Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/kernel/sys.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/export.h>
9#include <linux/mm.h>
10#include <linux/mm_inline.h>
11#include <linux/utsname.h>
12#include <linux/mman.h>
13#include <linux/reboot.h>
14#include <linux/prctl.h>
15#include <linux/highuid.h>
16#include <linux/fs.h>
17#include <linux/kmod.h>
18#include <linux/ksm.h>
19#include <linux/perf_event.h>
20#include <linux/resource.h>
21#include <linux/kernel.h>
22#include <linux/workqueue.h>
23#include <linux/capability.h>
24#include <linux/device.h>
25#include <linux/key.h>
26#include <linux/times.h>
27#include <linux/posix-timers.h>
28#include <linux/security.h>
29#include <linux/random.h>
30#include <linux/suspend.h>
31#include <linux/tty.h>
32#include <linux/signal.h>
33#include <linux/cn_proc.h>
34#include <linux/getcpu.h>
35#include <linux/task_io_accounting_ops.h>
36#include <linux/seccomp.h>
37#include <linux/cpu.h>
38#include <linux/personality.h>
39#include <linux/ptrace.h>
40#include <linux/fs_struct.h>
41#include <linux/file.h>
42#include <linux/mount.h>
43#include <linux/gfp.h>
44#include <linux/syscore_ops.h>
45#include <linux/version.h>
46#include <linux/ctype.h>
47#include <linux/syscall_user_dispatch.h>
48
49#include <linux/compat.h>
50#include <linux/syscalls.h>
51#include <linux/kprobes.h>
52#include <linux/user_namespace.h>
53#include <linux/time_namespace.h>
54#include <linux/binfmts.h>
55
56#include <linux/sched.h>
57#include <linux/sched/autogroup.h>
58#include <linux/sched/loadavg.h>
59#include <linux/sched/stat.h>
60#include <linux/sched/mm.h>
61#include <linux/sched/coredump.h>
62#include <linux/sched/task.h>
63#include <linux/sched/cputime.h>
64#include <linux/rcupdate.h>
65#include <linux/uidgid.h>
66#include <linux/cred.h>
67
68#include <linux/nospec.h>
69
70#include <linux/kmsg_dump.h>
71/* Move somewhere else to avoid recompiling? */
72#include <generated/utsrelease.h>
73
74#include <linux/uaccess.h>
75#include <asm/io.h>
76#include <asm/unistd.h>
77
78#include "uid16.h"
79
80#ifndef SET_UNALIGN_CTL
81# define SET_UNALIGN_CTL(a, b) (-EINVAL)
82#endif
83#ifndef GET_UNALIGN_CTL
84# define GET_UNALIGN_CTL(a, b) (-EINVAL)
85#endif
86#ifndef SET_FPEMU_CTL
87# define SET_FPEMU_CTL(a, b) (-EINVAL)
88#endif
89#ifndef GET_FPEMU_CTL
90# define GET_FPEMU_CTL(a, b) (-EINVAL)
91#endif
92#ifndef SET_FPEXC_CTL
93# define SET_FPEXC_CTL(a, b) (-EINVAL)
94#endif
95#ifndef GET_FPEXC_CTL
96# define GET_FPEXC_CTL(a, b) (-EINVAL)
97#endif
98#ifndef GET_ENDIAN
99# define GET_ENDIAN(a, b) (-EINVAL)
100#endif
101#ifndef SET_ENDIAN
102# define SET_ENDIAN(a, b) (-EINVAL)
103#endif
104#ifndef GET_TSC_CTL
105# define GET_TSC_CTL(a) (-EINVAL)
106#endif
107#ifndef SET_TSC_CTL
108# define SET_TSC_CTL(a) (-EINVAL)
109#endif
110#ifndef GET_FP_MODE
111# define GET_FP_MODE(a) (-EINVAL)
112#endif
113#ifndef SET_FP_MODE
114# define SET_FP_MODE(a,b) (-EINVAL)
115#endif
116#ifndef SVE_SET_VL
117# define SVE_SET_VL(a) (-EINVAL)
118#endif
119#ifndef SVE_GET_VL
120# define SVE_GET_VL() (-EINVAL)
121#endif
122#ifndef SME_SET_VL
123# define SME_SET_VL(a) (-EINVAL)
124#endif
125#ifndef SME_GET_VL
126# define SME_GET_VL() (-EINVAL)
127#endif
128#ifndef PAC_RESET_KEYS
129# define PAC_RESET_KEYS(a, b) (-EINVAL)
130#endif
131#ifndef PAC_SET_ENABLED_KEYS
132# define PAC_SET_ENABLED_KEYS(a, b, c) (-EINVAL)
133#endif
134#ifndef PAC_GET_ENABLED_KEYS
135# define PAC_GET_ENABLED_KEYS(a) (-EINVAL)
136#endif
137#ifndef SET_TAGGED_ADDR_CTRL
138# define SET_TAGGED_ADDR_CTRL(a) (-EINVAL)
139#endif
140#ifndef GET_TAGGED_ADDR_CTRL
141# define GET_TAGGED_ADDR_CTRL() (-EINVAL)
142#endif
143#ifndef RISCV_V_SET_CONTROL
144# define RISCV_V_SET_CONTROL(a) (-EINVAL)
145#endif
146#ifndef RISCV_V_GET_CONTROL
147# define RISCV_V_GET_CONTROL() (-EINVAL)
148#endif
149
150/*
151 * this is where the system-wide overflow UID and GID are defined, for
152 * architectures that now have 32-bit UID/GID but didn't in the past
153 */
154
155int overflowuid = DEFAULT_OVERFLOWUID;
156int overflowgid = DEFAULT_OVERFLOWGID;
157
158EXPORT_SYMBOL(overflowuid);
159EXPORT_SYMBOL(overflowgid);
160
161/*
162 * the same as above, but for filesystems which can only store a 16-bit
163 * UID and GID. as such, this is needed on all architectures
164 */
165
166int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
167int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
168
169EXPORT_SYMBOL(fs_overflowuid);
170EXPORT_SYMBOL(fs_overflowgid);
171
172/*
173 * Returns true if current's euid is same as p's uid or euid,
174 * or has CAP_SYS_NICE to p's user_ns.
175 *
176 * Called with rcu_read_lock, creds are safe
177 */
178static bool set_one_prio_perm(struct task_struct *p)
179{
180 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
181
182 if (uid_eq(pcred->uid, cred->euid) ||
183 uid_eq(pcred->euid, cred->euid))
184 return true;
185 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
186 return true;
187 return false;
188}
189
190/*
191 * set the priority of a task
192 * - the caller must hold the RCU read lock
193 */
194static int set_one_prio(struct task_struct *p, int niceval, int error)
195{
196 int no_nice;
197
198 if (!set_one_prio_perm(p)) {
199 error = -EPERM;
200 goto out;
201 }
202 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
203 error = -EACCES;
204 goto out;
205 }
206 no_nice = security_task_setnice(p, niceval);
207 if (no_nice) {
208 error = no_nice;
209 goto out;
210 }
211 if (error == -ESRCH)
212 error = 0;
213 set_user_nice(p, niceval);
214out:
215 return error;
216}
217
218SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
219{
220 struct task_struct *g, *p;
221 struct user_struct *user;
222 const struct cred *cred = current_cred();
223 int error = -EINVAL;
224 struct pid *pgrp;
225 kuid_t uid;
226
227 if (which > PRIO_USER || which < PRIO_PROCESS)
228 goto out;
229
230 /* normalize: avoid signed division (rounding problems) */
231 error = -ESRCH;
232 if (niceval < MIN_NICE)
233 niceval = MIN_NICE;
234 if (niceval > MAX_NICE)
235 niceval = MAX_NICE;
236
237 rcu_read_lock();
238 switch (which) {
239 case PRIO_PROCESS:
240 if (who)
241 p = find_task_by_vpid(who);
242 else
243 p = current;
244 if (p)
245 error = set_one_prio(p, niceval, error);
246 break;
247 case PRIO_PGRP:
248 if (who)
249 pgrp = find_vpid(who);
250 else
251 pgrp = task_pgrp(current);
252 read_lock(&tasklist_lock);
253 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
254 error = set_one_prio(p, niceval, error);
255 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
256 read_unlock(&tasklist_lock);
257 break;
258 case PRIO_USER:
259 uid = make_kuid(cred->user_ns, who);
260 user = cred->user;
261 if (!who)
262 uid = cred->uid;
263 else if (!uid_eq(uid, cred->uid)) {
264 user = find_user(uid);
265 if (!user)
266 goto out_unlock; /* No processes for this user */
267 }
268 for_each_process_thread(g, p) {
269 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
270 error = set_one_prio(p, niceval, error);
271 }
272 if (!uid_eq(uid, cred->uid))
273 free_uid(user); /* For find_user() */
274 break;
275 }
276out_unlock:
277 rcu_read_unlock();
278out:
279 return error;
280}
281
282/*
283 * Ugh. To avoid negative return values, "getpriority()" will
284 * not return the normal nice-value, but a negated value that
285 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
286 * to stay compatible.
287 */
288SYSCALL_DEFINE2(getpriority, int, which, int, who)
289{
290 struct task_struct *g, *p;
291 struct user_struct *user;
292 const struct cred *cred = current_cred();
293 long niceval, retval = -ESRCH;
294 struct pid *pgrp;
295 kuid_t uid;
296
297 if (which > PRIO_USER || which < PRIO_PROCESS)
298 return -EINVAL;
299
300 rcu_read_lock();
301 switch (which) {
302 case PRIO_PROCESS:
303 if (who)
304 p = find_task_by_vpid(who);
305 else
306 p = current;
307 if (p) {
308 niceval = nice_to_rlimit(task_nice(p));
309 if (niceval > retval)
310 retval = niceval;
311 }
312 break;
313 case PRIO_PGRP:
314 if (who)
315 pgrp = find_vpid(who);
316 else
317 pgrp = task_pgrp(current);
318 read_lock(&tasklist_lock);
319 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
320 niceval = nice_to_rlimit(task_nice(p));
321 if (niceval > retval)
322 retval = niceval;
323 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
324 read_unlock(&tasklist_lock);
325 break;
326 case PRIO_USER:
327 uid = make_kuid(cred->user_ns, who);
328 user = cred->user;
329 if (!who)
330 uid = cred->uid;
331 else if (!uid_eq(uid, cred->uid)) {
332 user = find_user(uid);
333 if (!user)
334 goto out_unlock; /* No processes for this user */
335 }
336 for_each_process_thread(g, p) {
337 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
338 niceval = nice_to_rlimit(task_nice(p));
339 if (niceval > retval)
340 retval = niceval;
341 }
342 }
343 if (!uid_eq(uid, cred->uid))
344 free_uid(user); /* for find_user() */
345 break;
346 }
347out_unlock:
348 rcu_read_unlock();
349
350 return retval;
351}
352
353/*
354 * Unprivileged users may change the real gid to the effective gid
355 * or vice versa. (BSD-style)
356 *
357 * If you set the real gid at all, or set the effective gid to a value not
358 * equal to the real gid, then the saved gid is set to the new effective gid.
359 *
360 * This makes it possible for a setgid program to completely drop its
361 * privileges, which is often a useful assertion to make when you are doing
362 * a security audit over a program.
363 *
364 * The general idea is that a program which uses just setregid() will be
365 * 100% compatible with BSD. A program which uses just setgid() will be
366 * 100% compatible with POSIX with saved IDs.
367 *
368 * SMP: There are not races, the GIDs are checked only by filesystem
369 * operations (as far as semantic preservation is concerned).
370 */
371#ifdef CONFIG_MULTIUSER
372long __sys_setregid(gid_t rgid, gid_t egid)
373{
374 struct user_namespace *ns = current_user_ns();
375 const struct cred *old;
376 struct cred *new;
377 int retval;
378 kgid_t krgid, kegid;
379
380 krgid = make_kgid(ns, rgid);
381 kegid = make_kgid(ns, egid);
382
383 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
384 return -EINVAL;
385 if ((egid != (gid_t) -1) && !gid_valid(kegid))
386 return -EINVAL;
387
388 new = prepare_creds();
389 if (!new)
390 return -ENOMEM;
391 old = current_cred();
392
393 retval = -EPERM;
394 if (rgid != (gid_t) -1) {
395 if (gid_eq(old->gid, krgid) ||
396 gid_eq(old->egid, krgid) ||
397 ns_capable_setid(old->user_ns, CAP_SETGID))
398 new->gid = krgid;
399 else
400 goto error;
401 }
402 if (egid != (gid_t) -1) {
403 if (gid_eq(old->gid, kegid) ||
404 gid_eq(old->egid, kegid) ||
405 gid_eq(old->sgid, kegid) ||
406 ns_capable_setid(old->user_ns, CAP_SETGID))
407 new->egid = kegid;
408 else
409 goto error;
410 }
411
412 if (rgid != (gid_t) -1 ||
413 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
414 new->sgid = new->egid;
415 new->fsgid = new->egid;
416
417 retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
418 if (retval < 0)
419 goto error;
420
421 return commit_creds(new);
422
423error:
424 abort_creds(new);
425 return retval;
426}
427
428SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
429{
430 return __sys_setregid(rgid, egid);
431}
432
433/*
434 * setgid() is implemented like SysV w/ SAVED_IDS
435 *
436 * SMP: Same implicit races as above.
437 */
438long __sys_setgid(gid_t gid)
439{
440 struct user_namespace *ns = current_user_ns();
441 const struct cred *old;
442 struct cred *new;
443 int retval;
444 kgid_t kgid;
445
446 kgid = make_kgid(ns, gid);
447 if (!gid_valid(kgid))
448 return -EINVAL;
449
450 new = prepare_creds();
451 if (!new)
452 return -ENOMEM;
453 old = current_cred();
454
455 retval = -EPERM;
456 if (ns_capable_setid(old->user_ns, CAP_SETGID))
457 new->gid = new->egid = new->sgid = new->fsgid = kgid;
458 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
459 new->egid = new->fsgid = kgid;
460 else
461 goto error;
462
463 retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
464 if (retval < 0)
465 goto error;
466
467 return commit_creds(new);
468
469error:
470 abort_creds(new);
471 return retval;
472}
473
474SYSCALL_DEFINE1(setgid, gid_t, gid)
475{
476 return __sys_setgid(gid);
477}
478
479/*
480 * change the user struct in a credentials set to match the new UID
481 */
482static int set_user(struct cred *new)
483{
484 struct user_struct *new_user;
485
486 new_user = alloc_uid(new->uid);
487 if (!new_user)
488 return -EAGAIN;
489
490 free_uid(new->user);
491 new->user = new_user;
492 return 0;
493}
494
495static void flag_nproc_exceeded(struct cred *new)
496{
497 if (new->ucounts == current_ucounts())
498 return;
499
500 /*
501 * We don't fail in case of NPROC limit excess here because too many
502 * poorly written programs don't check set*uid() return code, assuming
503 * it never fails if called by root. We may still enforce NPROC limit
504 * for programs doing set*uid()+execve() by harmlessly deferring the
505 * failure to the execve() stage.
506 */
507 if (is_rlimit_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
508 new->user != INIT_USER)
509 current->flags |= PF_NPROC_EXCEEDED;
510 else
511 current->flags &= ~PF_NPROC_EXCEEDED;
512}
513
514/*
515 * Unprivileged users may change the real uid to the effective uid
516 * or vice versa. (BSD-style)
517 *
518 * If you set the real uid at all, or set the effective uid to a value not
519 * equal to the real uid, then the saved uid is set to the new effective uid.
520 *
521 * This makes it possible for a setuid program to completely drop its
522 * privileges, which is often a useful assertion to make when you are doing
523 * a security audit over a program.
524 *
525 * The general idea is that a program which uses just setreuid() will be
526 * 100% compatible with BSD. A program which uses just setuid() will be
527 * 100% compatible with POSIX with saved IDs.
528 */
529long __sys_setreuid(uid_t ruid, uid_t euid)
530{
531 struct user_namespace *ns = current_user_ns();
532 const struct cred *old;
533 struct cred *new;
534 int retval;
535 kuid_t kruid, keuid;
536
537 kruid = make_kuid(ns, ruid);
538 keuid = make_kuid(ns, euid);
539
540 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
541 return -EINVAL;
542 if ((euid != (uid_t) -1) && !uid_valid(keuid))
543 return -EINVAL;
544
545 new = prepare_creds();
546 if (!new)
547 return -ENOMEM;
548 old = current_cred();
549
550 retval = -EPERM;
551 if (ruid != (uid_t) -1) {
552 new->uid = kruid;
553 if (!uid_eq(old->uid, kruid) &&
554 !uid_eq(old->euid, kruid) &&
555 !ns_capable_setid(old->user_ns, CAP_SETUID))
556 goto error;
557 }
558
559 if (euid != (uid_t) -1) {
560 new->euid = keuid;
561 if (!uid_eq(old->uid, keuid) &&
562 !uid_eq(old->euid, keuid) &&
563 !uid_eq(old->suid, keuid) &&
564 !ns_capable_setid(old->user_ns, CAP_SETUID))
565 goto error;
566 }
567
568 if (!uid_eq(new->uid, old->uid)) {
569 retval = set_user(new);
570 if (retval < 0)
571 goto error;
572 }
573 if (ruid != (uid_t) -1 ||
574 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
575 new->suid = new->euid;
576 new->fsuid = new->euid;
577
578 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
579 if (retval < 0)
580 goto error;
581
582 retval = set_cred_ucounts(new);
583 if (retval < 0)
584 goto error;
585
586 flag_nproc_exceeded(new);
587 return commit_creds(new);
588
589error:
590 abort_creds(new);
591 return retval;
592}
593
594SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
595{
596 return __sys_setreuid(ruid, euid);
597}
598
599/*
600 * setuid() is implemented like SysV with SAVED_IDS
601 *
602 * Note that SAVED_ID's is deficient in that a setuid root program
603 * like sendmail, for example, cannot set its uid to be a normal
604 * user and then switch back, because if you're root, setuid() sets
605 * the saved uid too. If you don't like this, blame the bright people
606 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
607 * will allow a root program to temporarily drop privileges and be able to
608 * regain them by swapping the real and effective uid.
