!retval) {
664 		retval = put_user(euid, euidp);
665 		if (!retval)
666 			return put_user(suid, suidp);
667 	}
668 	return retval;
669 }
670 
671 /*
672  * Same as above, but for rgid, egid, sgid.
673  */
674 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
675 {
676 	struct user_namespace *ns = current_user_ns();
677 	const struct cred *old;
678 	struct cred *new;
679 	int retval;
680 	kgid_t krgid, kegid, ksgid;
681 
682 	krgid = make_kgid(ns, rgid);
683 	kegid = make_kgid(ns, egid);
684 	ksgid = make_kgid(ns, sgid);
685 
686 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
687 		return -EINVAL;
688 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
689 		return -EINVAL;
690 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
691 		return -EINVAL;
692 
693 	new = prepare_creds();
694 	if (!new)
695 		return -ENOMEM;
696 	old = current_cred();
697 
698 	retval = -EPERM;
699 	if (!ns_capable(old->user_ns, CAP_SETGID)) {
700 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
701 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
702 			goto error;
703 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
704 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
705 			goto error;
706 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
707 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
708 			goto error;
709 	}
710 
711 	if (rgid != (gid_t) -1)
712 		new->gid = krgid;
713 	if (egid != (gid_t) -1)
714 		new->egid = kegid;
715 	if (sgid != (gid_t) -1)
716 		new->sgid = ksgid;
717 	new->fsgid = new->egid;
718 
719 	return commit_creds(new);
720 
721 error:
722 	abort_creds(new);
723 	return retval;
724 }
725 
726 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
727 {
728 	const struct cred *cred = current_cred();
729 	int retval;
730 	gid_t rgid, egid, sgid;
731 
732 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
733 	egid = from_kgid_munged(cred->user_ns, cred->egid);
734 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
735 
736 	retval = put_user(rgid, rgidp);
737 	if (!retval) {
738 		retval = put_user(egid, egidp);
739 		if (!retval)
740 			retval = put_user(sgid, sgidp);
741 	}
742 
743 	return retval;
744 }
745 
746 
747 /*
748  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
749  * is used for "access()" and for the NFS daemon (letting nfsd stay at
750  * whatever uid it wants to). It normally shadows "euid", except when
751  * explicitly set by setfsuid() or for access..
752  */
753 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
754 {
755 	const struct cred *old;
756 	struct cred *new;
757 	uid_t old_fsuid;
758 	kuid_t kuid;
759 
760 	old = current_cred();
761 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
762 
763 	kuid = make_kuid(old->user_ns, uid);
764 	if (!uid_valid(kuid))
765 		return old_fsuid;
766 
767 	new = prepare_creds();
768 	if (!new)
769 		return old_fsuid;
770 
771 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
772 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
773 	    ns_capable(old->user_ns, CAP_SETUID)) {
774 		if (!uid_eq(kuid, old->fsuid)) {
775 			new->fsuid = kuid;
776 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
777 				goto change_okay;
778 		}
779 	}
780 
781 	abort_creds(new);
782 	return old_fsuid;
783 
784 change_okay:
785 	commit_creds(new);
786 	return old_fsuid;
787 }
788 
789 /*
790  * Samma på svenska..
791  */
792 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
793 {
794 	const struct cred *old;
795 	struct cred *new;
796 	gid_t old_fsgid;
797 	kgid_t kgid;
798 
799 	old = current_cred();
800 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
801 
802 	kgid = make_kgid(old->user_ns, gid);
803 	if (!gid_valid(kgid))
804 		return old_fsgid;
805 
806 	new = prepare_creds();
807 	if (!new)
808 		return old_fsgid;
809 
810 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
811 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
812 	    ns_capable(old->user_ns, CAP_SETGID)) {
813 		if (!gid_eq(kgid, old->fsgid)) {
814 			new->fsgid = kgid;
815 			goto change_okay;
816 		}
817 	}
818 
819 	abort_creds(new);
820 	return old_fsgid;
821 
822 change_okay:
823 	commit_creds(new);
824 	return old_fsgid;
825 }
826 #endif /* CONFIG_MULTIUSER */
827 
828 /**
829  * sys_getpid - return the thread group id of the current process
830  *
831  * Note, despite the name, this returns the tgid not the pid.  The tgid and
832  * the pid are identical unless CLONE_THREAD was specified on clone() in
833  * which case the tgid is the same in all threads of the same group.
834  *
835  * This is SMP safe as current->tgid does not change.
836  */
837 SYSCALL_DEFINE0(getpid)
838 {
839 	return task_tgid_vnr(current);
840 }
841 
842 /* Thread ID - the internal kernel "pid" */
843 SYSCALL_DEFINE0(gettid)
844 {
845 	return task_pid_vnr(current);
846 }
847 
848 /*
849  * Accessing ->real_parent is not SMP-safe, it could
850  * change from under us. However, we can use a stale
851  * value of ->real_parent under rcu_read_lock(), see
852  * release_task()->call_rcu(delayed_put_task_struct).
853  */
854 SYSCALL_DEFINE0(getppid)
855 {
856 	int pid;
857 
858 	rcu_read_lock();
859 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
860 	rcu_read_unlock();
861 
862 	return pid;
863 }
864 
865 SYSCALL_DEFINE0(getuid)
866 {
867 	/* Only we change this so SMP safe */
868 	return from_kuid_munged(current_user_ns(), current_uid());
869 }
870 
871 SYSCALL_DEFINE0(geteuid)
872 {
873 	/* Only we change this so SMP safe */
874 	return from_kuid_munged(current_user_ns(), current_euid());
875 }
876 
877 SYSCALL_DEFINE0(getgid)
878 {
879 	/* Only we change this so SMP safe */
880 	return from_kgid_munged(current_user_ns(), current_gid());
881 }
882 
883 SYSCALL_DEFINE0(getegid)
884 {
885 	/* Only we change this so SMP safe */
886 	return from_kgid_munged(current_user_ns(), current_egid());
887 }
888 
889 void do_sys_times(struct tms *tms)
890 {
891 	u64 tgutime, tgstime, cutime, cstime;
892 
893 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
894 	cutime = current->signal->cutime;
895 	cstime = current->signal->cstime;
896 	tms->tms_utime = nsec_to_clock_t(tgutime);
897 	tms->tms_stime = nsec_to_clock_t(tgstime);
898 	tms->tms_cutime = nsec_to_clock_t(cutime);
899 	tms->tms_cstime = nsec_to_clock_t(cstime);
900 }
901 
902 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
903 {
904 	if (tbuf) {
905 		struct tms tmp;
906 
907 		do_sys_times(&tmp);
908 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
909 			return -EFAULT;
910 	}
911 	force_successful_syscall_return();
912 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
913 }
914 
915 /*
916  * This needs some heavy checking ...
917  * I just haven't the stomach for it. I also don't fully
918  * understand sessions/pgrp etc. Let somebody who does explain it.
919  *
920  * OK, I think I have the protection semantics right.... this is really
921  * only important on a multi-user system anyway, to make sure one user
922  * can't send a signal to a process owned by another.  -TYT, 12/12/91
923  *
924  * !PF_FORKNOEXEC check to conform completely to POSIX.
