]> git.ipfire.org Git - thirdparty/kernel/stable.git/blob - kernel/sys.c
projects / thirdparty / kernel / stable.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
sched/headers: Prepare for new header dependencies before moving code to <linux/sched...
[thirdparty/kernel/stable.git] / kernel / sys.c
1 /*
2 * linux/kernel/sys.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 #include <linux/export.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
14 #include <linux/fs.h>
15 #include <linux/kmod.h>
16 #include <linux/perf_event.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/workqueue.h>
20 #include <linux/capability.h>
21 #include <linux/device.h>
22 #include <linux/key.h>
23 #include <linux/times.h>
24 #include <linux/posix-timers.h>
25 #include <linux/security.h>
26 #include <linux/dcookies.h>
27 #include <linux/suspend.h>
28 #include <linux/tty.h>
29 #include <linux/signal.h>
30 #include <linux/cn_proc.h>
31 #include <linux/getcpu.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/seccomp.h>
34 #include <linux/cpu.h>
35 #include <linux/personality.h>
36 #include <linux/ptrace.h>
37 #include <linux/fs_struct.h>
38 #include <linux/file.h>
39 #include <linux/mount.h>
40 #include <linux/gfp.h>
41 #include <linux/syscore_ops.h>
42 #include <linux/version.h>
43 #include <linux/ctype.h>
44
45 #include <linux/compat.h>
46 #include <linux/syscalls.h>
47 #include <linux/kprobes.h>
48 #include <linux/user_namespace.h>
49 #include <linux/binfmts.h>
50
51 #include <linux/sched.h>
52 #include <linux/sched/autogroup.h>
53 #include <linux/sched/loadavg.h>
54 #include <linux/sched/mm.h>
55 #include <linux/rcupdate.h>
56 #include <linux/uidgid.h>
57 #include <linux/cred.h>
58
59 #include <linux/kmsg_dump.h>
60 /* Move somewhere else to avoid recompiling? */
61 #include <generated/utsrelease.h>
62
63 #include <linux/uaccess.h>
64 #include <asm/io.h>
65 #include <asm/unistd.h>
66
67 #ifndef SET_UNALIGN_CTL
68 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
69 #endif
70 #ifndef GET_UNALIGN_CTL
71 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
72 #endif
73 #ifndef SET_FPEMU_CTL
74 # define SET_FPEMU_CTL(a, b) (-EINVAL)
75 #endif
76 #ifndef GET_FPEMU_CTL
77 # define GET_FPEMU_CTL(a, b) (-EINVAL)
78 #endif
79 #ifndef SET_FPEXC_CTL
80 # define SET_FPEXC_CTL(a, b) (-EINVAL)
81 #endif
82 #ifndef GET_FPEXC_CTL
83 # define GET_FPEXC_CTL(a, b) (-EINVAL)
84 #endif
85 #ifndef GET_ENDIAN
86 # define GET_ENDIAN(a, b) (-EINVAL)
87 #endif
88 #ifndef SET_ENDIAN
89 # define SET_ENDIAN(a, b) (-EINVAL)
90 #endif
91 #ifndef GET_TSC_CTL
92 # define GET_TSC_CTL(a) (-EINVAL)
93 #endif
94 #ifndef SET_TSC_CTL
95 # define SET_TSC_CTL(a) (-EINVAL)
96 #endif
97 #ifndef MPX_ENABLE_MANAGEMENT
98 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
99 #endif
100 #ifndef MPX_DISABLE_MANAGEMENT
101 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
102 #endif
103 #ifndef GET_FP_MODE
104 # define GET_FP_MODE(a) (-EINVAL)
105 #endif
106 #ifndef SET_FP_MODE
107 # define SET_FP_MODE(a,b) (-EINVAL)
108 #endif
109
110 /*
111 * this is where the system-wide overflow UID and GID are defined, for
112 * architectures that now have 32-bit UID/GID but didn't in the past
113 */
114
115 int overflowuid = DEFAULT_OVERFLOWUID;
116 int overflowgid = DEFAULT_OVERFLOWGID;
117
118 EXPORT_SYMBOL(overflowuid);
119 EXPORT_SYMBOL(overflowgid);
120
121 /*
122 * the same as above, but for filesystems which can only store a 16-bit
123 * UID and GID. as such, this is needed on all architectures
124 */
125
126 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
127 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
128
129 EXPORT_SYMBOL(fs_overflowuid);
130 EXPORT_SYMBOL(fs_overflowgid);
131
132 /*
133 * Returns true if current's euid is same as p's uid or euid,
134 * or has CAP_SYS_NICE to p's user_ns.
135 *
136 * Called with rcu_read_lock, creds are safe
137 */
138 static bool set_one_prio_perm(struct task_struct *p)
139 {
140 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
141
142 if (uid_eq(pcred->uid, cred->euid) ||
143 uid_eq(pcred->euid, cred->euid))
144 return true;
145 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
146 return true;
147 return false;
148 }
149
150 /*
151 * set the priority of a task
152 * - the caller must hold the RCU read lock
153 */
154 static int set_one_prio(struct task_struct *p, int niceval, int error)
155 {
156 int no_nice;
157
158 if (!set_one_prio_perm(p)) {
159 error = -EPERM;
160 goto out;
161 }
162 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
163 error = -EACCES;
164 goto out;
165 }
166 no_nice = security_task_setnice(p, niceval);
167 if (no_nice) {
168 error = no_nice;
169 goto out;
170 }
171 if (error == -ESRCH)
172 error = 0;
173 set_user_nice(p, niceval);
174 out:
175 return error;
176 }
177
178 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
179 {
180 struct task_struct *g, *p;
181 struct user_struct *user;
182 const struct cred *cred = current_cred();
183 int error = -EINVAL;
184 struct pid *pgrp;
185 kuid_t uid;
186
187 if (which > PRIO_USER || which < PRIO_PROCESS)
188 goto out;
189
190 /* normalize: avoid signed division (rounding problems) */
191 error = -ESRCH;
192 if (niceval < MIN_NICE)
193 niceval = MIN_NICE;
194 if (niceval > MAX_NICE)
195 niceval = MAX_NICE;
196
197 rcu_read_lock();
198 read_lock(&tasklist_lock);
199 switch (which) {
200 case PRIO_PROCESS:
201 if (who)
202 p = find_task_by_vpid(who);
203 else
204 p = current;
205 if (p)
206 error = set_one_prio(p, niceval, error);
207 break;
208 case PRIO_PGRP:
209 if (who)
210 pgrp = find_vpid(who);
211 else
212 pgrp = task_pgrp(current);
213 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
214 error = set_one_prio(p, niceval, error);
215 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
216 break;
217 case PRIO_USER:
218 uid = make_kuid(cred->user_ns, who);
219 user = cred->user;
220 if (!who)
221 uid = cred->uid;
222 else if (!uid_eq(uid, cred->uid)) {
223 user = find_user(uid);
224 if (!user)
225 goto out_unlock; /* No processes for this user */
226 }
227 do_each_thread(g, p) {
228 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
229 error = set_one_prio(p, niceval, error);
230 } while_each_thread(g, p);
231 if (!uid_eq(uid, cred->uid))
232 free_uid(user); /* For find_user() */
233 break;
234 }
235 out_unlock:
236 read_unlock(&tasklist_lock);
237 rcu_read_unlock();
238 out:
239 return error;
240 }
241
242 /*
243 * Ugh. To avoid negative return values, "getpriority()" will
244 * not return the normal nice-value, but a negated value that
245 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
246 * to stay compatible.
247 */
248 SYSCALL_DEFINE2(getpriority, int, which, int, who)
249 {
250 struct task_struct *g, *p;
251 struct user_struct *user;
252 const struct cred *cred = current_cred();
253 long niceval, retval = -ESRCH;
254 struct pid *pgrp;
255 kuid_t uid;
256
257 if (which > PRIO_USER || which < PRIO_PROCESS)
258 return -EINVAL;
259
260 rcu_read_lock();
261 read_lock(&tasklist_lock);
262 switch (which) {
263 case PRIO_PROCESS:
264 if (who)
265 p = find_task_by_vpid(who);
266 else
267 p = current;
268 if (p) {
269 niceval = nice_to_rlimit(task_nice(p));
270 if (niceval > retval)
271 retval = niceval;
272 }
273 break;
274 case PRIO_PGRP:
275 if (who)
276 pgrp = find_vpid(who);
277 else
278 pgrp = task_pgrp(current);
279 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
280 niceval = nice_to_rlimit(task_nice(p));
281 if (niceval > retval)
282 retval = niceval;
283 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
284 break;
285 case PRIO_USER:
286 uid = make_kuid(cred->user_ns, who);
287 user = cred->user;
288 if (!who)
289 uid = cred->uid;
290 else if (!uid_eq(uid, cred->uid)) {
291 user = find_user(uid);
292 if (!user)
293 goto out_unlock; /* No processes for this user */
294 }
295 do_each_thread(g, p) {
296 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
297 niceval = nice_to_rlimit(task_nice(p));
298 if (niceval > retval)
299 retval = niceval;
300 }
301 } while_each_thread(g, p);
302 if (!uid_eq(uid, cred->uid))
303 free_uid(user); /* for find_user() */
304 break;
305 }
306 out_unlock:
307 read_unlock(&tasklist_lock);
308 rcu_read_unlock();
309
310 return retval;
311 }
312
313 /*
314 * Unprivileged users may change the real gid to the effective gid
315 * or vice versa. (BSD-style)
316 *
317 * If you set the real gid at all, or set the effective gid to a value not
318 * equal to the real gid, then the saved gid is set to the new effective gid.
