?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
documentation: RCU grace-period memory ordering guarantees
[linux/fpc-iii.git] / kernel / sys.c
blob9aebc293501330ffdfef0b6ad53df155ae6b31ca
1 /*
2 * linux/kernel/sys.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 #include <linux/export.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
14 #include <linux/fs.h>
15 #include <linux/kmod.h>
16 #include <linux/perf_event.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/workqueue.h>
20 #include <linux/capability.h>
21 #include <linux/device.h>
22 #include <linux/key.h>
23 #include <linux/times.h>
24 #include <linux/posix-timers.h>
25 #include <linux/security.h>
26 #include <linux/dcookies.h>
27 #include <linux/suspend.h>
28 #include <linux/tty.h>
29 #include <linux/signal.h>
30 #include <linux/cn_proc.h>
31 #include <linux/getcpu.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/seccomp.h>
34 #include <linux/cpu.h>
35 #include <linux/personality.h>
36 #include <linux/ptrace.h>
37 #include <linux/fs_struct.h>
38 #include <linux/file.h>
39 #include <linux/mount.h>
40 #include <linux/gfp.h>
41 #include <linux/syscore_ops.h>
42 #include <linux/version.h>
43 #include <linux/ctype.h>
44
45 #include <linux/compat.h>
46 #include <linux/syscalls.h>
47 #include <linux/kprobes.h>
48 #include <linux/user_namespace.h>
49 #include <linux/binfmts.h>
50
51 #include <linux/sched.h>
52 #include <linux/sched/autogroup.h>
53 #include <linux/sched/loadavg.h>
54 #include <linux/sched/stat.h>
55 #include <linux/sched/mm.h>
56 #include <linux/sched/coredump.h>
57 #include <linux/sched/task.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/rcupdate.h>
60 #include <linux/uidgid.h>
61 #include <linux/cred.h>
62
63 #include <linux/kmsg_dump.h>
64 /* Move somewhere else to avoid recompiling? */
65 #include <generated/utsrelease.h>
66
67 #include <linux/uaccess.h>
68 #include <asm/io.h>
69 #include <asm/unistd.h>
70
71 #ifndef SET_UNALIGN_CTL
72 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
73 #endif
74 #ifndef GET_UNALIGN_CTL
75 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
76 #endif
77 #ifndef SET_FPEMU_CTL
78 # define SET_FPEMU_CTL(a, b) (-EINVAL)
79 #endif
80 #ifndef GET_FPEMU_CTL
81 # define GET_FPEMU_CTL(a, b) (-EINVAL)
82 #endif
83 #ifndef SET_FPEXC_CTL
84 # define SET_FPEXC_CTL(a, b) (-EINVAL)
85 #endif
86 #ifndef GET_FPEXC_CTL
87 # define GET_FPEXC_CTL(a, b) (-EINVAL)
88 #endif
89 #ifndef GET_ENDIAN
90 # define GET_ENDIAN(a, b) (-EINVAL)
91 #endif
92 #ifndef SET_ENDIAN
93 # define SET_ENDIAN(a, b) (-EINVAL)
94 #endif
95 #ifndef GET_TSC_CTL
96 # define GET_TSC_CTL(a) (-EINVAL)
97 #endif
98 #ifndef SET_TSC_CTL
99 # define SET_TSC_CTL(a) (-EINVAL)
100 #endif
101 #ifndef MPX_ENABLE_MANAGEMENT
102 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
103 #endif
104 #ifndef MPX_DISABLE_MANAGEMENT
105 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
106 #endif
107 #ifndef GET_FP_MODE
108 # define GET_FP_MODE(a) (-EINVAL)
109 #endif
110 #ifndef SET_FP_MODE
111 # define SET_FP_MODE(a,b) (-EINVAL)
112 #endif
113
114 /*
115 * this is where the system-wide overflow UID and GID are defined, for
116 * architectures that now have 32-bit UID/GID but didn't in the past
117 */
118
119 int overflowuid = DEFAULT_OVERFLOWUID;
120 int overflowgid = DEFAULT_OVERFLOWGID;
121
122 EXPORT_SYMBOL(overflowuid);
123 EXPORT_SYMBOL(overflowgid);
124
125 /*
126 * the same as above, but for filesystems which can only store a 16-bit
127 * UID and GID. as such, this is needed on all architectures
128 */
129
130 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
131 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
132
133 EXPORT_SYMBOL(fs_overflowuid);
134 EXPORT_SYMBOL(fs_overflowgid);
135
136 /*
137 * Returns true if current's euid is same as p's uid or euid,
138 * or has CAP_SYS_NICE to p's user_ns.
139 *
140 * Called with rcu_read_lock, creds are safe
141 */
142 static bool set_one_prio_perm(struct task_struct *p)
143 {
144 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
145
146 if (uid_eq(pcred->uid, cred->euid) ||
147 uid_eq(pcred->euid, cred->euid))
148 return true;
149 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
150 return true;
151 return false;
152 }
153
154 /*
155 * set the priority of a task
156 * - the caller must hold the RCU read lock
157 */
158 static int set_one_prio(struct task_struct *p, int niceval, int error)
159 {
160 int no_nice;
161
162 if (!set_one_prio_perm(p)) {
163 error = -EPERM;
164 goto out;
165 }
166 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
167 error = -EACCES;
168 goto out;
169 }
170 no_nice = security_task_setnice(p, niceval);
171 if (no_nice) {
172 error = no_nice;
173 goto out;
174 }
175 if (error == -ESRCH)
176 error = 0;
177 set_user_nice(p, niceval);
178 out:
179 return error;
180 }
181
182 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
183 {
184 struct task_struct *g, *p;
185 struct user_struct *user;
186 const struct cred *cred = current_cred();
187 int error = -EINVAL;
188 struct pid *pgrp;
189 kuid_t uid;
190
191 if (which > PRIO_USER || which < PRIO_PROCESS)
192 goto out;
193
194 /* normalize: avoid signed division (rounding problems) */
195 error = -ESRCH;
196 if (niceval < MIN_NICE)
197 niceval = MIN_NICE;
198 if (niceval > MAX_NICE)
199 niceval = MAX_NICE;
200
201 rcu_read_lock();
202 read_lock(&tasklist_lock);
203 switch (which) {
204 case PRIO_PROCESS:
205 if (who)
206 p = find_task_by_vpid(who);
207 else
208 p = current;
209 if (p)
210 error = set_one_prio(p, niceval, error);
211 break;
212 case PRIO_PGRP:
213 if (who)
214 pgrp = find_vpid(who);
215 else
216 pgrp = task_pgrp(current);
217 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
218 error = set_one_prio(p, niceval, error);
219 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
220 break;
221 case PRIO_USER:
222 uid = make_kuid(cred->user_ns, who);
223 user = cred->user;
224 if (!who)
225 uid = cred->uid;
226 else if (!uid_eq(uid, cred->uid)) {
227 user = find_user(uid);
228 if (!user)
229 goto out_unlock; /* No processes for this user */
230 }
231 do_each_thread(g, p) {
232 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
233 error = set_one_prio(p, niceval, error);
234 } while_each_thread(g, p);
235 if (!uid_eq(uid, cred->uid))
236 free_uid(user); /* For find_user() */
237 break;
238 }
239 out_unlock:
240 read_unlock(&tasklist_lock);
241 rcu_read_unlock();
242 out:
243 return error;
244 }
245
246 /*
247 * Ugh. To avoid negative return values, "getpriority()" will
248 * not return the normal nice-value, but a negated value that
249 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
250 * to stay compatible.
251 */
252 SYSCALL_DEFINE2(getpriority, int, which, int, who)
253 {
254 struct task_struct *g, *p;
255 struct user_struct *user;
256 const struct cred *cred = current_cred();
257 long niceval, retval = -ESRCH;
258 struct pid *pgrp;
259 kuid_t uid;
260
261 if (which > PRIO_USER || which < PRIO_PROCESS)
262 return -EINVAL;
263
264 rcu_read_lock();
265 read_lock(&tasklist_lock);
266 switch (which) {
267 case PRIO_PROCESS:
268 if (who)
269 p = find_task_by_vpid(who);
270 else
271 p = current;
272 if (p) {
273 niceval = nice_to_rlimit(task_nice(p));
274 if (niceval > retval)
275 retval = niceval;
276 }
277 break;
278 case PRIO_PGRP:
279 if (who)
280 pgrp = find_vpid(who);
281 else
282 pgrp = task_pgrp(current);
283 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
284 niceval = nice_to_rlimit(task_nice(p));
285 if (niceval > retval)
286 retval = niceval;
287 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
288 break;
289 case PRIO_USER:
290 uid = make_kuid(cred->user_ns, who);
291 user = cred->user;
292 if (!who)
293 uid = cred->uid;
294 else if (!uid_eq(uid, cred->uid)) {
295 user = find_user(uid);
296 if (!user)
297 goto out_unlock; /* No processes for this user */
298 }
299 do_each_thread(g, p) {
300 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
301 niceval = nice_to_rlimit(task_nice(p));
302 if (niceval > retval)
303 retval = niceval;
304 }
305 } while_each_thread(g, p);
306 if (!uid_eq(uid, cred->uid))
307 free_uid(user); /* for find_user() */
308 break;
309 }
310 out_unlock:
311 read_unlock(&tasklist_lock);
312 rcu_read_unlock();
313
314 return retval;
315 }
316
317 /*
318 * Unprivileged users may change the real gid to the effective gid
319 * or vice versa. (BSD-style)
320 *
321 * If you set the real gid at all, or set the effective gid to a value not
322 * equal to the real gid, then the saved gid is set to the new effective gid.
