xref: /openbmc/linux/kernel/sys.c (revision 867a0e05)
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/rcupdate.h>
53 #include <linux/uidgid.h>
54 #include <linux/cred.h>
55 
56 #include <linux/kmsg_dump.h>
57 /* Move somewhere else to avoid recompiling? */
58 #include <generated/utsrelease.h>
59 
60 #include <asm/uaccess.h>
61 #include <asm/io.h>
62 #include <asm/unistd.h>
63 
64 #ifndef SET_UNALIGN_CTL
65 # define SET_UNALIGN_CTL(a, b)	(-EINVAL)
66 #endif
67 #ifndef GET_UNALIGN_CTL
68 # define GET_UNALIGN_CTL(a, b)	(-EINVAL)
69 #endif
70 #ifndef SET_FPEMU_CTL
71 # define SET_FPEMU_CTL(a, b)	(-EINVAL)
72 #endif
73 #ifndef GET_FPEMU_CTL
74 # define GET_FPEMU_CTL(a, b)	(-EINVAL)
75 #endif
76 #ifndef SET_FPEXC_CTL
77 # define SET_FPEXC_CTL(a, b)	(-EINVAL)
78 #endif
79 #ifndef GET_FPEXC_CTL
80 # define GET_FPEXC_CTL(a, b)	(-EINVAL)
81 #endif
82 #ifndef GET_ENDIAN
83 # define GET_ENDIAN(a, b)	(-EINVAL)
84 #endif
85 #ifndef SET_ENDIAN
86 # define SET_ENDIAN(a, b)	(-EINVAL)
87 #endif
88 #ifndef GET_TSC_CTL
89 # define GET_TSC_CTL(a)		(-EINVAL)
90 #endif
91 #ifndef SET_TSC_CTL
92 # define SET_TSC_CTL(a)		(-EINVAL)
93 #endif
94 #ifndef MPX_ENABLE_MANAGEMENT
95 # define MPX_ENABLE_MANAGEMENT(a)	(-EINVAL)
96 #endif
97 #ifndef MPX_DISABLE_MANAGEMENT
98 # define MPX_DISABLE_MANAGEMENT(a)	(-EINVAL)
99 #endif
100 
101 /*
102  * this is where the system-wide overflow UID and GID are defined, for
103  * architectures that now have 32-bit UID/GID but didn't in the past
104  */
105 
106 int overflowuid = DEFAULT_OVERFLOWUID;
107 int overflowgid = DEFAULT_OVERFLOWGID;
108 
109 EXPORT_SYMBOL(overflowuid);
110 EXPORT_SYMBOL(overflowgid);
111 
112 /*
113  * the same as above, but for filesystems which can only store a 16-bit
114  * UID and GID. as such, this is needed on all architectures
115  */
116 
117 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
118 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
119 
120 EXPORT_SYMBOL(fs_overflowuid);
121 EXPORT_SYMBOL(fs_overflowgid);
122 
123 /*
124  * Returns true if current's euid is same as p's uid or euid,
125  * or has CAP_SYS_NICE to p's user_ns.
126  *
127  * Called with rcu_read_lock, creds are safe
128  */
129 static bool set_one_prio_perm(struct task_struct *p)
130 {
131 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
132 
133 	if (uid_eq(pcred->uid,  cred->euid) ||
134 	    uid_eq(pcred->euid, cred->euid))
135 		return true;
136 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
137 		return true;
138 	return false;
139 }
140 
141 /*
142  * set the priority of a task
143  * - the caller must hold the RCU read lock
144  */
145 static int set_one_prio(struct task_struct *p, int niceval, int error)
146 {
147 	int no_nice;
148 
149 	if (!set_one_prio_perm(p)) {
150 		error = -EPERM;
151 		goto out;
152 	}
153 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
154 		error = -EACCES;
155 		goto out;
156 	}
157 	no_nice = security_task_setnice(p, niceval);
158 	if (no_nice) {
159 		error = no_nice;
160 		goto out;
161 	}
162 	if (error == -ESRCH)
163 		error = 0;
164 	set_user_nice(p, niceval);
165 out:
166 	return error;
167 }
168 
169 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
170 {
171 	struct task_struct *g, *p;
172 	struct user_struct *user;
173 	const struct cred *cred = current_cred();
174 	int error = -EINVAL;
175 	struct pid *pgrp;
176 	kuid_t uid;
177 
178 	if (which > PRIO_USER || which < PRIO_PROCESS)
179 		goto out;
180 
181 	/* normalize: avoid signed division (rounding problems) */
182 	error = -ESRCH;
183 	if (niceval < MIN_NICE)
184 		niceval = MIN_NICE;
185 	if (niceval > MAX_NICE)
186 		niceval = MAX_NICE;
187 
188 	rcu_read_lock();
189 	read_lock(&tasklist_lock);
190 	switch (which) {
191 	case PRIO_PROCESS:
192 		if (who)
193 			p = find_task_by_vpid(who);
194 		else
195 			p = current;
196 		if (p)
197 			error = set_one_prio(p, niceval, error);
198 		break;
199 	case PRIO_PGRP:
200 		if (who)
201 			pgrp = find_vpid(who);
202 		else
203 			pgrp = task_pgrp(current);
204 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
205 			error = set_one_prio(p, niceval, error);
206 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
207 		break;
208 	case PRIO_USER:
209 		uid = make_kuid(cred->user_ns, who);
210 		user = cred->user;
211 		if (!who)
212 			uid = cred->uid;
213 		else if (!uid_eq(uid, cred->uid)) {
214 			user = find_user(uid);
215 			if (!user)
216 				goto out_unlock;	/* No processes for this user */
217 		}
218 		do_each_thread(g, p) {
219 			if (uid_eq(task_uid(p), uid))
220 				error = set_one_prio(p, niceval, error);
221 		} while_each_thread(g, p);
222 		if (!uid_eq(uid, cred->uid))
223 			free_uid(user);		/* For find_user() */
224 		break;
225 	}
226 out_unlock:
227 	read_unlock(&tasklist_lock);
228 	rcu_read_unlock();
229 out:
230 	return error;
231 }
232 
233 /*
234  * Ugh. To avoid negative return values, "getpriority()" will
235  * not return the normal nice-value, but a negated value that
236  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
237  * to stay compatible.
238  */
239 SYSCALL_DEFINE2(getpriority, int, which, int, who)
240 {
241 	struct task_struct *g, *p;
242 	struct user_struct *user;
243 	const struct cred *cred = current_cred();
244 	long niceval, retval = -ESRCH;
245 	struct pid *pgrp;
246 	kuid_t uid;
247 
248 	if (which > PRIO_USER || which < PRIO_PROCESS)
249 		return -EINVAL;
250 
251 	rcu_read_lock();
252 	read_lock(&tasklist_lock);
253 	switch (which) {
254 	case PRIO_PROCESS:
255 		if (who)
256 			p = find_task_by_vpid(who);
257 		else
258 			p = current;
259 		if (p) {
260 			niceval = nice_to_rlimit(task_nice(p));
261 			if (niceval > retval)
262 				retval = niceval;
263 		}
264 		break;
265 	case PRIO_PGRP:
266 		if (who)
267 			pgrp = find_vpid(who);
268 		else
269 			pgrp = task_pgrp(current);
270 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
271 			niceval = nice_to_rlimit(task_nice(p));
272 			if (niceval > retval)
273 				retval = niceval;
274 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
275 		break;
276 	case PRIO_USER:
277 		uid = make_kuid(cred->user_ns, who);
278 		user = cred->user;
279 		if (!who)
280 			uid = cred->uid;
281 		else if (!uid_eq(uid, cred->uid)) {
282 			user = find_user(uid);
283 			if (!user)
284 				goto out_unlock;	/* No processes for this user */
285 		}
286 		do_each_thread(g, p) {
287 			if (uid_eq(task_uid(p), uid)) {
288 				niceval = nice_to_rlimit(task_nice(p));
289 				if (niceval > retval)
290 					retval = niceval;
291 			}
292 		} while_each_thread(g, p);
293 		if (!uid_eq(uid, cred->uid))
294 			free_uid(user);		/* for find_user() */
295 		break;
296 	}
297 out_unlock:
298 	read_unlock(&tasklist_lock);
299 	rcu_read_unlock();
300 
301 	return retval;
302 }
303 
304 /*
305  * Unprivileged users may change the real gid to the effective gid
306  * or vice versa.  (BSD-style)
307  *
308  * If you set the real gid at all, or set the effective gid to a value not
309  * equal to the real gid, then the saved gid is set to the new effective gid.
310  *
311  * This makes it possible for a setgid program to completely drop its
312  * privileges, which is often a useful assertion to make when you are doing
313  * a security audit over a program.
314  *
315  * The general idea is that a program which uses just setregid() will be
316  * 100% compatible with BSD.  A program which uses just setgid() will be
317  * 100% compatible with POSIX with saved IDs.
318  *
319  * SMP: There are not races, the GIDs are checked only by filesystem
320  *      operations (as far as semantic preservation is concerned).