319 *
320 * This makes it possible for a setgid program to completely drop its
321 * privileges, which is often a useful assertion to make when you are doing
322 * a security audit over a program.
323 *
324 * The general idea is that a program which uses just setregid() will be
325 * 100% compatible with BSD. A program which uses just setgid() will be
326 * 100% compatible with POSIX with saved IDs.
327 *
328 * SMP: There are not races, the GIDs are checked only by filesystem
329 * operations (as far as semantic preservation is concerned).
330 */
331 #ifdef CONFIG_MULTIUSER
332 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
333 {
334 struct user_namespace *ns = current_user_ns();
335 const struct cred *old;
336 struct cred *new;
337 int retval;
338 kgid_t krgid, kegid;
339
340 krgid = make_kgid(ns, rgid);
341 kegid = make_kgid(ns, egid);
342
343 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
344 return -EINVAL;
345 if ((egid != (gid_t) -1) && !gid_valid(kegid))
346 return -EINVAL;
347
348 new = prepare_creds();
349 if (!new)
350 return -ENOMEM;
351 old = current_cred();
352
353 retval = -EPERM;
354 if (rgid != (gid_t) -1) {
355 if (gid_eq(old->gid, krgid) ||
356 gid_eq(old->egid, krgid) ||
357 ns_capable(old->user_ns, CAP_SETGID))
358 new->gid = krgid;
359 else
360 goto error;
361 }
362 if (egid != (gid_t) -1) {
363 if (gid_eq(old->gid, kegid) ||
364 gid_eq(old->egid, kegid) ||
365 gid_eq(old->sgid, kegid) ||
366 ns_capable(old->user_ns, CAP_SETGID))
367 new->egid = kegid;
368 else
369 goto error;
370 }
371
372 if (rgid != (gid_t) -1 ||
373 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
374 new->sgid = new->egid;
375 new->fsgid = new->egid;
376
377 return commit_creds(new);
378
379 error:
380 abort_creds(new);
381 return retval;
382 }
383
384 /*
385 * setgid() is implemented like SysV w/ SAVED_IDS
386 *
387 * SMP: Same implicit races as above.
388 */
389 SYSCALL_DEFINE1(setgid, gid_t, gid)
390 {
391 struct user_namespace *ns = current_user_ns();
392 const struct cred *old;
393 struct cred *new;
394 int retval;
395 kgid_t kgid;
396
397 kgid = make_kgid(ns, gid);
398 if (!gid_valid(kgid))
399 return -EINVAL;
400
401 new = prepare_creds();
402 if (!new)
403 return -ENOMEM;
404 old = current_cred();
405
406 retval = -EPERM;
407 if (ns_capable(old->user_ns, CAP_SETGID))
408 new->gid = new->egid = new->sgid = new->fsgid = kgid;
409 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
410 new->egid = new->fsgid = kgid;
411 else
412 goto error;
413
414 return commit_creds(new);
415
416 error:
417 abort_creds(new);
418 return retval;
419 }
420
421 /*
422 * change the user struct in a credentials set to match the new UID
423 */
424 static int set_user(struct cred *new)
425 {
426 struct user_struct *new_user;
427
428 new_user = alloc_uid(new->uid);
429 if (!new_user)
430 return -EAGAIN;
431
432 /*
433 * We don't fail in case of NPROC limit excess here because too many
434 * poorly written programs don't check set*uid() return code, assuming
435 * it never fails if called by root. We may still enforce NPROC limit
436 * for programs doing set*uid()+execve() by harmlessly deferring the
437 * failure to the execve() stage.
438 */
439 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
440 new_user != INIT_USER)
441 current->flags |= PF_NPROC_EXCEEDED;
442 else
443 current->flags &= ~PF_NPROC_EXCEEDED;
444
445 free_uid(new->user);
446 new->user = new_user;
447 return 0;
448 }
449
450 /*
451 * Unprivileged users may change the real uid to the effective uid
452 * or vice versa. (BSD-style)
453 *
454 * If you set the real uid at all, or set the effective uid to a value not
455 * equal to the real uid, then the saved uid is set to the new effective uid.
456 *
457 * This makes it possible for a setuid program to completely drop its
458 * privileges, which is often a useful assertion to make when you are doing
459 * a security audit over a program.
460 *
461 * The general idea is that a program which uses just setreuid() will be
462 * 100% compatible with BSD. A program which uses just setuid() will be
463 * 100% compatible with POSIX with saved IDs.
464 */
465 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
466 {
467 struct user_namespace *ns = current_user_ns();
468 const struct cred *old;
469 struct cred *new;
470 int retval;
471 kuid_t kruid, keuid;
472
473 kruid = make_kuid(ns, ruid);
474 keuid = make_kuid(ns, euid);
475
476 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
477 return -EINVAL;
478 if ((euid != (uid_t) -1) && !uid_valid(keuid))
479 return -EINVAL;
480
481 new = prepare_creds();
482 if (!new)
483 return -ENOMEM;
484 old = current_cred();
485
486 retval = -EPERM;
487 if (ruid != (uid_t) -1) {
488 new->uid = kruid;
489 if (!uid_eq(old->uid, kruid) &&
490 !uid_eq(old->euid, kruid) &&
491 !ns_capable(old->user_ns, CAP_SETUID))
492 goto error;
493 }
494
495 if (euid != (uid_t) -1) {
496 new->euid = keuid;
497 if (!uid_eq(old->uid, keuid) &&
498 !uid_eq(old->euid, keuid) &&
499 !uid_eq(old->suid, keuid) &&
500 !ns_capable(old->user_ns, CAP_SETUID))
501 goto error;
502 }
503
504 if (!uid_eq(new->uid, old->uid)) {
505 retval = set_user(new);
506 if (retval < 0)
507 goto error;
508 }
509 if (ruid != (uid_t) -1 ||
510 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
511 new->suid = new->euid;
512 new->fsuid = new->euid;
513
514 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
515 if (retval < 0)
516 goto error;
517
518 return commit_creds(new);
519
520 error:
521 abort_creds(new);
522 return retval;
523 }
524
525 /*
526 * setuid() is implemented like SysV with SAVED_IDS
527 *
528 * Note that SAVED_ID's is deficient in that a setuid root program
529 * like sendmail, for example, cannot set its uid to be a normal
530 * user and then switch back, because if you're root, setuid() sets
531 * the saved uid too. If you don't like this, blame the bright people
532 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
533 * will allow a root program to temporarily drop privileges and be able to
534 * regain them by swapping the real and effective uid.
535 */
536 SYSCALL_DEFINE1(setuid, uid_t, uid)
537 {
538 struct user_namespace *ns = current_user_ns();
539 const struct cred *old;
540 struct cred *new;
541 int retval;
542 kuid_t kuid;
543
544 kuid = make_kuid(ns, uid);
545 if (!uid_valid(kuid))
546 return -EINVAL;
547
548 new = prepare_creds();
549 if (!new)
550 return -ENOMEM;
551 old = current_cred();
552
553 retval = -EPERM;
554 if (ns_capable(old->user_ns, CAP_SETUID)) {
555 new->suid = new->uid = kuid;
556 if (!uid_eq(kuid, old->uid)) {
557 retval = set_user(new);
558 if (retval < 0)
559 goto error;
560 }
561 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
562 goto error;
563 }
564
565 new->fsuid = new->euid = kuid;
566
567 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
568 if (retval < 0)
569 goto error;
570
571 return commit_creds(new);
572
573 error:
574 abort_creds(new);
575 return retval;
576 }
577
578
579 /*
580 * This function implements a generic ability to update ruid, euid,
581 * and suid. This allows you to implement the 4.4 compatible seteuid().
582 */
583 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
584 {
585 struct user_namespace *ns = current_user_ns();
586 const struct cred *old;
587 struct cred *new;
588 int retval;
589 kuid_t kruid, keuid, ksuid;
590
591 kruid = make_kuid(ns, ruid);
592 keuid = make_kuid(ns, euid);
593 ksuid = make_kuid(ns, suid);
594
595 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
596 return -EINVAL;
597
598 if ((euid != (uid_t) -1) && !uid_valid(keuid))
599 return -EINVAL;
600
601 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
602 return -EINVAL;
603
604 new = prepare_creds();
605 if (!new)
606 return -ENOMEM;
607
608 old = current_cred();
609
610 retval = -EPERM;
611 if (!ns_capable(old->user_ns, CAP_SETUID)) {
612 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
613 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
614 goto error;
615 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
616 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
617 goto error;
618 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
619 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
620 goto error;
621 }
622
623 if (ruid != (uid_t) -1) {
624 new->uid = kruid;
625 if (!uid_eq(kruid, old->uid)) {
626 retval = set_user(new);
627 if (retval < 0)
628 goto error;
629 }
630 }
631 if (euid != (uid_t) -1)
632 new->euid = keuid;
633 if (suid != (uid_t) -1)
634 new->suid = ksuid;
635 new->fsuid = new->euid;
636
637 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
638 if (retval < 0)
639 goto error;
640
641 return commit_creds(new);
642
643 error:
644 abort_creds(new);
645 return retval;
646 }
647
648 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
649 {
650 const struct cred *cred = current_cred();
651 int retval;
652 uid_t ruid, euid, suid;
653
654 ruid = from_kuid_munged(cred->user_ns, cred->uid);
655 euid = from_kuid_munged(cred->user_ns, cred->euid);
656 suid = from_kuid_munged(cred->user_ns, cred->suid);
657
658 retval = put_user(ruid, ruidp);
659 if (!retval) {
660 retval = put_user(euid, euidp);
661 if (!retval)
662 return put_user(suid, suidp);
663 }
664 return retval;
665 }
666
667 /*
668 * Same as above, but for rgid, egid, sgid.