609 */
610long __sys_setuid(uid_t uid)
611{
612 struct user_namespace *ns = current_user_ns();
613 const struct cred *old;
614 struct cred *new;
615 int retval;
616 kuid_t kuid;
617
618 kuid = make_kuid(ns, uid);
619 if (!uid_valid(kuid))
620 return -EINVAL;
621
622 new = prepare_creds();
623 if (!new)
624 return -ENOMEM;
625 old = current_cred();
626
627 retval = -EPERM;
628 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
629 new->suid = new->uid = kuid;
630 if (!uid_eq(kuid, old->uid)) {
631 retval = set_user(new);
632 if (retval < 0)
633 goto error;
634 }
635 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
636 goto error;
637 }
638
639 new->fsuid = new->euid = kuid;
640
641 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
642 if (retval < 0)
643 goto error;
644
645 retval = set_cred_ucounts(new);
646 if (retval < 0)
647 goto error;
648
649 flag_nproc_exceeded(new);
650 return commit_creds(new);
651
652error:
653 abort_creds(new);
654 return retval;
655}
656
657SYSCALL_DEFINE1(setuid, uid_t, uid)
658{
659 return __sys_setuid(uid);
660}
661
662
663/*
664 * This function implements a generic ability to update ruid, euid,
665 * and suid. This allows you to implement the 4.4 compatible seteuid().
666 */
667long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
668{
669 struct user_namespace *ns = current_user_ns();
670 const struct cred *old;
671 struct cred *new;
672 int retval;
673 kuid_t kruid, keuid, ksuid;
674 bool ruid_new, euid_new, suid_new;
675
676 kruid = make_kuid(ns, ruid);
677 keuid = make_kuid(ns, euid);
678 ksuid = make_kuid(ns, suid);
679
680 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
681 return -EINVAL;
682
683 if ((euid != (uid_t) -1) && !uid_valid(keuid))
684 return -EINVAL;
685
686 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
687 return -EINVAL;
688
689 old = current_cred();
690
691 /* check for no-op */
692 if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
693 (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
694 uid_eq(keuid, old->fsuid))) &&
695 (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
696 return 0;
697
698 ruid_new = ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
699 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
700 euid_new = euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
701 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
702 suid_new = suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
703 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
704 if ((ruid_new || euid_new || suid_new) &&
705 !ns_capable_setid(old->user_ns, CAP_SETUID))
706 return -EPERM;
707
708 new = prepare_creds();
709 if (!new)
710 return -ENOMEM;
711
712 if (ruid != (uid_t) -1) {
713 new->uid = kruid;
714 if (!uid_eq(kruid, old->uid)) {
715 retval = set_user(new);
716 if (retval < 0)
717 goto error;
718 }
719 }
720 if (euid != (uid_t) -1)
721 new->euid = keuid;
722 if (suid != (uid_t) -1)
723 new->suid = ksuid;
724 new->fsuid = new->euid;
725
726 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
727 if (retval < 0)
728 goto error;
729
730 retval = set_cred_ucounts(new);
731 if (retval < 0)
732 goto error;
733
734 flag_nproc_exceeded(new);
735 return commit_creds(new);
736
737error:
738 abort_creds(new);
739 return retval;
740}
741
742SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
743{
744 return __sys_setresuid(ruid, euid, suid);
745}
746
747SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
748{
749 const struct cred *cred = current_cred();
750 int retval;
751 uid_t ruid, euid, suid;
752
753 ruid = from_kuid_munged(cred->user_ns, cred->uid);
754 euid = from_kuid_munged(cred->user_ns, cred->euid);
755 suid = from_kuid_munged(cred->user_ns, cred->suid);
756
757 retval = put_user(ruid, ruidp);
758 if (!retval) {
759 retval = put_user(euid, euidp);
760 if (!retval)
761 return put_user(suid, suidp);
762 }
763 return retval;
764}
765
766/*
767 * Same as above, but for rgid, egid, sgid.
768 */
769long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
770{
771 struct user_namespace *ns = current_user_ns();
772 const struct cred *old;
773 struct cred *new;
774 int retval;
775 kgid_t krgid, kegid, ksgid;
776 bool rgid_new, egid_new, sgid_new;
777
778 krgid = make_kgid(ns, rgid);
779 kegid = make_kgid(ns, egid);
780 ksgid = make_kgid(ns, sgid);
781
782 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
783 return -EINVAL;
784 if ((egid != (gid_t) -1) && !gid_valid(kegid))
785 return -EINVAL;
786 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
787 return -EINVAL;
788
789 old = current_cred();
790
791 /* check for no-op */
792 if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
793 (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
794 gid_eq(kegid, old->fsgid))) &&
795 (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
796 return 0;
797
798 rgid_new = rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
799 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
800 egid_new = egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
801 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
802 sgid_new = sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
803 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
804 if ((rgid_new || egid_new || sgid_new) &&
805 !ns_capable_setid(old->user_ns, CAP_SETGID))
806 return -EPERM;
807
808 new = prepare_creds();
809 if (!new)
810 return -ENOMEM;
811
812 if (rgid != (gid_t) -1)
813 new->gid = krgid;
814 if (egid != (gid_t) -1)
815 new->egid = kegid;
816 if (sgid != (gid_t) -1)
817 new->sgid = ksgid;
818 new->fsgid = new->egid;
819
820 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
821 if (retval < 0)
822 goto error;
823
824 return commit_creds(new);
825
826error:
827 abort_creds(new);
828 return retval;
829}
830
831SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
832{
833 return __sys_setresgid(rgid, egid, sgid);
834}
835
836SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
837{
838 const struct cred *cred = current_cred();
839 int retval;
840 gid_t rgid, egid, sgid;
841
842 rgid = from_kgid_munged(cred->user_ns, cred->gid);
843 egid = from_kgid_munged(cred->user_ns, cred->egid);
844 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
845
846 retval = put_user(rgid, rgidp);
847 if (!retval) {
848 retval = put_user(egid, egidp);
849 if (!retval)
850 retval = put_user(sgid, sgidp);
851 }
852
853 return retval;
854}
855
856
857/*
858 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
859 * is used for "access()" and for the NFS daemon (letting nfsd stay at
860 * whatever uid it wants to). It normally shadows "euid", except when
861 * explicitly set by setfsuid() or for access..
862 */
863long __sys_setfsuid(uid_t uid)
864{
865 const struct cred *old;
866 struct cred *new;
867 uid_t old_fsuid;
868 kuid_t kuid;
869
870 old = current_cred();
871 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
872
873 kuid = make_kuid(old->user_ns, uid);
874 if (!uid_valid(kuid))
875 return old_fsuid;
876
877 new = prepare_creds();
878 if (!new)
879 return old_fsuid;
880
881 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
882 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
883 ns_capable_setid(old->user_ns, CAP_SETUID)) {
884 if (!uid_eq(kuid, old->fsuid)) {
885 new->fsuid = kuid;
886 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
887 goto change_okay;
888 }
889 }
890
891 abort_creds(new);
892 return old_fsuid;
893
894change_okay:
895 commit_creds(new);
896 return old_fsuid;
897}
898
899SYSCALL_DEFINE1(setfsuid, uid_t, uid)
900{
901 return __sys_setfsuid(uid);
902}
903
904/*
905 * Samma på svenska..
906 */
907long __sys_setfsgid(gid_t gid)
908{
909 const struct cred *old;
910 struct cred *new;
911 gid_t old_fsgid;
912 kgid_t kgid;
913
914 old = current_cred();
915 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
916
917 kgid = make_kgid(old->user_ns, gid);
918 if (!gid_valid(kgid))
919 return old_fsgid;
920
921 new = prepare_creds();
922 if (!new)
923 return old_fsgid;
924
925 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
926 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
927 ns_capable_setid(old->user_ns, CAP_SETGID)) {
928 if (!gid_eq(kgid, old->fsgid)) {
929 new->fsgid = kgid;
930 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
931 goto change_okay;
932 }
933 }
934
935 abort_creds(new);
936 return old_fsgid;
937
938change_okay:
939 commit_creds(new);
940 return old_fsgid;
941}
942
943SYSCALL_DEFINE1(setfsgid, gid_t, gid)
944{
945 return __sys_setfsgid(gid);
946}
947#endif /* CONFIG_MULTIUSER */
948
949/**
950 * sys_getpid - return the thread group id of the current process
951 *
952 * Note, despite the name, this returns the tgid not the pid. The tgid and
953 * the pid are identical unless CLONE_THREAD was specified on clone() in
954 * which case the tgid is the same in all threads of the same group.
955 *
956 * This is SMP safe as current->tgid does not change.
957 */
958SYSCALL_DEFINE0(getpid)
959{
960 return task_tgid_vnr(current);
961}
962
963/* Thread ID - the internal kernel "pid" */
964SYSCALL_DEFINE0(gettid)
965{
966 return task_pid_vnr(current);
967}
968
969/*
970 * Accessing ->real_parent is not SMP-safe, it could
971 * change from under us. However, we can use a stale
972 * value of ->real_parent under rcu_read_lock(), see
973 * release_task()->call_rcu(delayed_put_task_struct).
974 */
975SYSCALL_DEFINE0(getppid)
976{
977 int pid;
978
979 rcu_read_lock();
980 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
981 rcu_read_unlock();
982
983 return pid;
984}
985
986SYSCALL_DEFINE0(getuid)
987{
988 /* Only we change this so SMP safe */
989 return from_kuid_munged(current_user_ns(), current_uid());
990}
991
992SYSCALL_DEFINE0(geteuid)
993{
994 /* Only we change this so SMP safe */
995 return from_kuid_munged(current_user_ns(), current_euid());
996}
997
998SYSCALL_DEFINE0(getgid)
999{
1000 /* Only we change this so SMP safe */
1001 return from_kgid_munged(current_user_ns(), current_gid());
1002}
1003
1004SYSCALL_DEFINE0(getegid)
1005{
1006 /* Only we change this so SMP safe */
1007 return from_kgid_munged(current_user_ns(), current_egid());
1008}
1009
1010static void do_sys_times(struct tms *tms)
1011{
1012 u64 tgutime, tgstime, cutime, cstime;
1013
1014 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1015 cutime = current->signal->cutime;
1016 cstime = current->signal->cstime;
1017 tms->tms_utime = nsec_to_clock_t(tgutime);
1018 tms->tms_stime = nsec_to_clock_t(tgstime);
1019 tms->tms_cutime = nsec_to_clock_t(cutime);
1020 tms->tms_cstime = nsec_to_clock_t(cstime);
1021}
1022
1023SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1024{
1025 if (tbuf) {
1026 struct tms tmp;
1027
1028 do_sys_times(&tmp);
1029 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1030 return -EFAULT;
1031 }
1032 force_successful_syscall_return();
1033 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1034}
1035
1036#ifdef CONFIG_COMPAT
1037static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1038{
1039 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1040}
1041
1042COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1043{
1044 if (tbuf) {
1045 struct tms tms;
1046 struct compat_tms tmp;
1047
1048 do_sys_times(&tms);
1049 /* Convert our struct tms to the compat version. */
1050 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1051 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1052 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1053 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1054 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1055 return -EFAULT;
1056 }
1057 force_successful_syscall_return();
1058 return compat_jiffies_to_clock_t(jiffies);
1059}
1060#endif
1061
1062/*
1063 * This needs some heavy checking ...
1064 * I just haven't the stomach for it. I also don't fully
1065 * understand sessions/pgrp etc. Let somebody who does explain it.
1066 *
1067 * OK, I think I have the protection semantics right.... this is really
1068 * only important on a multi-user system anyway, to make sure one user
1069 * can't send a signal to a process owned by another. -TYT, 12/12/91
1070 *
1071 * !PF_FORKNOEXEC check to conform completely to POSIX.
1072 */
1073SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1074{
1075 struct task_struct *p;
1076 struct task_struct *group_leader = current->group_leader;
1077 struct pid *pgrp;
1078 int err;
1079
1080 if (!pid)
1081 pid = task_pid_vnr(group_leader);
1082 if (!pgid)
1083 pgid = pid;
1084 if (pgid < 0)
1085 return -EINVAL;
1086 rcu_read_lock();
1087
1088 /* From this point forward we keep holding onto the tasklist lock
1089 * so that our parent does not change from under us. -DaveM
1090 */
1091 write_lock_irq(&tasklist_lock);
1092
1093 err = -ESRCH;
1094 p = find_task_by_vpid(pid);
1095 if (!p)
1096 goto out;
1097
1098 err = -EINVAL;
1099 if (!thread_group_leader(p))
1100 goto out;
1101
1102 if (same_thread_group(p->real_parent, group_leader)) {
1103 err = -EPERM;
1104 if (task_session(p) != task_session(group_leader))
1105 goto out;
1106 err = -EACCES;
1107 if (!(p->flags & PF_FORKNOEXEC))
1108 goto out;
1109 } else {
1110 err = -ESRCH;
1111 if (p != group_leader)
1112 goto out;
1113 }
1114
1115 err = -EPERM;
1116 if (p->signal->leader)
1117 goto out;
1118
1119 pgrp = task_pid(p);
1120 if (pgid != pid) {
1121 struct task_struct *g;
1122
1123 pgrp = find_vpid(pgid);
1124 g = pid_task(pgrp, PIDTYPE_PGID);
1125 if (!g || task_session(g) != task_session(group_leader))
1126 goto out;
1127 }
1128
1129 err = security_task_setpgid(p, pgid);
1130 if (err)
1131 goto out;
1132
1133 if (task_pgrp(p) != pgrp)
1134 change_pid(p, PIDTYPE_PGID, pgrp);
1135
1136 err = 0;
1137out:
1138 /* All paths lead to here, thus we are safe. -DaveM */
1139 write_unlock_irq(&tasklist_lock);
1140 rcu_read_unlock();
1141 return err;
1142}
1143
1144static int do_getpgid(pid_t pid)
1145{
1146 struct task_struct *p;
1147 struct pid *grp;
1148 int retval;
1149
1150 rcu_read_lock();
1151 if (!pid)
1152 grp = task_pgrp(current);
1153 else {
1154 retval = -ESRCH;
1155 p = find_task_by_vpid(pid);
1156 if (!p)
1157 goto out;
1158 grp = task_pgrp(p);
1159 if (!grp)
1160 goto out;
1161
1162 retval = security_task_getpgid(p);
1163 if (retval)
1164 goto out;
1165 }
1166 retval = pid_vnr(grp);
1167out:
1168 rcu_read_unlock();
1169 return retval;
1170}
1171
1172SYSCALL_DEFINE1(getpgid, pid_t, pid)
1173{
1174 return do_getpgid(pid);
1175}
1176
1177#ifdef __ARCH_WANT_SYS_GETPGRP
1178
1179SYSCALL_DEFINE0(getpgrp)
1180{
1181 return do_getpgid(0);
1182}
1183
1184#endif
1185
1186SYSCALL_DEFINE1(getsid, pid_t, pid)
1187{
1188 struct task_struct *p;
1189 struct pid *sid;
1190 int retval;
1191
1192 rcu_read_lock();
1193 if (!pid)
1194 sid = task_session(current);
1195 else {
1196 retval = -ESRCH;
1197 p = find_task_by_vpid(pid);
1198 if (!p)
1199 goto out;
1200 sid = task_session(p);
1201 if (!sid)
1202 goto out;
1203
1204 retval = security_task_getsid(p);
1205 if (retval)
1206 goto out;
1207 }
1208 retval = pid_vnr(sid);
1209out:
1210 rcu_read_unlock();
1211 return retval;
1212}
1213
1214static void set_special_pids(struct pid *pid)
1215{
1216 struct task_struct *curr = current->group_leader;
1217
1218 if (task_session(curr) != pid)
1219 change_pid(curr, PIDTYPE_SID, pid);
1220
1221 if (task_pgrp(curr) != pid)
1222 change_pid(curr, PIDTYPE_PGID, pid);
1223}
1224
1225int ksys_setsid(void)
1226{
1227 struct task_struct *group_leader = current->group_leader;
1228 struct pid *sid = task_pid(group_leader);
1229 pid_t session = pid_vnr(sid);
1230 int err = -EPERM;
1231
1232 write_lock_irq(&tasklist_lock);
1233 /* Fail if I am already a session leader */
1234 if (group_leader->signal->leader)
1235 goto out;
1236
1237 /* Fail if a process group id already exists that equals the
1238 * proposed session id.
1239 */
1240 if (pid_task(sid, PIDTYPE_PGID))
1241 goto out;
1242
1243 group_leader->signal->leader = 1;
1244 set_special_pids(sid);
1245
1246 proc_clear_tty(group_leader);
1247
1248 err = session;
1249out:
1250 write_unlock_irq(&tasklist_lock);
1251 if (err > 0) {
1252 proc_sid_connector(group_leader);
1253 sched_autogroup_create_attach(group_leader);
1254 }
1255 return err;
1256}
1257
1258SYSCALL_DEFINE0(setsid)
1259{
1260 return ksys_setsid();
1261}
1262
1263DECLARE_RWSEM(uts_sem);
1264
1265#ifdef COMPAT_UTS_MACHINE
1266#define override_architecture(name) \
1267 (personality(current->personality) == PER_LINUX32 && \
1268 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1269 sizeof(COMPAT_UTS_MACHINE)))
1270#else
1271#define override_architecture(name) 0
1272#endif
1273
1274/*
1275 * Work around broken programs that cannot handle "Linux 3.0".
1276 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1277 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1278 * 2.6.60.
