?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
powerpc/crypto: add compression support to arch vec
[linux-2.6.git] / kernel / sys.c
blob241507f23eca097871bec58f5976476f352f3d75
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/kexec.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/file.h>
40 #include <linux/mount.h>
41 #include <linux/gfp.h>
42 #include <linux/syscore_ops.h>
43 #include <linux/version.h>
44 #include <linux/ctype.h>
45
46 #include <linux/compat.h>
47 #include <linux/syscalls.h>
48 #include <linux/kprobes.h>
49 #include <linux/user_namespace.h>
50
51 #include <linux/kmsg_dump.h>
52 /* Move somewhere else to avoid recompiling? */
53 #include <generated/utsrelease.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/io.h>
57 #include <asm/unistd.h>
58
59 #ifndef SET_UNALIGN_CTL
60 # define SET_UNALIGN_CTL(a,b) (-EINVAL)
61 #endif
62 #ifndef GET_UNALIGN_CTL
63 # define GET_UNALIGN_CTL(a,b) (-EINVAL)
64 #endif
65 #ifndef SET_FPEMU_CTL
66 # define SET_FPEMU_CTL(a,b) (-EINVAL)
67 #endif
68 #ifndef GET_FPEMU_CTL
69 # define GET_FPEMU_CTL(a,b) (-EINVAL)
70 #endif
71 #ifndef SET_FPEXC_CTL
72 # define SET_FPEXC_CTL(a,b) (-EINVAL)
73 #endif
74 #ifndef GET_FPEXC_CTL
75 # define GET_FPEXC_CTL(a,b) (-EINVAL)
76 #endif
77 #ifndef GET_ENDIAN
78 # define GET_ENDIAN(a,b) (-EINVAL)
79 #endif
80 #ifndef SET_ENDIAN
81 # define SET_ENDIAN(a,b) (-EINVAL)
82 #endif
83 #ifndef GET_TSC_CTL
84 # define GET_TSC_CTL(a) (-EINVAL)
85 #endif
86 #ifndef SET_TSC_CTL
87 # define SET_TSC_CTL(a) (-EINVAL)
88 #endif
89
90 /*
91 * this is where the system-wide overflow UID and GID are defined, for
92 * architectures that now have 32-bit UID/GID but didn't in the past
93 */
94
95 int overflowuid = DEFAULT_OVERFLOWUID;
96 int overflowgid = DEFAULT_OVERFLOWGID;
97
98 EXPORT_SYMBOL(overflowuid);
99 EXPORT_SYMBOL(overflowgid);
100
101 /*
102 * the same as above, but for filesystems which can only store a 16-bit
103 * UID and GID. as such, this is needed on all architectures
104 */
105
106 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
107 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
108
109 EXPORT_SYMBOL(fs_overflowuid);
110 EXPORT_SYMBOL(fs_overflowgid);
111
112 /*
113 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
114 */
115
116 int C_A_D = 1;
117 struct pid *cad_pid;
118 EXPORT_SYMBOL(cad_pid);
119
120 /*
121 * If set, this is used for preparing the system to power off.
122 */
123
124 void (*pm_power_off_prepare)(void);
125
126 /*
127 * Returns true if current's euid is same as p's uid or euid,
128 * or has CAP_SYS_NICE to p's user_ns.
129 *
130 * Called with rcu_read_lock, creds are safe
131 */
132 static bool set_one_prio_perm(struct task_struct *p)
133 {
134 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
135
136 if (uid_eq(pcred->uid, cred->euid) ||
137 uid_eq(pcred->euid, cred->euid))
138 return true;
139 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
140 return true;
141 return false;
142 }
143
144 /*
145 * set the priority of a task
146 * - the caller must hold the RCU read lock
147 */
148 static int set_one_prio(struct task_struct *p, int niceval, int error)
149 {
150 int no_nice;
151
152 if (!set_one_prio_perm(p)) {
153 error = -EPERM;
154 goto out;
155 }
156 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
157 error = -EACCES;
158 goto out;
159 }
160 no_nice = security_task_setnice(p, niceval);
161 if (no_nice) {
162 error = no_nice;
163 goto out;
164 }
165 if (error == -ESRCH)
166 error = 0;
167 set_user_nice(p, niceval);
168 out:
169 return error;
170 }
171
172 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
173 {
174 struct task_struct *g, *p;
175 struct user_struct *user;
176 const struct cred *cred = current_cred();
177 int error = -EINVAL;
178 struct pid *pgrp;
179 kuid_t uid;
180
181 if (which > PRIO_USER || which < PRIO_PROCESS)
182 goto out;
183
184 /* normalize: avoid signed division (rounding problems) */
185 error = -ESRCH;
186 if (niceval < -20)
187 niceval = -20;
188 if (niceval > 19)
189 niceval = 19;
190
191 rcu_read_lock();
192 read_lock(&tasklist_lock);
193 switch (which) {
194 case PRIO_PROCESS:
195 if (who)
196 p = find_task_by_vpid(who);
197 else
198 p = current;
199 if (p)
200 error = set_one_prio(p, niceval, error);
201 break;
202 case PRIO_PGRP:
203 if (who)
204 pgrp = find_vpid(who);
205 else
206 pgrp = task_pgrp(current);
207 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
208 error = set_one_prio(p, niceval, error);
209 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
210 break;
211 case PRIO_USER:
212 uid = make_kuid(cred->user_ns, who);
213 user = cred->user;
214 if (!who)
215 uid = cred->uid;
216 else if (!uid_eq(uid, cred->uid) &&
217 !(user = find_user(uid)))
218 goto out_unlock; /* No processes for this user */
219
220 do_each_thread(g, p) {
221 if (uid_eq(task_uid(p), uid))
222 error = set_one_prio(p, niceval, error);
223 } while_each_thread(g, p);
224 if (!uid_eq(uid, cred->uid))
225 free_uid(user); /* For find_user() */
226 break;
227 }
228 out_unlock:
229 read_unlock(&tasklist_lock);
230 rcu_read_unlock();
231 out:
232 return error;
233 }
234
235 /*
236 * Ugh. To avoid negative return values, "getpriority()" will
237 * not return the normal nice-value, but a negated value that
238 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
239 * to stay compatible.
240 */
241 SYSCALL_DEFINE2(getpriority, int, which, int, who)
242 {
243 struct task_struct *g, *p;
244 struct user_struct *user;
245 const struct cred *cred = current_cred();
246 long niceval, retval = -ESRCH;
247 struct pid *pgrp;
248 kuid_t uid;
249
250 if (which > PRIO_USER || which < PRIO_PROCESS)
251 return -EINVAL;
252
253 rcu_read_lock();
254 read_lock(&tasklist_lock);
255 switch (which) {
256 case PRIO_PROCESS:
257 if (who)
258 p = find_task_by_vpid(who);
259 else
260 p = current;
261 if (p) {
262 niceval = 20 - task_nice(p);
263 if (niceval > retval)
264 retval = niceval;
265 }
266 break;
267 case PRIO_PGRP:
268 if (who)
269 pgrp = find_vpid(who);
270 else
271 pgrp = task_pgrp(current);
272 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
273 niceval = 20 - task_nice(p);
274 if (niceval > retval)
275 retval = niceval;
276 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
277 break;
278 case PRIO_USER:
279 uid = make_kuid(cred->user_ns, who);
280 user = cred->user;
281 if (!who)
282 uid = cred->uid;
283 else if (!uid_eq(uid, cred->uid) &&
284 !(user = find_user(uid)))
285 goto out_unlock; /* No processes for this user */
286
287 do_each_thread(g, p) {
288 if (uid_eq(task_uid(p), uid)) {
289 niceval = 20 - task_nice(p);
290 if (niceval > retval)
291 retval = niceval;
292 }
293 } while_each_thread(g, p);
294 if (!uid_eq(uid, cred->uid))
295 free_uid(user); /* for find_user() */
296 break;
297 }
298 out_unlock:
299 read_unlock(&tasklist_lock);
300 rcu_read_unlock();
301
302 return retval;
303 }
304
305 /**
306 * emergency_restart - reboot the system
307 *
308 * Without shutting down any hardware or taking any locks
309 * reboot the system. This is called when we know we are in
310 * trouble so this is our best effort to reboot. This is
311 * safe to call in interrupt context.
