]> 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
perf: Fix sample_max_stack maximum check
[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 int errno = 0;
1201
1202 down_read(&uts_sem);
1203 if (copy_to_user(name, utsname(), sizeof *name))
1204 errno = -EFAULT;
1205 up_read(&uts_sem);
1206
1207 if (!errno && override_release(name->release, sizeof(name->release)))
1208 errno = -EFAULT;
1209 if (!errno && override_architecture(name))
1210 errno = -EFAULT;
1211 return errno;
1212 }
1213
1214 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1215 /*
1216 * Old cruft
1217 */
1218 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1219 {
1220 int error = 0;
1221
1222 if (!name)
1223 return -EFAULT;
1224
1225 down_read(&uts_sem);
1226 if (copy_to_user(name, utsname(), sizeof(*name)))
1227 error = -EFAULT;
1228 up_read(&uts_sem);
1229
1230 if (!error && override_release(name->release, sizeof(name->release)))
1231 error = -EFAULT;
1232 if (!error && override_architecture(name))
1233 error = -EFAULT;
1234 return error;
1235 }
1236
1237 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1238 {
1239 int error;
1240
1241 if (!name)
1242 return -EFAULT;
1243 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1244 return -EFAULT;
1245
1246 down_read(&uts_sem);
1247 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1248 __OLD_UTS_LEN);
1249 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1250 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1251 __OLD_UTS_LEN);
1252 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1253 error |= __copy_to_user(&name->release, &utsname()->release,
1254 __OLD_UTS_LEN);
1255 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1256 error |= __copy_to_user(&name->version, &utsname()->version,
1257 __OLD_UTS_LEN);
1258 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1259 error |= __copy_to_user(&name->machine, &utsname()->machine,
1260 __OLD_UTS_LEN);
1261 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1262 up_read(&uts_sem);
1263
1264 if (!error && override_architecture(name))
1265 error = -EFAULT;
1266 if (!error && override_release(name->release, sizeof(name->release)))
1267 error = -EFAULT;
1268 return error ? -EFAULT : 0;
1269 }
1270 #endif
1271
1272 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1273 {
1274 int errno;
1275 char tmp[__NEW_UTS_LEN];
1276
1277 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1278 return -EPERM;
1279
1280 if (len < 0 || len > __NEW_UTS_LEN)
1281 return -EINVAL;
1282 down_write(&uts_sem);
1283 errno = -EFAULT;
1284 if (!copy_from_user(tmp, name, len)) {
1285 struct new_utsname *u = utsname();
1286
1287 memcpy(u->nodename, tmp, len);
1288 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1289 errno = 0;
1290 uts_proc_notify(UTS_PROC_HOSTNAME);
1291 }
1292 up_write(&uts_sem);
1293 return errno;
1294 }
1295
1296 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1297
1298 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1299 {
1300 int i, errno;
1301 struct new_utsname *u;
1302
1303 if (len < 0)
1304 return -EINVAL;
1305 down_read(&uts_sem);
1306 u = utsname();
1307 i = 1 + strlen(u->nodename);
1308 if (i > len)
1309 i = len;
1310 errno = 0;
1311 if (copy_to_user(name, u->nodename, i))
1312 errno = -EFAULT;
1313 up_read(&uts_sem);
1314 return errno;
1315 }
1316
1317 #endif
1318
1319 /*
1320 * Only setdomainname; getdomainname can be implemented by calling
1321 * uname()
1322 */
1323 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1324 {
1325 int errno;
1326 char tmp[__NEW_UTS_LEN];
1327
1328 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1329 return -EPERM;
1330 if (len < 0 || len > __NEW_UTS_LEN)
1331 return -EINVAL;
1332
1333 down_write(&uts_sem);
1334 errno = -EFAULT;
1335 if (!copy_from_user(tmp, name, len)) {
1336 struct new_utsname *u = utsname();
1337
1338 memcpy(u->domainname, tmp, len);
1339 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1340 errno = 0;
1341 uts_proc_notify(UTS_PROC_DOMAINNAME);
1342 }
1343 up_write(&uts_sem);
1344 return errno;
1345 }
1346
1347 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1348 {
1349 struct rlimit value;
1350 int ret;
1351
1352 ret = do_prlimit(current, resource, NULL, &value);
1353 if (!ret)
1354 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1355
1356 return ret;
1357 }
1358
1359 #ifdef CONFIG_COMPAT
1360
1361 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1362 struct compat_rlimit __user *, rlim)
1363 {
1364 struct rlimit r;
1365 struct compat_rlimit r32;
1366
1367 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1368 return -EFAULT;
1369
1370 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1371 r.rlim_cur = RLIM_INFINITY;
1372 else
1373 r.rlim_cur = r32.rlim_cur;
1374 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1375 r.rlim_max = RLIM_INFINITY;
1376 else
1377 r.rlim_max = r32.rlim_max;
1378 return do_prlimit(current, resource, &r, NULL);
1379 }
1380
1381 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1382 struct compat_rlimit __user *, rlim)
1383 {
1384 struct rlimit r;
1385 int ret;
1386
1387 ret = do_prlimit(current, resource, NULL, &r);
1388 if (!ret) {
1389 struct compat_rlimit r32;
1390 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1391 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1392 else
1393 r32.rlim_cur = r.rlim_cur;
1394 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1395 r32.rlim_max = COMPAT_RLIM_INFINITY;
1396 else
1397 r32.rlim_max = r.rlim_max;
1398
1399 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1400 return -EFAULT;
1401 }
1402 return ret;
1403 }
1404
1405 #endif
1406
1407 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1408
1409 /*
1410 * Back compatibility for getrlimit. Needed for some apps.