1279 */
1280static int override_release(char __user *release, size_t len)
1281{
1282 int ret = 0;
1283
1284 if (current->personality & UNAME26) {
1285 const char *rest = UTS_RELEASE;
1286 char buf[65] = { 0 };
1287 int ndots = 0;
1288 unsigned v;
1289 size_t copy;
1290
1291 while (*rest) {
1292 if (*rest == '.' && ++ndots >= 3)
1293 break;
1294 if (!isdigit(*rest) && *rest != '.')
1295 break;
1296 rest++;
1297 }
1298 v = LINUX_VERSION_PATCHLEVEL + 60;
1299 copy = clamp_t(size_t, len, 1, sizeof(buf));
1300 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1301 ret = copy_to_user(release, buf, copy + 1);
1302 }
1303 return ret;
1304}
1305
1306SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1307{
1308 struct new_utsname tmp;
1309
1310 down_read(&uts_sem);
1311 memcpy(&tmp, utsname(), sizeof(tmp));
1312 up_read(&uts_sem);
1313 if (copy_to_user(name, &tmp, sizeof(tmp)))
1314 return -EFAULT;
1315
1316 if (override_release(name->release, sizeof(name->release)))
1317 return -EFAULT;
1318 if (override_architecture(name))
1319 return -EFAULT;
1320 return 0;
1321}
1322
1323#ifdef __ARCH_WANT_SYS_OLD_UNAME
1324/*
1325 * Old cruft
1326 */
1327SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1328{
1329 struct old_utsname tmp;
1330
1331 if (!name)
1332 return -EFAULT;
1333
1334 down_read(&uts_sem);
1335 memcpy(&tmp, utsname(), sizeof(tmp));
1336 up_read(&uts_sem);
1337 if (copy_to_user(name, &tmp, sizeof(tmp)))
1338 return -EFAULT;
1339
1340 if (override_release(name->release, sizeof(name->release)))
1341 return -EFAULT;
1342 if (override_architecture(name))
1343 return -EFAULT;
1344 return 0;
1345}
1346
1347SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1348{
1349 struct oldold_utsname tmp;
1350
1351 if (!name)
1352 return -EFAULT;
1353
1354 memset(&tmp, 0, sizeof(tmp));
1355
1356 down_read(&uts_sem);
1357 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1358 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1359 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1360 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1361 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1362 up_read(&uts_sem);
1363 if (copy_to_user(name, &tmp, sizeof(tmp)))
1364 return -EFAULT;
1365
1366 if (override_architecture(name))
1367 return -EFAULT;
1368 if (override_release(name->release, sizeof(name->release)))
1369 return -EFAULT;
1370 return 0;
1371}
1372#endif
1373
1374SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1375{
1376 int errno;
1377 char tmp[__NEW_UTS_LEN];
1378
1379 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1380 return -EPERM;
1381
1382 if (len < 0 || len > __NEW_UTS_LEN)
1383 return -EINVAL;
1384 errno = -EFAULT;
1385 if (!copy_from_user(tmp, name, len)) {
1386 struct new_utsname *u;
1387
1388 add_device_randomness(tmp, len);
1389 down_write(&uts_sem);
1390 u = utsname();
1391 memcpy(u->nodename, tmp, len);
1392 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1393 errno = 0;
1394 uts_proc_notify(UTS_PROC_HOSTNAME);
1395 up_write(&uts_sem);
1396 }
1397 return errno;
1398}
1399
1400#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1401
1402SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1403{
1404 int i;
1405 struct new_utsname *u;
1406 char tmp[__NEW_UTS_LEN + 1];
1407
1408 if (len < 0)
1409 return -EINVAL;
1410 down_read(&uts_sem);
1411 u = utsname();
1412 i = 1 + strlen(u->nodename);
1413 if (i > len)
1414 i = len;
1415 memcpy(tmp, u->nodename, i);
1416 up_read(&uts_sem);
1417 if (copy_to_user(name, tmp, i))
1418 return -EFAULT;
1419 return 0;
1420}
1421
1422#endif
1423
1424/*
1425 * Only setdomainname; getdomainname can be implemented by calling
1426 * uname()
1427 */
1428SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1429{
1430 int errno;
1431 char tmp[__NEW_UTS_LEN];
1432
1433 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1434 return -EPERM;
1435 if (len < 0 || len > __NEW_UTS_LEN)
1436 return -EINVAL;
1437
1438 errno = -EFAULT;
1439 if (!copy_from_user(tmp, name, len)) {
1440 struct new_utsname *u;
1441
1442 add_device_randomness(tmp, len);
1443 down_write(&uts_sem);
1444 u = utsname();
1445 memcpy(u->domainname, tmp, len);
1446 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1447 errno = 0;
1448 uts_proc_notify(UTS_PROC_DOMAINNAME);
1449 up_write(&uts_sem);
1450 }
1451 return errno;
1452}
1453
1454/* make sure you are allowed to change @tsk limits before calling this */
1455static int do_prlimit(struct task_struct *tsk, unsigned int resource,
1456 struct rlimit *new_rlim, struct rlimit *old_rlim)
1457{
1458 struct rlimit *rlim;
1459 int retval = 0;
1460
1461 if (resource >= RLIM_NLIMITS)
1462 return -EINVAL;
1463 resource = array_index_nospec(resource, RLIM_NLIMITS);
1464
1465 if (new_rlim) {
1466 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1467 return -EINVAL;
1468 if (resource == RLIMIT_NOFILE &&
1469 new_rlim->rlim_max > sysctl_nr_open)
1470 return -EPERM;
1471 }
1472
1473 /* Holding a refcount on tsk protects tsk->signal from disappearing. */
1474 rlim = tsk->signal->rlim + resource;
1475 task_lock(tsk->group_leader);
1476 if (new_rlim) {
1477 /*
1478 * Keep the capable check against init_user_ns until cgroups can
1479 * contain all limits.
1480 */
1481 if (new_rlim->rlim_max > rlim->rlim_max &&
1482 !capable(CAP_SYS_RESOURCE))
1483 retval = -EPERM;
1484 if (!retval)
1485 retval = security_task_setrlimit(tsk, resource, new_rlim);
1486 }
1487 if (!retval) {
1488 if (old_rlim)
1489 *old_rlim = *rlim;
1490 if (new_rlim)
1491 *rlim = *new_rlim;
1492 }
1493 task_unlock(tsk->group_leader);
1494
1495 /*
1496 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1497 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1498 * ignores the rlimit.
1499 */
1500 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1501 new_rlim->rlim_cur != RLIM_INFINITY &&
1502 IS_ENABLED(CONFIG_POSIX_TIMERS)) {
1503 /*
1504 * update_rlimit_cpu can fail if the task is exiting, but there
1505 * may be other tasks in the thread group that are not exiting,
1506 * and they need their cpu timers adjusted.
1507 *
1508 * The group_leader is the last task to be released, so if we
1509 * cannot update_rlimit_cpu on it, then the entire process is
1510 * exiting and we do not need to update at all.
1511 */
1512 update_rlimit_cpu(tsk->group_leader, new_rlim->rlim_cur);
1513 }
1514
1515 return retval;
1516}
1517
1518SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1519{
1520 struct rlimit value;
1521 int ret;
1522
1523 ret = do_prlimit(current, resource, NULL, &value);
1524 if (!ret)
1525 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1526
1527 return ret;
1528}
1529
1530#ifdef CONFIG_COMPAT
1531
1532COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1533 struct compat_rlimit __user *, rlim)
1534{
1535 struct rlimit r;
1536 struct compat_rlimit r32;
1537
1538 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1539 return -EFAULT;
1540
1541 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1542 r.rlim_cur = RLIM_INFINITY;
1543 else
1544 r.rlim_cur = r32.rlim_cur;
1545 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1546 r.rlim_max = RLIM_INFINITY;
1547 else
1548 r.rlim_max = r32.rlim_max;
1549 return do_prlimit(current, resource, &r, NULL);
1550}
1551
1552COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1553 struct compat_rlimit __user *, rlim)
1554{
1555 struct rlimit r;
1556 int ret;
1557
1558 ret = do_prlimit(current, resource, NULL, &r);
1559 if (!ret) {
1560 struct compat_rlimit r32;
1561 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1562 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1563 else
1564 r32.rlim_cur = r.rlim_cur;
1565 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1566 r32.rlim_max = COMPAT_RLIM_INFINITY;
1567 else
1568 r32.rlim_max = r.rlim_max;
1569
1570 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1571 return -EFAULT;
1572 }
1573 return ret;
1574}
1575
1576#endif
1577
1578#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1579
1580/*
1581 * Back compatibility for getrlimit. Needed for some apps.
1582 */
1583SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1584 struct rlimit __user *, rlim)
1585{
1586 struct rlimit x;
1587 if (resource >= RLIM_NLIMITS)
1588 return -EINVAL;
1589
1590 resource = array_index_nospec(resource, RLIM_NLIMITS);
1591 task_lock(current->group_leader);
1592 x = current->signal->rlim[resource];
1593 task_unlock(current->group_leader);
1594 if (x.rlim_cur > 0x7FFFFFFF)
1595 x.rlim_cur = 0x7FFFFFFF;
1596 if (x.rlim_max > 0x7FFFFFFF)
1597 x.rlim_max = 0x7FFFFFFF;
1598 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1599}
1600
1601#ifdef CONFIG_COMPAT
1602COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1603 struct compat_rlimit __user *, rlim)
1604{
1605 struct rlimit r;
1606
1607 if (resource >= RLIM_NLIMITS)
1608 return -EINVAL;
1609
1610 resource = array_index_nospec(resource, RLIM_NLIMITS);
1611 task_lock(current->group_leader);
1612 r = current->signal->rlim[resource];
1613 task_unlock(current->group_leader);
1614 if (r.rlim_cur > 0x7FFFFFFF)
1615 r.rlim_cur = 0x7FFFFFFF;
1616 if (r.rlim_max > 0x7FFFFFFF)
1617 r.rlim_max = 0x7FFFFFFF;
1618
1619 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1620 put_user(r.rlim_max, &rlim->rlim_max))
1621 return -EFAULT;
1622 return 0;
1623}
1624#endif
1625
1626#endif
1627
1628static inline bool rlim64_is_infinity(__u64 rlim64)
1629{
1630#if BITS_PER_LONG < 64
1631 return rlim64 >= ULONG_MAX;
1632#else
1633 return rlim64 == RLIM64_INFINITY;
1634#endif
1635}
1636
1637static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1638{
1639 if (rlim->rlim_cur == RLIM_INFINITY)
1640 rlim64->rlim_cur = RLIM64_INFINITY;
1641 else
1642 rlim64->rlim_cur = rlim->rlim_cur;
1643 if (rlim->rlim_max == RLIM_INFINITY)
1644 rlim64->rlim_max = RLIM64_INFINITY;
1645 else
1646 rlim64->rlim_max = rlim->rlim_max;
1647}
1648
1649static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1650{
1651 if (rlim64_is_infinity(rlim64->rlim_cur))
1652 rlim->rlim_cur = RLIM_INFINITY;
1653 else
1654 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1655 if (rlim64_is_infinity(rlim64->rlim_max))
1656 rlim->rlim_max = RLIM_INFINITY;
1657 else
1658 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1659}
1660
1661/* rcu lock must be held */
1662static int check_prlimit_permission(struct task_struct *task,
1663 unsigned int flags)
1664{
1665 const struct cred *cred = current_cred(), *tcred;
1666 bool id_match;
1667
1668 if (current == task)
1669 return 0;
1670
1671 tcred = __task_cred(task);
1672 id_match = (uid_eq(cred->uid, tcred->euid) &&
1673 uid_eq(cred->uid, tcred->suid) &&
1674 uid_eq(cred->uid, tcred->uid) &&
1675 gid_eq(cred->gid, tcred->egid) &&
1676 gid_eq(cred->gid, tcred->sgid) &&
1677 gid_eq(cred->gid, tcred->gid));
1678 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1679 return -EPERM;
1680
1681 return security_task_prlimit(cred, tcred, flags);
1682}
1683
1684SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1685 const struct rlimit64 __user *, new_rlim,
1686 struct rlimit64 __user *, old_rlim)
1687{
1688 struct rlimit64 old64, new64;
1689 struct rlimit old, new;
1690 struct task_struct *tsk;
1691 unsigned int checkflags = 0;
1692 int ret;
1693
1694 if (old_rlim)
1695 checkflags |= LSM_PRLIMIT_READ;
1696
1697 if (new_rlim) {
1698 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1699 return -EFAULT;
1700 rlim64_to_rlim(&new64, &new);
1701 checkflags |= LSM_PRLIMIT_WRITE;
1702 }
1703
1704 rcu_read_lock();
1705 tsk = pid ? find_task_by_vpid(pid) : current;
1706 if (!tsk) {
1707 rcu_read_unlock();
1708 return -ESRCH;
1709 }
1710 ret = check_prlimit_permission(tsk, checkflags);
1711 if (ret) {
1712 rcu_read_unlock();
1713 return ret;
1714 }
1715 get_task_struct(tsk);
1716 rcu_read_unlock();
1717
1718 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1719 old_rlim ? &old : NULL);
1720
1721 if (!ret && old_rlim) {
1722 rlim_to_rlim64(&old, &old64);
1723 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1724 ret = -EFAULT;
1725 }
1726
1727 put_task_struct(tsk);
1728 return ret;
1729}
1730
1731SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1732{
1733 struct rlimit new_rlim;
1734
1735 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1736 return -EFAULT;
1737 return do_prlimit(current, resource, &new_rlim, NULL);
1738}
1739
1740/*
1741 * It would make sense to put struct rusage in the task_struct,
1742 * except that would make the task_struct be *really big*. After
1743 * task_struct gets moved into malloc'ed memory, it would
1744 * make sense to do this. It will make moving the rest of the information
1745 * a lot simpler! (Which we're not doing right now because we're not
1746 * measuring them yet).
1747 *
1748 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1749 * races with threads incrementing their own counters. But since word
1750 * reads are atomic, we either get new values or old values and we don't
1751 * care which for the sums. We always take the siglock to protect reading
1752 * the c* fields from p->signal from races with exit.c updating those
1753 * fields when reaping, so a sample either gets all the additions of a
1754 * given child after it's reaped, or none so this sample is before reaping.
1755 *
1756 * Locking:
1757 * We need to take the siglock for CHILDEREN, SELF and BOTH
1758 * for the cases current multithreaded, non-current single threaded
1759 * non-current multithreaded. Thread traversal is now safe with
1760 * the siglock held.
1761 * Strictly speaking, we donot need to take the siglock if we are current and
1762 * single threaded, as no one else can take our signal_struct away, no one
1763 * else can reap the children to update signal->c* counters, and no one else
1764 * can race with the signal-> fields. If we do not take any lock, the
1765 * signal-> fields could be read out of order while another thread was just
1766 * exiting. So we should place a read memory barrier when we avoid the lock.
1767 * On the writer side, write memory barrier is implied in __exit_signal
1768 * as __exit_signal releases the siglock spinlock after updating the signal->
1769 * fields. But we don't do this yet to keep things simple.
1770 *
1771 */
1772
1773static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1774{
1775 r->ru_nvcsw += t->nvcsw;
1776 r->ru_nivcsw += t->nivcsw;
1777 r->ru_minflt += t->min_flt;
1778 r->ru_majflt += t->maj_flt;
1779 r->ru_inblock += task_io_get_inblock(t);
1780 r->ru_oublock += task_io_get_oublock(t);
1781}
1782
1783void getrusage(struct task_struct *p, int who, struct rusage *r)
1784{
1785 struct task_struct *t;
1786 unsigned long flags;
1787 u64 tgutime, tgstime, utime, stime;
1788 unsigned long maxrss;
1789 struct mm_struct *mm;
1790 struct signal_struct *sig = p->signal;
1791 unsigned int seq = 0;
1792
1793retry:
1794 memset(r, 0, sizeof(*r));
1795 utime = stime = 0;
1796 maxrss = 0;
1797
1798 if (who == RUSAGE_THREAD) {
1799 task_cputime_adjusted(current, &utime, &stime);
1800 accumulate_thread_rusage(p, r);
1801 maxrss = sig->maxrss;
1802 goto out_thread;
1803 }
1804
1805 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
1806
1807 switch (who) {
1808 case RUSAGE_BOTH:
1809 case RUSAGE_CHILDREN:
1810 utime = sig->cutime;
1811 stime = sig->cstime;
1812 r->ru_nvcsw = sig->cnvcsw;
1813 r->ru_nivcsw = sig->cnivcsw;
1814 r->ru_minflt = sig->cmin_flt;
1815 r->ru_majflt = sig->cmaj_flt;
1816 r->ru_inblock = sig->cinblock;
1817 r->ru_oublock = sig->coublock;
1818 maxrss = sig->cmaxrss;
1819
1820 if (who == RUSAGE_CHILDREN)
1821 break;
1822 fallthrough;
1823
1824 case RUSAGE_SELF:
1825 r->ru_nvcsw += sig->nvcsw;
1826 r->ru_nivcsw += sig->nivcsw;
1827 r->ru_minflt += sig->min_flt;
1828 r->ru_majflt += sig->maj_flt;
1829 r->ru_inblock += sig->inblock;
1830 r->ru_oublock += sig->oublock;
1831 if (maxrss < sig->maxrss)
1832 maxrss = sig->maxrss;
1833
1834 rcu_read_lock();
1835 __for_each_thread(sig, t)
1836 accumulate_thread_rusage(t, r);
1837 rcu_read_unlock();
1838
1839 break;
1840
1841 default:
1842 BUG();
1843 }
1844
1845 if (need_seqretry(&sig->stats_lock, seq)) {
1846 seq = 1;
1847 goto retry;
1848 }
1849 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
1850
1851 if (who == RUSAGE_CHILDREN)
1852 goto out_children;
1853
1854 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1855 utime += tgutime;
1856 stime += tgstime;
1857
1858out_thread:
1859 mm = get_task_mm(p);
1860 if (mm) {
1861 setmax_mm_hiwater_rss(&maxrss, mm);
1862 mmput(mm);
1863 }
1864
1865out_children:
1866 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1867 r->ru_utime = ns_to_kernel_old_timeval(utime);
1868 r->ru_stime = ns_to_kernel_old_timeval(stime);
1869}
1870
1871SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1872{
1873 struct rusage r;
1874
1875 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1876 who != RUSAGE_THREAD)
1877 return -EINVAL;
1878
1879 getrusage(current, who, &r);
1880 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1881}
1882
1883#ifdef CONFIG_COMPAT
1884COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1885{
1886 struct rusage r;
1887
1888 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1889 who != RUSAGE_THREAD)
1890 return -EINVAL;
1891
1892 getrusage(current, who, &r);
1893 return put_compat_rusage(&r, ru);
1894}
1895#endif
1896
1897SYSCALL_DEFINE1(umask, int, mask)
1898{
1899 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1900 return mask;
1901}
1902
1903static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1904{
1905 struct fd exe;
1906 struct inode *inode;
1907 int err;
1908
1909 exe = fdget(fd);
1910 if (!exe.file)
1911 return -EBADF;
1912
1913 inode = file_inode(exe.file);
1914
1915 /*
1916 * Because the original mm->exe_file points to executable file, make
1917 * sure that this one is executable as well, to avoid breaking an
1918 * overall picture.