312 */
313 void emergency_restart(void)
314 {
315 kmsg_dump(KMSG_DUMP_EMERG);
316 machine_emergency_restart();
317 }
318 EXPORT_SYMBOL_GPL(emergency_restart);
319
320 void kernel_restart_prepare(char *cmd)
321 {
322 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
323 system_state = SYSTEM_RESTART;
324 usermodehelper_disable();
325 device_shutdown();
326 syscore_shutdown();
327 }
328
329 /**
330 * register_reboot_notifier - Register function to be called at reboot time
331 * @nb: Info about notifier function to be called
332 *
333 * Registers a function with the list of functions
334 * to be called at reboot time.
335 *
336 * Currently always returns zero, as blocking_notifier_chain_register()
337 * always returns zero.
338 */
339 int register_reboot_notifier(struct notifier_block *nb)
340 {
341 return blocking_notifier_chain_register(&reboot_notifier_list, nb);
342 }
343 EXPORT_SYMBOL(register_reboot_notifier);
344
345 /**
346 * unregister_reboot_notifier - Unregister previously registered reboot notifier
347 * @nb: Hook to be unregistered
348 *
349 * Unregisters a previously registered reboot
350 * notifier function.
351 *
352 * Returns zero on success, or %-ENOENT on failure.
353 */
354 int unregister_reboot_notifier(struct notifier_block *nb)
355 {
356 return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
357 }
358 EXPORT_SYMBOL(unregister_reboot_notifier);
359
360 /**
361 * kernel_restart - reboot the system
362 * @cmd: pointer to buffer containing command to execute for restart
363 * or %NULL
364 *
365 * Shutdown everything and perform a clean reboot.
366 * This is not safe to call in interrupt context.
367 */
368 void kernel_restart(char *cmd)
369 {
370 kernel_restart_prepare(cmd);
371 if (!cmd)
372 printk(KERN_EMERG "Restarting system.\n");
373 else
374 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
375 kmsg_dump(KMSG_DUMP_RESTART);
376 machine_restart(cmd);
377 }
378 EXPORT_SYMBOL_GPL(kernel_restart);
379
380 static void kernel_shutdown_prepare(enum system_states state)
381 {
382 blocking_notifier_call_chain(&reboot_notifier_list,
383 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
384 system_state = state;
385 usermodehelper_disable();
386 device_shutdown();
387 }
388 /**
389 * kernel_halt - halt the system
390 *
391 * Shutdown everything and perform a clean system halt.
392 */
393 void kernel_halt(void)
394 {
395 kernel_shutdown_prepare(SYSTEM_HALT);
396 syscore_shutdown();
397 printk(KERN_EMERG "System halted.\n");
398 kmsg_dump(KMSG_DUMP_HALT);
399 machine_halt();
400 }
401
402 EXPORT_SYMBOL_GPL(kernel_halt);
403
404 /**
405 * kernel_power_off - power_off the system
406 *
407 * Shutdown everything and perform a clean system power_off.
408 */
409 void kernel_power_off(void)
410 {
411 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
412 if (pm_power_off_prepare)
413 pm_power_off_prepare();
414 disable_nonboot_cpus();
415 syscore_shutdown();
416 printk(KERN_EMERG "Power down.\n");
417 kmsg_dump(KMSG_DUMP_POWEROFF);
418 machine_power_off();
419 }
420 EXPORT_SYMBOL_GPL(kernel_power_off);
421
422 static DEFINE_MUTEX(reboot_mutex);
423
424 /*
425 * Reboot system call: for obvious reasons only root may call it,
426 * and even root needs to set up some magic numbers in the registers
427 * so that some mistake won't make this reboot the whole machine.
428 * You can also set the meaning of the ctrl-alt-del-key here.
429 *
430 * reboot doesn't sync: do that yourself before calling this.
431 */
432 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
433 void __user *, arg)
434 {
435 char buffer[256];
436 int ret = 0;
437
438 /* We only trust the superuser with rebooting the system. */
439 if (!capable(CAP_SYS_BOOT))
440 return -EPERM;
441
442 /* For safety, we require "magic" arguments. */
443 if (magic1 != LINUX_REBOOT_MAGIC1 ||
444 (magic2 != LINUX_REBOOT_MAGIC2 &&
445 magic2 != LINUX_REBOOT_MAGIC2A &&
446 magic2 != LINUX_REBOOT_MAGIC2B &&
447 magic2 != LINUX_REBOOT_MAGIC2C))
448 return -EINVAL;
449
450 /*
451 * If pid namespaces are enabled and the current task is in a child
452 * pid_namespace, the command is handled by reboot_pid_ns() which will
453 * call do_exit().
454 */
455 ret = reboot_pid_ns(task_active_pid_ns(current), cmd);
456 if (ret)
457 return ret;
458
459 /* Instead of trying to make the power_off code look like
460 * halt when pm_power_off is not set do it the easy way.
461 */
462 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
463 cmd = LINUX_REBOOT_CMD_HALT;
464
465 mutex_lock(&reboot_mutex);
466 switch (cmd) {
467 case LINUX_REBOOT_CMD_RESTART:
468 kernel_restart(NULL);
469 break;
470
471 case LINUX_REBOOT_CMD_CAD_ON:
472 C_A_D = 1;
473 break;
474
475 case LINUX_REBOOT_CMD_CAD_OFF:
476 C_A_D = 0;
477 break;
478
479 case LINUX_REBOOT_CMD_HALT:
480 kernel_halt();
481 do_exit(0);
482 panic("cannot halt");
483
484 case LINUX_REBOOT_CMD_POWER_OFF:
485 kernel_power_off();
486 do_exit(0);
487 break;
488
489 case LINUX_REBOOT_CMD_RESTART2:
490 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
491 ret = -EFAULT;
492 break;
493 }
494 buffer[sizeof(buffer) - 1] = '\0';
495
496 kernel_restart(buffer);
497 break;
498
499 #ifdef CONFIG_KEXEC
500 case LINUX_REBOOT_CMD_KEXEC:
501 ret = kernel_kexec();
502 break;
503 #endif
504
505 #ifdef CONFIG_HIBERNATION
506 case LINUX_REBOOT_CMD_SW_SUSPEND:
507 ret = hibernate();
508 break;
509 #endif
510
511 default:
512 ret = -EINVAL;
513 break;
514 }
515 mutex_unlock(&reboot_mutex);
516 return ret;
517 }
518
519 static void deferred_cad(struct work_struct *dummy)
520 {
521 kernel_restart(NULL);
522 }
523
524 /*
525 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
526 * As it's called within an interrupt, it may NOT sync: the only choice
527 * is whether to reboot at once, or just ignore the ctrl-alt-del.
528 */
529 void ctrl_alt_del(void)
530 {
531 static DECLARE_WORK(cad_work, deferred_cad);
532
533 if (C_A_D)
534 schedule_work(&cad_work);
535 else
536 kill_cad_pid(SIGINT, 1);
537 }
538
539 /*
540 * Unprivileged users may change the real gid to the effective gid
541 * or vice versa. (BSD-style)
542 *
543 * If you set the real gid at all, or set the effective gid to a value not
544 * equal to the real gid, then the saved gid is set to the new effective gid.
545 *
546 * This makes it possible for a setgid program to completely drop its
547 * privileges, which is often a useful assertion to make when you are doing
548 * a security audit over a program.
549 *
550 * The general idea is that a program which uses just setregid() will be
551 * 100% compatible with BSD. A program which uses just setgid() will be
552 * 100% compatible with POSIX with saved IDs.
553 *
554 * SMP: There are not races, the GIDs are checked only by filesystem
555 * operations (as far as semantic preservation is concerned).
