projects / GitHub / MotorolaMobilityLLC / kernel-slsi.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge 4.14.44 into android-4.14
[GitHub/MotorolaMobilityLLC/kernel-slsi.git] / kernel / sys.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/kernel/sys.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8 #include <linux/export.h>
9 #include <linux/mm.h>
10 #include <linux/utsname.h>
11 #include <linux/mman.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
15 #include <linux/fs.h>
16 #include <linux/kmod.h>
17 #include <linux/perf_event.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/file.h>
40 #include <linux/mount.h>
41 #include <linux/gfp.h>
42 #include <linux/syscore_ops.h>
43 #include <linux/version.h>
44 #include <linux/ctype.h>
45 #include <linux/mm.h>
46 #include <linux/mempolicy.h>
47
48 #include <linux/compat.h>
49 #include <linux/syscalls.h>
50 #include <linux/kprobes.h>
51 #include <linux/user_namespace.h>
52 #include <linux/binfmts.h>
53
54 #include <linux/sched.h>
55 #include <linux/sched/autogroup.h>
56 #include <linux/sched/loadavg.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/mm.h>
59 #include <linux/sched/coredump.h>
60 #include <linux/sched/task.h>
61 #include <linux/sched/cputime.h>
62 #include <linux/rcupdate.h>
63 #include <linux/uidgid.h>
64 #include <linux/cred.h>
65
66 #include <linux/nospec.h>
67
68 #include <linux/kmsg_dump.h>
69 /* Move somewhere else to avoid recompiling? */
70 #include <generated/utsrelease.h>
71
72 #include <linux/uaccess.h>
73 #include <asm/io.h>
74 #include <asm/unistd.h>
75
76 #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
119 /*
120 * this is where the system-wide overflow UID and GID are defined, for
121 * architectures that now have 32-bit UID/GID but didn't in the past
122 */
123
124 int overflowuid = DEFAULT_OVERFLOWUID;
125 int overflowgid = DEFAULT_OVERFLOWGID;
126
127 EXPORT_SYMBOL(overflowuid);
128 EXPORT_SYMBOL(overflowgid);
129
130 /*
131 * the same as above, but for filesystems which can only store a 16-bit
132 * UID and GID. as such, this is needed on all architectures
133 */
134
135 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
136 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
137
138 EXPORT_SYMBOL(fs_overflowuid);
139 EXPORT_SYMBOL(fs_overflowgid);
140
141 /*
142 * Returns true if current's euid is same as p's uid or euid,
143 * or has CAP_SYS_NICE to p's user_ns.
144 *
145 * Called with rcu_read_lock, creds are safe
146 */
147 static bool set_one_prio_perm(struct task_struct *p)
148 {
149 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
150
151 if (uid_eq(pcred->uid, cred->euid) ||
152 uid_eq(pcred->euid, cred->euid))
153 return true;
154 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
155 return true;
156 return false;
157 }
158
159 /*
160 * set the priority of a task
161 * - the caller must hold the RCU read lock
162 */
163 static int set_one_prio(struct task_struct *p, int niceval, int error)
164 {
165 int no_nice;
166
167 if (!set_one_prio_perm(p)) {
168 error = -EPERM;
169 goto out;
170 }
171 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
172 error = -EACCES;
173 goto out;
174 }
175 no_nice = security_task_setnice(p, niceval);
176 if (no_nice) {
177 error = no_nice;
178 goto out;
179 }
180 if (error == -ESRCH)
181 error = 0;
182 set_user_nice(p, niceval);
183 out:
184 return error;
185 }
186
187 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
188 {
189 struct task_struct *g, *p;
190 struct user_struct *user;
191 const struct cred *cred = current_cred();
192 int error = -EINVAL;
193 struct pid *pgrp;
194 kuid_t uid;
195
196 if (which > PRIO_USER || which < PRIO_PROCESS)
197 goto out;
198
199 /* normalize: avoid signed division (rounding problems) */
200 error = -ESRCH;
201 if (niceval < MIN_NICE)
202 niceval = MIN_NICE;
203 if (niceval > MAX_NICE)
204 niceval = MAX_NICE;
205
206 rcu_read_lock();
207 read_lock(&tasklist_lock);
208 switch (which) {
209 case PRIO_PROCESS:
210 if (who)
211 p = find_task_by_vpid(who);
212 else
213 p = current;
214 if (p)
215 error = set_one_prio(p, niceval, error);
216 break;
217 case PRIO_PGRP:
218 if (who)
219 pgrp = find_vpid(who);
220 else
221 pgrp = task_pgrp(current);
222 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
223 error = set_one_prio(p, niceval, error);
224 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
225 break;
226 case PRIO_USER:
227 uid = make_kuid(cred->user_ns, who);
228 user = cred->user;
229 if (!who)
230 uid = cred->uid;
231 else if (!uid_eq(uid, cred->uid)) {
232 user = find_user(uid);
233 if (!user)
234 goto out_unlock; /* No processes for this user */
235 }
236 do_each_thread(g, p) {
237 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
238 error = set_one_prio(p, niceval, error);
239 } while_each_thread(g, p);
240 if (!uid_eq(uid, cred->uid))
241 free_uid(user); /* For find_user() */
242 break;
243 }
244 out_unlock:
245 read_unlock(&tasklist_lock);
246 rcu_read_unlock();
247 out:
248 return error;
249 }
250
251 /*
252 * Ugh. To avoid negative return values, "getpriority()" will
253 * not return the normal nice-value, but a negated value that
254 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
255 * to stay compatible.
256 */
257 SYSCALL_DEFINE2(getpriority, int, which, int, who)
258 {
259 struct task_struct *g, *p;
260 struct user_struct *user;
261 const struct cred *cred = current_cred();
262 long niceval, retval = -ESRCH;
263 struct pid *pgrp;
264 kuid_t uid;
265
266 if (which > PRIO_USER || which < PRIO_PROCESS)
267 return -EINVAL;
268
269 rcu_read_lock();
270 read_lock(&tasklist_lock);
271 switch (which) {
272 case PRIO_PROCESS:
273 if (who)
274 p = find_task_by_vpid(who);
275 else
276 p = current;
277 if (p) {
278 niceval = nice_to_rlimit(task_nice(p));
279 if (niceval > retval)
280 retval = niceval;
281 }
282 break;
283 case PRIO_PGRP:
284 if (who)
285 pgrp = find_vpid(who);
286 else
287 pgrp = task_pgrp(current);
288 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
289 niceval = nice_to_rlimit(task_nice(p));
290 if (niceval > retval)
291 retval = niceval;
292 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
293 break;
294 case PRIO_USER:
295 uid = make_kuid(cred->user_ns, who);
296 user = cred->user;
297 if (!who)
298 uid = cred->uid;
299 else if (!uid_eq(uid, cred->uid)) {
300 user = find_user(uid);
301 if (!user)
302 goto out_unlock; /* No processes for this user */
303 }
304 do_each_thread(g, p) {
305 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
306 niceval = nice_to_rlimit(task_nice(p));
307 if (niceval > retval)
308 retval = niceval;
309 }
310 } while_each_thread(g, p);
311 if (!uid_eq(uid, cred->uid))
312 free_uid(user); /* for find_user() */
313 break;
314 }
315 out_unlock:
316 read_unlock(&tasklist_lock);
317 rcu_read_unlock();
318
319 return retval;
320 }
321
322 /*
323 * Unprivileged users may change the real gid to the effective gid
324 * or vice versa. (BSD-style)
325 *
326 * If you set the real gid at all, or set the effective gid to a value not
327 * equal to the real gid, then the saved gid is set to the new effective gid.
328 *
329 * This makes it possible for a setgid program to completely drop its
330 * privileges, which is often a useful assertion to make when you are doing
331 * a security audit over a program.
332 *
333 * The general idea is that a program which uses just setregid() will be
334 * 100% compatible with BSD. A program which uses just setgid() will be
335 * 100% compatible with POSIX with saved IDs.
336 *
337 * SMP: There are not races, the GIDs are checked only by filesystem
338 * operations (as far as semantic preservation is concerned).
339 */
340 #ifdef CONFIG_MULTIUSER
341 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
342 {
343 struct user_namespace *ns = current_user_ns();
344 const struct cred *old;
345 struct cred *new;
346 int retval;
347 kgid_t krgid, kegid;
348
349 krgid = make_kgid(ns, rgid);
350 kegid = make_kgid(ns, egid);
351
352 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
353 return -EINVAL;
354 if ((egid != (gid_t) -1) && !gid_valid(kegid))
355 return -EINVAL;
356
357 new = prepare_creds();
358 if (!new)
359 return -ENOMEM;
360 old = current_cred();
361
362 retval = -EPERM;
363 if (rgid != (gid_t) -1) {
364 if (gid_eq(old->gid, krgid) ||
365 gid_eq(old->egid, krgid) ||
366 ns_capable(old->user_ns, CAP_SETGID))
367 new->gid = krgid;
368 else
369 goto error;
370 }
371 if (egid != (gid_t) -1) {
372 if (gid_eq(old->gid, kegid) ||
373 gid_eq(old->egid, kegid) ||
374 gid_eq(old->sgid, kegid) ||
375 ns_capable(old->user_ns, CAP_SETGID))
376 new->egid = kegid;
377 else
378 goto error;
379 }
380
381 if (rgid != (gid_t) -1 ||
382 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
383 new->sgid = new->egid;
384 new->fsgid = new->egid;
385
386 return commit_creds(new);
387
388 error:
389 abort_creds(new);
390 return retval;
391 }
392
393 /*
394 * setgid() is implemented like SysV w/ SAVED_IDS
395 *
396 * SMP: Same implicit races as above.