925  */
926 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
927 {
928 	struct task_struct *p;
929 	struct task_struct *group_leader = current->group_leader;
930 	struct pid *pgrp;
931 	int err;
932 
933 	if (!pid)
934 		pid = task_pid_vnr(group_leader);
935 	if (!pgid)
936 		pgid = pid;
937 	if (pgid < 0)
938 		return -EINVAL;
939 	rcu_read_lock();
940 
941 	/* From this point forward we keep holding onto the tasklist lock
942 	 * so that our parent does not change from under us. -DaveM
943 	 */
944 	write_lock_irq(&tasklist_lock);
945 
946 	err = -ESRCH;
947 	p = find_task_by_vpid(pid);
948 	if (!p)
949 		goto out;
950 
951 	err = -EINVAL;
952 	if (!thread_group_leader(p))
953 		goto out;
954 
955 	if (same_thread_group(p->real_parent, group_leader)) {
956 		err = -EPERM;
957 		if (task_session(p) != task_session(group_leader))
958 			goto out;
959 		err = -EACCES;
960 		if (!(p->flags & PF_FORKNOEXEC))
961 			goto out;
962 	} else {
963 		err = -ESRCH;
964 		if (p != group_leader)
965 			goto out;
966 	}
967 
968 	err = -EPERM;
969 	if (p->signal->leader)
970 		goto out;
971 
972 	pgrp = task_pid(p);
973 	if (pgid != pid) {
974 		struct task_struct *g;
975 
976 		pgrp = find_vpid(pgid);
977 		g = pid_task(pgrp, PIDTYPE_PGID);
978 		if (!g || task_session(g) != task_session(group_leader))
979 			goto out;
980 	}
981 
982 	err = security_task_setpgid(p, pgid);
983 	if (err)
984 		goto out;
985 
986 	if (task_pgrp(p) != pgrp)
987 		change_pid(p, PIDTYPE_PGID, pgrp);
988 
989 	err = 0;
990 out:
991 	/* All paths lead to here, thus we are safe. -DaveM */
992 	write_unlock_irq(&tasklist_lock);
993 	rcu_read_unlock();
994 	return err;
995 }
996 
997 SYSCALL_DEFINE1(getpgid, pid_t, pid)
998 {
999 	struct task_struct *p;
1000 	struct pid *grp;
1001 	int retval;
1002 
1003 	rcu_read_lock();
1004 	if (!pid)
1005 		grp = task_pgrp(current);
1006 	else {
1007 		retval = -ESRCH;
1008 		p = find_task_by_vpid(pid);
1009 		if (!p)
1010 			goto out;
1011 		grp = task_pgrp(p);
1012 		if (!grp)
1013 			goto out;
1014 
1015 		retval = security_task_getpgid(p);
1016 		if (retval)
1017 			goto out;
1018 	}
1019 	retval = pid_vnr(grp);
1020 out:
1021 	rcu_read_unlock();
1022 	return retval;
1023 }
1024 
1025 #ifdef __ARCH_WANT_SYS_GETPGRP
1026 
1027 SYSCALL_DEFINE0(getpgrp)
1028 {
1029 	return sys_getpgid(0);
1030 }
1031 
1032 #endif
1033 
1034 SYSCALL_DEFINE1(getsid, pid_t, pid)
1035 {
1036 	struct task_struct *p;
1037 	struct pid *sid;
1038 	int retval;
1039 
1040 	rcu_read_lock();
1041 	if (!pid)
1042 		sid = task_session(current);
1043 	else {
1044 		retval = -ESRCH;
1045 		p = find_task_by_vpid(pid);
1046 		if (!p)
1047 			goto out;
1048 		sid = task_session(p);
1049 		if (!sid)
1050 			goto out;
1051 
1052 		retval = security_task_getsid(p);
1053 		if (retval)
1054 			goto out;
1055 	}
1056 	retval = pid_vnr(sid);
1057 out:
1058 	rcu_read_unlock();
1059 	return retval;
1060 }
1061 
1062 static void set_special_pids(struct pid *pid)
1063 {
1064 	struct task_struct *curr = current->group_leader;
1065 
1066 	if (task_session(curr) != pid)
1067 		change_pid(curr, PIDTYPE_SID, pid);
1068 
1069 	if (task_pgrp(curr) != pid)
1070 		change_pid(curr, PIDTYPE_PGID, pid);
1071 }
1072 
1073 SYSCALL_DEFINE0(setsid)
1074 {
1075 	struct task_struct *group_leader = current->group_leader;
1076 	struct pid *sid = task_pid(group_leader);
1077 	pid_t session = pid_vnr(sid);
1078 	int err = -EPERM;
1079 
1080 	write_lock_irq(&tasklist_lock);
1081 	/* Fail if I am already a session leader */
1082 	if (group_leader->signal->leader)
1083 		goto out;
1084 
1085 	/* Fail if a process group id already exists that equals the
1086 	 * proposed session id.
1087 	 */
1088 	if (pid_task(sid, PIDTYPE_PGID))
1089 		goto out;
1090 
1091 	group_leader->signal->leader = 1;
1092 	set_special_pids(sid);
1093 
1094 	proc_clear_tty(group_leader);
1095 
1096 	err = session;
1097 out:
1098 	write_unlock_irq(&tasklist_lock);
1099 	if (err > 0) {
1100 		proc_sid_connector(group_leader);
1101 		sched_autogroup_create_attach(group_leader);
1102 	}
1103 	return err;
1104 }
1105 
1106 DECLARE_RWSEM(uts_sem);
1107 
1108 #ifdef COMPAT_UTS_MACHINE
1109 #define override_architecture(name) \
1110 	(personality(current->personality) == PER_LINUX32 && \
1111 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1112 		      sizeof(COMPAT_UTS_MACHINE)))
1113 #else
1114 #define override_architecture(name)	0
1115 #endif
1116 
1117 /*
1118  * Work around broken programs that cannot handle "Linux 3.0".
1119  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1120  * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1121  */
1122 static int override_release(char __user *release, size_t len)
1123 {
1124 	int ret = 0;
1125 
1126 	if (current->personality & UNAME26) {
1127 		const char *rest = UTS_RELEASE;
1128 		char buf[65] = { 0 };
1129 		int ndots = 0;
1130 		unsigned v;
1131 		size_t copy;
1132 
1133 		while (*rest) {
1134 			if (*rest == '.' && ++ndots >= 3)
1135 				break;
1136 			if (!isdigit(*rest) && *rest != '.')
