projects/linux·blob v3.12-rc3 · refs

2163 lines · 50.2 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/kexec.h>
20#include <linux/workqueue.h>
21#include <linux/capability.h>
22#include <linux/device.h>
23#include <linux/key.h>
24#include <linux/times.h>
25#include <linux/posix-timers.h>
26#include <linux/security.h>
27#include <linux/dcookies.h>
28#include <linux/suspend.h>
29#include <linux/tty.h>
30#include <linux/signal.h>
31#include <linux/cn_proc.h>
32#include <linux/getcpu.h>
33#include <linux/task_io_accounting_ops.h>
34#include <linux/seccomp.h>
35#include <linux/cpu.h>
36#include <linux/personality.h>
37#include <linux/ptrace.h>
38#include <linux/fs_struct.h>
39#include <linux/file.h>
40#include <linux/mount.h>
41#include <linux/gfp.h>
42#include <linux/syscore_ops.h>
43#include <linux/version.h>
44#include <linux/ctype.h>
45
46#include <linux/compat.h>
47#include <linux/syscalls.h>
48#include <linux/kprobes.h>
49#include <linux/user_namespace.h>
50#include <linux/binfmts.h>
51
52#include <linux/sched.h>
53#include <linux/rcupdate.h>
54#include <linux/uidgid.h>
55#include <linux/cred.h>
56
57#include <linux/kmsg_dump.h>
58/* Move somewhere else to avoid recompiling? */
59#include <generated/utsrelease.h>
60
61#include <asm/uaccess.h>
62#include <asm/io.h>
63#include <asm/unistd.h>
64
65#ifndef SET_UNALIGN_CTL
66# define SET_UNALIGN_CTL(a,b) (-EINVAL)
67#endif
68#ifndef GET_UNALIGN_CTL
69# define GET_UNALIGN_CTL(a,b) (-EINVAL)
70#endif
71#ifndef SET_FPEMU_CTL
72# define SET_FPEMU_CTL(a,b) (-EINVAL)
73#endif
74#ifndef GET_FPEMU_CTL
75# define GET_FPEMU_CTL(a,b) (-EINVAL)
76#endif
77#ifndef SET_FPEXC_CTL
78# define SET_FPEXC_CTL(a,b) (-EINVAL)
79#endif
80#ifndef GET_FPEXC_CTL
81# define GET_FPEXC_CTL(a,b) (-EINVAL)
82#endif
83#ifndef GET_ENDIAN
84# define GET_ENDIAN(a,b) (-EINVAL)
85#endif
86#ifndef SET_ENDIAN
87# define SET_ENDIAN(a,b) (-EINVAL)
88#endif
89#ifndef GET_TSC_CTL
90# define GET_TSC_CTL(a) (-EINVAL)
91#endif
92#ifndef SET_TSC_CTL
93# define SET_TSC_CTL(a) (-EINVAL)
94#endif
95
96/*
97 * this is where the system-wide overflow UID and GID are defined, for
98 * architectures that now have 32-bit UID/GID but didn't in the past
99 */
100
101int overflowuid = DEFAULT_OVERFLOWUID;
102int overflowgid = DEFAULT_OVERFLOWGID;
103
104EXPORT_SYMBOL(overflowuid);
105EXPORT_SYMBOL(overflowgid);
106
107/*
108 * the same as above, but for filesystems which can only store a 16-bit
109 * UID and GID. as such, this is needed on all architectures
110 */
111
112int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
113int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
114
115EXPORT_SYMBOL(fs_overflowuid);
116EXPORT_SYMBOL(fs_overflowgid);
117
118/*
119 * Returns true if current's euid is same as p's uid or euid,
120 * or has CAP_SYS_NICE to p's user_ns.
121 *
122 * Called with rcu_read_lock, creds are safe
123 */
124static bool set_one_prio_perm(struct task_struct *p)
125{
126 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
127
128 if (uid_eq(pcred->uid, cred->euid) ||
129 uid_eq(pcred->euid, cred->euid))
130 return true;
131 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
132 return true;
133 return false;
134}
135
136/*
137 * set the priority of a task
138 * - the caller must hold the RCU read lock
139 */
140static int set_one_prio(struct task_struct *p, int niceval, int error)
141{
142 int no_nice;
143
144 if (!set_one_prio_perm(p)) {
145 error = -EPERM;
146 goto out;
147 }
148 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
149 error = -EACCES;
150 goto out;
151 }
152 no_nice = security_task_setnice(p, niceval);
153 if (no_nice) {
154 error = no_nice;
155 goto out;
156 }
157 if (error == -ESRCH)
158 error = 0;
159 set_user_nice(p, niceval);
160out:
161 return error;
162}
163
164SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
165{
166 struct task_struct *g, *p;
167 struct user_struct *user;
168 const struct cred *cred = current_cred();
169 int error = -EINVAL;
170 struct pid *pgrp;
171 kuid_t uid;
172
173 if (which > PRIO_USER || which < PRIO_PROCESS)
174 goto out;
175
176 /* normalize: avoid signed division (rounding problems) */
177 error = -ESRCH;
178 if (niceval < -20)
179 niceval = -20;
180 if (niceval > 19)
181 niceval = 19;
182
183 rcu_read_lock();
184 read_lock(&tasklist_lock);
185 switch (which) {
186 case PRIO_PROCESS:
187 if (who)
188 p = find_task_by_vpid(who);
189 else
190 p = current;
191 if (p)
192 error = set_one_prio(p, niceval, error);
193 break;
194 case PRIO_PGRP:
195 if (who)
196 pgrp = find_vpid(who);
197 else
198 pgrp = task_pgrp(current);
199 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
200 error = set_one_prio(p, niceval, error);
201 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
202 break;
203 case PRIO_USER:
204 uid = make_kuid(cred->user_ns, who);
205 user = cred->user;
206 if (!who)
207 uid = cred->uid;
208 else if (!uid_eq(uid, cred->uid) &&
209 !(user = find_user(uid)))
210 goto out_unlock; /* No processes for this user */
211
212 do_each_thread(g, p) {
213 if (uid_eq(task_uid(p), uid))
214 error = set_one_prio(p, niceval, error);
215 } while_each_thread(g, p);
216 if (!uid_eq(uid, cred->uid))
217 free_uid(user); /* For find_user() */
218 break;
219 }
220out_unlock:
221 read_unlock(&tasklist_lock);
222 rcu_read_unlock();
223out:
224 return error;
225}
226
227/*
228 * Ugh. To avoid negative return values, "getpriority()" will
229 * not return the normal nice-value, but a negated value that
230 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
231 * to stay compatible.
232 */
233SYSCALL_DEFINE2(getpriority, int, which, int, who)
234{
235 struct task_struct *g, *p;
236 struct user_struct *user;
237 const struct cred *cred = current_cred();
238 long niceval, retval = -ESRCH;
239 struct pid *pgrp;
240 kuid_t uid;
241
242 if (which > PRIO_USER || which < PRIO_PROCESS)
243 return -EINVAL;
244
245 rcu_read_lock();
246 read_lock(&tasklist_lock);
247 switch (which) {
248 case PRIO_PROCESS:
249 if (who)
250 p = find_task_by_vpid(who);
251 else
252 p = current;
253 if (p) {
254 niceval = 20 - task_nice(p);
255 if (niceval > retval)
256 retval = niceval;
257 }
258 break;
259 case PRIO_PGRP:
260 if (who)
261 pgrp = find_vpid(who);
262 else
263 pgrp = task_pgrp(current);
264 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
265 niceval = 20 - task_nice(p);
266 if (niceval > retval)
267 retval = niceval;
268 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
269 break;
270 case PRIO_USER:
271 uid = make_kuid(cred->user_ns, who);
272 user = cred->user;
273 if (!who)
274 uid = cred->uid;
275 else if (!uid_eq(uid, cred->uid) &&
276 !(user = find_user(uid)))
277 goto out_unlock; /* No processes for this user */
278
279 do_each_thread(g, p) {
280 if (uid_eq(task_uid(p), uid)) {
281 niceval = 20 - task_nice(p);
282 if (niceval > retval)
283 retval = niceval;
284 }
285 } while_each_thread(g, p);
286 if (!uid_eq(uid, cred->uid))
287 free_uid(user); /* for find_user() */
288 break;
289 }
290out_unlock:
291 read_unlock(&tasklist_lock);
292 rcu_read_unlock();
293
294 return retval;
295}
296
297/*
298 * Unprivileged users may change the real gid to the effective gid
299 * or vice versa. (BSD-style)
300 *
301 * If you set the real gid at all, or set the effective gid to a value not
302 * equal to the real gid, then the saved gid is set to the new effective gid.
