?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
rtc_sysfs_show_hctosys(): display 0 if resume failed
[linux-2.6.git] / kernel / sys.c
blobc5cb5b99cb8152808f569f6b993010a33e646bce
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 disable_nonboot_cpus();
372 if (!cmd)
373 printk(KERN_EMERG "Restarting system.\n");
374 else
375 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
376 kmsg_dump(KMSG_DUMP_RESTART);
377 machine_restart(cmd);
378 }
379 EXPORT_SYMBOL_GPL(kernel_restart);
380
381 static void kernel_shutdown_prepare(enum system_states state)
382 {
383 blocking_notifier_call_chain(&reboot_notifier_list,
384 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
385 system_state = state;
386 usermodehelper_disable();
387 device_shutdown();
388 }
389 /**
390 * kernel_halt - halt the system
391 *
392 * Shutdown everything and perform a clean system halt.
393 */
394 void kernel_halt(void)
395 {
396 kernel_shutdown_prepare(SYSTEM_HALT);
397 syscore_shutdown();
398 printk(KERN_EMERG "System halted.\n");
399 kmsg_dump(KMSG_DUMP_HALT);
400 machine_halt();
401 }
402
403 EXPORT_SYMBOL_GPL(kernel_halt);
404
405 /**
406 * kernel_power_off - power_off the system
407 *
408 * Shutdown everything and perform a clean system power_off.
409 */
410 void kernel_power_off(void)
411 {
412 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
413 if (pm_power_off_prepare)
414 pm_power_off_prepare();
415 disable_nonboot_cpus();
416 syscore_shutdown();
417 printk(KERN_EMERG "Power down.\n");
418 kmsg_dump(KMSG_DUMP_POWEROFF);
419 machine_power_off();
420 }
421 EXPORT_SYMBOL_GPL(kernel_power_off);
422
423 static DEFINE_MUTEX(reboot_mutex);
424
425 /*
426 * Reboot system call: for obvious reasons only root may call it,
427 * and even root needs to set up some magic numbers in the registers
428 * so that some mistake won't make this reboot the whole machine.
429 * You can also set the meaning of the ctrl-alt-del-key here.
430 *
431 * reboot doesn't sync: do that yourself before calling this.
432 */
433 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
434 void __user *, arg)
435 {
436 char buffer[256];
437 int ret = 0;
438
439 /* We only trust the superuser with rebooting the system. */
440 if (!capable(CAP_SYS_BOOT))
441 return -EPERM;
442
443 /* For safety, we require "magic" arguments. */
444 if (magic1 != LINUX_REBOOT_MAGIC1 ||
445 (magic2 != LINUX_REBOOT_MAGIC2 &&
446 magic2 != LINUX_REBOOT_MAGIC2A &&
447 magic2 != LINUX_REBOOT_MAGIC2B &&
448 magic2 != LINUX_REBOOT_MAGIC2C))
449 return -EINVAL;
450
451 /*
452 * If pid namespaces are enabled and the current task is in a child
453 * pid_namespace, the command is handled by reboot_pid_ns() which will
454 * call do_exit().
455 */
456 ret = reboot_pid_ns(task_active_pid_ns(current), cmd);
457 if (ret)
458 return ret;
459
460 /* Instead of trying to make the power_off code look like
461 * halt when pm_power_off is not set do it the easy way.
462 */
463 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
464 cmd = LINUX_REBOOT_CMD_HALT;
465
466 mutex_lock(&reboot_mutex);
467 switch (cmd) {
468 case LINUX_REBOOT_CMD_RESTART:
469 kernel_restart(NULL);
470 break;
471
472 case LINUX_REBOOT_CMD_CAD_ON:
473 C_A_D = 1;
474 break;
475
476 case LINUX_REBOOT_CMD_CAD_OFF:
477 C_A_D = 0;
478 break;
479
480 case LINUX_REBOOT_CMD_HALT:
481 kernel_halt();
482 do_exit(0);
483 panic("cannot halt");
484
485 case LINUX_REBOOT_CMD_POWER_OFF:
486 kernel_power_off();
487 do_exit(0);
488 break;
489
490 case LINUX_REBOOT_CMD_RESTART2:
491 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
492 ret = -EFAULT;
493 break;
494 }
495 buffer[sizeof(buffer) - 1] = '\0';
496
497 kernel_restart(buffer);
498 break;
499
500 #ifdef CONFIG_KEXEC
501 case LINUX_REBOOT_CMD_KEXEC:
502 ret = kernel_kexec();
503 break;
504 #endif
505
506 #ifdef CONFIG_HIBERNATION
507 case LINUX_REBOOT_CMD_SW_SUSPEND:
508 ret = hibernate();
509 break;
510 #endif
511
512 default:
513 ret = -EINVAL;
514 break;
515 }
516 mutex_unlock(&reboot_mutex);
517 return ret;
518 }
519
520 static void deferred_cad(struct work_struct *dummy)
521 {
522 kernel_restart(NULL);
523 }
524
525 /*
526 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
527 * As it's called within an interrupt, it may NOT sync: the only choice
528 * is whether to reboot at once, or just ignore the ctrl-alt-del.
529 */
530 void ctrl_alt_del(void)
531 {
532 static DECLARE_WORK(cad_work, deferred_cad);
533
534 if (C_A_D)
535 schedule_work(&cad_work);
536 else
537 kill_cad_pid(SIGINT, 1);
538 }
539
540 /*
541 * Unprivileged users may change the real gid to the effective gid
542 * or vice versa. (BSD-style)
543 *
544 * If you set the real gid at all, or set the effective gid to a value not
545 * equal to the real gid, then the saved gid is set to the new effective gid.
546 *
547 * This makes it possible for a setgid program to completely drop its
548 * privileges, which is often a useful assertion to make when you are doing
549 * a security audit over a program.
550 *
551 * The general idea is that a program which uses just setregid() will be
552 * 100% compatible with BSD. A program which uses just setgid() will be
553 * 100% compatible with POSIX with saved IDs.
554 *
555 * SMP: There are not races, the GIDs are checked only by filesystem
556 * operations (as far as semantic preservation is concerned).
557 */
558 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
559 {
560 struct user_namespace *ns = current_user_ns();
561 const struct cred *old;
562 struct cred *new;
563 int retval;
564 kgid_t krgid, kegid;
565
566 krgid = make_kgid(ns, rgid);
567 kegid = make_kgid(ns, egid);
568
569 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
570 return -EINVAL;
571 if ((egid != (gid_t) -1) && !gid_valid(kegid))
572 return -EINVAL;
573
574 new = prepare_creds();
575 if (!new)
576 return -ENOMEM;
577 old = current_cred();
578
579 retval = -EPERM;
580 if (rgid != (gid_t) -1) {
581 if (gid_eq(old->gid, krgid) ||
582 gid_eq(old->egid, krgid) ||
583 nsown_capable(CAP_SETGID))
584 new->gid = krgid;
585 else
586 goto error;
587 }
588 if (egid != (gid_t) -1) {
589 if (gid_eq(old->gid, kegid) ||
590 gid_eq(old->egid, kegid) ||
591 gid_eq(old->sgid, kegid) ||
592 nsown_capable(CAP_SETGID))
593 new->egid = kegid;
594 else
595 goto error;
596 }
597
598 if (rgid != (gid_t) -1 ||
599 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
600 new->sgid = new->egid;
601 new->fsgid = new->egid;
602
603 return commit_creds(new);
604
605 error:
606 abort_creds(new);
607 return retval;
608 }
609
610 /*
611 * setgid() is implemented like SysV w/ SAVED_IDS
612 *
613 * SMP: Same implicit races as above.