323 *
324 * This makes it possible for a setgid program to completely drop its
325 * privileges, which is often a useful assertion to make when you are doing
326 * a security audit over a program.
327 *
328 * The general idea is that a program which uses just setregid() will be
329 * 100% compatible with BSD. A program which uses just setgid() will be
330 * 100% compatible with POSIX with saved IDs.
331 *
332 * SMP: There are not races, the GIDs are checked only by filesystem
333 * operations (as far as semantic preservation is concerned).
334 */
335 #ifdef CONFIG_MULTIUSER
336 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
337 {
338 struct user_namespace *ns = current_user_ns();
339 const struct cred *old;
340 struct cred *new;
341 int retval;
342 kgid_t krgid, kegid;
343
344 krgid = make_kgid(ns, rgid);
345 kegid = make_kgid(ns, egid);
346
347 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
348 return -EINVAL;
349 if ((egid != (gid_t) -1) && !gid_valid(kegid))
350 return -EINVAL;
351
352 new = prepare_creds();
353 if (!new)
354 return -ENOMEM;
355 old = current_cred();
356
357 retval = -EPERM;
358 if (rgid != (gid_t) -1) {
359 if (gid_eq(old->gid, krgid) ||
360 gid_eq(old->egid, krgid) ||
361 ns_capable(old->user_ns, CAP_SETGID))
362 new->gid = krgid;
363 else
364 goto error;
365 }
366 if (egid != (gid_t) -1) {
367 if (gid_eq(old->gid, kegid) ||
368 gid_eq(old->egid, kegid) ||
369 gid_eq(old->sgid, kegid) ||
370 ns_capable(old->user_ns, CAP_SETGID))
371 new->egid = kegid;
372 else
373 goto error;
374 }
375
376 if (rgid != (gid_t) -1 ||
377 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
378 new->sgid = new->egid;
379 new->fsgid = new->egid;
380
381 return commit_creds(new);
382
383 error:
384 abort_creds(new);
385 return retval;
386 }
387
388 /*
389 * setgid() is implemented like SysV w/ SAVED_IDS
390 *
391 * SMP: Same implicit races as above.
392 */
393 SYSCALL_DEFINE1(setgid, gid_t, gid)
394 {
395 struct user_namespace *ns = current_user_ns();
396 const struct cred *old;
397 struct cred *new;
398 int retval;
399 kgid_t kgid;
400
401 kgid = make_kgid(ns, gid);
402 if (!gid_valid(kgid))
403 return -EINVAL;
404
405 new = prepare_creds();
406 if (!new)
407 return -ENOMEM;
408 old = current_cred();
409
410 retval = -EPERM;
411 if (ns_capable(old->user_ns, CAP_SETGID))
412 new->gid = new->egid = new->sgid = new->fsgid = kgid;
413 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
414 new->egid = new->fsgid = kgid;
415 else
416 goto error;
417
418 return commit_creds(new);
419
420 error:
421 abort_creds(new);
422 return retval;
423 }
424
425 /*
426 * change the user struct in a credentials set to match the new UID
427 */
428 static int set_user(struct cred *new)
429 {
430 struct user_struct *new_user;
431
432 new_user = alloc_uid(new->uid);
433 if (!new_user)
434 return -EAGAIN;
435
436 /*
437 * We don't fail in case of NPROC limit excess here because too many
438 * poorly written programs don't check set*uid() return code, assuming
439 * it never fails if called by root. We may still enforce NPROC limit
440 * for programs doing set*uid()+execve() by harmlessly deferring the
441 * failure to the execve() stage.
442 */
443 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
444 new_user != INIT_USER)
445 current->flags |= PF_NPROC_EXCEEDED;
446 else
447 current->flags &= ~PF_NPROC_EXCEEDED;
448
449 free_uid(new->user);
450 new->user = new_user;
451 return 0;
452 }
453
454 /*
455 * Unprivileged users may change the real uid to the effective uid
456 * or vice versa. (BSD-style)
457 *
458 * If you set the real uid at all, or set the effective uid to a value not
459 * equal to the real uid, then the saved uid is set to the new effective uid.
460 *
461 * This makes it possible for a setuid program to completely drop its
462 * privileges, which is often a useful assertion to make when you are doing
463 * a security audit over a program.
464 *
465 * The general idea is that a program which uses just setreuid() will be
466 * 100% compatible with BSD. A program which uses just setuid() will be
467 * 100% compatible with POSIX with saved IDs.
468 */
469 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
470 {
471 struct user_namespace *ns = current_user_ns();
472 const struct cred *old;
473 struct cred *new;
474 int retval;
475 kuid_t kruid, keuid;
476
477 kruid = make_kuid(ns, ruid);
478 keuid = make_kuid(ns, euid);
479
480 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
481 return -EINVAL;
482 if ((euid != (uid_t) -1) && !uid_valid(keuid))
483 return -EINVAL;
484
485 new = prepare_creds();
486 if (!new)
487 return -ENOMEM;
488 old = current_cred();
489
490 retval = -EPERM;
491 if (ruid != (uid_t) -1) {
492 new->uid = kruid;
493 if (!uid_eq(old->uid, kruid) &&
494 !uid_eq(old->euid, kruid) &&
495 !ns_capable(old->user_ns, CAP_SETUID))
496 goto error;
497 }
498
499 if (euid != (uid_t) -1) {
500 new->euid = keuid;
501 if (!uid_eq(old->uid, keuid) &&
502 !uid_eq(old->euid, keuid) &&
503 !uid_eq(old->suid, keuid) &&
504 !ns_capable(old->user_ns, CAP_SETUID))
505 goto error;
506 }
507
508 if (!uid_eq(new->uid, old->uid)) {
509 retval = set_user(new);
510 if (retval < 0)
511 goto error;
512 }
513 if (ruid != (uid_t) -1 ||
514 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
515 new->suid = new->euid;
516 new->fsuid = new->euid;
517
518 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
519 if (retval < 0)
520 goto error;
521
522 return commit_creds(new);
523
524 error:
525 abort_creds(new);
526 return retval;
527 }
528
529 /*
530 * setuid() is implemented like SysV with SAVED_IDS
531 *
532 * Note that SAVED_ID's is deficient in that a setuid root program
533 * like sendmail, for example, cannot set its uid to be a normal
534 * user and then switch back, because if you're root, setuid() sets
535 * the saved uid too. If you don't like this, blame the bright people
536 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
537 * will allow a root program to temporarily drop privileges and be able to
538 * regain them by swapping the real and effective uid.
539 */
540 SYSCALL_DEFINE1(setuid, uid_t, uid)
541 {
542 struct user_namespace *ns = current_user_ns();
543 const struct cred *old;
544 struct cred *new;
545 int retval;
546 kuid_t kuid;
547
548 kuid = make_kuid(ns, uid);
549 if (!uid_valid(kuid))
550 return -EINVAL;
551
552 new = prepare_creds();
553 if (!new)
554 return -ENOMEM;
555 old = current_cred();
556
557 retval = -EPERM;
558 if (ns_capable(old->user_ns, CAP_SETUID)) {
559 new->suid = new->uid = kuid;
560 if (!uid_eq(kuid, old->uid)) {
561 retval = set_user(new);
562 if (retval < 0)
563 goto error;
564 }
565 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
566 goto error;
567 }
568
569 new->fsuid = new->euid = kuid;
570
571 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
572 if (retval < 0)
573 goto error;
574
575 return commit_creds(new);
576
577 error:
578 abort_creds(new);
579 return retval;
580 }
581
582
583 /*
584 * This function implements a generic ability to update ruid, euid,
585 * and suid. This allows you to implement the 4.4 compatible seteuid().
586 */
587 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
588 {
589 struct user_namespace *ns = current_user_ns();
590 const struct cred *old;
591 struct cred *new;
592 int retval;
593 kuid_t kruid, keuid, ksuid;
594
595 kruid = make_kuid(ns, ruid);
596 keuid = make_kuid(ns, euid);
597 ksuid = make_kuid(ns, suid);
598
599 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
600 return -EINVAL;
601
602 if ((euid != (uid_t) -1) && !uid_valid(keuid))
603 return -EINVAL;
604
605 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
606 return -EINVAL;
607
608 new = prepare_creds();
609 if (!new)
610 return -ENOMEM;
611
612 old = current_cred();
613
614 retval = -EPERM;
615 if (!ns_capable(old->user_ns, CAP_SETUID)) {
616 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
617 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
618 goto error;
619 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
620 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
621 goto error;
622 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
623 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
624 goto error;
625 }
626
627 if (ruid != (uid_t) -1) {
628 new->uid = kruid;
629 if (!uid_eq(kruid, old->uid)) {
630 retval = set_user(new);
631 if (retval < 0)
632 goto error;
633 }
634 }
635 if (euid != (uid_t) -1)
636 new->euid = keuid;
637 if (suid != (uid_t) -1)
638 new->suid = ksuid;
639 new->fsuid = new->euid;
640
641 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
642 if (retval < 0)
643 goto error;
644
645 return commit_creds(new);
646
647 error:
648 abort_creds(new);
649 return retval;
650 }
651
652 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
653 {
654 const struct cred *cred = current_cred();
655 int retval;
656 uid_t ruid, euid, suid;
657
658 ruid = from_kuid_munged(cred->user_ns, cred->uid);
659 euid = from_kuid_munged(cred->user_ns, cred->euid);
660 suid = from_kuid_munged(cred->user_ns, cred->suid);
661
662 retval = put_user(ruid, ruidp);
663 if (!retval) {
664 retval = put_user(euid, euidp);
665 if (!retval)
666 return put_user(suid, suidp);
667 }
668 return retval;
669 }
670
671 /*
672 * Same as above, but for rgid, egid, sgid.