556 */
557 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
558 {
559 struct user_namespace *ns = current_user_ns();
560 const struct cred *old;
561 struct cred *new;
562 int retval;
563 kgid_t krgid, kegid;
564
565 krgid = make_kgid(ns, rgid);
566 kegid = make_kgid(ns, egid);
567
568 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
569 return -EINVAL;
570 if ((egid != (gid_t) -1) && !gid_valid(kegid))
571 return -EINVAL;
572
573 new = prepare_creds();
574 if (!new)
575 return -ENOMEM;
576 old = current_cred();
577
578 retval = -EPERM;
579 if (rgid != (gid_t) -1) {
580 if (gid_eq(old->gid, krgid) ||
581 gid_eq(old->egid, krgid) ||
582 nsown_capable(CAP_SETGID))
583 new->gid = krgid;
584 else
585 goto error;
586 }
587 if (egid != (gid_t) -1) {
588 if (gid_eq(old->gid, kegid) ||
589 gid_eq(old->egid, kegid) ||
590 gid_eq(old->sgid, kegid) ||
591 nsown_capable(CAP_SETGID))
592 new->egid = kegid;
593 else
594 goto error;
595 }
596
597 if (rgid != (gid_t) -1 ||
598 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
599 new->sgid = new->egid;
600 new->fsgid = new->egid;
601
602 return commit_creds(new);
603
604 error:
605 abort_creds(new);
606 return retval;
607 }
608
609 /*
610 * setgid() is implemented like SysV w/ SAVED_IDS
611 *
612 * SMP: Same implicit races as above.
613 */
614 SYSCALL_DEFINE1(setgid, gid_t, gid)
615 {
616 struct user_namespace *ns = current_user_ns();
617 const struct cred *old;
618 struct cred *new;
619 int retval;
620 kgid_t kgid;
621
622 kgid = make_kgid(ns, gid);
623 if (!gid_valid(kgid))
624 return -EINVAL;
625
626 new = prepare_creds();
627 if (!new)
628 return -ENOMEM;
629 old = current_cred();
630
631 retval = -EPERM;
632 if (nsown_capable(CAP_SETGID))
633 new->gid = new->egid = new->sgid = new->fsgid = kgid;
634 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
635 new->egid = new->fsgid = kgid;
636 else
637 goto error;
638
639 return commit_creds(new);
640
641 error:
642 abort_creds(new);
643 return retval;
644 }
645
646 /*
647 * change the user struct in a credentials set to match the new UID
648 */
649 static int set_user(struct cred *new)
650 {
651 struct user_struct *new_user;
652
653 new_user = alloc_uid(new->uid);
654 if (!new_user)
655 return -EAGAIN;
656
657 /*
658 * We don't fail in case of NPROC limit excess here because too many
659 * poorly written programs don't check set*uid() return code, assuming
660 * it never fails if called by root. We may still enforce NPROC limit
661 * for programs doing set*uid()+execve() by harmlessly deferring the
662 * failure to the execve() stage.
663 */
664 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
665 new_user != INIT_USER)
666 current->flags |= PF_NPROC_EXCEEDED;
667 else
668 current->flags &= ~PF_NPROC_EXCEEDED;
669
670 free_uid(new->user);
671 new->user = new_user;
672 return 0;
673 }
674
675 /*
676 * Unprivileged users may change the real uid to the effective uid
677 * or vice versa. (BSD-style)
678 *
679 * If you set the real uid at all, or set the effective uid to a value not
680 * equal to the real uid, then the saved uid is set to the new effective uid.
681 *
682 * This makes it possible for a setuid program to completely drop its
683 * privileges, which is often a useful assertion to make when you are doing
684 * a security audit over a program.
685 *
686 * The general idea is that a program which uses just setreuid() will be
687 * 100% compatible with BSD. A program which uses just setuid() will be
688 * 100% compatible with POSIX with saved IDs.
689 */
690 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
691 {
692 struct user_namespace *ns = current_user_ns();
693 const struct cred *old;
694 struct cred *new;
695 int retval;
696 kuid_t kruid, keuid;
697
698 kruid = make_kuid(ns, ruid);
699 keuid = make_kuid(ns, euid);
700
701 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
702 return -EINVAL;
703 if ((euid != (uid_t) -1) && !uid_valid(keuid))
704 return -EINVAL;
705
706 new = prepare_creds();
707 if (!new)
708 return -ENOMEM;
709 old = current_cred();
710
711 retval = -EPERM;
712 if (ruid != (uid_t) -1) {
713 new->uid = kruid;
714 if (!uid_eq(old->uid, kruid) &&
715 !uid_eq(old->euid, kruid) &&
716 !nsown_capable(CAP_SETUID))
717 goto error;
718 }
719
720 if (euid != (uid_t) -1) {
721 new->euid = keuid;
722 if (!uid_eq(old->uid, keuid) &&
723 !uid_eq(old->euid, keuid) &&
724 !uid_eq(old->suid, keuid) &&
725 !nsown_capable(CAP_SETUID))
726 goto error;
727 }
728
729 if (!uid_eq(new->uid, old->uid)) {
730 retval = set_user(new);
731 if (retval < 0)
732 goto error;
733 }
734 if (ruid != (uid_t) -1 ||
735 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
736 new->suid = new->euid;
737 new->fsuid = new->euid;
738
739 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
740 if (retval < 0)
741 goto error;
742
743 return commit_creds(new);
744
745 error:
746 abort_creds(new);
747 return retval;
748 }
749
750 /*
751 * setuid() is implemented like SysV with SAVED_IDS
752 *
753 * Note that SAVED_ID's is deficient in that a setuid root program
754 * like sendmail, for example, cannot set its uid to be a normal
755 * user and then switch back, because if you're root, setuid() sets
756 * the saved uid too. If you don't like this, blame the bright people
757 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
758 * will allow a root program to temporarily drop privileges and be able to
759 * regain them by swapping the real and effective uid.
760 */
761 SYSCALL_DEFINE1(setuid, uid_t, uid)
762 {
763 struct user_namespace *ns = current_user_ns();
764 const struct cred *old;
765 struct cred *new;
766 int retval;
767 kuid_t kuid;
768
769 kuid = make_kuid(ns, uid);
770 if (!uid_valid(kuid))
771 return -EINVAL;
772
773 new = prepare_creds();
774 if (!new)
775 return -ENOMEM;
776 old = current_cred();
777
778 retval = -EPERM;
779 if (nsown_capable(CAP_SETUID)) {
780 new->suid = new->uid = kuid;
781 if (!uid_eq(kuid, old->uid)) {
782 retval = set_user(new);
783 if (retval < 0)
784 goto error;
785 }
786 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
787 goto error;
788 }
789
790 new->fsuid = new->euid = kuid;
791
792 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
793 if (retval < 0)
794 goto error;
795
796 return commit_creds(new);
797
798 error:
799 abort_creds(new);
800 return retval;
801 }
802
803
804 /*
805 * This function implements a generic ability to update ruid, euid,
806 * and suid. This allows you to implement the 4.4 compatible seteuid().
807 */
808 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
809 {
810 struct user_namespace *ns = current_user_ns();
811 const struct cred *old;
812 struct cred *new;
813 int retval;
814 kuid_t kruid, keuid, ksuid;
815
816 kruid = make_kuid(ns, ruid);
817 keuid = make_kuid(ns, euid);
818 ksuid = make_kuid(ns, suid);
819
820 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
821 return -EINVAL;
822
823 if ((euid != (uid_t) -1) && !uid_valid(keuid))
824 return -EINVAL;
825
826 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
827 return -EINVAL;
828
829 new = prepare_creds();
830 if (!new)
831 return -ENOMEM;
832
833 old = current_cred();
834
835 retval = -EPERM;
836 if (!nsown_capable(CAP_SETUID)) {
837 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
838 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
839 goto error;
840 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
841 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
842 goto error;
843 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
844 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
845 goto error;
846 }
847
848 if (ruid != (uid_t) -1) {
849 new->uid = kruid;
850 if (!uid_eq(kruid, old->uid)) {
851 retval = set_user(new);
852 if (retval < 0)
853 goto error;
854 }
855 }
856 if (euid != (uid_t) -1)
857 new->euid = keuid;
858 if (suid != (uid_t) -1)
859 new->suid = ksuid;
860 new->fsuid = new->euid;
861
862 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
863 if (retval < 0)
864 goto error;
865
866 return commit_creds(new);
867
868 error:
869 abort_creds(new);
870 return retval;
871 }
872
873 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
874 {
875 const struct cred *cred = current_cred();
876 int retval;
877 uid_t ruid, euid, suid;
878
879 ruid = from_kuid_munged(cred->user_ns, cred->uid);
880 euid = from_kuid_munged(cred->user_ns, cred->euid);
881 suid = from_kuid_munged(cred->user_ns, cred->suid);
882
883 if (!(retval = put_user(ruid, ruidp)) &&
884 !(retval = put_user(euid, euidp)))
885 retval = put_user(suid, suidp);
886
887 return retval;
888 }
889
890 /*
891 * Same as above, but for rgid, egid, sgid.
