xref: /linux/kernel/sys.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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 #ifndef GET_FP_MODE
101 # define GET_FP_MODE(a)		(-EINVAL)
102 #endif
103 #ifndef SET_FP_MODE
104 # define SET_FP_MODE(a,b)	(-EINVAL)
105 #endif
106 
107 /*
108  * this is where the system-wide overflow UID and GID are defined, for
109  * architectures that now have 32-bit UID/GID but didn't in the past
110  */
111 
112 int overflowuid = DEFAULT_OVERFLOWUID;
113 int overflowgid = DEFAULT_OVERFLOWGID;
114 
115 EXPORT_SYMBOL(overflowuid);
116 EXPORT_SYMBOL(overflowgid);
117 
118 /*
119  * the same as above, but for filesystems which can only store a 16-bit
120  * UID and GID. as such, this is needed on all architectures
121  */
122 
123 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
124 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
125 
126 EXPORT_SYMBOL(fs_overflowuid);
127 EXPORT_SYMBOL(fs_overflowgid);
128 
129 /*
130  * Returns true if current's euid is same as p's uid or euid,
131  * or has CAP_SYS_NICE to p's user_ns.
132  *
133  * Called with rcu_read_lock, creds are safe
134  */
135 static bool set_one_prio_perm(struct task_struct *p)
136 {
137 	const struct cred *cred = current_cred(), *pcred = __task_cred(p);
138 
139 	if (uid_eq(pcred->uid,  cred->euid) ||
140 	    uid_eq(pcred->euid, cred->euid))
141 		return true;
142 	if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
143 		return true;
144 	return false;
145 }
146 
147 /*
148  * set the priority of a task
149  * - the caller must hold the RCU read lock
150  */
151 static int set_one_prio(struct task_struct *p, int niceval, int error)
152 {
153 	int no_nice;
154 
155 	if (!set_one_prio_perm(p)) {
156 		error = -EPERM;
157 		goto out;
158 	}
159 	if (niceval < task_nice(p) && !can_nice(p, niceval)) {
160 		error = -EACCES;
161 		goto out;
162 	}
163 	no_nice = security_task_setnice(p, niceval);
164 	if (no_nice) {
165 		error = no_nice;
166 		goto out;
167 	}
168 	if (error == -ESRCH)
169 		error = 0;
170 	set_user_nice(p, niceval);
171 out:
172 	return error;
173 }
174 
175 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
176 {
177 	struct task_struct *g, *p;
178 	struct user_struct *user;
179 	const struct cred *cred = current_cred();
180 	int error = -EINVAL;
181 	struct pid *pgrp;
182 	kuid_t uid;
183 
184 	if (which > PRIO_USER || which < PRIO_PROCESS)
185 		goto out;
186 
187 	/* normalize: avoid signed division (rounding problems) */
188 	error = -ESRCH;
189 	if (niceval < MIN_NICE)
190 		niceval = MIN_NICE;
191 	if (niceval > MAX_NICE)
192 		niceval = MAX_NICE;
193 
194 	rcu_read_lock();
195 	read_lock(&tasklist_lock);
196 	switch (which) {
197 	case PRIO_PROCESS:
198 		if (who)
199 			p = find_task_by_vpid(who);
200 		else
201 			p = current;
202 		if (p)
203 			error = set_one_prio(p, niceval, error);
204 		break;
205 	case PRIO_PGRP:
206 		if (who)
207 			pgrp = find_vpid(who);
208 		else
209 			pgrp = task_pgrp(current);
210 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
211 			error = set_one_prio(p, niceval, error);
212 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
213 		break;
214 	case PRIO_USER:
215 		uid = make_kuid(cred->user_ns, who);
216 		user = cred->user;
217 		if (!who)
218 			uid = cred->uid;
219 		else if (!uid_eq(uid, cred->uid)) {
220 			user = find_user(uid);
221 			if (!user)
222 				goto out_unlock;	/* No processes for this user */
223 		}
224 		do_each_thread(g, p) {
225 			if (uid_eq(task_uid(p), uid))
226 				error = set_one_prio(p, niceval, error);
227 		} while_each_thread(g, p);
228 		if (!uid_eq(uid, cred->uid))
229 			free_uid(user);		/* For find_user() */
230 		break;
231 	}
232 out_unlock:
233 	read_unlock(&tasklist_lock);
234 	rcu_read_unlock();
235 out:
236 	return error;
237 }
238 
239 /*
240  * Ugh. To avoid negative return values, "getpriority()" will
241  * not return the normal nice-value, but a negated value that
242  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
243  * to stay compatible.
244  */
245 SYSCALL_DEFINE2(getpriority, int, which, int, who)
246 {
247 	struct task_struct *g, *p;
248 	struct user_struct *user;
249 	const struct cred *cred = current_cred();
250 	long niceval, retval = -ESRCH;
251 	struct pid *pgrp;
252 	kuid_t uid;
253 
254 	if (which > PRIO_USER || which < PRIO_PROCESS)
255 		return -EINVAL;
256 
257 	rcu_read_lock();
258 	read_lock(&tasklist_lock);
259 	switch (which) {
260 	case PRIO_PROCESS:
261 		if (who)
262 			p = find_task_by_vpid(who);
263 		else
264 			p = current;
265 		if (p) {
266 			niceval = nice_to_rlimit(task_nice(p));
267 			if (niceval > retval)
268 				retval = niceval;
269 		}
270 		break;
271 	case PRIO_PGRP:
272 		if (who)
273 			pgrp = find_vpid(who);
274 		else
275 			pgrp = task_pgrp(current);
276 		do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
277 			niceval = nice_to_rlimit(task_nice(p));
278 			if (niceval > retval)
279 				retval = niceval;
280 		} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
281 		break;
282 	case PRIO_USER:
283 		uid = make_kuid(cred->user_ns, who);
284 		user = cred->user;
285 		if (!who)
286 			uid = cred->uid;
287 		else if (!uid_eq(uid, cred->uid)) {
288 			user = find_user(uid);
289 			if (!user)
290 				goto out_unlock;	/* No processes for this user */
291 		}
292 		do_each_thread(g, p) {
293 			if (uid_eq(task_uid(p), uid)) {
294 				niceval = nice_to_rlimit(task_nice(p));
295 				if (niceval > retval)
296 					retval = niceval;
297 			}
298 		} while_each_thread(g, p);
299 		if (!uid_eq(uid, cred->uid))
300 			free_uid(user);		/* for find_user() */
301 		break;
302 	}
303 out_unlock:
304 	read_unlock(&tasklist_lock);
305 	rcu_read_unlock();
306 
307 	return retval;
308 }
309 
310 /*
311  * Unprivileged users may change the real gid to the effective gid
312  * or vice versa.  (BSD-style)
313  *
314  * If you set the real gid at all, or set the effective gid to a value not
315  * equal to the real gid, then the saved gid is set to the new effective gid.
316  *
317  * This makes it possible for a setgid program to completely drop its
318  * privileges, which is often a useful assertion to make when you are doing
319  * a security audit over a program.
320  *
321  * The general idea is that a program which uses just setregid() will be
322  * 100% compatible with BSD.  A program which uses just setgid() will be
323  * 100% compatible with POSIX with saved IDs.
324  *
325  * SMP: There are not races, the GIDs are checked only by filesystem
326  *      operations (as far as semantic preservation is concerned).
327  */
328 #ifdef CONFIG_MULTIUSER
329 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
330 {
331 	struct user_namespace *ns = current_user_ns();
332 	const struct cred *old;
333 	struct cred *new;
334 	int retval;
335 	kgid_t krgid, kegid;
336 
337 	krgid = make_kgid(ns, rgid);
338 	kegid = make_kgid(ns, egid);
339 
340 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
341 		return -EINVAL;
342 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
343 		return -EINVAL;
344 
345 	new = prepare_creds();
346 	if (!new)
347 		return -ENOMEM;
348 	old = current_cred();
349 
350 	retval = -EPERM;
351 	if (rgid != (gid_t) -1) {
352 		if (gid_eq(old->gid, krgid) ||
353 		    gid_eq(old->egid, krgid) ||
354 		    ns_capable(old->user_ns, CAP_SETGID))
355 			new->gid = krgid;
356 		else
357 			goto error;
358 	}
359 	if (egid != (gid_t) -1) {
360 		if (gid_eq(old->gid, kegid) ||
361 		    gid_eq(old->egid, kegid) ||
362 		    gid_eq(old->sgid, kegid) ||
363 		    ns_capable(old->user_ns, CAP_SETGID))
364 			new->egid = kegid;
365 		else
366 			goto error;
367 	}
368 
369 	if (rgid != (gid_t) -1 ||
370 	    (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
371 		new->sgid = new->egid;
372 	new->fsgid = new->egid;
373 
374 	return commit_creds(new);
375 
376 error:
377 	abort_creds(new);
378 	return retval;
379 }
380 
381 /*
382  * setgid() is implemented like SysV w/ SAVED_IDS
383  *
384  * SMP: Same implicit races as above.