1137 				break;
1138 			rest++;
1139 		}
1140 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1141 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1142 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1143 		ret = copy_to_user(release, buf, copy + 1);
1144 	}
1145 	return ret;
1146 }
1147 
1148 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1149 {
1150 	int errno = 0;
1151 
1152 	down_read(&uts_sem);
1153 	if (copy_to_user(name, utsname(), sizeof *name))
1154 		errno = -EFAULT;
1155 	up_read(&uts_sem);
1156 
1157 	if (!errno && override_release(name->release, sizeof(name->release)))
1158 		errno = -EFAULT;
1159 	if (!errno && override_architecture(name))
1160 		errno = -EFAULT;
1161 	return errno;
1162 }
1163 
1164 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1165 /*
1166  * Old cruft
1167  */
1168 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1169 {
1170 	int error = 0;
1171 
1172 	if (!name)
1173 		return -EFAULT;
1174 
1175 	down_read(&uts_sem);
1176 	if (copy_to_user(name, utsname(), sizeof(*name)))
1177 		error = -EFAULT;
1178 	up_read(&uts_sem);
1179 
1180 	if (!error && override_release(name->release, sizeof(name->release)))
1181 		error = -EFAULT;
1182 	if (!error && override_architecture(name))
1183 		error = -EFAULT;
1184 	return error;
1185 }
1186 
1187 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1188 {
1189 	int error;
1190 
1191 	if (!name)
1192 		return -EFAULT;
1193 	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1194 		return -EFAULT;
1195 
1196 	down_read(&uts_sem);
1197 	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1198 			       __OLD_UTS_LEN);
1199 	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1200 	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1201 				__OLD_UTS_LEN);
1202 	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1203 	error |= __copy_to_user(&name->release, &utsname()->release,
1204 				__OLD_UTS_LEN);
1205 	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1206 	error |= __copy_to_user(&name->version, &utsname()->version,
1207 				__OLD_UTS_LEN);
1208 	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1209 	error |= __copy_to_user(&name->machine, &utsname()->machine,
1210 				__OLD_UTS_LEN);
1211 	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1212 	up_read(&uts_sem);
1213 
1214 	if (!error && override_architecture(name))
1215 		error = -EFAULT;
1216 	if (!error && override_release(name->release, sizeof(name->release)))
1217 		error = -EFAULT;
1218 	return error ? -EFAULT : 0;
1219 }
1220 #endif
1221 
1222 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1223 {
1224 	int errno;
1225 	char tmp[__NEW_UTS_LEN];
1226 
1227 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1228 		return -EPERM;
1229 
1230 	if (len < 0 || len > __NEW_UTS_LEN)
1231 		return -EINVAL;
1232 	down_write(&uts_sem);
1233 	errno = -EFAULT;
1234 	if (!copy_from_user(tmp, name, len)) {
1235 		struct new_utsname *u = utsname();
1236 
1237 		memcpy(u->nodename, tmp, len);
1238 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1239 		errno = 0;
1240 		uts_proc_notify(UTS_PROC_HOSTNAME);
1241 	}
1242 	up_write(&uts_sem);
1243 	return errno;
1244 }
1245 
1246 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1247 
1248 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1249 {
1250 	int i, errno;
1251 	struct new_utsname *u;
1252 
1253 	if (len < 0)
1254 		return -EINVAL;
1255 	down_read(&uts_sem);
1256 	u = utsname();
1257 	i = 1 + strlen(u->nodename);
1258 	if (i > len)
1259 		i = len;
1260 	errno = 0;
1261 	if (copy_to_user(name, u->nodename, i))
1262 		errno = -EFAULT;
1263 	up_read(&uts_sem);
1264 	return errno;
1265 }
1266 
1267 #endif
1268 
1269 /*
1270  * Only setdomainname; getdomainname can be implemented by calling
1271  * uname()
1272  */
1273 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1274 {
1275 	int errno;
1276 	char tmp[__NEW_UTS_LEN];
1277 
1278 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1279 		return -EPERM;
1280 	if (len < 0 || len > __NEW_UTS_LEN)
1281 		return -EINVAL;
1282 
1283 	down_write(&uts_sem);
1284 	errno = -EFAULT;
1285 	if (!copy_from_user(tmp, name, len)) {
1286 		struct new_utsname *u = utsname();
1287 
1288 		memcpy(u->domainname, tmp, len);
1289 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1290 		errno = 0;
1291 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1292 	}
1293 	up_write(&uts_sem);
1294 	return errno;
1295 }
1296 
1297 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1298 {
1299 	struct rlimit value;
1300 	int ret;
1301 
1302 	ret = do_prlimit(current, resource, NULL, &value);
1303 	if (!ret)
1304 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1305 
1306 	return ret;
1307 }
1308 
1309 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1310 
1311 /*
1312  *	Back compatibility for getrlimit. Needed for some apps.
1313  */
1314 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1315 		struct rlimit __user *, rlim)
1316 {
1317 	struct rlimit x;
1318 	if (resource >= RLIM_NLIMITS)
1319 		return -EINVAL;
1320 
1321 	task_lock(current->group_leader);
1322 	x = current->signal->rlim[resource];
1323 	task_unlock(current->group_leader);
1324 	if (x.rlim_cur > 0x7FFFFFFF)
1325 		x.rlim_cur = 0x7FFFFFFF;
1326 	if (x.rlim_max > 0x7FFFFFFF)
1327 		x.rlim_max = 0x7FFFFFFF;
1328 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1329 }
1330 
1331 #endif
1332 
1333 static inline bool rlim64_is_infinity(__u64 rlim64)
1334 {
1335 #if BITS_PER_LONG < 64
1336 	return rlim64 >= ULONG_MAX;
1337 #else
1338 	return rlim64 == RLIM64_INFINITY;
1339 #endif
1340 }
1341 
1342 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1343 {
1344 	if (rlim->rlim_cur == RLIM_INFINITY)
1345 		rlim64->rlim_cur = RLIM64_INFINITY;
1346 	else
1347 		rlim64->rlim_cur = rlim->rlim_cur;
1348 	if (rlim->rlim_max == RLIM_INFINITY)
1349 		rlim64->rlim_max = RLIM64_INFINITY;
1350 	else
1351 		rlim64->rlim_max = rlim->rlim_max;
1352 }
1353 
1354 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1355 {
1356 	if (rlim64_is_infinity(rlim64->rlim_cur))
1357 		rlim->rlim_cur = RLIM_INFINITY;
1358 	else
1359 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1360 	if (rlim64_is_infinity(rlim64->rlim_max))
1361 		rlim->rlim_max = RLIM_INFINITY;
1362 	else
1363 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1364 }
1365 
1366 /* make sure you are allowed to change @tsk limits before calling this */
1367 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1368 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1369 {
1370 	struct rlimit *rlim;
1371 	int retval = 0;
1372 
1373 	if (resource >= RLIM_NLIMITS)
1374 		return -EINVAL;
1375 	if (new_rlim) {
1376 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1377 			return -EINVAL;
1378 		if (resource == RLIMIT_NOFILE &&
1379 				new_rlim->rlim_max > sysctl_nr_open)
1380 			return -EPERM;
1381 	}
1382 
1383 	/* protect tsk->signal and tsk->sighand from disappearing */
1384 	read_lock(&tasklist_lock);
1385 	if (!tsk->sighand) {
1386 		retval = -ESRCH;
1387 		goto out;
1388 	}
1389 
1390 	rlim = tsk->signal->rlim + resource;
1391 	task_lock(tsk->group_leader);
1392 	if (new_rlim) {
1393 		/* Keep the capable check against init_user_ns until
1394 		   cgroups can contain all limits */
1395 		if (new_rlim->rlim_max > rlim->rlim_max &&
1396 				!capable(CAP_SYS_RESOURCE))
1397 			retval = -EPERM;
1398 		if (!retval)
1399 			retval = security_task_setrlimit(tsk->group_leader,
1400 					resource, new_rlim);
1401 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1402 			/*
1403 			 * The caller is asking for an immediate RLIMIT_CPU
1404 			 * expiry.  But we use the zero value to mean "it was
1405 			 * never set".  So let's cheat and make it one second
1406 			 * instead
1407 			 */
1408 			new_rlim->rlim_cur = 1;
1409 		}
1410 	}
1411 	if (!retval) {
1412 		if (old_rlim)
1413 			*old_rlim = *rlim;
1414 		if (new_rlim)
1415 			*rlim = *new_rlim;
1416 	}
1417 	task_unlock(tsk->group_leader);
1418 
1419 	/*