321  */
322 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
323 {
324 	struct user_namespace *ns = current_user_ns();
325 	const struct cred *old;
326 	struct cred *new;
327 	int retval;
328 	kgid_t krgid, kegid;
329 
330 	krgid = make_kgid(ns, rgid);
331 	kegid = make_kgid(ns, egid);
332 
333 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
334 		return -EINVAL;
335 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
336 		return -EINVAL;
337 
338 	new = prepare_creds();
339 	if (!new)
340 		return -ENOMEM;
341 	old = current_cred();
342 
343 	retval = -EPERM;
344 	if (rgid != (gid_t) -1) {
345 		if (gid_eq(old->gid, krgid) ||
346 		    gid_eq(old->egid, krgid) ||
347 		    ns_capable(old->user_ns, CAP_SETGID))
348 			new->gid = krgid;
349 		else
350 			goto error;
351 	}
352 	if (egid != (gid_t) -1) {
353 		if (gid_eq(old->gid, kegid) ||
354 		    gid_eq(old->egid, kegid) ||
355 		    gid_eq(old->sgid, kegid) ||
356 		    ns_capable(old->user_ns, CAP_SETGID))
357 			new->egid = kegid;
358 		else
359 			goto error;
360 	}
361 
362 	if (rgid != (gid_t) -1 ||
363 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
364 		new->sgid = new->egid;
365 	new->fsgid = new->egid;
366 
367 	return commit_creds(new);
368 
369 error:
370 	abort_creds(new);
371 	return retval;
372 }
373 
374 /*
375  * setgid() is implemented like SysV w/ SAVED_IDS
376  *
377  * SMP: Same implicit races as above.
378  */
379 SYSCALL_DEFINE1(setgid, gid_t, gid)
380 {
381 	struct user_namespace *ns = current_user_ns();
382 	const struct cred *old;
383 	struct cred *new;
384 	int retval;
385 	kgid_t kgid;
386 
387 	kgid = make_kgid(ns, gid);
388 	if (!gid_valid(kgid))
389 		return -EINVAL;
390 
391 	new = prepare_creds();
392 	if (!new)
393 		return -ENOMEM;
394 	old = current_cred();
395 
396 	retval = -EPERM;
397 	if (ns_capable(old->user_ns, CAP_SETGID))
398 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
399 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
400 		new->egid = new->fsgid = kgid;
401 	else
402 		goto error;
403 
404 	return commit_creds(new);
405 
406 error:
407 	abort_creds(new);
408 	return retval;
409 }
410 
411 /*
412  * change the user struct in a credentials set to match the new UID
413  */
414 static int set_user(struct cred *new)
415 {
416 	struct user_struct *new_user;
417 
418 	new_user = alloc_uid(new->uid);
419 	if (!new_user)
420 		return -EAGAIN;
421 
422 	/*
423 	 * We don't fail in case of NPROC limit excess here because too many
424 	 * poorly written programs don't check set*uid() return code, assuming
425 	 * it never fails if called by root.  We may still enforce NPROC limit
426 	 * for programs doing set*uid()+execve() by harmlessly deferring the
427 	 * failure to the execve() stage.
428 	 */
429 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
430 			new_user != INIT_USER)
431 		current->flags |= PF_NPROC_EXCEEDED;
432 	else
433 		current->flags &= ~PF_NPROC_EXCEEDED;
434 
435 	free_uid(new->user);
436 	new->user = new_user;
437 	return 0;
438 }
439 
440 /*
441  * Unprivileged users may change the real uid to the effective uid
442  * or vice versa.  (BSD-style)
443  *
444  * If you set the real uid at all, or set the effective uid to a value not
445  * equal to the real uid, then the saved uid is set to the new effective uid.
446  *
447  * This makes it possible for a setuid program to completely drop its
448  * privileges, which is often a useful assertion to make when you are doing
449  * a security audit over a program.
450  *
451  * The general idea is that a program which uses just setreuid() will be
452  * 100% compatible with BSD.  A program which uses just setuid() will be
453  * 100% compatible with POSIX with saved IDs.
454  */
455 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
456 {
457 	struct user_namespace *ns = current_user_ns();
458 	const struct cred *old;
459 	struct cred *new;
460 	int retval;
461 	kuid_t kruid, keuid;
462 
463 	kruid = make_kuid(ns, ruid);
464 	keuid = make_kuid(ns, euid);
465 
466 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
467 		return -EINVAL;
468 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
469 		return -EINVAL;
470 
471 	new = prepare_creds();
472 	if (!new)
473 		return -ENOMEM;
474 	old = current_cred();
475 
476 	retval = -EPERM;
477 	if (ruid != (uid_t) -1) {
478 		new->uid = kruid;
479 		if (!uid_eq(old->uid, kruid) &&
480 		    !uid_eq(old->euid, kruid) &&
481 		    !ns_capable(old->user_ns, CAP_SETUID))
482 			goto error;
483 	}
484 
485 	if (euid != (uid_t) -1) {
486 		new->euid = keuid;
487 		if (!uid_eq(old->uid, keuid) &&
488 		    !uid_eq(old->euid, keuid) &&
489 		    !uid_eq(old->suid, keuid) &&
490 		    !ns_capable(old->user_ns, CAP_SETUID))
491 			goto error;
492 	}
493 
494 	if (!uid_eq(new->uid, old->uid)) {
495 		retval = set_user(new);
496 		if (retval < 0)
497 			goto error;
498 	}
499 	if (ruid != (uid_t) -1 ||
500 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
501 		new->suid = new->euid;
502 	new->fsuid = new->euid;
503 
504 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
505 	if (retval < 0)
506 		goto error;
507 
508 	return commit_creds(new);
509 
510 error:
511 	abort_creds(new);
512 	return retval;
513 }
514 
515 /*
516  * setuid() is implemented like SysV with SAVED_IDS
517  *
518  * Note that SAVED_ID's is deficient in that a setuid root program
519  * like sendmail, for example, cannot set its uid to be a normal
520  * user and then switch back, because if you're root, setuid() sets
521  * the saved uid too.  If you don't like this, blame the bright people
522  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
523  * will allow a root program to temporarily drop privileges and be able to
524  * regain them by swapping the real and effective uid.
525  */
526 SYSCALL_DEFINE1(setuid, uid_t, uid)
527 {
528 	struct user_namespace *ns = current_user_ns();
529 	const struct cred *old;
530 	struct cred *new;
531 	int retval;
532 	kuid_t kuid;
533 
534 	kuid = make_kuid(ns, uid);
535 	if (!uid_valid(kuid))
536 		return -EINVAL;
537 
538 	new = prepare_creds();
539 	if (!new)
540 		return -ENOMEM;
541 	old = current_cred();
542 
543 	retval = -EPERM;
544 	if (ns_capable(old->user_ns, CAP_SETUID)) {
545 		new->suid = new->uid = kuid;
546 		if (!uid_eq(kuid, old->uid)) {
547 			retval = set_user(new);
548 			if (retval < 0)
549 				goto error;
550 		}
551 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
552 		goto error;
553 	}
554 
555 	new->fsuid = new->euid = kuid;
556 
557 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
558 	if (retval < 0)
559 		goto error;
560 
561 	return commit_creds(new);
562 
563 error:
564 	abort_creds(new);
565 	return retval;
566 }
567 
568 
569 /*
570  * This function implements a generic ability to update ruid, euid,
571  * and suid.  This allows you to implement the 4.4 compatible seteuid().
572  */
573 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
574 {
575 	struct user_namespace *ns = current_user_ns();
576 	const struct cred *old;
577 	struct cred *new;
578 	int retval;
579 	kuid_t kruid, keuid, ksuid;
580 
581 	kruid = make_kuid(ns, ruid);
582 	keuid = make_kuid(ns, euid);
583 	ksuid = make_kuid(ns, suid);
584 
585 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
586 		return -EINVAL;
587 
588 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
589 		return -EINVAL;
590 
591 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
592 		return -EINVAL;
593 
594 	new = prepare_creds();
595 	if (!new)
596 		return -ENOMEM;
597 
598 	old = current_cred();
599 
600 	retval = -EPERM;
601 	if (!ns_capable(old->user_ns, CAP_SETUID)) {
602 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
603 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
604 			goto error;
605 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
606 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
607 			goto error;
608 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
609 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
610 			goto error;
611 	}
612 
613 	if (ruid != (uid_t) -1) {
614 		new->uid = kruid;
615 		if (!uid_eq(kruid, old->uid)) {
616 			retval = set_user(new);
617 			if (retval < 0)
618 				goto error;
619 		}
620 	}
621 	if (euid != (uid_t) -1)
622 		new->euid = keuid;
623 	if (suid != (uid_t) -1)
624 		new->suid = ksuid;
625 	new->fsuid = new->euid;
626 
627 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
628 	if (retval < 0)
629 		goto error;
630 
631 	return commit_creds(new);
632 
633 error:
634 	abort_creds(new);
635 	return retval;
636 }
637 
638 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
639 {
640 	const struct cred *cred = current_cred();
641 	int retval;
642 	uid_t ruid, euid, suid;
643 
644 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
645 	euid = from_kuid_munged(cred->user_ns, cred->euid);
646 	suid = from_kuid_munged(cred->user_ns, cred->suid);
647 
648 	retval = put_user(ruid, ruidp);
649 	if (!retval) {
650 		retval = put_user(euid, euidp);
651 		if (!retval)
652 			return put_user(suid, suidp);
653 	}
654 	return retval;
655 }
656 
657 /*
658  * Same as above, but for rgid, egid, sgid.
