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#ifndef RISCV_SET_ICACHE_FLUSH_CTX
150# define RISCV_SET_ICACHE_FLUSH_CTX(a, b) (-EINVAL)
151#endif
152#ifndef PPC_GET_DEXCR_ASPECT
153# define PPC_GET_DEXCR_ASPECT(a, b) (-EINVAL)
154#endif
155#ifndef PPC_SET_DEXCR_ASPECT
156# define PPC_SET_DEXCR_ASPECT(a, b, c) (-EINVAL)
157#endif
158
159/*
160 * this is where the system-wide overflow UID and GID are defined, for
161 * architectures that now have 32-bit UID/GID but didn't in the past
162 */
163
164int overflowuid = DEFAULT_OVERFLOWUID;
165int overflowgid = DEFAULT_OVERFLOWGID;
166
167EXPORT_SYMBOL(overflowuid);
168EXPORT_SYMBOL(overflowgid);
169
170/*
171 * the same as above, but for filesystems which can only store a 16-bit
172 * UID and GID. as such, this is needed on all architectures
173 */
174
175int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
176int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
177
178EXPORT_SYMBOL(fs_overflowuid);
179EXPORT_SYMBOL(fs_overflowgid);
180
181/*
182 * Returns true if current's euid is same as p's uid or euid,
183 * or has CAP_SYS_NICE to p's user_ns.
184 *
185 * Called with rcu_read_lock, creds are safe
186 */
187static bool set_one_prio_perm(struct task_struct *p)
188{
189 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
190
191 if (uid_eq(pcred->uid, cred->euid) ||
192 uid_eq(pcred->euid, cred->euid))
193 return true;
194 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
195 return true;
196 return false;
197}
198
199/*
200 * set the priority of a task
201 * - the caller must hold the RCU read lock
202 */
203static int set_one_prio(struct task_struct *p, int niceval, int error)
204{
205 int no_nice;
206
207 if (!set_one_prio_perm(p)) {
208 error = -EPERM;
209 goto out;
210 }
211 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
212 error = -EACCES;
213 goto out;
214 }
215 no_nice = security_task_setnice(p, niceval);
216 if (no_nice) {
217 error = no_nice;
218 goto out;
219 }
220 if (error == -ESRCH)
221 error = 0;
222 set_user_nice(p, niceval);
223out:
224 return error;
225}
226
227SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
228{
229 struct task_struct *g, *p;
230 struct user_struct *user;
231 const struct cred *cred = current_cred();
232 int error = -EINVAL;
233 struct pid *pgrp;
234 kuid_t uid;
235
236 if (which > PRIO_USER || which < PRIO_PROCESS)
237 goto out;
238
239 /* normalize: avoid signed division (rounding problems) */
240 error = -ESRCH;
241 if (niceval < MIN_NICE)
242 niceval = MIN_NICE;
243 if (niceval > MAX_NICE)
244 niceval = MAX_NICE;
245
246 rcu_read_lock();
247 switch (which) {
248 case PRIO_PROCESS:
249 if (who)
250 p = find_task_by_vpid(who);
251 else
252 p = current;
253 if (p)
254 error = set_one_prio(p, niceval, error);
255 break;
256 case PRIO_PGRP:
257 if (who)
258 pgrp = find_vpid(who);
259 else
260 pgrp = task_pgrp(current);
261 read_lock(&tasklist_lock);
262 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
263 error = set_one_prio(p, niceval, error);
264 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
265 read_unlock(&tasklist_lock);
266 break;
267 case PRIO_USER:
268 uid = make_kuid(cred->user_ns, who);
269 user = cred->user;
270 if (!who)
271 uid = cred->uid;
272 else if (!uid_eq(uid, cred->uid)) {
273 user = find_user(uid);
274 if (!user)
275 goto out_unlock; /* No processes for this user */
276 }
277 for_each_process_thread(g, p) {
278 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
279 error = set_one_prio(p, niceval, error);
280 }
281 if (!uid_eq(uid, cred->uid))
282 free_uid(user); /* For find_user() */
283 break;
284 }
285out_unlock:
286 rcu_read_unlock();
287out:
288 return error;
289}
290
291/*
292 * Ugh. To avoid negative return values, "getpriority()" will
293 * not return the normal nice-value, but a negated value that
294 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
295 * to stay compatible.
296 */
297SYSCALL_DEFINE2(getpriority, int, which, int, who)
298{
299 struct task_struct *g, *p;
300 struct user_struct *user;
301 const struct cred *cred = current_cred();
302 long niceval, retval = -ESRCH;
303 struct pid *pgrp;
304 kuid_t uid;
305
306 if (which > PRIO_USER || which < PRIO_PROCESS)
307 return -EINVAL;
308
309 rcu_read_lock();
310 switch (which) {
311 case PRIO_PROCESS:
312 if (who)
313 p = find_task_by_vpid(who);
314 else
315 p = current;
316 if (p) {
317 niceval = nice_to_rlimit(task_nice(p));
318 if (niceval > retval)
319 retval = niceval;
320 }
321 break;
322 case PRIO_PGRP:
323 if (who)
324 pgrp = find_vpid(who);
325 else
326 pgrp = task_pgrp(current);
327 read_lock(&tasklist_lock);
328 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
329 niceval = nice_to_rlimit(task_nice(p));
330 if (niceval > retval)
331 retval = niceval;
332 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
333 read_unlock(&tasklist_lock);
334 break;
335 case PRIO_USER:
336 uid = make_kuid(cred->user_ns, who);
337 user = cred->user;
338 if (!who)
339 uid = cred->uid;
340 else if (!uid_eq(uid, cred->uid)) {
341 user = find_user(uid);
342 if (!user)
343 goto out_unlock; /* No processes for this user */
344 }
345 for_each_process_thread(g, p) {
346 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
347 niceval = nice_to_rlimit(task_nice(p));
348 if (niceval > retval)
349 retval = niceval;
350 }
351 }
352 if (!uid_eq(uid, cred->uid))
353 free_uid(user); /* for find_user() */
354 break;
355 }
356out_unlock:
357 rcu_read_unlock();
358
359 return retval;
360}
361
362/*
363 * Unprivileged users may change the real gid to the effective gid
364 * or vice versa. (BSD-style)
365 *
366 * If you set the real gid at all, or set the effective gid to a value not
367 * equal to the real gid, then the saved gid is set to the new effective gid.
368 *
369 * This makes it possible for a setgid program to completely drop its
370 * privileges, which is often a useful assertion to make when you are doing
371 * a security audit over a program.
372 *
373 * The general idea is that a program which uses just setregid() will be
374 * 100% compatible with BSD. A program which uses just setgid() will be
375 * 100% compatible with POSIX with saved IDs.
376 *
377 * SMP: There are not races, the GIDs are checked only by filesystem
378 * operations (as far as semantic preservation is concerned).
379 */
380#ifdef CONFIG_MULTIUSER
381long __sys_setregid(gid_t rgid, gid_t egid)
382{
383 struct user_namespace *ns = current_user_ns();
384 const struct cred *old;
385 struct cred *new;
386 int retval;
387 kgid_t krgid, kegid;
388
389 krgid = make_kgid(ns, rgid);
390 kegid = make_kgid(ns, egid);
391
392 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
393 return -EINVAL;
394 if ((egid != (gid_t) -1) && !gid_valid(kegid))
395 return -EINVAL;
396
397 new = prepare_creds();
398 if (!new)
399 return -ENOMEM;
400 old = current_cred();
401
402 retval = -EPERM;
403 if (rgid != (gid_t) -1) {
404 if (gid_eq(old->gid, krgid) ||
405 gid_eq(old->egid, krgid) ||
406 ns_capable_setid(old->user_ns, CAP_SETGID))
407 new->gid = krgid;
408 else
409 goto error;
410 }
411 if (egid != (gid_t) -1) {
412 if (gid_eq(old->gid, kegid) ||
413 gid_eq(old->egid, kegid) ||
414 gid_eq(old->sgid, kegid) ||
415 ns_capable_setid(old->user_ns, CAP_SETGID))
416 new->egid = kegid;
417 else
418 goto error;
419 }
420
421 if (rgid != (gid_t) -1 ||
422 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
423 new->sgid = new->egid;
424 new->fsgid = new->egid;
425
426 retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
427 if (retval < 0)
428 goto error;
429
430 return commit_creds(new);
431
432error:
433 abort_creds(new);
434 return retval;
435}
436
437SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
438{
439 return __sys_setregid(rgid, egid);
440}
441
442/*
443 * setgid() is implemented like SysV w/ SAVED_IDS
444 *
445 * SMP: Same implicit races as above.
446 */
447long __sys_setgid(gid_t gid)
448{
449 struct user_namespace *ns = current_user_ns();
450 const struct cred *old;
451 struct cred *new;
452 int retval;
453 kgid_t kgid;
454
455 kgid = make_kgid(ns, gid);
456 if (!gid_valid(kgid))
457 return -EINVAL;
458
459 new = prepare_creds();
460 if (!new)
461 return -ENOMEM;
462 old = current_cred();
463
464 retval = -EPERM;
465 if (ns_capable_setid(old->user_ns, CAP_SETGID))
466 new->gid = new->egid = new->sgid = new->fsgid = kgid;
467 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
468 new->egid = new->fsgid = kgid;
469 else
470 goto error;
471
472 retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
473 if (retval < 0)
474 goto error;
475
476 return commit_creds(new);
477
478error:
479 abort_creds(new);
480 return retval;
481}
482
483SYSCALL_DEFINE1(setgid, gid_t, gid)
484{
485 return __sys_setgid(gid);
486}
487
488/*
489 * change the user struct in a credentials set to match the new UID
490 */
491static int set_user(struct cred *new)
492{
493 struct user_struct *new_user;
494
495 new_user = alloc_uid(new->uid);
496 if (!new_user)
497 return -EAGAIN;
498
499 free_uid(new->user);
500 new->user = new_user;
501 return 0;
502}
503
504static void flag_nproc_exceeded(struct cred *new)
505{
506 if (new->ucounts == current_ucounts())
507 return;
508
509 /*
510 * We don't fail in case of NPROC limit excess here because too many
511 * poorly written programs don't check set*uid() return code, assuming
512 * it never fails if called by root. We may still enforce NPROC limit
513 * for programs doing set*uid()+execve() by harmlessly deferring the
514 * failure to the execve() stage.
515 */
516 if (is_rlimit_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
517 new->user != INIT_USER)
518 current->flags |= PF_NPROC_EXCEEDED;
519 else
520 current->flags &= ~PF_NPROC_EXCEEDED;
521}
522
523/*
524 * Unprivileged users may change the real uid to the effective uid
525 * or vice versa. (BSD-style)
526 *
527 * If you set the real uid at all, or set the effective uid to a value not
528 * equal to the real uid, then the saved uid is set to the new effective uid.
