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