659  */
660 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
661 {
662 	struct user_namespace *ns = current_user_ns();
663 	const struct cred *old;
664 	struct cred *new;
665 	int retval;
666 	kgid_t krgid, kegid, ksgid;
667 
668 	krgid = make_kgid(ns, rgid);
669 	kegid = make_kgid(ns, egid);
670 	ksgid = make_kgid(ns, sgid);
671 
672 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
673 		return -EINVAL;
674 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
675 		return -EINVAL;
676 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
677 		return -EINVAL;
678 
679 	new = prepare_creds();
680 	if (!new)
681 		return -ENOMEM;
682 	old = current_cred();
683 
684 	retval = -EPERM;
685 	if (!ns_capable(old->user_ns, CAP_SETGID)) {
686 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
687 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
688 			goto error;
689 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
690 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
691 			goto error;
692 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
693 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
694 			goto error;
695 	}
696 
697 	if (rgid != (gid_t) -1)
698 		new->gid = krgid;
699 	if (egid != (gid_t) -1)
700 		new->egid = kegid;
701 	if (sgid != (gid_t) -1)
702 		new->sgid = ksgid;
703 	new->fsgid = new->egid;
704 
705 	return commit_creds(new);
706 
707 error:
708 	abort_creds(new);
709 	return retval;
710 }
711 
712 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
713 {
714 	const struct cred *cred = current_cred();
715 	int retval;
716 	gid_t rgid, egid, sgid;
717 
718 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
719 	egid = from_kgid_munged(cred->user_ns, cred->egid);
720 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
721 
722 	retval = put_user(rgid, rgidp);
723 	if (!retval) {
724 		retval = put_user(egid, egidp);
725 		if (!retval)
726 			retval = put_user(sgid, sgidp);
727 	}
728 
729 	return retval;
730 }
731 
732 
733 /*
734  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
735  * is used for "access()" and for the NFS daemon (letting nfsd stay at
736  * whatever uid it wants to). It normally shadows "euid", except when
737  * explicitly set by setfsuid() or for access..
738  */
739 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
740 {
741 	const struct cred *old;
742 	struct cred *new;
743 	uid_t old_fsuid;
744 	kuid_t kuid;
745 
746 	old = current_cred();
747 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
748 
749 	kuid = make_kuid(old->user_ns, uid);
750 	if (!uid_valid(kuid))
751 		return old_fsuid;
752 
753 	new = prepare_creds();
754 	if (!new)
755 		return old_fsuid;
756 
757 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
758 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
759 	    ns_capable(old->user_ns, CAP_SETUID)) {
760 		if (!uid_eq(kuid, old->fsuid)) {
761 			new->fsuid = kuid;
762 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
763 				goto change_okay;
764 		}
765 	}
766 
767 	abort_creds(new);
768 	return old_fsuid;
769 
770 change_okay:
771 	commit_creds(new);
772 	return old_fsuid;
773 }
774 
775 /*
776  * Samma på svenska..
777  */
778 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
779 {
780 	const struct cred *old;
781 	struct cred *new;
782 	gid_t old_fsgid;
783 	kgid_t kgid;
784 
785 	old = current_cred();
786 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
787 
788 	kgid = make_kgid(old->user_ns, gid);
789 	if (!gid_valid(kgid))
790 		return old_fsgid;
791 
792 	new = prepare_creds();
793 	if (!new)
794 		return old_fsgid;
795 
796 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
797 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
798 	    ns_capable(old->user_ns, CAP_SETGID)) {
799 		if (!gid_eq(kgid, old->fsgid)) {
800 			new->fsgid = kgid;
801 			goto change_okay;
802 		}
803 	}
804 
805 	abort_creds(new);
806 	return old_fsgid;
807 
808 change_okay:
809 	commit_creds(new);
810 	return old_fsgid;
811 }
812 
813 /**
814  * sys_getpid - return the thread group id of the current process
815  *
816  * Note, despite the name, this returns the tgid not the pid.  The tgid and
817  * the pid are identical unless CLONE_THREAD was specified on clone() in
818  * which case the tgid is the same in all threads of the same group.
819  *
820  * This is SMP safe as current->tgid does not change.
821  */
822 SYSCALL_DEFINE0(getpid)
823 {
824 	return task_tgid_vnr(current);
825 }
826 
827 /* Thread ID - the internal kernel "pid" */
828 SYSCALL_DEFINE0(gettid)
829 {
830 	return task_pid_vnr(current);
831 }
832 
833 /*
834  * Accessing ->real_parent is not SMP-safe, it could
835  * change from under us. However, we can use a stale
836  * value of ->real_parent under rcu_read_lock(), see
837  * release_task()->call_rcu(delayed_put_task_struct).
838  */
839 SYSCALL_DEFINE0(getppid)
840 {
841 	int pid;
842 
843 	rcu_read_lock();
844 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
845 	rcu_read_unlock();
846 
847 	return pid;
848 }
849 
850 SYSCALL_DEFINE0(getuid)
851 {
852 	/* Only we change this so SMP safe */
853 	return from_kuid_munged(current_user_ns(), current_uid());
854 }
855 
856 SYSCALL_DEFINE0(geteuid)
857 {
858 	/* Only we change this so SMP safe */
859 	return from_kuid_munged(current_user_ns(), current_euid());
860 }
861 
862 SYSCALL_DEFINE0(getgid)
863 {
864 	/* Only we change this so SMP safe */
865 	return from_kgid_munged(current_user_ns(), current_gid());
866 }
867 
868 SYSCALL_DEFINE0(getegid)
869 {
870 	/* Only we change this so SMP safe */
871 	return from_kgid_munged(current_user_ns(), current_egid());
872 }
873 
874 void do_sys_times(struct tms *tms)
875 {
876 	cputime_t tgutime, tgstime, cutime, cstime;
877 
878 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
879 	cutime = current->signal->cutime;
880 	cstime = current->signal->cstime;
881 	tms->tms_utime = cputime_to_clock_t(tgutime);
882 	tms->tms_stime = cputime_to_clock_t(tgstime);
883 	tms->tms_cutime = cputime_to_clock_t(cutime);
884 	tms->tms_cstime = cputime_to_clock_t(cstime);
885 }
886 
887 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
888 {
889 	if (tbuf) {
890 		struct tms tmp;
891 
892 		do_sys_times(&tmp);
893 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
894 			return -EFAULT;
895 	}
896 	force_successful_syscall_return();
897 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
898 }
899 
900 /*
901  * This needs some heavy checking ...
902  * I just haven't the stomach for it. I also don't fully
903  * understand sessions/pgrp etc. Let somebody who does explain it.
904  *
905  * OK, I think I have the protection semantics right.... this is really
906  * only important on a multi-user system anyway, to make sure one user
907  * can't send a signal to a process owned by another.  -TYT, 12/12/91
908  *
909  * !PF_FORKNOEXEC check to conform completely to POSIX.
910  */
911 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
912 {
913 	struct task_struct *p;
914 	struct task_struct *group_leader = current->group_leader;
915 	struct pid *pgrp;
916 	int err;
917 
918 	if (!pid)
919 		pid = task_pid_vnr(group_leader);
920 	if (!pgid)
921 		pgid = pid;
922 	if (pgid < 0)
923 		return -EINVAL;
924 	rcu_read_lock();
925 
926 	/* From this point forward we keep holding onto the tasklist lock
927 	 * so that our parent does not change from under us. -DaveM
928 	 */
929 	write_lock_irq(&tasklist_lock);
930 
931 	err = -ESRCH;
932 	p = find_task_by_vpid(pid);
933 	if (!p)
934 		goto out;
935 
936 	err = -EINVAL;
937 	if (!thread_group_leader(p))
938 		goto out;
939 
940 	if (same_thread_group(p->real_parent, group_leader)) {
941 		err = -EPERM;
942 		if (task_session(p) != task_session(group_leader))
943 			goto out;
944 		err = -EACCES;
945 		if (!(p->flags & PF_FORKNOEXEC))
946 			goto out;
947 	} else {
948 		err = -ESRCH;
949 		if (p != group_leader)
950 			goto out;
951 	}
952 
953 	err = -EPERM;
954 	if (p->signal->leader)
955 		goto out;
956 
957 	pgrp = task_pid(p);
958 	if (pgid != pid) {
959 		struct task_struct *g;
960 
961 		pgrp = find_vpid(pgid);
962 		g = pid_task(pgrp, PIDTYPE_PGID);
963 		if (!g || task_session(g) != task_session(group_leader))
964 			goto out;
965 	}
966 
967 	err = security_task_setpgid(p, pgid);
968 	if (err)
969 		goto out;
970 
971 	if (task_pgrp(p) != pgrp)
972 		change_pid(p, PIDTYPE_PGID, pgrp);
973 
974 	err = 0;
975 out:
976 	/* All paths lead to here, thus we are safe. -DaveM */
977 	write_unlock_irq(&tasklist_lock);
978 	rcu_read_unlock();
979 	return err;
980 }
981 
982 SYSCALL_DEFINE1(getpgid, pid_t, pid)
983 {
984 	struct task_struct *p;
985 	struct pid *grp;
986 	int retval;
987 
988 	rcu_read_lock();
989 	if (!pid)
990 		grp = task_pgrp(current);
991 	else {
992 		retval = -ESRCH;
993 		p = find_task_by_vpid(pid);
994 		if (!p)
995 			goto out;
996 		grp = task_pgrp(p);
997 		if (!grp)
998 			goto out;
999 
1000 		retval = security_task_getpgid(p);
1001 		if (retval)
1002 			goto out;
1003 	}
1004 	retval = pid_vnr(grp);
1005 out:
1006 	rcu_read_unlock();
1007 	return retval;
1008 }
1009 
1010 #ifdef __ARCH_WANT_SYS_GETPGRP
1011 
1012 SYSCALL_DEFINE0(getpgrp)
1013 {
1014 	return sys_getpgid(0);
1015 }
1016 
1017 #endif
1018 
1019 SYSCALL_DEFINE1(getsid, pid_t, pid)
1020 {
1021 	struct task_struct *p;
1022 	struct pid *sid;
1023 	int retval;
1024 
1025 	rcu_read_lock();
1026 	if (!pid)
1027 		sid = task_session(current);
1028 	else {
1029 		retval = -ESRCH;
1030 		p = find_task_by_vpid(pid);
1031 		if (!p)
1032 			goto out;
1033 		sid = task_session(p);
1034 		if (!sid)
1035 			goto out;
1036 
1037 		retval = security_task_getsid(p);
1038 		if (retval)
1039 			goto out;
1040 	}
1041 	retval = pid_vnr(sid);
1042 out:
1043 	rcu_read_unlock();
1044 	return retval;
1045 }
1046 
1047 static void set_special_pids(struct pid *pid)
1048 {
1049 	struct task_struct *curr = current->group_leader;
1050 
1051 	if (task_session(curr) != pid)
1052 		change_pid(curr, PIDTYPE_SID, pid);
1053 
1054 	if (task_pgrp(curr) != pid)
1055 		change_pid(curr, PIDTYPE_PGID, pid);
1056 }
1057 
1058 SYSCALL_DEFINE0(setsid)
1059 {
1060 	struct task_struct *group_leader = current->group_leader;
1061 	struct pid *sid = task_pid(group_leader);
1062 	pid_t session = pid_vnr(sid);
1063 	int err = -EPERM;
1064 
1065 	write_lock_irq(&tasklist_lock);
1066 	/* Fail if I am already a session leader */
1067 	if (group_leader->signal->leader)
1068 		goto out;
1069 
1070 	/* Fail if a process group id already exists that equals the
1071 	 * proposed session id.