303 *
304 * This makes it possible for a setgid program to completely drop its
305 * privileges, which is often a useful assertion to make when you are doing
306 * a security audit over a program.
307 *
308 * The general idea is that a program which uses just setregid() will be
309 * 100% compatible with BSD. A program which uses just setgid() will be
310 * 100% compatible with POSIX with saved IDs.
311 *
312 * SMP: There are not races, the GIDs are checked only by filesystem
313 * operations (as far as semantic preservation is concerned).
314 */
315SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
316{
317 struct user_namespace *ns = current_user_ns();
318 const struct cred *old;
319 struct cred *new;
320 int retval;
321 kgid_t krgid, kegid;
322
323 krgid = make_kgid(ns, rgid);
324 kegid = make_kgid(ns, egid);
325
326 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
327 return -EINVAL;
328 if ((egid != (gid_t) -1) && !gid_valid(kegid))
329 return -EINVAL;
330
331 new = prepare_creds();
332 if (!new)
333 return -ENOMEM;
334 old = current_cred();
335
336 retval = -EPERM;
337 if (rgid != (gid_t) -1) {
338 if (gid_eq(old->gid, krgid) ||
339 gid_eq(old->egid, krgid) ||
340 ns_capable(old->user_ns, CAP_SETGID))
341 new->gid = krgid;
342 else
343 goto error;
344 }
345 if (egid != (gid_t) -1) {
346 if (gid_eq(old->gid, kegid) ||
347 gid_eq(old->egid, kegid) ||
348 gid_eq(old->sgid, kegid) ||
349 ns_capable(old->user_ns, CAP_SETGID))
350 new->egid = kegid;
351 else
352 goto error;
353 }
354
355 if (rgid != (gid_t) -1 ||
356 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
357 new->sgid = new->egid;
358 new->fsgid = new->egid;
359
360 return commit_creds(new);
361
362error:
363 abort_creds(new);
364 return retval;
365}
366
367/*
368 * setgid() is implemented like SysV w/ SAVED_IDS
369 *
370 * SMP: Same implicit races as above.
371 */
372SYSCALL_DEFINE1(setgid, gid_t, gid)
373{
374 struct user_namespace *ns = current_user_ns();
375 const struct cred *old;
376 struct cred *new;
377 int retval;
378 kgid_t kgid;
379
380 kgid = make_kgid(ns, gid);
381 if (!gid_valid(kgid))
382 return -EINVAL;
383
384 new = prepare_creds();
385 if (!new)
386 return -ENOMEM;
387 old = current_cred();
388
389 retval = -EPERM;
390 if (ns_capable(old->user_ns, CAP_SETGID))
391 new->gid = new->egid = new->sgid = new->fsgid = kgid;
392 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
393 new->egid = new->fsgid = kgid;
394 else
395 goto error;
396
397 return commit_creds(new);
398
399error:
400 abort_creds(new);
401 return retval;
402}
403
404/*
405 * change the user struct in a credentials set to match the new UID
406 */
407static int set_user(struct cred *new)
408{
409 struct user_struct *new_user;
410
411 new_user = alloc_uid(new->uid);
412 if (!new_user)
413 return -EAGAIN;
414
415 /*
416 * We don't fail in case of NPROC limit excess here because too many
417 * poorly written programs don't check set*uid() return code, assuming
418 * it never fails if called by root. We may still enforce NPROC limit
419 * for programs doing set*uid()+execve() by harmlessly deferring the
420 * failure to the execve() stage.
421 */
422 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
423 new_user != INIT_USER)
424 current->flags |= PF_NPROC_EXCEEDED;
425 else
426 current->flags &= ~PF_NPROC_EXCEEDED;
427
428 free_uid(new->user);
429 new->user = new_user;
430 return 0;
431}
432
433/*
434 * Unprivileged users may change the real uid to the effective uid
435 * or vice versa. (BSD-style)
436 *
437 * If you set the real uid at all, or set the effective uid to a value not
438 * equal to the real uid, then the saved uid is set to the new effective uid.
439 *
440 * This makes it possible for a setuid program to completely drop its
441 * privileges, which is often a useful assertion to make when you are doing
442 * a security audit over a program.
443 *
444 * The general idea is that a program which uses just setreuid() will be
445 * 100% compatible with BSD. A program which uses just setuid() will be
446 * 100% compatible with POSIX with saved IDs.
447 */
448SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
449{
450 struct user_namespace *ns = current_user_ns();
451 const struct cred *old;
452 struct cred *new;
453 int retval;
454 kuid_t kruid, keuid;
455
456 kruid = make_kuid(ns, ruid);
457 keuid = make_kuid(ns, euid);
458
459 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
460 return -EINVAL;
461 if ((euid != (uid_t) -1) && !uid_valid(keuid))
462 return -EINVAL;
463
464 new = prepare_creds();
465 if (!new)
466 return -ENOMEM;
467 old = current_cred();
468
469 retval = -EPERM;
470 if (ruid != (uid_t) -1) {
471 new->uid = kruid;
472 if (!uid_eq(old->uid, kruid) &&
473 !uid_eq(old->euid, kruid) &&
474 !ns_capable(old->user_ns, CAP_SETUID))
475 goto error;
476 }
477
478 if (euid != (uid_t) -1) {
479 new->euid = keuid;
480 if (!uid_eq(old->uid, keuid) &&
481 !uid_eq(old->euid, keuid) &&
482 !uid_eq(old->suid, keuid) &&
483 !ns_capable(old->user_ns, CAP_SETUID))
484 goto error;
485 }
486
487 if (!uid_eq(new->uid, old->uid)) {
488 retval = set_user(new);
489 if (retval < 0)
490 goto error;
491 }
492 if (ruid != (uid_t) -1 ||
493 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
494 new->suid = new->euid;
495 new->fsuid = new->euid;
496
497 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
498 if (retval < 0)
499 goto error;
500
501 return commit_creds(new);
502
503error:
504 abort_creds(new);
505 return retval;
506}
507
508/*
509 * setuid() is implemented like SysV with SAVED_IDS
510 *
511 * Note that SAVED_ID's is deficient in that a setuid root program
512 * like sendmail, for example, cannot set its uid to be a normal
513 * user and then switch back, because if you're root, setuid() sets
514 * the saved uid too. If you don't like this, blame the bright people
515 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
516 * will allow a root program to temporarily drop privileges and be able to
517 * regain them by swapping the real and effective uid.
518 */
519SYSCALL_DEFINE1(setuid, uid_t, uid)
520{
521 struct user_namespace *ns = current_user_ns();
522 const struct cred *old;
523 struct cred *new;
524 int retval;
525 kuid_t kuid;
526
527 kuid = make_kuid(ns, uid);
528 if (!uid_valid(kuid))
529 return -EINVAL;
530
531 new = prepare_creds();
532 if (!new)
533 return -ENOMEM;
534 old = current_cred();
535
536 retval = -EPERM;
537 if (ns_capable(old->user_ns, CAP_SETUID)) {
538 new->suid = new->uid = kuid;
539 if (!uid_eq(kuid, old->uid)) {
540 retval = set_user(new);
541 if (retval < 0)
542 goto error;
543 }
544 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
545 goto error;
546 }
547
548 new->fsuid = new->euid = kuid;
549
550 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
551 if (retval < 0)
552 goto error;
553
554 return commit_creds(new);
555
556error:
557 abort_creds(new);
558 return retval;
559}
560
561
562/*
563 * This function implements a generic ability to update ruid, euid,
564 * and suid. This allows you to implement the 4.4 compatible seteuid().