1420 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1421 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1422 	 * very long-standing error, and fixing it now risks breakage of
1423 	 * applications, so we live with it
1424 	 */
1425 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1426 	     new_rlim->rlim_cur != RLIM_INFINITY &&
1427 	     IS_ENABLED(CONFIG_POSIX_TIMERS))
1428 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1429 out:
1430 	read_unlock(&tasklist_lock);
1431 	return retval;
1432 }
1433 
1434 /* rcu lock must be held */
1435 static int check_prlimit_permission(struct task_struct *task)
1436 {
1437 	const struct cred *cred = current_cred(), *tcred;
1438 
1439 	if (current == task)
1440 		return 0;
1441 
1442 	tcred = __task_cred(task);
1443 	if (uid_eq(cred->uid, tcred->euid) &&
1444 	    uid_eq(cred->uid, tcred->suid) &&
1445 	    uid_eq(cred->uid, tcred->uid)  &&
1446 	    gid_eq(cred->gid, tcred->egid) &&
1447 	    gid_eq(cred->gid, tcred->sgid) &&
1448 	    gid_eq(cred->gid, tcred->gid))
1449 		return 0;
1450 	if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1451 		return 0;
1452 
1453 	return -EPERM;
1454 }
1455 
1456 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1457 		const struct rlimit64 __user *, new_rlim,
1458 		struct rlimit64 __user *, old_rlim)
1459 {
1460 	struct rlimit64 old64, new64;
1461 	struct rlimit old, new;
1462 	struct task_struct *tsk;
1463 	int ret;
1464 
1465 	if (new_rlim) {
1466 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1467 			return -EFAULT;
1468 		rlim64_to_rlim(&new64, &new);
1469 	}
1470 
1471 	rcu_read_lock();
1472 	tsk = pid ? find_task_by_vpid(pid) : current;
1473 	if (!tsk) {
1474 		rcu_read_unlock();
1475 		return -ESRCH;
1476 	}
1477 	ret = check_prlimit_permission(tsk);
1478 	if (ret) {
1479 		rcu_read_unlock();
1480 		return ret;
1481 	}
1482 	get_task_struct(tsk);
1483 	rcu_read_unlock();
1484 
1485 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1486 			old_rlim ? &old : NULL);
1487 
1488 	if (!ret && old_rlim) {
1489 		rlim_to_rlim64(&old, &old64);
1490 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1491 			ret = -EFAULT;
1492 	}
1493 
1494 	put_task_struct(tsk);
1495 	return ret;
1496 }
1497 
1498 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1499 {
1500 	struct rlimit new_rlim;
1501 
1502 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1503 		return -EFAULT;
1504 	return do_prlimit(current, resource, &new_rlim, NULL);
1505 }
1506 
1507 /*
1508  * It would make sense to put struct rusage in the task_struct,
1509  * except that would make the task_struct be *really big*.  After
1510  * task_struct gets moved into malloc'ed memory, it would
1511  * make sense to do this.  It will make moving the rest of the information
1512  * a lot simpler!  (Which we're not doing right now because we're not
1513  * measuring them yet).
1514  *
1515  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1516  * races with threads incrementing their own counters.  But since word
1517  * reads are atomic, we either get new values or old values and we don't
1518  * care which for the sums.  We always take the siglock to protect reading
1519  * the c* fields from p->signal from races with exit.c updating those
1520  * fields when reaping, so a sample either gets all the additions of a
1521  * given child after it's reaped, or none so this sample is before reaping.
1522  *
1523  * Locking:
1524  * We need to take the siglock for CHILDEREN, SELF and BOTH
1525  * for  the cases current multithreaded, non-current single threaded
1526  * non-current multithreaded.  Thread traversal is now safe with
1527  * the siglock held.
1528  * Strictly speaking, we donot need to take the siglock if we are current and
1529  * single threaded,  as no one else can take our signal_struct away, no one
1530  * else can  reap the  children to update signal->c* counters, and no one else
1531  * can race with the signal-> fields. If we do not take any lock, the
1532  * signal-> fields could be read out of order while another thread was just
1533  * exiting. So we should  place a read memory barrier when we avoid the lock.
1534  * On the writer side,  write memory barrier is implied in  __exit_signal
1535  * as __exit_signal releases  the siglock spinlock after updating the signal->
1536  * fields. But we don't do this yet to keep things simple.
1537  *
1538  */
1539 
1540 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1541 {
1542 	r->ru_nvcsw += t->nvcsw;
1543 	r->ru_nivcsw += t->nivcsw;
1544 	r->ru_minflt += t->min_flt;
1545 	r->ru_majflt += t->maj_flt;
1546 	r->ru_inblock += task_io_get_inblock(t);
1547 	r->ru_oublock += task_io_get_oublock(t);
1548 }
1549 
1550 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1551 {
1552 	struct task_struct *t;
1553 	unsigned long flags;
1554 	u64 tgutime, tgstime, utime, stime;
1555 	unsigned long maxrss = 0;
1556 
1557 	memset((char *)r, 0, sizeof (*r));
1558 	utime = stime = 0;
1559 
1560 	if (who == RUSAGE_THREAD) {
1561 		task_cputime_adjusted(current, &utime, &stime);
1562 		accumulate_thread_rusage(p, r);
1563 		maxrss = p->signal->maxrss;
1564 		goto out;
1565 	}
1566 
1567 	if (!lock_task_sighand(p, &flags))
1568 		return;
1569 
1570 	switch (who) {
1571 	case RUSAGE_BOTH:
1572 	case RUSAGE_CHILDREN:
1573 		utime = p->signal->cutime;
1574 		stime = p->signal->cstime;
1575 		r->ru_nvcsw = p->signal->cnvcsw;
1576 		r->ru_nivcsw = p->signal->cnivcsw;
1577 		r->ru_minflt = p->signal->cmin_flt;
1578 		r->ru_majflt = p->signal->cmaj_flt;
1579 		r->ru_inblock = p->signal->cinblock;
1580 		r->ru_oublock = p->signal->coublock;
1581 		maxrss = p->signal->cmaxrss;
1582 
1583 		if (who == RUSAGE_CHILDREN)
1584 			break;
1585 
1586 	case RUSAGE_SELF:
1587 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1588 		utime += tgutime;
1589 		stime += tgstime;
1590 		r->ru_nvcsw += p->signal->nvcsw;
1591 		r->ru_nivcsw += p->signal->nivcsw;
1592 		r->ru_minflt += p->signal->min_flt;
1593 		r->ru_majflt += p->signal->maj_flt;
1594 		r->ru_inblock += p->signal->inblock;
1595 		r->ru_oublock += p->signal->oublock;
1596 		if (maxrss < p->signal->maxrss)
1597 			maxrss = p->signal->maxrss;
1598 		t = p;
1599 		do {
1600 			accumulate_thread_rusage(t, r);
1601 		} while_each_thread(p, t);
1602 		break;
1603 
1604 	default:
1605 		BUG();
1606 	}
1607 	unlock_task_sighand(p, &flags);
1608 
1609 out:
1610 	r->ru_utime = ns_to_timeval(utime);
1611 	r->ru_stime = ns_to_timeval(stime);
1612 
1613 	if (who != RUSAGE_CHILDREN) {
1614 		struct mm_struct *mm = get_task_mm(p);
1615 
1616 		if (mm) {
1617 			setmax_mm_hiwater_rss(&maxrss, mm);
1618 			mmput(mm);
1619 		}
1620 	}
1621 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1622 }
1623 
1624 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1625 {
1626 	struct rusage r;
1627 
1628 	k_getrusage(p, who, &r);
1629 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1630 }
1631 
1632 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1633 {
1634 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1635 	    who != RUSAGE_THREAD)
1636 		return -EINVAL;
1637 	return getrusage(current, who, ru);
1638 }
1639 
1640 #ifdef CONFIG_COMPAT
1641 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1642 {
1643 	struct rusage r;
1644 
1645 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1646 	    who != RUSAGE_THREAD)
1647 		return -EINVAL;
1648 
1649 	k_getrusage(current, who, &r);
1650 	return put_compat_rusage(&r, ru);
1651 }
1652 #endif
1653 
1654 SYSCALL_DEFINE1(umask, int, mask)
1655 {
1656 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1657 	return mask;
1658 }
1659 
1660 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1661 {
1662 	struct fd exe;
1663 	struct file *old_exe, *exe_file;
1664 	struct inode *inode;
1665 	int err;
1666 
1667 	exe = fdget(fd);
1668 	if (!exe.file)
1669 		return -EBADF;
1670 
1671 	inode = file_inode(exe.file);
1672 
1673 	/*
1674 	 * Because the original mm->exe_file points to executable file, make
1675 	 * sure that this one is executable as well, to avoid breaking an
1676 	 * overall picture.