892 */
893 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
894 {
895 struct user_namespace *ns = current_user_ns();
896 const struct cred *old;
897 struct cred *new;
898 int retval;
899 kgid_t krgid, kegid, ksgid;
900
901 krgid = make_kgid(ns, rgid);
902 kegid = make_kgid(ns, egid);
903 ksgid = make_kgid(ns, sgid);
904
905 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
906 return -EINVAL;
907 if ((egid != (gid_t) -1) && !gid_valid(kegid))
908 return -EINVAL;
909 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
910 return -EINVAL;
911
912 new = prepare_creds();
913 if (!new)
914 return -ENOMEM;
915 old = current_cred();
916
917 retval = -EPERM;
918 if (!nsown_capable(CAP_SETGID)) {
919 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
920 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
921 goto error;
922 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
923 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
924 goto error;
925 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
926 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
927 goto error;
928 }
929
930 if (rgid != (gid_t) -1)
931 new->gid = krgid;
932 if (egid != (gid_t) -1)
933 new->egid = kegid;
934 if (sgid != (gid_t) -1)
935 new->sgid = ksgid;
936 new->fsgid = new->egid;
937
938 return commit_creds(new);
939
940 error:
941 abort_creds(new);
942 return retval;
943 }
944
945 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
946 {
947 const struct cred *cred = current_cred();
948 int retval;
949 gid_t rgid, egid, sgid;
950
951 rgid = from_kgid_munged(cred->user_ns, cred->gid);
952 egid = from_kgid_munged(cred->user_ns, cred->egid);
953 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
954
955 if (!(retval = put_user(rgid, rgidp)) &&
956 !(retval = put_user(egid, egidp)))
957 retval = put_user(sgid, sgidp);
958
959 return retval;
960 }
961
962
963 /*
964 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
965 * is used for "access()" and for the NFS daemon (letting nfsd stay at
966 * whatever uid it wants to). It normally shadows "euid", except when
967 * explicitly set by setfsuid() or for access..
968 */
969 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
970 {
971 const struct cred *old;
972 struct cred *new;
973 uid_t old_fsuid;
974 kuid_t kuid;
975
976 old = current_cred();
977 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
978
979 kuid = make_kuid(old->user_ns, uid);
980 if (!uid_valid(kuid))
981 return old_fsuid;
982
983 new = prepare_creds();
984 if (!new)
985 return old_fsuid;
986
987 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
988 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
989 nsown_capable(CAP_SETUID)) {
990 if (!uid_eq(kuid, old->fsuid)) {
991 new->fsuid = kuid;
992 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
993 goto change_okay;
994 }
995 }
996
997 abort_creds(new);
998 return old_fsuid;
999
1000 change_okay:
1001 commit_creds(new);
1002 return old_fsuid;
1003 }
1004
1005 /*
1006 * Samma på svenska..
1007 */
1008 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
1009 {
1010 const struct cred *old;
1011 struct cred *new;
1012 gid_t old_fsgid;
1013 kgid_t kgid;
1014
1015 old = current_cred();
1016 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
1017
1018 kgid = make_kgid(old->user_ns, gid);
1019 if (!gid_valid(kgid))
1020 return old_fsgid;
1021
1022 new = prepare_creds();
1023 if (!new)
1024 return old_fsgid;
1025
1026 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
1027 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
1028 nsown_capable(CAP_SETGID)) {
1029 if (!gid_eq(kgid, old->fsgid)) {
1030 new->fsgid = kgid;
1031 goto change_okay;
1032 }
1033 }
1034
1035 abort_creds(new);
1036 return old_fsgid;
1037
1038 change_okay:
1039 commit_creds(new);
1040 return old_fsgid;
1041 }
1042
1043 void do_sys_times(struct tms *tms)
1044 {
1045 cputime_t tgutime, tgstime, cutime, cstime;
1046
1047 spin_lock_irq(&current->sighand->siglock);
1048 thread_group_times(current, &tgutime, &tgstime);
1049 cutime = current->signal->cutime;
1050 cstime = current->signal->cstime;
1051 spin_unlock_irq(&current->sighand->siglock);
1052 tms->tms_utime = cputime_to_clock_t(tgutime);
1053 tms->tms_stime = cputime_to_clock_t(tgstime);
1054 tms->tms_cutime = cputime_to_clock_t(cutime);
1055 tms->tms_cstime = cputime_to_clock_t(cstime);
1056 }
1057
1058 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1059 {
1060 if (tbuf) {
1061 struct tms tmp;
1062
1063 do_sys_times(&tmp);
1064 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1065 return -EFAULT;
1066 }
1067 force_successful_syscall_return();
1068 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1069 }
1070
1071 /*
1072 * This needs some heavy checking ...
1073 * I just haven't the stomach for it. I also don't fully
1074 * understand sessions/pgrp etc. Let somebody who does explain it.
1075 *
1076 * OK, I think I have the protection semantics right.... this is really
1077 * only important on a multi-user system anyway, to make sure one user
1078 * can't send a signal to a process owned by another. -TYT, 12/12/91
1079 *
1080 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1081 * LBT 04.03.94
1082 */
1083 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1084 {
1085 struct task_struct *p;
1086 struct task_struct *group_leader = current->group_leader;
1087 struct pid *pgrp;
1088 int err;
1089
1090 if (!pid)
1091 pid = task_pid_vnr(group_leader);
1092 if (!pgid)
1093 pgid = pid;
1094 if (pgid < 0)
1095 return -EINVAL;
1096 rcu_read_lock();
1097
1098 /* From this point forward we keep holding onto the tasklist lock
1099 * so that our parent does not change from under us. -DaveM
1100 */
1101 write_lock_irq(&tasklist_lock);
1102
1103 err = -ESRCH;
1104 p = find_task_by_vpid(pid);
1105 if (!p)
1106 goto out;
1107
1108 err = -EINVAL;
1109 if (!thread_group_leader(p))
1110 goto out;
1111
1112 if (same_thread_group(p->real_parent, group_leader)) {
1113 err = -EPERM;
1114 if (task_session(p) != task_session(group_leader))
1115 goto out;
1116 err = -EACCES;
1117 if (p->did_exec)
1118 goto out;
1119 } else {
1120 err = -ESRCH;
1121 if (p != group_leader)
1122 goto out;
1123 }
1124
1125 err = -EPERM;
1126 if (p->signal->leader)
1127 goto out;
1128
1129 pgrp = task_pid(p);
1130 if (pgid != pid) {
1131 struct task_struct *g;
1132
1133 pgrp = find_vpid(pgid);
1134 g = pid_task(pgrp, PIDTYPE_PGID);
1135 if (!g || task_session(g) != task_session(group_leader))
1136 goto out;
1137 }
1138
1139 err = security_task_setpgid(p, pgid);
1140 if (err)
1141 goto out;
1142
1143 if (task_pgrp(p) != pgrp)
1144 change_pid(p, PIDTYPE_PGID, pgrp);
1145
1146 err = 0;
1147 out:
1148 /* All paths lead to here, thus we are safe. -DaveM */
1149 write_unlock_irq(&tasklist_lock);
1150 rcu_read_unlock();
1151 return err;
1152 }
1153
1154 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1155 {
1156 struct task_struct *p;
1157 struct pid *grp;
1158 int retval;
1159
1160 rcu_read_lock();
1161 if (!pid)
1162 grp = task_pgrp(current);
1163 else {
1164 retval = -ESRCH;
1165 p = find_task_by_vpid(pid);
1166 if (!p)
1167 goto out;
1168 grp = task_pgrp(p);
1169 if (!grp)
1170 goto out;
1171
1172 retval = security_task_getpgid(p);
1173 if (retval)
1174 goto out;
1175 }
1176 retval = pid_vnr(grp);
1177 out:
1178 rcu_read_unlock();
1179 return retval;
1180 }
1181
1182 #ifdef __ARCH_WANT_SYS_GETPGRP
1183
1184 SYSCALL_DEFINE0(getpgrp)
1185 {
1186 return sys_getpgid(0);
1187 }
1188
1189 #endif
1190
1191 SYSCALL_DEFINE1(getsid, pid_t, pid)
1192 {
1193 struct task_struct *p;
1194 struct pid *sid;
1195 int retval;
1196
1197 rcu_read_lock();
1198 if (!pid)
1199 sid = task_session(current);
1200 else {
1201 retval = -ESRCH;
1202 p = find_task_by_vpid(pid);
1203 if (!p)
1204 goto out;
1205 sid = task_session(p);
1206 if (!sid)
1207 goto out;
1208
1209 retval = security_task_getsid(p);
1210 if (retval)
1211 goto out;
1212 }
1213 retval = pid_vnr(sid);
1214 out:
1215 rcu_read_unlock();
1216 return retval;
1217 }
1218
1219 SYSCALL_DEFINE0(setsid)
1220 {
1221 struct task_struct *group_leader = current->group_leader;
1222 struct pid *sid = task_pid(group_leader);
1223 pid_t session = pid_vnr(sid);
1224 int err = -EPERM;
1225
1226 write_lock_irq(&tasklist_lock);
1227 /* Fail if I am already a session leader */
1228 if (group_leader->signal->leader)
1229 goto out;
1230
1231 /* Fail if a process group id already exists that equals the
1232 * proposed session id.
