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// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/kernel/sys.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/export.h>
9#include <linux/mm.h>
10#include <linux/mm_inline.h>
11#include <linux/utsname.h>
12#include <linux/mman.h>
13#include <linux/reboot.h>
14#include <linux/prctl.h>
15#include <linux/highuid.h>
16#include <linux/fs.h>
17#include <linux/kmod.h>
18#include <linux/ksm.h>
19#include <linux/perf_event.h>
20#include <linux/resource.h>
21#include <linux/kernel.h>
22#include <linux/workqueue.h>
23#include <linux/capability.h>
24#include <linux/device.h>
25#include <linux/key.h>
26#include <linux/times.h>
27#include <linux/posix-timers.h>
28#include <linux/security.h>
29#include <linux/random.h>
30#include <linux/suspend.h>
31#include <linux/tty.h>
32#include <linux/signal.h>
33#include <linux/cn_proc.h>
34#include <linux/getcpu.h>
35#include <linux/task_io_accounting_ops.h>
36#include <linux/seccomp.h>
37#include <linux/cpu.h>
38#include <linux/personality.h>
39#include <linux/ptrace.h>
40#include <linux/fs_struct.h>
41#include <linux/file.h>
42#include <linux/mount.h>
43#include <linux/gfp.h>
44#include <linux/syscore_ops.h>
45#include <linux/version.h>
46#include <linux/ctype.h>
47#include <linux/syscall_user_dispatch.h>
48
49#include <linux/compat.h>
50#include <linux/syscalls.h>
51#include <linux/kprobes.h>
52#include <linux/user_namespace.h>
53#include <linux/time_namespace.h>
54#include <linux/binfmts.h>
55
56#include <linux/sched.h>
57#include <linux/sched/autogroup.h>
58#include <linux/sched/loadavg.h>
59#include <linux/sched/stat.h>
60#include <linux/sched/mm.h>
61#include <linux/sched/coredump.h>
62#include <linux/sched/task.h>
63#include <linux/sched/cputime.h>
64#include <linux/rcupdate.h>
65#include <linux/uidgid.h>
66#include <linux/cred.h>
67
68#include <linux/nospec.h>
69
70#include <linux/kmsg_dump.h>
71/* Move somewhere else to avoid recompiling? */
72#include <generated/utsrelease.h>
73
74#include <linux/uaccess.h>
75#include <asm/io.h>
76#include <asm/unistd.h>
77
78#include "uid16.h"
79
80#ifndef SET_UNALIGN_CTL
81# define SET_UNALIGN_CTL(a, b) (-EINVAL)
82#endif
83#ifndef GET_UNALIGN_CTL
84# define GET_UNALIGN_CTL(a, b) (-EINVAL)
85#endif
86#ifndef SET_FPEMU_CTL
87# define SET_FPEMU_CTL(a, b) (-EINVAL)
88#endif
89#ifndef GET_FPEMU_CTL
90# define GET_FPEMU_CTL(a, b) (-EINVAL)
91#endif
92#ifndef SET_FPEXC_CTL
93# define SET_FPEXC_CTL(a, b) (-EINVAL)
94#endif
95#ifndef GET_FPEXC_CTL
96# define GET_FPEXC_CTL(a, b) (-EINVAL)
97#endif
98#ifndef GET_ENDIAN
99# define GET_ENDIAN(a, b) (-EINVAL)
100#endif
101#ifndef SET_ENDIAN
102# define SET_ENDIAN(a, b) (-EINVAL)
103#endif
104#ifndef GET_TSC_CTL
105# define GET_TSC_CTL(a) (-EINVAL)
106#endif
107#ifndef SET_TSC_CTL
108# define SET_TSC_CTL(a) (-EINVAL)
109#endif
110#ifndef GET_FP_MODE
111# define GET_FP_MODE(a) (-EINVAL)
112#endif
113#ifndef SET_FP_MODE
114# define SET_FP_MODE(a,b) (-EINVAL)
115#endif
116#ifndef SVE_SET_VL
117# define SVE_SET_VL(a) (-EINVAL)
118#endif
119#ifndef SVE_GET_VL
120# define SVE_GET_VL() (-EINVAL)
121#endif
122#ifndef SME_SET_VL
123# define SME_SET_VL(a) (-EINVAL)
124#endif
125#ifndef SME_GET_VL
126# define SME_GET_VL() (-EINVAL)
127#endif
128#ifndef PAC_RESET_KEYS
129# define PAC_RESET_KEYS(a, b) (-EINVAL)
130#endif
131#ifndef PAC_SET_ENABLED_KEYS
132# define PAC_SET_ENABLED_KEYS(a, b, c) (-EINVAL)
133#endif
134#ifndef PAC_GET_ENABLED_KEYS
135# define PAC_GET_ENABLED_KEYS(a) (-EINVAL)
136#endif
137#ifndef SET_TAGGED_ADDR_CTRL
138# define SET_TAGGED_ADDR_CTRL(a) (-EINVAL)
139#endif
140#ifndef GET_TAGGED_ADDR_CTRL
141# define GET_TAGGED_ADDR_CTRL() (-EINVAL)
142#endif
143#ifndef RISCV_V_SET_CONTROL
144# define RISCV_V_SET_CONTROL(a) (-EINVAL)
145#endif
146#ifndef RISCV_V_GET_CONTROL
147# define RISCV_V_GET_CONTROL() (-EINVAL)
148#endif
149
150/*
151 * this is where the system-wide overflow UID and GID are defined, for
152 * architectures that now have 32-bit UID/GID but didn't in the past
153 */
154
155int overflowuid = DEFAULT_OVERFLOWUID;
156int overflowgid = DEFAULT_OVERFLOWGID;
157
158EXPORT_SYMBOL(overflowuid);
159EXPORT_SYMBOL(overflowgid);
160
161/*
162 * the same as above, but for filesystems which can only store a 16-bit
163 * UID and GID. as such, this is needed on all architectures
164 */
165
166int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
167int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
168
169EXPORT_SYMBOL(fs_overflowuid);
170EXPORT_SYMBOL(fs_overflowgid);
171
172/*
173 * Returns true if current's euid is same as p's uid or euid,
174 * or has CAP_SYS_NICE to p's user_ns.
175 *
176 * Called with rcu_read_lock, creds are safe
177 */
178static bool set_one_prio_perm(struct task_struct *p)
179{
180 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
181
182 if (uid_eq(pcred->uid, cred->euid) ||
183 uid_eq(pcred->euid, cred->euid))
184 return true;
185 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
186 return true;
187 return false;
188}
189
190/*
191 * set the priority of a task
192 * - the caller must hold the RCU read lock
193 */
194static int set_one_prio(struct task_struct *p, int niceval, int error)
195{
196 int no_nice;
197
198 if (!set_one_prio_perm(p)) {
199 error = -EPERM;
200 goto out;
201 }
202 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
203 error = -EACCES;
204 goto out;
205 }
206 no_nice = security_task_setnice(p, niceval);
207 if (no_nice) {
208 error = no_nice;
209 goto out;
210 }
211 if (error == -ESRCH)
212 error = 0;
213 set_user_nice(p, niceval);
214out:
215 return error;
216}
217
218SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
219{
220 struct task_struct *g, *p;
221 struct user_struct *user;
222 const struct cred *cred = current_cred();
223 int error = -EINVAL;
224 struct pid *pgrp;
225 kuid_t uid;
226
227 if (which > PRIO_USER || which < PRIO_PROCESS)
228 goto out;
229
230 /* normalize: avoid signed division (rounding problems) */
231 error = -ESRCH;
232 if (niceval < MIN_NICE)
233 niceval = MIN_NICE;
234 if (niceval > MAX_NICE)
235 niceval = MAX_NICE;
236
237 rcu_read_lock();
238 switch (which) {
239 case PRIO_PROCESS:
240 if (who)
241 p = find_task_by_vpid(who);
242 else
243 p = current;
244 if (p)
245 error = set_one_prio(p, niceval, error);
246 break;
247 case PRIO_PGRP:
248 if (who)
249 pgrp = find_vpid(who);
250 else
251 pgrp = task_pgrp(current);
252 read_lock(&tasklist_lock);
253 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
254 error = set_one_prio(p, niceval, error);
255 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
256 read_unlock(&tasklist_lock);
257 break;
258 case PRIO_USER:
259 uid = make_kuid(cred->user_ns, who);
260 user = cred->user;
261 if (!who)
262 uid = cred->uid;
263 else if (!uid_eq(uid, cred->uid)) {
264 user = find_user(uid);
265 if (!user)
266 goto out_unlock; /* No processes for this user */
267 }
268 for_each_process_thread(g, p) {
269 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
270 error = set_one_prio(p, niceval, error);
271 }
272 if (!uid_eq(uid, cred->uid))
273 free_uid(user); /* For find_user() */
274 break;
275 }
276out_unlock:
277 rcu_read_unlock();
278out:
279 return error;
280}
281
282/*
283 * Ugh. To avoid negative return values, "getpriority()" will
284 * not return the normal nice-value, but a negated value that
285 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
286 * to stay compatible.
287 */
288SYSCALL_DEFINE2(getpriority, int, which, int, who)
289{
290 struct task_struct *g, *p;
291 struct user_struct *user;
292 const struct cred *cred = current_cred();
293 long niceval, retval = -ESRCH;
294 struct pid *pgrp;
295 kuid_t uid;
296
297 if (which > PRIO_USER || which < PRIO_PROCESS)
298 return -EINVAL;
299
300 rcu_read_lock();
301 switch (which) {
302 case PRIO_PROCESS:
303 if (who)
304 p = find_task_by_vpid(who);
305 else
306 p = current;
307 if (p) {
308 niceval = nice_to_rlimit(task_nice(p));
309 if (niceval > retval)
310 retval = niceval;
311 }
312 break;
313 case PRIO_PGRP:
314 if (who)
315 pgrp = find_vpid(who);
316 else
317 pgrp = task_pgrp(current);
318 read_lock(&tasklist_lock);
319 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
320 niceval = nice_to_rlimit(task_nice(p));
321 if (niceval > retval)
322 retval = niceval;
323 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
324 read_unlock(&tasklist_lock);
325 break;
326 case PRIO_USER:
327 uid = make_kuid(cred->user_ns, who);
328 user = cred->user;
329 if (!who)
330 uid = cred->uid;
331 else if (!uid_eq(uid, cred->uid)) {
332 user = find_user(uid);
333 if (!user)
334 goto out_unlock; /* No processes for this user */
335 }
336 for_each_process_thread(g, p) {
337 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
338 niceval = nice_to_rlimit(task_nice(p));
339 if (niceval > retval)
340 retval = niceval;
341 }
342 }
343 if (!uid_eq(uid, cred->uid))
344 free_uid(user); /* for find_user() */
345 break;
346 }
347out_unlock:
348 rcu_read_unlock();
349
350 return retval;
351}
352
353/*
354 * Unprivileged users may change the real gid to the effective gid
355 * or vice versa. (BSD-style)
356 *
357 * If you set the real gid at all, or set the effective gid to a value not
358 * equal to the real gid, then the saved gid is set to the new effective gid.
