?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dm thin metadata: fix __udivdi3 undefined on 32-bit
[linux/fpc-iii.git] / kernel / sys.c
blobe2446ade79ba700dd427b98f26af30ca67090fa2
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/rcupdate.h>
53 #include <linux/uidgid.h>
54 #include <linux/cred.h>
55
56 #include <linux/nospec.h>
57
58 #include <linux/kmsg_dump.h>
59 /* Move somewhere else to avoid recompiling? */
60 #include <generated/utsrelease.h>
61
62 #include <asm/uaccess.h>
63 #include <asm/io.h>
64 #include <asm/unistd.h>
65
66 #ifndef SET_UNALIGN_CTL
67 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
68 #endif
69 #ifndef GET_UNALIGN_CTL
70 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
71 #endif
72 #ifndef SET_FPEMU_CTL
73 # define SET_FPEMU_CTL(a, b) (-EINVAL)
74 #endif
75 #ifndef GET_FPEMU_CTL
76 # define GET_FPEMU_CTL(a, b) (-EINVAL)
77 #endif
78 #ifndef SET_FPEXC_CTL
79 # define SET_FPEXC_CTL(a, b) (-EINVAL)
80 #endif
81 #ifndef GET_FPEXC_CTL
82 # define GET_FPEXC_CTL(a, b) (-EINVAL)
83 #endif
84 #ifndef GET_ENDIAN
85 # define GET_ENDIAN(a, b) (-EINVAL)
86 #endif
87 #ifndef SET_ENDIAN
88 # define SET_ENDIAN(a, b) (-EINVAL)
89 #endif
90 #ifndef GET_TSC_CTL
91 # define GET_TSC_CTL(a) (-EINVAL)
92 #endif
93 #ifndef SET_TSC_CTL
94 # define SET_TSC_CTL(a) (-EINVAL)
95 #endif
96 #ifndef MPX_ENABLE_MANAGEMENT
97 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
98 #endif
99 #ifndef MPX_DISABLE_MANAGEMENT
100 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
101 #endif
102 #ifndef GET_FP_MODE
103 # define GET_FP_MODE(a) (-EINVAL)
104 #endif
105 #ifndef SET_FP_MODE
106 # define SET_FP_MODE(a,b) (-EINVAL)
107 #endif
108
109 /*
110 * this is where the system-wide overflow UID and GID are defined, for
111 * architectures that now have 32-bit UID/GID but didn't in the past
112 */
113
114 int overflowuid = DEFAULT_OVERFLOWUID;
115 int overflowgid = DEFAULT_OVERFLOWGID;
116
117 EXPORT_SYMBOL(overflowuid);
118 EXPORT_SYMBOL(overflowgid);
119
120 /*
121 * the same as above, but for filesystems which can only store a 16-bit
122 * UID and GID. as such, this is needed on all architectures
123 */
124
125 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
126 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
127
128 EXPORT_SYMBOL(fs_overflowuid);
129 EXPORT_SYMBOL(fs_overflowgid);
130
131 /*
132 * Returns true if current's euid is same as p's uid or euid,
133 * or has CAP_SYS_NICE to p's user_ns.
134 *
135 * Called with rcu_read_lock, creds are safe
136 */
137 static bool set_one_prio_perm(struct task_struct *p)
138 {
139 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
140
141 if (uid_eq(pcred->uid, cred->euid) ||
142 uid_eq(pcred->euid, cred->euid))
143 return true;
144 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
145 return true;
146 return false;
147 }
148
149 /*
150 * set the priority of a task
151 * - the caller must hold the RCU read lock
152 */
153 static int set_one_prio(struct task_struct *p, int niceval, int error)
154 {
155 int no_nice;
156
157 if (!set_one_prio_perm(p)) {
158 error = -EPERM;
159 goto out;
160 }
161 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
162 error = -EACCES;
163 goto out;
164 }
165 no_nice = security_task_setnice(p, niceval);
166 if (no_nice) {
167 error = no_nice;
168 goto out;
169 }
170 if (error == -ESRCH)
171 error = 0;
172 set_user_nice(p, niceval);
173 out:
174 return error;
175 }
176
177 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
178 {
179 struct task_struct *g, *p;
180 struct user_struct *user;
181 const struct cred *cred = current_cred();
182 int error = -EINVAL;
183 struct pid *pgrp;
184 kuid_t uid;
185
186 if (which > PRIO_USER || which < PRIO_PROCESS)
187 goto out;
188
189 /* normalize: avoid signed division (rounding problems) */
190 error = -ESRCH;
191 if (niceval < MIN_NICE)
192 niceval = MIN_NICE;
193 if (niceval > MAX_NICE)
194 niceval = MAX_NICE;
195
196 rcu_read_lock();
197 read_lock(&tasklist_lock);
198 switch (which) {
199 case PRIO_PROCESS:
200 if (who)
201 p = find_task_by_vpid(who);
202 else
203 p = current;
204 if (p)
205 error = set_one_prio(p, niceval, error);
206 break;
207 case PRIO_PGRP:
208 if (who)
209 pgrp = find_vpid(who);
210 else
211 pgrp = task_pgrp(current);
212 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
213 error = set_one_prio(p, niceval, error);
214 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
215 break;
216 case PRIO_USER:
217 uid = make_kuid(cred->user_ns, who);
218 user = cred->user;
219 if (!who)
220 uid = cred->uid;
221 else if (!uid_eq(uid, cred->uid)) {
222 user = find_user(uid);
223 if (!user)
224 goto out_unlock; /* No processes for this user */
225 }
226 do_each_thread(g, p) {
227 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
228 error = set_one_prio(p, niceval, error);
229 } while_each_thread(g, p);
230 if (!uid_eq(uid, cred->uid))
231 free_uid(user); /* For find_user() */
232 break;
233 }
234 out_unlock:
235 read_unlock(&tasklist_lock);
236 rcu_read_unlock();
237 out:
238 return error;
239 }
240
241 /*
242 * Ugh. To avoid negative return values, "getpriority()" will
243 * not return the normal nice-value, but a negated value that
244 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
245 * to stay compatible.
246 */
247 SYSCALL_DEFINE2(getpriority, int, which, int, who)
248 {
249 struct task_struct *g, *p;
250 struct user_struct *user;
251 const struct cred *cred = current_cred();
252 long niceval, retval = -ESRCH;
253 struct pid *pgrp;
254 kuid_t uid;
255
256 if (which > PRIO_USER || which < PRIO_PROCESS)
257 return -EINVAL;
258
259 rcu_read_lock();
260 read_lock(&tasklist_lock);
261 switch (which) {
262 case PRIO_PROCESS:
263 if (who)
264 p = find_task_by_vpid(who);
265 else
266 p = current;
267 if (p) {
268 niceval = nice_to_rlimit(task_nice(p));
269 if (niceval > retval)
270 retval = niceval;
271 }
272 break;
273 case PRIO_PGRP:
274 if (who)
275 pgrp = find_vpid(who);
276 else
277 pgrp = task_pgrp(current);
278 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
279 niceval = nice_to_rlimit(task_nice(p));
280 if (niceval > retval)
281 retval = niceval;
282 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
283 break;
284 case PRIO_USER:
285 uid = make_kuid(cred->user_ns, who);
286 user = cred->user;
287 if (!who)
288 uid = cred->uid;
289 else if (!uid_eq(uid, cred->uid)) {
290 user = find_user(uid);
291 if (!user)
292 goto out_unlock; /* No processes for this user */
293 }
294 do_each_thread(g, p) {
295 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
296 niceval = nice_to_rlimit(task_nice(p));
297 if (niceval > retval)
298 retval = niceval;
299 }
300 } while_each_thread(g, p);
301 if (!uid_eq(uid, cred->uid))
302 free_uid(user); /* for find_user() */
303 break;
304 }
305 out_unlock:
306 read_unlock(&tasklist_lock);
307 rcu_read_unlock();
308
309 return retval;
310 }
311
312 /*
313 * Unprivileged users may change the real gid to the effective gid
314 * or vice versa. (BSD-style)
315 *
316 * If you set the real gid at all, or set the effective gid to a value not
317 * equal to the real gid, then the saved gid is set to the new effective gid.
318 *
319 * This makes it possible for a setgid program to completely drop its
320 * privileges, which is often a useful assertion to make when you are doing
321 * a security audit over a program.
322 *
323 * The general idea is that a program which uses just setregid() will be
324 * 100% compatible with BSD. A program which uses just setgid() will be
325 * 100% compatible with POSIX with saved IDs.
326 *
327 * SMP: There are not races, the GIDs are checked only by filesystem
328 * operations (as far as semantic preservation is concerned).