1919 */
1920 err = -EACCES;
1921 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1922 goto exit;
1923
1924 err = file_permission(exe.file, MAY_EXEC);
1925 if (err)
1926 goto exit;
1927
1928 err = replace_mm_exe_file(mm, exe.file);
1929exit:
1930 fdput(exe);
1931 return err;
1932}
1933
1934/*
1935 * Check arithmetic relations of passed addresses.
1936 *
1937 * WARNING: we don't require any capability here so be very careful
1938 * in what is allowed for modification from userspace.
1939 */
1940static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1941{
1942 unsigned long mmap_max_addr = TASK_SIZE;
1943 int error = -EINVAL, i;
1944
1945 static const unsigned char offsets[] = {
1946 offsetof(struct prctl_mm_map, start_code),
1947 offsetof(struct prctl_mm_map, end_code),
1948 offsetof(struct prctl_mm_map, start_data),
1949 offsetof(struct prctl_mm_map, end_data),
1950 offsetof(struct prctl_mm_map, start_brk),
1951 offsetof(struct prctl_mm_map, brk),
1952 offsetof(struct prctl_mm_map, start_stack),
1953 offsetof(struct prctl_mm_map, arg_start),
1954 offsetof(struct prctl_mm_map, arg_end),
1955 offsetof(struct prctl_mm_map, env_start),
1956 offsetof(struct prctl_mm_map, env_end),
1957 };
1958
1959 /*
1960 * Make sure the members are not somewhere outside
1961 * of allowed address space.
1962 */
1963 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1964 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1965
1966 if ((unsigned long)val >= mmap_max_addr ||
1967 (unsigned long)val < mmap_min_addr)
1968 goto out;
1969 }
1970
1971 /*
1972 * Make sure the pairs are ordered.
1973 */
1974#define __prctl_check_order(__m1, __op, __m2) \
1975 ((unsigned long)prctl_map->__m1 __op \
1976 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1977 error = __prctl_check_order(start_code, <, end_code);
1978 error |= __prctl_check_order(start_data,<=, end_data);
1979 error |= __prctl_check_order(start_brk, <=, brk);
1980 error |= __prctl_check_order(arg_start, <=, arg_end);
1981 error |= __prctl_check_order(env_start, <=, env_end);
1982 if (error)
1983 goto out;
1984#undef __prctl_check_order
1985
1986 error = -EINVAL;
1987
1988 /*
1989 * Neither we should allow to override limits if they set.
1990 */
1991 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1992 prctl_map->start_brk, prctl_map->end_data,
1993 prctl_map->start_data))
1994 goto out;
1995
1996 error = 0;
1997out:
1998 return error;
1999}
2000
2001#ifdef CONFIG_CHECKPOINT_RESTORE
2002static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
2003{
2004 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
2005 unsigned long user_auxv[AT_VECTOR_SIZE];
2006 struct mm_struct *mm = current->mm;
2007 int error;
2008
2009 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2010 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
2011
2012 if (opt == PR_SET_MM_MAP_SIZE)
2013 return put_user((unsigned int)sizeof(prctl_map),
2014 (unsigned int __user *)addr);
2015
2016 if (data_size != sizeof(prctl_map))
2017 return -EINVAL;
2018
2019 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2020 return -EFAULT;
2021
2022 error = validate_prctl_map_addr(&prctl_map);
2023 if (error)
2024 return error;
2025
2026 if (prctl_map.auxv_size) {
2027 /*
2028 * Someone is trying to cheat the auxv vector.
2029 */
2030 if (!prctl_map.auxv ||
2031 prctl_map.auxv_size > sizeof(mm->saved_auxv))
2032 return -EINVAL;
2033
2034 memset(user_auxv, 0, sizeof(user_auxv));
2035 if (copy_from_user(user_auxv,
2036 (const void __user *)prctl_map.auxv,
2037 prctl_map.auxv_size))
2038 return -EFAULT;
2039
2040 /* Last entry must be AT_NULL as specification requires */
2041 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2042 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2043 }
2044
2045 if (prctl_map.exe_fd != (u32)-1) {
2046 /*
2047 * Check if the current user is checkpoint/restore capable.
2048 * At the time of this writing, it checks for CAP_SYS_ADMIN
2049 * or CAP_CHECKPOINT_RESTORE.
2050 * Note that a user with access to ptrace can masquerade an
2051 * arbitrary program as any executable, even setuid ones.
2052 * This may have implications in the tomoyo subsystem.
2053 */
2054 if (!checkpoint_restore_ns_capable(current_user_ns()))
2055 return -EPERM;
2056
2057 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2058 if (error)
2059 return error;
2060 }
2061
2062 /*
2063 * arg_lock protects concurrent updates but we still need mmap_lock for
2064 * read to exclude races with sys_brk.
2065 */
2066 mmap_read_lock(mm);
2067
2068 /*
2069 * We don't validate if these members are pointing to
2070 * real present VMAs because application may have correspond
2071 * VMAs already unmapped and kernel uses these members for statistics
2072 * output in procfs mostly, except
2073 *
2074 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2075 * for VMAs when updating these members so anything wrong written
2076 * here cause kernel to swear at userspace program but won't lead
2077 * to any problem in kernel itself
2078 */
2079
2080 spin_lock(&mm->arg_lock);
2081 mm->start_code = prctl_map.start_code;
2082 mm->end_code = prctl_map.end_code;
2083 mm->start_data = prctl_map.start_data;
2084 mm->end_data = prctl_map.end_data;
2085 mm->start_brk = prctl_map.start_brk;
2086 mm->brk = prctl_map.brk;
2087 mm->start_stack = prctl_map.start_stack;
2088 mm->arg_start = prctl_map.arg_start;
2089 mm->arg_end = prctl_map.arg_end;
2090 mm->env_start = prctl_map.env_start;
2091 mm->env_end = prctl_map.env_end;
2092 spin_unlock(&mm->arg_lock);
2093
2094 /*
2095 * Note this update of @saved_auxv is lockless thus
2096 * if someone reads this member in procfs while we're
2097 * updating -- it may get partly updated results. It's
2098 * known and acceptable trade off: we leave it as is to
2099 * not introduce additional locks here making the kernel
2100 * more complex.
2101 */
2102 if (prctl_map.auxv_size)
2103 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2104
2105 mmap_read_unlock(mm);
2106 return 0;
2107}
2108#endif /* CONFIG_CHECKPOINT_RESTORE */
2109
2110static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2111 unsigned long len)
2112{
2113 /*
2114 * This doesn't move the auxiliary vector itself since it's pinned to
2115 * mm_struct, but it permits filling the vector with new values. It's
2116 * up to the caller to provide sane values here, otherwise userspace
2117 * tools which use this vector might be unhappy.
2118 */
2119 unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2120
2121 if (len > sizeof(user_auxv))
2122 return -EINVAL;
2123
2124 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2125 return -EFAULT;
2126
2127 /* Make sure the last entry is always AT_NULL */
2128 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2129 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2130
2131 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2132
2133 task_lock(current);
2134 memcpy(mm->saved_auxv, user_auxv, len);
2135 task_unlock(current);
2136
2137 return 0;
2138}
2139
2140static int prctl_set_mm(int opt, unsigned long addr,
2141 unsigned long arg4, unsigned long arg5)
2142{
2143 struct mm_struct *mm = current->mm;
2144 struct prctl_mm_map prctl_map = {
2145 .auxv = NULL,
2146 .auxv_size = 0,
2147 .exe_fd = -1,
2148 };
2149 struct vm_area_struct *vma;
2150 int error;
2151
2152 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2153 opt != PR_SET_MM_MAP &&
2154 opt != PR_SET_MM_MAP_SIZE)))
2155 return -EINVAL;
2156
2157#ifdef CONFIG_CHECKPOINT_RESTORE
2158 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2159 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2160#endif
2161
2162 if (!capable(CAP_SYS_RESOURCE))
2163 return -EPERM;
2164
2165 if (opt == PR_SET_MM_EXE_FILE)
2166 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2167
2168 if (opt == PR_SET_MM_AUXV)
2169 return prctl_set_auxv(mm, addr, arg4);
2170
2171 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2172 return -EINVAL;
2173
2174 error = -EINVAL;
2175
2176 /*
2177 * arg_lock protects concurrent updates of arg boundaries, we need
2178 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2179 * validation.
2180 */
2181 mmap_read_lock(mm);
2182 vma = find_vma(mm, addr);
2183
2184 spin_lock(&mm->arg_lock);
2185 prctl_map.start_code = mm->start_code;
2186 prctl_map.end_code = mm->end_code;
2187 prctl_map.start_data = mm->start_data;
2188 prctl_map.end_data = mm->end_data;
2189 prctl_map.start_brk = mm->start_brk;
2190 prctl_map.brk = mm->brk;
2191 prctl_map.start_stack = mm->start_stack;
2192 prctl_map.arg_start = mm->arg_start;
2193 prctl_map.arg_end = mm->arg_end;
2194 prctl_map.env_start = mm->env_start;
2195 prctl_map.env_end = mm->env_end;
2196
2197 switch (opt) {
2198 case PR_SET_MM_START_CODE:
2199 prctl_map.start_code = addr;
2200 break;
2201 case PR_SET_MM_END_CODE:
2202 prctl_map.end_code = addr;
2203 break;
2204 case PR_SET_MM_START_DATA:
2205 prctl_map.start_data = addr;
2206 break;
2207 case PR_SET_MM_END_DATA:
2208 prctl_map.end_data = addr;
2209 break;
2210 case PR_SET_MM_START_STACK:
2211 prctl_map.start_stack = addr;
2212 break;
2213 case PR_SET_MM_START_BRK:
2214 prctl_map.start_brk = addr;
2215 break;
2216 case PR_SET_MM_BRK:
2217 prctl_map.brk = addr;
2218 break;
2219 case PR_SET_MM_ARG_START:
2220 prctl_map.arg_start = addr;
2221 break;
2222 case PR_SET_MM_ARG_END:
2223 prctl_map.arg_end = addr;
2224 break;
2225 case PR_SET_MM_ENV_START:
2226 prctl_map.env_start = addr;
2227 break;
2228 case PR_SET_MM_ENV_END:
2229 prctl_map.env_end = addr;
2230 break;
2231 default:
2232 goto out;
2233 }
2234
2235 error = validate_prctl_map_addr(&prctl_map);
2236 if (error)
2237 goto out;
2238
2239 switch (opt) {
2240 /*
2241 * If command line arguments and environment
2242 * are placed somewhere else on stack, we can
2243 * set them up here, ARG_START/END to setup
2244 * command line arguments and ENV_START/END
2245 * for environment.
2246 */
2247 case PR_SET_MM_START_STACK:
2248 case PR_SET_MM_ARG_START:
2249 case PR_SET_MM_ARG_END:
2250 case PR_SET_MM_ENV_START:
2251 case PR_SET_MM_ENV_END:
2252 if (!vma) {
2253 error = -EFAULT;
2254 goto out;
2255 }
2256 }
2257
2258 mm->start_code = prctl_map.start_code;
2259 mm->end_code = prctl_map.end_code;
2260 mm->start_data = prctl_map.start_data;
2261 mm->end_data = prctl_map.end_data;
2262 mm->start_brk = prctl_map.start_brk;
2263 mm->brk = prctl_map.brk;
2264 mm->start_stack = prctl_map.start_stack;
2265 mm->arg_start = prctl_map.arg_start;
2266 mm->arg_end = prctl_map.arg_end;
2267 mm->env_start = prctl_map.env_start;
2268 mm->env_end = prctl_map.env_end;
2269
2270 error = 0;
2271out:
2272 spin_unlock(&mm->arg_lock);
2273 mmap_read_unlock(mm);
2274 return error;
2275}
2276
2277#ifdef CONFIG_CHECKPOINT_RESTORE
2278static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2279{
2280 return put_user(me->clear_child_tid, tid_addr);
2281}
2282#else
2283static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2284{
2285 return -EINVAL;
2286}
2287#endif
2288
2289static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2290{
2291 /*
2292 * If task has has_child_subreaper - all its descendants
2293 * already have these flag too and new descendants will
2294 * inherit it on fork, skip them.
2295 *
2296 * If we've found child_reaper - skip descendants in
2297 * it's subtree as they will never get out pidns.