1411 */
1412 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1413 struct rlimit __user *, rlim)
1414 {
1415 struct rlimit x;
1416 if (resource >= RLIM_NLIMITS)
1417 return -EINVAL;
1418
1419 task_lock(current->group_leader);
1420 x = current->signal->rlim[resource];
1421 task_unlock(current->group_leader);
1422 if (x.rlim_cur > 0x7FFFFFFF)
1423 x.rlim_cur = 0x7FFFFFFF;
1424 if (x.rlim_max > 0x7FFFFFFF)
1425 x.rlim_max = 0x7FFFFFFF;
1426 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1427 }
1428
1429 #ifdef CONFIG_COMPAT
1430 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1431 struct compat_rlimit __user *, rlim)
1432 {
1433 struct rlimit r;
1434
1435 if (resource >= RLIM_NLIMITS)
1436 return -EINVAL;
1437
1438 task_lock(current->group_leader);
1439 r = current->signal->rlim[resource];
1440 task_unlock(current->group_leader);
1441 if (r.rlim_cur > 0x7FFFFFFF)
1442 r.rlim_cur = 0x7FFFFFFF;
1443 if (r.rlim_max > 0x7FFFFFFF)
1444 r.rlim_max = 0x7FFFFFFF;
1445
1446 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1447 put_user(r.rlim_max, &rlim->rlim_max))
1448 return -EFAULT;
1449 return 0;
1450 }
1451 #endif
1452
1453 #endif
1454
1455 static inline bool rlim64_is_infinity(__u64 rlim64)
1456 {
1457 #if BITS_PER_LONG < 64
1458 return rlim64 >= ULONG_MAX;
1459 #else
1460 return rlim64 == RLIM64_INFINITY;
1461 #endif
1462 }
1463
1464 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1465 {
1466 if (rlim->rlim_cur == RLIM_INFINITY)
1467 rlim64->rlim_cur = RLIM64_INFINITY;
1468 else
1469 rlim64->rlim_cur = rlim->rlim_cur;
1470 if (rlim->rlim_max == RLIM_INFINITY)
1471 rlim64->rlim_max = RLIM64_INFINITY;
1472 else
1473 rlim64->rlim_max = rlim->rlim_max;
1474 }
1475
1476 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1477 {
1478 if (rlim64_is_infinity(rlim64->rlim_cur))
1479 rlim->rlim_cur = RLIM_INFINITY;
1480 else
1481 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1482 if (rlim64_is_infinity(rlim64->rlim_max))
1483 rlim->rlim_max = RLIM_INFINITY;
1484 else
1485 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1486 }
1487
1488 /* make sure you are allowed to change @tsk limits before calling this */
1489 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1490 struct rlimit *new_rlim, struct rlimit *old_rlim)
1491 {
1492 struct rlimit *rlim;
1493 int retval = 0;
1494
1495 if (resource >= RLIM_NLIMITS)
1496 return -EINVAL;
1497 if (new_rlim) {
1498 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1499 return -EINVAL;
1500 if (resource == RLIMIT_NOFILE &&
1501 new_rlim->rlim_max > sysctl_nr_open)
1502 return -EPERM;
1503 }
1504
1505 /* protect tsk->signal and tsk->sighand from disappearing */
1506 read_lock(&tasklist_lock);
1507 if (!tsk->sighand) {
1508 retval = -ESRCH;
1509 goto out;
1510 }
1511
1512 rlim = tsk->signal->rlim + resource;
1513 task_lock(tsk->group_leader);
1514 if (new_rlim) {
1515 /* Keep the capable check against init_user_ns until
1516 cgroups can contain all limits */
1517 if (new_rlim->rlim_max > rlim->rlim_max &&
1518 !capable(CAP_SYS_RESOURCE))
1519 retval = -EPERM;
1520 if (!retval)
1521 retval = security_task_setrlimit(tsk, resource, new_rlim);
1522 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1523 /*
1524 * The caller is asking for an immediate RLIMIT_CPU
1525 * expiry. But we use the zero value to mean "it was
1526 * never set". So let's cheat and make it one second
1527 * instead
1528 */
1529 new_rlim->rlim_cur = 1;
1530 }
1531 }
1532 if (!retval) {
1533 if (old_rlim)
1534 *old_rlim = *rlim;
1535 if (new_rlim)
1536 *rlim = *new_rlim;
1537 }
1538 task_unlock(tsk->group_leader);
1539
1540 /*
1541 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1542 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1543 * very long-standing error, and fixing it now risks breakage of
1544 * applications, so we live with it
1545 */
1546 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1547 new_rlim->rlim_cur != RLIM_INFINITY &&
1548 IS_ENABLED(CONFIG_POSIX_TIMERS))
1549 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1550 out:
1551 read_unlock(&tasklist_lock);
1552 return retval;
1553 }
1554
1555 /* rcu lock must be held */
1556 static int check_prlimit_permission(struct task_struct *task,
1557 unsigned int flags)
1558 {
1559 const struct cred *cred = current_cred(), *tcred;
1560 bool id_match;
1561
1562 if (current == task)
1563 return 0;
1564
1565 tcred = __task_cred(task);
1566 id_match = (uid_eq(cred->uid, tcred->euid) &&
1567 uid_eq(cred->uid, tcred->suid) &&
1568 uid_eq(cred->uid, tcred->uid) &&
1569 gid_eq(cred->gid, tcred->egid) &&
1570 gid_eq(cred->gid, tcred->sgid) &&
1571 gid_eq(cred->gid, tcred->gid));
1572 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1573 return -EPERM;
1574
1575 return security_task_prlimit(cred, tcred, flags);
1576 }
1577
1578 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1579 const struct rlimit64 __user *, new_rlim,
1580 struct rlimit64 __user *, old_rlim)
1581 {
1582 struct rlimit64 old64, new64;
1583 struct rlimit old, new;