385  */
386 SYSCALL_DEFINE1(setgid, gid_t, gid)
387 {
388 	struct user_namespace *ns = current_user_ns();
389 	const struct cred *old;
390 	struct cred *new;
391 	int retval;
392 	kgid_t kgid;
393 
394 	kgid = make_kgid(ns, gid);
395 	if (!gid_valid(kgid))
396 		return -EINVAL;
397 
398 	new = prepare_creds();
399 	if (!new)
400 		return -ENOMEM;
401 	old = current_cred();
402 
403 	retval = -EPERM;
404 	if (ns_capable(old->user_ns, CAP_SETGID))
405 		new->gid = new->egid = new->sgid = new->fsgid = kgid;
406 	else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
407 		new->egid = new->fsgid = kgid;
408 	else
409 		goto error;
410 
411 	return commit_creds(new);
412 
413 error:
414 	abort_creds(new);
415 	return retval;
416 }
417 
418 /*
419  * change the user struct in a credentials set to match the new UID
420  */
421 static int set_user(struct cred *new)
422 {
423 	struct user_struct *new_user;
424 
425 	new_user = alloc_uid(new->uid);
426 	if (!new_user)
427 		return -EAGAIN;
428 
429 	/*
430 	 * We don't fail in case of NPROC limit excess here because too many
431 	 * poorly written programs don't check set*uid() return code, assuming
432 	 * it never fails if called by root.  We may still enforce NPROC limit
433 	 * for programs doing set*uid()+execve() by harmlessly deferring the
434 	 * failure to the execve() stage.
435 	 */
436 	if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
437 			new_user != INIT_USER)
438 		current->flags |= PF_NPROC_EXCEEDED;
439 	else
440 		current->flags &= ~PF_NPROC_EXCEEDED;
441 
442 	free_uid(new->user);
443 	new->user = new_user;
444 	return 0;
445 }
446 
447 /*
448  * Unprivileged users may change the real uid to the effective uid
449  * or vice versa.  (BSD-style)
450  *
451  * If you set the real uid at all, or set the effective uid to a value not
452  * equal to the real uid, then the saved uid is set to the new effective uid.
453  *
454  * This makes it possible for a setuid program to completely drop its
455  * privileges, which is often a useful assertion to make when you are doing
456  * a security audit over a program.
457  *
458  * The general idea is that a program which uses just setreuid() will be
459  * 100% compatible with BSD.  A program which uses just setuid() will be
460  * 100% compatible with POSIX with saved IDs.
461  */
462 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
463 {
464 	struct user_namespace *ns = current_user_ns();
465 	const struct cred *old;
466 	struct cred *new;
467 	int retval;
468 	kuid_t kruid, keuid;
469 
470 	kruid = make_kuid(ns, ruid);
471 	keuid = make_kuid(ns, euid);
472 
473 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
474 		return -EINVAL;
475 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
476 		return -EINVAL;
477 
478 	new = prepare_creds();
479 	if (!new)
480 		return -ENOMEM;
481 	old = current_cred();
482 
483 	retval = -EPERM;
484 	if (ruid != (uid_t) -1) {
485 		new->uid = kruid;
486 		if (!uid_eq(old->uid, kruid) &&
487 		    !uid_eq(old->euid, kruid) &&
488 		    !ns_capable(old->user_ns, CAP_SETUID))
489 			goto error;
490 	}
491 
492 	if (euid != (uid_t) -1) {
493 		new->euid = keuid;
494 		if (!uid_eq(old->uid, keuid) &&
495 		    !uid_eq(old->euid, keuid) &&
496 		    !uid_eq(old->suid, keuid) &&
497 		    !ns_capable(old->user_ns, CAP_SETUID))
498 			goto error;
499 	}
500 
501 	if (!uid_eq(new->uid, old->uid)) {
502 		retval = set_user(new);
503 		if (retval < 0)
504 			goto error;
505 	}
506 	if (ruid != (uid_t) -1 ||
507 	    (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
508 		new->suid = new->euid;
509 	new->fsuid = new->euid;
510 
511 	retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
512 	if (retval < 0)
513 		goto error;
514 
515 	return commit_creds(new);
516 
517 error:
518 	abort_creds(new);
519 	return retval;
520 }
521 
522 /*
523  * setuid() is implemented like SysV with SAVED_IDS
524  *
525  * Note that SAVED_ID's is deficient in that a setuid root program
526  * like sendmail, for example, cannot set its uid to be a normal
527  * user and then switch back, because if you're root, setuid() sets
528  * the saved uid too.  If you don't like this, blame the bright people
529  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
530  * will allow a root program to temporarily drop privileges and be able to
531  * regain them by swapping the real and effective uid.
532  */
533 SYSCALL_DEFINE1(setuid, uid_t, uid)
534 {
535 	struct user_namespace *ns = current_user_ns();
536 	const struct cred *old;
537 	struct cred *new;
538 	int retval;
539 	kuid_t kuid;
540 
541 	kuid = make_kuid(ns, uid);
542 	if (!uid_valid(kuid))
543 		return -EINVAL;
544 
545 	new = prepare_creds();
546 	if (!new)
547 		return -ENOMEM;
548 	old = current_cred();
549 
550 	retval = -EPERM;
551 	if (ns_capable(old->user_ns, CAP_SETUID)) {
552 		new->suid = new->uid = kuid;
553 		if (!uid_eq(kuid, old->uid)) {
554 			retval = set_user(new);
555 			if (retval < 0)
556 				goto error;
557 		}
558 	} else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
559 		goto error;
560 	}
561 
562 	new->fsuid = new->euid = kuid;
563 
564 	retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
565 	if (retval < 0)
566 		goto error;
567 
568 	return commit_creds(new);
569 
570 error:
571 	abort_creds(new);
572 	return retval;
573 }
574 
575 
576 /*
577  * This function implements a generic ability to update ruid, euid,
578  * and suid.  This allows you to implement the 4.4 compatible seteuid().
579  */
580 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
581 {
582 	struct user_namespace *ns = current_user_ns();
583 	const struct cred *old;
584 	struct cred *new;
585 	int retval;
586 	kuid_t kruid, keuid, ksuid;
587 
588 	kruid = make_kuid(ns, ruid);
589 	keuid = make_kuid(ns, euid);
590 	ksuid = make_kuid(ns, suid);
591 
592 	if ((ruid != (uid_t) -1) && !uid_valid(kruid))
593 		return -EINVAL;
594 
595 	if ((euid != (uid_t) -1) && !uid_valid(keuid))
596 		return -EINVAL;
597 
598 	if ((suid != (uid_t) -1) && !uid_valid(ksuid))
599 		return -EINVAL;
600 
601 	new = prepare_creds();
602 	if (!new)
603 		return -ENOMEM;
604 
605 	old = current_cred();
606 
607 	retval = -EPERM;
608 	if (!ns_capable(old->user_ns, CAP_SETUID)) {
609 		if (ruid != (uid_t) -1        && !uid_eq(kruid, old->uid) &&
610 		    !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
611 			goto error;
612 		if (euid != (uid_t) -1        && !uid_eq(keuid, old->uid) &&
613 		    !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
614 			goto error;
615 		if (suid != (uid_t) -1        && !uid_eq(ksuid, old->uid) &&
616 		    !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
617 			goto error;
618 	}
619 
620 	if (ruid != (uid_t) -1) {
621 		new->uid = kruid;
622 		if (!uid_eq(kruid, old->uid)) {
623 			retval = set_user(new);
624 			if (retval < 0)
625 				goto error;
626 		}
627 	}
628 	if (euid != (uid_t) -1)
629 		new->euid = keuid;
630 	if (suid != (uid_t) -1)
631 		new->suid = ksuid;
632 	new->fsuid = new->euid;
633 
634 	retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
635 	if (retval < 0)
636 		goto error;
637 
638 	return commit_creds(new);
639 
640 error:
641 	abort_creds(new);
642 	return retval;
643 }
644 
645 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
646 {
647 	const struct cred *cred = current_cred();
648 	int retval;
649 	uid_t ruid, euid, suid;
650 
651 	ruid = from_kuid_munged(cred->user_ns, cred->uid);
652 	euid = from_kuid_munged(cred->user_ns, cred->euid);
653 	suid = from_kuid_munged(cred->user_ns, cred->suid);
654 
655 	retval = put_user(ruid, ruidp);
656 	if (!retval) {
657 		retval = put_user(euid, euidp);
658 		if (!retval)
659 			return put_user(suid, suidp);
660 	}
661 	return retval;
662 }
663 
664 /*
665  * Same as above, but for rgid, egid, sgid.
666  */
667 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
668 {
669 	struct user_namespace *ns = current_user_ns();
670 	const struct cred *old;
671 	struct cred *new;
672 	int retval;
673 	kgid_t krgid, kegid, ksgid;
674 
675 	krgid = make_kgid(ns, rgid);
676 	kegid = make_kgid(ns, egid);
677 	ksgid = make_kgid(ns, sgid);
678 
679 	if ((rgid != (gid_t) -1) && !gid_valid(krgid))
680 		return -EINVAL;
681 	if ((egid != (gid_t) -1) && !gid_valid(kegid))
682 		return -EINVAL;
683 	if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
684 		return -EINVAL;
685 
686 	new = prepare_creds();
687 	if (!new)
688 		return -ENOMEM;
689 	old = current_cred();
690 
691 	retval = -EPERM;
692 	if (!ns_capable(old->user_ns, CAP_SETGID)) {
693 		if (rgid != (gid_t) -1        && !gid_eq(krgid, old->gid) &&
694 		    !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
695 			goto error;
696 		if (egid != (gid_t) -1        && !gid_eq(kegid, old->gid) &&
697 		    !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
698 			goto error;
699 		if (sgid != (gid_t) -1        && !gid_eq(ksgid, old->gid) &&
700 		    !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
701 			goto error;
702 	}
703 
704 	if (rgid != (gid_t) -1)
705 		new->gid = krgid;
706 	if (egid != (gid_t) -1)
707 		new->egid = kegid;
708 	if (sgid != (gid_t) -1)
709 		new->sgid = ksgid;
710 	new->fsgid = new->egid;
711 
712 	return commit_creds(new);
713 
714 error:
715 	abort_creds(new);
716 	return retval;
717 }
718 
719 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
720 {
721 	const struct cred *cred = current_cred();
722 	int retval;
723 	gid_t rgid, egid, sgid;
724 
725 	rgid = from_kgid_munged(cred->user_ns, cred->gid);
726 	egid = from_kgid_munged(cred->user_ns, cred->egid);
727 	sgid = from_kgid_munged(cred->user_ns, cred->sgid);
728 
729 	retval = put_user(rgid, rgidp);
730 	if (!retval) {
731 		retval = put_user(egid, egidp);
732 		if (!retval)
733 			retval = put_user(sgid, sgidp);
734 	}
735 
736 	return retval;
737 }
738 
739 
740 /*
741  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
742  * is used for "access()" and for the NFS daemon (letting nfsd stay at
743  * whatever uid it wants to). It normally shadows "euid", except when
744  * explicitly set by setfsuid() or for access..