1233 */
1234 if (pid_task(sid, PIDTYPE_PGID))
1235 goto out;
1236
1237 group_leader->signal->leader = 1;
1238 __set_special_pids(sid);
1239
1240 proc_clear_tty(group_leader);
1241
1242 err = session;
1243 out:
1244 write_unlock_irq(&tasklist_lock);
1245 if (err > 0) {
1246 proc_sid_connector(group_leader);
1247 sched_autogroup_create_attach(group_leader);
1248 }
1249 return err;
1250 }
1251
1252 DECLARE_RWSEM(uts_sem);
1253
1254 #ifdef COMPAT_UTS_MACHINE
1255 #define override_architecture(name) \
1256 (personality(current->personality) == PER_LINUX32 && \
1257 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1258 sizeof(COMPAT_UTS_MACHINE)))
1259 #else
1260 #define override_architecture(name) 0
1261 #endif
1262
1263 /*
1264 * Work around broken programs that cannot handle "Linux 3.0".
1265 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1266 */
1267 static int override_release(char __user *release, int len)
1268 {
1269 int ret = 0;
1270 char buf[65];
1271
1272 if (current->personality & UNAME26) {
1273 char *rest = UTS_RELEASE;
1274 int ndots = 0;
1275 unsigned v;
1276
1277 while (*rest) {
1278 if (*rest == '.' && ++ndots >= 3)
1279 break;
1280 if (!isdigit(*rest) && *rest != '.')
1281 break;
1282 rest++;
1283 }
1284 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1285 snprintf(buf, len, "2.6.%u%s", v, rest);
1286 ret = copy_to_user(release, buf, len);
1287 }
1288 return ret;
1289 }
1290
1291 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1292 {
1293 int errno = 0;
1294
1295 down_read(&uts_sem);
1296 if (copy_to_user(name, utsname(), sizeof *name))
1297 errno = -EFAULT;
1298 up_read(&uts_sem);
1299
1300 if (!errno && override_release(name->release, sizeof(name->release)))
1301 errno = -EFAULT;
1302 if (!errno && override_architecture(name))
1303 errno = -EFAULT;
1304 return errno;
1305 }
1306
1307 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1308 /*
1309 * Old cruft
1310 */
1311 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1312 {
1313 int error = 0;
1314
1315 if (!name)
1316 return -EFAULT;
1317
1318 down_read(&uts_sem);
1319 if (copy_to_user(name, utsname(), sizeof(*name)))
1320 error = -EFAULT;
1321 up_read(&uts_sem);
1322
1323 if (!error && override_release(name->release, sizeof(name->release)))
1324 error = -EFAULT;
1325 if (!error && override_architecture(name))
1326 error = -EFAULT;
1327 return error;
1328 }
1329
1330 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1331 {
1332 int error;
1333
1334 if (!name)
1335 return -EFAULT;
1336 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1337 return -EFAULT;
1338
1339 down_read(&uts_sem);
1340 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1341 __OLD_UTS_LEN);
1342 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1343 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1344 __OLD_UTS_LEN);
1345 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1346 error |= __copy_to_user(&name->release, &utsname()->release,
1347 __OLD_UTS_LEN);
1348 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1349 error |= __copy_to_user(&name->version, &utsname()->version,
1350 __OLD_UTS_LEN);
1351 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1352 error |= __copy_to_user(&name->machine, &utsname()->machine,
1353 __OLD_UTS_LEN);
1354 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1355 up_read(&uts_sem);
1356
1357 if (!error && override_architecture(name))
1358 error = -EFAULT;
1359 if (!error && override_release(name->release, sizeof(name->release)))
1360 error = -EFAULT;
1361 return error ? -EFAULT : 0;
1362 }
1363 #endif
1364
1365 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1366 {
1367 int errno;
1368 char tmp[__NEW_UTS_LEN];
1369
1370 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1371 return -EPERM;
1372
1373 if (len < 0 || len > __NEW_UTS_LEN)
1374 return -EINVAL;
1375 down_write(&uts_sem);
1376 errno = -EFAULT;
1377 if (!copy_from_user(tmp, name, len)) {
1378 struct new_utsname *u = utsname();
1379
1380 memcpy(u->nodename, tmp, len);
1381 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1382 errno = 0;
1383 uts_proc_notify(UTS_PROC_HOSTNAME);
1384 }
1385 up_write(&uts_sem);
1386 return errno;
1387 }
1388
1389 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1390
1391 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1392 {
1393 int i, errno;
1394 struct new_utsname *u;
1395
1396 if (len < 0)
1397 return -EINVAL;
1398 down_read(&uts_sem);
1399 u = utsname();
1400 i = 1 + strlen(u->nodename);
1401 if (i > len)
1402 i = len;
1403 errno = 0;
1404 if (copy_to_user(name, u->nodename, i))
1405 errno = -EFAULT;
1406 up_read(&uts_sem);
1407 return errno;
1408 }
1409
1410 #endif
1411
1412 /*
1413 * Only setdomainname; getdomainname can be implemented by calling
1414 * uname()
1415 */
1416 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1417 {
1418 int errno;
1419 char tmp[__NEW_UTS_LEN];
1420
1421 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1422 return -EPERM;
1423 if (len < 0 || len > __NEW_UTS_LEN)
1424 return -EINVAL;
1425
1426 down_write(&uts_sem);
1427 errno = -EFAULT;
1428 if (!copy_from_user(tmp, name, len)) {
1429 struct new_utsname *u = utsname();
1430
1431 memcpy(u->domainname, tmp, len);
1432 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1433 errno = 0;
1434 uts_proc_notify(UTS_PROC_DOMAINNAME);
1435 }
1436 up_write(&uts_sem);
1437 return errno;
1438 }
1439
1440 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1441 {
1442 struct rlimit value;
1443 int ret;
1444
1445 ret = do_prlimit(current, resource, NULL, &value);
1446 if (!ret)
1447 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1448
1449 return ret;
1450 }
1451
1452 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1453
1454 /*
1455 * Back compatibility for getrlimit. Needed for some apps.