673 */
674 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
675 {
676 struct user_namespace *ns = current_user_ns();
677 const struct cred *old;
678 struct cred *new;
679 int retval;
680 kgid_t krgid, kegid, ksgid;
681
682 krgid = make_kgid(ns, rgid);
683 kegid = make_kgid(ns, egid);
684 ksgid = make_kgid(ns, sgid);
685
686 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
687 return -EINVAL;
688 if ((egid != (gid_t) -1) && !gid_valid(kegid))
689 return -EINVAL;
690 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
691 return -EINVAL;
692
693 new = prepare_creds();
694 if (!new)
695 return -ENOMEM;
696 old = current_cred();
697
698 retval = -EPERM;
699 if (!ns_capable(old->user_ns, CAP_SETGID)) {
700 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
701 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
702 goto error;
703 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
704 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
705 goto error;
706 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
707 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
708 goto error;
709 }
710
711 if (rgid != (gid_t) -1)
712 new->gid = krgid;
713 if (egid != (gid_t) -1)
714 new->egid = kegid;
715 if (sgid != (gid_t) -1)
716 new->sgid = ksgid;
717 new->fsgid = new->egid;
718
719 return commit_creds(new);
720
721 error:
722 abort_creds(new);
723 return retval;
724 }
725
726 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
727 {
728 const struct cred *cred = current_cred();
729 int retval;
730 gid_t rgid, egid, sgid;
731
732 rgid = from_kgid_munged(cred->user_ns, cred->gid);
733 egid = from_kgid_munged(cred->user_ns, cred->egid);
734 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
735
736 retval = put_user(rgid, rgidp);
737 if (!retval) {
738 retval = put_user(egid, egidp);
739 if (!retval)
740 retval = put_user(sgid, sgidp);
741 }
742
743 return retval;
744 }
745
746
747 /*
748 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
749 * is used for "access()" and for the NFS daemon (letting nfsd stay at
750 * whatever uid it wants to). It normally shadows "euid", except when
751 * explicitly set by setfsuid() or for access..
752 */
753 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
754 {
755 const struct cred *old;
756 struct cred *new;
757 uid_t old_fsuid;
758 kuid_t kuid;
759
760 old = current_cred();
761 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
762
763 kuid = make_kuid(old->user_ns, uid);
764 if (!uid_valid(kuid))
765 return old_fsuid;
766
767 new = prepare_creds();
768 if (!new)
769 return old_fsuid;
770
771 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
772 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
773 ns_capable(old->user_ns, CAP_SETUID)) {
774 if (!uid_eq(kuid, old->fsuid)) {
775 new->fsuid = kuid;
776 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
777 goto change_okay;
778 }
779 }
780
781 abort_creds(new);
782 return old_fsuid;
783
784 change_okay:
785 commit_creds(new);
786 return old_fsuid;
787 }
788
789 /*
790 * Samma på svenska..
791 */
792 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
793 {
794 const struct cred *old;
795 struct cred *new;
796 gid_t old_fsgid;
797 kgid_t kgid;
798
799 old = current_cred();
800 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
801
802 kgid = make_kgid(old->user_ns, gid);
803 if (!gid_valid(kgid))
804 return old_fsgid;
805
806 new = prepare_creds();
807 if (!new)
808 return old_fsgid;
809
810 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
811 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
812 ns_capable(old->user_ns, CAP_SETGID)) {
813 if (!gid_eq(kgid, old->fsgid)) {
814 new->fsgid = kgid;
815 goto change_okay;
816 }
817 }
818
819 abort_creds(new);
820 return old_fsgid;
821
822 change_okay:
823 commit_creds(new);
824 return old_fsgid;
825 }
826 #endif /* CONFIG_MULTIUSER */
827
828 /**
829 * sys_getpid - return the thread group id of the current process
830 *
831 * Note, despite the name, this returns the tgid not the pid. The tgid and
832 * the pid are identical unless CLONE_THREAD was specified on clone() in
833 * which case the tgid is the same in all threads of the same group.
834 *
835 * This is SMP safe as current->tgid does not change.
836 */
837 SYSCALL_DEFINE0(getpid)
838 {
839 return task_tgid_vnr(current);
840 }
841
842 /* Thread ID - the internal kernel "pid" */
843 SYSCALL_DEFINE0(gettid)
844 {
845 return task_pid_vnr(current);
846 }
847
848 /*
849 * Accessing ->real_parent is not SMP-safe, it could
850 * change from under us. However, we can use a stale
851 * value of ->real_parent under rcu_read_lock(), see
852 * release_task()->call_rcu(delayed_put_task_struct).
853 */
854 SYSCALL_DEFINE0(getppid)
855 {
856 int pid;
857
858 rcu_read_lock();
859 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
860 rcu_read_unlock();
861
862 return pid;
863 }
864
865 SYSCALL_DEFINE0(getuid)
866 {
867 /* Only we change this so SMP safe */
868 return from_kuid_munged(current_user_ns(), current_uid());
869 }
870
871 SYSCALL_DEFINE0(geteuid)
872 {
873 /* Only we change this so SMP safe */
874 return from_kuid_munged(current_user_ns(), current_euid());
875 }
876
877 SYSCALL_DEFINE0(getgid)
878 {
879 /* Only we change this so SMP safe */
880 return from_kgid_munged(current_user_ns(), current_gid());
881 }
882
883 SYSCALL_DEFINE0(getegid)
884 {
885 /* Only we change this so SMP safe */
886 return from_kgid_munged(current_user_ns(), current_egid());
887 }
888
889 static void do_sys_times(struct tms *tms)
890 {
891 u64 tgutime, tgstime, cutime, cstime;
892
893 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
894 cutime = current->signal->cutime;
895 cstime = current->signal->cstime;
896 tms->tms_utime = nsec_to_clock_t(tgutime);
897 tms->tms_stime = nsec_to_clock_t(tgstime);
898 tms->tms_cutime = nsec_to_clock_t(cutime);
899 tms->tms_cstime = nsec_to_clock_t(cstime);
900 }
901
902 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
903 {
904 if (tbuf) {
905 struct tms tmp;
906
907 do_sys_times(&tmp);
908 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
909 return -EFAULT;
910 }
911 force_successful_syscall_return();
912 return (long) jiffies_64_to_clock_t(get_jiffies_64());
913 }
914
915 #ifdef CONFIG_COMPAT
916 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
917 {
918 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
919 }
920
921 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
922 {
923 if (tbuf) {
924 struct tms tms;
925 struct compat_tms tmp;
926
927 do_sys_times(&tms);
928 /* Convert our struct tms to the compat version. */
929 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
930 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
931 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
932 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
933 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
934 return -EFAULT;
935 }
936 force_successful_syscall_return();
937 return compat_jiffies_to_clock_t(jiffies);
938 }
939 #endif
940
941 /*
942 * This needs some heavy checking ...
943 * I just haven't the stomach for it. I also don't fully
944 * understand sessions/pgrp etc. Let somebody who does explain it.
945 *
946 * OK, I think I have the protection semantics right.... this is really
947 * only important on a multi-user system anyway, to make sure one user
948 * can't send a signal to a process owned by another. -TYT, 12/12/91
949 *
950 * !PF_FORKNOEXEC check to conform completely to POSIX.