397 */
398 SYSCALL_DEFINE1(setgid, gid_t, gid)
399 {
400 struct user_namespace *ns = current_user_ns();
401 const struct cred *old;
402 struct cred *new;
403 int retval;
404 kgid_t kgid;
405
406 kgid = make_kgid(ns, gid);
407 if (!gid_valid(kgid))
408 return -EINVAL;
409
410 new = prepare_creds();
411 if (!new)
412 return -ENOMEM;
413 old = current_cred();
414
415 retval = -EPERM;
416 if (ns_capable(old->user_ns, CAP_SETGID))
417 new->gid = new->egid = new->sgid = new->fsgid = kgid;
418 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
419 new->egid = new->fsgid = kgid;
420 else
421 goto error;
422
423 return commit_creds(new);
424
425 error:
426 abort_creds(new);
427 return retval;
428 }
429
430 /*
431 * change the user struct in a credentials set to match the new UID
432 */
433 static int set_user(struct cred *new)
434 {
435 struct user_struct *new_user;
436
437 new_user = alloc_uid(new->uid);
438 if (!new_user)
439 return -EAGAIN;
440
441 /*
442 * We don't fail in case of NPROC limit excess here because too many
443 * poorly written programs don't check set*uid() return code, assuming
444 * it never fails if called by root. We may still enforce NPROC limit
445 * for programs doing set*uid()+execve() by harmlessly deferring the
446 * failure to the execve() stage.
447 */
448 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
449 new_user != INIT_USER)
450 current->flags |= PF_NPROC_EXCEEDED;
451 else
452 current->flags &= ~PF_NPROC_EXCEEDED;
453
454 free_uid(new->user);
455 new->user = new_user;
456 return 0;
457 }
458
459 /*
460 * Unprivileged users may change the real uid to the effective uid
461 * or vice versa. (BSD-style)
462 *
463 * If you set the real uid at all, or set the effective uid to a value not
464 * equal to the real uid, then the saved uid is set to the new effective uid.
465 *
466 * This makes it possible for a setuid program to completely drop its
467 * privileges, which is often a useful assertion to make when you are doing
468 * a security audit over a program.
469 *
470 * The general idea is that a program which uses just setreuid() will be
471 * 100% compatible with BSD. A program which uses just setuid() will be
472 * 100% compatible with POSIX with saved IDs.
473 */
474 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
475 {
476 struct user_namespace *ns = current_user_ns();
477 const struct cred *old;
478 struct cred *new;
479 int retval;
480 kuid_t kruid, keuid;
481
482 kruid = make_kuid(ns, ruid);
483 keuid = make_kuid(ns, euid);
484
485 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
486 return -EINVAL;
487 if ((euid != (uid_t) -1) && !uid_valid(keuid))
488 return -EINVAL;
489
490 new = prepare_creds();
491 if (!new)
492 return -ENOMEM;
493 old = current_cred();
494
495 retval = -EPERM;
496 if (ruid != (uid_t) -1) {
497 new->uid = kruid;
498 if (!uid_eq(old->uid, kruid) &&
499 !uid_eq(old->euid, kruid) &&
500 !ns_capable(old->user_ns, CAP_SETUID))
501 goto error;
502 }
503
504 if (euid != (uid_t) -1) {
505 new->euid = keuid;
506 if (!uid_eq(old->uid, keuid) &&
507 !uid_eq(old->euid, keuid) &&
508 !uid_eq(old->suid, keuid) &&
509 !ns_capable(old->user_ns, CAP_SETUID))
510 goto error;
511 }
512
513 if (!uid_eq(new->uid, old->uid)) {
514 retval = set_user(new);
515 if (retval < 0)
516 goto error;
517 }
518 if (ruid != (uid_t) -1 ||
519 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
520 new->suid = new->euid;
521 new->fsuid = new->euid;
522
523 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
524 if (retval < 0)
525 goto error;
526
527 return commit_creds(new);
528
529 error:
530 abort_creds(new);
531 return retval;
532 }
533
534 /*
535 * setuid() is implemented like SysV with SAVED_IDS
536 *
537 * Note that SAVED_ID's is deficient in that a setuid root program
538 * like sendmail, for example, cannot set its uid to be a normal
539 * user and then switch back, because if you're root, setuid() sets
540 * the saved uid too. If you don't like this, blame the bright people
541 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
542 * will allow a root program to temporarily drop privileges and be able to
543 * regain them by swapping the real and effective uid.
544 */
545 SYSCALL_DEFINE1(setuid, uid_t, uid)
546 {
547 struct user_namespace *ns = current_user_ns();
548 const struct cred *old;
549 struct cred *new;
550 int retval;
551 kuid_t kuid;
552
553 kuid = make_kuid(ns, uid);
554 if (!uid_valid(kuid))
555 return -EINVAL;
556
557 new = prepare_creds();
558 if (!new)
559 return -ENOMEM;
560 old = current_cred();
561
562 retval = -EPERM;
563 if (ns_capable(old->user_ns, CAP_SETUID)) {
564 new->suid = new->uid = kuid;
565 if (!uid_eq(kuid, old->uid)) {
566 retval = set_user(new);
567 if (retval < 0)
568 goto error;
569 }
570 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
571 goto error;
572 }
573
574 new->fsuid = new->euid = kuid;
575
576 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
577 if (retval < 0)
578 goto error;
579
580 return commit_creds(new);
581
582 error:
583 abort_creds(new);
584 return retval;
585 }
586
587
588 /*
589 * This function implements a generic ability to update ruid, euid,
590 * and suid. This allows you to implement the 4.4 compatible seteuid().
591 */
592 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
593 {
594 struct user_namespace *ns = current_user_ns();
595 const struct cred *old;
596 struct cred *new;
597 int retval;
598 kuid_t kruid, keuid, ksuid;
599
600 kruid = make_kuid(ns, ruid);
601 keuid = make_kuid(ns, euid);
602 ksuid = make_kuid(ns, suid);
603
604 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
605 return -EINVAL;
606
607 if ((euid != (uid_t) -1) && !uid_valid(keuid))
608 return -EINVAL;
609
610 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
611 return -EINVAL;
612
613 new = prepare_creds();
614 if (!new)
615 return -ENOMEM;
616
617 old = current_cred();
618
619 retval = -EPERM;
620 if (!ns_capable(old->user_ns, CAP_SETUID)) {
621 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
622 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
623 goto error;
624 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
625 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
626 goto error;
627 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
628 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
629 goto error;
630 }
631
632 if (ruid != (uid_t) -1) {
633 new->uid = kruid;
634 if (!uid_eq(kruid, old->uid)) {
635 retval = set_user(new);
636 if (retval < 0)
637 goto error;
638 }
639 }
640 if (euid != (uid_t) -1)
641 new->euid = keuid;
642 if (suid != (uid_t) -1)
643 new->suid = ksuid;
644 new->fsuid = new->euid;
645
646 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
647 if (retval < 0)
648 goto error;
649
650 return commit_creds(new);
651
652 error:
653 abort_creds(new);
654 return retval;
655 }
656
657 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
658 {
659 const struct cred *cred = current_cred();
660 int retval;
661 uid_t ruid, euid, suid;
662
663 ruid = from_kuid_munged(cred->user_ns, cred->uid);
664 euid = from_kuid_munged(cred->user_ns, cred->euid);
665 suid = from_kuid_munged(cred->user_ns, cred->suid);
666
667 retval = put_user(ruid, ruidp);
668 if (!retval) {
669 retval = put_user(euid, euidp);
670 if (!retval)
671 return put_user(suid, suidp);
672 }
673 return retval;
674 }
675
676 /*
677 * Same as above, but for rgid, egid, sgid.
678 */
679 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
680 {
681 struct user_namespace *ns = current_user_ns();
682 const struct cred *old;
683 struct cred *new;
684 int retval;
685 kgid_t krgid, kegid, ksgid;
686
687 krgid = make_kgid(ns, rgid);
688 kegid = make_kgid(ns, egid);
689 ksgid = make_kgid(ns, sgid);
690
691 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
692 return -EINVAL;
693 if ((egid != (gid_t) -1) && !gid_valid(kegid))
694 return -EINVAL;
695 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
696 return -EINVAL;
697
698 new = prepare_creds();
699 if (!new)
700 return -ENOMEM;
701 old = current_cred();
702
703 retval = -EPERM;
704 if (!ns_capable(old->user_ns, CAP_SETGID)) {
705 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
706 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
707 goto error;
708 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
709 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
710 goto error;
711 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
712 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
713 goto error;
714 }
715
716 if (rgid != (gid_t) -1)
717 new->gid = krgid;
718 if (egid != (gid_t) -1)
719 new->egid = kegid;
720 if (sgid != (gid_t) -1)
721 new->sgid = ksgid;
722 new->fsgid = new->egid;
723
724 return commit_creds(new);
725
726 error:
727 abort_creds(new);
728 return retval;
729 }
730
731 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
732 {
733 const struct cred *cred = current_cred();
734 int retval;
735 gid_t rgid, egid, sgid;
736
737 rgid = from_kgid_munged(cred->user_ns, cred->gid);
738 egid = from_kgid_munged(cred->user_ns, cred->egid);
739 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
740
741 retval = put_user(rgid, rgidp);
742 if (!retval) {
743 retval = put_user(egid, egidp);
744 if (!retval)
745 retval = put_user(sgid, sgidp);
746 }
747
748 return retval;
749 }
750
751
752 /*
753 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
754 * is used for "access()" and for the NFS daemon (letting nfsd stay at
755 * whatever uid it wants to). It normally shadows "euid", except when
756 * explicitly set by setfsuid() or for access..
757 */
758 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
759 {
760 const struct cred *old;
761 struct cred *new;
762 uid_t old_fsuid;
763 kuid_t kuid;
764
765 old = current_cred();
766 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
767
768 kuid = make_kuid(old->user_ns, uid);
769 if (!uid_valid(kuid))
770 return old_fsuid;
771
772 new = prepare_creds();
773 if (!new)
774 return old_fsuid;
775
776 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
777 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
778 ns_capable(old->user_ns, CAP_SETUID)) {
779 if (!uid_eq(kuid, old->fsuid)) {
780 new->fsuid = kuid;
781 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
782 goto change_okay;
783 }
784 }
785
786 abort_creds(new);
787 return old_fsuid;
788
789 change_okay:
790 commit_creds(new);
791 return old_fsuid;
792 }
793
794 /*
795 * Samma på svenska..