529 *
530 * This makes it possible for a setuid program to completely drop its
531 * privileges, which is often a useful assertion to make when you are doing
532 * a security audit over a program.
533 *
534 * The general idea is that a program which uses just setreuid() will be
535 * 100% compatible with BSD. A program which uses just setuid() will be
536 * 100% compatible with POSIX with saved IDs.
537 */
538long __sys_setreuid(uid_t ruid, uid_t euid)
539{
540 struct user_namespace *ns = current_user_ns();
541 const struct cred *old;
542 struct cred *new;
543 int retval;
544 kuid_t kruid, keuid;
545
546 kruid = make_kuid(ns, ruid);
547 keuid = make_kuid(ns, euid);
548
549 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
550 return -EINVAL;
551 if ((euid != (uid_t) -1) && !uid_valid(keuid))
552 return -EINVAL;
553
554 new = prepare_creds();
555 if (!new)
556 return -ENOMEM;
557 old = current_cred();
558
559 retval = -EPERM;
560 if (ruid != (uid_t) -1) {
561 new->uid = kruid;
562 if (!uid_eq(old->uid, kruid) &&
563 !uid_eq(old->euid, kruid) &&
564 !ns_capable_setid(old->user_ns, CAP_SETUID))
565 goto error;
566 }
567
568 if (euid != (uid_t) -1) {
569 new->euid = keuid;
570 if (!uid_eq(old->uid, keuid) &&
571 !uid_eq(old->euid, keuid) &&
572 !uid_eq(old->suid, keuid) &&
573 !ns_capable_setid(old->user_ns, CAP_SETUID))
574 goto error;
575 }
576
577 if (!uid_eq(new->uid, old->uid)) {
578 retval = set_user(new);
579 if (retval < 0)
580 goto error;
581 }
582 if (ruid != (uid_t) -1 ||
583 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
584 new->suid = new->euid;
585 new->fsuid = new->euid;
586
587 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
588 if (retval < 0)
589 goto error;
590
591 retval = set_cred_ucounts(new);
592 if (retval < 0)
593 goto error;
594
595 flag_nproc_exceeded(new);
596 return commit_creds(new);
597
598error:
599 abort_creds(new);
600 return retval;
601}
602
603SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
604{
605 return __sys_setreuid(ruid, euid);
606}
607
608/*
609 * setuid() is implemented like SysV with SAVED_IDS
610 *
611 * Note that SAVED_ID's is deficient in that a setuid root program
612 * like sendmail, for example, cannot set its uid to be a normal
613 * user and then switch back, because if you're root, setuid() sets
614 * the saved uid too. If you don't like this, blame the bright people
615 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
616 * will allow a root program to temporarily drop privileges and be able to
617 * regain them by swapping the real and effective uid.
618 */
619long __sys_setuid(uid_t uid)
620{
621 struct user_namespace *ns = current_user_ns();
622 const struct cred *old;
623 struct cred *new;
624 int retval;
625 kuid_t kuid;
626
627 kuid = make_kuid(ns, uid);
628 if (!uid_valid(kuid))
629 return -EINVAL;
630
631 new = prepare_creds();
632 if (!new)
633 return -ENOMEM;
634 old = current_cred();
635
636 retval = -EPERM;
637 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
638 new->suid = new->uid = kuid;
639 if (!uid_eq(kuid, old->uid)) {
640 retval = set_user(new);
641 if (retval < 0)
642 goto error;
643 }
644 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
645 goto error;
646 }
647
648 new->fsuid = new->euid = kuid;
649
650 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
651 if (retval < 0)
652 goto error;
653
654 retval = set_cred_ucounts(new);
655 if (retval < 0)
656 goto error;
657
658 flag_nproc_exceeded(new);
659 return commit_creds(new);
660
661error:
662 abort_creds(new);
663 return retval;
664}
665
666SYSCALL_DEFINE1(setuid, uid_t, uid)
667{
668 return __sys_setuid(uid);
669}
670
671
672/*
673 * This function implements a generic ability to update ruid, euid,
674 * and suid. This allows you to implement the 4.4 compatible seteuid().
675 */
676long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
677{
678 struct user_namespace *ns = current_user_ns();
679 const struct cred *old;
680 struct cred *new;
681 int retval;
682 kuid_t kruid, keuid, ksuid;
683 bool ruid_new, euid_new, suid_new;
684
685 kruid = make_kuid(ns, ruid);
686 keuid = make_kuid(ns, euid);
687 ksuid = make_kuid(ns, suid);
688
689 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
690 return -EINVAL;
691
692 if ((euid != (uid_t) -1) && !uid_valid(keuid))
693 return -EINVAL;
694
695 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
696 return -EINVAL;
697
698 old = current_cred();
699
700 /* check for no-op */
701 if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
702 (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
703 uid_eq(keuid, old->fsuid))) &&
704 (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
705 return 0;
706
707 ruid_new = ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
708 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
709 euid_new = euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
710 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
711 suid_new = suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
712 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
713 if ((ruid_new || euid_new || suid_new) &&
714 !ns_capable_setid(old->user_ns, CAP_SETUID))
715 return -EPERM;
716
717 new = prepare_creds();
718 if (!new)
719 return -ENOMEM;
720
721 if (ruid != (uid_t) -1) {
722 new->uid = kruid;
723 if (!uid_eq(kruid, old->uid)) {
724 retval = set_user(new);
725 if (retval < 0)
726 goto error;
727 }
728 }
729 if (euid != (uid_t) -1)
730 new->euid = keuid;
731 if (suid != (uid_t) -1)
732 new->suid = ksuid;
733 new->fsuid = new->euid;
734
735 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
736 if (retval < 0)
737 goto error;
738
739 retval = set_cred_ucounts(new);
740 if (retval < 0)
741 goto error;
742
743 flag_nproc_exceeded(new);
744 return commit_creds(new);
745
746error:
747 abort_creds(new);
748 return retval;
749}
750
751SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
752{
753 return __sys_setresuid(ruid, euid, suid);
754}
755
756SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
757{
758 const struct cred *cred = current_cred();
759 int retval;
760 uid_t ruid, euid, suid;
761
762 ruid = from_kuid_munged(cred->user_ns, cred->uid);
763 euid = from_kuid_munged(cred->user_ns, cred->euid);
764 suid = from_kuid_munged(cred->user_ns, cred->suid);
765
766 retval = put_user(ruid, ruidp);
767 if (!retval) {
768 retval = put_user(euid, euidp);
769 if (!retval)
770 return put_user(suid, suidp);
771 }
772 return retval;
773}
774
775/*
776 * Same as above, but for rgid, egid, sgid.
777 */
778long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
779{
780 struct user_namespace *ns = current_user_ns();
781 const struct cred *old;
782 struct cred *new;
783 int retval;
784 kgid_t krgid, kegid, ksgid;
785 bool rgid_new, egid_new, sgid_new;
786
787 krgid = make_kgid(ns, rgid);
788 kegid = make_kgid(ns, egid);
789 ksgid = make_kgid(ns, sgid);
790
791 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
792 return -EINVAL;
793 if ((egid != (gid_t) -1) && !gid_valid(kegid))
794 return -EINVAL;
795 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
796 return -EINVAL;
797
798 old = current_cred();
799
800 /* check for no-op */
801 if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
802 (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
803 gid_eq(kegid, old->fsgid))) &&
804 (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
805 return 0;
806
807 rgid_new = rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
808 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
809 egid_new = egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
810 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
811 sgid_new = sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
812 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
813 if ((rgid_new || egid_new || sgid_new) &&
814 !ns_capable_setid(old->user_ns, CAP_SETGID))
815 return -EPERM;
816
817 new = prepare_creds();
818 if (!new)
819 return -ENOMEM;
820
821 if (rgid != (gid_t) -1)
822 new->gid = krgid;
823 if (egid != (gid_t) -1)
824 new->egid = kegid;
825 if (sgid != (gid_t) -1)
826 new->sgid = ksgid;
827 new->fsgid = new->egid;
828
829 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
830 if (retval < 0)
831 goto error;
832
833 return commit_creds(new);
834
835error:
836 abort_creds(new);
837 return retval;
838}
839
840SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
841{
842 return __sys_setresgid(rgid, egid, sgid);
843}
844
845SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
846{
847 const struct cred *cred = current_cred();
848 int retval;
849 gid_t rgid, egid, sgid;
850
851 rgid = from_kgid_munged(cred->user_ns, cred->gid);
852 egid = from_kgid_munged(cred->user_ns, cred->egid);
853 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
854
855 retval = put_user(rgid, rgidp);
856 if (!retval) {
857 retval = put_user(egid, egidp);
858 if (!retval)
859 retval = put_user(sgid, sgidp);
860 }
861
862 return retval;
863}
864
865
866/*
867 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
868 * is used for "access()" and for the NFS daemon (letting nfsd stay at
869 * whatever uid it wants to). It normally shadows "euid", except when
870 * explicitly set by setfsuid() or for access..
871 */
872long __sys_setfsuid(uid_t uid)
873{
874 const struct cred *old;
875 struct cred *new;
876 uid_t old_fsuid;
877 kuid_t kuid;
878
879 old = current_cred();
880 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
881
882 kuid = make_kuid(old->user_ns, uid);
883 if (!uid_valid(kuid))
884 return old_fsuid;
885
886 new = prepare_creds();
887 if (!new)
888 return old_fsuid;
889
890 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
891 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
892 ns_capable_setid(old->user_ns, CAP_SETUID)) {
893 if (!uid_eq(kuid, old->fsuid)) {
894 new->fsuid = kuid;
895 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
896 goto change_okay;
897 }
898 }
899
900 abort_creds(new);
901 return old_fsuid;
902
903change_okay:
904 commit_creds(new);
905 return old_fsuid;
906}
907
908SYSCALL_DEFINE1(setfsuid, uid_t, uid)
909{
910 return __sys_setfsuid(uid);
911}
912
913/*
914 * Samma på svenska..
915 */
916long __sys_setfsgid(gid_t gid)
917{
918 const struct cred *old;
919 struct cred *new;
920 gid_t old_fsgid;
921 kgid_t kgid;
922
923 old = current_cred();
924 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
925
926 kgid = make_kgid(old->user_ns, gid);
927 if (!gid_valid(kgid))
928 return old_fsgid;
929
930 new = prepare_creds();
931 if (!new)
932 return old_fsgid;
933
934 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
935 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
936 ns_capable_setid(old->user_ns, CAP_SETGID)) {
937 if (!gid_eq(kgid, old->fsgid)) {
938 new->fsgid = kgid;
939 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
940 goto change_okay;
941 }
942 }
943
944 abort_creds(new);
945 return old_fsgid;
946
947change_okay:
948 commit_creds(new);
949 return old_fsgid;
950}
951
952SYSCALL_DEFINE1(setfsgid, gid_t, gid)
953{
954 return __sys_setfsgid(gid);
955}
956#endif /* CONFIG_MULTIUSER */
957
958/**
959 * sys_getpid - return the thread group id of the current process
960 *
961 * Note, despite the name, this returns the tgid not the pid. The tgid and
962 * the pid are identical unless CLONE_THREAD was specified on clone() in
963 * which case the tgid is the same in all threads of the same group.