565 */
566SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
567{
568 struct user_namespace *ns = current_user_ns();
569 const struct cred *old;
570 struct cred *new;
571 int retval;
572 kuid_t kruid, keuid, ksuid;
573
574 kruid = make_kuid(ns, ruid);
575 keuid = make_kuid(ns, euid);
576 ksuid = make_kuid(ns, suid);
577
578 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
579 return -EINVAL;
580
581 if ((euid != (uid_t) -1) && !uid_valid(keuid))
582 return -EINVAL;
583
584 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
585 return -EINVAL;
586
587 new = prepare_creds();
588 if (!new)
589 return -ENOMEM;
590
591 old = current_cred();
592
593 retval = -EPERM;
594 if (!ns_capable(old->user_ns, CAP_SETUID)) {
595 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
596 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
597 goto error;
598 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
599 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
600 goto error;
601 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
602 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
603 goto error;
604 }
605
606 if (ruid != (uid_t) -1) {
607 new->uid = kruid;
608 if (!uid_eq(kruid, old->uid)) {
609 retval = set_user(new);
610 if (retval < 0)
611 goto error;
612 }
613 }
614 if (euid != (uid_t) -1)
615 new->euid = keuid;
616 if (suid != (uid_t) -1)
617 new->suid = ksuid;
618 new->fsuid = new->euid;
619
620 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
621 if (retval < 0)
622 goto error;
623
624 return commit_creds(new);
625
626error:
627 abort_creds(new);
628 return retval;
629}
630
631SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
632{
633 const struct cred *cred = current_cred();
634 int retval;
635 uid_t ruid, euid, suid;
636
637 ruid = from_kuid_munged(cred->user_ns, cred->uid);
638 euid = from_kuid_munged(cred->user_ns, cred->euid);
639 suid = from_kuid_munged(cred->user_ns, cred->suid);
640
641 if (!(retval = put_user(ruid, ruidp)) &&
642 !(retval = put_user(euid, euidp)))
643 retval = put_user(suid, suidp);
644
645 return retval;
646}
647
648/*
649 * Same as above, but for rgid, egid, sgid.
650 */
651SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
652{
653 struct user_namespace *ns = current_user_ns();
654 const struct cred *old;
655 struct cred *new;
656 int retval;
657 kgid_t krgid, kegid, ksgid;
658
659 krgid = make_kgid(ns, rgid);
660 kegid = make_kgid(ns, egid);
661 ksgid = make_kgid(ns, sgid);
662
663 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
664 return -EINVAL;
665 if ((egid != (gid_t) -1) && !gid_valid(kegid))
666 return -EINVAL;
667 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
668 return -EINVAL;
669
670 new = prepare_creds();
671 if (!new)
672 return -ENOMEM;
673 old = current_cred();
674
675 retval = -EPERM;
676 if (!ns_capable(old->user_ns, CAP_SETGID)) {
677 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
678 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
679 goto error;
680 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
681 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
682 goto error;
683 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
684 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
685 goto error;
686 }
687
688 if (rgid != (gid_t) -1)
689 new->gid = krgid;
690 if (egid != (gid_t) -1)
691 new->egid = kegid;
692 if (sgid != (gid_t) -1)
693 new->sgid = ksgid;
694 new->fsgid = new->egid;
695
696 return commit_creds(new);
697
698error:
699 abort_creds(new);
700 return retval;
701}
702
703SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
704{
705 const struct cred *cred = current_cred();
706 int retval;
707 gid_t rgid, egid, sgid;
708
709 rgid = from_kgid_munged(cred->user_ns, cred->gid);
710 egid = from_kgid_munged(cred->user_ns, cred->egid);
711 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
712
713 if (!(retval = put_user(rgid, rgidp)) &&
714 !(retval = put_user(egid, egidp)))
715 retval = put_user(sgid, sgidp);
716
717 return retval;
718}
719
720
721/*
722 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
723 * is used for "access()" and for the NFS daemon (letting nfsd stay at
724 * whatever uid it wants to). It normally shadows "euid", except when
725 * explicitly set by setfsuid() or for access..
726 */
727SYSCALL_DEFINE1(setfsuid, uid_t, uid)
728{
729 const struct cred *old;
730 struct cred *new;
731 uid_t old_fsuid;
732 kuid_t kuid;
733
734 old = current_cred();
735 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
736
737 kuid = make_kuid(old->user_ns, uid);
738 if (!uid_valid(kuid))
739 return old_fsuid;
740
741 new = prepare_creds();
742 if (!new)
743 return old_fsuid;
744
745 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
746 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
747 ns_capable(old->user_ns, CAP_SETUID)) {
748 if (!uid_eq(kuid, old->fsuid)) {
749 new->fsuid = kuid;
750 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
751 goto change_okay;
752 }
753 }
754
755 abort_creds(new);
756 return old_fsuid;
757
758change_okay:
759 commit_creds(new);
760 return old_fsuid;
761}
762
763/*
764 * Samma på svenska..
765 */
766SYSCALL_DEFINE1(setfsgid, gid_t, gid)
767{
768 const struct cred *old;
769 struct cred *new;
770 gid_t old_fsgid;
771 kgid_t kgid;
772
773 old = current_cred();
774 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
775
776 kgid = make_kgid(old->user_ns, gid);
777 if (!gid_valid(kgid))
778 return old_fsgid;
779
780 new = prepare_creds();
781 if (!new)
782 return old_fsgid;
783
784 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
785 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
786 ns_capable(old->user_ns, CAP_SETGID)) {
787 if (!gid_eq(kgid, old->fsgid)) {
788 new->fsgid = kgid;
789 goto change_okay;
790 }
791 }
792
793 abort_creds(new);
794 return old_fsgid;
795
796change_okay:
797 commit_creds(new);
798 return old_fsgid;
799}
800
801/**
802 * sys_getpid - return the thread group id of the current process
803 *
804 * Note, despite the name, this returns the tgid not the pid. The tgid and
805 * the pid are identical unless CLONE_THREAD was specified on clone() in
806 * which case the tgid is the same in all threads of the same group.
807 *
808 * This is SMP safe as current->tgid does not change.
809 */
810SYSCALL_DEFINE0(getpid)
811{
812 return task_tgid_vnr(current);
813}
814
815/* Thread ID - the internal kernel "pid" */
816SYSCALL_DEFINE0(gettid)
817{
818 return task_pid_vnr(current);
819}
820
821/*
822 * Accessing ->real_parent is not SMP-safe, it could
823 * change from under us. However, we can use a stale
824 * value of ->real_parent under rcu_read_lock(), see
825 * release_task()->call_rcu(delayed_put_task_struct).
826 */
827SYSCALL_DEFINE0(getppid)
828{
829 int pid;
830
831 rcu_read_lock();
832 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
833 rcu_read_unlock();
834
835 return pid;
836}
837
838SYSCALL_DEFINE0(getuid)
839{
840 /* Only we change this so SMP safe */
841 return from_kuid_munged(current_user_ns(), current_uid());
842}
843
844SYSCALL_DEFINE0(geteuid)
845{
846 /* Only we change this so SMP safe */
847 return from_kuid_munged(current_user_ns(), current_euid());
848}
849
850SYSCALL_DEFINE0(getgid)
851{
852 /* Only we change this so SMP safe */
853 return from_kgid_munged(current_user_ns(), current_gid());
854}
855
856SYSCALL_DEFINE0(getegid)
857{
858 /* Only we change this so SMP safe */
859 return from_kgid_munged(current_user_ns(), current_egid());
860}
861
862void do_sys_times(struct tms *tms)
863{
864 cputime_t tgutime, tgstime, cutime, cstime;
865
866 spin_lock_irq(&current->sighand->siglock);
867 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
868 cutime = current->signal->cutime;
869 cstime = current->signal->cstime;
870 spin_unlock_irq(&current->sighand->siglock);
871 tms->tms_utime = cputime_to_clock_t(tgutime);
872 tms->tms_stime = cputime_to_clock_t(tgstime);
873 tms->tms_cutime = cputime_to_clock_t(cutime);
874 tms->tms_cstime = cputime_to_clock_t(cstime);
875}
876
877SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
878{
879 if (tbuf) {
880 struct tms tmp;
881
882 do_sys_times(&tmp);
883 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
884 return -EFAULT;
885 }
886 force_successful_syscall_return();
887 return (long) jiffies_64_to_clock_t(get_jiffies_64());
888}
889
890/*
891 * This needs some heavy checking ...