745  */
746 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
747 {
748 	const struct cred *old;
749 	struct cred *new;
750 	uid_t old_fsuid;
751 	kuid_t kuid;
752 
753 	old = current_cred();
754 	old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
755 
756 	kuid = make_kuid(old->user_ns, uid);
757 	if (!uid_valid(kuid))
758 		return old_fsuid;
759 
760 	new = prepare_creds();
761 	if (!new)
762 		return old_fsuid;
763 
764 	if (uid_eq(kuid, old->uid)  || uid_eq(kuid, old->euid)  ||
765 	    uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
766 	    ns_capable(old->user_ns, CAP_SETUID)) {
767 		if (!uid_eq(kuid, old->fsuid)) {
768 			new->fsuid = kuid;
769 			if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
770 				goto change_okay;
771 		}
772 	}
773 
774 	abort_creds(new);
775 	return old_fsuid;
776 
777 change_okay:
778 	commit_creds(new);
779 	return old_fsuid;
780 }
781 
782 /*
783  * Samma på svenska..
784  */
785 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
786 {
787 	const struct cred *old;
788 	struct cred *new;
789 	gid_t old_fsgid;
790 	kgid_t kgid;
791 
792 	old = current_cred();
793 	old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
794 
795 	kgid = make_kgid(old->user_ns, gid);
796 	if (!gid_valid(kgid))
797 		return old_fsgid;
798 
799 	new = prepare_creds();
800 	if (!new)
801 		return old_fsgid;
802 
803 	if (gid_eq(kgid, old->gid)  || gid_eq(kgid, old->egid)  ||
804 	    gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
805 	    ns_capable(old->user_ns, CAP_SETGID)) {
806 		if (!gid_eq(kgid, old->fsgid)) {
807 			new->fsgid = kgid;
808 			goto change_okay;
809 		}
810 	}
811 
812 	abort_creds(new);
813 	return old_fsgid;
814 
815 change_okay:
816 	commit_creds(new);
817 	return old_fsgid;
818 }
819 #endif /* CONFIG_MULTIUSER */
820 
821 /**
822  * sys_getpid - return the thread group id of the current process
823  *
824  * Note, despite the name, this returns the tgid not the pid.  The tgid and
825  * the pid are identical unless CLONE_THREAD was specified on clone() in
826  * which case the tgid is the same in all threads of the same group.
827  *
828  * This is SMP safe as current->tgid does not change.
829  */
830 SYSCALL_DEFINE0(getpid)
831 {
832 	return task_tgid_vnr(current);
833 }
834 
835 /* Thread ID - the internal kernel "pid" */
836 SYSCALL_DEFINE0(gettid)
837 {
838 	return task_pid_vnr(current);
839 }
840 
841 /*
842  * Accessing ->real_parent is not SMP-safe, it could
843  * change from under us. However, we can use a stale
844  * value of ->real_parent under rcu_read_lock(), see
845  * release_task()->call_rcu(delayed_put_task_struct).
846  */
847 SYSCALL_DEFINE0(getppid)
848 {
849 	int pid;
850 
851 	rcu_read_lock();
852 	pid = task_tgid_vnr(rcu_dereference(current->real_parent));
853 	rcu_read_unlock();
854 
855 	return pid;
856 }
857 
858 SYSCALL_DEFINE0(getuid)
859 {
860 	/* Only we change this so SMP safe */
861 	return from_kuid_munged(current_user_ns(), current_uid());
862 }
863 
864 SYSCALL_DEFINE0(geteuid)
865 {
866 	/* Only we change this so SMP safe */
867 	return from_kuid_munged(current_user_ns(), current_euid());
868 }
869 
870 SYSCALL_DEFINE0(getgid)
871 {
872 	/* Only we change this so SMP safe */
873 	return from_kgid_munged(current_user_ns(), current_gid());
874 }
875 
876 SYSCALL_DEFINE0(getegid)
877 {
878 	/* Only we change this so SMP safe */
879 	return from_kgid_munged(current_user_ns(), current_egid());
880 }
881 
882 void do_sys_times(struct tms *tms)
883 {
884 	cputime_t tgutime, tgstime, cutime, cstime;
885 
886 	thread_group_cputime_adjusted(current, &tgutime, &tgstime);
887 	cutime = current->signal->cutime;
888 	cstime = current->signal->cstime;
889 	tms->tms_utime = cputime_to_clock_t(tgutime);
890 	tms->tms_stime = cputime_to_clock_t(tgstime);
891 	tms->tms_cutime = cputime_to_clock_t(cutime);
892 	tms->tms_cstime = cputime_to_clock_t(cstime);
893 }
894 
895 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
896 {
897 	if (tbuf) {
898 		struct tms tmp;
899 
900 		do_sys_times(&tmp);
901 		if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
902 			return -EFAULT;
903 	}
904 	force_successful_syscall_return();
905 	return (long) jiffies_64_to_clock_t(get_jiffies_64());
906 }
907 
908 /*
909  * This needs some heavy checking ...
910  * I just haven't the stomach for it. I also don't fully
911  * understand sessions/pgrp etc. Let somebody who does explain it.
912  *
913  * OK, I think I have the protection semantics right.... this is really
914  * only important on a multi-user system anyway, to make sure one user
915  * can't send a signal to a process owned by another.  -TYT, 12/12/91
916  *
917  * !PF_FORKNOEXEC check to conform completely to POSIX.
918  */
919 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
920 {
921 	struct task_struct *p;
922 	struct task_struct *group_leader = current->group_leader;
923 	struct pid *pgrp;
924 	int err;
925 
926 	if (!pid)
927 		pid = task_pid_vnr(group_leader);
928 	if (!pgid)
929 		pgid = pid;
930 	if (pgid < 0)
931 		return -EINVAL;
932 	rcu_read_lock();
933 
934 	/* From this point forward we keep holding onto the tasklist lock
935 	 * so that our parent does not change from under us. -DaveM
936 	 */
937 	write_lock_irq(&tasklist_lock);
938 
939 	err = -ESRCH;
940 	p = find_task_by_vpid(pid);
941 	if (!p)
942 		goto out;
943 
944 	err = -EINVAL;
945 	if (!thread_group_leader(p))
946 		goto out;
947 
948 	if (same_thread_group(p->real_parent, group_leader)) {
949 		err = -EPERM;
950 		if (task_session(p) != task_session(group_leader))
951 			goto out;
952 		err = -EACCES;
953 		if (!(p->flags & PF_FORKNOEXEC))
954 			goto out;
955 	} else {
956 		err = -ESRCH;
957 		if (p != group_leader)
958 			goto out;
959 	}
960 
961 	err = -EPERM;
962 	if (p->signal->leader)
963 		goto out;
964 
965 	pgrp = task_pid(p);
966 	if (pgid != pid) {
967 		struct task_struct *g;
968 
969 		pgrp = find_vpid(pgid);
970 		g = pid_task(pgrp, PIDTYPE_PGID);
971 		if (!g || task_session(g) != task_session(group_leader))
972 			goto out;
973 	}
974 
975 	err = security_task_setpgid(p, pgid);
976 	if (err)
977 		goto out;
978 
979 	if (task_pgrp(p) != pgrp)
980 		change_pid(p, PIDTYPE_PGID, pgrp);
981 
982 	err = 0;
983 out:
984 	/* All paths lead to here, thus we are safe. -DaveM */
985 	write_unlock_irq(&tasklist_lock);
986 	rcu_read_unlock();
987 	return err;
988 }
989 
990 SYSCALL_DEFINE1(getpgid, pid_t, pid)
991 {
992 	struct task_struct *p;
993 	struct pid *grp;
994 	int retval;
995 
996 	rcu_read_lock();
997 	if (!pid)
998 		grp = task_pgrp(current);
999 	else {
1000 		retval = -ESRCH;
1001 		p = find_task_by_vpid(pid);
1002 		if (!p)
1003 			goto out;
1004 		grp = task_pgrp(p);
1005 		if (!grp)
1006 			goto out;
1007 
1008 		retval = security_task_getpgid(p);
1009 		if (retval)
1010 			goto out;
1011 	}
1012 	retval = pid_vnr(grp);
1013 out:
1014 	rcu_read_unlock();
1015 	return retval;
1016 }
1017 
1018 #ifdef __ARCH_WANT_SYS_GETPGRP
1019 
1020 SYSCALL_DEFINE0(getpgrp)
1021 {
1022 	return sys_getpgid(0);
1023 }
1024 
1025 #endif
1026 
1027 SYSCALL_DEFINE1(getsid, pid_t, pid)
1028 {
1029 	struct task_struct *p;
1030 	struct pid *sid;
1031 	int retval;
1032 
1033 	rcu_read_lock();
1034 	if (!pid)
1035 		sid = task_session(current);
1036 	else {
1037 		retval = -ESRCH;
1038 		p = find_task_by_vpid(pid);
1039 		if (!p)
1040 			goto out;
1041 		sid = task_session(p);
1042 		if (!sid)
1043 			goto out;
1044 
1045 		retval = security_task_getsid(p);
1046 		if (retval)
1047 			goto out;
1048 	}
1049 	retval = pid_vnr(sid);
1050 out:
1051 	rcu_read_unlock();
1052 	return retval;
1053 }
1054 
1055 static void set_special_pids(struct pid *pid)
1056 {
1057 	struct task_struct *curr = current->group_leader;
1058 
1059 	if (task_session(curr) != pid)
1060 		change_pid(curr, PIDTYPE_SID, pid);
1061 
1062 	if (task_pgrp(curr) != pid)
1063 		change_pid(curr, PIDTYPE_PGID, pid);
1064 }
1065 
1066 SYSCALL_DEFINE0(setsid)
1067 {
1068 	struct task_struct *group_leader = current->group_leader;
1069 	struct pid *sid = task_pid(group_leader);
1070 	pid_t session = pid_vnr(sid);
1071 	int err = -EPERM;
1072 
1073 	write_lock_irq(&tasklist_lock);
1074 	/* Fail if I am already a session leader */
1075 	if (group_leader->signal->leader)
1076 		goto out;
1077 
1078 	/* Fail if a process group id already exists that equals the
1079 	 * proposed session id.