964 *
965 * This is SMP safe as current->tgid does not change.
966 */
967SYSCALL_DEFINE0(getpid)
968{
969 return task_tgid_vnr(current);
970}
971
972/* Thread ID - the internal kernel "pid" */
973SYSCALL_DEFINE0(gettid)
974{
975 return task_pid_vnr(current);
976}
977
978/*
979 * Accessing ->real_parent is not SMP-safe, it could
980 * change from under us. However, we can use a stale
981 * value of ->real_parent under rcu_read_lock(), see
982 * release_task()->call_rcu(delayed_put_task_struct).
983 */
984SYSCALL_DEFINE0(getppid)
985{
986 int pid;
987
988 rcu_read_lock();
989 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
990 rcu_read_unlock();
991
992 return pid;
993}
994
995SYSCALL_DEFINE0(getuid)
996{
997 /* Only we change this so SMP safe */
998 return from_kuid_munged(current_user_ns(), current_uid());
999}
1000
1001SYSCALL_DEFINE0(geteuid)
1002{
1003 /* Only we change this so SMP safe */
1004 return from_kuid_munged(current_user_ns(), current_euid());
1005}
1006
1007SYSCALL_DEFINE0(getgid)
1008{
1009 /* Only we change this so SMP safe */
1010 return from_kgid_munged(current_user_ns(), current_gid());
1011}
1012
1013SYSCALL_DEFINE0(getegid)
1014{
1015 /* Only we change this so SMP safe */
1016 return from_kgid_munged(current_user_ns(), current_egid());
1017}
1018
1019static void do_sys_times(struct tms *tms)
1020{
1021 u64 tgutime, tgstime, cutime, cstime;
1022
1023 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1024 cutime = current->signal->cutime;
1025 cstime = current->signal->cstime;
1026 tms->tms_utime = nsec_to_clock_t(tgutime);
1027 tms->tms_stime = nsec_to_clock_t(tgstime);
1028 tms->tms_cutime = nsec_to_clock_t(cutime);
1029 tms->tms_cstime = nsec_to_clock_t(cstime);
1030}
1031
1032SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1033{
1034 if (tbuf) {
1035 struct tms tmp;
1036
1037 do_sys_times(&tmp);
1038 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1039 return -EFAULT;
1040 }
1041 force_successful_syscall_return();
1042 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1043}
1044
1045#ifdef CONFIG_COMPAT
1046static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1047{
1048 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1049}
1050
1051COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1052{
1053 if (tbuf) {
1054 struct tms tms;
1055 struct compat_tms tmp;
1056
1057 do_sys_times(&tms);
1058 /* Convert our struct tms to the compat version. */
1059 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1060 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1061 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1062 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1063 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1064 return -EFAULT;
1065 }
1066 force_successful_syscall_return();
1067 return compat_jiffies_to_clock_t(jiffies);
1068}
1069#endif
1070
1071/*
1072 * This needs some heavy checking ...
1073 * I just haven't the stomach for it. I also don't fully
1074 * understand sessions/pgrp etc. Let somebody who does explain it.
1075 *
1076 * OK, I think I have the protection semantics right.... this is really
1077 * only important on a multi-user system anyway, to make sure one user
1078 * can't send a signal to a process owned by another. -TYT, 12/12/91
1079 *
1080 * !PF_FORKNOEXEC check to conform completely to POSIX.
1081 */
1082SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1083{
1084 struct task_struct *p;
1085 struct task_struct *group_leader = current->group_leader;
1086 struct pid *pgrp;
1087 int err;
1088
1089 if (!pid)
1090 pid = task_pid_vnr(group_leader);
1091 if (!pgid)
1092 pgid = pid;
1093 if (pgid < 0)
1094 return -EINVAL;
1095 rcu_read_lock();
1096
1097 /* From this point forward we keep holding onto the tasklist lock
1098 * so that our parent does not change from under us. -DaveM
1099 */
1100 write_lock_irq(&tasklist_lock);
1101
1102 err = -ESRCH;
1103 p = find_task_by_vpid(pid);
1104 if (!p)
1105 goto out;
1106
1107 err = -EINVAL;
1108 if (!thread_group_leader(p))
1109 goto out;
1110
1111 if (same_thread_group(p->real_parent, group_leader)) {
1112 err = -EPERM;
1113 if (task_session(p) != task_session(group_leader))
1114 goto out;
1115 err = -EACCES;
1116 if (!(p->flags & PF_FORKNOEXEC))
1117 goto out;
1118 } else {
1119 err = -ESRCH;
1120 if (p != group_leader)
1121 goto out;
1122 }
1123
1124 err = -EPERM;
1125 if (p->signal->leader)
1126 goto out;
1127
1128 pgrp = task_pid(p);
1129 if (pgid != pid) {
1130 struct task_struct *g;
1131
1132 pgrp = find_vpid(pgid);
1133 g = pid_task(pgrp, PIDTYPE_PGID);
1134 if (!g || task_session(g) != task_session(group_leader))
1135 goto out;
1136 }
1137
1138 err = security_task_setpgid(p, pgid);
1139 if (err)
1140 goto out;
1141
1142 if (task_pgrp(p) != pgrp)
1143 change_pid(p, PIDTYPE_PGID, pgrp);
1144
1145 err = 0;
1146out:
1147 /* All paths lead to here, thus we are safe. -DaveM */
1148 write_unlock_irq(&tasklist_lock);
1149 rcu_read_unlock();
1150 return err;
1151}
1152
1153static int do_getpgid(pid_t pid)
1154{
1155 struct task_struct *p;
1156 struct pid *grp;
1157 int retval;
1158
1159 rcu_read_lock();
1160 if (!pid)
1161 grp = task_pgrp(current);
1162 else {
1163 retval = -ESRCH;
1164 p = find_task_by_vpid(pid);
1165 if (!p)
1166 goto out;
1167 grp = task_pgrp(p);
1168 if (!grp)
1169 goto out;
1170
1171 retval = security_task_getpgid(p);
1172 if (retval)
1173 goto out;
1174 }
1175 retval = pid_vnr(grp);
1176out:
1177 rcu_read_unlock();
1178 return retval;
1179}
1180
1181SYSCALL_DEFINE1(getpgid, pid_t, pid)
1182{
1183 return do_getpgid(pid);
1184}
1185
1186#ifdef __ARCH_WANT_SYS_GETPGRP
1187
1188SYSCALL_DEFINE0(getpgrp)
1189{
1190 return do_getpgid(0);
1191}
1192
1193#endif
1194
1195SYSCALL_DEFINE1(getsid, pid_t, pid)
1196{
1197 struct task_struct *p;
1198 struct pid *sid;
1199 int retval;
1200
1201 rcu_read_lock();
1202 if (!pid)
1203 sid = task_session(current);
1204 else {
1205 retval = -ESRCH;
1206 p = find_task_by_vpid(pid);
1207 if (!p)
1208 goto out;
1209 sid = task_session(p);
1210 if (!sid)
1211 goto out;
1212
1213 retval = security_task_getsid(p);
1214 if (retval)
1215 goto out;
1216 }
1217 retval = pid_vnr(sid);
1218out:
1219 rcu_read_unlock();
1220 return retval;
1221}
1222
1223static void set_special_pids(struct pid *pid)
1224{
1225 struct task_struct *curr = current->group_leader;
1226
1227 if (task_session(curr) != pid)
1228 change_pid(curr, PIDTYPE_SID, pid);
1229
1230 if (task_pgrp(curr) != pid)
1231 change_pid(curr, PIDTYPE_PGID, pid);
1232}
1233
1234int ksys_setsid(void)
1235{
1236 struct task_struct *group_leader = current->group_leader;
1237 struct pid *sid = task_pid(group_leader);
1238 pid_t session = pid_vnr(sid);
1239 int err = -EPERM;
1240
1241 write_lock_irq(&tasklist_lock);
1242 /* Fail if I am already a session leader */
1243 if (group_leader->signal->leader)
1244 goto out;
1245
1246 /* Fail if a process group id already exists that equals the
1247 * proposed session id.
1248 */
1249 if (pid_task(sid, PIDTYPE_PGID))
1250 goto out;
1251
1252 group_leader->signal->leader = 1;
1253 set_special_pids(sid);
1254
1255 proc_clear_tty(group_leader);
1256
1257 err = session;
1258out:
1259 write_unlock_irq(&tasklist_lock);
1260 if (err > 0) {
1261 proc_sid_connector(group_leader);
1262 sched_autogroup_create_attach(group_leader);
1263 }
1264 return err;
1265}
1266
1267SYSCALL_DEFINE0(setsid)
1268{
1269 return ksys_setsid();
1270}
1271
1272DECLARE_RWSEM(uts_sem);
1273
1274#ifdef COMPAT_UTS_MACHINE
1275#define override_architecture(name) \
1276 (personality(current->personality) == PER_LINUX32 && \
1277 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1278 sizeof(COMPAT_UTS_MACHINE)))
1279#else
1280#define override_architecture(name) 0
1281#endif
1282
1283/*
1284 * Work around broken programs that cannot handle "Linux 3.0".
1285 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1286 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1287 * 2.6.60.
1288 */
1289static int override_release(char __user *release, size_t len)
1290{
1291 int ret = 0;
1292
1293 if (current->personality & UNAME26) {
1294 const char *rest = UTS_RELEASE;
1295 char buf[65] = { 0 };
1296 int ndots = 0;
1297 unsigned v;
1298 size_t copy;
1299
1300 while (*rest) {
1301 if (*rest == '.' && ++ndots >= 3)
1302 break;
1303 if (!isdigit(*rest) && *rest != '.')