1584 struct task_struct *tsk;
1585 unsigned int checkflags = 0;
1586 int ret;
1587
1588 if (old_rlim)
1589 checkflags |= LSM_PRLIMIT_READ;
1590
1591 if (new_rlim) {
1592 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1593 return -EFAULT;
1594 rlim64_to_rlim(&new64, &new);
1595 checkflags |= LSM_PRLIMIT_WRITE;
1596 }
1597
1598 rcu_read_lock();
1599 tsk = pid ? find_task_by_vpid(pid) : current;
1600 if (!tsk) {
1601 rcu_read_unlock();
1602 return -ESRCH;
1603 }
1604 ret = check_prlimit_permission(tsk, checkflags);
1605 if (ret) {
1606 rcu_read_unlock();
1607 return ret;
1608 }
1609 get_task_struct(tsk);
1610 rcu_read_unlock();
1611
1612 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1613 old_rlim ? &old : NULL);
1614
1615 if (!ret && old_rlim) {
1616 rlim_to_rlim64(&old, &old64);
1617 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1618 ret = -EFAULT;
1619 }
1620
1621 put_task_struct(tsk);
1622 return ret;
1623 }
1624
1625 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1626 {
1627 struct rlimit new_rlim;
1628
1629 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1630 return -EFAULT;
1631 return do_prlimit(current, resource, &new_rlim, NULL);
1632 }
1633
1634 /*
1635 * It would make sense to put struct rusage in the task_struct,
1636 * except that would make the task_struct be *really big*. After
1637 * task_struct gets moved into malloc'ed memory, it would
1638 * make sense to do this. It will make moving the rest of the information
1639 * a lot simpler! (Which we're not doing right now because we're not
1640 * measuring them yet).
1641 *
1642 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1643 * races with threads incrementing their own counters. But since word
1644 * reads are atomic, we either get new values or old values and we don't
1645 * care which for the sums. We always take the siglock to protect reading
1646 * the c* fields from p->signal from races with exit.c updating those
1647 * fields when reaping, so a sample either gets all the additions of a
1648 * given child after it's reaped, or none so this sample is before reaping.
1649 *
1650 * Locking:
1651 * We need to take the siglock for CHILDEREN, SELF and BOTH
1652 * for the cases current multithreaded, non-current single threaded
1653 * non-current multithreaded. Thread traversal is now safe with
1654 * the siglock held.
1655 * Strictly speaking, we donot need to take the siglock if we are current and
1656 * single threaded, as no one else can take our signal_struct away, no one
1657 * else can reap the children to update signal->c* counters, and no one else
1658 * can race with the signal-> fields. If we do not take any lock, the
1659 * signal-> fields could be read out of order while another thread was just
1660 * exiting. So we should place a read memory barrier when we avoid the lock.
1661 * On the writer side, write memory barrier is implied in __exit_signal
1662 * as __exit_signal releases the siglock spinlock after updating the signal->
1663 * fields. But we don't do this yet to keep things simple.
1664 *
1665 */
1666
1667 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1668 {
1669 r->ru_nvcsw += t->nvcsw;
1670 r->ru_nivcsw += t->nivcsw;
1671 r->ru_minflt += t->min_flt;
1672 r->ru_majflt += t->maj_flt;
1673 r->ru_inblock += task_io_get_inblock(t);
1674 r->ru_oublock += task_io_get_oublock(t);
1675 }
1676
1677 void getrusage(struct task_struct *p, int who, struct rusage *r)
1678 {
1679 struct task_struct *t;
1680 unsigned long flags;
1681 u64 tgutime, tgstime, utime, stime;
1682 unsigned long maxrss = 0;
1683
1684 memset((char *)r, 0, sizeof (*r));
1685 utime = stime = 0;
1686
1687 if (who == RUSAGE_THREAD) {
1688 task_cputime_adjusted(current, &utime, &stime);
1689 accumulate_thread_rusage(p, r);
1690 maxrss = p->signal->maxrss;
1691 goto out;
1692 }
1693
1694 if (!lock_task_sighand(p, &flags))
1695 return;
1696
1697 switch (who) {
1698 case RUSAGE_BOTH:
1699 case RUSAGE_CHILDREN:
1700 utime = p->signal->cutime;
1701 stime = p->signal->cstime;
1702 r->ru_nvcsw = p->signal->cnvcsw;
1703 r->ru_nivcsw = p->signal->cnivcsw;
1704 r->ru_minflt = p->signal->cmin_flt;
1705 r->ru_majflt = p->signal->cmaj_flt;
1706 r->ru_inblock = p->signal->cinblock;
1707 r->ru_oublock = p->signal->coublock;
1708 maxrss = p->signal->cmaxrss;
1709
1710 if (who == RUSAGE_CHILDREN)
1711 break;
1712
1713 case RUSAGE_SELF:
1714 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1715 utime += tgutime;
1716 stime += tgstime;
1717 r->ru_nvcsw += p->signal->nvcsw;
1718 r->ru_nivcsw += p->signal->nivcsw;
1719 r->ru_minflt += p->signal->min_flt;
1720 r->ru_majflt += p->signal->maj_flt;
1721 r->ru_inblock += p->signal->inblock;
1722 r->ru_oublock += p->signal->oublock;
1723 if (maxrss < p->signal->maxrss)
1724 maxrss = p->signal->maxrss;
1725 t = p;
1726 do {
1727 accumulate_thread_rusage(t, r);
1728 } while_each_thread(p, t);
1729 break;
1730
1731 default:
1732 BUG();
1733 }
1734 unlock_task_sighand(p, &flags);
1735
1736 out:
1737 r->ru_utime = ns_to_timeval(utime);
1738 r->ru_stime = ns_to_timeval(stime);