2298 */
2299 if (p->signal->has_child_subreaper ||
2300 is_child_reaper(task_pid(p)))
2301 return 0;
2302
2303 p->signal->has_child_subreaper = 1;
2304 return 1;
2305}
2306
2307int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2308{
2309 return -EINVAL;
2310}
2311
2312int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2313 unsigned long ctrl)
2314{
2315 return -EINVAL;
2316}
2317
2318#define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2319
2320#ifdef CONFIG_ANON_VMA_NAME
2321
2322#define ANON_VMA_NAME_MAX_LEN 80
2323#define ANON_VMA_NAME_INVALID_CHARS "\\`$[]"
2324
2325static inline bool is_valid_name_char(char ch)
2326{
2327 /* printable ascii characters, excluding ANON_VMA_NAME_INVALID_CHARS */
2328 return ch > 0x1f && ch < 0x7f &&
2329 !strchr(ANON_VMA_NAME_INVALID_CHARS, ch);
2330}
2331
2332static int prctl_set_vma(unsigned long opt, unsigned long addr,
2333 unsigned long size, unsigned long arg)
2334{
2335 struct mm_struct *mm = current->mm;
2336 const char __user *uname;
2337 struct anon_vma_name *anon_name = NULL;
2338 int error;
2339
2340 switch (opt) {
2341 case PR_SET_VMA_ANON_NAME:
2342 uname = (const char __user *)arg;
2343 if (uname) {
2344 char *name, *pch;
2345
2346 name = strndup_user(uname, ANON_VMA_NAME_MAX_LEN);
2347 if (IS_ERR(name))
2348 return PTR_ERR(name);
2349
2350 for (pch = name; *pch != '\0'; pch++) {
2351 if (!is_valid_name_char(*pch)) {
2352 kfree(name);
2353 return -EINVAL;
2354 }
2355 }
2356 /* anon_vma has its own copy */
2357 anon_name = anon_vma_name_alloc(name);
2358 kfree(name);
2359 if (!anon_name)
2360 return -ENOMEM;
2361
2362 }
2363
2364 mmap_write_lock(mm);
2365 error = madvise_set_anon_name(mm, addr, size, anon_name);
2366 mmap_write_unlock(mm);
2367 anon_vma_name_put(anon_name);
2368 break;
2369 default:
2370 error = -EINVAL;
2371 }
2372
2373 return error;
2374}
2375
2376#else /* CONFIG_ANON_VMA_NAME */
2377static int prctl_set_vma(unsigned long opt, unsigned long start,
2378 unsigned long size, unsigned long arg)
2379{
2380 return -EINVAL;
2381}
2382#endif /* CONFIG_ANON_VMA_NAME */
2383
2384static inline unsigned long get_current_mdwe(void)
2385{
2386 unsigned long ret = 0;
2387
2388 if (test_bit(MMF_HAS_MDWE, ¤t->mm->flags))
2389 ret |= PR_MDWE_REFUSE_EXEC_GAIN;
2390 if (test_bit(MMF_HAS_MDWE_NO_INHERIT, ¤t->mm->flags))
2391 ret |= PR_MDWE_NO_INHERIT;
2392
2393 return ret;
2394}
2395
2396static inline int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
2397 unsigned long arg4, unsigned long arg5)
2398{
2399 unsigned long current_bits;
2400
2401 if (arg3 || arg4 || arg5)
2402 return -EINVAL;
2403
2404 if (bits & ~(PR_MDWE_REFUSE_EXEC_GAIN | PR_MDWE_NO_INHERIT))
2405 return -EINVAL;
2406
2407 /* NO_INHERIT only makes sense with REFUSE_EXEC_GAIN */
2408 if (bits & PR_MDWE_NO_INHERIT && !(bits & PR_MDWE_REFUSE_EXEC_GAIN))
2409 return -EINVAL;
2410
2411 /* PARISC cannot allow mdwe as it needs writable stacks */
2412 if (IS_ENABLED(CONFIG_PARISC))
2413 return -EINVAL;
2414
2415 current_bits = get_current_mdwe();
2416 if (current_bits && current_bits != bits)
2417 return -EPERM; /* Cannot unset the flags */
2418
2419 if (bits & PR_MDWE_NO_INHERIT)
2420 set_bit(MMF_HAS_MDWE_NO_INHERIT, ¤t->mm->flags);
2421 if (bits & PR_MDWE_REFUSE_EXEC_GAIN)
2422 set_bit(MMF_HAS_MDWE, ¤t->mm->flags);
2423
2424 return 0;
2425}
2426
2427static inline int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
2428 unsigned long arg4, unsigned long arg5)
2429{
2430 if (arg2 || arg3 || arg4 || arg5)
2431 return -EINVAL;
2432 return get_current_mdwe();
2433}
2434
2435static int prctl_get_auxv(void __user *addr, unsigned long len)
2436{
2437 struct mm_struct *mm = current->mm;
2438 unsigned long size = min_t(unsigned long, sizeof(mm->saved_auxv), len);
2439
2440 if (size && copy_to_user(addr, mm->saved_auxv, size))
2441 return -EFAULT;
2442 return sizeof(mm->saved_auxv);
2443}
2444
2445SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2446 unsigned long, arg4, unsigned long, arg5)
2447{
2448 struct task_struct *me = current;
2449 unsigned char comm[sizeof(me->comm)];
2450 long error;
2451
2452 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2453 if (error != -ENOSYS)
2454 return error;
2455
2456 error = 0;
2457 switch (option) {
2458 case PR_SET_PDEATHSIG:
2459 if (!valid_signal(arg2)) {
2460 error = -EINVAL;
2461 break;
2462 }
2463 me->pdeath_signal = arg2;
2464 break;
2465 case PR_GET_PDEATHSIG:
2466 error = put_user(me->pdeath_signal, (int __user *)arg2);
2467 break;
2468 case PR_GET_DUMPABLE:
2469 error = get_dumpable(me->mm);
2470 break;
2471 case PR_SET_DUMPABLE:
2472 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2473 error = -EINVAL;
2474 break;
2475 }
2476 set_dumpable(me->mm, arg2);
2477 break;
2478
2479 case PR_SET_UNALIGN:
2480 error = SET_UNALIGN_CTL(me, arg2);
2481 break;
2482 case PR_GET_UNALIGN:
2483 error = GET_UNALIGN_CTL(me, arg2);
2484 break;
2485 case PR_SET_FPEMU:
2486 error = SET_FPEMU_CTL(me, arg2);
2487 break;
2488 case PR_GET_FPEMU:
2489 error = GET_FPEMU_CTL(me, arg2);
2490 break;
2491 case PR_SET_FPEXC:
2492 error = SET_FPEXC_CTL(me, arg2);
2493 break;
2494 case PR_GET_FPEXC:
2495 error = GET_FPEXC_CTL(me, arg2);
2496 break;
2497 case PR_GET_TIMING:
2498 error = PR_TIMING_STATISTICAL;
2499 break;
2500 case PR_SET_TIMING:
2501 if (arg2 != PR_TIMING_STATISTICAL)
2502 error = -EINVAL;
2503 break;
2504 case PR_SET_NAME:
2505 comm[sizeof(me->comm) - 1] = 0;
2506 if (strncpy_from_user(comm, (char __user *)arg2,
2507 sizeof(me->comm) - 1) < 0)
2508 return -EFAULT;
2509 set_task_comm(me, comm);
2510 proc_comm_connector(me);
2511 break;
2512 case PR_GET_NAME:
2513 get_task_comm(comm, me);
2514 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2515 return -EFAULT;
2516 break;
2517 case PR_GET_ENDIAN:
2518 error = GET_ENDIAN(me, arg2);
2519 break;
2520 case PR_SET_ENDIAN:
2521 error = SET_ENDIAN(me, arg2);
2522 break;
2523 case PR_GET_SECCOMP:
2524 error = prctl_get_seccomp();
2525 break;
2526 case PR_SET_SECCOMP:
2527 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2528 break;
2529 case PR_GET_TSC:
2530 error = GET_TSC_CTL(arg2);
2531 break;
2532 case PR_SET_TSC:
2533 error = SET_TSC_CTL(arg2);
2534 break;
2535 case PR_TASK_PERF_EVENTS_DISABLE:
2536 error = perf_event_task_disable();
2537 break;
2538 case PR_TASK_PERF_EVENTS_ENABLE:
2539 error = perf_event_task_enable();
2540 break;
2541 case PR_GET_TIMERSLACK:
2542 if (current->timer_slack_ns > ULONG_MAX)
2543 error = ULONG_MAX;
2544 else
2545 error = current->timer_slack_ns;
2546 break;
2547 case PR_SET_TIMERSLACK:
2548 if (arg2 <= 0)
2549 current->timer_slack_ns =
2550 current->default_timer_slack_ns;
2551 else
2552 current->timer_slack_ns = arg2;
2553 break;
2554 case PR_MCE_KILL:
2555 if (arg4 | arg5)
2556 return -EINVAL;
2557 switch (arg2) {
2558 case PR_MCE_KILL_CLEAR:
2559 if (arg3 != 0)
2560 return -EINVAL;
2561 current->flags &= ~PF_MCE_PROCESS;
2562 break;
2563 case PR_MCE_KILL_SET:
2564 current->flags |= PF_MCE_PROCESS;
2565 if (arg3 == PR_MCE_KILL_EARLY)
2566 current->flags |= PF_MCE_EARLY;
2567 else if (arg3 == PR_MCE_KILL_LATE)
2568 current->flags &= ~PF_MCE_EARLY;
2569 else if (arg3 == PR_MCE_KILL_DEFAULT)
2570 current->flags &=
2571 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2572 else
2573 return -EINVAL;
2574 break;
2575 default:
2576 return -EINVAL;
2577 }
2578 break;
2579 case PR_MCE_KILL_GET:
2580 if (arg2 | arg3 | arg4 | arg5)
2581 return -EINVAL;
2582 if (current->flags & PF_MCE_PROCESS)
2583 error = (current->flags & PF_MCE_EARLY) ?
2584 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2585 else
2586 error = PR_MCE_KILL_DEFAULT;
2587 break;
2588 case PR_SET_MM:
2589 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2590 break;
2591 case PR_GET_TID_ADDRESS:
2592 error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2593 break;
2594 case PR_SET_CHILD_SUBREAPER:
2595 me->signal->is_child_subreaper = !!arg2;
2596 if (!arg2)
2597 break;
2598
2599 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2600 break;
2601 case PR_GET_CHILD_SUBREAPER:
2602 error = put_user(me->signal->is_child_subreaper,
2603 (int __user *)arg2);
2604 break;
2605 case PR_SET_NO_NEW_PRIVS:
2606 if (arg2 != 1 || arg3 || arg4 || arg5)
2607 return -EINVAL;
2608
2609 task_set_no_new_privs(current);
2610 break;
2611 case PR_GET_NO_NEW_PRIVS:
2612 if (arg2 || arg3 || arg4 || arg5)
2613 return -EINVAL;
2614 return task_no_new_privs(current) ? 1 : 0;
2615 case PR_GET_THP_DISABLE:
2616 if (arg2 || arg3 || arg4 || arg5)
2617 return -EINVAL;
2618 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2619 break;
2620 case PR_SET_THP_DISABLE:
2621 if (arg3 || arg4 || arg5)
2622 return -EINVAL;
2623 if (mmap_write_lock_killable(me->mm))
2624 return -EINTR;
2625 if (arg2)
2626 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2627 else
2628 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2629 mmap_write_unlock(me->mm);
2630 break;
2631 case PR_MPX_ENABLE_MANAGEMENT:
2632 case PR_MPX_DISABLE_MANAGEMENT:
2633 /* No longer implemented: */
2634 return -EINVAL;
2635 case PR_SET_FP_MODE:
2636 error = SET_FP_MODE(me, arg2);
2637 break;
2638 case PR_GET_FP_MODE:
2639 error = GET_FP_MODE(me);
2640 break;
2641 case PR_SVE_SET_VL:
2642 error = SVE_SET_VL(arg2);
2643 break;
2644 case PR_SVE_GET_VL:
2645 error = SVE_GET_VL();
2646 break;
2647 case PR_SME_SET_VL:
2648 error = SME_SET_VL(arg2);
2649 break;
2650 case PR_SME_GET_VL:
2651 error = SME_GET_VL();
2652 break;
2653 case PR_GET_SPECULATION_CTRL:
2654 if (arg3 || arg4 || arg5)
2655 return -EINVAL;
2656 error = arch_prctl_spec_ctrl_get(me, arg2);
2657 break;
2658 case PR_SET_SPECULATION_CTRL:
2659 if (arg4 || arg5)
2660 return -EINVAL;
2661 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2662 break;
2663 case PR_PAC_RESET_KEYS:
2664 if (arg3 || arg4 || arg5)
2665 return -EINVAL;
2666 error = PAC_RESET_KEYS(me, arg2);
2667 break;
2668 case PR_PAC_SET_ENABLED_KEYS:
2669 if (arg4 || arg5)
2670 return -EINVAL;
2671 error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2672 break;
2673 case PR_PAC_GET_ENABLED_KEYS:
2674 if (arg2 || arg3 || arg4 || arg5)
2675 return -EINVAL;
2676 error = PAC_GET_ENABLED_KEYS(me);
2677 break;
2678 case PR_SET_TAGGED_ADDR_CTRL:
2679 if (arg3 || arg4 || arg5)
2680 return -EINVAL;
2681 error = SET_TAGGED_ADDR_CTRL(arg2);
2682 break;
2683 case PR_GET_TAGGED_ADDR_CTRL:
2684 if (arg2 || arg3 || arg4 || arg5)
2685 return -EINVAL;
2686 error = GET_TAGGED_ADDR_CTRL();
2687 break;
2688 case PR_SET_IO_FLUSHER:
2689 if (!capable(CAP_SYS_RESOURCE))
2690 return -EPERM;
2691
2692 if (arg3 || arg4 || arg5)
2693 return -EINVAL;
2694
2695 if (arg2 == 1)
2696 current->flags |= PR_IO_FLUSHER;
2697 else if (!arg2)
2698 current->flags &= ~PR_IO_FLUSHER;
2699 else
2700 return -EINVAL;
2701 break;
2702 case PR_GET_IO_FLUSHER:
2703 if (!capable(CAP_SYS_RESOURCE))
2704 return -EPERM;
2705
2706 if (arg2 || arg3 || arg4 || arg5)
2707 return -EINVAL;
2708
2709 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2710 break;
2711 case PR_SET_SYSCALL_USER_DISPATCH:
2712 error = set_syscall_user_dispatch(arg2, arg3, arg4,
2713 (char __user *) arg5);
2714 break;
2715#ifdef CONFIG_SCHED_CORE
2716 case PR_SCHED_CORE:
2717 error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2718 break;
2719#endif
2720 case PR_SET_MDWE:
2721 error = prctl_set_mdwe(arg2, arg3, arg4, arg5);
2722 break;
2723 case PR_GET_MDWE:
2724 error = prctl_get_mdwe(arg2, arg3, arg4, arg5);
2725 break;
2726 case PR_SET_VMA:
2727 error = prctl_set_vma(arg2, arg3, arg4, arg5);
2728 break;
2729 case PR_GET_AUXV:
2730 if (arg4 || arg5)
2731 return -EINVAL;
2732 error = prctl_get_auxv((void __user *)arg2, arg3);
2733 break;
2734#ifdef CONFIG_KSM
2735 case PR_SET_MEMORY_MERGE:
2736 if (arg3 || arg4 || arg5)
2737 return -EINVAL;
2738 if (mmap_write_lock_killable(me->mm))
2739 return -EINTR;
2740
2741 if (arg2)
2742 error = ksm_enable_merge_any(me->mm);
2743 else
2744 error = ksm_disable_merge_any(me->mm);
2745 mmap_write_unlock(me->mm);
2746 break;
2747 case PR_GET_MEMORY_MERGE:
2748 if (arg2 || arg3 || arg4 || arg5)
2749 return -EINVAL;
2750
2751 error = !!test_bit(MMF_VM_MERGE_ANY, &me->mm->flags);
2752 break;
2753#endif
2754 case PR_RISCV_V_SET_CONTROL:
2755 error = RISCV_V_SET_CONTROL(arg2);
2756 break;
2757 case PR_RISCV_V_GET_CONTROL:
2758 error = RISCV_V_GET_CONTROL();
2759 break;
2760 default:
2761 error = -EINVAL;
2762 break;
2763 }
2764 return error;
2765}
2766
2767SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2768 struct getcpu_cache __user *, unused)
2769{
2770 int err = 0;
2771 int cpu = raw_smp_processor_id();
2772
2773 if (cpup)
2774 err |= put_user(cpu, cpup);
2775 if (nodep)
2776 err |= put_user(cpu_to_node(cpu), nodep);
2777 return err ? -EFAULT : 0;
2778}
2779
2780/**
2781 * do_sysinfo - fill in sysinfo struct
2782 * @info: pointer to buffer to fill
2783 */
2784static int do_sysinfo(struct sysinfo *info)
2785{
2786 unsigned long mem_total, sav_total;
2787 unsigned int mem_unit, bitcount;
2788 struct timespec64 tp;
2789
2790 memset(info, 0, sizeof(struct sysinfo));
2791
2792 ktime_get_boottime_ts64(&tp);
2793 timens_add_boottime(&tp);
2794 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2795
2796 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2797
2798 info->procs = nr_threads;
2799
2800 si_meminfo(info);
2801 si_swapinfo(info);
2802
2803 /*
2804 * If the sum of all the available memory (i.e. ram + swap)
2805 * is less than can be stored in a 32 bit unsigned long then
2806 * we can be binary compatible with 2.2.x kernels. If not,
2807 * well, in that case 2.2.x was broken anyways...
2808 *
2809 * -Erik Andersen <andersee@debian.org>
2810 */
2811
2812 mem_total = info->totalram + info->totalswap;
2813 if (mem_total < info->totalram || mem_total < info->totalswap)
2814 goto out;
2815 bitcount = 0;
2816 mem_unit = info->mem_unit;
2817 while (mem_unit > 1) {
2818 bitcount++;
2819 mem_unit >>= 1;
2820 sav_total = mem_total;
2821 mem_total <<= 1;
2822 if (mem_total < sav_total)
2823 goto out;
2824 }
2825
2826 /*
2827 * If mem_total did not overflow, multiply all memory values by
2828 * info->mem_unit and set it to 1. This leaves things compatible
2829 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2830 * kernels...
2831 */
2832
2833 info->mem_unit = 1;
2834 info->totalram <<= bitcount;
2835 info->freeram <<= bitcount;
2836 info->sharedram <<= bitcount;
2837 info->bufferram <<= bitcount;
2838 info->totalswap <<= bitcount;
2839 info->freeswap <<= bitcount;
2840 info->totalhigh <<= bitcount;
2841 info->freehigh <<= bitcount;
2842
2843out:
2844 return 0;
2845}
2846
2847SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2848{
2849 struct sysinfo val;
2850
2851 do_sysinfo(&val);
2852
2853 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2854 return -EFAULT;
2855
2856 return 0;
2857}
2858
2859#ifdef CONFIG_COMPAT
2860struct compat_sysinfo {
2861 s32 uptime;
2862 u32 loads[3];
2863 u32 totalram;
2864 u32 freeram;
2865 u32 sharedram;
2866 u32 bufferram;
2867 u32 totalswap;
2868 u32 freeswap;
2869 u16 procs;
2870 u16 pad;
2871 u32 totalhigh;
2872 u32 freehigh;
2873 u32 mem_unit;
2874 char _f[20-2*sizeof(u32)-sizeof(int)];
2875};
2876
2877COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2878{
2879 struct sysinfo s;
2880 struct compat_sysinfo s_32;
2881
2882 do_sysinfo(&s);
2883
2884 /* Check to see if any memory value is too large for 32-bit and scale
2885 * down if needed
2886 */
2887 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2888 int bitcount = 0;
2889
2890 while (s.mem_unit < PAGE_SIZE) {
2891 s.mem_unit <<= 1;
2892 bitcount++;
2893 }
2894
2895 s.totalram >>= bitcount;
2896 s.freeram >>= bitcount;
2897 s.sharedram >>= bitcount;
2898 s.bufferram >>= bitcount;
2899 s.totalswap >>= bitcount;
2900 s.freeswap >>= bitcount;
2901 s.totalhigh >>= bitcount;
2902 s.freehigh >>= bitcount;
2903 }
2904
2905 memset(&s_32, 0, sizeof(s_32));
2906 s_32.uptime = s.uptime;
2907 s_32.loads[0] = s.loads[0];
2908 s_32.loads[1] = s.loads[1];
2909 s_32.loads[2] = s.loads[2];
2910 s_32.totalram = s.totalram;
2911 s_32.freeram = s.freeram;
2912 s_32.sharedram = s.sharedram;
2913 s_32.bufferram = s.bufferram;
2914 s_32.totalswap = s.totalswap;
2915 s_32.freeswap = s.freeswap;
2916 s_32.procs = s.procs;
2917 s_32.totalhigh = s.totalhigh;
2918 s_32.freehigh = s.freehigh;
2919 s_32.mem_unit = s.mem_unit;
2920 if (copy_to_user(info, &s_32, sizeof(s_32)))
2921 return -EFAULT;
2922 return 0;
2923}
2924#endif /* CONFIG_COMPAT */
1/*
2 * linux/kernel/sys.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/module.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/perf_event.h>
16#include <linux/resource.h>
17#include <linux/kernel.h>
18#include <linux/kexec.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/gfp.h>
39#include <linux/syscore_ops.h>
40#include <linux/version.h>
41#include <linux/ctype.h>
42
43#include <linux/compat.h>
44#include <linux/syscalls.h>
45#include <linux/kprobes.h>
46#include <linux/user_namespace.h>
47
48#include <linux/kmsg_dump.h>
49/* Move somewhere else to avoid recompiling? */
50#include <generated/utsrelease.h>
51
52#include <asm/uaccess.h>
53#include <asm/io.h>
54#include <asm/unistd.h>
55
56#ifndef SET_UNALIGN_CTL
57# define SET_UNALIGN_CTL(a,b) (-EINVAL)
58#endif
59#ifndef GET_UNALIGN_CTL
60# define GET_UNALIGN_CTL(a,b) (-EINVAL)
61#endif
62#ifndef SET_FPEMU_CTL
63# define SET_FPEMU_CTL(a,b) (-EINVAL)
64#endif
65#ifndef GET_FPEMU_CTL
66# define GET_FPEMU_CTL(a,b) (-EINVAL)
67#endif
68#ifndef SET_FPEXC_CTL
69# define SET_FPEXC_CTL(a,b) (-EINVAL)
70#endif
71#ifndef GET_FPEXC_CTL
72# define GET_FPEXC_CTL(a,b) (-EINVAL)
73#endif
74#ifndef GET_ENDIAN
75# define GET_ENDIAN(a,b) (-EINVAL)
76#endif
77#ifndef SET_ENDIAN
78# define SET_ENDIAN(a,b) (-EINVAL)
79#endif
80#ifndef GET_TSC_CTL
81# define GET_TSC_CTL(a) (-EINVAL)
82#endif
83#ifndef SET_TSC_CTL
84# define SET_TSC_CTL(a) (-EINVAL)
85#endif
86
87/*
88 * this is where the system-wide overflow UID and GID are defined, for
89 * architectures that now have 32-bit UID/GID but didn't in the past
90 */
91
92int overflowuid = DEFAULT_OVERFLOWUID;
93int overflowgid = DEFAULT_OVERFLOWGID;
94
95#ifdef CONFIG_UID16
96EXPORT_SYMBOL(overflowuid);
97EXPORT_SYMBOL(overflowgid);
98#endif
99
100/*
101 * the same as above, but for filesystems which can only store a 16-bit
102 * UID and GID. as such, this is needed on all architectures
103 */
104
105int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
106int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
107
108EXPORT_SYMBOL(fs_overflowuid);
109EXPORT_SYMBOL(fs_overflowgid);
110
111/*
112 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
113 */
114
115int C_A_D = 1;
116struct pid *cad_pid;
117EXPORT_SYMBOL(cad_pid);
118
119/*
120 * If set, this is used for preparing the system to power off.