1304 break;
1305 rest++;
1306 }
1307 v = LINUX_VERSION_PATCHLEVEL + 60;
1308 copy = clamp_t(size_t, len, 1, sizeof(buf));
1309 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1310 ret = copy_to_user(release, buf, copy + 1);
1311 }
1312 return ret;
1313}
1314
1315SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1316{
1317 struct new_utsname tmp;
1318
1319 down_read(&uts_sem);
1320 memcpy(&tmp, utsname(), sizeof(tmp));
1321 up_read(&uts_sem);
1322 if (copy_to_user(name, &tmp, sizeof(tmp)))
1323 return -EFAULT;
1324
1325 if (override_release(name->release, sizeof(name->release)))
1326 return -EFAULT;
1327 if (override_architecture(name))
1328 return -EFAULT;
1329 return 0;
1330}
1331
1332#ifdef __ARCH_WANT_SYS_OLD_UNAME
1333/*
1334 * Old cruft
1335 */
1336SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1337{
1338 struct old_utsname tmp;
1339
1340 if (!name)
1341 return -EFAULT;
1342
1343 down_read(&uts_sem);
1344 memcpy(&tmp, utsname(), sizeof(tmp));
1345 up_read(&uts_sem);
1346 if (copy_to_user(name, &tmp, sizeof(tmp)))
1347 return -EFAULT;
1348
1349 if (override_release(name->release, sizeof(name->release)))
1350 return -EFAULT;
1351 if (override_architecture(name))
1352 return -EFAULT;
1353 return 0;
1354}
1355
1356SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1357{
1358 struct oldold_utsname tmp;
1359
1360 if (!name)
1361 return -EFAULT;
1362
1363 memset(&tmp, 0, sizeof(tmp));
1364
1365 down_read(&uts_sem);
1366 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1367 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1368 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1369 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1370 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1371 up_read(&uts_sem);
1372 if (copy_to_user(name, &tmp, sizeof(tmp)))
1373 return -EFAULT;
1374
1375 if (override_architecture(name))
1376 return -EFAULT;
1377 if (override_release(name->release, sizeof(name->release)))
1378 return -EFAULT;
1379 return 0;
1380}
1381#endif
1382
1383SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1384{
1385 int errno;
1386 char tmp[__NEW_UTS_LEN];
1387
1388 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1389 return -EPERM;
1390
1391 if (len < 0 || len > __NEW_UTS_LEN)
1392 return -EINVAL;
1393 errno = -EFAULT;
1394 if (!copy_from_user(tmp, name, len)) {
1395 struct new_utsname *u;
1396
1397 add_device_randomness(tmp, len);
1398 down_write(&uts_sem);
1399 u = utsname();
1400 memcpy(u->nodename, tmp, len);
1401 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1402 errno = 0;
1403 uts_proc_notify(UTS_PROC_HOSTNAME);
1404 up_write(&uts_sem);
1405 }
1406 return errno;
1407}
1408
1409#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1410
1411SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1412{
1413 int i;
1414 struct new_utsname *u;
1415 char tmp[__NEW_UTS_LEN + 1];
1416
1417 if (len < 0)
1418 return -EINVAL;
1419 down_read(&uts_sem);
1420 u = utsname();
1421 i = 1 + strlen(u->nodename);
1422 if (i > len)
1423 i = len;
1424 memcpy(tmp, u->nodename, i);
1425 up_read(&uts_sem);
1426 if (copy_to_user(name, tmp, i))
1427 return -EFAULT;
1428 return 0;
1429}
1430
1431#endif
1432
1433/*
1434 * Only setdomainname; getdomainname can be implemented by calling
1435 * uname()
1436 */
1437SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1438{
1439 int errno;
1440 char tmp[__NEW_UTS_LEN];
1441
1442 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1443 return -EPERM;
1444 if (len < 0 || len > __NEW_UTS_LEN)
1445 return -EINVAL;
1446
1447 errno = -EFAULT;
1448 if (!copy_from_user(tmp, name, len)) {
1449 struct new_utsname *u;
1450
1451 add_device_randomness(tmp, len);
1452 down_write(&uts_sem);
1453 u = utsname();
1454 memcpy(u->domainname, tmp, len);
1455 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1456 errno = 0;
1457 uts_proc_notify(UTS_PROC_DOMAINNAME);
1458 up_write(&uts_sem);
1459 }
1460 return errno;
1461}
1462
1463/* make sure you are allowed to change @tsk limits before calling this */
1464static int do_prlimit(struct task_struct *tsk, unsigned int resource,
1465 struct rlimit *new_rlim, struct rlimit *old_rlim)
1466{
1467 struct rlimit *rlim;
1468 int retval = 0;
1469
1470 if (resource >= RLIM_NLIMITS)
1471 return -EINVAL;
1472 resource = array_index_nospec(resource, RLIM_NLIMITS);
1473
1474 if (new_rlim) {
1475 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1476 return -EINVAL;
1477 if (resource == RLIMIT_NOFILE &&
1478 new_rlim->rlim_max > sysctl_nr_open)
1479 return -EPERM;
1480 }
1481
1482 /* Holding a refcount on tsk protects tsk->signal from disappearing. */
1483 rlim = tsk->signal->rlim + resource;
1484 task_lock(tsk->group_leader);
1485 if (new_rlim) {
1486 /*
1487 * Keep the capable check against init_user_ns until cgroups can
1488 * contain all limits.
1489 */
1490 if (new_rlim->rlim_max > rlim->rlim_max &&
1491 !capable(CAP_SYS_RESOURCE))
1492 retval = -EPERM;
1493 if (!retval)
1494 retval = security_task_setrlimit(tsk, resource, new_rlim);
1495 }
1496 if (!retval) {
1497 if (old_rlim)
1498 *old_rlim = *rlim;
1499 if (new_rlim)
1500 *rlim = *new_rlim;
1501 }
1502 task_unlock(tsk->group_leader);
1503
1504 /*
1505 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1506 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1507 * ignores the rlimit.
1508 */
1509 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1510 new_rlim->rlim_cur != RLIM_INFINITY &&
1511 IS_ENABLED(CONFIG_POSIX_TIMERS)) {
1512 /*
1513 * update_rlimit_cpu can fail if the task is exiting, but there
1514 * may be other tasks in the thread group that are not exiting,
1515 * and they need their cpu timers adjusted.
1516 *
1517 * The group_leader is the last task to be released, so if we
1518 * cannot update_rlimit_cpu on it, then the entire process is
1519 * exiting and we do not need to update at all.
1520 */
1521 update_rlimit_cpu(tsk->group_leader, new_rlim->rlim_cur);
1522 }
1523
1524 return retval;
1525}
1526
1527SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1528{
1529 struct rlimit value;
1530 int ret;
1531
1532 ret = do_prlimit(current, resource, NULL, &value);
1533 if (!ret)
1534 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1535
1536 return ret;
1537}
1538
1539#ifdef CONFIG_COMPAT
1540
1541COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1542 struct compat_rlimit __user *, rlim)
1543{
1544 struct rlimit r;
1545 struct compat_rlimit r32;
1546
1547 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1548 return -EFAULT;
1549
1550 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1551 r.rlim_cur = RLIM_INFINITY;
1552 else
1553 r.rlim_cur = r32.rlim_cur;
1554 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1555 r.rlim_max = RLIM_INFINITY;
1556 else
1557 r.rlim_max = r32.rlim_max;
1558 return do_prlimit(current, resource, &r, NULL);
1559}
1560
1561COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1562 struct compat_rlimit __user *, rlim)
1563{
1564 struct rlimit r;
1565 int ret;
1566
1567 ret = do_prlimit(current, resource, NULL, &r);
1568 if (!ret) {
1569 struct compat_rlimit r32;
1570 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1571 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1572 else
1573 r32.rlim_cur = r.rlim_cur;
1574 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1575 r32.rlim_max = COMPAT_RLIM_INFINITY;
1576 else
1577 r32.rlim_max = r.rlim_max;
1578
1579 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1580 return -EFAULT;
1581 }
1582 return ret;
1583}
1584
1585#endif
1586
1587#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1588
1589/*
1590 * Back compatibility for getrlimit. Needed for some apps.
1591 */
1592SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1593 struct rlimit __user *, rlim)
1594{
1595 struct rlimit x;
1596 if (resource >= RLIM_NLIMITS)
1597 return -EINVAL;
1598
1599 resource = array_index_nospec(resource, RLIM_NLIMITS);
1600 task_lock(current->group_leader);
1601 x = current->signal->rlim[resource];
1602 task_unlock(current->group_leader);
1603 if (x.rlim_cur > 0x7FFFFFFF)
1604 x.rlim_cur = 0x7FFFFFFF;
1605 if (x.rlim_max > 0x7FFFFFFF)
1606 x.rlim_max = 0x7FFFFFFF;
1607 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1608}
1609
1610#ifdef CONFIG_COMPAT
1611COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1612 struct compat_rlimit __user *, rlim)
1613{
1614 struct rlimit r;
1615
1616 if (resource >= RLIM_NLIMITS)
1617 return -EINVAL;
1618
1619 resource = array_index_nospec(resource, RLIM_NLIMITS);
1620 task_lock(current->group_leader);
1621 r = current->signal->rlim[resource];
1622 task_unlock(current->group_leader);
1623 if (r.rlim_cur > 0x7FFFFFFF)
1624 r.rlim_cur = 0x7FFFFFFF;
1625 if (r.rlim_max > 0x7FFFFFFF)
1626 r.rlim_max = 0x7FFFFFFF;
1627
1628 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1629 put_user(r.rlim_max, &rlim->rlim_max))
1630 return -EFAULT;
1631 return 0;
1632}
1633#endif
1634
1635#endif
1636
1637static inline bool rlim64_is_infinity(__u64 rlim64)
1638{
1639#if BITS_PER_LONG < 64
1640 return rlim64 >= ULONG_MAX;
1641#else
1642 return rlim64 == RLIM64_INFINITY;
1643#endif
1644}
1645
1646static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1647{
1648 if (rlim->rlim_cur == RLIM_INFINITY)
1649 rlim64->rlim_cur = RLIM64_INFINITY;
1650 else
1651 rlim64->rlim_cur = rlim->rlim_cur;
1652 if (rlim->rlim_max == RLIM_INFINITY)
1653 rlim64->rlim_max = RLIM64_INFINITY;
1654 else
1655 rlim64->rlim_max = rlim->rlim_max;
1656}
1657
1658static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1659{
1660 if (rlim64_is_infinity(rlim64->rlim_cur))
1661 rlim->rlim_cur = RLIM_INFINITY;
1662 else
1663 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1664 if (rlim64_is_infinity(rlim64->rlim_max))
1665 rlim->rlim_max = RLIM_INFINITY;
1666 else
1667 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1668}
1669
1670/* rcu lock must be held */
1671static int check_prlimit_permission(struct task_struct *task,
1672 unsigned int flags)
1673{
1674 const struct cred *cred = current_cred(), *tcred;
1675 bool id_match;
1676
1677 if (current == task)
1678 return 0;
1679
1680 tcred = __task_cred(task);
1681 id_match = (uid_eq(cred->uid, tcred->euid) &&
1682 uid_eq(cred->uid, tcred->suid) &&
1683 uid_eq(cred->uid, tcred->uid) &&
1684 gid_eq(cred->gid, tcred->egid) &&
1685 gid_eq(cred->gid, tcred->sgid) &&
1686 gid_eq(cred->gid, tcred->gid));
1687 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1688 return -EPERM;
1689
1690 return security_task_prlimit(cred, tcred, flags);
1691}
1692
1693SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1694 const struct rlimit64 __user *, new_rlim,
1695 struct rlimit64 __user *, old_rlim)
1696{
1697 struct rlimit64 old64, new64;
1698 struct rlimit old, new;
1699 struct task_struct *tsk;
1700 unsigned int checkflags = 0;
1701 int ret;
1702
1703 if (old_rlim)
1704 checkflags |= LSM_PRLIMIT_READ;
1705
1706 if (new_rlim) {
1707 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1708 return -EFAULT;
1709 rlim64_to_rlim(&new64, &new);
1710 checkflags |= LSM_PRLIMIT_WRITE;
1711 }
1712
1713 rcu_read_lock();
1714 tsk = pid ? find_task_by_vpid(pid) : current;
1715 if (!tsk) {
1716 rcu_read_unlock();
1717 return -ESRCH;
1718 }
1719 ret = check_prlimit_permission(tsk, checkflags);
1720 if (ret) {
1721 rcu_read_unlock();
1722 return ret;
1723 }
1724 get_task_struct(tsk);
1725 rcu_read_unlock();
1726
1727 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1728 old_rlim ? &old : NULL);
1729
1730 if (!ret && old_rlim) {
1731 rlim_to_rlim64(&old, &old64);
1732 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1733 ret = -EFAULT;
1734 }
1735
1736 put_task_struct(tsk);
1737 return ret;
1738}
1739
1740SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1741{
1742 struct rlimit new_rlim;
1743
1744 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1745 return -EFAULT;
1746 return do_prlimit(current, resource, &new_rlim, NULL);
1747}
1748
1749/*
1750 * It would make sense to put struct rusage in the task_struct,
1751 * except that would make the task_struct be *really big*. After
1752 * task_struct gets moved into malloc'ed memory, it would
1753 * make sense to do this. It will make moving the rest of the information
1754 * a lot simpler! (Which we're not doing right now because we're not
1755 * measuring them yet).