796 */
797 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
798 {
799 const struct cred *old;
800 struct cred *new;
801 gid_t old_fsgid;
802 kgid_t kgid;
803
804 old = current_cred();
805 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
806
807 kgid = make_kgid(old->user_ns, gid);
808 if (!gid_valid(kgid))
809 return old_fsgid;
810
811 new = prepare_creds();
812 if (!new)
813 return old_fsgid;
814
815 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
816 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
817 ns_capable(old->user_ns, CAP_SETGID)) {
818 if (!gid_eq(kgid, old->fsgid)) {
819 new->fsgid = kgid;
820 goto change_okay;
821 }
822 }
823
824 abort_creds(new);
825 return old_fsgid;
826
827 change_okay:
828 commit_creds(new);
829 return old_fsgid;
830 }
831 #endif /* CONFIG_MULTIUSER */
832
833 /**
834 * sys_getpid - return the thread group id of the current process
835 *
836 * Note, despite the name, this returns the tgid not the pid. The tgid and
837 * the pid are identical unless CLONE_THREAD was specified on clone() in
838 * which case the tgid is the same in all threads of the same group.
839 *
840 * This is SMP safe as current->tgid does not change.
841 */
842 SYSCALL_DEFINE0(getpid)
843 {
844 return task_tgid_vnr(current);
845 }
846
847 /* Thread ID - the internal kernel "pid" */
848 SYSCALL_DEFINE0(gettid)
849 {
850 return task_pid_vnr(current);
851 }
852
853 /*
854 * Accessing ->real_parent is not SMP-safe, it could
855 * change from under us. However, we can use a stale
856 * value of ->real_parent under rcu_read_lock(), see
857 * release_task()->call_rcu(delayed_put_task_struct).
858 */
859 SYSCALL_DEFINE0(getppid)
860 {
861 int pid;
862
863 rcu_read_lock();
864 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
865 rcu_read_unlock();
866
867 return pid;
868 }
869
870 SYSCALL_DEFINE0(getuid)
871 {
872 /* Only we change this so SMP safe */
873 return from_kuid_munged(current_user_ns(), current_uid());
874 }
875
876 SYSCALL_DEFINE0(geteuid)
877 {
878 /* Only we change this so SMP safe */
879 return from_kuid_munged(current_user_ns(), current_euid());
880 }
881
882 SYSCALL_DEFINE0(getgid)
883 {
884 /* Only we change this so SMP safe */
885 return from_kgid_munged(current_user_ns(), current_gid());
886 }
887
888 SYSCALL_DEFINE0(getegid)
889 {
890 /* Only we change this so SMP safe */
891 return from_kgid_munged(current_user_ns(), current_egid());
892 }
893
894 static void do_sys_times(struct tms *tms)
895 {
896 u64 tgutime, tgstime, cutime, cstime;
897
898 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
899 cutime = current->signal->cutime;
900 cstime = current->signal->cstime;
901 tms->tms_utime = nsec_to_clock_t(tgutime);
902 tms->tms_stime = nsec_to_clock_t(tgstime);
903 tms->tms_cutime = nsec_to_clock_t(cutime);
904 tms->tms_cstime = nsec_to_clock_t(cstime);
905 }
906
907 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
908 {
909 if (tbuf) {
910 struct tms tmp;
911
912 do_sys_times(&tmp);
913 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
914 return -EFAULT;
915 }
916 force_successful_syscall_return();
917 return (long) jiffies_64_to_clock_t(get_jiffies_64());
918 }
919
920 #ifdef CONFIG_COMPAT
921 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
922 {
923 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
924 }
925
926 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
927 {
928 if (tbuf) {
929 struct tms tms;
930 struct compat_tms tmp;
931
932 do_sys_times(&tms);
933 /* Convert our struct tms to the compat version. */
934 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
935 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
936 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
937 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
938 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
939 return -EFAULT;
940 }
941 force_successful_syscall_return();
942 return compat_jiffies_to_clock_t(jiffies);
943 }
944 #endif
945
946 /*
947 * This needs some heavy checking ...
948 * I just haven't the stomach for it. I also don't fully
949 * understand sessions/pgrp etc. Let somebody who does explain it.
950 *
951 * OK, I think I have the protection semantics right.... this is really
952 * only important on a multi-user system anyway, to make sure one user
953 * can't send a signal to a process owned by another. -TYT, 12/12/91
954 *
955 * !PF_FORKNOEXEC check to conform completely to POSIX.
956 */
957 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
958 {
959 struct task_struct *p;
960 struct task_struct *group_leader = current->group_leader;
961 struct pid *pgrp;
962 int err;
963
964 if (!pid)
965 pid = task_pid_vnr(group_leader);
966 if (!pgid)
967 pgid = pid;
968 if (pgid < 0)
969 return -EINVAL;
970 rcu_read_lock();
971
972 /* From this point forward we keep holding onto the tasklist lock
973 * so that our parent does not change from under us. -DaveM
974 */
975 write_lock_irq(&tasklist_lock);
976
977 err = -ESRCH;
978 p = find_task_by_vpid(pid);
979 if (!p)
980 goto out;
981
982 err = -EINVAL;
983 if (!thread_group_leader(p))
984 goto out;
985
986 if (same_thread_group(p->real_parent, group_leader)) {
987 err = -EPERM;
988 if (task_session(p) != task_session(group_leader))
989 goto out;
990 err = -EACCES;
991 if (!(p->flags & PF_FORKNOEXEC))
992 goto out;
993 } else {
994 err = -ESRCH;
995 if (p != group_leader)
996 goto out;
997 }
998
999 err = -EPERM;
1000 if (p->signal->leader)
1001 goto out;
1002
1003 pgrp = task_pid(p);
1004 if (pgid != pid) {
1005 struct task_struct *g;
1006
1007 pgrp = find_vpid(pgid);
1008 g = pid_task(pgrp, PIDTYPE_PGID);
1009 if (!g || task_session(g) != task_session(group_leader))
1010 goto out;
1011 }
1012
1013 err = security_task_setpgid(p, pgid);
1014 if (err)
1015 goto out;
1016
1017 if (task_pgrp(p) != pgrp)
1018 change_pid(p, PIDTYPE_PGID, pgrp);
1019
1020 err = 0;
1021 out:
1022 /* All paths lead to here, thus we are safe. -DaveM */
1023 write_unlock_irq(&tasklist_lock);
1024 rcu_read_unlock();
1025 return err;
1026 }
1027
1028 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1029 {
1030 struct task_struct *p;
1031 struct pid *grp;
1032 int retval;
1033
1034 rcu_read_lock();
1035 if (!pid)
1036 grp = task_pgrp(current);
1037 else {
1038 retval = -ESRCH;
1039 p = find_task_by_vpid(pid);
1040 if (!p)
1041 goto out;
1042 grp = task_pgrp(p);
1043 if (!grp)
1044 goto out;
1045
1046 retval = security_task_getpgid(p);
1047 if (retval)
1048 goto out;
1049 }
1050 retval = pid_vnr(grp);
1051 out:
1052 rcu_read_unlock();
1053 return retval;
1054 }
1055
1056 #ifdef __ARCH_WANT_SYS_GETPGRP
1057
1058 SYSCALL_DEFINE0(getpgrp)
1059 {
1060 return sys_getpgid(0);
1061 }
1062
1063 #endif
1064
1065 SYSCALL_DEFINE1(getsid, pid_t, pid)
1066 {
1067 struct task_struct *p;
1068 struct pid *sid;
1069 int retval;
1070
1071 rcu_read_lock();
1072 if (!pid)
1073 sid = task_session(current);
1074 else {
1075 retval = -ESRCH;
1076 p = find_task_by_vpid(pid);
1077 if (!p)
1078 goto out;
1079 sid = task_session(p);
1080 if (!sid)
1081 goto out;
1082
1083 retval = security_task_getsid(p);
1084 if (retval)
1085 goto out;
1086 }
1087 retval = pid_vnr(sid);
1088 out:
1089 rcu_read_unlock();
1090 return retval;
1091 }
1092
1093 static void set_special_pids(struct pid *pid)
1094 {
1095 struct task_struct *curr = current->group_leader;
1096
1097 if (task_session(curr) != pid)
1098 change_pid(curr, PIDTYPE_SID, pid);
1099
1100 if (task_pgrp(curr) != pid)
1101 change_pid(curr, PIDTYPE_PGID, pid);
1102 }
1103
1104 SYSCALL_DEFINE0(setsid)
1105 {
1106 struct task_struct *group_leader = current->group_leader;
1107 struct pid *sid = task_pid(group_leader);
1108 pid_t session = pid_vnr(sid);
1109 int err = -EPERM;
1110
1111 write_lock_irq(&tasklist_lock);
1112 /* Fail if I am already a session leader */
1113 if (group_leader->signal->leader)
1114 goto out;
1115
1116 /* Fail if a process group id already exists that equals the
1117 * proposed session id.
1118 */
1119 if (pid_task(sid, PIDTYPE_PGID))
1120 goto out;
1121
1122 group_leader->signal->leader = 1;
1123 set_special_pids(sid);
1124
1125 proc_clear_tty(group_leader);
1126
1127 err = session;
1128 out:
1129 write_unlock_irq(&tasklist_lock);
1130 if (err > 0) {
1131 proc_sid_connector(group_leader);
1132 sched_autogroup_create_attach(group_leader);
1133 }
1134 return err;
1135 }
1136
1137 DECLARE_RWSEM(uts_sem);
1138
1139 #ifdef COMPAT_UTS_MACHINE
1140 #define override_architecture(name) \
1141 (personality(current->personality) == PER_LINUX32 && \
1142 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1143 sizeof(COMPAT_UTS_MACHINE)))
1144 #else
1145 #define override_architecture(name) 0
1146 #endif
1147
1148 /*
1149 * Work around broken programs that cannot handle "Linux 3.0".