1739
1740 if (who != RUSAGE_CHILDREN) {
1741 struct mm_struct *mm = get_task_mm(p);
1742
1743 if (mm) {
1744 setmax_mm_hiwater_rss(&maxrss, mm);
1745 mmput(mm);
1746 }
1747 }
1748 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1749 }
1750
1751 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1752 {
1753 struct rusage r;
1754
1755 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1756 who != RUSAGE_THREAD)
1757 return -EINVAL;
1758
1759 getrusage(current, who, &r);
1760 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1761 }
1762
1763 #ifdef CONFIG_COMPAT
1764 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1765 {
1766 struct rusage r;
1767
1768 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1769 who != RUSAGE_THREAD)
1770 return -EINVAL;
1771
1772 getrusage(current, who, &r);
1773 return put_compat_rusage(&r, ru);
1774 }
1775 #endif
1776
1777 SYSCALL_DEFINE1(umask, int, mask)
1778 {
1779 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1780 return mask;
1781 }
1782
1783 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1784 {
1785 struct fd exe;
1786 struct file *old_exe, *exe_file;
1787 struct inode *inode;
1788 int err;
1789
1790 exe = fdget(fd);
1791 if (!exe.file)
1792 return -EBADF;
1793
1794 inode = file_inode(exe.file);
1795
1796 /*
1797 * Because the original mm->exe_file points to executable file, make
1798 * sure that this one is executable as well, to avoid breaking an
1799 * overall picture.
1800 */
1801 err = -EACCES;
1802 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1803 goto exit;
1804
1805 err = inode_permission(inode, MAY_EXEC);
1806 if (err)
1807 goto exit;
1808
1809 /*
1810 * Forbid mm->exe_file change if old file still mapped.
1811 */
1812 exe_file = get_mm_exe_file(mm);
1813 err = -EBUSY;
1814 if (exe_file) {
1815 struct vm_area_struct *vma;
1816
1817 down_read(&mm->mmap_sem);
1818 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1819 if (!vma->vm_file)
1820 continue;
1821 if (path_equal(&vma->vm_file->f_path,
1822 &exe_file->f_path))
1823 goto exit_err;
1824 }
1825
1826 up_read(&mm->mmap_sem);
1827 fput(exe_file);
1828 }
1829
1830 err = 0;
1831 /* set the new file, lockless */
1832 get_file(exe.file);
1833 old_exe = xchg(&mm->exe_file, exe.file);
1834 if (old_exe)
1835 fput(old_exe);
1836 exit:
1837 fdput(exe);
1838 return err;
1839 exit_err:
1840 up_read(&mm->mmap_sem);
1841 fput(exe_file);
1842 goto exit;
1843 }
1844
1845 /*
1846 * WARNING: we don't require any capability here so be very careful
1847 * in what is allowed for modification from userspace.
1848 */
1849 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1850 {
1851 unsigned long mmap_max_addr = TASK_SIZE;
1852 struct mm_struct *mm = current->mm;
1853 int error = -EINVAL, i;
1854
1855 static const unsigned char offsets[] = {
1856 offsetof(struct prctl_mm_map, start_code),
1857 offsetof(struct prctl_mm_map, end_code),
1858 offsetof(struct prctl_mm_map, start_data),
1859 offsetof(struct prctl_mm_map, end_data),
1860 offsetof(struct prctl_mm_map, start_brk),
1861 offsetof(struct prctl_mm_map, brk),
1862 offsetof(struct prctl_mm_map, start_stack),
1863 offsetof(struct prctl_mm_map, arg_start),
1864 offsetof(struct prctl_mm_map, arg_end),
1865 offsetof(struct prctl_mm_map, env_start),
1866 offsetof(struct prctl_mm_map, env_end),
1867 };
1868
1869 /*
1870 * Make sure the members are not somewhere outside
1871 * of allowed address space.
1872 */
1873 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1874 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1875
1876 if ((unsigned long)val >= mmap_max_addr ||
1877 (unsigned long)val < mmap_min_addr)
1878 goto out;
1879 }
1880
1881 /*
1882 * Make sure the pairs are ordered.
1883 */
1884 #define __prctl_check_order(__m1, __op, __m2) \
1885 ((unsigned long)prctl_map->__m1 __op \
1886 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1887 error = __prctl_check_order(start_code, <, end_code);
1888 error |= __prctl_check_order(start_data, <, end_data);
1889 error |= __prctl_check_order(start_brk, <=, brk);
1890 error |= __prctl_check_order(arg_start, <=, arg_end);
1891 error |= __prctl_check_order(env_start, <=, env_end);
1892 if (error)
1893 goto out;
1894 #undef __prctl_check_order
1895
1896 error = -EINVAL;
1897
1898 /*
1899 * @brk should be after @end_data in traditional maps.
1900 */
1901 if (prctl_map->start_brk <= prctl_map->end_data ||
1902 prctl_map->brk <= prctl_map->end_data)
1903 goto out;
1904
1905 /*
1906 * Neither we should allow to override limits if they set.
1907 */
1908 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1909 prctl_map->start_brk, prctl_map->end_data,
1910 prctl_map->start_data))
1911 goto out;
1912
1913 /*
1914 * Someone is trying to cheat the auxv vector.
1915 */
1916 if (prctl_map->auxv_size) {
1917 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1918 goto out;
1919 }
1920
1921 /*
1922 * Finally, make sure the caller has the rights to
1923 * change /proc/pid/exe link: only local sys admin should
1924 * be allowed to.