1756 *
1757 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1758 * races with threads incrementing their own counters. But since word
1759 * reads are atomic, we either get new values or old values and we don't
1760 * care which for the sums. We always take the siglock to protect reading
1761 * the c* fields from p->signal from races with exit.c updating those
1762 * fields when reaping, so a sample either gets all the additions of a
1763 * given child after it's reaped, or none so this sample is before reaping.
1764 *
1765 * Locking:
1766 * We need to take the siglock for CHILDEREN, SELF and BOTH
1767 * for the cases current multithreaded, non-current single threaded
1768 * non-current multithreaded. Thread traversal is now safe with
1769 * the siglock held.
1770 * Strictly speaking, we donot need to take the siglock if we are current and
1771 * single threaded, as no one else can take our signal_struct away, no one
1772 * else can reap the children to update signal->c* counters, and no one else
1773 * can race with the signal-> fields. If we do not take any lock, the
1774 * signal-> fields could be read out of order while another thread was just
1775 * exiting. So we should place a read memory barrier when we avoid the lock.
1776 * On the writer side, write memory barrier is implied in __exit_signal
1777 * as __exit_signal releases the siglock spinlock after updating the signal->
1778 * fields. But we don't do this yet to keep things simple.
1779 *
1780 */
1781
1782static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1783{
1784 r->ru_nvcsw += t->nvcsw;
1785 r->ru_nivcsw += t->nivcsw;
1786 r->ru_minflt += t->min_flt;
1787 r->ru_majflt += t->maj_flt;
1788 r->ru_inblock += task_io_get_inblock(t);
1789 r->ru_oublock += task_io_get_oublock(t);
1790}
1791
1792void getrusage(struct task_struct *p, int who, struct rusage *r)
1793{
1794 struct task_struct *t;
1795 unsigned long flags;
1796 u64 tgutime, tgstime, utime, stime;
1797 unsigned long maxrss;
1798 struct mm_struct *mm;
1799 struct signal_struct *sig = p->signal;
1800 unsigned int seq = 0;
1801
1802retry:
1803 memset(r, 0, sizeof(*r));
1804 utime = stime = 0;
1805 maxrss = 0;
1806
1807 if (who == RUSAGE_THREAD) {
1808 task_cputime_adjusted(current, &utime, &stime);
1809 accumulate_thread_rusage(p, r);
1810 maxrss = sig->maxrss;
1811 goto out_thread;
1812 }
1813
1814 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
1815
1816 switch (who) {
1817 case RUSAGE_BOTH:
1818 case RUSAGE_CHILDREN:
1819 utime = sig->cutime;
1820 stime = sig->cstime;
1821 r->ru_nvcsw = sig->cnvcsw;
1822 r->ru_nivcsw = sig->cnivcsw;
1823 r->ru_minflt = sig->cmin_flt;
1824 r->ru_majflt = sig->cmaj_flt;
1825 r->ru_inblock = sig->cinblock;
1826 r->ru_oublock = sig->coublock;
1827 maxrss = sig->cmaxrss;
1828
1829 if (who == RUSAGE_CHILDREN)
1830 break;
1831 fallthrough;
1832
1833 case RUSAGE_SELF:
1834 r->ru_nvcsw += sig->nvcsw;
1835 r->ru_nivcsw += sig->nivcsw;
1836 r->ru_minflt += sig->min_flt;
1837 r->ru_majflt += sig->maj_flt;
1838 r->ru_inblock += sig->inblock;
1839 r->ru_oublock += sig->oublock;
1840 if (maxrss < sig->maxrss)
1841 maxrss = sig->maxrss;
1842
1843 rcu_read_lock();
1844 __for_each_thread(sig, t)
1845 accumulate_thread_rusage(t, r);
1846 rcu_read_unlock();
1847
1848 break;
1849
1850 default:
1851 BUG();
1852 }
1853
1854 if (need_seqretry(&sig->stats_lock, seq)) {
1855 seq = 1;
1856 goto retry;
1857 }
1858 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
1859
1860 if (who == RUSAGE_CHILDREN)
1861 goto out_children;
1862
1863 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1864 utime += tgutime;
1865 stime += tgstime;
1866
1867out_thread:
1868 mm = get_task_mm(p);
1869 if (mm) {
1870 setmax_mm_hiwater_rss(&maxrss, mm);
1871 mmput(mm);
1872 }
1873
1874out_children:
1875 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1876 r->ru_utime = ns_to_kernel_old_timeval(utime);
1877 r->ru_stime = ns_to_kernel_old_timeval(stime);
1878}
1879
1880SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1881{
1882 struct rusage r;
1883
1884 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1885 who != RUSAGE_THREAD)
1886 return -EINVAL;
1887
1888 getrusage(current, who, &r);
1889 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1890}
1891
1892#ifdef CONFIG_COMPAT
1893COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1894{
1895 struct rusage r;
1896
1897 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1898 who != RUSAGE_THREAD)
1899 return -EINVAL;
1900
1901 getrusage(current, who, &r);
1902 return put_compat_rusage(&r, ru);
1903}
1904#endif
1905
1906SYSCALL_DEFINE1(umask, int, mask)
1907{
1908 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1909 return mask;
1910}
1911
1912static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1913{
1914 CLASS(fd, exe)(fd);
1915 struct inode *inode;
1916 int err;
1917
1918 if (fd_empty(exe))
1919 return -EBADF;
1920
1921 inode = file_inode(fd_file(exe));
1922
1923 /*
1924 * Because the original mm->exe_file points to executable file, make
1925 * sure that this one is executable as well, to avoid breaking an
1926 * overall picture.
1927 */
1928 if (!S_ISREG(inode->i_mode) || path_noexec(&fd_file(exe)->f_path))
1929 return -EACCES;
1930
1931 err = file_permission(fd_file(exe), MAY_EXEC);
1932 if (err)
1933 return err;
1934
1935 return replace_mm_exe_file(mm, fd_file(exe));
1936}
1937
1938/*
1939 * Check arithmetic relations of passed addresses.
1940 *
1941 * WARNING: we don't require any capability here so be very careful
1942 * in what is allowed for modification from userspace.
1943 */
1944static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1945{
1946 unsigned long mmap_max_addr = TASK_SIZE;
1947 int error = -EINVAL, i;
1948
1949 static const unsigned char offsets[] = {
1950 offsetof(struct prctl_mm_map, start_code),
1951 offsetof(struct prctl_mm_map, end_code),
1952 offsetof(struct prctl_mm_map, start_data),
1953 offsetof(struct prctl_mm_map, end_data),
1954 offsetof(struct prctl_mm_map, start_brk),
1955 offsetof(struct prctl_mm_map, brk),
1956 offsetof(struct prctl_mm_map, start_stack),
1957 offsetof(struct prctl_mm_map, arg_start),
1958 offsetof(struct prctl_mm_map, arg_end),
1959 offsetof(struct prctl_mm_map, env_start),
1960 offsetof(struct prctl_mm_map, env_end),
1961 };
1962
1963 /*
1964 * Make sure the members are not somewhere outside
1965 * of allowed address space.
1966 */
1967 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1968 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1969
1970 if ((unsigned long)val >= mmap_max_addr ||
1971 (unsigned long)val < mmap_min_addr)
1972 goto out;
1973 }
1974
1975 /*
1976 * Make sure the pairs are ordered.
1977 */
1978#define __prctl_check_order(__m1, __op, __m2) \
1979 ((unsigned long)prctl_map->__m1 __op \
1980 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1981 error = __prctl_check_order(start_code, <, end_code);
1982 error |= __prctl_check_order(start_data,<=, end_data);
1983 error |= __prctl_check_order(start_brk, <=, brk);
1984 error |= __prctl_check_order(arg_start, <=, arg_end);
1985 error |= __prctl_check_order(env_start, <=, env_end);
1986 if (error)
1987 goto out;
1988#undef __prctl_check_order
1989
1990 error = -EINVAL;
1991
1992 /*
1993 * Neither we should allow to override limits if they set.
1994 */
1995 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1996 prctl_map->start_brk, prctl_map->end_data,
1997 prctl_map->start_data))
1998 goto out;
1999
2000 error = 0;
2001out:
2002 return error;
2003}
2004
2005#ifdef CONFIG_CHECKPOINT_RESTORE
2006static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
2007{
2008 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
2009 unsigned long user_auxv[AT_VECTOR_SIZE];
2010 struct mm_struct *mm = current->mm;
2011 int error;
2012
2013 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2014 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
2015
2016 if (opt == PR_SET_MM_MAP_SIZE)
2017 return put_user((unsigned int)sizeof(prctl_map),
2018 (unsigned int __user *)addr);
2019
2020 if (data_size != sizeof(prctl_map))
2021 return -EINVAL;
2022
2023 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2024 return -EFAULT;
2025
2026 error = validate_prctl_map_addr(&prctl_map);
2027 if (error)
2028 return error;
2029
2030 if (prctl_map.auxv_size) {
2031 /*
2032 * Someone is trying to cheat the auxv vector.