329 */
330 #ifdef CONFIG_MULTIUSER
331 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
332 {
333 struct user_namespace *ns = current_user_ns();
334 const struct cred *old;
335 struct cred *new;
336 int retval;
337 kgid_t krgid, kegid;
338
339 krgid = make_kgid(ns, rgid);
340 kegid = make_kgid(ns, egid);
341
342 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
343 return -EINVAL;
344 if ((egid != (gid_t) -1) && !gid_valid(kegid))
345 return -EINVAL;
346
347 new = prepare_creds();
348 if (!new)
349 return -ENOMEM;
350 old = current_cred();
351
352 retval = -EPERM;
353 if (rgid != (gid_t) -1) {
354 if (gid_eq(old->gid, krgid) ||
355 gid_eq(old->egid, krgid) ||
356 ns_capable(old->user_ns, CAP_SETGID))
357 new->gid = krgid;
358 else
359 goto error;
360 }
361 if (egid != (gid_t) -1) {
362 if (gid_eq(old->gid, kegid) ||
363 gid_eq(old->egid, kegid) ||
364 gid_eq(old->sgid, kegid) ||
365 ns_capable(old->user_ns, CAP_SETGID))
366 new->egid = kegid;
367 else
368 goto error;
369 }
370
371 if (rgid != (gid_t) -1 ||
372 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
373 new->sgid = new->egid;
374 new->fsgid = new->egid;
375
376 return commit_creds(new);
377
378 error:
379 abort_creds(new);
380 return retval;
381 }
382
383 /*
384 * setgid() is implemented like SysV w/ SAVED_IDS
385 *
386 * SMP: Same implicit races as above.
387 */
388 SYSCALL_DEFINE1(setgid, gid_t, gid)
389 {
390 struct user_namespace *ns = current_user_ns();
391 const struct cred *old;
392 struct cred *new;
393 int retval;
394 kgid_t kgid;
395
396 kgid = make_kgid(ns, gid);
397 if (!gid_valid(kgid))
398 return -EINVAL;
399
400 new = prepare_creds();
401 if (!new)
402 return -ENOMEM;
403 old = current_cred();
404
405 retval = -EPERM;
406 if (ns_capable(old->user_ns, CAP_SETGID))
407 new->gid = new->egid = new->sgid = new->fsgid = kgid;
408 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
409 new->egid = new->fsgid = kgid;
410 else
411 goto error;
412
413 return commit_creds(new);
414
415 error:
416 abort_creds(new);
417 return retval;
418 }
419
420 /*
421 * change the user struct in a credentials set to match the new UID
422 */
423 static int set_user(struct cred *new)
424 {
425 struct user_struct *new_user;
426
427 new_user = alloc_uid(new->uid);
428 if (!new_user)
429 return -EAGAIN;
430
431 /*
432 * We don't fail in case of NPROC limit excess here because too many
433 * poorly written programs don't check set*uid() return code, assuming
434 * it never fails if called by root. We may still enforce NPROC limit
435 * for programs doing set*uid()+execve() by harmlessly deferring the
436 * failure to the execve() stage.
437 */
438 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
439 new_user != INIT_USER)
440 current->flags |= PF_NPROC_EXCEEDED;
441 else
442 current->flags &= ~PF_NPROC_EXCEEDED;
443
444 free_uid(new->user);
445 new->user = new_user;
446 return 0;
447 }
448
449 /*
450 * Unprivileged users may change the real uid to the effective uid
451 * or vice versa. (BSD-style)
452 *
453 * If you set the real uid at all, or set the effective uid to a value not
454 * equal to the real uid, then the saved uid is set to the new effective uid.
455 *
456 * This makes it possible for a setuid program to completely drop its
457 * privileges, which is often a useful assertion to make when you are doing
458 * a security audit over a program.
459 *
460 * The general idea is that a program which uses just setreuid() will be
461 * 100% compatible with BSD. A program which uses just setuid() will be
462 * 100% compatible with POSIX with saved IDs.
463 */
464 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
465 {
466 struct user_namespace *ns = current_user_ns();
467 const struct cred *old;
468 struct cred *new;
469 int retval;
470 kuid_t kruid, keuid;
471
472 kruid = make_kuid(ns, ruid);
473 keuid = make_kuid(ns, euid);
474
475 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
476 return -EINVAL;
477 if ((euid != (uid_t) -1) && !uid_valid(keuid))
478 return -EINVAL;
479
480 new = prepare_creds();
481 if (!new)
482 return -ENOMEM;
483 old = current_cred();
484
485 retval = -EPERM;
486 if (ruid != (uid_t) -1) {
487 new->uid = kruid;
488 if (!uid_eq(old->uid, kruid) &&
489 !uid_eq(old->euid, kruid) &&
490 !ns_capable(old->user_ns, CAP_SETUID))
491 goto error;
492 }
493
494 if (euid != (uid_t) -1) {
495 new->euid = keuid;
496 if (!uid_eq(old->uid, keuid) &&
497 !uid_eq(old->euid, keuid) &&
498 !uid_eq(old->suid, keuid) &&
499 !ns_capable(old->user_ns, CAP_SETUID))
500 goto error;
501 }
502
503 if (!uid_eq(new->uid, old->uid)) {
504 retval = set_user(new);
505 if (retval < 0)
506 goto error;
507 }
508 if (ruid != (uid_t) -1 ||
509 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
510 new->suid = new->euid;
511 new->fsuid = new->euid;
512
513 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
514 if (retval < 0)
515 goto error;
516
517 return commit_creds(new);
518
519 error:
520 abort_creds(new);
521 return retval;
522 }
523
524 /*
525 * setuid() is implemented like SysV with SAVED_IDS
526 *
527 * Note that SAVED_ID's is deficient in that a setuid root program
528 * like sendmail, for example, cannot set its uid to be a normal
529 * user and then switch back, because if you're root, setuid() sets
530 * the saved uid too. If you don't like this, blame the bright people
531 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
532 * will allow a root program to temporarily drop privileges and be able to
533 * regain them by swapping the real and effective uid.
534 */
535 SYSCALL_DEFINE1(setuid, uid_t, uid)
536 {
537 struct user_namespace *ns = current_user_ns();
538 const struct cred *old;
539 struct cred *new;
540 int retval;
541 kuid_t kuid;
542
543 kuid = make_kuid(ns, uid);
544 if (!uid_valid(kuid))
545 return -EINVAL;
546
547 new = prepare_creds();
548 if (!new)
549 return -ENOMEM;
550 old = current_cred();
551
552 retval = -EPERM;
553 if (ns_capable(old->user_ns, CAP_SETUID)) {
554 new->suid = new->uid = kuid;
555 if (!uid_eq(kuid, old->uid)) {
556 retval = set_user(new);
557 if (retval < 0)
558 goto error;
559 }
560 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
561 goto error;
562 }
563
564 new->fsuid = new->euid = kuid;
565
566 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
567 if (retval < 0)
568 goto error;
569
570 return commit_creds(new);
571
572 error:
573 abort_creds(new);
574 return retval;
575 }
576
577
578 /*
579 * This function implements a generic ability to update ruid, euid,
580 * and suid. This allows you to implement the 4.4 compatible seteuid().
581 */
582 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
583 {
584 struct user_namespace *ns = current_user_ns();
585 const struct cred *old;
586 struct cred *new;
587 int retval;
588 kuid_t kruid, keuid, ksuid;
589
590 kruid = make_kuid(ns, ruid);
591 keuid = make_kuid(ns, euid);
592 ksuid = make_kuid(ns, suid);
593
594 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
595 return -EINVAL;
596
597 if ((euid != (uid_t) -1) && !uid_valid(keuid))
598 return -EINVAL;
599
600 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
601 return -EINVAL;
602
603 new = prepare_creds();
604 if (!new)
605 return -ENOMEM;
606
607 old = current_cred();
608
609 retval = -EPERM;
610 if (!ns_capable(old->user_ns, CAP_SETUID)) {
611 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
612 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
613 goto error;
614 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
615 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
616 goto error;
617 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
618 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
619 goto error;
620 }
621
622 if (ruid != (uid_t) -1) {
623 new->uid = kruid;
624 if (!uid_eq(kruid, old->uid)) {
625 retval = set_user(new);
626 if (retval < 0)
627 goto error;
628 }
629 }
630 if (euid != (uid_t) -1)
631 new->euid = keuid;
632 if (suid != (uid_t) -1)
633 new->suid = ksuid;
634 new->fsuid = new->euid;
635
636 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
637 if (retval < 0)
638 goto error;
639
640 return commit_creds(new);
641
642 error:
643 abort_creds(new);
644 return retval;
645 }
646
647 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
648 {
649 const struct cred *cred = current_cred();
650 int retval;
651 uid_t ruid, euid, suid;
652
653 ruid = from_kuid_munged(cred->user_ns, cred->uid);
654 euid = from_kuid_munged(cred->user_ns, cred->euid);
655 suid = from_kuid_munged(cred->user_ns, cred->suid);
656
657 retval = put_user(ruid, ruidp);
658 if (!retval) {
659 retval = put_user(euid, euidp);
660 if (!retval)
661 return put_user(suid, suidp);
662 }
663 return retval;
664 }
665
666 /*
667 * Same as above, but for rgid, egid, sgid.
