projects/linux·blob v4.0-rc1 · refs

2396 lines · 56.5 KiB

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