951 */
952 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
953 {
954 struct task_struct *p;
955 struct task_struct *group_leader = current->group_leader;
956 struct pid *pgrp;
957 int err;
958
959 if (!pid)
960 pid = task_pid_vnr(group_leader);
961 if (!pgid)
962 pgid = pid;
963 if (pgid < 0)
964 return -EINVAL;
965 rcu_read_lock();
966
967 /* From this point forward we keep holding onto the tasklist lock
968 * so that our parent does not change from under us. -DaveM
969 */
970 write_lock_irq(&tasklist_lock);
971
972 err = -ESRCH;
973 p = find_task_by_vpid(pid);
974 if (!p)
975 goto out;
976
977 err = -EINVAL;
978 if (!thread_group_leader(p))
979 goto out;
980
981 if (same_thread_group(p->real_parent, group_leader)) {
982 err = -EPERM;
983 if (task_session(p) != task_session(group_leader))
984 goto out;
985 err = -EACCES;
986 if (!(p->flags & PF_FORKNOEXEC))
987 goto out;
988 } else {
989 err = -ESRCH;
990 if (p != group_leader)
991 goto out;
992 }
993
994 err = -EPERM;
995 if (p->signal->leader)
996 goto out;
997
998 pgrp = task_pid(p);
999 if (pgid != pid) {
1000 struct task_struct *g;
1001
1002 pgrp = find_vpid(pgid);
1003 g = pid_task(pgrp, PIDTYPE_PGID);
1004 if (!g || task_session(g) != task_session(group_leader))
1005 goto out;
1006 }
1007
1008 err = security_task_setpgid(p, pgid);
1009 if (err)
1010 goto out;
1011
1012 if (task_pgrp(p) != pgrp)
1013 change_pid(p, PIDTYPE_PGID, pgrp);
1014
1015 err = 0;
1016 out:
1017 /* All paths lead to here, thus we are safe. -DaveM */
1018 write_unlock_irq(&tasklist_lock);
1019 rcu_read_unlock();
1020 return err;
1021 }
1022
1023 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1024 {
1025 struct task_struct *p;
1026 struct pid *grp;
1027 int retval;
1028
1029 rcu_read_lock();
1030 if (!pid)
1031 grp = task_pgrp(current);
1032 else {
1033 retval = -ESRCH;
1034 p = find_task_by_vpid(pid);
1035 if (!p)
1036 goto out;
1037 grp = task_pgrp(p);
1038 if (!grp)
1039 goto out;
1040
1041 retval = security_task_getpgid(p);
1042 if (retval)
1043 goto out;
1044 }
1045 retval = pid_vnr(grp);
1046 out:
1047 rcu_read_unlock();
1048 return retval;
1049 }
1050
1051 #ifdef __ARCH_WANT_SYS_GETPGRP
1052
1053 SYSCALL_DEFINE0(getpgrp)
1054 {
1055 return sys_getpgid(0);
1056 }
1057
1058 #endif
1059
1060 SYSCALL_DEFINE1(getsid, pid_t, pid)
1061 {
1062 struct task_struct *p;
1063 struct pid *sid;
1064 int retval;
1065
1066 rcu_read_lock();
1067 if (!pid)
1068 sid = task_session(current);
1069 else {
1070 retval = -ESRCH;
1071 p = find_task_by_vpid(pid);
1072 if (!p)
1073 goto out;
1074 sid = task_session(p);
1075 if (!sid)
1076 goto out;
1077
1078 retval = security_task_getsid(p);
1079 if (retval)
1080 goto out;
1081 }
1082 retval = pid_vnr(sid);
1083 out:
1084 rcu_read_unlock();
1085 return retval;
1086 }
1087
1088 static void set_special_pids(struct pid *pid)
1089 {
1090 struct task_struct *curr = current->group_leader;
1091
1092 if (task_session(curr) != pid)
1093 change_pid(curr, PIDTYPE_SID, pid);
1094
1095 if (task_pgrp(curr) != pid)
1096 change_pid(curr, PIDTYPE_PGID, pid);
1097 }
1098
1099 SYSCALL_DEFINE0(setsid)
1100 {
1101 struct task_struct *group_leader = current->group_leader;
1102 struct pid *sid = task_pid(group_leader);
1103 pid_t session = pid_vnr(sid);
1104 int err = -EPERM;
1105
1106 write_lock_irq(&tasklist_lock);
1107 /* Fail if I am already a session leader */
1108 if (group_leader->signal->leader)
1109 goto out;
1110
1111 /* Fail if a process group id already exists that equals the
1112 * proposed session id.
1113 */
1114 if (pid_task(sid, PIDTYPE_PGID))
1115 goto out;
1116
1117 group_leader->signal->leader = 1;
1118 set_special_pids(sid);
1119
1120 proc_clear_tty(group_leader);
1121
1122 err = session;
1123 out:
1124 write_unlock_irq(&tasklist_lock);
1125 if (err > 0) {
1126 proc_sid_connector(group_leader);
1127 sched_autogroup_create_attach(group_leader);
1128 }
1129 return err;
1130 }
1131
1132 DECLARE_RWSEM(uts_sem);
1133
1134 #ifdef COMPAT_UTS_MACHINE
1135 #define override_architecture(name) \
1136 (personality(current->personality) == PER_LINUX32 && \
1137 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1138 sizeof(COMPAT_UTS_MACHINE)))
1139 #else
1140 #define override_architecture(name) 0
1141 #endif
1142
1143 /*
1144 * Work around broken programs that cannot handle "Linux 3.0".
1145 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1146 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1147 */
1148 static int override_release(char __user *release, size_t len)
1149 {
1150 int ret = 0;
1151
1152 if (current->personality & UNAME26) {
1153 const char *rest = UTS_RELEASE;
1154 char buf[65] = { 0 };
1155 int ndots = 0;
1156 unsigned v;
1157 size_t copy;
1158
1159 while (*rest) {
1160 if (*rest == '.' && ++ndots >= 3)
1161 break;
1162 if (!isdigit(*rest) && *rest != '.')
1163 break;
1164 rest++;
1165 }
1166 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1167 copy = clamp_t(size_t, len, 1, sizeof(buf));
1168 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1169 ret = copy_to_user(release, buf, copy + 1);
1170 }
1171 return ret;
1172 }
1173
1174 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1175 {
1176 int errno = 0;
1177
1178 down_read(&uts_sem);
1179 if (copy_to_user(name, utsname(), sizeof *name))
1180 errno = -EFAULT;
1181 up_read(&uts_sem);
1182
1183 if (!errno && override_release(name->release, sizeof(name->release)))
1184 errno = -EFAULT;
1185 if (!errno && override_architecture(name))
1186 errno = -EFAULT;
1187 return errno;
1188 }
1189
1190 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1191 /*
1192 * Old cruft
1193 */
1194 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1195 {
1196 int error = 0;
1197
1198 if (!name)
1199 return -EFAULT;
1200
1201 down_read(&uts_sem);
1202 if (copy_to_user(name, utsname(), sizeof(*name)))
1203 error = -EFAULT;
1204 up_read(&uts_sem);
1205
1206 if (!error && override_release(name->release, sizeof(name->release)))
1207 error = -EFAULT;
1208 if (!error && override_architecture(name))
1209 error = -EFAULT;
1210 return error;
1211 }
1212
1213 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1214 {
1215 int error;
1216
1217 if (!name)
1218 return -EFAULT;
1219 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1220 return -EFAULT;
1221
1222 down_read(&uts_sem);
1223 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1224 __OLD_UTS_LEN);
1225 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1226 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1227 __OLD_UTS_LEN);
1228 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1229 error |= __copy_to_user(&name->release, &utsname()->release,
1230 __OLD_UTS_LEN);
1231 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1232 error |= __copy_to_user(&name->version, &utsname()->version,
1233 __OLD_UTS_LEN);
1234 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1235 error |= __copy_to_user(&name->machine, &utsname()->machine,
1236 __OLD_UTS_LEN);
1237 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1238 up_read(&uts_sem);
1239
1240 if (!error && override_architecture(name))
1241 error = -EFAULT;
1242 if (!error && override_release(name->release, sizeof(name->release)))
1243 error = -EFAULT;
1244 return error ? -EFAULT : 0;
1245 }
1246 #endif
1247
1248 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1249 {
1250 int errno;
1251 char tmp[__NEW_UTS_LEN];
1252
1253 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1254 return -EPERM;
1255
1256 if (len < 0 || len > __NEW_UTS_LEN)
1257 return -EINVAL;
1258 down_write(&uts_sem);
1259 errno = -EFAULT;
1260 if (!copy_from_user(tmp, name, len)) {
1261 struct new_utsname *u = utsname();
1262
1263 memcpy(u->nodename, tmp, len);
1264 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1265 errno = 0;
1266 uts_proc_notify(UTS_PROC_HOSTNAME);
1267 }
1268 up_write(&uts_sem);
1269 return errno;
1270 }
1271
1272 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1273
1274 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1275 {
1276 int i, errno;
1277 struct new_utsname *u;
1278
1279 if (len < 0)
1280 return -EINVAL;
1281 down_read(&uts_sem);
1282 u = utsname();
1283 i = 1 + strlen(u->nodename);
1284 if (i > len)
1285 i = len;
1286 errno = 0;
1287 if (copy_to_user(name, u->nodename, i))
1288 errno = -EFAULT;
1289 up_read(&uts_sem);
1290 return errno;
1291 }
1292
1293 #endif
1294
1295 /*
1296 * Only setdomainname; getdomainname can be implemented by calling
1297 * uname()
1298 */
1299 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1300 {
1301 int errno;
1302 char tmp[__NEW_UTS_LEN];
1303
1304 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1305 return -EPERM;
1306 if (len < 0 || len > __NEW_UTS_LEN)
1307 return -EINVAL;
1308
1309 down_write(&uts_sem);
1310 errno = -EFAULT;
1311 if (!copy_from_user(tmp, name, len)) {
1312 struct new_utsname *u = utsname();
1313
1314 memcpy(u->domainname, tmp, len);
1315 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1316 errno = 0;
1317 uts_proc_notify(UTS_PROC_DOMAINNAME);
1318 }
1319 up_write(&uts_sem);
1320 return errno;
1321 }
1322
1323 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1324 {
1325 struct rlimit value;
1326 int ret;
1327
1328 ret = do_prlimit(current, resource, NULL, &value);
1329 if (!ret)
1330 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1331
1332 return ret;
1333 }
1334
1335 #ifdef CONFIG_COMPAT
1336
1337 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1338 struct compat_rlimit __user *, rlim)
1339 {
1340 struct rlimit r;
1341 struct compat_rlimit r32;
1342
1343 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1344 return -EFAULT;
1345
1346 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1347 r.rlim_cur = RLIM_INFINITY;
1348 else
1349 r.rlim_cur = r32.rlim_cur;
1350 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1351 r.rlim_max = RLIM_INFINITY;
1352 else
1353 r.rlim_max = r32.rlim_max;
1354 return do_prlimit(current, resource, &r, NULL);
1355 }
1356
1357 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1358 struct compat_rlimit __user *, rlim)
1359 {
1360 struct rlimit r;
1361 int ret;
1362
1363 ret = do_prlimit(current, resource, NULL, &r);
1364 if (!ret) {
1365 struct compat_rlimit r32;
1366 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1367 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1368 else
1369 r32.rlim_cur = r.rlim_cur;
1370 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1371 r32.rlim_max = COMPAT_RLIM_INFINITY;
1372 else
1373 r32.rlim_max = r.rlim_max;
1374
1375 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1376 return -EFAULT;
1377 }
1378 return ret;
1379 }
1380
1381 #endif
1382
1383 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1384
1385 /*
1386 * Back compatibility for getrlimit. Needed for some apps.