669 */
670 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
671 {
672 struct user_namespace *ns = current_user_ns();
673 const struct cred *old;
674 struct cred *new;
675 int retval;
676 kgid_t krgid, kegid, ksgid;
677
678 krgid = make_kgid(ns, rgid);
679 kegid = make_kgid(ns, egid);
680 ksgid = make_kgid(ns, sgid);
681
682 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
683 return -EINVAL;
684 if ((egid != (gid_t) -1) && !gid_valid(kegid))
685 return -EINVAL;
686 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
687 return -EINVAL;
688
689 new = prepare_creds();
690 if (!new)
691 return -ENOMEM;
692 old = current_cred();
693
694 retval = -EPERM;
695 if (!ns_capable(old->user_ns, CAP_SETGID)) {
696 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
697 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
698 goto error;
699 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
700 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
701 goto error;
702 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
703 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
704 goto error;
705 }
706
707 if (rgid != (gid_t) -1)
708 new->gid = krgid;
709 if (egid != (gid_t) -1)
710 new->egid = kegid;
711 if (sgid != (gid_t) -1)
712 new->sgid = ksgid;
713 new->fsgid = new->egid;
714
715 return commit_creds(new);
716
717 error:
718 abort_creds(new);
719 return retval;
720 }
721
722 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
723 {
724 const struct cred *cred = current_cred();
725 int retval;
726 gid_t rgid, egid, sgid;
727
728 rgid = from_kgid_munged(cred->user_ns, cred->gid);
729 egid = from_kgid_munged(cred->user_ns, cred->egid);
730 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
731
732 retval = put_user(rgid, rgidp);
733 if (!retval) {
734 retval = put_user(egid, egidp);
735 if (!retval)
736 retval = put_user(sgid, sgidp);
737 }
738
739 return retval;
740 }
741
742
743 /*
744 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
745 * is used for "access()" and for the NFS daemon (letting nfsd stay at
746 * whatever uid it wants to). It normally shadows "euid", except when
747 * explicitly set by setfsuid() or for access..
748 */
749 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
750 {
751 const struct cred *old;
752 struct cred *new;
753 uid_t old_fsuid;
754 kuid_t kuid;
755
756 old = current_cred();
757 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
758
759 kuid = make_kuid(old->user_ns, uid);
760 if (!uid_valid(kuid))
761 return old_fsuid;
762
763 new = prepare_creds();
764 if (!new)
765 return old_fsuid;
766
767 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
768 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
769 ns_capable(old->user_ns, CAP_SETUID)) {
770 if (!uid_eq(kuid, old->fsuid)) {
771 new->fsuid = kuid;
772 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
773 goto change_okay;
774 }
775 }
776
777 abort_creds(new);
778 return old_fsuid;
779
780 change_okay:
781 commit_creds(new);
782 return old_fsuid;
783 }
784
785 /*
786 * Samma på svenska..
787 */
788 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
789 {
790 const struct cred *old;
791 struct cred *new;
792 gid_t old_fsgid;
793 kgid_t kgid;
794
795 old = current_cred();
796 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
797
798 kgid = make_kgid(old->user_ns, gid);
799 if (!gid_valid(kgid))
800 return old_fsgid;
801
802 new = prepare_creds();
803 if (!new)
804 return old_fsgid;
805
806 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
807 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
808 ns_capable(old->user_ns, CAP_SETGID)) {
809 if (!gid_eq(kgid, old->fsgid)) {
810 new->fsgid = kgid;
811 goto change_okay;
812 }
813 }
814
815 abort_creds(new);
816 return old_fsgid;
817
818 change_okay:
819 commit_creds(new);
820 return old_fsgid;
821 }
822 #endif /* CONFIG_MULTIUSER */
823
824 /**
825 * sys_getpid - return the thread group id of the current process
826 *
827 * Note, despite the name, this returns the tgid not the pid. The tgid and
828 * the pid are identical unless CLONE_THREAD was specified on clone() in
829 * which case the tgid is the same in all threads of the same group.
830 *
831 * This is SMP safe as current->tgid does not change.
832 */
833 SYSCALL_DEFINE0(getpid)
834 {
835 return task_tgid_vnr(current);
836 }
837
838 /* Thread ID - the internal kernel "pid" */
839 SYSCALL_DEFINE0(gettid)
840 {
841 return task_pid_vnr(current);
842 }
843
844 /*
845 * Accessing ->real_parent is not SMP-safe, it could
846 * change from under us. However, we can use a stale
847 * value of ->real_parent under rcu_read_lock(), see
848 * release_task()->call_rcu(delayed_put_task_struct).
849 */
850 SYSCALL_DEFINE0(getppid)
851 {
852 int pid;
853
854 rcu_read_lock();
855 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
856 rcu_read_unlock();
857
858 return pid;
859 }
860
861 SYSCALL_DEFINE0(getuid)
862 {
863 /* Only we change this so SMP safe */
864 return from_kuid_munged(current_user_ns(), current_uid());
865 }
866
867 SYSCALL_DEFINE0(geteuid)
868 {
869 /* Only we change this so SMP safe */
870 return from_kuid_munged(current_user_ns(), current_euid());
871 }
872
873 SYSCALL_DEFINE0(getgid)
874 {
875 /* Only we change this so SMP safe */
876 return from_kgid_munged(current_user_ns(), current_gid());
877 }
878
879 SYSCALL_DEFINE0(getegid)
880 {
881 /* Only we change this so SMP safe */
882 return from_kgid_munged(current_user_ns(), current_egid());
883 }
884
885 void do_sys_times(struct tms *tms)
886 {
887 u64 tgutime, tgstime, cutime, cstime;
888
889 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
890 cutime = current->signal->cutime;
891 cstime = current->signal->cstime;
892 tms->tms_utime = nsec_to_clock_t(tgutime);
893 tms->tms_stime = nsec_to_clock_t(tgstime);
894 tms->tms_cutime = nsec_to_clock_t(cutime);
895 tms->tms_cstime = nsec_to_clock_t(cstime);
896 }
897
898 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
899 {
900 if (tbuf) {
901 struct tms tmp;
902
903 do_sys_times(&tmp);
904 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
905 return -EFAULT;
906 }
907 force_successful_syscall_return();
908 return (long) jiffies_64_to_clock_t(get_jiffies_64());
909 }
910
911 /*
912 * This needs some heavy checking ...
913 * I just haven't the stomach for it. I also don't fully
914 * understand sessions/pgrp etc. Let somebody who does explain it.
915 *
916 * OK, I think I have the protection semantics right.... this is really
917 * only important on a multi-user system anyway, to make sure one user
918 * can't send a signal to a process owned by another. -TYT, 12/12/91
919 *
920 * !PF_FORKNOEXEC check to conform completely to POSIX.