359 *
360 * This makes it possible for a setgid program to completely drop its
361 * privileges, which is often a useful assertion to make when you are doing
362 * a security audit over a program.
363 *
364 * The general idea is that a program which uses just setregid() will be
365 * 100% compatible with BSD. A program which uses just setgid() will be
366 * 100% compatible with POSIX with saved IDs.
367 *
368 * SMP: There are not races, the GIDs are checked only by filesystem
369 * operations (as far as semantic preservation is concerned).
370 */
371#ifdef CONFIG_MULTIUSER
372long __sys_setregid(gid_t rgid, gid_t egid)
373{
374 struct user_namespace *ns = current_user_ns();
375 const struct cred *old;
376 struct cred *new;
377 int retval;
378 kgid_t krgid, kegid;
379
380 krgid = make_kgid(ns, rgid);
381 kegid = make_kgid(ns, egid);
382
383 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
384 return -EINVAL;
385 if ((egid != (gid_t) -1) && !gid_valid(kegid))
386 return -EINVAL;
387
388 new = prepare_creds();
389 if (!new)
390 return -ENOMEM;
391 old = current_cred();
392
393 retval = -EPERM;
394 if (rgid != (gid_t) -1) {
395 if (gid_eq(old->gid, krgid) ||
396 gid_eq(old->egid, krgid) ||
397 ns_capable_setid(old->user_ns, CAP_SETGID))
398 new->gid = krgid;
399 else
400 goto error;
401 }
402 if (egid != (gid_t) -1) {
403 if (gid_eq(old->gid, kegid) ||
404 gid_eq(old->egid, kegid) ||
405 gid_eq(old->sgid, kegid) ||
406 ns_capable_setid(old->user_ns, CAP_SETGID))
407 new->egid = kegid;
408 else
409 goto error;
410 }
411
412 if (rgid != (gid_t) -1 ||
413 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
414 new->sgid = new->egid;
415 new->fsgid = new->egid;
416
417 retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
418 if (retval < 0)
419 goto error;
420
421 return commit_creds(new);
422
423error:
424 abort_creds(new);
425 return retval;
426}
427
428SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
429{
430 return __sys_setregid(rgid, egid);
431}
432
433/*
434 * setgid() is implemented like SysV w/ SAVED_IDS
435 *
436 * SMP: Same implicit races as above.
437 */
438long __sys_setgid(gid_t gid)
439{
440 struct user_namespace *ns = current_user_ns();
441 const struct cred *old;
442 struct cred *new;
443 int retval;
444 kgid_t kgid;
445
446 kgid = make_kgid(ns, gid);
447 if (!gid_valid(kgid))
448 return -EINVAL;
449
450 new = prepare_creds();
451 if (!new)
452 return -ENOMEM;
453 old = current_cred();
454
455 retval = -EPERM;
456 if (ns_capable_setid(old->user_ns, CAP_SETGID))
457 new->gid = new->egid = new->sgid = new->fsgid = kgid;
458 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
459 new->egid = new->fsgid = kgid;
460 else
461 goto error;
462
463 retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
464 if (retval < 0)
465 goto error;
466
467 return commit_creds(new);
468
469error:
470 abort_creds(new);
471 return retval;
472}
473
474SYSCALL_DEFINE1(setgid, gid_t, gid)
475{
476 return __sys_setgid(gid);
477}
478
479/*
480 * change the user struct in a credentials set to match the new UID
481 */
482static int set_user(struct cred *new)
483{
484 struct user_struct *new_user;
485
486 new_user = alloc_uid(new->uid);
487 if (!new_user)
488 return -EAGAIN;
489
490 free_uid(new->user);
491 new->user = new_user;
492 return 0;
493}
494
495static void flag_nproc_exceeded(struct cred *new)
496{
497 if (new->ucounts == current_ucounts())
498 return;
499
500 /*
501 * We don't fail in case of NPROC limit excess here because too many
502 * poorly written programs don't check set*uid() return code, assuming
503 * it never fails if called by root. We may still enforce NPROC limit
504 * for programs doing set*uid()+execve() by harmlessly deferring the
505 * failure to the execve() stage.
506 */
507 if (is_rlimit_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
508 new->user != INIT_USER)
509 current->flags |= PF_NPROC_EXCEEDED;
510 else
511 current->flags &= ~PF_NPROC_EXCEEDED;
512}
513
514/*
515 * Unprivileged users may change the real uid to the effective uid
516 * or vice versa. (BSD-style)
517 *
518 * If you set the real uid at all, or set the effective uid to a value not
519 * equal to the real uid, then the saved uid is set to the new effective uid.
520 *
521 * This makes it possible for a setuid program to completely drop its
522 * privileges, which is often a useful assertion to make when you are doing
523 * a security audit over a program.
524 *
525 * The general idea is that a program which uses just setreuid() will be
526 * 100% compatible with BSD. A program which uses just setuid() will be
527 * 100% compatible with POSIX with saved IDs.
528 */
529long __sys_setreuid(uid_t ruid, uid_t euid)
530{
531 struct user_namespace *ns = current_user_ns();
532 const struct cred *old;
533 struct cred *new;
534 int retval;
535 kuid_t kruid, keuid;
536
537 kruid = make_kuid(ns, ruid);
538 keuid = make_kuid(ns, euid);
539
540 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
541 return -EINVAL;
542 if ((euid != (uid_t) -1) && !uid_valid(keuid))
543 return -EINVAL;
544
545 new = prepare_creds();
546 if (!new)
547 return -ENOMEM;
548 old = current_cred();
549
550 retval = -EPERM;
551 if (ruid != (uid_t) -1) {
552 new->uid = kruid;
553 if (!uid_eq(old->uid, kruid) &&
554 !uid_eq(old->euid, kruid) &&
555 !ns_capable_setid(old->user_ns, CAP_SETUID))
556 goto error;
557 }
558
559 if (euid != (uid_t) -1) {
560 new->euid = keuid;
561 if (!uid_eq(old->uid, keuid) &&
562 !uid_eq(old->euid, keuid) &&
563 !uid_eq(old->suid, keuid) &&
564 !ns_capable_setid(old->user_ns, CAP_SETUID))
565 goto error;
566 }
567
568 if (!uid_eq(new->uid, old->uid)) {
569 retval = set_user(new);
570 if (retval < 0)
571 goto error;
572 }
573 if (ruid != (uid_t) -1 ||
574 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
575 new->suid = new->euid;
576 new->fsuid = new->euid;
577
578 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
579 if (retval < 0)
580 goto error;
581
582 retval = set_cred_ucounts(new);
583 if (retval < 0)
584 goto error;
585
586 flag_nproc_exceeded(new);
587 return commit_creds(new);
588
589error:
590 abort_creds(new);
591 return retval;
592}
593
594SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
595{
596 return __sys_setreuid(ruid, euid);
597}
598
599/*
600 * setuid() is implemented like SysV with SAVED_IDS
601 *
602 * Note that SAVED_ID's is deficient in that a setuid root program
603 * like sendmail, for example, cannot set its uid to be a normal
604 * user and then switch back, because if you're root, setuid() sets
605 * the saved uid too. If you don't like this, blame the bright people
606 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
607 * will allow a root program to temporarily drop privileges and be able to
608 * regain them by swapping the real and effective uid.
609 */
610long __sys_setuid(uid_t uid)
611{
612 struct user_namespace *ns = current_user_ns();
613 const struct cred *old;
614 struct cred *new;
615 int retval;
616 kuid_t kuid;
617
618 kuid = make_kuid(ns, uid);
619 if (!uid_valid(kuid))
620 return -EINVAL;
621
622 new = prepare_creds();
623 if (!new)
624 return -ENOMEM;
625 old = current_cred();
626
627 retval = -EPERM;
628 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
629 new->suid = new->uid = kuid;
630 if (!uid_eq(kuid, old->uid)) {
631 retval = set_user(new);
632 if (retval < 0)
633 goto error;
634 }
635 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
636 goto error;
637 }
638
639 new->fsuid = new->euid = kuid;
640
641 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
642 if (retval < 0)
643 goto error;
644
645 retval = set_cred_ucounts(new);
646 if (retval < 0)
647 goto error;
648
649 flag_nproc_exceeded(new);
650 return commit_creds(new);
651
652error:
653 abort_creds(new);
654 return retval;
655}
656
657SYSCALL_DEFINE1(setuid, uid_t, uid)
658{
659 return __sys_setuid(uid);
660}
661
662
663/*
664 * This function implements a generic ability to update ruid, euid,
665 * and suid. This allows you to implement the 4.4 compatible seteuid().
666 */
667long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
668{
669 struct user_namespace *ns = current_user_ns();
670 const struct cred *old;
671 struct cred *new;
672 int retval;
673 kuid_t kruid, keuid, ksuid;
674 bool ruid_new, euid_new, suid_new;
675
676 kruid = make_kuid(ns, ruid);
677 keuid = make_kuid(ns, euid);
678 ksuid = make_kuid(ns, suid);
679
680 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
681 return -EINVAL;
682
683 if ((euid != (uid_t) -1) && !uid_valid(keuid))
684 return -EINVAL;
685
686 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
687 return -EINVAL;
688
689 old = current_cred();
690
691 /* check for no-op */
692 if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
693 (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
694 uid_eq(keuid, old->fsuid))) &&
695 (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
696 return 0;
697
698 ruid_new = ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
699 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
700 euid_new = euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
701 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
702 suid_new = suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
703 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
704 if ((ruid_new || euid_new || suid_new) &&
705 !ns_capable_setid(old->user_ns, CAP_SETUID))
706 return -EPERM;
707
708 new = prepare_creds();
709 if (!new)
710 return -ENOMEM;
711
712 if (ruid != (uid_t) -1) {
713 new->uid = kruid;
714 if (!uid_eq(kruid, old->uid)) {
715 retval = set_user(new);
716 if (retval < 0)
717 goto error;
718 }
719 }
720 if (euid != (uid_t) -1)
721 new->euid = keuid;
722 if (suid != (uid_t) -1)
723 new->suid = ksuid;
724 new->fsuid = new->euid;
725
726 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
727 if (retval < 0)
728 goto error;
729
730 retval = set_cred_ucounts(new);
731 if (retval < 0)
732 goto error;
733
734 flag_nproc_exceeded(new);
735 return commit_creds(new);
736
737error:
738 abort_creds(new);
739 return retval;
740}
741
742SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
743{
744 return __sys_setresuid(ruid, euid, suid);
745}
746
747SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
748{
749 const struct cred *cred = current_cred();
750 int retval;
751 uid_t ruid, euid, suid;
752
753 ruid = from_kuid_munged(cred->user_ns, cred->uid);
754 euid = from_kuid_munged(cred->user_ns, cred->euid);
755 suid = from_kuid_munged(cred->user_ns, cred->suid);
756
757 retval = put_user(ruid, ruidp);
758 if (!retval) {
759 retval = put_user(euid, euidp);
760 if (!retval)
761 return put_user(suid, suidp);
762 }
763 return retval;
764}
765
766/*
767 * Same as above, but for rgid, egid, sgid.