1677 	 */
1678 	err = -EACCES;
1679 	if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1680 		goto exit;
1681 
1682 	err = inode_permission(inode, MAY_EXEC);
1683 	if (err)
1684 		goto exit;
1685 
1686 	/*
1687 	 * Forbid mm->exe_file change if old file still mapped.
1688 	 */
1689 	exe_file = get_mm_exe_file(mm);
1690 	err = -EBUSY;
1691 	if (exe_file) {
1692 		struct vm_area_struct *vma;
1693 
1694 		down_read(&mm->mmap_sem);
1695 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
1696 			if (!vma->vm_file)
1697 				continue;
1698 			if (path_equal(&vma->vm_file->f_path,
1699 				       &exe_file->f_path))
1700 				goto exit_err;
1701 		}
1702 
1703 		up_read(&mm->mmap_sem);
1704 		fput(exe_file);
1705 	}
1706 
1707 	err = 0;
1708 	/* set the new file, lockless */
1709 	get_file(exe.file);
1710 	old_exe = xchg(&mm->exe_file, exe.file);
1711 	if (old_exe)
1712 		fput(old_exe);
1713 exit:
1714 	fdput(exe);
1715 	return err;
1716 exit_err:
1717 	up_read(&mm->mmap_sem);
1718 	fput(exe_file);
1719 	goto exit;
1720 }
1721 
1722 /*
1723  * WARNING: we don't require any capability here so be very careful
1724  * in what is allowed for modification from userspace.
1725  */
1726 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1727 {
1728 	unsigned long mmap_max_addr = TASK_SIZE;
1729 	struct mm_struct *mm = current->mm;
1730 	int error = -EINVAL, i;
1731 
1732 	static const unsigned char offsets[] = {
1733 		offsetof(struct prctl_mm_map, start_code),
1734 		offsetof(struct prctl_mm_map, end_code),
1735 		offsetof(struct prctl_mm_map, start_data),
1736 		offsetof(struct prctl_mm_map, end_data),
1737 		offsetof(struct prctl_mm_map, start_brk),
1738 		offsetof(struct prctl_mm_map, brk),
1739 		offsetof(struct prctl_mm_map, start_stack),
1740 		offsetof(struct prctl_mm_map, arg_start),
1741 		offsetof(struct prctl_mm_map, arg_end),
1742 		offsetof(struct prctl_mm_map, env_start),
1743 		offsetof(struct prctl_mm_map, env_end),
1744 	};
1745 
1746 	/*
1747 	 * Make sure the members are not somewhere outside
1748 	 * of allowed address space.
1749 	 */
1750 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1751 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1752 
1753 		if ((unsigned long)val >= mmap_max_addr ||
1754 		    (unsigned long)val < mmap_min_addr)
1755 			goto out;
1756 	}
1757 
1758 	/*
1759 	 * Make sure the pairs are ordered.
1760 	 */
1761 #define __prctl_check_order(__m1, __op, __m2)				\
1762 	((unsigned long)prctl_map->__m1 __op				\
1763 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1764 	error  = __prctl_check_order(start_code, <, end_code);
1765 	error |= __prctl_check_order(start_data, <, end_data);
1766 	error |= __prctl_check_order(start_brk, <=, brk);
1767 	error |= __prctl_check_order(arg_start, <=, arg_end);
1768 	error |= __prctl_check_order(env_start, <=, env_end);
1769 	if (error)
1770 		goto out;
1771 #undef __prctl_check_order
1772 
1773 	error = -EINVAL;
1774 
1775 	/*
1776 	 * @brk should be after @end_data in traditional maps.
1777 	 */
1778 	if (prctl_map->start_brk <= prctl_map->end_data ||
1779 	    prctl_map->brk <= prctl_map->end_data)
1780 		goto out;
1781 
1782 	/*
1783 	 * Neither we should allow to override limits if they set.
1784 	 */
1785 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1786 			      prctl_map->start_brk, prctl_map->end_data,
1787 			      prctl_map->start_data))
1788 			goto out;
1789 
1790 	/*
1791 	 * Someone is trying to cheat the auxv vector.
1792 	 */
1793 	if (prctl_map->auxv_size) {
1794 		if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1795 			goto out;
1796 	}
1797 
1798 	/*
1799 	 * Finally, make sure the caller has the rights to
1800 	 * change /proc/pid/exe link: only local root should
1801 	 * be allowed to.