614 */
615 SYSCALL_DEFINE1(setgid, gid_t, gid)
616 {
617 struct user_namespace *ns = current_user_ns();
618 const struct cred *old;
619 struct cred *new;
620 int retval;
621 kgid_t kgid;
622
623 kgid = make_kgid(ns, gid);
624 if (!gid_valid(kgid))
625 return -EINVAL;
626
627 new = prepare_creds();
628 if (!new)
629 return -ENOMEM;
630 old = current_cred();
631
632 retval = -EPERM;
633 if (nsown_capable(CAP_SETGID))
634 new->gid = new->egid = new->sgid = new->fsgid = kgid;
635 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
636 new->egid = new->fsgid = kgid;
637 else
638 goto error;
639
640 return commit_creds(new);
641
642 error:
643 abort_creds(new);
644 return retval;
645 }
646
647 /*
648 * change the user struct in a credentials set to match the new UID
649 */
650 static int set_user(struct cred *new)
651 {
652 struct user_struct *new_user;
653
654 new_user = alloc_uid(new->uid);
655 if (!new_user)
656 return -EAGAIN;
657
658 /*
659 * We don't fail in case of NPROC limit excess here because too many
660 * poorly written programs don't check set*uid() return code, assuming
661 * it never fails if called by root. We may still enforce NPROC limit
662 * for programs doing set*uid()+execve() by harmlessly deferring the
663 * failure to the execve() stage.
664 */
665 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
666 new_user != INIT_USER)
667 current->flags |= PF_NPROC_EXCEEDED;
668 else
669 current->flags &= ~PF_NPROC_EXCEEDED;
670
671 free_uid(new->user);
672 new->user = new_user;
673 return 0;
674 }
675
676 /*
677 * Unprivileged users may change the real uid to the effective uid
678 * or vice versa. (BSD-style)
679 *
680 * If you set the real uid at all, or set the effective uid to a value not
681 * equal to the real uid, then the saved uid is set to the new effective uid.
682 *
683 * This makes it possible for a setuid program to completely drop its
684 * privileges, which is often a useful assertion to make when you are doing
685 * a security audit over a program.
686 *
687 * The general idea is that a program which uses just setreuid() will be
688 * 100% compatible with BSD. A program which uses just setuid() will be
689 * 100% compatible with POSIX with saved IDs.
690 */
691 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
692 {
693 struct user_namespace *ns = current_user_ns();
694 const struct cred *old;
695 struct cred *new;
696 int retval;
697 kuid_t kruid, keuid;
698
699 kruid = make_kuid(ns, ruid);
700 keuid = make_kuid(ns, euid);
701
702 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
703 return -EINVAL;
704 if ((euid != (uid_t) -1) && !uid_valid(keuid))
705 return -EINVAL;
706
707 new = prepare_creds();
708 if (!new)
709 return -ENOMEM;
710 old = current_cred();
711
712 retval = -EPERM;
713 if (ruid != (uid_t) -1) {
714 new->uid = kruid;
715 if (!uid_eq(old->uid, kruid) &&
716 !uid_eq(old->euid, kruid) &&
717 !nsown_capable(CAP_SETUID))
718 goto error;
719 }
720
721 if (euid != (uid_t) -1) {
722 new->euid = keuid;
723 if (!uid_eq(old->uid, keuid) &&
724 !uid_eq(old->euid, keuid) &&
725 !uid_eq(old->suid, keuid) &&
726 !nsown_capable(CAP_SETUID))
727 goto error;
728 }
729
730 if (!uid_eq(new->uid, old->uid)) {
731 retval = set_user(new);
732 if (retval < 0)
733 goto error;
734 }
735 if (ruid != (uid_t) -1 ||
736 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
737 new->suid = new->euid;
738 new->fsuid = new->euid;
739
740 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
741 if (retval < 0)
742 goto error;
743
744 return commit_creds(new);
745
746 error:
747 abort_creds(new);
748 return retval;
749 }
750
751 /*
752 * setuid() is implemented like SysV with SAVED_IDS
753 *
754 * Note that SAVED_ID's is deficient in that a setuid root program
755 * like sendmail, for example, cannot set its uid to be a normal
756 * user and then switch back, because if you're root, setuid() sets
757 * the saved uid too. If you don't like this, blame the bright people
758 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
759 * will allow a root program to temporarily drop privileges and be able to
760 * regain them by swapping the real and effective uid.
761 */
762 SYSCALL_DEFINE1(setuid, uid_t, uid)
763 {
764 struct user_namespace *ns = current_user_ns();
765 const struct cred *old;
766 struct cred *new;
767 int retval;
768 kuid_t kuid;
769
770 kuid = make_kuid(ns, uid);
771 if (!uid_valid(kuid))
772 return -EINVAL;
773
774 new = prepare_creds();
775 if (!new)
776 return -ENOMEM;
777 old = current_cred();
778
779 retval = -EPERM;
780 if (nsown_capable(CAP_SETUID)) {
781 new->suid = new->uid = kuid;
782 if (!uid_eq(kuid, old->uid)) {
783 retval = set_user(new);
784 if (retval < 0)
785 goto error;
786 }
787 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
788 goto error;
789 }
790
791 new->fsuid = new->euid = kuid;
792
793 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
794 if (retval < 0)
795 goto error;
796
797 return commit_creds(new);
798
799 error:
800 abort_creds(new);
801 return retval;
802 }
803
804
805 /*
806 * This function implements a generic ability to update ruid, euid,
807 * and suid. This allows you to implement the 4.4 compatible seteuid().