1925 */
1926 if (prctl_map->exe_fd != (u32)-1) {
1927 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1928 goto out;
1929 }
1930
1931 error = 0;
1932 out:
1933 return error;
1934 }
1935
1936 #ifdef CONFIG_CHECKPOINT_RESTORE
1937 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1938 {
1939 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1940 unsigned long user_auxv[AT_VECTOR_SIZE];
1941 struct mm_struct *mm = current->mm;
1942 int error;
1943
1944 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1945 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1946
1947 if (opt == PR_SET_MM_MAP_SIZE)
1948 return put_user((unsigned int)sizeof(prctl_map),
1949 (unsigned int __user *)addr);
1950
1951 if (data_size != sizeof(prctl_map))
1952 return -EINVAL;
1953
1954 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1955 return -EFAULT;
1956
1957 error = validate_prctl_map(&prctl_map);
1958 if (error)
1959 return error;
1960
1961 if (prctl_map.auxv_size) {
1962 memset(user_auxv, 0, sizeof(user_auxv));
1963 if (copy_from_user(user_auxv,
1964 (const void __user *)prctl_map.auxv,
1965 prctl_map.auxv_size))
1966 return -EFAULT;
1967
1968 /* Last entry must be AT_NULL as specification requires */
1969 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1970 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1971 }
1972
1973 if (prctl_map.exe_fd != (u32)-1) {
1974 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1975 if (error)
1976 return error;
1977 }
1978
1979 down_write(&mm->mmap_sem);
1980
1981 /*
1982 * We don't validate if these members are pointing to
1983 * real present VMAs because application may have correspond
1984 * VMAs already unmapped and kernel uses these members for statistics
1985 * output in procfs mostly, except
1986 *
1987 * - @start_brk/@brk which are used in do_brk but kernel lookups
1988 * for VMAs when updating these memvers so anything wrong written
1989 * here cause kernel to swear at userspace program but won't lead
1990 * to any problem in kernel itself
1991 */
1992
1993 mm->start_code = prctl_map.start_code;
1994 mm->end_code = prctl_map.end_code;
1995 mm->start_data = prctl_map.start_data;
1996 mm->end_data = prctl_map.end_data;
1997 mm->start_brk = prctl_map.start_brk;
1998 mm->brk = prctl_map.brk;
1999 mm->start_stack = prctl_map.start_stack;
2000 mm->arg_start = prctl_map.arg_start;
2001 mm->arg_end = prctl_map.arg_end;
2002 mm->env_start = prctl_map.env_start;
2003 mm->env_end = prctl_map.env_end;
2004
2005 /*
2006 * Note this update of @saved_auxv is lockless thus
2007 * if someone reads this member in procfs while we're
2008 * updating -- it may get partly updated results. It's
2009 * known and acceptable trade off: we leave it as is to
2010 * not introduce additional locks here making the kernel
2011 * more complex.
2012 */
2013 if (prctl_map.auxv_size)
2014 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2015
2016 up_write(&mm->mmap_sem);
2017 return 0;
2018 }
2019 #endif /* CONFIG_CHECKPOINT_RESTORE */
2020
2021 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2022 unsigned long len)
2023 {
2024 /*
2025 * This doesn't move the auxiliary vector itself since it's pinned to
2026 * mm_struct, but it permits filling the vector with new values. It's
2027 * up to the caller to provide sane values here, otherwise userspace
2028 * tools which use this vector might be unhappy.
2029 */
2030 unsigned long user_auxv[AT_VECTOR_SIZE];
2031
2032 if (len > sizeof(user_auxv))
2033 return -EINVAL;
2034
2035 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2036 return -EFAULT;
2037
2038 /* Make sure the last entry is always AT_NULL */
2039 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2040 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2041
2042 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2043
2044 task_lock(current);
2045 memcpy(mm->saved_auxv, user_auxv, len);
2046 task_unlock(current);
2047
2048 return 0;
2049 }
2050
2051 static int prctl_set_mm(int opt, unsigned long addr,
2052 unsigned long arg4, unsigned long arg5)
2053 {
2054 struct mm_struct *mm = current->mm;
2055 struct prctl_mm_map prctl_map;
2056 struct vm_area_struct *vma;
2057 int error;
2058
2059 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2060 opt != PR_SET_MM_MAP &&
2061 opt != PR_SET_MM_MAP_SIZE)))
2062 return -EINVAL;
2063
2064 #ifdef CONFIG_CHECKPOINT_RESTORE
2065 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2066 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2067 #endif
2068
2069 if (!capable(CAP_SYS_RESOURCE))
2070 return -EPERM;
2071
2072 if (opt == PR_SET_MM_EXE_FILE)
2073 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2074