1080 	 */
1081 	if (pid_task(sid, PIDTYPE_PGID))
1082 		goto out;
1083 
1084 	group_leader->signal->leader = 1;
1085 	set_special_pids(sid);
1086 
1087 	proc_clear_tty(group_leader);
1088 
1089 	err = session;
1090 out:
1091 	write_unlock_irq(&tasklist_lock);
1092 	if (err > 0) {
1093 		proc_sid_connector(group_leader);
1094 		sched_autogroup_create_attach(group_leader);
1095 	}
1096 	return err;
1097 }
1098 
1099 DECLARE_RWSEM(uts_sem);
1100 
1101 #ifdef COMPAT_UTS_MACHINE
1102 #define override_architecture(name) \
1103 	(personality(current->personality) == PER_LINUX32 && \
1104 	 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1105 		      sizeof(COMPAT_UTS_MACHINE)))
1106 #else
1107 #define override_architecture(name)	0
1108 #endif
1109 
1110 /*
1111  * Work around broken programs that cannot handle "Linux 3.0".
1112  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1113  * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1114  */
1115 static int override_release(char __user *release, size_t len)
1116 {
1117 	int ret = 0;
1118 
1119 	if (current->personality & UNAME26) {
1120 		const char *rest = UTS_RELEASE;
1121 		char buf[65] = { 0 };
1122 		int ndots = 0;
1123 		unsigned v;
1124 		size_t copy;
1125 
1126 		while (*rest) {
1127 			if (*rest == '.' && ++ndots >= 3)
1128 				break;
1129 			if (!isdigit(*rest) && *rest != '.')
1130 				break;
1131 			rest++;
1132 		}
1133 		v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1134 		copy = clamp_t(size_t, len, 1, sizeof(buf));
1135 		copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1136 		ret = copy_to_user(release, buf, copy + 1);
1137 	}
1138 	return ret;
1139 }
1140 
1141 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1142 {
1143 	int errno = 0;
1144 
1145 	down_read(&uts_sem);
1146 	if (copy_to_user(name, utsname(), sizeof *name))
1147 		errno = -EFAULT;
1148 	up_read(&uts_sem);
1149 
1150 	if (!errno && override_release(name->release, sizeof(name->release)))
1151 		errno = -EFAULT;
1152 	if (!errno && override_architecture(name))
1153 		errno = -EFAULT;
1154 	return errno;
1155 }
1156 
1157 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1158 /*
1159  * Old cruft
1160  */
1161 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1162 {
1163 	int error = 0;
1164 
1165 	if (!name)
1166 		return -EFAULT;
1167 
1168 	down_read(&uts_sem);
1169 	if (copy_to_user(name, utsname(), sizeof(*name)))
1170 		error = -EFAULT;
1171 	up_read(&uts_sem);
1172 
1173 	if (!error && override_release(name->release, sizeof(name->release)))
1174 		error = -EFAULT;
1175 	if (!error && override_architecture(name))
1176 		error = -EFAULT;
1177 	return error;
1178 }
1179 
1180 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1181 {
1182 	int error;
1183 
1184 	if (!name)
1185 		return -EFAULT;
1186 	if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1187 		return -EFAULT;
1188 
1189 	down_read(&uts_sem);
1190 	error = __copy_to_user(&name->sysname, &utsname()->sysname,
1191 			       __OLD_UTS_LEN);
1192 	error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1193 	error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1194 				__OLD_UTS_LEN);
1195 	error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1196 	error |= __copy_to_user(&name->release, &utsname()->release,
1197 				__OLD_UTS_LEN);
1198 	error |= __put_user(0, name->release + __OLD_UTS_LEN);
1199 	error |= __copy_to_user(&name->version, &utsname()->version,
1200 				__OLD_UTS_LEN);
1201 	error |= __put_user(0, name->version + __OLD_UTS_LEN);
1202 	error |= __copy_to_user(&name->machine, &utsname()->machine,
1203 				__OLD_UTS_LEN);
1204 	error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1205 	up_read(&uts_sem);
1206 
1207 	if (!error && override_architecture(name))
1208 		error = -EFAULT;
1209 	if (!error && override_release(name->release, sizeof(name->release)))
1210 		error = -EFAULT;
1211 	return error ? -EFAULT : 0;
1212 }
1213 #endif
1214 
1215 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1216 {
1217 	int errno;
1218 	char tmp[__NEW_UTS_LEN];
1219 
1220 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1221 		return -EPERM;
1222 
1223 	if (len < 0 || len > __NEW_UTS_LEN)
1224 		return -EINVAL;
1225 	down_write(&uts_sem);
1226 	errno = -EFAULT;
1227 	if (!copy_from_user(tmp, name, len)) {
1228 		struct new_utsname *u = utsname();
1229 
1230 		memcpy(u->nodename, tmp, len);
1231 		memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1232 		errno = 0;
1233 		uts_proc_notify(UTS_PROC_HOSTNAME);
1234 	}
1235 	up_write(&uts_sem);
1236 	return errno;
1237 }
1238 
1239 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1240 
1241 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1242 {
1243 	int i, errno;
1244 	struct new_utsname *u;
1245 
1246 	if (len < 0)
1247 		return -EINVAL;
1248 	down_read(&uts_sem);
1249 	u = utsname();
1250 	i = 1 + strlen(u->nodename);
1251 	if (i > len)
1252 		i = len;
1253 	errno = 0;
1254 	if (copy_to_user(name, u->nodename, i))
1255 		errno = -EFAULT;
1256 	up_read(&uts_sem);
1257 	return errno;
1258 }
1259 
1260 #endif
1261 
1262 /*
1263  * Only setdomainname; getdomainname can be implemented by calling
1264  * uname()
1265  */
1266 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1267 {
1268 	int errno;
1269 	char tmp[__NEW_UTS_LEN];
1270 
1271 	if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1272 		return -EPERM;
1273 	if (len < 0 || len > __NEW_UTS_LEN)
1274 		return -EINVAL;
1275 
1276 	down_write(&uts_sem);
1277 	errno = -EFAULT;
1278 	if (!copy_from_user(tmp, name, len)) {
1279 		struct new_utsname *u = utsname();
1280 
1281 		memcpy(u->domainname, tmp, len);
1282 		memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1283 		errno = 0;
1284 		uts_proc_notify(UTS_PROC_DOMAINNAME);
1285 	}
1286 	up_write(&uts_sem);
1287 	return errno;
1288 }
1289 
1290 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1291 {
1292 	struct rlimit value;
1293 	int ret;
1294 
1295 	ret = do_prlimit(current, resource, NULL, &value);
1296 	if (!ret)
1297 		ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1298 
1299 	return ret;
1300 }
1301 
1302 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1303 
1304 /*
1305  *	Back compatibility for getrlimit. Needed for some apps.