1150 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1151 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1152 */
1153 static int override_release(char __user *release, size_t len)
1154 {
1155 int ret = 0;
1156
1157 if (current->personality & UNAME26) {
1158 const char *rest = UTS_RELEASE;
1159 char buf[65] = { 0 };
1160 int ndots = 0;
1161 unsigned v;
1162 size_t copy;
1163
1164 while (*rest) {
1165 if (*rest == '.' && ++ndots >= 3)
1166 break;
1167 if (!isdigit(*rest) && *rest != '.')
1168 break;
1169 rest++;
1170 }
1171 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1172 copy = clamp_t(size_t, len, 1, sizeof(buf));
1173 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1174 ret = copy_to_user(release, buf, copy + 1);
1175 }
1176 return ret;
1177 }
1178
1179 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1180 {
1181 int errno = 0;
1182
1183 down_read(&uts_sem);
1184 if (copy_to_user(name, utsname(), sizeof *name))
1185 errno = -EFAULT;
1186 up_read(&uts_sem);
1187
1188 if (!errno && override_release(name->release, sizeof(name->release)))
1189 errno = -EFAULT;
1190 if (!errno && override_architecture(name))
1191 errno = -EFAULT;
1192 return errno;
1193 }
1194
1195 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1196 /*
1197 * Old cruft
1198 */
1199 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1200 {
1201 int error = 0;
1202
1203 if (!name)
1204 return -EFAULT;
1205
1206 down_read(&uts_sem);
1207 if (copy_to_user(name, utsname(), sizeof(*name)))
1208 error = -EFAULT;
1209 up_read(&uts_sem);
1210
1211 if (!error && override_release(name->release, sizeof(name->release)))
1212 error = -EFAULT;
1213 if (!error && override_architecture(name))
1214 error = -EFAULT;
1215 return error;
1216 }
1217
1218 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1219 {
1220 int error;
1221
1222 if (!name)
1223 return -EFAULT;
1224 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1225 return -EFAULT;
1226
1227 down_read(&uts_sem);
1228 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1229 __OLD_UTS_LEN);
1230 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1231 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1232 __OLD_UTS_LEN);
1233 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1234 error |= __copy_to_user(&name->release, &utsname()->release,
1235 __OLD_UTS_LEN);
1236 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1237 error |= __copy_to_user(&name->version, &utsname()->version,
1238 __OLD_UTS_LEN);
1239 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1240 error |= __copy_to_user(&name->machine, &utsname()->machine,
1241 __OLD_UTS_LEN);
1242 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1243 up_read(&uts_sem);
1244
1245 if (!error && override_architecture(name))
1246 error = -EFAULT;
1247 if (!error && override_release(name->release, sizeof(name->release)))
1248 error = -EFAULT;
1249 return error ? -EFAULT : 0;
1250 }
1251 #endif
1252
1253 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1254 {
1255 int errno;
1256 char tmp[__NEW_UTS_LEN];
1257
1258 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1259 return -EPERM;
1260
1261 if (len < 0 || len > __NEW_UTS_LEN)
1262 return -EINVAL;
1263 down_write(&uts_sem);
1264 errno = -EFAULT;
1265 if (!copy_from_user(tmp, name, len)) {
1266 struct new_utsname *u = utsname();
1267
1268 memcpy(u->nodename, tmp, len);
1269 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1270 errno = 0;
1271 uts_proc_notify(UTS_PROC_HOSTNAME);
1272 }
1273 up_write(&uts_sem);
1274 return errno;
1275 }
1276
1277 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1278
1279 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1280 {
1281 int i, errno;
1282 struct new_utsname *u;
1283
1284 if (len < 0)
1285 return -EINVAL;
1286 down_read(&uts_sem);
1287 u = utsname();
1288 i = 1 + strlen(u->nodename);
1289 if (i > len)
1290 i = len;
1291 errno = 0;
1292 if (copy_to_user(name, u->nodename, i))
1293 errno = -EFAULT;
1294 up_read(&uts_sem);
1295 return errno;
1296 }
1297
1298 #endif
1299
1300 /*
1301 * Only setdomainname; getdomainname can be implemented by calling
1302 * uname()
1303 */
1304 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1305 {
1306 int errno;
1307 char tmp[__NEW_UTS_LEN];
1308
1309 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1310 return -EPERM;
1311 if (len < 0 || len > __NEW_UTS_LEN)
1312 return -EINVAL;
1313
1314 down_write(&uts_sem);
1315 errno = -EFAULT;
1316 if (!copy_from_user(tmp, name, len)) {
1317 struct new_utsname *u = utsname();
1318
1319 memcpy(u->domainname, tmp, len);
1320 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1321 errno = 0;
1322 uts_proc_notify(UTS_PROC_DOMAINNAME);
1323 }
1324 up_write(&uts_sem);
1325 return errno;
1326 }
1327
1328 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1329 {
1330 struct rlimit value;
1331 int ret;
1332
1333 ret = do_prlimit(current, resource, NULL, &value);
1334 if (!ret)
1335 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1336
1337 return ret;
1338 }
1339
1340 #ifdef CONFIG_COMPAT
1341
1342 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1343 struct compat_rlimit __user *, rlim)
1344 {
1345 struct rlimit r;
1346 struct compat_rlimit r32;
1347
1348 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1349 return -EFAULT;
1350
1351 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1352 r.rlim_cur = RLIM_INFINITY;
1353 else
1354 r.rlim_cur = r32.rlim_cur;
1355 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1356 r.rlim_max = RLIM_INFINITY;
1357 else
1358 r.rlim_max = r32.rlim_max;
1359 return do_prlimit(current, resource, &r, NULL);
1360 }
1361
1362 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1363 struct compat_rlimit __user *, rlim)
1364 {
1365 struct rlimit r;
1366 int ret;
1367
1368 ret = do_prlimit(current, resource, NULL, &r);
1369 if (!ret) {
1370 struct compat_rlimit r32;
1371 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1372 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1373 else
1374 r32.rlim_cur = r.rlim_cur;
1375 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1376 r32.rlim_max = COMPAT_RLIM_INFINITY;
1377 else
1378 r32.rlim_max = r.rlim_max;
1379
1380 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1381 return -EFAULT;
1382 }
1383 return ret;
1384 }
1385
1386 #endif
1387
1388 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1389
1390 /*
1391 * Back compatibility for getrlimit. Needed for some apps.
1392 */
1393 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1394 struct rlimit __user *, rlim)
1395 {
1396 struct rlimit x;
1397 if (resource >= RLIM_NLIMITS)
1398 return -EINVAL;
1399
1400 task_lock(current->group_leader);
1401 x = current->signal->rlim[resource];
1402 task_unlock(current->group_leader);
1403 if (x.rlim_cur > 0x7FFFFFFF)
1404 x.rlim_cur = 0x7FFFFFFF;
1405 if (x.rlim_max > 0x7FFFFFFF)
1406 x.rlim_max = 0x7FFFFFFF;
1407 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1408 }
1409
1410 #ifdef CONFIG_COMPAT
1411 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1412 struct compat_rlimit __user *, rlim)
1413 {
1414 struct rlimit r;
1415
1416 if (resource >= RLIM_NLIMITS)
1417 return -EINVAL;
1418
1419 task_lock(current->group_leader);
1420 r = current->signal->rlim[resource];
1421 task_unlock(current->group_leader);
1422 if (r.rlim_cur > 0x7FFFFFFF)
1423 r.rlim_cur = 0x7FFFFFFF;
1424 if (r.rlim_max > 0x7FFFFFFF)
1425 r.rlim_max = 0x7FFFFFFF;
1426
1427 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1428 put_user(r.rlim_max, &rlim->rlim_max))
1429 return -EFAULT;
1430 return 0;
1431 }
1432 #endif
1433
1434 #endif
1435
1436 static inline bool rlim64_is_infinity(__u64 rlim64)
1437 {
1438 #if BITS_PER_LONG < 64
1439 return rlim64 >= ULONG_MAX;
1440 #else
1441 return rlim64 == RLIM64_INFINITY;
1442 #endif
1443 }
1444
1445 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1446 {
1447 if (rlim->rlim_cur == RLIM_INFINITY)
1448 rlim64->rlim_cur = RLIM64_INFINITY;
1449 else
1450 rlim64->rlim_cur = rlim->rlim_cur;
1451 if (rlim->rlim_max == RLIM_INFINITY)
1452 rlim64->rlim_max = RLIM64_INFINITY;
1453 else
1454 rlim64->rlim_max = rlim->rlim_max;
1455 }
1456
1457 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1458 {
1459 if (rlim64_is_infinity(rlim64->rlim_cur))
1460 rlim->rlim_cur = RLIM_INFINITY;
1461 else
1462 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1463 if (rlim64_is_infinity(rlim64->rlim_max))
1464 rlim->rlim_max = RLIM_INFINITY;
1465 else
1466 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1467 }
1468
1469 /* make sure you are allowed to change @tsk limits before calling this */
1470 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1471 struct rlimit *new_rlim, struct rlimit *old_rlim)
1472 {
1473 struct rlimit *rlim;
1474 int retval = 0;
1475
1476 if (resource >= RLIM_NLIMITS)
1477 return -EINVAL;
1478 if (new_rlim) {
1479 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1480 return -EINVAL;
1481 if (resource == RLIMIT_NOFILE &&
1482 new_rlim->rlim_max > sysctl_nr_open)
1483 return -EPERM;
1484 }
1485
1486 /* protect tsk->signal and tsk->sighand from disappearing */