768 */
769long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
770{
771 struct user_namespace *ns = current_user_ns();
772 const struct cred *old;
773 struct cred *new;
774 int retval;
775 kgid_t krgid, kegid, ksgid;
776 bool rgid_new, egid_new, sgid_new;
777
778 krgid = make_kgid(ns, rgid);
779 kegid = make_kgid(ns, egid);
780 ksgid = make_kgid(ns, sgid);
781
782 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
783 return -EINVAL;
784 if ((egid != (gid_t) -1) && !gid_valid(kegid))
785 return -EINVAL;
786 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
787 return -EINVAL;
788
789 old = current_cred();
790
791 /* check for no-op */
792 if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
793 (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
794 gid_eq(kegid, old->fsgid))) &&
795 (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
796 return 0;
797
798 rgid_new = rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
799 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
800 egid_new = egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
801 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
802 sgid_new = sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
803 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
804 if ((rgid_new || egid_new || sgid_new) &&
805 !ns_capable_setid(old->user_ns, CAP_SETGID))
806 return -EPERM;
807
808 new = prepare_creds();
809 if (!new)
810 return -ENOMEM;
811
812 if (rgid != (gid_t) -1)
813 new->gid = krgid;
814 if (egid != (gid_t) -1)
815 new->egid = kegid;
816 if (sgid != (gid_t) -1)
817 new->sgid = ksgid;
818 new->fsgid = new->egid;
819
820 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
821 if (retval < 0)
822 goto error;
823
824 return commit_creds(new);
825
826error:
827 abort_creds(new);
828 return retval;
829}
830
831SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
832{
833 return __sys_setresgid(rgid, egid, sgid);
834}
835
836SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
837{
838 const struct cred *cred = current_cred();
839 int retval;
840 gid_t rgid, egid, sgid;
841
842 rgid = from_kgid_munged(cred->user_ns, cred->gid);
843 egid = from_kgid_munged(cred->user_ns, cred->egid);
844 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
845
846 retval = put_user(rgid, rgidp);
847 if (!retval) {
848 retval = put_user(egid, egidp);
849 if (!retval)
850 retval = put_user(sgid, sgidp);
851 }
852
853 return retval;
854}
855
856
857/*
858 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
859 * is used for "access()" and for the NFS daemon (letting nfsd stay at
860 * whatever uid it wants to). It normally shadows "euid", except when
861 * explicitly set by setfsuid() or for access..
862 */
863long __sys_setfsuid(uid_t uid)
864{
865 const struct cred *old;
866 struct cred *new;
867 uid_t old_fsuid;
868 kuid_t kuid;
869
870 old = current_cred();
871 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
872
873 kuid = make_kuid(old->user_ns, uid);
874 if (!uid_valid(kuid))
875 return old_fsuid;
876
877 new = prepare_creds();
878 if (!new)
879 return old_fsuid;
880
881 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
882 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
883 ns_capable_setid(old->user_ns, CAP_SETUID)) {
884 if (!uid_eq(kuid, old->fsuid)) {
885 new->fsuid = kuid;
886 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
887 goto change_okay;
888 }
889 }
890
891 abort_creds(new);
892 return old_fsuid;
893
894change_okay:
895 commit_creds(new);
896 return old_fsuid;
897}
898
899SYSCALL_DEFINE1(setfsuid, uid_t, uid)
900{
901 return __sys_setfsuid(uid);
902}
903
904/*
905 * Samma på svenska..
906 */
907long __sys_setfsgid(gid_t gid)
908{
909 const struct cred *old;
910 struct cred *new;
911 gid_t old_fsgid;
912 kgid_t kgid;
913
914 old = current_cred();
915 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
916
917 kgid = make_kgid(old->user_ns, gid);
918 if (!gid_valid(kgid))
919 return old_fsgid;
920
921 new = prepare_creds();
922 if (!new)
923 return old_fsgid;
924
925 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
926 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
927 ns_capable_setid(old->user_ns, CAP_SETGID)) {
928 if (!gid_eq(kgid, old->fsgid)) {
929 new->fsgid = kgid;
930 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
931 goto change_okay;
932 }
933 }
934
935 abort_creds(new);
936 return old_fsgid;
937
938change_okay:
939 commit_creds(new);
940 return old_fsgid;
941}
942
943SYSCALL_DEFINE1(setfsgid, gid_t, gid)
944{
945 return __sys_setfsgid(gid);
946}
947#endif /* CONFIG_MULTIUSER */
948
949/**
950 * sys_getpid - return the thread group id of the current process
951 *
952 * Note, despite the name, this returns the tgid not the pid. The tgid and
953 * the pid are identical unless CLONE_THREAD was specified on clone() in
954 * which case the tgid is the same in all threads of the same group.
955 *
956 * This is SMP safe as current->tgid does not change.
957 */
958SYSCALL_DEFINE0(getpid)
959{
960 return task_tgid_vnr(current);
961}
962
963/* Thread ID - the internal kernel "pid" */
964SYSCALL_DEFINE0(gettid)
965{
966 return task_pid_vnr(current);
967}
968
969/*
970 * Accessing ->real_parent is not SMP-safe, it could
971 * change from under us. However, we can use a stale
972 * value of ->real_parent under rcu_read_lock(), see
973 * release_task()->call_rcu(delayed_put_task_struct).
974 */
975SYSCALL_DEFINE0(getppid)
976{
977 int pid;
978
979 rcu_read_lock();
980 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
981 rcu_read_unlock();
982
983 return pid;
984}
985
986SYSCALL_DEFINE0(getuid)
987{
988 /* Only we change this so SMP safe */
989 return from_kuid_munged(current_user_ns(), current_uid());
990}
991
992SYSCALL_DEFINE0(geteuid)
993{
994 /* Only we change this so SMP safe */
995 return from_kuid_munged(current_user_ns(), current_euid());
996}
997
998SYSCALL_DEFINE0(getgid)
999{
1000 /* Only we change this so SMP safe */
1001 return from_kgid_munged(current_user_ns(), current_gid());
1002}
1003
1004SYSCALL_DEFINE0(getegid)
1005{
1006 /* Only we change this so SMP safe */
1007 return from_kgid_munged(current_user_ns(), current_egid());
1008}
1009
1010static void do_sys_times(struct tms *tms)
1011{
1012 u64 tgutime, tgstime, cutime, cstime;
1013
1014 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1015 cutime = current->signal->cutime;
1016 cstime = current->signal->cstime;
1017 tms->tms_utime = nsec_to_clock_t(tgutime);
1018 tms->tms_stime = nsec_to_clock_t(tgstime);
1019 tms->tms_cutime = nsec_to_clock_t(cutime);
1020 tms->tms_cstime = nsec_to_clock_t(cstime);
1021}
1022
1023SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1024{
1025 if (tbuf) {
1026 struct tms tmp;
1027
1028 do_sys_times(&tmp);
1029 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1030 return -EFAULT;
1031 }
1032 force_successful_syscall_return();
1033 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1034}
1035
1036#ifdef CONFIG_COMPAT
1037static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1038{
1039 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1040}
1041
1042COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1043{
1044 if (tbuf) {
1045 struct tms tms;
1046 struct compat_tms tmp;
1047
1048 do_sys_times(&tms);
1049 /* Convert our struct tms to the compat version. */
1050 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1051 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1052 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1053 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1054 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1055 return -EFAULT;
1056 }
1057 force_successful_syscall_return();
1058 return compat_jiffies_to_clock_t(jiffies);
1059}
1060#endif
1061
1062/*
1063 * This needs some heavy checking ...
1064 * I just haven't the stomach for it. I also don't fully
1065 * understand sessions/pgrp etc. Let somebody who does explain it.
1066 *
1067 * OK, I think I have the protection semantics right.... this is really
1068 * only important on a multi-user system anyway, to make sure one user
1069 * can't send a signal to a process owned by another. -TYT, 12/12/91
1070 *
1071 * !PF_FORKNOEXEC check to conform completely to POSIX.
1072 */
1073SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1074{
1075 struct task_struct *p;
1076 struct task_struct *group_leader = current->group_leader;
1077 struct pid *pgrp;
1078 int err;
1079
1080 if (!pid)
1081 pid = task_pid_vnr(group_leader);
1082 if (!pgid)
1083 pgid = pid;
1084 if (pgid < 0)
1085 return -EINVAL;
1086 rcu_read_lock();
1087
1088 /* From this point forward we keep holding onto the tasklist lock
1089 * so that our parent does not change from under us. -DaveM
1090 */
1091 write_lock_irq(&tasklist_lock);
1092
1093 err = -ESRCH;
1094 p = find_task_by_vpid(pid);
1095 if (!p)
1096 goto out;
1097
1098 err = -EINVAL;
1099 if (!thread_group_leader(p))
1100 goto out;
1101
1102 if (same_thread_group(p->real_parent, group_leader)) {
1103 err = -EPERM;
1104 if (task_session(p) != task_session(group_leader))
1105 goto out;
1106 err = -EACCES;
1107 if (!(p->flags & PF_FORKNOEXEC))
1108 goto out;
1109 } else {
1110 err = -ESRCH;
1111 if (p != group_leader)
1112 goto out;
1113 }
1114
1115 err = -EPERM;
1116 if (p->signal->leader)
1117 goto out;
1118
1119 pgrp = task_pid(p);
1120 if (pgid != pid) {
1121 struct task_struct *g;
1122
1123 pgrp = find_vpid(pgid);
1124 g = pid_task(pgrp, PIDTYPE_PGID);
1125 if (!g || task_session(g) != task_session(group_leader))
1126 goto out;
1127 }
1128
1129 err = security_task_setpgid(p, pgid);
1130 if (err)
1131 goto out;
1132
1133 if (task_pgrp(p) != pgrp)
1134 change_pid(p, PIDTYPE_PGID, pgrp);
1135
1136 err = 0;
1137out:
1138 /* All paths lead to here, thus we are safe. -DaveM */
1139 write_unlock_irq(&tasklist_lock);
1140 rcu_read_unlock();
1141 return err;
1142}
1143
1144static int do_getpgid(pid_t pid)
1145{
1146 struct task_struct *p;
1147 struct pid *grp;
1148 int retval;
1149
1150 rcu_read_lock();
1151 if (!pid)
1152 grp = task_pgrp(current);
1153 else {
1154 retval = -ESRCH;
1155 p = find_task_by_vpid(pid);
1156 if (!p)
1157 goto out;
1158 grp = task_pgrp(p);
1159 if (!grp)
1160 goto out;
1161
1162 retval = security_task_getpgid(p);
1163 if (retval)
1164 goto out;
1165 }
1166 retval = pid_vnr(grp);
1167out:
1168 rcu_read_unlock();
1169 return retval;
1170}
1171
1172SYSCALL_DEFINE1(getpgid, pid_t, pid)
1173{
1174 return do_getpgid(pid);
1175}
1176
1177#ifdef __ARCH_WANT_SYS_GETPGRP
1178
1179SYSCALL_DEFINE0(getpgrp)
1180{
1181 return do_getpgid(0);
1182}
1183
1184#endif
1185
1186SYSCALL_DEFINE1(getsid, pid_t, pid)
1187{
1188 struct task_struct *p;
1189 struct pid *sid;
1190 int retval;
1191
1192 rcu_read_lock();
1193 if (!pid)
1194 sid = task_session(current);
1195 else {
1196 retval = -ESRCH;
1197 p = find_task_by_vpid(pid);
1198 if (!p)
1199 goto out;
1200 sid = task_session(p);
1201 if (!sid)
1202 goto out;
1203
1204 retval = security_task_getsid(p);
1205 if (retval)
1206 goto out;
1207 }
1208 retval = pid_vnr(sid);
1209out:
1210 rcu_read_unlock();
1211 return retval;
1212}
1213
1214static void set_special_pids(struct pid *pid)
1215{
1216 struct task_struct *curr = current->group_leader;
1217
1218 if (task_session(curr) != pid)
1219 change_pid(curr, PIDTYPE_SID, pid);
1220
1221 if (task_pgrp(curr) != pid)
1222 change_pid(curr, PIDTYPE_PGID, pid);
1223}
1224
1225int ksys_setsid(void)
1226{
1227 struct task_struct *group_leader = current->group_leader;
1228 struct pid *sid = task_pid(group_leader);
1229 pid_t session = pid_vnr(sid);
1230 int err = -EPERM;
1231
1232 write_lock_irq(&tasklist_lock);
1233 /* Fail if I am already a session leader */
1234 if (group_leader->signal->leader)
1235 goto out;
1236
1237 /* Fail if a process group id already exists that equals the
1238 * proposed session id.