668 */
669 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
670 {
671 struct user_namespace *ns = current_user_ns();
672 const struct cred *old;
673 struct cred *new;
674 int retval;
675 kgid_t krgid, kegid, ksgid;
676
677 krgid = make_kgid(ns, rgid);
678 kegid = make_kgid(ns, egid);
679 ksgid = make_kgid(ns, sgid);
680
681 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
682 return -EINVAL;
683 if ((egid != (gid_t) -1) && !gid_valid(kegid))
684 return -EINVAL;
685 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
686 return -EINVAL;
687
688 new = prepare_creds();
689 if (!new)
690 return -ENOMEM;
691 old = current_cred();
692
693 retval = -EPERM;
694 if (!ns_capable(old->user_ns, CAP_SETGID)) {
695 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
696 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
697 goto error;
698 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
699 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
700 goto error;
701 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
702 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
703 goto error;
704 }
705
706 if (rgid != (gid_t) -1)
707 new->gid = krgid;
708 if (egid != (gid_t) -1)
709 new->egid = kegid;
710 if (sgid != (gid_t) -1)
711 new->sgid = ksgid;
712 new->fsgid = new->egid;
713
714 return commit_creds(new);
715
716 error:
717 abort_creds(new);
718 return retval;
719 }
720
721 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
722 {
723 const struct cred *cred = current_cred();
724 int retval;
725 gid_t rgid, egid, sgid;
726
727 rgid = from_kgid_munged(cred->user_ns, cred->gid);
728 egid = from_kgid_munged(cred->user_ns, cred->egid);
729 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
730
731 retval = put_user(rgid, rgidp);
732 if (!retval) {
733 retval = put_user(egid, egidp);
734 if (!retval)
735 retval = put_user(sgid, sgidp);
736 }
737
738 return retval;
739 }
740
741
742 /*
743 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
744 * is used for "access()" and for the NFS daemon (letting nfsd stay at
745 * whatever uid it wants to). It normally shadows "euid", except when
746 * explicitly set by setfsuid() or for access..
747 */
748 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
749 {
750 const struct cred *old;
751 struct cred *new;
752 uid_t old_fsuid;
753 kuid_t kuid;
754
755 old = current_cred();
756 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
757
758 kuid = make_kuid(old->user_ns, uid);
759 if (!uid_valid(kuid))
760 return old_fsuid;
761
762 new = prepare_creds();
763 if (!new)
764 return old_fsuid;
765
766 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
767 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
768 ns_capable(old->user_ns, CAP_SETUID)) {
769 if (!uid_eq(kuid, old->fsuid)) {
770 new->fsuid = kuid;
771 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
772 goto change_okay;
773 }
774 }
775
776 abort_creds(new);
777 return old_fsuid;
778
779 change_okay:
780 commit_creds(new);
781 return old_fsuid;
782 }
783
784 /*
785 * Samma på svenska..
786 */
787 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
788 {
789 const struct cred *old;
790 struct cred *new;
791 gid_t old_fsgid;
792 kgid_t kgid;
793
794 old = current_cred();
795 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
796
797 kgid = make_kgid(old->user_ns, gid);
798 if (!gid_valid(kgid))
799 return old_fsgid;
800
801 new = prepare_creds();
802 if (!new)
803 return old_fsgid;
804
805 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
806 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
807 ns_capable(old->user_ns, CAP_SETGID)) {
808 if (!gid_eq(kgid, old->fsgid)) {
809 new->fsgid = kgid;
810 goto change_okay;
811 }
812 }
813
814 abort_creds(new);
815 return old_fsgid;
816
817 change_okay:
818 commit_creds(new);
819 return old_fsgid;
820 }
821 #endif /* CONFIG_MULTIUSER */
822
823 /**
824 * sys_getpid - return the thread group id of the current process
825 *
826 * Note, despite the name, this returns the tgid not the pid. The tgid and
827 * the pid are identical unless CLONE_THREAD was specified on clone() in
828 * which case the tgid is the same in all threads of the same group.
829 *
830 * This is SMP safe as current->tgid does not change.
831 */
832 SYSCALL_DEFINE0(getpid)
833 {
834 return task_tgid_vnr(current);
835 }
836
837 /* Thread ID - the internal kernel "pid" */
838 SYSCALL_DEFINE0(gettid)
839 {
840 return task_pid_vnr(current);
841 }
842
843 /*
844 * Accessing ->real_parent is not SMP-safe, it could
845 * change from under us. However, we can use a stale
846 * value of ->real_parent under rcu_read_lock(), see
847 * release_task()->call_rcu(delayed_put_task_struct).
848 */
849 SYSCALL_DEFINE0(getppid)
850 {
851 int pid;
852
853 rcu_read_lock();
854 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
855 rcu_read_unlock();
856
857 return pid;
858 }
859
860 SYSCALL_DEFINE0(getuid)
861 {
862 /* Only we change this so SMP safe */
863 return from_kuid_munged(current_user_ns(), current_uid());
864 }
865
866 SYSCALL_DEFINE0(geteuid)
867 {
868 /* Only we change this so SMP safe */
869 return from_kuid_munged(current_user_ns(), current_euid());
870 }
871
872 SYSCALL_DEFINE0(getgid)
873 {
874 /* Only we change this so SMP safe */
875 return from_kgid_munged(current_user_ns(), current_gid());
876 }
877
878 SYSCALL_DEFINE0(getegid)
879 {
880 /* Only we change this so SMP safe */
881 return from_kgid_munged(current_user_ns(), current_egid());
882 }
883
884 void do_sys_times(struct tms *tms)
885 {
886 cputime_t tgutime, tgstime, cutime, cstime;
887
888 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
889 cutime = current->signal->cutime;
890 cstime = current->signal->cstime;
891 tms->tms_utime = cputime_to_clock_t(tgutime);
892 tms->tms_stime = cputime_to_clock_t(tgstime);
893 tms->tms_cutime = cputime_to_clock_t(cutime);
894 tms->tms_cstime = cputime_to_clock_t(cstime);
895 }
896
897 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
898 {
899 if (tbuf) {
900 struct tms tmp;
901
902 do_sys_times(&tmp);
903 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
904 return -EFAULT;
905 }
906 force_successful_syscall_return();
907 return (long) jiffies_64_to_clock_t(get_jiffies_64());
908 }
909
910 /*
911 * This needs some heavy checking ...
912 * I just haven't the stomach for it. I also don't fully
913 * understand sessions/pgrp etc. Let somebody who does explain it.
914 *
915 * OK, I think I have the protection semantics right.... this is really
916 * only important on a multi-user system anyway, to make sure one user
917 * can't send a signal to a process owned by another. -TYT, 12/12/91
918 *
919 * !PF_FORKNOEXEC check to conform completely to POSIX.
920 */
921 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
922 {
923 struct task_struct *p;
924 struct task_struct *group_leader = current->group_leader;
925 struct pid *pgrp;
926 int err;
927
928 if (!pid)
929 pid = task_pid_vnr(group_leader);
930 if (!pgid)
931 pgid = pid;
932 if (pgid < 0)
933 return -EINVAL;
934 rcu_read_lock();
935
936 /* From this point forward we keep holding onto the tasklist lock
937 * so that our parent does not change from under us. -DaveM
938 */
939 write_lock_irq(&tasklist_lock);
940
941 err = -ESRCH;
942 p = find_task_by_vpid(pid);
943 if (!p)
944 goto out;
945
946 err = -EINVAL;
947 if (!thread_group_leader(p))
948 goto out;
949
950 if (same_thread_group(p->real_parent, group_leader)) {
951 err = -EPERM;
952 if (task_session(p) != task_session(group_leader))
953 goto out;
954 err = -EACCES;
955 if (!(p->flags & PF_FORKNOEXEC))
956 goto out;
957 } else {
958 err = -ESRCH;
959 if (p != group_leader)
960 goto out;
961 }
962
963 err = -EPERM;
964 if (p->signal->leader)
965 goto out;
966
967 pgrp = task_pid(p);
968 if (pgid != pid) {
969 struct task_struct *g;
970
971 pgrp = find_vpid(pgid);
972 g = pid_task(pgrp, PIDTYPE_PGID);
973 if (!g || task_session(g) != task_session(group_leader))
974 goto out;
975 }
976
977 err = security_task_setpgid(p, pgid);
978 if (err)
979 goto out;
980
981 if (task_pgrp(p) != pgrp)
982 change_pid(p, PIDTYPE_PGID, pgrp);
983
984 err = 0;
985 out:
986 /* All paths lead to here, thus we are safe. -DaveM */
987 write_unlock_irq(&tasklist_lock);
988 rcu_read_unlock();
989 return err;
990 }
991
992 SYSCALL_DEFINE1(getpgid, pid_t, pid)
993 {
994 struct task_struct *p;
995 struct pid *grp;
996 int retval;
997
998 rcu_read_lock();
999 if (!pid)
1000 grp = task_pgrp(current);
1001 else {
1002 retval = -ESRCH;
1003 p = find_task_by_vpid(pid);
1004 if (!p)
1005 goto out;
1006 grp = task_pgrp(p);
1007 if (!grp)
1008 goto out;
1009
1010 retval = security_task_getpgid(p);
1011 if (retval)
1012 goto out;
1013 }
1014 retval = pid_vnr(grp);
1015 out:
1016 rcu_read_unlock();
1017 return retval;
1018 }
1019
1020 #ifdef __ARCH_WANT_SYS_GETPGRP
1021
1022 SYSCALL_DEFINE0(getpgrp)
1023 {
1024 return sys_getpgid(0);
1025 }
1026
1027 #endif
1028
1029 SYSCALL_DEFINE1(getsid, pid_t, pid)
1030 {
1031 struct task_struct *p;
1032 struct pid *sid;
1033 int retval;
1034
1035 rcu_read_lock();
1036 if (!pid)
1037 sid = task_session(current);
1038 else {
1039 retval = -ESRCH;
1040 p = find_task_by_vpid(pid);
1041 if (!p)
1042 goto out;
1043 sid = task_session(p);
1044 if (!sid)
1045 goto out;
1046
1047 retval = security_task_getsid(p);
1048 if (retval)
1049 goto out;
1050 }
1051 retval = pid_vnr(sid);
1052 out:
1053 rcu_read_unlock();
1054 return retval;
1055 }
1056
1057 static void set_special_pids(struct pid *pid)
1058 {
1059 struct task_struct *curr = current->group_leader;
1060
1061 if (task_session(curr) != pid)
1062 change_pid(curr, PIDTYPE_SID, pid);
1063
1064 if (task_pgrp(curr) != pid)
1065 change_pid(curr, PIDTYPE_PGID, pid);
1066 }
1067
1068 SYSCALL_DEFINE0(setsid)
1069 {
1070 struct task_struct *group_leader = current->group_leader;
1071 struct pid *sid = task_pid(group_leader);
1072 pid_t session = pid_vnr(sid);
1073 int err = -EPERM;
1074
1075 write_lock_irq(&tasklist_lock);
1076 /* Fail if I am already a session leader */
1077 if (group_leader->signal->leader)
1078 goto out;
1079
1080 /* Fail if a process group id already exists that equals the
1081 * proposed session id.