121 */
122
123void (*pm_power_off_prepare)(void);
124
125/*
126 * Returns true if current's euid is same as p's uid or euid,
127 * or has CAP_SYS_NICE to p's user_ns.
128 *
129 * Called with rcu_read_lock, creds are safe
130 */
131static bool set_one_prio_perm(struct task_struct *p)
132{
133 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
134
135 if (pcred->user->user_ns == cred->user->user_ns &&
136 (pcred->uid == cred->euid ||
137 pcred->euid == cred->euid))
138 return true;
139 if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
140 return true;
141 return false;
142}
143
144/*
145 * set the priority of a task
146 * - the caller must hold the RCU read lock
147 */
148static int set_one_prio(struct task_struct *p, int niceval, int error)
149{
150 int no_nice;
151
152 if (!set_one_prio_perm(p)) {
153 error = -EPERM;
154 goto out;
155 }
156 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
157 error = -EACCES;
158 goto out;
159 }
160 no_nice = security_task_setnice(p, niceval);
161 if (no_nice) {
162 error = no_nice;
163 goto out;
164 }
165 if (error == -ESRCH)
166 error = 0;
167 set_user_nice(p, niceval);
168out:
169 return error;
170}
171
172SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
173{
174 struct task_struct *g, *p;
175 struct user_struct *user;
176 const struct cred *cred = current_cred();
177 int error = -EINVAL;
178 struct pid *pgrp;
179
180 if (which > PRIO_USER || which < PRIO_PROCESS)
181 goto out;
182
183 /* normalize: avoid signed division (rounding problems) */
184 error = -ESRCH;
185 if (niceval < -20)
186 niceval = -20;
187 if (niceval > 19)
188 niceval = 19;
189
190 rcu_read_lock();
191 read_lock(&tasklist_lock);
192 switch (which) {
193 case PRIO_PROCESS:
194 if (who)
195 p = find_task_by_vpid(who);
196 else
197 p = current;
198 if (p)
199 error = set_one_prio(p, niceval, error);
200 break;
201 case PRIO_PGRP:
202 if (who)
203 pgrp = find_vpid(who);
204 else
205 pgrp = task_pgrp(current);
206 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
207 error = set_one_prio(p, niceval, error);
208 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
209 break;
210 case PRIO_USER:
211 user = (struct user_struct *) cred->user;
212 if (!who)
213 who = cred->uid;
214 else if ((who != cred->uid) &&
215 !(user = find_user(who)))
216 goto out_unlock; /* No processes for this user */
217
218 do_each_thread(g, p) {
219 if (__task_cred(p)->uid == who)
220 error = set_one_prio(p, niceval, error);
221 } while_each_thread(g, p);
222 if (who != cred->uid)
223 free_uid(user); /* For find_user() */
224 break;
225 }
226out_unlock:
227 read_unlock(&tasklist_lock);
228 rcu_read_unlock();
229out:
230 return error;
231}
232
233/*
234 * Ugh. To avoid negative return values, "getpriority()" will
235 * not return the normal nice-value, but a negated value that
236 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
237 * to stay compatible.
238 */
239SYSCALL_DEFINE2(getpriority, int, which, int, who)
240{
241 struct task_struct *g, *p;
242 struct user_struct *user;
243 const struct cred *cred = current_cred();
244 long niceval, retval = -ESRCH;
245 struct pid *pgrp;
246
247 if (which > PRIO_USER || which < PRIO_PROCESS)
248 return -EINVAL;
249
250 rcu_read_lock();
251 read_lock(&tasklist_lock);
252 switch (which) {
253 case PRIO_PROCESS:
254 if (who)
255 p = find_task_by_vpid(who);
256 else
257 p = current;
258 if (p) {
259 niceval = 20 - task_nice(p);
260 if (niceval > retval)
261 retval = niceval;
262 }
263 break;
264 case PRIO_PGRP:
265 if (who)
266 pgrp = find_vpid(who);
267 else
268 pgrp = task_pgrp(current);
269 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
270 niceval = 20 - task_nice(p);
271 if (niceval > retval)
272 retval = niceval;
273 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
274 break;
275 case PRIO_USER:
276 user = (struct user_struct *) cred->user;
277 if (!who)
278 who = cred->uid;
279 else if ((who != cred->uid) &&
280 !(user = find_user(who)))
281 goto out_unlock; /* No processes for this user */
282
283 do_each_thread(g, p) {
284 if (__task_cred(p)->uid == who) {
285 niceval = 20 - task_nice(p);
286 if (niceval > retval)
287 retval = niceval;
288 }
289 } while_each_thread(g, p);
290 if (who != cred->uid)
291 free_uid(user); /* for find_user() */
292 break;
293 }
294out_unlock:
295 read_unlock(&tasklist_lock);
296 rcu_read_unlock();
297
298 return retval;
299}
300
301/**
302 * emergency_restart - reboot the system
303 *
304 * Without shutting down any hardware or taking any locks
305 * reboot the system. This is called when we know we are in
306 * trouble so this is our best effort to reboot. This is
307 * safe to call in interrupt context.
308 */
309void emergency_restart(void)
310{
311 kmsg_dump(KMSG_DUMP_EMERG);
312 machine_emergency_restart();
313}
314EXPORT_SYMBOL_GPL(emergency_restart);
315
316void kernel_restart_prepare(char *cmd)
317{
318 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
319 system_state = SYSTEM_RESTART;
320 usermodehelper_disable();
321 device_shutdown();
322 syscore_shutdown();
323}
324
325/**
326 * register_reboot_notifier - Register function to be called at reboot time
327 * @nb: Info about notifier function to be called
328 *
329 * Registers a function with the list of functions
330 * to be called at reboot time.
331 *
332 * Currently always returns zero, as blocking_notifier_chain_register()
333 * always returns zero.
334 */
335int register_reboot_notifier(struct notifier_block *nb)
336{
337 return blocking_notifier_chain_register(&reboot_notifier_list, nb);
338}
339EXPORT_SYMBOL(register_reboot_notifier);
340
341/**
342 * unregister_reboot_notifier - Unregister previously registered reboot notifier
343 * @nb: Hook to be unregistered
344 *
345 * Unregisters a previously registered reboot
346 * notifier function.
347 *
348 * Returns zero on success, or %-ENOENT on failure.
349 */
350int unregister_reboot_notifier(struct notifier_block *nb)
351{
352 return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
353}
354EXPORT_SYMBOL(unregister_reboot_notifier);
355
356/**
357 * kernel_restart - reboot the system
358 * @cmd: pointer to buffer containing command to execute for restart
359 * or %NULL
360 *
361 * Shutdown everything and perform a clean reboot.
362 * This is not safe to call in interrupt context.
363 */
364void kernel_restart(char *cmd)
365{
366 kernel_restart_prepare(cmd);
367 if (!cmd)
368 printk(KERN_EMERG "Restarting system.\n");
369 else
370 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
371 kmsg_dump(KMSG_DUMP_RESTART);
372 machine_restart(cmd);
373}
374EXPORT_SYMBOL_GPL(kernel_restart);
375
376static void kernel_shutdown_prepare(enum system_states state)
377{
378 blocking_notifier_call_chain(&reboot_notifier_list,
379 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
380 system_state = state;
381 usermodehelper_disable();
382 device_shutdown();
383}
384/**
385 * kernel_halt - halt the system
386 *
387 * Shutdown everything and perform a clean system halt.
388 */
389void kernel_halt(void)
390{
391 kernel_shutdown_prepare(SYSTEM_HALT);
392 syscore_shutdown();
393 printk(KERN_EMERG "System halted.\n");
394 kmsg_dump(KMSG_DUMP_HALT);
395 machine_halt();
396}
397
398EXPORT_SYMBOL_GPL(kernel_halt);
399
400/**
401 * kernel_power_off - power_off the system
402 *
403 * Shutdown everything and perform a clean system power_off.
404 */
405void kernel_power_off(void)
406{
407 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
408 if (pm_power_off_prepare)
409 pm_power_off_prepare();
410 disable_nonboot_cpus();
411 syscore_shutdown();
412 printk(KERN_EMERG "Power down.\n");
413 kmsg_dump(KMSG_DUMP_POWEROFF);
414 machine_power_off();
415}
416EXPORT_SYMBOL_GPL(kernel_power_off);
417
418static DEFINE_MUTEX(reboot_mutex);
419
420/*
421 * Reboot system call: for obvious reasons only root may call it,
422 * and even root needs to set up some magic numbers in the registers
423 * so that some mistake won't make this reboot the whole machine.
424 * You can also set the meaning of the ctrl-alt-del-key here.
425 *
426 * reboot doesn't sync: do that yourself before calling this.
427 */
428SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
429 void __user *, arg)
430{
431 char buffer[256];
432 int ret = 0;
433
434 /* We only trust the superuser with rebooting the system. */
435 if (!capable(CAP_SYS_BOOT))
436 return -EPERM;
437
438 /* For safety, we require "magic" arguments. */
439 if (magic1 != LINUX_REBOOT_MAGIC1 ||
440 (magic2 != LINUX_REBOOT_MAGIC2 &&
441 magic2 != LINUX_REBOOT_MAGIC2A &&
442 magic2 != LINUX_REBOOT_MAGIC2B &&
443 magic2 != LINUX_REBOOT_MAGIC2C))
444 return -EINVAL;
445
446 /* Instead of trying to make the power_off code look like
447 * halt when pm_power_off is not set do it the easy way.
448 */
449 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
450 cmd = LINUX_REBOOT_CMD_HALT;
451
452 mutex_lock(&reboot_mutex);
453 switch (cmd) {
454 case LINUX_REBOOT_CMD_RESTART:
455 kernel_restart(NULL);
456 break;
457
458 case LINUX_REBOOT_CMD_CAD_ON:
459 C_A_D = 1;
460 break;
461
462 case LINUX_REBOOT_CMD_CAD_OFF:
463 C_A_D = 0;
464 break;
465
466 case LINUX_REBOOT_CMD_HALT:
467 kernel_halt();
468 do_exit(0);
469 panic("cannot halt");
470
471 case LINUX_REBOOT_CMD_POWER_OFF:
472 kernel_power_off();
473 do_exit(0);
474 break;
475
476 case LINUX_REBOOT_CMD_RESTART2:
477 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
478 ret = -EFAULT;
479 break;
480 }
481 buffer[sizeof(buffer) - 1] = '\0';
482
483 kernel_restart(buffer);
484 break;
485
486#ifdef CONFIG_KEXEC
487 case LINUX_REBOOT_CMD_KEXEC:
488 ret = kernel_kexec();
489 break;
490#endif
491
492#ifdef CONFIG_HIBERNATION
493 case LINUX_REBOOT_CMD_SW_SUSPEND:
494 ret = hibernate();
495 break;
496#endif
497
498 default:
499 ret = -EINVAL;
500 break;
501 }
502 mutex_unlock(&reboot_mutex);
503 return ret;
504}
505
506static void deferred_cad(struct work_struct *dummy)
507{
508 kernel_restart(NULL);
509}
510
511/*
512 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
513 * As it's called within an interrupt, it may NOT sync: the only choice
514 * is whether to reboot at once, or just ignore the ctrl-alt-del.
515 */
516void ctrl_alt_del(void)
517{
518 static DECLARE_WORK(cad_work, deferred_cad);
519
520 if (C_A_D)
521 schedule_work(&cad_work);
522 else
523 kill_cad_pid(SIGINT, 1);
524}
525
526/*
527 * Unprivileged users may change the real gid to the effective gid
528 * or vice versa. (BSD-style)
529 *
530 * If you set the real gid at all, or set the effective gid to a value not
531 * equal to the real gid, then the saved gid is set to the new effective gid.
532 *
533 * This makes it possible for a setgid program to completely drop its
534 * privileges, which is often a useful assertion to make when you are doing
535 * a security audit over a program.
536 *
537 * The general idea is that a program which uses just setregid() will be
538 * 100% compatible with BSD. A program which uses just setgid() will be
539 * 100% compatible with POSIX with saved IDs.
540 *
541 * SMP: There are not races, the GIDs are checked only by filesystem
542 * operations (as far as semantic preservation is concerned).
543 */
544SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
545{
546 const struct cred *old;
547 struct cred *new;
548 int retval;
549
550 new = prepare_creds();
551 if (!new)
552 return -ENOMEM;
553 old = current_cred();
554
555 retval = -EPERM;
556 if (rgid != (gid_t) -1) {
557 if (old->gid == rgid ||
558 old->egid == rgid ||
559 nsown_capable(CAP_SETGID))
560 new->gid = rgid;
561 else
562 goto error;
563 }
564 if (egid != (gid_t) -1) {
565 if (old->gid == egid ||
566 old->egid == egid ||
567 old->sgid == egid ||
568 nsown_capable(CAP_SETGID))
569 new->egid = egid;
570 else
571 goto error;
572 }
573
574 if (rgid != (gid_t) -1 ||
575 (egid != (gid_t) -1 && egid != old->gid))
576 new->sgid = new->egid;
577 new->fsgid = new->egid;
578
579 return commit_creds(new);
580
581error:
582 abort_creds(new);
583 return retval;
584}
585
586/*
587 * setgid() is implemented like SysV w/ SAVED_IDS
588 *
589 * SMP: Same implicit races as above.
590 */
591SYSCALL_DEFINE1(setgid, gid_t, gid)
592{
593 const struct cred *old;
594 struct cred *new;
595 int retval;
596
597 new = prepare_creds();
598 if (!new)
599 return -ENOMEM;
600 old = current_cred();
601
602 retval = -EPERM;
603 if (nsown_capable(CAP_SETGID))
604 new->gid = new->egid = new->sgid = new->fsgid = gid;
605 else if (gid == old->gid || gid == old->sgid)
606 new->egid = new->fsgid = gid;
607 else
608 goto error;
609
610 return commit_creds(new);
611
612error:
613 abort_creds(new);
614 return retval;
615}
616
617/*
618 * change the user struct in a credentials set to match the new UID
619 */
620static int set_user(struct cred *new)
621{
622 struct user_struct *new_user;
623
624 new_user = alloc_uid(current_user_ns(), new->uid);
625 if (!new_user)
626 return -EAGAIN;
627
628 /*
629 * We don't fail in case of NPROC limit excess here because too many
630 * poorly written programs don't check set*uid() return code, assuming
631 * it never fails if called by root. We may still enforce NPROC limit
632 * for programs doing set*uid()+execve() by harmlessly deferring the
633 * failure to the execve() stage.
634 */
635 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
636 new_user != INIT_USER)
637 current->flags |= PF_NPROC_EXCEEDED;
638 else
639 current->flags &= ~PF_NPROC_EXCEEDED;
640
641 free_uid(new->user);
642 new->user = new_user;
643 return 0;
644}
645
646/*
647 * Unprivileged users may change the real uid to the effective uid
648 * or vice versa. (BSD-style)
649 *
650 * If you set the real uid at all, or set the effective uid to a value not
651 * equal to the real uid, then the saved uid is set to the new effective uid.
652 *
653 * This makes it possible for a setuid program to completely drop its
654 * privileges, which is often a useful assertion to make when you are doing
655 * a security audit over a program.
656 *
657 * The general idea is that a program which uses just setreuid() will be
658 * 100% compatible with BSD. A program which uses just setuid() will be
659 * 100% compatible with POSIX with saved IDs.