1487 read_lock(&tasklist_lock);
1488 if (!tsk->sighand) {
1489 retval = -ESRCH;
1490 goto out;
1491 }
1492
1493 rlim = tsk->signal->rlim + resource;
1494 task_lock(tsk->group_leader);
1495 if (new_rlim) {
1496 /* Keep the capable check against init_user_ns until
1497 cgroups can contain all limits */
1498 if (new_rlim->rlim_max > rlim->rlim_max &&
1499 !capable(CAP_SYS_RESOURCE))
1500 retval = -EPERM;
1501 if (!retval)
1502 retval = security_task_setrlimit(tsk, resource, new_rlim);
1503 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1504 /*
1505 * The caller is asking for an immediate RLIMIT_CPU
1506 * expiry. But we use the zero value to mean "it was
1507 * never set". So let's cheat and make it one second
1508 * instead
1509 */
1510 new_rlim->rlim_cur = 1;
1511 }
1512 }
1513 if (!retval) {
1514 if (old_rlim)
1515 *old_rlim = *rlim;
1516 if (new_rlim)
1517 *rlim = *new_rlim;
1518 }
1519 task_unlock(tsk->group_leader);
1520
1521 /*
1522 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1523 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1524 * very long-standing error, and fixing it now risks breakage of
1525 * applications, so we live with it
1526 */
1527 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1528 new_rlim->rlim_cur != RLIM_INFINITY &&
1529 IS_ENABLED(CONFIG_POSIX_TIMERS))
1530 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1531 out:
1532 read_unlock(&tasklist_lock);
1533 return retval;
1534 }
1535
1536 /* rcu lock must be held */
1537 static int check_prlimit_permission(struct task_struct *task,
1538 unsigned int flags)
1539 {
1540 const struct cred *cred = current_cred(), *tcred;
1541 bool id_match;
1542
1543 if (current == task)
1544 return 0;
1545
1546 tcred = __task_cred(task);
1547 id_match = (uid_eq(cred->uid, tcred->euid) &&
1548 uid_eq(cred->uid, tcred->suid) &&
1549 uid_eq(cred->uid, tcred->uid) &&
1550 gid_eq(cred->gid, tcred->egid) &&
1551 gid_eq(cred->gid, tcred->sgid) &&
1552 gid_eq(cred->gid, tcred->gid));
1553 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1554 return -EPERM;
1555
1556 return security_task_prlimit(cred, tcred, flags);
1557 }
1558
1559 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1560 const struct rlimit64 __user *, new_rlim,
1561 struct rlimit64 __user *, old_rlim)
1562 {
1563 struct rlimit64 old64, new64;
1564 struct rlimit old, new;
1565 struct task_struct *tsk;
1566 unsigned int checkflags = 0;
1567 int ret;
1568
1569 if (old_rlim)
1570 checkflags |= LSM_PRLIMIT_READ;
1571
1572 if (new_rlim) {
1573 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1574 return -EFAULT;
1575 rlim64_to_rlim(&new64, &new);
1576 checkflags |= LSM_PRLIMIT_WRITE;
1577 }
1578
1579 rcu_read_lock();
1580 tsk = pid ? find_task_by_vpid(pid) : current;
1581 if (!tsk) {
1582 rcu_read_unlock();
1583 return -ESRCH;
1584 }
1585 ret = check_prlimit_permission(tsk, checkflags);
1586 if (ret) {
1587 rcu_read_unlock();
1588 return ret;
1589 }
1590 get_task_struct(tsk);
1591 rcu_read_unlock();
1592
1593 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1594 old_rlim ? &old : NULL);
1595
1596 if (!ret && old_rlim) {
1597 rlim_to_rlim64(&old, &old64);
1598 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1599 ret = -EFAULT;
1600 }
1601
1602 put_task_struct(tsk);
1603 return ret;
1604 }
1605
1606 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1607 {
1608 struct rlimit new_rlim;
1609
1610 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1611 return -EFAULT;
1612 return do_prlimit(current, resource, &new_rlim, NULL);
1613 }
1614
1615 /*
1616 * It would make sense to put struct rusage in the task_struct,
1617 * except that would make the task_struct be *really big*. After
1618 * task_struct gets moved into malloc'ed memory, it would
1619 * make sense to do this. It will make moving the rest of the information
1620 * a lot simpler! (Which we're not doing right now because we're not
1621 * measuring them yet).
1622 *
1623 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1624 * races with threads incrementing their own counters. But since word
1625 * reads are atomic, we either get new values or old values and we don't
1626 * care which for the sums. We always take the siglock to protect reading
1627 * the c* fields from p->signal from races with exit.c updating those
1628 * fields when reaping, so a sample either gets all the additions of a
1629 * given child after it's reaped, or none so this sample is before reaping.
1630 *
1631 * Locking:
1632 * We need to take the siglock for CHILDEREN, SELF and BOTH
1633 * for the cases current multithreaded, non-current single threaded
1634 * non-current multithreaded. Thread traversal is now safe with
1635 * the siglock held.
1636 * Strictly speaking, we donot need to take the siglock if we are current and
1637 * single threaded, as no one else can take our signal_struct away, no one
1638 * else can reap the children to update signal->c* counters, and no one else
1639 * can race with the signal-> fields. If we do not take any lock, the
1640 * signal-> fields could be read out of order while another thread was just
1641 * exiting. So we should place a read memory barrier when we avoid the lock.
1642 * On the writer side, write memory barrier is implied in __exit_signal
1643 * as __exit_signal releases the siglock spinlock after updating the signal->
1644 * fields. But we don't do this yet to keep things simple.
1645 *
1646 */
1647
1648 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1649 {
1650 r->ru_nvcsw += t->nvcsw;
1651 r->ru_nivcsw += t->nivcsw;
1652 r->ru_minflt += t->min_flt;
1653 r->ru_majflt += t->maj_flt;
1654 r->ru_inblock += task_io_get_inblock(t);
1655 r->ru_oublock += task_io_get_oublock(t);
1656 }
1657
1658 void getrusage(struct task_struct *p, int who, struct rusage *r)
1659 {
1660 struct task_struct *t;
1661 unsigned long flags;
1662 u64 tgutime, tgstime, utime, stime;
1663 unsigned long maxrss = 0;
1664
1665 memset((char *)r, 0, sizeof (*r));
1666 utime = stime = 0;
1667
1668 if (who == RUSAGE_THREAD) {
1669 task_cputime_adjusted(current, &utime, &stime);
1670 accumulate_thread_rusage(p, r);
1671 maxrss = p->signal->maxrss;
1672 goto out;
1673 }
1674
1675 if (!lock_task_sighand(p, &flags))
1676 return;
1677
1678 switch (who) {
1679 case RUSAGE_BOTH:
1680 case RUSAGE_CHILDREN:
1681 utime = p->signal->cutime;
1682 stime = p->signal->cstime;
1683 r->ru_nvcsw = p->signal->cnvcsw;
1684 r->ru_nivcsw = p->signal->cnivcsw;
1685 r->ru_minflt = p->signal->cmin_flt;
1686 r->ru_majflt = p->signal->cmaj_flt;
1687 r->ru_inblock = p->signal->cinblock;
1688 r->ru_oublock = p->signal->coublock;
1689 maxrss = p->signal->cmaxrss;
1690
1691 if (who == RUSAGE_CHILDREN)
1692 break;
1693
1694 case RUSAGE_SELF:
1695 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1696 utime += tgutime;
1697 stime += tgstime;
1698 r->ru_nvcsw += p->signal->nvcsw;
1699 r->ru_nivcsw += p->signal->nivcsw;
1700 r->ru_minflt += p->signal->min_flt;
1701 r->ru_majflt += p->signal->maj_flt;
1702 r->ru_inblock += p->signal->inblock;
1703 r->ru_oublock += p->signal->oublock;
1704 if (maxrss < p->signal->maxrss)
1705 maxrss = p->signal->maxrss;
1706 t = p;
1707 do {
1708 accumulate_thread_rusage(t, r);
1709 } while_each_thread(p, t);
1710 break;
1711
1712 default:
1713 BUG();
1714 }
1715 unlock_task_sighand(p, &flags);
1716
1717 out:
1718 r->ru_utime = ns_to_timeval(utime);
1719 r->ru_stime = ns_to_timeval(stime);
1720
1721 if (who != RUSAGE_CHILDREN) {
1722 struct mm_struct *mm = get_task_mm(p);
1723
1724 if (mm) {
1725 setmax_mm_hiwater_rss(&maxrss, mm);
1726 mmput(mm);
1727 }
1728 }
1729 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1730 }
1731
1732 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1733 {
1734 struct rusage r;
1735
1736 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1737 who != RUSAGE_THREAD)
1738 return -EINVAL;
1739
1740 getrusage(current, who, &r);
1741 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1742 }
1743
1744 #ifdef CONFIG_COMPAT
1745 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1746 {
1747 struct rusage r;
1748
1749 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1750 who != RUSAGE_THREAD)
1751 return -EINVAL;
1752
1753 getrusage(current, who, &r);
1754 return put_compat_rusage(&r, ru);
1755 }
1756 #endif
1757
1758 SYSCALL_DEFINE1(umask, int, mask)
1759 {
1760 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1761 return mask;
1762 }
1763
1764 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1765 {
1766 struct fd exe;
1767 struct file *old_exe, *exe_file;
1768 struct inode *inode;
1769 int err;
1770
1771 exe = fdget(fd);
1772 if (!exe.file)
1773 return -EBADF;
1774
1775 inode = file_inode(exe.file);
1776
1777 /*
1778 * Because the original mm->exe_file points to executable file, make
1779 * sure that this one is executable as well, to avoid breaking an
1780 * overall picture.
1781 */
1782 err = -EACCES;
1783 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1784 goto exit;
1785
1786 err = inode_permission(inode, MAY_EXEC);
1787 if (err)
1788 goto exit;
1789
1790 /*
1791 * Forbid mm->exe_file change if old file still mapped.