1082 */
1083 if (pid_task(sid, PIDTYPE_PGID))
1084 goto out;
1085
1086 group_leader->signal->leader = 1;
1087 set_special_pids(sid);
1088
1089 proc_clear_tty(group_leader);
1090
1091 err = session;
1092 out:
1093 write_unlock_irq(&tasklist_lock);
1094 if (err > 0) {
1095 proc_sid_connector(group_leader);
1096 sched_autogroup_create_attach(group_leader);
1097 }
1098 return err;
1099 }
1100
1101 DECLARE_RWSEM(uts_sem);
1102
1103 #ifdef COMPAT_UTS_MACHINE
1104 #define override_architecture(name) \
1105 (personality(current->personality) == PER_LINUX32 && \
1106 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1107 sizeof(COMPAT_UTS_MACHINE)))
1108 #else
1109 #define override_architecture(name) 0
1110 #endif
1111
1112 /*
1113 * Work around broken programs that cannot handle "Linux 3.0".
1114 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1115 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1116 */
1117 static int override_release(char __user *release, size_t len)
1118 {
1119 int ret = 0;
1120
1121 if (current->personality & UNAME26) {
1122 const char *rest = UTS_RELEASE;
1123 char buf[65] = { 0 };
1124 int ndots = 0;
1125 unsigned v;
1126 size_t copy;
1127
1128 while (*rest) {
1129 if (*rest == '.' && ++ndots >= 3)
1130 break;
1131 if (!isdigit(*rest) && *rest != '.')
1132 break;
1133 rest++;
1134 }
1135 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1136 copy = clamp_t(size_t, len, 1, sizeof(buf));
1137 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1138 ret = copy_to_user(release, buf, copy + 1);
1139 }
1140 return ret;
1141 }
1142
1143 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1144 {
1145 struct new_utsname tmp;
1146
1147 down_read(&uts_sem);
1148 memcpy(&tmp, utsname(), sizeof(tmp));
1149 up_read(&uts_sem);
1150 if (copy_to_user(name, &tmp, sizeof(tmp)))
1151 return -EFAULT;
1152
1153 if (override_release(name->release, sizeof(name->release)))
1154 return -EFAULT;
1155 if (override_architecture(name))
1156 return -EFAULT;
1157 return 0;
1158 }
1159
1160 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1161 /*
1162 * Old cruft
1163 */
1164 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1165 {
1166 struct old_utsname tmp;
1167
1168 if (!name)
1169 return -EFAULT;
1170
1171 down_read(&uts_sem);
1172 memcpy(&tmp, utsname(), sizeof(tmp));
1173 up_read(&uts_sem);
1174 if (copy_to_user(name, &tmp, sizeof(tmp)))
1175 return -EFAULT;
1176
1177 if (override_release(name->release, sizeof(name->release)))
1178 return -EFAULT;
1179 if (override_architecture(name))
1180 return -EFAULT;
1181 return 0;
1182 }
1183
1184 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1185 {
1186 struct oldold_utsname tmp = {};
1187
1188 if (!name)
1189 return -EFAULT;
1190
1191 down_read(&uts_sem);
1192 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1193 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1194 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1195 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1196 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1197 up_read(&uts_sem);
1198 if (copy_to_user(name, &tmp, sizeof(tmp)))
1199 return -EFAULT;
1200
1201 if (override_architecture(name))
1202 return -EFAULT;
1203 if (override_release(name->release, sizeof(name->release)))
1204 return -EFAULT;
1205 return 0;
1206 }
1207 #endif
1208
1209 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1210 {
1211 int errno;
1212 char tmp[__NEW_UTS_LEN];
1213
1214 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1215 return -EPERM;
1216
1217 if (len < 0 || len > __NEW_UTS_LEN)
1218 return -EINVAL;
1219 errno = -EFAULT;
1220 if (!copy_from_user(tmp, name, len)) {
1221 struct new_utsname *u;
1222
1223 down_write(&uts_sem);
1224 u = utsname();
1225 memcpy(u->nodename, tmp, len);
1226 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1227 errno = 0;
1228 uts_proc_notify(UTS_PROC_HOSTNAME);
1229 up_write(&uts_sem);
1230 }
1231 return errno;
1232 }
1233
1234 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1235
1236 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1237 {
1238 int i;
1239 struct new_utsname *u;
1240 char tmp[__NEW_UTS_LEN + 1];
1241
1242 if (len < 0)
1243 return -EINVAL;
1244 down_read(&uts_sem);
1245 u = utsname();
1246 i = 1 + strlen(u->nodename);
1247 if (i > len)
1248 i = len;
1249 memcpy(tmp, u->nodename, i);
1250 up_read(&uts_sem);
1251 if (copy_to_user(name, tmp, i))
1252 return -EFAULT;
1253 return 0;
1254 }
1255
1256 #endif
1257
1258 /*
1259 * Only setdomainname; getdomainname can be implemented by calling
1260 * uname()
1261 */
1262 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1263 {
1264 int errno;
1265 char tmp[__NEW_UTS_LEN];
1266
1267 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1268 return -EPERM;
1269 if (len < 0 || len > __NEW_UTS_LEN)
1270 return -EINVAL;
1271
1272 errno = -EFAULT;
1273 if (!copy_from_user(tmp, name, len)) {
1274 struct new_utsname *u;
1275
1276 down_write(&uts_sem);
1277 u = utsname();
1278 memcpy(u->domainname, tmp, len);
1279 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1280 errno = 0;
1281 uts_proc_notify(UTS_PROC_DOMAINNAME);
1282 up_write(&uts_sem);
1283 }
1284 return errno;
1285 }
1286
1287 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1288 {
1289 struct rlimit value;
1290 int ret;
1291
1292 ret = do_prlimit(current, resource, NULL, &value);
1293 if (!ret)
1294 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1295
1296 return ret;
1297 }
1298
1299 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1300
1301 /*
1302 * Back compatibility for getrlimit. Needed for some apps.