660 */
661SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
662{
663 const struct cred *old;
664 struct cred *new;
665 int retval;
666
667 new = prepare_creds();
668 if (!new)
669 return -ENOMEM;
670 old = current_cred();
671
672 retval = -EPERM;
673 if (ruid != (uid_t) -1) {
674 new->uid = ruid;
675 if (old->uid != ruid &&
676 old->euid != ruid &&
677 !nsown_capable(CAP_SETUID))
678 goto error;
679 }
680
681 if (euid != (uid_t) -1) {
682 new->euid = euid;
683 if (old->uid != euid &&
684 old->euid != euid &&
685 old->suid != euid &&
686 !nsown_capable(CAP_SETUID))
687 goto error;
688 }
689
690 if (new->uid != old->uid) {
691 retval = set_user(new);
692 if (retval < 0)
693 goto error;
694 }
695 if (ruid != (uid_t) -1 ||
696 (euid != (uid_t) -1 && euid != old->uid))
697 new->suid = new->euid;
698 new->fsuid = new->euid;
699
700 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
701 if (retval < 0)
702 goto error;
703
704 return commit_creds(new);
705
706error:
707 abort_creds(new);
708 return retval;
709}
710
711/*
712 * setuid() is implemented like SysV with SAVED_IDS
713 *
714 * Note that SAVED_ID's is deficient in that a setuid root program
715 * like sendmail, for example, cannot set its uid to be a normal
716 * user and then switch back, because if you're root, setuid() sets
717 * the saved uid too. If you don't like this, blame the bright people
718 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
719 * will allow a root program to temporarily drop privileges and be able to
720 * regain them by swapping the real and effective uid.
721 */
722SYSCALL_DEFINE1(setuid, uid_t, uid)
723{
724 const struct cred *old;
725 struct cred *new;
726 int retval;
727
728 new = prepare_creds();
729 if (!new)
730 return -ENOMEM;
731 old = current_cred();
732
733 retval = -EPERM;
734 if (nsown_capable(CAP_SETUID)) {
735 new->suid = new->uid = uid;
736 if (uid != old->uid) {
737 retval = set_user(new);
738 if (retval < 0)
739 goto error;
740 }
741 } else if (uid != old->uid && uid != new->suid) {
742 goto error;
743 }
744
745 new->fsuid = new->euid = uid;
746
747 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
748 if (retval < 0)
749 goto error;
750
751 return commit_creds(new);
752
753error:
754 abort_creds(new);
755 return retval;
756}
757
758
759/*
760 * This function implements a generic ability to update ruid, euid,
761 * and suid. This allows you to implement the 4.4 compatible seteuid().
762 */
763SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
764{
765 const struct cred *old;
766 struct cred *new;
767 int retval;
768
769 new = prepare_creds();
770 if (!new)
771 return -ENOMEM;
772
773 old = current_cred();
774
775 retval = -EPERM;
776 if (!nsown_capable(CAP_SETUID)) {
777 if (ruid != (uid_t) -1 && ruid != old->uid &&
778 ruid != old->euid && ruid != old->suid)
779 goto error;
780 if (euid != (uid_t) -1 && euid != old->uid &&
781 euid != old->euid && euid != old->suid)
782 goto error;
783 if (suid != (uid_t) -1 && suid != old->uid &&
784 suid != old->euid && suid != old->suid)
785 goto error;
786 }
787
788 if (ruid != (uid_t) -1) {
789 new->uid = ruid;
790 if (ruid != old->uid) {
791 retval = set_user(new);
792 if (retval < 0)
793 goto error;
794 }
795 }
796 if (euid != (uid_t) -1)
797 new->euid = euid;
798 if (suid != (uid_t) -1)
799 new->suid = suid;
800 new->fsuid = new->euid;
801
802 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
803 if (retval < 0)
804 goto error;
805
806 return commit_creds(new);
807
808error:
809 abort_creds(new);
810 return retval;
811}
812
813SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
814{
815 const struct cred *cred = current_cred();
816 int retval;
817
818 if (!(retval = put_user(cred->uid, ruid)) &&
819 !(retval = put_user(cred->euid, euid)))
820 retval = put_user(cred->suid, suid);
821
822 return retval;
823}
824
825/*
826 * Same as above, but for rgid, egid, sgid.
827 */
828SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
829{
830 const struct cred *old;
831 struct cred *new;
832 int retval;
833
834 new = prepare_creds();
835 if (!new)
836 return -ENOMEM;
837 old = current_cred();
838
839 retval = -EPERM;
840 if (!nsown_capable(CAP_SETGID)) {
841 if (rgid != (gid_t) -1 && rgid != old->gid &&
842 rgid != old->egid && rgid != old->sgid)
843 goto error;
844 if (egid != (gid_t) -1 && egid != old->gid &&
845 egid != old->egid && egid != old->sgid)
846 goto error;
847 if (sgid != (gid_t) -1 && sgid != old->gid &&
848 sgid != old->egid && sgid != old->sgid)
849 goto error;
850 }
851
852 if (rgid != (gid_t) -1)
853 new->gid = rgid;
854 if (egid != (gid_t) -1)
855 new->egid = egid;
856 if (sgid != (gid_t) -1)
857 new->sgid = sgid;
858 new->fsgid = new->egid;
859
860 return commit_creds(new);
861
862error:
863 abort_creds(new);
864 return retval;
865}
866
867SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
868{
869 const struct cred *cred = current_cred();
870 int retval;
871
872 if (!(retval = put_user(cred->gid, rgid)) &&
873 !(retval = put_user(cred->egid, egid)))
874 retval = put_user(cred->sgid, sgid);
875
876 return retval;
877}
878
879
880/*
881 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
882 * is used for "access()" and for the NFS daemon (letting nfsd stay at
883 * whatever uid it wants to). It normally shadows "euid", except when
884 * explicitly set by setfsuid() or for access..
885 */
886SYSCALL_DEFINE1(setfsuid, uid_t, uid)
887{
888 const struct cred *old;
889 struct cred *new;
890 uid_t old_fsuid;
891
892 new = prepare_creds();
893 if (!new)
894 return current_fsuid();
895 old = current_cred();
896 old_fsuid = old->fsuid;
897
898 if (uid == old->uid || uid == old->euid ||
899 uid == old->suid || uid == old->fsuid ||
900 nsown_capable(CAP_SETUID)) {
901 if (uid != old_fsuid) {
902 new->fsuid = uid;
903 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
904 goto change_okay;
905 }
906 }
907
908 abort_creds(new);
909 return old_fsuid;
910
911change_okay:
912 commit_creds(new);
913 return old_fsuid;
914}
915
916/*
917 * Samma på svenska..
918 */
919SYSCALL_DEFINE1(setfsgid, gid_t, gid)
920{
921 const struct cred *old;
922 struct cred *new;
923 gid_t old_fsgid;
924
925 new = prepare_creds();
926 if (!new)
927 return current_fsgid();
928 old = current_cred();
929 old_fsgid = old->fsgid;
930
931 if (gid == old->gid || gid == old->egid ||
932 gid == old->sgid || gid == old->fsgid ||
933 nsown_capable(CAP_SETGID)) {
934 if (gid != old_fsgid) {
935 new->fsgid = gid;
936 goto change_okay;
937 }
938 }
939
940 abort_creds(new);
941 return old_fsgid;
942
943change_okay:
944 commit_creds(new);
945 return old_fsgid;
946}
947
948void do_sys_times(struct tms *tms)
949{
950 cputime_t tgutime, tgstime, cutime, cstime;
951
952 spin_lock_irq(¤t->sighand->siglock);
953 thread_group_times(current, &tgutime, &tgstime);
954 cutime = current->signal->cutime;
955 cstime = current->signal->cstime;
956 spin_unlock_irq(¤t->sighand->siglock);
957 tms->tms_utime = cputime_to_clock_t(tgutime);
958 tms->tms_stime = cputime_to_clock_t(tgstime);
959 tms->tms_cutime = cputime_to_clock_t(cutime);
960 tms->tms_cstime = cputime_to_clock_t(cstime);
961}
962
963SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
964{
965 if (tbuf) {
966 struct tms tmp;
967
968 do_sys_times(&tmp);
969 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
970 return -EFAULT;
971 }
972 force_successful_syscall_return();
973 return (long) jiffies_64_to_clock_t(get_jiffies_64());
974}
975
976/*
977 * This needs some heavy checking ...
978 * I just haven't the stomach for it. I also don't fully
979 * understand sessions/pgrp etc. Let somebody who does explain it.
980 *
981 * OK, I think I have the protection semantics right.... this is really
982 * only important on a multi-user system anyway, to make sure one user
983 * can't send a signal to a process owned by another. -TYT, 12/12/91
984 *
985 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
986 * LBT 04.03.94
987 */
988SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
989{
990 struct task_struct *p;
991 struct task_struct *group_leader = current->group_leader;
992 struct pid *pgrp;
993 int err;
994
995 if (!pid)
996 pid = task_pid_vnr(group_leader);
997 if (!pgid)
998 pgid = pid;
999 if (pgid < 0)
1000 return -EINVAL;
1001 rcu_read_lock();
1002
1003 /* From this point forward we keep holding onto the tasklist lock
1004 * so that our parent does not change from under us. -DaveM
1005 */
1006 write_lock_irq(&tasklist_lock);
1007
1008 err = -ESRCH;
1009 p = find_task_by_vpid(pid);
1010 if (!p)
1011 goto out;
1012
1013 err = -EINVAL;
1014 if (!thread_group_leader(p))
1015 goto out;
1016
1017 if (same_thread_group(p->real_parent, group_leader)) {
1018 err = -EPERM;
1019 if (task_session(p) != task_session(group_leader))
1020 goto out;
1021 err = -EACCES;
1022 if (p->did_exec)
1023 goto out;
1024 } else {
1025 err = -ESRCH;
1026 if (p != group_leader)
1027 goto out;
1028 }
1029
1030 err = -EPERM;
1031 if (p->signal->leader)
1032 goto out;
1033
1034 pgrp = task_pid(p);
1035 if (pgid != pid) {
1036 struct task_struct *g;
1037
1038 pgrp = find_vpid(pgid);
1039 g = pid_task(pgrp, PIDTYPE_PGID);
1040 if (!g || task_session(g) != task_session(group_leader))
1041 goto out;
1042 }
1043
1044 err = security_task_setpgid(p, pgid);
1045 if (err)
1046 goto out;
1047
1048 if (task_pgrp(p) != pgrp)
1049 change_pid(p, PIDTYPE_PGID, pgrp);
1050
1051 err = 0;
1052out:
1053 /* All paths lead to here, thus we are safe. -DaveM */
1054 write_unlock_irq(&tasklist_lock);
1055 rcu_read_unlock();
1056 return err;
1057}
1058
1059SYSCALL_DEFINE1(getpgid, pid_t, pid)
1060{
1061 struct task_struct *p;
1062 struct pid *grp;
1063 int retval;
1064
1065 rcu_read_lock();
1066 if (!pid)
1067 grp = task_pgrp(current);
1068 else {
1069 retval = -ESRCH;
1070 p = find_task_by_vpid(pid);
1071 if (!p)
1072 goto out;
1073 grp = task_pgrp(p);
1074 if (!grp)
1075 goto out;
1076
1077 retval = security_task_getpgid(p);
1078 if (retval)
1079 goto out;
1080 }
1081 retval = pid_vnr(grp);
1082out:
1083 rcu_read_unlock();
1084 return retval;
1085}
1086
1087#ifdef __ARCH_WANT_SYS_GETPGRP
1088
1089SYSCALL_DEFINE0(getpgrp)
1090{
1091 return sys_getpgid(0);
1092}
1093
1094#endif
1095
1096SYSCALL_DEFINE1(getsid, pid_t, pid)
1097{
1098 struct task_struct *p;
1099 struct pid *sid;
1100 int retval;
1101
1102 rcu_read_lock();
1103 if (!pid)
1104 sid = task_session(current);
1105 else {
1106 retval = -ESRCH;
1107 p = find_task_by_vpid(pid);
1108 if (!p)
1109 goto out;
1110 sid = task_session(p);
1111 if (!sid)
1112 goto out;
1113
1114 retval = security_task_getsid(p);
1115 if (retval)
1116 goto out;
1117 }
1118 retval = pid_vnr(sid);
1119out:
1120 rcu_read_unlock();
1121 return retval;
1122}
1123
1124SYSCALL_DEFINE0(setsid)
1125{
1126 struct task_struct *group_leader = current->group_leader;
1127 struct pid *sid = task_pid(group_leader);
1128 pid_t session = pid_vnr(sid);
1129 int err = -EPERM;
1130
1131 write_lock_irq(&tasklist_lock);
1132 /* Fail if I am already a session leader */
1133 if (group_leader->signal->leader)
1134 goto out;
1135
1136 /* Fail if a process group id already exists that equals the
1137 * proposed session id.
1138 */
1139 if (pid_task(sid, PIDTYPE_PGID))
1140 goto out;
1141
1142 group_leader->signal->leader = 1;
1143 __set_special_pids(sid);
1144
1145 proc_clear_tty(group_leader);
1146
1147 err = session;
1148out:
1149 write_unlock_irq(&tasklist_lock);
1150 if (err > 0) {
1151 proc_sid_connector(group_leader);
1152 sched_autogroup_create_attach(group_leader);
1153 }
1154 return err;
1155}
1156
1157DECLARE_RWSEM(uts_sem);
1158
1159#ifdef COMPAT_UTS_MACHINE
1160#define override_architecture(name) \
1161 (personality(current->personality) == PER_LINUX32 && \
1162 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1163 sizeof(COMPAT_UTS_MACHINE)))
1164#else
1165#define override_architecture(name) 0
1166#endif
1167
1168/*
1169 * Work around broken programs that cannot handle "Linux 3.0".
1170 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1171 */
1172static int override_release(char __user *release, int len)
1173{
1174 int ret = 0;
1175 char buf[65];
1176
1177 if (current->personality & UNAME26) {
1178 char *rest = UTS_RELEASE;
1179 int ndots = 0;
1180 unsigned v;
1181
1182 while (*rest) {
1183 if (*rest == '.' && ++ndots >= 3)
1184 break;
1185 if (!isdigit(*rest) && *rest != '.')
1186 break;
1187 rest++;
1188 }
1189 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1190 snprintf(buf, len, "2.6.%u%s", v, rest);
1191 ret = copy_to_user(release, buf, len);
1192 }
1193 return ret;
1194}
1195
1196SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1197{
1198 int errno = 0;
1199
1200 down_read(&uts_sem);
1201 if (copy_to_user(name, utsname(), sizeof *name))
1202 errno = -EFAULT;
1203 up_read(&uts_sem);
1204
1205 if (!errno && override_release(name->release, sizeof(name->release)))
1206 errno = -EFAULT;
1207 if (!errno && override_architecture(name))
1208 errno = -EFAULT;
1209 return errno;
1210}
1211
1212#ifdef __ARCH_WANT_SYS_OLD_UNAME
1213/*
1214 * Old cruft
1215 */
1216SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1217{
1218 int error = 0;
1219
1220 if (!name)
1221 return -EFAULT;
1222
1223 down_read(&uts_sem);
1224 if (copy_to_user(name, utsname(), sizeof(*name)))
1225 error = -EFAULT;
1226 up_read(&uts_sem);
1227
1228 if (!error && override_release(name->release, sizeof(name->release)))
1229 error = -EFAULT;
1230 if (!error && override_architecture(name))
1231 error = -EFAULT;
1232 return error;
1233}
1234
1235SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1236{
1237 int error;
1238
1239 if (!name)
1240 return -EFAULT;
1241 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1242 return -EFAULT;
1243
1244 down_read(&uts_sem);
1245 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1246 __OLD_UTS_LEN);
1247 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1248 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1249 __OLD_UTS_LEN);
1250 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1251 error |= __copy_to_user(&name->release, &utsname()->release,
1252 __OLD_UTS_LEN);
1253 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1254 error |= __copy_to_user(&name->version, &utsname()->version,
1255 __OLD_UTS_LEN);
1256 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1257 error |= __copy_to_user(&name->machine, &utsname()->machine,
1258 __OLD_UTS_LEN);
1259 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1260 up_read(&uts_sem);
1261
1262 if (!error && override_architecture(name))
1263 error = -EFAULT;
1264 if (!error && override_release(name->release, sizeof(name->release)))
1265 error = -EFAULT;
1266 return error ? -EFAULT : 0;
1267}
1268#endif
1269
1270SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1271{
1272 int errno;
1273 char tmp[__NEW_UTS_LEN];
1274
1275 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1276 return -EPERM;
1277
1278 if (len < 0 || len > __NEW_UTS_LEN)
1279 return -EINVAL;
1280 down_write(&uts_sem);
1281 errno = -EFAULT;
1282 if (!copy_from_user(tmp, name, len)) {
1283 struct new_utsname *u = utsname();
1284
1285 memcpy(u->nodename, tmp, len);
1286 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1287 errno = 0;
1288 }
1289 up_write(&uts_sem);
1290 return errno;
1291}
1292
1293#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1294
1295SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1296{
1297 int i, errno;
1298 struct new_utsname *u;
1299
1300 if (len < 0)
1301 return -EINVAL;
1302 down_read(&uts_sem);
1303 u = utsname();
1304 i = 1 + strlen(u->nodename);
1305 if (i > len)
1306 i = len;
1307 errno = 0;
1308 if (copy_to_user(name, u->nodename, i))
1309 errno = -EFAULT;
1310 up_read(&uts_sem);
1311 return errno;
1312}
1313
1314#endif
1315
1316/*
1317 * Only setdomainname; getdomainname can be implemented by calling
1318 * uname()
1319 */
1320SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1321{
1322 int errno;
1323 char tmp[__NEW_UTS_LEN];
1324
1325 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1326 return -EPERM;
1327 if (len < 0 || len > __NEW_UTS_LEN)
1328 return -EINVAL;
1329
1330 down_write(&uts_sem);
1331 errno = -EFAULT;
1332 if (!copy_from_user(tmp, name, len)) {
1333 struct new_utsname *u = utsname();
1334
1335 memcpy(u->domainname, tmp, len);
1336 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1337 errno = 0;
1338 }
1339 up_write(&uts_sem);
1340 return errno;
1341}
1342
1343SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1344{
1345 struct rlimit value;
1346 int ret;
1347
1348 ret = do_prlimit(current, resource, NULL, &value);
1349 if (!ret)
1350 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1351
1352 return ret;
1353}
1354
1355#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1356
1357/*
1358 * Back compatibility for getrlimit. Needed for some apps.