1306  */
1307 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1308 		struct rlimit __user *, rlim)
1309 {
1310 	struct rlimit x;
1311 	if (resource >= RLIM_NLIMITS)
1312 		return -EINVAL;
1313 
1314 	task_lock(current->group_leader);
1315 	x = current->signal->rlim[resource];
1316 	task_unlock(current->group_leader);
1317 	if (x.rlim_cur > 0x7FFFFFFF)
1318 		x.rlim_cur = 0x7FFFFFFF;
1319 	if (x.rlim_max > 0x7FFFFFFF)
1320 		x.rlim_max = 0x7FFFFFFF;
1321 	return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1322 }
1323 
1324 #endif
1325 
1326 static inline bool rlim64_is_infinity(__u64 rlim64)
1327 {
1328 #if BITS_PER_LONG < 64
1329 	return rlim64 >= ULONG_MAX;
1330 #else
1331 	return rlim64 == RLIM64_INFINITY;
1332 #endif
1333 }
1334 
1335 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1336 {
1337 	if (rlim->rlim_cur == RLIM_INFINITY)
1338 		rlim64->rlim_cur = RLIM64_INFINITY;
1339 	else
1340 		rlim64->rlim_cur = rlim->rlim_cur;
1341 	if (rlim->rlim_max == RLIM_INFINITY)
1342 		rlim64->rlim_max = RLIM64_INFINITY;
1343 	else
1344 		rlim64->rlim_max = rlim->rlim_max;
1345 }
1346 
1347 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1348 {
1349 	if (rlim64_is_infinity(rlim64->rlim_cur))
1350 		rlim->rlim_cur = RLIM_INFINITY;
1351 	else
1352 		rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1353 	if (rlim64_is_infinity(rlim64->rlim_max))
1354 		rlim->rlim_max = RLIM_INFINITY;
1355 	else
1356 		rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1357 }
1358 
1359 /* make sure you are allowed to change @tsk limits before calling this */
1360 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1361 		struct rlimit *new_rlim, struct rlimit *old_rlim)
1362 {
1363 	struct rlimit *rlim;
1364 	int retval = 0;
1365 
1366 	if (resource >= RLIM_NLIMITS)
1367 		return -EINVAL;
1368 	if (new_rlim) {
1369 		if (new_rlim->rlim_cur > new_rlim->rlim_max)
1370 			return -EINVAL;
1371 		if (resource == RLIMIT_NOFILE &&
1372 				new_rlim->rlim_max > sysctl_nr_open)
1373 			return -EPERM;
1374 	}
1375 
1376 	/* protect tsk->signal and tsk->sighand from disappearing */
1377 	read_lock(&tasklist_lock);
1378 	if (!tsk->sighand) {
1379 		retval = -ESRCH;
1380 		goto out;
1381 	}
1382 
1383 	rlim = tsk->signal->rlim + resource;
1384 	task_lock(tsk->group_leader);
1385 	if (new_rlim) {
1386 		/* Keep the capable check against init_user_ns until
1387 		   cgroups can contain all limits */
1388 		if (new_rlim->rlim_max > rlim->rlim_max &&
1389 				!capable(CAP_SYS_RESOURCE))
1390 			retval = -EPERM;
1391 		if (!retval)
1392 			retval = security_task_setrlimit(tsk->group_leader,
1393 					resource, new_rlim);
1394 		if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1395 			/*
1396 			 * The caller is asking for an immediate RLIMIT_CPU
1397 			 * expiry.  But we use the zero value to mean "it was
1398 			 * never set".  So let's cheat and make it one second
1399 			 * instead
1400 			 */
1401 			new_rlim->rlim_cur = 1;
1402 		}
1403 	}
1404 	if (!retval) {
1405 		if (old_rlim)
1406 			*old_rlim = *rlim;
1407 		if (new_rlim)
1408 			*rlim = *new_rlim;
1409 	}
1410 	task_unlock(tsk->group_leader);
1411 
1412 	/*
1413 	 * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1414 	 * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1415 	 * very long-standing error, and fixing it now risks breakage of
1416 	 * applications, so we live with it
1417 	 */
1418 	 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1419 			 new_rlim->rlim_cur != RLIM_INFINITY)
1420 		update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1421 out:
1422 	read_unlock(&tasklist_lock);
1423 	return retval;
1424 }
1425 
1426 /* rcu lock must be held */
1427 static int check_prlimit_permission(struct task_struct *task)
1428 {
1429 	const struct cred *cred = current_cred(), *tcred;
1430 
1431 	if (current == task)
1432 		return 0;
1433 
1434 	tcred = __task_cred(task);
1435 	if (uid_eq(cred->uid, tcred->euid) &&
1436 	    uid_eq(cred->uid, tcred->suid) &&
1437 	    uid_eq(cred->uid, tcred->uid)  &&
1438 	    gid_eq(cred->gid, tcred->egid) &&
1439 	    gid_eq(cred->gid, tcred->sgid) &&
1440 	    gid_eq(cred->gid, tcred->gid))
1441 		return 0;
1442 	if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1443 		return 0;
1444 
1445 	return -EPERM;
1446 }
1447 
1448 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1449 		const struct rlimit64 __user *, new_rlim,
1450 		struct rlimit64 __user *, old_rlim)
1451 {
1452 	struct rlimit64 old64, new64;
1453 	struct rlimit old, new;
1454 	struct task_struct *tsk;
1455 	int ret;
1456 
1457 	if (new_rlim) {
1458 		if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1459 			return -EFAULT;
1460 		rlim64_to_rlim(&new64, &new);
1461 	}
1462 
1463 	rcu_read_lock();
1464 	tsk = pid ? find_task_by_vpid(pid) : current;
1465 	if (!tsk) {
1466 		rcu_read_unlock();
1467 		return -ESRCH;
1468 	}
1469 	ret = check_prlimit_permission(tsk);
1470 	if (ret) {
1471 		rcu_read_unlock();
1472 		return ret;
1473 	}
1474 	get_task_struct(tsk);
1475 	rcu_read_unlock();
1476 
1477 	ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1478 			old_rlim ? &old : NULL);
1479 
1480 	if (!ret && old_rlim) {
1481 		rlim_to_rlim64(&old, &old64);
1482 		if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1483 			ret = -EFAULT;
1484 	}
1485 
1486 	put_task_struct(tsk);
1487 	return ret;
1488 }
1489 
1490 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1491 {
1492 	struct rlimit new_rlim;
1493 
1494 	if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1495 		return -EFAULT;
1496 	return do_prlimit(current, resource, &new_rlim, NULL);
1497 }
1498 
1499 /*
1500  * It would make sense to put struct rusage in the task_struct,
1501  * except that would make the task_struct be *really big*.  After
1502  * task_struct gets moved into malloc'ed memory, it would
1503  * make sense to do this.  It will make moving the rest of the information
1504  * a lot simpler!  (Which we're not doing right now because we're not
1505  * measuring them yet).
1506  *
1507  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1508  * races with threads incrementing their own counters.  But since word
1509  * reads are atomic, we either get new values or old values and we don't
1510  * care which for the sums.  We always take the siglock to protect reading
1511  * the c* fields from p->signal from races with exit.c updating those
1512  * fields when reaping, so a sample either gets all the additions of a
1513  * given child after it's reaped, or none so this sample is before reaping.
1514  *
1515  * Locking:
1516  * We need to take the siglock for CHILDEREN, SELF and BOTH
1517  * for  the cases current multithreaded, non-current single threaded
1518  * non-current multithreaded.  Thread traversal is now safe with
1519  * the siglock held.
1520  * Strictly speaking, we donot need to take the siglock if we are current and
1521  * single threaded,  as no one else can take our signal_struct away, no one
1522  * else can  reap the  children to update signal->c* counters, and no one else
1523  * can race with the signal-> fields. If we do not take any lock, the
1524  * signal-> fields could be read out of order while another thread was just
1525  * exiting. So we should  place a read memory barrier when we avoid the lock.
1526  * On the writer side,  write memory barrier is implied in  __exit_signal
1527  * as __exit_signal releases  the siglock spinlock after updating the signal->
1528  * fields. But we don't do this yet to keep things simple.
1529  *
1530  */
1531 
1532 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1533 {
1534 	r->ru_nvcsw += t->nvcsw;
1535 	r->ru_nivcsw += t->nivcsw;
1536 	r->ru_minflt += t->min_flt;
1537 	r->ru_majflt += t->maj_flt;
1538 	r->ru_inblock += task_io_get_inblock(t);
1539 	r->ru_oublock += task_io_get_oublock(t);
1540 }
1541 
1542 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1543 {
1544 	struct task_struct *t;
1545 	unsigned long flags;
1546 	cputime_t tgutime, tgstime, utime, stime;
1547 	unsigned long maxrss = 0;
1548 
1549 	memset((char *)r, 0, sizeof (*r));
1550 	utime = stime = 0;
1551 
1552 	if (who == RUSAGE_THREAD) {
1553 		task_cputime_adjusted(current, &utime, &stime);
1554 		accumulate_thread_rusage(p, r);
1555 		maxrss = p->signal->maxrss;
1556 		goto out;
1557 	}
1558 
1559 	if (!lock_task_sighand(p, &flags))
1560 		return;
1561 
1562 	switch (who) {
1563 	case RUSAGE_BOTH:
1564 	case RUSAGE_CHILDREN:
1565 		utime = p->signal->cutime;
1566 		stime = p->signal->cstime;
1567 		r->ru_nvcsw = p->signal->cnvcsw;
1568 		r->ru_nivcsw = p->signal->cnivcsw;
1569 		r->ru_minflt = p->signal->cmin_flt;
1570 		r->ru_majflt = p->signal->cmaj_flt;
1571 		r->ru_inblock = p->signal->cinblock;
1572 		r->ru_oublock = p->signal->coublock;
1573 		maxrss = p->signal->cmaxrss;
1574 
1575 		if (who == RUSAGE_CHILDREN)
1576 			break;
1577 
1578 	case RUSAGE_SELF:
1579 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1580 		utime += tgutime;
1581 		stime += tgstime;
1582 		r->ru_nvcsw += p->signal->nvcsw;
1583 		r->ru_nivcsw += p->signal->nivcsw;
1584 		r->ru_minflt += p->signal->min_flt;
1585 		r->ru_majflt += p->signal->maj_flt;
1586 		r->ru_inblock += p->signal->inblock;
1587 		r->ru_oublock += p->signal->oublock;
1588 		if (maxrss < p->signal->maxrss)
1589 			maxrss = p->signal->maxrss;
1590 		t = p;
1591 		do {
1592 			accumulate_thread_rusage(t, r);
1593 		} while_each_thread(p, t);
1594 		break;
1595 
1596 	default:
1597 		BUG();
1598 	}
1599 	unlock_task_sighand(p, &flags);
1600 
1601 out:
1602 	cputime_to_timeval(utime, &r->ru_utime);
1603 	cputime_to_timeval(stime, &r->ru_stime);
1604 
1605 	if (who != RUSAGE_CHILDREN) {
1606 		struct mm_struct *mm = get_task_mm(p);
1607 
1608 		if (mm) {
1609 			setmax_mm_hiwater_rss(&maxrss, mm);
1610 			mmput(mm);
1611 		}
1612 	}
1613 	r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1614 }
1615 
1616 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1617 {
1618 	struct rusage r;
1619 
1620 	k_getrusage(p, who, &r);
1621 	return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1622 }
1623 
1624 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1625 {
1626 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1627 	    who != RUSAGE_THREAD)
1628 		return -EINVAL;
1629 	return getrusage(current, who, ru);
1630 }
1631 
1632 #ifdef CONFIG_COMPAT
1633 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1634 {
1635 	struct rusage r;
1636 
1637 	if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1638 	    who != RUSAGE_THREAD)
1639 		return -EINVAL;
1640 
1641 	k_getrusage(current, who, &r);
1642 	return put_compat_rusage(&r, ru);
1643 }
1644 #endif
1645 
1646 SYSCALL_DEFINE1(umask, int, mask)
1647 {
1648 	mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1649 	return mask;
1650 }
1651 
1652 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1653 {
1654 	struct fd exe;
1655 	struct file *old_exe, *exe_file;
1656 	struct inode *inode;
1657 	int err;
1658 
1659 	exe = fdget(fd);
1660 	if (!exe.file)
1661 		return -EBADF;
1662 
1663 	inode = file_inode(exe.file);
1664 
1665 	/*
1666 	 * Because the original mm->exe_file points to executable file, make
1667 	 * sure that this one is executable as well, to avoid breaking an
1668 	 * overall picture.