1387 */
1388 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1389 struct rlimit __user *, rlim)
1390 {
1391 struct rlimit x;
1392 if (resource >= RLIM_NLIMITS)
1393 return -EINVAL;
1394
1395 task_lock(current->group_leader);
1396 x = current->signal->rlim[resource];
1397 task_unlock(current->group_leader);
1398 if (x.rlim_cur > 0x7FFFFFFF)
1399 x.rlim_cur = 0x7FFFFFFF;
1400 if (x.rlim_max > 0x7FFFFFFF)
1401 x.rlim_max = 0x7FFFFFFF;
1402 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1403 }
1404
1405 #ifdef CONFIG_COMPAT
1406 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1407 struct compat_rlimit __user *, rlim)
1408 {
1409 struct rlimit r;
1410
1411 if (resource >= RLIM_NLIMITS)
1412 return -EINVAL;
1413
1414 task_lock(current->group_leader);
1415 r = current->signal->rlim[resource];
1416 task_unlock(current->group_leader);
1417 if (r.rlim_cur > 0x7FFFFFFF)
1418 r.rlim_cur = 0x7FFFFFFF;
1419 if (r.rlim_max > 0x7FFFFFFF)
1420 r.rlim_max = 0x7FFFFFFF;
1421
1422 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1423 put_user(r.rlim_max, &rlim->rlim_max))
1424 return -EFAULT;
1425 return 0;
1426 }
1427 #endif
1428
1429 #endif
1430
1431 static inline bool rlim64_is_infinity(__u64 rlim64)
1432 {
1433 #if BITS_PER_LONG < 64
1434 return rlim64 >= ULONG_MAX;
1435 #else
1436 return rlim64 == RLIM64_INFINITY;
1437 #endif
1438 }
1439
1440 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1441 {
1442 if (rlim->rlim_cur == RLIM_INFINITY)
1443 rlim64->rlim_cur = RLIM64_INFINITY;
1444 else
1445 rlim64->rlim_cur = rlim->rlim_cur;
1446 if (rlim->rlim_max == RLIM_INFINITY)
1447 rlim64->rlim_max = RLIM64_INFINITY;
1448 else
1449 rlim64->rlim_max = rlim->rlim_max;
1450 }
1451
1452 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1453 {
1454 if (rlim64_is_infinity(rlim64->rlim_cur))
1455 rlim->rlim_cur = RLIM_INFINITY;
1456 else
1457 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1458 if (rlim64_is_infinity(rlim64->rlim_max))
1459 rlim->rlim_max = RLIM_INFINITY;
1460 else
1461 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1462 }
1463
1464 /* make sure you are allowed to change @tsk limits before calling this */
1465 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1466 struct rlimit *new_rlim, struct rlimit *old_rlim)
1467 {
1468 struct rlimit *rlim;
1469 int retval = 0;
1470
1471 if (resource >= RLIM_NLIMITS)
1472 return -EINVAL;
1473 if (new_rlim) {
1474 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1475 return -EINVAL;
1476 if (resource == RLIMIT_NOFILE &&
1477 new_rlim->rlim_max > sysctl_nr_open)
1478 return -EPERM;
1479 }
1480
1481 /* protect tsk->signal and tsk->sighand from disappearing */
1482 read_lock(&tasklist_lock);
1483 if (!tsk->sighand) {
1484 retval = -ESRCH;
1485 goto out;
1486 }
1487
1488 rlim = tsk->signal->rlim + resource;
1489 task_lock(tsk->group_leader);
1490 if (new_rlim) {
1491 /* Keep the capable check against init_user_ns until
1492 cgroups can contain all limits */
1493 if (new_rlim->rlim_max > rlim->rlim_max &&
1494 !capable(CAP_SYS_RESOURCE))
1495 retval = -EPERM;
1496 if (!retval)
1497 retval = security_task_setrlimit(tsk, resource, new_rlim);
1498 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1499 /*
1500 * The caller is asking for an immediate RLIMIT_CPU
1501 * expiry. But we use the zero value to mean "it was
1502 * never set". So let's cheat and make it one second
1503 * instead
1504 */
1505 new_rlim->rlim_cur = 1;
1506 }
1507 }
1508 if (!retval) {
1509 if (old_rlim)
1510 *old_rlim = *rlim;
1511 if (new_rlim)
1512 *rlim = *new_rlim;
1513 }
1514 task_unlock(tsk->group_leader);
1515
1516 /*
1517 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1518 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1519 * very long-standing error, and fixing it now risks breakage of
1520 * applications, so we live with it
1521 */
1522 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1523 new_rlim->rlim_cur != RLIM_INFINITY &&
1524 IS_ENABLED(CONFIG_POSIX_TIMERS))
1525 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1526 out:
1527 read_unlock(&tasklist_lock);
1528 return retval;
1529 }
1530
1531 /* rcu lock must be held */
1532 static int check_prlimit_permission(struct task_struct *task,
1533 unsigned int flags)
1534 {
1535 const struct cred *cred = current_cred(), *tcred;
1536 bool id_match;
1537
1538 if (current == task)
1539 return 0;
1540
1541 tcred = __task_cred(task);
1542 id_match = (uid_eq(cred->uid, tcred->euid) &&
1543 uid_eq(cred->uid, tcred->suid) &&
1544 uid_eq(cred->uid, tcred->uid) &&
1545 gid_eq(cred->gid, tcred->egid) &&
1546 gid_eq(cred->gid, tcred->sgid) &&
1547 gid_eq(cred->gid, tcred->gid));
1548 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1549 return -EPERM;
1550
1551 return security_task_prlimit(cred, tcred, flags);
1552 }
1553
1554 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1555 const struct rlimit64 __user *, new_rlim,
1556 struct rlimit64 __user *, old_rlim)
1557 {
1558 struct rlimit64 old64, new64;
1559 struct rlimit old, new;
1560 struct task_struct *tsk;
1561 unsigned int checkflags = 0;
1562 int ret;
1563
1564 if (old_rlim)
1565 checkflags |= LSM_PRLIMIT_READ;
1566
1567 if (new_rlim) {
1568 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1569 return -EFAULT;
1570 rlim64_to_rlim(&new64, &new);
1571 checkflags |= LSM_PRLIMIT_WRITE;
1572 }
1573
1574 rcu_read_lock();
1575 tsk = pid ? find_task_by_vpid(pid) : current;
1576 if (!tsk) {
1577 rcu_read_unlock();
1578 return -ESRCH;
1579 }
1580 ret = check_prlimit_permission(tsk, checkflags);
1581 if (ret) {
1582 rcu_read_unlock();
1583 return ret;
1584 }
1585 get_task_struct(tsk);
1586 rcu_read_unlock();
1587
1588 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1589 old_rlim ? &old : NULL);
1590
1591 if (!ret && old_rlim) {
1592 rlim_to_rlim64(&old, &old64);
1593 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1594 ret = -EFAULT;
1595 }
1596
1597 put_task_struct(tsk);
1598 return ret;
1599 }
1600
1601 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1602 {
1603 struct rlimit new_rlim;
1604
1605 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1606 return -EFAULT;
1607 return do_prlimit(current, resource, &new_rlim, NULL);
1608 }
1609
1610 /*
1611 * It would make sense to put struct rusage in the task_struct,
1612 * except that would make the task_struct be *really big*. After
1613 * task_struct gets moved into malloc'ed memory, it would
1614 * make sense to do this. It will make moving the rest of the information
1615 * a lot simpler! (Which we're not doing right now because we're not
1616 * measuring them yet).
1617 *
1618 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1619 * races with threads incrementing their own counters. But since word
1620 * reads are atomic, we either get new values or old values and we don't
1621 * care which for the sums. We always take the siglock to protect reading
1622 * the c* fields from p->signal from races with exit.c updating those
1623 * fields when reaping, so a sample either gets all the additions of a
1624 * given child after it's reaped, or none so this sample is before reaping.
1625 *
1626 * Locking:
1627 * We need to take the siglock for CHILDEREN, SELF and BOTH
1628 * for the cases current multithreaded, non-current single threaded
1629 * non-current multithreaded. Thread traversal is now safe with
1630 * the siglock held.
1631 * Strictly speaking, we donot need to take the siglock if we are current and
1632 * single threaded, as no one else can take our signal_struct away, no one
1633 * else can reap the children to update signal->c* counters, and no one else
1634 * can race with the signal-> fields. If we do not take any lock, the
1635 * signal-> fields could be read out of order while another thread was just
1636 * exiting. So we should place a read memory barrier when we avoid the lock.
1637 * On the writer side, write memory barrier is implied in __exit_signal
1638 * as __exit_signal releases the siglock spinlock after updating the signal->
1639 * fields. But we don't do this yet to keep things simple.