921 */
922 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
923 {
924 struct task_struct *p;
925 struct task_struct *group_leader = current->group_leader;
926 struct pid *pgrp;
927 int err;
928
929 if (!pid)
930 pid = task_pid_vnr(group_leader);
931 if (!pgid)
932 pgid = pid;
933 if (pgid < 0)
934 return -EINVAL;
935 rcu_read_lock();
936
937 /* From this point forward we keep holding onto the tasklist lock
938 * so that our parent does not change from under us. -DaveM
939 */
940 write_lock_irq(&tasklist_lock);
941
942 err = -ESRCH;
943 p = find_task_by_vpid(pid);
944 if (!p)
945 goto out;
946
947 err = -EINVAL;
948 if (!thread_group_leader(p))
949 goto out;
950
951 if (same_thread_group(p->real_parent, group_leader)) {
952 err = -EPERM;
953 if (task_session(p) != task_session(group_leader))
954 goto out;
955 err = -EACCES;
956 if (!(p->flags & PF_FORKNOEXEC))
957 goto out;
958 } else {
959 err = -ESRCH;
960 if (p != group_leader)
961 goto out;
962 }
963
964 err = -EPERM;
965 if (p->signal->leader)
966 goto out;
967
968 pgrp = task_pid(p);
969 if (pgid != pid) {
970 struct task_struct *g;
971
972 pgrp = find_vpid(pgid);
973 g = pid_task(pgrp, PIDTYPE_PGID);
974 if (!g || task_session(g) != task_session(group_leader))
975 goto out;
976 }
977
978 err = security_task_setpgid(p, pgid);
979 if (err)
980 goto out;
981
982 if (task_pgrp(p) != pgrp)
983 change_pid(p, PIDTYPE_PGID, pgrp);
984
985 err = 0;
986 out:
987 /* All paths lead to here, thus we are safe. -DaveM */
988 write_unlock_irq(&tasklist_lock);
989 rcu_read_unlock();
990 return err;
991 }
992
993 SYSCALL_DEFINE1(getpgid, pid_t, pid)
994 {
995 struct task_struct *p;
996 struct pid *grp;
997 int retval;
998
999 rcu_read_lock();
1000 if (!pid)
1001 grp = task_pgrp(current);
1002 else {
1003 retval = -ESRCH;
1004 p = find_task_by_vpid(pid);
1005 if (!p)
1006 goto out;
1007 grp = task_pgrp(p);
1008 if (!grp)
1009 goto out;
1010
1011 retval = security_task_getpgid(p);
1012 if (retval)
1013 goto out;
1014 }
1015 retval = pid_vnr(grp);
1016 out:
1017 rcu_read_unlock();
1018 return retval;
1019 }
1020
1021 #ifdef __ARCH_WANT_SYS_GETPGRP
1022
1023 SYSCALL_DEFINE0(getpgrp)
1024 {
1025 return sys_getpgid(0);
1026 }
1027
1028 #endif
1029
1030 SYSCALL_DEFINE1(getsid, pid_t, pid)
1031 {
1032 struct task_struct *p;
1033 struct pid *sid;
1034 int retval;
1035
1036 rcu_read_lock();
1037 if (!pid)
1038 sid = task_session(current);
1039 else {
1040 retval = -ESRCH;
1041 p = find_task_by_vpid(pid);
1042 if (!p)
1043 goto out;
1044 sid = task_session(p);
1045 if (!sid)
1046 goto out;
1047
1048 retval = security_task_getsid(p);
1049 if (retval)
1050 goto out;
1051 }
1052 retval = pid_vnr(sid);
1053 out:
1054 rcu_read_unlock();
1055 return retval;
1056 }
1057
1058 static void set_special_pids(struct pid *pid)
1059 {
1060 struct task_struct *curr = current->group_leader;
1061
1062 if (task_session(curr) != pid)
1063 change_pid(curr, PIDTYPE_SID, pid);
1064
1065 if (task_pgrp(curr) != pid)
1066 change_pid(curr, PIDTYPE_PGID, pid);
1067 }
1068
1069 SYSCALL_DEFINE0(setsid)
1070 {
1071 struct task_struct *group_leader = current->group_leader;
1072 struct pid *sid = task_pid(group_leader);
1073 pid_t session = pid_vnr(sid);
1074 int err = -EPERM;
1075
1076 write_lock_irq(&tasklist_lock);
1077 /* Fail if I am already a session leader */
1078 if (group_leader->signal->leader)
1079 goto out;
1080
1081 /* Fail if a process group id already exists that equals the
1082 * proposed session id.
1083 */
1084 if (pid_task(sid, PIDTYPE_PGID))
1085 goto out;
1086
1087 group_leader->signal->leader = 1;
1088 set_special_pids(sid);
1089
1090 proc_clear_tty(group_leader);
1091
1092 err = session;
1093 out:
1094 write_unlock_irq(&tasklist_lock);
1095 if (err > 0) {
1096 proc_sid_connector(group_leader);
1097 sched_autogroup_create_attach(group_leader);
1098 }
1099 return err;
1100 }
1101
1102 DECLARE_RWSEM(uts_sem);
1103
1104 #ifdef COMPAT_UTS_MACHINE
1105 #define override_architecture(name) \
1106 (personality(current->personality) == PER_LINUX32 && \
1107 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1108 sizeof(COMPAT_UTS_MACHINE)))
1109 #else
1110 #define override_architecture(name) 0
1111 #endif
1112
1113 /*
1114 * Work around broken programs that cannot handle "Linux 3.0".
1115 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1116 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1117 */
1118 static int override_release(char __user *release, size_t len)
1119 {
1120 int ret = 0;
1121
1122 if (current->personality & UNAME26) {
1123 const char *rest = UTS_RELEASE;
1124 char buf[65] = { 0 };
1125 int ndots = 0;
1126 unsigned v;
1127 size_t copy;
1128
1129 while (*rest) {
1130 if (*rest == '.' && ++ndots >= 3)
1131 break;
1132 if (!isdigit(*rest) && *rest != '.')
1133 break;
1134 rest++;
1135 }
1136 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1137 copy = clamp_t(size_t, len, 1, sizeof(buf));
1138 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1139 ret = copy_to_user(release, buf, copy + 1);
1140 }
1141 return ret;
1142 }
1143
1144 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1145 {
1146 int errno = 0;
1147
1148 down_read(&uts_sem);
1149 if (copy_to_user(name, utsname(), sizeof *name))
1150 errno = -EFAULT;
1151 up_read(&uts_sem);
1152
1153 if (!errno && override_release(name->release, sizeof(name->release)))
1154 errno = -EFAULT;
1155 if (!errno && override_architecture(name))
1156 errno = -EFAULT;
1157 return errno;
1158 }
1159
1160 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1161 /*
1162 * Old cruft
1163 */
1164 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1165 {
1166 int error = 0;
1167
1168 if (!name)
1169 return -EFAULT;
1170
1171 down_read(&uts_sem);
1172 if (copy_to_user(name, utsname(), sizeof(*name)))
1173 error = -EFAULT;
1174 up_read(&uts_sem);
1175
1176 if (!error && override_release(name->release, sizeof(name->release)))
1177 error = -EFAULT;
1178 if (!error && override_architecture(name))
1179 error = -EFAULT;
1180 return error;
1181 }
1182
1183 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1184 {
1185 int error;
1186
1187 if (!name)
1188 return -EFAULT;
1189 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1190 return -EFAULT;
1191
1192 down_read(&uts_sem);
1193 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1194 __OLD_UTS_LEN);
1195 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1196 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1197 __OLD_UTS_LEN);
1198 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1199 error |= __copy_to_user(&name->release, &utsname()->release,
1200 __OLD_UTS_LEN);
1201 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1202 error |= __copy_to_user(&name->version, &utsname()->version,
1203 __OLD_UTS_LEN);
1204 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1205 error |= __copy_to_user(&name->machine, &utsname()->machine,
1206 __OLD_UTS_LEN);
1207 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1208 up_read(&uts_sem);
1209
1210 if (!error && override_architecture(name))
1211 error = -EFAULT;
1212 if (!error && override_release(name->release, sizeof(name->release)))
1213 error = -EFAULT;
1214 return error ? -EFAULT : 0;
1215 }
1216 #endif
1217
1218 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1219 {
1220 int errno;
1221 char tmp[__NEW_UTS_LEN];
1222
1223 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1224 return -EPERM;
1225
1226 if (len < 0 || len > __NEW_UTS_LEN)
1227 return -EINVAL;
1228 down_write(&uts_sem);
1229 errno = -EFAULT;
1230 if (!copy_from_user(tmp, name, len)) {
1231 struct new_utsname *u = utsname();
1232
1233 memcpy(u->nodename, tmp, len);
1234 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1235 errno = 0;
1236 uts_proc_notify(UTS_PROC_HOSTNAME);
1237 }
1238 up_write(&uts_sem);
1239 return errno;
1240 }
1241
1242 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1243
1244 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1245 {
1246 int i, errno;
1247 struct new_utsname *u;
1248
1249 if (len < 0)
1250 return -EINVAL;
1251 down_read(&uts_sem);
1252 u = utsname();
1253 i = 1 + strlen(u->nodename);
1254 if (i > len)
1255 i = len;
1256 errno = 0;
1257 if (copy_to_user(name, u->nodename, i))
1258 errno = -EFAULT;
1259 up_read(&uts_sem);
1260 return errno;
1261 }
1262
1263 #endif
1264
1265 /*
1266 * Only setdomainname; getdomainname can be implemented by calling
1267 * uname()
1268 */
1269 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1270 {
1271 int errno;
1272 char tmp[__NEW_UTS_LEN];
1273
1274 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1275 return -EPERM;
1276 if (len < 0 || len > __NEW_UTS_LEN)
1277 return -EINVAL;
1278
1279 down_write(&uts_sem);
1280 errno = -EFAULT;
1281 if (!copy_from_user(tmp, name, len)) {
1282 struct new_utsname *u = utsname();
1283
1284 memcpy(u->domainname, tmp, len);
1285 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1286 errno = 0;
1287 uts_proc_notify(UTS_PROC_DOMAINNAME);
1288 }
1289 up_write(&uts_sem);
1290 return errno;
1291 }
1292
1293 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1294 {
1295 struct rlimit value;
1296 int ret;
1297
1298 ret = do_prlimit(current, resource, NULL, &value);
1299 if (!ret)
1300 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1301
1302 return ret;
1303 }
1304
1305 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1306
1307 /*
1308 * Back compatibility for getrlimit. Needed for some apps.