1669 	 */
1670 	err = -EACCES;
1671 	if (!S_ISREG(inode->i_mode)	||
1672 	    exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1673 		goto exit;
1674 
1675 	err = inode_permission(inode, MAY_EXEC);
1676 	if (err)
1677 		goto exit;
1678 
1679 	/*
1680 	 * Forbid mm->exe_file change if old file still mapped.
1681 	 */
1682 	exe_file = get_mm_exe_file(mm);
1683 	err = -EBUSY;
1684 	if (exe_file) {
1685 		struct vm_area_struct *vma;
1686 
1687 		down_read(&mm->mmap_sem);
1688 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
1689 			if (!vma->vm_file)
1690 				continue;
1691 			if (path_equal(&vma->vm_file->f_path,
1692 				       &exe_file->f_path))
1693 				goto exit_err;
1694 		}
1695 
1696 		up_read(&mm->mmap_sem);
1697 		fput(exe_file);
1698 	}
1699 
1700 	/*
1701 	 * The symlink can be changed only once, just to disallow arbitrary
1702 	 * transitions malicious software might bring in. This means one
1703 	 * could make a snapshot over all processes running and monitor
1704 	 * /proc/pid/exe changes to notice unusual activity if needed.
1705 	 */
1706 	err = -EPERM;
1707 	if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1708 		goto exit;
1709 
1710 	err = 0;
1711 	/* set the new file, lockless */
1712 	get_file(exe.file);
1713 	old_exe = xchg(&mm->exe_file, exe.file);
1714 	if (old_exe)
1715 		fput(old_exe);
1716 exit:
1717 	fdput(exe);
1718 	return err;
1719 exit_err:
1720 	up_read(&mm->mmap_sem);
1721 	fput(exe_file);
1722 	goto exit;
1723 }
1724 
1725 #ifdef CONFIG_CHECKPOINT_RESTORE
1726 /*
1727  * WARNING: we don't require any capability here so be very careful
1728  * in what is allowed for modification from userspace.
1729  */
1730 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1731 {
1732 	unsigned long mmap_max_addr = TASK_SIZE;
1733 	struct mm_struct *mm = current->mm;
1734 	int error = -EINVAL, i;
1735 
1736 	static const unsigned char offsets[] = {
1737 		offsetof(struct prctl_mm_map, start_code),
1738 		offsetof(struct prctl_mm_map, end_code),
1739 		offsetof(struct prctl_mm_map, start_data),
1740 		offsetof(struct prctl_mm_map, end_data),
1741 		offsetof(struct prctl_mm_map, start_brk),
1742 		offsetof(struct prctl_mm_map, brk),
1743 		offsetof(struct prctl_mm_map, start_stack),
1744 		offsetof(struct prctl_mm_map, arg_start),
1745 		offsetof(struct prctl_mm_map, arg_end),
1746 		offsetof(struct prctl_mm_map, env_start),
1747 		offsetof(struct prctl_mm_map, env_end),
1748 	};
1749 
1750 	/*
1751 	 * Make sure the members are not somewhere outside
1752 	 * of allowed address space.
1753 	 */
1754 	for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1755 		u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1756 
1757 		if ((unsigned long)val >= mmap_max_addr ||
1758 		    (unsigned long)val < mmap_min_addr)
1759 			goto out;
1760 	}
1761 
1762 	/*
1763 	 * Make sure the pairs are ordered.
1764 	 */
1765 #define __prctl_check_order(__m1, __op, __m2)				\
1766 	((unsigned long)prctl_map->__m1 __op				\
1767 	 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1768 	error  = __prctl_check_order(start_code, <, end_code);
1769 	error |= __prctl_check_order(start_data, <, end_data);
1770 	error |= __prctl_check_order(start_brk, <=, brk);
1771 	error |= __prctl_check_order(arg_start, <=, arg_end);
1772 	error |= __prctl_check_order(env_start, <=, env_end);
1773 	if (error)
1774 		goto out;
1775 #undef __prctl_check_order
1776 
1777 	error = -EINVAL;
1778 
1779 	/*
1780 	 * @brk should be after @end_data in traditional maps.
1781 	 */
1782 	if (prctl_map->start_brk <= prctl_map->end_data ||
1783 	    prctl_map->brk <= prctl_map->end_data)
1784 		goto out;
1785 
1786 	/*
1787 	 * Neither we should allow to override limits if they set.
1788 	 */
1789 	if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1790 			      prctl_map->start_brk, prctl_map->end_data,
1791 			      prctl_map->start_data))
1792 			goto out;
1793 
1794 	/*
1795 	 * Someone is trying to cheat the auxv vector.
1796 	 */
1797 	if (prctl_map->auxv_size) {
1798 		if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1799 			goto out;
1800 	}
1801 
1802 	/*
1803 	 * Finally, make sure the caller has the rights to
1804 	 * change /proc/pid/exe link: only local root should
1805 	 * be allowed to.
1806 	 */
1807 	if (prctl_map->exe_fd != (u32)-1) {
1808 		struct user_namespace *ns = current_user_ns();
1809 		const struct cred *cred = current_cred();
1810 
1811 		if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1812 		    !gid_eq(cred->gid, make_kgid(ns, 0)))
1813 			goto out;
1814 	}
1815 
1816 	error = 0;
1817 out:
1818 	return error;
1819 }
1820 
1821 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1822 {
1823 	struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1824 	unsigned long user_auxv[AT_VECTOR_SIZE];
1825 	struct mm_struct *mm = current->mm;
1826 	int error;
1827 
1828 	BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1829 	BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1830 
1831 	if (opt == PR_SET_MM_MAP_SIZE)
1832 		return put_user((unsigned int)sizeof(prctl_map),
1833 				(unsigned int __user *)addr);
1834 
1835 	if (data_size != sizeof(prctl_map))
1836 		return -EINVAL;
1837 
1838 	if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1839 		return -EFAULT;
1840 
1841 	error = validate_prctl_map(&prctl_map);
1842 	if (error)
1843 		return error;
1844 
1845 	if (prctl_map.auxv_size) {
1846 		memset(user_auxv, 0, sizeof(user_auxv));
1847 		if (copy_from_user(user_auxv,
1848 				   (const void __user *)prctl_map.auxv,
1849 				   prctl_map.auxv_size))
1850 			return -EFAULT;
1851 
1852 		/* Last entry must be AT_NULL as specification requires */
1853 		user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1854 		user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1855 	}
1856 
1857 	if (prctl_map.exe_fd != (u32)-1)
1858 		error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1859 	down_read(&mm->mmap_sem);
1860 	if (error)
1861 		goto out;
1862 
1863 	/*
1864 	 * We don't validate if these members are pointing to
1865 	 * real present VMAs because application may have correspond
1866 	 * VMAs already unmapped and kernel uses these members for statistics
1867 	 * output in procfs mostly, except
1868 	 *
1869 	 *  - @start_brk/@brk which are used in do_brk but kernel lookups
1870 	 *    for VMAs when updating these memvers so anything wrong written
1871 	 *    here cause kernel to swear at userspace program but won't lead
1872 	 *    to any problem in kernel itself
1873 	 */
1874 
1875 	mm->start_code	= prctl_map.start_code;
1876 	mm->end_code	= prctl_map.end_code;
1877 	mm->start_data	= prctl_map.start_data;
1878 	mm->end_data	= prctl_map.end_data;
1879 	mm->start_brk	= prctl_map.start_brk;
1880 	mm->brk		= prctl_map.brk;
1881 	mm->start_stack	= prctl_map.start_stack;
1882 	mm->arg_start	= prctl_map.arg_start;
1883 	mm->arg_end	= prctl_map.arg_end;
1884 	mm->env_start	= prctl_map.env_start;
1885 	mm->env_end	= prctl_map.env_end;
1886 
1887 	/*
1888 	 * Note this update of @saved_auxv is lockless thus
1889 	 * if someone reads this member in procfs while we're
1890 	 * updating -- it may get partly updated results. It's
1891 	 * known and acceptable trade off: we leave it as is to
1892 	 * not introduce additional locks here making the kernel
1893 	 * more complex.