1640 *
1641 */
1642
1643 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1644 {
1645 r->ru_nvcsw += t->nvcsw;
1646 r->ru_nivcsw += t->nivcsw;
1647 r->ru_minflt += t->min_flt;
1648 r->ru_majflt += t->maj_flt;
1649 r->ru_inblock += task_io_get_inblock(t);
1650 r->ru_oublock += task_io_get_oublock(t);
1651 }
1652
1653 void getrusage(struct task_struct *p, int who, struct rusage *r)
1654 {
1655 struct task_struct *t;
1656 unsigned long flags;
1657 u64 tgutime, tgstime, utime, stime;
1658 unsigned long maxrss = 0;
1659
1660 memset((char *)r, 0, sizeof (*r));
1661 utime = stime = 0;
1662
1663 if (who == RUSAGE_THREAD) {
1664 task_cputime_adjusted(current, &utime, &stime);
1665 accumulate_thread_rusage(p, r);
1666 maxrss = p->signal->maxrss;
1667 goto out;
1668 }
1669
1670 if (!lock_task_sighand(p, &flags))
1671 return;
1672
1673 switch (who) {
1674 case RUSAGE_BOTH:
1675 case RUSAGE_CHILDREN:
1676 utime = p->signal->cutime;
1677 stime = p->signal->cstime;
1678 r->ru_nvcsw = p->signal->cnvcsw;
1679 r->ru_nivcsw = p->signal->cnivcsw;
1680 r->ru_minflt = p->signal->cmin_flt;
1681 r->ru_majflt = p->signal->cmaj_flt;
1682 r->ru_inblock = p->signal->cinblock;
1683 r->ru_oublock = p->signal->coublock;
1684 maxrss = p->signal->cmaxrss;
1685
1686 if (who == RUSAGE_CHILDREN)
1687 break;
1688
1689 case RUSAGE_SELF:
1690 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1691 utime += tgutime;
1692 stime += tgstime;
1693 r->ru_nvcsw += p->signal->nvcsw;
1694 r->ru_nivcsw += p->signal->nivcsw;
1695 r->ru_minflt += p->signal->min_flt;
1696 r->ru_majflt += p->signal->maj_flt;
1697 r->ru_inblock += p->signal->inblock;
1698 r->ru_oublock += p->signal->oublock;
1699 if (maxrss < p->signal->maxrss)
1700 maxrss = p->signal->maxrss;
1701 t = p;
1702 do {
1703 accumulate_thread_rusage(t, r);
1704 } while_each_thread(p, t);
1705 break;
1706
1707 default:
1708 BUG();
1709 }
1710 unlock_task_sighand(p, &flags);
1711
1712 out:
1713 r->ru_utime = ns_to_timeval(utime);
1714 r->ru_stime = ns_to_timeval(stime);
1715
1716 if (who != RUSAGE_CHILDREN) {
1717 struct mm_struct *mm = get_task_mm(p);
1718
1719 if (mm) {
1720 setmax_mm_hiwater_rss(&maxrss, mm);
1721 mmput(mm);
1722 }
1723 }
1724 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1725 }
1726
1727 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1728 {
1729 struct rusage r;
1730
1731 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1732 who != RUSAGE_THREAD)
1733 return -EINVAL;
1734
1735 getrusage(current, who, &r);
1736 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1737 }
1738
1739 #ifdef CONFIG_COMPAT
1740 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1741 {
1742 struct rusage r;
1743
1744 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1745 who != RUSAGE_THREAD)
1746 return -EINVAL;
1747
1748 getrusage(current, who, &r);
1749 return put_compat_rusage(&r, ru);
1750 }
1751 #endif
1752
1753 SYSCALL_DEFINE1(umask, int, mask)
1754 {
1755 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1756 return mask;
1757 }
1758
1759 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1760 {
1761 struct fd exe;
1762 struct file *old_exe, *exe_file;
1763 struct inode *inode;
1764 int err;
1765
1766 exe = fdget(fd);
1767 if (!exe.file)
1768 return -EBADF;
1769
1770 inode = file_inode(exe.file);
1771
1772 /*
1773 * Because the original mm->exe_file points to executable file, make
1774 * sure that this one is executable as well, to avoid breaking an
1775 * overall picture.
1776 */
1777 err = -EACCES;
1778 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1779 goto exit;
1780
1781 err = inode_permission(inode, MAY_EXEC);
1782 if (err)
1783 goto exit;
1784
1785 /*
1786 * Forbid mm->exe_file change if old file still mapped.
1787 */
1788 exe_file = get_mm_exe_file(mm);
1789 err = -EBUSY;
1790 if (exe_file) {
1791 struct vm_area_struct *vma;
1792
1793 down_read(&mm->mmap_sem);
1794 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1795 if (!vma->vm_file)
1796 continue;
1797 if (path_equal(&vma->vm_file->f_path,
1798 &exe_file->f_path))
1799 goto exit_err;
1800 }
1801
1802 up_read(&mm->mmap_sem);
1803 fput(exe_file);
1804 }
1805
1806 err = 0;
1807 /* set the new file, lockless */
1808 get_file(exe.file);
1809 old_exe = xchg(&mm->exe_file, exe.file);
1810 if (old_exe)
1811 fput(old_exe);
1812 exit:
1813 fdput(exe);
1814 return err;
1815 exit_err:
1816 up_read(&mm->mmap_sem);
1817 fput(exe_file);
1818 goto exit;
1819 }
1820
1821 /*
1822 * WARNING: we don't require any capability here so be very careful
1823 * in what is allowed for modification from userspace.
1824 */
1825 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1826 {
1827 unsigned long mmap_max_addr = TASK_SIZE;
1828 struct mm_struct *mm = current->mm;
1829 int error = -EINVAL, i;
1830
1831 static const unsigned char offsets[] = {
1832 offsetof(struct prctl_mm_map, start_code),
1833 offsetof(struct prctl_mm_map, end_code),
1834 offsetof(struct prctl_mm_map, start_data),
1835 offsetof(struct prctl_mm_map, end_data),
1836 offsetof(struct prctl_mm_map, start_brk),
1837 offsetof(struct prctl_mm_map, brk),
1838 offsetof(struct prctl_mm_map, start_stack),
1839 offsetof(struct prctl_mm_map, arg_start),
1840 offsetof(struct prctl_mm_map, arg_end),
1841 offsetof(struct prctl_mm_map, env_start),
1842 offsetof(struct prctl_mm_map, env_end),
1843 };
1844
1845 /*
1846 * Make sure the members are not somewhere outside
1847 * of allowed address space.
1848 */
1849 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1850 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1851
1852 if ((unsigned long)val >= mmap_max_addr ||
1853 (unsigned long)val < mmap_min_addr)
1854 goto out;
1855 }
1856
1857 /*
1858 * Make sure the pairs are ordered.
1859 */
1860 #define __prctl_check_order(__m1, __op, __m2) \
1861 ((unsigned long)prctl_map->__m1 __op \
1862 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1863 error = __prctl_check_order(start_code, <, end_code);
1864 error |= __prctl_check_order(start_data, <, end_data);
1865 error |= __prctl_check_order(start_brk, <=, brk);
1866 error |= __prctl_check_order(arg_start, <=, arg_end);
1867 error |= __prctl_check_order(env_start, <=, env_end);
1868 if (error)
1869 goto out;
1870 #undef __prctl_check_order
1871
1872 error = -EINVAL;
1873
1874 /*
1875 * @brk should be after @end_data in traditional maps.
1876 */
1877 if (prctl_map->start_brk <= prctl_map->end_data ||
1878 prctl_map->brk <= prctl_map->end_data)
1879 goto out;
1880
1881 /*
1882 * Neither we should allow to override limits if they set.
1883 */
1884 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1885 prctl_map->start_brk, prctl_map->end_data,
1886 prctl_map->start_data))
1887 goto out;
1888
1889 /*
1890 * Someone is trying to cheat the auxv vector.
1891 */
1892 if (prctl_map->auxv_size) {
1893 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1894 goto out;
1895 }
1896
1897 /*
1898 * Finally, make sure the caller has the rights to
1899 * change /proc/pid/exe link: only local sys admin should
1900 * be allowed to.
1901 */
1902 if (prctl_map->exe_fd != (u32)-1) {
1903 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1904 goto out;
1905 }
1906
1907 error = 0;
1908 out:
1909 return error;
1910 }
1911
1912 #ifdef CONFIG_CHECKPOINT_RESTORE
1913 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1914 {
1915 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1916 unsigned long user_auxv[AT_VECTOR_SIZE];
1917 struct mm_struct *mm = current->mm;
1918 int error;
1919
1920 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1921 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1922
1923 if (opt == PR_SET_MM_MAP_SIZE)
1924 return put_user((unsigned int)sizeof(prctl_map),
1925 (unsigned int __user *)addr);
1926
1927 if (data_size != sizeof(prctl_map))
1928 return -EINVAL;
1929
1930 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1931 return -EFAULT;
1932
1933 error = validate_prctl_map(&prctl_map);
1934 if (error)
1935 return error;
1936
1937 if (prctl_map.auxv_size) {
1938 memset(user_auxv, 0, sizeof(user_auxv));
1939 if (copy_from_user(user_auxv,
1940 (const void __user *)prctl_map.auxv,
1941 prctl_map.auxv_size))
1942 return -EFAULT;
1943
1944 /* Last entry must be AT_NULL as specification requires */
1945 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1946 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1947 }
1948
1949 if (prctl_map.exe_fd != (u32)-1) {
1950 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1951 if (error)
1952 return error;
1953 }
1954
1955 down_write(&mm->mmap_sem);
1956
1957 /*
1958 * We don't validate if these members are pointing to
1959 * real present VMAs because application may have correspond
1960 * VMAs already unmapped and kernel uses these members for statistics
1961 * output in procfs mostly, except
1962 *
1963 * - @start_brk/@brk which are used in do_brk but kernel lookups
1964 * for VMAs when updating these memvers so anything wrong written
1965 * here cause kernel to swear at userspace program but won't lead
1966 * to any problem in kernel itself
1967 */
1968
1969 mm->start_code = prctl_map.start_code;
1970 mm->end_code = prctl_map.end_code;
1971 mm->start_data = prctl_map.start_data;
1972 mm->end_data = prctl_map.end_data;
1973 mm->start_brk = prctl_map.start_brk;
1974 mm->brk = prctl_map.brk;
1975 mm->start_stack = prctl_map.start_stack;
1976 mm->arg_start = prctl_map.arg_start;
1977 mm->arg_end = prctl_map.arg_end;
1978 mm->env_start = prctl_map.env_start;
1979 mm->env_end = prctl_map.env_end;
1980
1981 /*
1982 * Note this update of @saved_auxv is lockless thus
1983 * if someone reads this member in procfs while we're
1984 * updating -- it may get partly updated results. It's
1985 * known and acceptable trade off: we leave it as is to
1986 * not introduce additional locks here making the kernel
1987 * more complex.