1072 	 */
1073 	if (pid_task(sid, PIDTYPE_PGID))
1074 		goto out;
1075 
1076 	group_leader->signal->leader = 1;
1077 	set_special_pids(sid);
1078 
1079 	proc_clear_tty(group_leader);
1080 
1081 	err = session;
1082 out:
1083 	write_unlock_irq(&tasklist_lock);
1084 	if (err > 0) {
1085 		proc_sid_connector(group_leader);
1086 		sched_autogroup_create_attach(group_leader);
1087 	}
1088 	return err;
1089 }
1090 
1091 DECLARE_RWSEM(uts_sem);
1092 
1093 #ifdef COMPAT_UTS_MACHINE
1094 #define override_architecture(name) \
1095 	(personality(current->personality) == PER_LINUX32 && \
1096 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1097 		      sizeof(COMPAT_UTS_MACHINE)))
1098 #else
1099 #define override_architecture(name)	0
1100 #endif
1101 
1102 /*
1103  * Work around broken programs that cannot handle "Linux 3.0".
1104  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1105  */
1106 static int override_release(char __user *release, size_t len)
1107 {
1108 	int ret = 0;
1109 
1110 	if (current->personality & UNAME26) {
1111 		const char *rest = UTS_RELEASE;
1112 		char buf[65] = { 0 };
1113 		int ndots = 0;
1114 		unsigned v;
1115 		size_t copy;
1116 
1117 		while (*rest) {
1118 			if (*rest == '.' && ++ndots >= 3)
1119 				break;
1120 			if (!isdigit(*rest) && *rest != '.')
1121 				break;
1122 			rest++;
1123 		}
1124 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1125 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1126 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1127 		ret = copy_to_user(release, buf, copy + 1);
1128 	}
1129 	return ret;
1130 }
1131 
1132 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1133 {
1134 	int errno = 0;
1135 
1136 	down_read(&uts_sem);
1137 	if (copy_to_user(name, utsname(), sizeof *name))
1138 		errno = -EFAULT;
1139 	up_read(&uts_sem);
1140 
1141 	if (!errno && override_release(name->release, sizeof(name->release)))
1142 		errno = -EFAULT;
1143 	if (!errno && override_architecture(name))
1144 		errno = -EFAULT;
1145 	return errno;
1146 }
1147 
1148 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1149 /*
1150  * Old cruft
1151  */
1152 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1153 {
1154 	int error = 0;
1155 
1156 	if (!name)
1157 		return -EFAULT;
1158 
1159 	down_read(&uts_sem);
1160 	if (copy_to_user(name, utsname(), sizeof(*name)))
1161 		error = -EFAULT;
1162 	up_read(&uts_sem);
1163 
1164 	if (!error && override_release(name->release, sizeof(name->release)))
1165 		error = -EFAULT;
1166 	if (!error && override_architecture(name))
1167 		error = -EFAULT;
1168 	return error;
1169 }
1170 
1171 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1172 {
1173 	int error;
1174 
1175 	if (!name)
1176 		return -EFAULT;
1177 	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1178 		return -EFAULT;
1179 
1180 	down_read(&uts_sem);
1181 	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1182 			       __OLD_UTS_LEN);
1183 	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1184 	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1185 				__OLD_UTS_LEN);
1186 	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1187 	error |= __copy_to_user(&name->release, &utsname()->release,
1188 				__OLD_UTS_LEN);
1189 	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1190 	error |= __copy_to_user(&name->version, &utsname()->version,
1191 				__OLD_UTS_LEN);
1192 	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1193 	error |= __copy_to_user(&name->machine, &utsname()->machine,
1194 				__OLD_UTS_LEN);
1195 	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1196 	up_read(&uts_sem);
1197 
1198 	if (!error && override_architecture(name))
1199 		error = -EFAULT;
1200 	if (!error && override_release(name->release, sizeof(name->release)))
1201 		error = -EFAULT;
1202 	return error ? -EFAULT : 0;
1203 }
1204 #endif
1205 
1206 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1207 {
1208 	int errno;
1209 	char tmp[__NEW_UTS_LEN];
1210 
1211 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1212 		return -EPERM;
1213 
1214 	if (len < 0 || len > __NEW_UTS_LEN)
1215 		return -EINVAL;
1216 	down_write(&uts_sem);
1217 	errno = -EFAULT;
1218 	if (!copy_from_user(tmp, name, len)) {
1219 		struct new_utsname *u = utsname();
1220 
1221 		memcpy(u->nodename, tmp, len);
1222 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1223 		errno = 0;
1224 		uts_proc_notify(UTS_PROC_HOSTNAME);
1225 	}
1226 	up_write(&uts_sem);
1227 	return errno;
1228 }
1229 
1230 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1231 
1232 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1233 {
1234 	int i, errno;
1235 	struct new_utsname *u;
1236 
1237 	if (len < 0)
1238 		return -EINVAL;
1239 	down_read(&uts_sem);
1240 	u = utsname();
1241 	i = 1 + strlen(u->nodename);
1242 	if (i > len)
1243 		i = len;
1244 	errno = 0;
1245 	if (copy_to_user(name, u->nodename, i))
1246 		errno = -EFAULT;
1247 	up_read(&uts_sem);
1248 	return errno;
1249 }
1250 
1251 #endif
1252 
1253 /*
1254  * Only setdomainname; getdomainname can be implemented by calling
1255  * uname()
1256  */
1257 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1258 {
1259 	int errno;
1260 	char tmp[__NEW_UTS_LEN];
1261 
1262 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1263 		return -EPERM;
1264 	if (len < 0 || len > __NEW_UTS_LEN)
1265 		return -EINVAL;
1266 
1267 	down_write(&uts_sem);
1268 	errno = -EFAULT;
1269 	if (!copy_from_user(tmp, name, len)) {
1270 		struct new_utsname *u = utsname();
1271 
1272 		memcpy(u->domainname, tmp, len);
1273 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1274 		errno = 0;
1275 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1276 	}
1277 	up_write(&uts_sem);
1278 	return errno;
1279 }
1280 
1281 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1282 {
1283 	struct rlimit value;
1284 	int ret;
1285 
1286 	ret = do_prlimit(current, resource, NULL, &value);
1287 	if (!ret)
1288 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1289 
1290 	return ret;
1291 }
1292 
1293 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1294 
1295 /*
1296  *	Back compatibility for getrlimit. Needed for some apps.