1239 */
1240 if (pid_task(sid, PIDTYPE_PGID))
1241 goto out;
1242
1243 group_leader->signal->leader = 1;
1244 set_special_pids(sid);
1245
1246 proc_clear_tty(group_leader);
1247
1248 err = session;
1249out:
1250 write_unlock_irq(&tasklist_lock);
1251 if (err > 0) {
1252 proc_sid_connector(group_leader);
1253 sched_autogroup_create_attach(group_leader);
1254 }
1255 return err;
1256}
1257
1258SYSCALL_DEFINE0(setsid)
1259{
1260 return ksys_setsid();
1261}
1262
1263DECLARE_RWSEM(uts_sem);
1264
1265#ifdef COMPAT_UTS_MACHINE
1266#define override_architecture(name) \
1267 (personality(current->personality) == PER_LINUX32 && \
1268 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1269 sizeof(COMPAT_UTS_MACHINE)))
1270#else
1271#define override_architecture(name) 0
1272#endif
1273
1274/*
1275 * Work around broken programs that cannot handle "Linux 3.0".
1276 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1277 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1278 * 2.6.60.
1279 */
1280static int override_release(char __user *release, size_t len)
1281{
1282 int ret = 0;
1283
1284 if (current->personality & UNAME26) {
1285 const char *rest = UTS_RELEASE;
1286 char buf[65] = { 0 };
1287 int ndots = 0;
1288 unsigned v;
1289 size_t copy;
1290
1291 while (*rest) {
1292 if (*rest == '.' && ++ndots >= 3)
1293 break;
1294 if (!isdigit(*rest) && *rest != '.')
1295 break;
1296 rest++;
1297 }
1298 v = LINUX_VERSION_PATCHLEVEL + 60;
1299 copy = clamp_t(size_t, len, 1, sizeof(buf));
1300 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1301 ret = copy_to_user(release, buf, copy + 1);
1302 }
1303 return ret;
1304}
1305
1306SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1307{
1308 struct new_utsname tmp;
1309
1310 down_read(&uts_sem);
1311 memcpy(&tmp, utsname(), sizeof(tmp));
1312 up_read(&uts_sem);
1313 if (copy_to_user(name, &tmp, sizeof(tmp)))
1314 return -EFAULT;
1315
1316 if (override_release(name->release, sizeof(name->release)))
1317 return -EFAULT;
1318 if (override_architecture(name))
1319 return -EFAULT;
1320 return 0;
1321}
1322
1323#ifdef __ARCH_WANT_SYS_OLD_UNAME
1324/*
1325 * Old cruft
1326 */
1327SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1328{
1329 struct old_utsname tmp;
1330
1331 if (!name)
1332 return -EFAULT;
1333
1334 down_read(&uts_sem);
1335 memcpy(&tmp, utsname(), sizeof(tmp));
1336 up_read(&uts_sem);
1337 if (copy_to_user(name, &tmp, sizeof(tmp)))
1338 return -EFAULT;
1339
1340 if (override_release(name->release, sizeof(name->release)))
1341 return -EFAULT;
1342 if (override_architecture(name))
1343 return -EFAULT;
1344 return 0;
1345}
1346
1347SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1348{
1349 struct oldold_utsname tmp;
1350
1351 if (!name)
1352 return -EFAULT;
1353
1354 memset(&tmp, 0, sizeof(tmp));
1355
1356 down_read(&uts_sem);
1357 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1358 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1359 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1360 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1361 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1362 up_read(&uts_sem);
1363 if (copy_to_user(name, &tmp, sizeof(tmp)))
1364 return -EFAULT;
1365
1366 if (override_architecture(name))
1367 return -EFAULT;
1368 if (override_release(name->release, sizeof(name->release)))
1369 return -EFAULT;
1370 return 0;
1371}
1372#endif
1373
1374SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1375{
1376 int errno;
1377 char tmp[__NEW_UTS_LEN];
1378
1379 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1380 return -EPERM;
1381
1382 if (len < 0 || len > __NEW_UTS_LEN)
1383 return -EINVAL;
1384 errno = -EFAULT;
1385 if (!copy_from_user(tmp, name, len)) {
1386 struct new_utsname *u;
1387
1388 add_device_randomness(tmp, len);
1389 down_write(&uts_sem);
1390 u = utsname();
1391 memcpy(u->nodename, tmp, len);
1392 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1393 errno = 0;
1394 uts_proc_notify(UTS_PROC_HOSTNAME);
1395 up_write(&uts_sem);
1396 }
1397 return errno;
1398}
1399
1400#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1401
1402SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1403{
1404 int i;
1405 struct new_utsname *u;
1406 char tmp[__NEW_UTS_LEN + 1];
1407
1408 if (len < 0)
1409 return -EINVAL;
1410 down_read(&uts_sem);
1411 u = utsname();
1412 i = 1 + strlen(u->nodename);
1413 if (i > len)
1414 i = len;
1415 memcpy(tmp, u->nodename, i);
1416 up_read(&uts_sem);
1417 if (copy_to_user(name, tmp, i))
1418 return -EFAULT;
1419 return 0;
1420}
1421
1422#endif
1423
1424/*
1425 * Only setdomainname; getdomainname can be implemented by calling
1426 * uname()
1427 */
1428SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1429{
1430 int errno;
1431 char tmp[__NEW_UTS_LEN];
1432
1433 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1434 return -EPERM;
1435 if (len < 0 || len > __NEW_UTS_LEN)
1436 return -EINVAL;
1437
1438 errno = -EFAULT;
1439 if (!copy_from_user(tmp, name, len)) {
1440 struct new_utsname *u;
1441
1442 add_device_randomness(tmp, len);
1443 down_write(&uts_sem);
1444 u = utsname();
1445 memcpy(u->domainname, tmp, len);
1446 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1447 errno = 0;
1448 uts_proc_notify(UTS_PROC_DOMAINNAME);
1449 up_write(&uts_sem);
1450 }
1451 return errno;
1452}
1453
1454/* make sure you are allowed to change @tsk limits before calling this */
1455static int do_prlimit(struct task_struct *tsk, unsigned int resource,
1456 struct rlimit *new_rlim, struct rlimit *old_rlim)
1457{
1458 struct rlimit *rlim;
1459 int retval = 0;
1460
1461 if (resource >= RLIM_NLIMITS)
1462 return -EINVAL;
1463 resource = array_index_nospec(resource, RLIM_NLIMITS);
1464
1465 if (new_rlim) {
1466 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1467 return -EINVAL;
1468 if (resource == RLIMIT_NOFILE &&
1469 new_rlim->rlim_max > sysctl_nr_open)
1470 return -EPERM;
1471 }
1472
1473 /* Holding a refcount on tsk protects tsk->signal from disappearing. */
1474 rlim = tsk->signal->rlim + resource;
1475 task_lock(tsk->group_leader);
1476 if (new_rlim) {
1477 /*
1478 * Keep the capable check against init_user_ns until cgroups can
1479 * contain all limits.
1480 */
1481 if (new_rlim->rlim_max > rlim->rlim_max &&
1482 !capable(CAP_SYS_RESOURCE))
1483 retval = -EPERM;
1484 if (!retval)
1485 retval = security_task_setrlimit(tsk, resource, new_rlim);
1486 }
1487 if (!retval) {
1488 if (old_rlim)
1489 *old_rlim = *rlim;
1490 if (new_rlim)
1491 *rlim = *new_rlim;
1492 }
1493 task_unlock(tsk->group_leader);
1494
1495 /*
1496 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1497 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1498 * ignores the rlimit.
1499 */
1500 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1501 new_rlim->rlim_cur != RLIM_INFINITY &&
1502 IS_ENABLED(CONFIG_POSIX_TIMERS)) {
1503 /*
1504 * update_rlimit_cpu can fail if the task is exiting, but there
1505 * may be other tasks in the thread group that are not exiting,
1506 * and they need their cpu timers adjusted.
1507 *
1508 * The group_leader is the last task to be released, so if we
1509 * cannot update_rlimit_cpu on it, then the entire process is
1510 * exiting and we do not need to update at all.
1511 */
1512 update_rlimit_cpu(tsk->group_leader, new_rlim->rlim_cur);
1513 }
1514
1515 return retval;
1516}
1517
1518SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1519{
1520 struct rlimit value;
1521 int ret;
1522
1523 ret = do_prlimit(current, resource, NULL, &value);
1524 if (!ret)
1525 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1526
1527 return ret;
1528}
1529
1530#ifdef CONFIG_COMPAT
1531
1532COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1533 struct compat_rlimit __user *, rlim)
1534{
1535 struct rlimit r;
1536 struct compat_rlimit r32;
1537
1538 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1539 return -EFAULT;
1540
1541 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1542 r.rlim_cur = RLIM_INFINITY;
1543 else
1544 r.rlim_cur = r32.rlim_cur;
1545 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1546 r.rlim_max = RLIM_INFINITY;
1547 else
1548 r.rlim_max = r32.rlim_max;
1549 return do_prlimit(current, resource, &r, NULL);
1550}
1551
1552COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1553 struct compat_rlimit __user *, rlim)
1554{
1555 struct rlimit r;
1556 int ret;
1557
1558 ret = do_prlimit(current, resource, NULL, &r);
1559 if (!ret) {
1560 struct compat_rlimit r32;
1561 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1562 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1563 else
1564 r32.rlim_cur = r.rlim_cur;
1565 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1566 r32.rlim_max = COMPAT_RLIM_INFINITY;
1567 else
1568 r32.rlim_max = r.rlim_max;
1569
1570 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1571 return -EFAULT;
1572 }
1573 return ret;
1574}
1575
1576#endif
1577
1578#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1579
1580/*
1581 * Back compatibility for getrlimit. Needed for some apps.