1359 */
1360
1361SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1362 struct rlimit __user *, rlim)
1363{
1364 struct rlimit x;
1365 if (resource >= RLIM_NLIMITS)
1366 return -EINVAL;
1367
1368 task_lock(current->group_leader);
1369 x = current->signal->rlim[resource];
1370 task_unlock(current->group_leader);
1371 if (x.rlim_cur > 0x7FFFFFFF)
1372 x.rlim_cur = 0x7FFFFFFF;
1373 if (x.rlim_max > 0x7FFFFFFF)
1374 x.rlim_max = 0x7FFFFFFF;
1375 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1376}
1377
1378#endif
1379
1380static inline bool rlim64_is_infinity(__u64 rlim64)
1381{
1382#if BITS_PER_LONG < 64
1383 return rlim64 >= ULONG_MAX;
1384#else
1385 return rlim64 == RLIM64_INFINITY;
1386#endif
1387}
1388
1389static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1390{
1391 if (rlim->rlim_cur == RLIM_INFINITY)
1392 rlim64->rlim_cur = RLIM64_INFINITY;
1393 else
1394 rlim64->rlim_cur = rlim->rlim_cur;
1395 if (rlim->rlim_max == RLIM_INFINITY)
1396 rlim64->rlim_max = RLIM64_INFINITY;
1397 else
1398 rlim64->rlim_max = rlim->rlim_max;
1399}
1400
1401static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1402{
1403 if (rlim64_is_infinity(rlim64->rlim_cur))
1404 rlim->rlim_cur = RLIM_INFINITY;
1405 else
1406 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1407 if (rlim64_is_infinity(rlim64->rlim_max))
1408 rlim->rlim_max = RLIM_INFINITY;
1409 else
1410 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1411}
1412
1413/* make sure you are allowed to change @tsk limits before calling this */
1414int do_prlimit(struct task_struct *tsk, unsigned int resource,
1415 struct rlimit *new_rlim, struct rlimit *old_rlim)
1416{
1417 struct rlimit *rlim;
1418 int retval = 0;
1419
1420 if (resource >= RLIM_NLIMITS)
1421 return -EINVAL;
1422 if (new_rlim) {
1423 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1424 return -EINVAL;
1425 if (resource == RLIMIT_NOFILE &&
1426 new_rlim->rlim_max > sysctl_nr_open)
1427 return -EPERM;
1428 }
1429
1430 /* protect tsk->signal and tsk->sighand from disappearing */
1431 read_lock(&tasklist_lock);
1432 if (!tsk->sighand) {
1433 retval = -ESRCH;
1434 goto out;
1435 }
1436
1437 rlim = tsk->signal->rlim + resource;
1438 task_lock(tsk->group_leader);
1439 if (new_rlim) {
1440 /* Keep the capable check against init_user_ns until
1441 cgroups can contain all limits */
1442 if (new_rlim->rlim_max > rlim->rlim_max &&
1443 !capable(CAP_SYS_RESOURCE))
1444 retval = -EPERM;
1445 if (!retval)
1446 retval = security_task_setrlimit(tsk->group_leader,
1447 resource, new_rlim);
1448 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1449 /*
1450 * The caller is asking for an immediate RLIMIT_CPU
1451 * expiry. But we use the zero value to mean "it was
1452 * never set". So let's cheat and make it one second
1453 * instead
1454 */
1455 new_rlim->rlim_cur = 1;
1456 }
1457 }
1458 if (!retval) {
1459 if (old_rlim)
1460 *old_rlim = *rlim;
1461 if (new_rlim)
1462 *rlim = *new_rlim;
1463 }
1464 task_unlock(tsk->group_leader);
1465
1466 /*
1467 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1468 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1469 * very long-standing error, and fixing it now risks breakage of
1470 * applications, so we live with it
1471 */
1472 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1473 new_rlim->rlim_cur != RLIM_INFINITY)
1474 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1475out:
1476 read_unlock(&tasklist_lock);
1477 return retval;
1478}
1479
1480/* rcu lock must be held */
1481static int check_prlimit_permission(struct task_struct *task)
1482{
1483 const struct cred *cred = current_cred(), *tcred;
1484
1485 if (current == task)
1486 return 0;
1487
1488 tcred = __task_cred(task);
1489 if (cred->user->user_ns == tcred->user->user_ns &&
1490 (cred->uid == tcred->euid &&
1491 cred->uid == tcred->suid &&
1492 cred->uid == tcred->uid &&
1493 cred->gid == tcred->egid &&
1494 cred->gid == tcred->sgid &&
1495 cred->gid == tcred->gid))
1496 return 0;
1497 if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1498 return 0;
1499
1500 return -EPERM;
1501}
1502
1503SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1504 const struct rlimit64 __user *, new_rlim,
1505 struct rlimit64 __user *, old_rlim)
1506{
1507 struct rlimit64 old64, new64;
1508 struct rlimit old, new;
1509 struct task_struct *tsk;
1510 int ret;
1511
1512 if (new_rlim) {
1513 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1514 return -EFAULT;
1515 rlim64_to_rlim(&new64, &new);
1516 }
1517
1518 rcu_read_lock();
1519 tsk = pid ? find_task_by_vpid(pid) : current;
1520 if (!tsk) {
1521 rcu_read_unlock();
1522 return -ESRCH;
1523 }
1524 ret = check_prlimit_permission(tsk);
1525 if (ret) {
1526 rcu_read_unlock();
1527 return ret;
1528 }
1529 get_task_struct(tsk);
1530 rcu_read_unlock();
1531
1532 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1533 old_rlim ? &old : NULL);
1534
1535 if (!ret && old_rlim) {
1536 rlim_to_rlim64(&old, &old64);
1537 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1538 ret = -EFAULT;
1539 }
1540
1541 put_task_struct(tsk);
1542 return ret;
1543}
1544
1545SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1546{
1547 struct rlimit new_rlim;
1548
1549 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1550 return -EFAULT;
1551 return do_prlimit(current, resource, &new_rlim, NULL);
1552}
1553
1554/*
1555 * It would make sense to put struct rusage in the task_struct,
1556 * except that would make the task_struct be *really big*. After
1557 * task_struct gets moved into malloc'ed memory, it would
1558 * make sense to do this. It will make moving the rest of the information
1559 * a lot simpler! (Which we're not doing right now because we're not
1560 * measuring them yet).
1561 *
1562 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1563 * races with threads incrementing their own counters. But since word
1564 * reads are atomic, we either get new values or old values and we don't
1565 * care which for the sums. We always take the siglock to protect reading
1566 * the c* fields from p->signal from races with exit.c updating those
1567 * fields when reaping, so a sample either gets all the additions of a
1568 * given child after it's reaped, or none so this sample is before reaping.
1569 *
1570 * Locking:
1571 * We need to take the siglock for CHILDEREN, SELF and BOTH
1572 * for the cases current multithreaded, non-current single threaded
1573 * non-current multithreaded. Thread traversal is now safe with
1574 * the siglock held.
1575 * Strictly speaking, we donot need to take the siglock if we are current and
1576 * single threaded, as no one else can take our signal_struct away, no one
1577 * else can reap the children to update signal->c* counters, and no one else
1578 * can race with the signal-> fields. If we do not take any lock, the
1579 * signal-> fields could be read out of order while another thread was just
1580 * exiting. So we should place a read memory barrier when we avoid the lock.
1581 * On the writer side, write memory barrier is implied in __exit_signal
1582 * as __exit_signal releases the siglock spinlock after updating the signal->
1583 * fields. But we don't do this yet to keep things simple.
1584 *
1585 */
1586
1587static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1588{
1589 r->ru_nvcsw += t->nvcsw;
1590 r->ru_nivcsw += t->nivcsw;
1591 r->ru_minflt += t->min_flt;
1592 r->ru_majflt += t->maj_flt;
1593 r->ru_inblock += task_io_get_inblock(t);
1594 r->ru_oublock += task_io_get_oublock(t);
1595}
1596
1597static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1598{
1599 struct task_struct *t;
1600 unsigned long flags;
1601 cputime_t tgutime, tgstime, utime, stime;
1602 unsigned long maxrss = 0;
1603
1604 memset((char *) r, 0, sizeof *r);
1605 utime = stime = cputime_zero;
1606
1607 if (who == RUSAGE_THREAD) {
1608 task_times(current, &utime, &stime);
1609 accumulate_thread_rusage(p, r);
1610 maxrss = p->signal->maxrss;
1611 goto out;
1612 }
1613
1614 if (!lock_task_sighand(p, &flags))
1615 return;
1616
1617 switch (who) {
1618 case RUSAGE_BOTH:
1619 case RUSAGE_CHILDREN:
1620 utime = p->signal->cutime;
1621 stime = p->signal->cstime;
1622 r->ru_nvcsw = p->signal->cnvcsw;
1623 r->ru_nivcsw = p->signal->cnivcsw;
1624 r->ru_minflt = p->signal->cmin_flt;
1625 r->ru_majflt = p->signal->cmaj_flt;
1626 r->ru_inblock = p->signal->cinblock;
1627 r->ru_oublock = p->signal->coublock;
1628 maxrss = p->signal->cmaxrss;
1629
1630 if (who == RUSAGE_CHILDREN)
1631 break;
1632
1633 case RUSAGE_SELF:
1634 thread_group_times(p, &tgutime, &tgstime);
1635 utime = cputime_add(utime, tgutime);
1636 stime = cputime_add(stime, tgstime);
1637 r->ru_nvcsw += p->signal->nvcsw;
1638 r->ru_nivcsw += p->signal->nivcsw;
1639 r->ru_minflt += p->signal->min_flt;
1640 r->ru_majflt += p->signal->maj_flt;
1641 r->ru_inblock += p->signal->inblock;
1642 r->ru_oublock += p->signal->oublock;
1643 if (maxrss < p->signal->maxrss)
1644 maxrss = p->signal->maxrss;
1645 t = p;
1646 do {
1647 accumulate_thread_rusage(t, r);
1648 t = next_thread(t);
1649 } while (t != p);
1650 break;
1651
1652 default:
1653 BUG();
1654 }
1655 unlock_task_sighand(p, &flags);
1656
1657out:
1658 cputime_to_timeval(utime, &r->ru_utime);
1659 cputime_to_timeval(stime, &r->ru_stime);
1660
1661 if (who != RUSAGE_CHILDREN) {
1662 struct mm_struct *mm = get_task_mm(p);
1663 if (mm) {
1664 setmax_mm_hiwater_rss(&maxrss, mm);
1665 mmput(mm);
1666 }
1667 }
1668 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1669}
1670
1671int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1672{
1673 struct rusage r;
1674 k_getrusage(p, who, &r);
1675 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1676}
1677
1678SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1679{
1680 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1681 who != RUSAGE_THREAD)
1682 return -EINVAL;
1683 return getrusage(current, who, ru);
1684}
1685
1686SYSCALL_DEFINE1(umask, int, mask)
1687{
1688 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1689 return mask;
1690}
1691
1692SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1693 unsigned long, arg4, unsigned long, arg5)
1694{
1695 struct task_struct *me = current;
1696 unsigned char comm[sizeof(me->comm)];
1697 long error;
1698
1699 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1700 if (error != -ENOSYS)
1701 return error;
1702
1703 error = 0;
1704 switch (option) {
1705 case PR_SET_PDEATHSIG:
1706 if (!valid_signal(arg2)) {
1707 error = -EINVAL;
1708 break;
1709 }
1710 me->pdeath_signal = arg2;
1711 error = 0;
1712 break;
1713 case PR_GET_PDEATHSIG:
1714 error = put_user(me->pdeath_signal, (int __user *)arg2);
1715 break;
1716 case PR_GET_DUMPABLE:
1717 error = get_dumpable(me->mm);
1718 break;
1719 case PR_SET_DUMPABLE:
1720 if (arg2 < 0 || arg2 > 1) {
1721 error = -EINVAL;
1722 break;
1723 }
1724 set_dumpable(me->mm, arg2);
1725 error = 0;
1726 break;
1727
1728 case PR_SET_UNALIGN:
1729 error = SET_UNALIGN_CTL(me, arg2);
1730 break;
1731 case PR_GET_UNALIGN:
1732 error = GET_UNALIGN_CTL(me, arg2);
1733 break;
1734 case PR_SET_FPEMU:
1735 error = SET_FPEMU_CTL(me, arg2);
1736 break;
1737 case PR_GET_FPEMU:
1738 error = GET_FPEMU_CTL(me, arg2);
1739 break;
1740 case PR_SET_FPEXC:
1741 error = SET_FPEXC_CTL(me, arg2);
1742 break;
1743 case PR_GET_FPEXC:
1744 error = GET_FPEXC_CTL(me, arg2);
1745 break;
1746 case PR_GET_TIMING:
1747 error = PR_TIMING_STATISTICAL;
1748 break;
1749 case PR_SET_TIMING:
1750 if (arg2 != PR_TIMING_STATISTICAL)
1751 error = -EINVAL;
1752 else
1753 error = 0;
1754 break;
1755
1756 case PR_SET_NAME:
1757 comm[sizeof(me->comm)-1] = 0;
1758 if (strncpy_from_user(comm, (char __user *)arg2,
1759 sizeof(me->comm) - 1) < 0)
1760 return -EFAULT;
1761 set_task_comm(me, comm);
1762 return 0;
1763 case PR_GET_NAME:
1764 get_task_comm(comm, me);
1765 if (copy_to_user((char __user *)arg2, comm,
1766 sizeof(comm)))
1767 return -EFAULT;
1768 return 0;
1769 case PR_GET_ENDIAN:
1770 error = GET_ENDIAN(me, arg2);
1771 break;
1772 case PR_SET_ENDIAN:
1773 error = SET_ENDIAN(me, arg2);
1774 break;
1775
1776 case PR_GET_SECCOMP:
1777 error = prctl_get_seccomp();
1778 break;
1779 case PR_SET_SECCOMP:
1780 error = prctl_set_seccomp(arg2);
1781 break;
1782 case PR_GET_TSC:
1783 error = GET_TSC_CTL(arg2);
1784 break;
1785 case PR_SET_TSC:
1786 error = SET_TSC_CTL(arg2);
1787 break;
1788 case PR_TASK_PERF_EVENTS_DISABLE:
1789 error = perf_event_task_disable();
1790 break;
1791 case PR_TASK_PERF_EVENTS_ENABLE:
1792 error = perf_event_task_enable();
1793 break;
1794 case PR_GET_TIMERSLACK:
1795 error = current->timer_slack_ns;
1796 break;
1797 case PR_SET_TIMERSLACK:
1798 if (arg2 <= 0)
1799 current->timer_slack_ns =
1800 current->default_timer_slack_ns;
1801 else
1802 current->timer_slack_ns = arg2;
1803 error = 0;
1804 break;
1805 case PR_MCE_KILL:
1806 if (arg4 | arg5)
1807 return -EINVAL;
1808 switch (arg2) {
1809 case PR_MCE_KILL_CLEAR:
1810 if (arg3 != 0)
1811 return -EINVAL;
1812 current->flags &= ~PF_MCE_PROCESS;
1813 break;
1814 case PR_MCE_KILL_SET:
1815 current->flags |= PF_MCE_PROCESS;
1816 if (arg3 == PR_MCE_KILL_EARLY)
1817 current->flags |= PF_MCE_EARLY;
1818 else if (arg3 == PR_MCE_KILL_LATE)
1819 current->flags &= ~PF_MCE_EARLY;
1820 else if (arg3 == PR_MCE_KILL_DEFAULT)
1821 current->flags &=
1822 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1823 else
1824 return -EINVAL;
1825 break;
1826 default:
1827 return -EINVAL;
1828 }
1829 error = 0;
1830 break;
1831 case PR_MCE_KILL_GET:
1832 if (arg2 | arg3 | arg4 | arg5)
1833 return -EINVAL;
1834 if (current->flags & PF_MCE_PROCESS)
1835 error = (current->flags & PF_MCE_EARLY) ?
1836 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1837 else
1838 error = PR_MCE_KILL_DEFAULT;
1839 break;
1840 default:
1841 error = -EINVAL;
1842 break;
1843 }
1844 return error;
1845}
1846
1847SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1848 struct getcpu_cache __user *, unused)
1849{
1850 int err = 0;
1851 int cpu = raw_smp_processor_id();
1852 if (cpup)
1853 err |= put_user(cpu, cpup);
1854 if (nodep)
1855 err |= put_user(cpu_to_node(cpu), nodep);
1856 return err ? -EFAULT : 0;
1857}
1858
1859char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1860
1861static void argv_cleanup(struct subprocess_info *info)
1862{
1863 argv_free(info->argv);
1864}
1865
1866/**
1867 * orderly_poweroff - Trigger an orderly system poweroff
1868 * @force: force poweroff if command execution fails
1869 *
1870 * This may be called from any context to trigger a system shutdown.
1871 * If the orderly shutdown fails, it will force an immediate shutdown.
1872 */
1873int orderly_poweroff(bool force)
1874{
1875 int argc;
1876 char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1877 static char *envp[] = {
1878 "HOME=/",
1879 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1880 NULL
1881 };
1882 int ret = -ENOMEM;
1883 struct subprocess_info *info;
1884
1885 if (argv == NULL) {
1886 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1887 __func__, poweroff_cmd);
1888 goto out;
1889 }
1890
1891 info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1892 if (info == NULL) {
1893 argv_free(argv);
1894 goto out;
1895 }
1896
1897 call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1898
1899 ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1900
1901 out:
1902 if (ret && force) {
1903 printk(KERN_WARNING "Failed to start orderly shutdown: "
1904 "forcing the issue\n");
1905
1906 /* I guess this should try to kick off some daemon to
1907 sync and poweroff asap. Or not even bother syncing
1908 if we're doing an emergency shutdown? */
1909 emergency_sync();
1910 kernel_power_off();
1911 }
1912
1913 return ret;
1914}
1915EXPORT_SYMBOL_GPL(orderly_poweroff);