1792 */
1793 exe_file = get_mm_exe_file(mm);
1794 err = -EBUSY;
1795 if (exe_file) {
1796 struct vm_area_struct *vma;
1797
1798 down_read(&mm->mmap_sem);
1799 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1800 if (!vma->vm_file)
1801 continue;
1802 if (path_equal(&vma->vm_file->f_path,
1803 &exe_file->f_path))
1804 goto exit_err;
1805 }
1806
1807 up_read(&mm->mmap_sem);
1808 fput(exe_file);
1809 }
1810
1811 err = 0;
1812 /* set the new file, lockless */
1813 get_file(exe.file);
1814 old_exe = xchg(&mm->exe_file, exe.file);
1815 if (old_exe)
1816 fput(old_exe);
1817 exit:
1818 fdput(exe);
1819 return err;
1820 exit_err:
1821 up_read(&mm->mmap_sem);
1822 fput(exe_file);
1823 goto exit;
1824 }
1825
1826 /*
1827 * WARNING: we don't require any capability here so be very careful
1828 * in what is allowed for modification from userspace.
1829 */
1830 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1831 {
1832 unsigned long mmap_max_addr = TASK_SIZE;
1833 struct mm_struct *mm = current->mm;
1834 int error = -EINVAL, i;
1835
1836 static const unsigned char offsets[] = {
1837 offsetof(struct prctl_mm_map, start_code),
1838 offsetof(struct prctl_mm_map, end_code),
1839 offsetof(struct prctl_mm_map, start_data),
1840 offsetof(struct prctl_mm_map, end_data),
1841 offsetof(struct prctl_mm_map, start_brk),
1842 offsetof(struct prctl_mm_map, brk),
1843 offsetof(struct prctl_mm_map, start_stack),
1844 offsetof(struct prctl_mm_map, arg_start),
1845 offsetof(struct prctl_mm_map, arg_end),
1846 offsetof(struct prctl_mm_map, env_start),
1847 offsetof(struct prctl_mm_map, env_end),
1848 };
1849
1850 /*
1851 * Make sure the members are not somewhere outside
1852 * of allowed address space.
1853 */
1854 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1855 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1856
1857 if ((unsigned long)val >= mmap_max_addr ||
1858 (unsigned long)val < mmap_min_addr)
1859 goto out;
1860 }
1861
1862 /*
1863 * Make sure the pairs are ordered.
1864 */
1865 #define __prctl_check_order(__m1, __op, __m2) \
1866 ((unsigned long)prctl_map->__m1 __op \
1867 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1868 error = __prctl_check_order(start_code, <, end_code);
1869 error |= __prctl_check_order(start_data, <, end_data);
1870 error |= __prctl_check_order(start_brk, <=, brk);
1871 error |= __prctl_check_order(arg_start, <=, arg_end);
1872 error |= __prctl_check_order(env_start, <=, env_end);
1873 if (error)
1874 goto out;
1875 #undef __prctl_check_order
1876
1877 error = -EINVAL;
1878
1879 /*
1880 * @brk should be after @end_data in traditional maps.
1881 */
1882 if (prctl_map->start_brk <= prctl_map->end_data ||
1883 prctl_map->brk <= prctl_map->end_data)
1884 goto out;
1885
1886 /*
1887 * Neither we should allow to override limits if they set.
1888 */
1889 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1890 prctl_map->start_brk, prctl_map->end_data,
1891 prctl_map->start_data))
1892 goto out;
1893
1894 /*
1895 * Someone is trying to cheat the auxv vector.
1896 */
1897 if (prctl_map->auxv_size) {
1898 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1899 goto out;
1900 }
1901
1902 /*
1903 * Finally, make sure the caller has the rights to
1904 * change /proc/pid/exe link: only local sys admin should
1905 * be allowed to.
1906 */
1907 if (prctl_map->exe_fd != (u32)-1) {
1908 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1909 goto out;
1910 }
1911
1912 error = 0;
1913 out:
1914 return error;
1915 }
1916
1917 #ifdef CONFIG_CHECKPOINT_RESTORE
1918 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1919 {
1920 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1921 unsigned long user_auxv[AT_VECTOR_SIZE];
1922 struct mm_struct *mm = current->mm;
1923 int error;
1924
1925 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1926 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1927
1928 if (opt == PR_SET_MM_MAP_SIZE)
1929 return put_user((unsigned int)sizeof(prctl_map),
1930 (unsigned int __user *)addr);
1931
1932 if (data_size != sizeof(prctl_map))
1933 return -EINVAL;
1934
1935 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1936 return -EFAULT;
1937
1938 error = validate_prctl_map(&prctl_map);
1939 if (error)
1940 return error;
1941
1942 if (prctl_map.auxv_size) {
1943 memset(user_auxv, 0, sizeof(user_auxv));
1944 if (copy_from_user(user_auxv,
1945 (const void __user *)prctl_map.auxv,
1946 prctl_map.auxv_size))
1947 return -EFAULT;
1948
1949 /* Last entry must be AT_NULL as specification requires */
1950 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1951 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1952 }
1953
1954 if (prctl_map.exe_fd != (u32)-1) {
1955 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1956 if (error)
1957 return error;
1958 }
1959
1960 down_write(&mm->mmap_sem);
1961
1962 /*
1963 * We don't validate if these members are pointing to
1964 * real present VMAs because application may have correspond
1965 * VMAs already unmapped and kernel uses these members for statistics
1966 * output in procfs mostly, except
1967 *
1968 * - @start_brk/@brk which are used in do_brk but kernel lookups
1969 * for VMAs when updating these memvers so anything wrong written
1970 * here cause kernel to swear at userspace program but won't lead
1971 * to any problem in kernel itself
1972 */
1973
1974 mm->start_code = prctl_map.start_code;
1975 mm->end_code = prctl_map.end_code;
1976 mm->start_data = prctl_map.start_data;
1977 mm->end_data = prctl_map.end_data;
1978 mm->start_brk = prctl_map.start_brk;
1979 mm->brk = prctl_map.brk;
1980 mm->start_stack = prctl_map.start_stack;
1981 mm->arg_start = prctl_map.arg_start;
1982 mm->arg_end = prctl_map.arg_end;
1983 mm->env_start = prctl_map.env_start;
1984 mm->env_end = prctl_map.env_end;
1985
1986 /*
1987 * Note this update of @saved_auxv is lockless thus
1988 * if someone reads this member in procfs while we're
1989 * updating -- it may get partly updated results. It's
1990 * known and acceptable trade off: we leave it as is to
1991 * not introduce additional locks here making the kernel
1992 * more complex.
1993 */
1994 if (prctl_map.auxv_size)
1995 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1996
1997 up_write(&mm->mmap_sem);
1998 return 0;
1999 }
2000 #endif /* CONFIG_CHECKPOINT_RESTORE */
2001
2002 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2003 unsigned long len)
2004 {
2005 /*
2006 * This doesn't move the auxiliary vector itself since it's pinned to
2007 * mm_struct, but it permits filling the vector with new values. It's
2008 * up to the caller to provide sane values here, otherwise userspace
2009 * tools which use this vector might be unhappy.
2010 */
2011 unsigned long user_auxv[AT_VECTOR_SIZE];
2012
2013 if (len > sizeof(user_auxv))
2014 return -EINVAL;
2015
2016 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2017 return -EFAULT;
2018
2019 /* Make sure the last entry is always AT_NULL */
2020 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2021 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2022
2023 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2024
2025 task_lock(current);
2026 memcpy(mm->saved_auxv, user_auxv, len);
2027 task_unlock(current);
2028
2029 return 0;
2030 }
2031
2032 static int prctl_set_mm(int opt, unsigned long addr,
2033 unsigned long arg4, unsigned long arg5)
2034 {
2035 struct mm_struct *mm = current->mm;
2036 struct prctl_mm_map prctl_map;
2037 struct vm_area_struct *vma;
2038 int error;
2039
2040 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2041 opt != PR_SET_MM_MAP &&
2042 opt != PR_SET_MM_MAP_SIZE)))
2043 return -EINVAL;
2044
2045 #ifdef CONFIG_CHECKPOINT_RESTORE
2046 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2047 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2048 #endif
2049
2050 if (!capable(CAP_SYS_RESOURCE))
2051 return -EPERM;
2052
2053 if (opt == PR_SET_MM_EXE_FILE)
2054 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2055
2056 if (opt == PR_SET_MM_AUXV)
2057 return prctl_set_auxv(mm, addr, arg4);
2058
2059 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2060 return -EINVAL;
2061
2062 error = -EINVAL;
2063
2064 down_write(&mm->mmap_sem);
2065 vma = find_vma(mm, addr);
2066
2067 prctl_map.start_code = mm->start_code;
2068 prctl_map.end_code = mm->end_code;
2069 prctl_map.start_data = mm->start_data;
2070 prctl_map.end_data = mm->end_data;
2071 prctl_map.start_brk = mm->start_brk;
2072 prctl_map.brk = mm->brk;
2073 prctl_map.start_stack = mm->start_stack;
2074 prctl_map.arg_start = mm->arg_start;
2075 prctl_map.arg_end = mm->arg_end;
2076 prctl_map.env_start = mm->env_start;
2077 prctl_map.env_end = mm->env_end;
2078 prctl_map.auxv = NULL;
2079 prctl_map.auxv_size = 0;
2080 prctl_map.exe_fd = -1;
2081
2082 switch (opt) {
2083 case PR_SET_MM_START_CODE:
2084 prctl_map.start_code = addr;
2085 break;
2086 case PR_SET_MM_END_CODE:
2087 prctl_map.end_code = addr;
2088 break;
2089 case PR_SET_MM_START_DATA:
2090 prctl_map.start_data = addr;
2091 break;
2092 case PR_SET_MM_END_DATA:
2093 prctl_map.end_data = addr;
2094 break;
2095 case PR_SET_MM_START_STACK:
2096 prctl_map.start_stack = addr;
2097 break;
2098 case PR_SET_MM_START_BRK:
2099 prctl_map.start_brk = addr;
2100 break;
2101 case PR_SET_MM_BRK:
2102 prctl_map.brk = addr;
2103 break;
2104 case PR_SET_MM_ARG_START:
2105 prctl_map.arg_start = addr;
2106 break;
2107 case PR_SET_MM_ARG_END:
2108 prctl_map.arg_end = addr;
2109 break;
2110 case PR_SET_MM_ENV_START:
2111 prctl_map.env_start = addr;
2112 break;
2113 case PR_SET_MM_ENV_END:
2114 prctl_map.env_end = addr;
2115 break;
2116 default:
2117 goto out;
2118 }
2119
2120 error = validate_prctl_map(&prctl_map);
2121 if (error)
2122 goto out;
2123
2124 switch (opt) {
2125 /*
2126 * If command line arguments and environment
2127 * are placed somewhere else on stack, we can
2128 * set them up here, ARG_START/END to setup
2129 * command line argumets and ENV_START/END
2130 * for environment.
