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