1894 	 */
1895 	if (prctl_map.auxv_size)
1896 		memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1897 
1898 	error = 0;
1899 out:
1900 	up_read(&mm->mmap_sem);
1901 	return error;
1902 }
1903 #endif /* CONFIG_CHECKPOINT_RESTORE */
1904 
1905 static int prctl_set_mm(int opt, unsigned long addr,
1906 			unsigned long arg4, unsigned long arg5)
1907 {
1908 	struct mm_struct *mm = current->mm;
1909 	struct vm_area_struct *vma;
1910 	int error;
1911 
1912 	if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1913 			      opt != PR_SET_MM_MAP &&
1914 			      opt != PR_SET_MM_MAP_SIZE)))
1915 		return -EINVAL;
1916 
1917 #ifdef CONFIG_CHECKPOINT_RESTORE
1918 	if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1919 		return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1920 #endif
1921 
1922 	if (!capable(CAP_SYS_RESOURCE))
1923 		return -EPERM;
1924 
1925 	if (opt == PR_SET_MM_EXE_FILE)
1926 		return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1927 
1928 	if (addr >= TASK_SIZE || addr < mmap_min_addr)
1929 		return -EINVAL;
1930 
1931 	error = -EINVAL;
1932 
1933 	down_read(&mm->mmap_sem);
1934 	vma = find_vma(mm, addr);
1935 
1936 	switch (opt) {
1937 	case PR_SET_MM_START_CODE:
1938 		mm->start_code = addr;
1939 		break;
1940 	case PR_SET_MM_END_CODE:
1941 		mm->end_code = addr;
1942 		break;
1943 	case PR_SET_MM_START_DATA:
1944 		mm->start_data = addr;
1945 		break;
1946 	case PR_SET_MM_END_DATA:
1947 		mm->end_data = addr;
1948 		break;
1949 
1950 	case PR_SET_MM_START_BRK:
1951 		if (addr <= mm->end_data)
1952 			goto out;
1953 
1954 		if (check_data_rlimit(rlimit(RLIMIT_DATA), mm->brk, addr,
1955 				      mm->end_data, mm->start_data))
1956 			goto out;
1957 
1958 		mm->start_brk = addr;
1959 		break;
1960 
1961 	case PR_SET_MM_BRK:
1962 		if (addr <= mm->end_data)
1963 			goto out;
1964 
1965 		if (check_data_rlimit(rlimit(RLIMIT_DATA), addr, mm->start_brk,
1966 				      mm->end_data, mm->start_data))
1967 			goto out;
1968 
1969 		mm->brk = addr;
1970 		break;
1971 
1972 	/*
1973 	 * If command line arguments and environment
1974 	 * are placed somewhere else on stack, we can
1975 	 * set them up here, ARG_START/END to setup
1976 	 * command line argumets and ENV_START/END
1977 	 * for environment.
1978 	 */
1979 	case PR_SET_MM_START_STACK:
1980 	case PR_SET_MM_ARG_START:
1981 	case PR_SET_MM_ARG_END:
1982 	case PR_SET_MM_ENV_START:
1983 	case PR_SET_MM_ENV_END:
1984 		if (!vma) {
1985 			error = -EFAULT;
1986 			goto out;
1987 		}
1988 		if (opt == PR_SET_MM_START_STACK)
1989 			mm->start_stack = addr;
1990 		else if (opt == PR_SET_MM_ARG_START)
1991 			mm->arg_start = addr;
1992 		else if (opt == PR_SET_MM_ARG_END)
1993 			mm->arg_end = addr;
1994 		else if (opt == PR_SET_MM_ENV_START)
1995 			mm->env_start = addr;
1996 		else if (opt == PR_SET_MM_ENV_END)
1997 			mm->env_end = addr;
1998 		break;
1999 
2000 	/*
2001 	 * This doesn't move auxiliary vector itself
2002 	 * since it's pinned to mm_struct, but allow
2003 	 * to fill vector with new values. It's up
2004 	 * to a caller to provide sane values here
2005 	 * otherwise user space tools which use this
2006 	 * vector might be unhappy.
2007 	 */
2008 	case PR_SET_MM_AUXV: {
2009 		unsigned long user_auxv[AT_VECTOR_SIZE];
2010 
2011 		if (arg4 > sizeof(user_auxv))
2012 			goto out;
2013 		up_read(&mm->mmap_sem);
2014 
2015 		if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
2016 			return -EFAULT;
2017 
2018 		/* Make sure the last entry is always AT_NULL */
2019 		user_auxv[AT_VECTOR_SIZE - 2] = 0;
2020 		user_auxv[AT_VECTOR_SIZE - 1] = 0;
2021 
2022 		BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2023 
2024 		task_lock(current);
2025 		memcpy(mm->saved_auxv, user_auxv, arg4);
2026 		task_unlock(current);
2027 
2028 		return 0;
2029 	}
2030 	default:
2031 		goto out;
2032 	}
2033 
2034 	error = 0;
2035 out:
2036 	up_read(&mm->mmap_sem);
2037 	return error;
2038 }
2039 
2040 #ifdef CONFIG_CHECKPOINT_RESTORE
2041 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2042 {
2043 	return put_user(me->clear_child_tid, tid_addr);
2044 }
2045 #else
2046 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2047 {
2048 	return -EINVAL;
2049 }
2050 #endif
2051 
2052 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2053 		unsigned long, arg4, unsigned long, arg5)
2054 {
2055 	struct task_struct *me = current;
2056 	unsigned char comm[sizeof(me->comm)];
2057 	long error;
2058 
2059 	error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2060 	if (error != -ENOSYS)
2061 		return error;
2062 
2063 	error = 0;
2064 	switch (option) {
2065 	case PR_SET_PDEATHSIG:
2066 		if (!valid_signal(arg2)) {
2067 			error = -EINVAL;
2068 			break;
2069 		}
2070 		me->pdeath_signal = arg2;
2071 		break;
2072 	case PR_GET_PDEATHSIG:
2073 		error = put_user(me->pdeath_signal, (int __user *)arg2);
2074 		break;
2075 	case PR_GET_DUMPABLE:
2076 		error = get_dumpable(me->mm);
2077 		break;
2078 	case PR_SET_DUMPABLE:
2079 		if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2080 			error = -EINVAL;
2081 			break;
2082 		}
2083 		set_dumpable(me->mm, arg2);
2084 		break;
2085 
2086 	case PR_SET_UNALIGN:
2087 		error = SET_UNALIGN_CTL(me, arg2);
2088 		break;
2089 	case PR_GET_UNALIGN:
2090 		error = GET_UNALIGN_CTL(me, arg2);
2091 		break;
2092 	case PR_SET_FPEMU:
2093 		error = SET_FPEMU_CTL(me, arg2);
2094 		break;
2095 	case PR_GET_FPEMU:
2096 		error = GET_FPEMU_CTL(me, arg2);
2097 		break;
2098 	case PR_SET_FPEXC:
2099 		error = SET_FPEXC_CTL(me, arg2);
2100 		break;
2101 	case PR_GET_FPEXC:
2102 		error = GET_FPEXC_CTL(me, arg2);
2103 		break;
2104 	case PR_GET_TIMING:
2105 		error = PR_TIMING_STATISTICAL;
2106 		break;
2107 	case PR_SET_TIMING:
2108 		if (arg2 != PR_TIMING_STATISTICAL)
2109 			error = -EINVAL;
2110 		break;
2111 	case PR_SET_NAME:
2112 		comm[sizeof(me->comm) - 1] = 0;
2113 		if (strncpy_from_user(comm, (char __user *)arg2,
2114 				      sizeof(me->comm) - 1) < 0)
2115 			return -EFAULT;
2116 		set_task_comm(me, comm);
2117 		proc_comm_connector(me);
2118 		break;
2119 	case PR_GET_NAME:
2120 		get_task_comm(comm, me);
2121 		if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2122 			return -EFAULT;
2123 		break;
2124 	case PR_GET_ENDIAN:
2125 		error = GET_ENDIAN(me, arg2);
2126 		break;
2127 	case PR_SET_ENDIAN:
2128 		error = SET_ENDIAN(me, arg2);
2129 		break;
2130 	case PR_GET_SECCOMP:
2131 		error = prctl_get_seccomp();
2132 		break;
2133 	case PR_SET_SECCOMP:
2134 		error = prctl_set_seccomp(arg2, (char __user *)arg3);
2135 		break;
2136 	case PR_GET_TSC:
2137 		error = GET_TSC_CTL(arg2);
2138 		break;
2139 	case PR_SET_TSC:
2140 		error = SET_TSC_CTL(arg2);
2141 		break;
2142 	case PR_TASK_PERF_EVENTS_DISABLE:
2143 		error = perf_event_task_disable();
2144 		break;
2145 	case PR_TASK_PERF_EVENTS_ENABLE:
2146 		error = perf_event_task_enable();
2147 		break;
2148 	case PR_GET_TIMERSLACK:
2149 		error = current->timer_slack_ns;
2150 		break;
2151 	case PR_SET_TIMERSLACK:
2152 		if (arg2 <= 0)
2153 			current->timer_slack_ns =
2154 					current->default_timer_slack_ns;
2155 		else
2156 			current->timer_slack_ns = arg2;
2157 		break;
2158 	case PR_MCE_KILL:
2159 		if (arg4 | arg5)
2160 			return -EINVAL;
2161 		switch (arg2) {
2162 		case PR_MCE_KILL_CLEAR:
2163 			if (arg3 != 0)
2164 				return -EINVAL;
2165 			current->flags &= ~PF_MCE_PROCESS;
2166 			break;
2167 		case PR_MCE_KILL_SET:
2168 			current->flags |= PF_MCE_PROCESS;
2169 			if (arg3 == PR_MCE_KILL_EARLY)
2170 				current->flags |= PF_MCE_EARLY;
2171 			else if (arg3 == PR_MCE_KILL_LATE)
2172 				current->flags &= ~PF_MCE_EARLY;
2173 			else if (arg3 == PR_MCE_KILL_DEFAULT)
2174 				current->flags &=
2175 						~(PF_MCE_EARLY|PF_MCE_PROCESS);
2176 			else
2177 				return -EINVAL;
2178 			break;
2179 		default:
2180 			return -EINVAL;
2181 		}
2182 		break;
2183 	case PR_MCE_KILL_GET:
2184 		if (arg2 | arg3 | arg4 | arg5)
2185 			return -EINVAL;
2186 		if (current->flags & PF_MCE_PROCESS)
2187 			error = (current->flags & PF_MCE_EARLY) ?