1303 */
1304 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1305 struct rlimit __user *, rlim)
1306 {
1307 struct rlimit x;
1308 if (resource >= RLIM_NLIMITS)
1309 return -EINVAL;
1310
1311 resource = array_index_nospec(resource, RLIM_NLIMITS);
1312 task_lock(current->group_leader);
1313 x = current->signal->rlim[resource];
1314 task_unlock(current->group_leader);
1315 if (x.rlim_cur > 0x7FFFFFFF)
1316 x.rlim_cur = 0x7FFFFFFF;
1317 if (x.rlim_max > 0x7FFFFFFF)
1318 x.rlim_max = 0x7FFFFFFF;
1319 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1320 }
1321
1322 #endif
1323
1324 static inline bool rlim64_is_infinity(__u64 rlim64)
1325 {
1326 #if BITS_PER_LONG < 64
1327 return rlim64 >= ULONG_MAX;
1328 #else
1329 return rlim64 == RLIM64_INFINITY;
1330 #endif
1331 }
1332
1333 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1334 {
1335 if (rlim->rlim_cur == RLIM_INFINITY)
1336 rlim64->rlim_cur = RLIM64_INFINITY;
1337 else
1338 rlim64->rlim_cur = rlim->rlim_cur;
1339 if (rlim->rlim_max == RLIM_INFINITY)
1340 rlim64->rlim_max = RLIM64_INFINITY;
1341 else
1342 rlim64->rlim_max = rlim->rlim_max;
1343 }
1344
1345 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1346 {
1347 if (rlim64_is_infinity(rlim64->rlim_cur))
1348 rlim->rlim_cur = RLIM_INFINITY;
1349 else
1350 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1351 if (rlim64_is_infinity(rlim64->rlim_max))
1352 rlim->rlim_max = RLIM_INFINITY;
1353 else
1354 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1355 }
1356
1357 /* make sure you are allowed to change @tsk limits before calling this */
1358 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1359 struct rlimit *new_rlim, struct rlimit *old_rlim)
1360 {
1361 struct rlimit *rlim;
1362 int retval = 0;
1363
1364 if (resource >= RLIM_NLIMITS)
1365 return -EINVAL;
1366 if (new_rlim) {
1367 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1368 return -EINVAL;
1369 if (resource == RLIMIT_NOFILE &&
1370 new_rlim->rlim_max > sysctl_nr_open)
1371 return -EPERM;
1372 }
1373
1374 /* protect tsk->signal and tsk->sighand from disappearing */
1375 read_lock(&tasklist_lock);
1376 if (!tsk->sighand) {
1377 retval = -ESRCH;
1378 goto out;
1379 }
1380
1381 rlim = tsk->signal->rlim + resource;
1382 task_lock(tsk->group_leader);
1383 if (new_rlim) {
1384 /* Keep the capable check against init_user_ns until
1385 cgroups can contain all limits */
1386 if (new_rlim->rlim_max > rlim->rlim_max &&
1387 !capable(CAP_SYS_RESOURCE))
1388 retval = -EPERM;
1389 if (!retval)
1390 retval = security_task_setrlimit(tsk->group_leader,
1391 resource, new_rlim);
1392 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1393 /*
1394 * The caller is asking for an immediate RLIMIT_CPU
1395 * expiry. But we use the zero value to mean "it was
1396 * never set". So let's cheat and make it one second
1397 * instead
1398 */
1399 new_rlim->rlim_cur = 1;
1400 }
1401 }
1402 if (!retval) {
1403 if (old_rlim)
1404 *old_rlim = *rlim;
1405 if (new_rlim)
1406 *rlim = *new_rlim;
1407 }
1408 task_unlock(tsk->group_leader);
1409
1410 /*
1411 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1412 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1413 * very long-standing error, and fixing it now risks breakage of
1414 * applications, so we live with it
1415 */
1416 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1417 new_rlim->rlim_cur != RLIM_INFINITY)
1418 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1419 out:
1420 read_unlock(&tasklist_lock);
1421 return retval;
1422 }
1423
1424 /* rcu lock must be held */
1425 static int check_prlimit_permission(struct task_struct *task)
1426 {
1427 const struct cred *cred = current_cred(), *tcred;
1428
1429 if (current == task)
1430 return 0;
1431
1432 tcred = __task_cred(task);
1433 if (uid_eq(cred->uid, tcred->euid) &&
1434 uid_eq(cred->uid, tcred->suid) &&
1435 uid_eq(cred->uid, tcred->uid) &&
1436 gid_eq(cred->gid, tcred->egid) &&
1437 gid_eq(cred->gid, tcred->sgid) &&
1438 gid_eq(cred->gid, tcred->gid))
1439 return 0;
1440 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1441 return 0;
1442
1443 return -EPERM;
1444 }
1445
1446 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1447 const struct rlimit64 __user *, new_rlim,
1448 struct rlimit64 __user *, old_rlim)
1449 {
1450 struct rlimit64 old64, new64;
1451 struct rlimit old, new;
1452 struct task_struct *tsk;
1453 int ret;
1454
1455 if (new_rlim) {
1456 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1457 return -EFAULT;
1458 rlim64_to_rlim(&new64, &new);
1459 }
1460
1461 rcu_read_lock();
1462 tsk = pid ? find_task_by_vpid(pid) : current;
1463 if (!tsk) {
1464 rcu_read_unlock();
1465 return -ESRCH;
1466 }
1467 ret = check_prlimit_permission(tsk);
1468 if (ret) {
1469 rcu_read_unlock();
1470 return ret;
1471 }
1472 get_task_struct(tsk);
1473 rcu_read_unlock();
1474
1475 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1476 old_rlim ? &old : NULL);
1477
1478 if (!ret && old_rlim) {
1479 rlim_to_rlim64(&old, &old64);
1480 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1481 ret = -EFAULT;
1482 }
1483
1484 put_task_struct(tsk);
1485 return ret;
1486 }
1487
1488 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1489 {
1490 struct rlimit new_rlim;
1491
1492 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1493 return -EFAULT;
1494 return do_prlimit(current, resource, &new_rlim, NULL);
1495 }
1496
1497 /*
1498 * It would make sense to put struct rusage in the task_struct,
1499 * except that would make the task_struct be *really big*. After
1500 * task_struct gets moved into malloc'ed memory, it would
1501 * make sense to do this. It will make moving the rest of the information
1502 * a lot simpler! (Which we're not doing right now because we're not
1503 * measuring them yet).
1504 *
1505 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1506 * races with threads incrementing their own counters. But since word
1507 * reads are atomic, we either get new values or old values and we don't
1508 * care which for the sums. We always take the siglock to protect reading
1509 * the c* fields from p->signal from races with exit.c updating those
1510 * fields when reaping, so a sample either gets all the additions of a
1511 * given child after it's reaped, or none so this sample is before reaping.
1512 *
1513 * Locking:
1514 * We need to take the siglock for CHILDEREN, SELF and BOTH
1515 * for the cases current multithreaded, non-current single threaded
1516 * non-current multithreaded. Thread traversal is now safe with
1517 * the siglock held.
1518 * Strictly speaking, we donot need to take the siglock if we are current and
1519 * single threaded, as no one else can take our signal_struct away, no one
1520 * else can reap the children to update signal->c* counters, and no one else
1521 * can race with the signal-> fields. If we do not take any lock, the
1522 * signal-> fields could be read out of order while another thread was just
1523 * exiting. So we should place a read memory barrier when we avoid the lock.
1524 * On the writer side, write memory barrier is implied in __exit_signal
1525 * as __exit_signal releases the siglock spinlock after updating the signal->
1526 * fields. But we don't do this yet to keep things simple.
1527 *
1528 */
1529
1530 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1531 {
1532 r->ru_nvcsw += t->nvcsw;
1533 r->ru_nivcsw += t->nivcsw;
1534 r->ru_minflt += t->min_flt;
1535 r->ru_majflt += t->maj_flt;
1536 r->ru_inblock += task_io_get_inblock(t);
1537 r->ru_oublock += task_io_get_oublock(t);
1538 }
1539
1540 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1541 {
1542 struct task_struct *t;
1543 unsigned long flags;
1544 cputime_t tgutime, tgstime, utime, stime;
1545 unsigned long maxrss = 0;
1546
1547 memset((char *)r, 0, sizeof (*r));
1548 utime = stime = 0;
1549
1550 if (who == RUSAGE_THREAD) {
1551 task_cputime_adjusted(current, &utime, &stime);
1552 accumulate_thread_rusage(p, r);
1553 maxrss = p->signal->maxrss;
1554 goto out;
1555 }
1556
1557 if (!lock_task_sighand(p, &flags))
1558 return;
1559
1560 switch (who) {
1561 case RUSAGE_BOTH:
1562 case RUSAGE_CHILDREN:
1563 utime = p->signal->cutime;
1564 stime = p->signal->cstime;
1565 r->ru_nvcsw = p->signal->cnvcsw;
1566 r->ru_nivcsw = p->signal->cnivcsw;
1567 r->ru_minflt = p->signal->cmin_flt;
1568 r->ru_majflt = p->signal->cmaj_flt;
1569 r->ru_inblock = p->signal->cinblock;
1570 r->ru_oublock = p->signal->coublock;
1571 maxrss = p->signal->cmaxrss;
1572
1573 if (who == RUSAGE_CHILDREN)
1574 break;
1575
1576 case RUSAGE_SELF:
1577 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1578 utime += tgutime;
1579 stime += tgstime;
1580 r->ru_nvcsw += p->signal->nvcsw;
1581 r->ru_nivcsw += p->signal->nivcsw;
1582 r->ru_minflt += p->signal->min_flt;
1583 r->ru_majflt += p->signal->maj_flt;
1584 r->ru_inblock += p->signal->inblock;
1585 r->ru_oublock += p->signal->oublock;
1586 if (maxrss < p->signal->maxrss)
1587 maxrss = p->signal->maxrss;
1588 t = p;
1589 do {
1590 accumulate_thread_rusage(t, r);
1591 } while_each_thread(p, t);
1592 break;
1593
1594 default:
1595 BUG();
1596 }
1597 unlock_task_sighand(p, &flags);
1598
1599 out:
1600 cputime_to_timeval(utime, &r->ru_utime);
1601 cputime_to_timeval(stime, &r->ru_stime);
1602
1603 if (who != RUSAGE_CHILDREN) {
1604 struct mm_struct *mm = get_task_mm(p);
1605
1606 if (mm) {
1607 setmax_mm_hiwater_rss(&maxrss, mm);
1608 mmput(mm);
1609 }
1610 }
1611 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1612 }
1613
1614 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1615 {
1616 struct rusage r;
1617
1618 k_getrusage(p, who, &r);
1619 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1620 }
1621
1622 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1623 {
1624 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1625 who != RUSAGE_THREAD)
1626 return -EINVAL;
1627 return getrusage(current, who, ru);
1628 }
1629
1630 #ifdef CONFIG_COMPAT
1631 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1632 {
1633 struct rusage r;
1634
1635 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1636 who != RUSAGE_THREAD)
1637 return -EINVAL;
1638
1639 k_getrusage(current, who, &r);
1640 return put_compat_rusage(&r, ru);
1641 }
1642 #endif
1643
1644 SYSCALL_DEFINE1(umask, int, mask)
1645 {
1646 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1647 return mask;
1648 }
1649
1650 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1651 {
1652 struct fd exe;
1653 struct file *old_exe, *exe_file;
1654 struct inode *inode;
1655 int err;
1656
1657 exe = fdget(fd);
1658 if (!exe.file)
1659 return -EBADF;
1660
1661 inode = file_inode(exe.file);
1662
1663 /*
1664 * Because the original mm->exe_file points to executable file, make
1665 * sure that this one is executable as well, to avoid breaking an
1666 * overall picture.
