projects/linux·blob v5.11-rc3 · refs

2702 lines · 63.1 KiB

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