1309 */
1310 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1311 struct rlimit __user *, rlim)
1312 {
1313 struct rlimit x;
1314 if (resource >= RLIM_NLIMITS)
1315 return -EINVAL;
1316
1317 task_lock(current->group_leader);
1318 x = current->signal->rlim[resource];
1319 task_unlock(current->group_leader);
1320 if (x.rlim_cur > 0x7FFFFFFF)
1321 x.rlim_cur = 0x7FFFFFFF;
1322 if (x.rlim_max > 0x7FFFFFFF)
1323 x.rlim_max = 0x7FFFFFFF;
1324 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1325 }
1326
1327 #endif
1328
1329 static inline bool rlim64_is_infinity(__u64 rlim64)
1330 {
1331 #if BITS_PER_LONG < 64
1332 return rlim64 >= ULONG_MAX;
1333 #else
1334 return rlim64 == RLIM64_INFINITY;
1335 #endif
1336 }
1337
1338 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1339 {
1340 if (rlim->rlim_cur == RLIM_INFINITY)
1341 rlim64->rlim_cur = RLIM64_INFINITY;
1342 else
1343 rlim64->rlim_cur = rlim->rlim_cur;
1344 if (rlim->rlim_max == RLIM_INFINITY)
1345 rlim64->rlim_max = RLIM64_INFINITY;
1346 else
1347 rlim64->rlim_max = rlim->rlim_max;
1348 }
1349
1350 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1351 {
1352 if (rlim64_is_infinity(rlim64->rlim_cur))
1353 rlim->rlim_cur = RLIM_INFINITY;
1354 else
1355 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1356 if (rlim64_is_infinity(rlim64->rlim_max))
1357 rlim->rlim_max = RLIM_INFINITY;
1358 else
1359 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1360 }
1361
1362 /* make sure you are allowed to change @tsk limits before calling this */
1363 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1364 struct rlimit *new_rlim, struct rlimit *old_rlim)
1365 {
1366 struct rlimit *rlim;
1367 int retval = 0;
1368
1369 if (resource >= RLIM_NLIMITS)
1370 return -EINVAL;
1371 if (new_rlim) {
1372 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1373 return -EINVAL;
1374 if (resource == RLIMIT_NOFILE &&
1375 new_rlim->rlim_max > sysctl_nr_open)
1376 return -EPERM;
1377 }
1378
1379 /* protect tsk->signal and tsk->sighand from disappearing */
1380 read_lock(&tasklist_lock);
1381 if (!tsk->sighand) {
1382 retval = -ESRCH;
1383 goto out;
1384 }
1385
1386 rlim = tsk->signal->rlim + resource;
1387 task_lock(tsk->group_leader);
1388 if (new_rlim) {
1389 /* Keep the capable check against init_user_ns until
1390 cgroups can contain all limits */
1391 if (new_rlim->rlim_max > rlim->rlim_max &&
1392 !capable(CAP_SYS_RESOURCE))
1393 retval = -EPERM;
1394 if (!retval)
1395 retval = security_task_setrlimit(tsk->group_leader,
1396 resource, new_rlim);
1397 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1398 /*
1399 * The caller is asking for an immediate RLIMIT_CPU
1400 * expiry. But we use the zero value to mean "it was
1401 * never set". So let's cheat and make it one second
1402 * instead
1403 */
1404 new_rlim->rlim_cur = 1;
1405 }
1406 }
1407 if (!retval) {
1408 if (old_rlim)
1409 *old_rlim = *rlim;
1410 if (new_rlim)
1411 *rlim = *new_rlim;
1412 }
1413 task_unlock(tsk->group_leader);
1414
1415 /*
1416 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1417 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1418 * very long-standing error, and fixing it now risks breakage of
1419 * applications, so we live with it
1420 */
1421 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1422 new_rlim->rlim_cur != RLIM_INFINITY &&
1423 IS_ENABLED(CONFIG_POSIX_TIMERS))
1424 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1425 out:
1426 read_unlock(&tasklist_lock);
1427 return retval;
1428 }
1429
1430 /* rcu lock must be held */
1431 static int check_prlimit_permission(struct task_struct *task)
1432 {
1433 const struct cred *cred = current_cred(), *tcred;
1434
1435 if (current == task)
1436 return 0;
1437
1438 tcred = __task_cred(task);
1439 if (uid_eq(cred->uid, tcred->euid) &&
1440 uid_eq(cred->uid, tcred->suid) &&
1441 uid_eq(cred->uid, tcred->uid) &&
1442 gid_eq(cred->gid, tcred->egid) &&
1443 gid_eq(cred->gid, tcred->sgid) &&
1444 gid_eq(cred->gid, tcred->gid))
1445 return 0;
1446 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1447 return 0;
1448
1449 return -EPERM;
1450 }
1451
1452 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1453 const struct rlimit64 __user *, new_rlim,
1454 struct rlimit64 __user *, old_rlim)
1455 {
1456 struct rlimit64 old64, new64;
1457 struct rlimit old, new;
1458 struct task_struct *tsk;
1459 int ret;
1460
1461 if (new_rlim) {
1462 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1463 return -EFAULT;
1464 rlim64_to_rlim(&new64, &new);
1465 }
1466
1467 rcu_read_lock();
1468 tsk = pid ? find_task_by_vpid(pid) : current;
1469 if (!tsk) {
1470 rcu_read_unlock();
1471 return -ESRCH;
1472 }
1473 ret = check_prlimit_permission(tsk);
1474 if (ret) {
1475 rcu_read_unlock();
1476 return ret;
1477 }
1478 get_task_struct(tsk);
1479 rcu_read_unlock();
1480
1481 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1482 old_rlim ? &old : NULL);
1483
1484 if (!ret && old_rlim) {
1485 rlim_to_rlim64(&old, &old64);
1486 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1487 ret = -EFAULT;
1488 }
1489
1490 put_task_struct(tsk);
1491 return ret;
1492 }
1493
1494 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1495 {
1496 struct rlimit new_rlim;
1497
1498 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1499 return -EFAULT;
1500 return do_prlimit(current, resource, &new_rlim, NULL);
1501 }
1502
1503 /*
1504 * It would make sense to put struct rusage in the task_struct,
1505 * except that would make the task_struct be *really big*. After
1506 * task_struct gets moved into malloc'ed memory, it would
1507 * make sense to do this. It will make moving the rest of the information
1508 * a lot simpler! (Which we're not doing right now because we're not
1509 * measuring them yet).
1510 *
1511 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1512 * races with threads incrementing their own counters. But since word
1513 * reads are atomic, we either get new values or old values and we don't
1514 * care which for the sums. We always take the siglock to protect reading
1515 * the c* fields from p->signal from races with exit.c updating those
1516 * fields when reaping, so a sample either gets all the additions of a
1517 * given child after it's reaped, or none so this sample is before reaping.
1518 *
1519 * Locking:
1520 * We need to take the siglock for CHILDEREN, SELF and BOTH
1521 * for the cases current multithreaded, non-current single threaded
1522 * non-current multithreaded. Thread traversal is now safe with
1523 * the siglock held.
1524 * Strictly speaking, we donot need to take the siglock if we are current and
1525 * single threaded, as no one else can take our signal_struct away, no one
1526 * else can reap the children to update signal->c* counters, and no one else
1527 * can race with the signal-> fields. If we do not take any lock, the
1528 * signal-> fields could be read out of order while another thread was just
1529 * exiting. So we should place a read memory barrier when we avoid the lock.
1530 * On the writer side, write memory barrier is implied in __exit_signal
1531 * as __exit_signal releases the siglock spinlock after updating the signal->
1532 * fields. But we don't do this yet to keep things simple.
1533 *
1534 */
1535
1536 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1537 {
1538 r->ru_nvcsw += t->nvcsw;
1539 r->ru_nivcsw += t->nivcsw;
1540 r->ru_minflt += t->min_flt;
1541 r->ru_majflt += t->maj_flt;
1542 r->ru_inblock += task_io_get_inblock(t);
1543 r->ru_oublock += task_io_get_oublock(t);
1544 }
1545
1546 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1547 {
1548 struct task_struct *t;
1549 unsigned long flags;
1550 u64 tgutime, tgstime, utime, stime;
1551 unsigned long maxrss = 0;
1552
1553 memset((char *)r, 0, sizeof (*r));
1554 utime = stime = 0;
1555
1556 if (who == RUSAGE_THREAD) {
1557 task_cputime_adjusted(current, &utime, &stime);
1558 accumulate_thread_rusage(p, r);
1559 maxrss = p->signal->maxrss;
1560 goto out;
1561 }
1562
1563 if (!lock_task_sighand(p, &flags))
1564 return;
1565
1566 switch (who) {
1567 case RUSAGE_BOTH:
1568 case RUSAGE_CHILDREN:
1569 utime = p->signal->cutime;
1570 stime = p->signal->cstime;
1571 r->ru_nvcsw = p->signal->cnvcsw;
1572 r->ru_nivcsw = p->signal->cnivcsw;
1573 r->ru_minflt = p->signal->cmin_flt;
1574 r->ru_majflt = p->signal->cmaj_flt;
1575 r->ru_inblock = p->signal->cinblock;
1576 r->ru_oublock = p->signal->coublock;
1577 maxrss = p->signal->cmaxrss;
1578
1579 if (who == RUSAGE_CHILDREN)
1580 break;
1581
1582 case RUSAGE_SELF:
1583 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1584 utime += tgutime;
1585 stime += tgstime;
1586 r->ru_nvcsw += p->signal->nvcsw;
1587 r->ru_nivcsw += p->signal->nivcsw;
1588 r->ru_minflt += p->signal->min_flt;
1589 r->ru_majflt += p->signal->maj_flt;
1590 r->ru_inblock += p->signal->inblock;
1591 r->ru_oublock += p->signal->oublock;
1592 if (maxrss < p->signal->maxrss)
1593 maxrss = p->signal->maxrss;
1594 t = p;
1595 do {
1596 accumulate_thread_rusage(t, r);
1597 } while_each_thread(p, t);
1598 break;
1599
1600 default:
1601 BUG();
1602 }
1603 unlock_task_sighand(p, &flags);
1604
1605 out:
1606 r->ru_utime = ns_to_timeval(utime);
1607 r->ru_stime = ns_to_timeval(stime);
1608
1609 if (who != RUSAGE_CHILDREN) {
1610 struct mm_struct *mm = get_task_mm(p);
1611
1612 if (mm) {
1613 setmax_mm_hiwater_rss(&maxrss, mm);
1614 mmput(mm);
1615 }
1616 }
1617 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1618 }
1619
1620 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1621 {
1622 struct rusage r;
1623
1624 k_getrusage(p, who, &r);
1625 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1626 }
1627
1628 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1629 {
1630 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1631 who != RUSAGE_THREAD)
1632 return -EINVAL;
1633 return getrusage(current, who, ru);
1634 }
1635
1636 #ifdef CONFIG_COMPAT
1637 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1638 {
1639 struct rusage r;
1640
1641 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1642 who != RUSAGE_THREAD)
1643 return -EINVAL;
1644
1645 k_getrusage(current, who, &r);
1646 return put_compat_rusage(&r, ru);
1647 }
1648 #endif
1649
1650 SYSCALL_DEFINE1(umask, int, mask)
1651 {
1652 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1653 return mask;
1654 }
1655
1656 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1657 {
1658 struct fd exe;
1659 struct file *old_exe, *exe_file;
1660 struct inode *inode;
1661 int err;
1662
1663 exe = fdget(fd);
1664 if (!exe.file)
1665 return -EBADF;
1666
1667 inode = file_inode(exe.file);
1668
1669 /*
1670 * Because the original mm->exe_file points to executable file, make
1671 * sure that this one is executable as well, to avoid breaking an
1672 * overall picture.