1582 */
1583SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1584 struct rlimit __user *, rlim)
1585{
1586 struct rlimit x;
1587 if (resource >= RLIM_NLIMITS)
1588 return -EINVAL;
1589
1590 resource = array_index_nospec(resource, RLIM_NLIMITS);
1591 task_lock(current->group_leader);
1592 x = current->signal->rlim[resource];
1593 task_unlock(current->group_leader);
1594 if (x.rlim_cur > 0x7FFFFFFF)
1595 x.rlim_cur = 0x7FFFFFFF;
1596 if (x.rlim_max > 0x7FFFFFFF)
1597 x.rlim_max = 0x7FFFFFFF;
1598 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1599}
1600
1601#ifdef CONFIG_COMPAT
1602COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1603 struct compat_rlimit __user *, rlim)
1604{
1605 struct rlimit r;
1606
1607 if (resource >= RLIM_NLIMITS)
1608 return -EINVAL;
1609
1610 resource = array_index_nospec(resource, RLIM_NLIMITS);
1611 task_lock(current->group_leader);
1612 r = current->signal->rlim[resource];
1613 task_unlock(current->group_leader);
1614 if (r.rlim_cur > 0x7FFFFFFF)
1615 r.rlim_cur = 0x7FFFFFFF;
1616 if (r.rlim_max > 0x7FFFFFFF)
1617 r.rlim_max = 0x7FFFFFFF;
1618
1619 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1620 put_user(r.rlim_max, &rlim->rlim_max))
1621 return -EFAULT;
1622 return 0;
1623}
1624#endif
1625
1626#endif
1627
1628static inline bool rlim64_is_infinity(__u64 rlim64)
1629{
1630#if BITS_PER_LONG < 64
1631 return rlim64 >= ULONG_MAX;
1632#else
1633 return rlim64 == RLIM64_INFINITY;
1634#endif
1635}
1636
1637static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1638{
1639 if (rlim->rlim_cur == RLIM_INFINITY)
1640 rlim64->rlim_cur = RLIM64_INFINITY;
1641 else
1642 rlim64->rlim_cur = rlim->rlim_cur;
1643 if (rlim->rlim_max == RLIM_INFINITY)
1644 rlim64->rlim_max = RLIM64_INFINITY;
1645 else
1646 rlim64->rlim_max = rlim->rlim_max;
1647}
1648
1649static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1650{
1651 if (rlim64_is_infinity(rlim64->rlim_cur))
1652 rlim->rlim_cur = RLIM_INFINITY;
1653 else
1654 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1655 if (rlim64_is_infinity(rlim64->rlim_max))
1656 rlim->rlim_max = RLIM_INFINITY;
1657 else
1658 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1659}
1660
1661/* rcu lock must be held */
1662static int check_prlimit_permission(struct task_struct *task,
1663 unsigned int flags)
1664{
1665 const struct cred *cred = current_cred(), *tcred;
1666 bool id_match;
1667
1668 if (current == task)
1669 return 0;
1670
1671 tcred = __task_cred(task);
1672 id_match = (uid_eq(cred->uid, tcred->euid) &&
1673 uid_eq(cred->uid, tcred->suid) &&
1674 uid_eq(cred->uid, tcred->uid) &&
1675 gid_eq(cred->gid, tcred->egid) &&
1676 gid_eq(cred->gid, tcred->sgid) &&
1677 gid_eq(cred->gid, tcred->gid));
1678 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1679 return -EPERM;
1680
1681 return security_task_prlimit(cred, tcred, flags);
1682}
1683
1684SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1685 const struct rlimit64 __user *, new_rlim,
1686 struct rlimit64 __user *, old_rlim)
1687{
1688 struct rlimit64 old64, new64;
1689 struct rlimit old, new;
1690 struct task_struct *tsk;
1691 unsigned int checkflags = 0;
1692 int ret;
1693
1694 if (old_rlim)
1695 checkflags |= LSM_PRLIMIT_READ;
1696
1697 if (new_rlim) {
1698 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1699 return -EFAULT;
1700 rlim64_to_rlim(&new64, &new);
1701 checkflags |= LSM_PRLIMIT_WRITE;
1702 }
1703
1704 rcu_read_lock();
1705 tsk = pid ? find_task_by_vpid(pid) : current;
1706 if (!tsk) {
1707 rcu_read_unlock();
1708 return -ESRCH;
1709 }
1710 ret = check_prlimit_permission(tsk, checkflags);
1711 if (ret) {
1712 rcu_read_unlock();
1713 return ret;
1714 }
1715 get_task_struct(tsk);
1716 rcu_read_unlock();
1717
1718 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1719 old_rlim ? &old : NULL);
1720
1721 if (!ret && old_rlim) {
1722 rlim_to_rlim64(&old, &old64);
1723 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1724 ret = -EFAULT;
1725 }
1726
1727 put_task_struct(tsk);
1728 return ret;
1729}
1730
1731SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1732{
1733 struct rlimit new_rlim;
1734
1735 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1736 return -EFAULT;
1737 return do_prlimit(current, resource, &new_rlim, NULL);
1738}
1739
1740/*
1741 * It would make sense to put struct rusage in the task_struct,
1742 * except that would make the task_struct be *really big*. After
1743 * task_struct gets moved into malloc'ed memory, it would
1744 * make sense to do this. It will make moving the rest of the information
1745 * a lot simpler! (Which we're not doing right now because we're not
1746 * measuring them yet).
1747 *
1748 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1749 * races with threads incrementing their own counters. But since word
1750 * reads are atomic, we either get new values or old values and we don't
1751 * care which for the sums. We always take the siglock to protect reading
1752 * the c* fields from p->signal from races with exit.c updating those
1753 * fields when reaping, so a sample either gets all the additions of a
1754 * given child after it's reaped, or none so this sample is before reaping.
1755 *
1756 * Locking:
1757 * We need to take the siglock for CHILDEREN, SELF and BOTH
1758 * for the cases current multithreaded, non-current single threaded
1759 * non-current multithreaded. Thread traversal is now safe with
1760 * the siglock held.
1761 * Strictly speaking, we donot need to take the siglock if we are current and
1762 * single threaded, as no one else can take our signal_struct away, no one
1763 * else can reap the children to update signal->c* counters, and no one else
1764 * can race with the signal-> fields. If we do not take any lock, the
1765 * signal-> fields could be read out of order while another thread was just
1766 * exiting. So we should place a read memory barrier when we avoid the lock.
1767 * On the writer side, write memory barrier is implied in __exit_signal
1768 * as __exit_signal releases the siglock spinlock after updating the signal->
1769 * fields. But we don't do this yet to keep things simple.
1770 *
1771 */
1772
1773static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1774{
1775 r->ru_nvcsw += t->nvcsw;
1776 r->ru_nivcsw += t->nivcsw;
1777 r->ru_minflt += t->min_flt;
1778 r->ru_majflt += t->maj_flt;
1779 r->ru_inblock += task_io_get_inblock(t);
1780 r->ru_oublock += task_io_get_oublock(t);
1781}
1782
1783void getrusage(struct task_struct *p, int who, struct rusage *r)
1784{
1785 struct task_struct *t;
1786 unsigned long flags;
1787 u64 tgutime, tgstime, utime, stime;
1788 unsigned long maxrss;
1789 struct mm_struct *mm;
1790 struct signal_struct *sig = p->signal;
1791 unsigned int seq = 0;
1792
1793retry:
1794 memset(r, 0, sizeof(*r));
1795 utime = stime = 0;
1796 maxrss = 0;
1797
1798 if (who == RUSAGE_THREAD) {
1799 task_cputime_adjusted(current, &utime, &stime);
1800 accumulate_thread_rusage(p, r);
1801 maxrss = sig->maxrss;
1802 goto out_thread;
1803 }
1804
1805 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
1806
1807 switch (who) {
1808 case RUSAGE_BOTH:
1809 case RUSAGE_CHILDREN:
1810 utime = sig->cutime;
1811 stime = sig->cstime;
1812 r->ru_nvcsw = sig->cnvcsw;
1813 r->ru_nivcsw = sig->cnivcsw;
1814 r->ru_minflt = sig->cmin_flt;
1815 r->ru_majflt = sig->cmaj_flt;
1816 r->ru_inblock = sig->cinblock;
1817 r->ru_oublock = sig->coublock;
1818 maxrss = sig->cmaxrss;
1819
1820 if (who == RUSAGE_CHILDREN)
1821 break;
1822 fallthrough;
1823
1824 case RUSAGE_SELF:
1825 r->ru_nvcsw += sig->nvcsw;
1826 r->ru_nivcsw += sig->nivcsw;
1827 r->ru_minflt += sig->min_flt;
1828 r->ru_majflt += sig->maj_flt;
1829 r->ru_inblock += sig->inblock;
1830 r->ru_oublock += sig->oublock;
1831 if (maxrss < sig->maxrss)
1832 maxrss = sig->maxrss;
1833
1834 rcu_read_lock();
1835 __for_each_thread(sig, t)
1836 accumulate_thread_rusage(t, r);
1837 rcu_read_unlock();
1838
1839 break;
1840
1841 default:
1842 BUG();
1843 }
1844
1845 if (need_seqretry(&sig->stats_lock, seq)) {
1846 seq = 1;
1847 goto retry;
1848 }
1849 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
1850
1851 if (who == RUSAGE_CHILDREN)
1852 goto out_children;
1853
1854 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1855 utime += tgutime;
1856 stime += tgstime;
1857
1858out_thread:
1859 mm = get_task_mm(p);
1860 if (mm) {
1861 setmax_mm_hiwater_rss(&maxrss, mm);
1862 mmput(mm);
1863 }
1864
1865out_children:
1866 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1867 r->ru_utime = ns_to_kernel_old_timeval(utime);
1868 r->ru_stime = ns_to_kernel_old_timeval(stime);
1869}
1870
1871SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1872{
1873 struct rusage r;
1874
1875 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1876 who != RUSAGE_THREAD)
1877 return -EINVAL;
1878
1879 getrusage(current, who, &r);
1880 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1881}
1882
1883#ifdef CONFIG_COMPAT
1884COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1885{
1886 struct rusage r;
1887
1888 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1889 who != RUSAGE_THREAD)
1890 return -EINVAL;
1891
1892 getrusage(current, who, &r);
1893 return put_compat_rusage(&r, ru);
1894}
1895#endif
1896
1897SYSCALL_DEFINE1(umask, int, mask)
1898{
1899 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1900 return mask;
1901}
1902
1903static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1904{
1905 struct fd exe;
1906 struct inode *inode;
1907 int err;
1908
1909 exe = fdget(fd);
1910 if (!exe.file)
1911 return -EBADF;
1912
1913 inode = file_inode(exe.file);
1914
1915 /*
1916 * Because the original mm->exe_file points to executable file, make
1917 * sure that this one is executable as well, to avoid breaking an
1918 * overall picture.
1919 */
1920 err = -EACCES;
1921 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1922 goto exit;
1923
1924 err = file_permission(exe.file, MAY_EXEC);
1925 if (err)
1926 goto exit;
1927
1928 err = replace_mm_exe_file(mm, exe.file);
1929exit:
1930 fdput(exe);
1931 return err;
1932}
1933
1934/*
1935 * Check arithmetic relations of passed addresses.