892 * I just haven't the stomach for it. I also don't fully
893 * understand sessions/pgrp etc. Let somebody who does explain it.
894 *
895 * OK, I think I have the protection semantics right.... this is really
896 * only important on a multi-user system anyway, to make sure one user
897 * can't send a signal to a process owned by another. -TYT, 12/12/91
898 *
899 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
900 * LBT 04.03.94
901 */
902SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
903{
904 struct task_struct *p;
905 struct task_struct *group_leader = current->group_leader;
906 struct pid *pgrp;
907 int err;
908
909 if (!pid)
910 pid = task_pid_vnr(group_leader);
911 if (!pgid)
912 pgid = pid;
913 if (pgid < 0)
914 return -EINVAL;
915 rcu_read_lock();
916
917 /* From this point forward we keep holding onto the tasklist lock
918 * so that our parent does not change from under us. -DaveM
919 */
920 write_lock_irq(&tasklist_lock);
921
922 err = -ESRCH;
923 p = find_task_by_vpid(pid);
924 if (!p)
925 goto out;
926
927 err = -EINVAL;
928 if (!thread_group_leader(p))
929 goto out;
930
931 if (same_thread_group(p->real_parent, group_leader)) {
932 err = -EPERM;
933 if (task_session(p) != task_session(group_leader))
934 goto out;
935 err = -EACCES;
936 if (p->did_exec)
937 goto out;
938 } else {
939 err = -ESRCH;
940 if (p != group_leader)
941 goto out;
942 }
943
944 err = -EPERM;
945 if (p->signal->leader)
946 goto out;
947
948 pgrp = task_pid(p);
949 if (pgid != pid) {
950 struct task_struct *g;
951
952 pgrp = find_vpid(pgid);
953 g = pid_task(pgrp, PIDTYPE_PGID);
954 if (!g || task_session(g) != task_session(group_leader))
955 goto out;
956 }
957
958 err = security_task_setpgid(p, pgid);
959 if (err)
960 goto out;
961
962 if (task_pgrp(p) != pgrp)
963 change_pid(p, PIDTYPE_PGID, pgrp);
964
965 err = 0;
966out:
967 /* All paths lead to here, thus we are safe. -DaveM */
968 write_unlock_irq(&tasklist_lock);
969 rcu_read_unlock();
970 return err;
971}
972
973SYSCALL_DEFINE1(getpgid, pid_t, pid)
974{
975 struct task_struct *p;
976 struct pid *grp;
977 int retval;
978
979 rcu_read_lock();
980 if (!pid)
981 grp = task_pgrp(current);
982 else {
983 retval = -ESRCH;
984 p = find_task_by_vpid(pid);
985 if (!p)
986 goto out;
987 grp = task_pgrp(p);
988 if (!grp)
989 goto out;
990
991 retval = security_task_getpgid(p);
992 if (retval)
993 goto out;
994 }
995 retval = pid_vnr(grp);
996out:
997 rcu_read_unlock();
998 return retval;
999}
1000
1001#ifdef __ARCH_WANT_SYS_GETPGRP
1002
1003SYSCALL_DEFINE0(getpgrp)
1004{
1005 return sys_getpgid(0);
1006}
1007
1008#endif
1009
1010SYSCALL_DEFINE1(getsid, pid_t, pid)
1011{
1012 struct task_struct *p;
1013 struct pid *sid;
1014 int retval;
1015
1016 rcu_read_lock();
1017 if (!pid)
1018 sid = task_session(current);
1019 else {
1020 retval = -ESRCH;
1021 p = find_task_by_vpid(pid);
1022 if (!p)
1023 goto out;
1024 sid = task_session(p);
1025 if (!sid)
1026 goto out;
1027
1028 retval = security_task_getsid(p);
1029 if (retval)
1030 goto out;
1031 }
1032 retval = pid_vnr(sid);
1033out:
1034 rcu_read_unlock();
1035 return retval;
1036}
1037
1038static void set_special_pids(struct pid *pid)
1039{
1040 struct task_struct *curr = current->group_leader;
1041
1042 if (task_session(curr) != pid)
1043 change_pid(curr, PIDTYPE_SID, pid);
1044
1045 if (task_pgrp(curr) != pid)
1046 change_pid(curr, PIDTYPE_PGID, pid);
1047}
1048
1049SYSCALL_DEFINE0(setsid)
1050{
1051 struct task_struct *group_leader = current->group_leader;
1052 struct pid *sid = task_pid(group_leader);
1053 pid_t session = pid_vnr(sid);
1054 int err = -EPERM;
1055
1056 write_lock_irq(&tasklist_lock);
1057 /* Fail if I am already a session leader */
1058 if (group_leader->signal->leader)
1059 goto out;
1060
1061 /* Fail if a process group id already exists that equals the
1062 * proposed session id.
1063 */
1064 if (pid_task(sid, PIDTYPE_PGID))
1065 goto out;
1066
1067 group_leader->signal->leader = 1;
1068 set_special_pids(sid);
1069
1070 proc_clear_tty(group_leader);
1071
1072 err = session;
1073out:
1074 write_unlock_irq(&tasklist_lock);
1075 if (err > 0) {
1076 proc_sid_connector(group_leader);
1077 sched_autogroup_create_attach(group_leader);
1078 }
1079 return err;
1080}
1081
1082DECLARE_RWSEM(uts_sem);
1083
1084#ifdef COMPAT_UTS_MACHINE
1085#define override_architecture(name) \
1086 (personality(current->personality) == PER_LINUX32 && \
1087 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1088 sizeof(COMPAT_UTS_MACHINE)))
1089#else
1090#define override_architecture(name) 0
1091#endif
1092
1093/*
1094 * Work around broken programs that cannot handle "Linux 3.0".
1095 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1096 */
1097static int override_release(char __user *release, size_t len)
1098{
1099 int ret = 0;
1100
1101 if (current->personality & UNAME26) {
1102 const char *rest = UTS_RELEASE;
1103 char buf[65] = { 0 };
1104 int ndots = 0;
1105 unsigned v;
1106 size_t copy;
1107
1108 while (*rest) {
1109 if (*rest == '.' && ++ndots >= 3)
1110 break;
1111 if (!isdigit(*rest) && *rest != '.')