1802 	 */
1803 	if (prctl_map->exe_fd != (u32)-1) {
1804 		struct user_namespace *ns = current_user_ns();
1805 		const struct cred *cred = current_cred();
1806 
1807 		if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1808 		    !gid_eq(cred->gid, make_kgid(ns, 0)))
1809 			goto out;
1810 	}
1811 
1812 	error = 0;
1813 out:
1814 	return error;
1815 }
1816 
1817 #ifdef CONFIG_CHECKPOINT_RESTORE
1818 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1819 {
1820 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1821 	unsigned long user_auxv[AT_VECTOR_SIZE];
1822 	struct mm_struct *mm = current->mm;
1823 	int error;
1824 
1825 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1826 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1827 
1828 	if (opt == PR_SET_MM_MAP_SIZE)
1829 		return put_user((unsigned int)sizeof(prctl_map),
1830 				(unsigned int __user *)addr);
1831 
1832 	if (data_size != sizeof(prctl_map))
1833 		return -EINVAL;
1834 
1835 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1836 		return -EFAULT;
1837 
1838 	error = validate_prctl_map(&prctl_map);
1839 	if (error)
1840 		return error;
1841 
1842 	if (prctl_map.auxv_size) {
1843 		memset(user_auxv, 0, sizeof(user_auxv));
1844 		if (copy_from_user(user_auxv,
1845 				   (const void __user *)prctl_map.auxv,
1846 				   prctl_map.auxv_size))
1847 			return -EFAULT;
1848 
1849 		/* Last entry must be AT_NULL as specification requires */
1850 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1851 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1852 	}
1853 
1854 	if (prctl_map.exe_fd != (u32)-1) {
1855 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1856 		if (error)
1857 			return error;
1858 	}
1859 
1860 	down_write(&mm->mmap_sem);
1861 
1862 	/*
1863 	 * We don't validate if these members are pointing to
1864 	 * real present VMAs because application may have correspond
1865 	 * VMAs already unmapped and kernel uses these members for statistics
1866 	 * output in procfs mostly, except
1867 	 *
1868 	 *  - @start_brk/@brk which are used in do_brk but kernel lookups
1869 	 *    for VMAs when updating these memvers so anything wrong written
1870 	 *    here cause kernel to swear at userspace program but won't lead
1871 	 *    to any problem in kernel itself
1872 	 */
1873 
1874 	mm->start_code	= prctl_map.start_code;
1875 	mm->end_code	= prctl_map.end_code;
1876 	mm->start_data	= prctl_map.start_data;
1877 	mm->end_data	= prctl_map.end_data;
1878 	mm->start_brk	= prctl_map.start_brk;
1879 	mm->brk		= prctl_map.brk;
1880 	mm->start_stack	= prctl_map.start_stack;
1881 	mm->arg_start	= prctl_map.arg_start;
1882 	mm->arg_end	= prctl_map.arg_end;
1883 	mm->env_start	= prctl_map.env_start;
1884 	mm->env_end	= prctl_map.env_end;
1885 
1886 	/*
1887 	 * Note this update of @saved_auxv is lockless thus
1888 	 * if someone reads this member in procfs while we're
1889 	 * updating -- it may get partly updated results. It's
1890 	 * known and acceptable trade off: we leave it as is to
1891 	 * not introduce additional locks here making the kernel
1892 	 * more complex.
1893 	 */
1894 	if (prctl_map.auxv_size)
1895 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1896 
1897 	up_write(&mm->mmap_sem);
1898 	return 0;
1899 }
1900 #endif /* CONFIG_CHECKPOINT_RESTORE */
1901 
1902 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
1903 			  unsigned long len)
1904 {
1905 	/*
1906 	 * This doesn't move the auxiliary vector itself since it's pinned to
1907 	 * mm_struct, but it permits filling the vector with new values.  It's
1908 	 * up to the caller to provide sane values here, otherwise userspace
1909 	 * tools which use this vector might be unhappy.
1910 	 */
1911 	unsigned long user_auxv[AT_VECTOR_SIZE];
1912 
1913 	if (len > sizeof(user_auxv))
1914 		return -EINVAL;
1915 
1916 	if (copy_from_user(user_auxv, (const void __user *)addr, len))
1917 		return -EFAULT;
1918 
1919 	/* Make sure the last entry is always AT_NULL */
1920 	user_auxv[AT_VECTOR_SIZE - 2] = 0;
1921 	user_auxv[AT_VECTOR_SIZE - 1] = 0;
1922 
1923 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1924 
1925 	task_lock(current);
1926 	memcpy(mm->saved_auxv, user_auxv, len);
1927 	task_unlock(current);
1928 
1929 	return 0;
1930 }
1931 
1932 static int prctl_set_mm(int opt, unsigned long addr,
1933 			unsigned long arg4, unsigned long arg5)
1934 {
1935 	struct mm_struct *mm = current->mm;
1936 	struct prctl_mm_map prctl_map;
1937 	struct vm_area_struct *vma;
1938 	int error;
1939 
1940 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1941 			      opt != PR_SET_MM_MAP &&
1942 			      opt != PR_SET_MM_MAP_SIZE)))
1943 		return -EINVAL;
1944 
1945 #ifdef CONFIG_CHECKPOINT_RESTORE
1946 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1947 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1948 #endif
1949 
1950 	if (!capable(CAP_SYS_RESOURCE))
1951 		return -EPERM;
1952 
1953 	if (opt == PR_SET_MM_EXE_FILE)
1954 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1955 
1956 	if (opt == PR_SET_MM_AUXV)
1957 		return prctl_set_auxv(mm, addr, arg4);
1958 
1959 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
1960 		return -EINVAL;
1961 
1962 	error = -EINVAL;
1963 
1964 	down_write(&mm->mmap_sem);
1965 	vma = find_vma(mm, addr);
1966 
1967 	prctl_map.start_code	= mm->start_code;
1968 	prctl_map.end_code	= mm->end_code;
1969 	prctl_map.start_data	= mm->start_data;
1970 	prctl_map.end_data	= mm->end_data;
1971 	prctl_map.start_brk	= mm->start_brk;
1972 	prctl_map.brk		= mm->brk;
1973 	prctl_map.start_stack	= mm->start_stack;
1974 	prctl_map.arg_start	= mm->arg_start;
1975 	prctl_map.arg_end	= mm->arg_end;
1976 	prctl_map.env_start	= mm->env_start;
1977 	prctl_map.env_end	= mm->env_end;
1978 	prctl_map.auxv		= NULL;
1979 	prctl_map.auxv_size	= 0;
1980 	prctl_map.exe_fd	= -1;
1981 
1982 	switch (opt) {
1983 	case PR_SET_MM_START_CODE:
1984 		prctl_map.start_code = addr;
1985 		break;
1986 	case PR_SET_MM_END_CODE:
1987 		prctl_map.end_code = addr;
1988 		break;
1989 	case PR_SET_MM_START_DATA:
1990 		prctl_map.start_data = addr;
1991 		break;
1992 	case PR_SET_MM_END_DATA:
1993 		prctl_map.end_data = addr;
1994 		break;
1995 	case PR_SET_MM_START_STACK:
1996 		prctl_map.start_stack = addr;
1997 		break;
1998 	case PR_SET_MM_START_BRK:
1999 		prctl_map.start_brk = addr;
2000 		break;
2001 	case PR_SET_MM_BRK:
2002 		prctl_map.brk = addr;
2003 		break;
2004 	case PR_SET_MM_ARG_START:
2005 		prctl_map.arg_start = addr;
2006 		break;
2007 	case PR_SET_MM_ARG_END:
2008 		prctl_map.arg_end = addr;
2009 		break;
2010 	case PR_SET_MM_ENV_START:
2011 		prctl_map.env_start = addr;
2012 		break;
2013 	case PR_SET_MM_ENV_END:
2014 		prctl_map.env_end = addr;
2015 		break;
2016 	default:
2017 		goto out;
2018 	}
2019 
2020 	error = validate_prctl_map(&prctl_map);
2021 	if (error)
2022 		goto out;
2023 
2024 	switch (opt) {
2025 	/*
2026 	 * If command line arguments and environment
2027 	 * are placed somewhere else on stack, we can
2028 	 * set them up here, ARG_START/END to setup
2029 	 * command line argumets and ENV_START/END
2030 	 * for environment.