2075 if (opt == PR_SET_MM_AUXV)
2076 return prctl_set_auxv(mm, addr, arg4);
2077
2078 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2079 return -EINVAL;
2080
2081 error = -EINVAL;
2082
2083 down_write(&mm->mmap_sem);
2084 vma = find_vma(mm, addr);
2085
2086 prctl_map.start_code = mm->start_code;
2087 prctl_map.end_code = mm->end_code;
2088 prctl_map.start_data = mm->start_data;
2089 prctl_map.end_data = mm->end_data;
2090 prctl_map.start_brk = mm->start_brk;
2091 prctl_map.brk = mm->brk;
2092 prctl_map.start_stack = mm->start_stack;
2093 prctl_map.arg_start = mm->arg_start;
2094 prctl_map.arg_end = mm->arg_end;
2095 prctl_map.env_start = mm->env_start;
2096 prctl_map.env_end = mm->env_end;
2097 prctl_map.auxv = NULL;
2098 prctl_map.auxv_size = 0;
2099 prctl_map.exe_fd = -1;
2100
2101 switch (opt) {
2102 case PR_SET_MM_START_CODE:
2103 prctl_map.start_code = addr;
2104 break;
2105 case PR_SET_MM_END_CODE:
2106 prctl_map.end_code = addr;
2107 break;
2108 case PR_SET_MM_START_DATA:
2109 prctl_map.start_data = addr;
2110 break;
2111 case PR_SET_MM_END_DATA:
2112 prctl_map.end_data = addr;
2113 break;
2114 case PR_SET_MM_START_STACK:
2115 prctl_map.start_stack = addr;
2116 break;
2117 case PR_SET_MM_START_BRK:
2118 prctl_map.start_brk = addr;
2119 break;
2120 case PR_SET_MM_BRK:
2121 prctl_map.brk = addr;
2122 break;
2123 case PR_SET_MM_ARG_START:
2124 prctl_map.arg_start = addr;
2125 break;
2126 case PR_SET_MM_ARG_END:
2127 prctl_map.arg_end = addr;
2128 break;
2129 case PR_SET_MM_ENV_START:
2130 prctl_map.env_start = addr;
2131 break;
2132 case PR_SET_MM_ENV_END:
2133 prctl_map.env_end = addr;
2134 break;
2135 default:
2136 goto out;
2137 }
2138
2139 error = validate_prctl_map(&prctl_map);
2140 if (error)
2141 goto out;
2142
2143 switch (opt) {
2144 /*
2145 * If command line arguments and environment
2146 * are placed somewhere else on stack, we can
2147 * set them up here, ARG_START/END to setup
2148 * command line argumets and ENV_START/END
2149 * for environment.
2150 */
2151 case PR_SET_MM_START_STACK:
2152 case PR_SET_MM_ARG_START:
2153 case PR_SET_MM_ARG_END:
2154 case PR_SET_MM_ENV_START:
2155 case PR_SET_MM_ENV_END:
2156 if (!vma) {
2157 error = -EFAULT;
2158 goto out;
2159 }
2160 }
2161
2162 mm->start_code = prctl_map.start_code;
2163 mm->end_code = prctl_map.end_code;
2164 mm->start_data = prctl_map.start_data;
2165 mm->end_data = prctl_map.end_data;
2166 mm->start_brk = prctl_map.start_brk;
2167 mm->brk = prctl_map.brk;
2168 mm->start_stack = prctl_map.start_stack;
2169 mm->arg_start = prctl_map.arg_start;
2170 mm->arg_end = prctl_map.arg_end;
2171 mm->env_start = prctl_map.env_start;
2172 mm->env_end = prctl_map.env_end;
2173
2174 error = 0;
2175 out:
2176 up_write(&mm->mmap_sem);
2177 return error;
2178 }
2179
2180 #ifdef CONFIG_CHECKPOINT_RESTORE
2181 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2182 {
2183 return put_user(me->clear_child_tid, tid_addr);
2184 }
2185 #else
2186 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2187 {
2188 return -EINVAL;
2189 }
2190 #endif
2191
2192 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2193 {
2194 /*
2195 * If task has has_child_subreaper - all its decendants
2196 * already have these flag too and new decendants will
2197 * inherit it on fork, skip them.
2198 *
2199 * If we've found child_reaper - skip descendants in
2200 * it's subtree as they will never get out pidns.
2201 */
2202 if (p->signal->has_child_subreaper ||
2203 is_child_reaper(task_pid(p)))
2204 return 0;
2205
2206 p->signal->has_child_subreaper = 1;
2207 return 1;
2208 }
2209
2210 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2211 {
2212 return -EINVAL;
2213 }
2214
2215 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2216 unsigned long ctrl)
2217 {
2218 return -EINVAL;
2219 }
2220
2221 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2222 unsigned long, arg4, unsigned long, arg5)
2223 {
2224 struct task_struct *me = current;
2225 unsigned char comm[sizeof(me->comm)];
2226 long error;
2227
2228 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2229 if (error != -ENOSYS)
2230 return error;
2231
2232 error = 0;
2233 switch (option) {
2234 case PR_SET_PDEATHSIG:
2235 if (!valid_signal(arg2)) {
2236 error = -EINVAL;
2237 break;
2238 }
2239 me->pdeath_signal = arg2;
2240 break;
2241 case PR_GET_PDEATHSIG:
2242 error = put_user(me->pdeath_signal, (int __user *)arg2);
2243 break;
2244 case PR_GET_DUMPABLE:
2245 error = get_dumpable(me->mm);
2246 break;
2247 case PR_SET_DUMPABLE:
2248 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2249 error = -EINVAL;
2250 break;
2251 }
2252 set_dumpable(me->mm, arg2);
2253 break;
2254