1112 break;
1113 rest++;
1114 }
1115 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1116 copy = clamp_t(size_t, len, 1, sizeof(buf));
1117 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1118 ret = copy_to_user(release, buf, copy + 1);
1119 }
1120 return ret;
1121}
1122
1123SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1124{
1125 int errno = 0;
1126
1127 down_read(&uts_sem);
1128 if (copy_to_user(name, utsname(), sizeof *name))
1129 errno = -EFAULT;
1130 up_read(&uts_sem);
1131
1132 if (!errno && override_release(name->release, sizeof(name->release)))
1133 errno = -EFAULT;
1134 if (!errno && override_architecture(name))
1135 errno = -EFAULT;
1136 return errno;
1137}
1138
1139#ifdef __ARCH_WANT_SYS_OLD_UNAME
1140/*
1141 * Old cruft
1142 */
1143SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1144{
1145 int error = 0;
1146
1147 if (!name)
1148 return -EFAULT;
1149
1150 down_read(&uts_sem);
1151 if (copy_to_user(name, utsname(), sizeof(*name)))
1152 error = -EFAULT;
1153 up_read(&uts_sem);
1154
1155 if (!error && override_release(name->release, sizeof(name->release)))
1156 error = -EFAULT;
1157 if (!error && override_architecture(name))
1158 error = -EFAULT;
1159 return error;
1160}
1161
1162SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1163{
1164 int error;
1165
1166 if (!name)
1167 return -EFAULT;
1168 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1169 return -EFAULT;
1170
1171 down_read(&uts_sem);
1172 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1173 __OLD_UTS_LEN);
1174 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1175 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1176 __OLD_UTS_LEN);
1177 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1178 error |= __copy_to_user(&name->release, &utsname()->release,
1179 __OLD_UTS_LEN);
1180 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1181 error |= __copy_to_user(&name->version, &utsname()->version,
1182 __OLD_UTS_LEN);
1183 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1184 error |= __copy_to_user(&name->machine, &utsname()->machine,
1185 __OLD_UTS_LEN);
1186 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1187 up_read(&uts_sem);
1188
1189 if (!error && override_architecture(name))
1190 error = -EFAULT;
1191 if (!error && override_release(name->release, sizeof(name->release)))
1192 error = -EFAULT;
1193 return error ? -EFAULT : 0;
1194}
1195#endif
1196
1197SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1198{
1199 int errno;
1200 char tmp[__NEW_UTS_LEN];
1201
1202 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1203 return -EPERM;
1204
1205 if (len < 0 || len > __NEW_UTS_LEN)
1206 return -EINVAL;
1207 down_write(&uts_sem);
1208 errno = -EFAULT;
1209 if (!copy_from_user(tmp, name, len)) {
1210 struct new_utsname *u = utsname();
1211
1212 memcpy(u->nodename, tmp, len);
1213 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1214 errno = 0;
1215 uts_proc_notify(UTS_PROC_HOSTNAME);
1216 }
1217 up_write(&uts_sem);
1218 return errno;
1219}
1220
1221#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1222
1223SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1224{
1225 int i, errno;
1226 struct new_utsname *u;
1227
1228 if (len < 0)
1229 return -EINVAL;
1230 down_read(&uts_sem);
1231 u = utsname();
1232 i = 1 + strlen(u->nodename);
1233 if (i > len)
1234 i = len;
1235 errno = 0;
1236 if (copy_to_user(name, u->nodename, i))
1237 errno = -EFAULT;
1238 up_read(&uts_sem);
1239 return errno;
1240}
1241
1242#endif
1243
1244/*
1245 * Only setdomainname; getdomainname can be implemented by calling
1246 * uname()
1247 */
1248SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1249{
1250 int errno;
1251 char tmp[__NEW_UTS_LEN];
1252
1253 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1254 return -EPERM;
1255 if (len < 0 || len > __NEW_UTS_LEN)
1256 return -EINVAL;
1257
1258 down_write(&uts_sem);
1259 errno = -EFAULT;
1260 if (!copy_from_user(tmp, name, len)) {
1261 struct new_utsname *u = utsname();
1262
1263 memcpy(u->domainname, tmp, len);
1264 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1265 errno = 0;
1266 uts_proc_notify(UTS_PROC_DOMAINNAME);
1267 }
1268 up_write(&uts_sem);
1269 return errno;
1270}
1271
1272SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1273{
1274 struct rlimit value;
1275 int ret;
1276
1277 ret = do_prlimit(current, resource, NULL, &value);
1278 if (!ret)
1279 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1280
1281 return ret;
1282}
1283
1284#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1285
1286/*
1287 * Back compatibility for getrlimit. Needed for some apps.
1288 */
1289
1290SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1291 struct rlimit __user *, rlim)
1292{
1293 struct rlimit x;
1294 if (resource >= RLIM_NLIMITS)
1295 return -EINVAL;
1296
1297 task_lock(current->group_leader);
1298 x = current->signal->rlim[resource];
1299 task_unlock(current->group_leader);
1300 if (x.rlim_cur > 0x7FFFFFFF)
1301 x.rlim_cur = 0x7FFFFFFF;
1302 if (x.rlim_max > 0x7FFFFFFF)
1303 x.rlim_max = 0x7FFFFFFF;
1304 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1305}
1306
1307#endif
1308
1309static inline bool rlim64_is_infinity(__u64 rlim64)
1310{
1311#if BITS_PER_LONG < 64
1312 return rlim64 >= ULONG_MAX;
1313#else
1314 return rlim64 == RLIM64_INFINITY;
1315#endif
1316}
1317
1318static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1319{
1320 if (rlim->rlim_cur == RLIM_INFINITY)
1321 rlim64->rlim_cur = RLIM64_INFINITY;
1322 else
1323 rlim64->rlim_cur = rlim->rlim_cur;
1324 if (rlim->rlim_max == RLIM_INFINITY)
1325 rlim64->rlim_max = RLIM64_INFINITY;
1326 else
1327 rlim64->rlim_max = rlim->rlim_max;
1328}
1329
1330static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1331{
1332 if (rlim64_is_infinity(rlim64->rlim_cur))
1333 rlim->rlim_cur = RLIM_INFINITY;
1334 else
1335 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1336 if (rlim64_is_infinity(rlim64->rlim_max))
1337 rlim->rlim_max = RLIM_INFINITY;
1338 else
1339 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1340}
1341
1342/* make sure you are allowed to change @tsk limits before calling this */
1343int do_prlimit(struct task_struct *tsk, unsigned int resource,
1344 struct rlimit *new_rlim, struct rlimit *old_rlim)
1345{
1346 struct rlimit *rlim;
1347 int retval = 0;
1348
1349 if (resource >= RLIM_NLIMITS)
1350 return -EINVAL;
1351 if (new_rlim) {
1352 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1353 return -EINVAL;
1354 if (resource == RLIMIT_NOFILE &&
1355 new_rlim->rlim_max > sysctl_nr_open)
1356 return -EPERM;
1357 }
1358
1359 /* protect tsk->signal and tsk->sighand from disappearing */
1360 read_lock(&tasklist_lock);
1361 if (!tsk->sighand) {
1362 retval = -ESRCH;
1363 goto out;
1364 }
1365
1366 rlim = tsk->signal->rlim + resource;
1367 task_lock(tsk->group_leader);
1368 if (new_rlim) {
1369 /* Keep the capable check against init_user_ns until
1370 cgroups can contain all limits */
1371 if (new_rlim->rlim_max > rlim->rlim_max &&
1372 !capable(CAP_SYS_RESOURCE))
1373 retval = -EPERM;
1374 if (!retval)
1375 retval = security_task_setrlimit(tsk->group_leader,
1376 resource, new_rlim);
1377 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1378 /*
1379 * The caller is asking for an immediate RLIMIT_CPU
1380 * expiry. But we use the zero value to mean "it was
1381 * never set". So let's cheat and make it one second
1382 * instead
1383 */
1384 new_rlim->rlim_cur = 1;
1385 }
1386 }
1387 if (!retval) {
1388 if (old_rlim)
1389 *old_rlim = *rlim;
1390 if (new_rlim)
1391 *rlim = *new_rlim;
1392 }
1393 task_unlock(tsk->group_leader);
1394
1395 /*
1396 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1397 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1398 * very long-standing error, and fixing it now risks breakage of
1399 * applications, so we live with it
1400 */
1401 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1402 new_rlim->rlim_cur != RLIM_INFINITY)
1403 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1404out:
1405 read_unlock(&tasklist_lock);
1406 return retval;
1407}
1408
1409/* rcu lock must be held */
1410static int check_prlimit_permission(struct task_struct *task)
1411{
1412 const struct cred *cred = current_cred(), *tcred;
1413
1414 if (current == task)
1415 return 0;
1416
1417 tcred = __task_cred(task);
1418 if (uid_eq(cred->uid, tcred->euid) &&
1419 uid_eq(cred->uid, tcred->suid) &&
1420 uid_eq(cred->uid, tcred->uid) &&
1421 gid_eq(cred->gid, tcred->egid) &&
1422 gid_eq(cred->gid, tcred->sgid) &&
1423 gid_eq(cred->gid, tcred->gid))
1424 return 0;
1425 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1426 return 0;
1427
1428 return -EPERM;
1429}
1430
1431SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1432 const struct rlimit64 __user *, new_rlim,
1433 struct rlimit64 __user *, old_rlim)
1434{
1435 struct rlimit64 old64, new64;
1436 struct rlimit old, new;
1437 struct task_struct *tsk;
1438 int ret;
1439
1440 if (new_rlim) {
1441 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1442 return -EFAULT;
1443 rlim64_to_rlim(&new64, &new);
1444 }
1445
1446 rcu_read_lock();
1447 tsk = pid ? find_task_by_vpid(pid) : current;
1448 if (!tsk) {
1449 rcu_read_unlock();
1450 return -ESRCH;
1451 }
1452 ret = check_prlimit_permission(tsk);
1453 if (ret) {
1454 rcu_read_unlock();
1455 return ret;
1456 }
1457 get_task_struct(tsk);
1458 rcu_read_unlock();
1459
1460 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1461 old_rlim ? &old : NULL);
1462
1463 if (!ret && old_rlim) {
1464 rlim_to_rlim64(&old, &old64);
1465 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1466 ret = -EFAULT;
1467 }
1468
1469 put_task_struct(tsk);
1470 return ret;
1471}
1472
1473SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1474{
1475 struct rlimit new_rlim;
1476
1477 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1478 return -EFAULT;
1479 return do_prlimit(current, resource, &new_rlim, NULL);
1480}
1481
1482/*
1483 * It would make sense to put struct rusage in the task_struct,
1484 * except that would make the task_struct be *really big*. After
1485 * task_struct gets moved into malloc'ed memory, it would
1486 * make sense to do this. It will make moving the rest of the information
1487 * a lot simpler! (Which we're not doing right now because we're not
1488 * measuring them yet).