2131 */
2132 case PR_SET_MM_START_STACK:
2133 case PR_SET_MM_ARG_START:
2134 case PR_SET_MM_ARG_END:
2135 case PR_SET_MM_ENV_START:
2136 case PR_SET_MM_ENV_END:
2137 if (!vma) {
2138 error = -EFAULT;
2139 goto out;
2140 }
2141 }
2142
2143 mm->start_code = prctl_map.start_code;
2144 mm->end_code = prctl_map.end_code;
2145 mm->start_data = prctl_map.start_data;
2146 mm->end_data = prctl_map.end_data;
2147 mm->start_brk = prctl_map.start_brk;
2148 mm->brk = prctl_map.brk;
2149 mm->start_stack = prctl_map.start_stack;
2150 mm->arg_start = prctl_map.arg_start;
2151 mm->arg_end = prctl_map.arg_end;
2152 mm->env_start = prctl_map.env_start;
2153 mm->env_end = prctl_map.env_end;
2154
2155 error = 0;
2156 out:
2157 up_write(&mm->mmap_sem);
2158 return error;
2159 }
2160
2161 #ifdef CONFIG_CHECKPOINT_RESTORE
2162 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2163 {
2164 return put_user(me->clear_child_tid, tid_addr);
2165 }
2166 #else
2167 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2168 {
2169 return -EINVAL;
2170 }
2171 #endif
2172
2173 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2174 {
2175 /*
2176 * If task has has_child_subreaper - all its decendants
2177 * already have these flag too and new decendants will
2178 * inherit it on fork, skip them.
2179 *
2180 * If we've found child_reaper - skip descendants in
2181 * it's subtree as they will never get out pidns.
2182 */
2183 if (p->signal->has_child_subreaper ||
2184 is_child_reaper(task_pid(p)))
2185 return 0;
2186
2187 p->signal->has_child_subreaper = 1;
2188 return 1;
2189 }
2190
2191 #ifdef CONFIG_MMU
2192 static int prctl_update_vma_anon_name(struct vm_area_struct *vma,
2193 struct vm_area_struct **prev,
2194 unsigned long start, unsigned long end,
2195 const char __user *name_addr)
2196 {
2197 struct mm_struct *mm = vma->vm_mm;
2198 int error = 0;
2199 pgoff_t pgoff;
2200
2201 if (name_addr == vma_get_anon_name(vma)) {
2202 *prev = vma;
2203 goto out;
2204 }
2205
2206 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
2207 *prev = vma_merge(mm, *prev, start, end, vma->vm_flags, vma->anon_vma,
2208 vma->vm_file, pgoff, vma_policy(vma),
2209 vma->vm_userfaultfd_ctx, name_addr);
2210 if (*prev) {
2211 vma = *prev;
2212 goto success;
2213 }
2214
2215 *prev = vma;
2216
2217 if (start != vma->vm_start) {
2218 error = split_vma(mm, vma, start, 1);
2219 if (error)
2220 goto out;
2221 }
2222
2223 if (end != vma->vm_end) {
2224 error = split_vma(mm, vma, end, 0);
2225 if (error)
2226 goto out;
2227 }
2228
2229 success:
2230 if (!vma->vm_file)
2231 vma->anon_name = name_addr;
2232
2233 out:
2234 if (error == -ENOMEM)
2235 error = -EAGAIN;
2236 return error;
2237 }
2238
2239 static int prctl_set_vma_anon_name(unsigned long start, unsigned long end,
2240 unsigned long arg)
2241 {
2242 unsigned long tmp;
2243 struct vm_area_struct *vma, *prev;
2244 int unmapped_error = 0;
2245 int error = -EINVAL;
2246
2247 /*
2248 * If the interval [start,end) covers some unmapped address
2249 * ranges, just ignore them, but return -ENOMEM at the end.
2250 * - this matches the handling in madvise.
2251 */
2252 vma = find_vma_prev(current->mm, start, &prev);
2253 if (vma && start > vma->vm_start)
2254 prev = vma;
2255
2256 for (;;) {
2257 /* Still start < end. */
2258 error = -ENOMEM;
2259 if (!vma)
2260 return error;
2261
2262 /* Here start < (end|vma->vm_end). */
2263 if (start < vma->vm_start) {
2264 unmapped_error = -ENOMEM;
2265 start = vma->vm_start;
2266 if (start >= end)
2267 return error;
2268 }
2269
2270 /* Here vma->vm_start <= start < (end|vma->vm_end) */
2271 tmp = vma->vm_end;
2272 if (end < tmp)
2273 tmp = end;
2274
2275 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
2276 error = prctl_update_vma_anon_name(vma, &prev, start, tmp,
2277 (const char __user *)arg);
2278 if (error)
2279 return error;
2280 start = tmp;
2281 if (prev && start < prev->vm_end)
2282 start = prev->vm_end;
2283 error = unmapped_error;
2284 if (start >= end)
2285 return error;
2286 if (prev)
2287 vma = prev->vm_next;
2288 else /* madvise_remove dropped mmap_sem */
2289 vma = find_vma(current->mm, start);
2290 }
2291 }
2292
2293 static int prctl_set_vma(unsigned long opt, unsigned long start,
2294 unsigned long len_in, unsigned long arg)
2295 {
2296 struct mm_struct *mm = current->mm;
2297 int error;
2298 unsigned long len;
2299 unsigned long end;
2300
2301 if (start & ~PAGE_MASK)
2302 return -EINVAL;
2303 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
2304
2305 /* Check to see whether len was rounded up from small -ve to zero */
2306 if (len_in && !len)
2307 return -EINVAL;
2308
2309 end = start + len;
2310 if (end < start)
2311 return -EINVAL;
2312
2313 if (end == start)
2314 return 0;
2315
2316 down_write(&mm->mmap_sem);
2317
2318 switch (opt) {
2319 case PR_SET_VMA_ANON_NAME:
2320 error = prctl_set_vma_anon_name(start, end, arg);
2321 break;
2322 default:
2323 error = -EINVAL;
2324 }
2325
2326 up_write(&mm->mmap_sem);
2327
2328 return error;
2329 }
2330 #else /* CONFIG_MMU */
2331 static int prctl_set_vma(unsigned long opt, unsigned long start,
2332 unsigned long len_in, unsigned long arg)
2333 {
2334 return -EINVAL;
2335 }
2336 #endif
2337
2338 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2339 {
2340 return -EINVAL;
2341 }
2342
2343 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2344 unsigned long ctrl)
2345 {
2346 return -EINVAL;
2347 }
2348
2349 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2350 unsigned long, arg4, unsigned long, arg5)
2351 {
2352 struct task_struct *me = current;
2353 unsigned char comm[sizeof(me->comm)];
2354 long error;
2355
2356 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2357 if (error != -ENOSYS)
2358 return error;
2359
2360 error = 0;
2361 switch (option) {
2362 case PR_SET_PDEATHSIG:
2363 if (!valid_signal(arg2)) {
2364 error = -EINVAL;
2365 break;
2366 }
2367 me->pdeath_signal = arg2;
2368 break;
2369 case PR_GET_PDEATHSIG:
2370 error = put_user(me->pdeath_signal, (int __user *)arg2);
2371 break;
2372 case PR_GET_DUMPABLE:
2373 error = get_dumpable(me->mm);
2374 break;
2375 case PR_SET_DUMPABLE:
2376 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2377 error = -EINVAL;
2378 break;
2379 }
2380 set_dumpable(me->mm, arg2);
2381 break;
2382
2383 case PR_SET_UNALIGN:
2384 error = SET_UNALIGN_CTL(me, arg2);
2385 break;
2386 case PR_GET_UNALIGN:
2387 error = GET_UNALIGN_CTL(me, arg2);
2388 break;
2389 case PR_SET_FPEMU:
2390 error = SET_FPEMU_CTL(me, arg2);
2391 break;
2392 case PR_GET_FPEMU:
2393 error = GET_FPEMU_CTL(me, arg2);
2394 break;
2395 case PR_SET_FPEXC:
2396 error = SET_FPEXC_CTL(me, arg2);
2397 break;
2398 case PR_GET_FPEXC:
2399 error = GET_FPEXC_CTL(me, arg2);
2400 break;
2401 case PR_GET_TIMING:
2402 error = PR_TIMING_STATISTICAL;
2403 break;
2404 case PR_SET_TIMING:
2405 if (arg2 != PR_TIMING_STATISTICAL)
2406 error = -EINVAL;
2407 break;
2408 case PR_SET_NAME:
2409 comm[sizeof(me->comm) - 1] = 0;
2410 if (strncpy_from_user(comm, (char __user *)arg2,
2411 sizeof(me->comm) - 1) < 0)
2412 return -EFAULT;
2413 set_task_comm(me, comm);
2414 proc_comm_connector(me);
2415 break;
2416 case PR_GET_NAME:
2417 get_task_comm(comm, me);
2418 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2419 return -EFAULT;
2420 break;
2421 case PR_GET_ENDIAN:
2422 error = GET_ENDIAN(me, arg2);
2423 break;
2424 case PR_SET_ENDIAN:
2425 error = SET_ENDIAN(me, arg2);
2426 break;
2427 case PR_GET_SECCOMP:
2428 error = prctl_get_seccomp();
2429 break;
2430 case PR_SET_SECCOMP:
2431 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2432 break;
2433 case PR_GET_TSC:
2434 error = GET_TSC_CTL(arg2);
2435 break;
2436 case PR_SET_TSC:
2437 error = SET_TSC_CTL(arg2);
2438 break;
2439 case PR_TASK_PERF_EVENTS_DISABLE:
2440 error = perf_event_task_disable();
2441 break;
2442 case PR_TASK_PERF_EVENTS_ENABLE:
2443 error = perf_event_task_enable();
2444 break;
2445 case PR_GET_TIMERSLACK:
2446 if (current->timer_slack_ns > ULONG_MAX)
2447 error = ULONG_MAX;
2448 else
2449 error = current->timer_slack_ns;
2450 break;
2451 case PR_SET_TIMERSLACK:
2452 if (arg2 <= 0)
2453 current->timer_slack_ns =
2454 current->default_timer_slack_ns;
2455 else
2456 current->timer_slack_ns = arg2;
2457 break;
2458 case PR_MCE_KILL:
2459 if (arg4 | arg5)
2460 return -EINVAL;
2461 switch (arg2) {
2462 case PR_MCE_KILL_CLEAR:
2463 if (arg3 != 0)
2464 return -EINVAL;
2465 current->flags &= ~PF_MCE_PROCESS;
2466 break;
2467 case PR_MCE_KILL_SET:
2468 current->flags |= PF_MCE_PROCESS;
2469 if (arg3 == PR_MCE_KILL_EARLY)
2470 current->flags |= PF_MCE_EARLY;
2471 else if (arg3 == PR_MCE_KILL_LATE)
2472 current->flags &= ~PF_MCE_EARLY;
2473 else if (arg3 == PR_MCE_KILL_DEFAULT)
2474 current->flags &=
2475 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2476 else
2477 return -EINVAL;
2478 break;
2479 default:
2480 return -EINVAL;
2481 }
2482 break;
2483 case PR_MCE_KILL_GET:
2484 if (arg2 | arg3 | arg4 | arg5)
2485 return -EINVAL;
2486 if (current->flags & PF_MCE_PROCESS)
2487 error = (current->flags & PF_MCE_EARLY) ?
