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