1297  */
1298 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1299 		struct rlimit __user *, rlim)
1300 {
1301 	struct rlimit x;
1302 	if (resource >= RLIM_NLIMITS)
1303 		return -EINVAL;
1304 
1305 	task_lock(current->group_leader);
1306 	x = current->signal->rlim[resource];
1307 	task_unlock(current->group_leader);
1308 	if (x.rlim_cur > 0x7FFFFFFF)
1309 		x.rlim_cur = 0x7FFFFFFF;
1310 	if (x.rlim_max > 0x7FFFFFFF)
1311 		x.rlim_max = 0x7FFFFFFF;
1312 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1313 }
1314 
1315 #endif
1316 
1317 static inline bool rlim64_is_infinity(__u64 rlim64)
1318 {
1319 #if BITS_PER_LONG < 64
1320 	return rlim64 >= ULONG_MAX;
1321 #else
1322 	return rlim64 == RLIM64_INFINITY;
1323 #endif
1324 }
1325 
1326 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1327 {
1328 	if (rlim->rlim_cur == RLIM_INFINITY)
1329 		rlim64->rlim_cur = RLIM64_INFINITY;
1330 	else
1331 		rlim64->rlim_cur = rlim->rlim_cur;
1332 	if (rlim->rlim_max == RLIM_INFINITY)
1333 		rlim64->rlim_max = RLIM64_INFINITY;
1334 	else
1335 		rlim64->rlim_max = rlim->rlim_max;
1336 }
1337 
1338 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1339 {
1340 	if (rlim64_is_infinity(rlim64->rlim_cur))
1341 		rlim->rlim_cur = RLIM_INFINITY;
1342 	else
1343 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1344 	if (rlim64_is_infinity(rlim64->rlim_max))
1345 		rlim->rlim_max = RLIM_INFINITY;
1346 	else
1347 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1348 }
1349 
1350 /* make sure you are allowed to change @tsk limits before calling this */
1351 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1352 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1353 {
1354 	struct rlimit *rlim;
1355 	int retval = 0;
1356 
1357 	if (resource >= RLIM_NLIMITS)
1358 		return -EINVAL;
1359 	if (new_rlim) {
1360 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1361 			return -EINVAL;
1362 		if (resource == RLIMIT_NOFILE &&
1363 				new_rlim->rlim_max > sysctl_nr_open)
1364 			return -EPERM;
1365 	}
1366 
1367 	/* protect tsk->signal and tsk->sighand from disappearing */
1368 	read_lock(&tasklist_lock);
1369 	if (!tsk->sighand) {
1370 		retval = -ESRCH;
1371 		goto out;
1372 	}
1373 
1374 	rlim = tsk->signal->rlim + resource;
1375 	task_lock(tsk->group_leader);
1376 	if (new_rlim) {
1377 		/* Keep the capable check against init_user_ns until
1378 		   cgroups can contain all limits */
1379 		if (new_rlim->rlim_max > rlim->rlim_max &&
1380 				!capable(CAP_SYS_RESOURCE))
1381 			retval = -EPERM;
1382 		if (!retval)
1383 			retval = security_task_setrlimit(tsk->group_leader,
1384 					resource, new_rlim);
1385 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1386 			/*
1387 			 * The caller is asking for an immediate RLIMIT_CPU
1388 			 * expiry.  But we use the zero value to mean "it was
1389 			 * never set".  So let's cheat and make it one second
1390 			 * instead
1391 			 */
1392 			new_rlim->rlim_cur = 1;
1393 		}
1394 	}
1395 	if (!retval) {
1396 		if (old_rlim)
1397 			*old_rlim = *rlim;
1398 		if (new_rlim)
1399 			*rlim = *new_rlim;
1400 	}
1401 	task_unlock(tsk->group_leader);
1402 
1403 	/*
1404 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1405 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1406 	 * very long-standing error, and fixing it now risks breakage of
1407 	 * applications, so we live with it
1408 	 */
1409 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1410 			 new_rlim->rlim_cur != RLIM_INFINITY)
1411 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1412 out:
1413 	read_unlock(&tasklist_lock);
1414 	return retval;
1415 }
1416 
1417 /* rcu lock must be held */
1418 static int check_prlimit_permission(struct task_struct *task)
1419 {
1420 	const struct cred *cred = current_cred(), *tcred;
1421 
1422 	if (current == task)
1423 		return 0;
1424 
1425 	tcred = __task_cred(task);
1426 	if (uid_eq(cred->uid, tcred->euid) &&
1427 	    uid_eq(cred->uid, tcred->suid) &&
1428 	    uid_eq(cred->uid, tcred->uid)  &&
1429 	    gid_eq(cred->gid, tcred->egid) &&
1430 	    gid_eq(cred->gid, tcred->sgid) &&
1431 	    gid_eq(cred->gid, tcred->gid))
1432 		return 0;
1433 	if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1434 		return 0;
1435 
1436 	return -EPERM;
1437 }
1438 
1439 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1440 		const struct rlimit64 __user *, new_rlim,
1441 		struct rlimit64 __user *, old_rlim)
1442 {
1443 	struct rlimit64 old64, new64;
1444 	struct rlimit old, new;
1445 	struct task_struct *tsk;
1446 	int ret;
1447 
1448 	if (new_rlim) {
1449 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1450 			return -EFAULT;
1451 		rlim64_to_rlim(&new64, &new);
1452 	}
1453 
1454 	rcu_read_lock();
1455 	tsk = pid ? find_task_by_vpid(pid) : current;
1456 	if (!tsk) {
1457 		rcu_read_unlock();
1458 		return -ESRCH;
1459 	}
1460 	ret = check_prlimit_permission(tsk);
1461 	if (ret) {
1462 		rcu_read_unlock();
1463 		return ret;
1464 	}
1465 	get_task_struct(tsk);
1466 	rcu_read_unlock();
1467 
1468 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1469 			old_rlim ? &old : NULL);
1470 
1471 	if (!ret && old_rlim) {
1472 		rlim_to_rlim64(&old, &old64);
1473 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1474 			ret = -EFAULT;
1475 	}
1476 
1477 	put_task_struct(tsk);
1478 	return ret;
1479 }
1480 
1481 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1482 {
1483 	struct rlimit new_rlim;
1484 
1485 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1486 		return -EFAULT;
1487 	return do_prlimit(current, resource, &new_rlim, NULL);
1488 }
1489 
1490 /*
1491  * It would make sense to put struct rusage in the task_struct,
1492  * except that would make the task_struct be *really big*.  After
1493  * task_struct gets moved into malloc'ed memory, it would
1494  * make sense to do this.  It will make moving the rest of the information
1495  * a lot simpler!  (Which we're not doing right now because we're not
1496  * measuring them yet).
1497  *
1498  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1499  * races with threads incrementing their own counters.  But since word
1500  * reads are atomic, we either get new values or old values and we don't
1501  * care which for the sums.  We always take the siglock to protect reading
1502  * the c* fields from p->signal from races with exit.c updating those
1503  * fields when reaping, so a sample either gets all the additions of a
1504  * given child after it's reaped, or none so this sample is before reaping.
1505  *
1506  * Locking:
1507  * We need to take the siglock for CHILDEREN, SELF and BOTH
1508  * for  the cases current multithreaded, non-current single threaded
1509  * non-current multithreaded.  Thread traversal is now safe with
1510  * the siglock held.
1511  * Strictly speaking, we donot need to take the siglock if we are current and
1512  * single threaded,  as no one else can take our signal_struct away, no one
1513  * else can  reap the  children to update signal->c* counters, and no one else
1514  * can race with the signal-> fields. If we do not take any lock, the
1515  * signal-> fields could be read out of order while another thread was just
1516  * exiting. So we should  place a read memory barrier when we avoid the lock.
1517  * On the writer side,  write memory barrier is implied in  __exit_signal
1518  * as __exit_signal releases  the siglock spinlock after updating the signal->
1519  * fields. But we don't do this yet to keep things simple.