1673 */
1674 err = -EACCES;
1675 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1676 goto exit;
1677
1678 err = inode_permission(inode, MAY_EXEC);
1679 if (err)
1680 goto exit;
1681
1682 /*
1683 * Forbid mm->exe_file change if old file still mapped.
1684 */
1685 exe_file = get_mm_exe_file(mm);
1686 err = -EBUSY;
1687 if (exe_file) {
1688 struct vm_area_struct *vma;
1689
1690 down_read(&mm->mmap_sem);
1691 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1692 if (!vma->vm_file)
1693 continue;
1694 if (path_equal(&vma->vm_file->f_path,
1695 &exe_file->f_path))
1696 goto exit_err;
1697 }
1698
1699 up_read(&mm->mmap_sem);
1700 fput(exe_file);
1701 }
1702
1703 err = 0;
1704 /* set the new file, lockless */
1705 get_file(exe.file);
1706 old_exe = xchg(&mm->exe_file, exe.file);
1707 if (old_exe)
1708 fput(old_exe);
1709 exit:
1710 fdput(exe);
1711 return err;
1712 exit_err:
1713 up_read(&mm->mmap_sem);
1714 fput(exe_file);
1715 goto exit;
1716 }
1717
1718 /*
1719 * WARNING: we don't require any capability here so be very careful
1720 * in what is allowed for modification from userspace.
1721 */
1722 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1723 {
1724 unsigned long mmap_max_addr = TASK_SIZE;
1725 struct mm_struct *mm = current->mm;
1726 int error = -EINVAL, i;
1727
1728 static const unsigned char offsets[] = {
1729 offsetof(struct prctl_mm_map, start_code),
1730 offsetof(struct prctl_mm_map, end_code),
1731 offsetof(struct prctl_mm_map, start_data),
1732 offsetof(struct prctl_mm_map, end_data),
1733 offsetof(struct prctl_mm_map, start_brk),
1734 offsetof(struct prctl_mm_map, brk),
1735 offsetof(struct prctl_mm_map, start_stack),
1736 offsetof(struct prctl_mm_map, arg_start),
1737 offsetof(struct prctl_mm_map, arg_end),
1738 offsetof(struct prctl_mm_map, env_start),
1739 offsetof(struct prctl_mm_map, env_end),
1740 };
1741
1742 /*
1743 * Make sure the members are not somewhere outside
1744 * of allowed address space.
1745 */
1746 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1747 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1748
1749 if ((unsigned long)val >= mmap_max_addr ||
1750 (unsigned long)val < mmap_min_addr)
1751 goto out;
1752 }
1753
1754 /*
1755 * Make sure the pairs are ordered.
1756 */
1757 #define __prctl_check_order(__m1, __op, __m2) \
1758 ((unsigned long)prctl_map->__m1 __op \
1759 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1760 error = __prctl_check_order(start_code, <, end_code);
1761 error |= __prctl_check_order(start_data, <, end_data);
1762 error |= __prctl_check_order(start_brk, <=, brk);
1763 error |= __prctl_check_order(arg_start, <=, arg_end);
1764 error |= __prctl_check_order(env_start, <=, env_end);
1765 if (error)
1766 goto out;
1767 #undef __prctl_check_order
1768
1769 error = -EINVAL;
1770
1771 /*
1772 * @brk should be after @end_data in traditional maps.
1773 */
1774 if (prctl_map->start_brk <= prctl_map->end_data ||
1775 prctl_map->brk <= prctl_map->end_data)
1776 goto out;
1777
1778 /*
1779 * Neither we should allow to override limits if they set.
1780 */
1781 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1782 prctl_map->start_brk, prctl_map->end_data,
1783 prctl_map->start_data))
1784 goto out;
1785
1786 /*
1787 * Someone is trying to cheat the auxv vector.
1788 */
1789 if (prctl_map->auxv_size) {
1790 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1791 goto out;
1792 }
1793
1794 /*
1795 * Finally, make sure the caller has the rights to
1796 * change /proc/pid/exe link: only local root should
1797 * be allowed to.
1798 */
1799 if (prctl_map->exe_fd != (u32)-1) {
1800 struct user_namespace *ns = current_user_ns();
1801 const struct cred *cred = current_cred();
1802
1803 if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1804 !gid_eq(cred->gid, make_kgid(ns, 0)))
1805 goto out;
1806 }
1807
1808 error = 0;
1809 out:
1810 return error;
1811 }
1812
1813 #ifdef CONFIG_CHECKPOINT_RESTORE
1814 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1815 {
1816 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1817 unsigned long user_auxv[AT_VECTOR_SIZE];
1818 struct mm_struct *mm = current->mm;
1819 int error;
1820
1821 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1822 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1823
1824 if (opt == PR_SET_MM_MAP_SIZE)
1825 return put_user((unsigned int)sizeof(prctl_map),
1826 (unsigned int __user *)addr);
1827
1828 if (data_size != sizeof(prctl_map))
1829 return -EINVAL;
1830
1831 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1832 return -EFAULT;
1833
1834 error = validate_prctl_map(&prctl_map);
1835 if (error)
1836 return error;
1837
1838 if (prctl_map.auxv_size) {
1839 memset(user_auxv, 0, sizeof(user_auxv));
1840 if (copy_from_user(user_auxv,
1841 (const void __user *)prctl_map.auxv,
1842 prctl_map.auxv_size))
1843 return -EFAULT;
1844
1845 /* Last entry must be AT_NULL as specification requires */
1846 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1847 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1848 }
1849
1850 if (prctl_map.exe_fd != (u32)-1) {
1851 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1852 if (error)
1853 return error;
1854 }
1855
1856 down_write(&mm->mmap_sem);
1857
1858 /*
1859 * We don't validate if these members are pointing to
1860 * real present VMAs because application may have correspond
1861 * VMAs already unmapped and kernel uses these members for statistics
1862 * output in procfs mostly, except
1863 *
1864 * - @start_brk/@brk which are used in do_brk but kernel lookups
1865 * for VMAs when updating these memvers so anything wrong written
1866 * here cause kernel to swear at userspace program but won't lead
1867 * to any problem in kernel itself
1868 */
1869
1870 mm->start_code = prctl_map.start_code;
1871 mm->end_code = prctl_map.end_code;
1872 mm->start_data = prctl_map.start_data;
1873 mm->end_data = prctl_map.end_data;
1874 mm->start_brk = prctl_map.start_brk;
1875 mm->brk = prctl_map.brk;
1876 mm->start_stack = prctl_map.start_stack;
1877 mm->arg_start = prctl_map.arg_start;
1878 mm->arg_end = prctl_map.arg_end;
1879 mm->env_start = prctl_map.env_start;
1880 mm->env_end = prctl_map.env_end;
1881
1882 /*
1883 * Note this update of @saved_auxv is lockless thus
1884 * if someone reads this member in procfs while we're
1885 * updating -- it may get partly updated results. It's
1886 * known and acceptable trade off: we leave it as is to
1887 * not introduce additional locks here making the kernel
1888 * more complex.
1889 */
1890 if (prctl_map.auxv_size)
1891 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1892
1893 up_write(&mm->mmap_sem);
1894 return 0;
1895 }
1896 #endif /* CONFIG_CHECKPOINT_RESTORE */
1897
1898 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
1899 unsigned long len)
1900 {
1901 /*
1902 * This doesn't move the auxiliary vector itself since it's pinned to
1903 * mm_struct, but it permits filling the vector with new values. It's
1904 * up to the caller to provide sane values here, otherwise userspace
1905 * tools which use this vector might be unhappy.