2255 case PR_SET_UNALIGN:
2256 error = SET_UNALIGN_CTL(me, arg2);
2257 break;
2258 case PR_GET_UNALIGN:
2259 error = GET_UNALIGN_CTL(me, arg2);
2260 break;
2261 case PR_SET_FPEMU:
2262 error = SET_FPEMU_CTL(me, arg2);
2263 break;
2264 case PR_GET_FPEMU:
2265 error = GET_FPEMU_CTL(me, arg2);
2266 break;
2267 case PR_SET_FPEXC:
2268 error = SET_FPEXC_CTL(me, arg2);
2269 break;
2270 case PR_GET_FPEXC:
2271 error = GET_FPEXC_CTL(me, arg2);
2272 break;
2273 case PR_GET_TIMING:
2274 error = PR_TIMING_STATISTICAL;
2275 break;
2276 case PR_SET_TIMING:
2277 if (arg2 != PR_TIMING_STATISTICAL)
2278 error = -EINVAL;
2279 break;
2280 case PR_SET_NAME:
2281 comm[sizeof(me->comm) - 1] = 0;
2282 if (strncpy_from_user(comm, (char __user *)arg2,
2283 sizeof(me->comm) - 1) < 0)
2284 return -EFAULT;
2285 set_task_comm(me, comm);
2286 proc_comm_connector(me);
2287 break;
2288 case PR_GET_NAME:
2289 get_task_comm(comm, me);
2290 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2291 return -EFAULT;
2292 break;
2293 case PR_GET_ENDIAN:
2294 error = GET_ENDIAN(me, arg2);
2295 break;
2296 case PR_SET_ENDIAN:
2297 error = SET_ENDIAN(me, arg2);
2298 break;
2299 case PR_GET_SECCOMP:
2300 error = prctl_get_seccomp();
2301 break;
2302 case PR_SET_SECCOMP:
2303 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2304 break;
2305 case PR_GET_TSC:
2306 error = GET_TSC_CTL(arg2);
2307 break;
2308 case PR_SET_TSC:
2309 error = SET_TSC_CTL(arg2);
2310 break;
2311 case PR_TASK_PERF_EVENTS_DISABLE:
2312 error = perf_event_task_disable();
2313 break;
2314 case PR_TASK_PERF_EVENTS_ENABLE:
2315 error = perf_event_task_enable();
2316 break;
2317 case PR_GET_TIMERSLACK:
2318 if (current->timer_slack_ns > ULONG_MAX)
2319 error = ULONG_MAX;
2320 else
2321 error = current->timer_slack_ns;
2322 break;
2323 case PR_SET_TIMERSLACK:
2324 if (arg2 <= 0)
2325 current->timer_slack_ns =
2326 current->default_timer_slack_ns;
2327 else
2328 current->timer_slack_ns = arg2;
2329 break;
2330 case PR_MCE_KILL:
2331 if (arg4 | arg5)
2332 return -EINVAL;
2333 switch (arg2) {
2334 case PR_MCE_KILL_CLEAR:
2335 if (arg3 != 0)
2336 return -EINVAL;
2337 current->flags &= ~PF_MCE_PROCESS;
2338 break;
2339 case PR_MCE_KILL_SET:
2340 current->flags |= PF_MCE_PROCESS;
2341 if (arg3 == PR_MCE_KILL_EARLY)
2342 current->flags |= PF_MCE_EARLY;
2343 else if (arg3 == PR_MCE_KILL_LATE)
2344 current->flags &= ~PF_MCE_EARLY;
2345 else if (arg3 == PR_MCE_KILL_DEFAULT)
2346 current->flags &=
2347 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2348 else
2349 return -EINVAL;
2350 break;
2351 default:
2352 return -EINVAL;
2353 }
2354 break;
2355 case PR_MCE_KILL_GET:
2356 if (arg2 | arg3 | arg4 | arg5)
2357 return -EINVAL;
2358 if (current->flags & PF_MCE_PROCESS)
2359 error = (current->flags & PF_MCE_EARLY) ?
2360 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2361 else
2362 error = PR_MCE_KILL_DEFAULT;
2363 break;
2364 case PR_SET_MM:
2365 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2366 break;
2367 case PR_GET_TID_ADDRESS:
2368 error = prctl_get_tid_address(me, (int __user **)arg2);
2369 break;
2370 case PR_SET_CHILD_SUBREAPER:
2371 me->signal->is_child_subreaper = !!arg2;
2372 if (!arg2)
2373 break;
2374
2375 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2376 break;
2377 case PR_GET_CHILD_SUBREAPER:
2378 error = put_user(me->signal->is_child_subreaper,
2379 (int __user *)arg2);
2380 break;
2381 case PR_SET_NO_NEW_PRIVS:
2382 if (arg2 != 1 || arg3 || arg4 || arg5)
2383 return -EINVAL;
2384
2385 task_set_no_new_privs(current);
2386 break;
2387 case PR_GET_NO_NEW_PRIVS:
2388 if (arg2 || arg3 || arg4 || arg5)
2389 return -EINVAL;
2390 return task_no_new_privs(current) ? 1 : 0;
2391 case PR_GET_THP_DISABLE:
2392 if (arg2 || arg3 || arg4 || arg5)
2393 return -EINVAL;
2394 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2395 break;
2396 case PR_SET_THP_DISABLE:
2397 if (arg3 || arg4 || arg5)
2398 return -EINVAL;
2399 if (down_write_killable(&me->mm->mmap_sem))
2400 return -EINTR;
2401 if (arg2)
2402 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2403 else
2404 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2405 up_write(&me->mm->mmap_sem);
2406 break;
2407 case PR_MPX_ENABLE_MANAGEMENT:
2408 if (arg2 || arg3 || arg4 || arg5)
2409 return -EINVAL;
2410 error = MPX_ENABLE_MANAGEMENT();
2411 break;
2412 case PR_MPX_DISABLE_MANAGEMENT:
2413 if (arg2 || arg3 || arg4 || arg5)
2414 return -EINVAL;
2415 error = MPX_DISABLE_MANAGEMENT();
2416 break;
2417 case PR_SET_FP_MODE:
2418 error = SET_FP_MODE(me, arg2);
2419 break;
2420 case PR_GET_FP_MODE:
2421 error = GET_FP_MODE(me);
2422 break;
2423 case PR_SVE_SET_VL:
2424 error = SVE_SET_VL(arg2);