1520  *
1521  */
1522 
1523 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1524 {
1525 	r->ru_nvcsw += t->nvcsw;
1526 	r->ru_nivcsw += t->nivcsw;
1527 	r->ru_minflt += t->min_flt;
1528 	r->ru_majflt += t->maj_flt;
1529 	r->ru_inblock += task_io_get_inblock(t);
1530 	r->ru_oublock += task_io_get_oublock(t);
1531 }
1532 
1533 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1534 {
1535 	struct task_struct *t;
1536 	unsigned long flags;
1537 	cputime_t tgutime, tgstime, utime, stime;
1538 	unsigned long maxrss = 0;
1539 
1540 	memset((char *)r, 0, sizeof (*r));
1541 	utime = stime = 0;
1542 
1543 	if (who == RUSAGE_THREAD) {
1544 		task_cputime_adjusted(current, &utime, &stime);
1545 		accumulate_thread_rusage(p, r);
1546 		maxrss = p->signal->maxrss;
1547 		goto out;
1548 	}
1549 
1550 	if (!lock_task_sighand(p, &flags))
1551 		return;
1552 
1553 	switch (who) {
1554 	case RUSAGE_BOTH:
1555 	case RUSAGE_CHILDREN:
1556 		utime = p->signal->cutime;
1557 		stime = p->signal->cstime;
1558 		r->ru_nvcsw = p->signal->cnvcsw;
1559 		r->ru_nivcsw = p->signal->cnivcsw;
1560 		r->ru_minflt = p->signal->cmin_flt;
1561 		r->ru_majflt = p->signal->cmaj_flt;
1562 		r->ru_inblock = p->signal->cinblock;
1563 		r->ru_oublock = p->signal->coublock;
1564 		maxrss = p->signal->cmaxrss;
1565 
1566 		if (who == RUSAGE_CHILDREN)
1567 			break;
1568 
1569 	case RUSAGE_SELF:
1570 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1571 		utime += tgutime;
1572 		stime += tgstime;
1573 		r->ru_nvcsw += p->signal->nvcsw;
1574 		r->ru_nivcsw += p->signal->nivcsw;
1575 		r->ru_minflt += p->signal->min_flt;
1576 		r->ru_majflt += p->signal->maj_flt;
1577 		r->ru_inblock += p->signal->inblock;
1578 		r->ru_oublock += p->signal->oublock;
1579 		if (maxrss < p->signal->maxrss)
1580 			maxrss = p->signal->maxrss;
1581 		t = p;
1582 		do {
1583 			accumulate_thread_rusage(t, r);
1584 		} while_each_thread(p, t);
1585 		break;
1586 
1587 	default:
1588 		BUG();
1589 	}
1590 	unlock_task_sighand(p, &flags);
1591 
1592 out:
1593 	cputime_to_timeval(utime, &r->ru_utime);
1594 	cputime_to_timeval(stime, &r->ru_stime);
1595 
1596 	if (who != RUSAGE_CHILDREN) {
1597 		struct mm_struct *mm = get_task_mm(p);
1598 
1599 		if (mm) {
1600 			setmax_mm_hiwater_rss(&maxrss, mm);
1601 			mmput(mm);
1602 		}
1603 	}
1604 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1605 }
1606 
1607 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1608 {
1609 	struct rusage r;
1610 
1611 	k_getrusage(p, who, &r);
1612 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1613 }
1614 
1615 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1616 {
1617 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1618 	    who != RUSAGE_THREAD)
1619 		return -EINVAL;
1620 	return getrusage(current, who, ru);
1621 }
1622 
1623 #ifdef CONFIG_COMPAT
1624 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1625 {
1626 	struct rusage r;
1627 
1628 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1629 	    who != RUSAGE_THREAD)
1630 		return -EINVAL;
1631 
1632 	k_getrusage(current, who, &r);
1633 	return put_compat_rusage(&r, ru);
1634 }
1635 #endif
1636 
1637 SYSCALL_DEFINE1(umask, int, mask)
1638 {
1639 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1640 	return mask;
1641 }
1642 
1643 static int prctl_set_mm_exe_file_locked(struct mm_struct *mm, unsigned int fd)
1644 {
1645 	struct fd exe;
1646 	struct inode *inode;
1647 	int err;
1648 
1649 	VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
1650 
1651 	exe = fdget(fd);
1652 	if (!exe.file)
1653 		return -EBADF;
1654 
1655 	inode = file_inode(exe.file);
1656 
1657 	/*
1658 	 * Because the original mm->exe_file points to executable file, make
1659 	 * sure that this one is executable as well, to avoid breaking an
1660 	 * overall picture.
1661 	 */
1662 	err = -EACCES;
1663 	if (!S_ISREG(inode->i_mode)	||
1664 	    exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1665 		goto exit;
1666 
1667 	err = inode_permission(inode, MAY_EXEC);
1668 	if (err)
1669 		goto exit;
1670 
1671 	/*
1672 	 * Forbid mm->exe_file change if old file still mapped.
1673 	 */
1674 	err = -EBUSY;
1675 	if (mm->exe_file) {
1676 		struct vm_area_struct *vma;
1677 
1678 		for (vma = mm->mmap; vma; vma = vma->vm_next)
1679 			if (vma->vm_file &&
1680 			    path_equal(&vma->vm_file->f_path,
1681 				       &mm->exe_file->f_path))
1682 				goto exit;
1683 	}
1684 
1685 	/*
1686 	 * The symlink can be changed only once, just to disallow arbitrary
1687 	 * transitions malicious software might bring in. This means one
1688 	 * could make a snapshot over all processes running and monitor
1689 	 * /proc/pid/exe changes to notice unusual activity if needed.
1690 	 */
1691 	err = -EPERM;
1692 	if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1693 		goto exit;
1694 
1695 	err = 0;
1696 	set_mm_exe_file(mm, exe.file);	/* this grabs a reference to exe.file */
1697 exit:
1698 	fdput(exe);
1699 	return err;
1700 }
1701 
1702 #ifdef CONFIG_CHECKPOINT_RESTORE
1703 /*
1704  * WARNING: we don't require any capability here so be very careful
1705  * in what is allowed for modification from userspace.
1706  */
1707 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1708 {
1709 	unsigned long mmap_max_addr = TASK_SIZE;
1710 	struct mm_struct *mm = current->mm;
1711 	int error = -EINVAL, i;
1712 
1713 	static const unsigned char offsets[] = {
1714 		offsetof(struct prctl_mm_map, start_code),
1715 		offsetof(struct prctl_mm_map, end_code),
1716 		offsetof(struct prctl_mm_map, start_data),
1717 		offsetof(struct prctl_mm_map, end_data),
1718 		offsetof(struct prctl_mm_map, start_brk),
1719 		offsetof(struct prctl_mm_map, brk),
1720 		offsetof(struct prctl_mm_map, start_stack),
1721 		offsetof(struct prctl_mm_map, arg_start),
1722 		offsetof(struct prctl_mm_map, arg_end),
1723 		offsetof(struct prctl_mm_map, env_start),
1724 		offsetof(struct prctl_mm_map, env_end),
1725 	};
1726 
1727 	/*
1728 	 * Make sure the members are not somewhere outside
1729 	 * of allowed address space.
1730 	 */
1731 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1732 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1733 
1734 		if ((unsigned long)val >= mmap_max_addr ||
1735 		    (unsigned long)val < mmap_min_addr)
1736 			goto out;
1737 	}
1738 
1739 	/*
1740 	 * Make sure the pairs are ordered.
1741 	 */
1742 #define __prctl_check_order(__m1, __op, __m2)				\
1743 	((unsigned long)prctl_map->__m1 __op				\
1744 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1745 	error  = __prctl_check_order(start_code, <, end_code);
1746 	error |= __prctl_check_order(start_data, <, end_data);
1747 	error |= __prctl_check_order(start_brk, <=, brk);
1748 	error |= __prctl_check_order(arg_start, <=, arg_end);
1749 	error |= __prctl_check_order(env_start, <=, env_end);
1750 	if (error)
1751 		goto out;
1752 #undef __prctl_check_order
1753 
1754 	error = -EINVAL;
1755 
1756 	/*
1757 	 * @brk should be after @end_data in traditional maps.
1758 	 */
1759 	if (prctl_map->start_brk <= prctl_map->end_data ||
1760 	    prctl_map->brk <= prctl_map->end_data)
1761 		goto out;
1762 
1763 	/*
1764 	 * Neither we should allow to override limits if they set.
1765 	 */
1766 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1767 			      prctl_map->start_brk, prctl_map->end_data,
1768 			      prctl_map->start_data))
1769 			goto out;
1770 
1771 	/*
1772 	 * Someone is trying to cheat the auxv vector.
1773 	 */
1774 	if (prctl_map->auxv_size) {
1775 		if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1776 			goto out;
1777 	}
1778 
1779 	/*
1780 	 * Finally, make sure the caller has the rights to
1781 	 * change /proc/pid/exe link: only local root should
1782 	 * be allowed to.
1783 	 */
1784 	if (prctl_map->exe_fd != (u32)-1) {
1785 		struct user_namespace *ns = current_user_ns();
1786 		const struct cred *cred = current_cred();
1787 
1788 		if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1789 		    !gid_eq(cred->gid, make_kgid(ns, 0)))
1790 			goto out;
1791 	}
1792 
1793 	error = 0;
1794 out:
1795 	return error;
1796 }
1797 
1798 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1799 {
1800 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1801 	unsigned long user_auxv[AT_VECTOR_SIZE];
1802 	struct mm_struct *mm = current->mm;
1803 	int error;
1804 
1805 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1806 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1807 
1808 	if (opt == PR_SET_MM_MAP_SIZE)
1809 		return put_user((unsigned int)sizeof(prctl_map),
1810 				(unsigned int __user *)addr);
1811 
1812 	if (data_size != sizeof(prctl_map))
1813 		return -EINVAL;
1814 
1815 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1816 		return -EFAULT;
1817 
1818 	error = validate_prctl_map(&prctl_map);
1819 	if (error)
1820 		return error;
1821 
1822 	if (prctl_map.auxv_size) {
1823 		memset(user_auxv, 0, sizeof(user_auxv));
1824 		if (copy_from_user(user_auxv,
1825 				   (const void __user *)prctl_map.auxv,
1826 				   prctl_map.auxv_size))
1827 			return -EFAULT;
1828 
1829 		/* Last entry must be AT_NULL as specification requires */
1830 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1831 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1832 	}
1833 
1834 	down_write(&mm->mmap_sem);
1835 	if (prctl_map.exe_fd != (u32)-1)
1836 		error = prctl_set_mm_exe_file_locked(mm, prctl_map.exe_fd);
1837 	downgrade_write(&mm->mmap_sem);
1838 	if (error)
1839 		goto out;
1840 
1841 	/*
1842 	 * We don't validate if these members are pointing to
1843 	 * real present VMAs because application may have correspond
1844 	 * VMAs already unmapped and kernel uses these members for statistics
1845 	 * output in procfs mostly, except
1846 	 *
1847 	 *  - @start_brk/@brk which are used in do_brk but kernel lookups
1848 	 *    for VMAs when updating these memvers so anything wrong written
1849 	 *    here cause kernel to swear at userspace program but won't lead
1850 	 *    to any problem in kernel itself
1851 	 */
1852 
1853 	mm->start_code	= prctl_map.start_code;
1854 	mm->end_code	= prctl_map.end_code;
1855 	mm->start_data	= prctl_map.start_data;
1856 	mm->end_data	= prctl_map.end_data;
1857 	mm->start_brk	= prctl_map.start_brk;
1858 	mm->brk		= prctl_map.brk;
1859 	mm->start_stack	= prctl_map.start_stack;
1860 	mm->arg_start	= prctl_map.arg_start;
1861 	mm->arg_end	= prctl_map.arg_end;
1862 	mm->env_start	= prctl_map.env_start;
1863 	mm->env_end	= prctl_map.env_end;
1864 
1865 	/*
1866 	 * Note this update of @saved_auxv is lockless thus
1867 	 * if someone reads this member in procfs while we're
1868 	 * updating -- it may get partly updated results. It's
1869 	 * known and acceptable trade off: we leave it as is to
1870 	 * not introduce additional locks here making the kernel
1871 	 * more complex.