1936 *
1937 * WARNING: we don't require any capability here so be very careful
1938 * in what is allowed for modification from userspace.
1939 */
1940static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1941{
1942 unsigned long mmap_max_addr = TASK_SIZE;
1943 int error = -EINVAL, i;
1944
1945 static const unsigned char offsets[] = {
1946 offsetof(struct prctl_mm_map, start_code),
1947 offsetof(struct prctl_mm_map, end_code),
1948 offsetof(struct prctl_mm_map, start_data),
1949 offsetof(struct prctl_mm_map, end_data),
1950 offsetof(struct prctl_mm_map, start_brk),
1951 offsetof(struct prctl_mm_map, brk),
1952 offsetof(struct prctl_mm_map, start_stack),
1953 offsetof(struct prctl_mm_map, arg_start),
1954 offsetof(struct prctl_mm_map, arg_end),
1955 offsetof(struct prctl_mm_map, env_start),
1956 offsetof(struct prctl_mm_map, env_end),
1957 };
1958
1959 /*
1960 * Make sure the members are not somewhere outside
1961 * of allowed address space.
1962 */
1963 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1964 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1965
1966 if ((unsigned long)val >= mmap_max_addr ||
1967 (unsigned long)val < mmap_min_addr)
1968 goto out;
1969 }
1970
1971 /*
1972 * Make sure the pairs are ordered.
1973 */
1974#define __prctl_check_order(__m1, __op, __m2) \
1975 ((unsigned long)prctl_map->__m1 __op \
1976 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1977 error = __prctl_check_order(start_code, <, end_code);
1978 error |= __prctl_check_order(start_data,<=, end_data);
1979 error |= __prctl_check_order(start_brk, <=, brk);
1980 error |= __prctl_check_order(arg_start, <=, arg_end);
1981 error |= __prctl_check_order(env_start, <=, env_end);
1982 if (error)
1983 goto out;
1984#undef __prctl_check_order
1985
1986 error = -EINVAL;
1987
1988 /*
1989 * Neither we should allow to override limits if they set.
1990 */
1991 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1992 prctl_map->start_brk, prctl_map->end_data,
1993 prctl_map->start_data))
1994 goto out;
1995
1996 error = 0;
1997out:
1998 return error;
1999}
2000
2001#ifdef CONFIG_CHECKPOINT_RESTORE
2002static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
2003{
2004 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
2005 unsigned long user_auxv[AT_VECTOR_SIZE];
2006 struct mm_struct *mm = current->mm;
2007 int error;
2008
2009 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2010 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
2011
2012 if (opt == PR_SET_MM_MAP_SIZE)
2013 return put_user((unsigned int)sizeof(prctl_map),
2014 (unsigned int __user *)addr);
2015
2016 if (data_size != sizeof(prctl_map))
2017 return -EINVAL;
2018
2019 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2020 return -EFAULT;
2021
2022 error = validate_prctl_map_addr(&prctl_map);
2023 if (error)
2024 return error;
2025
2026 if (prctl_map.auxv_size) {
2027 /*
2028 * Someone is trying to cheat the auxv vector.
2029 */
2030 if (!prctl_map.auxv ||
2031 prctl_map.auxv_size > sizeof(mm->saved_auxv))
2032 return -EINVAL;
2033
2034 memset(user_auxv, 0, sizeof(user_auxv));
2035 if (copy_from_user(user_auxv,
2036 (const void __user *)prctl_map.auxv,
2037 prctl_map.auxv_size))
2038 return -EFAULT;
2039
2040 /* Last entry must be AT_NULL as specification requires */
2041 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2042 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2043 }
2044
2045 if (prctl_map.exe_fd != (u32)-1) {
2046 /*
2047 * Check if the current user is checkpoint/restore capable.
2048 * At the time of this writing, it checks for CAP_SYS_ADMIN
2049 * or CAP_CHECKPOINT_RESTORE.
2050 * Note that a user with access to ptrace can masquerade an
2051 * arbitrary program as any executable, even setuid ones.
2052 * This may have implications in the tomoyo subsystem.
2053 */
2054 if (!checkpoint_restore_ns_capable(current_user_ns()))
2055 return -EPERM;
2056
2057 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2058 if (error)
2059 return error;
2060 }
2061
2062 /*
2063 * arg_lock protects concurrent updates but we still need mmap_lock for
2064 * read to exclude races with sys_brk.
2065 */
2066 mmap_read_lock(mm);
2067
2068 /*
2069 * We don't validate if these members are pointing to
2070 * real present VMAs because application may have correspond
2071 * VMAs already unmapped and kernel uses these members for statistics
2072 * output in procfs mostly, except
2073 *
2074 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2075 * for VMAs when updating these members so anything wrong written
2076 * here cause kernel to swear at userspace program but won't lead
2077 * to any problem in kernel itself
2078 */
2079
2080 spin_lock(&mm->arg_lock);
2081 mm->start_code = prctl_map.start_code;
2082 mm->end_code = prctl_map.end_code;
2083 mm->start_data = prctl_map.start_data;
2084 mm->end_data = prctl_map.end_data;
2085 mm->start_brk = prctl_map.start_brk;
2086 mm->brk = prctl_map.brk;
2087 mm->start_stack = prctl_map.start_stack;
2088 mm->arg_start = prctl_map.arg_start;
2089 mm->arg_end = prctl_map.arg_end;
2090 mm->env_start = prctl_map.env_start;
2091 mm->env_end = prctl_map.env_end;
2092 spin_unlock(&mm->arg_lock);
2093
2094 /*
2095 * Note this update of @saved_auxv is lockless thus
2096 * if someone reads this member in procfs while we're
2097 * updating -- it may get partly updated results. It's
2098 * known and acceptable trade off: we leave it as is to
2099 * not introduce additional locks here making the kernel
2100 * more complex.
2101 */
2102 if (prctl_map.auxv_size)
2103 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2104
2105 mmap_read_unlock(mm);
2106 return 0;
2107}
2108#endif /* CONFIG_CHECKPOINT_RESTORE */
2109
2110static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2111 unsigned long len)
2112{
2113 /*
2114 * This doesn't move the auxiliary vector itself since it's pinned to
2115 * mm_struct, but it permits filling the vector with new values. It's
2116 * up to the caller to provide sane values here, otherwise userspace
2117 * tools which use this vector might be unhappy.
2118 */
2119 unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2120
2121 if (len > sizeof(user_auxv))
2122 return -EINVAL;
2123
2124 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2125 return -EFAULT;
2126
2127 /* Make sure the last entry is always AT_NULL */
2128 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2129 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2130
2131 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2132
2133 task_lock(current);
2134 memcpy(mm->saved_auxv, user_auxv, len);
2135 task_unlock(current);
2136
2137 return 0;
2138}
2139
2140static int prctl_set_mm(int opt, unsigned long addr,
2141 unsigned long arg4, unsigned long arg5)
2142{
2143 struct mm_struct *mm = current->mm;
2144 struct prctl_mm_map prctl_map = {
2145 .auxv = NULL,
2146 .auxv_size = 0,
2147 .exe_fd = -1,
2148 };
2149 struct vm_area_struct *vma;
2150 int error;
2151
2152 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2153 opt != PR_SET_MM_MAP &&
2154 opt != PR_SET_MM_MAP_SIZE)))
2155 return -EINVAL;
2156
2157#ifdef CONFIG_CHECKPOINT_RESTORE
2158 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2159 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2160#endif
2161
2162 if (!capable(CAP_SYS_RESOURCE))
2163 return -EPERM;
2164
2165 if (opt == PR_SET_MM_EXE_FILE)
2166 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2167
2168 if (opt == PR_SET_MM_AUXV)
2169 return prctl_set_auxv(mm, addr, arg4);
2170
2171 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2172 return -EINVAL;
2173
2174 error = -EINVAL;
2175
2176 /*
2177 * arg_lock protects concurrent updates of arg boundaries, we need
2178 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2179 * validation.
2180 */
2181 mmap_read_lock(mm);
2182 vma = find_vma(mm, addr);
2183
2184 spin_lock(&mm->arg_lock);
2185 prctl_map.start_code = mm->start_code;
2186 prctl_map.end_code = mm->end_code;
2187 prctl_map.start_data = mm->start_data;
2188 prctl_map.end_data = mm->end_data;
2189 prctl_map.start_brk = mm->start_brk;
2190 prctl_map.brk = mm->brk;
2191 prctl_map.start_stack = mm->start_stack;
2192 prctl_map.arg_start = mm->arg_start;
2193 prctl_map.arg_end = mm->arg_end;
2194 prctl_map.env_start = mm->env_start;
2195 prctl_map.env_end = mm->env_end;
2196
2197 switch (opt) {
2198 case PR_SET_MM_START_CODE:
2199 prctl_map.start_code = addr;
2200 break;
2201 case PR_SET_MM_END_CODE:
2202 prctl_map.end_code = addr;
2203 break;
2204 case PR_SET_MM_START_DATA:
2205 prctl_map.start_data = addr;
2206 break;
2207 case PR_SET_MM_END_DATA:
2208 prctl_map.end_data = addr;
2209 break;
2210 case PR_SET_MM_START_STACK:
2211 prctl_map.start_stack = addr;
2212 break;
2213 case PR_SET_MM_START_BRK:
2214 prctl_map.start_brk = addr;
2215 break;
2216 case PR_SET_MM_BRK:
2217 prctl_map.brk = addr;
2218 break;
2219 case PR_SET_MM_ARG_START:
2220 prctl_map.arg_start = addr;
2221 break;
2222 case PR_SET_MM_ARG_END:
2223 prctl_map.arg_end = addr;
2224 break;
2225 case PR_SET_MM_ENV_START:
2226 prctl_map.env_start = addr;
2227 break;
2228 case PR_SET_MM_ENV_END:
2229 prctl_map.env_end = addr;
2230 break;
2231 default:
2232 goto out;
2233 }
2234
2235 error = validate_prctl_map_addr(&prctl_map);
2236 if (error)
2237 goto out;
2238
2239 switch (opt) {
2240 /*
2241 * If command line arguments and environment
2242 * are placed somewhere else on stack, we can
2243 * set them up here, ARG_START/END to setup
2244 * command line arguments and ENV_START/END
2245 * for environment.
2246 */
2247 case PR_SET_MM_START_STACK:
2248 case PR_SET_MM_ARG_START:
2249 case PR_SET_MM_ARG_END:
2250 case PR_SET_MM_ENV_START:
2251 case PR_SET_MM_ENV_END:
2252 if (!vma) {
2253 error = -EFAULT;
2254 goto out;
2255 }
2256 }
2257
2258 mm->start_code = prctl_map.start_code;
2259 mm->end_code = prctl_map.end_code;
2260 mm->start_data = prctl_map.start_data;
2261 mm->end_data = prctl_map.end_data;
2262 mm->start_brk = prctl_map.start_brk;
2263 mm->brk = prctl_map.brk;
2264 mm->start_stack = prctl_map.start_stack;
2265 mm->arg_start = prctl_map.arg_start;
2266 mm->arg_end = prctl_map.arg_end;
2267 mm->env_start = prctl_map.env_start;
2268 mm->env_end = prctl_map.env_end;
2269
2270 error = 0;
2271out:
2272 spin_unlock(&mm->arg_lock);
2273 mmap_read_unlock(mm);
2274 return error;
2275}
2276
2277#ifdef CONFIG_CHECKPOINT_RESTORE
2278static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2279{
2280 return put_user(me->clear_child_tid, tid_addr);
2281}
2282#else
2283static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2284{
2285 return -EINVAL;
2286}
2287#endif
2288
2289static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2290{
2291 /*
2292 * If task has has_child_subreaper - all its descendants
2293 * already have these flag too and new descendants will
2294 * inherit it on fork, skip them.