1906 */
1907 unsigned long user_auxv[AT_VECTOR_SIZE];
1908
1909 if (len > sizeof(user_auxv))
1910 return -EINVAL;
1911
1912 if (copy_from_user(user_auxv, (const void __user *)addr, len))
1913 return -EFAULT;
1914
1915 /* Make sure the last entry is always AT_NULL */
1916 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1917 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1918
1919 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1920
1921 task_lock(current);
1922 memcpy(mm->saved_auxv, user_auxv, len);
1923 task_unlock(current);
1924
1925 return 0;
1926 }
1927
1928 static int prctl_set_mm(int opt, unsigned long addr,
1929 unsigned long arg4, unsigned long arg5)
1930 {
1931 struct mm_struct *mm = current->mm;
1932 struct prctl_mm_map prctl_map;
1933 struct vm_area_struct *vma;
1934 int error;
1935
1936 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1937 opt != PR_SET_MM_MAP &&
1938 opt != PR_SET_MM_MAP_SIZE)))
1939 return -EINVAL;
1940
1941 #ifdef CONFIG_CHECKPOINT_RESTORE
1942 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1943 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1944 #endif
1945
1946 if (!capable(CAP_SYS_RESOURCE))
1947 return -EPERM;
1948
1949 if (opt == PR_SET_MM_EXE_FILE)
1950 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1951
1952 if (opt == PR_SET_MM_AUXV)
1953 return prctl_set_auxv(mm, addr, arg4);
1954
1955 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1956 return -EINVAL;
1957
1958 error = -EINVAL;
1959
1960 down_write(&mm->mmap_sem);
1961 vma = find_vma(mm, addr);
1962
1963 prctl_map.start_code = mm->start_code;
1964 prctl_map.end_code = mm->end_code;
1965 prctl_map.start_data = mm->start_data;
1966 prctl_map.end_data = mm->end_data;
1967 prctl_map.start_brk = mm->start_brk;
1968 prctl_map.brk = mm->brk;
1969 prctl_map.start_stack = mm->start_stack;
1970 prctl_map.arg_start = mm->arg_start;
1971 prctl_map.arg_end = mm->arg_end;
1972 prctl_map.env_start = mm->env_start;
1973 prctl_map.env_end = mm->env_end;
1974 prctl_map.auxv = NULL;
1975 prctl_map.auxv_size = 0;
1976 prctl_map.exe_fd = -1;
1977
1978 switch (opt) {
1979 case PR_SET_MM_START_CODE:
1980 prctl_map.start_code = addr;
1981 break;
1982 case PR_SET_MM_END_CODE:
1983 prctl_map.end_code = addr;
1984 break;
1985 case PR_SET_MM_START_DATA:
1986 prctl_map.start_data = addr;
1987 break;
1988 case PR_SET_MM_END_DATA:
1989 prctl_map.end_data = addr;
1990 break;
1991 case PR_SET_MM_START_STACK:
1992 prctl_map.start_stack = addr;
1993 break;
1994 case PR_SET_MM_START_BRK:
1995 prctl_map.start_brk = addr;
1996 break;
1997 case PR_SET_MM_BRK:
1998 prctl_map.brk = addr;
1999 break;
2000 case PR_SET_MM_ARG_START:
2001 prctl_map.arg_start = addr;
2002 break;
2003 case PR_SET_MM_ARG_END:
2004 prctl_map.arg_end = addr;
2005 break;
2006 case PR_SET_MM_ENV_START:
2007 prctl_map.env_start = addr;
2008 break;
2009 case PR_SET_MM_ENV_END:
2010 prctl_map.env_end = addr;
2011 break;
2012 default:
2013 goto out;
2014 }
2015
2016 error = validate_prctl_map(&prctl_map);
2017 if (error)
2018 goto out;
2019
2020 switch (opt) {
2021 /*
2022 * If command line arguments and environment
2023 * are placed somewhere else on stack, we can
2024 * set them up here, ARG_START/END to setup
2025 * command line argumets and ENV_START/END
2026 * for environment.
2027 */
2028 case PR_SET_MM_START_STACK:
2029 case PR_SET_MM_ARG_START:
2030 case PR_SET_MM_ARG_END:
2031 case PR_SET_MM_ENV_START:
2032 case PR_SET_MM_ENV_END:
2033 if (!vma) {
2034 error = -EFAULT;
2035 goto out;
2036 }
2037 }
2038
2039 mm->start_code = prctl_map.start_code;
2040 mm->end_code = prctl_map.end_code;
2041 mm->start_data = prctl_map.start_data;
2042 mm->end_data = prctl_map.end_data;
2043 mm->start_brk = prctl_map.start_brk;
2044 mm->brk = prctl_map.brk;
2045 mm->start_stack = prctl_map.start_stack;
2046 mm->arg_start = prctl_map.arg_start;
2047 mm->arg_end = prctl_map.arg_end;
2048 mm->env_start = prctl_map.env_start;
2049 mm->env_end = prctl_map.env_end;
2050
2051 error = 0;
2052 out:
2053 up_write(&mm->mmap_sem);
2054 return error;
2055 }
2056
2057 #ifdef CONFIG_CHECKPOINT_RESTORE
2058 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2059 {
2060 return put_user(me->clear_child_tid, tid_addr);
2061 }
2062 #else
2063 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2064 {
2065 return -EINVAL;
2066 }
2067 #endif
2068
2069 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2070 {
2071 /*
2072 * If task has has_child_subreaper - all its decendants
2073 * already have these flag too and new decendants will
2074 * inherit it on fork, skip them.
2075 *
2076 * If we've found child_reaper - skip descendants in
2077 * it's subtree as they will never get out pidns.
2078 */
2079 if (p->signal->has_child_subreaper ||
2080 is_child_reaper(task_pid(p)))
2081 return 0;
2082
2083 p->signal->has_child_subreaper = 1;
2084 return 1;
2085 }
2086
2087 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2088 unsigned long, arg4, unsigned long, arg5)
2089 {
2090 struct task_struct *me = current;
2091 unsigned char comm[sizeof(me->comm)];
2092 long error;
2093
2094 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2095 if (error != -ENOSYS)
2096 return error;
2097
2098 error = 0;
2099 switch (option) {
2100 case PR_SET_PDEATHSIG:
2101 if (!valid_signal(arg2)) {
2102 error = -EINVAL;
2103 break;
2104 }
2105 me->pdeath_signal = arg2;
2106 break;
2107 case PR_GET_PDEATHSIG:
2108 error = put_user(me->pdeath_signal, (int __user *)arg2);
2109 break;
2110 case PR_GET_DUMPABLE:
2111 error = get_dumpable(me->mm);
2112 break;
2113 case PR_SET_DUMPABLE:
2114 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2115 error = -EINVAL;
2116 break;
2117 }
2118 set_dumpable(me->mm, arg2);
2119 break;
2120
2121 case PR_SET_UNALIGN:
2122 error = SET_UNALIGN_CTL(me, arg2);
2123 break;
2124 case PR_GET_UNALIGN:
2125 error = GET_UNALIGN_CTL(me, arg2);
2126 break;
2127 case PR_SET_FPEMU:
2128 error = SET_FPEMU_CTL(me, arg2);
2129 break;
2130 case PR_GET_FPEMU:
2131 error = GET_FPEMU_CTL(me, arg2);
2132 break;
2133 case PR_SET_FPEXC:
2134 error = SET_FPEXC_CTL(me, arg2);
2135 break;
2136 case PR_GET_FPEXC:
2137 error = GET_FPEXC_CTL(me, arg2);
2138 break;
2139 case PR_GET_TIMING:
2140 error = PR_TIMING_STATISTICAL;
2141 break;
2142 case PR_SET_TIMING:
2143 if (arg2 != PR_TIMING_STATISTICAL)
2144 error = -EINVAL;
2145 break;
2146 case PR_SET_NAME:
2147 comm[sizeof(me->comm) - 1] = 0;
2148 if (strncpy_from_user(comm, (char __user *)arg2,
2149 sizeof(me->comm) - 1) < 0)
2150 return -EFAULT;
2151 set_task_comm(me, comm);
2152 proc_comm_connector(me);
2153 break;
2154 case PR_GET_NAME:
2155 get_task_comm(comm, me);
2156 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2157 return -EFAULT;
2158 break;
2159 case PR_GET_ENDIAN:
2160 error = GET_ENDIAN(me, arg2);
2161 break;
2162 case PR_SET_ENDIAN:
2163 error = SET_ENDIAN(me, arg2);
2164 break;
2165 case PR_GET_SECCOMP:
2166 error = prctl_get_seccomp();
2167 break;
2168 case PR_SET_SECCOMP:
2169 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2170 break;
2171 case PR_GET_TSC:
2172 error = GET_TSC_CTL(arg2);
2173 break;
2174 case PR_SET_TSC:
2175 error = SET_TSC_CTL(arg2);
2176 break;
2177 case PR_TASK_PERF_EVENTS_DISABLE:
2178 error = perf_event_task_disable();
2179 break;
2180 case PR_TASK_PERF_EVENTS_ENABLE:
2181 error = perf_event_task_enable();
2182 break;
2183 case PR_GET_TIMERSLACK:
2184 if (current->timer_slack_ns > ULONG_MAX)
2185 error = ULONG_MAX;
2186 else
2187 error = current->timer_slack_ns;
2188 break;
2189 case PR_SET_TIMERSLACK:
2190 if (arg2 <= 0)
2191 current->timer_slack_ns =
2192 current->default_timer_slack_ns;
2193 else
2194 current->timer_slack_ns = arg2;
2195 break;
2196 case PR_MCE_KILL:
2197 if (arg4 | arg5)
2198 return -EINVAL;
2199 switch (arg2) {
2200 case PR_MCE_KILL_CLEAR:
2201 if (arg3 != 0)
2202 return -EINVAL;
2203 current->flags &= ~PF_MCE_PROCESS;
2204 break;
2205 case PR_MCE_KILL_SET:
2206 current->flags |= PF_MCE_PROCESS;
2207 if (arg3 == PR_MCE_KILL_EARLY)
2208 current->flags |= PF_MCE_EARLY;
2209 else if (arg3 == PR_MCE_KILL_LATE)
2210 current->flags &= ~PF_MCE_EARLY;
2211 else if (arg3 == PR_MCE_KILL_DEFAULT)
2212 current->flags &=
2213 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2214 else
2215 return -EINVAL;
2216 break;
2217 default:
2218 return -EINVAL;
2219 }
2220 break;
2221 case PR_MCE_KILL_GET:
2222 if (arg2 | arg3 | arg4 | arg5)
2223 return -EINVAL;
2224 if (current->flags & PF_MCE_PROCESS)
2225 error = (current->flags & PF_MCE_EARLY) ?