1667 */
1668 err = -EACCES;
1669 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1670 goto exit;
1671
1672 err = inode_permission(inode, MAY_EXEC);
1673 if (err)
1674 goto exit;
1675
1676 /*
1677 * Forbid mm->exe_file change if old file still mapped.
1678 */
1679 exe_file = get_mm_exe_file(mm);
1680 err = -EBUSY;
1681 if (exe_file) {
1682 struct vm_area_struct *vma;
1683
1684 down_read(&mm->mmap_sem);
1685 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1686 if (!vma->vm_file)
1687 continue;
1688 if (path_equal(&vma->vm_file->f_path,
1689 &exe_file->f_path))
1690 goto exit_err;
1691 }
1692
1693 up_read(&mm->mmap_sem);
1694 fput(exe_file);
1695 }
1696
1697 /*
1698 * The symlink can be changed only once, just to disallow arbitrary
1699 * transitions malicious software might bring in. This means one
1700 * could make a snapshot over all processes running and monitor
1701 * /proc/pid/exe changes to notice unusual activity if needed.
1702 */
1703 err = -EPERM;
1704 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1705 goto exit;
1706
1707 err = 0;
1708 /* set the new file, lockless */
1709 get_file(exe.file);
1710 old_exe = xchg(&mm->exe_file, exe.file);
1711 if (old_exe)
1712 fput(old_exe);
1713 exit:
1714 fdput(exe);
1715 return err;
1716 exit_err:
1717 up_read(&mm->mmap_sem);
1718 fput(exe_file);
1719 goto exit;
1720 }
1721
1722 /*
1723 * WARNING: we don't require any capability here so be very careful
1724 * in what is allowed for modification from userspace.
1725 */
1726 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1727 {
1728 unsigned long mmap_max_addr = TASK_SIZE;
1729 struct mm_struct *mm = current->mm;
1730 int error = -EINVAL, i;
1731
1732 static const unsigned char offsets[] = {
1733 offsetof(struct prctl_mm_map, start_code),
1734 offsetof(struct prctl_mm_map, end_code),
1735 offsetof(struct prctl_mm_map, start_data),
1736 offsetof(struct prctl_mm_map, end_data),
1737 offsetof(struct prctl_mm_map, start_brk),
1738 offsetof(struct prctl_mm_map, brk),
1739 offsetof(struct prctl_mm_map, start_stack),
1740 offsetof(struct prctl_mm_map, arg_start),
1741 offsetof(struct prctl_mm_map, arg_end),
1742 offsetof(struct prctl_mm_map, env_start),
1743 offsetof(struct prctl_mm_map, env_end),
1744 };
1745
1746 /*
1747 * Make sure the members are not somewhere outside
1748 * of allowed address space.
1749 */
1750 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1751 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1752
1753 if ((unsigned long)val >= mmap_max_addr ||
1754 (unsigned long)val < mmap_min_addr)
1755 goto out;
1756 }
1757
1758 /*
1759 * Make sure the pairs are ordered.
1760 */
1761 #define __prctl_check_order(__m1, __op, __m2) \
1762 ((unsigned long)prctl_map->__m1 __op \
1763 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1764 error = __prctl_check_order(start_code, <, end_code);
1765 error |= __prctl_check_order(start_data, <, end_data);
1766 error |= __prctl_check_order(start_brk, <=, brk);
1767 error |= __prctl_check_order(arg_start, <=, arg_end);
1768 error |= __prctl_check_order(env_start, <=, env_end);
1769 if (error)
1770 goto out;
1771 #undef __prctl_check_order
1772
1773 error = -EINVAL;
1774
1775 /*
1776 * @brk should be after @end_data in traditional maps.
1777 */
1778 if (prctl_map->start_brk <= prctl_map->end_data ||
1779 prctl_map->brk <= prctl_map->end_data)
1780 goto out;
1781
1782 /*
1783 * Neither we should allow to override limits if they set.
1784 */
1785 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1786 prctl_map->start_brk, prctl_map->end_data,
1787 prctl_map->start_data))
1788 goto out;
1789
1790 /*
1791 * Someone is trying to cheat the auxv vector.
1792 */
1793 if (prctl_map->auxv_size) {
1794 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1795 goto out;
1796 }
1797
1798 /*
1799 * Finally, make sure the caller has the rights to
1800 * change /proc/pid/exe link: only local root should
1801 * be allowed to.
1802 */
1803 if (prctl_map->exe_fd != (u32)-1) {
1804 struct user_namespace *ns = current_user_ns();
1805 const struct cred *cred = current_cred();
1806
1807 if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1808 !gid_eq(cred->gid, make_kgid(ns, 0)))
1809 goto out;
1810 }
1811
1812 error = 0;
1813 out:
1814 return error;
1815 }
1816
1817 #ifdef CONFIG_CHECKPOINT_RESTORE
1818 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1819 {
1820 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1821 unsigned long user_auxv[AT_VECTOR_SIZE];
1822 struct mm_struct *mm = current->mm;
1823 int error;
1824
1825 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1826 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1827
1828 if (opt == PR_SET_MM_MAP_SIZE)
1829 return put_user((unsigned int)sizeof(prctl_map),
1830 (unsigned int __user *)addr);
1831
1832 if (data_size != sizeof(prctl_map))
1833 return -EINVAL;
1834
1835 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1836 return -EFAULT;
1837
1838 error = validate_prctl_map(&prctl_map);
1839 if (error)
1840 return error;
1841
1842 if (prctl_map.auxv_size) {
1843 memset(user_auxv, 0, sizeof(user_auxv));
1844 if (copy_from_user(user_auxv,
1845 (const void __user *)prctl_map.auxv,
1846 prctl_map.auxv_size))
1847 return -EFAULT;
1848
1849 /* Last entry must be AT_NULL as specification requires */
1850 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1851 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1852 }
1853
1854 if (prctl_map.exe_fd != (u32)-1) {
1855 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1856 if (error)
1857 return error;
1858 }
1859
1860 down_write(&mm->mmap_sem);
1861
1862 /*
1863 * We don't validate if these members are pointing to
1864 * real present VMAs because application may have correspond
1865 * VMAs already unmapped and kernel uses these members for statistics
1866 * output in procfs mostly, except
1867 *
1868 * - @start_brk/@brk which are used in do_brk but kernel lookups
1869 * for VMAs when updating these memvers so anything wrong written
1870 * here cause kernel to swear at userspace program but won't lead
1871 * to any problem in kernel itself
1872 */
1873
1874 mm->start_code = prctl_map.start_code;
1875 mm->end_code = prctl_map.end_code;
1876 mm->start_data = prctl_map.start_data;
1877 mm->end_data = prctl_map.end_data;
1878 mm->start_brk = prctl_map.start_brk;
1879 mm->brk = prctl_map.brk;
1880 mm->start_stack = prctl_map.start_stack;
1881 mm->arg_start = prctl_map.arg_start;
1882 mm->arg_end = prctl_map.arg_end;
1883 mm->env_start = prctl_map.env_start;
1884 mm->env_end = prctl_map.env_end;
1885
1886 /*
1887 * Note this update of @saved_auxv is lockless thus
1888 * if someone reads this member in procfs while we're
1889 * updating -- it may get partly updated results. It's
1890 * known and acceptable trade off: we leave it as is to
1891 * not introduce additional locks here making the kernel
1892 * more complex.
1893 */
1894 if (prctl_map.auxv_size)
1895 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1896
1897 up_write(&mm->mmap_sem);
1898 return 0;
1899 }
1900 #endif /* CONFIG_CHECKPOINT_RESTORE */
1901
1902 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
1903 unsigned long len)
1904 {
1905 /*
1906 * This doesn't move the auxiliary vector itself since it's pinned to
1907 * mm_struct, but it permits filling the vector with new values. It's
1908 * up to the caller to provide sane values here, otherwise userspace
1909 * tools which use this vector might be unhappy.