2295 *
2296 * If we've found child_reaper - skip descendants in
2297 * it's subtree as they will never get out pidns.
2298 */
2299 if (p->signal->has_child_subreaper ||
2300 is_child_reaper(task_pid(p)))
2301 return 0;
2302
2303 p->signal->has_child_subreaper = 1;
2304 return 1;
2305}
2306
2307int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2308{
2309 return -EINVAL;
2310}
2311
2312int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2313 unsigned long ctrl)
2314{
2315 return -EINVAL;
2316}
2317
2318#define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2319
2320#ifdef CONFIG_ANON_VMA_NAME
2321
2322#define ANON_VMA_NAME_MAX_LEN 80
2323#define ANON_VMA_NAME_INVALID_CHARS "\\`$[]"
2324
2325static inline bool is_valid_name_char(char ch)
2326{
2327 /* printable ascii characters, excluding ANON_VMA_NAME_INVALID_CHARS */
2328 return ch > 0x1f && ch < 0x7f &&
2329 !strchr(ANON_VMA_NAME_INVALID_CHARS, ch);
2330}
2331
2332static int prctl_set_vma(unsigned long opt, unsigned long addr,
2333 unsigned long size, unsigned long arg)
2334{
2335 struct mm_struct *mm = current->mm;
2336 const char __user *uname;
2337 struct anon_vma_name *anon_name = NULL;
2338 int error;
2339
2340 switch (opt) {
2341 case PR_SET_VMA_ANON_NAME:
2342 uname = (const char __user *)arg;
2343 if (uname) {
2344 char *name, *pch;
2345
2346 name = strndup_user(uname, ANON_VMA_NAME_MAX_LEN);
2347 if (IS_ERR(name))
2348 return PTR_ERR(name);
2349
2350 for (pch = name; *pch != '\0'; pch++) {
2351 if (!is_valid_name_char(*pch)) {
2352 kfree(name);
2353 return -EINVAL;
2354 }
2355 }
2356 /* anon_vma has its own copy */
2357 anon_name = anon_vma_name_alloc(name);
2358 kfree(name);
2359 if (!anon_name)
2360 return -ENOMEM;
2361
2362 }
2363
2364 mmap_write_lock(mm);
2365 error = madvise_set_anon_name(mm, addr, size, anon_name);
2366 mmap_write_unlock(mm);
2367 anon_vma_name_put(anon_name);
2368 break;
2369 default:
2370 error = -EINVAL;
2371 }
2372
2373 return error;
2374}
2375
2376#else /* CONFIG_ANON_VMA_NAME */
2377static int prctl_set_vma(unsigned long opt, unsigned long start,
2378 unsigned long size, unsigned long arg)
2379{
2380 return -EINVAL;
2381}
2382#endif /* CONFIG_ANON_VMA_NAME */
2383
2384static inline unsigned long get_current_mdwe(void)
2385{
2386 unsigned long ret = 0;
2387
2388 if (test_bit(MMF_HAS_MDWE, ¤t->mm->flags))
2389 ret |= PR_MDWE_REFUSE_EXEC_GAIN;
2390 if (test_bit(MMF_HAS_MDWE_NO_INHERIT, ¤t->mm->flags))
2391 ret |= PR_MDWE_NO_INHERIT;
2392
2393 return ret;
2394}
2395
2396static inline int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
2397 unsigned long arg4, unsigned long arg5)
2398{
2399 unsigned long current_bits;
2400
2401 if (arg3 || arg4 || arg5)
2402 return -EINVAL;
2403
2404 if (bits & ~(PR_MDWE_REFUSE_EXEC_GAIN | PR_MDWE_NO_INHERIT))
2405 return -EINVAL;
2406
2407 /* NO_INHERIT only makes sense with REFUSE_EXEC_GAIN */
2408 if (bits & PR_MDWE_NO_INHERIT && !(bits & PR_MDWE_REFUSE_EXEC_GAIN))
2409 return -EINVAL;
2410
2411 /*
2412 * EOPNOTSUPP might be more appropriate here in principle, but
2413 * existing userspace depends on EINVAL specifically.
2414 */
2415 if (!arch_memory_deny_write_exec_supported())
2416 return -EINVAL;
2417
2418 current_bits = get_current_mdwe();
2419 if (current_bits && current_bits != bits)
2420 return -EPERM; /* Cannot unset the flags */
2421
2422 if (bits & PR_MDWE_NO_INHERIT)
2423 set_bit(MMF_HAS_MDWE_NO_INHERIT, ¤t->mm->flags);
2424 if (bits & PR_MDWE_REFUSE_EXEC_GAIN)
2425 set_bit(MMF_HAS_MDWE, ¤t->mm->flags);
2426
2427 return 0;
2428}
2429
2430static inline int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
2431 unsigned long arg4, unsigned long arg5)
2432{
2433 if (arg2 || arg3 || arg4 || arg5)
2434 return -EINVAL;
2435 return get_current_mdwe();
2436}
2437
2438static int prctl_get_auxv(void __user *addr, unsigned long len)
2439{
2440 struct mm_struct *mm = current->mm;
2441 unsigned long size = min_t(unsigned long, sizeof(mm->saved_auxv), len);
2442
2443 if (size && copy_to_user(addr, mm->saved_auxv, size))
2444 return -EFAULT;
2445 return sizeof(mm->saved_auxv);
2446}
2447
2448SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2449 unsigned long, arg4, unsigned long, arg5)
2450{
2451 struct task_struct *me = current;
2452 unsigned char comm[sizeof(me->comm)];
2453 long error;
2454
2455 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2456 if (error != -ENOSYS)
2457 return error;
2458
2459 error = 0;
2460 switch (option) {
2461 case PR_SET_PDEATHSIG:
2462 if (!valid_signal(arg2)) {
2463 error = -EINVAL;
2464 break;
2465 }
2466 me->pdeath_signal = arg2;
2467 break;
2468 case PR_GET_PDEATHSIG:
2469 error = put_user(me->pdeath_signal, (int __user *)arg2);
2470 break;
2471 case PR_GET_DUMPABLE:
2472 error = get_dumpable(me->mm);
2473 break;
2474 case PR_SET_DUMPABLE:
2475 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2476 error = -EINVAL;
2477 break;
2478 }
2479 set_dumpable(me->mm, arg2);
2480 break;
2481
2482 case PR_SET_UNALIGN:
2483 error = SET_UNALIGN_CTL(me, arg2);
2484 break;
2485 case PR_GET_UNALIGN:
2486 error = GET_UNALIGN_CTL(me, arg2);
2487 break;
2488 case PR_SET_FPEMU:
2489 error = SET_FPEMU_CTL(me, arg2);
2490 break;
2491 case PR_GET_FPEMU:
2492 error = GET_FPEMU_CTL(me, arg2);
2493 break;
2494 case PR_SET_FPEXC:
2495 error = SET_FPEXC_CTL(me, arg2);
2496 break;
2497 case PR_GET_FPEXC:
2498 error = GET_FPEXC_CTL(me, arg2);
2499 break;
2500 case PR_GET_TIMING:
2501 error = PR_TIMING_STATISTICAL;
2502 break;
2503 case PR_SET_TIMING:
2504 if (arg2 != PR_TIMING_STATISTICAL)
2505 error = -EINVAL;
2506 break;
2507 case PR_SET_NAME:
2508 comm[sizeof(me->comm) - 1] = 0;
2509 if (strncpy_from_user(comm, (char __user *)arg2,
2510 sizeof(me->comm) - 1) < 0)
2511 return -EFAULT;
2512 set_task_comm(me, comm);
2513 proc_comm_connector(me);
2514 break;
2515 case PR_GET_NAME:
2516 get_task_comm(comm, me);
2517 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2518 return -EFAULT;
2519 break;
2520 case PR_GET_ENDIAN:
2521 error = GET_ENDIAN(me, arg2);
2522 break;
2523 case PR_SET_ENDIAN:
2524 error = SET_ENDIAN(me, arg2);
2525 break;
2526 case PR_GET_SECCOMP:
2527 error = prctl_get_seccomp();
2528 break;
2529 case PR_SET_SECCOMP:
2530 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2531 break;
2532 case PR_GET_TSC:
2533 error = GET_TSC_CTL(arg2);
2534 break;
2535 case PR_SET_TSC:
2536 error = SET_TSC_CTL(arg2);
2537 break;
2538 case PR_TASK_PERF_EVENTS_DISABLE:
2539 error = perf_event_task_disable();
2540 break;
2541 case PR_TASK_PERF_EVENTS_ENABLE:
2542 error = perf_event_task_enable();
2543 break;
2544 case PR_GET_TIMERSLACK:
2545 if (current->timer_slack_ns > ULONG_MAX)
2546 error = ULONG_MAX;
2547 else
2548 error = current->timer_slack_ns;
2549 break;
2550 case PR_SET_TIMERSLACK:
2551 if (arg2 <= 0)
2552 current->timer_slack_ns =
2553 current->default_timer_slack_ns;
2554 else
2555 current->timer_slack_ns = arg2;
2556 break;
2557 case PR_MCE_KILL:
2558 if (arg4 | arg5)
2559 return -EINVAL;
2560 switch (arg2) {
2561 case PR_MCE_KILL_CLEAR:
2562 if (arg3 != 0)
2563 return -EINVAL;
2564 current->flags &= ~PF_MCE_PROCESS;
2565 break;
2566 case PR_MCE_KILL_SET:
2567 current->flags |= PF_MCE_PROCESS;
2568 if (arg3 == PR_MCE_KILL_EARLY)
2569 current->flags |= PF_MCE_EARLY;
2570 else if (arg3 == PR_MCE_KILL_LATE)
2571 current->flags &= ~PF_MCE_EARLY;
2572 else if (arg3 == PR_MCE_KILL_DEFAULT)
2573 current->flags &=
2574 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2575 else
2576 return -EINVAL;
2577 break;
2578 default:
2579 return -EINVAL;
2580 }
2581 break;
2582 case PR_MCE_KILL_GET:
2583 if (arg2 | arg3 | arg4 | arg5)
2584 return -EINVAL;
2585 if (current->flags & PF_MCE_PROCESS)
2586 error = (current->flags & PF_MCE_EARLY) ?