1988 */
1989 if (prctl_map.auxv_size)
1990 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1991
1992 up_write(&mm->mmap_sem);
1993 return 0;
1994 }
1995 #endif /* CONFIG_CHECKPOINT_RESTORE */
1996
1997 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
1998 unsigned long len)
1999 {
2000 /*
2001 * This doesn't move the auxiliary vector itself since it's pinned to
2002 * mm_struct, but it permits filling the vector with new values. It's
2003 * up to the caller to provide sane values here, otherwise userspace
2004 * tools which use this vector might be unhappy.
2005 */
2006 unsigned long user_auxv[AT_VECTOR_SIZE];
2007
2008 if (len > sizeof(user_auxv))
2009 return -EINVAL;
2010
2011 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2012 return -EFAULT;
2013
2014 /* Make sure the last entry is always AT_NULL */
2015 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2016 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2017
2018 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2019
2020 task_lock(current);
2021 memcpy(mm->saved_auxv, user_auxv, len);
2022 task_unlock(current);
2023
2024 return 0;
2025 }
2026
2027 static int prctl_set_mm(int opt, unsigned long addr,
2028 unsigned long arg4, unsigned long arg5)
2029 {
2030 struct mm_struct *mm = current->mm;
2031 struct prctl_mm_map prctl_map;
2032 struct vm_area_struct *vma;
2033 int error;
2034
2035 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2036 opt != PR_SET_MM_MAP &&
2037 opt != PR_SET_MM_MAP_SIZE)))
2038 return -EINVAL;
2039
2040 #ifdef CONFIG_CHECKPOINT_RESTORE
2041 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2042 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2043 #endif
2044
2045 if (!capable(CAP_SYS_RESOURCE))
2046 return -EPERM;
2047
2048 if (opt == PR_SET_MM_EXE_FILE)
2049 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2050
2051 if (opt == PR_SET_MM_AUXV)
2052 return prctl_set_auxv(mm, addr, arg4);
2053
2054 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2055 return -EINVAL;
2056
2057 error = -EINVAL;
2058
2059 down_write(&mm->mmap_sem);
2060 vma = find_vma(mm, addr);
2061
2062 prctl_map.start_code = mm->start_code;
2063 prctl_map.end_code = mm->end_code;
2064 prctl_map.start_data = mm->start_data;
2065 prctl_map.end_data = mm->end_data;
2066 prctl_map.start_brk = mm->start_brk;
2067 prctl_map.brk = mm->brk;
2068 prctl_map.start_stack = mm->start_stack;
2069 prctl_map.arg_start = mm->arg_start;
2070 prctl_map.arg_end = mm->arg_end;
2071 prctl_map.env_start = mm->env_start;
2072 prctl_map.env_end = mm->env_end;
2073 prctl_map.auxv = NULL;
2074 prctl_map.auxv_size = 0;
2075 prctl_map.exe_fd = -1;
2076
2077 switch (opt) {
2078 case PR_SET_MM_START_CODE:
2079 prctl_map.start_code = addr;
2080 break;
2081 case PR_SET_MM_END_CODE:
2082 prctl_map.end_code = addr;
2083 break;
2084 case PR_SET_MM_START_DATA:
2085 prctl_map.start_data = addr;
2086 break;
2087 case PR_SET_MM_END_DATA:
2088 prctl_map.end_data = addr;
2089 break;
2090 case PR_SET_MM_START_STACK:
2091 prctl_map.start_stack = addr;
2092 break;
2093 case PR_SET_MM_START_BRK:
2094 prctl_map.start_brk = addr;
2095 break;
2096 case PR_SET_MM_BRK:
2097 prctl_map.brk = addr;
2098 break;
2099 case PR_SET_MM_ARG_START:
2100 prctl_map.arg_start = addr;
2101 break;
2102 case PR_SET_MM_ARG_END:
2103 prctl_map.arg_end = addr;
2104 break;
2105 case PR_SET_MM_ENV_START:
2106 prctl_map.env_start = addr;
2107 break;
2108 case PR_SET_MM_ENV_END:
2109 prctl_map.env_end = addr;
2110 break;
2111 default:
2112 goto out;
2113 }
2114
2115 error = validate_prctl_map(&prctl_map);
2116 if (error)
2117 goto out;
2118
2119 switch (opt) {
2120 /*
2121 * If command line arguments and environment
2122 * are placed somewhere else on stack, we can
2123 * set them up here, ARG_START/END to setup
2124 * command line argumets and ENV_START/END
2125 * for environment.
2126 */
2127 case PR_SET_MM_START_STACK:
2128 case PR_SET_MM_ARG_START:
2129 case PR_SET_MM_ARG_END:
2130 case PR_SET_MM_ENV_START:
2131 case PR_SET_MM_ENV_END:
2132 if (!vma) {
2133 error = -EFAULT;
2134 goto out;
2135 }
2136 }
2137
2138 mm->start_code = prctl_map.start_code;
2139 mm->end_code = prctl_map.end_code;
2140 mm->start_data = prctl_map.start_data;
2141 mm->end_data = prctl_map.end_data;
2142 mm->start_brk = prctl_map.start_brk;
2143 mm->brk = prctl_map.brk;
2144 mm->start_stack = prctl_map.start_stack;
2145 mm->arg_start = prctl_map.arg_start;
2146 mm->arg_end = prctl_map.arg_end;
2147 mm->env_start = prctl_map.env_start;
2148 mm->env_end = prctl_map.env_end;
2149
2150 error = 0;
2151 out:
2152 up_write(&mm->mmap_sem);
2153 return error;
2154 }
2155
2156 #ifdef CONFIG_CHECKPOINT_RESTORE
2157 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2158 {
2159 return put_user(me->clear_child_tid, tid_addr);
2160 }
2161 #else
2162 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2163 {
2164 return -EINVAL;
2165 }
2166 #endif
2167
2168 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2169 {
2170 /*
2171 * If task has has_child_subreaper - all its decendants
2172 * already have these flag too and new decendants will
2173 * inherit it on fork, skip them.
2174 *
2175 * If we've found child_reaper - skip descendants in
2176 * it's subtree as they will never get out pidns.
2177 */
2178 if (p->signal->has_child_subreaper ||
2179 is_child_reaper(task_pid(p)))
2180 return 0;
2181
2182 p->signal->has_child_subreaper = 1;
2183 return 1;
2184 }
2185
2186 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2187 unsigned long, arg4, unsigned long, arg5)
2188 {
2189 struct task_struct *me = current;
2190 unsigned char comm[sizeof(me->comm)];
2191 long error;
2192
2193 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2194 if (error != -ENOSYS)
2195 return error;
2196
2197 error = 0;
2198 switch (option) {
2199 case PR_SET_PDEATHSIG:
2200 if (!valid_signal(arg2)) {
2201 error = -EINVAL;
2202 break;
2203 }
2204 me->pdeath_signal = arg2;
2205 break;
2206 case PR_GET_PDEATHSIG:
2207 error = put_user(me->pdeath_signal, (int __user *)arg2);
2208 break;
2209 case PR_GET_DUMPABLE:
2210 error = get_dumpable(me->mm);
2211 break;
2212 case PR_SET_DUMPABLE:
2213 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2214 error = -EINVAL;
2215 break;
2216 }
2217 set_dumpable(me->mm, arg2);
2218 break;
2219
2220 case PR_SET_UNALIGN:
2221 error = SET_UNALIGN_CTL(me, arg2);
2222 break;
2223 case PR_GET_UNALIGN:
2224 error = GET_UNALIGN_CTL(me, arg2);
2225 break;
2226 case PR_SET_FPEMU:
2227 error = SET_FPEMU_CTL(me, arg2);
2228 break;
2229 case PR_GET_FPEMU:
2230 error = GET_FPEMU_CTL(me, arg2);
2231 break;
2232 case PR_SET_FPEXC:
2233 error = SET_FPEXC_CTL(me, arg2);
2234 break;
2235 case PR_GET_FPEXC:
2236 error = GET_FPEXC_CTL(me, arg2);
2237 break;
2238 case PR_GET_TIMING:
2239 error = PR_TIMING_STATISTICAL;
2240 break;
2241 case PR_SET_TIMING:
2242 if (arg2 != PR_TIMING_STATISTICAL)
2243 error = -EINVAL;
2244 break;
2245 case PR_SET_NAME:
2246 comm[sizeof(me->comm) - 1] = 0;
2247 if (strncpy_from_user(comm, (char __user *)arg2,
2248 sizeof(me->comm) - 1) < 0)
2249 return -EFAULT;
2250 set_task_comm(me, comm);
2251 proc_comm_connector(me);
2252 break;
2253 case PR_GET_NAME:
2254 get_task_comm(comm, me);
2255 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2256 return -EFAULT;
2257 break;
2258 case PR_GET_ENDIAN:
2259 error = GET_ENDIAN(me, arg2);
2260 break;
2261 case PR_SET_ENDIAN:
2262 error = SET_ENDIAN(me, arg2);
2263 break;
2264 case PR_GET_SECCOMP:
2265 error = prctl_get_seccomp();
2266 break;
2267 case PR_SET_SECCOMP:
2268 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2269 break;
2270 case PR_GET_TSC:
2271 error = GET_TSC_CTL(arg2);
2272 break;
2273 case PR_SET_TSC:
2274 error = SET_TSC_CTL(arg2);
2275 break;
2276 case PR_TASK_PERF_EVENTS_DISABLE:
2277 error = perf_event_task_disable();
2278 break;
2279 case PR_TASK_PERF_EVENTS_ENABLE:
2280 error = perf_event_task_enable();
2281 break;
2282 case PR_GET_TIMERSLACK:
2283 if (current->timer_slack_ns > ULONG_MAX)
2284 error = ULONG_MAX;
2285 else
2286 error = current->timer_slack_ns;
2287 break;
2288 case PR_SET_TIMERSLACK:
2289 if (arg2 <= 0)
2290 current->timer_slack_ns =
2291 current->default_timer_slack_ns;
2292 else
2293 current->timer_slack_ns = arg2;
2294 break;
2295 case PR_MCE_KILL:
2296 if (arg4 | arg5)
2297 return -EINVAL;
2298 switch (arg2) {
2299 case PR_MCE_KILL_CLEAR:
2300 if (arg3 != 0)
2301 return -EINVAL;
2302 current->flags &= ~PF_MCE_PROCESS;
2303 break;
2304 case PR_MCE_KILL_SET:
2305 current->flags |= PF_MCE_PROCESS;
2306 if (arg3 == PR_MCE_KILL_EARLY)
2307 current->flags |= PF_MCE_EARLY;
2308 else if (arg3 == PR_MCE_KILL_LATE)
2309 current->flags &= ~PF_MCE_EARLY;
2310 else if (arg3 == PR_MCE_KILL_DEFAULT)
2311 current->flags &=
2312 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2313 else
2314 return -EINVAL;
2315 break;
2316 default:
2317 return -EINVAL;
2318 }
2319 break;
2320 case PR_MCE_KILL_GET:
2321 if (arg2 | arg3 | arg4 | arg5)
2322 return -EINVAL;
2323 if (current->flags & PF_MCE_PROCESS)
2324 error = (current->flags & PF_MCE_EARLY) ?