1456 */
1457
1458 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1459 struct rlimit __user *, rlim)
1460 {
1461 struct rlimit x;
1462 if (resource >= RLIM_NLIMITS)
1463 return -EINVAL;
1464
1465 task_lock(current->group_leader);
1466 x = current->signal->rlim[resource];
1467 task_unlock(current->group_leader);
1468 if (x.rlim_cur > 0x7FFFFFFF)
1469 x.rlim_cur = 0x7FFFFFFF;
1470 if (x.rlim_max > 0x7FFFFFFF)
1471 x.rlim_max = 0x7FFFFFFF;
1472 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1473 }
1474
1475 #endif
1476
1477 static inline bool rlim64_is_infinity(__u64 rlim64)
1478 {
1479 #if BITS_PER_LONG < 64
1480 return rlim64 >= ULONG_MAX;
1481 #else
1482 return rlim64 == RLIM64_INFINITY;
1483 #endif
1484 }
1485
1486 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1487 {
1488 if (rlim->rlim_cur == RLIM_INFINITY)
1489 rlim64->rlim_cur = RLIM64_INFINITY;
1490 else
1491 rlim64->rlim_cur = rlim->rlim_cur;
1492 if (rlim->rlim_max == RLIM_INFINITY)
1493 rlim64->rlim_max = RLIM64_INFINITY;
1494 else
1495 rlim64->rlim_max = rlim->rlim_max;
1496 }
1497
1498 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1499 {
1500 if (rlim64_is_infinity(rlim64->rlim_cur))
1501 rlim->rlim_cur = RLIM_INFINITY;
1502 else
1503 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1504 if (rlim64_is_infinity(rlim64->rlim_max))
1505 rlim->rlim_max = RLIM_INFINITY;
1506 else
1507 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1508 }
1509
1510 /* make sure you are allowed to change @tsk limits before calling this */
1511 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1512 struct rlimit *new_rlim, struct rlimit *old_rlim)
1513 {
1514 struct rlimit *rlim;
1515 int retval = 0;
1516
1517 if (resource >= RLIM_NLIMITS)
1518 return -EINVAL;
1519 if (new_rlim) {
1520 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1521 return -EINVAL;
1522 if (resource == RLIMIT_NOFILE &&
1523 new_rlim->rlim_max > sysctl_nr_open)
1524 return -EPERM;
1525 }
1526
1527 /* protect tsk->signal and tsk->sighand from disappearing */
1528 read_lock(&tasklist_lock);
1529 if (!tsk->sighand) {
1530 retval = -ESRCH;
1531 goto out;
1532 }
1533
1534 rlim = tsk->signal->rlim + resource;
1535 task_lock(tsk->group_leader);
1536 if (new_rlim) {
1537 /* Keep the capable check against init_user_ns until
1538 cgroups can contain all limits */
1539 if (new_rlim->rlim_max > rlim->rlim_max &&
1540 !capable(CAP_SYS_RESOURCE))
1541 retval = -EPERM;
1542 if (!retval)
1543 retval = security_task_setrlimit(tsk->group_leader,
1544 resource, new_rlim);
1545 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1546 /*
1547 * The caller is asking for an immediate RLIMIT_CPU
1548 * expiry. But we use the zero value to mean "it was
1549 * never set". So let's cheat and make it one second
1550 * instead
1551 */
1552 new_rlim->rlim_cur = 1;
1553 }
1554 }
1555 if (!retval) {
1556 if (old_rlim)
1557 *old_rlim = *rlim;
1558 if (new_rlim)
1559 *rlim = *new_rlim;
1560 }
1561 task_unlock(tsk->group_leader);
1562
1563 /*
1564 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1565 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1566 * very long-standing error, and fixing it now risks breakage of
1567 * applications, so we live with it
1568 */
1569 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1570 new_rlim->rlim_cur != RLIM_INFINITY)
1571 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1572 out:
1573 read_unlock(&tasklist_lock);
1574 return retval;
1575 }
1576
1577 /* rcu lock must be held */
1578 static int check_prlimit_permission(struct task_struct *task)
1579 {
1580 const struct cred *cred = current_cred(), *tcred;
1581
1582 if (current == task)
1583 return 0;
1584
1585 tcred = __task_cred(task);
1586 if (uid_eq(cred->uid, tcred->euid) &&
1587 uid_eq(cred->uid, tcred->suid) &&
1588 uid_eq(cred->uid, tcred->uid) &&
1589 gid_eq(cred->gid, tcred->egid) &&
1590 gid_eq(cred->gid, tcred->sgid) &&
1591 gid_eq(cred->gid, tcred->gid))
1592 return 0;
1593 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1594 return 0;
1595
1596 return -EPERM;
1597 }
1598
1599 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1600 const struct rlimit64 __user *, new_rlim,
1601 struct rlimit64 __user *, old_rlim)
1602 {
1603 struct rlimit64 old64, new64;
1604 struct rlimit old, new;
1605 struct task_struct *tsk;
1606 int ret;
1607
1608 if (new_rlim) {
1609 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1610 return -EFAULT;
1611 rlim64_to_rlim(&new64, &new);
1612 }
1613
1614 rcu_read_lock();
1615 tsk = pid ? find_task_by_vpid(pid) : current;
1616 if (!tsk) {
1617 rcu_read_unlock();
1618 return -ESRCH;
1619 }
1620 ret = check_prlimit_permission(tsk);
1621 if (ret) {
1622 rcu_read_unlock();
1623 return ret;
1624 }
1625 get_task_struct(tsk);
1626 rcu_read_unlock();
1627
1628 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1629 old_rlim ? &old : NULL);
1630
1631 if (!ret && old_rlim) {
1632 rlim_to_rlim64(&old, &old64);
1633 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1634 ret = -EFAULT;
1635 }
1636
1637 put_task_struct(tsk);
1638 return ret;
1639 }
1640
1641 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1642 {
1643 struct rlimit new_rlim;
1644
1645 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1646 return -EFAULT;
1647 return do_prlimit(current, resource, &new_rlim, NULL);
1648 }
1649
1650 /*
1651 * It would make sense to put struct rusage in the task_struct,
1652 * except that would make the task_struct be *really big*. After
1653 * task_struct gets moved into malloc'ed memory, it would
1654 * make sense to do this. It will make moving the rest of the information
1655 * a lot simpler! (Which we're not doing right now because we're not
1656 * measuring them yet).
1657 *
1658 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1659 * races with threads incrementing their own counters. But since word
1660 * reads are atomic, we either get new values or old values and we don't
1661 * care which for the sums. We always take the siglock to protect reading
1662 * the c* fields from p->signal from races with exit.c updating those
1663 * fields when reaping, so a sample either gets all the additions of a
1664 * given child after it's reaped, or none so this sample is before reaping.
1665 *
1666 * Locking:
1667 * We need to take the siglock for CHILDEREN, SELF and BOTH
1668 * for the cases current multithreaded, non-current single threaded
1669 * non-current multithreaded. Thread traversal is now safe with
1670 * the siglock held.
1671 * Strictly speaking, we donot need to take the siglock if we are current and
1672 * single threaded, as no one else can take our signal_struct away, no one
1673 * else can reap the children to update signal->c* counters, and no one else
1674 * can race with the signal-> fields. If we do not take any lock, the
1675 * signal-> fields could be read out of order while another thread was just
1676 * exiting. So we should place a read memory barrier when we avoid the lock.
1677 * On the writer side, write memory barrier is implied in __exit_signal
1678 * as __exit_signal releases the siglock spinlock after updating the signal->
1679 * fields. But we don't do this yet to keep things simple.