2031 	 */
2032 	case PR_SET_MM_START_STACK:
2033 	case PR_SET_MM_ARG_START:
2034 	case PR_SET_MM_ARG_END:
2035 	case PR_SET_MM_ENV_START:
2036 	case PR_SET_MM_ENV_END:
2037 		if (!vma) {
2038 			error = -EFAULT;
2039 			goto out;
2040 		}
2041 	}
2042 
2043 	mm->start_code	= prctl_map.start_code;
2044 	mm->end_code	= prctl_map.end_code;
2045 	mm->start_data	= prctl_map.start_data;
2046 	mm->end_data	= prctl_map.end_data;
2047 	mm->start_brk	= prctl_map.start_brk;
2048 	mm->brk		= prctl_map.brk;
2049 	mm->start_stack	= prctl_map.start_stack;
2050 	mm->arg_start	= prctl_map.arg_start;
2051 	mm->arg_end	= prctl_map.arg_end;
2052 	mm->env_start	= prctl_map.env_start;
2053 	mm->env_end	= prctl_map.env_end;
2054 
2055 	error = 0;
2056 out:
2057 	up_write(&mm->mmap_sem);
2058 	return error;
2059 }
2060 
2061 #ifdef CONFIG_CHECKPOINT_RESTORE
2062 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2063 {
2064 	return put_user(me->clear_child_tid, tid_addr);
2065 }
2066 #else
2067 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2068 {
2069 	return -EINVAL;
2070 }
2071 #endif
2072 
2073 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2074 {
2075 	/*
2076 	 * If task has has_child_subreaper - all its decendants
2077 	 * already have these flag too and new decendants will
2078 	 * inherit it on fork, skip them.
2079 	 *
2080 	 * If we've found child_reaper - skip descendants in
2081 	 * it's subtree as they will never get out pidns.
2082 	 */
2083 	if (p->signal->has_child_subreaper ||
2084 	    is_child_reaper(task_pid(p)))
2085 		return 0;
2086 
2087 	p->signal->has_child_subreaper = 1;
2088 	return 1;
2089 }
2090 
2091 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2092 		unsigned long, arg4, unsigned long, arg5)
2093 {
2094 	struct task_struct *me = current;
2095 	unsigned char comm[sizeof(me->comm)];
2096 	long error;
2097 
2098 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2099 	if (error != -ENOSYS)
2100 		return error;
2101 
2102 	error = 0;
2103 	switch (option) {
2104 	case PR_SET_PDEATHSIG:
2105 		if (!valid_signal(arg2)) {
2106 			error = -EINVAL;
2107 			break;
2108 		}
2109 		me->pdeath_signal = arg2;
2110 		break;
2111 	case PR_GET_PDEATHSIG:
2112 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2113 		break;
2114 	case PR_GET_DUMPABLE:
2115 		error = get_dumpable(me->mm);
2116 		break;
2117 	case PR_SET_DUMPABLE:
2118 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2119 			error = -EINVAL;
2120 			break;
2121 		}
2122 		set_dumpable(me->mm, arg2);
2123 		break;
2124 
2125 	case PR_SET_UNALIGN:
2126 		error = SET_UNALIGN_CTL(me, arg2);
2127 		break;
2128 	case PR_GET_UNALIGN:
2129 		error = GET_UNALIGN_CTL(me, arg2);
2130 		break;
2131 	case PR_SET_FPEMU:
2132 		error = SET_FPEMU_CTL(me, arg2);
2133 		break;
2134 	case PR_GET_FPEMU:
2135 		error = GET_FPEMU_CTL(me, arg2);
2136 		break;
2137 	case PR_SET_FPEXC:
2138 		error = SET_FPEXC_CTL(me, arg2);
2139 		break;
2140 	case PR_GET_FPEXC:
2141 		error = GET_FPEXC_CTL(me, arg2);
2142 		break;
2143 	case PR_GET_TIMING:
2144 		error = PR_TIMING_STATISTICAL;
2145 		break;
2146 	case PR_SET_TIMING:
2147 		if (arg2 != PR_TIMING_STATISTICAL)
2148 			error = -EINVAL;
2149 		break;
2150 	case PR_SET_NAME:
2151 		comm[sizeof(me->comm) - 1] = 0;
2152 		if (strncpy_from_user(comm, (char __user *)arg2,
2153 				      sizeof(me->comm) - 1) < 0)
2154 			return -EFAULT;
2155 		set_task_comm(me, comm);
2156 		proc_comm_connector(me);
2157 		break;
2158 	case PR_GET_NAME:
2159 		get_task_comm(comm, me);
2160 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2161 			return -EFAULT;
2162 		break;
2163 	case PR_GET_ENDIAN:
2164 		error = GET_ENDIAN(me, arg2);
2165 		break;
2166 	case PR_SET_ENDIAN:
2167 		error = SET_ENDIAN(me, arg2);
2168 		break;
2169 	case PR_GET_SECCOMP:
2170 		error = prctl_get_seccomp();
2171 		break;
2172 	case PR_SET_SECCOMP:
2173 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2174 		break;
2175 	case PR_GET_TSC:
2176 		error = GET_TSC_CTL(arg2);
2177 		break;
2178 	case PR_SET_TSC:
2179 		error = SET_TSC_CTL(arg2);
2180 		break;
2181 	case PR_TASK_PERF_EVENTS_DISABLE:
2182 		error = perf_event_task_disable();
2183 		break;
2184 	case PR_TASK_PERF_EVENTS_ENABLE:
2185 		error = perf_event_task_enable();
2186 		break;
2187 	case PR_GET_TIMERSLACK:
2188 		if (current->timer_slack_ns > ULONG_MAX)
2189 			error = ULONG_MAX;
2190 		else
2191 			error = current->timer_slack_ns;
2192 		break;
2193 	case PR_SET_TIMERSLACK:
2194 		if (arg2 <= 0)
2195 			current->timer_slack_ns =
2196 					current->default_timer_slack_ns;
2197 		else
2198 			current->timer_slack_ns = arg2;
2199 		break;
2200 	case PR_MCE_KILL:
2201 		if (arg4 | arg5)
2202 			return -EINVAL;
2203 		switch (arg2) {
2204 		case PR_MCE_KILL_CLEAR:
2205 			if (arg3 != 0)
2206 				return -EINVAL;
2207 			current->flags &= ~PF_MCE_PROCESS;
2208 			break;
2209 		case PR_MCE_KILL_SET:
2210 			current->flags |= PF_MCE_PROCESS;
2211 			if (arg3 == PR_MCE_KILL_EARLY)
2212 				current->flags |= PF_MCE_EARLY;
2213 			else if (arg3 == PR_MCE_KILL_LATE)
2214 				current->flags &= ~PF_MCE_EARLY;
2215 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2216 				current->flags &=
2217 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2218 			else
2219 				return -EINVAL;
2220 			break;
2221 		default:
2222 			return -EINVAL;
2223 		}
2224 		break;
2225 	case PR_MCE_KILL_GET:
2226 		if (arg2 | arg3 | arg4 | arg5)
2227 			return -EINVAL;
2228 		if (current->flags & PF_MCE_PROCESS)
2229 			error = (current->flags & PF_MCE_EARLY) ?