808 */
809 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
810 {
811 struct user_namespace *ns = current_user_ns();
812 const struct cred *old;
813 struct cred *new;
814 int retval;
815 kuid_t kruid, keuid, ksuid;
816
817 kruid = make_kuid(ns, ruid);
818 keuid = make_kuid(ns, euid);
819 ksuid = make_kuid(ns, suid);
820
821 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
822 return -EINVAL;
823
824 if ((euid != (uid_t) -1) && !uid_valid(keuid))
825 return -EINVAL;
826
827 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
828 return -EINVAL;
829
830 new = prepare_creds();
831 if (!new)
832 return -ENOMEM;
833
834 old = current_cred();
835
836 retval = -EPERM;
837 if (!nsown_capable(CAP_SETUID)) {
838 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
839 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
840 goto error;
841 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
842 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
843 goto error;
844 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
845 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
846 goto error;
847 }
848
849 if (ruid != (uid_t) -1) {
850 new->uid = kruid;
851 if (!uid_eq(kruid, old->uid)) {
852 retval = set_user(new);
853 if (retval < 0)
854 goto error;
855 }
856 }
857 if (euid != (uid_t) -1)
858 new->euid = keuid;
859 if (suid != (uid_t) -1)
860 new->suid = ksuid;
861 new->fsuid = new->euid;
862
863 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
864 if (retval < 0)
865 goto error;
866
867 return commit_creds(new);
868
869 error:
870 abort_creds(new);
871 return retval;
872 }
873
874 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
875 {
876 const struct cred *cred = current_cred();
877 int retval;
878 uid_t ruid, euid, suid;
879
880 ruid = from_kuid_munged(cred->user_ns, cred->uid);
881 euid = from_kuid_munged(cred->user_ns, cred->euid);
882 suid = from_kuid_munged(cred->user_ns, cred->suid);
883
884 if (!(retval = put_user(ruid, ruidp)) &&
885 !(retval = put_user(euid, euidp)))
886 retval = put_user(suid, suidp);
887
888 return retval;
889 }
890
891 /*
892 * Same as above, but for rgid, egid, sgid.
893 */
894 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
895 {
896 struct user_namespace *ns = current_user_ns();
897 const struct cred *old;
898 struct cred *new;
899 int retval;
900 kgid_t krgid, kegid, ksgid;
901
902 krgid = make_kgid(ns, rgid);
903 kegid = make_kgid(ns, egid);
904 ksgid = make_kgid(ns, sgid);
905
906 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
907 return -EINVAL;
908 if ((egid != (gid_t) -1) && !gid_valid(kegid))
909 return -EINVAL;
910 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
911 return -EINVAL;
912
913 new = prepare_creds();
914 if (!new)
915 return -ENOMEM;
916 old = current_cred();
917
918 retval = -EPERM;
919 if (!nsown_capable(CAP_SETGID)) {
920 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
921 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
922 goto error;
923 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
924 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
925 goto error;
926 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
927 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
928 goto error;
929 }
930
931 if (rgid != (gid_t) -1)
932 new->gid = krgid;
933 if (egid != (gid_t) -1)
934 new->egid = kegid;
935 if (sgid != (gid_t) -1)
936 new->sgid = ksgid;
937 new->fsgid = new->egid;
938
939 return commit_creds(new);
940
941 error:
942 abort_creds(new);
943 return retval;
944 }
945
946 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
947 {
948 const struct cred *cred = current_cred();
949 int retval;
950 gid_t rgid, egid, sgid;
951
952 rgid = from_kgid_munged(cred->user_ns, cred->gid);
953 egid = from_kgid_munged(cred->user_ns, cred->egid);
954 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
955
956 if (!(retval = put_user(rgid, rgidp)) &&
957 !(retval = put_user(egid, egidp)))
958 retval = put_user(sgid, sgidp);
959
960 return retval;
961 }
962
963
964 /*
965 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
966 * is used for "access()" and for the NFS daemon (letting nfsd stay at
967 * whatever uid it wants to). It normally shadows "euid", except when
968 * explicitly set by setfsuid() or for access..
969 */
970 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
971 {
972 const struct cred *old;
973 struct cred *new;
974 uid_t old_fsuid;
975 kuid_t kuid;
976
977 old = current_cred();
978 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
979
980 kuid = make_kuid(old->user_ns, uid);
981 if (!uid_valid(kuid))
982 return old_fsuid;
983
984 new = prepare_creds();
985 if (!new)
986 return old_fsuid;
987
988 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
989 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
990 nsown_capable(CAP_SETUID)) {
991 if (!uid_eq(kuid, old->fsuid)) {
992 new->fsuid = kuid;
993 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
994 goto change_okay;
995 }
996 }
997
998 abort_creds(new);
999 return old_fsuid;
1000
1001 change_okay:
1002 commit_creds(new);
1003 return old_fsuid;
1004 }
1005
1006 /*
1007 * Samma på svenska..
1008 */
1009 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
1010 {
1011 const struct cred *old;
1012 struct cred *new;
1013 gid_t old_fsgid;
1014 kgid_t kgid;
1015
1016 old = current_cred();
1017 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
1018
1019 kgid = make_kgid(old->user_ns, gid);
1020 if (!gid_valid(kgid))
1021 return old_fsgid;
1022
1023 new = prepare_creds();
1024 if (!new)
1025 return old_fsgid;
1026
1027 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
1028 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
1029 nsown_capable(CAP_SETGID)) {
1030 if (!gid_eq(kgid, old->fsgid)) {
1031 new->fsgid = kgid;
1032 goto change_okay;
1033 }
1034 }
1035
1036 abort_creds(new);
1037 return old_fsgid;
1038
1039 change_okay:
1040 commit_creds(new);
1041 return old_fsgid;
1042 }
1043
1044 void do_sys_times(struct tms *tms)
1045 {
1046 cputime_t tgutime, tgstime, cutime, cstime;
1047
1048 spin_lock_irq(&current->sighand->siglock);
1049 thread_group_times(current, &tgutime, &tgstime);
1050 cutime = current->signal->cutime;
1051 cstime = current->signal->cstime;
1052 spin_unlock_irq(&current->sighand->siglock);
1053 tms->tms_utime = cputime_to_clock_t(tgutime);
1054 tms->tms_stime = cputime_to_clock_t(tgstime);
1055 tms->tms_cutime = cputime_to_clock_t(cutime);
1056 tms->tms_cstime = cputime_to_clock_t(cstime);
1057 }
1058
1059 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1060 {
1061 if (tbuf) {
1062 struct tms tmp;
1063
1064 do_sys_times(&tmp);
1065 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1066 return -EFAULT;
1067 }
1068 force_successful_syscall_return();
1069 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1070 }
1071
1072 /*
1073 * This needs some heavy checking ...
1074 * I just haven't the stomach for it. I also don't fully
1075 * understand sessions/pgrp etc. Let somebody who does explain it.