1680 *
1681 */
1682
1683 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1684 {
1685 r->ru_nvcsw += t->nvcsw;
1686 r->ru_nivcsw += t->nivcsw;
1687 r->ru_minflt += t->min_flt;
1688 r->ru_majflt += t->maj_flt;
1689 r->ru_inblock += task_io_get_inblock(t);
1690 r->ru_oublock += task_io_get_oublock(t);
1691 }
1692
1693 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1694 {
1695 struct task_struct *t;
1696 unsigned long flags;
1697 cputime_t tgutime, tgstime, utime, stime;
1698 unsigned long maxrss = 0;
1699
1700 memset((char *) r, 0, sizeof *r);
1701 utime = stime = 0;
1702
1703 if (who == RUSAGE_THREAD) {
1704 task_times(current, &utime, &stime);
1705 accumulate_thread_rusage(p, r);
1706 maxrss = p->signal->maxrss;
1707 goto out;
1708 }
1709
1710 if (!lock_task_sighand(p, &flags))
1711 return;
1712
1713 switch (who) {
1714 case RUSAGE_BOTH:
1715 case RUSAGE_CHILDREN:
1716 utime = p->signal->cutime;
1717 stime = p->signal->cstime;
1718 r->ru_nvcsw = p->signal->cnvcsw;
1719 r->ru_nivcsw = p->signal->cnivcsw;
1720 r->ru_minflt = p->signal->cmin_flt;
1721 r->ru_majflt = p->signal->cmaj_flt;
1722 r->ru_inblock = p->signal->cinblock;
1723 r->ru_oublock = p->signal->coublock;
1724 maxrss = p->signal->cmaxrss;
1725
1726 if (who == RUSAGE_CHILDREN)
1727 break;
1728
1729 case RUSAGE_SELF:
1730 thread_group_times(p, &tgutime, &tgstime);
1731 utime += tgutime;
1732 stime += tgstime;
1733 r->ru_nvcsw += p->signal->nvcsw;
1734 r->ru_nivcsw += p->signal->nivcsw;
1735 r->ru_minflt += p->signal->min_flt;
1736 r->ru_majflt += p->signal->maj_flt;
1737 r->ru_inblock += p->signal->inblock;
1738 r->ru_oublock += p->signal->oublock;
1739 if (maxrss < p->signal->maxrss)
1740 maxrss = p->signal->maxrss;
1741 t = p;
1742 do {
1743 accumulate_thread_rusage(t, r);
1744 t = next_thread(t);
1745 } while (t != p);
1746 break;
1747
1748 default:
1749 BUG();
1750 }
1751 unlock_task_sighand(p, &flags);
1752
1753 out:
1754 cputime_to_timeval(utime, &r->ru_utime);
1755 cputime_to_timeval(stime, &r->ru_stime);
1756
1757 if (who != RUSAGE_CHILDREN) {
1758 struct mm_struct *mm = get_task_mm(p);
1759 if (mm) {
1760 setmax_mm_hiwater_rss(&maxrss, mm);
1761 mmput(mm);
1762 }
1763 }
1764 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1765 }
1766
1767 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1768 {
1769 struct rusage r;
1770 k_getrusage(p, who, &r);
1771 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1772 }
1773
1774 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1775 {
1776 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1777 who != RUSAGE_THREAD)
1778 return -EINVAL;
1779 return getrusage(current, who, ru);
1780 }
1781
1782 SYSCALL_DEFINE1(umask, int, mask)
1783 {
1784 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1785 return mask;
1786 }
1787
1788 #ifdef CONFIG_CHECKPOINT_RESTORE
1789 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1790 {
1791 struct file *exe_file;
1792 struct dentry *dentry;
1793 int err;
1794
1795 exe_file = fget(fd);
1796 if (!exe_file)
1797 return -EBADF;
1798
1799 dentry = exe_file->f_path.dentry;
1800
1801 /*
1802 * Because the original mm->exe_file points to executable file, make
1803 * sure that this one is executable as well, to avoid breaking an
1804 * overall picture.
1805 */
1806 err = -EACCES;
1807 if (!S_ISREG(dentry->d_inode->i_mode) ||
1808 exe_file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1809 goto exit;
1810
1811 err = inode_permission(dentry->d_inode, MAY_EXEC);
1812 if (err)
1813 goto exit;
1814
1815 down_write(&mm->mmap_sem);
1816
1817 /*
1818 * Forbid mm->exe_file change if old file still mapped.
1819 */
1820 err = -EBUSY;
1821 if (mm->exe_file) {
1822 struct vm_area_struct *vma;
1823
1824 for (vma = mm->mmap; vma; vma = vma->vm_next)
1825 if (vma->vm_file &&
1826 path_equal(&vma->vm_file->f_path,
1827 &mm->exe_file->f_path))
1828 goto exit_unlock;
1829 }
1830
1831 /*
1832 * The symlink can be changed only once, just to disallow arbitrary
1833 * transitions malicious software might bring in. This means one
1834 * could make a snapshot over all processes running and monitor
1835 * /proc/pid/exe changes to notice unusual activity if needed.
1836 */
1837 err = -EPERM;
1838 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1839 goto exit_unlock;
1840
1841 err = 0;
1842 set_mm_exe_file(mm, exe_file);
1843 exit_unlock:
1844 up_write(&mm->mmap_sem);
1845
1846 exit:
1847 fput(exe_file);
1848 return err;
1849 }
1850
1851 static int prctl_set_mm(int opt, unsigned long addr,
1852 unsigned long arg4, unsigned long arg5)
1853 {
1854 unsigned long rlim = rlimit(RLIMIT_DATA);
1855 struct mm_struct *mm = current->mm;
1856 struct vm_area_struct *vma;
1857 int error;
1858
1859 if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
1860 return -EINVAL;
1861
1862 if (!capable(CAP_SYS_RESOURCE))
1863 return -EPERM;
1864
1865 if (opt == PR_SET_MM_EXE_FILE)
1866 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1867
1868 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1869 return -EINVAL;
1870
1871 error = -EINVAL;
1872
1873 down_read(&mm->mmap_sem);
1874 vma = find_vma(mm, addr);
1875
1876 switch (opt) {
1877 case PR_SET_MM_START_CODE:
1878 mm->start_code = addr;
1879 break;
1880 case PR_SET_MM_END_CODE:
1881 mm->end_code = addr;
1882 break;
1883 case PR_SET_MM_START_DATA:
1884 mm->start_data = addr;
1885 break;
1886 case PR_SET_MM_END_DATA:
1887 mm->end_data = addr;
1888 break;
1889
1890 case PR_SET_MM_START_BRK:
1891 if (addr <= mm->end_data)
1892 goto out;
1893
1894 if (rlim < RLIM_INFINITY &&
1895 (mm->brk - addr) +
1896 (mm->end_data - mm->start_data) > rlim)
1897 goto out;
1898
1899 mm->start_brk = addr;
1900 break;
1901
1902 case PR_SET_MM_BRK:
1903 if (addr <= mm->end_data)
1904 goto out;
1905
1906 if (rlim < RLIM_INFINITY &&
1907 (addr - mm->start_brk) +
1908 (mm->end_data - mm->start_data) > rlim)
1909 goto out;
1910
1911 mm->brk = addr;
1912 break;
1913
1914 /*
1915 * If command line arguments and environment
1916 * are placed somewhere else on stack, we can
1917 * set them up here, ARG_START/END to setup
1918 * command line argumets and ENV_START/END
1919 * for environment.
1920 */
1921 case PR_SET_MM_START_STACK:
1922 case PR_SET_MM_ARG_START:
1923 case PR_SET_MM_ARG_END:
1924 case PR_SET_MM_ENV_START:
1925 case PR_SET_MM_ENV_END:
1926 if (!vma) {
1927 error = -EFAULT;
1928 goto out;
1929 }
1930 if (opt == PR_SET_MM_START_STACK)
1931 mm->start_stack = addr;
1932 else if (opt == PR_SET_MM_ARG_START)
1933 mm->arg_start = addr;
1934 else if (opt == PR_SET_MM_ARG_END)
1935 mm->arg_end = addr;
1936 else if (opt == PR_SET_MM_ENV_START)
1937 mm->env_start = addr;
1938 else if (opt == PR_SET_MM_ENV_END)
1939 mm->env_end = addr;
1940 break;
1941
1942 /*
1943 * This doesn't move auxiliary vector itself
1944 * since it's pinned to mm_struct, but allow
1945 * to fill vector with new values. It's up
1946 * to a caller to provide sane values here
1947 * otherwise user space tools which use this
1948 * vector might be unhappy.