1910 */
1911 unsigned long user_auxv[AT_VECTOR_SIZE];
1912
1913 if (len > sizeof(user_auxv))
1914 return -EINVAL;
1915
1916 if (copy_from_user(user_auxv, (const void __user *)addr, len))
1917 return -EFAULT;
1918
1919 /* Make sure the last entry is always AT_NULL */
1920 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1921 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1922
1923 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1924
1925 task_lock(current);
1926 memcpy(mm->saved_auxv, user_auxv, len);
1927 task_unlock(current);
1928
1929 return 0;
1930 }
1931
1932 static int prctl_set_mm(int opt, unsigned long addr,
1933 unsigned long arg4, unsigned long arg5)
1934 {
1935 struct mm_struct *mm = current->mm;
1936 struct prctl_mm_map prctl_map;
1937 struct vm_area_struct *vma;
1938 int error;
1939
1940 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1941 opt != PR_SET_MM_MAP &&
1942 opt != PR_SET_MM_MAP_SIZE)))
1943 return -EINVAL;
1944
1945 #ifdef CONFIG_CHECKPOINT_RESTORE
1946 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1947 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1948 #endif
1949
1950 if (!capable(CAP_SYS_RESOURCE))
1951 return -EPERM;
1952
1953 if (opt == PR_SET_MM_EXE_FILE)
1954 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1955
1956 if (opt == PR_SET_MM_AUXV)
1957 return prctl_set_auxv(mm, addr, arg4);
1958
1959 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1960 return -EINVAL;
1961
1962 error = -EINVAL;
1963
1964 down_write(&mm->mmap_sem);
1965 vma = find_vma(mm, addr);
1966
1967 prctl_map.start_code = mm->start_code;
1968 prctl_map.end_code = mm->end_code;
1969 prctl_map.start_data = mm->start_data;
1970 prctl_map.end_data = mm->end_data;
1971 prctl_map.start_brk = mm->start_brk;
1972 prctl_map.brk = mm->brk;
1973 prctl_map.start_stack = mm->start_stack;
1974 prctl_map.arg_start = mm->arg_start;
1975 prctl_map.arg_end = mm->arg_end;
1976 prctl_map.env_start = mm->env_start;
1977 prctl_map.env_end = mm->env_end;
1978 prctl_map.auxv = NULL;
1979 prctl_map.auxv_size = 0;
1980 prctl_map.exe_fd = -1;
1981
1982 switch (opt) {
1983 case PR_SET_MM_START_CODE:
1984 prctl_map.start_code = addr;
1985 break;
1986 case PR_SET_MM_END_CODE:
1987 prctl_map.end_code = addr;
1988 break;
1989 case PR_SET_MM_START_DATA:
1990 prctl_map.start_data = addr;
1991 break;
1992 case PR_SET_MM_END_DATA:
1993 prctl_map.end_data = addr;
1994 break;
1995 case PR_SET_MM_START_STACK:
1996 prctl_map.start_stack = addr;
1997 break;
1998 case PR_SET_MM_START_BRK:
1999 prctl_map.start_brk = addr;
2000 break;
2001 case PR_SET_MM_BRK:
2002 prctl_map.brk = addr;
2003 break;
2004 case PR_SET_MM_ARG_START:
2005 prctl_map.arg_start = addr;
2006 break;
2007 case PR_SET_MM_ARG_END:
2008 prctl_map.arg_end = addr;
2009 break;
2010 case PR_SET_MM_ENV_START:
2011 prctl_map.env_start = addr;
2012 break;
2013 case PR_SET_MM_ENV_END:
2014 prctl_map.env_end = addr;
2015 break;
2016 default:
2017 goto out;
2018 }
2019
2020 error = validate_prctl_map(&prctl_map);
2021 if (error)
2022 goto out;
2023
2024 switch (opt) {
2025 /*
2026 * If command line arguments and environment
2027 * are placed somewhere else on stack, we can
2028 * set them up here, ARG_START/END to setup
2029 * command line argumets and ENV_START/END
2030 * for environment.
2031 */
2032 case PR_SET_MM_START_STACK:
2033 case PR_SET_MM_ARG_START:
2034 case PR_SET_MM_ARG_END:
2035 case PR_SET_MM_ENV_START:
2036 case PR_SET_MM_ENV_END:
2037 if (!vma) {
2038 error = -EFAULT;
2039 goto out;
2040 }
2041 }
2042
2043 mm->start_code = prctl_map.start_code;
2044 mm->end_code = prctl_map.end_code;
2045 mm->start_data = prctl_map.start_data;
2046 mm->end_data = prctl_map.end_data;
2047 mm->start_brk = prctl_map.start_brk;
2048 mm->brk = prctl_map.brk;
2049 mm->start_stack = prctl_map.start_stack;
2050 mm->arg_start = prctl_map.arg_start;
2051 mm->arg_end = prctl_map.arg_end;
2052 mm->env_start = prctl_map.env_start;
2053 mm->env_end = prctl_map.env_end;
2054
2055 error = 0;
2056 out:
2057 up_write(&mm->mmap_sem);
2058 return error;
2059 }
2060
2061 #ifdef CONFIG_CHECKPOINT_RESTORE
2062 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2063 {
2064 return put_user(me->clear_child_tid, tid_addr);
2065 }
2066 #else
2067 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2068 {
2069 return -EINVAL;
2070 }
2071 #endif
2072
2073 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2074 {
2075 return -EINVAL;
2076 }
2077
2078 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2079 unsigned long ctrl)
2080 {
2081 return -EINVAL;
2082 }
2083
2084 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2085 unsigned long, arg4, unsigned long, arg5)
2086 {
2087 struct task_struct *me = current;
2088 unsigned char comm[sizeof(me->comm)];
2089 long error;
2090
2091 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2092 if (error != -ENOSYS)
2093 return error;
2094
2095 error = 0;
2096 switch (option) {
2097 case PR_SET_PDEATHSIG:
2098 if (!valid_signal(arg2)) {
2099 error = -EINVAL;
2100 break;
2101 }
2102 me->pdeath_signal = arg2;
2103 break;
2104 case PR_GET_PDEATHSIG:
2105 error = put_user(me->pdeath_signal, (int __user *)arg2);
2106 break;
2107 case PR_GET_DUMPABLE:
2108 error = get_dumpable(me->mm);
2109 break;
2110 case PR_SET_DUMPABLE:
2111 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2112 error = -EINVAL;
2113 break;
2114 }
2115 set_dumpable(me->mm, arg2);
2116 break;
2117
2118 case PR_SET_UNALIGN:
2119 error = SET_UNALIGN_CTL(me, arg2);
2120 break;
2121 case PR_GET_UNALIGN:
2122 error = GET_UNALIGN_CTL(me, arg2);
2123 break;
2124 case PR_SET_FPEMU:
2125 error = SET_FPEMU_CTL(me, arg2);
2126 break;
2127 case PR_GET_FPEMU:
2128 error = GET_FPEMU_CTL(me, arg2);
2129 break;
2130 case PR_SET_FPEXC:
2131 error = SET_FPEXC_CTL(me, arg2);
2132 break;
2133 case PR_GET_FPEXC:
2134 error = GET_FPEXC_CTL(me, arg2);
2135 break;
2136 case PR_GET_TIMING:
2137 error = PR_TIMING_STATISTICAL;
2138 break;
2139 case PR_SET_TIMING:
2140 if (arg2 != PR_TIMING_STATISTICAL)
2141 error = -EINVAL;
2142 break;
2143 case PR_SET_NAME:
2144 comm[sizeof(me->comm) - 1] = 0;
2145 if (strncpy_from_user(comm, (char __user *)arg2,
2146 sizeof(me->comm) - 1) < 0)
2147 return -EFAULT;
2148 set_task_comm(me, comm);
2149 proc_comm_connector(me);
2150 break;
2151 case PR_GET_NAME:
2152 get_task_comm(comm, me);
2153 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2154 return -EFAULT;
2155 break;
2156 case PR_GET_ENDIAN:
2157 error = GET_ENDIAN(me, arg2);
2158 break;
2159 case PR_SET_ENDIAN:
2160 error = SET_ENDIAN(me, arg2);
2161 break;
2162 case PR_GET_SECCOMP:
2163 error = prctl_get_seccomp();
2164 break;
2165 case PR_SET_SECCOMP:
2166 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2167 break;
2168 case PR_GET_TSC:
2169 error = GET_TSC_CTL(arg2);
2170 break;
2171 case PR_SET_TSC:
2172 error = SET_TSC_CTL(arg2);
2173 break;
2174 case PR_TASK_PERF_EVENTS_DISABLE:
2175 error = perf_event_task_disable();
2176 break;
2177 case PR_TASK_PERF_EVENTS_ENABLE:
2178 error = perf_event_task_enable();
2179 break;
2180 case PR_GET_TIMERSLACK:
2181 error = current->timer_slack_ns;
2182 break;
2183 case PR_SET_TIMERSLACK:
2184 if (arg2 <= 0)
2185 current->timer_slack_ns =
2186 current->default_timer_slack_ns;
2187 else
2188 current->timer_slack_ns = arg2;
2189 break;
2190 case PR_MCE_KILL:
2191 if (arg4 | arg5)
2192 return -EINVAL;
2193 switch (arg2) {
2194 case PR_MCE_KILL_CLEAR:
2195 if (arg3 != 0)
2196 return -EINVAL;
2197 current->flags &= ~PF_MCE_PROCESS;
2198 break;
2199 case PR_MCE_KILL_SET:
2200 current->flags |= PF_MCE_PROCESS;
2201 if (arg3 == PR_MCE_KILL_EARLY)
2202 current->flags |= PF_MCE_EARLY;
2203 else if (arg3 == PR_MCE_KILL_LATE)
2204 current->flags &= ~PF_MCE_EARLY;
2205 else if (arg3 == PR_MCE_KILL_DEFAULT)
2206 current->flags &=
2207 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2208 else
2209 return -EINVAL;
2210 break;
2211 default:
2212 return -EINVAL;
2213 }
2214 break;
2215 case PR_MCE_KILL_GET:
2216 if (arg2 | arg3 | arg4 | arg5)
2217 return -EINVAL;
2218 if (current->flags & PF_MCE_PROCESS)
2219 error = (current->flags & PF_MCE_EARLY) ?