1076 *
1077 * OK, I think I have the protection semantics right.... this is really
1078 * only important on a multi-user system anyway, to make sure one user
1079 * can't send a signal to a process owned by another. -TYT, 12/12/91
1080 *
1081 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1082 * LBT 04.03.94
1083 */
1084 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1085 {
1086 struct task_struct *p;
1087 struct task_struct *group_leader = current->group_leader;
1088 struct pid *pgrp;
1089 int err;
1090
1091 if (!pid)
1092 pid = task_pid_vnr(group_leader);
1093 if (!pgid)
1094 pgid = pid;
1095 if (pgid < 0)
1096 return -EINVAL;
1097 rcu_read_lock();
1098
1099 /* From this point forward we keep holding onto the tasklist lock
1100 * so that our parent does not change from under us. -DaveM
1101 */
1102 write_lock_irq(&tasklist_lock);
1103
1104 err = -ESRCH;
1105 p = find_task_by_vpid(pid);
1106 if (!p)
1107 goto out;
1108
1109 err = -EINVAL;
1110 if (!thread_group_leader(p))
1111 goto out;
1112
1113 if (same_thread_group(p->real_parent, group_leader)) {
1114 err = -EPERM;
1115 if (task_session(p) != task_session(group_leader))
1116 goto out;
1117 err = -EACCES;
1118 if (p->did_exec)
1119 goto out;
1120 } else {
1121 err = -ESRCH;
1122 if (p != group_leader)
1123 goto out;
1124 }
1125
1126 err = -EPERM;
1127 if (p->signal->leader)
1128 goto out;
1129
1130 pgrp = task_pid(p);
1131 if (pgid != pid) {
1132 struct task_struct *g;
1133
1134 pgrp = find_vpid(pgid);
1135 g = pid_task(pgrp, PIDTYPE_PGID);
1136 if (!g || task_session(g) != task_session(group_leader))
1137 goto out;
1138 }
1139
1140 err = security_task_setpgid(p, pgid);
1141 if (err)
1142 goto out;
1143
1144 if (task_pgrp(p) != pgrp)
1145 change_pid(p, PIDTYPE_PGID, pgrp);
1146
1147 err = 0;
1148 out:
1149 /* All paths lead to here, thus we are safe. -DaveM */
1150 write_unlock_irq(&tasklist_lock);
1151 rcu_read_unlock();
1152 return err;
1153 }
1154
1155 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1156 {
1157 struct task_struct *p;
1158 struct pid *grp;
1159 int retval;
1160
1161 rcu_read_lock();
1162 if (!pid)
1163 grp = task_pgrp(current);
1164 else {
1165 retval = -ESRCH;
1166 p = find_task_by_vpid(pid);
1167 if (!p)
1168 goto out;
1169 grp = task_pgrp(p);
1170 if (!grp)
1171 goto out;
1172
1173 retval = security_task_getpgid(p);
1174 if (retval)
1175 goto out;
1176 }
1177 retval = pid_vnr(grp);
1178 out:
1179 rcu_read_unlock();
1180 return retval;
1181 }
1182
1183 #ifdef __ARCH_WANT_SYS_GETPGRP
1184
1185 SYSCALL_DEFINE0(getpgrp)
1186 {
1187 return sys_getpgid(0);
1188 }
1189
1190 #endif
1191
1192 SYSCALL_DEFINE1(getsid, pid_t, pid)
1193 {
1194 struct task_struct *p;
1195 struct pid *sid;
1196 int retval;
1197
1198 rcu_read_lock();
1199 if (!pid)
1200 sid = task_session(current);
1201 else {
1202 retval = -ESRCH;
1203 p = find_task_by_vpid(pid);
1204 if (!p)
1205 goto out;
1206 sid = task_session(p);
1207 if (!sid)
1208 goto out;
1209
1210 retval = security_task_getsid(p);
1211 if (retval)
1212 goto out;
1213 }
1214 retval = pid_vnr(sid);
1215 out:
1216 rcu_read_unlock();
1217 return retval;
1218 }
1219
1220 SYSCALL_DEFINE0(setsid)
1221 {
1222 struct task_struct *group_leader = current->group_leader;
1223 struct pid *sid = task_pid(group_leader);
1224 pid_t session = pid_vnr(sid);
1225 int err = -EPERM;
1226
1227 write_lock_irq(&tasklist_lock);
1228 /* Fail if I am already a session leader */
1229 if (group_leader->signal->leader)
1230 goto out;
1231
1232 /* Fail if a process group id already exists that equals the
1233 * proposed session id.
1234 */
1235 if (pid_task(sid, PIDTYPE_PGID))
1236 goto out;
1237
1238 group_leader->signal->leader = 1;
1239 __set_special_pids(sid);
1240
1241 proc_clear_tty(group_leader);
1242
1243 err = session;
1244 out:
1245 write_unlock_irq(&tasklist_lock);
1246 if (err > 0) {
1247 proc_sid_connector(group_leader);
1248 sched_autogroup_create_attach(group_leader);
1249 }
1250 return err;
1251 }
1252
1253 DECLARE_RWSEM(uts_sem);
1254
1255 #ifdef COMPAT_UTS_MACHINE
1256 #define override_architecture(name) \
1257 (personality(current->personality) == PER_LINUX32 && \
1258 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1259 sizeof(COMPAT_UTS_MACHINE)))
1260 #else
1261 #define override_architecture(name) 0
1262 #endif
1263
1264 /*
1265 * Work around broken programs that cannot handle "Linux 3.0".
1266 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1267 */
1268 static int override_release(char __user *release, int len)
1269 {
1270 int ret = 0;
1271 char buf[65];
1272
1273 if (current->personality & UNAME26) {
1274 char *rest = UTS_RELEASE;
1275 int ndots = 0;
1276 unsigned v;
1277
1278 while (*rest) {
1279 if (*rest == '.' && ++ndots >= 3)
1280 break;
1281 if (!isdigit(*rest) && *rest != '.')
1282 break;
1283 rest++;
1284 }
1285 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1286 snprintf(buf, len, "2.6.%u%s", v, rest);
1287 ret = copy_to_user(release, buf, len);
1288 }
1289 return ret;
1290 }
1291
1292 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1293 {
1294 int errno = 0;
1295
1296 down_read(&uts_sem);
1297 if (copy_to_user(name, utsname(), sizeof *name))
1298 errno = -EFAULT;
1299 up_read(&uts_sem);
1300
1301 if (!errno && override_release(name->release, sizeof(name->release)))
1302 errno = -EFAULT;
1303 if (!errno && override_architecture(name))
1304 errno = -EFAULT;
1305 return errno;
1306 }
1307
1308 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1309 /*
1310 * Old cruft
1311 */
1312 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1313 {
1314 int error = 0;
1315
1316 if (!name)
1317 return -EFAULT;
1318
1319 down_read(&uts_sem);
1320 if (copy_to_user(name, utsname(), sizeof(*name)))
1321 error = -EFAULT;
1322 up_read(&uts_sem);
1323
1324 if (!error && override_release(name->release, sizeof(name->release)))
1325 error = -EFAULT;
1326 if (!error && override_architecture(name))
1327 error = -EFAULT;
1328 return error;
1329 }
1330
1331 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1332 {
1333 int error;
1334
1335 if (!name)
1336 return -EFAULT;
1337 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1338 return -EFAULT;
1339
1340 down_read(&uts_sem);
1341 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1342 __OLD_UTS_LEN);
1343 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1344 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1345 __OLD_UTS_LEN);
1346 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1347 error |= __copy_to_user(&name->release, &utsname()->release,
1348 __OLD_UTS_LEN);
1349 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1350 error |= __copy_to_user(&name->version, &utsname()->version,
1351 __OLD_UTS_LEN);
1352 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1353 error |= __copy_to_user(&name->machine, &utsname()->machine,
1354 __OLD_UTS_LEN);
1355 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1356 up_read(&uts_sem);
1357
1358 if (!error && override_architecture(name))
1359 error = -EFAULT;
1360 if (!error && override_release(name->release, sizeof(name->release)))
1361 error = -EFAULT;
1362 return error ? -EFAULT : 0;
1363 }
1364 #endif
1365
1366 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1367 {
1368 int errno;
1369 char tmp[__NEW_UTS_LEN];
1370
1371 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1372 return -EPERM;
1373
1374 if (len < 0 || len > __NEW_UTS_LEN)
1375 return -EINVAL;
1376 down_write(&uts_sem);
1377 errno = -EFAULT;
1378 if (!copy_from_user(tmp, name, len)) {
1379 struct new_utsname *u = utsname();