2033 */
2034 if (!prctl_map.auxv ||
2035 prctl_map.auxv_size > sizeof(mm->saved_auxv))
2036 return -EINVAL;
2037
2038 memset(user_auxv, 0, sizeof(user_auxv));
2039 if (copy_from_user(user_auxv,
2040 (const void __user *)prctl_map.auxv,
2041 prctl_map.auxv_size))
2042 return -EFAULT;
2043
2044 /* Last entry must be AT_NULL as specification requires */
2045 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2046 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2047 }
2048
2049 if (prctl_map.exe_fd != (u32)-1) {
2050 /*
2051 * Check if the current user is checkpoint/restore capable.
2052 * At the time of this writing, it checks for CAP_SYS_ADMIN
2053 * or CAP_CHECKPOINT_RESTORE.
2054 * Note that a user with access to ptrace can masquerade an
2055 * arbitrary program as any executable, even setuid ones.
2056 * This may have implications in the tomoyo subsystem.
2057 */
2058 if (!checkpoint_restore_ns_capable(current_user_ns()))
2059 return -EPERM;
2060
2061 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2062 if (error)
2063 return error;
2064 }
2065
2066 /*
2067 * arg_lock protects concurrent updates but we still need mmap_lock for
2068 * read to exclude races with sys_brk.
2069 */
2070 mmap_read_lock(mm);
2071
2072 /*
2073 * We don't validate if these members are pointing to
2074 * real present VMAs because application may have correspond
2075 * VMAs already unmapped and kernel uses these members for statistics
2076 * output in procfs mostly, except
2077 *
2078 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2079 * for VMAs when updating these members so anything wrong written
2080 * here cause kernel to swear at userspace program but won't lead
2081 * to any problem in kernel itself
2082 */
2083
2084 spin_lock(&mm->arg_lock);
2085 mm->start_code = prctl_map.start_code;
2086 mm->end_code = prctl_map.end_code;
2087 mm->start_data = prctl_map.start_data;
2088 mm->end_data = prctl_map.end_data;
2089 mm->start_brk = prctl_map.start_brk;
2090 mm->brk = prctl_map.brk;
2091 mm->start_stack = prctl_map.start_stack;
2092 mm->arg_start = prctl_map.arg_start;
2093 mm->arg_end = prctl_map.arg_end;
2094 mm->env_start = prctl_map.env_start;
2095 mm->env_end = prctl_map.env_end;
2096 spin_unlock(&mm->arg_lock);
2097
2098 /*
2099 * Note this update of @saved_auxv is lockless thus
2100 * if someone reads this member in procfs while we're
2101 * updating -- it may get partly updated results. It's
2102 * known and acceptable trade off: we leave it as is to
2103 * not introduce additional locks here making the kernel
2104 * more complex.
2105 */
2106 if (prctl_map.auxv_size)
2107 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2108
2109 mmap_read_unlock(mm);
2110 return 0;
2111}
2112#endif /* CONFIG_CHECKPOINT_RESTORE */
2113
2114static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2115 unsigned long len)
2116{
2117 /*
2118 * This doesn't move the auxiliary vector itself since it's pinned to
2119 * mm_struct, but it permits filling the vector with new values. It's
2120 * up to the caller to provide sane values here, otherwise userspace
2121 * tools which use this vector might be unhappy.
2122 */
2123 unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2124
2125 if (len > sizeof(user_auxv))
2126 return -EINVAL;
2127
2128 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2129 return -EFAULT;
2130
2131 /* Make sure the last entry is always AT_NULL */
2132 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2133 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2134
2135 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2136
2137 task_lock(current);
2138 memcpy(mm->saved_auxv, user_auxv, len);
2139 task_unlock(current);
2140
2141 return 0;
2142}
2143
2144static int prctl_set_mm(int opt, unsigned long addr,
2145 unsigned long arg4, unsigned long arg5)
2146{
2147 struct mm_struct *mm = current->mm;
2148 struct prctl_mm_map prctl_map = {
2149 .auxv = NULL,
2150 .auxv_size = 0,
2151 .exe_fd = -1,
2152 };
2153 struct vm_area_struct *vma;
2154 int error;
2155
2156 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2157 opt != PR_SET_MM_MAP &&
2158 opt != PR_SET_MM_MAP_SIZE)))
2159 return -EINVAL;
2160
2161#ifdef CONFIG_CHECKPOINT_RESTORE
2162 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2163 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2164#endif
2165
2166 if (!capable(CAP_SYS_RESOURCE))
2167 return -EPERM;
2168
2169 if (opt == PR_SET_MM_EXE_FILE)
2170 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2171
2172 if (opt == PR_SET_MM_AUXV)
2173 return prctl_set_auxv(mm, addr, arg4);
2174
2175 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2176 return -EINVAL;
2177
2178 error = -EINVAL;
2179
2180 /*
2181 * arg_lock protects concurrent updates of arg boundaries, we need
2182 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2183 * validation.
2184 */
2185 mmap_read_lock(mm);
2186 vma = find_vma(mm, addr);
2187
2188 spin_lock(&mm->arg_lock);
2189 prctl_map.start_code = mm->start_code;
2190 prctl_map.end_code = mm->end_code;
2191 prctl_map.start_data = mm->start_data;
2192 prctl_map.end_data = mm->end_data;
2193 prctl_map.start_brk = mm->start_brk;
2194 prctl_map.brk = mm->brk;
2195 prctl_map.start_stack = mm->start_stack;
2196 prctl_map.arg_start = mm->arg_start;
2197 prctl_map.arg_end = mm->arg_end;
2198 prctl_map.env_start = mm->env_start;
2199 prctl_map.env_end = mm->env_end;
2200
2201 switch (opt) {
2202 case PR_SET_MM_START_CODE:
2203 prctl_map.start_code = addr;
2204 break;
2205 case PR_SET_MM_END_CODE:
2206 prctl_map.end_code = addr;
2207 break;
2208 case PR_SET_MM_START_DATA:
2209 prctl_map.start_data = addr;
2210 break;
2211 case PR_SET_MM_END_DATA:
2212 prctl_map.end_data = addr;
2213 break;
2214 case PR_SET_MM_START_STACK:
2215 prctl_map.start_stack = addr;
2216 break;
2217 case PR_SET_MM_START_BRK:
2218 prctl_map.start_brk = addr;
2219 break;
2220 case PR_SET_MM_BRK:
2221 prctl_map.brk = addr;
2222 break;
2223 case PR_SET_MM_ARG_START:
2224 prctl_map.arg_start = addr;
2225 break;
2226 case PR_SET_MM_ARG_END:
2227 prctl_map.arg_end = addr;
2228 break;
2229 case PR_SET_MM_ENV_START:
2230 prctl_map.env_start = addr;
2231 break;
2232 case PR_SET_MM_ENV_END:
2233 prctl_map.env_end = addr;
2234 break;
2235 default:
2236 goto out;
2237 }
2238
2239 error = validate_prctl_map_addr(&prctl_map);
2240 if (error)
2241 goto out;
2242
2243 switch (opt) {
2244 /*
2245 * If command line arguments and environment
2246 * are placed somewhere else on stack, we can
2247 * set them up here, ARG_START/END to setup
2248 * command line arguments and ENV_START/END
2249 * for environment.
2250 */
2251 case PR_SET_MM_START_STACK:
2252 case PR_SET_MM_ARG_START:
2253 case PR_SET_MM_ARG_END:
2254 case PR_SET_MM_ENV_START:
2255 case PR_SET_MM_ENV_END:
2256 if (!vma) {
2257 error = -EFAULT;
2258 goto out;
2259 }
2260 }
2261
2262 mm->start_code = prctl_map.start_code;
2263 mm->end_code = prctl_map.end_code;
2264 mm->start_data = prctl_map.start_data;
2265 mm->end_data = prctl_map.end_data;
2266 mm->start_brk = prctl_map.start_brk;
2267 mm->brk = prctl_map.brk;
2268 mm->start_stack = prctl_map.start_stack;
2269 mm->arg_start = prctl_map.arg_start;
2270 mm->arg_end = prctl_map.arg_end;
2271 mm->env_start = prctl_map.env_start;
2272 mm->env_end = prctl_map.env_end;
2273
2274 error = 0;
2275out:
2276 spin_unlock(&mm->arg_lock);
2277 mmap_read_unlock(mm);
2278 return error;
2279}
2280
2281#ifdef CONFIG_CHECKPOINT_RESTORE
2282static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2283{
2284 return put_user(me->clear_child_tid, tid_addr);
2285}
2286#else
2287static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2288{
2289 return -EINVAL;
2290}
2291#endif
2292
2293static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2294{
2295 /*
2296 * If task has has_child_subreaper - all its descendants
2297 * already have these flag too and new descendants will
2298 * inherit it on fork, skip them.
2299 *
2300 * If we've found child_reaper - skip descendants in
2301 * it's subtree as they will never get out pidns.
