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