2587 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2588 else
2589 error = PR_MCE_KILL_DEFAULT;
2590 break;
2591 case PR_SET_MM:
2592 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2593 break;
2594 case PR_GET_TID_ADDRESS:
2595 error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2596 break;
2597 case PR_SET_CHILD_SUBREAPER:
2598 me->signal->is_child_subreaper = !!arg2;
2599 if (!arg2)
2600 break;
2601
2602 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2603 break;
2604 case PR_GET_CHILD_SUBREAPER:
2605 error = put_user(me->signal->is_child_subreaper,
2606 (int __user *)arg2);
2607 break;
2608 case PR_SET_NO_NEW_PRIVS:
2609 if (arg2 != 1 || arg3 || arg4 || arg5)
2610 return -EINVAL;
2611
2612 task_set_no_new_privs(current);
2613 break;
2614 case PR_GET_NO_NEW_PRIVS:
2615 if (arg2 || arg3 || arg4 || arg5)
2616 return -EINVAL;
2617 return task_no_new_privs(current) ? 1 : 0;
2618 case PR_GET_THP_DISABLE:
2619 if (arg2 || arg3 || arg4 || arg5)
2620 return -EINVAL;
2621 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2622 break;
2623 case PR_SET_THP_DISABLE:
2624 if (arg3 || arg4 || arg5)
2625 return -EINVAL;
2626 if (mmap_write_lock_killable(me->mm))
2627 return -EINTR;
2628 if (arg2)
2629 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2630 else
2631 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2632 mmap_write_unlock(me->mm);
2633 break;
2634 case PR_MPX_ENABLE_MANAGEMENT:
2635 case PR_MPX_DISABLE_MANAGEMENT:
2636 /* No longer implemented: */
2637 return -EINVAL;
2638 case PR_SET_FP_MODE:
2639 error = SET_FP_MODE(me, arg2);
2640 break;
2641 case PR_GET_FP_MODE:
2642 error = GET_FP_MODE(me);
2643 break;
2644 case PR_SVE_SET_VL:
2645 error = SVE_SET_VL(arg2);
2646 break;
2647 case PR_SVE_GET_VL:
2648 error = SVE_GET_VL();
2649 break;
2650 case PR_SME_SET_VL:
2651 error = SME_SET_VL(arg2);
2652 break;
2653 case PR_SME_GET_VL:
2654 error = SME_GET_VL();
2655 break;
2656 case PR_GET_SPECULATION_CTRL:
2657 if (arg3 || arg4 || arg5)
2658 return -EINVAL;
2659 error = arch_prctl_spec_ctrl_get(me, arg2);
2660 break;
2661 case PR_SET_SPECULATION_CTRL:
2662 if (arg4 || arg5)
2663 return -EINVAL;
2664 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2665 break;
2666 case PR_PAC_RESET_KEYS:
2667 if (arg3 || arg4 || arg5)
2668 return -EINVAL;
2669 error = PAC_RESET_KEYS(me, arg2);
2670 break;
2671 case PR_PAC_SET_ENABLED_KEYS:
2672 if (arg4 || arg5)
2673 return -EINVAL;
2674 error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2675 break;
2676 case PR_PAC_GET_ENABLED_KEYS:
2677 if (arg2 || arg3 || arg4 || arg5)
2678 return -EINVAL;
2679 error = PAC_GET_ENABLED_KEYS(me);
2680 break;
2681 case PR_SET_TAGGED_ADDR_CTRL:
2682 if (arg3 || arg4 || arg5)
2683 return -EINVAL;
2684 error = SET_TAGGED_ADDR_CTRL(arg2);
2685 break;
2686 case PR_GET_TAGGED_ADDR_CTRL:
2687 if (arg2 || arg3 || arg4 || arg5)
2688 return -EINVAL;
2689 error = GET_TAGGED_ADDR_CTRL();
2690 break;
2691 case PR_SET_IO_FLUSHER:
2692 if (!capable(CAP_SYS_RESOURCE))
2693 return -EPERM;
2694
2695 if (arg3 || arg4 || arg5)
2696 return -EINVAL;
2697
2698 if (arg2 == 1)
2699 current->flags |= PR_IO_FLUSHER;
2700 else if (!arg2)
2701 current->flags &= ~PR_IO_FLUSHER;
2702 else
2703 return -EINVAL;
2704 break;
2705 case PR_GET_IO_FLUSHER:
2706 if (!capable(CAP_SYS_RESOURCE))
2707 return -EPERM;
2708
2709 if (arg2 || arg3 || arg4 || arg5)
2710 return -EINVAL;
2711
2712 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2713 break;
2714 case PR_SET_SYSCALL_USER_DISPATCH:
2715 error = set_syscall_user_dispatch(arg2, arg3, arg4,
2716 (char __user *) arg5);
2717 break;
2718#ifdef CONFIG_SCHED_CORE
2719 case PR_SCHED_CORE:
2720 error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2721 break;
2722#endif
2723 case PR_SET_MDWE:
2724 error = prctl_set_mdwe(arg2, arg3, arg4, arg5);
2725 break;
2726 case PR_GET_MDWE:
2727 error = prctl_get_mdwe(arg2, arg3, arg4, arg5);
2728 break;
2729 case PR_SET_VMA:
2730 error = prctl_set_vma(arg2, arg3, arg4, arg5);
2731 break;
2732 case PR_GET_AUXV:
2733 if (arg4 || arg5)
2734 return -EINVAL;
2735 error = prctl_get_auxv((void __user *)arg2, arg3);
2736 break;
2737#ifdef CONFIG_KSM
2738 case PR_SET_MEMORY_MERGE:
2739 if (arg3 || arg4 || arg5)
2740 return -EINVAL;
2741 if (mmap_write_lock_killable(me->mm))
2742 return -EINTR;
2743
2744 if (arg2)
2745 error = ksm_enable_merge_any(me->mm);
2746 else
2747 error = ksm_disable_merge_any(me->mm);
2748 mmap_write_unlock(me->mm);
2749 break;
2750 case PR_GET_MEMORY_MERGE:
2751 if (arg2 || arg3 || arg4 || arg5)
2752 return -EINVAL;
2753
2754 error = !!test_bit(MMF_VM_MERGE_ANY, &me->mm->flags);
2755 break;
2756#endif
2757 case PR_RISCV_V_SET_CONTROL:
2758 error = RISCV_V_SET_CONTROL(arg2);
2759 break;
2760 case PR_RISCV_V_GET_CONTROL:
2761 error = RISCV_V_GET_CONTROL();
2762 break;
2763 default:
2764 error = -EINVAL;
2765 break;
2766 }
2767 return error;
2768}
2769
2770SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2771 struct getcpu_cache __user *, unused)
2772{
2773 int err = 0;
2774 int cpu = raw_smp_processor_id();
2775
2776 if (cpup)
2777 err |= put_user(cpu, cpup);
2778 if (nodep)
2779 err |= put_user(cpu_to_node(cpu), nodep);
2780 return err ? -EFAULT : 0;
2781}
2782
2783/**
2784 * do_sysinfo - fill in sysinfo struct
2785 * @info: pointer to buffer to fill
2786 */
2787static int do_sysinfo(struct sysinfo *info)
2788{
2789 unsigned long mem_total, sav_total;
2790 unsigned int mem_unit, bitcount;
2791 struct timespec64 tp;
2792
2793 memset(info, 0, sizeof(struct sysinfo));
2794
2795 ktime_get_boottime_ts64(&tp);
2796 timens_add_boottime(&tp);
2797 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2798
2799 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2800
2801 info->procs = nr_threads;
2802
2803 si_meminfo(info);
2804 si_swapinfo(info);
2805
2806 /*
2807 * If the sum of all the available memory (i.e. ram + swap)
2808 * is less than can be stored in a 32 bit unsigned long then
2809 * we can be binary compatible with 2.2.x kernels. If not,
2810 * well, in that case 2.2.x was broken anyways...
2811 *
2812 * -Erik Andersen <andersee@debian.org>
2813 */
2814
2815 mem_total = info->totalram + info->totalswap;
2816 if (mem_total < info->totalram || mem_total < info->totalswap)
2817 goto out;
2818 bitcount = 0;
2819 mem_unit = info->mem_unit;
2820 while (mem_unit > 1) {
2821 bitcount++;
2822 mem_unit >>= 1;
2823 sav_total = mem_total;
2824 mem_total <<= 1;
2825 if (mem_total < sav_total)
2826 goto out;
2827 }
2828
2829 /*
2830 * If mem_total did not overflow, multiply all memory values by
2831 * info->mem_unit and set it to 1. This leaves things compatible
2832 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2833 * kernels...
2834 */
2835
2836 info->mem_unit = 1;
2837 info->totalram <<= bitcount;
2838 info->freeram <<= bitcount;
2839 info->sharedram <<= bitcount;
2840 info->bufferram <<= bitcount;
2841 info->totalswap <<= bitcount;
2842 info->freeswap <<= bitcount;
2843 info->totalhigh <<= bitcount;
2844 info->freehigh <<= bitcount;
2845
2846out:
2847 return 0;
2848}
2849
2850SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2851{
2852 struct sysinfo val;
2853
2854 do_sysinfo(&val);
2855
2856 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2857 return -EFAULT;
2858
2859 return 0;
2860}
2861
2862#ifdef CONFIG_COMPAT
2863struct compat_sysinfo {
2864 s32 uptime;
2865 u32 loads[3];
2866 u32 totalram;
2867 u32 freeram;
2868 u32 sharedram;
2869 u32 bufferram;
2870 u32 totalswap;
2871 u32 freeswap;
2872 u16 procs;
2873 u16 pad;
2874 u32 totalhigh;
2875 u32 freehigh;
2876 u32 mem_unit;
2877 char _f[20-2*sizeof(u32)-sizeof(int)];
2878};
2879
2880COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2881{
2882 struct sysinfo s;
2883 struct compat_sysinfo s_32;
2884
2885 do_sysinfo(&s);
2886
2887 /* Check to see if any memory value is too large for 32-bit and scale
2888 * down if needed
2889 */
2890 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2891 int bitcount = 0;
2892
2893 while (s.mem_unit < PAGE_SIZE) {
2894 s.mem_unit <<= 1;
2895 bitcount++;
2896 }
2897
2898 s.totalram >>= bitcount;
2899 s.freeram >>= bitcount;
2900 s.sharedram >>= bitcount;
2901 s.bufferram >>= bitcount;
2902 s.totalswap >>= bitcount;
2903 s.freeswap >>= bitcount;
2904 s.totalhigh >>= bitcount;
2905 s.freehigh >>= bitcount;
2906 }
2907
2908 memset(&s_32, 0, sizeof(s_32));
2909 s_32.uptime = s.uptime;
2910 s_32.loads[0] = s.loads[0];
2911 s_32.loads[1] = s.loads[1];
2912 s_32.loads[2] = s.loads[2];
2913 s_32.totalram = s.totalram;
2914 s_32.freeram = s.freeram;
2915 s_32.sharedram = s.sharedram;
2916 s_32.bufferram = s.bufferram;
2917 s_32.totalswap = s.totalswap;
2918 s_32.freeswap = s.freeswap;
2919 s_32.procs = s.procs;
2920 s_32.totalhigh = s.totalhigh;
2921 s_32.freehigh = s.freehigh;
2922 s_32.mem_unit = s.mem_unit;
2923 if (copy_to_user(info, &s_32, sizeof(s_32)))
2924 return -EFAULT;
2925 return 0;
2926}
2927#endif /* CONFIG_COMPAT */