2302 */
2303 if (p->signal->has_child_subreaper ||
2304 is_child_reaper(task_pid(p)))
2305 return 0;
2306
2307 p->signal->has_child_subreaper = 1;
2308 return 1;
2309}
2310
2311int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2312{
2313 return -EINVAL;
2314}
2315
2316int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2317 unsigned long ctrl)
2318{
2319 return -EINVAL;
2320}
2321
2322int __weak arch_get_shadow_stack_status(struct task_struct *t, unsigned long __user *status)
2323{
2324 return -EINVAL;
2325}
2326
2327int __weak arch_set_shadow_stack_status(struct task_struct *t, unsigned long status)
2328{
2329 return -EINVAL;
2330}
2331
2332int __weak arch_lock_shadow_stack_status(struct task_struct *t, unsigned long status)
2333{
2334 return -EINVAL;
2335}
2336
2337#define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2338
2339#ifdef CONFIG_ANON_VMA_NAME
2340
2341#define ANON_VMA_NAME_MAX_LEN 80
2342#define ANON_VMA_NAME_INVALID_CHARS "\\`$[]"
2343
2344static inline bool is_valid_name_char(char ch)
2345{
2346 /* printable ascii characters, excluding ANON_VMA_NAME_INVALID_CHARS */
2347 return ch > 0x1f && ch < 0x7f &&
2348 !strchr(ANON_VMA_NAME_INVALID_CHARS, ch);
2349}
2350
2351static int prctl_set_vma(unsigned long opt, unsigned long addr,
2352 unsigned long size, unsigned long arg)
2353{
2354 struct mm_struct *mm = current->mm;
2355 const char __user *uname;
2356 struct anon_vma_name *anon_name = NULL;
2357 int error;
2358
2359 switch (opt) {
2360 case PR_SET_VMA_ANON_NAME:
2361 uname = (const char __user *)arg;
2362 if (uname) {
2363 char *name, *pch;
2364
2365 name = strndup_user(uname, ANON_VMA_NAME_MAX_LEN);
2366 if (IS_ERR(name))
2367 return PTR_ERR(name);
2368
2369 for (pch = name; *pch != '\0'; pch++) {
2370 if (!is_valid_name_char(*pch)) {
2371 kfree(name);
2372 return -EINVAL;
2373 }
2374 }
2375 /* anon_vma has its own copy */
2376 anon_name = anon_vma_name_alloc(name);
2377 kfree(name);
2378 if (!anon_name)
2379 return -ENOMEM;
2380
2381 }
2382
2383 mmap_write_lock(mm);
2384 error = madvise_set_anon_name(mm, addr, size, anon_name);
2385 mmap_write_unlock(mm);
2386 anon_vma_name_put(anon_name);
2387 break;
2388 default:
2389 error = -EINVAL;
2390 }
2391
2392 return error;
2393}
2394
2395#else /* CONFIG_ANON_VMA_NAME */
2396static int prctl_set_vma(unsigned long opt, unsigned long start,
2397 unsigned long size, unsigned long arg)
2398{
2399 return -EINVAL;
2400}
2401#endif /* CONFIG_ANON_VMA_NAME */
2402
2403static inline unsigned long get_current_mdwe(void)
2404{
2405 unsigned long ret = 0;
2406
2407 if (test_bit(MMF_HAS_MDWE, ¤t->mm->flags))
2408 ret |= PR_MDWE_REFUSE_EXEC_GAIN;
2409 if (test_bit(MMF_HAS_MDWE_NO_INHERIT, ¤t->mm->flags))
2410 ret |= PR_MDWE_NO_INHERIT;
2411
2412 return ret;
2413}
2414
2415static inline int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
2416 unsigned long arg4, unsigned long arg5)
2417{
2418 unsigned long current_bits;
2419
2420 if (arg3 || arg4 || arg5)
2421 return -EINVAL;
2422
2423 if (bits & ~(PR_MDWE_REFUSE_EXEC_GAIN | PR_MDWE_NO_INHERIT))
2424 return -EINVAL;
2425
2426 /* NO_INHERIT only makes sense with REFUSE_EXEC_GAIN */
2427 if (bits & PR_MDWE_NO_INHERIT && !(bits & PR_MDWE_REFUSE_EXEC_GAIN))
2428 return -EINVAL;
2429
2430 /*
2431 * EOPNOTSUPP might be more appropriate here in principle, but
2432 * existing userspace depends on EINVAL specifically.
2433 */
2434 if (!arch_memory_deny_write_exec_supported())
2435 return -EINVAL;
2436
2437 current_bits = get_current_mdwe();
2438 if (current_bits && current_bits != bits)
2439 return -EPERM; /* Cannot unset the flags */
2440
2441 if (bits & PR_MDWE_NO_INHERIT)
2442 set_bit(MMF_HAS_MDWE_NO_INHERIT, ¤t->mm->flags);
2443 if (bits & PR_MDWE_REFUSE_EXEC_GAIN)
2444 set_bit(MMF_HAS_MDWE, ¤t->mm->flags);
2445
2446 return 0;
2447}
2448
2449static inline int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
2450 unsigned long arg4, unsigned long arg5)
2451{
2452 if (arg2 || arg3 || arg4 || arg5)
2453 return -EINVAL;
2454 return get_current_mdwe();
2455}
2456
2457static int prctl_get_auxv(void __user *addr, unsigned long len)
2458{
2459 struct mm_struct *mm = current->mm;
2460 unsigned long size = min_t(unsigned long, sizeof(mm->saved_auxv), len);
2461
2462 if (size && copy_to_user(addr, mm->saved_auxv, size))
2463 return -EFAULT;
2464 return sizeof(mm->saved_auxv);
2465}
2466
2467SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2468 unsigned long, arg4, unsigned long, arg5)
2469{
2470 struct task_struct *me = current;
2471 unsigned char comm[sizeof(me->comm)];
2472 long error;
2473
2474 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2475 if (error != -ENOSYS)
2476 return error;
2477
2478 error = 0;
2479 switch (option) {
2480 case PR_SET_PDEATHSIG:
2481 if (!valid_signal(arg2)) {
2482 error = -EINVAL;
2483 break;
2484 }
2485 me->pdeath_signal = arg2;
2486 break;
2487 case PR_GET_PDEATHSIG:
2488 error = put_user(me->pdeath_signal, (int __user *)arg2);
2489 break;
2490 case PR_GET_DUMPABLE:
2491 error = get_dumpable(me->mm);
2492 break;
2493 case PR_SET_DUMPABLE:
2494 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2495 error = -EINVAL;
2496 break;
2497 }
2498 set_dumpable(me->mm, arg2);
2499 break;
2500
2501 case PR_SET_UNALIGN:
2502 error = SET_UNALIGN_CTL(me, arg2);
2503 break;
2504 case PR_GET_UNALIGN:
2505 error = GET_UNALIGN_CTL(me, arg2);
2506 break;
2507 case PR_SET_FPEMU:
2508 error = SET_FPEMU_CTL(me, arg2);
2509 break;
2510 case PR_GET_FPEMU:
2511 error = GET_FPEMU_CTL(me, arg2);
2512 break;
2513 case PR_SET_FPEXC:
2514 error = SET_FPEXC_CTL(me, arg2);
2515 break;
2516 case PR_GET_FPEXC:
2517 error = GET_FPEXC_CTL(me, arg2);
2518 break;
2519 case PR_GET_TIMING:
2520 error = PR_TIMING_STATISTICAL;
2521 break;
2522 case PR_SET_TIMING:
2523 if (arg2 != PR_TIMING_STATISTICAL)
2524 error = -EINVAL;
2525 break;
2526 case PR_SET_NAME:
2527 comm[sizeof(me->comm) - 1] = 0;
2528 if (strncpy_from_user(comm, (char __user *)arg2,
2529 sizeof(me->comm) - 1) < 0)
2530 return -EFAULT;
2531 set_task_comm(me, comm);
2532 proc_comm_connector(me);
2533 break;
2534 case PR_GET_NAME:
2535 get_task_comm(comm, me);
2536 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2537 return -EFAULT;
2538 break;
2539 case PR_GET_ENDIAN:
2540 error = GET_ENDIAN(me, arg2);
2541 break;
2542 case PR_SET_ENDIAN:
2543 error = SET_ENDIAN(me, arg2);
2544 break;
2545 case PR_GET_SECCOMP:
2546 error = prctl_get_seccomp();
2547 break;
2548 case PR_SET_SECCOMP:
2549 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2550 break;
2551 case PR_GET_TSC:
2552 error = GET_TSC_CTL(arg2);
2553 break;
2554 case PR_SET_TSC:
2555 error = SET_TSC_CTL(arg2);
2556 break;
2557 case PR_TASK_PERF_EVENTS_DISABLE:
2558 error = perf_event_task_disable();
2559 break;
2560 case PR_TASK_PERF_EVENTS_ENABLE:
2561 error = perf_event_task_enable();
2562 break;
2563 case PR_GET_TIMERSLACK:
2564 if (current->timer_slack_ns > ULONG_MAX)
2565 error = ULONG_MAX;
2566 else
2567 error = current->timer_slack_ns;
2568 break;
2569 case PR_SET_TIMERSLACK:
2570 if (rt_or_dl_task_policy(current))
2571 break;
2572 if (arg2 <= 0)
2573 current->timer_slack_ns =
2574 current->default_timer_slack_ns;
2575 else
2576 current->timer_slack_ns = arg2;
2577 break;
2578 case PR_MCE_KILL:
2579 if (arg4 | arg5)
2580 return -EINVAL;
2581 switch (arg2) {
2582 case PR_MCE_KILL_CLEAR:
2583 if (arg3 != 0)
2584 return -EINVAL;
2585 current->flags &= ~PF_MCE_PROCESS;
2586 break;
2587 case PR_MCE_KILL_SET:
2588 current->flags |= PF_MCE_PROCESS;
2589 if (arg3 == PR_MCE_KILL_EARLY)
2590 current->flags |= PF_MCE_EARLY;
2591 else if (arg3 == PR_MCE_KILL_LATE)
2592 current->flags &= ~PF_MCE_EARLY;
2593 else if (arg3 == PR_MCE_KILL_DEFAULT)
2594 current->flags &=
2595 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2596 else
2597 return -EINVAL;
2598 break;
2599 default:
2600 return -EINVAL;
2601 }
2602 break;
2603 case PR_MCE_KILL_GET:
2604 if (arg2 | arg3 | arg4 | arg5)
2605 return -EINVAL;
2606 if (current->flags & PF_MCE_PROCESS)
2607 error = (current->flags & PF_MCE_EARLY) ?