1380
1381 memcpy(u->nodename, tmp, len);
1382 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1383 errno = 0;
1384 uts_proc_notify(UTS_PROC_HOSTNAME);
1385 }
1386 up_write(&uts_sem);
1387 return errno;
1388 }
1389
1390 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1391
1392 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1393 {
1394 int i, errno;
1395 struct new_utsname *u;
1396
1397 if (len < 0)
1398 return -EINVAL;
1399 down_read(&uts_sem);
1400 u = utsname();
1401 i = 1 + strlen(u->nodename);
1402 if (i > len)
1403 i = len;
1404 errno = 0;
1405 if (copy_to_user(name, u->nodename, i))
1406 errno = -EFAULT;
1407 up_read(&uts_sem);
1408 return errno;
1409 }
1410
1411 #endif
1412
1413 /*
1414 * Only setdomainname; getdomainname can be implemented by calling
1415 * uname()
1416 */
1417 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1418 {
1419 int errno;
1420 char tmp[__NEW_UTS_LEN];
1421
1422 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1423 return -EPERM;
1424 if (len < 0 || len > __NEW_UTS_LEN)
1425 return -EINVAL;
1426
1427 down_write(&uts_sem);
1428 errno = -EFAULT;
1429 if (!copy_from_user(tmp, name, len)) {
1430 struct new_utsname *u = utsname();
1431
1432 memcpy(u->domainname, tmp, len);
1433 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1434 errno = 0;
1435 uts_proc_notify(UTS_PROC_DOMAINNAME);
1436 }
1437 up_write(&uts_sem);
1438 return errno;
1439 }
1440
1441 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1442 {
1443 struct rlimit value;
1444 int ret;
1445
1446 ret = do_prlimit(current, resource, NULL, &value);
1447 if (!ret)
1448 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1449
1450 return ret;
1451 }
1452
1453 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1454
1455 /*
1456 * Back compatibility for getrlimit. Needed for some apps.
1457 */
1458
1459 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1460 struct rlimit __user *, rlim)
1461 {
1462 struct rlimit x;
1463 if (resource >= RLIM_NLIMITS)
1464 return -EINVAL;
1465
1466 task_lock(current->group_leader);
1467 x = current->signal->rlim[resource];
1468 task_unlock(current->group_leader);
1469 if (x.rlim_cur > 0x7FFFFFFF)
1470 x.rlim_cur = 0x7FFFFFFF;
1471 if (x.rlim_max > 0x7FFFFFFF)
1472 x.rlim_max = 0x7FFFFFFF;
1473 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1474 }
1475
1476 #endif
1477
1478 static inline bool rlim64_is_infinity(__u64 rlim64)
1479 {
1480 #if BITS_PER_LONG < 64
1481 return rlim64 >= ULONG_MAX;
1482 #else
1483 return rlim64 == RLIM64_INFINITY;
1484 #endif
1485 }
1486
1487 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1488 {
1489 if (rlim->rlim_cur == RLIM_INFINITY)
1490 rlim64->rlim_cur = RLIM64_INFINITY;
1491 else
1492 rlim64->rlim_cur = rlim->rlim_cur;
1493 if (rlim->rlim_max == RLIM_INFINITY)
1494 rlim64->rlim_max = RLIM64_INFINITY;
1495 else
1496 rlim64->rlim_max = rlim->rlim_max;
1497 }
1498
1499 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1500 {
1501 if (rlim64_is_infinity(rlim64->rlim_cur))
1502 rlim->rlim_cur = RLIM_INFINITY;
1503 else
1504 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1505 if (rlim64_is_infinity(rlim64->rlim_max))
1506 rlim->rlim_max = RLIM_INFINITY;
1507 else
1508 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1509 }
1510
1511 /* make sure you are allowed to change @tsk limits before calling this */
1512 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1513 struct rlimit *new_rlim, struct rlimit *old_rlim)
1514 {
1515 struct rlimit *rlim;
1516 int retval = 0;
1517
1518 if (resource >= RLIM_NLIMITS)
1519 return -EINVAL;
1520 if (new_rlim) {
1521 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1522 return -EINVAL;
1523 if (resource == RLIMIT_NOFILE &&
1524 new_rlim->rlim_max > sysctl_nr_open)
1525 return -EPERM;
1526 }
1527
1528 /* protect tsk->signal and tsk->sighand from disappearing */
1529 read_lock(&tasklist_lock);
1530 if (!tsk->sighand) {
1531 retval = -ESRCH;
1532 goto out;
1533 }
1534
1535 rlim = tsk->signal->rlim + resource;
1536 task_lock(tsk->group_leader);
1537 if (new_rlim) {
1538 /* Keep the capable check against init_user_ns until
1539 cgroups can contain all limits */
1540 if (new_rlim->rlim_max > rlim->rlim_max &&
1541 !capable(CAP_SYS_RESOURCE))
1542 retval = -EPERM;
1543 if (!retval)
1544 retval = security_task_setrlimit(tsk->group_leader,
1545 resource, new_rlim);
1546 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1547 /*
1548 * The caller is asking for an immediate RLIMIT_CPU
1549 * expiry. But we use the zero value to mean "it was
1550 * never set". So let's cheat and make it one second
1551 * instead
1552 */
1553 new_rlim->rlim_cur = 1;
1554 }
1555 }
1556 if (!retval) {
1557 if (old_rlim)
1558 *old_rlim = *rlim;
1559 if (new_rlim)
1560 *rlim = *new_rlim;
1561 }
1562 task_unlock(tsk->group_leader);
1563
1564 /*
1565 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1566 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1567 * very long-standing error, and fixing it now risks breakage of
1568 * applications, so we live with it
1569 */
1570 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1571 new_rlim->rlim_cur != RLIM_INFINITY)
1572 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1573 out:
1574 read_unlock(&tasklist_lock);
1575 return retval;
1576 }
1577
1578 /* rcu lock must be held */
1579 static int check_prlimit_permission(struct task_struct *task)
1580 {
1581 const struct cred *cred = current_cred(), *tcred;
1582
1583 if (current == task)
1584 return 0;
1585
1586 tcred = __task_cred(task);
1587 if (uid_eq(cred->uid, tcred->euid) &&
1588 uid_eq(cred->uid, tcred->suid) &&
1589 uid_eq(cred->uid, tcred->uid) &&
1590 gid_eq(cred->gid, tcred->egid) &&
1591 gid_eq(cred->gid, tcred->sgid) &&
1592 gid_eq(cred->gid, tcred->gid))
1593 return 0;
1594 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1595 return 0;
1596
1597 return -EPERM;
1598 }
1599
1600 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1601 const struct rlimit64 __user *, new_rlim,
1602 struct rlimit64 __user *, old_rlim)
1603 {
1604 struct rlimit64 old64, new64;
1605 struct rlimit old, new;
1606 struct task_struct *tsk;
1607 int ret;
1608
1609 if (new_rlim) {
1610 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1611 return -EFAULT;
1612 rlim64_to_rlim(&new64, &new);
1613 }
1614
1615 rcu_read_lock();
1616 tsk = pid ? find_task_by_vpid(pid) : current;
1617 if (!tsk) {
1618 rcu_read_unlock();
1619 return -ESRCH;
1620 }
1621 ret = check_prlimit_permission(tsk);
1622 if (ret) {
1623 rcu_read_unlock();
1624 return ret;
1625 }
1626 get_task_struct(tsk);
1627 rcu_read_unlock();
1628
1629 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1630 old_rlim ? &old : NULL);
1631
1632 if (!ret && old_rlim) {
1633 rlim_to_rlim64(&old, &old64);
1634 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1635 ret = -EFAULT;
1636 }
1637
1638 put_task_struct(tsk);
1639 return ret;
1640 }
1641
1642 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1643 {
1644 struct rlimit new_rlim;
1645
1646 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1647 return -EFAULT;
1648 return do_prlimit(current, resource, &new_rlim, NULL);
1649 }
1650
1651 /*
1652 * It would make sense to put struct rusage in the task_struct,
1653 * except that would make the task_struct be *really big*. After
1654 * task_struct gets moved into malloc'ed memory, it would
1655 * make sense to do this. It will make moving the rest of the information
1656 * a lot simpler! (Which we're not doing right now because we're not
1657 * measuring them yet).