2325 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2326 else
2327 error = PR_MCE_KILL_DEFAULT;
2328 break;
2329 case PR_SET_MM:
2330 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2331 break;
2332 case PR_GET_TID_ADDRESS:
2333 error = prctl_get_tid_address(me, (int __user **)arg2);
2334 break;
2335 case PR_SET_CHILD_SUBREAPER:
2336 me->signal->is_child_subreaper = !!arg2;
2337 if (!arg2)
2338 break;
2339
2340 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2341 break;
2342 case PR_GET_CHILD_SUBREAPER:
2343 error = put_user(me->signal->is_child_subreaper,
2344 (int __user *)arg2);
2345 break;
2346 case PR_SET_NO_NEW_PRIVS:
2347 if (arg2 != 1 || arg3 || arg4 || arg5)
2348 return -EINVAL;
2349
2350 task_set_no_new_privs(current);
2351 break;
2352 case PR_GET_NO_NEW_PRIVS:
2353 if (arg2 || arg3 || arg4 || arg5)
2354 return -EINVAL;
2355 return task_no_new_privs(current) ? 1 : 0;
2356 case PR_GET_THP_DISABLE:
2357 if (arg2 || arg3 || arg4 || arg5)
2358 return -EINVAL;
2359 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2360 break;
2361 case PR_SET_THP_DISABLE:
2362 if (arg3 || arg4 || arg5)
2363 return -EINVAL;
2364 if (down_write_killable(&me->mm->mmap_sem))
2365 return -EINTR;
2366 if (arg2)
2367 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2368 else
2369 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2370 up_write(&me->mm->mmap_sem);
2371 break;
2372 case PR_MPX_ENABLE_MANAGEMENT:
2373 if (arg2 || arg3 || arg4 || arg5)
2374 return -EINVAL;
2375 error = MPX_ENABLE_MANAGEMENT();
2376 break;
2377 case PR_MPX_DISABLE_MANAGEMENT:
2378 if (arg2 || arg3 || arg4 || arg5)
2379 return -EINVAL;
2380 error = MPX_DISABLE_MANAGEMENT();
2381 break;
2382 case PR_SET_FP_MODE:
2383 error = SET_FP_MODE(me, arg2);
2384 break;
2385 case PR_GET_FP_MODE:
2386 error = GET_FP_MODE(me);
2387 break;
2388 default:
2389 error = -EINVAL;
2390 break;
2391 }
2392 return error;
2393 }
2394
2395 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2396 struct getcpu_cache __user *, unused)
2397 {
2398 int err = 0;
2399 int cpu = raw_smp_processor_id();
2400
2401 if (cpup)
2402 err |= put_user(cpu, cpup);
2403 if (nodep)
2404 err |= put_user(cpu_to_node(cpu), nodep);
2405 return err ? -EFAULT : 0;
2406 }
2407
2408 /**
2409 * do_sysinfo - fill in sysinfo struct
2410 * @info: pointer to buffer to fill
2411 */
2412 static int do_sysinfo(struct sysinfo *info)
2413 {
2414 unsigned long mem_total, sav_total;
2415 unsigned int mem_unit, bitcount;
2416 struct timespec tp;
2417
2418 memset(info, 0, sizeof(struct sysinfo));
2419
2420 get_monotonic_boottime(&tp);
2421 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2422
2423 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2424
2425 info->procs = nr_threads;
2426
2427 si_meminfo(info);
2428 si_swapinfo(info);
2429
2430 /*
2431 * If the sum of all the available memory (i.e. ram + swap)
2432 * is less than can be stored in a 32 bit unsigned long then
2433 * we can be binary compatible with 2.2.x kernels. If not,
2434 * well, in that case 2.2.x was broken anyways...
2435 *
2436 * -Erik Andersen <andersee@debian.org>
2437 */
2438
2439 mem_total = info->totalram + info->totalswap;
2440 if (mem_total < info->totalram || mem_total < info->totalswap)
2441 goto out;
2442 bitcount = 0;
2443 mem_unit = info->mem_unit;
2444 while (mem_unit > 1) {
2445 bitcount++;
2446 mem_unit >>= 1;
2447 sav_total = mem_total;
2448 mem_total <<= 1;
2449 if (mem_total < sav_total)
2450 goto out;
2451 }
2452
2453 /*
2454 * If mem_total did not overflow, multiply all memory values by
2455 * info->mem_unit and set it to 1. This leaves things compatible
2456 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2457 * kernels...
2458 */
2459
2460 info->mem_unit = 1;
2461 info->totalram <<= bitcount;
2462 info->freeram <<= bitcount;
2463 info->sharedram <<= bitcount;
2464 info->bufferram <<= bitcount;
2465 info->totalswap <<= bitcount;
2466 info->freeswap <<= bitcount;
2467 info->totalhigh <<= bitcount;
2468 info->freehigh <<= bitcount;
2469
2470 out:
2471 return 0;
2472 }
2473
2474 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2475 {
2476 struct sysinfo val;
2477
2478 do_sysinfo(&val);
2479
2480 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2481 return -EFAULT;
2482
2483 return 0;
2484 }
2485
2486 #ifdef CONFIG_COMPAT
2487 struct compat_sysinfo {
2488 s32 uptime;
2489 u32 loads[3];
2490 u32 totalram;
2491 u32 freeram;
2492 u32 sharedram;
2493 u32 bufferram;
2494 u32 totalswap;
2495 u32 freeswap;
2496 u16 procs;
2497 u16 pad;
2498 u32 totalhigh;
2499 u32 freehigh;
2500 u32 mem_unit;
2501 char _f[20-2*sizeof(u32)-sizeof(int)];
2502 };
2503
2504 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2505 {
2506 struct sysinfo s;
2507
2508 do_sysinfo(&s);
2509
2510 /* Check to see if any memory value is too large for 32-bit and scale
2511 * down if needed
2512 */
2513 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2514 int bitcount = 0;
2515
2516 while (s.mem_unit < PAGE_SIZE) {
2517 s.mem_unit <<= 1;
2518 bitcount++;
2519 }
2520
2521 s.totalram >>= bitcount;
2522 s.freeram >>= bitcount;
2523 s.sharedram >>= bitcount;
2524 s.bufferram >>= bitcount;
2525 s.totalswap >>= bitcount;
2526 s.freeswap >>= bitcount;
2527 s.totalhigh >>= bitcount;
2528 s.freehigh >>= bitcount;
2529 }
2530
2531 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2532 __put_user(s.uptime, &info->uptime) ||
2533 __put_user(s.loads[0], &info->loads[0]) ||
2534 __put_user(s.loads[1], &info->loads[1]) ||
2535 __put_user(s.loads[2], &info->loads[2]) ||
2536 __put_user(s.totalram, &info->totalram) ||
2537 __put_user(s.freeram, &info->freeram) ||
2538 __put_user(s.sharedram, &info->sharedram) ||
2539 __put_user(s.bufferram, &info->bufferram) ||
2540 __put_user(s.totalswap, &info->totalswap) ||
2541 __put_user(s.freeswap, &info->freeswap) ||
2542 __put_user(s.procs, &info->procs) ||
2543 __put_user(s.totalhigh, &info->totalhigh) ||
2544 __put_user(s.freehigh, &info->freehigh) ||
2545 __put_user(s.mem_unit, &info->mem_unit))
2546 return -EFAULT;
2547
2548 return 0;
2549 }
2550 #endif /* CONFIG_COMPAT */
Linux with added FPC-III support