2226 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2227 else
2228 error = PR_MCE_KILL_DEFAULT;
2229 break;
2230 case PR_SET_MM:
2231 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2232 break;
2233 case PR_GET_TID_ADDRESS:
2234 error = prctl_get_tid_address(me, (int __user **)arg2);
2235 break;
2236 case PR_SET_CHILD_SUBREAPER:
2237 me->signal->is_child_subreaper = !!arg2;
2238 if (!arg2)
2239 break;
2240
2241 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2242 break;
2243 case PR_GET_CHILD_SUBREAPER:
2244 error = put_user(me->signal->is_child_subreaper,
2245 (int __user *)arg2);
2246 break;
2247 case PR_SET_NO_NEW_PRIVS:
2248 if (arg2 != 1 || arg3 || arg4 || arg5)
2249 return -EINVAL;
2250
2251 task_set_no_new_privs(current);
2252 break;
2253 case PR_GET_NO_NEW_PRIVS:
2254 if (arg2 || arg3 || arg4 || arg5)
2255 return -EINVAL;
2256 return task_no_new_privs(current) ? 1 : 0;
2257 case PR_GET_THP_DISABLE:
2258 if (arg2 || arg3 || arg4 || arg5)
2259 return -EINVAL;
2260 error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2261 break;
2262 case PR_SET_THP_DISABLE:
2263 if (arg3 || arg4 || arg5)
2264 return -EINVAL;
2265 if (down_write_killable(&me->mm->mmap_sem))
2266 return -EINTR;
2267 if (arg2)
2268 me->mm->def_flags |= VM_NOHUGEPAGE;
2269 else
2270 me->mm->def_flags &= ~VM_NOHUGEPAGE;
2271 up_write(&me->mm->mmap_sem);
2272 break;
2273 case PR_MPX_ENABLE_MANAGEMENT:
2274 if (arg2 || arg3 || arg4 || arg5)
2275 return -EINVAL;
2276 error = MPX_ENABLE_MANAGEMENT();
2277 break;
2278 case PR_MPX_DISABLE_MANAGEMENT:
2279 if (arg2 || arg3 || arg4 || arg5)
2280 return -EINVAL;
2281 error = MPX_DISABLE_MANAGEMENT();
2282 break;
2283 case PR_SET_FP_MODE:
2284 error = SET_FP_MODE(me, arg2);
2285 break;
2286 case PR_GET_FP_MODE:
2287 error = GET_FP_MODE(me);
2288 break;
2289 default:
2290 error = -EINVAL;
2291 break;
2292 }
2293 return error;
2294 }
2295
2296 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2297 struct getcpu_cache __user *, unused)
2298 {
2299 int err = 0;
2300 int cpu = raw_smp_processor_id();
2301
2302 if (cpup)
2303 err |= put_user(cpu, cpup);
2304 if (nodep)
2305 err |= put_user(cpu_to_node(cpu), nodep);
2306 return err ? -EFAULT : 0;
2307 }
2308
2309 /**
2310 * do_sysinfo - fill in sysinfo struct
2311 * @info: pointer to buffer to fill
2312 */
2313 static int do_sysinfo(struct sysinfo *info)
2314 {
2315 unsigned long mem_total, sav_total;
2316 unsigned int mem_unit, bitcount;
2317 struct timespec tp;
2318
2319 memset(info, 0, sizeof(struct sysinfo));
2320
2321 get_monotonic_boottime(&tp);
2322 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2323
2324 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2325
2326 info->procs = nr_threads;
2327
2328 si_meminfo(info);
2329 si_swapinfo(info);
2330
2331 /*
2332 * If the sum of all the available memory (i.e. ram + swap)
2333 * is less than can be stored in a 32 bit unsigned long then
2334 * we can be binary compatible with 2.2.x kernels. If not,
2335 * well, in that case 2.2.x was broken anyways...
2336 *
2337 * -Erik Andersen <andersee@debian.org>
2338 */
2339
2340 mem_total = info->totalram + info->totalswap;
2341 if (mem_total < info->totalram || mem_total < info->totalswap)
2342 goto out;
2343 bitcount = 0;
2344 mem_unit = info->mem_unit;
2345 while (mem_unit > 1) {
2346 bitcount++;
2347 mem_unit >>= 1;
2348 sav_total = mem_total;
2349 mem_total <<= 1;
2350 if (mem_total < sav_total)
2351 goto out;
2352 }
2353
2354 /*
2355 * If mem_total did not overflow, multiply all memory values by
2356 * info->mem_unit and set it to 1. This leaves things compatible
2357 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2358 * kernels...
2359 */
2360
2361 info->mem_unit = 1;
2362 info->totalram <<= bitcount;
2363 info->freeram <<= bitcount;
2364 info->sharedram <<= bitcount;
2365 info->bufferram <<= bitcount;
2366 info->totalswap <<= bitcount;
2367 info->freeswap <<= bitcount;
2368 info->totalhigh <<= bitcount;
2369 info->freehigh <<= bitcount;
2370
2371 out:
2372 return 0;
2373 }
2374
2375 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2376 {
2377 struct sysinfo val;
2378
2379 do_sysinfo(&val);
2380
2381 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2382 return -EFAULT;
2383
2384 return 0;
2385 }
2386
2387 #ifdef CONFIG_COMPAT
2388 struct compat_sysinfo {
2389 s32 uptime;
2390 u32 loads[3];
2391 u32 totalram;
2392 u32 freeram;
2393 u32 sharedram;
2394 u32 bufferram;
2395 u32 totalswap;
2396 u32 freeswap;
2397 u16 procs;
2398 u16 pad;
2399 u32 totalhigh;
2400 u32 freehigh;
2401 u32 mem_unit;
2402 char _f[20-2*sizeof(u32)-sizeof(int)];
2403 };
2404
2405 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2406 {
2407 struct sysinfo s;
2408
2409 do_sysinfo(&s);
2410
2411 /* Check to see if any memory value is too large for 32-bit and scale
2412 * down if needed
2413 */
2414 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2415 int bitcount = 0;
2416
2417 while (s.mem_unit < PAGE_SIZE) {
2418 s.mem_unit <<= 1;
2419 bitcount++;
2420 }
2421
2422 s.totalram >>= bitcount;
2423 s.freeram >>= bitcount;
2424 s.sharedram >>= bitcount;
2425 s.bufferram >>= bitcount;
2426 s.totalswap >>= bitcount;
2427 s.freeswap >>= bitcount;
2428 s.totalhigh >>= bitcount;
2429 s.freehigh >>= bitcount;
2430 }
2431
2432 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2433 __put_user(s.uptime, &info->uptime) ||
2434 __put_user(s.loads[0], &info->loads[0]) ||
2435 __put_user(s.loads[1], &info->loads[1]) ||
2436 __put_user(s.loads[2], &info->loads[2]) ||
2437 __put_user(s.totalram, &info->totalram) ||
2438 __put_user(s.freeram, &info->freeram) ||
2439 __put_user(s.sharedram, &info->sharedram) ||
2440 __put_user(s.bufferram, &info->bufferram) ||
2441 __put_user(s.totalswap, &info->totalswap) ||
2442 __put_user(s.freeswap, &info->freeswap) ||
2443 __put_user(s.procs, &info->procs) ||
2444 __put_user(s.totalhigh, &info->totalhigh) ||
2445 __put_user(s.freehigh, &info->freehigh) ||
2446 __put_user(s.mem_unit, &info->mem_unit))
2447 return -EFAULT;
2448
2449 return 0;
2450 }
2451 #endif /* CONFIG_COMPAT */
Mirror https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git