2230 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2231 		else
2232 			error = PR_MCE_KILL_DEFAULT;
2233 		break;
2234 	case PR_SET_MM:
2235 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2236 		break;
2237 	case PR_GET_TID_ADDRESS:
2238 		error = prctl_get_tid_address(me, (int __user **)arg2);
2239 		break;
2240 	case PR_SET_CHILD_SUBREAPER:
2241 		me->signal->is_child_subreaper = !!arg2;
2242 		if (!arg2)
2243 			break;
2244 
2245 		walk_process_tree(me, propagate_has_child_subreaper, NULL);
2246 		break;
2247 	case PR_GET_CHILD_SUBREAPER:
2248 		error = put_user(me->signal->is_child_subreaper,
2249 				 (int __user *)arg2);
2250 		break;
2251 	case PR_SET_NO_NEW_PRIVS:
2252 		if (arg2 != 1 || arg3 || arg4 || arg5)
2253 			return -EINVAL;
2254 
2255 		task_set_no_new_privs(current);
2256 		break;
2257 	case PR_GET_NO_NEW_PRIVS:
2258 		if (arg2 || arg3 || arg4 || arg5)
2259 			return -EINVAL;
2260 		return task_no_new_privs(current) ? 1 : 0;
2261 	case PR_GET_THP_DISABLE:
2262 		if (arg2 || arg3 || arg4 || arg5)
2263 			return -EINVAL;
2264 		error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2265 		break;
2266 	case PR_SET_THP_DISABLE:
2267 		if (arg3 || arg4 || arg5)
2268 			return -EINVAL;
2269 		if (down_write_killable(&me->mm->mmap_sem))
2270 			return -EINTR;
2271 		if (arg2)
2272 			me->mm->def_flags |= VM_NOHUGEPAGE;
2273 		else
2274 			me->mm->def_flags &= ~VM_NOHUGEPAGE;
2275 		up_write(&me->mm->mmap_sem);
2276 		break;
2277 	case PR_MPX_ENABLE_MANAGEMENT:
2278 		if (arg2 || arg3 || arg4 || arg5)
2279 			return -EINVAL;
2280 		error = MPX_ENABLE_MANAGEMENT();
2281 		break;
2282 	case PR_MPX_DISABLE_MANAGEMENT:
2283 		if (arg2 || arg3 || arg4 || arg5)
2284 			return -EINVAL;
2285 		error = MPX_DISABLE_MANAGEMENT();
2286 		break;
2287 	case PR_SET_FP_MODE:
2288 		error = SET_FP_MODE(me, arg2);
2289 		break;
2290 	case PR_GET_FP_MODE:
2291 		error = GET_FP_MODE(me);
2292 		break;
2293 	default:
2294 		error = -EINVAL;
2295 		break;
2296 	}
2297 	return error;
2298 }
2299 
2300 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2301 		struct getcpu_cache __user *, unused)
2302 {
2303 	int err = 0;
2304 	int cpu = raw_smp_processor_id();
2305 
2306 	if (cpup)
2307 		err |= put_user(cpu, cpup);
2308 	if (nodep)
2309 		err |= put_user(cpu_to_node(cpu), nodep);
2310 	return err ? -EFAULT : 0;
2311 }
2312 
2313 /**
2314  * do_sysinfo - fill in sysinfo struct
2315  * @info: pointer to buffer to fill
2316  */
2317 static int do_sysinfo(struct sysinfo *info)
2318 {
2319 	unsigned long mem_total, sav_total;
2320 	unsigned int mem_unit, bitcount;
2321 	struct timespec tp;
2322 
2323 	memset(info, 0, sizeof(struct sysinfo));
2324 
2325 	get_monotonic_boottime(&tp);
2326 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2327 
2328 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2329 
2330 	info->procs = nr_threads;
2331 
2332 	si_meminfo(info);
2333 	si_swapinfo(info);
2334 
2335 	/*
2336 	 * If the sum of all the available memory (i.e. ram + swap)
2337 	 * is less than can be stored in a 32 bit unsigned long then
2338 	 * we can be binary compatible with 2.2.x kernels.  If not,
2339 	 * well, in that case 2.2.x was broken anyways...
2340 	 *
2341 	 *  -Erik Andersen <andersee@debian.org>
2342 	 */
2343 
2344 	mem_total = info->totalram + info->totalswap;
2345 	if (mem_total < info->totalram || mem_total < info->totalswap)
2346 		goto out;
2347 	bitcount = 0;
2348 	mem_unit = info->mem_unit;
2349 	while (mem_unit > 1) {
2350 		bitcount++;
2351 		mem_unit >>= 1;
2352 		sav_total = mem_total;
2353 		mem_total <<= 1;
2354 		if (mem_total < sav_total)
2355 			goto out;
2356 	}
2357 
2358 	/*
2359 	 * If mem_total did not overflow, multiply all memory values by
2360 	 * info->mem_unit and set it to 1.  This leaves things compatible
2361 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2362 	 * kernels...
2363 	 */
2364 
2365 	info->mem_unit = 1;
2366 	info->totalram <<= bitcount;
2367 	info->freeram <<= bitcount;
2368 	info->sharedram <<= bitcount;
2369 	info->bufferram <<= bitcount;
2370 	info->totalswap <<= bitcount;
2371 	info->freeswap <<= bitcount;
2372 	info->totalhigh <<= bitcount;
2373 	info->freehigh <<= bitcount;
2374 
2375 out:
2376 	return 0;
2377 }
2378 
2379 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2380 {
2381 	struct sysinfo val;
2382 
2383 	do_sysinfo(&val);
2384 
2385 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2386 		return -EFAULT;
2387 
2388 	return 0;
2389 }
2390 
2391 #ifdef CONFIG_COMPAT
2392 struct compat_sysinfo {
2393 	s32 uptime;
2394 	u32 loads[3];
2395 	u32 totalram;
2396 	u32 freeram;
2397 	u32 sharedram;
2398 	u32 bufferram;
2399 	u32 totalswap;
2400 	u32 freeswap;
2401 	u16 procs;
2402 	u16 pad;
2403 	u32 totalhigh;
2404 	u32 freehigh;
2405 	u32 mem_unit;
2406 	char _f[20-2*sizeof(u32)-sizeof(int)];
2407 };
2408 
2409 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2410 {
2411 	struct sysinfo s;
2412 
2413 	do_sysinfo(&s);
2414 
2415 	/* Check to see if any memory value is too large for 32-bit and scale
2416 	 *  down if needed
2417 	 */
2418 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2419 		int bitcount = 0;
2420 
2421 		while (s.mem_unit < PAGE_SIZE) {
2422 			s.mem_unit <<= 1;
2423 			bitcount++;
2424 		}
2425 
2426 		s.totalram >>= bitcount;
2427 		s.freeram >>= bitcount;
2428 		s.sharedram >>= bitcount;
2429 		s.bufferram >>= bitcount;
2430 		s.totalswap >>= bitcount;
2431 		s.freeswap >>= bitcount;
2432 		s.totalhigh >>= bitcount;
2433 		s.freehigh >>= bitcount;
2434 	}
2435 
2436 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2437 	    __put_user(s.uptime, &info->uptime) ||
2438 	    __put_user(s.loads[0], &info->loads[0]) ||
2439 	    __put_user(s.loads[1], &info->loads[1]) ||
2440 	    __put_user(s.loads[2], &info->loads[2]) ||
2441 	    __put_user(s.totalram, &info->totalram) ||
2442 	    __put_user(s.freeram, &info->freeram) ||
2443 	    __put_user(s.sharedram, &info->sharedram) ||
2444 	    __put_user(s.bufferram, &info->bufferram) ||
2445 	    __put_user(s.totalswap, &info->totalswap) ||
2446 	    __put_user(s.freeswap, &info->freeswap) ||
2447 	    __put_user(s.procs, &info->procs) ||
2448 	    __put_user(s.totalhigh, &info->totalhigh) ||
2449 	    __put_user(s.freehigh, &info->freehigh) ||
2450 	    __put_user(s.mem_unit, &info->mem_unit))
2451 		return -EFAULT;
2452 
2453 	return 0;
2454 }
2455 #endif /* CONFIG_COMPAT */
2456