1658 *
1659 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1660 * races with threads incrementing their own counters. But since word
1661 * reads are atomic, we either get new values or old values and we don't
1662 * care which for the sums. We always take the siglock to protect reading
1663 * the c* fields from p->signal from races with exit.c updating those
1664 * fields when reaping, so a sample either gets all the additions of a
1665 * given child after it's reaped, or none so this sample is before reaping.
1666 *
1667 * Locking:
1668 * We need to take the siglock for CHILDEREN, SELF and BOTH
1669 * for the cases current multithreaded, non-current single threaded
1670 * non-current multithreaded. Thread traversal is now safe with
1671 * the siglock held.
1672 * Strictly speaking, we donot need to take the siglock if we are current and
1673 * single threaded, as no one else can take our signal_struct away, no one
1674 * else can reap the children to update signal->c* counters, and no one else
1675 * can race with the signal-> fields. If we do not take any lock, the
1676 * signal-> fields could be read out of order while another thread was just
1677 * exiting. So we should place a read memory barrier when we avoid the lock.
1678 * On the writer side, write memory barrier is implied in __exit_signal
1679 * as __exit_signal releases the siglock spinlock after updating the signal->
1680 * fields. But we don't do this yet to keep things simple.
1681 *
1682 */
1683
1684 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1685 {
1686 r->ru_nvcsw += t->nvcsw;
1687 r->ru_nivcsw += t->nivcsw;
1688 r->ru_minflt += t->min_flt;
1689 r->ru_majflt += t->maj_flt;
1690 r->ru_inblock += task_io_get_inblock(t);
1691 r->ru_oublock += task_io_get_oublock(t);
1692 }
1693
1694 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1695 {
1696 struct task_struct *t;
1697 unsigned long flags;
1698 cputime_t tgutime, tgstime, utime, stime;
1699 unsigned long maxrss = 0;
1700
1701 memset((char *) r, 0, sizeof *r);
1702 utime = stime = 0;
1703
1704 if (who == RUSAGE_THREAD) {
1705 task_times(current, &utime, &stime);
1706 accumulate_thread_rusage(p, r);
1707 maxrss = p->signal->maxrss;
1708 goto out;
1709 }
1710
1711 if (!lock_task_sighand(p, &flags))
1712 return;
1713
1714 switch (who) {
1715 case RUSAGE_BOTH:
1716 case RUSAGE_CHILDREN:
1717 utime = p->signal->cutime;
1718 stime = p->signal->cstime;
1719 r->ru_nvcsw = p->signal->cnvcsw;
1720 r->ru_nivcsw = p->signal->cnivcsw;
1721 r->ru_minflt = p->signal->cmin_flt;
1722 r->ru_majflt = p->signal->cmaj_flt;
1723 r->ru_inblock = p->signal->cinblock;
1724 r->ru_oublock = p->signal->coublock;
1725 maxrss = p->signal->cmaxrss;
1726
1727 if (who == RUSAGE_CHILDREN)
1728 break;
1729
1730 case RUSAGE_SELF:
1731 thread_group_times(p, &tgutime, &tgstime);
1732 utime += tgutime;
1733 stime += tgstime;
1734 r->ru_nvcsw += p->signal->nvcsw;
1735 r->ru_nivcsw += p->signal->nivcsw;
1736 r->ru_minflt += p->signal->min_flt;
1737 r->ru_majflt += p->signal->maj_flt;
1738 r->ru_inblock += p->signal->inblock;
1739 r->ru_oublock += p->signal->oublock;
1740 if (maxrss < p->signal->maxrss)
1741 maxrss = p->signal->maxrss;
1742 t = p;
1743 do {
1744 accumulate_thread_rusage(t, r);
1745 t = next_thread(t);
1746 } while (t != p);
1747 break;
1748
1749 default:
1750 BUG();
1751 }
1752 unlock_task_sighand(p, &flags);
1753
1754 out:
1755 cputime_to_timeval(utime, &r->ru_utime);
1756 cputime_to_timeval(stime, &r->ru_stime);
1757
1758 if (who != RUSAGE_CHILDREN) {
1759 struct mm_struct *mm = get_task_mm(p);
1760 if (mm) {
1761 setmax_mm_hiwater_rss(&maxrss, mm);
1762 mmput(mm);
1763 }
1764 }
1765 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1766 }
1767
1768 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1769 {
1770 struct rusage r;
1771 k_getrusage(p, who, &r);
1772 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1773 }
1774
1775 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1776 {
1777 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1778 who != RUSAGE_THREAD)
1779 return -EINVAL;
1780 return getrusage(current, who, ru);
1781 }
1782
1783 SYSCALL_DEFINE1(umask, int, mask)
1784 {
1785 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1786 return mask;
1787 }
1788
1789 #ifdef CONFIG_CHECKPOINT_RESTORE
1790 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1791 {
1792 struct fd exe;
1793 struct dentry *dentry;
1794 int err;
1795
1796 exe = fdget(fd);
1797 if (!exe.file)
1798 return -EBADF;
1799
1800 dentry = exe.file->f_path.dentry;
1801
1802 /*
1803 * Because the original mm->exe_file points to executable file, make
1804 * sure that this one is executable as well, to avoid breaking an
1805 * overall picture.