2220 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2221 else
2222 error = PR_MCE_KILL_DEFAULT;
2223 break;
2224 case PR_SET_MM:
2225 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2226 break;
2227 case PR_GET_TID_ADDRESS:
2228 error = prctl_get_tid_address(me, (int __user **)arg2);
2229 break;
2230 case PR_SET_CHILD_SUBREAPER:
2231 me->signal->is_child_subreaper = !!arg2;
2232 break;
2233 case PR_GET_CHILD_SUBREAPER:
2234 error = put_user(me->signal->is_child_subreaper,
2235 (int __user *)arg2);
2236 break;
2237 case PR_SET_NO_NEW_PRIVS:
2238 if (arg2 != 1 || arg3 || arg4 || arg5)
2239 return -EINVAL;
2240
2241 task_set_no_new_privs(current);
2242 break;
2243 case PR_GET_NO_NEW_PRIVS:
2244 if (arg2 || arg3 || arg4 || arg5)
2245 return -EINVAL;
2246 return task_no_new_privs(current) ? 1 : 0;
2247 case PR_GET_THP_DISABLE:
2248 if (arg2 || arg3 || arg4 || arg5)
2249 return -EINVAL;
2250 error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2251 break;
2252 case PR_SET_THP_DISABLE:
2253 if (arg3 || arg4 || arg5)
2254 return -EINVAL;
2255 down_write(&me->mm->mmap_sem);
2256 if (arg2)
2257 me->mm->def_flags |= VM_NOHUGEPAGE;
2258 else
2259 me->mm->def_flags &= ~VM_NOHUGEPAGE;
2260 up_write(&me->mm->mmap_sem);
2261 break;
2262 case PR_MPX_ENABLE_MANAGEMENT:
2263 if (arg2 || arg3 || arg4 || arg5)
2264 return -EINVAL;
2265 error = MPX_ENABLE_MANAGEMENT();
2266 break;
2267 case PR_MPX_DISABLE_MANAGEMENT:
2268 if (arg2 || arg3 || arg4 || arg5)
2269 return -EINVAL;
2270 error = MPX_DISABLE_MANAGEMENT();
2271 break;
2272 case PR_SET_FP_MODE:
2273 error = SET_FP_MODE(me, arg2);
2274 break;
2275 case PR_GET_FP_MODE:
2276 error = GET_FP_MODE(me);
2277 break;
2278 case PR_GET_SPECULATION_CTRL:
2279 if (arg3 || arg4 || arg5)
2280 return -EINVAL;
2281 error = arch_prctl_spec_ctrl_get(me, arg2);
2282 break;
2283 case PR_SET_SPECULATION_CTRL:
2284 if (arg4 || arg5)
2285 return -EINVAL;
2286 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2287 break;
2288 default:
2289 error = -EINVAL;
2290 break;
2291 }
2292 return error;
2293 }
2294
2295 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2296 struct getcpu_cache __user *, unused)
2297 {
2298 int err = 0;
2299 int cpu = raw_smp_processor_id();
2300
2301 if (cpup)
2302 err |= put_user(cpu, cpup);
2303 if (nodep)
2304 err |= put_user(cpu_to_node(cpu), nodep);
2305 return err ? -EFAULT : 0;
2306 }
2307
2308 /**
2309 * do_sysinfo - fill in sysinfo struct
2310 * @info: pointer to buffer to fill
2311 */
2312 static int do_sysinfo(struct sysinfo *info)
2313 {
2314 unsigned long mem_total, sav_total;
2315 unsigned int mem_unit, bitcount;
2316 struct timespec tp;
2317
2318 memset(info, 0, sizeof(struct sysinfo));
2319
2320 get_monotonic_boottime(&tp);
2321 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2322
2323 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2324
2325 info->procs = nr_threads;
2326
2327 si_meminfo(info);
2328 si_swapinfo(info);
2329
2330 /*
2331 * If the sum of all the available memory (i.e. ram + swap)
2332 * is less than can be stored in a 32 bit unsigned long then
2333 * we can be binary compatible with 2.2.x kernels. If not,
2334 * well, in that case 2.2.x was broken anyways...
2335 *
2336 * -Erik Andersen <andersee@debian.org>
2337 */
2338
2339 mem_total = info->totalram + info->totalswap;
2340 if (mem_total < info->totalram || mem_total < info->totalswap)
2341 goto out;
2342 bitcount = 0;
2343 mem_unit = info->mem_unit;
2344 while (mem_unit > 1) {
2345 bitcount++;
2346 mem_unit >>= 1;
2347 sav_total = mem_total;
2348 mem_total <<= 1;
2349 if (mem_total < sav_total)
2350 goto out;
2351 }
2352
2353 /*
2354 * If mem_total did not overflow, multiply all memory values by
2355 * info->mem_unit and set it to 1. This leaves things compatible
2356 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2357 * kernels...
2358 */
2359
2360 info->mem_unit = 1;
2361 info->totalram <<= bitcount;
2362 info->freeram <<= bitcount;
2363 info->sharedram <<= bitcount;
2364 info->bufferram <<= bitcount;
2365 info->totalswap <<= bitcount;
2366 info->freeswap <<= bitcount;
2367 info->totalhigh <<= bitcount;
2368 info->freehigh <<= bitcount;
2369
2370 out:
2371 return 0;
2372 }
2373
2374 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2375 {
2376 struct sysinfo val;
2377
2378 do_sysinfo(&val);
2379
2380 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2381 return -EFAULT;
2382
2383 return 0;
2384 }
2385
2386 #ifdef CONFIG_COMPAT
2387 struct compat_sysinfo {
2388 s32 uptime;
2389 u32 loads[3];
2390 u32 totalram;
2391 u32 freeram;
2392 u32 sharedram;
2393 u32 bufferram;
2394 u32 totalswap;
2395 u32 freeswap;
2396 u16 procs;
2397 u16 pad;
2398 u32 totalhigh;
2399 u32 freehigh;
2400 u32 mem_unit;
2401 char _f[20-2*sizeof(u32)-sizeof(int)];
2402 };
2403
2404 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2405 {
2406 struct sysinfo s;
2407
2408 do_sysinfo(&s);
2409
2410 /* Check to see if any memory value is too large for 32-bit and scale
2411 * down if needed
2412 */
2413 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2414 int bitcount = 0;
2415
2416 while (s.mem_unit < PAGE_SIZE) {
2417 s.mem_unit <<= 1;
2418 bitcount++;
2419 }
2420
2421 s.totalram >>= bitcount;
2422 s.freeram >>= bitcount;
2423 s.sharedram >>= bitcount;
2424 s.bufferram >>= bitcount;
2425 s.totalswap >>= bitcount;
2426 s.freeswap >>= bitcount;
2427 s.totalhigh >>= bitcount;
2428 s.freehigh >>= bitcount;
2429 }
2430
2431 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2432 __put_user(s.uptime, &info->uptime) ||
2433 __put_user(s.loads[0], &info->loads[0]) ||
2434 __put_user(s.loads[1], &info->loads[1]) ||
2435 __put_user(s.loads[2], &info->loads[2]) ||
2436 __put_user(s.totalram, &info->totalram) ||
2437 __put_user(s.freeram, &info->freeram) ||
2438 __put_user(s.sharedram, &info->sharedram) ||
2439 __put_user(s.bufferram, &info->bufferram) ||
2440 __put_user(s.totalswap, &info->totalswap) ||
2441 __put_user(s.freeswap, &info->freeswap) ||
2442 __put_user(s.procs, &info->procs) ||
2443 __put_user(s.totalhigh, &info->totalhigh) ||
2444 __put_user(s.freehigh, &info->freehigh) ||
2445 __put_user(s.mem_unit, &info->mem_unit))
2446 return -EFAULT;
2447
2448 return 0;
2449 }
2450 #endif /* CONFIG_COMPAT */
Linux with added FPC-III support