2608 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2609 else
2610 error = PR_MCE_KILL_DEFAULT;
2611 break;
2612 case PR_SET_MM:
2613 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2614 break;
2615 case PR_GET_TID_ADDRESS:
2616 error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2617 break;
2618 case PR_SET_CHILD_SUBREAPER:
2619 me->signal->is_child_subreaper = !!arg2;
2620 if (!arg2)
2621 break;
2622
2623 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2624 break;
2625 case PR_GET_CHILD_SUBREAPER:
2626 error = put_user(me->signal->is_child_subreaper,
2627 (int __user *)arg2);
2628 break;
2629 case PR_SET_NO_NEW_PRIVS:
2630 if (arg2 != 1 || arg3 || arg4 || arg5)
2631 return -EINVAL;
2632
2633 task_set_no_new_privs(current);
2634 break;
2635 case PR_GET_NO_NEW_PRIVS:
2636 if (arg2 || arg3 || arg4 || arg5)
2637 return -EINVAL;
2638 return task_no_new_privs(current) ? 1 : 0;
2639 case PR_GET_THP_DISABLE:
2640 if (arg2 || arg3 || arg4 || arg5)
2641 return -EINVAL;
2642 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2643 break;
2644 case PR_SET_THP_DISABLE:
2645 if (arg3 || arg4 || arg5)
2646 return -EINVAL;
2647 if (mmap_write_lock_killable(me->mm))
2648 return -EINTR;
2649 if (arg2)
2650 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2651 else
2652 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2653 mmap_write_unlock(me->mm);
2654 break;
2655 case PR_MPX_ENABLE_MANAGEMENT:
2656 case PR_MPX_DISABLE_MANAGEMENT:
2657 /* No longer implemented: */
2658 return -EINVAL;
2659 case PR_SET_FP_MODE:
2660 error = SET_FP_MODE(me, arg2);
2661 break;
2662 case PR_GET_FP_MODE:
2663 error = GET_FP_MODE(me);
2664 break;
2665 case PR_SVE_SET_VL:
2666 error = SVE_SET_VL(arg2);
2667 break;
2668 case PR_SVE_GET_VL:
2669 error = SVE_GET_VL();
2670 break;
2671 case PR_SME_SET_VL:
2672 error = SME_SET_VL(arg2);
2673 break;
2674 case PR_SME_GET_VL:
2675 error = SME_GET_VL();
2676 break;
2677 case PR_GET_SPECULATION_CTRL:
2678 if (arg3 || arg4 || arg5)
2679 return -EINVAL;
2680 error = arch_prctl_spec_ctrl_get(me, arg2);
2681 break;
2682 case PR_SET_SPECULATION_CTRL:
2683 if (arg4 || arg5)
2684 return -EINVAL;
2685 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2686 break;
2687 case PR_PAC_RESET_KEYS:
2688 if (arg3 || arg4 || arg5)
2689 return -EINVAL;
2690 error = PAC_RESET_KEYS(me, arg2);
2691 break;
2692 case PR_PAC_SET_ENABLED_KEYS:
2693 if (arg4 || arg5)
2694 return -EINVAL;
2695 error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2696 break;
2697 case PR_PAC_GET_ENABLED_KEYS:
2698 if (arg2 || arg3 || arg4 || arg5)
2699 return -EINVAL;
2700 error = PAC_GET_ENABLED_KEYS(me);
2701 break;
2702 case PR_SET_TAGGED_ADDR_CTRL:
2703 if (arg3 || arg4 || arg5)
2704 return -EINVAL;
2705 error = SET_TAGGED_ADDR_CTRL(arg2);
2706 break;
2707 case PR_GET_TAGGED_ADDR_CTRL:
2708 if (arg2 || arg3 || arg4 || arg5)
2709 return -EINVAL;
2710 error = GET_TAGGED_ADDR_CTRL();
2711 break;
2712 case PR_SET_IO_FLUSHER:
2713 if (!capable(CAP_SYS_RESOURCE))
2714 return -EPERM;
2715
2716 if (arg3 || arg4 || arg5)
2717 return -EINVAL;
2718
2719 if (arg2 == 1)
2720 current->flags |= PR_IO_FLUSHER;
2721 else if (!arg2)
2722 current->flags &= ~PR_IO_FLUSHER;
2723 else
2724 return -EINVAL;
2725 break;
2726 case PR_GET_IO_FLUSHER:
2727 if (!capable(CAP_SYS_RESOURCE))
2728 return -EPERM;
2729
2730 if (arg2 || arg3 || arg4 || arg5)
2731 return -EINVAL;
2732
2733 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2734 break;
2735 case PR_SET_SYSCALL_USER_DISPATCH:
2736 error = set_syscall_user_dispatch(arg2, arg3, arg4,
2737 (char __user *) arg5);
2738 break;
2739#ifdef CONFIG_SCHED_CORE
2740 case PR_SCHED_CORE:
2741 error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2742 break;
2743#endif
2744 case PR_SET_MDWE:
2745 error = prctl_set_mdwe(arg2, arg3, arg4, arg5);
2746 break;
2747 case PR_GET_MDWE:
2748 error = prctl_get_mdwe(arg2, arg3, arg4, arg5);
2749 break;
2750 case PR_PPC_GET_DEXCR:
2751 if (arg3 || arg4 || arg5)
2752 return -EINVAL;
2753 error = PPC_GET_DEXCR_ASPECT(me, arg2);
2754 break;
2755 case PR_PPC_SET_DEXCR:
2756 if (arg4 || arg5)
2757 return -EINVAL;
2758 error = PPC_SET_DEXCR_ASPECT(me, arg2, arg3);
2759 break;
2760 case PR_SET_VMA:
2761 error = prctl_set_vma(arg2, arg3, arg4, arg5);
2762 break;
2763 case PR_GET_AUXV:
2764 if (arg4 || arg5)
2765 return -EINVAL;
2766 error = prctl_get_auxv((void __user *)arg2, arg3);
2767 break;
2768#ifdef CONFIG_KSM
2769 case PR_SET_MEMORY_MERGE:
2770 if (arg3 || arg4 || arg5)
2771 return -EINVAL;
2772 if (mmap_write_lock_killable(me->mm))
2773 return -EINTR;
2774
2775 if (arg2)
2776 error = ksm_enable_merge_any(me->mm);
2777 else
2778 error = ksm_disable_merge_any(me->mm);
2779 mmap_write_unlock(me->mm);
2780 break;
2781 case PR_GET_MEMORY_MERGE:
2782 if (arg2 || arg3 || arg4 || arg5)
2783 return -EINVAL;
2784
2785 error = !!test_bit(MMF_VM_MERGE_ANY, &me->mm->flags);
2786 break;
2787#endif
2788 case PR_RISCV_V_SET_CONTROL:
2789 error = RISCV_V_SET_CONTROL(arg2);
2790 break;
2791 case PR_RISCV_V_GET_CONTROL:
2792 error = RISCV_V_GET_CONTROL();
2793 break;
2794 case PR_RISCV_SET_ICACHE_FLUSH_CTX:
2795 error = RISCV_SET_ICACHE_FLUSH_CTX(arg2, arg3);
2796 break;
2797 case PR_GET_SHADOW_STACK_STATUS:
2798 if (arg3 || arg4 || arg5)
2799 return -EINVAL;
2800 error = arch_get_shadow_stack_status(me, (unsigned long __user *) arg2);
2801 break;
2802 case PR_SET_SHADOW_STACK_STATUS:
2803 if (arg3 || arg4 || arg5)
2804 return -EINVAL;
2805 error = arch_set_shadow_stack_status(me, arg2);
2806 break;
2807 case PR_LOCK_SHADOW_STACK_STATUS:
2808 if (arg3 || arg4 || arg5)
2809 return -EINVAL;
2810 error = arch_lock_shadow_stack_status(me, arg2);
2811 break;
2812 default:
2813 error = -EINVAL;
2814 break;
2815 }
2816 return error;
2817}
2818
2819SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2820 struct getcpu_cache __user *, unused)
2821{
2822 int err = 0;
2823 int cpu = raw_smp_processor_id();
2824
2825 if (cpup)
2826 err |= put_user(cpu, cpup);
2827 if (nodep)
2828 err |= put_user(cpu_to_node(cpu), nodep);
2829 return err ? -EFAULT : 0;
2830}
2831
2832/**
2833 * do_sysinfo - fill in sysinfo struct
2834 * @info: pointer to buffer to fill
2835 */
2836static int do_sysinfo(struct sysinfo *info)
2837{
2838 unsigned long mem_total, sav_total;
2839 unsigned int mem_unit, bitcount;
2840 struct timespec64 tp;
2841
2842 memset(info, 0, sizeof(struct sysinfo));
2843
2844 ktime_get_boottime_ts64(&tp);
2845 timens_add_boottime(&tp);
2846 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2847
2848 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2849
2850 info->procs = nr_threads;
2851
2852 si_meminfo(info);
2853 si_swapinfo(info);
2854
2855 /*
2856 * If the sum of all the available memory (i.e. ram + swap)
2857 * is less than can be stored in a 32 bit unsigned long then
2858 * we can be binary compatible with 2.2.x kernels. If not,
2859 * well, in that case 2.2.x was broken anyways...
2860 *
2861 * -Erik Andersen <andersee@debian.org>
2862 */
2863
2864 mem_total = info->totalram + info->totalswap;
2865 if (mem_total < info->totalram || mem_total < info->totalswap)
2866 goto out;
2867 bitcount = 0;
2868 mem_unit = info->mem_unit;
2869 while (mem_unit > 1) {
2870 bitcount++;
2871 mem_unit >>= 1;
2872 sav_total = mem_total;
2873 mem_total <<= 1;
2874 if (mem_total < sav_total)
2875 goto out;
2876 }
2877
2878 /*
2879 * If mem_total did not overflow, multiply all memory values by
2880 * info->mem_unit and set it to 1. This leaves things compatible
2881 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2882 * kernels...
2883 */
2884
2885 info->mem_unit = 1;
2886 info->totalram <<= bitcount;
2887 info->freeram <<= bitcount;
2888 info->sharedram <<= bitcount;
2889 info->bufferram <<= bitcount;
2890 info->totalswap <<= bitcount;
2891 info->freeswap <<= bitcount;
2892 info->totalhigh <<= bitcount;
2893 info->freehigh <<= bitcount;
2894
2895out:
2896 return 0;
2897}
2898
2899SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2900{
2901 struct sysinfo val;
2902
2903 do_sysinfo(&val);
2904
2905 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2906 return -EFAULT;
2907
2908 return 0;
2909}
2910
2911#ifdef CONFIG_COMPAT
2912struct compat_sysinfo {
2913 s32 uptime;
2914 u32 loads[3];
2915 u32 totalram;
2916 u32 freeram;
2917 u32 sharedram;
2918 u32 bufferram;
2919 u32 totalswap;
2920 u32 freeswap;
2921 u16 procs;
2922 u16 pad;
2923 u32 totalhigh;
2924 u32 freehigh;
2925 u32 mem_unit;
2926 char _f[20-2*sizeof(u32)-sizeof(int)];
2927};
2928
2929COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2930{
2931 struct sysinfo s;
2932 struct compat_sysinfo s_32;
2933
2934 do_sysinfo(&s);
2935
2936 /* Check to see if any memory value is too large for 32-bit and scale
2937 * down if needed
2938 */
2939 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2940 int bitcount = 0;
2941
2942 while (s.mem_unit < PAGE_SIZE) {
2943 s.mem_unit <<= 1;
2944 bitcount++;
2945 }
2946
2947 s.totalram >>= bitcount;
2948 s.freeram >>= bitcount;
2949 s.sharedram >>= bitcount;
2950 s.bufferram >>= bitcount;
2951 s.totalswap >>= bitcount;
2952 s.freeswap >>= bitcount;
2953 s.totalhigh >>= bitcount;
2954 s.freehigh >>= bitcount;
2955 }
2956
2957 memset(&s_32, 0, sizeof(s_32));
2958 s_32.uptime = s.uptime;
2959 s_32.loads[0] = s.loads[0];
2960 s_32.loads[1] = s.loads[1];
2961 s_32.loads[2] = s.loads[2];
2962 s_32.totalram = s.totalram;
2963 s_32.freeram = s.freeram;
2964 s_32.sharedram = s.sharedram;
2965 s_32.bufferram = s.bufferram;
2966 s_32.totalswap = s.totalswap;
2967 s_32.freeswap = s.freeswap;
2968 s_32.procs = s.procs;
2969 s_32.totalhigh = s.totalhigh;
2970 s_32.freehigh = s.freehigh;
2971 s_32.mem_unit = s.mem_unit;
2972 if (copy_to_user(info, &s_32, sizeof(s_32)))
2973 return -EFAULT;
2974 return 0;
2975}
2976#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);