1872 	 */
1873 	if (prctl_map.auxv_size)
1874 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1875 
1876 	error = 0;
1877 out:
1878 	up_read(&mm->mmap_sem);
1879 	return error;
1880 }
1881 #endif /* CONFIG_CHECKPOINT_RESTORE */
1882 
1883 static int prctl_set_mm(int opt, unsigned long addr,
1884 			unsigned long arg4, unsigned long arg5)
1885 {
1886 	struct mm_struct *mm = current->mm;
1887 	struct vm_area_struct *vma;
1888 	int error;
1889 
1890 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1891 			      opt != PR_SET_MM_MAP &&
1892 			      opt != PR_SET_MM_MAP_SIZE)))
1893 		return -EINVAL;
1894 
1895 #ifdef CONFIG_CHECKPOINT_RESTORE
1896 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1897 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1898 #endif
1899 
1900 	if (!capable(CAP_SYS_RESOURCE))
1901 		return -EPERM;
1902 
1903 	if (opt == PR_SET_MM_EXE_FILE) {
1904 		down_write(&mm->mmap_sem);
1905 		error = prctl_set_mm_exe_file_locked(mm, (unsigned int)addr);
1906 		up_write(&mm->mmap_sem);
1907 		return error;
1908 	}
1909 
1910 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
1911 		return -EINVAL;
1912 
1913 	error = -EINVAL;
1914 
1915 	down_read(&mm->mmap_sem);
1916 	vma = find_vma(mm, addr);
1917 
1918 	switch (opt) {
1919 	case PR_SET_MM_START_CODE:
1920 		mm->start_code = addr;
1921 		break;
1922 	case PR_SET_MM_END_CODE:
1923 		mm->end_code = addr;
1924 		break;
1925 	case PR_SET_MM_START_DATA:
1926 		mm->start_data = addr;
1927 		break;
1928 	case PR_SET_MM_END_DATA:
1929 		mm->end_data = addr;
1930 		break;
1931 
1932 	case PR_SET_MM_START_BRK:
1933 		if (addr <= mm->end_data)
1934 			goto out;
1935 
1936 		if (check_data_rlimit(rlimit(RLIMIT_DATA), mm->brk, addr,
1937 				      mm->end_data, mm->start_data))
1938 			goto out;
1939 
1940 		mm->start_brk = addr;
1941 		break;
1942 
1943 	case PR_SET_MM_BRK:
1944 		if (addr <= mm->end_data)
1945 			goto out;
1946 
1947 		if (check_data_rlimit(rlimit(RLIMIT_DATA), addr, mm->start_brk,
1948 				      mm->end_data, mm->start_data))
1949 			goto out;
1950 
1951 		mm->brk = addr;
1952 		break;
1953 
1954 	/*
1955 	 * If command line arguments and environment
1956 	 * are placed somewhere else on stack, we can
1957 	 * set them up here, ARG_START/END to setup
1958 	 * command line argumets and ENV_START/END
1959 	 * for environment.
1960 	 */
1961 	case PR_SET_MM_START_STACK:
1962 	case PR_SET_MM_ARG_START:
1963 	case PR_SET_MM_ARG_END:
1964 	case PR_SET_MM_ENV_START:
1965 	case PR_SET_MM_ENV_END:
1966 		if (!vma) {
1967 			error = -EFAULT;
1968 			goto out;
1969 		}
1970 		if (opt == PR_SET_MM_START_STACK)
1971 			mm->start_stack = addr;
1972 		else if (opt == PR_SET_MM_ARG_START)
1973 			mm->arg_start = addr;
1974 		else if (opt == PR_SET_MM_ARG_END)
1975 			mm->arg_end = addr;
1976 		else if (opt == PR_SET_MM_ENV_START)
1977 			mm->env_start = addr;
1978 		else if (opt == PR_SET_MM_ENV_END)
1979 			mm->env_end = addr;
1980 		break;
1981 
1982 	/*
1983 	 * This doesn't move auxiliary vector itself
1984 	 * since it's pinned to mm_struct, but allow
1985 	 * to fill vector with new values. It's up
1986 	 * to a caller to provide sane values here
1987 	 * otherwise user space tools which use this
1988 	 * vector might be unhappy.
1989 	 */
1990 	case PR_SET_MM_AUXV: {
1991 		unsigned long user_auxv[AT_VECTOR_SIZE];
1992 
1993 		if (arg4 > sizeof(user_auxv))
1994 			goto out;
1995 		up_read(&mm->mmap_sem);
1996 
1997 		if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
1998 			return -EFAULT;
1999 
2000 		/* Make sure the last entry is always AT_NULL */
2001 		user_auxv[AT_VECTOR_SIZE - 2] = 0;
2002 		user_auxv[AT_VECTOR_SIZE - 1] = 0;
2003 
2004 		BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2005 
2006 		task_lock(current);
2007 		memcpy(mm->saved_auxv, user_auxv, arg4);
2008 		task_unlock(current);
2009 
2010 		return 0;
2011 	}
2012 	default:
2013 		goto out;
2014 	}
2015 
2016 	error = 0;
2017 out:
2018 	up_read(&mm->mmap_sem);
2019 	return error;
2020 }
2021 
2022 #ifdef CONFIG_CHECKPOINT_RESTORE
2023 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2024 {
2025 	return put_user(me->clear_child_tid, tid_addr);
2026 }
2027 #else
2028 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2029 {
2030 	return -EINVAL;
2031 }
2032 #endif
2033 
2034 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2035 		unsigned long, arg4, unsigned long, arg5)
2036 {
2037 	struct task_struct *me = current;
2038 	unsigned char comm[sizeof(me->comm)];
2039 	long error;
2040 
2041 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2042 	if (error != -ENOSYS)
2043 		return error;
2044 
2045 	error = 0;
2046 	switch (option) {
2047 	case PR_SET_PDEATHSIG:
2048 		if (!valid_signal(arg2)) {
2049 			error = -EINVAL;
2050 			break;
2051 		}
2052 		me->pdeath_signal = arg2;
2053 		break;
2054 	case PR_GET_PDEATHSIG:
2055 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2056 		break;
2057 	case PR_GET_DUMPABLE:
2058 		error = get_dumpable(me->mm);
2059 		break;
2060 	case PR_SET_DUMPABLE:
2061 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2062 			error = -EINVAL;
2063 			break;
2064 		}
2065 		set_dumpable(me->mm, arg2);
2066 		break;
2067 
2068 	case PR_SET_UNALIGN:
2069 		error = SET_UNALIGN_CTL(me, arg2);
2070 		break;
2071 	case PR_GET_UNALIGN:
2072 		error = GET_UNALIGN_CTL(me, arg2);
2073 		break;
2074 	case PR_SET_FPEMU:
2075 		error = SET_FPEMU_CTL(me, arg2);
2076 		break;
2077 	case PR_GET_FPEMU:
2078 		error = GET_FPEMU_CTL(me, arg2);
2079 		break;
2080 	case PR_SET_FPEXC:
2081 		error = SET_FPEXC_CTL(me, arg2);
2082 		break;
2083 	case PR_GET_FPEXC:
2084 		error = GET_FPEXC_CTL(me, arg2);
2085 		break;
2086 	case PR_GET_TIMING:
2087 		error = PR_TIMING_STATISTICAL;
2088 		break;
2089 	case PR_SET_TIMING:
2090 		if (arg2 != PR_TIMING_STATISTICAL)
2091 			error = -EINVAL;
2092 		break;
2093 	case PR_SET_NAME:
2094 		comm[sizeof(me->comm) - 1] = 0;
2095 		if (strncpy_from_user(comm, (char __user *)arg2,
2096 				      sizeof(me->comm) - 1) < 0)
2097 			return -EFAULT;
2098 		set_task_comm(me, comm);
2099 		proc_comm_connector(me);
2100 		break;
2101 	case PR_GET_NAME:
2102 		get_task_comm(comm, me);
2103 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2104 			return -EFAULT;
2105 		break;
2106 	case PR_GET_ENDIAN:
2107 		error = GET_ENDIAN(me, arg2);
2108 		break;
2109 	case PR_SET_ENDIAN:
2110 		error = SET_ENDIAN(me, arg2);
2111 		break;
2112 	case PR_GET_SECCOMP:
2113 		error = prctl_get_seccomp();
2114 		break;
2115 	case PR_SET_SECCOMP:
2116 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2117 		break;
2118 	case PR_GET_TSC:
2119 		error = GET_TSC_CTL(arg2);
2120 		break;
2121 	case PR_SET_TSC:
2122 		error = SET_TSC_CTL(arg2);
2123 		break;
2124 	case PR_TASK_PERF_EVENTS_DISABLE:
2125 		error = perf_event_task_disable();
2126 		break;
2127 	case PR_TASK_PERF_EVENTS_ENABLE:
2128 		error = perf_event_task_enable();
2129 		break;
2130 	case PR_GET_TIMERSLACK:
2131 		error = current->timer_slack_ns;
2132 		break;
2133 	case PR_SET_TIMERSLACK:
2134 		if (arg2 <= 0)
2135 			current->timer_slack_ns =
2136 					current->default_timer_slack_ns;
2137 		else
2138 			current->timer_slack_ns = arg2;
2139 		break;
2140 	case PR_MCE_KILL:
2141 		if (arg4 | arg5)
2142 			return -EINVAL;
2143 		switch (arg2) {
2144 		case PR_MCE_KILL_CLEAR:
2145 			if (arg3 != 0)
2146 				return -EINVAL;
2147 			current->flags &= ~PF_MCE_PROCESS;
2148 			break;
2149 		case PR_MCE_KILL_SET:
2150 			current->flags |= PF_MCE_PROCESS;
2151 			if (arg3 == PR_MCE_KILL_EARLY)
2152 				current->flags |= PF_MCE_EARLY;
2153 			else if (arg3 == PR_MCE_KILL_LATE)
2154 				current->flags &= ~PF_MCE_EARLY;
2155 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2156 				current->flags &=
2157 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2158 			else
2159 				return -EINVAL;
2160 			break;
2161 		default:
2162 			return -EINVAL;
2163 		}
2164 		break;
2165 	case PR_MCE_KILL_GET:
2166 		if (arg2 | arg3 | arg4 | arg5)
2167 			return -EINVAL;
2168 		if (current->flags & PF_MCE_PROCESS)
2169 			error = (current->flags & PF_MCE_EARLY) ?