2188 				PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2189 		else
2190 			error = PR_MCE_KILL_DEFAULT;
2191 		break;
2192 	case PR_SET_MM:
2193 		error = prctl_set_mm(arg2, arg3, arg4, arg5);
2194 		break;
2195 	case PR_GET_TID_ADDRESS:
2196 		error = prctl_get_tid_address(me, (int __user **)arg2);
2197 		break;
2198 	case PR_SET_CHILD_SUBREAPER:
2199 		me->signal->is_child_subreaper = !!arg2;
2200 		break;
2201 	case PR_GET_CHILD_SUBREAPER:
2202 		error = put_user(me->signal->is_child_subreaper,
2203 				 (int __user *)arg2);
2204 		break;
2205 	case PR_SET_NO_NEW_PRIVS:
2206 		if (arg2 != 1 || arg3 || arg4 || arg5)
2207 			return -EINVAL;
2208 
2209 		task_set_no_new_privs(current);
2210 		break;
2211 	case PR_GET_NO_NEW_PRIVS:
2212 		if (arg2 || arg3 || arg4 || arg5)
2213 			return -EINVAL;
2214 		return task_no_new_privs(current) ? 1 : 0;
2215 	case PR_GET_THP_DISABLE:
2216 		if (arg2 || arg3 || arg4 || arg5)
2217 			return -EINVAL;
2218 		error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2219 		break;
2220 	case PR_SET_THP_DISABLE:
2221 		if (arg3 || arg4 || arg5)
2222 			return -EINVAL;
2223 		down_write(&me->mm->mmap_sem);
2224 		if (arg2)
2225 			me->mm->def_flags |= VM_NOHUGEPAGE;
2226 		else
2227 			me->mm->def_flags &= ~VM_NOHUGEPAGE;
2228 		up_write(&me->mm->mmap_sem);
2229 		break;
2230 	case PR_MPX_ENABLE_MANAGEMENT:
2231 		if (arg2 || arg3 || arg4 || arg5)
2232 			return -EINVAL;
2233 		error = MPX_ENABLE_MANAGEMENT(me);
2234 		break;
2235 	case PR_MPX_DISABLE_MANAGEMENT:
2236 		if (arg2 || arg3 || arg4 || arg5)
2237 			return -EINVAL;
2238 		error = MPX_DISABLE_MANAGEMENT(me);
2239 		break;
2240 	case PR_SET_FP_MODE:
2241 		error = SET_FP_MODE(me, arg2);
2242 		break;
2243 	case PR_GET_FP_MODE:
2244 		error = GET_FP_MODE(me);
2245 		break;
2246 	default:
2247 		error = -EINVAL;
2248 		break;
2249 	}
2250 	return error;
2251 }
2252 
2253 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2254 		struct getcpu_cache __user *, unused)
2255 {
2256 	int err = 0;
2257 	int cpu = raw_smp_processor_id();
2258 
2259 	if (cpup)
2260 		err |= put_user(cpu, cpup);
2261 	if (nodep)
2262 		err |= put_user(cpu_to_node(cpu), nodep);
2263 	return err ? -EFAULT : 0;
2264 }
2265 
2266 /**
2267  * do_sysinfo - fill in sysinfo struct
2268  * @info: pointer to buffer to fill
2269  */
2270 static int do_sysinfo(struct sysinfo *info)
2271 {
2272 	unsigned long mem_total, sav_total;
2273 	unsigned int mem_unit, bitcount;
2274 	struct timespec tp;
2275 
2276 	memset(info, 0, sizeof(struct sysinfo));
2277 
2278 	get_monotonic_boottime(&tp);
2279 	info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2280 
2281 	get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2282 
2283 	info->procs = nr_threads;
2284 
2285 	si_meminfo(info);
2286 	si_swapinfo(info);
2287 
2288 	/*
2289 	 * If the sum of all the available memory (i.e. ram + swap)
2290 	 * is less than can be stored in a 32 bit unsigned long then
2291 	 * we can be binary compatible with 2.2.x kernels.  If not,
2292 	 * well, in that case 2.2.x was broken anyways...
2293 	 *
2294 	 *  -Erik Andersen <andersee@debian.org>
2295 	 */
2296 
2297 	mem_total = info->totalram + info->totalswap;
2298 	if (mem_total < info->totalram || mem_total < info->totalswap)
2299 		goto out;
2300 	bitcount = 0;
2301 	mem_unit = info->mem_unit;
2302 	while (mem_unit > 1) {
2303 		bitcount++;
2304 		mem_unit >>= 1;
2305 		sav_total = mem_total;
2306 		mem_total <<= 1;
2307 		if (mem_total < sav_total)
2308 			goto out;
2309 	}
2310 
2311 	/*
2312 	 * If mem_total did not overflow, multiply all memory values by
2313 	 * info->mem_unit and set it to 1.  This leaves things compatible
2314 	 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2315 	 * kernels...
2316 	 */
2317 
2318 	info->mem_unit = 1;
2319 	info->totalram <<= bitcount;
2320 	info->freeram <<= bitcount;
2321 	info->sharedram <<= bitcount;
2322 	info->bufferram <<= bitcount;
2323 	info->totalswap <<= bitcount;
2324 	info->freeswap <<= bitcount;
2325 	info->totalhigh <<= bitcount;
2326 	info->freehigh <<= bitcount;
2327 
2328 out:
2329 	return 0;
2330 }
2331 
2332 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2333 {
2334 	struct sysinfo val;
2335 
2336 	do_sysinfo(&val);
2337 
2338 	if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2339 		return -EFAULT;
2340 
2341 	return 0;
2342 }
2343 
2344 #ifdef CONFIG_COMPAT
2345 struct compat_sysinfo {
2346 	s32 uptime;
2347 	u32 loads[3];
2348 	u32 totalram;
2349 	u32 freeram;
2350 	u32 sharedram;
2351 	u32 bufferram;
2352 	u32 totalswap;
2353 	u32 freeswap;
2354 	u16 procs;
2355 	u16 pad;
2356 	u32 totalhigh;
2357 	u32 freehigh;
2358 	u32 mem_unit;
2359 	char _f[20-2*sizeof(u32)-sizeof(int)];
2360 };
2361 
2362 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2363 {
2364 	struct sysinfo s;
2365 
2366 	do_sysinfo(&s);
2367 
2368 	/* Check to see if any memory value is too large for 32-bit and scale
2369 	 *  down if needed
2370 	 */
2371 	if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2372 		int bitcount = 0;
2373 
2374 		while (s.mem_unit < PAGE_SIZE) {
2375 			s.mem_unit <<= 1;
2376 			bitcount++;
2377 		}
2378 
2379 		s.totalram >>= bitcount;
2380 		s.freeram >>= bitcount;
2381 		s.sharedram >>= bitcount;
2382 		s.bufferram >>= bitcount;
2383 		s.totalswap >>= bitcount;
2384 		s.freeswap >>= bitcount;
2385 		s.totalhigh >>= bitcount;
2386 		s.freehigh >>= bitcount;
2387 	}
2388 
2389 	if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2390 	    __put_user(s.uptime, &info->uptime) ||
2391 	    __put_user(s.loads[0], &info->loads[0]) ||
2392 	    __put_user(s.loads[1], &info->loads[1]) ||
2393 	    __put_user(s.loads[2], &info->loads[2]) ||
2394 	    __put_user(s.totalram, &info->totalram) ||
2395 	    __put_user(s.freeram, &info->freeram) ||
2396 	    __put_user(s.sharedram, &info->sharedram) ||
2397 	    __put_user(s.bufferram, &info->bufferram) ||
2398 	    __put_user(s.totalswap, &info->totalswap) ||
2399 	    __put_user(s.freeswap, &info->freeswap) ||
2400 	    __put_user(s.procs, &info->procs) ||
2401 	    __put_user(s.totalhigh, &info->totalhigh) ||
2402 	    __put_user(s.freehigh, &info->freehigh) ||
2403 	    __put_user(s.mem_unit, &info->mem_unit))
2404 		return -EFAULT;
2405 
2406 	return 0;
2407 }
2408 #endif /* CONFIG_COMPAT */
2409