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