2488 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2489 else
2490 error = PR_MCE_KILL_DEFAULT;
2491 break;
2492 case PR_SET_MM:
2493 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2494 break;
2495 case PR_GET_TID_ADDRESS:
2496 error = prctl_get_tid_address(me, (int __user **)arg2);
2497 break;
2498 case PR_SET_CHILD_SUBREAPER:
2499 me->signal->is_child_subreaper = !!arg2;
2500 if (!arg2)
2501 break;
2502
2503 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2504 break;
2505 case PR_GET_CHILD_SUBREAPER:
2506 error = put_user(me->signal->is_child_subreaper,
2507 (int __user *)arg2);
2508 break;
2509 case PR_SET_NO_NEW_PRIVS:
2510 if (arg2 != 1 || arg3 || arg4 || arg5)
2511 return -EINVAL;
2512
2513 task_set_no_new_privs(current);
2514 break;
2515 case PR_GET_NO_NEW_PRIVS:
2516 if (arg2 || arg3 || arg4 || arg5)
2517 return -EINVAL;
2518 return task_no_new_privs(current) ? 1 : 0;
2519 case PR_GET_THP_DISABLE:
2520 if (arg2 || arg3 || arg4 || arg5)
2521 return -EINVAL;
2522 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2523 break;
2524 case PR_SET_THP_DISABLE:
2525 if (arg3 || arg4 || arg5)
2526 return -EINVAL;
2527 if (down_write_killable(&me->mm->mmap_sem))
2528 return -EINTR;
2529 if (arg2)
2530 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2531 else
2532 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2533 up_write(&me->mm->mmap_sem);
2534 break;
2535 case PR_MPX_ENABLE_MANAGEMENT:
2536 if (arg2 || arg3 || arg4 || arg5)
2537 return -EINVAL;
2538 error = MPX_ENABLE_MANAGEMENT();
2539 break;
2540 case PR_MPX_DISABLE_MANAGEMENT:
2541 if (arg2 || arg3 || arg4 || arg5)
2542 return -EINVAL;
2543 error = MPX_DISABLE_MANAGEMENT();
2544 break;
2545 case PR_SET_FP_MODE:
2546 error = SET_FP_MODE(me, arg2);
2547 break;
2548 case PR_GET_FP_MODE:
2549 error = GET_FP_MODE(me);
2550 break;
2551 case PR_GET_SPECULATION_CTRL:
2552 if (arg3 || arg4 || arg5)
2553 return -EINVAL;
2554 error = arch_prctl_spec_ctrl_get(me, arg2);
2555 break;
2556 case PR_SET_SPECULATION_CTRL:
2557 if (arg4 || arg5)
2558 return -EINVAL;
2559 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2560 break;
2561 case PR_SET_VMA:
2562 error = prctl_set_vma(arg2, arg3, arg4, arg5);
2563 break;
2564 default:
2565 error = -EINVAL;
2566 break;
2567 }
2568 return error;
2569 }
2570
2571 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2572 struct getcpu_cache __user *, unused)
2573 {
2574 int err = 0;
2575 int cpu = raw_smp_processor_id();
2576
2577 if (cpup)
2578 err |= put_user(cpu, cpup);
2579 if (nodep)
2580 err |= put_user(cpu_to_node(cpu), nodep);
2581 return err ? -EFAULT : 0;
2582 }
2583
2584 /**
2585 * do_sysinfo - fill in sysinfo struct
2586 * @info: pointer to buffer to fill
2587 */
2588 static int do_sysinfo(struct sysinfo *info)
2589 {
2590 unsigned long mem_total, sav_total;
2591 unsigned int mem_unit, bitcount;
2592 struct timespec tp;
2593
2594 memset(info, 0, sizeof(struct sysinfo));
2595
2596 get_monotonic_boottime(&tp);
2597 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2598
2599 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2600
2601 info->procs = nr_threads;
2602
2603 si_meminfo(info);
2604 si_swapinfo(info);
2605
2606 /*
2607 * If the sum of all the available memory (i.e. ram + swap)
2608 * is less than can be stored in a 32 bit unsigned long then
2609 * we can be binary compatible with 2.2.x kernels. If not,
2610 * well, in that case 2.2.x was broken anyways...
2611 *
2612 * -Erik Andersen <andersee@debian.org>
2613 */
2614
2615 mem_total = info->totalram + info->totalswap;
2616 if (mem_total < info->totalram || mem_total < info->totalswap)
2617 goto out;
2618 bitcount = 0;
2619 mem_unit = info->mem_unit;
2620 while (mem_unit > 1) {
2621 bitcount++;
2622 mem_unit >>= 1;
2623 sav_total = mem_total;
2624 mem_total <<= 1;
2625 if (mem_total < sav_total)
2626 goto out;
2627 }
2628
2629 /*
2630 * If mem_total did not overflow, multiply all memory values by
2631 * info->mem_unit and set it to 1. This leaves things compatible
2632 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2633 * kernels...
2634 */
2635
2636 info->mem_unit = 1;
2637 info->totalram <<= bitcount;
2638 info->freeram <<= bitcount;
2639 info->sharedram <<= bitcount;
2640 info->bufferram <<= bitcount;
2641 info->totalswap <<= bitcount;
2642 info->freeswap <<= bitcount;
2643 info->totalhigh <<= bitcount;
2644 info->freehigh <<= bitcount;
2645
2646 out:
2647 return 0;
2648 }
2649
2650 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2651 {
2652 struct sysinfo val;
2653
2654 do_sysinfo(&val);
2655
2656 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2657 return -EFAULT;
2658
2659 return 0;
2660 }
2661
2662 #ifdef CONFIG_COMPAT
2663 struct compat_sysinfo {
2664 s32 uptime;
2665 u32 loads[3];
2666 u32 totalram;
2667 u32 freeram;
2668 u32 sharedram;
2669 u32 bufferram;
2670 u32 totalswap;
2671 u32 freeswap;
2672 u16 procs;
2673 u16 pad;
2674 u32 totalhigh;
2675 u32 freehigh;
2676 u32 mem_unit;
2677 char _f[20-2*sizeof(u32)-sizeof(int)];
2678 };
2679
2680 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2681 {
2682 struct sysinfo s;
2683
2684 do_sysinfo(&s);
2685
2686 /* Check to see if any memory value is too large for 32-bit and scale
2687 * down if needed
2688 */
2689 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2690 int bitcount = 0;
2691
2692 while (s.mem_unit < PAGE_SIZE) {
2693 s.mem_unit <<= 1;
2694 bitcount++;
2695 }
2696
2697 s.totalram >>= bitcount;
2698 s.freeram >>= bitcount;
2699 s.sharedram >>= bitcount;
2700 s.bufferram >>= bitcount;
2701 s.totalswap >>= bitcount;
2702 s.freeswap >>= bitcount;
2703 s.totalhigh >>= bitcount;
2704 s.freehigh >>= bitcount;
2705 }
2706
2707 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2708 __put_user(s.uptime, &info->uptime) ||
2709 __put_user(s.loads[0], &info->loads[0]) ||
2710 __put_user(s.loads[1], &info->loads[1]) ||
2711 __put_user(s.loads[2], &info->loads[2]) ||
2712 __put_user(s.totalram, &info->totalram) ||
2713 __put_user(s.freeram, &info->freeram) ||
2714 __put_user(s.sharedram, &info->sharedram) ||
2715 __put_user(s.bufferram, &info->bufferram) ||
2716 __put_user(s.totalswap, &info->totalswap) ||
2717 __put_user(s.freeswap, &info->freeswap) ||
2718 __put_user(s.procs, &info->procs) ||
2719 __put_user(s.totalhigh, &info->totalhigh) ||
2720 __put_user(s.freehigh, &info->freehigh) ||
2721 __put_user(s.mem_unit, &info->mem_unit))
2722 return -EFAULT;
2723
2724 return 0;
2725 }
2726 #endif /* CONFIG_COMPAT */
Motorola kernel-slsi