1489 *
1490 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1491 * races with threads incrementing their own counters. But since word
1492 * reads are atomic, we either get new values or old values and we don't
1493 * care which for the sums. We always take the siglock to protect reading
1494 * the c* fields from p->signal from races with exit.c updating those
1495 * fields when reaping, so a sample either gets all the additions of a
1496 * given child after it's reaped, or none so this sample is before reaping.
1497 *
1498 * Locking:
1499 * We need to take the siglock for CHILDEREN, SELF and BOTH
1500 * for the cases current multithreaded, non-current single threaded
1501 * non-current multithreaded. Thread traversal is now safe with
1502 * the siglock held.
1503 * Strictly speaking, we donot need to take the siglock if we are current and
1504 * single threaded, as no one else can take our signal_struct away, no one
1505 * else can reap the children to update signal->c* counters, and no one else
1506 * can race with the signal-> fields. If we do not take any lock, the
1507 * signal-> fields could be read out of order while another thread was just
1508 * exiting. So we should place a read memory barrier when we avoid the lock.
1509 * On the writer side, write memory barrier is implied in __exit_signal
1510 * as __exit_signal releases the siglock spinlock after updating the signal->
1511 * fields. But we don't do this yet to keep things simple.
1512 *
1513 */
1514
1515static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1516{
1517 r->ru_nvcsw += t->nvcsw;
1518 r->ru_nivcsw += t->nivcsw;
1519 r->ru_minflt += t->min_flt;
1520 r->ru_majflt += t->maj_flt;
1521 r->ru_inblock += task_io_get_inblock(t);
1522 r->ru_oublock += task_io_get_oublock(t);
1523}
1524
1525static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1526{
1527 struct task_struct *t;
1528 unsigned long flags;
1529 cputime_t tgutime, tgstime, utime, stime;
1530 unsigned long maxrss = 0;
1531
1532 memset((char *) r, 0, sizeof *r);
1533 utime = stime = 0;
1534
1535 if (who == RUSAGE_THREAD) {
1536 task_cputime_adjusted(current, &utime, &stime);
1537 accumulate_thread_rusage(p, r);
1538 maxrss = p->signal->maxrss;
1539 goto out;
1540 }
1541
1542 if (!lock_task_sighand(p, &flags))
1543 return;
1544
1545 switch (who) {
1546 case RUSAGE_BOTH:
1547 case RUSAGE_CHILDREN:
1548 utime = p->signal->cutime;
1549 stime = p->signal->cstime;
1550 r->ru_nvcsw = p->signal->cnvcsw;
1551 r->ru_nivcsw = p->signal->cnivcsw;
1552 r->ru_minflt = p->signal->cmin_flt;
1553 r->ru_majflt = p->signal->cmaj_flt;
1554 r->ru_inblock = p->signal->cinblock;
1555 r->ru_oublock = p->signal->coublock;
1556 maxrss = p->signal->cmaxrss;
1557
1558 if (who == RUSAGE_CHILDREN)
1559 break;
1560
1561 case RUSAGE_SELF:
1562 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1563 utime += tgutime;
1564 stime += tgstime;
1565 r->ru_nvcsw += p->signal->nvcsw;
1566 r->ru_nivcsw += p->signal->nivcsw;
1567 r->ru_minflt += p->signal->min_flt;
1568 r->ru_majflt += p->signal->maj_flt;
1569 r->ru_inblock += p->signal->inblock;
1570 r->ru_oublock += p->signal->oublock;
1571 if (maxrss < p->signal->maxrss)
1572 maxrss = p->signal->maxrss;
1573 t = p;
1574 do {
1575 accumulate_thread_rusage(t, r);
1576 t = next_thread(t);
1577 } while (t != p);
1578 break;
1579
1580 default:
1581 BUG();
1582 }
1583 unlock_task_sighand(p, &flags);
1584
1585out:
1586 cputime_to_timeval(utime, &r->ru_utime);
1587 cputime_to_timeval(stime, &r->ru_stime);
1588
1589 if (who != RUSAGE_CHILDREN) {
1590 struct mm_struct *mm = get_task_mm(p);
1591 if (mm) {
1592 setmax_mm_hiwater_rss(&maxrss, mm);
1593 mmput(mm);
1594 }
1595 }
1596 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1597}
1598
1599int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1600{
1601 struct rusage r;
1602 k_getrusage(p, who, &r);
1603 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1604}
1605
1606SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1607{
1608 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1609 who != RUSAGE_THREAD)
1610 return -EINVAL;
1611 return getrusage(current, who, ru);
1612}
1613
1614#ifdef CONFIG_COMPAT
1615COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1616{
1617 struct rusage r;
1618
1619 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1620 who != RUSAGE_THREAD)
1621 return -EINVAL;
1622
1623 k_getrusage(current, who, &r);
1624 return put_compat_rusage(&r, ru);
1625}
1626#endif
1627
1628SYSCALL_DEFINE1(umask, int, mask)
1629{
1630 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1631 return mask;
1632}
1633
1634static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1635{
1636 struct fd exe;
1637 struct inode *inode;
1638 int err;
1639
1640 exe = fdget(fd);
1641 if (!exe.file)
1642 return -EBADF;
1643
1644 inode = file_inode(exe.file);
1645
1646 /*
1647 * Because the original mm->exe_file points to executable file, make
1648 * sure that this one is executable as well, to avoid breaking an
1649 * overall picture.
1650 */
1651 err = -EACCES;
1652 if (!S_ISREG(inode->i_mode) ||
1653 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1654 goto exit;
1655
1656 err = inode_permission(inode, MAY_EXEC);
1657 if (err)
1658 goto exit;
1659
1660 down_write(&mm->mmap_sem);
1661
1662 /*
1663 * Forbid mm->exe_file change if old file still mapped.
1664 */
1665 err = -EBUSY;
1666 if (mm->exe_file) {
1667 struct vm_area_struct *vma;
1668
1669 for (vma = mm->mmap; vma; vma = vma->vm_next)
1670 if (vma->vm_file &&
1671 path_equal(&vma->vm_file->f_path,
1672 &mm->exe_file->f_path))
1673 goto exit_unlock;
1674 }
1675
1676 /*
1677 * The symlink can be changed only once, just to disallow arbitrary
1678 * transitions malicious software might bring in. This means one
1679 * could make a snapshot over all processes running and monitor
1680 * /proc/pid/exe changes to notice unusual activity if needed.