2425 break;
2426 case PR_SVE_GET_VL:
2427 error = SVE_GET_VL();
2428 break;
2429 case PR_GET_SPECULATION_CTRL:
2430 if (arg3 || arg4 || arg5)
2431 return -EINVAL;
2432 error = arch_prctl_spec_ctrl_get(me, arg2);
2433 break;
2434 case PR_SET_SPECULATION_CTRL:
2435 if (arg4 || arg5)
2436 return -EINVAL;
2437 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2438 break;
2439 default:
2440 error = -EINVAL;
2441 break;
2442 }
2443 return error;
2444 }
2445
2446 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2447 struct getcpu_cache __user *, unused)
2448 {
2449 int err = 0;
2450 int cpu = raw_smp_processor_id();
2451
2452 if (cpup)
2453 err |= put_user(cpu, cpup);
2454 if (nodep)
2455 err |= put_user(cpu_to_node(cpu), nodep);
2456 return err ? -EFAULT : 0;
2457 }
2458
2459 /**
2460 * do_sysinfo - fill in sysinfo struct
2461 * @info: pointer to buffer to fill
2462 */
2463 static int do_sysinfo(struct sysinfo *info)
2464 {
2465 unsigned long mem_total, sav_total;
2466 unsigned int mem_unit, bitcount;
2467 struct timespec tp;
2468
2469 memset(info, 0, sizeof(struct sysinfo));
2470
2471 get_monotonic_boottime(&tp);
2472 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2473
2474 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2475
2476 info->procs = nr_threads;
2477
2478 si_meminfo(info);
2479 si_swapinfo(info);
2480
2481 /*
2482 * If the sum of all the available memory (i.e. ram + swap)
2483 * is less than can be stored in a 32 bit unsigned long then
2484 * we can be binary compatible with 2.2.x kernels. If not,
2485 * well, in that case 2.2.x was broken anyways...
2486 *
2487 * -Erik Andersen <andersee@debian.org>
2488 */
2489
2490 mem_total = info->totalram + info->totalswap;
2491 if (mem_total < info->totalram || mem_total < info->totalswap)
2492 goto out;
2493 bitcount = 0;
2494 mem_unit = info->mem_unit;
2495 while (mem_unit > 1) {
2496 bitcount++;
2497 mem_unit >>= 1;
2498 sav_total = mem_total;
2499 mem_total <<= 1;
2500 if (mem_total < sav_total)
2501 goto out;
2502 }
2503
2504 /*
2505 * If mem_total did not overflow, multiply all memory values by
2506 * info->mem_unit and set it to 1. This leaves things compatible
2507 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2508 * kernels...
2509 */
2510
2511 info->mem_unit = 1;
2512 info->totalram <<= bitcount;
2513 info->freeram <<= bitcount;
2514 info->sharedram <<= bitcount;
2515 info->bufferram <<= bitcount;
2516 info->totalswap <<= bitcount;
2517 info->freeswap <<= bitcount;
2518 info->totalhigh <<= bitcount;
2519 info->freehigh <<= bitcount;
2520
2521 out:
2522 return 0;
2523 }
2524
2525 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2526 {
2527 struct sysinfo val;
2528
2529 do_sysinfo(&val);
2530
2531 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2532 return -EFAULT;
2533
2534 return 0;
2535 }
2536
2537 #ifdef CONFIG_COMPAT
2538 struct compat_sysinfo {
2539 s32 uptime;
2540 u32 loads[3];
2541 u32 totalram;
2542 u32 freeram;
2543 u32 sharedram;
2544 u32 bufferram;
2545 u32 totalswap;
2546 u32 freeswap;
2547 u16 procs;
2548 u16 pad;
2549 u32 totalhigh;
2550 u32 freehigh;
2551 u32 mem_unit;
2552 char _f[20-2*sizeof(u32)-sizeof(int)];
2553 };
2554
2555 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2556 {
2557 struct sysinfo s;
2558
2559 do_sysinfo(&s);
2560
2561 /* Check to see if any memory value is too large for 32-bit and scale
2562 * down if needed
2563 */
2564 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2565 int bitcount = 0;
2566
2567 while (s.mem_unit < PAGE_SIZE) {
2568 s.mem_unit <<= 1;
2569 bitcount++;
2570 }
2571
2572 s.totalram >>= bitcount;
2573 s.freeram >>= bitcount;
2574 s.sharedram >>= bitcount;
2575 s.bufferram >>= bitcount;
2576 s.totalswap >>= bitcount;
2577 s.freeswap >>= bitcount;
2578 s.totalhigh >>= bitcount;
2579 s.freehigh >>= bitcount;
2580 }
2581
2582 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2583 __put_user(s.uptime, &info->uptime) ||
2584 __put_user(s.loads[0], &info->loads[0]) ||
2585 __put_user(s.loads[1], &info->loads[1]) ||
2586 __put_user(s.loads[2], &info->loads[2]) ||
2587 __put_user(s.totalram, &info->totalram) ||
2588 __put_user(s.freeram, &info->freeram) ||
2589 __put_user(s.sharedram, &info->sharedram) ||
2590 __put_user(s.bufferram, &info->bufferram) ||
2591 __put_user(s.totalswap, &info->totalswap) ||
2592 __put_user(s.freeswap, &info->freeswap) ||
2593 __put_user(s.procs, &info->procs) ||
2594 __put_user(s.totalhigh, &info->totalhigh) ||
2595 __put_user(s.freehigh, &info->freehigh) ||
2596 __put_user(s.mem_unit, &info->mem_unit))
2597 return -EFAULT;
2598
2599 return 0;
2600 }
2601 #endif /* CONFIG_COMPAT */
ubuntu-bionic-kernel mirror