1806 */
1807 err = -EACCES;
1808 if (!S_ISREG(dentry->d_inode->i_mode) ||
1809 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1810 goto exit;
1811
1812 err = inode_permission(dentry->d_inode, MAY_EXEC);
1813 if (err)
1814 goto exit;
1815
1816 down_write(&mm->mmap_sem);
1817
1818 /*
1819 * Forbid mm->exe_file change if old file still mapped.
1820 */
1821 err = -EBUSY;
1822 if (mm->exe_file) {
1823 struct vm_area_struct *vma;
1824
1825 for (vma = mm->mmap; vma; vma = vma->vm_next)
1826 if (vma->vm_file &&
1827 path_equal(&vma->vm_file->f_path,
1828 &mm->exe_file->f_path))
1829 goto exit_unlock;
1830 }
1831
1832 /*
1833 * The symlink can be changed only once, just to disallow arbitrary
1834 * transitions malicious software might bring in. This means one
1835 * could make a snapshot over all processes running and monitor
1836 * /proc/pid/exe changes to notice unusual activity if needed.
1837 */
1838 err = -EPERM;
1839 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1840 goto exit_unlock;
1841
1842 err = 0;
1843 set_mm_exe_file(mm, exe.file); /* this grabs a reference to exe.file */
1844 exit_unlock:
1845 up_write(&mm->mmap_sem);
1846
1847 exit:
1848 fdput(exe);
1849 return err;
1850 }
1851
1852 static int prctl_set_mm(int opt, unsigned long addr,
1853 unsigned long arg4, unsigned long arg5)
1854 {
1855 unsigned long rlim = rlimit(RLIMIT_DATA);
1856 struct mm_struct *mm = current->mm;
1857 struct vm_area_struct *vma;
1858 int error;
1859
1860 if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
1861 return -EINVAL;
1862
1863 if (!capable(CAP_SYS_RESOURCE))
1864 return -EPERM;
1865
1866 if (opt == PR_SET_MM_EXE_FILE)
1867 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1868
1869 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1870 return -EINVAL;
1871
1872 error = -EINVAL;
1873
1874 down_read(&mm->mmap_sem);
1875 vma = find_vma(mm, addr);
1876
1877 switch (opt) {
1878 case PR_SET_MM_START_CODE:
1879 mm->start_code = addr;
1880 break;
1881 case PR_SET_MM_END_CODE:
1882 mm->end_code = addr;
1883 break;
1884 case PR_SET_MM_START_DATA:
1885 mm->start_data = addr;
1886 break;
1887 case PR_SET_MM_END_DATA:
1888 mm->end_data = addr;
1889 break;
1890
1891 case PR_SET_MM_START_BRK:
1892 if (addr <= mm->end_data)
1893 goto out;
1894
1895 if (rlim < RLIM_INFINITY &&
1896 (mm->brk - addr) +
1897 (mm->end_data - mm->start_data) > rlim)
1898 goto out;
1899
1900 mm->start_brk = addr;
1901 break;
1902
1903 case PR_SET_MM_BRK:
1904 if (addr <= mm->end_data)
1905 goto out;
1906
1907 if (rlim < RLIM_INFINITY &&
1908 (addr - mm->start_brk) +
1909 (mm->end_data - mm->start_data) > rlim)
1910 goto out;
1911
1912 mm->brk = addr;
1913 break;
1914
1915 /*
1916 * If command line arguments and environment
1917 * are placed somewhere else on stack, we can
1918 * set them up here, ARG_START/END to setup
1919 * command line argumets and ENV_START/END
1920 * for environment.
1921 */
1922 case PR_SET_MM_START_STACK:
1923 case PR_SET_MM_ARG_START:
1924 case PR_SET_MM_ARG_END:
1925 case PR_SET_MM_ENV_START:
1926 case PR_SET_MM_ENV_END:
1927 if (!vma) {
1928 error = -EFAULT;
1929 goto out;
1930 }
1931 if (opt == PR_SET_MM_START_STACK)
1932 mm->start_stack = addr;
1933 else if (opt == PR_SET_MM_ARG_START)
1934 mm->arg_start = addr;
1935 else if (opt == PR_SET_MM_ARG_END)
1936 mm->arg_end = addr;
1937 else if (opt == PR_SET_MM_ENV_START)
1938 mm->env_start = addr;
1939 else if (opt == PR_SET_MM_ENV_END)
1940 mm->env_end = addr;
1941 break;
1942
1943 /*
1944 * This doesn't move auxiliary vector itself
1945 * since it's pinned to mm_struct, but allow
1946 * to fill vector with new values. It's up
1947 * to a caller to provide sane values here
1948 * otherwise user space tools which use this
1949 * vector might be unhappy.
1950 */
1951 case PR_SET_MM_AUXV: {
1952 unsigned long user_auxv[AT_VECTOR_SIZE];
1953
1954 if (arg4 > sizeof(user_auxv))
1955 goto out;
1956 up_read(&mm->mmap_sem);
1957
1958 if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
1959 return -EFAULT;
1960
1961 /* Make sure the last entry is always AT_NULL */
1962 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1963 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1964
1965 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1966
1967 task_lock(current);
1968 memcpy(mm->saved_auxv, user_auxv, arg4);
1969 task_unlock(current);
1970
1971 return 0;
1972 }
1973 default:
1974 goto out;
1975 }
1976
1977 error = 0;
1978 out:
1979 up_read(&mm->mmap_sem);
1980 return error;
1981 }
1982
1983 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1984 {
1985 return put_user(me->clear_child_tid, tid_addr);
1986 }
1987
1988 #else /* CONFIG_CHECKPOINT_RESTORE */
1989 static int prctl_set_mm(int opt, unsigned long addr,
1990 unsigned long arg4, unsigned long arg5)
1991 {
1992 return -EINVAL;
1993 }
1994 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1995 {
1996 return -EINVAL;
1997 }
1998 #endif
1999
2000 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2001 unsigned long, arg4, unsigned long, arg5)
2002 {
2003 struct task_struct *me = current;
2004 unsigned char comm[sizeof(me->comm)];
2005 long error;
2006
2007 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2008 if (error != -ENOSYS)
2009 return error;
2010
2011 error = 0;
2012 switch (option) {
2013 case PR_SET_PDEATHSIG:
2014 if (!valid_signal(arg2)) {
2015 error = -EINVAL;
2016 break;
2017 }
2018 me->pdeath_signal = arg2;
2019 break;
2020 case PR_GET_PDEATHSIG:
2021 error = put_user(me->pdeath_signal, (int __user *)arg2);
2022 break;
2023 case PR_GET_DUMPABLE:
2024 error = get_dumpable(me->mm);
2025 break;
2026 case PR_SET_DUMPABLE:
2027 if (arg2 < 0 || arg2 > 1) {
2028 error = -EINVAL;
2029 break;
2030 }
2031 set_dumpable(me->mm, arg2);
2032 break;
2033
2034 case PR_SET_UNALIGN:
2035 error = SET_UNALIGN_CTL(me, arg2);
2036 break;
2037 case PR_GET_UNALIGN:
2038 error = GET_UNALIGN_CTL(me, arg2);
2039 break;
2040 case PR_SET_FPEMU:
2041 error = SET_FPEMU_CTL(me, arg2);
2042 break;
2043 case PR_GET_FPEMU:
2044 error = GET_FPEMU_CTL(me, arg2);
2045 break;
2046 case PR_SET_FPEXC:
2047 error = SET_FPEXC_CTL(me, arg2);
2048 break;
2049 case PR_GET_FPEXC:
2050 error = GET_FPEXC_CTL(me, arg2);
2051 break;
2052 case PR_GET_TIMING:
2053 error = PR_TIMING_STATISTICAL;
2054 break;
2055 case PR_SET_TIMING:
2056 if (arg2 != PR_TIMING_STATISTICAL)
2057 error = -EINVAL;
2058 break;
2059 case PR_SET_NAME:
2060 comm[sizeof(me->comm)-1] = 0;
2061 if (strncpy_from_user(comm, (char __user *)arg2,
2062 sizeof(me->comm) - 1) < 0)
2063 return -EFAULT;
2064 set_task_comm(me, comm);
2065 proc_comm_connector(me);
2066 break;
2067 case PR_GET_NAME:
2068 get_task_comm(comm, me);
2069 if (copy_to_user((char __user *)arg2, comm,
2070 sizeof(comm)))
2071 return -EFAULT;
2072 break;
2073 case PR_GET_ENDIAN:
2074 error = GET_ENDIAN(me, arg2);
2075 break;
2076 case PR_SET_ENDIAN:
2077 error = SET_ENDIAN(me, arg2);
2078 break;
2079 case PR_GET_SECCOMP:
2080 error = prctl_get_seccomp();
2081 break;
2082 case PR_SET_SECCOMP:
2083 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2084 break;
2085 case PR_GET_TSC:
2086 error = GET_TSC_CTL(arg2);
2087 break;
2088 case PR_SET_TSC:
2089 error = SET_TSC_CTL(arg2);
2090 break;
2091 case PR_TASK_PERF_EVENTS_DISABLE:
2092 error = perf_event_task_disable();
2093 break;
2094 case PR_TASK_PERF_EVENTS_ENABLE:
2095 error = perf_event_task_enable();
2096 break;
2097 case PR_GET_TIMERSLACK:
2098 error = current->timer_slack_ns;
2099 break;
2100 case PR_SET_TIMERSLACK:
2101 if (arg2 <= 0)
2102 current->timer_slack_ns =
2103 current->default_timer_slack_ns;
2104 else
2105 current->timer_slack_ns = arg2;
2106 break;
2107 case PR_MCE_KILL:
2108 if (arg4 | arg5)
2109 return -EINVAL;
2110 switch (arg2) {
2111 case PR_MCE_KILL_CLEAR:
2112 if (arg3 != 0)
2113 return -EINVAL;
2114 current->flags &= ~PF_MCE_PROCESS;
2115 break;
2116 case PR_MCE_KILL_SET:
2117 current->flags |= PF_MCE_PROCESS;
2118 if (arg3 == PR_MCE_KILL_EARLY)
2119 current->flags |= PF_MCE_EARLY;
2120 else if (arg3 == PR_MCE_KILL_LATE)
2121 current->flags &= ~PF_MCE_EARLY;
2122 else if (arg3 == PR_MCE_KILL_DEFAULT)
2123 current->flags &=
2124 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2125 else
2126 return -EINVAL;
2127 break;
2128 default:
2129 return -EINVAL;
2130 }
2131 break;
2132 case PR_MCE_KILL_GET:
2133 if (arg2 | arg3 | arg4 | arg5)
2134 return -EINVAL;
2135 if (current->flags & PF_MCE_PROCESS)
2136 error = (current->flags & PF_MCE_EARLY) ?
2137 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2138 else
2139 error = PR_MCE_KILL_DEFAULT;
2140 break;
2141 case PR_SET_MM:
2142 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2143 break;
2144 case PR_GET_TID_ADDRESS:
2145 error = prctl_get_tid_address(me, (int __user **)arg2);
2146 break;
2147 case PR_SET_CHILD_SUBREAPER:
2148 me->signal->is_child_subreaper = !!arg2;
2149 break;
2150 case PR_GET_CHILD_SUBREAPER:
2151 error = put_user(me->signal->is_child_subreaper,
2152 (int __user *) arg2);
2153 break;
2154 case PR_SET_NO_NEW_PRIVS:
2155 if (arg2 != 1 || arg3 || arg4 || arg5)
2156 return -EINVAL;
2157
2158 current->no_new_privs = 1;
2159 break;
2160 case PR_GET_NO_NEW_PRIVS:
2161 if (arg2 || arg3 || arg4 || arg5)
2162 return -EINVAL;
2163 return current->no_new_privs ? 1 : 0;
2164 default:
2165 error = -EINVAL;
2166 break;
2167 }
2168 return error;
2169 }
2170
2171 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2172 struct getcpu_cache __user *, unused)
2173 {
2174 int err = 0;
2175 int cpu = raw_smp_processor_id();
2176 if (cpup)
2177 err |= put_user(cpu, cpup);
2178 if (nodep)
2179 err |= put_user(cpu_to_node(cpu), nodep);
2180 return err ? -EFAULT : 0;
2181 }
2182
2183 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
2184
2185 static void argv_cleanup(struct subprocess_info *info)
2186 {
2187 argv_free(info->argv);
2188 }
2189
2190 static int __orderly_poweroff(void)
2191 {
2192 int argc;
2193 char **argv;
2194 static char *envp[] = {
2195 "HOME=/",
2196 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
2197 NULL
2198 };
2199 int ret;
2200
2201 argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
2202 if (argv == NULL) {
2203 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
2204 __func__, poweroff_cmd);
2205 return -ENOMEM;
2206 }
2207
2208 ret = call_usermodehelper_fns(argv[0], argv, envp, UMH_WAIT_EXEC,
2209 NULL, argv_cleanup, NULL);
2210 if (ret == -ENOMEM)
2211 argv_free(argv);
2212
2213 return ret;
2214 }
2215
2216 /**
2217 * orderly_poweroff - Trigger an orderly system poweroff
2218 * @force: force poweroff if command execution fails
2219 *
2220 * This may be called from any context to trigger a system shutdown.
2221 * If the orderly shutdown fails, it will force an immediate shutdown.
2222 */
2223 int orderly_poweroff(bool force)
2224 {
2225 int ret = __orderly_poweroff();
2226
2227 if (ret && force) {
2228 printk(KERN_WARNING "Failed to start orderly shutdown: "
2229 "forcing the issue\n");
2230
2231 /*
2232 * I guess this should try to kick off some daemon to sync and
2233 * poweroff asap. Or not even bother syncing if we're doing an
2234 * emergency shutdown?
2235 */
2236 emergency_sync();
2237 kernel_power_off();
2238 }
2239
2240 return ret;
2241 }
2242 EXPORT_SYMBOL_GPL(orderly_poweroff);
Linus' kernel tree