1681 */
1682 err = -EPERM;
1683 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1684 goto exit_unlock;
1685
1686 err = 0;
1687 set_mm_exe_file(mm, exe.file); /* this grabs a reference to exe.file */
1688exit_unlock:
1689 up_write(&mm->mmap_sem);
1690
1691exit:
1692 fdput(exe);
1693 return err;
1694}
1695
1696static int prctl_set_mm(int opt, unsigned long addr,
1697 unsigned long arg4, unsigned long arg5)
1698{
1699 unsigned long rlim = rlimit(RLIMIT_DATA);
1700 struct mm_struct *mm = current->mm;
1701 struct vm_area_struct *vma;
1702 int error;
1703
1704 if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
1705 return -EINVAL;
1706
1707 if (!capable(CAP_SYS_RESOURCE))
1708 return -EPERM;
1709
1710 if (opt == PR_SET_MM_EXE_FILE)
1711 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1712
1713 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1714 return -EINVAL;
1715
1716 error = -EINVAL;
1717
1718 down_read(&mm->mmap_sem);
1719 vma = find_vma(mm, addr);
1720
1721 switch (opt) {
1722 case PR_SET_MM_START_CODE:
1723 mm->start_code = addr;
1724 break;
1725 case PR_SET_MM_END_CODE:
1726 mm->end_code = addr;
1727 break;
1728 case PR_SET_MM_START_DATA:
1729 mm->start_data = addr;
1730 break;
1731 case PR_SET_MM_END_DATA:
1732 mm->end_data = addr;
1733 break;
1734
1735 case PR_SET_MM_START_BRK:
1736 if (addr <= mm->end_data)
1737 goto out;
1738
1739 if (rlim < RLIM_INFINITY &&
1740 (mm->brk - addr) +
1741 (mm->end_data - mm->start_data) > rlim)
1742 goto out;
1743
1744 mm->start_brk = addr;
1745 break;
1746
1747 case PR_SET_MM_BRK:
1748 if (addr <= mm->end_data)
1749 goto out;
1750
1751 if (rlim < RLIM_INFINITY &&
1752 (addr - mm->start_brk) +
1753 (mm->end_data - mm->start_data) > rlim)
1754 goto out;
1755
1756 mm->brk = addr;
1757 break;
1758
1759 /*
1760 * If command line arguments and environment
1761 * are placed somewhere else on stack, we can
1762 * set them up here, ARG_START/END to setup
1763 * command line argumets and ENV_START/END
1764 * for environment.
1765 */
1766 case PR_SET_MM_START_STACK:
1767 case PR_SET_MM_ARG_START:
1768 case PR_SET_MM_ARG_END:
1769 case PR_SET_MM_ENV_START:
1770 case PR_SET_MM_ENV_END:
1771 if (!vma) {
1772 error = -EFAULT;
1773 goto out;
1774 }
1775 if (opt == PR_SET_MM_START_STACK)
1776 mm->start_stack = addr;
1777 else if (opt == PR_SET_MM_ARG_START)
1778 mm->arg_start = addr;
1779 else if (opt == PR_SET_MM_ARG_END)
1780 mm->arg_end = addr;
1781 else if (opt == PR_SET_MM_ENV_START)
1782 mm->env_start = addr;
1783 else if (opt == PR_SET_MM_ENV_END)
1784 mm->env_end = addr;
1785 break;
1786
1787 /*
1788 * This doesn't move auxiliary vector itself
1789 * since it's pinned to mm_struct, but allow
1790 * to fill vector with new values. It's up
1791 * to a caller to provide sane values here
1792 * otherwise user space tools which use this
1793 * vector might be unhappy.
1794 */
1795 case PR_SET_MM_AUXV: {
1796 unsigned long user_auxv[AT_VECTOR_SIZE];
1797
1798 if (arg4 > sizeof(user_auxv))
1799 goto out;
1800 up_read(&mm->mmap_sem);
1801
1802 if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
1803 return -EFAULT;
1804
1805 /* Make sure the last entry is always AT_NULL */
1806 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1807 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1808
1809 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1810
1811 task_lock(current);
1812 memcpy(mm->saved_auxv, user_auxv, arg4);
1813 task_unlock(current);
1814
1815 return 0;
1816 }
1817 default:
1818 goto out;
1819 }
1820
1821 error = 0;
1822out:
1823 up_read(&mm->mmap_sem);
1824 return error;
1825}
1826
1827#ifdef CONFIG_CHECKPOINT_RESTORE
1828static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1829{
1830 return put_user(me->clear_child_tid, tid_addr);
1831}
1832#else
1833static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1834{
1835 return -EINVAL;
1836}
1837#endif
1838
1839SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1840 unsigned long, arg4, unsigned long, arg5)
1841{
1842 struct task_struct *me = current;
1843 unsigned char comm[sizeof(me->comm)];
1844 long error;
1845
1846 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1847 if (error != -ENOSYS)
1848 return error;
1849
1850 error = 0;
1851 switch (option) {
1852 case PR_SET_PDEATHSIG:
1853 if (!valid_signal(arg2)) {
1854 error = -EINVAL;
1855 break;
1856 }
1857 me->pdeath_signal = arg2;
1858 break;
1859 case PR_GET_PDEATHSIG:
1860 error = put_user(me->pdeath_signal, (int __user *)arg2);
1861 break;
1862 case PR_GET_DUMPABLE:
1863 error = get_dumpable(me->mm);
1864 break;
1865 case PR_SET_DUMPABLE:
1866 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
1867 error = -EINVAL;
1868 break;
1869 }
1870 set_dumpable(me->mm, arg2);
1871 break;
1872
1873 case PR_SET_UNALIGN:
1874 error = SET_UNALIGN_CTL(me, arg2);
1875 break;
1876 case PR_GET_UNALIGN:
1877 error = GET_UNALIGN_CTL(me, arg2);
1878 break;
1879 case PR_SET_FPEMU:
1880 error = SET_FPEMU_CTL(me, arg2);
1881 break;
1882 case PR_GET_FPEMU:
1883 error = GET_FPEMU_CTL(me, arg2);
1884 break;
1885 case PR_SET_FPEXC:
1886 error = SET_FPEXC_CTL(me, arg2);
1887 break;
1888 case PR_GET_FPEXC:
1889 error = GET_FPEXC_CTL(me, arg2);
1890 break;
1891 case PR_GET_TIMING:
1892 error = PR_TIMING_STATISTICAL;
1893 break;
1894 case PR_SET_TIMING:
1895 if (arg2 != PR_TIMING_STATISTICAL)
1896 error = -EINVAL;
1897 break;
1898 case PR_SET_NAME:
1899 comm[sizeof(me->comm) - 1] = 0;
1900 if (strncpy_from_user(comm, (char __user *)arg2,
1901 sizeof(me->comm) - 1) < 0)
1902 return -EFAULT;
1903 set_task_comm(me, comm);
1904 proc_comm_connector(me);
1905 break;
1906 case PR_GET_NAME:
1907 get_task_comm(comm, me);
1908 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
1909 return -EFAULT;
1910 break;
1911 case PR_GET_ENDIAN:
1912 error = GET_ENDIAN(me, arg2);
1913 break;
1914 case PR_SET_ENDIAN:
1915 error = SET_ENDIAN(me, arg2);
1916 break;
1917 case PR_GET_SECCOMP:
1918 error = prctl_get_seccomp();
1919 break;
1920 case PR_SET_SECCOMP:
1921 error = prctl_set_seccomp(arg2, (char __user *)arg3);
1922 break;
1923 case PR_GET_TSC:
1924 error = GET_TSC_CTL(arg2);
1925 break;
1926 case PR_SET_TSC:
1927 error = SET_TSC_CTL(arg2);
1928 break;
1929 case PR_TASK_PERF_EVENTS_DISABLE:
1930 error = perf_event_task_disable();
1931 break;
1932 case PR_TASK_PERF_EVENTS_ENABLE:
1933 error = perf_event_task_enable();
1934 break;
1935 case PR_GET_TIMERSLACK:
1936 error = current->timer_slack_ns;
1937 break;
1938 case PR_SET_TIMERSLACK:
1939 if (arg2 <= 0)
1940 current->timer_slack_ns =
1941 current->default_timer_slack_ns;
1942 else
1943 current->timer_slack_ns = arg2;
1944 break;
1945 case PR_MCE_KILL:
1946 if (arg4 | arg5)
1947 return -EINVAL;
1948 switch (arg2) {
1949 case PR_MCE_KILL_CLEAR:
1950 if (arg3 != 0)
1951 return -EINVAL;
1952 current->flags &= ~PF_MCE_PROCESS;
1953 break;
1954 case PR_MCE_KILL_SET:
1955 current->flags |= PF_MCE_PROCESS;
1956 if (arg3 == PR_MCE_KILL_EARLY)
1957 current->flags |= PF_MCE_EARLY;
1958 else if (arg3 == PR_MCE_KILL_LATE)
1959 current->flags &= ~PF_MCE_EARLY;
1960 else if (arg3 == PR_MCE_KILL_DEFAULT)
1961 current->flags &=
1962 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1963 else
1964 return -EINVAL;
1965 break;
1966 default:
1967 return -EINVAL;
1968 }
1969 break;
1970 case PR_MCE_KILL_GET:
1971 if (arg2 | arg3 | arg4 | arg5)
1972 return -EINVAL;
1973 if (current->flags & PF_MCE_PROCESS)
1974 error = (current->flags & PF_MCE_EARLY) ?