1949 */
1950 case PR_SET_MM_AUXV: {
1951 unsigned long user_auxv[AT_VECTOR_SIZE];
1952
1953 if (arg4 > sizeof(user_auxv))
1954 goto out;
1955 up_read(&mm->mmap_sem);
1956
1957 if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
1958 return -EFAULT;
1959
1960 /* Make sure the last entry is always AT_NULL */
1961 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1962 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1963
1964 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1965
1966 task_lock(current);
1967 memcpy(mm->saved_auxv, user_auxv, arg4);
1968 task_unlock(current);
1969
1970 return 0;
1971 }
1972 default:
1973 goto out;
1974 }
1975
1976 error = 0;
1977 out:
1978 up_read(&mm->mmap_sem);
1979 return error;
1980 }
1981
1982 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1983 {
1984 return put_user(me->clear_child_tid, tid_addr);
1985 }
1986
1987 #else /* CONFIG_CHECKPOINT_RESTORE */
1988 static int prctl_set_mm(int opt, unsigned long addr,
1989 unsigned long arg4, unsigned long arg5)
1990 {
1991 return -EINVAL;
1992 }
1993 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1994 {
1995 return -EINVAL;
1996 }
1997 #endif
1998
1999 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2000 unsigned long, arg4, unsigned long, arg5)
2001 {
2002 struct task_struct *me = current;
2003 unsigned char comm[sizeof(me->comm)];
2004 long error;
2005
2006 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2007 if (error != -ENOSYS)
2008 return error;
2009
2010 error = 0;
2011 switch (option) {
2012 case PR_SET_PDEATHSIG:
2013 if (!valid_signal(arg2)) {
2014 error = -EINVAL;
2015 break;
2016 }
2017 me->pdeath_signal = arg2;
2018 break;
2019 case PR_GET_PDEATHSIG:
2020 error = put_user(me->pdeath_signal, (int __user *)arg2);
2021 break;
2022 case PR_GET_DUMPABLE:
2023 error = get_dumpable(me->mm);
2024 break;
2025 case PR_SET_DUMPABLE:
2026 if (arg2 < 0 || arg2 > 1) {
2027 error = -EINVAL;
2028 break;
2029 }
2030 set_dumpable(me->mm, arg2);
2031 break;
2032
2033 case PR_SET_UNALIGN:
2034 error = SET_UNALIGN_CTL(me, arg2);
2035 break;
2036 case PR_GET_UNALIGN:
2037 error = GET_UNALIGN_CTL(me, arg2);
2038 break;
2039 case PR_SET_FPEMU:
2040 error = SET_FPEMU_CTL(me, arg2);
2041 break;
2042 case PR_GET_FPEMU:
2043 error = GET_FPEMU_CTL(me, arg2);
2044 break;
2045 case PR_SET_FPEXC:
2046 error = SET_FPEXC_CTL(me, arg2);
2047 break;
2048 case PR_GET_FPEXC:
2049 error = GET_FPEXC_CTL(me, arg2);
2050 break;
2051 case PR_GET_TIMING:
2052 error = PR_TIMING_STATISTICAL;
2053 break;
2054 case PR_SET_TIMING:
2055 if (arg2 != PR_TIMING_STATISTICAL)
2056 error = -EINVAL;
2057 break;
2058 case PR_SET_NAME:
2059 comm[sizeof(me->comm)-1] = 0;
2060 if (strncpy_from_user(comm, (char __user *)arg2,
2061 sizeof(me->comm) - 1) < 0)
2062 return -EFAULT;
2063 set_task_comm(me, comm);
2064 proc_comm_connector(me);
2065 break;
2066 case PR_GET_NAME:
2067 get_task_comm(comm, me);
2068 if (copy_to_user((char __user *)arg2, comm,
2069 sizeof(comm)))
2070 return -EFAULT;
2071 break;
2072 case PR_GET_ENDIAN:
2073 error = GET_ENDIAN(me, arg2);
2074 break;
2075 case PR_SET_ENDIAN:
2076 error = SET_ENDIAN(me, arg2);
2077 break;
2078 case PR_GET_SECCOMP:
2079 error = prctl_get_seccomp();
2080 break;
2081 case PR_SET_SECCOMP:
2082 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2083 break;
2084 case PR_GET_TSC:
2085 error = GET_TSC_CTL(arg2);
2086 break;
2087 case PR_SET_TSC:
2088 error = SET_TSC_CTL(arg2);
2089 break;
2090 case PR_TASK_PERF_EVENTS_DISABLE:
2091 error = perf_event_task_disable();
2092 break;
2093 case PR_TASK_PERF_EVENTS_ENABLE:
2094 error = perf_event_task_enable();
2095 break;
2096 case PR_GET_TIMERSLACK:
2097 error = current->timer_slack_ns;
2098 break;
2099 case PR_SET_TIMERSLACK:
2100 if (arg2 <= 0)
2101 current->timer_slack_ns =
2102 current->default_timer_slack_ns;
2103 else
2104 current->timer_slack_ns = arg2;
2105 break;
2106 case PR_MCE_KILL:
2107 if (arg4 | arg5)
2108 return -EINVAL;
2109 switch (arg2) {
2110 case PR_MCE_KILL_CLEAR:
2111 if (arg3 != 0)
2112 return -EINVAL;
2113 current->flags &= ~PF_MCE_PROCESS;
2114 break;
2115 case PR_MCE_KILL_SET:
2116 current->flags |= PF_MCE_PROCESS;
2117 if (arg3 == PR_MCE_KILL_EARLY)
2118 current->flags |= PF_MCE_EARLY;
2119 else if (arg3 == PR_MCE_KILL_LATE)
2120 current->flags &= ~PF_MCE_EARLY;
2121 else if (arg3 == PR_MCE_KILL_DEFAULT)
2122 current->flags &=
2123 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2124 else
2125 return -EINVAL;
2126 break;
2127 default:
2128 return -EINVAL;
2129 }
2130 break;
2131 case PR_MCE_KILL_GET:
2132 if (arg2 | arg3 | arg4 | arg5)
2133 return -EINVAL;
2134 if (current->flags & PF_MCE_PROCESS)
2135 error = (current->flags & PF_MCE_EARLY) ?
2136 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2137 else
2138 error = PR_MCE_KILL_DEFAULT;
2139 break;
2140 case PR_SET_MM:
2141 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2142 break;
2143 case PR_GET_TID_ADDRESS:
2144 error = prctl_get_tid_address(me, (int __user **)arg2);
2145 break;
2146 case PR_SET_CHILD_SUBREAPER:
2147 me->signal->is_child_subreaper = !!arg2;
2148 break;
2149 case PR_GET_CHILD_SUBREAPER:
2150 error = put_user(me->signal->is_child_subreaper,
2151 (int __user *) arg2);
2152 break;
2153 case PR_SET_NO_NEW_PRIVS:
2154 if (arg2 != 1 || arg3 || arg4 || arg5)
2155 return -EINVAL;
2156
2157 current->no_new_privs = 1;
2158 break;
2159 case PR_GET_NO_NEW_PRIVS:
2160 if (arg2 || arg3 || arg4 || arg5)
2161 return -EINVAL;
2162 return current->no_new_privs ? 1 : 0;
2163 default:
2164 error = -EINVAL;
2165 break;
2166 }
2167 return error;
2168 }
2169
2170 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2171 struct getcpu_cache __user *, unused)
2172 {
2173 int err = 0;
2174 int cpu = raw_smp_processor_id();
2175 if (cpup)
2176 err |= put_user(cpu, cpup);
2177 if (nodep)
2178 err |= put_user(cpu_to_node(cpu), nodep);
2179 return err ? -EFAULT : 0;
2180 }
2181
2182 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
2183
2184 static void argv_cleanup(struct subprocess_info *info)
2185 {
2186 argv_free(info->argv);
2187 }
2188
2189 static int __orderly_poweroff(void)
2190 {
2191 int argc;
2192 char **argv;
2193 static char *envp[] = {
2194 "HOME=/",
2195 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
2196 NULL
2197 };
2198 int ret;
2199
2200 argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
2201 if (argv == NULL) {
2202 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
2203 __func__, poweroff_cmd);
2204 return -ENOMEM;
2205 }
2206
2207 ret = call_usermodehelper_fns(argv[0], argv, envp, UMH_NO_WAIT,
2208 NULL, argv_cleanup, NULL);
2209 if (ret == -ENOMEM)
2210 argv_free(argv);
2211
2212 return ret;
2213 }
2214
2215 /**
2216 * orderly_poweroff - Trigger an orderly system poweroff
2217 * @force: force poweroff if command execution fails
2218 *
2219 * This may be called from any context to trigger a system shutdown.
2220 * If the orderly shutdown fails, it will force an immediate shutdown.
2221 */
2222 int orderly_poweroff(bool force)
2223 {
2224 int ret = __orderly_poweroff();
2225
2226 if (ret && force) {
2227 printk(KERN_WARNING "Failed to start orderly shutdown: "
2228 "forcing the issue\n");
2229
2230 /*
2231 * I guess this should try to kick off some daemon to sync and
2232 * poweroff asap. Or not even bother syncing if we're doing an
2233 * emergency shutdown?
2234 */
2235 emergency_sync();
2236 kernel_power_off();
2237 }
2238
2239 return ret;
2240 }
2241 EXPORT_SYMBOL_GPL(orderly_poweroff);
Linus' kernel tree