2170 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2171 		else
2172 			error = PR_MCE_KILL_DEFAULT;
2173 		break;
2174 	case PR_SET_MM:
2175 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2176 		break;
2177 	case PR_GET_TID_ADDRESS:
2178 		error = prctl_get_tid_address(me, (int __user **)arg2);
2179 		break;
2180 	case PR_SET_CHILD_SUBREAPER:
2181 		me->signal->is_child_subreaper = !!arg2;
2182 		break;
2183 	case PR_GET_CHILD_SUBREAPER:
2184 		error = put_user(me->signal->is_child_subreaper,
2185 				 (int __user *)arg2);
2186 		break;
2187 	case PR_SET_NO_NEW_PRIVS:
2188 		if (arg2 != 1 || arg3 || arg4 || arg5)
2189 			return -EINVAL;
2190 
2191 		task_set_no_new_privs(current);
2192 		break;
2193 	case PR_GET_NO_NEW_PRIVS:
2194 		if (arg2 || arg3 || arg4 || arg5)
2195 			return -EINVAL;
2196 		return task_no_new_privs(current) ? 1 : 0;
2197 	case PR_GET_THP_DISABLE:
2198 		if (arg2 || arg3 || arg4 || arg5)
2199 			return -EINVAL;
2200 		error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2201 		break;
2202 	case PR_SET_THP_DISABLE:
2203 		if (arg3 || arg4 || arg5)
2204 			return -EINVAL;
2205 		down_write(&me->mm->mmap_sem);
2206 		if (arg2)
2207 			me->mm->def_flags |= VM_NOHUGEPAGE;
2208 		else
2209 			me->mm->def_flags &= ~VM_NOHUGEPAGE;
2210 		up_write(&me->mm->mmap_sem);
2211 		break;
2212 	case PR_MPX_ENABLE_MANAGEMENT:
2213 		error = MPX_ENABLE_MANAGEMENT(me);
2214 		break;
2215 	case PR_MPX_DISABLE_MANAGEMENT:
2216 		error = MPX_DISABLE_MANAGEMENT(me);
2217 		break;
2218 	default:
2219 		error = -EINVAL;
2220 		break;
2221 	}
2222 	return error;
2223 }
2224 
2225 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2226 		struct getcpu_cache __user *, unused)
2227 {
2228 	int err = 0;
2229 	int cpu = raw_smp_processor_id();
2230 
2231 	if (cpup)
2232 		err |= put_user(cpu, cpup);
2233 	if (nodep)
2234 		err |= put_user(cpu_to_node(cpu), nodep);
2235 	return err ? -EFAULT : 0;
2236 }
2237 
2238 /**
2239  * do_sysinfo - fill in sysinfo struct
2240  * @info: pointer to buffer to fill
2241  */
2242 static int do_sysinfo(struct sysinfo *info)
2243 {
2244 	unsigned long mem_total, sav_total;
2245 	unsigned int mem_unit, bitcount;
2246 	struct timespec tp;
2247 
2248 	memset(info, 0, sizeof(struct sysinfo));
2249 
2250 	get_monotonic_boottime(&tp);
2251 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2252 
2253 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2254 
2255 	info->procs = nr_threads;
2256 
2257 	si_meminfo(info);
2258 	si_swapinfo(info);
2259 
2260 	/*
2261 	 * If the sum of all the available memory (i.e. ram + swap)
2262 	 * is less than can be stored in a 32 bit unsigned long then
2263 	 * we can be binary compatible with 2.2.x kernels.  If not,
2264 	 * well, in that case 2.2.x was broken anyways...
2265 	 *
2266 	 *  -Erik Andersen <andersee@debian.org>
2267 	 */
2268 
2269 	mem_total = info->totalram + info->totalswap;
2270 	if (mem_total < info->totalram || mem_total < info->totalswap)
2271 		goto out;
2272 	bitcount = 0;
2273 	mem_unit = info->mem_unit;
2274 	while (mem_unit > 1) {
2275 		bitcount++;
2276 		mem_unit >>= 1;
2277 		sav_total = mem_total;
2278 		mem_total <<= 1;
2279 		if (mem_total < sav_total)
2280 			goto out;
2281 	}
2282 
2283 	/*
2284 	 * If mem_total did not overflow, multiply all memory values by
2285 	 * info->mem_unit and set it to 1.  This leaves things compatible
2286 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2287 	 * kernels...
2288 	 */
2289 
2290 	info->mem_unit = 1;
2291 	info->totalram <<= bitcount;
2292 	info->freeram <<= bitcount;
2293 	info->sharedram <<= bitcount;
2294 	info->bufferram <<= bitcount;
2295 	info->totalswap <<= bitcount;
2296 	info->freeswap <<= bitcount;
2297 	info->totalhigh <<= bitcount;
2298 	info->freehigh <<= bitcount;
2299 
2300 out:
2301 	return 0;
2302 }
2303 
2304 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2305 {
2306 	struct sysinfo val;
2307 
2308 	do_sysinfo(&val);
2309 
2310 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2311 		return -EFAULT;
2312 
2313 	return 0;
2314 }
2315 
2316 #ifdef CONFIG_COMPAT
2317 struct compat_sysinfo {
2318 	s32 uptime;
2319 	u32 loads[3];
2320 	u32 totalram;
2321 	u32 freeram;
2322 	u32 sharedram;
2323 	u32 bufferram;
2324 	u32 totalswap;
2325 	u32 freeswap;
2326 	u16 procs;
2327 	u16 pad;
2328 	u32 totalhigh;
2329 	u32 freehigh;
2330 	u32 mem_unit;
2331 	char _f[20-2*sizeof(u32)-sizeof(int)];
2332 };
2333 
2334 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2335 {
2336 	struct sysinfo s;
2337 
2338 	do_sysinfo(&s);
2339 
2340 	/* Check to see if any memory value is too large for 32-bit and scale
2341 	 *  down if needed
2342 	 */
2343 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2344 		int bitcount = 0;
2345 
2346 		while (s.mem_unit < PAGE_SIZE) {
2347 			s.mem_unit <<= 1;
2348 			bitcount++;
2349 		}
2350 
2351 		s.totalram >>= bitcount;
2352 		s.freeram >>= bitcount;
2353 		s.sharedram >>= bitcount;
2354 		s.bufferram >>= bitcount;
2355 		s.totalswap >>= bitcount;
2356 		s.freeswap >>= bitcount;
2357 		s.totalhigh >>= bitcount;
2358 		s.freehigh >>= bitcount;
2359 	}
2360 
2361 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2362 	    __put_user(s.uptime, &info->uptime) ||
2363 	    __put_user(s.loads[0], &info->loads[0]) ||
2364 	    __put_user(s.loads[1], &info->loads[1]) ||
2365 	    __put_user(s.loads[2], &info->loads[2]) ||
2366 	    __put_user(s.totalram, &info->totalram) ||
2367 	    __put_user(s.freeram, &info->freeram) ||
2368 	    __put_user(s.sharedram, &info->sharedram) ||
2369 	    __put_user(s.bufferram, &info->bufferram) ||
2370 	    __put_user(s.totalswap, &info->totalswap) ||
2371 	    __put_user(s.freeswap, &info->freeswap) ||
2372 	    __put_user(s.procs, &info->procs) ||
2373 	    __put_user(s.totalhigh, &info->totalhigh) ||
2374 	    __put_user(s.freehigh, &info->freehigh) ||
2375 	    __put_user(s.mem_unit, &info->mem_unit))
2376 		return -EFAULT;
2377 
2378 	return 0;
2379 }
2380 #endif /* CONFIG_COMPAT */
2381