1975 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1976 else
1977 error = PR_MCE_KILL_DEFAULT;
1978 break;
1979 case PR_SET_MM:
1980 error = prctl_set_mm(arg2, arg3, arg4, arg5);
1981 break;
1982 case PR_GET_TID_ADDRESS:
1983 error = prctl_get_tid_address(me, (int __user **)arg2);
1984 break;
1985 case PR_SET_CHILD_SUBREAPER:
1986 me->signal->is_child_subreaper = !!arg2;
1987 break;
1988 case PR_GET_CHILD_SUBREAPER:
1989 error = put_user(me->signal->is_child_subreaper,
1990 (int __user *)arg2);
1991 break;
1992 case PR_SET_NO_NEW_PRIVS:
1993 if (arg2 != 1 || arg3 || arg4 || arg5)
1994 return -EINVAL;
1995
1996 current->no_new_privs = 1;
1997 break;
1998 case PR_GET_NO_NEW_PRIVS:
1999 if (arg2 || arg3 || arg4 || arg5)
2000 return -EINVAL;
2001 return current->no_new_privs ? 1 : 0;
2002 default:
2003 error = -EINVAL;
2004 break;
2005 }
2006 return error;
2007}
2008
2009SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2010 struct getcpu_cache __user *, unused)
2011{
2012 int err = 0;
2013 int cpu = raw_smp_processor_id();
2014 if (cpup)
2015 err |= put_user(cpu, cpup);
2016 if (nodep)
2017 err |= put_user(cpu_to_node(cpu), nodep);
2018 return err ? -EFAULT : 0;
2019}
2020
2021/**
2022 * do_sysinfo - fill in sysinfo struct
2023 * @info: pointer to buffer to fill
2024 */
2025static int do_sysinfo(struct sysinfo *info)
2026{
2027 unsigned long mem_total, sav_total;
2028 unsigned int mem_unit, bitcount;
2029 struct timespec tp;
2030
2031 memset(info, 0, sizeof(struct sysinfo));
2032
2033 get_monotonic_boottime(&tp);
2034 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2035
2036 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2037
2038 info->procs = nr_threads;
2039
2040 si_meminfo(info);
2041 si_swapinfo(info);
2042
2043 /*
2044 * If the sum of all the available memory (i.e. ram + swap)
2045 * is less than can be stored in a 32 bit unsigned long then
2046 * we can be binary compatible with 2.2.x kernels. If not,
2047 * well, in that case 2.2.x was broken anyways...
2048 *
2049 * -Erik Andersen <[email protected]>
2050 */
2051
2052 mem_total = info->totalram + info->totalswap;
2053 if (mem_total < info->totalram || mem_total < info->totalswap)
2054 goto out;
2055 bitcount = 0;
2056 mem_unit = info->mem_unit;
2057 while (mem_unit > 1) {
2058 bitcount++;
2059 mem_unit >>= 1;
2060 sav_total = mem_total;
2061 mem_total <<= 1;
2062 if (mem_total < sav_total)
2063 goto out;
2064 }
2065
2066 /*
2067 * If mem_total did not overflow, multiply all memory values by
2068 * info->mem_unit and set it to 1. This leaves things compatible
2069 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2070 * kernels...
2071 */
2072
2073 info->mem_unit = 1;
2074 info->totalram <<= bitcount;
2075 info->freeram <<= bitcount;
2076 info->sharedram <<= bitcount;
2077 info->bufferram <<= bitcount;
2078 info->totalswap <<= bitcount;
2079 info->freeswap <<= bitcount;
2080 info->totalhigh <<= bitcount;
2081 info->freehigh <<= bitcount;
2082
2083out:
2084 return 0;
2085}
2086
2087SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2088{
2089 struct sysinfo val;
2090
2091 do_sysinfo(&val);
2092
2093 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2094 return -EFAULT;
2095
2096 return 0;
2097}
2098
2099#ifdef CONFIG_COMPAT
2100struct compat_sysinfo {
2101 s32 uptime;
2102 u32 loads[3];
2103 u32 totalram;
2104 u32 freeram;
2105 u32 sharedram;
2106 u32 bufferram;
2107 u32 totalswap;
2108 u32 freeswap;
2109 u16 procs;
2110 u16 pad;
2111 u32 totalhigh;
2112 u32 freehigh;
2113 u32 mem_unit;
2114 char _f[20-2*sizeof(u32)-sizeof(int)];
2115};
2116
2117COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2118{
2119 struct sysinfo s;
2120
2121 do_sysinfo(&s);
2122
2123 /* Check to see if any memory value is too large for 32-bit and scale
2124 * down if needed
2125 */
2126 if ((s.totalram >> 32) || (s.totalswap >> 32)) {
2127 int bitcount = 0;
2128
2129 while (s.mem_unit < PAGE_SIZE) {
2130 s.mem_unit <<= 1;
2131 bitcount++;
2132 }
2133
2134 s.totalram >>= bitcount;
2135 s.freeram >>= bitcount;
2136 s.sharedram >>= bitcount;
2137 s.bufferram >>= bitcount;
2138 s.totalswap >>= bitcount;
2139 s.freeswap >>= bitcount;
2140 s.totalhigh >>= bitcount;
2141 s.freehigh >>= bitcount;
2142 }
2143
2144 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2145 __put_user(s.uptime, &info->uptime) ||
2146 __put_user(s.loads[0], &info->loads[0]) ||
2147 __put_user(s.loads[1], &info->loads[1]) ||
2148 __put_user(s.loads[2], &info->loads[2]) ||
2149 __put_user(s.totalram, &info->totalram) ||
2150 __put_user(s.freeram, &info->freeram) ||
2151 __put_user(s.sharedram, &info->sharedram) ||
2152 __put_user(s.bufferram, &info->bufferram) ||
2153 __put_user(s.totalswap, &info->totalswap) ||
2154 __put_user(s.freeswap, &info->freeswap) ||
2155 __put_user(s.procs, &info->procs) ||
2156 __put_user(s.totalhigh, &info->totalhigh) ||
2157 __put_user(s.freehigh, &info->freehigh) ||
2158 __put_user(s.mem_unit, &info->mem_unit))
2159 return -EFAULT;
2160
2161 return 0;
2162}
2163#endif /* CONFIG_COMPAT */